ctl.c revision 273163
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl.c 273163 2014-10-16 08:42:17Z 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_debugconf_subpage debugconf_page_default = {
113	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
114	DBGCNF_SUBPAGE_CODE,		/* subpage */
115	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
116	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
117	DBGCNF_VERSION,			/* page_version */
118	{CTL_TIME_IO_DEFAULT_SECS>>8,
119	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
120};
121
122static struct copan_debugconf_subpage debugconf_page_changeable = {
123	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
124	DBGCNF_SUBPAGE_CODE,		/* subpage */
125	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
126	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
127	0,				/* page_version */
128	{0xff,0xff},			/* ctl_time_io_secs */
129};
130
131static struct scsi_da_rw_recovery_page rw_er_page_default = {
132	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
133	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
134	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
135	/*read_retry_count*/0,
136	/*correction_span*/0,
137	/*head_offset_count*/0,
138	/*data_strobe_offset_cnt*/0,
139	/*byte8*/0,
140	/*write_retry_count*/0,
141	/*reserved2*/0,
142	/*recovery_time_limit*/{0, 0},
143};
144
145static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
146	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
147	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
148	/*byte3*/0,
149	/*read_retry_count*/0,
150	/*correction_span*/0,
151	/*head_offset_count*/0,
152	/*data_strobe_offset_cnt*/0,
153	/*byte8*/0,
154	/*write_retry_count*/0,
155	/*reserved2*/0,
156	/*recovery_time_limit*/{0, 0},
157};
158
159static struct scsi_format_page format_page_default = {
160	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
161	/*page_length*/sizeof(struct scsi_format_page) - 2,
162	/*tracks_per_zone*/ {0, 0},
163	/*alt_sectors_per_zone*/ {0, 0},
164	/*alt_tracks_per_zone*/ {0, 0},
165	/*alt_tracks_per_lun*/ {0, 0},
166	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
167			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
168	/*bytes_per_sector*/ {0, 0},
169	/*interleave*/ {0, 0},
170	/*track_skew*/ {0, 0},
171	/*cylinder_skew*/ {0, 0},
172	/*flags*/ SFP_HSEC,
173	/*reserved*/ {0, 0, 0}
174};
175
176static struct scsi_format_page format_page_changeable = {
177	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
178	/*page_length*/sizeof(struct scsi_format_page) - 2,
179	/*tracks_per_zone*/ {0, 0},
180	/*alt_sectors_per_zone*/ {0, 0},
181	/*alt_tracks_per_zone*/ {0, 0},
182	/*alt_tracks_per_lun*/ {0, 0},
183	/*sectors_per_track*/ {0, 0},
184	/*bytes_per_sector*/ {0, 0},
185	/*interleave*/ {0, 0},
186	/*track_skew*/ {0, 0},
187	/*cylinder_skew*/ {0, 0},
188	/*flags*/ 0,
189	/*reserved*/ {0, 0, 0}
190};
191
192static struct scsi_rigid_disk_page rigid_disk_page_default = {
193	/*page_code*/SMS_RIGID_DISK_PAGE,
194	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
195	/*cylinders*/ {0, 0, 0},
196	/*heads*/ CTL_DEFAULT_HEADS,
197	/*start_write_precomp*/ {0, 0, 0},
198	/*start_reduced_current*/ {0, 0, 0},
199	/*step_rate*/ {0, 0},
200	/*landing_zone_cylinder*/ {0, 0, 0},
201	/*rpl*/ SRDP_RPL_DISABLED,
202	/*rotational_offset*/ 0,
203	/*reserved1*/ 0,
204	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
205			   CTL_DEFAULT_ROTATION_RATE & 0xff},
206	/*reserved2*/ {0, 0}
207};
208
209static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
210	/*page_code*/SMS_RIGID_DISK_PAGE,
211	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
212	/*cylinders*/ {0, 0, 0},
213	/*heads*/ 0,
214	/*start_write_precomp*/ {0, 0, 0},
215	/*start_reduced_current*/ {0, 0, 0},
216	/*step_rate*/ {0, 0},
217	/*landing_zone_cylinder*/ {0, 0, 0},
218	/*rpl*/ 0,
219	/*rotational_offset*/ 0,
220	/*reserved1*/ 0,
221	/*rotation_rate*/ {0, 0},
222	/*reserved2*/ {0, 0}
223};
224
225static struct scsi_caching_page caching_page_default = {
226	/*page_code*/SMS_CACHING_PAGE,
227	/*page_length*/sizeof(struct scsi_caching_page) - 2,
228	/*flags1*/ SCP_DISC | SCP_WCE,
229	/*ret_priority*/ 0,
230	/*disable_pf_transfer_len*/ {0xff, 0xff},
231	/*min_prefetch*/ {0, 0},
232	/*max_prefetch*/ {0xff, 0xff},
233	/*max_pf_ceiling*/ {0xff, 0xff},
234	/*flags2*/ 0,
235	/*cache_segments*/ 0,
236	/*cache_seg_size*/ {0, 0},
237	/*reserved*/ 0,
238	/*non_cache_seg_size*/ {0, 0, 0}
239};
240
241static struct scsi_caching_page caching_page_changeable = {
242	/*page_code*/SMS_CACHING_PAGE,
243	/*page_length*/sizeof(struct scsi_caching_page) - 2,
244	/*flags1*/ SCP_WCE | SCP_RCD,
245	/*ret_priority*/ 0,
246	/*disable_pf_transfer_len*/ {0, 0},
247	/*min_prefetch*/ {0, 0},
248	/*max_prefetch*/ {0, 0},
249	/*max_pf_ceiling*/ {0, 0},
250	/*flags2*/ 0,
251	/*cache_segments*/ 0,
252	/*cache_seg_size*/ {0, 0},
253	/*reserved*/ 0,
254	/*non_cache_seg_size*/ {0, 0, 0}
255};
256
257static struct scsi_control_page control_page_default = {
258	/*page_code*/SMS_CONTROL_MODE_PAGE,
259	/*page_length*/sizeof(struct scsi_control_page) - 2,
260	/*rlec*/0,
261	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
262	/*eca_and_aen*/0,
263	/*flags4*/SCP_TAS,
264	/*aen_holdoff_period*/{0, 0},
265	/*busy_timeout_period*/{0, 0},
266	/*extended_selftest_completion_time*/{0, 0}
267};
268
269static struct scsi_control_page control_page_changeable = {
270	/*page_code*/SMS_CONTROL_MODE_PAGE,
271	/*page_length*/sizeof(struct scsi_control_page) - 2,
272	/*rlec*/SCP_DSENSE,
273	/*queue_flags*/SCP_QUEUE_ALG_MASK,
274	/*eca_and_aen*/SCP_SWP,
275	/*flags4*/0,
276	/*aen_holdoff_period*/{0, 0},
277	/*busy_timeout_period*/{0, 0},
278	/*extended_selftest_completion_time*/{0, 0}
279};
280
281static struct scsi_info_exceptions_page ie_page_default = {
282	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
283	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
284	/*info_flags*/SIEP_FLAGS_DEXCPT,
285	/*mrie*/0,
286	/*interval_timer*/{0, 0, 0, 0},
287	/*report_count*/{0, 0, 0, 0}
288};
289
290static struct scsi_info_exceptions_page ie_page_changeable = {
291	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
292	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
293	/*info_flags*/0,
294	/*mrie*/0,
295	/*interval_timer*/{0, 0, 0, 0},
296	/*report_count*/{0, 0, 0, 0}
297};
298
299static struct scsi_logical_block_provisioning_page lbp_page_default = {
300	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
301	/*subpage_code*/0x02,
302	/*page_length*/{0, sizeof(struct scsi_logical_block_provisioning_page) - 4},
303	/*flags*/0,
304	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
305	/*descr*/{}
306};
307
308static struct scsi_logical_block_provisioning_page lbp_page_changeable = {
309	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
310	/*subpage_code*/0x02,
311	/*page_length*/{0, sizeof(struct scsi_logical_block_provisioning_page) - 4},
312	/*flags*/0,
313	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
314	/*descr*/{}
315};
316
317/*
318 * XXX KDM move these into the softc.
319 */
320static int rcv_sync_msg;
321static int persis_offset;
322static uint8_t ctl_pause_rtr;
323static int     ctl_is_single = 1;
324
325SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
326static int worker_threads = -1;
327SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
328    &worker_threads, 1, "Number of worker threads");
329static int ctl_debug = CTL_DEBUG_NONE;
330SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
331    &ctl_debug, 0, "Enabled debug flags");
332
333/*
334 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
335 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
336 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
337 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
338 */
339#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
340
341static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
342				  int param);
343static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
344static int ctl_init(void);
345void ctl_shutdown(void);
346static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
347static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
348static void ctl_ioctl_online(void *arg);
349static void ctl_ioctl_offline(void *arg);
350static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
351static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
352static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
353static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
354static int ctl_ioctl_submit_wait(union ctl_io *io);
355static void ctl_ioctl_datamove(union ctl_io *io);
356static void ctl_ioctl_done(union ctl_io *io);
357static void ctl_ioctl_hard_startstop_callback(void *arg,
358					      struct cfi_metatask *metatask);
359static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
360static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
361			      struct ctl_ooa *ooa_hdr,
362			      struct ctl_ooa_entry *kern_entries);
363static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
364		     struct thread *td);
365static uint32_t ctl_map_lun(int port_num, uint32_t lun);
366static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
367#ifdef unused
368static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
369				   uint32_t targ_target, uint32_t targ_lun,
370				   int can_wait);
371static void ctl_kfree_io(union ctl_io *io);
372#endif /* unused */
373static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
374			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
375static int ctl_free_lun(struct ctl_lun *lun);
376static void ctl_create_lun(struct ctl_be_lun *be_lun);
377/**
378static void ctl_failover_change_pages(struct ctl_softc *softc,
379				      struct ctl_scsiio *ctsio, int master);
380**/
381
382static int ctl_do_mode_select(union ctl_io *io);
383static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
384			   uint64_t res_key, uint64_t sa_res_key,
385			   uint8_t type, uint32_t residx,
386			   struct ctl_scsiio *ctsio,
387			   struct scsi_per_res_out *cdb,
388			   struct scsi_per_res_out_parms* param);
389static void ctl_pro_preempt_other(struct ctl_lun *lun,
390				  union ctl_ha_msg *msg);
391static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
392static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
394static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
395static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
396static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
397static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
398					 int alloc_len);
399static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
400					 int alloc_len);
401static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
402static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
403static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
404static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
405static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
406static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
407static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
408    union ctl_io *pending_io, union ctl_io *ooa_io);
409static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
410				union ctl_io *starting_io);
411static int ctl_check_blocked(struct ctl_lun *lun);
412static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
413				struct ctl_lun *lun,
414				const struct ctl_cmd_entry *entry,
415				struct ctl_scsiio *ctsio);
416//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
417static void ctl_failover(void);
418static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
419			       struct ctl_scsiio *ctsio);
420static int ctl_scsiio(struct ctl_scsiio *ctsio);
421
422static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
423static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
424			    ctl_ua_type ua_type);
425static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
426			 ctl_ua_type ua_type);
427static int ctl_abort_task(union ctl_io *io);
428static int ctl_abort_task_set(union ctl_io *io);
429static int ctl_i_t_nexus_reset(union ctl_io *io);
430static void ctl_run_task(union ctl_io *io);
431#ifdef CTL_IO_DELAY
432static void ctl_datamove_timer_wakeup(void *arg);
433static void ctl_done_timer_wakeup(void *arg);
434#endif /* CTL_IO_DELAY */
435
436static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
437static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
438static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
439static void ctl_datamove_remote_write(union ctl_io *io);
440static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
441static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
442static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
443static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
444				    ctl_ha_dt_cb callback);
445static void ctl_datamove_remote_read(union ctl_io *io);
446static void ctl_datamove_remote(union ctl_io *io);
447static int ctl_process_done(union ctl_io *io);
448static void ctl_lun_thread(void *arg);
449static void ctl_work_thread(void *arg);
450static void ctl_enqueue_incoming(union ctl_io *io);
451static void ctl_enqueue_rtr(union ctl_io *io);
452static void ctl_enqueue_done(union ctl_io *io);
453static void ctl_enqueue_isc(union ctl_io *io);
454static const struct ctl_cmd_entry *
455    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
456static const struct ctl_cmd_entry *
457    ctl_validate_command(struct ctl_scsiio *ctsio);
458static int ctl_cmd_applicable(uint8_t lun_type,
459    const struct ctl_cmd_entry *entry);
460
461/*
462 * Load the serialization table.  This isn't very pretty, but is probably
463 * the easiest way to do it.
464 */
465#include "ctl_ser_table.c"
466
467/*
468 * We only need to define open, close and ioctl routines for this driver.
469 */
470static struct cdevsw ctl_cdevsw = {
471	.d_version =	D_VERSION,
472	.d_flags =	0,
473	.d_open =	ctl_open,
474	.d_close =	ctl_close,
475	.d_ioctl =	ctl_ioctl,
476	.d_name =	"ctl",
477};
478
479
480MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
481MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
482
483static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
484
485static moduledata_t ctl_moduledata = {
486	"ctl",
487	ctl_module_event_handler,
488	NULL
489};
490
491DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
492MODULE_VERSION(ctl, 1);
493
494static struct ctl_frontend ioctl_frontend =
495{
496	.name = "ioctl",
497};
498
499static void
500ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
501			    union ctl_ha_msg *msg_info)
502{
503	struct ctl_scsiio *ctsio;
504
505	if (msg_info->hdr.original_sc == NULL) {
506		printf("%s: original_sc == NULL!\n", __func__);
507		/* XXX KDM now what? */
508		return;
509	}
510
511	ctsio = &msg_info->hdr.original_sc->scsiio;
512	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
513	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
514	ctsio->io_hdr.status = msg_info->hdr.status;
515	ctsio->scsi_status = msg_info->scsi.scsi_status;
516	ctsio->sense_len = msg_info->scsi.sense_len;
517	ctsio->sense_residual = msg_info->scsi.sense_residual;
518	ctsio->residual = msg_info->scsi.residual;
519	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
520	       sizeof(ctsio->sense_data));
521	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
522	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
523	ctl_enqueue_isc((union ctl_io *)ctsio);
524}
525
526static void
527ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
528				union ctl_ha_msg *msg_info)
529{
530	struct ctl_scsiio *ctsio;
531
532	if (msg_info->hdr.serializing_sc == NULL) {
533		printf("%s: serializing_sc == NULL!\n", __func__);
534		/* XXX KDM now what? */
535		return;
536	}
537
538	ctsio = &msg_info->hdr.serializing_sc->scsiio;
539#if 0
540	/*
541	 * Attempt to catch the situation where an I/O has
542	 * been freed, and we're using it again.
543	 */
544	if (ctsio->io_hdr.io_type == 0xff) {
545		union ctl_io *tmp_io;
546		tmp_io = (union ctl_io *)ctsio;
547		printf("%s: %p use after free!\n", __func__,
548		       ctsio);
549		printf("%s: type %d msg %d cdb %x iptl: "
550		       "%d:%d:%d:%d tag 0x%04x "
551		       "flag %#x status %x\n",
552			__func__,
553			tmp_io->io_hdr.io_type,
554			tmp_io->io_hdr.msg_type,
555			tmp_io->scsiio.cdb[0],
556			tmp_io->io_hdr.nexus.initid.id,
557			tmp_io->io_hdr.nexus.targ_port,
558			tmp_io->io_hdr.nexus.targ_target.id,
559			tmp_io->io_hdr.nexus.targ_lun,
560			(tmp_io->io_hdr.io_type ==
561			CTL_IO_TASK) ?
562			tmp_io->taskio.tag_num :
563			tmp_io->scsiio.tag_num,
564		        tmp_io->io_hdr.flags,
565			tmp_io->io_hdr.status);
566	}
567#endif
568	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
569	ctl_enqueue_isc((union ctl_io *)ctsio);
570}
571
572/*
573 * ISC (Inter Shelf Communication) event handler.  Events from the HA
574 * subsystem come in here.
575 */
576static void
577ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
578{
579	struct ctl_softc *ctl_softc;
580	union ctl_io *io;
581	struct ctl_prio *presio;
582	ctl_ha_status isc_status;
583
584	ctl_softc = control_softc;
585	io = NULL;
586
587
588#if 0
589	printf("CTL: Isc Msg event %d\n", event);
590#endif
591	if (event == CTL_HA_EVT_MSG_RECV) {
592		union ctl_ha_msg msg_info;
593
594		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
595					     sizeof(msg_info), /*wait*/ 0);
596#if 0
597		printf("CTL: msg_type %d\n", msg_info.msg_type);
598#endif
599		if (isc_status != 0) {
600			printf("Error receiving message, status = %d\n",
601			       isc_status);
602			return;
603		}
604
605		switch (msg_info.hdr.msg_type) {
606		case CTL_MSG_SERIALIZE:
607#if 0
608			printf("Serialize\n");
609#endif
610			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
611			if (io == NULL) {
612				printf("ctl_isc_event_handler: can't allocate "
613				       "ctl_io!\n");
614				/* Bad Juju */
615				/* Need to set busy and send msg back */
616				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
617				msg_info.hdr.status = CTL_SCSI_ERROR;
618				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
619				msg_info.scsi.sense_len = 0;
620			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
621				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
622				}
623				goto bailout;
624			}
625			ctl_zero_io(io);
626			// populate ctsio from msg_info
627			io->io_hdr.io_type = CTL_IO_SCSI;
628			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
629			io->io_hdr.original_sc = msg_info.hdr.original_sc;
630#if 0
631			printf("pOrig %x\n", (int)msg_info.original_sc);
632#endif
633			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
634					    CTL_FLAG_IO_ACTIVE;
635			/*
636			 * If we're in serialization-only mode, we don't
637			 * want to go through full done processing.  Thus
638			 * the COPY flag.
639			 *
640			 * XXX KDM add another flag that is more specific.
641			 */
642			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
643				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
644			io->io_hdr.nexus = msg_info.hdr.nexus;
645#if 0
646			printf("targ %d, port %d, iid %d, lun %d\n",
647			       io->io_hdr.nexus.targ_target.id,
648			       io->io_hdr.nexus.targ_port,
649			       io->io_hdr.nexus.initid.id,
650			       io->io_hdr.nexus.targ_lun);
651#endif
652			io->scsiio.tag_num = msg_info.scsi.tag_num;
653			io->scsiio.tag_type = msg_info.scsi.tag_type;
654			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
655			       CTL_MAX_CDBLEN);
656			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
657				const struct ctl_cmd_entry *entry;
658
659				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
660				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
661				io->io_hdr.flags |=
662					entry->flags & CTL_FLAG_DATA_MASK;
663			}
664			ctl_enqueue_isc(io);
665			break;
666
667		/* Performed on the Originating SC, XFER mode only */
668		case CTL_MSG_DATAMOVE: {
669			struct ctl_sg_entry *sgl;
670			int i, j;
671
672			io = msg_info.hdr.original_sc;
673			if (io == NULL) {
674				printf("%s: original_sc == NULL!\n", __func__);
675				/* XXX KDM do something here */
676				break;
677			}
678			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
679			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
680			/*
681			 * Keep track of this, we need to send it back over
682			 * when the datamove is complete.
683			 */
684			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
685
686			if (msg_info.dt.sg_sequence == 0) {
687				/*
688				 * XXX KDM we use the preallocated S/G list
689				 * here, but we'll need to change this to
690				 * dynamic allocation if we need larger S/G
691				 * lists.
692				 */
693				if (msg_info.dt.kern_sg_entries >
694				    sizeof(io->io_hdr.remote_sglist) /
695				    sizeof(io->io_hdr.remote_sglist[0])) {
696					printf("%s: number of S/G entries "
697					    "needed %u > allocated num %zd\n",
698					    __func__,
699					    msg_info.dt.kern_sg_entries,
700					    sizeof(io->io_hdr.remote_sglist)/
701					    sizeof(io->io_hdr.remote_sglist[0]));
702
703					/*
704					 * XXX KDM send a message back to
705					 * the other side to shut down the
706					 * DMA.  The error will come back
707					 * through via the normal channel.
708					 */
709					break;
710				}
711				sgl = io->io_hdr.remote_sglist;
712				memset(sgl, 0,
713				       sizeof(io->io_hdr.remote_sglist));
714
715				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
716
717				io->scsiio.kern_sg_entries =
718					msg_info.dt.kern_sg_entries;
719				io->scsiio.rem_sg_entries =
720					msg_info.dt.kern_sg_entries;
721				io->scsiio.kern_data_len =
722					msg_info.dt.kern_data_len;
723				io->scsiio.kern_total_len =
724					msg_info.dt.kern_total_len;
725				io->scsiio.kern_data_resid =
726					msg_info.dt.kern_data_resid;
727				io->scsiio.kern_rel_offset =
728					msg_info.dt.kern_rel_offset;
729				/*
730				 * Clear out per-DMA flags.
731				 */
732				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
733				/*
734				 * Add per-DMA flags that are set for this
735				 * particular DMA request.
736				 */
737				io->io_hdr.flags |= msg_info.dt.flags &
738						    CTL_FLAG_RDMA_MASK;
739			} else
740				sgl = (struct ctl_sg_entry *)
741					io->scsiio.kern_data_ptr;
742
743			for (i = msg_info.dt.sent_sg_entries, j = 0;
744			     i < (msg_info.dt.sent_sg_entries +
745			     msg_info.dt.cur_sg_entries); i++, j++) {
746				sgl[i].addr = msg_info.dt.sg_list[j].addr;
747				sgl[i].len = msg_info.dt.sg_list[j].len;
748
749#if 0
750				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
751				       __func__,
752				       msg_info.dt.sg_list[j].addr,
753				       msg_info.dt.sg_list[j].len,
754				       sgl[i].addr, sgl[i].len, j, i);
755#endif
756			}
757#if 0
758			memcpy(&sgl[msg_info.dt.sent_sg_entries],
759			       msg_info.dt.sg_list,
760			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
761#endif
762
763			/*
764			 * If this is the last piece of the I/O, we've got
765			 * the full S/G list.  Queue processing in the thread.
766			 * Otherwise wait for the next piece.
767			 */
768			if (msg_info.dt.sg_last != 0)
769				ctl_enqueue_isc(io);
770			break;
771		}
772		/* Performed on the Serializing (primary) SC, XFER mode only */
773		case CTL_MSG_DATAMOVE_DONE: {
774			if (msg_info.hdr.serializing_sc == NULL) {
775				printf("%s: serializing_sc == NULL!\n",
776				       __func__);
777				/* XXX KDM now what? */
778				break;
779			}
780			/*
781			 * We grab the sense information here in case
782			 * there was a failure, so we can return status
783			 * back to the initiator.
784			 */
785			io = msg_info.hdr.serializing_sc;
786			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
787			io->io_hdr.status = msg_info.hdr.status;
788			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
789			io->scsiio.sense_len = msg_info.scsi.sense_len;
790			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
791			io->io_hdr.port_status = msg_info.scsi.fetd_status;
792			io->scsiio.residual = msg_info.scsi.residual;
793			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
794			       sizeof(io->scsiio.sense_data));
795			ctl_enqueue_isc(io);
796			break;
797		}
798
799		/* Preformed on Originating SC, SER_ONLY mode */
800		case CTL_MSG_R2R:
801			io = msg_info.hdr.original_sc;
802			if (io == NULL) {
803				printf("%s: Major Bummer\n", __func__);
804				return;
805			} else {
806#if 0
807				printf("pOrig %x\n",(int) ctsio);
808#endif
809			}
810			io->io_hdr.msg_type = CTL_MSG_R2R;
811			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
812			ctl_enqueue_isc(io);
813			break;
814
815		/*
816		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
817		 * mode.
818		 * Performed on the Originating (i.e. secondary) SC in XFER
819		 * mode
820		 */
821		case CTL_MSG_FINISH_IO:
822			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
823				ctl_isc_handler_finish_xfer(ctl_softc,
824							    &msg_info);
825			else
826				ctl_isc_handler_finish_ser_only(ctl_softc,
827								&msg_info);
828			break;
829
830		/* Preformed on Originating SC */
831		case CTL_MSG_BAD_JUJU:
832			io = msg_info.hdr.original_sc;
833			if (io == NULL) {
834				printf("%s: Bad JUJU!, original_sc is NULL!\n",
835				       __func__);
836				break;
837			}
838			ctl_copy_sense_data(&msg_info, io);
839			/*
840			 * IO should have already been cleaned up on other
841			 * SC so clear this flag so we won't send a message
842			 * back to finish the IO there.
843			 */
844			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
845			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
846
847			/* io = msg_info.hdr.serializing_sc; */
848			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
849			ctl_enqueue_isc(io);
850			break;
851
852		/* Handle resets sent from the other side */
853		case CTL_MSG_MANAGE_TASKS: {
854			struct ctl_taskio *taskio;
855			taskio = (struct ctl_taskio *)ctl_alloc_io(
856				(void *)ctl_softc->othersc_pool);
857			if (taskio == NULL) {
858				printf("ctl_isc_event_handler: can't allocate "
859				       "ctl_io!\n");
860				/* Bad Juju */
861				/* should I just call the proper reset func
862				   here??? */
863				goto bailout;
864			}
865			ctl_zero_io((union ctl_io *)taskio);
866			taskio->io_hdr.io_type = CTL_IO_TASK;
867			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
868			taskio->io_hdr.nexus = msg_info.hdr.nexus;
869			taskio->task_action = msg_info.task.task_action;
870			taskio->tag_num = msg_info.task.tag_num;
871			taskio->tag_type = msg_info.task.tag_type;
872#ifdef CTL_TIME_IO
873			taskio->io_hdr.start_time = time_uptime;
874			getbintime(&taskio->io_hdr.start_bt);
875#if 0
876			cs_prof_gettime(&taskio->io_hdr.start_ticks);
877#endif
878#endif /* CTL_TIME_IO */
879			ctl_run_task((union ctl_io *)taskio);
880			break;
881		}
882		/* Persistent Reserve action which needs attention */
883		case CTL_MSG_PERS_ACTION:
884			presio = (struct ctl_prio *)ctl_alloc_io(
885				(void *)ctl_softc->othersc_pool);
886			if (presio == NULL) {
887				printf("ctl_isc_event_handler: can't allocate "
888				       "ctl_io!\n");
889				/* Bad Juju */
890				/* Need to set busy and send msg back */
891				goto bailout;
892			}
893			ctl_zero_io((union ctl_io *)presio);
894			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
895			presio->pr_msg = msg_info.pr;
896			ctl_enqueue_isc((union ctl_io *)presio);
897			break;
898		case CTL_MSG_SYNC_FE:
899			rcv_sync_msg = 1;
900			break;
901		default:
902		        printf("How did I get here?\n");
903		}
904	} else if (event == CTL_HA_EVT_MSG_SENT) {
905		if (param != CTL_HA_STATUS_SUCCESS) {
906			printf("Bad status from ctl_ha_msg_send status %d\n",
907			       param);
908		}
909		return;
910	} else if (event == CTL_HA_EVT_DISCONNECT) {
911		printf("CTL: Got a disconnect from Isc\n");
912		return;
913	} else {
914		printf("ctl_isc_event_handler: Unknown event %d\n", event);
915		return;
916	}
917
918bailout:
919	return;
920}
921
922static void
923ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
924{
925	struct scsi_sense_data *sense;
926
927	sense = &dest->scsiio.sense_data;
928	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
929	dest->scsiio.scsi_status = src->scsi.scsi_status;
930	dest->scsiio.sense_len = src->scsi.sense_len;
931	dest->io_hdr.status = src->hdr.status;
932}
933
934static int
935ctl_init(void)
936{
937	struct ctl_softc *softc;
938	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
939	struct ctl_port *port;
940        uint8_t sc_id =0;
941	int i, error, retval;
942	//int isc_retval;
943
944	retval = 0;
945	ctl_pause_rtr = 0;
946        rcv_sync_msg = 0;
947
948	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
949			       M_WAITOK | M_ZERO);
950	softc = control_softc;
951
952	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
953			      "cam/ctl");
954
955	softc->dev->si_drv1 = softc;
956
957	/*
958	 * By default, return a "bad LUN" peripheral qualifier for unknown
959	 * LUNs.  The user can override this default using the tunable or
960	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
961	 */
962	softc->inquiry_pq_no_lun = 1;
963	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
964			  &softc->inquiry_pq_no_lun);
965	sysctl_ctx_init(&softc->sysctl_ctx);
966	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
967		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
968		CTLFLAG_RD, 0, "CAM Target Layer");
969
970	if (softc->sysctl_tree == NULL) {
971		printf("%s: unable to allocate sysctl tree\n", __func__);
972		destroy_dev(softc->dev);
973		free(control_softc, M_DEVBUF);
974		control_softc = NULL;
975		return (ENOMEM);
976	}
977
978	SYSCTL_ADD_INT(&softc->sysctl_ctx,
979		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
980		       "inquiry_pq_no_lun", CTLFLAG_RW,
981		       &softc->inquiry_pq_no_lun, 0,
982		       "Report no lun possible for invalid LUNs");
983
984	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
985	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
986	softc->open_count = 0;
987
988	/*
989	 * Default to actually sending a SYNCHRONIZE CACHE command down to
990	 * the drive.
991	 */
992	softc->flags = CTL_FLAG_REAL_SYNC;
993
994	/*
995	 * In Copan's HA scheme, the "master" and "slave" roles are
996	 * figured out through the slot the controller is in.  Although it
997	 * is an active/active system, someone has to be in charge.
998 	 */
999#ifdef NEEDTOPORT
1000        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1001#endif
1002
1003        if (sc_id == 0) {
1004		softc->flags |= CTL_FLAG_MASTER_SHELF;
1005		persis_offset = 0;
1006	} else
1007		persis_offset = CTL_MAX_INITIATORS;
1008
1009	/*
1010	 * XXX KDM need to figure out where we want to get our target ID
1011	 * and WWID.  Is it different on each port?
1012	 */
1013	softc->target.id = 0;
1014	softc->target.wwid[0] = 0x12345678;
1015	softc->target.wwid[1] = 0x87654321;
1016	STAILQ_INIT(&softc->lun_list);
1017	STAILQ_INIT(&softc->pending_lun_queue);
1018	STAILQ_INIT(&softc->fe_list);
1019	STAILQ_INIT(&softc->port_list);
1020	STAILQ_INIT(&softc->be_list);
1021	STAILQ_INIT(&softc->io_pools);
1022	ctl_tpc_init(softc);
1023
1024	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1025			    &internal_pool)!= 0){
1026		printf("ctl: can't allocate %d entry internal pool, "
1027		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1028		return (ENOMEM);
1029	}
1030
1031	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1032			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1033		printf("ctl: can't allocate %d entry emergency pool, "
1034		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1035		ctl_pool_free(internal_pool);
1036		return (ENOMEM);
1037	}
1038
1039	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1040	                    &other_pool) != 0)
1041	{
1042		printf("ctl: can't allocate %d entry other SC pool, "
1043		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1044		ctl_pool_free(internal_pool);
1045		ctl_pool_free(emergency_pool);
1046		return (ENOMEM);
1047	}
1048
1049	softc->internal_pool = internal_pool;
1050	softc->emergency_pool = emergency_pool;
1051	softc->othersc_pool = other_pool;
1052
1053	if (worker_threads <= 0)
1054		worker_threads = max(1, mp_ncpus / 4);
1055	if (worker_threads > CTL_MAX_THREADS)
1056		worker_threads = CTL_MAX_THREADS;
1057
1058	for (i = 0; i < worker_threads; i++) {
1059		struct ctl_thread *thr = &softc->threads[i];
1060
1061		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1062		thr->ctl_softc = softc;
1063		STAILQ_INIT(&thr->incoming_queue);
1064		STAILQ_INIT(&thr->rtr_queue);
1065		STAILQ_INIT(&thr->done_queue);
1066		STAILQ_INIT(&thr->isc_queue);
1067
1068		error = kproc_kthread_add(ctl_work_thread, thr,
1069		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1070		if (error != 0) {
1071			printf("error creating CTL work thread!\n");
1072			ctl_pool_free(internal_pool);
1073			ctl_pool_free(emergency_pool);
1074			ctl_pool_free(other_pool);
1075			return (error);
1076		}
1077	}
1078	error = kproc_kthread_add(ctl_lun_thread, softc,
1079	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1080	if (error != 0) {
1081		printf("error creating CTL lun thread!\n");
1082		ctl_pool_free(internal_pool);
1083		ctl_pool_free(emergency_pool);
1084		ctl_pool_free(other_pool);
1085		return (error);
1086	}
1087	if (bootverbose)
1088		printf("ctl: CAM Target Layer loaded\n");
1089
1090	/*
1091	 * Initialize the ioctl front end.
1092	 */
1093	ctl_frontend_register(&ioctl_frontend);
1094	port = &softc->ioctl_info.port;
1095	port->frontend = &ioctl_frontend;
1096	sprintf(softc->ioctl_info.port_name, "ioctl");
1097	port->port_type = CTL_PORT_IOCTL;
1098	port->num_requested_ctl_io = 100;
1099	port->port_name = softc->ioctl_info.port_name;
1100	port->port_online = ctl_ioctl_online;
1101	port->port_offline = ctl_ioctl_offline;
1102	port->onoff_arg = &softc->ioctl_info;
1103	port->lun_enable = ctl_ioctl_lun_enable;
1104	port->lun_disable = ctl_ioctl_lun_disable;
1105	port->targ_lun_arg = &softc->ioctl_info;
1106	port->fe_datamove = ctl_ioctl_datamove;
1107	port->fe_done = ctl_ioctl_done;
1108	port->max_targets = 15;
1109	port->max_target_id = 15;
1110
1111	if (ctl_port_register(&softc->ioctl_info.port,
1112	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1113		printf("ctl: ioctl front end registration failed, will "
1114		       "continue anyway\n");
1115	}
1116
1117#ifdef CTL_IO_DELAY
1118	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1119		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1120		       sizeof(struct callout), CTL_TIMER_BYTES);
1121		return (EINVAL);
1122	}
1123#endif /* CTL_IO_DELAY */
1124
1125	return (0);
1126}
1127
1128void
1129ctl_shutdown(void)
1130{
1131	struct ctl_softc *softc;
1132	struct ctl_lun *lun, *next_lun;
1133	struct ctl_io_pool *pool;
1134
1135	softc = (struct ctl_softc *)control_softc;
1136
1137	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1138		printf("ctl: ioctl front end deregistration failed\n");
1139
1140	mtx_lock(&softc->ctl_lock);
1141
1142	/*
1143	 * Free up each LUN.
1144	 */
1145	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1146		next_lun = STAILQ_NEXT(lun, links);
1147		ctl_free_lun(lun);
1148	}
1149
1150	mtx_unlock(&softc->ctl_lock);
1151
1152	ctl_frontend_deregister(&ioctl_frontend);
1153
1154	/*
1155	 * This will rip the rug out from under any FETDs or anyone else
1156	 * that has a pool allocated.  Since we increment our module
1157	 * refcount any time someone outside the main CTL module allocates
1158	 * a pool, we shouldn't have any problems here.  The user won't be
1159	 * able to unload the CTL module until client modules have
1160	 * successfully unloaded.
1161	 */
1162	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1163		ctl_pool_free(pool);
1164
1165#if 0
1166	ctl_shutdown_thread(softc->work_thread);
1167	mtx_destroy(&softc->queue_lock);
1168#endif
1169
1170	ctl_tpc_shutdown(softc);
1171	mtx_destroy(&softc->pool_lock);
1172	mtx_destroy(&softc->ctl_lock);
1173
1174	destroy_dev(softc->dev);
1175
1176	sysctl_ctx_free(&softc->sysctl_ctx);
1177
1178	free(control_softc, M_DEVBUF);
1179	control_softc = NULL;
1180
1181	if (bootverbose)
1182		printf("ctl: CAM Target Layer unloaded\n");
1183}
1184
1185static int
1186ctl_module_event_handler(module_t mod, int what, void *arg)
1187{
1188
1189	switch (what) {
1190	case MOD_LOAD:
1191		return (ctl_init());
1192	case MOD_UNLOAD:
1193		return (EBUSY);
1194	default:
1195		return (EOPNOTSUPP);
1196	}
1197}
1198
1199/*
1200 * XXX KDM should we do some access checks here?  Bump a reference count to
1201 * prevent a CTL module from being unloaded while someone has it open?
1202 */
1203static int
1204ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1205{
1206	return (0);
1207}
1208
1209static int
1210ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1211{
1212	return (0);
1213}
1214
1215int
1216ctl_port_enable(ctl_port_type port_type)
1217{
1218	struct ctl_softc *softc;
1219	struct ctl_port *port;
1220
1221	if (ctl_is_single == 0) {
1222		union ctl_ha_msg msg_info;
1223		int isc_retval;
1224
1225#if 0
1226		printf("%s: HA mode, synchronizing frontend enable\n",
1227		        __func__);
1228#endif
1229		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1230	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1231		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1232			printf("Sync msg send error retval %d\n", isc_retval);
1233		}
1234		if (!rcv_sync_msg) {
1235			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1236			        sizeof(msg_info), 1);
1237		}
1238#if 0
1239        	printf("CTL:Frontend Enable\n");
1240	} else {
1241		printf("%s: single mode, skipping frontend synchronization\n",
1242		        __func__);
1243#endif
1244	}
1245
1246	softc = control_softc;
1247
1248	STAILQ_FOREACH(port, &softc->port_list, links) {
1249		if (port_type & port->port_type)
1250		{
1251#if 0
1252			printf("port %d\n", port->targ_port);
1253#endif
1254			ctl_port_online(port);
1255		}
1256	}
1257
1258	return (0);
1259}
1260
1261int
1262ctl_port_disable(ctl_port_type port_type)
1263{
1264	struct ctl_softc *softc;
1265	struct ctl_port *port;
1266
1267	softc = control_softc;
1268
1269	STAILQ_FOREACH(port, &softc->port_list, links) {
1270		if (port_type & port->port_type)
1271			ctl_port_offline(port);
1272	}
1273
1274	return (0);
1275}
1276
1277/*
1278 * Returns 0 for success, 1 for failure.
1279 * Currently the only failure mode is if there aren't enough entries
1280 * allocated.  So, in case of a failure, look at num_entries_dropped,
1281 * reallocate and try again.
1282 */
1283int
1284ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1285	      int *num_entries_filled, int *num_entries_dropped,
1286	      ctl_port_type port_type, int no_virtual)
1287{
1288	struct ctl_softc *softc;
1289	struct ctl_port *port;
1290	int entries_dropped, entries_filled;
1291	int retval;
1292	int i;
1293
1294	softc = control_softc;
1295
1296	retval = 0;
1297	entries_filled = 0;
1298	entries_dropped = 0;
1299
1300	i = 0;
1301	mtx_lock(&softc->ctl_lock);
1302	STAILQ_FOREACH(port, &softc->port_list, links) {
1303		struct ctl_port_entry *entry;
1304
1305		if ((port->port_type & port_type) == 0)
1306			continue;
1307
1308		if ((no_virtual != 0)
1309		 && (port->virtual_port != 0))
1310			continue;
1311
1312		if (entries_filled >= num_entries_alloced) {
1313			entries_dropped++;
1314			continue;
1315		}
1316		entry = &entries[i];
1317
1318		entry->port_type = port->port_type;
1319		strlcpy(entry->port_name, port->port_name,
1320			sizeof(entry->port_name));
1321		entry->physical_port = port->physical_port;
1322		entry->virtual_port = port->virtual_port;
1323		entry->wwnn = port->wwnn;
1324		entry->wwpn = port->wwpn;
1325
1326		i++;
1327		entries_filled++;
1328	}
1329
1330	mtx_unlock(&softc->ctl_lock);
1331
1332	if (entries_dropped > 0)
1333		retval = 1;
1334
1335	*num_entries_dropped = entries_dropped;
1336	*num_entries_filled = entries_filled;
1337
1338	return (retval);
1339}
1340
1341static void
1342ctl_ioctl_online(void *arg)
1343{
1344	struct ctl_ioctl_info *ioctl_info;
1345
1346	ioctl_info = (struct ctl_ioctl_info *)arg;
1347
1348	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1349}
1350
1351static void
1352ctl_ioctl_offline(void *arg)
1353{
1354	struct ctl_ioctl_info *ioctl_info;
1355
1356	ioctl_info = (struct ctl_ioctl_info *)arg;
1357
1358	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1359}
1360
1361/*
1362 * Remove an initiator by port number and initiator ID.
1363 * Returns 0 for success, -1 for failure.
1364 */
1365int
1366ctl_remove_initiator(struct ctl_port *port, int iid)
1367{
1368	struct ctl_softc *softc = control_softc;
1369
1370	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1371
1372	if (iid > CTL_MAX_INIT_PER_PORT) {
1373		printf("%s: initiator ID %u > maximun %u!\n",
1374		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1375		return (-1);
1376	}
1377
1378	mtx_lock(&softc->ctl_lock);
1379	port->wwpn_iid[iid].in_use--;
1380	port->wwpn_iid[iid].last_use = time_uptime;
1381	mtx_unlock(&softc->ctl_lock);
1382
1383	return (0);
1384}
1385
1386/*
1387 * Add an initiator to the initiator map.
1388 * Returns iid for success, < 0 for failure.
1389 */
1390int
1391ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1392{
1393	struct ctl_softc *softc = control_softc;
1394	time_t best_time;
1395	int i, best;
1396
1397	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1398
1399	if (iid >= CTL_MAX_INIT_PER_PORT) {
1400		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1401		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1402		free(name, M_CTL);
1403		return (-1);
1404	}
1405
1406	mtx_lock(&softc->ctl_lock);
1407
1408	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1409		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1410			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1411				iid = i;
1412				break;
1413			}
1414			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1415			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1416				iid = i;
1417				break;
1418			}
1419		}
1420	}
1421
1422	if (iid < 0) {
1423		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1424			if (port->wwpn_iid[i].in_use == 0 &&
1425			    port->wwpn_iid[i].wwpn == 0 &&
1426			    port->wwpn_iid[i].name == NULL) {
1427				iid = i;
1428				break;
1429			}
1430		}
1431	}
1432
1433	if (iid < 0) {
1434		best = -1;
1435		best_time = INT32_MAX;
1436		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1437			if (port->wwpn_iid[i].in_use == 0) {
1438				if (port->wwpn_iid[i].last_use < best_time) {
1439					best = i;
1440					best_time = port->wwpn_iid[i].last_use;
1441				}
1442			}
1443		}
1444		iid = best;
1445	}
1446
1447	if (iid < 0) {
1448		mtx_unlock(&softc->ctl_lock);
1449		free(name, M_CTL);
1450		return (-2);
1451	}
1452
1453	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1454		/*
1455		 * This is not an error yet.
1456		 */
1457		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1458#if 0
1459			printf("%s: port %d iid %u WWPN %#jx arrived"
1460			    " again\n", __func__, port->targ_port,
1461			    iid, (uintmax_t)wwpn);
1462#endif
1463			goto take;
1464		}
1465		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1466		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1467#if 0
1468			printf("%s: port %d iid %u name '%s' arrived"
1469			    " again\n", __func__, port->targ_port,
1470			    iid, name);
1471#endif
1472			goto take;
1473		}
1474
1475		/*
1476		 * This is an error, but what do we do about it?  The
1477		 * driver is telling us we have a new WWPN for this
1478		 * initiator ID, so we pretty much need to use it.
1479		 */
1480		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1481		    " but WWPN %#jx '%s' is still at that address\n",
1482		    __func__, port->targ_port, iid, wwpn, name,
1483		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1484		    port->wwpn_iid[iid].name);
1485
1486		/*
1487		 * XXX KDM clear have_ca and ua_pending on each LUN for
1488		 * this initiator.
1489		 */
1490	}
1491take:
1492	free(port->wwpn_iid[iid].name, M_CTL);
1493	port->wwpn_iid[iid].name = name;
1494	port->wwpn_iid[iid].wwpn = wwpn;
1495	port->wwpn_iid[iid].in_use++;
1496	mtx_unlock(&softc->ctl_lock);
1497
1498	return (iid);
1499}
1500
1501static int
1502ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1503{
1504	int len;
1505
1506	switch (port->port_type) {
1507	case CTL_PORT_FC:
1508	{
1509		struct scsi_transportid_fcp *id =
1510		    (struct scsi_transportid_fcp *)buf;
1511		if (port->wwpn_iid[iid].wwpn == 0)
1512			return (0);
1513		memset(id, 0, sizeof(*id));
1514		id->format_protocol = SCSI_PROTO_FC;
1515		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1516		return (sizeof(*id));
1517	}
1518	case CTL_PORT_ISCSI:
1519	{
1520		struct scsi_transportid_iscsi_port *id =
1521		    (struct scsi_transportid_iscsi_port *)buf;
1522		if (port->wwpn_iid[iid].name == NULL)
1523			return (0);
1524		memset(id, 0, 256);
1525		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1526		    SCSI_PROTO_ISCSI;
1527		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1528		len = roundup2(min(len, 252), 4);
1529		scsi_ulto2b(len, id->additional_length);
1530		return (sizeof(*id) + len);
1531	}
1532	case CTL_PORT_SAS:
1533	{
1534		struct scsi_transportid_sas *id =
1535		    (struct scsi_transportid_sas *)buf;
1536		if (port->wwpn_iid[iid].wwpn == 0)
1537			return (0);
1538		memset(id, 0, sizeof(*id));
1539		id->format_protocol = SCSI_PROTO_SAS;
1540		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1541		return (sizeof(*id));
1542	}
1543	default:
1544	{
1545		struct scsi_transportid_spi *id =
1546		    (struct scsi_transportid_spi *)buf;
1547		memset(id, 0, sizeof(*id));
1548		id->format_protocol = SCSI_PROTO_SPI;
1549		scsi_ulto2b(iid, id->scsi_addr);
1550		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1551		return (sizeof(*id));
1552	}
1553	}
1554}
1555
1556static int
1557ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1558{
1559	return (0);
1560}
1561
1562static int
1563ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1564{
1565	return (0);
1566}
1567
1568/*
1569 * Data movement routine for the CTL ioctl frontend port.
1570 */
1571static int
1572ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1573{
1574	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1575	struct ctl_sg_entry ext_entry, kern_entry;
1576	int ext_sglen, ext_sg_entries, kern_sg_entries;
1577	int ext_sg_start, ext_offset;
1578	int len_to_copy, len_copied;
1579	int kern_watermark, ext_watermark;
1580	int ext_sglist_malloced;
1581	int i, j;
1582
1583	ext_sglist_malloced = 0;
1584	ext_sg_start = 0;
1585	ext_offset = 0;
1586
1587	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1588
1589	/*
1590	 * If this flag is set, fake the data transfer.
1591	 */
1592	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1593		ctsio->ext_data_filled = ctsio->ext_data_len;
1594		goto bailout;
1595	}
1596
1597	/*
1598	 * To simplify things here, if we have a single buffer, stick it in
1599	 * a S/G entry and just make it a single entry S/G list.
1600	 */
1601	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1602		int len_seen;
1603
1604		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1605
1606		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1607							   M_WAITOK);
1608		ext_sglist_malloced = 1;
1609		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1610				   ext_sglen) != 0) {
1611			ctl_set_internal_failure(ctsio,
1612						 /*sks_valid*/ 0,
1613						 /*retry_count*/ 0);
1614			goto bailout;
1615		}
1616		ext_sg_entries = ctsio->ext_sg_entries;
1617		len_seen = 0;
1618		for (i = 0; i < ext_sg_entries; i++) {
1619			if ((len_seen + ext_sglist[i].len) >=
1620			     ctsio->ext_data_filled) {
1621				ext_sg_start = i;
1622				ext_offset = ctsio->ext_data_filled - len_seen;
1623				break;
1624			}
1625			len_seen += ext_sglist[i].len;
1626		}
1627	} else {
1628		ext_sglist = &ext_entry;
1629		ext_sglist->addr = ctsio->ext_data_ptr;
1630		ext_sglist->len = ctsio->ext_data_len;
1631		ext_sg_entries = 1;
1632		ext_sg_start = 0;
1633		ext_offset = ctsio->ext_data_filled;
1634	}
1635
1636	if (ctsio->kern_sg_entries > 0) {
1637		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1638		kern_sg_entries = ctsio->kern_sg_entries;
1639	} else {
1640		kern_sglist = &kern_entry;
1641		kern_sglist->addr = ctsio->kern_data_ptr;
1642		kern_sglist->len = ctsio->kern_data_len;
1643		kern_sg_entries = 1;
1644	}
1645
1646
1647	kern_watermark = 0;
1648	ext_watermark = ext_offset;
1649	len_copied = 0;
1650	for (i = ext_sg_start, j = 0;
1651	     i < ext_sg_entries && j < kern_sg_entries;) {
1652		uint8_t *ext_ptr, *kern_ptr;
1653
1654		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1655				      kern_sglist[j].len - kern_watermark);
1656
1657		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1658		ext_ptr = ext_ptr + ext_watermark;
1659		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1660			/*
1661			 * XXX KDM fix this!
1662			 */
1663			panic("need to implement bus address support");
1664#if 0
1665			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1666#endif
1667		} else
1668			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1669		kern_ptr = kern_ptr + kern_watermark;
1670
1671		kern_watermark += len_to_copy;
1672		ext_watermark += len_to_copy;
1673
1674		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1675		     CTL_FLAG_DATA_IN) {
1676			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1677					 "bytes to user\n", len_to_copy));
1678			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1679					 "to %p\n", kern_ptr, ext_ptr));
1680			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1681				ctl_set_internal_failure(ctsio,
1682							 /*sks_valid*/ 0,
1683							 /*retry_count*/ 0);
1684				goto bailout;
1685			}
1686		} else {
1687			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1688					 "bytes from user\n", len_to_copy));
1689			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1690					 "to %p\n", ext_ptr, kern_ptr));
1691			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1692				ctl_set_internal_failure(ctsio,
1693							 /*sks_valid*/ 0,
1694							 /*retry_count*/0);
1695				goto bailout;
1696			}
1697		}
1698
1699		len_copied += len_to_copy;
1700
1701		if (ext_sglist[i].len == ext_watermark) {
1702			i++;
1703			ext_watermark = 0;
1704		}
1705
1706		if (kern_sglist[j].len == kern_watermark) {
1707			j++;
1708			kern_watermark = 0;
1709		}
1710	}
1711
1712	ctsio->ext_data_filled += len_copied;
1713
1714	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1715			 "kern_sg_entries: %d\n", ext_sg_entries,
1716			 kern_sg_entries));
1717	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1718			 "kern_data_len = %d\n", ctsio->ext_data_len,
1719			 ctsio->kern_data_len));
1720
1721
1722	/* XXX KDM set residual?? */
1723bailout:
1724
1725	if (ext_sglist_malloced != 0)
1726		free(ext_sglist, M_CTL);
1727
1728	return (CTL_RETVAL_COMPLETE);
1729}
1730
1731/*
1732 * Serialize a command that went down the "wrong" side, and so was sent to
1733 * this controller for execution.  The logic is a little different than the
1734 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1735 * sent back to the other side, but in the success case, we execute the
1736 * command on this side (XFER mode) or tell the other side to execute it
1737 * (SER_ONLY mode).
1738 */
1739static int
1740ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1741{
1742	struct ctl_softc *ctl_softc;
1743	union ctl_ha_msg msg_info;
1744	struct ctl_lun *lun;
1745	int retval = 0;
1746	uint32_t targ_lun;
1747
1748	ctl_softc = control_softc;
1749
1750	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1751	lun = ctl_softc->ctl_luns[targ_lun];
1752	if (lun==NULL)
1753	{
1754		/*
1755		 * Why isn't LUN defined? The other side wouldn't
1756		 * send a cmd if the LUN is undefined.
1757		 */
1758		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1759
1760		/* "Logical unit not supported" */
1761		ctl_set_sense_data(&msg_info.scsi.sense_data,
1762				   lun,
1763				   /*sense_format*/SSD_TYPE_NONE,
1764				   /*current_error*/ 1,
1765				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1766				   /*asc*/ 0x25,
1767				   /*ascq*/ 0x00,
1768				   SSD_ELEM_NONE);
1769
1770		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1771		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1772		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1773		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1774		msg_info.hdr.serializing_sc = NULL;
1775		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1776	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1777				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1778		}
1779		return(1);
1780
1781	}
1782
1783	mtx_lock(&lun->lun_lock);
1784    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1785
1786	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1787		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1788		 ooa_links))) {
1789	case CTL_ACTION_BLOCK:
1790		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1791		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1792				  blocked_links);
1793		break;
1794	case CTL_ACTION_PASS:
1795	case CTL_ACTION_SKIP:
1796		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1797			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1798			ctl_enqueue_rtr((union ctl_io *)ctsio);
1799		} else {
1800
1801			/* send msg back to other side */
1802			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1803			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1804			msg_info.hdr.msg_type = CTL_MSG_R2R;
1805#if 0
1806			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1807#endif
1808		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1809			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1810			}
1811		}
1812		break;
1813	case CTL_ACTION_OVERLAP:
1814		/* OVERLAPPED COMMANDS ATTEMPTED */
1815		ctl_set_sense_data(&msg_info.scsi.sense_data,
1816				   lun,
1817				   /*sense_format*/SSD_TYPE_NONE,
1818				   /*current_error*/ 1,
1819				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1820				   /*asc*/ 0x4E,
1821				   /*ascq*/ 0x00,
1822				   SSD_ELEM_NONE);
1823
1824		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1825		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1826		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1827		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1828		msg_info.hdr.serializing_sc = NULL;
1829		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1830#if 0
1831		printf("BAD JUJU:Major Bummer Overlap\n");
1832#endif
1833		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1834		retval = 1;
1835		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1836		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1837		}
1838		break;
1839	case CTL_ACTION_OVERLAP_TAG:
1840		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1841		ctl_set_sense_data(&msg_info.scsi.sense_data,
1842				   lun,
1843				   /*sense_format*/SSD_TYPE_NONE,
1844				   /*current_error*/ 1,
1845				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1846				   /*asc*/ 0x4D,
1847				   /*ascq*/ ctsio->tag_num & 0xff,
1848				   SSD_ELEM_NONE);
1849
1850		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1851		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1852		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1853		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1854		msg_info.hdr.serializing_sc = NULL;
1855		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1856#if 0
1857		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1858#endif
1859		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1860		retval = 1;
1861		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1862		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1863		}
1864		break;
1865	case CTL_ACTION_ERROR:
1866	default:
1867		/* "Internal target failure" */
1868		ctl_set_sense_data(&msg_info.scsi.sense_data,
1869				   lun,
1870				   /*sense_format*/SSD_TYPE_NONE,
1871				   /*current_error*/ 1,
1872				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1873				   /*asc*/ 0x44,
1874				   /*ascq*/ 0x00,
1875				   SSD_ELEM_NONE);
1876
1877		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1878		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1879		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1880		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1881		msg_info.hdr.serializing_sc = NULL;
1882		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1883#if 0
1884		printf("BAD JUJU:Major Bummer HW Error\n");
1885#endif
1886		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1887		retval = 1;
1888		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1889		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1890		}
1891		break;
1892	}
1893	mtx_unlock(&lun->lun_lock);
1894	return (retval);
1895}
1896
1897static int
1898ctl_ioctl_submit_wait(union ctl_io *io)
1899{
1900	struct ctl_fe_ioctl_params params;
1901	ctl_fe_ioctl_state last_state;
1902	int done, retval;
1903
1904	retval = 0;
1905
1906	bzero(&params, sizeof(params));
1907
1908	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1909	cv_init(&params.sem, "ctlioccv");
1910	params.state = CTL_IOCTL_INPROG;
1911	last_state = params.state;
1912
1913	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1914
1915	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1916
1917	/* This shouldn't happen */
1918	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1919		return (retval);
1920
1921	done = 0;
1922
1923	do {
1924		mtx_lock(&params.ioctl_mtx);
1925		/*
1926		 * Check the state here, and don't sleep if the state has
1927		 * already changed (i.e. wakeup has already occured, but we
1928		 * weren't waiting yet).
