ctl.c revision 273981
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl.c 273981 2014-11-02 17:31:10Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/ctype.h>
50#include <sys/kernel.h>
51#include <sys/types.h>
52#include <sys/kthread.h>
53#include <sys/bio.h>
54#include <sys/fcntl.h>
55#include <sys/lock.h>
56#include <sys/module.h>
57#include <sys/mutex.h>
58#include <sys/condvar.h>
59#include <sys/malloc.h>
60#include <sys/conf.h>
61#include <sys/ioccom.h>
62#include <sys/queue.h>
63#include <sys/sbuf.h>
64#include <sys/smp.h>
65#include <sys/endian.h>
66#include <sys/sysctl.h>
67
68#include <cam/cam.h>
69#include <cam/scsi/scsi_all.h>
70#include <cam/scsi/scsi_da.h>
71#include <cam/ctl/ctl_io.h>
72#include <cam/ctl/ctl.h>
73#include <cam/ctl/ctl_frontend.h>
74#include <cam/ctl/ctl_frontend_internal.h>
75#include <cam/ctl/ctl_util.h>
76#include <cam/ctl/ctl_backend.h>
77#include <cam/ctl/ctl_ioctl.h>
78#include <cam/ctl/ctl_ha.h>
79#include <cam/ctl/ctl_private.h>
80#include <cam/ctl/ctl_debug.h>
81#include <cam/ctl/ctl_scsi_all.h>
82#include <cam/ctl/ctl_error.h>
83
84struct ctl_softc *control_softc = NULL;
85
86/*
87 * Size and alignment macros needed for Copan-specific HA hardware.  These
88 * can go away when the HA code is re-written, and uses busdma for any
89 * hardware.
90 */
91#define	CTL_ALIGN_8B(target, source, type)				\
92	if (((uint32_t)source & 0x7) != 0)				\
93		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
94	else								\
95		target = (type)source;
96
97#define	CTL_SIZE_8B(target, size)					\
98	if ((size & 0x7) != 0)						\
99		target = size + (0x8 - (size & 0x7));			\
100	else								\
101		target = size;
102
103#define CTL_ALIGN_8B_MARGIN	16
104
105/*
106 * Template mode pages.
107 */
108
109/*
110 * Note that these are default values only.  The actual values will be
111 * filled in when the user does a mode sense.
112 */
113static struct copan_debugconf_subpage debugconf_page_default = {
114	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
115	DBGCNF_SUBPAGE_CODE,		/* subpage */
116	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
117	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
118	DBGCNF_VERSION,			/* page_version */
119	{CTL_TIME_IO_DEFAULT_SECS>>8,
120	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
121};
122
123static struct copan_debugconf_subpage debugconf_page_changeable = {
124	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
125	DBGCNF_SUBPAGE_CODE,		/* subpage */
126	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
127	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
128	0,				/* page_version */
129	{0xff,0xff},			/* ctl_time_io_secs */
130};
131
132static struct scsi_da_rw_recovery_page rw_er_page_default = {
133	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
134	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
135	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
136	/*read_retry_count*/0,
137	/*correction_span*/0,
138	/*head_offset_count*/0,
139	/*data_strobe_offset_cnt*/0,
140	/*byte8*/0,
141	/*write_retry_count*/0,
142	/*reserved2*/0,
143	/*recovery_time_limit*/{0, 0},
144};
145
146static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
147	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
148	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
149	/*byte3*/0,
150	/*read_retry_count*/0,
151	/*correction_span*/0,
152	/*head_offset_count*/0,
153	/*data_strobe_offset_cnt*/0,
154	/*byte8*/0,
155	/*write_retry_count*/0,
156	/*reserved2*/0,
157	/*recovery_time_limit*/{0, 0},
158};
159
160static struct scsi_format_page format_page_default = {
161	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
162	/*page_length*/sizeof(struct scsi_format_page) - 2,
163	/*tracks_per_zone*/ {0, 0},
164	/*alt_sectors_per_zone*/ {0, 0},
165	/*alt_tracks_per_zone*/ {0, 0},
166	/*alt_tracks_per_lun*/ {0, 0},
167	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
168			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
169	/*bytes_per_sector*/ {0, 0},
170	/*interleave*/ {0, 0},
171	/*track_skew*/ {0, 0},
172	/*cylinder_skew*/ {0, 0},
173	/*flags*/ SFP_HSEC,
174	/*reserved*/ {0, 0, 0}
175};
176
177static struct scsi_format_page format_page_changeable = {
178	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
179	/*page_length*/sizeof(struct scsi_format_page) - 2,
180	/*tracks_per_zone*/ {0, 0},
181	/*alt_sectors_per_zone*/ {0, 0},
182	/*alt_tracks_per_zone*/ {0, 0},
183	/*alt_tracks_per_lun*/ {0, 0},
184	/*sectors_per_track*/ {0, 0},
185	/*bytes_per_sector*/ {0, 0},
186	/*interleave*/ {0, 0},
187	/*track_skew*/ {0, 0},
188	/*cylinder_skew*/ {0, 0},
189	/*flags*/ 0,
190	/*reserved*/ {0, 0, 0}
191};
192
193static struct scsi_rigid_disk_page rigid_disk_page_default = {
194	/*page_code*/SMS_RIGID_DISK_PAGE,
195	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
196	/*cylinders*/ {0, 0, 0},
197	/*heads*/ CTL_DEFAULT_HEADS,
198	/*start_write_precomp*/ {0, 0, 0},
199	/*start_reduced_current*/ {0, 0, 0},
200	/*step_rate*/ {0, 0},
201	/*landing_zone_cylinder*/ {0, 0, 0},
202	/*rpl*/ SRDP_RPL_DISABLED,
203	/*rotational_offset*/ 0,
204	/*reserved1*/ 0,
205	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
206			   CTL_DEFAULT_ROTATION_RATE & 0xff},
207	/*reserved2*/ {0, 0}
208};
209
210static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
211	/*page_code*/SMS_RIGID_DISK_PAGE,
212	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
213	/*cylinders*/ {0, 0, 0},
214	/*heads*/ 0,
215	/*start_write_precomp*/ {0, 0, 0},
216	/*start_reduced_current*/ {0, 0, 0},
217	/*step_rate*/ {0, 0},
218	/*landing_zone_cylinder*/ {0, 0, 0},
219	/*rpl*/ 0,
220	/*rotational_offset*/ 0,
221	/*reserved1*/ 0,
222	/*rotation_rate*/ {0, 0},
223	/*reserved2*/ {0, 0}
224};
225
226static struct scsi_caching_page caching_page_default = {
227	/*page_code*/SMS_CACHING_PAGE,
228	/*page_length*/sizeof(struct scsi_caching_page) - 2,
229	/*flags1*/ SCP_DISC | SCP_WCE,
230	/*ret_priority*/ 0,
231	/*disable_pf_transfer_len*/ {0xff, 0xff},
232	/*min_prefetch*/ {0, 0},
233	/*max_prefetch*/ {0xff, 0xff},
234	/*max_pf_ceiling*/ {0xff, 0xff},
235	/*flags2*/ 0,
236	/*cache_segments*/ 0,
237	/*cache_seg_size*/ {0, 0},
238	/*reserved*/ 0,
239	/*non_cache_seg_size*/ {0, 0, 0}
240};
241
242static struct scsi_caching_page caching_page_changeable = {
243	/*page_code*/SMS_CACHING_PAGE,
244	/*page_length*/sizeof(struct scsi_caching_page) - 2,
245	/*flags1*/ SCP_WCE | SCP_RCD,
246	/*ret_priority*/ 0,
247	/*disable_pf_transfer_len*/ {0, 0},
248	/*min_prefetch*/ {0, 0},
249	/*max_prefetch*/ {0, 0},
250	/*max_pf_ceiling*/ {0, 0},
251	/*flags2*/ 0,
252	/*cache_segments*/ 0,
253	/*cache_seg_size*/ {0, 0},
254	/*reserved*/ 0,
255	/*non_cache_seg_size*/ {0, 0, 0}
256};
257
258static struct scsi_control_page control_page_default = {
259	/*page_code*/SMS_CONTROL_MODE_PAGE,
260	/*page_length*/sizeof(struct scsi_control_page) - 2,
261	/*rlec*/0,
262	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
263	/*eca_and_aen*/0,
264	/*flags4*/SCP_TAS,
265	/*aen_holdoff_period*/{0, 0},
266	/*busy_timeout_period*/{0, 0},
267	/*extended_selftest_completion_time*/{0, 0}
268};
269
270static struct scsi_control_page control_page_changeable = {
271	/*page_code*/SMS_CONTROL_MODE_PAGE,
272	/*page_length*/sizeof(struct scsi_control_page) - 2,
273	/*rlec*/SCP_DSENSE,
274	/*queue_flags*/SCP_QUEUE_ALG_MASK,
275	/*eca_and_aen*/SCP_SWP,
276	/*flags4*/0,
277	/*aen_holdoff_period*/{0, 0},
278	/*busy_timeout_period*/{0, 0},
279	/*extended_selftest_completion_time*/{0, 0}
280};
281
282static struct scsi_info_exceptions_page ie_page_default = {
283	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
284	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
285	/*info_flags*/SIEP_FLAGS_DEXCPT,
286	/*mrie*/0,
287	/*interval_timer*/{0, 0, 0, 0},
288	/*report_count*/{0, 0, 0, 0}
289};
290
291static struct scsi_info_exceptions_page ie_page_changeable = {
292	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
293	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
294	/*info_flags*/0,
295	/*mrie*/0,
296	/*interval_timer*/{0, 0, 0, 0},
297	/*report_count*/{0, 0, 0, 0}
298};
299
300static struct scsi_logical_block_provisioning_page lbp_page_default = {
301	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
302	/*subpage_code*/0x02,
303	/*page_length*/{0, sizeof(struct scsi_logical_block_provisioning_page) - 4},
304	/*flags*/0,
305	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
306	/*descr*/{}
307};
308
309static struct scsi_logical_block_provisioning_page lbp_page_changeable = {
310	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
311	/*subpage_code*/0x02,
312	/*page_length*/{0, sizeof(struct scsi_logical_block_provisioning_page) - 4},
313	/*flags*/0,
314	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
315	/*descr*/{}
316};
317
318/*
319 * XXX KDM move these into the softc.