1929		 */
1930		if (params.state == last_state) {
1931			/* XXX KDM cv_wait_sig instead? */
1932			cv_wait(&params.sem, &params.ioctl_mtx);
1933		}
1934		last_state = params.state;
1935
1936		switch (params.state) {
1937		case CTL_IOCTL_INPROG:
1938			/* Why did we wake up? */
1939			/* XXX KDM error here? */
1940			mtx_unlock(&params.ioctl_mtx);
1941			break;
1942		case CTL_IOCTL_DATAMOVE:
1943			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1944
1945			/*
1946			 * change last_state back to INPROG to avoid
1947			 * deadlock on subsequent data moves.
1948			 */
1949			params.state = last_state = CTL_IOCTL_INPROG;
1950
1951			mtx_unlock(&params.ioctl_mtx);
1952			ctl_ioctl_do_datamove(&io->scsiio);
1953			/*
1954			 * Note that in some cases, most notably writes,
1955			 * this will queue the I/O and call us back later.
1956			 * In other cases, generally reads, this routine
1957			 * will immediately call back and wake us up,
1958			 * probably using our own context.
1959			 */
1960			io->scsiio.be_move_done(io);
1961			break;
1962		case CTL_IOCTL_DONE:
1963			mtx_unlock(&params.ioctl_mtx);
1964			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1965			done = 1;
1966			break;
1967		default:
1968			mtx_unlock(&params.ioctl_mtx);
1969			/* XXX KDM error here? */
1970			break;
1971		}
1972	} while (done == 0);
1973
1974	mtx_destroy(&params.ioctl_mtx);
1975	cv_destroy(&params.sem);
1976
1977	return (CTL_RETVAL_COMPLETE);
1978}
1979
1980static void
1981ctl_ioctl_datamove(union ctl_io *io)
1982{
1983	struct ctl_fe_ioctl_params *params;
1984
1985	params = (struct ctl_fe_ioctl_params *)
1986		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1987
1988	mtx_lock(&params->ioctl_mtx);
1989	params->state = CTL_IOCTL_DATAMOVE;
1990	cv_broadcast(&params->sem);
1991	mtx_unlock(&params->ioctl_mtx);
1992}
1993
1994static void
1995ctl_ioctl_done(union ctl_io *io)
1996{
1997	struct ctl_fe_ioctl_params *params;
1998
1999	params = (struct ctl_fe_ioctl_params *)
2000		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2001
2002	mtx_lock(&params->ioctl_mtx);
2003	params->state = CTL_IOCTL_DONE;
2004	cv_broadcast(&params->sem);
2005	mtx_unlock(&params->ioctl_mtx);
2006}
2007
2008static void
2009ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2010{
2011	struct ctl_fe_ioctl_startstop_info *sd_info;
2012
2013	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2014
2015	sd_info->hs_info.status = metatask->status;
2016	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2017	sd_info->hs_info.luns_complete =
2018		metatask->taskinfo.startstop.luns_complete;
2019	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2020
2021	cv_broadcast(&sd_info->sem);
2022}
2023
2024static void
2025ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2026{
2027	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2028
2029	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2030
2031	mtx_lock(fe_bbr_info->lock);
2032	fe_bbr_info->bbr_info->status = metatask->status;
2033	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2034	fe_bbr_info->wakeup_done = 1;
2035	mtx_unlock(fe_bbr_info->lock);
2036
2037	cv_broadcast(&fe_bbr_info->sem);
2038}
2039
2040/*
2041 * Returns 0 for success, errno for failure.
2042 */
2043static int
2044ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2045		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2046{
2047	union ctl_io *io;
2048	int retval;
2049
2050	retval = 0;
2051
2052	mtx_lock(&lun->lun_lock);
2053	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2054	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2055	     ooa_links)) {
2056		struct ctl_ooa_entry *entry;
2057
2058		/*
2059		 * If we've got more than we can fit, just count the
2060		 * remaining entries.
2061		 */
2062		if (*cur_fill_num >= ooa_hdr->alloc_num)
2063			continue;
2064
2065		entry = &kern_entries[*cur_fill_num];
2066
2067		entry->tag_num = io->scsiio.tag_num;
2068		entry->lun_num = lun->lun;
2069#ifdef CTL_TIME_IO
2070		entry->start_bt = io->io_hdr.start_bt;
2071#endif
2072		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2073		entry->cdb_len = io->scsiio.cdb_len;
2074		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2075			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2076
2077		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2078			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2079
2080		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2081			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2082
2083		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2084			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2085
2086		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2087			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2088	}
2089	mtx_unlock(&lun->lun_lock);
2090
2091	return (retval);
2092}
2093
2094static void *
2095ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2096		 size_t error_str_len)
2097{
2098	void *kptr;
2099
2100	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2101
2102	if (copyin(user_addr, kptr, len) != 0) {
2103		snprintf(error_str, error_str_len, "Error copying %d bytes "
2104			 "from user address %p to kernel address %p", len,
2105			 user_addr, kptr);
2106		free(kptr, M_CTL);
2107		return (NULL);
2108	}
2109
2110	return (kptr);
2111}
2112
2113static void
2114ctl_free_args(int num_args, struct ctl_be_arg *args)
2115{
2116	int i;
2117
2118	if (args == NULL)
2119		return;
2120
2121	for (i = 0; i < num_args; i++) {
2122		free(args[i].kname, M_CTL);
2123		free(args[i].kvalue, M_CTL);
2124	}
2125
2126	free(args, M_CTL);
2127}
2128
2129static struct ctl_be_arg *
2130ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2131		char *error_str, size_t error_str_len)
2132{
2133	struct ctl_be_arg *args;
2134	int i;
2135
2136	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2137				error_str, error_str_len);
2138
2139	if (args == NULL)
2140		goto bailout;
2141
2142	for (i = 0; i < num_args; i++) {
2143		args[i].kname = NULL;
2144		args[i].kvalue = NULL;
2145	}
2146
2147	for (i = 0; i < num_args; i++) {
2148		uint8_t *tmpptr;
2149
2150		args[i].kname = ctl_copyin_alloc(args[i].name,
2151			args[i].namelen, error_str, error_str_len);
2152		if (args[i].kname == NULL)
2153			goto bailout;
2154
2155		if (args[i].kname[args[i].namelen - 1] != '\0') {
2156			snprintf(error_str, error_str_len, "Argument %d "
2157				 "name is not NUL-terminated", i);
2158			goto bailout;
2159		}
2160
2161		if (args[i].flags & CTL_BEARG_RD) {
2162			tmpptr = ctl_copyin_alloc(args[i].value,
2163				args[i].vallen, error_str, error_str_len);
2164			if (tmpptr == NULL)
2165				goto bailout;
2166			if ((args[i].flags & CTL_BEARG_ASCII)
2167			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2168				snprintf(error_str, error_str_len, "Argument "
2169				    "%d value is not NUL-terminated", i);
2170				goto bailout;
2171			}
2172			args[i].kvalue = tmpptr;
2173		} else {
2174			args[i].kvalue = malloc(args[i].vallen,
2175			    M_CTL, M_WAITOK | M_ZERO);
2176		}
2177	}
2178
2179	return (args);
2180bailout:
2181
2182	ctl_free_args(num_args, args);
2183
2184	return (NULL);
2185}
2186
2187static void
2188ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2189{
2190	int i;
2191
2192	for (i = 0; i < num_args; i++) {
2193		if (args[i].flags & CTL_BEARG_WR)
2194			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2195	}
2196}
2197
2198/*
2199 * Escape characters that are illegal or not recommended in XML.
2200 */
2201int
2202ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2203{
2204	int retval;
2205
2206	retval = 0;
2207
2208	for (; *str; str++) {
2209		switch (*str) {
2210		case '&':
2211			retval = sbuf_printf(sb, "&amp;");
2212			break;
2213		case '>':
2214			retval = sbuf_printf(sb, "&gt;");
2215			break;
2216		case '<':
2217			retval = sbuf_printf(sb, "&lt;");
2218			break;
2219		default:
2220			retval = sbuf_putc(sb, *str);
2221			break;
2222		}
2223
2224		if (retval != 0)
2225			break;
2226
2227	}
2228
2229	return (retval);
2230}
2231
2232static void
2233ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2234{
2235	struct scsi_vpd_id_descriptor *desc;
2236	int i;
2237
2238	if (id == NULL || id->len < 4)
2239		return;
2240	desc = (struct scsi_vpd_id_descriptor *)id->data;
2241	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2242	case SVPD_ID_TYPE_T10:
2243		sbuf_printf(sb, "t10.");
2244		break;
2245	case SVPD_ID_TYPE_EUI64:
2246		sbuf_printf(sb, "eui.");
2247		break;
2248	case SVPD_ID_TYPE_NAA:
2249		sbuf_printf(sb, "naa.");
2250		break;
2251	case SVPD_ID_TYPE_SCSI_NAME:
2252		break;
2253	}
2254	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2255	case SVPD_ID_CODESET_BINARY:
2256		for (i = 0; i < desc->length; i++)
2257			sbuf_printf(sb, "%02x", desc->identifier[i]);
2258		break;
2259	case SVPD_ID_CODESET_ASCII:
2260		sbuf_printf(sb, "%.*s", (int)desc->length,
2261		    (char *)desc->identifier);
2262		break;
2263	case SVPD_ID_CODESET_UTF8:
2264		sbuf_printf(sb, "%s", (char *)desc->identifier);
2265		break;
2266	}
2267}
2268
2269static int
2270ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2271	  struct thread *td)
2272{
2273	struct ctl_softc *softc;
2274	int retval;
2275
2276	softc = control_softc;
2277
2278	retval = 0;
2279
2280	switch (cmd) {
2281	case CTL_IO: {
2282		union ctl_io *io;
2283		void *pool_tmp;
2284
2285		/*
2286		 * If we haven't been "enabled", don't allow any SCSI I/O
2287		 * to this FETD.
2288		 */
2289		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2290			retval = EPERM;
2291			break;
2292		}
2293
2294		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2295		if (io == NULL) {
2296			printf("ctl_ioctl: can't allocate ctl_io!\n");
2297			retval = ENOSPC;
2298			break;
2299		}
2300
2301		/*
2302		 * Need to save the pool reference so it doesn't get
2303		 * spammed by the user's ctl_io.
2304		 */
2305		pool_tmp = io->io_hdr.pool;
2306
2307		memcpy(io, (void *)addr, sizeof(*io));
2308
2309		io->io_hdr.pool = pool_tmp;
2310		/*
2311		 * No status yet, so make sure the status is set properly.
2312		 */
2313		io->io_hdr.status = CTL_STATUS_NONE;
2314
2315		/*
2316		 * The user sets the initiator ID, target and LUN IDs.
2317		 */
2318		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2319		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2320		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2321		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2322			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2323
2324		retval = ctl_ioctl_submit_wait(io);
2325
2326		if (retval != 0) {
2327			ctl_free_io(io);
2328			break;
2329		}
2330
2331		memcpy((void *)addr, io, sizeof(*io));
2332
2333		/* return this to our pool */
2334		ctl_free_io(io);
2335
2336		break;
2337	}
2338	case CTL_ENABLE_PORT:
2339	case CTL_DISABLE_PORT:
2340	case CTL_SET_PORT_WWNS: {
2341		struct ctl_port *port;
2342		struct ctl_port_entry *entry;
2343
2344		entry = (struct ctl_port_entry *)addr;
2345
2346		mtx_lock(&softc->ctl_lock);
2347		STAILQ_FOREACH(port, &softc->port_list, links) {
2348			int action, done;
2349
2350			action = 0;
2351			done = 0;
2352
2353			if ((entry->port_type == CTL_PORT_NONE)
2354			 && (entry->targ_port == port->targ_port)) {
2355				/*
2356				 * If the user only wants to enable or
2357				 * disable or set WWNs on a specific port,
2358				 * do the operation and we're done.
2359				 */
2360				action = 1;
2361				done = 1;
2362			} else if (entry->port_type & port->port_type) {
2363				/*
2364				 * Compare the user's type mask with the
2365				 * particular frontend type to see if we
2366				 * have a match.
2367				 */
2368				action = 1;
2369				done = 0;
2370
2371				/*
2372				 * Make sure the user isn't trying to set
2373				 * WWNs on multiple ports at the same time.
2374				 */
2375				if (cmd == CTL_SET_PORT_WWNS) {
2376					printf("%s: Can't set WWNs on "
2377					       "multiple ports\n", __func__);
2378					retval = EINVAL;
2379					break;
2380				}
2381			}
2382			if (action != 0) {
2383				/*
2384				 * XXX KDM we have to drop the lock here,
2385				 * because the online/offline operations
2386				 * can potentially block.  We need to
2387				 * reference count the frontends so they
2388				 * can't go away,
2389				 */
2390				mtx_unlock(&softc->ctl_lock);
2391
2392				if (cmd == CTL_ENABLE_PORT) {
2393					struct ctl_lun *lun;
2394
2395					STAILQ_FOREACH(lun, &softc->lun_list,
2396						       links) {
2397						port->lun_enable(port->targ_lun_arg,
2398						    lun->target,
2399						    lun->lun);
2400					}
2401
2402					ctl_port_online(port);
2403				} else if (cmd == CTL_DISABLE_PORT) {
2404					struct ctl_lun *lun;
2405
2406					ctl_port_offline(port);
2407
2408					STAILQ_FOREACH(lun, &softc->lun_list,
2409						       links) {
2410						port->lun_disable(
2411						    port->targ_lun_arg,
2412						    lun->target,
2413						    lun->lun);
2414					}
2415				}
2416
2417				mtx_lock(&softc->ctl_lock);
2418
2419				if (cmd == CTL_SET_PORT_WWNS)
2420					ctl_port_set_wwns(port,
2421					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2422					    1 : 0, entry->wwnn,
2423					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2424					    1 : 0, entry->wwpn);
2425			}
2426			if (done != 0)
2427				break;
2428		}
2429		mtx_unlock(&softc->ctl_lock);
2430		break;
2431	}
2432	case CTL_GET_PORT_LIST: {
2433		struct ctl_port *port;
2434		struct ctl_port_list *list;
2435		int i;
2436
2437		list = (struct ctl_port_list *)addr;
2438
2439		if (list->alloc_len != (list->alloc_num *
2440		    sizeof(struct ctl_port_entry))) {
2441			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2442			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2443			       "%zu\n", __func__, list->alloc_len,
2444			       list->alloc_num, sizeof(struct ctl_port_entry));
2445			retval = EINVAL;
2446			break;
2447		}
2448		list->fill_len = 0;
2449		list->fill_num = 0;
2450		list->dropped_num = 0;
2451		i = 0;
2452		mtx_lock(&softc->ctl_lock);
2453		STAILQ_FOREACH(port, &softc->port_list, links) {
2454			struct ctl_port_entry entry, *list_entry;
2455
2456			if (list->fill_num >= list->alloc_num) {
2457				list->dropped_num++;
2458				continue;
2459			}
2460
2461			entry.port_type = port->port_type;
2462			strlcpy(entry.port_name, port->port_name,
2463				sizeof(entry.port_name));
2464			entry.targ_port = port->targ_port;
2465			entry.physical_port = port->physical_port;
2466			entry.virtual_port = port->virtual_port;
2467			entry.wwnn = port->wwnn;
2468			entry.wwpn = port->wwpn;
2469			if (port->status & CTL_PORT_STATUS_ONLINE)
2470				entry.online = 1;
2471			else
2472				entry.online = 0;
2473
2474			list_entry = &list->entries[i];
2475
2476			retval = copyout(&entry, list_entry, sizeof(entry));
2477			if (retval != 0) {
2478				printf("%s: CTL_GET_PORT_LIST: copyout "
2479				       "returned %d\n", __func__, retval);
2480				break;
2481			}
2482			i++;
2483			list->fill_num++;
2484			list->fill_len += sizeof(entry);
2485		}
2486		mtx_unlock(&softc->ctl_lock);
2487
2488		/*
2489		 * If this is non-zero, we had a copyout fault, so there's
2490		 * probably no point in attempting to set the status inside
2491		 * the structure.
2492		 */
2493		if (retval != 0)
2494			break;
2495
2496		if (list->dropped_num > 0)
2497			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2498		else
2499			list->status = CTL_PORT_LIST_OK;
2500		break;
2501	}
2502	case CTL_DUMP_OOA: {
2503		struct ctl_lun *lun;
2504		union ctl_io *io;
2505		char printbuf[128];
2506		struct sbuf sb;
2507
2508		mtx_lock(&softc->ctl_lock);
2509		printf("Dumping OOA queues:\n");
2510		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2511			mtx_lock(&lun->lun_lock);
2512			for (io = (union ctl_io *)TAILQ_FIRST(
2513			     &lun->ooa_queue); io != NULL;
2514			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2515			     ooa_links)) {
2516				sbuf_new(&sb, printbuf, sizeof(printbuf),
2517					 SBUF_FIXEDLEN);
2518				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2519					    (intmax_t)lun->lun,
2520					    io->scsiio.tag_num,
2521					    (io->io_hdr.flags &
2522					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2523					    (io->io_hdr.flags &
2524					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2525					    (io->io_hdr.flags &
2526					    CTL_FLAG_ABORT) ? " ABORT" : "",
2527			                    (io->io_hdr.flags &
2528		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2529				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2530				sbuf_finish(&sb);
2531				printf("%s\n", sbuf_data(&sb));
2532			}
2533			mtx_unlock(&lun->lun_lock);
2534		}
2535		printf("OOA queues dump done\n");
2536		mtx_unlock(&softc->ctl_lock);
2537		break;
2538	}
2539	case CTL_GET_OOA: {
2540		struct ctl_lun *lun;
2541		struct ctl_ooa *ooa_hdr;
2542		struct ctl_ooa_entry *entries;
2543		uint32_t cur_fill_num;
2544
2545		ooa_hdr = (struct ctl_ooa *)addr;
2546
2547		if ((ooa_hdr->alloc_len == 0)
2548		 || (ooa_hdr->alloc_num == 0)) {
2549			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2550			       "must be non-zero\n", __func__,
2551			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2552			retval = EINVAL;
2553			break;
2554		}
2555
2556		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2557		    sizeof(struct ctl_ooa_entry))) {
2558			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2559			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2560			       __func__, ooa_hdr->alloc_len,
2561			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2562			retval = EINVAL;
2563			break;
2564		}
2565
2566		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2567		if (entries == NULL) {
2568			printf("%s: could not allocate %d bytes for OOA "
2569			       "dump\n", __func__, ooa_hdr->alloc_len);
2570			retval = ENOMEM;
2571			break;
2572		}
2573
2574		mtx_lock(&softc->ctl_lock);
2575		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2576		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2577		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2578			mtx_unlock(&softc->ctl_lock);
2579			free(entries, M_CTL);
2580			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2581			       __func__, (uintmax_t)ooa_hdr->lun_num);
2582			retval = EINVAL;
2583			break;
2584		}
2585
2586		cur_fill_num = 0;
2587
2588		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2589			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2590				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2591					ooa_hdr, entries);
2592				if (retval != 0)
2593					break;
2594			}
2595			if (retval != 0) {
2596				mtx_unlock(&softc->ctl_lock);
2597				free(entries, M_CTL);
2598				break;
2599			}
2600		} else {
2601			lun = softc->ctl_luns[ooa_hdr->lun_num];
2602
2603			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2604						    entries);
2605		}
2606		mtx_unlock(&softc->ctl_lock);
2607
2608		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2609		ooa_hdr->fill_len = ooa_hdr->fill_num *
2610			sizeof(struct ctl_ooa_entry);
2611		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2612		if (retval != 0) {
2613			printf("%s: error copying out %d bytes for OOA dump\n",
2614			       __func__, ooa_hdr->fill_len);
2615		}
2616
2617		getbintime(&ooa_hdr->cur_bt);
2618
2619		if (cur_fill_num > ooa_hdr->alloc_num) {
2620			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2621			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2622		} else {
2623			ooa_hdr->dropped_num = 0;
2624			ooa_hdr->status = CTL_OOA_OK;
2625		}
2626
2627		free(entries, M_CTL);
2628		break;
2629	}
2630	case CTL_CHECK_OOA: {
2631		union ctl_io *io;
2632		struct ctl_lun *lun;
2633		struct ctl_ooa_info *ooa_info;
2634
2635
2636		ooa_info = (struct ctl_ooa_info *)addr;
2637
2638		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2639			ooa_info->status = CTL_OOA_INVALID_LUN;
2640			break;
2641		}
2642		mtx_lock(&softc->ctl_lock);
2643		lun = softc->ctl_luns[ooa_info->lun_id];
2644		if (lun == NULL) {
2645			mtx_unlock(&softc->ctl_lock);
2646			ooa_info->status = CTL_OOA_INVALID_LUN;
2647			break;
2648		}
2649		mtx_lock(&lun->lun_lock);
2650		mtx_unlock(&softc->ctl_lock);
2651		ooa_info->num_entries = 0;
2652		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2653		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2654		     &io->io_hdr, ooa_links)) {
2655			ooa_info->num_entries++;
2656		}
2657		mtx_unlock(&lun->lun_lock);
2658
2659		ooa_info->status = CTL_OOA_SUCCESS;
2660
2661		break;
2662	}
2663	case CTL_HARD_START:
2664	case CTL_HARD_STOP: {
2665		struct ctl_fe_ioctl_startstop_info ss_info;
2666		struct cfi_metatask *metatask;
2667		struct mtx hs_mtx;
2668
2669		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2670
2671		cv_init(&ss_info.sem, "hard start/stop cv" );
2672
2673		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2674		if (metatask == NULL) {
2675			retval = ENOMEM;
2676			mtx_destroy(&hs_mtx);
2677			break;
2678		}
2679
2680		if (cmd == CTL_HARD_START)
2681			metatask->tasktype = CFI_TASK_STARTUP;
2682		else
2683			metatask->tasktype = CFI_TASK_SHUTDOWN;
2684
2685		metatask->callback = ctl_ioctl_hard_startstop_callback;
2686		metatask->callback_arg = &ss_info;
2687
2688		cfi_action(metatask);
2689
2690		/* Wait for the callback */
2691		mtx_lock(&hs_mtx);
2692		cv_wait_sig(&ss_info.sem, &hs_mtx);
2693		mtx_unlock(&hs_mtx);
2694
2695		/*
2696		 * All information has been copied from the metatask by the
2697		 * time cv_broadcast() is called, so we free the metatask here.
2698		 */
2699		cfi_free_metatask(metatask);
2700
2701		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2702
2703		mtx_destroy(&hs_mtx);
2704		break;
2705	}
2706	case CTL_BBRREAD: {
2707		struct ctl_bbrread_info *bbr_info;
2708		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2709		struct mtx bbr_mtx;
2710		struct cfi_metatask *metatask;
2711
2712		bbr_info = (struct ctl_bbrread_info *)addr;
2713
2714		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2715
2716		bzero(&bbr_mtx, sizeof(bbr_mtx));
2717		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2718
2719		fe_bbr_info.bbr_info = bbr_info;
2720		fe_bbr_info.lock = &bbr_mtx;
2721
2722		cv_init(&fe_bbr_info.sem, "BBR read cv");
2723		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2724
2725		if (metatask == NULL) {
2726			mtx_destroy(&bbr_mtx);
2727			cv_destroy(&fe_bbr_info.sem);
2728			retval = ENOMEM;
2729			break;
2730		}
2731		metatask->tasktype = CFI_TASK_BBRREAD;
2732		metatask->callback = ctl_ioctl_bbrread_callback;
2733		metatask->callback_arg = &fe_bbr_info;
2734		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2735		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2736		metatask->taskinfo.bbrread.len = bbr_info->len;
2737
2738		cfi_action(metatask);
2739
2740		mtx_lock(&bbr_mtx);
2741		while (fe_bbr_info.wakeup_done == 0)
2742			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2743		mtx_unlock(&bbr_mtx);
2744
2745		bbr_info->status = metatask->status;
2746		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2747		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2748		memcpy(&bbr_info->sense_data,
2749		       &metatask->taskinfo.bbrread.sense_data,
2750		       ctl_min(sizeof(bbr_info->sense_data),
2751			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2752
2753		cfi_free_metatask(metatask);
2754
2755		mtx_destroy(&bbr_mtx);
2756		cv_destroy(&fe_bbr_info.sem);
2757
2758		break;
2759	}
2760	case CTL_DELAY_IO: {
2761		struct ctl_io_delay_info *delay_info;
2762#ifdef CTL_IO_DELAY
2763		struct ctl_lun *lun;
2764#endif /* CTL_IO_DELAY */
2765
2766		delay_info = (struct ctl_io_delay_info *)addr;
2767
2768#ifdef CTL_IO_DELAY
2769		mtx_lock(&softc->ctl_lock);
2770
2771		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2772		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2773			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2774		} else {
2775			lun = softc->ctl_luns[delay_info->lun_id];
2776			mtx_lock(&lun->lun_lock);
2777
2778			delay_info->status = CTL_DELAY_STATUS_OK;
2779
2780			switch (delay_info->delay_type) {
2781			case CTL_DELAY_TYPE_CONT:
2782				break;
2783			case CTL_DELAY_TYPE_ONESHOT:
2784				break;
2785			default:
2786				delay_info->status =
2787					CTL_DELAY_STATUS_INVALID_TYPE;
2788				break;
2789			}
2790
2791			switch (delay_info->delay_loc) {
2792			case CTL_DELAY_LOC_DATAMOVE:
2793				lun->delay_info.datamove_type =
2794					delay_info->delay_type;
2795				lun->delay_info.datamove_delay =
2796					delay_info->delay_secs;
2797				break;
2798			case CTL_DELAY_LOC_DONE:
2799				lun->delay_info.done_type =
2800					delay_info->delay_type;
2801				lun->delay_info.done_delay =
2802					delay_info->delay_secs;
2803				break;
2804			default:
2805				delay_info->status =
2806					CTL_DELAY_STATUS_INVALID_LOC;
2807				break;
2808			}
2809			mtx_unlock(&lun->lun_lock);
2810		}
2811
2812		mtx_unlock(&softc->ctl_lock);
2813#else
2814		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2815#endif /* CTL_IO_DELAY */
2816		break;
2817	}
2818	case CTL_REALSYNC_SET: {
2819		int *syncstate;
2820
2821		syncstate = (int *)addr;
2822
2823		mtx_lock(&softc->ctl_lock);
2824		switch (*syncstate) {
2825		case 0:
2826			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2827			break;
2828		case 1:
2829			softc->flags |= CTL_FLAG_REAL_SYNC;
2830			break;
2831		default:
2832			retval = EINVAL;
2833			break;
2834		}
2835		mtx_unlock(&softc->ctl_lock);
2836		break;
2837	}
2838	case CTL_REALSYNC_GET: {
2839		int *syncstate;
2840
2841		syncstate = (int*)addr;
2842
2843		mtx_lock(&softc->ctl_lock);
2844		if (softc->flags & CTL_FLAG_REAL_SYNC)
2845			*syncstate = 1;
2846		else
2847			*syncstate = 0;
2848		mtx_unlock(&softc->ctl_lock);
2849
2850		break;
2851	}
2852	case CTL_SETSYNC:
2853	case CTL_GETSYNC: {
2854		struct ctl_sync_info *sync_info;
2855		struct ctl_lun *lun;
2856
2857		sync_info = (struct ctl_sync_info *)addr;
2858
2859		mtx_lock(&softc->ctl_lock);
2860		lun = softc->ctl_luns[sync_info->lun_id];
2861		if (lun == NULL) {
2862			mtx_unlock(&softc->ctl_lock);
2863			sync_info->status = CTL_GS_SYNC_NO_LUN;
2864		}
2865		/*
2866		 * Get or set the sync interval.  We're not bounds checking
2867		 * in the set case, hopefully the user won't do something
2868		 * silly.
2869		 */
2870		mtx_lock(&lun->lun_lock);
2871		mtx_unlock(&softc->ctl_lock);
2872		if (cmd == CTL_GETSYNC)
2873			sync_info->sync_interval = lun->sync_interval;
2874		else
2875			lun->sync_interval = sync_info->sync_interval;
2876		mtx_unlock(&lun->lun_lock);
2877
2878		sync_info->status = CTL_GS_SYNC_OK;
2879
2880		break;
2881	}
2882	case CTL_GETSTATS: {
2883		struct ctl_stats *stats;
2884		struct ctl_lun *lun;
2885		int i;
2886
2887		stats = (struct ctl_stats *)addr;
2888
2889		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2890		     stats->alloc_len) {
2891			stats->status = CTL_SS_NEED_MORE_SPACE;
2892			stats->num_luns = softc->num_luns;
2893			break;
2894		}
2895		/*
2896		 * XXX KDM no locking here.  If the LUN list changes,
2897		 * things can blow up.
2898		 */
2899		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2900		     i++, lun = STAILQ_NEXT(lun, links)) {
2901			retval = copyout(&lun->stats, &stats->lun_stats[i],
2902					 sizeof(lun->stats));
2903			if (retval != 0)
2904				break;
2905		}
2906		stats->num_luns = softc->num_luns;
2907		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2908				 softc->num_luns;
2909		stats->status = CTL_SS_OK;
2910#ifdef CTL_TIME_IO
2911		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2912#else
2913		stats->flags = CTL_STATS_FLAG_NONE;
2914#endif
2915		getnanouptime(&stats->timestamp);
2916		break;
2917	}
2918	case CTL_ERROR_INJECT: {
2919		struct ctl_error_desc *err_desc, *new_err_desc;
2920		struct ctl_lun *lun;
2921
2922		err_desc = (struct ctl_error_desc *)addr;
2923
2924		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2925				      M_WAITOK | M_ZERO);
2926		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2927
2928		mtx_lock(&softc->ctl_lock);
2929		lun = softc->ctl_luns[err_desc->lun_id];
2930		if (lun == NULL) {
2931			mtx_unlock(&softc->ctl_lock);
2932			free(new_err_desc, M_CTL);
2933			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2934			       __func__, (uintmax_t)err_desc->lun_id);
2935			retval = EINVAL;
2936			break;
2937		}
2938		mtx_lock(&lun->lun_lock);
2939		mtx_unlock(&softc->ctl_lock);
2940
2941		/*
2942		 * We could do some checking here to verify the validity
2943		 * of the request, but given the complexity of error
2944		 * injection requests, the checking logic would be fairly
2945		 * complex.
2946		 *
2947		 * For now, if the request is invalid, it just won't get
2948		 * executed and might get deleted.
2949		 */
2950		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2951
2952		/*
2953		 * XXX KDM check to make sure the serial number is unique,
2954		 * in case we somehow manage to wrap.  That shouldn't
2955		 * happen for a very long time, but it's the right thing to
2956		 * do.
2957		 */
2958		new_err_desc->serial = lun->error_serial;
2959		err_desc->serial = lun->error_serial;
2960		lun->error_serial++;
2961
2962		mtx_unlock(&lun->lun_lock);
2963		break;
2964	}
2965	case CTL_ERROR_INJECT_DELETE: {
2966		struct ctl_error_desc *delete_desc, *desc, *desc2;
2967		struct ctl_lun *lun;
2968		int delete_done;
2969
2970		delete_desc = (struct ctl_error_desc *)addr;
2971		delete_done = 0;
2972
2973		mtx_lock(&softc->ctl_lock);
2974		lun = softc->ctl_luns[delete_desc->lun_id];
2975		if (lun == NULL) {
2976			mtx_unlock(&softc->ctl_lock);
2977			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2978			       __func__, (uintmax_t)delete_desc->lun_id);
2979			retval = EINVAL;
2980			break;
2981		}
2982		mtx_lock(&lun->lun_lock);
2983		mtx_unlock(&softc->ctl_lock);
2984		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2985			if (desc->serial != delete_desc->serial)
2986				continue;
2987
2988			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2989				      links);
2990			free(desc, M_CTL);
2991			delete_done = 1;
2992		}
2993		mtx_unlock(&lun->lun_lock);
2994		if (delete_done == 0) {
2995			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2996			       "error serial %ju on LUN %u\n", __func__,
2997			       delete_desc->serial, delete_desc->lun_id);
2998			retval = EINVAL;
2999			break;
3000		}
3001		break;
3002	}
3003	case CTL_DUMP_STRUCTS: {
3004		int i, j, k, idx;
3005		struct ctl_port *port;
3006		struct ctl_frontend *fe;
3007
3008		mtx_lock(&softc->ctl_lock);
3009		printf("CTL Persistent Reservation information start:\n");
3010		for (i = 0; i < CTL_MAX_LUNS; i++) {
3011			struct ctl_lun *lun;
3012
3013			lun = softc->ctl_luns[i];
3014
3015			if ((lun == NULL)
3016			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3017				continue;
3018
3019			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3020				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3021					idx = j * CTL_MAX_INIT_PER_PORT + k;
3022					if (lun->pr_keys[idx] == 0)
3023						continue;
3024					printf("  LUN %d port %d iid %d key "
3025					       "%#jx\n", i, j, k,
3026					       (uintmax_t)lun->pr_keys[idx]);
3027				}
3028			}
3029		}
3030		printf("CTL Persistent Reservation information end\n");
3031		printf("CTL Ports:\n");
3032		STAILQ_FOREACH(port, &softc->port_list, links) {
3033			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3034			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3035			       port->frontend->name, port->port_type,
3036			       port->physical_port, port->virtual_port,
3037			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3038			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3039				if (port->wwpn_iid[j].in_use == 0 &&
3040				    port->wwpn_iid[j].wwpn == 0 &&
3041				    port->wwpn_iid[j].name == NULL)
3042					continue;
3043
3044				printf("    iid %u use %d WWPN %#jx '%s'\n",
3045				    j, port->wwpn_iid[j].in_use,
3046				    (uintmax_t)port->wwpn_iid[j].wwpn,
3047				    port->wwpn_iid[j].name);
3048			}
3049		}
3050		printf("CTL Port information end\n");
3051		mtx_unlock(&softc->ctl_lock);
3052		/*
3053		 * XXX KDM calling this without a lock.  We'd likely want
3054		 * to drop the lock before calling the frontend's dump
3055		 * routine anyway.
3056		 */
3057		printf("CTL Frontends:\n");
3058		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3059			printf("  Frontend '%s'\n", fe->name);
3060			if (fe->fe_dump != NULL)
3061				fe->fe_dump();
3062		}
3063		printf("CTL Frontend information end\n");
3064		break;
3065	}
3066	case CTL_LUN_REQ: {
3067		struct ctl_lun_req *lun_req;
3068		struct ctl_backend_driver *backend;
3069
3070		lun_req = (struct ctl_lun_req *)addr;
3071
3072		backend = ctl_backend_find(lun_req->backend);
3073		if (backend == NULL) {
3074			lun_req->status = CTL_LUN_ERROR;
3075			snprintf(lun_req->error_str,
3076				 sizeof(lun_req->error_str),
3077				 "Backend \"%s\" not found.",
3078				 lun_req->backend);
3079			break;
3080		}
3081		if (lun_req->num_be_args > 0) {
3082			lun_req->kern_be_args = ctl_copyin_args(
3083				lun_req->num_be_args,
3084				lun_req->be_args,
3085				lun_req->error_str,
3086				sizeof(lun_req->error_str));
3087			if (lun_req->kern_be_args == NULL) {
3088				lun_req->status = CTL_LUN_ERROR;
3089				break;
3090			}
3091		}
3092
3093		retval = backend->ioctl(dev, cmd, addr, flag, td);
3094
3095		if (lun_req->num_be_args > 0) {
3096			ctl_copyout_args(lun_req->num_be_args,
3097				      lun_req->kern_be_args);
3098			ctl_free_args(lun_req->num_be_args,
3099				      lun_req->kern_be_args);
3100		}
3101		break;
3102	}
3103	case CTL_LUN_LIST: {
3104		struct sbuf *sb;
3105		struct ctl_lun *lun;
3106		struct ctl_lun_list *list;
3107		struct ctl_option *opt;
3108
3109		list = (struct ctl_lun_list *)addr;
3110
3111		/*
3112		 * Allocate a fixed length sbuf here, based on the length
3113		 * of the user's buffer.  We could allocate an auto-extending
3114		 * buffer, and then tell the user how much larger our
3115		 * amount of data is than his buffer, but that presents
3116		 * some problems:
3117		 *
3118		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3119		 *     we can't hold a lock while calling them with an
3120		 *     auto-extending buffer.
3121 		 *
3122		 * 2.  There is not currently a LUN reference counting
3123		 *     mechanism, outside of outstanding transactions on
3124		 *     the LUN's OOA queue.  So a LUN could go away on us
3125		 *     while we're getting the LUN number, backend-specific
3126		 *     information, etc.  Thus, given the way things
3127		 *     currently work, we need to hold the CTL lock while
3128		 *     grabbing LUN information.
3129		 *
3130		 * So, from the user's standpoint, the best thing to do is
3131		 * allocate what he thinks is a reasonable buffer length,
3132		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3133		 * double the buffer length and try again.  (And repeat
3134		 * that until he succeeds.)
3135		 */
3136		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3137		if (sb == NULL) {
3138			list->status = CTL_LUN_LIST_ERROR;
3139			snprintf(list->error_str, sizeof(list->error_str),
3140				 "Unable to allocate %d bytes for LUN list",
3141				 list->alloc_len);
3142			break;
3143		}
3144
3145		sbuf_printf(sb, "<ctllunlist>\n");
3146
3147		mtx_lock(&softc->ctl_lock);
3148		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3149			mtx_lock(&lun->lun_lock);
3150			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3151					     (uintmax_t)lun->lun);
3152
3153			/*
3154			 * Bail out as soon as we see that we've overfilled
3155			 * the buffer.
3156			 */
3157			if (retval != 0)
3158				break;
3159
3160			retval = sbuf_printf(sb, "\t<backend_type>%s"
3161					     "</backend_type>\n",
3162					     (lun->backend == NULL) ?  "none" :
3163					     lun->backend->name);
3164
3165			if (retval != 0)
3166				break;
3167
3168			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3169					     lun->be_lun->lun_type);
3170
3171			if (retval != 0)
3172				break;
3173
3174			if (lun->backend == NULL) {
3175				retval = sbuf_printf(sb, "</lun>\n");
3176				if (retval != 0)
3177					break;
3178				continue;
3179			}
3180
3181			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3182					     (lun->be_lun->maxlba > 0) ?
3183					     lun->be_lun->maxlba + 1 : 0);
3184
3185			if (retval != 0)
3186				break;
3187
3188			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3189					     lun->be_lun->blocksize);
3190
3191			if (retval != 0)
3192				break;
3193
3194			retval = sbuf_printf(sb, "\t<serial_number>");
3195
3196			if (retval != 0)
3197				break;
3198
3199			retval = ctl_sbuf_printf_esc(sb,
3200						     lun->be_lun->serial_num);
3201
3202			if (retval != 0)
3203				break;
3204
3205			retval = sbuf_printf(sb, "</serial_number>\n");
3206
3207			if (retval != 0)
3208				break;
3209
3210			retval = sbuf_printf(sb, "\t<device_id>");
3211
3212			if (retval != 0)
3213				break;
3214
3215			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3216
3217			if (retval != 0)
3218				break;
3219
3220			retval = sbuf_printf(sb, "</device_id>\n");
3221
3222			if (retval != 0)
3223				break;
3224
3225			if (lun->backend->lun_info != NULL) {
3226				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3227				if (retval != 0)
3228					break;
3229			}
3230			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3231				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3232				    opt->name, opt->value, opt->name);
3233				if (retval != 0)
3234					break;
3235			}
3236
3237			retval = sbuf_printf(sb, "</lun>\n");
3238
3239			if (retval != 0)
3240				break;
3241			mtx_unlock(&lun->lun_lock);
3242		}
3243		if (lun != NULL)
3244			mtx_unlock(&lun->lun_lock);
3245		mtx_unlock(&softc->ctl_lock);
3246
3247		if ((retval != 0)
3248		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3249			retval = 0;
3250			sbuf_delete(sb);
3251			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3252			snprintf(list->error_str, sizeof(list->error_str),
3253				 "Out of space, %d bytes is too small",
3254				 list->alloc_len);
3255			break;
3256		}
3257
3258		sbuf_finish(sb);
3259
3260		retval = copyout(sbuf_data(sb), list->lun_xml,
3261				 sbuf_len(sb) + 1);
3262
3263		list->fill_len = sbuf_len(sb) + 1;
3264		list->status = CTL_LUN_LIST_OK;
3265		sbuf_delete(sb);
3266		break;
3267	}
3268	case CTL_ISCSI: {
3269		struct ctl_iscsi *ci;
3270		struct ctl_frontend *fe;
3271
3272		ci = (struct ctl_iscsi *)addr;
3273
3274		fe = ctl_frontend_find("iscsi");
3275		if (fe == NULL) {
3276			ci->status = CTL_ISCSI_ERROR;
3277			snprintf(ci->error_str, sizeof(ci->error_str),
3278			    "Frontend \"iscsi\" not found.");
3279			break;
3280		}
3281
3282		retval = fe->ioctl(dev, cmd, addr, flag, td);
3283		break;
3284	}
3285	case CTL_PORT_REQ: {
3286		struct ctl_req *req;
3287		struct ctl_frontend *fe;
3288
3289		req = (struct ctl_req *)addr;
3290
3291		fe = ctl_frontend_find(req->driver);
3292		if (fe == NULL) {
3293			req->status = CTL_LUN_ERROR;
3294			snprintf(req->error_str, sizeof(req->error_str),
3295			    "Frontend \"%s\" not found.", req->driver);
3296			break;
3297		}
3298		if (req->num_args > 0) {
3299			req->kern_args = ctl_copyin_args(req->num_args,
3300			    req->args, req->error_str, sizeof(req->error_str));
3301			if (req->kern_args == NULL) {
3302				req->status = CTL_LUN_ERROR;
3303				break;
3304			}
3305		}
3306
3307		retval = fe->ioctl(dev, cmd, addr, flag, td);
3308
3309		if (req->num_args > 0) {
3310			ctl_copyout_args(req->num_args, req->kern_args);
3311			ctl_free_args(req->num_args, req->kern_args);
3312		}
3313		break;
3314	}
3315	case CTL_PORT_LIST: {
3316		struct sbuf *sb;
3317		struct ctl_port *port;
3318		struct ctl_lun_list *list;
3319		struct ctl_option *opt;
3320		int j;
3321
3322		list = (struct ctl_lun_list *)addr;
3323
3324		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3325		if (sb == NULL) {
3326			list->status = CTL_LUN_LIST_ERROR;
3327			snprintf(list->error_str, sizeof(list->error_str),
3328				 "Unable to allocate %d bytes for LUN list",
3329				 list->alloc_len);
3330			break;
3331		}
3332
3333		sbuf_printf(sb, "<ctlportlist>\n");
3334
3335		mtx_lock(&softc->ctl_lock);
3336		STAILQ_FOREACH(port, &softc->port_list, links) {
3337			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3338					     (uintmax_t)port->targ_port);
3339
3340			/*
3341			 * Bail out as soon as we see that we've overfilled
3342			 * the buffer.
3343			 */
3344			if (retval != 0)
3345				break;
3346
3347			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3348			    "</frontend_type>\n", port->frontend->name);
3349			if (retval != 0)
3350				break;
3351
3352			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3353					     port->port_type);
3354			if (retval != 0)
3355				break;
3356
3357			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3358			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3359			if (retval != 0)
3360				break;
3361
3362			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3363			    port->port_name);
3364			if (retval != 0)
3365				break;
3366
3367			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3368			    port->physical_port);
3369			if (retval != 0)
3370				break;
3371
3372			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3373			    port->virtual_port);
3374			if (retval != 0)
3375				break;
3376
3377			if (port->target_devid != NULL) {
3378				sbuf_printf(sb, "\t<target>");
3379				ctl_id_sbuf(port->target_devid, sb);
3380				sbuf_printf(sb, "</target>\n");
3381			}
3382
3383			if (port->port_devid != NULL) {
3384				sbuf_printf(sb, "\t<port>");
3385				ctl_id_sbuf(port->port_devid, sb);
3386				sbuf_printf(sb, "</port>\n");
3387			}
3388
3389			if (port->port_info != NULL) {
3390				retval = port->port_info(port->onoff_arg, sb);
3391				if (retval != 0)
3392					break;
3393			}
3394			STAILQ_FOREACH(opt, &port->options, links) {
3395				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3396				    opt->name, opt->value, opt->name);
3397				if (retval != 0)
3398					break;
3399			}
3400
3401			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3402				if (port->wwpn_iid[j].in_use == 0 ||
3403				    (port->wwpn_iid[j].wwpn == 0 &&
3404				     port->wwpn_iid[j].name == NULL))
3405					continue;
3406
3407				if (port->wwpn_iid[j].name != NULL)
3408					retval = sbuf_printf(sb,
3409					    "\t<initiator>%u %s</initiator>\n",
3410					    j, port->wwpn_iid[j].name);
3411				else
3412					retval = sbuf_printf(sb,
3413					    "\t<initiator>%u naa.%08jx</initiator>\n",
3414					    j, port->wwpn_iid[j].wwpn);
3415				if (retval != 0)
3416					break;
3417			}
3418			if (retval != 0)
3419				break;
3420
3421			retval = sbuf_printf(sb, "</targ_port>\n");
3422			if (retval != 0)
3423				break;
3424		}
3425		mtx_unlock(&softc->ctl_lock);
3426
3427		if ((retval != 0)
3428		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3429			retval = 0;
3430			sbuf_delete(sb);
3431			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3432			snprintf(list->error_str, sizeof(list->error_str),
3433				 "Out of space, %d bytes is too small",
3434				 list->alloc_len);
3435			break;
3436		}
3437
3438		sbuf_finish(sb);
3439
3440		retval = copyout(sbuf_data(sb), list->lun_xml,
3441				 sbuf_len(sb) + 1);
3442
3443		list->fill_len = sbuf_len(sb) + 1;
3444		list->status = CTL_LUN_LIST_OK;
3445		sbuf_delete(sb);
3446		break;
3447	}
3448	default: {
3449		/* XXX KDM should we fix this? */
3450#if 0
3451		struct ctl_backend_driver *backend;
3452		unsigned int type;
3453		int found;
3454
3455		found = 0;
3456
3457		/*
3458		 * We encode the backend type as the ioctl type for backend
3459		 * ioctls.  So parse it out here, and then search for a
3460		 * backend of this type.
3461		 */
3462		type = _IOC_TYPE(cmd);
3463
3464		STAILQ_FOREACH(backend, &softc->be_list, links) {
3465			if (backend->type == type) {
3466				found = 1;
3467				break;
3468			}
3469		}
3470		if (found == 0) {
3471			printf("ctl: unknown ioctl command %#lx or backend "
3472			       "%d\n", cmd, type);
3473			retval = EINVAL;
3474			break;
3475		}
3476		retval = backend->ioctl(dev, cmd, addr, flag, td);
3477#endif
3478		retval = ENOTTY;
3479		break;
3480	}
3481	}
3482	return (retval);
3483}
3484
3485uint32_t
3486ctl_get_initindex(struct ctl_nexus *nexus)
3487{
3488	if (nexus->targ_port < CTL_MAX_PORTS)
3489		return (nexus->initid.id +
3490			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3491	else
3492		return (nexus->initid.id +
3493		       ((nexus->targ_port - CTL_MAX_PORTS) *
3494			CTL_MAX_INIT_PER_PORT));
3495}
3496
3497uint32_t
3498ctl_get_resindex(struct ctl_nexus *nexus)
3499{
3500	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3501}
3502
3503uint32_t
3504ctl_port_idx(int port_num)
3505{
3506	if (port_num < CTL_MAX_PORTS)
3507		return(port_num);
3508	else
3509		return(port_num - CTL_MAX_PORTS);
3510}
3511
3512static uint32_t
3513ctl_map_lun(int port_num, uint32_t lun_id)
3514{
3515	struct ctl_port *port;
3516
3517	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3518	if (port == NULL)
3519		return (UINT32_MAX);
3520	if (port->lun_map == NULL)
3521		return (lun_id);
3522	return (port->lun_map(port->targ_lun_arg, lun_id));
3523}
3524
3525static uint32_t
3526ctl_map_lun_back(int port_num, uint32_t lun_id)
3527{
3528	struct ctl_port *port;
3529	uint32_t i;
3530
3531	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3532	if (port->lun_map == NULL)
3533		return (lun_id);
3534	for (i = 0; i < CTL_MAX_LUNS; i++) {
3535		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3536			return (i);
3537	}
3538	return (UINT32_MAX);
3539}
3540
3541/*
3542 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3543 * that are a power of 2.
3544 */
3545int
3546ctl_ffz(uint32_t *mask, uint32_t size)
3547{
3548	uint32_t num_chunks, num_pieces;
3549	int i, j;
3550
3551	num_chunks = (size >> 5);
3552	if (num_chunks == 0)
3553		num_chunks++;
3554	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3555
3556	for (i = 0; i < num_chunks; i++) {
3557		for (j = 0; j < num_pieces; j++) {
3558			if ((mask[i] & (1 << j)) == 0)
3559				return ((i << 5) + j);
3560		}
3561	}
3562
3563	return (-1);
3564}
3565
3566int
3567ctl_set_mask(uint32_t *mask, uint32_t bit)
3568{
3569	uint32_t chunk, piece;
3570
3571	chunk = bit >> 5;
3572	piece = bit % (sizeof(uint32_t) * 8);
3573
3574	if ((mask[chunk] & (1 << piece)) != 0)
3575		return (-1);
3576	else
3577		mask[chunk] |= (1 << piece);
3578
3579	return (0);
3580}
3581
3582int
3583ctl_clear_mask(uint32_t *mask, uint32_t bit)
3584{
3585	uint32_t chunk, piece;
3586
3587	chunk = bit >> 5;
3588	piece = bit % (sizeof(uint32_t) * 8);
3589
3590	if ((mask[chunk] & (1 << piece)) == 0)
3591		return (-1);
3592	else
3593		mask[chunk] &= ~(1 << piece);
3594
3595	return (0);
3596}
3597
3598int
3599ctl_is_set(uint32_t *mask, uint32_t bit)
3600{
3601	uint32_t chunk, piece;
3602
3603	chunk = bit >> 5;
3604	piece = bit % (sizeof(uint32_t) * 8);
3605
3606	if ((mask[chunk] & (1 << piece)) == 0)
3607		return (0);
3608	else
3609		return (1);
3610}
3611
3612#ifdef unused
3613/*
3614 * The bus, target and lun are optional, they can be filled in later.
3615 * can_wait is used to determine whether we can wait on the malloc or not.
3616 */
3617union ctl_io*
3618ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3619	      uint32_t targ_lun, int can_wait)
3620{
3621	union ctl_io *io;
3622
3623	if (can_wait)
3624		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3625	else
3626		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3627
3628	if (io != NULL) {
3629		io->io_hdr.io_type = io_type;
3630		io->io_hdr.targ_port = targ_port;
3631		/*
3632		 * XXX KDM this needs to change/go away.  We need to move
3633		 * to a preallocated pool of ctl_scsiio structures.
3634		 */
3635		io->io_hdr.nexus.targ_target.id = targ_target;
3636		io->io_hdr.nexus.targ_lun = targ_lun;
3637	}
3638
3639	return (io);
3640}
3641
3642void
3643ctl_kfree_io(union ctl_io *io)
3644{
3645	free(io, M_CTL);
3646}
3647#endif /* unused */
3648
3649/*
3650 * ctl_softc, pool_type, total_ctl_io are passed in.
3651 * npool is passed out.
3652 */
3653int
3654ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3655		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3656{
3657	uint32_t i;
3658	union ctl_io *cur_io, *next_io;
3659	struct ctl_io_pool *pool;
3660	int retval;
3661
3662	retval = 0;
3663
3664	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3665					    M_NOWAIT | M_ZERO);
3666	if (pool == NULL) {
3667		retval = ENOMEM;
3668		goto bailout;
3669	}
3670
3671	pool->type = pool_type;
3672	pool->ctl_softc = ctl_softc;
3673
3674	mtx_lock(&ctl_softc->pool_lock);
3675	pool->id = ctl_softc->cur_pool_id++;
3676	mtx_unlock(&ctl_softc->pool_lock);
3677
3678	pool->flags = CTL_POOL_FLAG_NONE;
3679	pool->refcount = 1;		/* Reference for validity. */
3680	STAILQ_INIT(&pool->free_queue);
3681
3682	/*
3683	 * XXX KDM other options here:
3684	 * - allocate a page at a time
3685	 * - allocate one big chunk of memory.
3686	 * Page allocation might work well, but would take a little more
3687	 * tracking.
3688	 */
3689	for (i = 0; i < total_ctl_io; i++) {
3690		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3691						M_NOWAIT);
3692		if (cur_io == NULL) {
3693			retval = ENOMEM;
3694			break;
3695		}
3696		cur_io->io_hdr.pool = pool;
3697		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3698		pool->total_ctl_io++;
3699		pool->free_ctl_io++;
3700	}
3701
3702	if (retval != 0) {
3703		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3704		     cur_io != NULL; cur_io = next_io) {
3705			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3706							      links);
3707			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3708				      ctl_io_hdr, links);
3709			free(cur_io, M_CTLIO);
3710		}
3711
3712		free(pool, M_CTL);
3713		goto bailout;
3714	}
3715	mtx_lock(&ctl_softc->pool_lock);
3716	ctl_softc->num_pools++;
3717	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3718	/*
3719	 * Increment our usage count if this is an external consumer, so we
3720	 * can't get unloaded until the external consumer (most likely a
3721	 * FETD) unloads and frees his pool.
3722	 *
3723	 * XXX KDM will this increment the caller's module use count, or
3724	 * mine?
3725	 */
3726#if 0
3727	if ((pool_type != CTL_POOL_EMERGENCY)
3728	 && (pool_type != CTL_POOL_INTERNAL)
3729	 && (pool_type != CTL_POOL_4OTHERSC))
3730		MOD_INC_USE_COUNT;
3731#endif
3732
3733	mtx_unlock(&ctl_softc->pool_lock);
3734
3735	*npool = pool;
3736
3737bailout:
3738
3739	return (retval);
3740}
3741
3742static int
3743ctl_pool_acquire(struct ctl_io_pool *pool)
3744{
3745
3746	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3747
3748	if (pool->flags & CTL_POOL_FLAG_INVALID)
3749		return (EINVAL);
3750
3751	pool->refcount++;
3752
3753	return (0);
3754}
3755
3756static void
3757ctl_pool_release(struct ctl_io_pool *pool)
3758{
3759	struct ctl_softc *ctl_softc = pool->ctl_softc;
3760	union ctl_io *io;
3761
3762	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3763
3764	if (--pool->refcount != 0)
3765		return;
3766
3767	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3768		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3769			      links);
3770		free(io, M_CTLIO);
3771	}
3772
3773	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3774	ctl_softc->num_pools--;
3775
3776	/*
3777	 * XXX KDM will this decrement the caller's usage count or mine?
3778	 */
3779#if 0
3780	if ((pool->type != CTL_POOL_EMERGENCY)
3781	 && (pool->type != CTL_POOL_INTERNAL)
3782	 && (pool->type != CTL_POOL_4OTHERSC))
3783		MOD_DEC_USE_COUNT;
3784#endif
3785
3786	free(pool, M_CTL);
3787}
3788
3789void
3790ctl_pool_free(struct ctl_io_pool *pool)
3791{
3792	struct ctl_softc *ctl_softc;
3793
3794	if (pool == NULL)
3795		return;
3796
3797	ctl_softc = pool->ctl_softc;
3798	mtx_lock(&ctl_softc->pool_lock);
3799	pool->flags |= CTL_POOL_FLAG_INVALID;
3800	ctl_pool_release(pool);
3801	mtx_unlock(&ctl_softc->pool_lock);
3802}
3803
3804/*
3805 * This routine does not block (except for spinlocks of course).
3806 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3807 * possible.
3808 */
3809union ctl_io *
3810ctl_alloc_io(void *pool_ref)
3811{
3812	union ctl_io *io;
3813	struct ctl_softc *ctl_softc;
3814	struct ctl_io_pool *pool, *npool;
3815	struct ctl_io_pool *emergency_pool;
3816
3817	pool = (struct ctl_io_pool *)pool_ref;
3818
3819	if (pool == NULL) {
3820		printf("%s: pool is NULL\n", __func__);
3821		return (NULL);
3822	}
3823
3824	emergency_pool = NULL;
3825
3826	ctl_softc = pool->ctl_softc;
3827
3828	mtx_lock(&ctl_softc->pool_lock);
3829	/*
3830	 * First, try to get the io structure from the user's pool.
3831	 */
3832	if (ctl_pool_acquire(pool) == 0) {
3833		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3834		if (io != NULL) {
3835			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3836			pool->total_allocated++;
3837			pool->free_ctl_io--;
3838			mtx_unlock(&ctl_softc->pool_lock);
3839			return (io);
3840		} else
3841			ctl_pool_release(pool);
3842	}
3843	/*
3844	 * If he doesn't have any io structures left, search for an
3845	 * emergency pool and grab one from there.
3846	 */
3847	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3848		if (npool->type != CTL_POOL_EMERGENCY)
3849			continue;
3850
3851		if (ctl_pool_acquire(npool) != 0)
3852			continue;
3853
3854		emergency_pool = npool;
3855
3856		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3857		if (io != NULL) {
3858			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3859			npool->total_allocated++;
3860			npool->free_ctl_io--;
3861			mtx_unlock(&ctl_softc->pool_lock);
3862			return (io);
3863		} else
3864			ctl_pool_release(npool);
3865	}
3866
3867	/* Drop the spinlock before we malloc */
3868	mtx_unlock(&ctl_softc->pool_lock);
3869
3870	/*
3871	 * The emergency pool (if it exists) didn't have one, so try an
3872	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3873	 */
3874	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3875	if (io != NULL) {
3876		/*
3877		 * If the emergency pool exists but is empty, add this
3878		 * ctl_io to its list when it gets freed.
3879		 */
3880		if (emergency_pool != NULL) {
3881			mtx_lock(&ctl_softc->pool_lock);
3882			if (ctl_pool_acquire(emergency_pool) == 0) {
3883				io->io_hdr.pool = emergency_pool;
3884				emergency_pool->total_ctl_io++;
3885				/*
3886				 * Need to bump this, otherwise
3887				 * total_allocated and total_freed won't
3888				 * match when we no longer have anything
3889				 * outstanding.
3890				 */
3891				emergency_pool->total_allocated++;
3892			}
3893			mtx_unlock(&ctl_softc->pool_lock);
3894		} else
3895			io->io_hdr.pool = NULL;
3896	}
3897
3898	return (io);
3899}
3900
3901void
3902ctl_free_io(union ctl_io *io)
3903{
3904	if (io == NULL)
3905		return;
3906
3907	/*
3908	 * If this ctl_io has a pool, return it to that pool.
3909	 */
3910	if (io->io_hdr.pool != NULL) {
3911		struct ctl_io_pool *pool;
3912
3913		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3914		mtx_lock(&pool->ctl_softc->pool_lock);
3915		io->io_hdr.io_type = 0xff;
3916		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3917		pool->total_freed++;
3918		pool->free_ctl_io++;
3919		ctl_pool_release(pool);
3920		mtx_unlock(&pool->ctl_softc->pool_lock);
3921	} else {
3922		/*
3923		 * Otherwise, just free it.  We probably malloced it and
3924		 * the emergency pool wasn't available.
3925		 */
3926		free(io, M_CTLIO);
3927	}
3928
3929}
3930
3931void
3932ctl_zero_io(union ctl_io *io)
3933{
3934	void *pool_ref;
3935
3936	if (io == NULL)
3937		return;
3938
3939	/*
3940	 * May need to preserve linked list pointers at some point too.
3941	 */
3942	pool_ref = io->io_hdr.pool;
3943
3944	memset(io, 0, sizeof(*io));
3945
3946	io->io_hdr.pool = pool_ref;
3947}
3948
3949/*
3950 * This routine is currently used for internal copies of ctl_ios that need
3951 * to persist for some reason after we've already returned status to the
3952 * FETD.  (Thus the flag set.)
3953 *
3954 * XXX XXX
3955 * Note that this makes a blind copy of all fields in the ctl_io, except
3956 * for the pool reference.  This includes any memory that has been
3957 * allocated!  That memory will no longer be valid after done has been
3958 * called, so this would be VERY DANGEROUS for command that actually does
3959 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3960 * start and stop commands, which don't transfer any data, so this is not a
3961 * problem.  If it is used for anything else, the caller would also need to
3962 * allocate data buffer space and this routine would need to be modified to
3963 * copy the data buffer(s) as well.
3964 */
3965void
3966ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3967{
3968	void *pool_ref;
3969
3970	if ((src == NULL)
3971	 || (dest == NULL))
3972		return;
3973
3974	/*
3975	 * May need to preserve linked list pointers at some point too.
3976	 */
3977	pool_ref = dest->io_hdr.pool;
3978
3979	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3980
3981	dest->io_hdr.pool = pool_ref;
3982	/*
3983	 * We need to know that this is an internal copy, and doesn't need
3984	 * to get passed back to the FETD that allocated it.
3985	 */
3986	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3987}
3988
3989/*
3990 * This routine could be used in the future to load default and/or saved
3991 * mode page parameters for a particuar lun.
3992 */
3993static int
3994ctl_init_page_index(struct ctl_lun *lun)
3995{
3996	int i;
3997	struct ctl_page_index *page_index;
3998	const char *value;
3999
4000	memcpy(&lun->mode_pages.index, page_index_template,
4001	       sizeof(page_index_template));
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_RW_ERROR_RECOVERY_PAGE: {
4018			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4019				panic("subpage is incorrect!");
4020			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4021			       &rw_er_page_default,
4022			       sizeof(rw_er_page_default));
4023			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4024			       &rw_er_page_changeable,
4025			       sizeof(rw_er_page_changeable));
4026			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4027			       &rw_er_page_default,
4028			       sizeof(rw_er_page_default));
4029			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4030			       &rw_er_page_default,
4031			       sizeof(rw_er_page_default));
4032			page_index->page_data =
4033				(uint8_t *)lun->mode_pages.rw_er_page;
4034			break;
4035		}
4036		case SMS_FORMAT_DEVICE_PAGE: {
4037			struct scsi_format_page *format_page;
4038
4039			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4040				panic("subpage is incorrect!");
4041
4042			/*
4043			 * Sectors per track are set above.  Bytes per
4044			 * sector need to be set here on a per-LUN basis.
4045			 */
4046			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4047			       &format_page_default,
4048			       sizeof(format_page_default));
4049			memcpy(&lun->mode_pages.format_page[
4050			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4051			       sizeof(format_page_changeable));
4052			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4053			       &format_page_default,
4054			       sizeof(format_page_default));
4055			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4056			       &format_page_default,
4057			       sizeof(format_page_default));
4058
4059			format_page = &lun->mode_pages.format_page[
4060				CTL_PAGE_CURRENT];
4061			scsi_ulto2b(lun->be_lun->blocksize,
4062				    format_page->bytes_per_sector);
4063
4064			format_page = &lun->mode_pages.format_page[
4065				CTL_PAGE_DEFAULT];
4066			scsi_ulto2b(lun->be_lun->blocksize,
4067				    format_page->bytes_per_sector);
4068
4069			format_page = &lun->mode_pages.format_page[
4070				CTL_PAGE_SAVED];
4071			scsi_ulto2b(lun->be_lun->blocksize,
4072				    format_page->bytes_per_sector);
4073
4074			page_index->page_data =
4075				(uint8_t *)lun->mode_pages.format_page;
4076			break;
4077		}
4078		case SMS_RIGID_DISK_PAGE: {
4079			struct scsi_rigid_disk_page *rigid_disk_page;
4080			uint32_t sectors_per_cylinder;
4081			uint64_t cylinders;
4082#ifndef	__XSCALE__
4083			int shift;
4084#endif /* !__XSCALE__ */
4085
4086			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4087				panic("invalid subpage value %d",
4088				      page_index->subpage);
4089
4090			/*
4091			 * Rotation rate and sectors per track are set
4092			 * above.  We calculate the cylinders here based on
4093			 * capacity.  Due to the number of heads and
4094			 * sectors per track we're using, smaller arrays
4095			 * may turn out to have 0 cylinders.  Linux and
4096			 * FreeBSD don't pay attention to these mode pages
4097			 * to figure out capacity, but Solaris does.  It
4098			 * seems to deal with 0 cylinders just fine, and
4099			 * works out a fake geometry based on the capacity.
4100			 */
4101			memcpy(&lun->mode_pages.rigid_disk_page[
4102			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4103			       sizeof(rigid_disk_page_default));
4104			memcpy(&lun->mode_pages.rigid_disk_page[
4105			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4106			       sizeof(rigid_disk_page_changeable));
4107			memcpy(&lun->mode_pages.rigid_disk_page[
4108			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4109			       sizeof(rigid_disk_page_default));
4110			memcpy(&lun->mode_pages.rigid_disk_page[
4111			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4112			       sizeof(rigid_disk_page_default));
4113
4114			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4115				CTL_DEFAULT_HEADS;
4116
4117			/*
4118			 * The divide method here will be more accurate,
4119			 * probably, but results in floating point being
4120			 * used in the kernel on i386 (__udivdi3()).  On the
4121			 * XScale, though, __udivdi3() is implemented in
4122			 * software.
4123			 *
4124			 * The shift method for cylinder calculation is
4125			 * accurate if sectors_per_cylinder is a power of
4126			 * 2.  Otherwise it might be slightly off -- you
4127			 * might have a bit of a truncation problem.
4128			 */
4129#ifdef	__XSCALE__
4130			cylinders = (lun->be_lun->maxlba + 1) /
4131				sectors_per_cylinder;
4132#else
4133			for (shift = 31; shift > 0; shift--) {
4134				if (sectors_per_cylinder & (1 << shift))
4135					break;
4136			}
4137			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4138#endif
4139
4140			/*
4141			 * We've basically got 3 bytes, or 24 bits for the
4142			 * cylinder size in the mode page.  If we're over,
4143			 * just round down to 2^24.
4144			 */
4145			if (cylinders > 0xffffff)
4146				cylinders = 0xffffff;
4147
4148			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4149				CTL_PAGE_CURRENT];
4150			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4151
4152			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4153				CTL_PAGE_DEFAULT];
4154			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4155
4156			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4157				CTL_PAGE_SAVED];
4158			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4159
4160			page_index->page_data =
4161				(uint8_t *)lun->mode_pages.rigid_disk_page;
4162			break;
4163		}
4164		case SMS_CACHING_PAGE: {
4165			struct scsi_caching_page *caching_page;
4166
4167			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4168				panic("invalid subpage value %d",
4169				      page_index->subpage);
4170			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4171			       &caching_page_default,
4172			       sizeof(caching_page_default));
4173			memcpy(&lun->mode_pages.caching_page[
4174			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4175			       sizeof(caching_page_changeable));
4176			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4177			       &caching_page_default,
4178			       sizeof(caching_page_default));
4179			caching_page = &lun->mode_pages.caching_page[
4180			    CTL_PAGE_SAVED];
4181			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4182			if (value != NULL && strcmp(value, "off") == 0)
4183				caching_page->flags1 &= ~SCP_WCE;
4184			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4185			if (value != NULL && strcmp(value, "off") == 0)
4186				caching_page->flags1 |= SCP_RCD;
4187			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4188			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4189			       sizeof(caching_page_default));
4190			page_index->page_data =
4191				(uint8_t *)lun->mode_pages.caching_page;
4192			break;
4193		}
4194		case SMS_CONTROL_MODE_PAGE: {
4195			struct scsi_control_page *control_page;
4196
4197			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4198				panic("invalid subpage value %d",
4199				      page_index->subpage);
4200
4201			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4202			       &control_page_default,
4203			       sizeof(control_page_default));
4204			memcpy(&lun->mode_pages.control_page[
4205			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4206			       sizeof(control_page_changeable));
4207			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4208			       &control_page_default,
4209			       sizeof(control_page_default));
4210			control_page = &lun->mode_pages.control_page[
4211			    CTL_PAGE_SAVED];
4212			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4213			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4214				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4215				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4216			}
4217			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4218			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4219			       sizeof(control_page_default));
4220			page_index->page_data =
4221				(uint8_t *)lun->mode_pages.control_page;
4222			break;
4223
4224		}
4225		case SMS_INFO_EXCEPTIONS_PAGE: {
4226			switch (page_index->subpage) {
4227			case SMS_SUBPAGE_PAGE_0:
4228				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4229				       &ie_page_default,
4230				       sizeof(ie_page_default));
4231				memcpy(&lun->mode_pages.ie_page[
4232				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4233				       sizeof(ie_page_changeable));
4234				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4235				       &ie_page_default,
4236				       sizeof(ie_page_default));
4237				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4238				       &ie_page_default,
4239				       sizeof(ie_page_default));
4240				page_index->page_data =
4241					(uint8_t *)lun->mode_pages.ie_page;
4242				break;
4243			case 0x02:
4244				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4245				       &lbp_page_default,
4246				       sizeof(lbp_page_default));
4247				memcpy(&lun->mode_pages.lbp_page[
4248				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4249				       sizeof(lbp_page_changeable));
4250				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4251				       &lbp_page_default,
4252				       sizeof(lbp_page_default));
4253				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4254				       &lbp_page_default,
4255				       sizeof(lbp_page_default));
4256				page_index->page_data =
4257					(uint8_t *)lun->mode_pages.lbp_page;
4258			}
4259			break;
4260		}
4261		case SMS_VENDOR_SPECIFIC_PAGE:{
4262			switch (page_index->subpage) {
4263			case DBGCNF_SUBPAGE_CODE: {
4264				struct copan_debugconf_subpage *current_page,
4265							       *saved_page;
4266
4267				memcpy(&lun->mode_pages.debugconf_subpage[
4268				       CTL_PAGE_CURRENT],
4269				       &debugconf_page_default,
4270				       sizeof(debugconf_page_default));
4271				memcpy(&lun->mode_pages.debugconf_subpage[
4272				       CTL_PAGE_CHANGEABLE],
4273				       &debugconf_page_changeable,
4274				       sizeof(debugconf_page_changeable));
4275				memcpy(&lun->mode_pages.debugconf_subpage[
4276				       CTL_PAGE_DEFAULT],
4277				       &debugconf_page_default,
4278				       sizeof(debugconf_page_default));
4279				memcpy(&lun->mode_pages.debugconf_subpage[
4280				       CTL_PAGE_SAVED],
4281				       &debugconf_page_default,
4282				       sizeof(debugconf_page_default));
4283				page_index->page_data =
4284					(uint8_t *)lun->mode_pages.debugconf_subpage;
4285
4286				current_page = (struct copan_debugconf_subpage *)
4287					(page_index->page_data +
4288					 (page_index->page_len *
4289					  CTL_PAGE_CURRENT));
4290				saved_page = (struct copan_debugconf_subpage *)
4291					(page_index->page_data +
4292					 (page_index->page_len *
4293					  CTL_PAGE_SAVED));
4294				break;
4295			}
4296			default:
4297				panic("invalid subpage value %d",
4298				      page_index->subpage);
4299				break;
4300			}
4301   			break;
4302		}
4303		default:
4304			panic("invalid page value %d",
4305			      page_index->page_code & SMPH_PC_MASK);
4306			break;
4307    	}
4308	}
4309
4310	return (CTL_RETVAL_COMPLETE);
4311}
4312
4313static int
4314ctl_init_log_page_index(struct ctl_lun *lun)
4315{
4316	struct ctl_page_index *page_index;
4317	int i, j, prev;
4318
4319	memcpy(&lun->log_pages.index, log_page_index_template,
4320	       sizeof(log_page_index_template));
4321
4322	prev = -1;
4323	for (i = 0, j = 0; i < CTL_NUM_LOG_PAGES; i++) {
4324
4325		page_index = &lun->log_pages.index[i];
4326		/*
4327		 * If this is a disk-only mode page, there's no point in
4328		 * setting it up.  For some pages, we have to have some
4329		 * basic information about the disk in order to calculate the
4330		 * mode page data.
4331		 */
4332		if ((lun->be_lun->lun_type != T_DIRECT)
4333		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4334			continue;
4335
4336		if (page_index->page_code != prev) {
4337			lun->log_pages.pages_page[j] = page_index->page_code;
4338			prev = page_index->page_code;
4339			j++;
4340		}
4341		lun->log_pages.subpages_page[i*2] = page_index->page_code;
4342		lun->log_pages.subpages_page[i*2+1] = page_index->subpage;
4343	}
4344	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4345	lun->log_pages.index[0].page_len = j;
4346	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4347	lun->log_pages.index[1].page_len = i * 2;
4348
4349	return (CTL_RETVAL_COMPLETE);
4350}
4351
4352/*
4353 * LUN allocation.
4354 *
4355 * Requirements:
4356 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4357 *   wants us to allocate the LUN and he can block.
4358 * - ctl_softc is always set
4359 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4360 *
4361 * Returns 0 for success, non-zero (errno) for failure.
4362 */
4363static int
4364ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4365	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4366{
4367	struct ctl_lun *nlun, *lun;
4368	struct ctl_port *port;
4369	struct scsi_vpd_id_descriptor *desc;
4370	struct scsi_vpd_id_t10 *t10id;
4371	const char *eui, *naa, *scsiname, *vendor, *value;
4372	int lun_number, i, lun_malloced;
4373	int devidlen, idlen1, idlen2 = 0, len;
4374
4375	if (be_lun == NULL)
4376		return (EINVAL);
4377
4378	/*
4379	 * We currently only support Direct Access or Processor LUN types.
4380	 */
4381	switch (be_lun->lun_type) {
4382	case T_DIRECT:
4383		break;
4384	case T_PROCESSOR:
4385		break;
4386	case T_SEQUENTIAL:
4387	case T_CHANGER:
4388	default:
4389		be_lun->lun_config_status(be_lun->be_lun,
4390					  CTL_LUN_CONFIG_FAILURE);
4391		break;
4392	}
4393	if (ctl_lun == NULL) {
4394		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4395		lun_malloced = 1;
4396	} else {
4397		lun_malloced = 0;
4398		lun = ctl_lun;
4399	}
4400
4401	memset(lun, 0, sizeof(*lun));
4402	if (lun_malloced)
4403		lun->flags = CTL_LUN_MALLOCED;
4404
4405	/* Generate LUN ID. */
4406	devidlen = max(CTL_DEVID_MIN_LEN,
4407	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4408	idlen1 = sizeof(*t10id) + devidlen;
4409	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4410	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4411	if (scsiname != NULL) {
4412		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4413		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4414	}
4415	eui = ctl_get_opt(&be_lun->options, "eui");
4416	if (eui != NULL) {
4417		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4418	}
4419	naa = ctl_get_opt(&be_lun->options, "naa");
4420	if (naa != NULL) {
4421		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4422	}
4423	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4424	    M_CTL, M_WAITOK | M_ZERO);
4425	lun->lun_devid->len = len;
4426	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4427	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4428	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4429	desc->length = idlen1;
4430	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4431	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4432	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4433		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4434	} else {
4435		strncpy(t10id->vendor, vendor,
4436		    min(sizeof(t10id->vendor), strlen(vendor)));
4437	}
4438	strncpy((char *)t10id->vendor_spec_id,
4439	    (char *)be_lun->device_id, devidlen);
4440	if (scsiname != NULL) {
4441		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4442		    desc->length);
4443		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4444		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4445		    SVPD_ID_TYPE_SCSI_NAME;
4446		desc->length = idlen2;
4447		strlcpy(desc->identifier, scsiname, idlen2);
4448	}
4449	if (eui != NULL) {
4450		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4451		    desc->length);
4452		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4453		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4454		    SVPD_ID_TYPE_EUI64;
4455		desc->length = 8;
4456		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4457	}
4458	if (naa != NULL) {
4459		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4460		    desc->length);
4461		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4462		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4463		    SVPD_ID_TYPE_NAA;
4464		desc->length = 8;
4465		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4466	}
4467
4468	mtx_lock(&ctl_softc->ctl_lock);
4469	/*
4470	 * See if the caller requested a particular LUN number.  If so, see
4471	 * if it is available.  Otherwise, allocate the first available LUN.
4472	 */
4473	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4474		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4475		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4476			mtx_unlock(&ctl_softc->ctl_lock);
4477			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4478				printf("ctl: requested LUN ID %d is higher "
4479				       "than CTL_MAX_LUNS - 1 (%d)\n",
4480				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4481			} else {
4482				/*
4483				 * XXX KDM return an error, or just assign
4484				 * another LUN ID in this case??
4485				 */
4486				printf("ctl: requested LUN ID %d is already "
4487				       "in use\n", be_lun->req_lun_id);
4488			}
4489			if (lun->flags & CTL_LUN_MALLOCED)
4490				free(lun, M_CTL);
4491			be_lun->lun_config_status(be_lun->be_lun,
4492						  CTL_LUN_CONFIG_FAILURE);
4493			return (ENOSPC);
4494		}
4495		lun_number = be_lun->req_lun_id;
4496	} else {
4497		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4498		if (lun_number == -1) {
4499			mtx_unlock(&ctl_softc->ctl_lock);
4500			printf("ctl: can't allocate LUN on target %ju, out of "
4501			       "LUNs\n", (uintmax_t)target_id.id);
4502			if (lun->flags & CTL_LUN_MALLOCED)
4503				free(lun, M_CTL);
4504			be_lun->lun_config_status(be_lun->be_lun,
4505						  CTL_LUN_CONFIG_FAILURE);
4506			return (ENOSPC);
4507		}
4508	}
4509	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4510
4511	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4512	lun->target = target_id;
4513	lun->lun = lun_number;
4514	lun->be_lun = be_lun;
4515	/*
4516	 * The processor LUN is always enabled.  Disk LUNs come on line
4517	 * disabled, and must be enabled by the backend.
4518	 */
4519	lun->flags |= CTL_LUN_DISABLED;
4520	lun->backend = be_lun->be;
4521	be_lun->ctl_lun = lun;
4522	be_lun->lun_id = lun_number;
4523	atomic_add_int(&be_lun->be->num_luns, 1);
4524	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4525		lun->flags |= CTL_LUN_OFFLINE;
4526
4527	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4528		lun->flags |= CTL_LUN_STOPPED;
4529
4530	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4531		lun->flags |= CTL_LUN_INOPERABLE;
4532
4533	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4534		lun->flags |= CTL_LUN_PRIMARY_SC;
4535
4536	value = ctl_get_opt(&be_lun->options, "readonly");
4537	if (value != NULL && strcmp(value, "on") == 0)
4538		lun->flags |= CTL_LUN_READONLY;
4539
4540	lun->ctl_softc = ctl_softc;
4541	TAILQ_INIT(&lun->ooa_queue);
4542	TAILQ_INIT(&lun->blocked_queue);
4543	STAILQ_INIT(&lun->error_list);
4544	ctl_tpc_lun_init(lun);
4545
4546	/*
4547	 * Initialize the mode and log page index.
4548	 */
4549	ctl_init_page_index(lun);
4550	ctl_init_log_page_index(lun);
4551
4552	/*
4553	 * Set the poweron UA for all initiators on this LUN only.
4554	 */
4555	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4556		lun->pending_ua[i] = CTL_UA_POWERON;
4557
4558	/*
4559	 * Now, before we insert this lun on the lun list, set the lun
4560	 * inventory changed UA for all other luns.
4561	 */
4562	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4563		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4564			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4565		}
4566	}
4567
4568	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4569
4570	ctl_softc->ctl_luns[lun_number] = lun;
4571
4572	ctl_softc->num_luns++;
4573
4574	/* Setup statistics gathering */
4575	lun->stats.device_type = be_lun->lun_type;
4576	lun->stats.lun_number = lun_number;
4577	if (lun->stats.device_type == T_DIRECT)
4578		lun->stats.blocksize = be_lun->blocksize;
4579	else
4580		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4581	for (i = 0;i < CTL_MAX_PORTS;i++)
4582		lun->stats.ports[i].targ_port = i;
4583
4584	mtx_unlock(&ctl_softc->ctl_lock);
4585
4586	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4587
4588	/*
4589	 * Run through each registered FETD and bring it online if it isn't
4590	 * already.  Enable the target ID if it hasn't been enabled, and
4591	 * enable this particular LUN.
4592	 */
4593	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4594		int retval;
4595
4596		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4597		if (retval != 0) {
4598			printf("ctl_alloc_lun: FETD %s port %d returned error "
4599			       "%d for lun_enable on target %ju lun %d\n",
4600			       port->port_name, port->targ_port, retval,
4601			       (uintmax_t)target_id.id, lun_number);
4602		} else
4603			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4604	}
4605	return (0);
4606}
4607
4608/*
4609 * Delete a LUN.
4610 * Assumptions:
4611 * - LUN has already been marked invalid and any pending I/O has been taken
4612 *   care of.
4613 */
4614static int
4615ctl_free_lun(struct ctl_lun *lun)
4616{
4617	struct ctl_softc *softc;
4618#if 0
4619	struct ctl_port *port;
4620#endif
4621	struct ctl_lun *nlun;
4622	int i;
4623
4624	softc = lun->ctl_softc;
4625
4626	mtx_assert(&softc->ctl_lock, MA_OWNED);
4627
4628	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4629
4630	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4631
4632	softc->ctl_luns[lun->lun] = NULL;
4633
4634	if (!TAILQ_EMPTY(&lun->ooa_queue))
4635		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4636
4637	softc->num_luns--;
4638
4639	/*
4640	 * XXX KDM this scheme only works for a single target/multiple LUN
4641	 * setup.  It needs to be revamped for a multiple target scheme.
4642	 *
4643	 * XXX KDM this results in port->lun_disable() getting called twice,
4644	 * once when ctl_disable_lun() is called, and a second time here.
4645	 * We really need to re-think the LUN disable semantics.  There
4646	 * should probably be several steps/levels to LUN removal:
4647	 *  - disable
4648	 *  - invalidate
4649	 *  - free
4650 	 *
4651	 * Right now we only have a disable method when communicating to
4652	 * the front end ports, at least for individual LUNs.
4653	 */
4654#if 0
4655	STAILQ_FOREACH(port, &softc->port_list, links) {
4656		int retval;
4657
4658		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4659					 lun->lun);
4660		if (retval != 0) {
4661			printf("ctl_free_lun: FETD %s port %d returned error "
4662			       "%d for lun_disable on target %ju lun %jd\n",
4663			       port->port_name, port->targ_port, retval,
4664			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4665		}
4666
4667		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4668			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4669
4670			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4671			if (retval != 0) {
4672				printf("ctl_free_lun: FETD %s port %d "
4673				       "returned error %d for targ_disable on "
4674				       "target %ju\n", port->port_name,
4675				       port->targ_port, retval,
4676				       (uintmax_t)lun->target.id);
4677			} else
4678				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4679
4680			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4681				continue;
4682
4683#if 0
4684			port->port_offline(port->onoff_arg);
4685			port->status &= ~CTL_PORT_STATUS_ONLINE;
4686#endif
4687		}
4688	}
4689#endif
4690
4691	/*
4692	 * Tell the backend to free resources, if this LUN has a backend.
4693	 */
4694	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4695	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4696
4697	ctl_tpc_lun_shutdown(lun);
4698	mtx_destroy(&lun->lun_lock);
4699	free(lun->lun_devid, M_CTL);
4700	if (lun->flags & CTL_LUN_MALLOCED)
4701		free(lun, M_CTL);
4702
4703	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4704		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4705			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4706		}
4707	}
4708
4709	return (0);
4710}
4711
4712static void
4713ctl_create_lun(struct ctl_be_lun *be_lun)
4714{
4715	struct ctl_softc *ctl_softc;
4716
4717	ctl_softc = control_softc;
4718
4719	/*
4720	 * ctl_alloc_lun() should handle all potential failure cases.
4721	 */
4722	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4723}
4724
4725int
4726ctl_add_lun(struct ctl_be_lun *be_lun)
4727{
4728	struct ctl_softc *ctl_softc = control_softc;
4729
4730	mtx_lock(&ctl_softc->ctl_lock);
4731	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4732	mtx_unlock(&ctl_softc->ctl_lock);
4733	wakeup(&ctl_softc->pending_lun_queue);
4734
4735	return (0);
4736}
4737
4738int
4739ctl_enable_lun(struct ctl_be_lun *be_lun)
4740{
4741	struct ctl_softc *ctl_softc;
4742	struct ctl_port *port, *nport;
4743	struct ctl_lun *lun;
4744	int retval;
4745
4746	ctl_softc = control_softc;
4747
4748	lun = (struct ctl_lun *)be_lun->ctl_lun;
4749
4750	mtx_lock(&ctl_softc->ctl_lock);
4751	mtx_lock(&lun->lun_lock);
4752	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4753		/*
4754		 * eh?  Why did we get called if the LUN is already
4755		 * enabled?
4756		 */
4757		mtx_unlock(&lun->lun_lock);
4758		mtx_unlock(&ctl_softc->ctl_lock);
4759		return (0);
4760	}
4761	lun->flags &= ~CTL_LUN_DISABLED;
4762	mtx_unlock(&lun->lun_lock);
4763
4764	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4765		nport = STAILQ_NEXT(port, links);
4766
4767		/*
4768		 * Drop the lock while we call the FETD's enable routine.
4769		 * This can lead to a callback into CTL (at least in the
4770		 * case of the internal initiator frontend.
4771		 */
4772		mtx_unlock(&ctl_softc->ctl_lock);
4773		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4774		mtx_lock(&ctl_softc->ctl_lock);
4775		if (retval != 0) {
4776			printf("%s: FETD %s port %d returned error "
4777			       "%d for lun_enable on target %ju lun %jd\n",
4778			       __func__, port->port_name, port->targ_port, retval,
4779			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4780		}
4781#if 0
4782		 else {
4783            /* NOTE:  TODO:  why does lun enable affect port status? */
4784			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4785		}
4786#endif
4787	}
4788
4789	mtx_unlock(&ctl_softc->ctl_lock);
4790
4791	return (0);
4792}
4793
4794int
4795ctl_disable_lun(struct ctl_be_lun *be_lun)
4796{
4797	struct ctl_softc *ctl_softc;
4798	struct ctl_port *port;
4799	struct ctl_lun *lun;
4800	int retval;
4801
4802	ctl_softc = control_softc;
4803
4804	lun = (struct ctl_lun *)be_lun->ctl_lun;
4805
4806	mtx_lock(&ctl_softc->ctl_lock);
4807	mtx_lock(&lun->lun_lock);
4808	if (lun->flags & CTL_LUN_DISABLED) {
4809		mtx_unlock(&lun->lun_lock);
4810		mtx_unlock(&ctl_softc->ctl_lock);
4811		return (0);
4812	}
4813	lun->flags |= CTL_LUN_DISABLED;
4814	mtx_unlock(&lun->lun_lock);
4815
4816	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4817		mtx_unlock(&ctl_softc->ctl_lock);
4818		/*
4819		 * Drop the lock before we call the frontend's disable
4820		 * routine, to avoid lock order reversals.
4821		 *
4822		 * XXX KDM what happens if the frontend list changes while
4823		 * we're traversing it?  It's unlikely, but should be handled.
4824		 */
4825		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4826					 lun->lun);
4827		mtx_lock(&ctl_softc->ctl_lock);
4828		if (retval != 0) {
4829			printf("ctl_alloc_lun: FETD %s port %d returned error "
4830			       "%d for lun_disable on target %ju lun %jd\n",
4831			       port->port_name, port->targ_port, retval,
4832			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4833		}
4834	}
4835
4836	mtx_unlock(&ctl_softc->ctl_lock);
4837
4838	return (0);
4839}
4840
4841int
4842ctl_start_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	lun->flags &= ~CTL_LUN_STOPPED;
4853	mtx_unlock(&lun->lun_lock);
4854
4855	return (0);
4856}
4857
4858int
4859ctl_stop_lun(struct ctl_be_lun *be_lun)
4860{
4861	struct ctl_softc *ctl_softc;
4862	struct ctl_lun *lun;
4863
4864	ctl_softc = control_softc;
4865
4866	lun = (struct ctl_lun *)be_lun->ctl_lun;
4867
4868	mtx_lock(&lun->lun_lock);
4869	lun->flags |= CTL_LUN_STOPPED;
4870	mtx_unlock(&lun->lun_lock);
4871
4872	return (0);
4873}
4874
4875int
4876ctl_lun_offline(struct ctl_be_lun *be_lun)
4877{
4878	struct ctl_softc *ctl_softc;
4879	struct ctl_lun *lun;
4880
4881	ctl_softc = control_softc;
4882
4883	lun = (struct ctl_lun *)be_lun->ctl_lun;
4884
4885	mtx_lock(&lun->lun_lock);
4886	lun->flags |= CTL_LUN_OFFLINE;
4887	mtx_unlock(&lun->lun_lock);
4888
4889	return (0);
4890}
4891
4892int
4893ctl_lun_online(struct ctl_be_lun *be_lun)
4894{
4895	struct ctl_softc *ctl_softc;
4896	struct ctl_lun *lun;
4897
4898	ctl_softc = control_softc;
4899
4900	lun = (struct ctl_lun *)be_lun->ctl_lun;
4901
4902	mtx_lock(&lun->lun_lock);
4903	lun->flags &= ~CTL_LUN_OFFLINE;
4904	mtx_unlock(&lun->lun_lock);
4905
4906	return (0);
4907}
4908
4909int
4910ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4911{
4912	struct ctl_softc *ctl_softc;
4913	struct ctl_lun *lun;
4914
4915	ctl_softc = control_softc;
4916
4917	lun = (struct ctl_lun *)be_lun->ctl_lun;
4918
4919	mtx_lock(&lun->lun_lock);
4920
4921	/*
4922	 * The LUN needs to be disabled before it can be marked invalid.
4923	 */
4924	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4925		mtx_unlock(&lun->lun_lock);
4926		return (-1);
4927	}
4928	/*
4929	 * Mark the LUN invalid.
4930	 */
4931	lun->flags |= CTL_LUN_INVALID;
4932
4933	/*
4934	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4935	 * If we have something in the OOA queue, we'll free it when the
4936	 * last I/O completes.
4937	 */
4938	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4939		mtx_unlock(&lun->lun_lock);
4940		mtx_lock(&ctl_softc->ctl_lock);
4941		ctl_free_lun(lun);
4942		mtx_unlock(&ctl_softc->ctl_lock);
4943	} else
4944		mtx_unlock(&lun->lun_lock);
4945
4946	return (0);
4947}
4948
4949int
4950ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4951{
4952	struct ctl_softc *ctl_softc;
4953	struct ctl_lun *lun;
4954
4955	ctl_softc = control_softc;
4956	lun = (struct ctl_lun *)be_lun->ctl_lun;
4957
4958	mtx_lock(&lun->lun_lock);
4959	lun->flags |= CTL_LUN_INOPERABLE;
4960	mtx_unlock(&lun->lun_lock);
4961
4962	return (0);
4963}
4964
4965int
4966ctl_lun_operable(struct ctl_be_lun *be_lun)
4967{
4968	struct ctl_softc *ctl_softc;
4969	struct ctl_lun *lun;
4970
4971	ctl_softc = control_softc;
4972	lun = (struct ctl_lun *)be_lun->ctl_lun;
4973
4974	mtx_lock(&lun->lun_lock);
4975	lun->flags &= ~CTL_LUN_INOPERABLE;
4976	mtx_unlock(&lun->lun_lock);
4977
4978	return (0);
4979}
4980
4981void
4982ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4983{
4984	struct ctl_lun *lun;
4985	struct ctl_softc *softc;
4986	int i;
4987
4988	softc = control_softc;
4989
4990	lun = (struct ctl_lun *)be_lun->ctl_lun;
4991
4992	mtx_lock(&lun->lun_lock);
4993
4994	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4995		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
4996
4997	mtx_unlock(&lun->lun_lock);
4998}
4999
5000/*
5001 * Backend "memory move is complete" callback for requests that never
5002 * make it down to say RAIDCore's configuration code.
5003 */
5004int
5005ctl_config_move_done(union ctl_io *io)
5006{
5007	int retval;
5008
5009	retval = CTL_RETVAL_COMPLETE;
5010
5011
5012	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5013	/*
5014	 * XXX KDM this shouldn't happen, but what if it does?
5015	 */
5016	if (io->io_hdr.io_type != CTL_IO_SCSI)
5017		panic("I/O type isn't CTL_IO_SCSI!");
5018
5019	if ((io->io_hdr.port_status == 0)
5020	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5021	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5022		io->io_hdr.status = CTL_SUCCESS;
5023	else if ((io->io_hdr.port_status != 0)
5024	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5025	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5026		/*
5027		 * For hardware error sense keys, the sense key
5028		 * specific value is defined to be a retry count,
5029		 * but we use it to pass back an internal FETD
5030		 * error code.  XXX KDM  Hopefully the FETD is only
5031		 * using 16 bits for an error code, since that's
5032		 * all the space we have in the sks field.
5033		 */
5034		ctl_set_internal_failure(&io->scsiio,
5035					 /*sks_valid*/ 1,
5036					 /*retry_count*/
5037					 io->io_hdr.port_status);
5038		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5039			free(io->scsiio.kern_data_ptr, M_CTL);
5040		ctl_done(io);
5041		goto bailout;
5042	}
5043
5044	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5045	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5046	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5047		/*
5048		 * XXX KDM just assuming a single pointer here, and not a
5049		 * S/G list.  If we start using S/G lists for config data,
5050		 * we'll need to know how to clean them up here as well.
5051		 */
5052		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5053			free(io->scsiio.kern_data_ptr, M_CTL);
5054		/* Hopefully the user has already set the status... */
5055		ctl_done(io);
5056	} else {
5057		/*
5058		 * XXX KDM now we need to continue data movement.  Some
5059		 * options:
5060		 * - call ctl_scsiio() again?  We don't do this for data
5061		 *   writes, because for those at least we know ahead of
5062		 *   time where the write will go and how long it is.  For
5063		 *   config writes, though, that information is largely
5064		 *   contained within the write itself, thus we need to
5065		 *   parse out the data again.
5066		 *
5067		 * - Call some other function once the data is in?
5068		 */
5069		if (ctl_debug & CTL_DEBUG_CDB_DATA)
5070			ctl_data_print(io);
5071
5072		/*
5073		 * XXX KDM call ctl_scsiio() again for now, and check flag
5074		 * bits to see whether we're allocated or not.
5075		 */
5076		retval = ctl_scsiio(&io->scsiio);
5077	}
5078bailout:
5079	return (retval);
5080}
5081
5082/*
5083 * This gets called by a backend driver when it is done with a
5084 * data_submit method.
5085 */
5086void
5087ctl_data_submit_done(union ctl_io *io)
5088{
5089	/*
5090	 * If the IO_CONT flag is set, we need to call the supplied
5091	 * function to continue processing the I/O, instead of completing
5092	 * the I/O just yet.
5093	 *
5094	 * If there is an error, though, we don't want to keep processing.
5095	 * Instead, just send status back to the initiator.
5096	 */
5097	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5098	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5099	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5100	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5101		io->scsiio.io_cont(io);
5102		return;
5103	}
5104	ctl_done(io);
5105}
5106
5107/*
5108 * This gets called by a backend driver when it is done with a
5109 * configuration write.
5110 */
5111void
5112ctl_config_write_done(union ctl_io *io)
5113{
5114	uint8_t *buf;
5115
5116	/*
5117	 * If the IO_CONT flag is set, we need to call the supplied
5118	 * function to continue processing the I/O, instead of completing
5119	 * the I/O just yet.
5120	 *
5121	 * If there is an error, though, we don't want to keep processing.
5122	 * Instead, just send status back to the initiator.
5123	 */
5124	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5125	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5126	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5127	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5128		io->scsiio.io_cont(io);
5129		return;
5130	}
5131	/*
5132	 * Since a configuration write can be done for commands that actually
5133	 * have data allocated, like write buffer, and commands that have
5134	 * no data, like start/stop unit, we need to check here.
5135	 */
5136	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5137		buf = io->scsiio.kern_data_ptr;
5138	else
5139		buf = NULL;
5140	ctl_done(io);
5141	if (buf)
5142		free(buf, M_CTL);
5143}
5144
5145/*
5146 * SCSI release command.
5147 */
5148int
5149ctl_scsi_release(struct ctl_scsiio *ctsio)
5150{
5151	int length, longid, thirdparty_id, resv_id;
5152	struct ctl_softc *ctl_softc;
5153	struct ctl_lun *lun;
5154	uint32_t residx;
5155
5156	length = 0;
5157	resv_id = 0;
5158
5159	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5160
5161	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5162	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5163	ctl_softc = control_softc;
5164
5165	switch (ctsio->cdb[0]) {
5166	case RELEASE_10: {
5167		struct scsi_release_10 *cdb;
5168
5169		cdb = (struct scsi_release_10 *)ctsio->cdb;
5170
5171		if (cdb->byte2 & SR10_LONGID)
5172			longid = 1;
5173		else
5174			thirdparty_id = cdb->thirdparty_id;
5175
5176		resv_id = cdb->resv_id;
5177		length = scsi_2btoul(cdb->length);
5178		break;
5179	}
5180	}
5181
5182
5183	/*
5184	 * XXX KDM right now, we only support LUN reservation.  We don't
5185	 * support 3rd party reservations, or extent reservations, which
5186	 * might actually need the parameter list.  If we've gotten this
5187	 * far, we've got a LUN reservation.  Anything else got kicked out
5188	 * above.  So, according to SPC, ignore the length.
5189	 */
5190	length = 0;
5191
5192	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5193	 && (length > 0)) {
5194		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5195		ctsio->kern_data_len = length;
5196		ctsio->kern_total_len = length;
5197		ctsio->kern_data_resid = 0;
5198		ctsio->kern_rel_offset = 0;
5199		ctsio->kern_sg_entries = 0;
5200		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5201		ctsio->be_move_done = ctl_config_move_done;
5202		ctl_datamove((union ctl_io *)ctsio);
5203
5204		return (CTL_RETVAL_COMPLETE);
5205	}
5206
5207	if (length > 0)
5208		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5209
5210	mtx_lock(&lun->lun_lock);
5211
5212	/*
5213	 * According to SPC, it is not an error for an intiator to attempt
5214	 * to release a reservation on a LUN that isn't reserved, or that
5215	 * is reserved by another initiator.  The reservation can only be
5216	 * released, though, by the initiator who made it or by one of
5217	 * several reset type events.
5218	 */
5219	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5220			lun->flags &= ~CTL_LUN_RESERVED;
5221
5222	mtx_unlock(&lun->lun_lock);
5223
5224	ctsio->scsi_status = SCSI_STATUS_OK;
5225	ctsio->io_hdr.status = CTL_SUCCESS;
5226
5227	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5228		free(ctsio->kern_data_ptr, M_CTL);
5229		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5230	}
5231
5232	ctl_done((union ctl_io *)ctsio);
5233	return (CTL_RETVAL_COMPLETE);
5234}
5235
5236int
5237ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5238{
5239	int extent, thirdparty, longid;
5240	int resv_id, length;
5241	uint64_t thirdparty_id;
5242	struct ctl_softc *ctl_softc;
5243	struct ctl_lun *lun;
5244	uint32_t residx;
5245
5246	extent = 0;
5247	thirdparty = 0;
5248	longid = 0;
5249	resv_id = 0;
5250	length = 0;
5251	thirdparty_id = 0;
5252
5253	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5254
5255	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5256	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5257	ctl_softc = control_softc;
5258
5259	switch (ctsio->cdb[0]) {
5260	case RESERVE_10: {
5261		struct scsi_reserve_10 *cdb;
5262
5263		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5264
5265		if (cdb->byte2 & SR10_LONGID)
5266			longid = 1;
5267		else
5268			thirdparty_id = cdb->thirdparty_id;
5269
5270		resv_id = cdb->resv_id;
5271		length = scsi_2btoul(cdb->length);
5272		break;
5273	}
5274	}
5275
5276	/*
5277	 * XXX KDM right now, we only support LUN reservation.  We don't
5278	 * support 3rd party reservations, or extent reservations, which
5279	 * might actually need the parameter list.  If we've gotten this
5280	 * far, we've got a LUN reservation.  Anything else got kicked out
5281	 * above.  So, according to SPC, ignore the length.
5282	 */
5283	length = 0;
5284
5285	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5286	 && (length > 0)) {
5287		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5288		ctsio->kern_data_len = length;
5289		ctsio->kern_total_len = length;
5290		ctsio->kern_data_resid = 0;
5291		ctsio->kern_rel_offset = 0;
5292		ctsio->kern_sg_entries = 0;
5293		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5294		ctsio->be_move_done = ctl_config_move_done;
5295		ctl_datamove((union ctl_io *)ctsio);
5296
5297		return (CTL_RETVAL_COMPLETE);
5298	}
5299
5300	if (length > 0)
5301		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5302
5303	mtx_lock(&lun->lun_lock);
5304	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5305		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5306		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5307		goto bailout;
5308	}
5309
5310	lun->flags |= CTL_LUN_RESERVED;
5311	lun->res_idx = residx;
5312
5313	ctsio->scsi_status = SCSI_STATUS_OK;
5314	ctsio->io_hdr.status = CTL_SUCCESS;
5315
5316bailout:
5317	mtx_unlock(&lun->lun_lock);
5318
5319	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5320		free(ctsio->kern_data_ptr, M_CTL);
5321		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5322	}
5323
5324	ctl_done((union ctl_io *)ctsio);
5325	return (CTL_RETVAL_COMPLETE);
5326}
5327
5328int
5329ctl_start_stop(struct ctl_scsiio *ctsio)
5330{
5331	struct scsi_start_stop_unit *cdb;
5332	struct ctl_lun *lun;
5333	struct ctl_softc *ctl_softc;
5334	int retval;
5335
5336	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5337
5338	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5339	ctl_softc = control_softc;
5340	retval = 0;
5341
5342	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5343
5344	/*
5345	 * XXX KDM
5346	 * We don't support the immediate bit on a stop unit.  In order to
5347	 * do that, we would need to code up a way to know that a stop is
5348	 * pending, and hold off any new commands until it completes, one
5349	 * way or another.  Then we could accept or reject those commands
5350	 * depending on its status.  We would almost need to do the reverse
5351	 * of what we do below for an immediate start -- return the copy of
5352	 * the ctl_io to the FETD with status to send to the host (and to
5353	 * free the copy!) and then free the original I/O once the stop
5354	 * actually completes.  That way, the OOA queue mechanism can work
5355	 * to block commands that shouldn't proceed.  Another alternative
5356	 * would be to put the copy in the queue in place of the original,
5357	 * and return the original back to the caller.  That could be
5358	 * slightly safer..
5359	 */
5360	if ((cdb->byte2 & SSS_IMMED)
5361	 && ((cdb->how & SSS_START) == 0)) {
5362		ctl_set_invalid_field(ctsio,
5363				      /*sks_valid*/ 1,
5364				      /*command*/ 1,
5365				      /*field*/ 1,
5366				      /*bit_valid*/ 1,
5367				      /*bit*/ 0);
5368		ctl_done((union ctl_io *)ctsio);
5369		return (CTL_RETVAL_COMPLETE);
5370	}
5371
5372	if ((lun->flags & CTL_LUN_PR_RESERVED)
5373	 && ((cdb->how & SSS_START)==0)) {
5374		uint32_t residx;
5375
5376		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5377		if (lun->pr_keys[residx] == 0
5378		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5379
5380			ctl_set_reservation_conflict(ctsio);
5381			ctl_done((union ctl_io *)ctsio);
5382			return (CTL_RETVAL_COMPLETE);
5383		}
5384	}
5385
5386	/*
5387	 * If there is no backend on this device, we can't start or stop
5388	 * it.  In theory we shouldn't get any start/stop commands in the
5389	 * first place at this level if the LUN doesn't have a backend.
5390	 * That should get stopped by the command decode code.
5391	 */
5392	if (lun->backend == NULL) {
5393		ctl_set_invalid_opcode(ctsio);
5394		ctl_done((union ctl_io *)ctsio);
5395		return (CTL_RETVAL_COMPLETE);
5396	}
5397
5398	/*
5399	 * XXX KDM Copan-specific offline behavior.
5400	 * Figure out a reasonable way to port this?
5401	 */
5402#ifdef NEEDTOPORT
5403	mtx_lock(&lun->lun_lock);
5404
5405	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5406	 && (lun->flags & CTL_LUN_OFFLINE)) {
5407		/*
5408		 * If the LUN is offline, and the on/offline bit isn't set,
5409		 * reject the start or stop.  Otherwise, let it through.
5410		 */
5411		mtx_unlock(&lun->lun_lock);
5412		ctl_set_lun_not_ready(ctsio);
5413		ctl_done((union ctl_io *)ctsio);
5414	} else {
5415		mtx_unlock(&lun->lun_lock);
5416#endif /* NEEDTOPORT */
5417		/*
5418		 * This could be a start or a stop when we're online,
5419		 * or a stop/offline or start/online.  A start or stop when
5420		 * we're offline is covered in the case above.
5421		 */
5422		/*
5423		 * In the non-immediate case, we send the request to
5424		 * the backend and return status to the user when
5425		 * it is done.
5426		 *
5427		 * In the immediate case, we allocate a new ctl_io
5428		 * to hold a copy of the request, and send that to
5429		 * the backend.  We then set good status on the
5430		 * user's request and return it immediately.
5431		 */
5432		if (cdb->byte2 & SSS_IMMED) {
5433			union ctl_io *new_io;
5434
5435			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5436			if (new_io == NULL) {
5437				ctl_set_busy(ctsio);
5438				ctl_done((union ctl_io *)ctsio);
5439			} else {
5440				ctl_copy_io((union ctl_io *)ctsio,
5441					    new_io);
5442				retval = lun->backend->config_write(new_io);
5443				ctl_set_success(ctsio);
5444				ctl_done((union ctl_io *)ctsio);
5445			}
5446		} else {
5447			retval = lun->backend->config_write(
5448				(union ctl_io *)ctsio);
5449		}
5450#ifdef NEEDTOPORT
5451	}
5452#endif
5453	return (retval);
5454}
5455
5456/*
5457 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5458 * we don't really do anything with the LBA and length fields if the user
5459 * passes them in.  Instead we'll just flush out the cache for the entire
5460 * LUN.