320 */
321static int rcv_sync_msg;
322static int persis_offset;
323static uint8_t ctl_pause_rtr;
324static int     ctl_is_single = 1;
325
326SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
327static int worker_threads = -1;
328TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
329SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
330    &worker_threads, 1, "Number of worker threads");
331static int ctl_debug = CTL_DEBUG_NONE;
332TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
333SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
334    &ctl_debug, 0, "Enabled debug flags");
335
336/*
337 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
338 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
339 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
340 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
341 */
342#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
343
344static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
345				  int param);
346static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
347static int ctl_init(void);
348void ctl_shutdown(void);
349static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
350static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
351static void ctl_ioctl_online(void *arg);
352static void ctl_ioctl_offline(void *arg);
353static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
354static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
355static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
356static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
357static int ctl_ioctl_submit_wait(union ctl_io *io);
358static void ctl_ioctl_datamove(union ctl_io *io);
359static void ctl_ioctl_done(union ctl_io *io);
360static void ctl_ioctl_hard_startstop_callback(void *arg,
361					      struct cfi_metatask *metatask);
362static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
363static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
364			      struct ctl_ooa *ooa_hdr,
365			      struct ctl_ooa_entry *kern_entries);
366static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
367		     struct thread *td);
368static uint32_t ctl_map_lun(int port_num, uint32_t lun);
369static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
370#ifdef unused
371static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
372				   uint32_t targ_target, uint32_t targ_lun,
373				   int can_wait);
374static void ctl_kfree_io(union ctl_io *io);
375#endif /* unused */
376static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
377			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
378static int ctl_free_lun(struct ctl_lun *lun);
379static void ctl_create_lun(struct ctl_be_lun *be_lun);
380/**
381static void ctl_failover_change_pages(struct ctl_softc *softc,
382				      struct ctl_scsiio *ctsio, int master);
383**/
384
385static int ctl_do_mode_select(union ctl_io *io);
386static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
387			   uint64_t res_key, uint64_t sa_res_key,
388			   uint8_t type, uint32_t residx,
389			   struct ctl_scsiio *ctsio,
390			   struct scsi_per_res_out *cdb,
391			   struct scsi_per_res_out_parms* param);
392static void ctl_pro_preempt_other(struct ctl_lun *lun,
393				  union ctl_ha_msg *msg);
394static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
395static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
396static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
397static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
398static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
399static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
400static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
401					 int alloc_len);
402static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
403					 int alloc_len);
404static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
405static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
406static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
407static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
408static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
409static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
410static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
411    union ctl_io *pending_io, union ctl_io *ooa_io);
412static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
413				union ctl_io *starting_io);
414static int ctl_check_blocked(struct ctl_lun *lun);
415static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
416				struct ctl_lun *lun,
417				const struct ctl_cmd_entry *entry,
418				struct ctl_scsiio *ctsio);
419//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
420static void ctl_failover(void);
421static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
422			       struct ctl_scsiio *ctsio);
423static int ctl_scsiio(struct ctl_scsiio *ctsio);
424
425static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
426static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
427			    ctl_ua_type ua_type);
428static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
429			 ctl_ua_type ua_type);
430static int ctl_abort_task(union ctl_io *io);
431static int ctl_abort_task_set(union ctl_io *io);
432static int ctl_i_t_nexus_reset(union ctl_io *io);
433static void ctl_run_task(union ctl_io *io);
434#ifdef CTL_IO_DELAY
435static void ctl_datamove_timer_wakeup(void *arg);
436static void ctl_done_timer_wakeup(void *arg);
437#endif /* CTL_IO_DELAY */
438
439static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
440static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
441static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
442static void ctl_datamove_remote_write(union ctl_io *io);
443static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
444static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
445static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
446static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
447				    ctl_ha_dt_cb callback);
448static void ctl_datamove_remote_read(union ctl_io *io);
449static void ctl_datamove_remote(union ctl_io *io);
450static int ctl_process_done(union ctl_io *io);
451static void ctl_lun_thread(void *arg);
452static void ctl_work_thread(void *arg);
453static void ctl_enqueue_incoming(union ctl_io *io);
454static void ctl_enqueue_rtr(union ctl_io *io);
455static void ctl_enqueue_done(union ctl_io *io);
456static void ctl_enqueue_isc(union ctl_io *io);
457static const struct ctl_cmd_entry *
458    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
459static const struct ctl_cmd_entry *
460    ctl_validate_command(struct ctl_scsiio *ctsio);
461static int ctl_cmd_applicable(uint8_t lun_type,
462    const struct ctl_cmd_entry *entry);
463
464/*
465 * Load the serialization table.  This isn't very pretty, but is probably
466 * the easiest way to do it.
467 */
468#include "ctl_ser_table.c"
469
470/*
471 * We only need to define open, close and ioctl routines for this driver.
472 */
473static struct cdevsw ctl_cdevsw = {
474	.d_version =	D_VERSION,
475	.d_flags =	0,
476	.d_open =	ctl_open,
477	.d_close =	ctl_close,
478	.d_ioctl =	ctl_ioctl,
479	.d_name =	"ctl",
480};
481
482
483MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
484MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
485
486static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
487
488static moduledata_t ctl_moduledata = {
489	"ctl",
490	ctl_module_event_handler,
491	NULL
492};
493
494DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
495MODULE_VERSION(ctl, 1);
496
497static struct ctl_frontend ioctl_frontend =
498{
499	.name = "ioctl",
500};
501
502static void
503ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
504			    union ctl_ha_msg *msg_info)
505{
506	struct ctl_scsiio *ctsio;
507
508	if (msg_info->hdr.original_sc == NULL) {
509		printf("%s: original_sc == NULL!\n", __func__);
510		/* XXX KDM now what? */
511		return;
512	}
513
514	ctsio = &msg_info->hdr.original_sc->scsiio;
515	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
516	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
517	ctsio->io_hdr.status = msg_info->hdr.status;
518	ctsio->scsi_status = msg_info->scsi.scsi_status;
519	ctsio->sense_len = msg_info->scsi.sense_len;
520	ctsio->sense_residual = msg_info->scsi.sense_residual;
521	ctsio->residual = msg_info->scsi.residual;
522	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
523	       sizeof(ctsio->sense_data));
524	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
525	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
526	ctl_enqueue_isc((union ctl_io *)ctsio);
527}
528
529static void
530ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
531				union ctl_ha_msg *msg_info)
532{
533	struct ctl_scsiio *ctsio;
534
535	if (msg_info->hdr.serializing_sc == NULL) {
536		printf("%s: serializing_sc == NULL!\n", __func__);
537		/* XXX KDM now what? */
538		return;
539	}
540
541	ctsio = &msg_info->hdr.serializing_sc->scsiio;
542#if 0
543	/*
544	 * Attempt to catch the situation where an I/O has
545	 * been freed, and we're using it again.
546	 */
547	if (ctsio->io_hdr.io_type == 0xff) {
548		union ctl_io *tmp_io;
549		tmp_io = (union ctl_io *)ctsio;
550		printf("%s: %p use after free!\n", __func__,
551		       ctsio);
552		printf("%s: type %d msg %d cdb %x iptl: "
553		       "%d:%d:%d:%d tag 0x%04x "
554		       "flag %#x status %x\n",
555			__func__,
556			tmp_io->io_hdr.io_type,
557			tmp_io->io_hdr.msg_type,
558			tmp_io->scsiio.cdb[0],
559			tmp_io->io_hdr.nexus.initid.id,
560			tmp_io->io_hdr.nexus.targ_port,
561			tmp_io->io_hdr.nexus.targ_target.id,
562			tmp_io->io_hdr.nexus.targ_lun,
563			(tmp_io->io_hdr.io_type ==
564			CTL_IO_TASK) ?
565			tmp_io->taskio.tag_num :
566			tmp_io->scsiio.tag_num,
567		        tmp_io->io_hdr.flags,
568			tmp_io->io_hdr.status);
569	}
570#endif
571	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
572	ctl_enqueue_isc((union ctl_io *)ctsio);
573}
574
575/*
576 * ISC (Inter Shelf Communication) event handler.  Events from the HA
577 * subsystem come in here.
578 */
579static void
580ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
581{
582	struct ctl_softc *ctl_softc;
583	union ctl_io *io;
584	struct ctl_prio *presio;
585	ctl_ha_status isc_status;
586
587	ctl_softc = control_softc;
588	io = NULL;
589
590
591#if 0
592	printf("CTL: Isc Msg event %d\n", event);
593#endif
594	if (event == CTL_HA_EVT_MSG_RECV) {
595		union ctl_ha_msg msg_info;
596
597		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
598					     sizeof(msg_info), /*wait*/ 0);
599#if 0
600		printf("CTL: msg_type %d\n", msg_info.msg_type);
601#endif
602		if (isc_status != 0) {
603			printf("Error receiving message, status = %d\n",
604			       isc_status);
605			return;
606		}
607
608		switch (msg_info.hdr.msg_type) {
609		case CTL_MSG_SERIALIZE:
610#if 0
611			printf("Serialize\n");
612#endif
613			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
614			if (io == NULL) {
615				printf("ctl_isc_event_handler: can't allocate "
616				       "ctl_io!\n");
617				/* Bad Juju */
618				/* Need to set busy and send msg back */
619				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
620				msg_info.hdr.status = CTL_SCSI_ERROR;
621				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
622				msg_info.scsi.sense_len = 0;
623			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
624				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
625				}
626				goto bailout;
627			}
628			ctl_zero_io(io);
629			// populate ctsio from msg_info
630			io->io_hdr.io_type = CTL_IO_SCSI;
631			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
632			io->io_hdr.original_sc = msg_info.hdr.original_sc;
633#if 0
634			printf("pOrig %x\n", (int)msg_info.original_sc);
635#endif
636			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
637					    CTL_FLAG_IO_ACTIVE;
638			/*
639			 * If we're in serialization-only mode, we don't
640			 * want to go through full done processing.  Thus
641			 * the COPY flag.