5461 */
5462int
5463ctl_sync_cache(struct ctl_scsiio *ctsio)
5464{
5465	struct ctl_lun *lun;
5466	struct ctl_softc *ctl_softc;
5467	uint64_t starting_lba;
5468	uint32_t block_count;
5469	int retval;
5470
5471	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5472
5473	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5474	ctl_softc = control_softc;
5475	retval = 0;
5476
5477	switch (ctsio->cdb[0]) {
5478	case SYNCHRONIZE_CACHE: {
5479		struct scsi_sync_cache *cdb;
5480		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5481
5482		starting_lba = scsi_4btoul(cdb->begin_lba);
5483		block_count = scsi_2btoul(cdb->lb_count);
5484		break;
5485	}
5486	case SYNCHRONIZE_CACHE_16: {
5487		struct scsi_sync_cache_16 *cdb;
5488		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5489
5490		starting_lba = scsi_8btou64(cdb->begin_lba);
5491		block_count = scsi_4btoul(cdb->lb_count);
5492		break;
5493	}
5494	default:
5495		ctl_set_invalid_opcode(ctsio);
5496		ctl_done((union ctl_io *)ctsio);
5497		goto bailout;
5498		break; /* NOTREACHED */
5499	}
5500
5501	/*
5502	 * We check the LBA and length, but don't do anything with them.
5503	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5504	 * get flushed.  This check will just help satisfy anyone who wants
5505	 * to see an error for an out of range LBA.
5506	 */
5507	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5508		ctl_set_lba_out_of_range(ctsio);
5509		ctl_done((union ctl_io *)ctsio);
5510		goto bailout;
5511	}
5512
5513	/*
5514	 * If this LUN has no backend, we can't flush the cache anyway.
5515	 */
5516	if (lun->backend == NULL) {
5517		ctl_set_invalid_opcode(ctsio);
5518		ctl_done((union ctl_io *)ctsio);
5519		goto bailout;
5520	}
5521
5522	/*
5523	 * Check to see whether we're configured to send the SYNCHRONIZE
5524	 * CACHE command directly to the back end.
5525	 */
5526	mtx_lock(&lun->lun_lock);
5527	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5528	 && (++(lun->sync_count) >= lun->sync_interval)) {
5529		lun->sync_count = 0;
5530		mtx_unlock(&lun->lun_lock);
5531		retval = lun->backend->config_write((union ctl_io *)ctsio);
5532	} else {
5533		mtx_unlock(&lun->lun_lock);
5534		ctl_set_success(ctsio);
5535		ctl_done((union ctl_io *)ctsio);
5536	}
5537
5538bailout:
5539
5540	return (retval);
5541}
5542
5543int
5544ctl_format(struct ctl_scsiio *ctsio)
5545{
5546	struct scsi_format *cdb;
5547	struct ctl_lun *lun;
5548	struct ctl_softc *ctl_softc;
5549	int length, defect_list_len;
5550
5551	CTL_DEBUG_PRINT(("ctl_format\n"));
5552
5553	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5554	ctl_softc = control_softc;
5555
5556	cdb = (struct scsi_format *)ctsio->cdb;
5557
5558	length = 0;
5559	if (cdb->byte2 & SF_FMTDATA) {
5560		if (cdb->byte2 & SF_LONGLIST)
5561			length = sizeof(struct scsi_format_header_long);
5562		else
5563			length = sizeof(struct scsi_format_header_short);
5564	}
5565
5566	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5567	 && (length > 0)) {
5568		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5569		ctsio->kern_data_len = length;
5570		ctsio->kern_total_len = length;
5571		ctsio->kern_data_resid = 0;
5572		ctsio->kern_rel_offset = 0;
5573		ctsio->kern_sg_entries = 0;
5574		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5575		ctsio->be_move_done = ctl_config_move_done;
5576		ctl_datamove((union ctl_io *)ctsio);
5577
5578		return (CTL_RETVAL_COMPLETE);
5579	}
5580
5581	defect_list_len = 0;
5582
5583	if (cdb->byte2 & SF_FMTDATA) {
5584		if (cdb->byte2 & SF_LONGLIST) {
5585			struct scsi_format_header_long *header;
5586
5587			header = (struct scsi_format_header_long *)
5588				ctsio->kern_data_ptr;
5589
5590			defect_list_len = scsi_4btoul(header->defect_list_len);
5591			if (defect_list_len != 0) {
5592				ctl_set_invalid_field(ctsio,
5593						      /*sks_valid*/ 1,
5594						      /*command*/ 0,
5595						      /*field*/ 2,
5596						      /*bit_valid*/ 0,
5597						      /*bit*/ 0);
5598				goto bailout;
5599			}
5600		} else {
5601			struct scsi_format_header_short *header;
5602
5603			header = (struct scsi_format_header_short *)
5604				ctsio->kern_data_ptr;
5605
5606			defect_list_len = scsi_2btoul(header->defect_list_len);
5607			if (defect_list_len != 0) {
5608				ctl_set_invalid_field(ctsio,
5609						      /*sks_valid*/ 1,
5610						      /*command*/ 0,
5611						      /*field*/ 2,
5612						      /*bit_valid*/ 0,
5613						      /*bit*/ 0);
5614				goto bailout;
5615			}
5616		}
5617	}
5618
5619	/*
5620	 * The format command will clear out the "Medium format corrupted"
5621	 * status if set by the configuration code.  That status is really
5622	 * just a way to notify the host that we have lost the media, and
5623	 * get them to issue a command that will basically make them think
5624	 * they're blowing away the media.
5625	 */
5626	mtx_lock(&lun->lun_lock);
5627	lun->flags &= ~CTL_LUN_INOPERABLE;
5628	mtx_unlock(&lun->lun_lock);
5629
5630	ctsio->scsi_status = SCSI_STATUS_OK;
5631	ctsio->io_hdr.status = CTL_SUCCESS;
5632bailout:
5633
5634	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5635		free(ctsio->kern_data_ptr, M_CTL);
5636		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5637	}
5638
5639	ctl_done((union ctl_io *)ctsio);
5640	return (CTL_RETVAL_COMPLETE);
5641}
5642
5643int
5644ctl_read_buffer(struct ctl_scsiio *ctsio)
5645{
5646	struct scsi_read_buffer *cdb;
5647	struct ctl_lun *lun;
5648	int buffer_offset, len;
5649	static uint8_t descr[4];
5650	static uint8_t echo_descr[4] = { 0 };
5651
5652	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5653
5654	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5655	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5656
5657	if (lun->flags & CTL_LUN_PR_RESERVED) {
5658		uint32_t residx;
5659
5660		/*
5661		 * XXX KDM need a lock here.
5662		 */
5663		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5664		if ((lun->res_type == SPR_TYPE_EX_AC
5665		  && residx != lun->pr_res_idx)
5666		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5667		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5668		  && lun->pr_keys[residx] == 0)) {
5669			ctl_set_reservation_conflict(ctsio);
5670			ctl_done((union ctl_io *)ctsio);
5671			return (CTL_RETVAL_COMPLETE);
5672	        }
5673	}
5674
5675	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5676	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5677	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5678		ctl_set_invalid_field(ctsio,
5679				      /*sks_valid*/ 1,
5680				      /*command*/ 1,
5681				      /*field*/ 1,
5682				      /*bit_valid*/ 1,
5683				      /*bit*/ 4);
5684		ctl_done((union ctl_io *)ctsio);
5685		return (CTL_RETVAL_COMPLETE);
5686	}
5687
5688	len = scsi_3btoul(cdb->length);
5689	buffer_offset = scsi_3btoul(cdb->offset);
5690
5691	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5692		ctl_set_invalid_field(ctsio,
5693				      /*sks_valid*/ 1,
5694				      /*command*/ 1,
5695				      /*field*/ 6,
5696				      /*bit_valid*/ 0,
5697				      /*bit*/ 0);
5698		ctl_done((union ctl_io *)ctsio);
5699		return (CTL_RETVAL_COMPLETE);
5700	}
5701
5702	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5703		descr[0] = 0;
5704		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5705		ctsio->kern_data_ptr = descr;
5706		len = min(len, sizeof(descr));
5707	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5708		ctsio->kern_data_ptr = echo_descr;
5709		len = min(len, sizeof(echo_descr));
5710	} else
5711		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5712	ctsio->kern_data_len = len;
5713	ctsio->kern_total_len = len;
5714	ctsio->kern_data_resid = 0;
5715	ctsio->kern_rel_offset = 0;
5716	ctsio->kern_sg_entries = 0;
5717	ctsio->be_move_done = ctl_config_move_done;
5718	ctl_datamove((union ctl_io *)ctsio);
5719
5720	return (CTL_RETVAL_COMPLETE);
5721}
5722
5723int
5724ctl_write_buffer(struct ctl_scsiio *ctsio)
5725{
5726	struct scsi_write_buffer *cdb;
5727	struct ctl_lun *lun;
5728	int buffer_offset, len;
5729
5730	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5731
5732	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5733	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5734
5735	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5736		ctl_set_invalid_field(ctsio,
5737				      /*sks_valid*/ 1,
5738				      /*command*/ 1,
5739				      /*field*/ 1,
5740				      /*bit_valid*/ 1,
5741				      /*bit*/ 4);
5742		ctl_done((union ctl_io *)ctsio);
5743		return (CTL_RETVAL_COMPLETE);
5744	}
5745
5746	len = scsi_3btoul(cdb->length);
5747	buffer_offset = scsi_3btoul(cdb->offset);
5748
5749	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5750		ctl_set_invalid_field(ctsio,
5751				      /*sks_valid*/ 1,
5752				      /*command*/ 1,
5753				      /*field*/ 6,
5754				      /*bit_valid*/ 0,
5755				      /*bit*/ 0);
5756		ctl_done((union ctl_io *)ctsio);
5757		return (CTL_RETVAL_COMPLETE);
5758	}
5759
5760	/*
5761	 * If we've got a kernel request that hasn't been malloced yet,
5762	 * malloc it and tell the caller the data buffer is here.
5763	 */
5764	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5765		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5766		ctsio->kern_data_len = len;
5767		ctsio->kern_total_len = len;
5768		ctsio->kern_data_resid = 0;
5769		ctsio->kern_rel_offset = 0;
5770		ctsio->kern_sg_entries = 0;
5771		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5772		ctsio->be_move_done = ctl_config_move_done;
5773		ctl_datamove((union ctl_io *)ctsio);
5774
5775		return (CTL_RETVAL_COMPLETE);
5776	}
5777
5778	ctl_done((union ctl_io *)ctsio);
5779
5780	return (CTL_RETVAL_COMPLETE);
5781}
5782
5783int
5784ctl_write_same(struct ctl_scsiio *ctsio)
5785{
5786	struct ctl_lun *lun;
5787	struct ctl_lba_len_flags *lbalen;
5788	uint64_t lba;
5789	uint32_t num_blocks;
5790	int len, retval;
5791	uint8_t byte2;
5792
5793	retval = CTL_RETVAL_COMPLETE;
5794
5795	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5796
5797	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5798
5799	switch (ctsio->cdb[0]) {
5800	case WRITE_SAME_10: {
5801		struct scsi_write_same_10 *cdb;
5802
5803		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5804
5805		lba = scsi_4btoul(cdb->addr);
5806		num_blocks = scsi_2btoul(cdb->length);
5807		byte2 = cdb->byte2;
5808		break;
5809	}
5810	case WRITE_SAME_16: {
5811		struct scsi_write_same_16 *cdb;
5812
5813		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5814
5815		lba = scsi_8btou64(cdb->addr);
5816		num_blocks = scsi_4btoul(cdb->length);
5817		byte2 = cdb->byte2;
5818		break;
5819	}
5820	default:
5821		/*
5822		 * We got a command we don't support.  This shouldn't
5823		 * happen, commands should be filtered out above us.
5824		 */
5825		ctl_set_invalid_opcode(ctsio);
5826		ctl_done((union ctl_io *)ctsio);
5827
5828		return (CTL_RETVAL_COMPLETE);
5829		break; /* NOTREACHED */
5830	}
5831
5832	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5833	if ((byte2 & SWS_UNMAP) == 0 &&
5834	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5835		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5836		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5837		ctl_done((union ctl_io *)ctsio);
5838		return (CTL_RETVAL_COMPLETE);
5839	}
5840
5841	/*
5842	 * The first check is to make sure we're in bounds, the second
5843	 * check is to catch wrap-around problems.  If the lba + num blocks
5844	 * is less than the lba, then we've wrapped around and the block
5845	 * range is invalid anyway.
5846	 */
5847	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5848	 || ((lba + num_blocks) < lba)) {
5849		ctl_set_lba_out_of_range(ctsio);
5850		ctl_done((union ctl_io *)ctsio);
5851		return (CTL_RETVAL_COMPLETE);
5852	}
5853
5854	/* Zero number of blocks means "to the last logical block" */
5855	if (num_blocks == 0) {
5856		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5857			ctl_set_invalid_field(ctsio,
5858					      /*sks_valid*/ 0,
5859					      /*command*/ 1,
5860					      /*field*/ 0,
5861					      /*bit_valid*/ 0,
5862					      /*bit*/ 0);
5863			ctl_done((union ctl_io *)ctsio);
5864			return (CTL_RETVAL_COMPLETE);
5865		}
5866		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5867	}
5868
5869	len = lun->be_lun->blocksize;
5870
5871	/*
5872	 * If we've got a kernel request that hasn't been malloced yet,
5873	 * malloc it and tell the caller the data buffer is here.
5874	 */
5875	if ((byte2 & SWS_NDOB) == 0 &&
5876	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5877		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5878		ctsio->kern_data_len = len;
5879		ctsio->kern_total_len = len;
5880		ctsio->kern_data_resid = 0;
5881		ctsio->kern_rel_offset = 0;
5882		ctsio->kern_sg_entries = 0;
5883		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5884		ctsio->be_move_done = ctl_config_move_done;
5885		ctl_datamove((union ctl_io *)ctsio);
5886
5887		return (CTL_RETVAL_COMPLETE);
5888	}
5889
5890	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5891	lbalen->lba = lba;
5892	lbalen->len = num_blocks;
5893	lbalen->flags = byte2;
5894	retval = lun->backend->config_write((union ctl_io *)ctsio);
5895
5896	return (retval);
5897}
5898
5899int
5900ctl_unmap(struct ctl_scsiio *ctsio)
5901{
5902	struct ctl_lun *lun;
5903	struct scsi_unmap *cdb;
5904	struct ctl_ptr_len_flags *ptrlen;
5905	struct scsi_unmap_header *hdr;
5906	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5907	uint64_t lba;
5908	uint32_t num_blocks;
5909	int len, retval;
5910	uint8_t byte2;
5911
5912	retval = CTL_RETVAL_COMPLETE;
5913
5914	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5915
5916	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5917	cdb = (struct scsi_unmap *)ctsio->cdb;
5918
5919	len = scsi_2btoul(cdb->length);
5920	byte2 = cdb->byte2;
5921
5922	/*
5923	 * If we've got a kernel request that hasn't been malloced yet,
5924	 * malloc it and tell the caller the data buffer is here.
5925	 */
5926	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5927		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5928		ctsio->kern_data_len = len;
5929		ctsio->kern_total_len = len;
5930		ctsio->kern_data_resid = 0;
5931		ctsio->kern_rel_offset = 0;
5932		ctsio->kern_sg_entries = 0;
5933		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5934		ctsio->be_move_done = ctl_config_move_done;
5935		ctl_datamove((union ctl_io *)ctsio);
5936
5937		return (CTL_RETVAL_COMPLETE);
5938	}
5939
5940	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5941	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5942	if (len < sizeof (*hdr) ||
5943	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5944	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5945	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5946		ctl_set_invalid_field(ctsio,
5947				      /*sks_valid*/ 0,
5948				      /*command*/ 0,
5949				      /*field*/ 0,
5950				      /*bit_valid*/ 0,
5951				      /*bit*/ 0);
5952		ctl_done((union ctl_io *)ctsio);
5953		return (CTL_RETVAL_COMPLETE);
5954	}
5955	len = scsi_2btoul(hdr->desc_length);
5956	buf = (struct scsi_unmap_desc *)(hdr + 1);
5957	end = buf + len / sizeof(*buf);
5958
5959	endnz = buf;
5960	for (range = buf; range < end; range++) {
5961		lba = scsi_8btou64(range->lba);
5962		num_blocks = scsi_4btoul(range->length);
5963		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5964		 || ((lba + num_blocks) < lba)) {
5965			ctl_set_lba_out_of_range(ctsio);
5966			ctl_done((union ctl_io *)ctsio);
5967			return (CTL_RETVAL_COMPLETE);
5968		}
5969		if (num_blocks != 0)
5970			endnz = range + 1;
5971	}
5972
5973	/*
5974	 * Block backend can not handle zero last range.
5975	 * Filter it out and return if there is nothing left.
5976	 */
5977	len = (uint8_t *)endnz - (uint8_t *)buf;
5978	if (len == 0) {
5979		ctl_set_success(ctsio);
5980		ctl_done((union ctl_io *)ctsio);
5981		return (CTL_RETVAL_COMPLETE);
5982	}
5983
5984	mtx_lock(&lun->lun_lock);
5985	ptrlen = (struct ctl_ptr_len_flags *)
5986	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5987	ptrlen->ptr = (void *)buf;
5988	ptrlen->len = len;
5989	ptrlen->flags = byte2;
5990	ctl_check_blocked(lun);
5991	mtx_unlock(&lun->lun_lock);
5992
5993	retval = lun->backend->config_write((union ctl_io *)ctsio);
5994	return (retval);
5995}
5996
5997/*
5998 * Note that this function currently doesn't actually do anything inside
5999 * CTL to enforce things if the DQue bit is turned on.
6000 *
6001 * Also note that this function can't be used in the default case, because
6002 * the DQue bit isn't set in the changeable mask for the control mode page
6003 * anyway.  This is just here as an example for how to implement a page
6004 * handler, and a placeholder in case we want to allow the user to turn
6005 * tagged queueing on and off.
6006 *
6007 * The D_SENSE bit handling is functional, however, and will turn
6008 * descriptor sense on and off for a given LUN.
6009 */
6010int
6011ctl_control_page_handler(struct ctl_scsiio *ctsio,
6012			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6013{
6014	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6015	struct ctl_lun *lun;
6016	struct ctl_softc *softc;
6017	int set_ua;
6018	uint32_t initidx;
6019
6020	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6021	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6022	set_ua = 0;
6023
6024	user_cp = (struct scsi_control_page *)page_ptr;
6025	current_cp = (struct scsi_control_page *)
6026		(page_index->page_data + (page_index->page_len *
6027		CTL_PAGE_CURRENT));
6028	saved_cp = (struct scsi_control_page *)
6029		(page_index->page_data + (page_index->page_len *
6030		CTL_PAGE_SAVED));
6031
6032	softc = control_softc;
6033
6034	mtx_lock(&lun->lun_lock);
6035	if (((current_cp->rlec & SCP_DSENSE) == 0)
6036	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6037		/*
6038		 * Descriptor sense is currently turned off and the user
6039		 * wants to turn it on.
6040		 */
6041		current_cp->rlec |= SCP_DSENSE;
6042		saved_cp->rlec |= SCP_DSENSE;
6043		lun->flags |= CTL_LUN_SENSE_DESC;
6044		set_ua = 1;
6045	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6046		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6047		/*
6048		 * Descriptor sense is currently turned on, and the user
6049		 * wants to turn it off.
6050		 */
6051		current_cp->rlec &= ~SCP_DSENSE;
6052		saved_cp->rlec &= ~SCP_DSENSE;
6053		lun->flags &= ~CTL_LUN_SENSE_DESC;
6054		set_ua = 1;
6055	}
6056	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6057	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6058		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6059		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6060		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6061		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6062		set_ua = 1;
6063	}
6064	if ((current_cp->eca_and_aen & SCP_SWP) !=
6065	    (user_cp->eca_and_aen & SCP_SWP)) {
6066		current_cp->eca_and_aen &= ~SCP_SWP;
6067		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6068		saved_cp->eca_and_aen &= ~SCP_SWP;
6069		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6070		set_ua = 1;
6071	}
6072	if (set_ua != 0) {
6073		int i;
6074		/*
6075		 * Let other initiators know that the mode
6076		 * parameters for this LUN have changed.
6077		 */
6078		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6079			if (i == initidx)
6080				continue;
6081
6082			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6083		}
6084	}
6085	mtx_unlock(&lun->lun_lock);
6086
6087	return (0);
6088}
6089
6090int
6091ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6092		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6093{
6094	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6095	struct ctl_lun *lun;
6096	int set_ua;
6097	uint32_t initidx;
6098
6099	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6100	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6101	set_ua = 0;
6102
6103	user_cp = (struct scsi_caching_page *)page_ptr;
6104	current_cp = (struct scsi_caching_page *)
6105		(page_index->page_data + (page_index->page_len *
6106		CTL_PAGE_CURRENT));
6107	saved_cp = (struct scsi_caching_page *)
6108		(page_index->page_data + (page_index->page_len *
6109		CTL_PAGE_SAVED));
6110
6111	mtx_lock(&lun->lun_lock);
6112	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6113	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6114		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6115		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6116		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6117		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6118		set_ua = 1;
6119	}
6120	if (set_ua != 0) {
6121		int i;
6122		/*
6123		 * Let other initiators know that the mode
6124		 * parameters for this LUN have changed.
6125		 */
6126		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6127			if (i == initidx)
6128				continue;
6129
6130			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6131		}
6132	}
6133	mtx_unlock(&lun->lun_lock);
6134
6135	return (0);
6136}
6137
6138int
6139ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6140				struct ctl_page_index *page_index,
6141				uint8_t *page_ptr)
6142{
6143	uint8_t *c;
6144	int i;
6145
6146	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6147	ctl_time_io_secs =
6148		(c[0] << 8) |
6149		(c[1] << 0) |
6150		0;
6151	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6152	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6153	printf("page data:");
6154	for (i=0; i<8; i++)
6155		printf(" %.2x",page_ptr[i]);
6156	printf("\n");
6157	return (0);
6158}
6159
6160int
6161ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6162			       struct ctl_page_index *page_index,
6163			       int pc)
6164{
6165	struct copan_debugconf_subpage *page;
6166
6167	page = (struct copan_debugconf_subpage *)page_index->page_data +
6168		(page_index->page_len * pc);
6169
6170	switch (pc) {
6171	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6172	case SMS_PAGE_CTRL_DEFAULT >> 6:
6173	case SMS_PAGE_CTRL_SAVED >> 6:
6174		/*
6175		 * We don't update the changable or default bits for this page.
6176		 */
6177		break;
6178	case SMS_PAGE_CTRL_CURRENT >> 6:
6179		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6180		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6181		break;
6182	default:
6183#ifdef NEEDTOPORT
6184		EPRINT(0, "Invalid PC %d!!", pc);
6185#endif /* NEEDTOPORT */
6186		break;
6187	}
6188	return (0);
6189}
6190
6191
6192static int
6193ctl_do_mode_select(union ctl_io *io)
6194{
6195	struct scsi_mode_page_header *page_header;
6196	struct ctl_page_index *page_index;
6197	struct ctl_scsiio *ctsio;
6198	int control_dev, page_len;
6199	int page_len_offset, page_len_size;
6200	union ctl_modepage_info *modepage_info;
6201	struct ctl_lun *lun;
6202	int *len_left, *len_used;
6203	int retval, i;
6204
6205	ctsio = &io->scsiio;
6206	page_index = NULL;
6207	page_len = 0;
6208	retval = CTL_RETVAL_COMPLETE;
6209
6210	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6211
6212	if (lun->be_lun->lun_type != T_DIRECT)
6213		control_dev = 1;
6214	else
6215		control_dev = 0;
6216
6217	modepage_info = (union ctl_modepage_info *)
6218		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6219	len_left = &modepage_info->header.len_left;
6220	len_used = &modepage_info->header.len_used;
6221
6222do_next_page:
6223
6224	page_header = (struct scsi_mode_page_header *)
6225		(ctsio->kern_data_ptr + *len_used);
6226
6227	if (*len_left == 0) {
6228		free(ctsio->kern_data_ptr, M_CTL);
6229		ctl_set_success(ctsio);
6230		ctl_done((union ctl_io *)ctsio);
6231		return (CTL_RETVAL_COMPLETE);
6232	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6233
6234		free(ctsio->kern_data_ptr, M_CTL);
6235		ctl_set_param_len_error(ctsio);
6236		ctl_done((union ctl_io *)ctsio);
6237		return (CTL_RETVAL_COMPLETE);
6238
6239	} else if ((page_header->page_code & SMPH_SPF)
6240		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6241
6242		free(ctsio->kern_data_ptr, M_CTL);
6243		ctl_set_param_len_error(ctsio);
6244		ctl_done((union ctl_io *)ctsio);
6245		return (CTL_RETVAL_COMPLETE);
6246	}
6247
6248
6249	/*
6250	 * XXX KDM should we do something with the block descriptor?
6251	 */
6252	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6253
6254		if ((control_dev != 0)
6255		 && (lun->mode_pages.index[i].page_flags &
6256		     CTL_PAGE_FLAG_DISK_ONLY))
6257			continue;
6258
6259		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6260		    (page_header->page_code & SMPH_PC_MASK))
6261			continue;
6262
6263		/*
6264		 * If neither page has a subpage code, then we've got a
6265		 * match.
6266		 */
6267		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6268		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6269			page_index = &lun->mode_pages.index[i];
6270			page_len = page_header->page_length;
6271			break;
6272		}
6273
6274		/*
6275		 * If both pages have subpages, then the subpage numbers
6276		 * have to match.
6277		 */
6278		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6279		  && (page_header->page_code & SMPH_SPF)) {
6280			struct scsi_mode_page_header_sp *sph;
6281
6282			sph = (struct scsi_mode_page_header_sp *)page_header;
6283
6284			if (lun->mode_pages.index[i].subpage ==
6285			    sph->subpage) {
6286				page_index = &lun->mode_pages.index[i];
6287				page_len = scsi_2btoul(sph->page_length);
6288				break;
6289			}
6290		}
6291	}
6292
6293	/*
6294	 * If we couldn't find the page, or if we don't have a mode select
6295	 * handler for it, send back an error to the user.
6296	 */
6297	if ((page_index == NULL)
6298	 || (page_index->select_handler == NULL)) {
6299		ctl_set_invalid_field(ctsio,
6300				      /*sks_valid*/ 1,
6301				      /*command*/ 0,
6302				      /*field*/ *len_used,
6303				      /*bit_valid*/ 0,
6304				      /*bit*/ 0);
6305		free(ctsio->kern_data_ptr, M_CTL);
6306		ctl_done((union ctl_io *)ctsio);
6307		return (CTL_RETVAL_COMPLETE);
6308	}
6309
6310	if (page_index->page_code & SMPH_SPF) {
6311		page_len_offset = 2;
6312		page_len_size = 2;
6313	} else {
6314		page_len_size = 1;
6315		page_len_offset = 1;
6316	}
6317
6318	/*
6319	 * If the length the initiator gives us isn't the one we specify in
6320	 * the mode page header, or if they didn't specify enough data in
6321	 * the CDB to avoid truncating this page, kick out the request.
6322	 */
6323	if ((page_len != (page_index->page_len - page_len_offset -
6324			  page_len_size))
6325	 || (*len_left < page_index->page_len)) {
6326
6327
6328		ctl_set_invalid_field(ctsio,
6329				      /*sks_valid*/ 1,
6330				      /*command*/ 0,
6331				      /*field*/ *len_used + page_len_offset,
6332				      /*bit_valid*/ 0,
6333				      /*bit*/ 0);
6334		free(ctsio->kern_data_ptr, M_CTL);
6335		ctl_done((union ctl_io *)ctsio);
6336		return (CTL_RETVAL_COMPLETE);
6337	}
6338
6339	/*
6340	 * Run through the mode page, checking to make sure that the bits
6341	 * the user changed are actually legal for him to change.
6342	 */
6343	for (i = 0; i < page_index->page_len; i++) {
6344		uint8_t *user_byte, *change_mask, *current_byte;
6345		int bad_bit;
6346		int j;
6347
6348		user_byte = (uint8_t *)page_header + i;
6349		change_mask = page_index->page_data +
6350			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6351		current_byte = page_index->page_data +
6352			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6353
6354		/*
6355		 * Check to see whether the user set any bits in this byte
6356		 * that he is not allowed to set.
6357		 */
6358		if ((*user_byte & ~(*change_mask)) ==
6359		    (*current_byte & ~(*change_mask)))
6360			continue;
6361
6362		/*
6363		 * Go through bit by bit to determine which one is illegal.
6364		 */
6365		bad_bit = 0;
6366		for (j = 7; j >= 0; j--) {
6367			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6368			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6369				bad_bit = i;
6370				break;
6371			}
6372		}
6373		ctl_set_invalid_field(ctsio,
6374				      /*sks_valid*/ 1,
6375				      /*command*/ 0,
6376				      /*field*/ *len_used + i,
6377				      /*bit_valid*/ 1,
6378				      /*bit*/ bad_bit);
6379		free(ctsio->kern_data_ptr, M_CTL);
6380		ctl_done((union ctl_io *)ctsio);
6381		return (CTL_RETVAL_COMPLETE);
6382	}
6383
6384	/*
6385	 * Decrement these before we call the page handler, since we may
6386	 * end up getting called back one way or another before the handler
6387	 * returns to this context.
6388	 */
6389	*len_left -= page_index->page_len;
6390	*len_used += page_index->page_len;
6391
6392	retval = page_index->select_handler(ctsio, page_index,
6393					    (uint8_t *)page_header);
6394
6395	/*
6396	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6397	 * wait until this queued command completes to finish processing
6398	 * the mode page.  If it returns anything other than
6399	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6400	 * already set the sense information, freed the data pointer, and
6401	 * completed the io for us.
6402	 */
6403	if (retval != CTL_RETVAL_COMPLETE)
6404		goto bailout_no_done;
6405
6406	/*
6407	 * If the initiator sent us more than one page, parse the next one.
6408	 */
6409	if (*len_left > 0)
6410		goto do_next_page;
6411
6412	ctl_set_success(ctsio);
6413	free(ctsio->kern_data_ptr, M_CTL);
6414	ctl_done((union ctl_io *)ctsio);
6415
6416bailout_no_done:
6417
6418	return (CTL_RETVAL_COMPLETE);
6419
6420}
6421
6422int
6423ctl_mode_select(struct ctl_scsiio *ctsio)
6424{
6425	int param_len, pf, sp;
6426	int header_size, bd_len;
6427	int len_left, len_used;
6428	struct ctl_page_index *page_index;
6429	struct ctl_lun *lun;
6430	int control_dev, page_len;
6431	union ctl_modepage_info *modepage_info;
6432	int retval;
6433
6434	pf = 0;
6435	sp = 0;
6436	page_len = 0;
6437	len_used = 0;
6438	len_left = 0;
6439	retval = 0;
6440	bd_len = 0;
6441	page_index = NULL;
6442
6443	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6444
6445	if (lun->be_lun->lun_type != T_DIRECT)
6446		control_dev = 1;
6447	else
6448		control_dev = 0;
6449
6450	switch (ctsio->cdb[0]) {
6451	case MODE_SELECT_6: {
6452		struct scsi_mode_select_6 *cdb;
6453
6454		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6455
6456		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6457		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6458
6459		param_len = cdb->length;
6460		header_size = sizeof(struct scsi_mode_header_6);
6461		break;
6462	}
6463	case MODE_SELECT_10: {
6464		struct scsi_mode_select_10 *cdb;
6465
6466		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6467
6468		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6469		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6470
6471		param_len = scsi_2btoul(cdb->length);
6472		header_size = sizeof(struct scsi_mode_header_10);
6473		break;
6474	}
6475	default:
6476		ctl_set_invalid_opcode(ctsio);
6477		ctl_done((union ctl_io *)ctsio);
6478		return (CTL_RETVAL_COMPLETE);
6479		break; /* NOTREACHED */
6480	}
6481
6482	/*
6483	 * From SPC-3:
6484	 * "A parameter list length of zero indicates that the Data-Out Buffer
6485	 * shall be empty. This condition shall not be considered as an error."
6486	 */
6487	if (param_len == 0) {
6488		ctl_set_success(ctsio);
6489		ctl_done((union ctl_io *)ctsio);
6490		return (CTL_RETVAL_COMPLETE);
6491	}
6492
6493	/*
6494	 * Since we'll hit this the first time through, prior to
6495	 * allocation, we don't need to free a data buffer here.
6496	 */
6497	if (param_len < header_size) {
6498		ctl_set_param_len_error(ctsio);
6499		ctl_done((union ctl_io *)ctsio);
6500		return (CTL_RETVAL_COMPLETE);
6501	}
6502
6503	/*
6504	 * Allocate the data buffer and grab the user's data.  In theory,
6505	 * we shouldn't have to sanity check the parameter list length here
6506	 * because the maximum size is 64K.  We should be able to malloc
6507	 * that much without too many problems.
6508	 */
6509	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6510		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6511		ctsio->kern_data_len = param_len;
6512		ctsio->kern_total_len = param_len;
6513		ctsio->kern_data_resid = 0;
6514		ctsio->kern_rel_offset = 0;
6515		ctsio->kern_sg_entries = 0;
6516		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6517		ctsio->be_move_done = ctl_config_move_done;
6518		ctl_datamove((union ctl_io *)ctsio);
6519
6520		return (CTL_RETVAL_COMPLETE);
6521	}
6522
6523	switch (ctsio->cdb[0]) {
6524	case MODE_SELECT_6: {
6525		struct scsi_mode_header_6 *mh6;
6526
6527		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6528		bd_len = mh6->blk_desc_len;
6529		break;
6530	}
6531	case MODE_SELECT_10: {
6532		struct scsi_mode_header_10 *mh10;
6533
6534		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6535		bd_len = scsi_2btoul(mh10->blk_desc_len);
6536		break;
6537	}
6538	default:
6539		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6540		break;
6541	}
6542
6543	if (param_len < (header_size + bd_len)) {
6544		free(ctsio->kern_data_ptr, M_CTL);
6545		ctl_set_param_len_error(ctsio);
6546		ctl_done((union ctl_io *)ctsio);
6547		return (CTL_RETVAL_COMPLETE);
6548	}
6549
6550	/*
6551	 * Set the IO_CONT flag, so that if this I/O gets passed to
6552	 * ctl_config_write_done(), it'll get passed back to
6553	 * ctl_do_mode_select() for further processing, or completion if
6554	 * we're all done.
6555	 */
6556	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6557	ctsio->io_cont = ctl_do_mode_select;
6558
6559	modepage_info = (union ctl_modepage_info *)
6560		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6561
6562	memset(modepage_info, 0, sizeof(*modepage_info));
6563
6564	len_left = param_len - header_size - bd_len;
6565	len_used = header_size + bd_len;
6566
6567	modepage_info->header.len_left = len_left;
6568	modepage_info->header.len_used = len_used;
6569
6570	return (ctl_do_mode_select((union ctl_io *)ctsio));
6571}
6572
6573int
6574ctl_mode_sense(struct ctl_scsiio *ctsio)
6575{
6576	struct ctl_lun *lun;
6577	int pc, page_code, dbd, llba, subpage;
6578	int alloc_len, page_len, header_len, total_len;
6579	struct scsi_mode_block_descr *block_desc;
6580	struct ctl_page_index *page_index;
6581	int control_dev;
6582
6583	dbd = 0;
6584	llba = 0;
6585	block_desc = NULL;
6586	page_index = NULL;
6587
6588	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6589
6590	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6591
6592	if (lun->be_lun->lun_type != T_DIRECT)
6593		control_dev = 1;
6594	else
6595		control_dev = 0;
6596
6597	if (lun->flags & CTL_LUN_PR_RESERVED) {
6598		uint32_t residx;
6599
6600		/*
6601		 * XXX KDM need a lock here.
6602		 */
6603		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6604		if ((lun->res_type == SPR_TYPE_EX_AC
6605		  && residx != lun->pr_res_idx)
6606		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6607		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6608		  && lun->pr_keys[residx] == 0)) {
6609			ctl_set_reservation_conflict(ctsio);
6610			ctl_done((union ctl_io *)ctsio);
6611			return (CTL_RETVAL_COMPLETE);
6612		}
6613	}
6614
6615	switch (ctsio->cdb[0]) {
6616	case MODE_SENSE_6: {
6617		struct scsi_mode_sense_6 *cdb;
6618
6619		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6620
6621		header_len = sizeof(struct scsi_mode_hdr_6);
6622		if (cdb->byte2 & SMS_DBD)
6623			dbd = 1;
6624		else
6625			header_len += sizeof(struct scsi_mode_block_descr);
6626
6627		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6628		page_code = cdb->page & SMS_PAGE_CODE;
6629		subpage = cdb->subpage;
6630		alloc_len = cdb->length;
6631		break;
6632	}
6633	case MODE_SENSE_10: {
6634		struct scsi_mode_sense_10 *cdb;
6635
6636		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6637
6638		header_len = sizeof(struct scsi_mode_hdr_10);
6639
6640		if (cdb->byte2 & SMS_DBD)
6641			dbd = 1;
6642		else
6643			header_len += sizeof(struct scsi_mode_block_descr);
6644		if (cdb->byte2 & SMS10_LLBAA)
6645			llba = 1;
6646		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6647		page_code = cdb->page & SMS_PAGE_CODE;
6648		subpage = cdb->subpage;
6649		alloc_len = scsi_2btoul(cdb->length);
6650		break;
6651	}
6652	default:
6653		ctl_set_invalid_opcode(ctsio);
6654		ctl_done((union ctl_io *)ctsio);
6655		return (CTL_RETVAL_COMPLETE);
6656		break; /* NOTREACHED */
6657	}
6658
6659	/*
6660	 * We have to make a first pass through to calculate the size of
6661	 * the pages that match the user's query.  Then we allocate enough
6662	 * memory to hold it, and actually copy the data into the buffer.
6663	 */
6664	switch (page_code) {
6665	case SMS_ALL_PAGES_PAGE: {
6666		int i;
6667
6668		page_len = 0;
6669
6670		/*
6671		 * At the moment, values other than 0 and 0xff here are
6672		 * reserved according to SPC-3.
6673		 */
6674		if ((subpage != SMS_SUBPAGE_PAGE_0)
6675		 && (subpage != SMS_SUBPAGE_ALL)) {
6676			ctl_set_invalid_field(ctsio,
6677					      /*sks_valid*/ 1,
6678					      /*command*/ 1,
6679					      /*field*/ 3,
6680					      /*bit_valid*/ 0,
6681					      /*bit*/ 0);
6682			ctl_done((union ctl_io *)ctsio);
6683			return (CTL_RETVAL_COMPLETE);
6684		}
6685
6686		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6687			if ((control_dev != 0)
6688			 && (lun->mode_pages.index[i].page_flags &
6689			     CTL_PAGE_FLAG_DISK_ONLY))
6690				continue;
6691
6692			/*
6693			 * We don't use this subpage if the user didn't
6694			 * request all subpages.
6695			 */
6696			if ((lun->mode_pages.index[i].subpage != 0)
6697			 && (subpage == SMS_SUBPAGE_PAGE_0))
6698				continue;
6699
6700#if 0
6701			printf("found page %#x len %d\n",
6702			       lun->mode_pages.index[i].page_code &
6703			       SMPH_PC_MASK,
6704			       lun->mode_pages.index[i].page_len);
6705#endif
6706			page_len += lun->mode_pages.index[i].page_len;
6707		}
6708		break;
6709	}
6710	default: {
6711		int i;
6712
6713		page_len = 0;
6714
6715		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6716			/* Look for the right page code */
6717			if ((lun->mode_pages.index[i].page_code &
6718			     SMPH_PC_MASK) != page_code)
6719				continue;
6720
6721			/* Look for the right subpage or the subpage wildcard*/
6722			if ((lun->mode_pages.index[i].subpage != subpage)
6723			 && (subpage != SMS_SUBPAGE_ALL))
6724				continue;
6725
6726			/* Make sure the page is supported for this dev type */
6727			if ((control_dev != 0)
6728			 && (lun->mode_pages.index[i].page_flags &
6729			     CTL_PAGE_FLAG_DISK_ONLY))
6730				continue;
6731
6732#if 0
6733			printf("found page %#x len %d\n",
6734			       lun->mode_pages.index[i].page_code &
6735			       SMPH_PC_MASK,
6736			       lun->mode_pages.index[i].page_len);
6737#endif
6738
6739			page_len += lun->mode_pages.index[i].page_len;
6740		}
6741
6742		if (page_len == 0) {
6743			ctl_set_invalid_field(ctsio,
6744					      /*sks_valid*/ 1,
6745					      /*command*/ 1,
6746					      /*field*/ 2,
6747					      /*bit_valid*/ 1,
6748					      /*bit*/ 5);
6749			ctl_done((union ctl_io *)ctsio);
6750			return (CTL_RETVAL_COMPLETE);
6751		}
6752		break;
6753	}
6754	}
6755
6756	total_len = header_len + page_len;
6757#if 0
6758	printf("header_len = %d, page_len = %d, total_len = %d\n",
6759	       header_len, page_len, total_len);
6760#endif
6761
6762	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6763	ctsio->kern_sg_entries = 0;
6764	ctsio->kern_data_resid = 0;
6765	ctsio->kern_rel_offset = 0;
6766	if (total_len < alloc_len) {
6767		ctsio->residual = alloc_len - total_len;
6768		ctsio->kern_data_len = total_len;
6769		ctsio->kern_total_len = total_len;
6770	} else {
6771		ctsio->residual = 0;
6772		ctsio->kern_data_len = alloc_len;
6773		ctsio->kern_total_len = alloc_len;
6774	}
6775
6776	switch (ctsio->cdb[0]) {
6777	case MODE_SENSE_6: {
6778		struct scsi_mode_hdr_6 *header;
6779
6780		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6781
6782		header->datalen = ctl_min(total_len - 1, 254);
6783		if (control_dev == 0) {
6784			header->dev_specific = 0x10; /* DPOFUA */
6785			if ((lun->flags & CTL_LUN_READONLY) ||
6786			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6787			    .eca_and_aen & SCP_SWP) != 0)
6788				    header->dev_specific |= 0x80; /* WP */
6789		}
6790		if (dbd)
6791			header->block_descr_len = 0;
6792		else
6793			header->block_descr_len =
6794				sizeof(struct scsi_mode_block_descr);
6795		block_desc = (struct scsi_mode_block_descr *)&header[1];
6796		break;
6797	}
6798	case MODE_SENSE_10: {
6799		struct scsi_mode_hdr_10 *header;
6800		int datalen;
6801
6802		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6803
6804		datalen = ctl_min(total_len - 2, 65533);
6805		scsi_ulto2b(datalen, header->datalen);
6806		if (control_dev == 0) {
6807			header->dev_specific = 0x10; /* DPOFUA */
6808			if ((lun->flags & CTL_LUN_READONLY) ||
6809			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6810			    .eca_and_aen & SCP_SWP) != 0)
6811				    header->dev_specific |= 0x80; /* WP */
6812		}
6813		if (dbd)
6814			scsi_ulto2b(0, header->block_descr_len);
6815		else
6816			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6817				    header->block_descr_len);
6818		block_desc = (struct scsi_mode_block_descr *)&header[1];
6819		break;
6820	}
6821	default:
6822		panic("invalid CDB type %#x", ctsio->cdb[0]);
6823		break; /* NOTREACHED */
6824	}
6825
6826	/*
6827	 * If we've got a disk, use its blocksize in the block
6828	 * descriptor.  Otherwise, just set it to 0.
6829	 */
6830	if (dbd == 0) {
6831		if (control_dev == 0)
6832			scsi_ulto3b(lun->be_lun->blocksize,
6833				    block_desc->block_len);
6834		else
6835			scsi_ulto3b(0, block_desc->block_len);
6836	}
6837
6838	switch (page_code) {
6839	case SMS_ALL_PAGES_PAGE: {
6840		int i, data_used;
6841
6842		data_used = header_len;
6843		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6844			struct ctl_page_index *page_index;
6845
6846			page_index = &lun->mode_pages.index[i];
6847
6848			if ((control_dev != 0)
6849			 && (page_index->page_flags &
6850			    CTL_PAGE_FLAG_DISK_ONLY))
6851				continue;
6852
6853			/*
6854			 * We don't use this subpage if the user didn't
6855			 * request all subpages.  We already checked (above)
6856			 * to make sure the user only specified a subpage
6857			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6858			 */
6859			if ((page_index->subpage != 0)
6860			 && (subpage == SMS_SUBPAGE_PAGE_0))
6861				continue;
6862
6863			/*
6864			 * Call the handler, if it exists, to update the
6865			 * page to the latest values.
6866			 */
6867			if (page_index->sense_handler != NULL)
6868				page_index->sense_handler(ctsio, page_index,pc);
6869
6870			memcpy(ctsio->kern_data_ptr + data_used,
6871			       page_index->page_data +
6872			       (page_index->page_len * pc),
6873			       page_index->page_len);
6874			data_used += page_index->page_len;
6875		}
6876		break;
6877	}
6878	default: {
6879		int i, data_used;
6880
6881		data_used = header_len;
6882
6883		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6884			struct ctl_page_index *page_index;
6885
6886			page_index = &lun->mode_pages.index[i];
6887
6888			/* Look for the right page code */
6889			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6890				continue;
6891
6892			/* Look for the right subpage or the subpage wildcard*/
6893			if ((page_index->subpage != subpage)
6894			 && (subpage != SMS_SUBPAGE_ALL))
6895				continue;
6896
6897			/* Make sure the page is supported for this dev type */
6898			if ((control_dev != 0)
6899			 && (page_index->page_flags &
6900			     CTL_PAGE_FLAG_DISK_ONLY))
6901				continue;
6902
6903			/*
6904			 * Call the handler, if it exists, to update the
6905			 * page to the latest values.
6906			 */
6907			if (page_index->sense_handler != NULL)
6908				page_index->sense_handler(ctsio, page_index,pc);
6909
6910			memcpy(ctsio->kern_data_ptr + data_used,
6911			       page_index->page_data +
6912			       (page_index->page_len * pc),
6913			       page_index->page_len);
6914			data_used += page_index->page_len;
6915		}
6916		break;
6917	}
6918	}
6919
6920	ctsio->scsi_status = SCSI_STATUS_OK;
6921
6922	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6923	ctsio->be_move_done = ctl_config_move_done;
6924	ctl_datamove((union ctl_io *)ctsio);
6925
6926	return (CTL_RETVAL_COMPLETE);
6927}
6928
6929int
6930ctl_log_sense(struct ctl_scsiio *ctsio)
6931{
6932	struct ctl_lun *lun;
6933	int i, pc, page_code, subpage;
6934	int alloc_len, total_len;
6935	struct ctl_page_index *page_index;
6936	struct scsi_log_sense *cdb;
6937	struct scsi_log_header *header;
6938
6939	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6940
6941	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6942	cdb = (struct scsi_log_sense *)ctsio->cdb;
6943	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6944	page_code = cdb->page & SLS_PAGE_CODE;
6945	subpage = cdb->subpage;
6946	alloc_len = scsi_2btoul(cdb->length);
6947
6948	page_index = NULL;
6949	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6950		page_index = &lun->log_pages.index[i];
6951
6952		/* Look for the right page code */
6953		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6954			continue;
6955
6956		/* Look for the right subpage or the subpage wildcard*/
6957		if (page_index->subpage != subpage)
6958			continue;
6959
6960		break;
6961	}
6962	if (i >= CTL_NUM_LOG_PAGES) {
6963		ctl_set_invalid_field(ctsio,
6964				      /*sks_valid*/ 1,
6965				      /*command*/ 1,
6966				      /*field*/ 2,
6967				      /*bit_valid*/ 0,
6968				      /*bit*/ 0);
6969		ctl_done((union ctl_io *)ctsio);
6970		return (CTL_RETVAL_COMPLETE);
6971	}
6972
6973	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6974
6975	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6976	ctsio->kern_sg_entries = 0;
6977	ctsio->kern_data_resid = 0;
6978	ctsio->kern_rel_offset = 0;
6979	if (total_len < alloc_len) {
6980		ctsio->residual = alloc_len - total_len;
6981		ctsio->kern_data_len = total_len;
6982		ctsio->kern_total_len = total_len;
6983	} else {
6984		ctsio->residual = 0;
6985		ctsio->kern_data_len = alloc_len;
6986		ctsio->kern_total_len = alloc_len;
6987	}
6988
6989	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6990	header->page = page_index->page_code;
6991	if (page_index->subpage) {
6992		header->page |= SL_SPF;
6993		header->subpage = page_index->subpage;
6994	}
6995	scsi_ulto2b(page_index->page_len, header->datalen);
6996
6997	/*
6998	 * Call the handler, if it exists, to update the
6999	 * page to the latest values.
7000	 */
7001	if (page_index->sense_handler != NULL)
7002		page_index->sense_handler(ctsio, page_index, pc);
7003
7004	memcpy(header + 1, page_index->page_data, page_index->page_len);
7005
7006	ctsio->scsi_status = SCSI_STATUS_OK;
7007	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7008	ctsio->be_move_done = ctl_config_move_done;
7009	ctl_datamove((union ctl_io *)ctsio);
7010
7011	return (CTL_RETVAL_COMPLETE);
7012}
7013
7014int
7015ctl_read_capacity(struct ctl_scsiio *ctsio)
7016{
7017	struct scsi_read_capacity *cdb;
7018	struct scsi_read_capacity_data *data;
7019	struct ctl_lun *lun;
7020	uint32_t lba;
7021
7022	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7023
7024	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7025
7026	lba = scsi_4btoul(cdb->addr);
7027	if (((cdb->pmi & SRC_PMI) == 0)
7028	 && (lba != 0)) {
7029		ctl_set_invalid_field(/*ctsio*/ ctsio,
7030				      /*sks_valid*/ 1,
7031				      /*command*/ 1,
7032				      /*field*/ 2,
7033				      /*bit_valid*/ 0,
7034				      /*bit*/ 0);
7035		ctl_done((union ctl_io *)ctsio);
7036		return (CTL_RETVAL_COMPLETE);
7037	}
7038
7039	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7040
7041	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7042	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7043	ctsio->residual = 0;
7044	ctsio->kern_data_len = sizeof(*data);
7045	ctsio->kern_total_len = sizeof(*data);
7046	ctsio->kern_data_resid = 0;
7047	ctsio->kern_rel_offset = 0;
7048	ctsio->kern_sg_entries = 0;
7049
7050	/*
7051	 * If the maximum LBA is greater than 0xfffffffe, the user must
7052	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7053	 * serivce action set.
7054	 */
7055	if (lun->be_lun->maxlba > 0xfffffffe)
7056		scsi_ulto4b(0xffffffff, data->addr);
7057	else
7058		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7059
7060	/*
7061	 * XXX KDM this may not be 512 bytes...
7062	 */
7063	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7064
7065	ctsio->scsi_status = SCSI_STATUS_OK;
7066
7067	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7068	ctsio->be_move_done = ctl_config_move_done;
7069	ctl_datamove((union ctl_io *)ctsio);
7070
7071	return (CTL_RETVAL_COMPLETE);
7072}
7073
7074int
7075ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7076{
7077	struct scsi_read_capacity_16 *cdb;
7078	struct scsi_read_capacity_data_long *data;
7079	struct ctl_lun *lun;
7080	uint64_t lba;
7081	uint32_t alloc_len;
7082
7083	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7084
7085	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7086
7087	alloc_len = scsi_4btoul(cdb->alloc_len);
7088	lba = scsi_8btou64(cdb->addr);
7089
7090	if ((cdb->reladr & SRC16_PMI)
7091	 && (lba != 0)) {
7092		ctl_set_invalid_field(/*ctsio*/ ctsio,
7093				      /*sks_valid*/ 1,
7094				      /*command*/ 1,
7095				      /*field*/ 2,
7096				      /*bit_valid*/ 0,
7097				      /*bit*/ 0);
7098		ctl_done((union ctl_io *)ctsio);
7099		return (CTL_RETVAL_COMPLETE);
7100	}
7101
7102	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7103
7104	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7105	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7106
7107	if (sizeof(*data) < alloc_len) {
7108		ctsio->residual = alloc_len - sizeof(*data);
7109		ctsio->kern_data_len = sizeof(*data);
7110		ctsio->kern_total_len = sizeof(*data);
7111	} else {
7112		ctsio->residual = 0;
7113		ctsio->kern_data_len = alloc_len;
7114		ctsio->kern_total_len = alloc_len;
7115	}
7116	ctsio->kern_data_resid = 0;
7117	ctsio->kern_rel_offset = 0;
7118	ctsio->kern_sg_entries = 0;
7119
7120	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7121	/* XXX KDM this may not be 512 bytes... */
7122	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7123	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7124	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7125	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7126		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7127
7128	ctsio->scsi_status = SCSI_STATUS_OK;
7129
7130	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7131	ctsio->be_move_done = ctl_config_move_done;
7132	ctl_datamove((union ctl_io *)ctsio);
7133
7134	return (CTL_RETVAL_COMPLETE);
7135}
7136
7137int
7138ctl_read_defect(struct ctl_scsiio *ctsio)
7139{
7140	struct scsi_read_defect_data_10 *ccb10;
7141	struct scsi_read_defect_data_12 *ccb12;
7142	struct scsi_read_defect_data_hdr_10 *data10;
7143	struct scsi_read_defect_data_hdr_12 *data12;
7144	struct ctl_lun *lun;
7145	uint32_t alloc_len, data_len;
7146	uint8_t format;
7147
7148	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7149
7150	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7151	if (lun->flags & CTL_LUN_PR_RESERVED) {
7152		uint32_t residx;
7153
7154		/*
7155		 * XXX KDM need a lock here.
7156		 */
7157		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7158		if ((lun->res_type == SPR_TYPE_EX_AC
7159		  && residx != lun->pr_res_idx)
7160		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
7161		   || lun->res_type == SPR_TYPE_EX_AC_AR)
7162		  && lun->pr_keys[residx] == 0)) {
7163			ctl_set_reservation_conflict(ctsio);
7164			ctl_done((union ctl_io *)ctsio);
7165			return (CTL_RETVAL_COMPLETE);
7166	        }
7167	}
7168
7169	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7170		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7171		format = ccb10->format;
7172		alloc_len = scsi_2btoul(ccb10->alloc_length);
7173		data_len = sizeof(*data10);
7174	} else {
7175		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7176		format = ccb12->format;
7177		alloc_len = scsi_4btoul(ccb12->alloc_length);
7178		data_len = sizeof(*data12);
7179	}
7180	if (alloc_len == 0) {
7181		ctl_set_success(ctsio);
7182		ctl_done((union ctl_io *)ctsio);
7183		return (CTL_RETVAL_COMPLETE);
7184	}
7185
7186	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7187	if (data_len < alloc_len) {
7188		ctsio->residual = alloc_len - data_len;
7189		ctsio->kern_data_len = data_len;
7190		ctsio->kern_total_len = data_len;
7191	} else {
7192		ctsio->residual = 0;
7193		ctsio->kern_data_len = alloc_len;
7194		ctsio->kern_total_len = alloc_len;
7195	}
7196	ctsio->kern_data_resid = 0;
7197	ctsio->kern_rel_offset = 0;
7198	ctsio->kern_sg_entries = 0;
7199
7200	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7201		data10 = (struct scsi_read_defect_data_hdr_10 *)
7202		    ctsio->kern_data_ptr;
7203		data10->format = format;
7204		scsi_ulto2b(0, data10->length);
7205	} else {
7206		data12 = (struct scsi_read_defect_data_hdr_12 *)
7207		    ctsio->kern_data_ptr;
7208		data12->format = format;
7209		scsi_ulto2b(0, data12->generation);
7210		scsi_ulto4b(0, data12->length);
7211	}
7212
7213	ctsio->scsi_status = SCSI_STATUS_OK;
7214	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7215	ctsio->be_move_done = ctl_config_move_done;
7216	ctl_datamove((union ctl_io *)ctsio);
7217	return (CTL_RETVAL_COMPLETE);
7218}
7219
7220int
7221ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7222{
7223	struct scsi_maintenance_in *cdb;
7224	int retval;
7225	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7226	int num_target_port_groups, num_target_ports, single;
7227	struct ctl_lun *lun;
7228	struct ctl_softc *softc;
7229	struct ctl_port *port;
7230	struct scsi_target_group_data *rtg_ptr;
7231	struct scsi_target_group_data_extended *rtg_ext_ptr;
7232	struct scsi_target_port_group_descriptor *tpg_desc;
7233
7234	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7235
7236	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7237	softc = control_softc;
7238	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7239
7240	retval = CTL_RETVAL_COMPLETE;
7241
7242	switch (cdb->byte2 & STG_PDF_MASK) {
7243	case STG_PDF_LENGTH:
7244		ext = 0;
7245		break;
7246	case STG_PDF_EXTENDED:
7247		ext = 1;
7248		break;
7249	default:
7250		ctl_set_invalid_field(/*ctsio*/ ctsio,
7251				      /*sks_valid*/ 1,
7252				      /*command*/ 1,
7253				      /*field*/ 2,
7254				      /*bit_valid*/ 1,
7255				      /*bit*/ 5);
7256		ctl_done((union ctl_io *)ctsio);
7257		return(retval);
7258	}
7259
7260	single = ctl_is_single;
7261	if (single)
7262		num_target_port_groups = 1;
7263	else
7264		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7265	num_target_ports = 0;
7266	mtx_lock(&softc->ctl_lock);
7267	STAILQ_FOREACH(port, &softc->port_list, links) {
7268		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7269			continue;
7270		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7271			continue;
7272		num_target_ports++;
7273	}
7274	mtx_unlock(&softc->ctl_lock);
7275
7276	if (ext)
7277		total_len = sizeof(struct scsi_target_group_data_extended);
7278	else
7279		total_len = sizeof(struct scsi_target_group_data);
7280	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7281		num_target_port_groups +
7282	    sizeof(struct scsi_target_port_descriptor) *
7283		num_target_ports * num_target_port_groups;
7284
7285	alloc_len = scsi_4btoul(cdb->length);
7286
7287	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7288
7289	ctsio->kern_sg_entries = 0;
7290
7291	if (total_len < alloc_len) {
7292		ctsio->residual = alloc_len - total_len;
7293		ctsio->kern_data_len = total_len;
7294		ctsio->kern_total_len = total_len;
7295	} else {
7296		ctsio->residual = 0;
7297		ctsio->kern_data_len = alloc_len;
7298		ctsio->kern_total_len = alloc_len;
7299	}
7300	ctsio->kern_data_resid = 0;
7301	ctsio->kern_rel_offset = 0;
7302
7303	if (ext) {
7304		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7305		    ctsio->kern_data_ptr;
7306		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7307		rtg_ext_ptr->format_type = 0x10;
7308		rtg_ext_ptr->implicit_transition_time = 0;
7309		tpg_desc = &rtg_ext_ptr->groups[0];
7310	} else {
7311		rtg_ptr = (struct scsi_target_group_data *)
7312		    ctsio->kern_data_ptr;
7313		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7314		tpg_desc = &rtg_ptr->groups[0];
7315	}
7316
7317	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7318	mtx_lock(&softc->ctl_lock);
7319	for (g = 0; g < num_target_port_groups; g++) {
7320		if (g == pg)
7321			tpg_desc->pref_state = TPG_PRIMARY |
7322			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7323		else
7324			tpg_desc->pref_state =
7325			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7326		tpg_desc->support = TPG_AO_SUP;
7327		if (!single)
7328			tpg_desc->support |= TPG_AN_SUP;
7329		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7330		tpg_desc->status = TPG_IMPLICIT;
7331		pc = 0;
7332		STAILQ_FOREACH(port, &softc->port_list, links) {
7333			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7334				continue;
7335			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7336			    CTL_MAX_LUNS)
7337				continue;
7338			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7339			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7340			    relative_target_port_identifier);
7341			pc++;
7342		}
7343		tpg_desc->target_port_count = pc;
7344		tpg_desc = (struct scsi_target_port_group_descriptor *)
7345		    &tpg_desc->descriptors[pc];
7346	}
7347	mtx_unlock(&softc->ctl_lock);
7348
7349	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7350	ctsio->be_move_done = ctl_config_move_done;
7351
7352	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7353			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7354			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7355			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7356			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7357
7358	ctl_datamove((union ctl_io *)ctsio);
7359	return(retval);
7360}
7361
7362int
7363ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7364{
7365	struct ctl_lun *lun;
7366	struct scsi_report_supported_opcodes *cdb;
7367	const struct ctl_cmd_entry *entry, *sentry;
7368	struct scsi_report_supported_opcodes_all *all;
7369	struct scsi_report_supported_opcodes_descr *descr;
7370	struct scsi_report_supported_opcodes_one *one;
7371	int retval;
7372	int alloc_len, total_len;
7373	int opcode, service_action, i, j, num;
7374
7375	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7376
7377	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7378	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7379
7380	retval = CTL_RETVAL_COMPLETE;
7381
7382	opcode = cdb->requested_opcode;
7383	service_action = scsi_2btoul(cdb->requested_service_action);
7384	switch (cdb->options & RSO_OPTIONS_MASK) {
7385	case RSO_OPTIONS_ALL:
7386		num = 0;
7387		for (i = 0; i < 256; i++) {
7388			entry = &ctl_cmd_table[i];
7389			if (entry->flags & CTL_CMD_FLAG_SA5) {
7390				for (j = 0; j < 32; j++) {
7391					sentry = &((const struct ctl_cmd_entry *)
7392					    entry->execute)[j];
7393					if (ctl_cmd_applicable(
7394					    lun->be_lun->lun_type, sentry))
7395						num++;
7396				}
7397			} else {
7398				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7399				    entry))
7400					num++;
7401			}
7402		}
7403		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7404		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7405		break;
7406	case RSO_OPTIONS_OC:
7407		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7408			ctl_set_invalid_field(/*ctsio*/ ctsio,
7409					      /*sks_valid*/ 1,
7410					      /*command*/ 1,
7411					      /*field*/ 2,
7412					      /*bit_valid*/ 1,
7413					      /*bit*/ 2);
7414			ctl_done((union ctl_io *)ctsio);
7415			return (CTL_RETVAL_COMPLETE);
7416		}
7417		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7418		break;
7419	case RSO_OPTIONS_OC_SA:
7420		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7421		    service_action >= 32) {
7422			ctl_set_invalid_field(/*ctsio*/ ctsio,
7423					      /*sks_valid*/ 1,
7424					      /*command*/ 1,
7425					      /*field*/ 2,
7426					      /*bit_valid*/ 1,
7427					      /*bit*/ 2);
7428			ctl_done((union ctl_io *)ctsio);
7429			return (CTL_RETVAL_COMPLETE);
7430		}
7431		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7432		break;
7433	default:
7434		ctl_set_invalid_field(/*ctsio*/ ctsio,
7435				      /*sks_valid*/ 1,
7436				      /*command*/ 1,
7437				      /*field*/ 2,
7438				      /*bit_valid*/ 1,
7439				      /*bit*/ 2);
7440		ctl_done((union ctl_io *)ctsio);
7441		return (CTL_RETVAL_COMPLETE);
7442	}
7443
7444	alloc_len = scsi_4btoul(cdb->length);
7445
7446	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7447
7448	ctsio->kern_sg_entries = 0;
7449
7450	if (total_len < alloc_len) {
7451		ctsio->residual = alloc_len - total_len;
7452		ctsio->kern_data_len = total_len;
7453		ctsio->kern_total_len = total_len;
7454	} else {
7455		ctsio->residual = 0;
7456		ctsio->kern_data_len = alloc_len;
7457		ctsio->kern_total_len = alloc_len;
7458	}
7459	ctsio->kern_data_resid = 0;
7460	ctsio->kern_rel_offset = 0;
7461
7462	switch (cdb->options & RSO_OPTIONS_MASK) {
7463	case RSO_OPTIONS_ALL:
7464		all = (struct scsi_report_supported_opcodes_all *)
7465		    ctsio->kern_data_ptr;
7466		num = 0;
7467		for (i = 0; i < 256; i++) {
7468			entry = &ctl_cmd_table[i];
7469			if (entry->flags & CTL_CMD_FLAG_SA5) {
7470				for (j = 0; j < 32; j++) {
7471					sentry = &((const struct ctl_cmd_entry *)
7472					    entry->execute)[j];
7473					if (!ctl_cmd_applicable(
7474					    lun->be_lun->lun_type, sentry))
7475						continue;
7476					descr = &all->descr[num++];
7477					descr->opcode = i;
7478					scsi_ulto2b(j, descr->service_action);
7479					descr->flags = RSO_SERVACTV;
7480					scsi_ulto2b(sentry->length,
7481					    descr->cdb_length);
7482				}
7483			} else {
7484				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7485				    entry))
7486					continue;
7487				descr = &all->descr[num++];
7488				descr->opcode = i;
7489				scsi_ulto2b(0, descr->service_action);
7490				descr->flags = 0;
7491				scsi_ulto2b(entry->length, descr->cdb_length);
7492			}
7493		}
7494		scsi_ulto4b(
7495		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7496		    all->length);
7497		break;
7498	case RSO_OPTIONS_OC:
7499		one = (struct scsi_report_supported_opcodes_one *)
7500		    ctsio->kern_data_ptr;
7501		entry = &ctl_cmd_table[opcode];
7502		goto fill_one;
7503	case RSO_OPTIONS_OC_SA:
7504		one = (struct scsi_report_supported_opcodes_one *)
7505		    ctsio->kern_data_ptr;
7506		entry = &ctl_cmd_table[opcode];
7507		entry = &((const struct ctl_cmd_entry *)
7508		    entry->execute)[service_action];
7509fill_one:
7510		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7511			one->support = 3;
7512			scsi_ulto2b(entry->length, one->cdb_length);
7513			one->cdb_usage[0] = opcode;
7514			memcpy(&one->cdb_usage[1], entry->usage,
7515			    entry->length - 1);
7516		} else
7517			one->support = 1;
7518		break;
7519	}
7520
7521	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7522	ctsio->be_move_done = ctl_config_move_done;
7523
7524	ctl_datamove((union ctl_io *)ctsio);
7525	return(retval);
7526}
7527
7528int
7529ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7530{
7531	struct ctl_lun *lun;
7532	struct scsi_report_supported_tmf *cdb;
7533	struct scsi_report_supported_tmf_data *data;
7534	int retval;
7535	int alloc_len, total_len;
7536
7537	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7538
7539	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7540	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7541
7542	retval = CTL_RETVAL_COMPLETE;
7543
7544	total_len = sizeof(struct scsi_report_supported_tmf_data);
7545	alloc_len = scsi_4btoul(cdb->length);
7546
7547	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7548
7549	ctsio->kern_sg_entries = 0;
7550
7551	if (total_len < alloc_len) {
7552		ctsio->residual = alloc_len - total_len;
7553		ctsio->kern_data_len = total_len;
7554		ctsio->kern_total_len = total_len;
7555	} else {
7556		ctsio->residual = 0;
7557		ctsio->kern_data_len = alloc_len;
7558		ctsio->kern_total_len = alloc_len;
7559	}
7560	ctsio->kern_data_resid = 0;
7561	ctsio->kern_rel_offset = 0;
7562
7563	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7564	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7565	data->byte2 |= RST_ITNRS;
7566
7567	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7568	ctsio->be_move_done = ctl_config_move_done;
7569
7570	ctl_datamove((union ctl_io *)ctsio);
7571	return (retval);
7572}
7573
7574int
7575ctl_report_timestamp(struct ctl_scsiio *ctsio)
7576{
7577	struct ctl_lun *lun;
7578	struct scsi_report_timestamp *cdb;
7579	struct scsi_report_timestamp_data *data;
7580	struct timeval tv;
7581	int64_t timestamp;
7582	int retval;
7583	int alloc_len, total_len;
7584
7585	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7586
7587	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7588	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7589
7590	retval = CTL_RETVAL_COMPLETE;
7591
7592	total_len = sizeof(struct scsi_report_timestamp_data);
7593	alloc_len = scsi_4btoul(cdb->length);
7594
7595	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7596
7597	ctsio->kern_sg_entries = 0;
7598
7599	if (total_len < alloc_len) {
7600		ctsio->residual = alloc_len - total_len;
7601		ctsio->kern_data_len = total_len;
7602		ctsio->kern_total_len = total_len;
7603	} else {
7604		ctsio->residual = 0;
7605		ctsio->kern_data_len = alloc_len;
7606		ctsio->kern_total_len = alloc_len;
7607	}
7608	ctsio->kern_data_resid = 0;
7609	ctsio->kern_rel_offset = 0;
7610
7611	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7612	scsi_ulto2b(sizeof(*data) - 2, data->length);
7613	data->origin = RTS_ORIG_OUTSIDE;
7614	getmicrotime(&tv);
7615	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7616	scsi_ulto4b(timestamp >> 16, data->timestamp);
7617	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7618
7619	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7620	ctsio->be_move_done = ctl_config_move_done;
7621
7622	ctl_datamove((union ctl_io *)ctsio);
7623	return (retval);
7624}
7625
7626int
7627ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7628{
7629	struct scsi_per_res_in *cdb;
7630	int alloc_len, total_len = 0;
7631	/* struct scsi_per_res_in_rsrv in_data; */
7632	struct ctl_lun *lun;
7633	struct ctl_softc *softc;
7634
7635	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7636
7637	softc = control_softc;
7638
7639	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7640
7641	alloc_len = scsi_2btoul(cdb->length);
7642
7643	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7644
7645retry:
7646	mtx_lock(&lun->lun_lock);
7647	switch (cdb->action) {
7648	case SPRI_RK: /* read keys */
7649		total_len = sizeof(struct scsi_per_res_in_keys) +
7650			lun->pr_key_count *
7651			sizeof(struct scsi_per_res_key);
7652		break;
7653	case SPRI_RR: /* read reservation */
7654		if (lun->flags & CTL_LUN_PR_RESERVED)
7655			total_len = sizeof(struct scsi_per_res_in_rsrv);
7656		else
7657			total_len = sizeof(struct scsi_per_res_in_header);
7658		break;
7659	case SPRI_RC: /* report capabilities */
7660		total_len = sizeof(struct scsi_per_res_cap);
7661		break;
7662	case SPRI_RS: /* read full status */
7663		total_len = sizeof(struct scsi_per_res_in_header) +
7664		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7665		    lun->pr_key_count;
7666		break;
7667	default:
7668		panic("Invalid PR type %x", cdb->action);
7669	}
7670	mtx_unlock(&lun->lun_lock);
7671
7672	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7673
7674	if (total_len < alloc_len) {
7675		ctsio->residual = alloc_len - total_len;
7676		ctsio->kern_data_len = total_len;
7677		ctsio->kern_total_len = total_len;
7678	} else {
7679		ctsio->residual = 0;
7680		ctsio->kern_data_len = alloc_len;
7681		ctsio->kern_total_len = alloc_len;
7682	}
7683
7684	ctsio->kern_data_resid = 0;
7685	ctsio->kern_rel_offset = 0;
7686	ctsio->kern_sg_entries = 0;
7687
7688	mtx_lock(&lun->lun_lock);
7689	switch (cdb->action) {
7690	case SPRI_RK: { // read keys
7691        struct scsi_per_res_in_keys *res_keys;
7692		int i, key_count;
7693
7694		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7695
7696		/*
7697		 * We had to drop the lock to allocate our buffer, which
7698		 * leaves time for someone to come in with another
7699		 * persistent reservation.  (That is unlikely, though,
7700		 * since this should be the only persistent reservation
7701		 * command active right now.)
7702		 */
7703		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7704		    (lun->pr_key_count *
7705		     sizeof(struct scsi_per_res_key)))){
7706			mtx_unlock(&lun->lun_lock);
7707			free(ctsio->kern_data_ptr, M_CTL);
7708			printf("%s: reservation length changed, retrying\n",
7709			       __func__);
7710			goto retry;
7711		}
7712
7713		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7714
7715		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7716			     lun->pr_key_count, res_keys->header.length);
7717
7718		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7719			if (lun->pr_keys[i] == 0)
7720				continue;
7721
7722			/*
7723			 * We used lun->pr_key_count to calculate the
7724			 * size to allocate.  If it turns out the number of
7725			 * initiators with the registered flag set is
7726			 * larger than that (i.e. they haven't been kept in
7727			 * sync), we've got a problem.
7728			 */
7729			if (key_count >= lun->pr_key_count) {
7730#ifdef NEEDTOPORT
7731				csevent_log(CSC_CTL | CSC_SHELF_SW |
7732					    CTL_PR_ERROR,
7733					    csevent_LogType_Fault,
7734					    csevent_AlertLevel_Yellow,
7735					    csevent_FRU_ShelfController,
7736					    csevent_FRU_Firmware,
7737				        csevent_FRU_Unknown,
7738					    "registered keys %d >= key "
7739					    "count %d", key_count,
7740					    lun->pr_key_count);
7741#endif
7742				key_count++;
7743				continue;
7744			}
7745			scsi_u64to8b(lun->pr_keys[i],
7746			    res_keys->keys[key_count].key);
7747			key_count++;
7748		}
7749		break;
7750	}
7751	case SPRI_RR: { // read reservation
7752		struct scsi_per_res_in_rsrv *res;
7753		int tmp_len, header_only;
7754
7755		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7756
7757		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7758
7759		if (lun->flags & CTL_LUN_PR_RESERVED)
7760		{
7761			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7762			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7763				    res->header.length);
7764			header_only = 0;
7765		} else {
7766			tmp_len = sizeof(struct scsi_per_res_in_header);
7767			scsi_ulto4b(0, res->header.length);
7768			header_only = 1;
7769		}
7770
7771		/*
7772		 * We had to drop the lock to allocate our buffer, which
7773		 * leaves time for someone to come in with another
7774		 * persistent reservation.  (That is unlikely, though,
7775		 * since this should be the only persistent reservation
7776		 * command active right now.)
7777		 */
7778		if (tmp_len != total_len) {
7779			mtx_unlock(&lun->lun_lock);
7780			free(ctsio->kern_data_ptr, M_CTL);
7781			printf("%s: reservation status changed, retrying\n",
7782			       __func__);
7783			goto retry;
7784		}
7785
7786		/*
7787		 * No reservation held, so we're done.
7788		 */
7789		if (header_only != 0)
7790			break;
7791
7792		/*
7793		 * If the registration is an All Registrants type, the key
7794		 * is 0, since it doesn't really matter.
7795		 */
7796		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7797			scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7798			    res->data.reservation);
7799		}
7800		res->data.scopetype = lun->res_type;
7801		break;
7802	}
7803	case SPRI_RC:     //report capabilities
7804	{
7805		struct scsi_per_res_cap *res_cap;
7806		uint16_t type_mask;
7807
7808		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7809		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7810		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7811		type_mask = SPRI_TM_WR_EX_AR |
7812			    SPRI_TM_EX_AC_RO |
7813			    SPRI_TM_WR_EX_RO |
7814			    SPRI_TM_EX_AC |
7815			    SPRI_TM_WR_EX |
7816			    SPRI_TM_EX_AC_AR;
7817		scsi_ulto2b(type_mask, res_cap->type_mask);
7818		break;
7819	}
7820	case SPRI_RS: { // read full status
7821		struct scsi_per_res_in_full *res_status;
7822		struct scsi_per_res_in_full_desc *res_desc;
7823		struct ctl_port *port;
7824		int i, len;
7825
7826		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7827
7828		/*
7829		 * We had to drop the lock to allocate our buffer, which
7830		 * leaves time for someone to come in with another
7831		 * persistent reservation.  (That is unlikely, though,
7832		 * since this should be the only persistent reservation
7833		 * command active right now.)
7834		 */
7835		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7836		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7837		     lun->pr_key_count)){
7838			mtx_unlock(&lun->lun_lock);
7839			free(ctsio->kern_data_ptr, M_CTL);
7840			printf("%s: reservation length changed, retrying\n",
7841			       __func__);
7842			goto retry;
7843		}
7844
7845		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7846
7847		res_desc = &res_status->desc[0];
7848		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7849			if (lun->pr_keys[i] == 0)
7850				continue;
7851
7852			scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
7853			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7854			    (lun->pr_res_idx == i ||
7855			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7856				res_desc->flags = SPRI_FULL_R_HOLDER;
7857				res_desc->scopetype = lun->res_type;
7858			}
7859			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7860			    res_desc->rel_trgt_port_id);
7861			len = 0;
7862			port = softc->ctl_ports[
7863			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7864			if (port != NULL)
7865				len = ctl_create_iid(port,
7866				    i % CTL_MAX_INIT_PER_PORT,
7867				    res_desc->transport_id);
7868			scsi_ulto4b(len, res_desc->additional_length);
7869			res_desc = (struct scsi_per_res_in_full_desc *)
7870			    &res_desc->transport_id[len];
7871		}
7872		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7873		    res_status->header.length);
7874		break;
7875	}
7876	default:
7877		/*
7878		 * This is a bug, because we just checked for this above,
7879		 * and should have returned an error.
7880		 */
7881		panic("Invalid PR type %x", cdb->action);
7882		break; /* NOTREACHED */
7883	}
7884	mtx_unlock(&lun->lun_lock);
7885
7886	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7887	ctsio->be_move_done = ctl_config_move_done;
7888
7889	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7890			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7891			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7892			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7893			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7894
7895	ctl_datamove((union ctl_io *)ctsio);
7896
7897	return (CTL_RETVAL_COMPLETE);
7898}
7899
7900/*
7901 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7902 * it should return.
7903 */
7904static int
7905ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7906		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7907		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7908		struct scsi_per_res_out_parms* param)
7909{
7910	union ctl_ha_msg persis_io;
7911	int retval, i;
7912	int isc_retval;
7913
7914	retval = 0;
7915
7916	mtx_lock(&lun->lun_lock);
7917	if (sa_res_key == 0) {
7918		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7919			/* validate scope and type */
7920			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7921			     SPR_LU_SCOPE) {
7922				mtx_unlock(&lun->lun_lock);
7923				ctl_set_invalid_field(/*ctsio*/ ctsio,
7924						      /*sks_valid*/ 1,
7925						      /*command*/ 1,
7926						      /*field*/ 2,
7927						      /*bit_valid*/ 1,
7928						      /*bit*/ 4);
7929				ctl_done((union ctl_io *)ctsio);
7930				return (1);
7931			}
7932
7933		        if (type>8 || type==2 || type==4 || type==0) {
7934				mtx_unlock(&lun->lun_lock);
7935				ctl_set_invalid_field(/*ctsio*/ ctsio,
7936       	           				      /*sks_valid*/ 1,
7937						      /*command*/ 1,
7938						      /*field*/ 2,
7939						      /*bit_valid*/ 1,
7940						      /*bit*/ 0);
7941				ctl_done((union ctl_io *)ctsio);
7942				return (1);
7943		        }
7944
7945			/*
7946			 * Unregister everybody else and build UA for
7947			 * them
7948			 */
7949			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7950				if (i == residx || lun->pr_keys[i] == 0)
7951					continue;
7952
7953				if (!persis_offset
7954				 && i <CTL_MAX_INITIATORS)
7955					lun->pending_ua[i] |=
7956						CTL_UA_REG_PREEMPT;
7957				else if (persis_offset
7958				      && i >= persis_offset)
7959					lun->pending_ua[i-persis_offset] |=
7960						CTL_UA_REG_PREEMPT;
7961				lun->pr_keys[i] = 0;
7962			}
7963			lun->pr_key_count = 1;
7964			lun->res_type = type;
7965			if (lun->res_type != SPR_TYPE_WR_EX_AR
7966			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7967				lun->pr_res_idx = residx;
7968
7969			/* send msg to other side */
7970			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7971			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7972			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7973			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7974			persis_io.pr.pr_info.res_type = type;
7975			memcpy(persis_io.pr.pr_info.sa_res_key,
7976			       param->serv_act_res_key,
7977			       sizeof(param->serv_act_res_key));
7978			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7979			     &persis_io, sizeof(persis_io), 0)) >
7980			     CTL_HA_STATUS_SUCCESS) {
7981				printf("CTL:Persis Out error returned "
7982				       "from ctl_ha_msg_send %d\n",
7983				       isc_retval);
7984			}
7985		} else {
7986			/* not all registrants */
7987			mtx_unlock(&lun->lun_lock);
7988			free(ctsio->kern_data_ptr, M_CTL);
7989			ctl_set_invalid_field(ctsio,
7990					      /*sks_valid*/ 1,
7991					      /*command*/ 0,
7992					      /*field*/ 8,
7993					      /*bit_valid*/ 0,
7994					      /*bit*/ 0);
7995			ctl_done((union ctl_io *)ctsio);
7996			return (1);
7997		}
7998	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7999		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8000		int found = 0;
8001
8002		if (res_key == sa_res_key) {
8003			/* special case */
8004			/*
8005			 * The spec implies this is not good but doesn't
8006			 * say what to do. There are two choices either
8007			 * generate a res conflict or check condition
8008			 * with illegal field in parameter data. Since
8009			 * that is what is done when the sa_res_key is
8010			 * zero I'll take that approach since this has
8011			 * to do with the sa_res_key.
8012			 */
8013			mtx_unlock(&lun->lun_lock);
8014			free(ctsio->kern_data_ptr, M_CTL);
8015			ctl_set_invalid_field(ctsio,
8016					      /*sks_valid*/ 1,
8017					      /*command*/ 0,
8018					      /*field*/ 8,
8019					      /*bit_valid*/ 0,
8020					      /*bit*/ 0);
8021			ctl_done((union ctl_io *)ctsio);
8022			return (1);
8023		}
8024
8025		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8026			if (lun->pr_keys[i] != sa_res_key)
8027				continue;
8028
8029			found = 1;
8030			lun->pr_keys[i] = 0;
8031			lun->pr_key_count--;
8032
8033			if (!persis_offset && i < CTL_MAX_INITIATORS)
8034				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8035			else if (persis_offset && i >= persis_offset)
8036				lun->pending_ua[i-persis_offset] |=
8037					CTL_UA_REG_PREEMPT;
8038		}
8039		if (!found) {
8040			mtx_unlock(&lun->lun_lock);
8041			free(ctsio->kern_data_ptr, M_CTL);
8042			ctl_set_reservation_conflict(ctsio);
8043			ctl_done((union ctl_io *)ctsio);
8044			return (CTL_RETVAL_COMPLETE);
8045		}
8046		/* send msg to other side */
8047		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8048		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8049		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8050		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8051		persis_io.pr.pr_info.res_type = type;
8052		memcpy(persis_io.pr.pr_info.sa_res_key,
8053		       param->serv_act_res_key,
8054		       sizeof(param->serv_act_res_key));
8055		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8056		     &persis_io, sizeof(persis_io), 0)) >
8057		     CTL_HA_STATUS_SUCCESS) {
8058			printf("CTL:Persis Out error returned from "
8059			       "ctl_ha_msg_send %d\n", isc_retval);
8060		}
8061	} else {
8062		/* Reserved but not all registrants */
8063		/* sa_res_key is res holder */
8064		if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8065			/* validate scope and type */
8066			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8067			     SPR_LU_SCOPE) {
8068				mtx_unlock(&lun->lun_lock);
8069				ctl_set_invalid_field(/*ctsio*/ ctsio,
8070						      /*sks_valid*/ 1,
8071						      /*command*/ 1,
8072						      /*field*/ 2,
8073						      /*bit_valid*/ 1,
8074						      /*bit*/ 4);
8075				ctl_done((union ctl_io *)ctsio);
8076				return (1);
8077			}
8078
8079			if (type>8 || type==2 || type==4 || type==0) {
8080				mtx_unlock(&lun->lun_lock);
8081				ctl_set_invalid_field(/*ctsio*/ ctsio,
8082						      /*sks_valid*/ 1,
8083						      /*command*/ 1,
8084						      /*field*/ 2,
8085						      /*bit_valid*/ 1,
8086						      /*bit*/ 0);
8087				ctl_done((union ctl_io *)ctsio);
8088				return (1);
8089			}
8090
8091			/*
8092			 * Do the following:
8093			 * if sa_res_key != res_key remove all
8094			 * registrants w/sa_res_key and generate UA
8095			 * for these registrants(Registrations
8096			 * Preempted) if it wasn't an exclusive
8097			 * reservation generate UA(Reservations
8098			 * Preempted) for all other registered nexuses
8099			 * if the type has changed. Establish the new
8100			 * reservation and holder. If res_key and
8101			 * sa_res_key are the same do the above
8102			 * except don't unregister the res holder.
8103			 */
8104
8105			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8106				if (i == residx || lun->pr_keys[i] == 0)
8107					continue;
8108
8109				if (sa_res_key == lun->pr_keys[i]) {
8110					lun->pr_keys[i] = 0;
8111					lun->pr_key_count--;
8112
8113					if (!persis_offset
8114					 && i < CTL_MAX_INITIATORS)
8115						lun->pending_ua[i] |=
8116							CTL_UA_REG_PREEMPT;
8117					else if (persis_offset
8118					      && i >= persis_offset)
8119						lun->pending_ua[i-persis_offset] |=
8120						  CTL_UA_REG_PREEMPT;
8121				} else if (type != lun->res_type
8122					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8123					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8124						if (!persis_offset
8125						 && i < CTL_MAX_INITIATORS)
8126							lun->pending_ua[i] |=
8127							CTL_UA_RES_RELEASE;
8128						else if (persis_offset
8129						      && i >= persis_offset)
8130							lun->pending_ua[
8131							i-persis_offset] |=
8132							CTL_UA_RES_RELEASE;
8133				}
8134			}
8135			lun->res_type = type;
8136			if (lun->res_type != SPR_TYPE_WR_EX_AR
8137			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8138				lun->pr_res_idx = residx;
8139			else
8140				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8141
8142			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8143			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8144			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8145			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8146			persis_io.pr.pr_info.res_type = type;
8147			memcpy(persis_io.pr.pr_info.sa_res_key,
8148			       param->serv_act_res_key,
8149			       sizeof(param->serv_act_res_key));
8150			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8151			     &persis_io, sizeof(persis_io), 0)) >
8152			     CTL_HA_STATUS_SUCCESS) {
8153				printf("CTL:Persis Out error returned "
8154				       "from ctl_ha_msg_send %d\n",
8155				       isc_retval);
8156			}
8157		} else {
8158			/*
8159			 * sa_res_key is not the res holder just
8160			 * remove registrants
8161			 */
8162			int found=0;
8163
8164			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8165				if (sa_res_key != lun->pr_keys[i])
8166					continue;
8167
8168				found = 1;
8169				lun->pr_keys[i] = 0;
8170				lun->pr_key_count--;
8171
8172				if (!persis_offset
8173				 && i < CTL_MAX_INITIATORS)
8174					lun->pending_ua[i] |=
8175						CTL_UA_REG_PREEMPT;
8176				else if (persis_offset
8177				      && i >= persis_offset)
8178					lun->pending_ua[i-persis_offset] |=
8179						CTL_UA_REG_PREEMPT;
8180			}
8181
8182			if (!found) {
8183				mtx_unlock(&lun->lun_lock);
8184				free(ctsio->kern_data_ptr, M_CTL);
8185				ctl_set_reservation_conflict(ctsio);
8186				ctl_done((union ctl_io *)ctsio);
8187		        	return (1);
8188			}
8189			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8190			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8191			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8192			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8193			persis_io.pr.pr_info.res_type = type;
8194			memcpy(persis_io.pr.pr_info.sa_res_key,
8195			       param->serv_act_res_key,
8196			       sizeof(param->serv_act_res_key));
8197			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8198			     &persis_io, sizeof(persis_io), 0)) >
8199			     CTL_HA_STATUS_SUCCESS) {
8200				printf("CTL:Persis Out error returned "
8201				       "from ctl_ha_msg_send %d\n",
8202				isc_retval);
8203			}
8204		}
8205	}
8206
8207	lun->PRGeneration++;
8208	mtx_unlock(&lun->lun_lock);
8209
8210	return (retval);
8211}
8212
8213static void
8214ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8215{
8216	uint64_t sa_res_key;
8217	int i;
8218
8219	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8220
8221	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8222	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8223	 || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8224		if (sa_res_key == 0) {
8225			/*
8226			 * Unregister everybody else and build UA for
8227			 * them
8228			 */
8229			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8230				if (i == msg->pr.pr_info.residx ||
8231				    lun->pr_keys[i] == 0)
8232					continue;
8233
8234				if (!persis_offset
8235				 && i < CTL_MAX_INITIATORS)
8236					lun->pending_ua[i] |=
8237						CTL_UA_REG_PREEMPT;
8238				else if (persis_offset && i >= persis_offset)
8239					lun->pending_ua[i - persis_offset] |=
8240						CTL_UA_REG_PREEMPT;
8241				lun->pr_keys[i] = 0;
8242			}
8243
8244			lun->pr_key_count = 1;
8245			lun->res_type = msg->pr.pr_info.res_type;
8246			if (lun->res_type != SPR_TYPE_WR_EX_AR
8247			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8248				lun->pr_res_idx = msg->pr.pr_info.residx;
8249		} else {
8250		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8251				if (sa_res_key == lun->pr_keys[i])
8252					continue;
8253
8254				lun->pr_keys[i] = 0;
8255				lun->pr_key_count--;
8256
8257				if (!persis_offset
8258				 && i < persis_offset)
8259					lun->pending_ua[i] |=
8260						CTL_UA_REG_PREEMPT;
8261				else if (persis_offset
8262				      && i >= persis_offset)
8263					lun->pending_ua[i - persis_offset] |=
8264						CTL_UA_REG_PREEMPT;
8265			}
8266		}
8267	} else {
8268		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8269			if (i == msg->pr.pr_info.residx ||
8270			    lun->pr_keys[i] == 0)
8271				continue;
8272
8273			if (sa_res_key == lun->pr_keys[i]) {
8274				lun->pr_keys[i] = 0;
8275				lun->pr_key_count--;
8276				if (!persis_offset
8277				 && i < CTL_MAX_INITIATORS)
8278					lun->pending_ua[i] |=
8279						CTL_UA_REG_PREEMPT;
8280				else if (persis_offset
8281				      && i >= persis_offset)
8282					lun->pending_ua[i - persis_offset] |=
8283						CTL_UA_REG_PREEMPT;
8284			} else if (msg->pr.pr_info.res_type != lun->res_type
8285				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8286				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8287					if (!persis_offset
8288					 && i < persis_offset)
8289						lun->pending_ua[i] |=
8290							CTL_UA_RES_RELEASE;
8291					else if (persis_offset
8292					      && i >= persis_offset)
8293					lun->pending_ua[i - persis_offset] |=
8294						CTL_UA_RES_RELEASE;
8295			}
8296		}
8297		lun->res_type = msg->pr.pr_info.res_type;
8298		if (lun->res_type != SPR_TYPE_WR_EX_AR
8299		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8300			lun->pr_res_idx = msg->pr.pr_info.residx;
8301		else
8302			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8303	}
8304	lun->PRGeneration++;
8305
8306}
8307
8308
8309int
8310ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8311{
8312	int retval;
8313	int isc_retval;
8314	u_int32_t param_len;
8315	struct scsi_per_res_out *cdb;
8316	struct ctl_lun *lun;
8317	struct scsi_per_res_out_parms* param;
8318	struct ctl_softc *softc;
8319	uint32_t residx;
8320	uint64_t res_key, sa_res_key;
8321	uint8_t type;
8322	union ctl_ha_msg persis_io;
8323	int    i;
8324
8325	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8326
8327	retval = CTL_RETVAL_COMPLETE;
8328
8329	softc = control_softc;
8330
8331	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8332	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8333
8334	/*
8335	 * We only support whole-LUN scope.  The scope & type are ignored for
8336	 * register, register and ignore existing key and clear.
8337	 * We sometimes ignore scope and type on preempts too!!
8338	 * Verify reservation type here as well.
8339	 */
8340	type = cdb->scope_type & SPR_TYPE_MASK;
8341	if ((cdb->action == SPRO_RESERVE)
8342	 || (cdb->action == SPRO_RELEASE)) {
8343		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8344			ctl_set_invalid_field(/*ctsio*/ ctsio,
8345					      /*sks_valid*/ 1,
8346					      /*command*/ 1,
8347					      /*field*/ 2,
8348					      /*bit_valid*/ 1,
8349					      /*bit*/ 4);
8350			ctl_done((union ctl_io *)ctsio);
8351			return (CTL_RETVAL_COMPLETE);
8352		}
8353
8354		if (type>8 || type==2 || type==4 || type==0) {
8355			ctl_set_invalid_field(/*ctsio*/ ctsio,
8356					      /*sks_valid*/ 1,
8357					      /*command*/ 1,
8358					      /*field*/ 2,
8359					      /*bit_valid*/ 1,
8360					      /*bit*/ 0);
8361			ctl_done((union ctl_io *)ctsio);
8362			return (CTL_RETVAL_COMPLETE);
8363		}
8364	}
8365
8366	param_len = scsi_4btoul(cdb->length);
8367
8368	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8369		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8370		ctsio->kern_data_len = param_len;
8371		ctsio->kern_total_len = param_len;
8372		ctsio->kern_data_resid = 0;
8373		ctsio->kern_rel_offset = 0;
8374		ctsio->kern_sg_entries = 0;
8375		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8376		ctsio->be_move_done = ctl_config_move_done;
8377		ctl_datamove((union ctl_io *)ctsio);
8378
8379		return (CTL_RETVAL_COMPLETE);
8380	}
8381
8382	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8383
8384	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8385	res_key = scsi_8btou64(param->res_key.key);
8386	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8387
8388	/*
8389	 * Validate the reservation key here except for SPRO_REG_IGNO
8390	 * This must be done for all other service actions
8391	 */
8392	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8393		mtx_lock(&lun->lun_lock);
8394		if (lun->pr_keys[residx] != 0) {
8395		    if (res_key != lun->pr_keys[residx]) {
8396				/*
8397				 * The current key passed in doesn't match
8398				 * the one the initiator previously
8399				 * registered.
8400				 */
8401				mtx_unlock(&lun->lun_lock);
8402				free(ctsio->kern_data_ptr, M_CTL);
8403				ctl_set_reservation_conflict(ctsio);
8404				ctl_done((union ctl_io *)ctsio);
8405				return (CTL_RETVAL_COMPLETE);
8406			}
8407		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8408			/*
8409			 * We are not registered
8410			 */
8411			mtx_unlock(&lun->lun_lock);
8412			free(ctsio->kern_data_ptr, M_CTL);
8413			ctl_set_reservation_conflict(ctsio);
8414			ctl_done((union ctl_io *)ctsio);
8415			return (CTL_RETVAL_COMPLETE);
8416		} else if (res_key != 0) {
8417			/*
8418			 * We are not registered and trying to register but
8419			 * the register key isn't zero.
8420			 */
8421			mtx_unlock(&lun->lun_lock);
8422			free(ctsio->kern_data_ptr, M_CTL);
8423			ctl_set_reservation_conflict(ctsio);
8424			ctl_done((union ctl_io *)ctsio);
8425			return (CTL_RETVAL_COMPLETE);
8426		}
8427		mtx_unlock(&lun->lun_lock);
8428	}
8429
8430	switch (cdb->action & SPRO_ACTION_MASK) {
8431	case SPRO_REGISTER:
8432	case SPRO_REG_IGNO: {
8433
8434#if 0
8435		printf("Registration received\n");
8436#endif
8437
8438		/*
8439		 * We don't support any of these options, as we report in
8440		 * the read capabilities request (see
8441		 * ctl_persistent_reserve_in(), above).
8442		 */
8443		if ((param->flags & SPR_SPEC_I_PT)
8444		 || (param->flags & SPR_ALL_TG_PT)
8445		 || (param->flags & SPR_APTPL)) {
8446			int bit_ptr;
8447
8448			if (param->flags & SPR_APTPL)
8449				bit_ptr = 0;
8450			else if (param->flags & SPR_ALL_TG_PT)
8451				bit_ptr = 2;
8452			else /* SPR_SPEC_I_PT */
8453				bit_ptr = 3;
8454
8455			free(ctsio->kern_data_ptr, M_CTL);
8456			ctl_set_invalid_field(ctsio,
8457					      /*sks_valid*/ 1,
8458					      /*command*/ 0,
8459					      /*field*/ 20,
8460					      /*bit_valid*/ 1,
8461					      /*bit*/ bit_ptr);
8462			ctl_done((union ctl_io *)ctsio);
8463			return (CTL_RETVAL_COMPLETE);
8464		}
8465
8466		mtx_lock(&lun->lun_lock);
8467
8468		/*
8469		 * The initiator wants to clear the
8470		 * key/unregister.
8471		 */
8472		if (sa_res_key == 0) {
8473			if ((res_key == 0
8474			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8475			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8476			  && lun->pr_keys[residx] == 0)) {
8477				mtx_unlock(&lun->lun_lock);
8478				goto done;
8479			}
8480
8481			lun->pr_keys[residx] = 0;
8482			lun->pr_key_count--;
8483
8484			if (residx == lun->pr_res_idx) {
8485				lun->flags &= ~CTL_LUN_PR_RESERVED;
8486				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8487
8488				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8489				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8490				 && lun->pr_key_count) {
8491					/*
8492					 * If the reservation is a registrants
8493					 * only type we need to generate a UA
8494					 * for other registered inits.  The
8495					 * sense code should be RESERVATIONS
8496					 * RELEASED
8497					 */
8498
8499					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8500						if (lun->pr_keys[
8501						    i + persis_offset] == 0)
8502							continue;
8503						lun->pending_ua[i] |=
8504							CTL_UA_RES_RELEASE;
8505					}
8506				}
8507				lun->res_type = 0;
8508			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8509				if (lun->pr_key_count==0) {
8510					lun->flags &= ~CTL_LUN_PR_RESERVED;
8511					lun->res_type = 0;
8512					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8513				}
8514			}
8515			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8516			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8517			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8518			persis_io.pr.pr_info.residx = residx;
8519			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8520			     &persis_io, sizeof(persis_io), 0 )) >
8521			     CTL_HA_STATUS_SUCCESS) {
8522				printf("CTL:Persis Out error returned from "
8523				       "ctl_ha_msg_send %d\n", isc_retval);
8524			}
8525		} else /* sa_res_key != 0 */ {
8526
8527			/*
8528			 * If we aren't registered currently then increment
8529			 * the key count and set the registered flag.
8530			 */
8531			if (lun->pr_keys[residx] == 0)
8532				lun->pr_key_count++;
8533			lun->pr_keys[residx] = sa_res_key;
8534
8535			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8536			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8537			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8538			persis_io.pr.pr_info.residx = residx;
8539			memcpy(persis_io.pr.pr_info.sa_res_key,
8540			       param->serv_act_res_key,
8541			       sizeof(param->serv_act_res_key));
8542			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8543			     &persis_io, sizeof(persis_io), 0)) >
8544			     CTL_HA_STATUS_SUCCESS) {
8545				printf("CTL:Persis Out error returned from "
8546				       "ctl_ha_msg_send %d\n", isc_retval);
8547			}
8548		}
8549		lun->PRGeneration++;
8550		mtx_unlock(&lun->lun_lock);
8551
8552		break;
8553	}
8554	case SPRO_RESERVE:
8555#if 0
8556                printf("Reserve executed type %d\n", type);
8557#endif
8558		mtx_lock(&lun->lun_lock);
8559		if (lun->flags & CTL_LUN_PR_RESERVED) {
8560			/*
8561			 * if this isn't the reservation holder and it's
8562			 * not a "all registrants" type or if the type is
8563			 * different then we have a conflict
8564			 */
8565			if ((lun->pr_res_idx != residx
8566			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8567			 || lun->res_type != type) {
8568				mtx_unlock(&lun->lun_lock);
8569				free(ctsio->kern_data_ptr, M_CTL);
8570				ctl_set_reservation_conflict(ctsio);
8571				ctl_done((union ctl_io *)ctsio);
8572				return (CTL_RETVAL_COMPLETE);
8573			}
8574			mtx_unlock(&lun->lun_lock);
8575		} else /* create a reservation */ {
8576			/*
8577			 * If it's not an "all registrants" type record
8578			 * reservation holder
8579			 */
8580			if (type != SPR_TYPE_WR_EX_AR
8581			 && type != SPR_TYPE_EX_AC_AR)
8582				lun->pr_res_idx = residx; /* Res holder */
8583			else
8584				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8585
8586			lun->flags |= CTL_LUN_PR_RESERVED;
8587			lun->res_type = type;
8588
8589			mtx_unlock(&lun->lun_lock);
8590
8591			/* send msg to other side */
8592			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8593			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8594			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8595			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8596			persis_io.pr.pr_info.res_type = type;
8597			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8598			     &persis_io, sizeof(persis_io), 0)) >
8599			     CTL_HA_STATUS_SUCCESS) {
8600				printf("CTL:Persis Out error returned from "
8601				       "ctl_ha_msg_send %d\n", isc_retval);
8602			}
8603		}
8604		break;
8605
8606	case SPRO_RELEASE:
8607		mtx_lock(&lun->lun_lock);
8608		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8609			/* No reservation exists return good status */
8610			mtx_unlock(&lun->lun_lock);
8611			goto done;
8612		}
8613		/*
8614		 * Is this nexus a reservation holder?
8615		 */
8616		if (lun->pr_res_idx != residx
8617		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8618			/*
8619			 * not a res holder return good status but
8620			 * do nothing
8621			 */
8622			mtx_unlock(&lun->lun_lock);
8623			goto done;
8624		}
8625
8626		if (lun->res_type != type) {
8627			mtx_unlock(&lun->lun_lock);
8628			free(ctsio->kern_data_ptr, M_CTL);
8629			ctl_set_illegal_pr_release(ctsio);
8630			ctl_done((union ctl_io *)ctsio);
8631			return (CTL_RETVAL_COMPLETE);
8632		}
8633
8634		/* okay to release */
8635		lun->flags &= ~CTL_LUN_PR_RESERVED;
8636		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8637		lun->res_type = 0;
8638
8639		/*
8640		 * if this isn't an exclusive access
8641		 * res generate UA for all other
8642		 * registrants.
8643		 */
8644		if (type != SPR_TYPE_EX_AC
8645		 && type != SPR_TYPE_WR_EX) {
8646			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8647				if (i == residx ||
8648				    lun->pr_keys[i + persis_offset] == 0)
8649					continue;
8650				lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8651			}
8652		}
8653		mtx_unlock(&lun->lun_lock);
8654		/* Send msg to other side */
8655		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8656		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8657		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8658		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8659		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8660			printf("CTL:Persis Out error returned from "
8661			       "ctl_ha_msg_send %d\n", isc_retval);
8662		}
8663		break;
8664
8665	case SPRO_CLEAR:
8666		/* send msg to other side */
8667
8668		mtx_lock(&lun->lun_lock);
8669		lun->flags &= ~CTL_LUN_PR_RESERVED;
8670		lun->res_type = 0;
8671		lun->pr_key_count = 0;
8672		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8673
8674		lun->pr_keys[residx] = 0;
8675
8676		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8677			if (lun->pr_keys[i] != 0) {
8678				if (!persis_offset && i < CTL_MAX_INITIATORS)
8679					lun->pending_ua[i] |=
8680						CTL_UA_RES_PREEMPT;
8681				else if (persis_offset && i >= persis_offset)
8682					lun->pending_ua[i-persis_offset] |=
8683					    CTL_UA_RES_PREEMPT;
8684
8685				lun->pr_keys[i] = 0;
8686			}
8687		lun->PRGeneration++;
8688		mtx_unlock(&lun->lun_lock);
8689		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8690		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8691		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8692		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8693		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8694			printf("CTL:Persis Out error returned from "
8695			       "ctl_ha_msg_send %d\n", isc_retval);
8696		}
8697		break;
8698
8699	case SPRO_PREEMPT: {
8700		int nretval;
8701
8702		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8703					  residx, ctsio, cdb, param);
8704		if (nretval != 0)
8705			return (CTL_RETVAL_COMPLETE);
8706		break;
8707	}
8708	default:
8709		panic("Invalid PR type %x", cdb->action);
8710	}
8711
8712done:
8713	free(ctsio->kern_data_ptr, M_CTL);
8714	ctl_set_success(ctsio);
8715	ctl_done((union ctl_io *)ctsio);
8716
8717	return (retval);
8718}
8719
8720/*
8721 * This routine is for handling a message from the other SC pertaining to
8722 * persistent reserve out. All the error checking will have been done
8723 * so only perorming the action need be done here to keep the two
8724 * in sync.
8725 */
8726static void
8727ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8728{
8729	struct ctl_lun *lun;
8730	struct ctl_softc *softc;
8731	int i;
8732	uint32_t targ_lun;
8733
8734	softc = control_softc;
8735
8736	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8737	lun = softc->ctl_luns[targ_lun];
8738	mtx_lock(&lun->lun_lock);
8739	switch(msg->pr.pr_info.action) {
8740	case CTL_PR_REG_KEY:
8741		if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8742			lun->pr_key_count++;
8743		lun->pr_keys[msg->pr.pr_info.residx] =
8744		    scsi_8btou64(msg->pr.pr_info.sa_res_key);
8745		lun->PRGeneration++;
8746		break;
8747
8748	case CTL_PR_UNREG_KEY:
8749		lun->pr_keys[msg->pr.pr_info.residx] = 0;
8750		lun->pr_key_count--;
8751
8752		/* XXX Need to see if the reservation has been released */
8753		/* if so do we need to generate UA? */
8754		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8755			lun->flags &= ~CTL_LUN_PR_RESERVED;
8756			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8757
8758			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8759			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8760			 && lun->pr_key_count) {
8761				/*
8762				 * If the reservation is a registrants
8763				 * only type we need to generate a UA
8764				 * for other registered inits.  The
8765				 * sense code should be RESERVATIONS
8766				 * RELEASED
8767				 */
8768
8769				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8770					if (lun->pr_keys[i+
8771					    persis_offset] == 0)
8772						continue;
8773
8774					lun->pending_ua[i] |=
8775						CTL_UA_RES_RELEASE;
8776				}
8777			}
8778			lun->res_type = 0;
8779		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8780			if (lun->pr_key_count==0) {
8781				lun->flags &= ~CTL_LUN_PR_RESERVED;
8782				lun->res_type = 0;
8783				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8784			}
8785		}
8786		lun->PRGeneration++;
8787		break;
8788
8789	case CTL_PR_RESERVE:
8790		lun->flags |= CTL_LUN_PR_RESERVED;
8791		lun->res_type = msg->pr.pr_info.res_type;
8792		lun->pr_res_idx = msg->pr.pr_info.residx;
8793
8794		break;
8795
8796	case CTL_PR_RELEASE:
8797		/*
8798		 * if this isn't an exclusive access res generate UA for all
8799		 * other registrants.
8800		 */
8801		if (lun->res_type != SPR_TYPE_EX_AC
8802		 && lun->res_type != SPR_TYPE_WR_EX) {
8803			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8804				if (lun->pr_keys[i+persis_offset] != 0)
8805					lun->pending_ua[i] |=
8806						CTL_UA_RES_RELEASE;
8807		}
8808
8809		lun->flags &= ~CTL_LUN_PR_RESERVED;
8810		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8811		lun->res_type = 0;
8812		break;
8813
8814	case CTL_PR_PREEMPT:
8815		ctl_pro_preempt_other(lun, msg);
8816		break;
8817	case CTL_PR_CLEAR:
8818		lun->flags &= ~CTL_LUN_PR_RESERVED;
8819		lun->res_type = 0;
8820		lun->pr_key_count = 0;
8821		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8822
8823		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8824			if (lun->pr_keys[i] == 0)
8825				continue;
8826			if (!persis_offset
8827			 && i < CTL_MAX_INITIATORS)
8828				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8829			else if (persis_offset
8830			      && i >= persis_offset)
8831				lun->pending_ua[i-persis_offset] |=
8832					CTL_UA_RES_PREEMPT;
8833			lun->pr_keys[i] = 0;
8834		}
8835		lun->PRGeneration++;
8836		break;
8837	}
8838
8839	mtx_unlock(&lun->lun_lock);
8840}
8841
8842int
8843ctl_read_write(struct ctl_scsiio *ctsio)
8844{
8845	struct ctl_lun *lun;
8846	struct ctl_lba_len_flags *lbalen;
8847	uint64_t lba;
8848	uint32_t num_blocks;
8849	int flags, retval;
8850	int isread;
8851
8852	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8853
8854	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8855
8856	flags = 0;
8857	retval = CTL_RETVAL_COMPLETE;
8858
8859	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8860	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8861	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8862		uint32_t residx;
8863
8864		/*
8865		 * XXX KDM need a lock here.