642			 *
643			 * XXX KDM add another flag that is more specific.
644			 */
645			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
646				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
647			io->io_hdr.nexus = msg_info.hdr.nexus;
648#if 0
649			printf("targ %d, port %d, iid %d, lun %d\n",
650			       io->io_hdr.nexus.targ_target.id,
651			       io->io_hdr.nexus.targ_port,
652			       io->io_hdr.nexus.initid.id,
653			       io->io_hdr.nexus.targ_lun);
654#endif
655			io->scsiio.tag_num = msg_info.scsi.tag_num;
656			io->scsiio.tag_type = msg_info.scsi.tag_type;
657			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
658			       CTL_MAX_CDBLEN);
659			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
660				const struct ctl_cmd_entry *entry;
661
662				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
663				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
664				io->io_hdr.flags |=
665					entry->flags & CTL_FLAG_DATA_MASK;
666			}
667			ctl_enqueue_isc(io);
668			break;
669
670		/* Performed on the Originating SC, XFER mode only */
671		case CTL_MSG_DATAMOVE: {
672			struct ctl_sg_entry *sgl;
673			int i, j;
674
675			io = msg_info.hdr.original_sc;
676			if (io == NULL) {
677				printf("%s: original_sc == NULL!\n", __func__);
678				/* XXX KDM do something here */
679				break;
680			}
681			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
682			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
683			/*
684			 * Keep track of this, we need to send it back over
685			 * when the datamove is complete.
686			 */
687			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
688
689			if (msg_info.dt.sg_sequence == 0) {
690				/*
691				 * XXX KDM we use the preallocated S/G list
692				 * here, but we'll need to change this to
693				 * dynamic allocation if we need larger S/G
694				 * lists.
695				 */
696				if (msg_info.dt.kern_sg_entries >
697				    sizeof(io->io_hdr.remote_sglist) /
698				    sizeof(io->io_hdr.remote_sglist[0])) {
699					printf("%s: number of S/G entries "
700					    "needed %u > allocated num %zd\n",
701					    __func__,
702					    msg_info.dt.kern_sg_entries,
703					    sizeof(io->io_hdr.remote_sglist)/
704					    sizeof(io->io_hdr.remote_sglist[0]));
705
706					/*
707					 * XXX KDM send a message back to
708					 * the other side to shut down the
709					 * DMA.  The error will come back
710					 * through via the normal channel.
711					 */
712					break;
713				}
714				sgl = io->io_hdr.remote_sglist;
715				memset(sgl, 0,
716				       sizeof(io->io_hdr.remote_sglist));
717
718				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
719
720				io->scsiio.kern_sg_entries =
721					msg_info.dt.kern_sg_entries;
722				io->scsiio.rem_sg_entries =
723					msg_info.dt.kern_sg_entries;
724				io->scsiio.kern_data_len =
725					msg_info.dt.kern_data_len;
726				io->scsiio.kern_total_len =
727					msg_info.dt.kern_total_len;
728				io->scsiio.kern_data_resid =
729					msg_info.dt.kern_data_resid;
730				io->scsiio.kern_rel_offset =
731					msg_info.dt.kern_rel_offset;
732				/*
733				 * Clear out per-DMA flags.
734				 */
735				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
736				/*
737				 * Add per-DMA flags that are set for this
738				 * particular DMA request.
739				 */
740				io->io_hdr.flags |= msg_info.dt.flags &
741						    CTL_FLAG_RDMA_MASK;
742			} else
743				sgl = (struct ctl_sg_entry *)
744					io->scsiio.kern_data_ptr;
745
746			for (i = msg_info.dt.sent_sg_entries, j = 0;
747			     i < (msg_info.dt.sent_sg_entries +
748			     msg_info.dt.cur_sg_entries); i++, j++) {
749				sgl[i].addr = msg_info.dt.sg_list[j].addr;
750				sgl[i].len = msg_info.dt.sg_list[j].len;
751
752#if 0
753				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
754				       __func__,
755				       msg_info.dt.sg_list[j].addr,
756				       msg_info.dt.sg_list[j].len,
757				       sgl[i].addr, sgl[i].len, j, i);
758#endif
759			}
760#if 0
761			memcpy(&sgl[msg_info.dt.sent_sg_entries],
762			       msg_info.dt.sg_list,
763			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
764#endif
765
766			/*
767			 * If this is the last piece of the I/O, we've got
768			 * the full S/G list.  Queue processing in the thread.
769			 * Otherwise wait for the next piece.
770			 */
771			if (msg_info.dt.sg_last != 0)
772				ctl_enqueue_isc(io);
773			break;
774		}
775		/* Performed on the Serializing (primary) SC, XFER mode only */
776		case CTL_MSG_DATAMOVE_DONE: {
777			if (msg_info.hdr.serializing_sc == NULL) {
778				printf("%s: serializing_sc == NULL!\n",
779				       __func__);
780				/* XXX KDM now what? */
781				break;
782			}
783			/*
784			 * We grab the sense information here in case
785			 * there was a failure, so we can return status
786			 * back to the initiator.
787			 */
788			io = msg_info.hdr.serializing_sc;
789			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
790			io->io_hdr.status = msg_info.hdr.status;
791			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
792			io->scsiio.sense_len = msg_info.scsi.sense_len;
793			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
794			io->io_hdr.port_status = msg_info.scsi.fetd_status;
795			io->scsiio.residual = msg_info.scsi.residual;
796			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
797			       sizeof(io->scsiio.sense_data));
798			ctl_enqueue_isc(io);
799			break;
800		}
801
802		/* Preformed on Originating SC, SER_ONLY mode */
803		case CTL_MSG_R2R:
804			io = msg_info.hdr.original_sc;
805			if (io == NULL) {
806				printf("%s: Major Bummer\n", __func__);
807				return;
808			} else {
809#if 0
810				printf("pOrig %x\n",(int) ctsio);
811#endif
812			}
813			io->io_hdr.msg_type = CTL_MSG_R2R;
814			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
815			ctl_enqueue_isc(io);
816			break;
817
818		/*
819		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
820		 * mode.
821		 * Performed on the Originating (i.e. secondary) SC in XFER
822		 * mode
823		 */
824		case CTL_MSG_FINISH_IO:
825			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
826				ctl_isc_handler_finish_xfer(ctl_softc,
827							    &msg_info);
828			else
829				ctl_isc_handler_finish_ser_only(ctl_softc,
830								&msg_info);
831			break;
832
833		/* Preformed on Originating SC */
834		case CTL_MSG_BAD_JUJU:
835			io = msg_info.hdr.original_sc;
836			if (io == NULL) {
837				printf("%s: Bad JUJU!, original_sc is NULL!\n",
838				       __func__);
839				break;
840			}
841			ctl_copy_sense_data(&msg_info, io);
842			/*
843			 * IO should have already been cleaned up on other
844			 * SC so clear this flag so we won't send a message
845			 * back to finish the IO there.
846			 */
847			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
848			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
849
850			/* io = msg_info.hdr.serializing_sc; */
851			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
852			ctl_enqueue_isc(io);
853			break;
854
855		/* Handle resets sent from the other side */
856		case CTL_MSG_MANAGE_TASKS: {
857			struct ctl_taskio *taskio;
858			taskio = (struct ctl_taskio *)ctl_alloc_io(
859				(void *)ctl_softc->othersc_pool);
860			if (taskio == NULL) {
861				printf("ctl_isc_event_handler: can't allocate "
862				       "ctl_io!\n");
863				/* Bad Juju */
864				/* should I just call the proper reset func
865				   here??? */
866				goto bailout;
867			}
868			ctl_zero_io((union ctl_io *)taskio);
869			taskio->io_hdr.io_type = CTL_IO_TASK;
870			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
871			taskio->io_hdr.nexus = msg_info.hdr.nexus;
872			taskio->task_action = msg_info.task.task_action;
873			taskio->tag_num = msg_info.task.tag_num;
874			taskio->tag_type = msg_info.task.tag_type;
875#ifdef CTL_TIME_IO
876			taskio->io_hdr.start_time = time_uptime;
877			getbintime(&taskio->io_hdr.start_bt);
878#if 0
879			cs_prof_gettime(&taskio->io_hdr.start_ticks);
880#endif
881#endif /* CTL_TIME_IO */
882			ctl_run_task((union ctl_io *)taskio);
883			break;
884		}
885		/* Persistent Reserve action which needs attention */
886		case CTL_MSG_PERS_ACTION:
887			presio = (struct ctl_prio *)ctl_alloc_io(
888				(void *)ctl_softc->othersc_pool);
889			if (presio == NULL) {
890				printf("ctl_isc_event_handler: can't allocate "
891				       "ctl_io!\n");
892				/* Bad Juju */
893				/* Need to set busy and send msg back */
894				goto bailout;
895			}
896			ctl_zero_io((union ctl_io *)presio);
897			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
898			presio->pr_msg = msg_info.pr;
899			ctl_enqueue_isc((union ctl_io *)presio);
900			break;
901		case CTL_MSG_SYNC_FE:
902			rcv_sync_msg = 1;
903			break;
904		default:
905		        printf("How did I get here?\n");
906		}
907	} else if (event == CTL_HA_EVT_MSG_SENT) {
908		if (param != CTL_HA_STATUS_SUCCESS) {
909			printf("Bad status from ctl_ha_msg_send status %d\n",
910			       param);
911		}
912		return;
913	} else if (event == CTL_HA_EVT_DISCONNECT) {
914		printf("CTL: Got a disconnect from Isc\n");
915		return;
916	} else {
917		printf("ctl_isc_event_handler: Unknown event %d\n", event);
918		return;
919	}
920
921bailout:
922	return;
923}
924
925static void
926ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
927{
928	struct scsi_sense_data *sense;
929
930	sense = &dest->scsiio.sense_data;
931	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
932	dest->scsiio.scsi_status = src->scsi.scsi_status;
933	dest->scsiio.sense_len = src->scsi.sense_len;
934	dest->io_hdr.status = src->hdr.status;
935}
936
937static int
938ctl_init(void)
939{
940	struct ctl_softc *softc;
941	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
942	struct ctl_port *port;
943        uint8_t sc_id =0;
944	int i, error, retval;
945	//int isc_retval;
946
947	retval = 0;
948	ctl_pause_rtr = 0;
949        rcv_sync_msg = 0;
950
951	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
952			       M_WAITOK | M_ZERO);
953	softc = control_softc;
954
955	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
956			      "cam/ctl");
957
958	softc->dev->si_drv1 = softc;
959
960	/*
961	 * By default, return a "bad LUN" peripheral qualifier for unknown
962	 * LUNs.  The user can override this default using the tunable or
963	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
964	 */
965	softc->inquiry_pq_no_lun = 1;
966	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
967			  &softc->inquiry_pq_no_lun);
968	sysctl_ctx_init(&softc->sysctl_ctx);
969	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
970		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
971		CTLFLAG_RD, 0, "CAM Target Layer");
972
973	if (softc->sysctl_tree == NULL) {
974		printf("%s: unable to allocate sysctl tree\n", __func__);
975		destroy_dev(softc->dev);
976		free(control_softc, M_DEVBUF);
977		control_softc = NULL;
978		return (ENOMEM);
979	}
980
981	SYSCTL_ADD_INT(&softc->sysctl_ctx,
982		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
983		       "inquiry_pq_no_lun", CTLFLAG_RW,
984		       &softc->inquiry_pq_no_lun, 0,
985		       "Report no lun possible for invalid LUNs");
986
987	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
988	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
989	softc->open_count = 0;
990
991	/*
992	 * Default to actually sending a SYNCHRONIZE CACHE command down to
993	 * the drive.
994	 */
995	softc->flags = CTL_FLAG_REAL_SYNC;
996
997	/*
998	 * In Copan's HA scheme, the "master" and "slave" roles are
999	 * figured out through the slot the controller is in.  Although it
1000	 * is an active/active system, someone has to be in charge.
1001 	 */
1002#ifdef NEEDTOPORT
1003        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1004#endif
1005
1006        if (sc_id == 0) {
1007		softc->flags |= CTL_FLAG_MASTER_SHELF;
1008		persis_offset = 0;
1009	} else
1010		persis_offset = CTL_MAX_INITIATORS;
1011
1012	/*
1013	 * XXX KDM need to figure out where we want to get our target ID
1014	 * and WWID.  Is it different on each port?
1015	 */
1016	softc->target.id = 0;
1017	softc->target.wwid[0] = 0x12345678;
1018	softc->target.wwid[1] = 0x87654321;
1019	STAILQ_INIT(&softc->lun_list);
1020	STAILQ_INIT(&softc->pending_lun_queue);
1021	STAILQ_INIT(&softc->fe_list);
1022	STAILQ_INIT(&softc->port_list);
1023	STAILQ_INIT(&softc->be_list);
1024	STAILQ_INIT(&softc->io_pools);
1025	ctl_tpc_init(softc);
1026
1027	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1028			    &internal_pool)!= 0){
1029		printf("ctl: can't allocate %d entry internal pool, "
1030		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1031		return (ENOMEM);
1032	}
1033
1034	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1035			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1036		printf("ctl: can't allocate %d entry emergency pool, "
1037		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1038		ctl_pool_free(internal_pool);
1039		return (ENOMEM);
1040	}
1041
1042	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1043	                    &other_pool) != 0)
1044	{
1045		printf("ctl: can't allocate %d entry other SC pool, "
1046		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1047		ctl_pool_free(internal_pool);
1048		ctl_pool_free(emergency_pool);
1049		return (ENOMEM);
1050	}
1051
1052	softc->internal_pool = internal_pool;
1053	softc->emergency_pool = emergency_pool;
1054	softc->othersc_pool = other_pool;
1055
1056	if (worker_threads <= 0)
1057		worker_threads = max(1, mp_ncpus / 4);
1058	if (worker_threads > CTL_MAX_THREADS)
1059		worker_threads = CTL_MAX_THREADS;
1060
1061	for (i = 0; i < worker_threads; i++) {
1062		struct ctl_thread *thr = &softc->threads[i];
1063
1064		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1065		thr->ctl_softc = softc;
1066		STAILQ_INIT(&thr->incoming_queue);
1067		STAILQ_INIT(&thr->rtr_queue);
1068		STAILQ_INIT(&thr->done_queue);
1069		STAILQ_INIT(&thr->isc_queue);
1070
1071		error = kproc_kthread_add(ctl_work_thread, thr,
1072		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1073		if (error != 0) {
1074			printf("error creating CTL work thread!\n");
1075			ctl_pool_free(internal_pool);
1076			ctl_pool_free(emergency_pool);
1077			ctl_pool_free(other_pool);
1078			return (error);
1079		}
1080	}
1081	error = kproc_kthread_add(ctl_lun_thread, softc,
1082	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1083	if (error != 0) {
1084		printf("error creating CTL lun thread!\n");
1085		ctl_pool_free(internal_pool);
1086		ctl_pool_free(emergency_pool);
1087		ctl_pool_free(other_pool);
1088		return (error);
1089	}
1090	if (bootverbose)
1091		printf("ctl: CAM Target Layer loaded\n");
1092
1093	/*
1094	 * Initialize the ioctl front end.
1095	 */
1096	ctl_frontend_register(&ioctl_frontend);
1097	port = &softc->ioctl_info.port;
1098	port->frontend = &ioctl_frontend;
1099	sprintf(softc->ioctl_info.port_name, "ioctl");
1100	port->port_type = CTL_PORT_IOCTL;
1101	port->num_requested_ctl_io = 100;
1102	port->port_name = softc->ioctl_info.port_name;
1103	port->port_online = ctl_ioctl_online;
1104	port->port_offline = ctl_ioctl_offline;
1105	port->onoff_arg = &softc->ioctl_info;
1106	port->lun_enable = ctl_ioctl_lun_enable;
1107	port->lun_disable = ctl_ioctl_lun_disable;
1108	port->targ_lun_arg = &softc->ioctl_info;
1109	port->fe_datamove = ctl_ioctl_datamove;
1110	port->fe_done = ctl_ioctl_done;
1111	port->max_targets = 15;
1112	port->max_target_id = 15;
1113
1114	if (ctl_port_register(&softc->ioctl_info.port,
1115	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1116		printf("ctl: ioctl front end registration failed, will "
1117		       "continue anyway\n");
1118	}
1119
1120#ifdef CTL_IO_DELAY
1121	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1122		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1123		       sizeof(struct callout), CTL_TIMER_BYTES);
1124		return (EINVAL);
1125	}
1126#endif /* CTL_IO_DELAY */
1127
1128	return (0);
1129}
1130
1131void
1132ctl_shutdown(void)
1133{
1134	struct ctl_softc *softc;
1135	struct ctl_lun *lun, *next_lun;
1136	struct ctl_io_pool *pool;
1137
1138	softc = (struct ctl_softc *)control_softc;
1139
1140	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1141		printf("ctl: ioctl front end deregistration failed\n");
1142
1143	mtx_lock(&softc->ctl_lock);
1144
1145	/*
1146	 * Free up each LUN.
1147	 */
1148	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1149		next_lun = STAILQ_NEXT(lun, links);
1150		ctl_free_lun(lun);
1151	}
1152
1153	mtx_unlock(&softc->ctl_lock);
1154
1155	ctl_frontend_deregister(&ioctl_frontend);
1156
1157	/*
1158	 * This will rip the rug out from under any FETDs or anyone else
1159	 * that has a pool allocated.  Since we increment our module
1160	 * refcount any time someone outside the main CTL module allocates
1161	 * a pool, we shouldn't have any problems here.  The user won't be
1162	 * able to unload the CTL module until client modules have
1163	 * successfully unloaded.
1164	 */
1165	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1166		ctl_pool_free(pool);
1167
1168#if 0
1169	ctl_shutdown_thread(softc->work_thread);
1170	mtx_destroy(&softc->queue_lock);
1171#endif
1172
1173	ctl_tpc_shutdown(softc);
1174	mtx_destroy(&softc->pool_lock);
1175	mtx_destroy(&softc->ctl_lock);
1176
1177	destroy_dev(softc->dev);
1178
1179	sysctl_ctx_free(&softc->sysctl_ctx);
1180
1181	free(control_softc, M_DEVBUF);
1182	control_softc = NULL;
1183
1184	if (bootverbose)
1185		printf("ctl: CAM Target Layer unloaded\n");
1186}
1187
1188static int
1189ctl_module_event_handler(module_t mod, int what, void *arg)
1190{
1191
1192	switch (what) {
1193	case MOD_LOAD:
1194		return (ctl_init());
1195	case MOD_UNLOAD:
1196		return (EBUSY);
1197	default:
1198		return (EOPNOTSUPP);
1199	}
1200}
1201
1202/*
1203 * XXX KDM should we do some access checks here?  Bump a reference count to
1204 * prevent a CTL module from being unloaded while someone has it open?