8866		 */
8867		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8868		if ((lun->res_type == SPR_TYPE_EX_AC
8869		  && residx != lun->pr_res_idx)
8870		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8871		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8872		  && lun->pr_keys[residx] == 0)) {
8873			ctl_set_reservation_conflict(ctsio);
8874			ctl_done((union ctl_io *)ctsio);
8875			return (CTL_RETVAL_COMPLETE);
8876	        }
8877	}
8878
8879	switch (ctsio->cdb[0]) {
8880	case READ_6:
8881	case WRITE_6: {
8882		struct scsi_rw_6 *cdb;
8883
8884		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8885
8886		lba = scsi_3btoul(cdb->addr);
8887		/* only 5 bits are valid in the most significant address byte */
8888		lba &= 0x1fffff;
8889		num_blocks = cdb->length;
8890		/*
8891		 * This is correct according to SBC-2.
8892		 */
8893		if (num_blocks == 0)
8894			num_blocks = 256;
8895		break;
8896	}
8897	case READ_10:
8898	case WRITE_10: {
8899		struct scsi_rw_10 *cdb;
8900
8901		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8902		if (cdb->byte2 & SRW10_FUA)
8903			flags |= CTL_LLF_FUA;
8904		if (cdb->byte2 & SRW10_DPO)
8905			flags |= CTL_LLF_DPO;
8906		lba = scsi_4btoul(cdb->addr);
8907		num_blocks = scsi_2btoul(cdb->length);
8908		break;
8909	}
8910	case WRITE_VERIFY_10: {
8911		struct scsi_write_verify_10 *cdb;
8912
8913		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8914		flags |= CTL_LLF_FUA;
8915		if (cdb->byte2 & SWV_DPO)
8916			flags |= CTL_LLF_DPO;
8917		lba = scsi_4btoul(cdb->addr);
8918		num_blocks = scsi_2btoul(cdb->length);
8919		break;
8920	}
8921	case READ_12:
8922	case WRITE_12: {
8923		struct scsi_rw_12 *cdb;
8924
8925		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8926		if (cdb->byte2 & SRW12_FUA)
8927			flags |= CTL_LLF_FUA;
8928		if (cdb->byte2 & SRW12_DPO)
8929			flags |= CTL_LLF_DPO;
8930		lba = scsi_4btoul(cdb->addr);
8931		num_blocks = scsi_4btoul(cdb->length);
8932		break;
8933	}
8934	case WRITE_VERIFY_12: {
8935		struct scsi_write_verify_12 *cdb;
8936
8937		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8938		flags |= CTL_LLF_FUA;
8939		if (cdb->byte2 & SWV_DPO)
8940			flags |= CTL_LLF_DPO;
8941		lba = scsi_4btoul(cdb->addr);
8942		num_blocks = scsi_4btoul(cdb->length);
8943		break;
8944	}
8945	case READ_16:
8946	case WRITE_16: {
8947		struct scsi_rw_16 *cdb;
8948
8949		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8950		if (cdb->byte2 & SRW12_FUA)
8951			flags |= CTL_LLF_FUA;
8952		if (cdb->byte2 & SRW12_DPO)
8953			flags |= CTL_LLF_DPO;
8954		lba = scsi_8btou64(cdb->addr);
8955		num_blocks = scsi_4btoul(cdb->length);
8956		break;
8957	}
8958	case WRITE_ATOMIC_16: {
8959		struct scsi_rw_16 *cdb;
8960
8961		if (lun->be_lun->atomicblock == 0) {
8962			ctl_set_invalid_opcode(ctsio);
8963			ctl_done((union ctl_io *)ctsio);
8964			return (CTL_RETVAL_COMPLETE);
8965		}
8966
8967		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8968		if (cdb->byte2 & SRW12_FUA)
8969			flags |= CTL_LLF_FUA;
8970		if (cdb->byte2 & SRW12_DPO)
8971			flags |= CTL_LLF_DPO;
8972		lba = scsi_8btou64(cdb->addr);
8973		num_blocks = scsi_4btoul(cdb->length);
8974		if (num_blocks > lun->be_lun->atomicblock) {
8975			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8976			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8977			    /*bit*/ 0);
8978			ctl_done((union ctl_io *)ctsio);
8979			return (CTL_RETVAL_COMPLETE);
8980		}
8981		break;
8982	}
8983	case WRITE_VERIFY_16: {
8984		struct scsi_write_verify_16 *cdb;
8985
8986		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8987		flags |= CTL_LLF_FUA;
8988		if (cdb->byte2 & SWV_DPO)
8989			flags |= CTL_LLF_DPO;
8990		lba = scsi_8btou64(cdb->addr);
8991		num_blocks = scsi_4btoul(cdb->length);
8992		break;
8993	}
8994	default:
8995		/*
8996		 * We got a command we don't support.  This shouldn't
8997		 * happen, commands should be filtered out above us.
8998		 */
8999		ctl_set_invalid_opcode(ctsio);
9000		ctl_done((union ctl_io *)ctsio);
9001
9002		return (CTL_RETVAL_COMPLETE);
9003		break; /* NOTREACHED */
9004	}
9005
9006	/*
9007	 * The first check is to make sure we're in bounds, the second
9008	 * check is to catch wrap-around problems.  If the lba + num blocks
9009	 * is less than the lba, then we've wrapped around and the block
9010	 * range is invalid anyway.
9011	 */
9012	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9013	 || ((lba + num_blocks) < lba)) {
9014		ctl_set_lba_out_of_range(ctsio);
9015		ctl_done((union ctl_io *)ctsio);
9016		return (CTL_RETVAL_COMPLETE);
9017	}
9018
9019	/*
9020	 * According to SBC-3, a transfer length of 0 is not an error.
9021	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9022	 * translates to 256 blocks for those commands.
9023	 */
9024	if (num_blocks == 0) {
9025		ctl_set_success(ctsio);
9026		ctl_done((union ctl_io *)ctsio);
9027		return (CTL_RETVAL_COMPLETE);
9028	}
9029
9030	/* Set FUA and/or DPO if caches are disabled. */
9031	if (isread) {
9032		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9033		    SCP_RCD) != 0)
9034			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9035	} else {
9036		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9037		    SCP_WCE) == 0)
9038			flags |= CTL_LLF_FUA;
9039	}
9040
9041	lbalen = (struct ctl_lba_len_flags *)
9042	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9043	lbalen->lba = lba;
9044	lbalen->len = num_blocks;
9045	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9046
9047	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9048	ctsio->kern_rel_offset = 0;
9049
9050	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9051
9052	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9053
9054	return (retval);
9055}
9056
9057static int
9058ctl_cnw_cont(union ctl_io *io)
9059{
9060	struct ctl_scsiio *ctsio;
9061	struct ctl_lun *lun;
9062	struct ctl_lba_len_flags *lbalen;
9063	int retval;
9064
9065	ctsio = &io->scsiio;
9066	ctsio->io_hdr.status = CTL_STATUS_NONE;
9067	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9068	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9069	lbalen = (struct ctl_lba_len_flags *)
9070	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9071	lbalen->flags &= ~CTL_LLF_COMPARE;
9072	lbalen->flags |= CTL_LLF_WRITE;
9073
9074	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9075	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9076	return (retval);
9077}
9078
9079int
9080ctl_cnw(struct ctl_scsiio *ctsio)
9081{
9082	struct ctl_lun *lun;
9083	struct ctl_lba_len_flags *lbalen;
9084	uint64_t lba;
9085	uint32_t num_blocks;
9086	int flags, retval;
9087
9088	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9089
9090	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9091
9092	flags = 0;
9093	retval = CTL_RETVAL_COMPLETE;
9094
9095	switch (ctsio->cdb[0]) {
9096	case COMPARE_AND_WRITE: {
9097		struct scsi_compare_and_write *cdb;
9098
9099		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9100		if (cdb->byte2 & SRW10_FUA)
9101			flags |= CTL_LLF_FUA;
9102		if (cdb->byte2 & SRW10_DPO)
9103			flags |= CTL_LLF_DPO;
9104		lba = scsi_8btou64(cdb->addr);
9105		num_blocks = cdb->length;
9106		break;
9107	}
9108	default:
9109		/*
9110		 * We got a command we don't support.  This shouldn't
9111		 * happen, commands should be filtered out above us.
9112		 */
9113		ctl_set_invalid_opcode(ctsio);
9114		ctl_done((union ctl_io *)ctsio);
9115
9116		return (CTL_RETVAL_COMPLETE);
9117		break; /* NOTREACHED */
9118	}
9119
9120	/*
9121	 * The first check is to make sure we're in bounds, the second
9122	 * check is to catch wrap-around problems.  If the lba + num blocks
9123	 * is less than the lba, then we've wrapped around and the block
9124	 * range is invalid anyway.
9125	 */
9126	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9127	 || ((lba + num_blocks) < lba)) {
9128		ctl_set_lba_out_of_range(ctsio);
9129		ctl_done((union ctl_io *)ctsio);
9130		return (CTL_RETVAL_COMPLETE);
9131	}
9132
9133	/*
9134	 * According to SBC-3, a transfer length of 0 is not an error.
9135	 */
9136	if (num_blocks == 0) {
9137		ctl_set_success(ctsio);
9138		ctl_done((union ctl_io *)ctsio);
9139		return (CTL_RETVAL_COMPLETE);
9140	}
9141
9142	/* Set FUA if write cache is disabled. */
9143	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9144	    SCP_WCE) == 0)
9145		flags |= CTL_LLF_FUA;
9146
9147	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9148	ctsio->kern_rel_offset = 0;
9149
9150	/*
9151	 * Set the IO_CONT flag, so that if this I/O gets passed to
9152	 * ctl_data_submit_done(), it'll get passed back to
9153	 * ctl_ctl_cnw_cont() for further processing.
9154	 */
9155	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9156	ctsio->io_cont = ctl_cnw_cont;
9157
9158	lbalen = (struct ctl_lba_len_flags *)
9159	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9160	lbalen->lba = lba;
9161	lbalen->len = num_blocks;
9162	lbalen->flags = CTL_LLF_COMPARE | flags;
9163
9164	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9165	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9166	return (retval);
9167}
9168
9169int
9170ctl_verify(struct ctl_scsiio *ctsio)
9171{
9172	struct ctl_lun *lun;
9173	struct ctl_lba_len_flags *lbalen;
9174	uint64_t lba;
9175	uint32_t num_blocks;
9176	int bytchk, flags;
9177	int retval;
9178
9179	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9180
9181	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9182
9183	bytchk = 0;
9184	flags = CTL_LLF_FUA;
9185	retval = CTL_RETVAL_COMPLETE;
9186
9187	switch (ctsio->cdb[0]) {
9188	case VERIFY_10: {
9189		struct scsi_verify_10 *cdb;
9190
9191		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9192		if (cdb->byte2 & SVFY_BYTCHK)
9193			bytchk = 1;
9194		if (cdb->byte2 & SVFY_DPO)
9195			flags |= CTL_LLF_DPO;
9196		lba = scsi_4btoul(cdb->addr);
9197		num_blocks = scsi_2btoul(cdb->length);
9198		break;
9199	}
9200	case VERIFY_12: {
9201		struct scsi_verify_12 *cdb;
9202
9203		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9204		if (cdb->byte2 & SVFY_BYTCHK)
9205			bytchk = 1;
9206		if (cdb->byte2 & SVFY_DPO)
9207			flags |= CTL_LLF_DPO;
9208		lba = scsi_4btoul(cdb->addr);
9209		num_blocks = scsi_4btoul(cdb->length);
9210		break;
9211	}
9212	case VERIFY_16: {
9213		struct scsi_rw_16 *cdb;
9214
9215		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9216		if (cdb->byte2 & SVFY_BYTCHK)
9217			bytchk = 1;
9218		if (cdb->byte2 & SVFY_DPO)
9219			flags |= CTL_LLF_DPO;
9220		lba = scsi_8btou64(cdb->addr);
9221		num_blocks = scsi_4btoul(cdb->length);
9222		break;
9223	}
9224	default:
9225		/*
9226		 * We got a command we don't support.  This shouldn't
9227		 * happen, commands should be filtered out above us.
9228		 */
9229		ctl_set_invalid_opcode(ctsio);
9230		ctl_done((union ctl_io *)ctsio);
9231		return (CTL_RETVAL_COMPLETE);
9232	}
9233
9234	/*
9235	 * The first check is to make sure we're in bounds, the second
9236	 * check is to catch wrap-around problems.  If the lba + num blocks
9237	 * is less than the lba, then we've wrapped around and the block
9238	 * range is invalid anyway.
9239	 */
9240	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9241	 || ((lba + num_blocks) < lba)) {
9242		ctl_set_lba_out_of_range(ctsio);
9243		ctl_done((union ctl_io *)ctsio);
9244		return (CTL_RETVAL_COMPLETE);
9245	}
9246
9247	/*
9248	 * According to SBC-3, a transfer length of 0 is not an error.
9249	 */
9250	if (num_blocks == 0) {
9251		ctl_set_success(ctsio);
9252		ctl_done((union ctl_io *)ctsio);
9253		return (CTL_RETVAL_COMPLETE);
9254	}
9255
9256	lbalen = (struct ctl_lba_len_flags *)
9257	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9258	lbalen->lba = lba;
9259	lbalen->len = num_blocks;
9260	if (bytchk) {
9261		lbalen->flags = CTL_LLF_COMPARE | flags;
9262		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9263	} else {
9264		lbalen->flags = CTL_LLF_VERIFY | flags;
9265		ctsio->kern_total_len = 0;
9266	}
9267	ctsio->kern_rel_offset = 0;
9268
9269	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9270	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9271	return (retval);
9272}
9273
9274int
9275ctl_report_luns(struct ctl_scsiio *ctsio)
9276{
9277	struct scsi_report_luns *cdb;
9278	struct scsi_report_luns_data *lun_data;
9279	struct ctl_lun *lun, *request_lun;
9280	int num_luns, retval;
9281	uint32_t alloc_len, lun_datalen;
9282	int num_filled, well_known;
9283	uint32_t initidx, targ_lun_id, lun_id;
9284
9285	retval = CTL_RETVAL_COMPLETE;
9286	well_known = 0;
9287
9288	cdb = (struct scsi_report_luns *)ctsio->cdb;
9289
9290	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9291
9292	mtx_lock(&control_softc->ctl_lock);
9293	num_luns = control_softc->num_luns;
9294	mtx_unlock(&control_softc->ctl_lock);
9295
9296	switch (cdb->select_report) {
9297	case RPL_REPORT_DEFAULT:
9298	case RPL_REPORT_ALL:
9299		break;
9300	case RPL_REPORT_WELLKNOWN:
9301		well_known = 1;
9302		num_luns = 0;
9303		break;
9304	default:
9305		ctl_set_invalid_field(ctsio,
9306				      /*sks_valid*/ 1,
9307				      /*command*/ 1,
9308				      /*field*/ 2,
9309				      /*bit_valid*/ 0,
9310				      /*bit*/ 0);
9311		ctl_done((union ctl_io *)ctsio);
9312		return (retval);
9313		break; /* NOTREACHED */
9314	}
9315
9316	alloc_len = scsi_4btoul(cdb->length);
9317	/*
9318	 * The initiator has to allocate at least 16 bytes for this request,
9319	 * so he can at least get the header and the first LUN.  Otherwise
9320	 * we reject the request (per SPC-3 rev 14, section 6.21).
9321	 */
9322	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9323	    sizeof(struct scsi_report_luns_lundata))) {
9324		ctl_set_invalid_field(ctsio,
9325				      /*sks_valid*/ 1,
9326				      /*command*/ 1,
9327				      /*field*/ 6,
9328				      /*bit_valid*/ 0,
9329				      /*bit*/ 0);
9330		ctl_done((union ctl_io *)ctsio);
9331		return (retval);
9332	}
9333
9334	request_lun = (struct ctl_lun *)
9335		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9336
9337	lun_datalen = sizeof(*lun_data) +
9338		(num_luns * sizeof(struct scsi_report_luns_lundata));
9339
9340	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9341	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9342	ctsio->kern_sg_entries = 0;
9343
9344	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9345
9346	mtx_lock(&control_softc->ctl_lock);
9347	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9348		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9349		if (lun_id >= CTL_MAX_LUNS)
9350			continue;
9351		lun = control_softc->ctl_luns[lun_id];
9352		if (lun == NULL)
9353			continue;
9354
9355		if (targ_lun_id <= 0xff) {
9356			/*
9357			 * Peripheral addressing method, bus number 0.
9358			 */
9359			lun_data->luns[num_filled].lundata[0] =
9360				RPL_LUNDATA_ATYP_PERIPH;
9361			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9362			num_filled++;
9363		} else if (targ_lun_id <= 0x3fff) {
9364			/*
9365			 * Flat addressing method.
9366			 */
9367			lun_data->luns[num_filled].lundata[0] =
9368				RPL_LUNDATA_ATYP_FLAT |
9369				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9370#ifdef OLDCTLHEADERS
9371				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9372				(targ_lun_id & SRLD_BUS_LUN_MASK);
9373#endif
9374			lun_data->luns[num_filled].lundata[1] =
9375#ifdef OLDCTLHEADERS
9376				targ_lun_id >> SRLD_BUS_LUN_BITS;
9377#endif
9378				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9379			num_filled++;
9380		} else {
9381			printf("ctl_report_luns: bogus LUN number %jd, "
9382			       "skipping\n", (intmax_t)targ_lun_id);
9383		}
9384		/*
9385		 * According to SPC-3, rev 14 section 6.21:
9386		 *
9387		 * "The execution of a REPORT LUNS command to any valid and
9388		 * installed logical unit shall clear the REPORTED LUNS DATA
9389		 * HAS CHANGED unit attention condition for all logical
9390		 * units of that target with respect to the requesting
9391		 * initiator. A valid and installed logical unit is one
9392		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9393		 * INQUIRY data (see 6.4.2)."
9394		 *
9395		 * If request_lun is NULL, the LUN this report luns command
9396		 * was issued to is either disabled or doesn't exist. In that
9397		 * case, we shouldn't clear any pending lun change unit
9398		 * attention.
9399		 */
9400		if (request_lun != NULL) {
9401			mtx_lock(&lun->lun_lock);
9402			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9403			mtx_unlock(&lun->lun_lock);
9404		}
9405	}
9406	mtx_unlock(&control_softc->ctl_lock);
9407
9408	/*
9409	 * It's quite possible that we've returned fewer LUNs than we allocated
9410	 * space for.  Trim it.
9411	 */
9412	lun_datalen = sizeof(*lun_data) +
9413		(num_filled * sizeof(struct scsi_report_luns_lundata));
9414
9415	if (lun_datalen < alloc_len) {
9416		ctsio->residual = alloc_len - lun_datalen;
9417		ctsio->kern_data_len = lun_datalen;
9418		ctsio->kern_total_len = lun_datalen;
9419	} else {
9420		ctsio->residual = 0;
9421		ctsio->kern_data_len = alloc_len;
9422		ctsio->kern_total_len = alloc_len;
9423	}
9424	ctsio->kern_data_resid = 0;
9425	ctsio->kern_rel_offset = 0;
9426	ctsio->kern_sg_entries = 0;
9427
9428	/*
9429	 * We set this to the actual data length, regardless of how much
9430	 * space we actually have to return results.  If the user looks at
9431	 * this value, he'll know whether or not he allocated enough space
9432	 * and reissue the command if necessary.  We don't support well
9433	 * known logical units, so if the user asks for that, return none.
9434	 */
9435	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9436
9437	/*
9438	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9439	 * this request.
9440	 */
9441	ctsio->scsi_status = SCSI_STATUS_OK;
9442
9443	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9444	ctsio->be_move_done = ctl_config_move_done;
9445	ctl_datamove((union ctl_io *)ctsio);
9446
9447	return (retval);
9448}
9449
9450int
9451ctl_request_sense(struct ctl_scsiio *ctsio)
9452{
9453	struct scsi_request_sense *cdb;
9454	struct scsi_sense_data *sense_ptr;
9455	struct ctl_lun *lun;
9456	uint32_t initidx;
9457	int have_error;
9458	scsi_sense_data_type sense_format;
9459
9460	cdb = (struct scsi_request_sense *)ctsio->cdb;
9461
9462	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9463
9464	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9465
9466	/*
9467	 * Determine which sense format the user wants.
9468	 */
9469	if (cdb->byte2 & SRS_DESC)
9470		sense_format = SSD_TYPE_DESC;
9471	else
9472		sense_format = SSD_TYPE_FIXED;
9473
9474	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9475	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9476	ctsio->kern_sg_entries = 0;
9477
9478	/*
9479	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9480	 * larger than the largest allowed value for the length field in the
9481	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9482	 */
9483	ctsio->residual = 0;
9484	ctsio->kern_data_len = cdb->length;
9485	ctsio->kern_total_len = cdb->length;
9486
9487	ctsio->kern_data_resid = 0;
9488	ctsio->kern_rel_offset = 0;
9489	ctsio->kern_sg_entries = 0;
9490
9491	/*
9492	 * If we don't have a LUN, we don't have any pending sense.
9493	 */
9494	if (lun == NULL)
9495		goto no_sense;
9496
9497	have_error = 0;
9498	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9499	/*
9500	 * Check for pending sense, and then for pending unit attentions.
9501	 * Pending sense gets returned first, then pending unit attentions.
9502	 */
9503	mtx_lock(&lun->lun_lock);
9504#ifdef CTL_WITH_CA
9505	if (ctl_is_set(lun->have_ca, initidx)) {
9506		scsi_sense_data_type stored_format;
9507
9508		/*
9509		 * Check to see which sense format was used for the stored
9510		 * sense data.
9511		 */
9512		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9513
9514		/*
9515		 * If the user requested a different sense format than the
9516		 * one we stored, then we need to convert it to the other
9517		 * format.  If we're going from descriptor to fixed format
9518		 * sense data, we may lose things in translation, depending
9519		 * on what options were used.
9520		 *
9521		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9522		 * for some reason we'll just copy it out as-is.
9523		 */
9524		if ((stored_format == SSD_TYPE_FIXED)
9525		 && (sense_format == SSD_TYPE_DESC))
9526			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9527			    &lun->pending_sense[initidx],
9528			    (struct scsi_sense_data_desc *)sense_ptr);
9529		else if ((stored_format == SSD_TYPE_DESC)
9530		      && (sense_format == SSD_TYPE_FIXED))
9531			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9532			    &lun->pending_sense[initidx],
9533			    (struct scsi_sense_data_fixed *)sense_ptr);
9534		else
9535			memcpy(sense_ptr, &lun->pending_sense[initidx],
9536			       ctl_min(sizeof(*sense_ptr),
9537			       sizeof(lun->pending_sense[initidx])));
9538
9539		ctl_clear_mask(lun->have_ca, initidx);
9540		have_error = 1;
9541	} else
9542#endif
9543	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9544		ctl_ua_type ua_type;
9545
9546		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9547				       sense_ptr, sense_format);
9548		if (ua_type != CTL_UA_NONE)
9549			have_error = 1;
9550	}
9551	mtx_unlock(&lun->lun_lock);
9552
9553	/*
9554	 * We already have a pending error, return it.
9555	 */
9556	if (have_error != 0) {
9557		/*
9558		 * We report the SCSI status as OK, since the status of the
9559		 * request sense command itself is OK.
9560		 */
9561		ctsio->scsi_status = SCSI_STATUS_OK;
9562
9563		/*
9564		 * We report 0 for the sense length, because we aren't doing
9565		 * autosense in this case.  We're reporting sense as
9566		 * parameter data.
9567		 */
9568		ctsio->sense_len = 0;
9569		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9570		ctsio->be_move_done = ctl_config_move_done;
9571		ctl_datamove((union ctl_io *)ctsio);
9572
9573		return (CTL_RETVAL_COMPLETE);
9574	}
9575
9576no_sense:
9577
9578	/*
9579	 * No sense information to report, so we report that everything is
9580	 * okay.
9581	 */
9582	ctl_set_sense_data(sense_ptr,
9583			   lun,
9584			   sense_format,
9585			   /*current_error*/ 1,
9586			   /*sense_key*/ SSD_KEY_NO_SENSE,
9587			   /*asc*/ 0x00,
9588			   /*ascq*/ 0x00,
9589			   SSD_ELEM_NONE);
9590
9591	ctsio->scsi_status = SCSI_STATUS_OK;
9592
9593	/*
9594	 * We report 0 for the sense length, because we aren't doing
9595	 * autosense in this case.  We're reporting sense as parameter data.
9596	 */
9597	ctsio->sense_len = 0;
9598	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9599	ctsio->be_move_done = ctl_config_move_done;
9600	ctl_datamove((union ctl_io *)ctsio);
9601
9602	return (CTL_RETVAL_COMPLETE);
9603}
9604
9605int
9606ctl_tur(struct ctl_scsiio *ctsio)
9607{
9608	struct ctl_lun *lun;
9609
9610	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9611
9612	CTL_DEBUG_PRINT(("ctl_tur\n"));
9613
9614	if (lun == NULL)
9615		return (EINVAL);
9616
9617	ctsio->scsi_status = SCSI_STATUS_OK;
9618	ctsio->io_hdr.status = CTL_SUCCESS;
9619
9620	ctl_done((union ctl_io *)ctsio);
9621
9622	return (CTL_RETVAL_COMPLETE);
9623}
9624
9625#ifdef notyet
9626static int
9627ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9628{
9629
9630}
9631#endif
9632
9633static int
9634ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9635{
9636	struct scsi_vpd_supported_pages *pages;
9637	int sup_page_size;
9638	struct ctl_lun *lun;
9639
9640	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9641
9642	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9643	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9644	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9645	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9646	ctsio->kern_sg_entries = 0;
9647
9648	if (sup_page_size < alloc_len) {
9649		ctsio->residual = alloc_len - sup_page_size;
9650		ctsio->kern_data_len = sup_page_size;
9651		ctsio->kern_total_len = sup_page_size;
9652	} else {
9653		ctsio->residual = 0;
9654		ctsio->kern_data_len = alloc_len;
9655		ctsio->kern_total_len = alloc_len;
9656	}
9657	ctsio->kern_data_resid = 0;
9658	ctsio->kern_rel_offset = 0;
9659	ctsio->kern_sg_entries = 0;
9660
9661	/*
9662	 * The control device is always connected.  The disk device, on the
9663	 * other hand, may not be online all the time.  Need to change this
9664	 * to figure out whether the disk device is actually online or not.
9665	 */
9666	if (lun != NULL)
9667		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9668				lun->be_lun->lun_type;
9669	else
9670		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9671
9672	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9673	/* Supported VPD pages */
9674	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9675	/* Serial Number */
9676	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9677	/* Device Identification */
9678	pages->page_list[2] = SVPD_DEVICE_ID;
9679	/* Extended INQUIRY Data */
9680	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9681	/* Mode Page Policy */
9682	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9683	/* SCSI Ports */
9684	pages->page_list[5] = SVPD_SCSI_PORTS;
9685	/* Third-party Copy */
9686	pages->page_list[6] = SVPD_SCSI_TPC;
9687	/* Block limits */
9688	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9689	/* Block Device Characteristics */
9690	pages->page_list[8] = SVPD_BDC;
9691	/* Logical Block Provisioning */
9692	pages->page_list[9] = SVPD_LBP;
9693
9694	ctsio->scsi_status = SCSI_STATUS_OK;
9695
9696	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9697	ctsio->be_move_done = ctl_config_move_done;
9698	ctl_datamove((union ctl_io *)ctsio);
9699
9700	return (CTL_RETVAL_COMPLETE);
9701}
9702
9703static int
9704ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9705{
9706	struct scsi_vpd_unit_serial_number *sn_ptr;
9707	struct ctl_lun *lun;
9708
9709	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9710
9711	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9712	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9713	ctsio->kern_sg_entries = 0;
9714
9715	if (sizeof(*sn_ptr) < alloc_len) {
9716		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9717		ctsio->kern_data_len = sizeof(*sn_ptr);
9718		ctsio->kern_total_len = sizeof(*sn_ptr);
9719	} else {
9720		ctsio->residual = 0;
9721		ctsio->kern_data_len = alloc_len;
9722		ctsio->kern_total_len = alloc_len;
9723	}
9724	ctsio->kern_data_resid = 0;
9725	ctsio->kern_rel_offset = 0;
9726	ctsio->kern_sg_entries = 0;
9727
9728	/*
9729	 * The control device is always connected.  The disk device, on the
9730	 * other hand, may not be online all the time.  Need to change this
9731	 * to figure out whether the disk device is actually online or not.
9732	 */
9733	if (lun != NULL)
9734		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9735				  lun->be_lun->lun_type;
9736	else
9737		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9738
9739	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9740	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9741	/*
9742	 * If we don't have a LUN, we just leave the serial number as
9743	 * all spaces.
9744	 */
9745	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9746	if (lun != NULL) {
9747		strncpy((char *)sn_ptr->serial_num,
9748			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9749	}
9750	ctsio->scsi_status = SCSI_STATUS_OK;
9751
9752	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9753	ctsio->be_move_done = ctl_config_move_done;
9754	ctl_datamove((union ctl_io *)ctsio);
9755
9756	return (CTL_RETVAL_COMPLETE);
9757}
9758
9759
9760static int
9761ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9762{
9763	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9764	struct ctl_lun *lun;
9765	int data_len;
9766
9767	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9768
9769	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9770	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9771	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9772	ctsio->kern_sg_entries = 0;
9773
9774	if (data_len < alloc_len) {
9775		ctsio->residual = alloc_len - data_len;
9776		ctsio->kern_data_len = data_len;
9777		ctsio->kern_total_len = data_len;
9778	} else {
9779		ctsio->residual = 0;
9780		ctsio->kern_data_len = alloc_len;
9781		ctsio->kern_total_len = alloc_len;
9782	}
9783	ctsio->kern_data_resid = 0;
9784	ctsio->kern_rel_offset = 0;
9785	ctsio->kern_sg_entries = 0;
9786
9787	/*
9788	 * The control device is always connected.  The disk device, on the
9789	 * other hand, may not be online all the time.
9790	 */
9791	if (lun != NULL)
9792		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9793				     lun->be_lun->lun_type;
9794	else
9795		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9796	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9797	eid_ptr->page_length = data_len - 4;
9798	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9799	eid_ptr->flags3 = SVPD_EID_V_SUP;
9800
9801	ctsio->scsi_status = SCSI_STATUS_OK;
9802	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9803	ctsio->be_move_done = ctl_config_move_done;
9804	ctl_datamove((union ctl_io *)ctsio);
9805
9806	return (CTL_RETVAL_COMPLETE);
9807}
9808
9809static int
9810ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9811{
9812	struct scsi_vpd_mode_page_policy *mpp_ptr;
9813	struct ctl_lun *lun;
9814	int data_len;
9815
9816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9817
9818	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9819	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9820
9821	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9822	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9823	ctsio->kern_sg_entries = 0;
9824
9825	if (data_len < alloc_len) {
9826		ctsio->residual = alloc_len - data_len;
9827		ctsio->kern_data_len = data_len;
9828		ctsio->kern_total_len = data_len;
9829	} else {
9830		ctsio->residual = 0;
9831		ctsio->kern_data_len = alloc_len;
9832		ctsio->kern_total_len = alloc_len;
9833	}
9834	ctsio->kern_data_resid = 0;
9835	ctsio->kern_rel_offset = 0;
9836	ctsio->kern_sg_entries = 0;
9837
9838	/*
9839	 * The control device is always connected.  The disk device, on the
9840	 * other hand, may not be online all the time.
9841	 */
9842	if (lun != NULL)
9843		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9844				     lun->be_lun->lun_type;
9845	else
9846		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9847	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9848	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9849	mpp_ptr->descr[0].page_code = 0x3f;
9850	mpp_ptr->descr[0].subpage_code = 0xff;
9851	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9852
9853	ctsio->scsi_status = SCSI_STATUS_OK;
9854	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9855	ctsio->be_move_done = ctl_config_move_done;
9856	ctl_datamove((union ctl_io *)ctsio);
9857
9858	return (CTL_RETVAL_COMPLETE);
9859}
9860
9861static int
9862ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9863{
9864	struct scsi_vpd_device_id *devid_ptr;
9865	struct scsi_vpd_id_descriptor *desc;
9866	struct ctl_softc *ctl_softc;
9867	struct ctl_lun *lun;
9868	struct ctl_port *port;
9869	int data_len;
9870	uint8_t proto;
9871
9872	ctl_softc = control_softc;
9873
9874	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9875	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9876
9877	data_len = sizeof(struct scsi_vpd_device_id) +
9878	    sizeof(struct scsi_vpd_id_descriptor) +
9879		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9880	    sizeof(struct scsi_vpd_id_descriptor) +
9881		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9882	if (lun && lun->lun_devid)
9883		data_len += lun->lun_devid->len;
9884	if (port->port_devid)
9885		data_len += port->port_devid->len;
9886	if (port->target_devid)
9887		data_len += port->target_devid->len;
9888
9889	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9890	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9891	ctsio->kern_sg_entries = 0;
9892
9893	if (data_len < alloc_len) {
9894		ctsio->residual = alloc_len - data_len;
9895		ctsio->kern_data_len = data_len;
9896		ctsio->kern_total_len = data_len;
9897	} else {
9898		ctsio->residual = 0;
9899		ctsio->kern_data_len = alloc_len;
9900		ctsio->kern_total_len = alloc_len;
9901	}
9902	ctsio->kern_data_resid = 0;
9903	ctsio->kern_rel_offset = 0;
9904	ctsio->kern_sg_entries = 0;
9905
9906	/*
9907	 * The control device is always connected.  The disk device, on the
9908	 * other hand, may not be online all the time.
9909	 */
9910	if (lun != NULL)
9911		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9912				     lun->be_lun->lun_type;
9913	else
9914		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9915	devid_ptr->page_code = SVPD_DEVICE_ID;
9916	scsi_ulto2b(data_len - 4, devid_ptr->length);
9917
9918	if (port->port_type == CTL_PORT_FC)
9919		proto = SCSI_PROTO_FC << 4;
9920	else if (port->port_type == CTL_PORT_ISCSI)
9921		proto = SCSI_PROTO_ISCSI << 4;
9922	else
9923		proto = SCSI_PROTO_SPI << 4;
9924	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9925
9926	/*
9927	 * We're using a LUN association here.  i.e., this device ID is a
9928	 * per-LUN identifier.
9929	 */
9930	if (lun && lun->lun_devid) {
9931		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9932		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9933		    lun->lun_devid->len);
9934	}
9935
9936	/*
9937	 * This is for the WWPN which is a port association.
9938	 */
9939	if (port->port_devid) {
9940		memcpy(desc, port->port_devid->data, port->port_devid->len);
9941		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9942		    port->port_devid->len);
9943	}
9944
9945	/*
9946	 * This is for the Relative Target Port(type 4h) identifier
9947	 */
9948	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9949	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9950	    SVPD_ID_TYPE_RELTARG;
9951	desc->length = 4;
9952	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9953	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9954	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9955
9956	/*
9957	 * This is for the Target Port Group(type 5h) identifier
9958	 */
9959	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9960	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9961	    SVPD_ID_TYPE_TPORTGRP;
9962	desc->length = 4;
9963	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9964	    &desc->identifier[2]);
9965	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9966	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9967
9968	/*
9969	 * This is for the Target identifier
9970	 */
9971	if (port->target_devid) {
9972		memcpy(desc, port->target_devid->data, port->target_devid->len);
9973	}
9974
9975	ctsio->scsi_status = SCSI_STATUS_OK;
9976	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9977	ctsio->be_move_done = ctl_config_move_done;
9978	ctl_datamove((union ctl_io *)ctsio);
9979
9980	return (CTL_RETVAL_COMPLETE);
9981}
9982
9983static int
9984ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9985{
9986	struct ctl_softc *softc = control_softc;
9987	struct scsi_vpd_scsi_ports *sp;
9988	struct scsi_vpd_port_designation *pd;
9989	struct scsi_vpd_port_designation_cont *pdc;
9990	struct ctl_lun *lun;
9991	struct ctl_port *port;
9992	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9993	int num_target_port_groups, single;
9994
9995	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9996
9997	single = ctl_is_single;
9998	if (single)
9999		num_target_port_groups = 1;
10000	else
10001		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10002	num_target_ports = 0;
10003	iid_len = 0;
10004	id_len = 0;
10005	mtx_lock(&softc->ctl_lock);
10006	STAILQ_FOREACH(port, &softc->port_list, links) {
10007		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10008			continue;
10009		if (lun != NULL &&
10010		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10011		    CTL_MAX_LUNS)
10012			continue;
10013		num_target_ports++;
10014		if (port->init_devid)
10015			iid_len += port->init_devid->len;
10016		if (port->port_devid)
10017			id_len += port->port_devid->len;
10018	}
10019	mtx_unlock(&softc->ctl_lock);
10020
10021	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10022	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10023	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10024	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10025	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10026	ctsio->kern_sg_entries = 0;
10027
10028	if (data_len < alloc_len) {
10029		ctsio->residual = alloc_len - data_len;
10030		ctsio->kern_data_len = data_len;
10031		ctsio->kern_total_len = data_len;
10032	} else {
10033		ctsio->residual = 0;
10034		ctsio->kern_data_len = alloc_len;
10035		ctsio->kern_total_len = alloc_len;
10036	}
10037	ctsio->kern_data_resid = 0;
10038	ctsio->kern_rel_offset = 0;
10039	ctsio->kern_sg_entries = 0;
10040
10041	/*
10042	 * The control device is always connected.  The disk device, on the
10043	 * other hand, may not be online all the time.  Need to change this
10044	 * to figure out whether the disk device is actually online or not.
10045	 */
10046	if (lun != NULL)
10047		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10048				  lun->be_lun->lun_type;
10049	else
10050		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10051
10052	sp->page_code = SVPD_SCSI_PORTS;
10053	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10054	    sp->page_length);
10055	pd = &sp->design[0];
10056
10057	mtx_lock(&softc->ctl_lock);
10058	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10059		pg = 0;
10060	else
10061		pg = 1;
10062	for (g = 0; g < num_target_port_groups; g++) {
10063		STAILQ_FOREACH(port, &softc->port_list, links) {
10064			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10065				continue;
10066			if (lun != NULL &&
10067			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10068			    CTL_MAX_LUNS)
10069				continue;
10070			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10071			scsi_ulto2b(p, pd->relative_port_id);
10072			if (port->init_devid && g == pg) {
10073				iid_len = port->init_devid->len;
10074				memcpy(pd->initiator_transportid,
10075				    port->init_devid->data, port->init_devid->len);
10076			} else
10077				iid_len = 0;
10078			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10079			pdc = (struct scsi_vpd_port_designation_cont *)
10080			    (&pd->initiator_transportid[iid_len]);
10081			if (port->port_devid && g == pg) {
10082				id_len = port->port_devid->len;
10083				memcpy(pdc->target_port_descriptors,
10084				    port->port_devid->data, port->port_devid->len);
10085			} else
10086				id_len = 0;
10087			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10088			pd = (struct scsi_vpd_port_designation *)
10089			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10090		}
10091	}
10092	mtx_unlock(&softc->ctl_lock);
10093
10094	ctsio->scsi_status = SCSI_STATUS_OK;
10095	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10096	ctsio->be_move_done = ctl_config_move_done;
10097	ctl_datamove((union ctl_io *)ctsio);
10098
10099	return (CTL_RETVAL_COMPLETE);
10100}
10101
10102static int
10103ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10104{
10105	struct scsi_vpd_block_limits *bl_ptr;
10106	struct ctl_lun *lun;
10107	int bs;
10108
10109	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10110
10111	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10112	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10113	ctsio->kern_sg_entries = 0;
10114
10115	if (sizeof(*bl_ptr) < alloc_len) {
10116		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10117		ctsio->kern_data_len = sizeof(*bl_ptr);
10118		ctsio->kern_total_len = sizeof(*bl_ptr);
10119	} else {
10120		ctsio->residual = 0;
10121		ctsio->kern_data_len = alloc_len;
10122		ctsio->kern_total_len = alloc_len;
10123	}
10124	ctsio->kern_data_resid = 0;
10125	ctsio->kern_rel_offset = 0;
10126	ctsio->kern_sg_entries = 0;
10127
10128	/*
10129	 * The control device is always connected.  The disk device, on the
10130	 * other hand, may not be online all the time.  Need to change this
10131	 * to figure out whether the disk device is actually online or not.
10132	 */
10133	if (lun != NULL)
10134		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10135				  lun->be_lun->lun_type;
10136	else
10137		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10138
10139	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10140	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10141	bl_ptr->max_cmp_write_len = 0xff;
10142	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10143	if (lun != NULL) {
10144		bs = lun->be_lun->blocksize;
10145		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10146		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10147			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10148			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10149			if (lun->be_lun->pblockexp != 0) {
10150				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10151				    bl_ptr->opt_unmap_grain);
10152				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10153				    bl_ptr->unmap_grain_align);
10154			}
10155		}
10156		scsi_ulto4b(lun->be_lun->atomicblock,
10157		    bl_ptr->max_atomic_transfer_length);
10158		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10159		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10160	}
10161	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10162
10163	ctsio->scsi_status = SCSI_STATUS_OK;
10164	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10165	ctsio->be_move_done = ctl_config_move_done;
10166	ctl_datamove((union ctl_io *)ctsio);
10167
10168	return (CTL_RETVAL_COMPLETE);
10169}
10170
10171static int
10172ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10173{
10174	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10175	struct ctl_lun *lun;
10176
10177	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10178
10179	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10180	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10181	ctsio->kern_sg_entries = 0;
10182
10183	if (sizeof(*bdc_ptr) < alloc_len) {
10184		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10185		ctsio->kern_data_len = sizeof(*bdc_ptr);
10186		ctsio->kern_total_len = sizeof(*bdc_ptr);
10187	} else {
10188		ctsio->residual = 0;
10189		ctsio->kern_data_len = alloc_len;
10190		ctsio->kern_total_len = alloc_len;
10191	}
10192	ctsio->kern_data_resid = 0;
10193	ctsio->kern_rel_offset = 0;
10194	ctsio->kern_sg_entries = 0;
10195
10196	/*
10197	 * The control device is always connected.  The disk device, on the
10198	 * other hand, may not be online all the time.  Need to change this
10199	 * to figure out whether the disk device is actually online or not.
10200	 */
10201	if (lun != NULL)
10202		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10203				  lun->be_lun->lun_type;
10204	else
10205		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10206	bdc_ptr->page_code = SVPD_BDC;
10207	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10208	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10209	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10210
10211	ctsio->scsi_status = SCSI_STATUS_OK;
10212	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10213	ctsio->be_move_done = ctl_config_move_done;
10214	ctl_datamove((union ctl_io *)ctsio);
10215
10216	return (CTL_RETVAL_COMPLETE);
10217}
10218
10219static int
10220ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10221{
10222	struct scsi_vpd_logical_block_prov *lbp_ptr;
10223	struct ctl_lun *lun;
10224
10225	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10226
10227	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10228	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10229	ctsio->kern_sg_entries = 0;
10230
10231	if (sizeof(*lbp_ptr) < alloc_len) {
10232		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10233		ctsio->kern_data_len = sizeof(*lbp_ptr);
10234		ctsio->kern_total_len = sizeof(*lbp_ptr);
10235	} else {
10236		ctsio->residual = 0;
10237		ctsio->kern_data_len = alloc_len;
10238		ctsio->kern_total_len = alloc_len;
10239	}
10240	ctsio->kern_data_resid = 0;
10241	ctsio->kern_rel_offset = 0;
10242	ctsio->kern_sg_entries = 0;
10243
10244	/*
10245	 * The control device is always connected.  The disk device, on the
10246	 * other hand, may not be online all the time.  Need to change this
10247	 * to figure out whether the disk device is actually online or not.
10248	 */
10249	if (lun != NULL)
10250		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10251				  lun->be_lun->lun_type;
10252	else
10253		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10254
10255	lbp_ptr->page_code = SVPD_LBP;
10256	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10257	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10258		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10259		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10260		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10261	}
10262
10263	ctsio->scsi_status = SCSI_STATUS_OK;
10264	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10265	ctsio->be_move_done = ctl_config_move_done;
10266	ctl_datamove((union ctl_io *)ctsio);
10267
10268	return (CTL_RETVAL_COMPLETE);
10269}
10270
10271static int
10272ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10273{
10274	struct scsi_inquiry *cdb;
10275	struct ctl_lun *lun;
10276	int alloc_len, retval;
10277
10278	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10279	cdb = (struct scsi_inquiry *)ctsio->cdb;
10280
10281	retval = CTL_RETVAL_COMPLETE;
10282
10283	alloc_len = scsi_2btoul(cdb->length);
10284
10285	switch (cdb->page_code) {
10286	case SVPD_SUPPORTED_PAGES:
10287		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10288		break;
10289	case SVPD_UNIT_SERIAL_NUMBER:
10290		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10291		break;
10292	case SVPD_DEVICE_ID:
10293		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10294		break;
10295	case SVPD_EXTENDED_INQUIRY_DATA:
10296		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10297		break;
10298	case SVPD_MODE_PAGE_POLICY:
10299		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10300		break;
10301	case SVPD_SCSI_PORTS:
10302		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10303		break;
10304	case SVPD_SCSI_TPC:
10305		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10306		break;
10307	case SVPD_BLOCK_LIMITS:
10308		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10309		break;
10310	case SVPD_BDC:
10311		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10312		break;
10313	case SVPD_LBP:
10314		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10315		break;
10316	default:
10317		ctl_set_invalid_field(ctsio,
10318				      /*sks_valid*/ 1,
10319				      /*command*/ 1,
10320				      /*field*/ 2,
10321				      /*bit_valid*/ 0,
10322				      /*bit*/ 0);
10323		ctl_done((union ctl_io *)ctsio);
10324		retval = CTL_RETVAL_COMPLETE;
10325		break;
10326	}
10327
10328	return (retval);
10329}
10330
10331static int
10332ctl_inquiry_std(struct ctl_scsiio *ctsio)
10333{
10334	struct scsi_inquiry_data *inq_ptr;
10335	struct scsi_inquiry *cdb;
10336	struct ctl_softc *ctl_softc;
10337	struct ctl_lun *lun;
10338	char *val;
10339	uint32_t alloc_len, data_len;
10340	ctl_port_type port_type;
10341
10342	ctl_softc = control_softc;
10343
10344	/*
10345	 * Figure out whether we're talking to a Fibre Channel port or not.
10346	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10347	 * SCSI front ends.
10348	 */
10349	port_type = ctl_softc->ctl_ports[
10350	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10351	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10352		port_type = CTL_PORT_SCSI;
10353
10354	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10355	cdb = (struct scsi_inquiry *)ctsio->cdb;
10356	alloc_len = scsi_2btoul(cdb->length);
10357
10358	/*
10359	 * We malloc the full inquiry data size here and fill it
10360	 * in.  If the user only asks for less, we'll give him
10361	 * that much.
10362	 */
10363	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10364	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10365	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10366	ctsio->kern_sg_entries = 0;
10367	ctsio->kern_data_resid = 0;
10368	ctsio->kern_rel_offset = 0;
10369
10370	if (data_len < alloc_len) {
10371		ctsio->residual = alloc_len - data_len;
10372		ctsio->kern_data_len = data_len;
10373		ctsio->kern_total_len = data_len;
10374	} else {
10375		ctsio->residual = 0;
10376		ctsio->kern_data_len = alloc_len;
10377		ctsio->kern_total_len = alloc_len;
10378	}
10379
10380	/*
10381	 * If we have a LUN configured, report it as connected.  Otherwise,
10382	 * report that it is offline or no device is supported, depending
10383	 * on the value of inquiry_pq_no_lun.
10384	 *
10385	 * According to the spec (SPC-4 r34), the peripheral qualifier
10386	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10387	 *
10388	 * "A peripheral device having the specified peripheral device type
10389	 * is not connected to this logical unit. However, the device
10390	 * server is capable of supporting the specified peripheral device
10391	 * type on this logical unit."
10392	 *
10393	 * According to the same spec, the peripheral qualifier
10394	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10395	 *
10396	 * "The device server is not capable of supporting a peripheral
10397	 * device on this logical unit. For this peripheral qualifier the
10398	 * peripheral device type shall be set to 1Fh. All other peripheral
10399	 * device type values are reserved for this peripheral qualifier."
10400	 *
10401	 * Given the text, it would seem that we probably want to report that
10402	 * the LUN is offline here.  There is no LUN connected, but we can
10403	 * support a LUN at the given LUN number.
10404	 *
10405	 * In the real world, though, it sounds like things are a little
10406	 * different:
10407	 *
10408	 * - Linux, when presented with a LUN with the offline peripheral
10409	 *   qualifier, will create an sg driver instance for it.  So when
10410	 *   you attach it to CTL, you wind up with a ton of sg driver
10411	 *   instances.  (One for every LUN that Linux bothered to probe.)
10412	 *   Linux does this despite the fact that it issues a REPORT LUNs
10413	 *   to LUN 0 to get the inventory of supported LUNs.
10414	 *
10415	 * - There is other anecdotal evidence (from Emulex folks) about
10416	 *   arrays that use the offline peripheral qualifier for LUNs that
10417	 *   are on the "passive" path in an active/passive array.
10418	 *
10419	 * So the solution is provide a hopefully reasonable default
10420	 * (return bad/no LUN) and allow the user to change the behavior
10421	 * with a tunable/sysctl variable.
10422	 */
10423	if (lun != NULL)
10424		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10425				  lun->be_lun->lun_type;
10426	else if (ctl_softc->inquiry_pq_no_lun == 0)
10427		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10428	else
10429		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10430
10431	/* RMB in byte 2 is 0 */
10432	inq_ptr->version = SCSI_REV_SPC4;
10433
10434	/*
10435	 * According to SAM-3, even if a device only supports a single
10436	 * level of LUN addressing, it should still set the HISUP bit:
10437	 *
10438	 * 4.9.1 Logical unit numbers overview
10439	 *
10440	 * All logical unit number formats described in this standard are
10441	 * hierarchical in structure even when only a single level in that
10442	 * hierarchy is used. The HISUP bit shall be set to one in the
10443	 * standard INQUIRY data (see SPC-2) when any logical unit number
10444	 * format described in this standard is used.  Non-hierarchical
10445	 * formats are outside the scope of this standard.
10446	 *
10447	 * Therefore we set the HiSup bit here.
10448	 *
10449	 * The reponse format is 2, per SPC-3.
10450	 */
10451	inq_ptr->response_format = SID_HiSup | 2;
10452
10453	inq_ptr->additional_length = data_len -
10454	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10455	CTL_DEBUG_PRINT(("additional_length = %d\n",
10456			 inq_ptr->additional_length));
10457
10458	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10459	/* 16 bit addressing */
10460	if (port_type == CTL_PORT_SCSI)
10461		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10462	/* XXX set the SID_MultiP bit here if we're actually going to
10463	   respond on multiple ports */
10464	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10465
10466	/* 16 bit data bus, synchronous transfers */
10467	if (port_type == CTL_PORT_SCSI)
10468		inq_ptr->flags = SID_WBus16 | SID_Sync;
10469	/*
10470	 * XXX KDM do we want to support tagged queueing on the control
10471	 * device at all?
10472	 */
10473	if ((lun == NULL)
10474	 || (lun->be_lun->lun_type != T_PROCESSOR))
10475		inq_ptr->flags |= SID_CmdQue;
10476	/*
10477	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10478	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10479	 * name and 4 bytes for the revision.
10480	 */
10481	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10482	    "vendor")) == NULL) {
10483		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10484	} else {
10485		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10486		strncpy(inq_ptr->vendor, val,
10487		    min(sizeof(inq_ptr->vendor), strlen(val)));
10488	}
10489	if (lun == NULL) {
10490		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10491		    sizeof(inq_ptr->product));
10492	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10493		switch (lun->be_lun->lun_type) {
10494		case T_DIRECT:
10495			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10496			    sizeof(inq_ptr->product));
10497			break;
10498		case T_PROCESSOR:
10499			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10500			    sizeof(inq_ptr->product));
10501			break;
10502		default:
10503			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10504			    sizeof(inq_ptr->product));
10505			break;
10506		}
10507	} else {
10508		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10509		strncpy(inq_ptr->product, val,
10510		    min(sizeof(inq_ptr->product), strlen(val)));
10511	}
10512
10513	/*
10514	 * XXX make this a macro somewhere so it automatically gets
10515	 * incremented when we make changes.
10516	 */
10517	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10518	    "revision")) == NULL) {
10519		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10520	} else {
10521		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10522		strncpy(inq_ptr->revision, val,
10523		    min(sizeof(inq_ptr->revision), strlen(val)));
10524	}
10525
10526	/*
10527	 * For parallel SCSI, we support double transition and single
10528	 * transition clocking.  We also support QAS (Quick Arbitration
10529	 * and Selection) and Information Unit transfers on both the
10530	 * control and array devices.