1205 */
1206static int
1207ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1208{
1209	return (0);
1210}
1211
1212static int
1213ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1214{
1215	return (0);
1216}
1217
1218int
1219ctl_port_enable(ctl_port_type port_type)
1220{
1221	struct ctl_softc *softc;
1222	struct ctl_port *port;
1223
1224	if (ctl_is_single == 0) {
1225		union ctl_ha_msg msg_info;
1226		int isc_retval;
1227
1228#if 0
1229		printf("%s: HA mode, synchronizing frontend enable\n",
1230		        __func__);
1231#endif
1232		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1233	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1234		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1235			printf("Sync msg send error retval %d\n", isc_retval);
1236		}
1237		if (!rcv_sync_msg) {
1238			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1239			        sizeof(msg_info), 1);
1240		}
1241#if 0
1242        	printf("CTL:Frontend Enable\n");
1243	} else {
1244		printf("%s: single mode, skipping frontend synchronization\n",
1245		        __func__);
1246#endif
1247	}
1248
1249	softc = control_softc;
1250
1251	STAILQ_FOREACH(port, &softc->port_list, links) {
1252		if (port_type & port->port_type)
1253		{
1254#if 0
1255			printf("port %d\n", port->targ_port);
1256#endif
1257			ctl_port_online(port);
1258		}
1259	}
1260
1261	return (0);
1262}
1263
1264int
1265ctl_port_disable(ctl_port_type port_type)
1266{
1267	struct ctl_softc *softc;
1268	struct ctl_port *port;
1269
1270	softc = control_softc;
1271
1272	STAILQ_FOREACH(port, &softc->port_list, links) {
1273		if (port_type & port->port_type)
1274			ctl_port_offline(port);
1275	}
1276
1277	return (0);
1278}
1279
1280/*
1281 * Returns 0 for success, 1 for failure.
1282 * Currently the only failure mode is if there aren't enough entries
1283 * allocated.  So, in case of a failure, look at num_entries_dropped,
1284 * reallocate and try again.
1285 */
1286int
1287ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1288	      int *num_entries_filled, int *num_entries_dropped,
1289	      ctl_port_type port_type, int no_virtual)
1290{
1291	struct ctl_softc *softc;
1292	struct ctl_port *port;
1293	int entries_dropped, entries_filled;
1294	int retval;
1295	int i;
1296
1297	softc = control_softc;
1298
1299	retval = 0;
1300	entries_filled = 0;
1301	entries_dropped = 0;
1302
1303	i = 0;
1304	mtx_lock(&softc->ctl_lock);
1305	STAILQ_FOREACH(port, &softc->port_list, links) {
1306		struct ctl_port_entry *entry;
1307
1308		if ((port->port_type & port_type) == 0)
1309			continue;
1310
1311		if ((no_virtual != 0)
1312		 && (port->virtual_port != 0))
1313			continue;
1314
1315		if (entries_filled >= num_entries_alloced) {
1316			entries_dropped++;
1317			continue;
1318		}
1319		entry = &entries[i];
1320
1321		entry->port_type = port->port_type;
1322		strlcpy(entry->port_name, port->port_name,
1323			sizeof(entry->port_name));
1324		entry->physical_port = port->physical_port;
1325		entry->virtual_port = port->virtual_port;
1326		entry->wwnn = port->wwnn;
1327		entry->wwpn = port->wwpn;
1328
1329		i++;
1330		entries_filled++;
1331	}
1332
1333	mtx_unlock(&softc->ctl_lock);
1334
1335	if (entries_dropped > 0)
1336		retval = 1;
1337
1338	*num_entries_dropped = entries_dropped;
1339	*num_entries_filled = entries_filled;
1340
1341	return (retval);
1342}
1343
1344static void
1345ctl_ioctl_online(void *arg)
1346{
1347	struct ctl_ioctl_info *ioctl_info;
1348
1349	ioctl_info = (struct ctl_ioctl_info *)arg;
1350
1351	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1352}
1353
1354static void
1355ctl_ioctl_offline(void *arg)
1356{
1357	struct ctl_ioctl_info *ioctl_info;
1358
1359	ioctl_info = (struct ctl_ioctl_info *)arg;
1360
1361	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1362}
1363
1364/*
1365 * Remove an initiator by port number and initiator ID.
1366 * Returns 0 for success, -1 for failure.
1367 */
1368int
1369ctl_remove_initiator(struct ctl_port *port, int iid)
1370{
1371	struct ctl_softc *softc = control_softc;
1372
1373	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1374
1375	if (iid > CTL_MAX_INIT_PER_PORT) {
1376		printf("%s: initiator ID %u > maximun %u!\n",
1377		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1378		return (-1);
1379	}
1380
1381	mtx_lock(&softc->ctl_lock);
1382	port->wwpn_iid[iid].in_use--;
1383	port->wwpn_iid[iid].last_use = time_uptime;
1384	mtx_unlock(&softc->ctl_lock);
1385
1386	return (0);
1387}
1388
1389/*
1390 * Add an initiator to the initiator map.
1391 * Returns iid for success, < 0 for failure.
1392 */
1393int
1394ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1395{
1396	struct ctl_softc *softc = control_softc;
1397	time_t best_time;
1398	int i, best;
1399
1400	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1401
1402	if (iid >= CTL_MAX_INIT_PER_PORT) {
1403		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1404		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1405		free(name, M_CTL);
1406		return (-1);
1407	}
1408
1409	mtx_lock(&softc->ctl_lock);
1410
1411	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1412		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1413			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1414				iid = i;
1415				break;
1416			}
1417			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1418			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1419				iid = i;
1420				break;
1421			}
1422		}
1423	}
1424
1425	if (iid < 0) {
1426		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1427			if (port->wwpn_iid[i].in_use == 0 &&
1428			    port->wwpn_iid[i].wwpn == 0 &&
1429			    port->wwpn_iid[i].name == NULL) {
1430				iid = i;
1431				break;
1432			}
1433		}
1434	}
1435
1436	if (iid < 0) {
1437		best = -1;
1438		best_time = INT32_MAX;
1439		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1440			if (port->wwpn_iid[i].in_use == 0) {
1441				if (port->wwpn_iid[i].last_use < best_time) {
1442					best = i;
1443					best_time = port->wwpn_iid[i].last_use;
1444				}
1445			}
1446		}
1447		iid = best;
1448	}
1449
1450	if (iid < 0) {
1451		mtx_unlock(&softc->ctl_lock);
1452		free(name, M_CTL);
1453		return (-2);
1454	}
1455
1456	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1457		/*
1458		 * This is not an error yet.
1459		 */
1460		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1461#if 0
1462			printf("%s: port %d iid %u WWPN %#jx arrived"
1463			    " again\n", __func__, port->targ_port,
1464			    iid, (uintmax_t)wwpn);
1465#endif
1466			goto take;
1467		}
1468		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1469		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1470#if 0
1471			printf("%s: port %d iid %u name '%s' arrived"
1472			    " again\n", __func__, port->targ_port,
1473			    iid, name);
1474#endif
1475			goto take;
1476		}
1477
1478		/*
1479		 * This is an error, but what do we do about it?  The
1480		 * driver is telling us we have a new WWPN for this
1481		 * initiator ID, so we pretty much need to use it.
1482		 */
1483		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1484		    " but WWPN %#jx '%s' is still at that address\n",
1485		    __func__, port->targ_port, iid, wwpn, name,
1486		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1487		    port->wwpn_iid[iid].name);
1488
1489		/*
1490		 * XXX KDM clear have_ca and ua_pending on each LUN for
1491		 * this initiator.
1492		 */
1493	}
1494take:
1495	free(port->wwpn_iid[iid].name, M_CTL);
1496	port->wwpn_iid[iid].name = name;
1497	port->wwpn_iid[iid].wwpn = wwpn;
1498	port->wwpn_iid[iid].in_use++;
1499	mtx_unlock(&softc->ctl_lock);
1500
1501	return (iid);
1502}
1503
1504static int
1505ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1506{
1507	int len;
1508
1509	switch (port->port_type) {
1510	case CTL_PORT_FC:
1511	{
1512		struct scsi_transportid_fcp *id =
1513		    (struct scsi_transportid_fcp *)buf;
1514		if (port->wwpn_iid[iid].wwpn == 0)
1515			return (0);
1516		memset(id, 0, sizeof(*id));
1517		id->format_protocol = SCSI_PROTO_FC;
1518		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1519		return (sizeof(*id));
1520	}
1521	case CTL_PORT_ISCSI:
1522	{
1523		struct scsi_transportid_iscsi_port *id =
1524		    (struct scsi_transportid_iscsi_port *)buf;
1525		if (port->wwpn_iid[iid].name == NULL)
1526			return (0);
1527		memset(id, 0, 256);
1528		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1529		    SCSI_PROTO_ISCSI;
1530		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1531		len = roundup2(min(len, 252), 4);
1532		scsi_ulto2b(len, id->additional_length);
1533		return (sizeof(*id) + len);
1534	}
1535	case CTL_PORT_SAS:
1536	{
1537		struct scsi_transportid_sas *id =
1538		    (struct scsi_transportid_sas *)buf;
1539		if (port->wwpn_iid[iid].wwpn == 0)
1540			return (0);
1541		memset(id, 0, sizeof(*id));
1542		id->format_protocol = SCSI_PROTO_SAS;
1543		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1544		return (sizeof(*id));
1545	}
1546	default:
1547	{
1548		struct scsi_transportid_spi *id =
1549		    (struct scsi_transportid_spi *)buf;
1550		memset(id, 0, sizeof(*id));
1551		id->format_protocol = SCSI_PROTO_SPI;
1552		scsi_ulto2b(iid, id->scsi_addr);
1553		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1554		return (sizeof(*id));
1555	}
1556	}
1557}
1558
1559static int
1560ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1561{
1562	return (0);
1563}
1564
1565static int
1566ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1567{
1568	return (0);
1569}
1570
1571/*
1572 * Data movement routine for the CTL ioctl frontend port.
1573 */
1574static int
1575ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1576{
1577	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1578	struct ctl_sg_entry ext_entry, kern_entry;
1579	int ext_sglen, ext_sg_entries, kern_sg_entries;
1580	int ext_sg_start, ext_offset;
1581	int len_to_copy, len_copied;
1582	int kern_watermark, ext_watermark;
1583	int ext_sglist_malloced;
1584	int i, j;
1585
1586	ext_sglist_malloced = 0;
1587	ext_sg_start = 0;
1588	ext_offset = 0;
1589
1590	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1591
1592	/*
1593	 * If this flag is set, fake the data transfer.
1594	 */
1595	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1596		ctsio->ext_data_filled = ctsio->ext_data_len;
1597		goto bailout;
1598	}
1599
1600	/*
1601	 * To simplify things here, if we have a single buffer, stick it in
1602	 * a S/G entry and just make it a single entry S/G list.
1603	 */
1604	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1605		int len_seen;
1606
1607		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1608
1609		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1610							   M_WAITOK);
1611		ext_sglist_malloced = 1;
1612		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1613				   ext_sglen) != 0) {
1614			ctl_set_internal_failure(ctsio,
1615						 /*sks_valid*/ 0,
1616						 /*retry_count*/ 0);
1617			goto bailout;
1618		}
1619		ext_sg_entries = ctsio->ext_sg_entries;
1620		len_seen = 0;
1621		for (i = 0; i < ext_sg_entries; i++) {
1622			if ((len_seen + ext_sglist[i].len) >=
1623			     ctsio->ext_data_filled) {
1624				ext_sg_start = i;
1625				ext_offset = ctsio->ext_data_filled - len_seen;
1626				break;
1627			}
1628			len_seen += ext_sglist[i].len;
1629		}
1630	} else {
1631		ext_sglist = &ext_entry;
1632		ext_sglist->addr = ctsio->ext_data_ptr;
1633		ext_sglist->len = ctsio->ext_data_len;
1634		ext_sg_entries = 1;
1635		ext_sg_start = 0;
1636		ext_offset = ctsio->ext_data_filled;
1637	}
1638
1639	if (ctsio->kern_sg_entries > 0) {
1640		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1641		kern_sg_entries = ctsio->kern_sg_entries;
1642	} else {
1643		kern_sglist = &kern_entry;
1644		kern_sglist->addr = ctsio->kern_data_ptr;
1645		kern_sglist->len = ctsio->kern_data_len;
1646		kern_sg_entries = 1;
1647	}
1648
1649
1650	kern_watermark = 0;
1651	ext_watermark = ext_offset;
1652	len_copied = 0;
1653	for (i = ext_sg_start, j = 0;
1654	     i < ext_sg_entries && j < kern_sg_entries;) {
1655		uint8_t *ext_ptr, *kern_ptr;
1656
1657		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1658				      kern_sglist[j].len - kern_watermark);
1659
1660		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1661		ext_ptr = ext_ptr + ext_watermark;
1662		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1663			/*
1664			 * XXX KDM fix this!
1665			 */
1666			panic("need to implement bus address support");
1667#if 0
1668			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1669#endif
1670		} else
1671			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1672		kern_ptr = kern_ptr + kern_watermark;
1673
1674		kern_watermark += len_to_copy;
1675		ext_watermark += len_to_copy;
1676
1677		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1678		     CTL_FLAG_DATA_IN) {
1679			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1680					 "bytes to user\n", len_to_copy));
1681			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1682					 "to %p\n", kern_ptr, ext_ptr));
1683			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1684				ctl_set_internal_failure(ctsio,
1685							 /*sks_valid*/ 0,
1686							 /*retry_count*/ 0);
1687				goto bailout;
1688			}
1689		} else {
1690			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1691					 "bytes from user\n", len_to_copy));
1692			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1693					 "to %p\n", ext_ptr, kern_ptr));
1694			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1695				ctl_set_internal_failure(ctsio,
1696							 /*sks_valid*/ 0,
1697							 /*retry_count*/0);
1698				goto bailout;
1699			}
1700		}
1701
1702		len_copied += len_to_copy;
1703
1704		if (ext_sglist[i].len == ext_watermark) {
1705			i++;
1706			ext_watermark = 0;
1707		}
1708
1709		if (kern_sglist[j].len == kern_watermark) {
1710			j++;
1711			kern_watermark = 0;
1712		}
1713	}
1714
1715	ctsio->ext_data_filled += len_copied;
1716
1717	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1718			 "kern_sg_entries: %d\n", ext_sg_entries,
1719			 kern_sg_entries));
1720	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1721			 "kern_data_len = %d\n", ctsio->ext_data_len,
1722			 ctsio->kern_data_len));
1723
1724
1725	/* XXX KDM set residual?? */
1726bailout:
1727
1728	if (ext_sglist_malloced != 0)
1729		free(ext_sglist, M_CTL);
1730
1731	return (CTL_RETVAL_COMPLETE);
1732}
1733
1734/*
1735 * Serialize a command that went down the "wrong" side, and so was sent to
1736 * this controller for execution.  The logic is a little different than the
1737 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1738 * sent back to the other side, but in the success case, we execute the
1739 * command on this side (XFER mode) or tell the other side to execute it
1740 * (SER_ONLY mode).
1741 */
1742static int
1743ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1744{
1745	struct ctl_softc *ctl_softc;
1746	union ctl_ha_msg msg_info;
1747	struct ctl_lun *lun;
1748	int retval = 0;
1749	uint32_t targ_lun;
1750
1751	ctl_softc = control_softc;
1752
1753	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1754	lun = ctl_softc->ctl_luns[targ_lun];
1755	if (lun==NULL)
1756	{
1757		/*
1758		 * Why isn't LUN defined? The other side wouldn't
1759		 * send a cmd if the LUN is undefined.
1760		 */
1761		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1762
1763		/* "Logical unit not supported" */
1764		ctl_set_sense_data(&msg_info.scsi.sense_data,
1765				   lun,
1766				   /*sense_format*/SSD_TYPE_NONE,
1767				   /*current_error*/ 1,
1768				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1769				   /*asc*/ 0x25,
1770				   /*ascq*/ 0x00,
1771				   SSD_ELEM_NONE);
1772
1773		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1774		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1775		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1776		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1777		msg_info.hdr.serializing_sc = NULL;
1778		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1779	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1780				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1781		}
1782		return(1);
1783
1784	}
1785
1786	mtx_lock(&lun->lun_lock);
1787    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1788
1789	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1790		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1791		 ooa_links))) {
1792	case CTL_ACTION_BLOCK:
1793		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1794		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1795				  blocked_links);
1796		break;
1797	case CTL_ACTION_PASS:
1798	case CTL_ACTION_SKIP:
1799		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1800			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1801			ctl_enqueue_rtr((union ctl_io *)ctsio);
1802		} else {
1803
1804			/* send msg back to other side */
1805			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1806			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1807			msg_info.hdr.msg_type = CTL_MSG_R2R;
1808#if 0
1809			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1810#endif
1811		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1812			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1813			}
1814		}
1815		break;
1816	case CTL_ACTION_OVERLAP:
1817		/* OVERLAPPED COMMANDS ATTEMPTED */
1818		ctl_set_sense_data(&msg_info.scsi.sense_data,
1819				   lun,
1820				   /*sense_format*/SSD_TYPE_NONE,
1821				   /*current_error*/ 1,
1822				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1823				   /*asc*/ 0x4E,
1824				   /*ascq*/ 0x00,
1825				   SSD_ELEM_NONE);
1826
1827		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1828		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1829		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1830		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1831		msg_info.hdr.serializing_sc = NULL;
1832		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1833#if 0
1834		printf("BAD JUJU:Major Bummer Overlap\n");
1835#endif
1836		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1837		retval = 1;
1838		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1839		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1840		}
1841		break;
1842	case CTL_ACTION_OVERLAP_TAG:
1843		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1844		ctl_set_sense_data(&msg_info.scsi.sense_data,
1845				   lun,
1846				   /*sense_format*/SSD_TYPE_NONE,
1847				   /*current_error*/ 1,
1848				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1849				   /*asc*/ 0x4D,
1850				   /*ascq*/ ctsio->tag_num & 0xff,
1851				   SSD_ELEM_NONE);
1852
1853		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1854		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1855		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1856		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1857		msg_info.hdr.serializing_sc = NULL;
1858		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1859#if 0
1860		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1861#endif
1862		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1863		retval = 1;
1864		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1865		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1866		}
1867		break;
1868	case CTL_ACTION_ERROR:
1869	default:
1870		/* "Internal target failure" */
1871		ctl_set_sense_data(&msg_info.scsi.sense_data,
1872				   lun,
1873				   /*sense_format*/SSD_TYPE_NONE,
1874				   /*current_error*/ 1,
1875				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1876				   /*asc*/ 0x44,
1877				   /*ascq*/ 0x00,
1878				   SSD_ELEM_NONE);
1879
1880		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1881		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1882		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1883		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1884		msg_info.hdr.serializing_sc = NULL;
1885		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1886#if 0
1887		printf("BAD JUJU:Major Bummer HW Error\n");
1888#endif
1889		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1890		retval = 1;
1891		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1892		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1893		}
1894		break;
1895	}
1896	mtx_unlock(&lun->lun_lock);
1897	return (retval);
1898}
1899
1900static int
1901ctl_ioctl_submit_wait(union ctl_io *io)
1902{
1903	struct ctl_fe_ioctl_params params;
1904	ctl_fe_ioctl_state last_state;
1905	int done, retval;
1906
1907	retval = 0;
1908
1909	bzero(&params, sizeof(params));
1910
1911	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1912	cv_init(&params.sem, "ctlioccv");
1913	params.state = CTL_IOCTL_INPROG;
1914	last_state = params.state;
1915
1916	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1917
1918	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1919
1920	/* This shouldn't happen */
1921	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1922		return (retval);
1923
1924	done = 0;
1925
1926	do {
1927		mtx_lock(&params.ioctl_mtx);
1928		/*
1929		 * Check the state here, and don't sleep if the state has
1930		 * already changed (i.e. wakeup has already occured, but we
1931		 * weren't waiting yet).