10531	 */
10532	if (port_type == CTL_PORT_SCSI)
10533		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10534				    SID_SPI_IUS;
10535
10536	/* SAM-5 (no version claimed) */
10537	scsi_ulto2b(0x00A0, inq_ptr->version1);
10538	/* SPC-4 (no version claimed) */
10539	scsi_ulto2b(0x0460, inq_ptr->version2);
10540	if (port_type == CTL_PORT_FC) {
10541		/* FCP-2 ANSI INCITS.350:2003 */
10542		scsi_ulto2b(0x0917, inq_ptr->version3);
10543	} else if (port_type == CTL_PORT_SCSI) {
10544		/* SPI-4 ANSI INCITS.362:200x */
10545		scsi_ulto2b(0x0B56, inq_ptr->version3);
10546	} else if (port_type == CTL_PORT_ISCSI) {
10547		/* iSCSI (no version claimed) */
10548		scsi_ulto2b(0x0960, inq_ptr->version3);
10549	} else if (port_type == CTL_PORT_SAS) {
10550		/* SAS (no version claimed) */
10551		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10552	}
10553
10554	if (lun == NULL) {
10555		/* SBC-4 (no version claimed) */
10556		scsi_ulto2b(0x0600, inq_ptr->version4);
10557	} else {
10558		switch (lun->be_lun->lun_type) {
10559		case T_DIRECT:
10560			/* SBC-4 (no version claimed) */
10561			scsi_ulto2b(0x0600, inq_ptr->version4);
10562			break;
10563		case T_PROCESSOR:
10564		default:
10565			break;
10566		}
10567	}
10568
10569	ctsio->scsi_status = SCSI_STATUS_OK;
10570	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10571	ctsio->be_move_done = ctl_config_move_done;
10572	ctl_datamove((union ctl_io *)ctsio);
10573	return (CTL_RETVAL_COMPLETE);
10574}
10575
10576int
10577ctl_inquiry(struct ctl_scsiio *ctsio)
10578{
10579	struct scsi_inquiry *cdb;
10580	int retval;
10581
10582	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10583
10584	cdb = (struct scsi_inquiry *)ctsio->cdb;
10585	if (cdb->byte2 & SI_EVPD)
10586		retval = ctl_inquiry_evpd(ctsio);
10587	else if (cdb->page_code == 0)
10588		retval = ctl_inquiry_std(ctsio);
10589	else {
10590		ctl_set_invalid_field(ctsio,
10591				      /*sks_valid*/ 1,
10592				      /*command*/ 1,
10593				      /*field*/ 2,
10594				      /*bit_valid*/ 0,
10595				      /*bit*/ 0);
10596		ctl_done((union ctl_io *)ctsio);
10597		return (CTL_RETVAL_COMPLETE);
10598	}
10599
10600	return (retval);
10601}
10602
10603/*
10604 * For known CDB types, parse the LBA and length.
10605 */
10606static int
10607ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10608{
10609	if (io->io_hdr.io_type != CTL_IO_SCSI)
10610		return (1);
10611
10612	switch (io->scsiio.cdb[0]) {
10613	case COMPARE_AND_WRITE: {
10614		struct scsi_compare_and_write *cdb;
10615
10616		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10617
10618		*lba = scsi_8btou64(cdb->addr);
10619		*len = cdb->length;
10620		break;
10621	}
10622	case READ_6:
10623	case WRITE_6: {
10624		struct scsi_rw_6 *cdb;
10625
10626		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10627
10628		*lba = scsi_3btoul(cdb->addr);
10629		/* only 5 bits are valid in the most significant address byte */
10630		*lba &= 0x1fffff;
10631		*len = cdb->length;
10632		break;
10633	}
10634	case READ_10:
10635	case WRITE_10: {
10636		struct scsi_rw_10 *cdb;
10637
10638		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10639
10640		*lba = scsi_4btoul(cdb->addr);
10641		*len = scsi_2btoul(cdb->length);
10642		break;
10643	}
10644	case WRITE_VERIFY_10: {
10645		struct scsi_write_verify_10 *cdb;
10646
10647		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10648
10649		*lba = scsi_4btoul(cdb->addr);
10650		*len = scsi_2btoul(cdb->length);
10651		break;
10652	}
10653	case READ_12:
10654	case WRITE_12: {
10655		struct scsi_rw_12 *cdb;
10656
10657		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10658
10659		*lba = scsi_4btoul(cdb->addr);
10660		*len = scsi_4btoul(cdb->length);
10661		break;
10662	}
10663	case WRITE_VERIFY_12: {
10664		struct scsi_write_verify_12 *cdb;
10665
10666		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10667
10668		*lba = scsi_4btoul(cdb->addr);
10669		*len = scsi_4btoul(cdb->length);
10670		break;
10671	}
10672	case READ_16:
10673	case WRITE_16:
10674	case WRITE_ATOMIC_16: {
10675		struct scsi_rw_16 *cdb;
10676
10677		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10678
10679		*lba = scsi_8btou64(cdb->addr);
10680		*len = scsi_4btoul(cdb->length);
10681		break;
10682	}
10683	case WRITE_VERIFY_16: {
10684		struct scsi_write_verify_16 *cdb;
10685
10686		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10687
10688		*lba = scsi_8btou64(cdb->addr);
10689		*len = scsi_4btoul(cdb->length);
10690		break;
10691	}
10692	case WRITE_SAME_10: {
10693		struct scsi_write_same_10 *cdb;
10694
10695		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10696
10697		*lba = scsi_4btoul(cdb->addr);
10698		*len = scsi_2btoul(cdb->length);
10699		break;
10700	}
10701	case WRITE_SAME_16: {
10702		struct scsi_write_same_16 *cdb;
10703
10704		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10705
10706		*lba = scsi_8btou64(cdb->addr);
10707		*len = scsi_4btoul(cdb->length);
10708		break;
10709	}
10710	case VERIFY_10: {
10711		struct scsi_verify_10 *cdb;
10712
10713		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10714
10715		*lba = scsi_4btoul(cdb->addr);
10716		*len = scsi_2btoul(cdb->length);
10717		break;
10718	}
10719	case VERIFY_12: {
10720		struct scsi_verify_12 *cdb;
10721
10722		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10723
10724		*lba = scsi_4btoul(cdb->addr);
10725		*len = scsi_4btoul(cdb->length);
10726		break;
10727	}
10728	case VERIFY_16: {
10729		struct scsi_verify_16 *cdb;
10730
10731		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10732
10733		*lba = scsi_8btou64(cdb->addr);
10734		*len = scsi_4btoul(cdb->length);
10735		break;
10736	}
10737	case UNMAP: {
10738		*lba = 0;
10739		*len = UINT64_MAX;
10740		break;
10741	}
10742	default:
10743		return (1);
10744		break; /* NOTREACHED */
10745	}
10746
10747	return (0);
10748}
10749
10750static ctl_action
10751ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10752{
10753	uint64_t endlba1, endlba2;
10754
10755	endlba1 = lba1 + len1 - 1;
10756	endlba2 = lba2 + len2 - 1;
10757
10758	if ((endlba1 < lba2)
10759	 || (endlba2 < lba1))
10760		return (CTL_ACTION_PASS);
10761	else
10762		return (CTL_ACTION_BLOCK);
10763}
10764
10765static int
10766ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10767{
10768	struct ctl_ptr_len_flags *ptrlen;
10769	struct scsi_unmap_desc *buf, *end, *range;
10770	uint64_t lba;
10771	uint32_t len;
10772
10773	/* If not UNMAP -- go other way. */
10774	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10775	    io->scsiio.cdb[0] != UNMAP)
10776		return (CTL_ACTION_ERROR);
10777
10778	/* If UNMAP without data -- block and wait for data. */
10779	ptrlen = (struct ctl_ptr_len_flags *)
10780	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10781	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10782	    ptrlen->ptr == NULL)
10783		return (CTL_ACTION_BLOCK);
10784
10785	/* UNMAP with data -- check for collision. */
10786	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10787	end = buf + ptrlen->len / sizeof(*buf);
10788	for (range = buf; range < end; range++) {
10789		lba = scsi_8btou64(range->lba);
10790		len = scsi_4btoul(range->length);
10791		if ((lba < lba2 + len2) && (lba + len > lba2))
10792			return (CTL_ACTION_BLOCK);
10793	}
10794	return (CTL_ACTION_PASS);
10795}
10796
10797static ctl_action
10798ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10799{
10800	uint64_t lba1, lba2;
10801	uint64_t len1, len2;
10802	int retval;
10803
10804	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10805		return (CTL_ACTION_ERROR);
10806
10807	retval = ctl_extent_check_unmap(io2, lba1, len1);
10808	if (retval != CTL_ACTION_ERROR)
10809		return (retval);
10810
10811	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10812		return (CTL_ACTION_ERROR);
10813
10814	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10815}
10816
10817static ctl_action
10818ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10819    union ctl_io *ooa_io)
10820{
10821	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10822	ctl_serialize_action *serialize_row;
10823
10824	/*
10825	 * The initiator attempted multiple untagged commands at the same
10826	 * time.  Can't do that.
10827	 */
10828	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10829	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10830	 && ((pending_io->io_hdr.nexus.targ_port ==
10831	      ooa_io->io_hdr.nexus.targ_port)
10832	  && (pending_io->io_hdr.nexus.initid.id ==
10833	      ooa_io->io_hdr.nexus.initid.id))
10834	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10835		return (CTL_ACTION_OVERLAP);
10836
10837	/*
10838	 * The initiator attempted to send multiple tagged commands with
10839	 * the same ID.  (It's fine if different initiators have the same
10840	 * tag ID.)
10841	 *
10842	 * Even if all of those conditions are true, we don't kill the I/O
10843	 * if the command ahead of us has been aborted.  We won't end up
10844	 * sending it to the FETD, and it's perfectly legal to resend a
10845	 * command with the same tag number as long as the previous
10846	 * instance of this tag number has been aborted somehow.
10847	 */
10848	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10849	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10850	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10851	 && ((pending_io->io_hdr.nexus.targ_port ==
10852	      ooa_io->io_hdr.nexus.targ_port)
10853	  && (pending_io->io_hdr.nexus.initid.id ==
10854	      ooa_io->io_hdr.nexus.initid.id))
10855	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10856		return (CTL_ACTION_OVERLAP_TAG);
10857
10858	/*
10859	 * If we get a head of queue tag, SAM-3 says that we should
10860	 * immediately execute it.
10861	 *
10862	 * What happens if this command would normally block for some other
10863	 * reason?  e.g. a request sense with a head of queue tag
10864	 * immediately after a write.  Normally that would block, but this
10865	 * will result in its getting executed immediately...
10866	 *
10867	 * We currently return "pass" instead of "skip", so we'll end up
10868	 * going through the rest of the queue to check for overlapped tags.
10869	 *
10870	 * XXX KDM check for other types of blockage first??
10871	 */
10872	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10873		return (CTL_ACTION_PASS);
10874
10875	/*
10876	 * Ordered tags have to block until all items ahead of them
10877	 * have completed.  If we get called with an ordered tag, we always
10878	 * block, if something else is ahead of us in the queue.
10879	 */
10880	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10881		return (CTL_ACTION_BLOCK);
10882
10883	/*
10884	 * Simple tags get blocked until all head of queue and ordered tags
10885	 * ahead of them have completed.  I'm lumping untagged commands in
10886	 * with simple tags here.  XXX KDM is that the right thing to do?
10887	 */
10888	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10889	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10890	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10891	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10892		return (CTL_ACTION_BLOCK);
10893
10894	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10895	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10896
10897	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10898
10899	switch (serialize_row[pending_entry->seridx]) {
10900	case CTL_SER_BLOCK:
10901		return (CTL_ACTION_BLOCK);
10902	case CTL_SER_EXTENT:
10903		return (ctl_extent_check(pending_io, ooa_io));
10904	case CTL_SER_EXTENTOPT:
10905		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10906		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10907			return (ctl_extent_check(pending_io, ooa_io));
10908		/* FALLTHROUGH */
10909	case CTL_SER_PASS:
10910		return (CTL_ACTION_PASS);
10911	case CTL_SER_BLOCKOPT:
10912		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10913		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10914			return (CTL_ACTION_BLOCK);
10915		return (CTL_ACTION_PASS);
10916	case CTL_SER_SKIP:
10917		return (CTL_ACTION_SKIP);
10918	default:
10919		panic("invalid serialization value %d",
10920		      serialize_row[pending_entry->seridx]);
10921	}
10922
10923	return (CTL_ACTION_ERROR);
10924}
10925
10926/*
10927 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10928 * Assumptions:
10929 * - pending_io is generally either incoming, or on the blocked queue
10930 * - starting I/O is the I/O we want to start the check with.
10931 */
10932static ctl_action
10933ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10934	      union ctl_io *starting_io)
10935{
10936	union ctl_io *ooa_io;
10937	ctl_action action;
10938
10939	mtx_assert(&lun->lun_lock, MA_OWNED);
10940
10941	/*
10942	 * Run back along the OOA queue, starting with the current
10943	 * blocked I/O and going through every I/O before it on the
10944	 * queue.  If starting_io is NULL, we'll just end up returning
10945	 * CTL_ACTION_PASS.
10946	 */
10947	for (ooa_io = starting_io; ooa_io != NULL;
10948	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10949	     ooa_links)){
10950
10951		/*
10952		 * This routine just checks to see whether
10953		 * cur_blocked is blocked by ooa_io, which is ahead
10954		 * of it in the queue.  It doesn't queue/dequeue
10955		 * cur_blocked.
10956		 */
10957		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10958		switch (action) {
10959		case CTL_ACTION_BLOCK:
10960		case CTL_ACTION_OVERLAP:
10961		case CTL_ACTION_OVERLAP_TAG:
10962		case CTL_ACTION_SKIP:
10963		case CTL_ACTION_ERROR:
10964			return (action);
10965			break; /* NOTREACHED */
10966		case CTL_ACTION_PASS:
10967			break;
10968		default:
10969			panic("invalid action %d", action);
10970			break;  /* NOTREACHED */
10971		}
10972	}
10973
10974	return (CTL_ACTION_PASS);
10975}
10976
10977/*
10978 * Assumptions:
10979 * - An I/O has just completed, and has been removed from the per-LUN OOA
10980 *   queue, so some items on the blocked queue may now be unblocked.
10981 */
10982static int
10983ctl_check_blocked(struct ctl_lun *lun)
10984{
10985	union ctl_io *cur_blocked, *next_blocked;
10986
10987	mtx_assert(&lun->lun_lock, MA_OWNED);
10988
10989	/*
10990	 * Run forward from the head of the blocked queue, checking each
10991	 * entry against the I/Os prior to it on the OOA queue to see if
10992	 * there is still any blockage.
10993	 *
10994	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10995	 * with our removing a variable on it while it is traversing the
10996	 * list.
10997	 */
10998	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10999	     cur_blocked != NULL; cur_blocked = next_blocked) {
11000		union ctl_io *prev_ooa;
11001		ctl_action action;
11002
11003		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11004							  blocked_links);
11005
11006		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11007						      ctl_ooaq, ooa_links);
11008
11009		/*
11010		 * If cur_blocked happens to be the first item in the OOA
11011		 * queue now, prev_ooa will be NULL, and the action
11012		 * returned will just be CTL_ACTION_PASS.
11013		 */
11014		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11015
11016		switch (action) {
11017		case CTL_ACTION_BLOCK:
11018			/* Nothing to do here, still blocked */
11019			break;
11020		case CTL_ACTION_OVERLAP:
11021		case CTL_ACTION_OVERLAP_TAG:
11022			/*
11023			 * This shouldn't happen!  In theory we've already
11024			 * checked this command for overlap...
11025			 */
11026			break;
11027		case CTL_ACTION_PASS:
11028		case CTL_ACTION_SKIP: {
11029			struct ctl_softc *softc;
11030			const struct ctl_cmd_entry *entry;
11031			uint32_t initidx;
11032			int isc_retval;
11033
11034			/*
11035			 * The skip case shouldn't happen, this transaction
11036			 * should have never made it onto the blocked queue.
11037			 */
11038			/*
11039			 * This I/O is no longer blocked, we can remove it
11040			 * from the blocked queue.  Since this is a TAILQ
11041			 * (doubly linked list), we can do O(1) removals
11042			 * from any place on the list.
11043			 */
11044			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11045				     blocked_links);
11046			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11047
11048			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11049				/*
11050				 * Need to send IO back to original side to
11051				 * run
11052				 */
11053				union ctl_ha_msg msg_info;
11054
11055				msg_info.hdr.original_sc =
11056					cur_blocked->io_hdr.original_sc;
11057				msg_info.hdr.serializing_sc = cur_blocked;
11058				msg_info.hdr.msg_type = CTL_MSG_R2R;
11059				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11060				     &msg_info, sizeof(msg_info), 0)) >
11061				     CTL_HA_STATUS_SUCCESS) {
11062					printf("CTL:Check Blocked error from "
11063					       "ctl_ha_msg_send %d\n",
11064					       isc_retval);
11065				}
11066				break;
11067			}
11068			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11069			softc = control_softc;
11070
11071			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11072
11073			/*
11074			 * Check this I/O for LUN state changes that may
11075			 * have happened while this command was blocked.
11076			 * The LUN state may have been changed by a command
11077			 * ahead of us in the queue, so we need to re-check
11078			 * for any states that can be caused by SCSI
11079			 * commands.
11080			 */
11081			if (ctl_scsiio_lun_check(softc, lun, entry,
11082						 &cur_blocked->scsiio) == 0) {
11083				cur_blocked->io_hdr.flags |=
11084				                      CTL_FLAG_IS_WAS_ON_RTR;
11085				ctl_enqueue_rtr(cur_blocked);
11086			} else
11087				ctl_done(cur_blocked);
11088			break;
11089		}
11090		default:
11091			/*
11092			 * This probably shouldn't happen -- we shouldn't
11093			 * get CTL_ACTION_ERROR, or anything else.
11094			 */
11095			break;
11096		}
11097	}
11098
11099	return (CTL_RETVAL_COMPLETE);
11100}
11101
11102/*
11103 * This routine (with one exception) checks LUN flags that can be set by
11104 * commands ahead of us in the OOA queue.  These flags have to be checked
11105 * when a command initially comes in, and when we pull a command off the
11106 * blocked queue and are preparing to execute it.  The reason we have to
11107 * check these flags for commands on the blocked queue is that the LUN
11108 * state may have been changed by a command ahead of us while we're on the
11109 * blocked queue.
11110 *
11111 * Ordering is somewhat important with these checks, so please pay
11112 * careful attention to the placement of any new checks.
11113 */
11114static int
11115ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11116    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11117{
11118	int retval;
11119	uint32_t residx;
11120
11121	retval = 0;
11122
11123	mtx_assert(&lun->lun_lock, MA_OWNED);
11124
11125	/*
11126	 * If this shelf is a secondary shelf controller, we have to reject
11127	 * any media access commands.
11128	 */
11129#if 0
11130	/* No longer needed for HA */
11131	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11132	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11133		ctl_set_lun_standby(ctsio);
11134		retval = 1;
11135		goto bailout;
11136	}
11137#endif
11138
11139	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11140		if (lun->flags & CTL_LUN_READONLY) {
11141			ctl_set_sense(ctsio, /*current_error*/ 1,
11142			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11143			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11144			retval = 1;
11145			goto bailout;
11146		}
11147		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11148		    .eca_and_aen & SCP_SWP) != 0) {
11149			ctl_set_sense(ctsio, /*current_error*/ 1,
11150			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11151			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11152			retval = 1;
11153			goto bailout;
11154		}
11155	}
11156
11157	/*
11158	 * Check for a reservation conflict.  If this command isn't allowed
11159	 * even on reserved LUNs, and if this initiator isn't the one who
11160	 * reserved us, reject the command with a reservation conflict.
11161	 */
11162	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11163	if ((lun->flags & CTL_LUN_RESERVED)
11164	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11165		if (lun->res_idx != residx) {
11166			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11167			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11168			retval = 1;
11169			goto bailout;
11170		}
11171	}
11172
11173	if ((lun->flags & CTL_LUN_PR_RESERVED)
11174	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11175		/*
11176		 * if we aren't registered or it's a res holder type
11177		 * reservation and this isn't the res holder then set a
11178		 * conflict.
11179		 * NOTE: Commands which might be allowed on write exclusive
11180		 * type reservations are checked in the particular command
11181		 * for a conflict. Read and SSU are the only ones.
11182		 */
11183		if (lun->pr_keys[residx] == 0
11184		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11185			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11186			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11187			retval = 1;
11188			goto bailout;
11189		}
11190
11191	}
11192
11193	if ((lun->flags & CTL_LUN_OFFLINE)
11194	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11195		ctl_set_lun_not_ready(ctsio);
11196		retval = 1;
11197		goto bailout;
11198	}
11199
11200	/*
11201	 * If the LUN is stopped, see if this particular command is allowed
11202	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11203	 */
11204	if ((lun->flags & CTL_LUN_STOPPED)
11205	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11206		/* "Logical unit not ready, initializing cmd. required" */
11207		ctl_set_lun_stopped(ctsio);
11208		retval = 1;
11209		goto bailout;
11210	}
11211
11212	if ((lun->flags & CTL_LUN_INOPERABLE)
11213	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11214		/* "Medium format corrupted" */
11215		ctl_set_medium_format_corrupted(ctsio);
11216		retval = 1;
11217		goto bailout;
11218	}
11219
11220bailout:
11221	return (retval);
11222
11223}
11224
11225static void
11226ctl_failover_io(union ctl_io *io, int have_lock)
11227{
11228	ctl_set_busy(&io->scsiio);
11229	ctl_done(io);
11230}
11231
11232static void
11233ctl_failover(void)
11234{
11235	struct ctl_lun *lun;
11236	struct ctl_softc *ctl_softc;
11237	union ctl_io *next_io, *pending_io;
11238	union ctl_io *io;
11239	int lun_idx;
11240	int i;
11241
11242	ctl_softc = control_softc;
11243
11244	mtx_lock(&ctl_softc->ctl_lock);
11245	/*
11246	 * Remove any cmds from the other SC from the rtr queue.  These
11247	 * will obviously only be for LUNs for which we're the primary.
11248	 * We can't send status or get/send data for these commands.
11249	 * Since they haven't been executed yet, we can just remove them.
11250	 * We'll either abort them or delete them below, depending on
11251	 * which HA mode we're in.
11252	 */
11253#ifdef notyet
11254	mtx_lock(&ctl_softc->queue_lock);
11255	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11256	     io != NULL; io = next_io) {
11257		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11258		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11259			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11260				      ctl_io_hdr, links);
11261	}
11262	mtx_unlock(&ctl_softc->queue_lock);
11263#endif
11264
11265	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11266		lun = ctl_softc->ctl_luns[lun_idx];
11267		if (lun==NULL)
11268			continue;
11269
11270		/*
11271		 * Processor LUNs are primary on both sides.
11272		 * XXX will this always be true?
11273		 */
11274		if (lun->be_lun->lun_type == T_PROCESSOR)
11275			continue;
11276
11277		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11278		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11279			printf("FAILOVER: primary lun %d\n", lun_idx);
11280		        /*
11281			 * Remove all commands from the other SC. First from the
11282			 * blocked queue then from the ooa queue. Once we have
11283			 * removed them. Call ctl_check_blocked to see if there
11284			 * is anything that can run.
11285			 */
11286			for (io = (union ctl_io *)TAILQ_FIRST(
11287			     &lun->blocked_queue); io != NULL; io = next_io) {
11288
11289		        	next_io = (union ctl_io *)TAILQ_NEXT(
11290				    &io->io_hdr, blocked_links);
11291
11292				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11293					TAILQ_REMOVE(&lun->blocked_queue,
11294						     &io->io_hdr,blocked_links);
11295					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11296					TAILQ_REMOVE(&lun->ooa_queue,
11297						     &io->io_hdr, ooa_links);
11298
11299					ctl_free_io(io);
11300				}
11301			}
11302
11303			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11304	     		     io != NULL; io = next_io) {
11305
11306		        	next_io = (union ctl_io *)TAILQ_NEXT(
11307				    &io->io_hdr, ooa_links);
11308
11309				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11310
11311					TAILQ_REMOVE(&lun->ooa_queue,
11312						&io->io_hdr,
11313					     	ooa_links);
11314
11315					ctl_free_io(io);
11316				}
11317			}
11318			ctl_check_blocked(lun);
11319		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11320			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11321
11322			printf("FAILOVER: primary lun %d\n", lun_idx);
11323			/*
11324			 * Abort all commands from the other SC.  We can't
11325			 * send status back for them now.  These should get
11326			 * cleaned up when they are completed or come out
11327			 * for a datamove operation.
11328			 */
11329			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11330	     		     io != NULL; io = next_io) {
11331		        	next_io = (union ctl_io *)TAILQ_NEXT(
11332					&io->io_hdr, ooa_links);
11333
11334				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11335					io->io_hdr.flags |= CTL_FLAG_ABORT;
11336			}
11337		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11338			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11339
11340			printf("FAILOVER: secondary lun %d\n", lun_idx);
11341
11342			lun->flags |= CTL_LUN_PRIMARY_SC;
11343
11344			/*
11345			 * We send all I/O that was sent to this controller
11346			 * and redirected to the other side back with
11347			 * busy status, and have the initiator retry it.
11348			 * Figuring out how much data has been transferred,
11349			 * etc. and picking up where we left off would be
11350			 * very tricky.
11351			 *
11352			 * XXX KDM need to remove I/O from the blocked
11353			 * queue as well!
11354			 */
11355			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11356			     &lun->ooa_queue); pending_io != NULL;
11357			     pending_io = next_io) {
11358
11359				next_io =  (union ctl_io *)TAILQ_NEXT(
11360					&pending_io->io_hdr, ooa_links);
11361
11362				pending_io->io_hdr.flags &=
11363					~CTL_FLAG_SENT_2OTHER_SC;
11364
11365				if (pending_io->io_hdr.flags &
11366				    CTL_FLAG_IO_ACTIVE) {
11367					pending_io->io_hdr.flags |=
11368						CTL_FLAG_FAILOVER;
11369				} else {
11370					ctl_set_busy(&pending_io->scsiio);
11371					ctl_done(pending_io);
11372				}
11373			}
11374
11375			/*
11376			 * Build Unit Attention
11377			 */
11378			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11379				lun->pending_ua[i] |=
11380				                     CTL_UA_ASYM_ACC_CHANGE;
11381			}
11382		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11383			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11384			printf("FAILOVER: secondary lun %d\n", lun_idx);
11385			/*
11386			 * if the first io on the OOA is not on the RtR queue
11387			 * add it.
11388			 */
11389			lun->flags |= CTL_LUN_PRIMARY_SC;
11390
11391			pending_io = (union ctl_io *)TAILQ_FIRST(
11392			    &lun->ooa_queue);
11393			if (pending_io==NULL) {
11394				printf("Nothing on OOA queue\n");
11395				continue;
11396			}
11397
11398			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11399			if ((pending_io->io_hdr.flags &
11400			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11401				pending_io->io_hdr.flags |=
11402				    CTL_FLAG_IS_WAS_ON_RTR;
11403				ctl_enqueue_rtr(pending_io);
11404			}
11405#if 0
11406			else
11407			{
11408				printf("Tag 0x%04x is running\n",
11409				      pending_io->scsiio.tag_num);
11410			}
11411#endif
11412
11413			next_io = (union ctl_io *)TAILQ_NEXT(
11414			    &pending_io->io_hdr, ooa_links);
11415			for (pending_io=next_io; pending_io != NULL;
11416			     pending_io = next_io) {
11417				pending_io->io_hdr.flags &=
11418				    ~CTL_FLAG_SENT_2OTHER_SC;
11419				next_io = (union ctl_io *)TAILQ_NEXT(
11420					&pending_io->io_hdr, ooa_links);
11421				if (pending_io->io_hdr.flags &
11422				    CTL_FLAG_IS_WAS_ON_RTR) {
11423#if 0
11424				        printf("Tag 0x%04x is running\n",
11425				      		pending_io->scsiio.tag_num);
11426#endif
11427					continue;
11428				}
11429
11430				switch (ctl_check_ooa(lun, pending_io,
11431			            (union ctl_io *)TAILQ_PREV(
11432				    &pending_io->io_hdr, ctl_ooaq,
11433				    ooa_links))) {
11434
11435				case CTL_ACTION_BLOCK:
11436					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11437							  &pending_io->io_hdr,
11438							  blocked_links);
11439					pending_io->io_hdr.flags |=
11440					    CTL_FLAG_BLOCKED;
11441					break;
11442				case CTL_ACTION_PASS:
11443				case CTL_ACTION_SKIP:
11444					pending_io->io_hdr.flags |=
11445					    CTL_FLAG_IS_WAS_ON_RTR;
11446					ctl_enqueue_rtr(pending_io);
11447					break;
11448				case CTL_ACTION_OVERLAP:
11449					ctl_set_overlapped_cmd(
11450					    (struct ctl_scsiio *)pending_io);
11451					ctl_done(pending_io);
11452					break;
11453				case CTL_ACTION_OVERLAP_TAG:
11454					ctl_set_overlapped_tag(
11455					    (struct ctl_scsiio *)pending_io,
11456					    pending_io->scsiio.tag_num & 0xff);
11457					ctl_done(pending_io);
11458					break;
11459				case CTL_ACTION_ERROR:
11460				default:
11461					ctl_set_internal_failure(
11462						(struct ctl_scsiio *)pending_io,
11463						0,  // sks_valid
11464						0); //retry count
11465					ctl_done(pending_io);
11466					break;
11467				}
11468			}
11469
11470			/*
11471			 * Build Unit Attention
11472			 */
11473			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11474				lun->pending_ua[i] |=
11475				                     CTL_UA_ASYM_ACC_CHANGE;
11476			}
11477		} else {
11478			panic("Unhandled HA mode failover, LUN flags = %#x, "
11479			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11480		}
11481	}
11482	ctl_pause_rtr = 0;
11483	mtx_unlock(&ctl_softc->ctl_lock);
11484}
11485
11486static int
11487ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11488{
11489	struct ctl_lun *lun;
11490	const struct ctl_cmd_entry *entry;
11491	uint32_t initidx, targ_lun;
11492	int retval;
11493
11494	retval = 0;
11495
11496	lun = NULL;
11497
11498	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11499	if ((targ_lun < CTL_MAX_LUNS)
11500	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11501		lun = ctl_softc->ctl_luns[targ_lun];
11502		/*
11503		 * If the LUN is invalid, pretend that it doesn't exist.
11504		 * It will go away as soon as all pending I/O has been
11505		 * completed.
11506		 */
11507		if (lun->flags & CTL_LUN_DISABLED) {
11508			lun = NULL;
11509		} else {
11510			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11511			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11512				lun->be_lun;
11513			if (lun->be_lun->lun_type == T_PROCESSOR) {
11514				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11515			}
11516
11517			/*
11518			 * Every I/O goes into the OOA queue for a
11519			 * particular LUN, and stays there until completion.
11520			 */
11521			mtx_lock(&lun->lun_lock);
11522			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11523			    ooa_links);
11524		}
11525	} else {
11526		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11527		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11528	}
11529
11530	/* Get command entry and return error if it is unsuppotyed. */
11531	entry = ctl_validate_command(ctsio);
11532	if (entry == NULL) {
11533		if (lun)
11534			mtx_unlock(&lun->lun_lock);
11535		return (retval);
11536	}
11537
11538	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11539	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11540
11541	/*
11542	 * Check to see whether we can send this command to LUNs that don't
11543	 * exist.  This should pretty much only be the case for inquiry
11544	 * and request sense.  Further checks, below, really require having
11545	 * a LUN, so we can't really check the command anymore.  Just put
11546	 * it on the rtr queue.
11547	 */
11548	if (lun == NULL) {
11549		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11550			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11551			ctl_enqueue_rtr((union ctl_io *)ctsio);
11552			return (retval);
11553		}
11554
11555		ctl_set_unsupported_lun(ctsio);
11556		ctl_done((union ctl_io *)ctsio);
11557		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11558		return (retval);
11559	} else {
11560		/*
11561		 * Make sure we support this particular command on this LUN.
11562		 * e.g., we don't support writes to the control LUN.
11563		 */
11564		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11565			mtx_unlock(&lun->lun_lock);
11566			ctl_set_invalid_opcode(ctsio);
11567			ctl_done((union ctl_io *)ctsio);
11568			return (retval);
11569		}
11570	}
11571
11572	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11573
11574#ifdef CTL_WITH_CA
11575	/*
11576	 * If we've got a request sense, it'll clear the contingent
11577	 * allegiance condition.  Otherwise, if we have a CA condition for
11578	 * this initiator, clear it, because it sent down a command other
11579	 * than request sense.
11580	 */
11581	if ((ctsio->cdb[0] != REQUEST_SENSE)
11582	 && (ctl_is_set(lun->have_ca, initidx)))
11583		ctl_clear_mask(lun->have_ca, initidx);
11584#endif
11585
11586	/*
11587	 * If the command has this flag set, it handles its own unit
11588	 * attention reporting, we shouldn't do anything.  Otherwise we
11589	 * check for any pending unit attentions, and send them back to the
11590	 * initiator.  We only do this when a command initially comes in,
11591	 * not when we pull it off the blocked queue.
11592	 *
11593	 * According to SAM-3, section 5.3.2, the order that things get
11594	 * presented back to the host is basically unit attentions caused
11595	 * by some sort of reset event, busy status, reservation conflicts
11596	 * or task set full, and finally any other status.
11597	 *
11598	 * One issue here is that some of the unit attentions we report
11599	 * don't fall into the "reset" category (e.g. "reported luns data
11600	 * has changed").  So reporting it here, before the reservation
11601	 * check, may be technically wrong.  I guess the only thing to do
11602	 * would be to check for and report the reset events here, and then
11603	 * check for the other unit attention types after we check for a
11604	 * reservation conflict.
11605	 *
11606	 * XXX KDM need to fix this
11607	 */
11608	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11609		ctl_ua_type ua_type;
11610
11611		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11612			scsi_sense_data_type sense_format;
11613
11614			if (lun != NULL)
11615				sense_format = (lun->flags &
11616				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11617				    SSD_TYPE_FIXED;
11618			else
11619				sense_format = SSD_TYPE_FIXED;
11620
11621			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11622			    &ctsio->sense_data, sense_format);
11623			if (ua_type != CTL_UA_NONE) {
11624				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11625				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11626						       CTL_AUTOSENSE;
11627				ctsio->sense_len = SSD_FULL_SIZE;
11628				mtx_unlock(&lun->lun_lock);
11629				ctl_done((union ctl_io *)ctsio);
11630				return (retval);
11631			}
11632		}
11633	}
11634
11635
11636	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11637		mtx_unlock(&lun->lun_lock);
11638		ctl_done((union ctl_io *)ctsio);
11639		return (retval);
11640	}
11641
11642	/*
11643	 * XXX CHD this is where we want to send IO to other side if
11644	 * this LUN is secondary on this SC. We will need to make a copy
11645	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11646	 * the copy we send as FROM_OTHER.
11647	 * We also need to stuff the address of the original IO so we can
11648	 * find it easily. Something similar will need be done on the other
11649	 * side so when we are done we can find the copy.
11650	 */
11651	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11652		union ctl_ha_msg msg_info;
11653		int isc_retval;
11654
11655		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11656
11657		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11658		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11659#if 0
11660		printf("1. ctsio %p\n", ctsio);
11661#endif
11662		msg_info.hdr.serializing_sc = NULL;
11663		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11664		msg_info.scsi.tag_num = ctsio->tag_num;
11665		msg_info.scsi.tag_type = ctsio->tag_type;
11666		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11667
11668		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11669
11670		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11671		    (void *)&msg_info, sizeof(msg_info), 0)) >
11672		    CTL_HA_STATUS_SUCCESS) {
11673			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11674			       isc_retval);
11675			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11676		} else {
11677#if 0
11678			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11679#endif
11680		}
11681
11682		/*
11683		 * XXX KDM this I/O is off the incoming queue, but hasn't
11684		 * been inserted on any other queue.  We may need to come
11685		 * up with a holding queue while we wait for serialization
11686		 * so that we have an idea of what we're waiting for from
11687		 * the other side.
11688		 */
11689		mtx_unlock(&lun->lun_lock);
11690		return (retval);
11691	}
11692
11693	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11694			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11695			      ctl_ooaq, ooa_links))) {
11696	case CTL_ACTION_BLOCK:
11697		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11698		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11699				  blocked_links);
11700		mtx_unlock(&lun->lun_lock);
11701		return (retval);
11702	case CTL_ACTION_PASS:
11703	case CTL_ACTION_SKIP:
11704		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11705		mtx_unlock(&lun->lun_lock);
11706		ctl_enqueue_rtr((union ctl_io *)ctsio);
11707		break;
11708	case CTL_ACTION_OVERLAP:
11709		mtx_unlock(&lun->lun_lock);
11710		ctl_set_overlapped_cmd(ctsio);
11711		ctl_done((union ctl_io *)ctsio);
11712		break;
11713	case CTL_ACTION_OVERLAP_TAG:
11714		mtx_unlock(&lun->lun_lock);
11715		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11716		ctl_done((union ctl_io *)ctsio);
11717		break;
11718	case CTL_ACTION_ERROR:
11719	default:
11720		mtx_unlock(&lun->lun_lock);
11721		ctl_set_internal_failure(ctsio,
11722					 /*sks_valid*/ 0,
11723					 /*retry_count*/ 0);
11724		ctl_done((union ctl_io *)ctsio);
11725		break;
11726	}
11727	return (retval);
11728}
11729
11730const struct ctl_cmd_entry *
11731ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11732{
11733	const struct ctl_cmd_entry *entry;
11734	int service_action;
11735
11736	entry = &ctl_cmd_table[ctsio->cdb[0]];
11737	if (sa)
11738		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11739	if (entry->flags & CTL_CMD_FLAG_SA5) {
11740		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11741		entry = &((const struct ctl_cmd_entry *)
11742		    entry->execute)[service_action];
11743	}
11744	return (entry);
11745}
11746
11747const struct ctl_cmd_entry *
11748ctl_validate_command(struct ctl_scsiio *ctsio)
11749{
11750	const struct ctl_cmd_entry *entry;
11751	int i, sa;
11752	uint8_t diff;
11753
11754	entry = ctl_get_cmd_entry(ctsio, &sa);
11755	if (entry->execute == NULL) {
11756		if (sa)
11757			ctl_set_invalid_field(ctsio,
11758					      /*sks_valid*/ 1,
11759					      /*command*/ 1,
11760					      /*field*/ 1,
11761					      /*bit_valid*/ 1,
11762					      /*bit*/ 4);
11763		else
11764			ctl_set_invalid_opcode(ctsio);
11765		ctl_done((union ctl_io *)ctsio);
11766		return (NULL);
11767	}
11768	KASSERT(entry->length > 0,
11769	    ("Not defined length for command 0x%02x/0x%02x",
11770	     ctsio->cdb[0], ctsio->cdb[1]));
11771	for (i = 1; i < entry->length; i++) {
11772		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11773		if (diff == 0)
11774			continue;
11775		ctl_set_invalid_field(ctsio,
11776				      /*sks_valid*/ 1,
11777				      /*command*/ 1,
11778				      /*field*/ i,
11779				      /*bit_valid*/ 1,
11780				      /*bit*/ fls(diff) - 1);
11781		ctl_done((union ctl_io *)ctsio);
11782		return (NULL);
11783	}
11784	return (entry);
11785}
11786
11787static int
11788ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11789{
11790
11791	switch (lun_type) {
11792	case T_PROCESSOR:
11793		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11794		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11795			return (0);
11796		break;
11797	case T_DIRECT:
11798		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11799		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11800			return (0);
11801		break;
11802	default:
11803		return (0);
11804	}
11805	return (1);
11806}
11807
11808static int
11809ctl_scsiio(struct ctl_scsiio *ctsio)
11810{
11811	int retval;
11812	const struct ctl_cmd_entry *entry;
11813
11814	retval = CTL_RETVAL_COMPLETE;
11815
11816	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11817
11818	entry = ctl_get_cmd_entry(ctsio, NULL);
11819
11820	/*
11821	 * If this I/O has been aborted, just send it straight to
11822	 * ctl_done() without executing it.
11823	 */
11824	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11825		ctl_done((union ctl_io *)ctsio);
11826		goto bailout;
11827	}
11828
11829	/*
11830	 * All the checks should have been handled by ctl_scsiio_precheck().
11831	 * We should be clear now to just execute the I/O.
11832	 */
11833	retval = entry->execute(ctsio);
11834
11835bailout:
11836	return (retval);
11837}
11838
11839/*
11840 * Since we only implement one target right now, a bus reset simply resets
11841 * our single target.
11842 */
11843static int
11844ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11845{
11846	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11847}
11848
11849static int
11850ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11851		 ctl_ua_type ua_type)
11852{
11853	struct ctl_lun *lun;
11854	int retval;
11855
11856	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11857		union ctl_ha_msg msg_info;
11858
11859		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11860		msg_info.hdr.nexus = io->io_hdr.nexus;
11861		if (ua_type==CTL_UA_TARG_RESET)
11862			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11863		else
11864			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11865		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11866		msg_info.hdr.original_sc = NULL;
11867		msg_info.hdr.serializing_sc = NULL;
11868		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11869		    (void *)&msg_info, sizeof(msg_info), 0)) {
11870		}
11871	}
11872	retval = 0;
11873
11874	mtx_lock(&ctl_softc->ctl_lock);
11875	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11876		retval += ctl_lun_reset(lun, io, ua_type);
11877	mtx_unlock(&ctl_softc->ctl_lock);
11878
11879	return (retval);
11880}
11881
11882/*
11883 * The LUN should always be set.  The I/O is optional, and is used to
11884 * distinguish between I/Os sent by this initiator, and by other
11885 * initiators.  We set unit attention for initiators other than this one.
11886 * SAM-3 is vague on this point.  It does say that a unit attention should
11887 * be established for other initiators when a LUN is reset (see section
11888 * 5.7.3), but it doesn't specifically say that the unit attention should
11889 * be established for this particular initiator when a LUN is reset.  Here
11890 * is the relevant text, from SAM-3 rev 8:
11891 *
11892 * 5.7.2 When a SCSI initiator port aborts its own tasks
11893 *
11894 * When a SCSI initiator port causes its own task(s) to be aborted, no
11895 * notification that the task(s) have been aborted shall be returned to
11896 * the SCSI initiator port other than the completion response for the
11897 * command or task management function action that caused the task(s) to
11898 * be aborted and notification(s) associated with related effects of the
11899 * action (e.g., a reset unit attention condition).
11900 *
11901 * XXX KDM for now, we're setting unit attention for all initiators.
11902 */
11903static int
11904ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11905{
11906	union ctl_io *xio;
11907#if 0
11908	uint32_t initindex;
11909#endif
11910	int i;
11911
11912	mtx_lock(&lun->lun_lock);
11913	/*
11914	 * Run through the OOA queue and abort each I/O.
11915	 */
11916#if 0
11917	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11918#endif
11919	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11920	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11921		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11922	}
11923
11924	/*
11925	 * This version sets unit attention for every
11926	 */
11927#if 0
11928	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11929	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11930		if (initindex == i)
11931			continue;
11932		lun->pending_ua[i] |= ua_type;
11933	}
11934#endif
11935
11936	/*
11937	 * A reset (any kind, really) clears reservations established with
11938	 * RESERVE/RELEASE.  It does not clear reservations established
11939	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11940	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11941	 * reservations made with the RESERVE/RELEASE commands, because
11942	 * those commands are obsolete in SPC-3.
11943	 */
11944	lun->flags &= ~CTL_LUN_RESERVED;
11945
11946	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11947#ifdef CTL_WITH_CA
11948		ctl_clear_mask(lun->have_ca, i);
11949#endif
11950		lun->pending_ua[i] |= ua_type;
11951	}
11952	mtx_unlock(&lun->lun_lock);
11953
11954	return (0);
11955}
11956
11957static void
11958ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11959    int other_sc)
11960{
11961	union ctl_io *xio;
11962
11963	mtx_assert(&lun->lun_lock, MA_OWNED);
11964
11965	/*
11966	 * Run through the OOA queue and attempt to find the given I/O.
11967	 * The target port, initiator ID, tag type and tag number have to
11968	 * match the values that we got from the initiator.  If we have an
11969	 * untagged command to abort, simply abort the first untagged command
11970	 * we come to.  We only allow one untagged command at a time of course.
11971	 */
11972	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11973	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11974
11975		if ((targ_port == UINT32_MAX ||
11976		     targ_port == xio->io_hdr.nexus.targ_port) &&
11977		    (init_id == UINT32_MAX ||
11978		     init_id == xio->io_hdr.nexus.initid.id)) {
11979			if (targ_port != xio->io_hdr.nexus.targ_port ||
11980			    init_id != xio->io_hdr.nexus.initid.id)
11981				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11982			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11983			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11984				union ctl_ha_msg msg_info;
11985
11986				msg_info.hdr.nexus = xio->io_hdr.nexus;
11987				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11988				msg_info.task.tag_num = xio->scsiio.tag_num;
11989				msg_info.task.tag_type = xio->scsiio.tag_type;
11990				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11991				msg_info.hdr.original_sc = NULL;
11992				msg_info.hdr.serializing_sc = NULL;
11993				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11994				    (void *)&msg_info, sizeof(msg_info), 0);
11995			}
11996		}
11997	}
11998}
11999
12000static int
12001ctl_abort_task_set(union ctl_io *io)
12002{
12003	struct ctl_softc *softc = control_softc;
12004	struct ctl_lun *lun;
12005	uint32_t targ_lun;
12006
12007	/*
12008	 * Look up the LUN.
12009	 */
12010	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12011	mtx_lock(&softc->ctl_lock);
12012	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12013		lun = softc->ctl_luns[targ_lun];
12014	else {
12015		mtx_unlock(&softc->ctl_lock);
12016		return (1);
12017	}
12018
12019	mtx_lock(&lun->lun_lock);
12020	mtx_unlock(&softc->ctl_lock);
12021	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12022		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12023		    io->io_hdr.nexus.initid.id,
12024		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12025	} else { /* CTL_TASK_CLEAR_TASK_SET */
12026		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12027		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12028	}
12029	mtx_unlock(&lun->lun_lock);
12030	return (0);
12031}
12032
12033static int
12034ctl_i_t_nexus_reset(union ctl_io *io)
12035{
12036	struct ctl_softc *softc = control_softc;
12037	struct ctl_lun *lun;
12038	uint32_t initindex, residx;
12039
12040	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12041	residx = ctl_get_resindex(&io->io_hdr.nexus);
12042	mtx_lock(&softc->ctl_lock);
12043	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12044		mtx_lock(&lun->lun_lock);
12045		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12046		    io->io_hdr.nexus.initid.id,
12047		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12048#ifdef CTL_WITH_CA
12049		ctl_clear_mask(lun->have_ca, initindex);
12050#endif
12051		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12052			lun->flags &= ~CTL_LUN_RESERVED;
12053		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12054		mtx_unlock(&lun->lun_lock);
12055	}
12056	mtx_unlock(&softc->ctl_lock);
12057	return (0);
12058}
12059
12060static int
12061ctl_abort_task(union ctl_io *io)
12062{
12063	union ctl_io *xio;
12064	struct ctl_lun *lun;
12065	struct ctl_softc *ctl_softc;
12066#if 0
12067	struct sbuf sb;
12068	char printbuf[128];
12069#endif
12070	int found;
12071	uint32_t targ_lun;
12072
12073	ctl_softc = control_softc;
12074	found = 0;
12075
12076	/*
12077	 * Look up the LUN.
12078	 */
12079	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12080	mtx_lock(&ctl_softc->ctl_lock);
12081	if ((targ_lun < CTL_MAX_LUNS)
12082	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12083		lun = ctl_softc->ctl_luns[targ_lun];
12084	else {
12085		mtx_unlock(&ctl_softc->ctl_lock);
12086		return (1);
12087	}
12088
12089#if 0
12090	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12091	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12092#endif
12093
12094	mtx_lock(&lun->lun_lock);
12095	mtx_unlock(&ctl_softc->ctl_lock);
12096	/*
12097	 * Run through the OOA queue and attempt to find the given I/O.
12098	 * The target port, initiator ID, tag type and tag number have to
12099	 * match the values that we got from the initiator.  If we have an
12100	 * untagged command to abort, simply abort the first untagged command
12101	 * we come to.  We only allow one untagged command at a time of course.
12102	 */
12103#if 0
12104	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12105#endif
12106	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12107	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12108#if 0
12109		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12110
12111		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12112			    lun->lun, xio->scsiio.tag_num,
12113			    xio->scsiio.tag_type,
12114			    (xio->io_hdr.blocked_links.tqe_prev
12115			    == NULL) ? "" : " BLOCKED",
12116			    (xio->io_hdr.flags &
12117			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12118			    (xio->io_hdr.flags &
12119			    CTL_FLAG_ABORT) ? " ABORT" : "",
12120			    (xio->io_hdr.flags &
12121			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12122		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12123		sbuf_finish(&sb);
12124		printf("%s\n", sbuf_data(&sb));
12125#endif
12126
12127		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12128		 && (xio->io_hdr.nexus.initid.id ==
12129		     io->io_hdr.nexus.initid.id)) {
12130			/*
12131			 * If the abort says that the task is untagged, the
12132			 * task in the queue must be untagged.  Otherwise,
12133			 * we just check to see whether the tag numbers
12134			 * match.  This is because the QLogic firmware
12135			 * doesn't pass back the tag type in an abort
12136			 * request.
12137			 */
12138#if 0
12139			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12140			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12141			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12142#endif
12143			/*
12144			 * XXX KDM we've got problems with FC, because it
12145			 * doesn't send down a tag type with aborts.  So we
12146			 * can only really go by the tag number...
12147			 * This may cause problems with parallel SCSI.
12148			 * Need to figure that out!!
12149			 */
12150			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12151				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12152				found = 1;
12153				if ((io->io_hdr.flags &
12154				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12155				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12156					union ctl_ha_msg msg_info;
12157
12158					io->io_hdr.flags |=
12159					                CTL_FLAG_SENT_2OTHER_SC;
12160					msg_info.hdr.nexus = io->io_hdr.nexus;
12161					msg_info.task.task_action =
12162						CTL_TASK_ABORT_TASK;
12163					msg_info.task.tag_num =
12164						io->taskio.tag_num;
12165					msg_info.task.tag_type =
12166						io->taskio.tag_type;
12167					msg_info.hdr.msg_type =
12168						CTL_MSG_MANAGE_TASKS;
12169					msg_info.hdr.original_sc = NULL;
12170					msg_info.hdr.serializing_sc = NULL;
12171#if 0
12172					printf("Sent Abort to other side\n");
12173#endif
12174					if (CTL_HA_STATUS_SUCCESS !=
12175					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12176		    				(void *)&msg_info,
12177						sizeof(msg_info), 0)) {
12178					}
12179				}
12180#if 0
12181				printf("ctl_abort_task: found I/O to abort\n");
12182#endif
12183				break;
12184			}
12185		}
12186	}
12187	mtx_unlock(&lun->lun_lock);
12188
12189	if (found == 0) {
12190		/*
12191		 * This isn't really an error.  It's entirely possible for
12192		 * the abort and command completion to cross on the wire.
12193		 * This is more of an informative/diagnostic error.
12194		 */
12195#if 0
12196		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12197		       "%d:%d:%d:%d tag %d type %d\n",
12198		       io->io_hdr.nexus.initid.id,
12199		       io->io_hdr.nexus.targ_port,
12200		       io->io_hdr.nexus.targ_target.id,
12201		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12202		       io->taskio.tag_type);
12203#endif
12204	}
12205	return (0);
12206}
12207
12208static void
12209ctl_run_task(union ctl_io *io)
12210{
12211	struct ctl_softc *ctl_softc = control_softc;
12212	int retval = 1;
12213	const char *task_desc;
12214
12215	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12216
12217	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12218	    ("ctl_run_task: Unextected io_type %d\n",
12219	     io->io_hdr.io_type));
12220
12221	task_desc = ctl_scsi_task_string(&io->taskio);
12222	if (task_desc != NULL) {
12223#ifdef NEEDTOPORT
12224		csevent_log(CSC_CTL | CSC_SHELF_SW |
12225			    CTL_TASK_REPORT,
12226			    csevent_LogType_Trace,
12227			    csevent_Severity_Information,
12228			    csevent_AlertLevel_Green,
12229			    csevent_FRU_Firmware,
12230			    csevent_FRU_Unknown,
12231			    "CTL: received task: %s",task_desc);
12232#endif
12233	} else {
12234#ifdef NEEDTOPORT
12235		csevent_log(CSC_CTL | CSC_SHELF_SW |
12236			    CTL_TASK_REPORT,
12237			    csevent_LogType_Trace,
12238			    csevent_Severity_Information,
12239			    csevent_AlertLevel_Green,
12240			    csevent_FRU_Firmware,
12241			    csevent_FRU_Unknown,
12242			    "CTL: received unknown task "
12243			    "type: %d (%#x)",
12244			    io->taskio.task_action,
12245			    io->taskio.task_action);
12246#endif
12247	}
12248	switch (io->taskio.task_action) {
12249	case CTL_TASK_ABORT_TASK:
12250		retval = ctl_abort_task(io);
12251		break;
12252	case CTL_TASK_ABORT_TASK_SET:
12253	case CTL_TASK_CLEAR_TASK_SET:
12254		retval = ctl_abort_task_set(io);
12255		break;
12256	case CTL_TASK_CLEAR_ACA:
12257		break;
12258	case CTL_TASK_I_T_NEXUS_RESET:
12259		retval = ctl_i_t_nexus_reset(io);
12260		break;
12261	case CTL_TASK_LUN_RESET: {
12262		struct ctl_lun *lun;
12263		uint32_t targ_lun;
12264
12265		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12266		mtx_lock(&ctl_softc->ctl_lock);
12267		if ((targ_lun < CTL_MAX_LUNS)
12268		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12269			lun = ctl_softc->ctl_luns[targ_lun];
12270		else {
12271			mtx_unlock(&ctl_softc->ctl_lock);
12272			retval = 1;
12273			break;
12274		}
12275
12276		if (!(io->io_hdr.flags &
12277		    CTL_FLAG_FROM_OTHER_SC)) {
12278			union ctl_ha_msg msg_info;
12279
12280			io->io_hdr.flags |=
12281				CTL_FLAG_SENT_2OTHER_SC;
12282			msg_info.hdr.msg_type =
12283				CTL_MSG_MANAGE_TASKS;
12284			msg_info.hdr.nexus = io->io_hdr.nexus;
12285			msg_info.task.task_action =
12286				CTL_TASK_LUN_RESET;
12287			msg_info.hdr.original_sc = NULL;
12288			msg_info.hdr.serializing_sc = NULL;
12289			if (CTL_HA_STATUS_SUCCESS !=
12290			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12291			    (void *)&msg_info,
12292			    sizeof(msg_info), 0)) {
12293			}
12294		}
12295
12296		retval = ctl_lun_reset(lun, io,
12297				       CTL_UA_LUN_RESET);
12298		mtx_unlock(&ctl_softc->ctl_lock);
12299		break;
12300	}
12301	case CTL_TASK_TARGET_RESET:
12302		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12303		break;
12304	case CTL_TASK_BUS_RESET:
12305		retval = ctl_bus_reset(ctl_softc, io);
12306		break;
12307	case CTL_TASK_PORT_LOGIN:
12308		break;
12309	case CTL_TASK_PORT_LOGOUT:
12310		break;
12311	default:
12312		printf("ctl_run_task: got unknown task management event %d\n",
12313		       io->taskio.task_action);
12314		break;
12315	}
12316	if (retval == 0)
12317		io->io_hdr.status = CTL_SUCCESS;
12318	else
12319		io->io_hdr.status = CTL_ERROR;
12320	ctl_done(io);
12321}
12322
12323/*
12324 * For HA operation.  Handle commands that come in from the other
12325 * controller.
12326 */
12327static void
12328ctl_handle_isc(union ctl_io *io)
12329{
12330	int free_io;
12331	struct ctl_lun *lun;
12332	struct ctl_softc *ctl_softc;
12333	uint32_t targ_lun;
12334
12335	ctl_softc = control_softc;
12336
12337	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12338	lun = ctl_softc->ctl_luns[targ_lun];
12339
12340	switch (io->io_hdr.msg_type) {
12341	case CTL_MSG_SERIALIZE:
12342		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12343		break;
12344	case CTL_MSG_R2R: {
12345		const struct ctl_cmd_entry *entry;
12346
12347		/*
12348		 * This is only used in SER_ONLY mode.
12349		 */
12350		free_io = 0;
12351		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12352		mtx_lock(&lun->lun_lock);
12353		if (ctl_scsiio_lun_check(ctl_softc, lun,
12354		    entry, (struct ctl_scsiio *)io) != 0) {
12355			mtx_unlock(&lun->lun_lock);
12356			ctl_done(io);
12357			break;
12358		}
12359		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12360		mtx_unlock(&lun->lun_lock);
12361		ctl_enqueue_rtr(io);
12362		break;
12363	}
12364	case CTL_MSG_FINISH_IO:
12365		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12366			free_io = 0;
12367			ctl_done(io);
12368		} else {
12369			free_io = 1;
12370			mtx_lock(&lun->lun_lock);
12371			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12372				     ooa_links);
12373			ctl_check_blocked(lun);
12374			mtx_unlock(&lun->lun_lock);
12375		}
12376		break;
12377	case CTL_MSG_PERS_ACTION:
12378		ctl_hndl_per_res_out_on_other_sc(
12379			(union ctl_ha_msg *)&io->presio.pr_msg);
12380		free_io = 1;
12381		break;
12382	case CTL_MSG_BAD_JUJU:
12383		free_io = 0;
12384		ctl_done(io);
12385		break;
12386	case CTL_MSG_DATAMOVE:
12387		/* Only used in XFER mode */
12388		free_io = 0;
12389		ctl_datamove_remote(io);
12390		break;
12391	case CTL_MSG_DATAMOVE_DONE:
12392		/* Only used in XFER mode */
12393		free_io = 0;
12394		io->scsiio.be_move_done(io);
12395		break;
12396	default:
12397		free_io = 1;
12398		printf("%s: Invalid message type %d\n",
12399		       __func__, io->io_hdr.msg_type);
12400		break;
12401	}
12402	if (free_io)
12403		ctl_free_io(io);
12404
12405}
12406
12407
12408/*
12409 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12410 * there is no match.
12411 */
12412static ctl_lun_error_pattern
12413ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12414{
12415	const struct ctl_cmd_entry *entry;
12416	ctl_lun_error_pattern filtered_pattern, pattern;
12417
12418	pattern = desc->error_pattern;
12419
12420	/*
12421	 * XXX KDM we need more data passed into this function to match a
12422	 * custom pattern, and we actually need to implement custom pattern
12423	 * matching.
12424	 */
12425	if (pattern & CTL_LUN_PAT_CMD)
12426		return (CTL_LUN_PAT_CMD);
12427
12428	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12429		return (CTL_LUN_PAT_ANY);
12430
12431	entry = ctl_get_cmd_entry(ctsio, NULL);
12432
12433	filtered_pattern = entry->pattern & pattern;
12434
12435	/*
12436	 * If the user requested specific flags in the pattern (e.g.
12437	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12438	 * flags.
12439	 *
12440	 * If the user did not specify any flags, it doesn't matter whether
12441	 * or not the command supports the flags.
12442	 */
12443	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12444	     (pattern & ~CTL_LUN_PAT_MASK))
12445		return (CTL_LUN_PAT_NONE);
12446
12447	/*
12448	 * If the user asked for a range check, see if the requested LBA
12449	 * range overlaps with this command's LBA range.
12450	 */
12451	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12452		uint64_t lba1;
12453		uint64_t len1;
12454		ctl_action action;
12455		int retval;
12456
12457		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12458		if (retval != 0)
12459			return (CTL_LUN_PAT_NONE);
12460
12461		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12462					      desc->lba_range.len);
12463		/*
12464		 * A "pass" means that the LBA ranges don't overlap, so
12465		 * this doesn't match the user's range criteria.
12466		 */
12467		if (action == CTL_ACTION_PASS)
12468			return (CTL_LUN_PAT_NONE);
12469	}
12470
12471	return (filtered_pattern);
12472}
12473
12474static void
12475ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12476{
12477	struct ctl_error_desc *desc, *desc2;
12478
12479	mtx_assert(&lun->lun_lock, MA_OWNED);
12480
12481	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12482		ctl_lun_error_pattern pattern;
12483		/*
12484		 * Check to see whether this particular command matches
12485		 * the pattern in the descriptor.
12486		 */
12487		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12488		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12489			continue;
12490
12491		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12492		case CTL_LUN_INJ_ABORTED:
12493			ctl_set_aborted(&io->scsiio);
12494			break;
12495		case CTL_LUN_INJ_MEDIUM_ERR:
12496			ctl_set_medium_error(&io->scsiio);
12497			break;
12498		case CTL_LUN_INJ_UA:
12499			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12500			 * OCCURRED */
12501			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12502			break;
12503		case CTL_LUN_INJ_CUSTOM:
12504			/*
12505			 * We're assuming the user knows what he is doing.
12506			 * Just copy the sense information without doing
12507			 * checks.
12508			 */
12509			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12510			      ctl_min(sizeof(desc->custom_sense),
12511				      sizeof(io->scsiio.sense_data)));
12512			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12513			io->scsiio.sense_len = SSD_FULL_SIZE;
12514			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12515			break;
12516		case CTL_LUN_INJ_NONE:
12517		default:
12518			/*
12519			 * If this is an error injection type we don't know
12520			 * about, clear the continuous flag (if it is set)
12521			 * so it will get deleted below.
12522			 */
12523			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12524			break;
12525		}
12526		/*
12527		 * By default, each error injection action is a one-shot
12528		 */
12529		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12530			continue;
12531
12532		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12533
12534		free(desc, M_CTL);
12535	}
12536}
12537
12538#ifdef CTL_IO_DELAY
12539static void
12540ctl_datamove_timer_wakeup(void *arg)
12541{
12542	union ctl_io *io;
12543
12544	io = (union ctl_io *)arg;
12545
12546	ctl_datamove(io);
12547}
12548#endif /* CTL_IO_DELAY */
12549
12550void
12551ctl_datamove(union ctl_io *io)
12552{
12553	void (*fe_datamove)(union ctl_io *io);
12554
12555	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12556
12557	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12558
12559#ifdef CTL_TIME_IO
12560	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12561		char str[256];
12562		char path_str[64];
12563		struct sbuf sb;
12564
12565		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12566		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12567
12568		sbuf_cat(&sb, path_str);
12569		switch (io->io_hdr.io_type) {
12570		case CTL_IO_SCSI:
12571			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12572			sbuf_printf(&sb, "\n");
12573			sbuf_cat(&sb, path_str);
12574			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12575				    io->scsiio.tag_num, io->scsiio.tag_type);
12576			break;
12577		case CTL_IO_TASK:
12578			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12579				    "Tag Type: %d\n", io->taskio.task_action,
12580				    io->taskio.tag_num, io->taskio.tag_type);
12581			break;
12582		default:
12583			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12584			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12585			break;
12586		}
12587		sbuf_cat(&sb, path_str);
12588		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12589			    (intmax_t)time_uptime - io->io_hdr.start_time);
12590		sbuf_finish(&sb);
12591		printf("%s", sbuf_data(&sb));
12592	}
12593#endif /* CTL_TIME_IO */
12594
12595#ifdef CTL_IO_DELAY
12596	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12597		struct ctl_lun *lun;
12598
12599		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12600
12601		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12602	} else {
12603		struct ctl_lun *lun;
12604
12605		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12606		if ((lun != NULL)
12607		 && (lun->delay_info.datamove_delay > 0)) {
12608			struct callout *callout;
12609
12610			callout = (struct callout *)&io->io_hdr.timer_bytes;
12611			callout_init(callout, /*mpsafe*/ 1);
12612			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12613			callout_reset(callout,
12614				      lun->delay_info.datamove_delay * hz,
12615				      ctl_datamove_timer_wakeup, io);
12616			if (lun->delay_info.datamove_type ==
12617			    CTL_DELAY_TYPE_ONESHOT)
12618				lun->delay_info.datamove_delay = 0;
12619			return;
12620		}
12621	}
12622#endif
12623
12624	/*
12625	 * This command has been aborted.  Set the port status, so we fail
12626	 * the data move.
12627	 */
12628	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12629		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12630		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12631		       io->io_hdr.nexus.targ_port,
12632		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12633		       io->io_hdr.nexus.targ_lun);
12634		io->io_hdr.port_status = 31337;
12635		/*
12636		 * Note that the backend, in this case, will get the
12637		 * callback in its context.  In other cases it may get
12638		 * called in the frontend's interrupt thread context.
12639		 */
12640		io->scsiio.be_move_done(io);
12641		return;
12642	}
12643
12644	/* Don't confuse frontend with zero length data move. */
12645	if (io->scsiio.kern_data_len == 0) {
12646		io->scsiio.be_move_done(io);
12647		return;
12648	}
12649
12650	/*
12651	 * If we're in XFER mode and this I/O is from the other shelf
12652	 * controller, we need to send the DMA to the other side to
12653	 * actually transfer the data to/from the host.  In serialize only
12654	 * mode the transfer happens below CTL and ctl_datamove() is only
12655	 * called on the machine that originally received the I/O.
12656	 */
12657	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12658	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12659		union ctl_ha_msg msg;
12660		uint32_t sg_entries_sent;
12661		int do_sg_copy;
12662		int i;
12663
12664		memset(&msg, 0, sizeof(msg));
12665		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12666		msg.hdr.original_sc = io->io_hdr.original_sc;
12667		msg.hdr.serializing_sc = io;
12668		msg.hdr.nexus = io->io_hdr.nexus;
12669		msg.dt.flags = io->io_hdr.flags;
12670		/*
12671		 * We convert everything into a S/G list here.  We can't
12672		 * pass by reference, only by value between controllers.
12673		 * So we can't pass a pointer to the S/G list, only as many
12674		 * S/G entries as we can fit in here.  If it's possible for
12675		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12676		 * then we need to break this up into multiple transfers.
12677		 */
12678		if (io->scsiio.kern_sg_entries == 0) {
12679			msg.dt.kern_sg_entries = 1;
12680			/*
12681			 * If this is in cached memory, flush the cache
12682			 * before we send the DMA request to the other
12683			 * controller.  We want to do this in either the
12684			 * read or the write case.  The read case is
12685			 * straightforward.  In the write case, we want to
12686			 * make sure nothing is in the local cache that
12687			 * could overwrite the DMAed data.
12688			 */
12689			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12690				/*
12691				 * XXX KDM use bus_dmamap_sync() here.
12692				 */
12693			}
12694
12695			/*
12696			 * Convert to a physical address if this is a
12697			 * virtual address.
12698			 */
12699			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12700				msg.dt.sg_list[0].addr =
12701					io->scsiio.kern_data_ptr;
12702			} else {
12703				/*
12704				 * XXX KDM use busdma here!
12705				 */
12706#if 0
12707				msg.dt.sg_list[0].addr = (void *)
12708					vtophys(io->scsiio.kern_data_ptr);
12709#endif
12710			}
12711
12712			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12713			do_sg_copy = 0;
12714		} else {
12715			struct ctl_sg_entry *sgl;
12716
12717			do_sg_copy = 1;
12718			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12719			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12720			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12721				/*
12722				 * XXX KDM use bus_dmamap_sync() here.
12723				 */
12724			}
12725		}
12726
12727		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12728		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12729		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12730		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12731		msg.dt.sg_sequence = 0;
12732
12733		/*
12734		 * Loop until we've sent all of the S/G entries.  On the
12735		 * other end, we'll recompose these S/G entries into one
12736		 * contiguous list before passing it to the
12737		 */
12738		for (sg_entries_sent = 0; sg_entries_sent <
12739		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12740			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12741				sizeof(msg.dt.sg_list[0])),
12742				msg.dt.kern_sg_entries - sg_entries_sent);
12743
12744			if (do_sg_copy != 0) {
12745				struct ctl_sg_entry *sgl;
12746				int j;
12747
12748				sgl = (struct ctl_sg_entry *)
12749					io->scsiio.kern_data_ptr;
12750				/*
12751				 * If this is in cached memory, flush the cache
12752				 * before we send the DMA request to the other
12753				 * controller.  We want to do this in either
12754				 * the * read or the write case.  The read
12755				 * case is straightforward.  In the write
12756				 * case, we want to make sure nothing is
12757				 * in the local cache that could overwrite
12758				 * the DMAed data.
12759				 */
12760
12761				for (i = sg_entries_sent, j = 0;
12762				     i < msg.dt.cur_sg_entries; i++, j++) {
12763					if ((io->io_hdr.flags &
12764					     CTL_FLAG_NO_DATASYNC) == 0) {
12765						/*
12766						 * XXX KDM use bus_dmamap_sync()
12767						 */
12768					}
12769					if ((io->io_hdr.flags &
12770					     CTL_FLAG_BUS_ADDR) == 0) {
12771						/*
12772						 * XXX KDM use busdma.
12773						 */
12774#if 0
12775						msg.dt.sg_list[j].addr =(void *)
12776						       vtophys(sgl[i].addr);
12777#endif
12778					} else {
12779						msg.dt.sg_list[j].addr =
12780							sgl[i].addr;
12781					}
12782					msg.dt.sg_list[j].len = sgl[i].len;
12783				}
12784			}
12785
12786			sg_entries_sent += msg.dt.cur_sg_entries;
12787			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12788				msg.dt.sg_last = 1;
12789			else
12790				msg.dt.sg_last = 0;
12791
12792			/*
12793			 * XXX KDM drop and reacquire the lock here?
12794			 */
12795			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12796			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12797				/*
12798				 * XXX do something here.
12799				 */
12800			}
12801
12802			msg.dt.sent_sg_entries = sg_entries_sent;
12803		}
12804		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12805		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12806			ctl_failover_io(io, /*have_lock*/ 0);
12807
12808	} else {
12809
12810		/*
12811		 * Lookup the fe_datamove() function for this particular
12812		 * front end.
12813		 */
12814		fe_datamove =
12815		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12816
12817		fe_datamove(io);
12818	}
12819}
12820
12821static void
12822ctl_send_datamove_done(union ctl_io *io, int have_lock)
12823{
12824	union ctl_ha_msg msg;
12825	int isc_status;
12826
12827	memset(&msg, 0, sizeof(msg));
12828
12829	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12830	msg.hdr.original_sc = io;
12831	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12832	msg.hdr.nexus = io->io_hdr.nexus;
12833	msg.hdr.status = io->io_hdr.status;
12834	msg.scsi.tag_num = io->scsiio.tag_num;
12835	msg.scsi.tag_type = io->scsiio.tag_type;
12836	msg.scsi.scsi_status = io->scsiio.scsi_status;
12837	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12838	       sizeof(io->scsiio.sense_data));
12839	msg.scsi.sense_len = io->scsiio.sense_len;
12840	msg.scsi.sense_residual = io->scsiio.sense_residual;
12841	msg.scsi.fetd_status = io->io_hdr.port_status;
12842	msg.scsi.residual = io->scsiio.residual;
12843	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12844
12845	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12846		ctl_failover_io(io, /*have_lock*/ have_lock);
12847		return;
12848	}
12849
12850	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12851	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12852		/* XXX do something if this fails */
12853	}
12854
12855}
12856
12857/*
12858 * The DMA to the remote side is done, now we need to tell the other side
12859 * we're done so it can continue with its data movement.
12860 */
12861static void
12862ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12863{
12864	union ctl_io *io;
12865
12866	io = rq->context;
12867
12868	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12869		printf("%s: ISC DMA write failed with error %d", __func__,
12870		       rq->ret);
12871		ctl_set_internal_failure(&io->scsiio,
12872					 /*sks_valid*/ 1,
12873					 /*retry_count*/ rq->ret);
12874	}
12875
12876	ctl_dt_req_free(rq);
12877
12878	/*
12879	 * In this case, we had to malloc the memory locally.  Free it.
12880	 */
12881	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12882		int i;
12883		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12884			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12885	}
12886	/*
12887	 * The data is in local and remote memory, so now we need to send
12888	 * status (good or back) back to the other side.
12889	 */
12890	ctl_send_datamove_done(io, /*have_lock*/ 0);
12891}
12892
12893/*
12894 * We've moved the data from the host/controller into local memory.  Now we
12895 * need to push it over to the remote controller's memory.
12896 */
12897static int
12898ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12899{
12900	int retval;
12901
12902	retval = 0;
12903
12904	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12905					  ctl_datamove_remote_write_cb);
12906
12907	return (retval);
12908}
12909
12910static void
12911ctl_datamove_remote_write(union ctl_io *io)
12912{
12913	int retval;
12914	void (*fe_datamove)(union ctl_io *io);
12915
12916	/*
12917	 * - Get the data from the host/HBA into local memory.
12918	 * - DMA memory from the local controller to the remote controller.
12919	 * - Send status back to the remote controller.
12920	 */
12921
12922	retval = ctl_datamove_remote_sgl_setup(io);
12923	if (retval != 0)
12924		return;
12925
12926	/* Switch the pointer over so the FETD knows what to do */
12927	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12928
12929	/*
12930	 * Use a custom move done callback, since we need to send completion
12931	 * back to the other controller, not to the backend on this side.
12932	 */
12933	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12934
12935	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12936
12937	fe_datamove(io);
12938
12939	return;
12940
12941}
12942
12943static int
12944ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12945{
12946#if 0
12947	char str[256];
12948	char path_str[64];
12949	struct sbuf sb;
12950#endif
12951
12952	/*
12953	 * In this case, we had to malloc the memory locally.  Free it.
12954	 */
12955	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12956		int i;
12957		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12958			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12959	}
12960
12961#if 0
12962	scsi_path_string(io, path_str, sizeof(path_str));
12963	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12964	sbuf_cat(&sb, path_str);
12965	scsi_command_string(&io->scsiio, NULL, &sb);
12966	sbuf_printf(&sb, "\n");
12967	sbuf_cat(&sb, path_str);
12968	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12969		    io->scsiio.tag_num, io->scsiio.tag_type);
12970	sbuf_cat(&sb, path_str);
12971	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12972		    io->io_hdr.flags, io->io_hdr.status);
12973	sbuf_finish(&sb);
12974	printk("%s", sbuf_data(&sb));
12975#endif
12976
12977
12978	/*
12979	 * The read is done, now we need to send status (good or bad) back
12980	 * to the other side.
12981	 */
12982	ctl_send_datamove_done(io, /*have_lock*/ 0);
12983
12984	return (0);
12985}
12986
12987static void
12988ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12989{
12990	union ctl_io *io;
12991	void (*fe_datamove)(union ctl_io *io);
12992
12993	io = rq->context;
12994
12995	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12996		printf("%s: ISC DMA read failed with error %d", __func__,
12997		       rq->ret);
12998		ctl_set_internal_failure(&io->scsiio,
12999					 /*sks_valid*/ 1,
13000					 /*retry_count*/ rq->ret);
13001	}
13002
13003	ctl_dt_req_free(rq);
13004
13005	/* Switch the pointer over so the FETD knows what to do */
13006	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13007
13008	/*
13009	 * Use a custom move done callback, since we need to send completion
13010	 * back to the other controller, not to the backend on this side.
13011	 */
13012	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13013
13014	/* XXX KDM add checks like the ones in ctl_datamove? */
13015
13016	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13017
13018	fe_datamove(io);
13019}
13020
13021static int
13022ctl_datamove_remote_sgl_setup(union ctl_io *io)
13023{
13024	struct ctl_sg_entry *local_sglist, *remote_sglist;
13025	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13026	struct ctl_softc *softc;
13027	int retval;
13028	int i;
13029
13030	retval = 0;
13031	softc = control_softc;
13032
13033	local_sglist = io->io_hdr.local_sglist;
13034	local_dma_sglist = io->io_hdr.local_dma_sglist;
13035	remote_sglist = io->io_hdr.remote_sglist;
13036	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13037
13038	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13039		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13040			local_sglist[i].len = remote_sglist[i].len;
13041
13042			/*
13043			 * XXX Detect the situation where the RS-level I/O
13044			 * redirector on the other side has already read the
13045			 * data off of the AOR RS on this side, and
13046			 * transferred it to remote (mirror) memory on the
13047			 * other side.  Since we already have the data in
13048			 * memory here, we just need to use it.
13049			 *
13050			 * XXX KDM this can probably be removed once we
13051			 * get the cache device code in and take the
13052			 * current AOR implementation out.
13053			 */
13054#ifdef NEEDTOPORT
13055			if ((remote_sglist[i].addr >=
13056			     (void *)vtophys(softc->mirr->addr))
13057			 && (remote_sglist[i].addr <
13058			     ((void *)vtophys(softc->mirr->addr) +
13059			     CacheMirrorOffset))) {
13060				local_sglist[i].addr = remote_sglist[i].addr -
13061					CacheMirrorOffset;
13062				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13063				     CTL_FLAG_DATA_IN)
13064					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13065			} else {
13066				local_sglist[i].addr = remote_sglist[i].addr +
13067					CacheMirrorOffset;
13068			}
13069#endif
13070#if 0
13071			printf("%s: local %p, remote %p, len %d\n",
13072			       __func__, local_sglist[i].addr,
13073			       remote_sglist[i].addr, local_sglist[i].len);
13074#endif
13075		}
13076	} else {
13077		uint32_t len_to_go;
13078
13079		/*
13080		 * In this case, we don't have automatically allocated
13081		 * memory for this I/O on this controller.  This typically
13082		 * happens with internal CTL I/O -- e.g. inquiry, mode
13083		 * sense, etc.  Anything coming from RAIDCore will have
13084		 * a mirror area available.
13085		 */
13086		len_to_go = io->scsiio.kern_data_len;
13087
13088		/*
13089		 * Clear the no datasync flag, we have to use malloced
13090		 * buffers.
13091		 */
13092		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13093
13094		/*
13095		 * The difficult thing here is that the size of the various
13096		 * S/G segments may be different than the size from the
13097		 * remote controller.  That'll make it harder when DMAing
13098		 * the data back to the other side.
13099		 */
13100		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13101		     sizeof(io->io_hdr.remote_sglist[0])) &&
13102		     (len_to_go > 0); i++) {
13103			local_sglist[i].len = ctl_min(len_to_go, 131072);
13104			CTL_SIZE_8B(local_dma_sglist[i].len,
13105				    local_sglist[i].len);
13106			local_sglist[i].addr =
13107				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13108
13109			local_dma_sglist[i].addr = local_sglist[i].addr;
13110
13111			if (local_sglist[i].addr == NULL) {
13112				int j;
13113
13114				printf("malloc failed for %zd bytes!",
13115				       local_dma_sglist[i].len);
13116				for (j = 0; j < i; j++) {
13117					free(local_sglist[j].addr, M_CTL);
13118				}
13119				ctl_set_internal_failure(&io->scsiio,
13120							 /*sks_valid*/ 1,
13121							 /*retry_count*/ 4857);
13122				retval = 1;
13123				goto bailout_error;
13124
13125			}
13126			/* XXX KDM do we need a sync here? */
13127
13128			len_to_go -= local_sglist[i].len;
13129		}
13130		/*
13131		 * Reset the number of S/G entries accordingly.  The
13132		 * original number of S/G entries is available in
13133		 * rem_sg_entries.
13134		 */
13135		io->scsiio.kern_sg_entries = i;
13136
13137#if 0
13138		printf("%s: kern_sg_entries = %d\n", __func__,
13139		       io->scsiio.kern_sg_entries);
13140		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13141			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13142			       local_sglist[i].addr, local_sglist[i].len,
13143			       local_dma_sglist[i].len);
13144#endif
13145	}
13146
13147
13148	return (retval);
13149
13150bailout_error:
13151
13152	ctl_send_datamove_done(io, /*have_lock*/ 0);
13153
13154	return (retval);
13155}
13156
13157static int
13158ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13159			 ctl_ha_dt_cb callback)
13160{
13161	struct ctl_ha_dt_req *rq;
13162	struct ctl_sg_entry *remote_sglist, *local_sglist;
13163	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13164	uint32_t local_used, remote_used, total_used;
13165	int retval;
13166	int i, j;
13167
13168	retval = 0;
13169
13170	rq = ctl_dt_req_alloc();
13171
13172	/*
13173	 * If we failed to allocate the request, and if the DMA didn't fail
13174	 * anyway, set busy status.  This is just a resource allocation
13175	 * failure.
13176	 */
13177	if ((rq == NULL)
13178	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13179		ctl_set_busy(&io->scsiio);
13180
13181	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13182
13183		if (rq != NULL)
13184			ctl_dt_req_free(rq);
13185
13186		/*
13187		 * The data move failed.  We need to return status back
13188		 * to the other controller.  No point in trying to DMA
13189		 * data to the remote controller.
13190		 */
13191
13192		ctl_send_datamove_done(io, /*have_lock*/ 0);
13193
13194		retval = 1;
13195
13196		goto bailout;
13197	}
13198
13199	local_sglist = io->io_hdr.local_sglist;
13200	local_dma_sglist = io->io_hdr.local_dma_sglist;
13201	remote_sglist = io->io_hdr.remote_sglist;
13202	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13203	local_used = 0;
13204	remote_used = 0;
13205	total_used = 0;
13206
13207	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13208		rq->ret = CTL_HA_STATUS_SUCCESS;
13209		rq->context = io;
13210		callback(rq);
13211		goto bailout;
13212	}
13213
13214	/*
13215	 * Pull/push the data over the wire from/to the other controller.
13216	 * This takes into account the possibility that the local and
13217	 * remote sglists may not be identical in terms of the size of
13218	 * the elements and the number of elements.
13219	 *
13220	 * One fundamental assumption here is that the length allocated for
13221	 * both the local and remote sglists is identical.  Otherwise, we've
13222	 * essentially got a coding error of some sort.
13223	 */
13224	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13225		int isc_ret;
13226		uint32_t cur_len, dma_length;
13227		uint8_t *tmp_ptr;
13228
13229		rq->id = CTL_HA_DATA_CTL;
13230		rq->command = command;
13231		rq->context = io;
13232
13233		/*
13234		 * Both pointers should be aligned.  But it is possible
13235		 * that the allocation length is not.  They should both
13236		 * also have enough slack left over at the end, though,
13237		 * to round up to the next 8 byte boundary.
13238		 */
13239		cur_len = ctl_min(local_sglist[i].len - local_used,
13240				  remote_sglist[j].len - remote_used);
13241
13242		/*
13243		 * In this case, we have a size issue and need to decrease
13244		 * the size, except in the case where we actually have less
13245		 * than 8 bytes left.  In that case, we need to increase
13246		 * the DMA length to get the last bit.
13247		 */
13248		if ((cur_len & 0x7) != 0) {
13249			if (cur_len > 0x7) {
13250				cur_len = cur_len - (cur_len & 0x7);
13251				dma_length = cur_len;
13252			} else {
13253				CTL_SIZE_8B(dma_length, cur_len);
13254			}
13255
13256		} else
13257			dma_length = cur_len;
13258
13259		/*
13260		 * If we had to allocate memory for this I/O, instead of using
13261		 * the non-cached mirror memory, we'll need to flush the cache
13262		 * before trying to DMA to the other controller.
13263		 *
13264		 * We could end up doing this multiple times for the same
13265		 * segment if we have a larger local segment than remote
13266		 * segment.  That shouldn't be an issue.
13267		 */
13268		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13269			/*
13270			 * XXX KDM use bus_dmamap_sync() here.
13271			 */
13272		}
13273
13274		rq->size = dma_length;
13275
13276		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13277		tmp_ptr += local_used;
13278
13279		/* Use physical addresses when talking to ISC hardware */
13280		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13281			/* XXX KDM use busdma */
13282#if 0
13283			rq->local = vtophys(tmp_ptr);
13284#endif
13285		} else
13286			rq->local = tmp_ptr;
13287
13288		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13289		tmp_ptr += remote_used;
13290		rq->remote = tmp_ptr;
13291
13292		rq->callback = NULL;
13293
13294		local_used += cur_len;
13295		if (local_used >= local_sglist[i].len) {
13296			i++;
13297			local_used = 0;
13298		}
13299
13300		remote_used += cur_len;
13301		if (remote_used >= remote_sglist[j].len) {
13302			j++;
13303			remote_used = 0;
13304		}
13305		total_used += cur_len;
13306
13307		if (total_used >= io->scsiio.kern_data_len)
13308			rq->callback = callback;
13309
13310		if ((rq->size & 0x7) != 0) {
13311			printf("%s: warning: size %d is not on 8b boundary\n",
13312			       __func__, rq->size);
13313		}
13314		if (((uintptr_t)rq->local & 0x7) != 0) {
13315			printf("%s: warning: local %p not on 8b boundary\n",
13316			       __func__, rq->local);
13317		}
13318		if (((uintptr_t)rq->remote & 0x7) != 0) {
13319			printf("%s: warning: remote %p not on 8b boundary\n",
13320			       __func__, rq->local);
13321		}
13322#if 0
13323		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13324		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13325		       rq->local, rq->remote, rq->size);
13326#endif
13327
13328		isc_ret = ctl_dt_single(rq);
13329		if (isc_ret == CTL_HA_STATUS_WAIT)
13330			continue;
13331
13332		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13333			rq->ret = CTL_HA_STATUS_SUCCESS;
13334		} else {
13335			rq->ret = isc_ret;
13336		}
13337		callback(rq);
13338		goto bailout;
13339	}
13340
13341bailout:
13342	return (retval);
13343
13344}
13345
13346static void
13347ctl_datamove_remote_read(union ctl_io *io)
13348{
13349	int retval;
13350	int i;
13351
13352	/*
13353	 * This will send an error to the other controller in the case of a
13354	 * failure.
13355	 */
13356	retval = ctl_datamove_remote_sgl_setup(io);
13357	if (retval != 0)
13358		return;
13359
13360	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13361					  ctl_datamove_remote_read_cb);
13362	if ((retval != 0)
13363	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13364		/*
13365		 * Make sure we free memory if there was an error..  The
13366		 * ctl_datamove_remote_xfer() function will send the
13367		 * datamove done message, or call the callback with an
13368		 * error if there is a problem.
13369		 */
13370		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13371			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13372	}
13373
13374	return;
13375}
13376
13377/*
13378 * Process a datamove request from the other controller.  This is used for
13379 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13380 * first.  Once that is complete, the data gets DMAed into the remote
13381 * controller's memory.  For reads, we DMA from the remote controller's
13382 * memory into our memory first, and then move it out to the FETD.
13383 */
13384static void
13385ctl_datamove_remote(union ctl_io *io)
13386{
13387	struct ctl_softc *softc;
13388
13389	softc = control_softc;
13390
13391	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13392
13393	/*
13394	 * Note that we look for an aborted I/O here, but don't do some of
13395	 * the other checks that ctl_datamove() normally does.
13396	 * We don't need to run the datamove delay code, since that should
13397	 * have been done if need be on the other controller.
13398	 */
13399	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13400		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13401		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13402		       io->io_hdr.nexus.targ_port,
13403		       io->io_hdr.nexus.targ_target.id,
13404		       io->io_hdr.nexus.targ_lun);
13405		io->io_hdr.port_status = 31338;
13406		ctl_send_datamove_done(io, /*have_lock*/ 0);
13407		return;
13408	}
13409
13410	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13411		ctl_datamove_remote_write(io);
13412	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13413		ctl_datamove_remote_read(io);
13414	} else {
13415		union ctl_ha_msg msg;
13416		struct scsi_sense_data *sense;
13417		uint8_t sks[3];
13418		int retry_count;
13419
13420		memset(&msg, 0, sizeof(msg));
13421
13422		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13423		msg.hdr.status = CTL_SCSI_ERROR;
13424		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13425
13426		retry_count = 4243;
13427
13428		sense = &msg.scsi.sense_data;
13429		sks[0] = SSD_SCS_VALID;
13430		sks[1] = (retry_count >> 8) & 0xff;
13431		sks[2] = retry_count & 0xff;
13432
13433		/* "Internal target failure" */
13434		scsi_set_sense_data(sense,
13435				    /*sense_format*/ SSD_TYPE_NONE,
13436				    /*current_error*/ 1,
13437				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13438				    /*asc*/ 0x44,
13439				    /*ascq*/ 0x00,
13440				    /*type*/ SSD_ELEM_SKS,
13441				    /*size*/ sizeof(sks),
13442				    /*data*/ sks,
13443				    SSD_ELEM_NONE);
13444
13445		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13446		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13447			ctl_failover_io(io, /*have_lock*/ 1);
13448			return;
13449		}
13450
13451		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13452		    CTL_HA_STATUS_SUCCESS) {
13453			/* XXX KDM what to do if this fails? */
13454		}
13455		return;
13456	}
13457
13458}
13459
13460static int
13461ctl_process_done(union ctl_io *io)
13462{
13463	struct ctl_lun *lun;
13464	struct ctl_softc *ctl_softc;
13465	void (*fe_done)(union ctl_io *io);
13466	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13467
13468	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13469
13470	fe_done =
13471	    control_softc->ctl_ports[targ_port]->fe_done;
13472
13473#ifdef CTL_TIME_IO
13474	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13475		char str[256];
13476		char path_str[64];
13477		struct sbuf sb;
13478
13479		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13480		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13481
13482		sbuf_cat(&sb, path_str);
13483		switch (io->io_hdr.io_type) {
13484		case CTL_IO_SCSI:
13485			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13486			sbuf_printf(&sb, "\n");
13487			sbuf_cat(&sb, path_str);
13488			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13489				    io->scsiio.tag_num, io->scsiio.tag_type);
13490			break;
13491		case CTL_IO_TASK:
13492			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13493				    "Tag Type: %d\n", io->taskio.task_action,
13494				    io->taskio.tag_num, io->taskio.tag_type);
13495			break;
13496		default:
13497			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13498			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13499			break;
13500		}
13501		sbuf_cat(&sb, path_str);
13502		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13503			    (intmax_t)time_uptime - io->io_hdr.start_time);
13504		sbuf_finish(&sb);
13505		printf("%s", sbuf_data(&sb));
13506	}
13507#endif /* CTL_TIME_IO */
13508
13509	switch (io->io_hdr.io_type) {
13510	case CTL_IO_SCSI:
13511		break;
13512	case CTL_IO_TASK:
13513		if (bootverbose || (ctl_debug & CTL_DEBUG_INFO))
13514			ctl_io_error_print(io, NULL);
13515		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13516			ctl_free_io(io);
13517		else
13518			fe_done(io);
13519		return (CTL_RETVAL_COMPLETE);
13520	default:
13521		panic("ctl_process_done: invalid io type %d\n",
13522		      io->io_hdr.io_type);
13523		break; /* NOTREACHED */
13524	}
13525
13526	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13527	if (lun == NULL) {
13528		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13529				 io->io_hdr.nexus.targ_mapped_lun));
13530		fe_done(io);
13531		goto bailout;
13532	}
13533	ctl_softc = lun->ctl_softc;
13534
13535	mtx_lock(&lun->lun_lock);
13536
13537	/*
13538	 * Check to see if we have any errors to inject here.  We only
13539	 * inject errors for commands that don't already have errors set.
13540	 */
13541	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13542	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13543		ctl_inject_error(lun, io);
13544
13545	/*
13546	 * XXX KDM how do we treat commands that aren't completed
13547	 * successfully?
13548	 *
13549	 * XXX KDM should we also track I/O latency?
13550	 */
13551	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13552	    io->io_hdr.io_type == CTL_IO_SCSI) {
13553#ifdef CTL_TIME_IO
13554		struct bintime cur_bt;
13555#endif
13556		int type;
13557
13558		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13559		    CTL_FLAG_DATA_IN)
13560			type = CTL_STATS_READ;
13561		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13562		    CTL_FLAG_DATA_OUT)
13563			type = CTL_STATS_WRITE;
13564		else
13565			type = CTL_STATS_NO_IO;
13566
13567		lun->stats.ports[targ_port].bytes[type] +=
13568		    io->scsiio.kern_total_len;
13569		lun->stats.ports[targ_port].operations[type]++;
13570#ifdef CTL_TIME_IO
13571		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13572		   &io->io_hdr.dma_bt);
13573		lun->stats.ports[targ_port].num_dmas[type] +=
13574		    io->io_hdr.num_dmas;
13575		getbintime(&cur_bt);
13576		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13577		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13578#endif
13579	}
13580
13581	/*
13582	 * Remove this from the OOA queue.
13583	 */
13584	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13585
13586	/*
13587	 * Run through the blocked queue on this LUN and see if anything
13588	 * has become unblocked, now that this transaction is done.
13589	 */
13590	ctl_check_blocked(lun);
13591
13592	/*
13593	 * If the LUN has been invalidated, free it if there is nothing
13594	 * left on its OOA queue.
13595	 */
13596	if ((lun->flags & CTL_LUN_INVALID)
13597	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13598		mtx_unlock(&lun->lun_lock);
13599		mtx_lock(&ctl_softc->ctl_lock);
13600		ctl_free_lun(lun);
13601		mtx_unlock(&ctl_softc->ctl_lock);
13602	} else
13603		mtx_unlock(&lun->lun_lock);
13604
13605	/*
13606	 * If this command has been aborted, make sure we set the status
13607	 * properly.  The FETD is responsible for freeing the I/O and doing
13608	 * whatever it needs to do to clean up its state.
13609	 */
13610	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13611		ctl_set_task_aborted(&io->scsiio);
13612
13613	/*
13614	 * If enabled, print command error status.
13615	 * We don't print UAs unless debugging was enabled explicitly.
13616	 */
13617	do {
13618		if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
13619			break;
13620		if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0)
13621			break;
13622		if ((ctl_debug & CTL_DEBUG_INFO) == 0 &&
13623		    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) &&
13624		     (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13625			int error_code, sense_key, asc, ascq;
13626
13627			scsi_extract_sense_len(&io->scsiio.sense_data,
13628			    io->scsiio.sense_len, &error_code, &sense_key,
13629			    &asc, &ascq, /*show_errors*/ 0);
13630			if (sense_key == SSD_KEY_UNIT_ATTENTION)
13631				break;
13632		}
13633
13634		ctl_io_error_print(io, NULL);
13635	} while (0);
13636
13637	/*
13638	 * Tell the FETD or the other shelf controller we're done with this
13639	 * command.  Note that only SCSI commands get to this point.  Task
13640	 * management commands are completed above.
13641	 *
13642	 * We only send status to the other controller if we're in XFER
13643	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13644	 * received the I/O (from CTL's perspective), and so the status is
13645	 * generated there.
13646	 *
13647	 * XXX KDM if we hold the lock here, we could cause a deadlock
13648	 * if the frontend comes back in in this context to queue
13649	 * something.
13650	 */
13651	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13652	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13653		union ctl_ha_msg msg;
13654
13655		memset(&msg, 0, sizeof(msg));
13656		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13657		msg.hdr.original_sc = io->io_hdr.original_sc;
13658		msg.hdr.nexus = io->io_hdr.nexus;
13659		msg.hdr.status = io->io_hdr.status;
13660		msg.scsi.scsi_status = io->scsiio.scsi_status;
13661		msg.scsi.tag_num = io->scsiio.tag_num;
13662		msg.scsi.tag_type = io->scsiio.tag_type;
13663		msg.scsi.sense_len = io->scsiio.sense_len;
13664		msg.scsi.sense_residual = io->scsiio.sense_residual;
13665		msg.scsi.residual = io->scsiio.residual;
13666		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13667		       sizeof(io->scsiio.sense_data));
13668		/*
13669		 * We copy this whether or not this is an I/O-related
13670		 * command.  Otherwise, we'd have to go and check to see
13671		 * whether it's a read/write command, and it really isn't
13672		 * worth it.
13673		 */
13674		memcpy(&msg.scsi.lbalen,
13675		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13676		       sizeof(msg.scsi.lbalen));
13677
13678		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13679				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13680			/* XXX do something here */
13681		}
13682
13683		ctl_free_io(io);
13684	} else
13685		fe_done(io);
13686
13687bailout:
13688
13689	return (CTL_RETVAL_COMPLETE);
13690}
13691
13692#ifdef CTL_WITH_CA
13693/*
13694 * Front end should call this if it doesn't do autosense.  When the request
13695 * sense comes back in from the initiator, we'll dequeue this and send it.
13696 */
13697int
13698ctl_queue_sense(union ctl_io *io)
13699{
13700	struct ctl_lun *lun;
13701	struct ctl_softc *ctl_softc;
13702	uint32_t initidx, targ_lun;
13703
13704	ctl_softc = control_softc;
13705
13706	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13707
13708	/*
13709	 * LUN lookup will likely move to the ctl_work_thread() once we
13710	 * have our new queueing infrastructure (that doesn't put things on
13711	 * a per-LUN queue initially).  That is so that we can handle
13712	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13713	 * can't deal with that right now.
13714	 */
13715	mtx_lock(&ctl_softc->ctl_lock);
13716
13717	/*
13718	 * If we don't have a LUN for this, just toss the sense
13719	 * information.
13720	 */
13721	targ_lun = io->io_hdr.nexus.targ_lun;
13722	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13723	if ((targ_lun < CTL_MAX_LUNS)
13724	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13725		lun = ctl_softc->ctl_luns[targ_lun];
13726	else
13727		goto bailout;
13728
13729	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13730
13731	mtx_lock(&lun->lun_lock);
13732	/*
13733	 * Already have CA set for this LUN...toss the sense information.
13734	 */
13735	if (ctl_is_set(lun->have_ca, initidx)) {
13736		mtx_unlock(&lun->lun_lock);
13737		goto bailout;
13738	}
13739
13740	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13741	       ctl_min(sizeof(lun->pending_sense[initidx]),
13742	       sizeof(io->scsiio.sense_data)));
13743	ctl_set_mask(lun->have_ca, initidx);
13744	mtx_unlock(&lun->lun_lock);
13745
13746bailout:
13747	mtx_unlock(&ctl_softc->ctl_lock);
13748
13749	ctl_free_io(io);
13750
13751	return (CTL_RETVAL_COMPLETE);
13752}
13753#endif
13754
13755/*
13756 * Primary command inlet from frontend ports.  All SCSI and task I/O
13757 * requests must go through this function.
13758 */
13759int
13760ctl_queue(union ctl_io *io)
13761{
13762	struct ctl_softc *ctl_softc;
13763
13764	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13765
13766	ctl_softc = control_softc;
13767
13768#ifdef CTL_TIME_IO
13769	io->io_hdr.start_time = time_uptime;
13770	getbintime(&io->io_hdr.start_bt);
13771#endif /* CTL_TIME_IO */
13772
13773	/* Map FE-specific LUN ID into global one. */
13774	io->io_hdr.nexus.targ_mapped_lun =
13775	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13776
13777	switch (io->io_hdr.io_type) {
13778	case CTL_IO_SCSI:
13779	case CTL_IO_TASK:
13780		if (ctl_debug & CTL_DEBUG_CDB)
13781			ctl_io_print(io);
13782		ctl_enqueue_incoming(io);
13783		break;
13784	default:
13785		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13786		return (EINVAL);
13787	}
13788
13789	return (CTL_RETVAL_COMPLETE);
13790}
13791
13792#ifdef CTL_IO_DELAY
13793static void
13794ctl_done_timer_wakeup(void *arg)
13795{
13796	union ctl_io *io;
13797
13798	io = (union ctl_io *)arg;
13799	ctl_done(io);
13800}
13801#endif /* CTL_IO_DELAY */
13802
13803void
13804ctl_done(union ctl_io *io)
13805{
13806	struct ctl_softc *ctl_softc;
13807
13808	ctl_softc = control_softc;
13809
13810	/*
13811	 * Enable this to catch duplicate completion issues.
13812	 */
13813#if 0
13814	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13815		printf("%s: type %d msg %d cdb %x iptl: "
13816		       "%d:%d:%d:%d tag 0x%04x "
13817		       "flag %#x status %x\n",
13818			__func__,
13819			io->io_hdr.io_type,
13820			io->io_hdr.msg_type,
13821			io->scsiio.cdb[0],
13822			io->io_hdr.nexus.initid.id,
13823			io->io_hdr.nexus.targ_port,
13824			io->io_hdr.nexus.targ_target.id,
13825			io->io_hdr.nexus.targ_lun,
13826			(io->io_hdr.io_type ==
13827			CTL_IO_TASK) ?
13828			io->taskio.tag_num :
13829			io->scsiio.tag_num,
13830		        io->io_hdr.flags,
13831			io->io_hdr.status);
13832	} else
13833		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13834#endif
13835
13836	/*
13837	 * This is an internal copy of an I/O, and should not go through
13838	 * the normal done processing logic.
13839	 */
13840	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13841		return;
13842
13843	/*
13844	 * We need to send a msg to the serializing shelf to finish the IO
13845	 * as well.  We don't send a finish message to the other shelf if
13846	 * this is a task management command.  Task management commands
13847	 * aren't serialized in the OOA queue, but rather just executed on
13848	 * both shelf controllers for commands that originated on that
13849	 * controller.
13850	 */
13851	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13852	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13853		union ctl_ha_msg msg_io;
13854
13855		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13856		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13857		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13858		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13859		}
13860		/* continue on to finish IO */
13861	}
13862#ifdef CTL_IO_DELAY
13863	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13864		struct ctl_lun *lun;
13865
13866		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13867
13868		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13869	} else {
13870		struct ctl_lun *lun;
13871
13872		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13873
13874		if ((lun != NULL)
13875		 && (lun->delay_info.done_delay > 0)) {
13876			struct callout *callout;
13877
13878			callout = (struct callout *)&io->io_hdr.timer_bytes;
13879			callout_init(callout, /*mpsafe*/ 1);
13880			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13881			callout_reset(callout,
13882				      lun->delay_info.done_delay * hz,
13883				      ctl_done_timer_wakeup, io);
13884			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13885				lun->delay_info.done_delay = 0;
13886			return;
13887		}
13888	}
13889#endif /* CTL_IO_DELAY */
13890
13891	ctl_enqueue_done(io);
13892}
13893
13894int
13895ctl_isc(struct ctl_scsiio *ctsio)
13896{
13897	struct ctl_lun *lun;
13898	int retval;
13899
13900	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13901
13902	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13903
13904	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13905
13906	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13907
13908	return (retval);
13909}
13910
13911
13912static void
13913ctl_work_thread(void *arg)
13914{
13915	struct ctl_thread *thr = (struct ctl_thread *)arg;
13916	struct ctl_softc *softc = thr->ctl_softc;
13917	union ctl_io *io;
13918	int retval;
13919
13920	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13921
13922	for (;;) {
13923		retval = 0;
13924
13925		/*
13926		 * We handle the queues in this order:
13927		 * - ISC
13928		 * - done queue (to free up resources, unblock other commands)
13929		 * - RtR queue
13930		 * - incoming queue
13931		 *
13932		 * If those queues are empty, we break out of the loop and
13933		 * go to sleep.
13934		 */
13935		mtx_lock(&thr->queue_lock);
13936		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13937		if (io != NULL) {
13938			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13939			mtx_unlock(&thr->queue_lock);
13940			ctl_handle_isc(io);
13941			continue;
13942		}
13943		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13944		if (io != NULL) {
13945			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13946			/* clear any blocked commands, call fe_done */
13947			mtx_unlock(&thr->queue_lock);
13948			retval = ctl_process_done(io);
13949			continue;
13950		}
13951		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13952		if (io != NULL) {
13953			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13954			mtx_unlock(&thr->queue_lock);
13955			if (io->io_hdr.io_type == CTL_IO_TASK)
13956				ctl_run_task(io);
13957			else
13958				ctl_scsiio_precheck(softc, &io->scsiio);
13959			continue;
13960		}
13961		if (!ctl_pause_rtr) {
13962			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13963			if (io != NULL) {
13964				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13965				mtx_unlock(&thr->queue_lock);
13966				retval = ctl_scsiio(&io->scsiio);
13967				if (retval != CTL_RETVAL_COMPLETE)
13968					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13969				continue;
13970			}
13971		}
13972
13973		/* Sleep until we have something to do. */
13974		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13975	}
13976}
13977
13978static void
13979ctl_lun_thread(void *arg)
13980{
13981	struct ctl_softc *softc = (struct ctl_softc *)arg;
13982	struct ctl_be_lun *be_lun;
13983	int retval;
13984
13985	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13986
13987	for (;;) {
13988		retval = 0;
13989		mtx_lock(&softc->ctl_lock);
13990		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13991		if (be_lun != NULL) {
13992			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13993			mtx_unlock(&softc->ctl_lock);
13994			ctl_create_lun(be_lun);
13995			continue;
13996		}
13997
13998		/* Sleep until we have something to do. */
13999		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14000		    PDROP | PRIBIO, "-", 0);
14001	}
14002}
14003
14004static void
14005ctl_enqueue_incoming(union ctl_io *io)
14006{
14007	struct ctl_softc *softc = control_softc;
14008	struct ctl_thread *thr;
14009	u_int idx;
14010
14011	idx = (io->io_hdr.nexus.targ_port * 127 +
14012	       io->io_hdr.nexus.initid.id) % worker_threads;
14013	thr = &softc->threads[idx];
14014	mtx_lock(&thr->queue_lock);
14015	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14016	mtx_unlock(&thr->queue_lock);
14017	wakeup(thr);
14018}
14019
14020static void
14021ctl_enqueue_rtr(union ctl_io *io)
14022{
14023	struct ctl_softc *softc = control_softc;
14024	struct ctl_thread *thr;
14025
14026	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14027	mtx_lock(&thr->queue_lock);
14028	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14029	mtx_unlock(&thr->queue_lock);
14030	wakeup(thr);
14031}
14032
14033static void
14034ctl_enqueue_done(union ctl_io *io)
14035{
14036	struct ctl_softc *softc = control_softc;
14037	struct ctl_thread *thr;
14038
14039	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14040	mtx_lock(&thr->queue_lock);
14041	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14042	mtx_unlock(&thr->queue_lock);
14043	wakeup(thr);
14044}
14045
14046static void
14047ctl_enqueue_isc(union ctl_io *io)
14048{
14049	struct ctl_softc *softc = control_softc;
14050	struct ctl_thread *thr;
14051
14052	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14053	mtx_lock(&thr->queue_lock);
14054	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14055	mtx_unlock(&thr->queue_lock);
14056	wakeup(thr);
14057}
14058
14059/* Initialization and failover */
14060
14061void
14062ctl_init_isc_msg(void)
14063{
14064	printf("CTL: Still calling this thing\n");
14065}
14066
14067/*
14068 * Init component
14069 * 	Initializes component into configuration defined by bootMode
14070 *	(see hasc-sv.c)
14071 *  	returns hasc_Status:
14072 * 		OK
14073 *		ERROR - fatal error
14074 */
14075static ctl_ha_comp_status
14076ctl_isc_init(struct ctl_ha_component *c)
14077{
14078	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14079
14080	c->status = ret;
14081	return ret;
14082}
14083
14084/* Start component
14085 * 	Starts component in state requested. If component starts successfully,
14086 *	it must set its own state to the requestrd state
14087 *	When requested state is HASC_STATE_HA, the component may refine it
14088 * 	by adding _SLAVE or _MASTER flags.
14089 *	Currently allowed state transitions are:
14090 *	UNKNOWN->HA		- initial startup
14091 *	UNKNOWN->SINGLE - initial startup when no parter detected
14092 *	HA->SINGLE		- failover
14093 * returns ctl_ha_comp_status:
14094 * 		OK	- component successfully started in requested state
14095 *		FAILED  - could not start the requested state, failover may
14096 * 			  be possible
14097 *		ERROR	- fatal error detected, no future startup possible
14098 */
14099static ctl_ha_comp_status
14100ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14101{
14102	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14103
14104	printf("%s: go\n", __func__);
14105
14106	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14107	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14108		ctl_is_single = 0;
14109		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14110		    != CTL_HA_STATUS_SUCCESS) {
14111			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14112			ret = CTL_HA_COMP_STATUS_ERROR;
14113		}
14114	} else if (CTL_HA_STATE_IS_HA(c->state)
14115		&& CTL_HA_STATE_IS_SINGLE(state)){
14116		// HA->SINGLE transition
14117	        ctl_failover();
14118		ctl_is_single = 1;
14119	} else {
14120		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14121		       c->state, state);
14122		ret = CTL_HA_COMP_STATUS_ERROR;
14123	}
14124	if (CTL_HA_STATE_IS_SINGLE(state))
14125		ctl_is_single = 1;
14126
14127	c->state = state;
14128	c->status = ret;
14129	return ret;
14130}
14131
14132/*
14133 * Quiesce component
14134 * The component must clear any error conditions (set status to OK) and
14135 * prepare itself to another Start call
14136 * returns ctl_ha_comp_status:
14137 * 	OK
14138 *	ERROR
14139 */
14140static ctl_ha_comp_status
14141ctl_isc_quiesce(struct ctl_ha_component *c)
14142{
14143	int ret = CTL_HA_COMP_STATUS_OK;
14144
14145	ctl_pause_rtr = 1;
14146	c->status = ret;
14147	return ret;
14148}
14149
14150struct ctl_ha_component ctl_ha_component_ctlisc =
14151{
14152	.name = "CTL ISC",
14153	.state = CTL_HA_STATE_UNKNOWN,
14154	.init = ctl_isc_init,
14155	.start = ctl_isc_start,
14156	.quiesce = ctl_isc_quiesce
14157};
14158
14159/*
14160 *  vim: ts=8
14161 */
14162