1932		 */
1933		if (params.state == last_state) {
1934			/* XXX KDM cv_wait_sig instead? */
1935			cv_wait(&params.sem, &params.ioctl_mtx);
1936		}
1937		last_state = params.state;
1938
1939		switch (params.state) {
1940		case CTL_IOCTL_INPROG:
1941			/* Why did we wake up? */
1942			/* XXX KDM error here? */
1943			mtx_unlock(&params.ioctl_mtx);
1944			break;
1945		case CTL_IOCTL_DATAMOVE:
1946			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1947
1948			/*
1949			 * change last_state back to INPROG to avoid
1950			 * deadlock on subsequent data moves.
1951			 */
1952			params.state = last_state = CTL_IOCTL_INPROG;
1953
1954			mtx_unlock(&params.ioctl_mtx);
1955			ctl_ioctl_do_datamove(&io->scsiio);
1956			/*
1957			 * Note that in some cases, most notably writes,
1958			 * this will queue the I/O and call us back later.
1959			 * In other cases, generally reads, this routine
1960			 * will immediately call back and wake us up,
1961			 * probably using our own context.
1962			 */
1963			io->scsiio.be_move_done(io);
1964			break;
1965		case CTL_IOCTL_DONE:
1966			mtx_unlock(&params.ioctl_mtx);
1967			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1968			done = 1;
1969			break;
1970		default:
1971			mtx_unlock(&params.ioctl_mtx);
1972			/* XXX KDM error here? */
1973			break;
1974		}
1975	} while (done == 0);
1976
1977	mtx_destroy(&params.ioctl_mtx);
1978	cv_destroy(&params.sem);
1979
1980	return (CTL_RETVAL_COMPLETE);
1981}
1982
1983static void
1984ctl_ioctl_datamove(union ctl_io *io)
1985{
1986	struct ctl_fe_ioctl_params *params;
1987
1988	params = (struct ctl_fe_ioctl_params *)
1989		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1990
1991	mtx_lock(&params->ioctl_mtx);
1992	params->state = CTL_IOCTL_DATAMOVE;
1993	cv_broadcast(&params->sem);
1994	mtx_unlock(&params->ioctl_mtx);
1995}
1996
1997static void
1998ctl_ioctl_done(union ctl_io *io)
1999{
2000	struct ctl_fe_ioctl_params *params;
2001
2002	params = (struct ctl_fe_ioctl_params *)
2003		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2004
2005	mtx_lock(&params->ioctl_mtx);
2006	params->state = CTL_IOCTL_DONE;
2007	cv_broadcast(&params->sem);
2008	mtx_unlock(&params->ioctl_mtx);
2009}
2010
2011static void
2012ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2013{
2014	struct ctl_fe_ioctl_startstop_info *sd_info;
2015
2016	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2017
2018	sd_info->hs_info.status = metatask->status;
2019	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2020	sd_info->hs_info.luns_complete =
2021		metatask->taskinfo.startstop.luns_complete;
2022	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2023
2024	cv_broadcast(&sd_info->sem);
2025}
2026
2027static void
2028ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2029{
2030	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2031
2032	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2033
2034	mtx_lock(fe_bbr_info->lock);
2035	fe_bbr_info->bbr_info->status = metatask->status;
2036	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2037	fe_bbr_info->wakeup_done = 1;
2038	mtx_unlock(fe_bbr_info->lock);
2039
2040	cv_broadcast(&fe_bbr_info->sem);
2041}
2042
2043/*
2044 * Returns 0 for success, errno for failure.
2045 */
2046static int
2047ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2048		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2049{
2050	union ctl_io *io;
2051	int retval;
2052
2053	retval = 0;
2054
2055	mtx_lock(&lun->lun_lock);
2056	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2057	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2058	     ooa_links)) {
2059		struct ctl_ooa_entry *entry;
2060
2061		/*
2062		 * If we've got more than we can fit, just count the
2063		 * remaining entries.
2064		 */
2065		if (*cur_fill_num >= ooa_hdr->alloc_num)
2066			continue;
2067
2068		entry = &kern_entries[*cur_fill_num];
2069
2070		entry->tag_num = io->scsiio.tag_num;
2071		entry->lun_num = lun->lun;
2072#ifdef CTL_TIME_IO
2073		entry->start_bt = io->io_hdr.start_bt;
2074#endif
2075		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2076		entry->cdb_len = io->scsiio.cdb_len;
2077		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2078			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2079
2080		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2081			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2082
2083		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2084			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2085
2086		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2087			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2088
2089		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2090			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2091	}
2092	mtx_unlock(&lun->lun_lock);
2093
2094	return (retval);
2095}
2096
2097static void *
2098ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2099		 size_t error_str_len)
2100{
2101	void *kptr;
2102
2103	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2104
2105	if (copyin(user_addr, kptr, len) != 0) {
2106		snprintf(error_str, error_str_len, "Error copying %d bytes "
2107			 "from user address %p to kernel address %p", len,
2108			 user_addr, kptr);
2109		free(kptr, M_CTL);
2110		return (NULL);
2111	}
2112
2113	return (kptr);
2114}
2115
2116static void
2117ctl_free_args(int num_args, struct ctl_be_arg *args)
2118{
2119	int i;
2120
2121	if (args == NULL)
2122		return;
2123
2124	for (i = 0; i < num_args; i++) {
2125		free(args[i].kname, M_CTL);
2126		free(args[i].kvalue, M_CTL);
2127	}
2128
2129	free(args, M_CTL);
2130}
2131
2132static struct ctl_be_arg *
2133ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2134		char *error_str, size_t error_str_len)
2135{
2136	struct ctl_be_arg *args;
2137	int i;
2138
2139	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2140				error_str, error_str_len);
2141
2142	if (args == NULL)
2143		goto bailout;
2144
2145	for (i = 0; i < num_args; i++) {
2146		args[i].kname = NULL;
2147		args[i].kvalue = NULL;
2148	}
2149
2150	for (i = 0; i < num_args; i++) {
2151		uint8_t *tmpptr;
2152
2153		args[i].kname = ctl_copyin_alloc(args[i].name,
2154			args[i].namelen, error_str, error_str_len);
2155		if (args[i].kname == NULL)
2156			goto bailout;
2157
2158		if (args[i].kname[args[i].namelen - 1] != '\0') {
2159			snprintf(error_str, error_str_len, "Argument %d "
2160				 "name is not NUL-terminated", i);
2161			goto bailout;
2162		}
2163
2164		if (args[i].flags & CTL_BEARG_RD) {
2165			tmpptr = ctl_copyin_alloc(args[i].value,
2166				args[i].vallen, error_str, error_str_len);
2167			if (tmpptr == NULL)
2168				goto bailout;
2169			if ((args[i].flags & CTL_BEARG_ASCII)
2170			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2171				snprintf(error_str, error_str_len, "Argument "
2172				    "%d value is not NUL-terminated", i);
2173				goto bailout;
2174			}
2175			args[i].kvalue = tmpptr;
2176		} else {
2177			args[i].kvalue = malloc(args[i].vallen,
2178			    M_CTL, M_WAITOK | M_ZERO);
2179		}
2180	}
2181
2182	return (args);
2183bailout:
2184
2185	ctl_free_args(num_args, args);
2186
2187	return (NULL);
2188}
2189
2190static void
2191ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2192{
2193	int i;
2194
2195	for (i = 0; i < num_args; i++) {
2196		if (args[i].flags & CTL_BEARG_WR)
2197			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2198	}
2199}
2200
2201/*
2202 * Escape characters that are illegal or not recommended in XML.
2203 */
2204int
2205ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2206{
2207	char *end = str + size;
2208	int retval;
2209
2210	retval = 0;
2211
2212	for (; *str && str < end; str++) {
2213		switch (*str) {
2214		case '&':
2215			retval = sbuf_printf(sb, "&amp;");
2216			break;
2217		case '>':
2218			retval = sbuf_printf(sb, "&gt;");
2219			break;
2220		case '<':
2221			retval = sbuf_printf(sb, "&lt;");
2222			break;
2223		default:
2224			retval = sbuf_putc(sb, *str);
2225			break;
2226		}
2227
2228		if (retval != 0)
2229			break;
2230
2231	}
2232
2233	return (retval);
2234}
2235
2236static void
2237ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2238{
2239	struct scsi_vpd_id_descriptor *desc;
2240	int i;
2241
2242	if (id == NULL || id->len < 4)
2243		return;
2244	desc = (struct scsi_vpd_id_descriptor *)id->data;
2245	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2246	case SVPD_ID_TYPE_T10:
2247		sbuf_printf(sb, "t10.");
2248		break;
2249	case SVPD_ID_TYPE_EUI64:
2250		sbuf_printf(sb, "eui.");
2251		break;
2252	case SVPD_ID_TYPE_NAA:
2253		sbuf_printf(sb, "naa.");
2254		break;
2255	case SVPD_ID_TYPE_SCSI_NAME:
2256		break;
2257	}
2258	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2259	case SVPD_ID_CODESET_BINARY:
2260		for (i = 0; i < desc->length; i++)
2261			sbuf_printf(sb, "%02x", desc->identifier[i]);
2262		break;
2263	case SVPD_ID_CODESET_ASCII:
2264		sbuf_printf(sb, "%.*s", (int)desc->length,
2265		    (char *)desc->identifier);
2266		break;
2267	case SVPD_ID_CODESET_UTF8:
2268		sbuf_printf(sb, "%s", (char *)desc->identifier);
2269		break;
2270	}
2271}
2272
2273static int
2274ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2275	  struct thread *td)
2276{
2277	struct ctl_softc *softc;
2278	int retval;
2279
2280	softc = control_softc;
2281
2282	retval = 0;
2283
2284	switch (cmd) {
2285	case CTL_IO: {
2286		union ctl_io *io;
2287		void *pool_tmp;
2288
2289		/*
2290		 * If we haven't been "enabled", don't allow any SCSI I/O
2291		 * to this FETD.
2292		 */
2293		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2294			retval = EPERM;
2295			break;
2296		}
2297
2298		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2299		if (io == NULL) {
2300			printf("ctl_ioctl: can't allocate ctl_io!\n");
2301			retval = ENOSPC;
2302			break;
2303		}
2304
2305		/*
2306		 * Need to save the pool reference so it doesn't get
2307		 * spammed by the user's ctl_io.
2308		 */
2309		pool_tmp = io->io_hdr.pool;
2310
2311		memcpy(io, (void *)addr, sizeof(*io));
2312
2313		io->io_hdr.pool = pool_tmp;
2314		/*
2315		 * No status yet, so make sure the status is set properly.
2316		 */
2317		io->io_hdr.status = CTL_STATUS_NONE;
2318
2319		/*
2320		 * The user sets the initiator ID, target and LUN IDs.
2321		 */
2322		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2323		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2324		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2325		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2326			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2327
2328		retval = ctl_ioctl_submit_wait(io);
2329
2330		if (retval != 0) {
2331			ctl_free_io(io);
2332			break;
2333		}
2334
2335		memcpy((void *)addr, io, sizeof(*io));
2336
2337		/* return this to our pool */
2338		ctl_free_io(io);
2339
2340		break;
2341	}
2342	case CTL_ENABLE_PORT:
2343	case CTL_DISABLE_PORT:
2344	case CTL_SET_PORT_WWNS: {
2345		struct ctl_port *port;
2346		struct ctl_port_entry *entry;
2347
2348		entry = (struct ctl_port_entry *)addr;
2349
2350		mtx_lock(&softc->ctl_lock);
2351		STAILQ_FOREACH(port, &softc->port_list, links) {
2352			int action, done;
2353
2354			action = 0;
2355			done = 0;
2356
2357			if ((entry->port_type == CTL_PORT_NONE)
2358			 && (entry->targ_port == port->targ_port)) {
2359				/*
2360				 * If the user only wants to enable or
2361				 * disable or set WWNs on a specific port,
2362				 * do the operation and we're done.
2363				 */
2364				action = 1;
2365				done = 1;
2366			} else if (entry->port_type & port->port_type) {
2367				/*
2368				 * Compare the user's type mask with the
2369				 * particular frontend type to see if we
2370				 * have a match.
2371				 */
2372				action = 1;
2373				done = 0;
2374
2375				/*
2376				 * Make sure the user isn't trying to set
2377				 * WWNs on multiple ports at the same time.
2378				 */
2379				if (cmd == CTL_SET_PORT_WWNS) {
2380					printf("%s: Can't set WWNs on "
2381					       "multiple ports\n", __func__);
2382					retval = EINVAL;
2383					break;
2384				}
2385			}
2386			if (action != 0) {
2387				/*
2388				 * XXX KDM we have to drop the lock here,
2389				 * because the online/offline operations
2390				 * can potentially block.  We need to
2391				 * reference count the frontends so they
2392				 * can't go away,
2393				 */
2394				mtx_unlock(&softc->ctl_lock);
2395
2396				if (cmd == CTL_ENABLE_PORT) {
2397					struct ctl_lun *lun;
2398
2399					STAILQ_FOREACH(lun, &softc->lun_list,
2400						       links) {
2401						port->lun_enable(port->targ_lun_arg,
2402						    lun->target,
2403						    lun->lun);
2404					}
2405
2406					ctl_port_online(port);
2407				} else if (cmd == CTL_DISABLE_PORT) {
2408					struct ctl_lun *lun;
2409
2410					ctl_port_offline(port);
2411
2412					STAILQ_FOREACH(lun, &softc->lun_list,
2413						       links) {
2414						port->lun_disable(
2415						    port->targ_lun_arg,
2416						    lun->target,
2417						    lun->lun);
2418					}
2419				}
2420
2421				mtx_lock(&softc->ctl_lock);
2422
2423				if (cmd == CTL_SET_PORT_WWNS)
2424					ctl_port_set_wwns(port,
2425					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2426					    1 : 0, entry->wwnn,
2427					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2428					    1 : 0, entry->wwpn);
2429			}
2430			if (done != 0)
2431				break;
2432		}
2433		mtx_unlock(&softc->ctl_lock);
2434		break;
2435	}
2436	case CTL_GET_PORT_LIST: {
2437		struct ctl_port *port;
2438		struct ctl_port_list *list;
2439		int i;
2440
2441		list = (struct ctl_port_list *)addr;
2442
2443		if (list->alloc_len != (list->alloc_num *
2444		    sizeof(struct ctl_port_entry))) {
2445			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2446			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2447			       "%zu\n", __func__, list->alloc_len,
2448			       list->alloc_num, sizeof(struct ctl_port_entry));
2449			retval = EINVAL;
2450			break;
2451		}
2452		list->fill_len = 0;
2453		list->fill_num = 0;
2454		list->dropped_num = 0;
2455		i = 0;
2456		mtx_lock(&softc->ctl_lock);
2457		STAILQ_FOREACH(port, &softc->port_list, links) {
2458			struct ctl_port_entry entry, *list_entry;
2459
2460			if (list->fill_num >= list->alloc_num) {
2461				list->dropped_num++;
2462				continue;
2463			}
2464
2465			entry.port_type = port->port_type;
2466			strlcpy(entry.port_name, port->port_name,
2467				sizeof(entry.port_name));
2468			entry.targ_port = port->targ_port;
2469			entry.physical_port = port->physical_port;
2470			entry.virtual_port = port->virtual_port;
2471			entry.wwnn = port->wwnn;
2472			entry.wwpn = port->wwpn;
2473			if (port->status & CTL_PORT_STATUS_ONLINE)
2474				entry.online = 1;
2475			else
2476				entry.online = 0;
2477
2478			list_entry = &list->entries[i];
2479
2480			retval = copyout(&entry, list_entry, sizeof(entry));
2481			if (retval != 0) {
2482				printf("%s: CTL_GET_PORT_LIST: copyout "
2483				       "returned %d\n", __func__, retval);
2484				break;
2485			}
2486			i++;
2487			list->fill_num++;
2488			list->fill_len += sizeof(entry);
2489		}
2490		mtx_unlock(&softc->ctl_lock);
2491
2492		/*
2493		 * If this is non-zero, we had a copyout fault, so there's
2494		 * probably no point in attempting to set the status inside
2495		 * the structure.
2496		 */
2497		if (retval != 0)
2498			break;
2499
2500		if (list->dropped_num > 0)
2501			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2502		else
2503			list->status = CTL_PORT_LIST_OK;
2504		break;
2505	}
2506	case CTL_DUMP_OOA: {
2507		struct ctl_lun *lun;
2508		union ctl_io *io;
2509		char printbuf[128];
2510		struct sbuf sb;
2511
2512		mtx_lock(&softc->ctl_lock);
2513		printf("Dumping OOA queues:\n");
2514		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2515			mtx_lock(&lun->lun_lock);
2516			for (io = (union ctl_io *)TAILQ_FIRST(
2517			     &lun->ooa_queue); io != NULL;
2518			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2519			     ooa_links)) {
2520				sbuf_new(&sb, printbuf, sizeof(printbuf),
2521					 SBUF_FIXEDLEN);
2522				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2523					    (intmax_t)lun->lun,
2524					    io->scsiio.tag_num,
2525					    (io->io_hdr.flags &
2526					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2527					    (io->io_hdr.flags &
2528					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2529					    (io->io_hdr.flags &
2530					    CTL_FLAG_ABORT) ? " ABORT" : "",
2531			                    (io->io_hdr.flags &
2532		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2533				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2534				sbuf_finish(&sb);
2535				printf("%s\n", sbuf_data(&sb));
2536			}
2537			mtx_unlock(&lun->lun_lock);
2538		}
2539		printf("OOA queues dump done\n");
2540		mtx_unlock(&softc->ctl_lock);
2541		break;
2542	}
2543	case CTL_GET_OOA: {
2544		struct ctl_lun *lun;
2545		struct ctl_ooa *ooa_hdr;
2546		struct ctl_ooa_entry *entries;
2547		uint32_t cur_fill_num;
2548
2549		ooa_hdr = (struct ctl_ooa *)addr;
2550
2551		if ((ooa_hdr->alloc_len == 0)
2552		 || (ooa_hdr->alloc_num == 0)) {
2553			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2554			       "must be non-zero\n", __func__,
2555			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2556			retval = EINVAL;
2557			break;
2558		}
2559
2560		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2561		    sizeof(struct ctl_ooa_entry))) {
2562			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2563			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2564			       __func__, ooa_hdr->alloc_len,
2565			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2566			retval = EINVAL;
2567			break;
2568		}
2569
2570		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2571		if (entries == NULL) {
2572			printf("%s: could not allocate %d bytes for OOA "
2573			       "dump\n", __func__, ooa_hdr->alloc_len);
2574			retval = ENOMEM;
2575			break;
2576		}
2577
2578		mtx_lock(&softc->ctl_lock);
2579		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2580		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2581		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2582			mtx_unlock(&softc->ctl_lock);
2583			free(entries, M_CTL);
2584			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2585			       __func__, (uintmax_t)ooa_hdr->lun_num);
2586			retval = EINVAL;
2587			break;
2588		}
2589
2590		cur_fill_num = 0;
2591
2592		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2593			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2594				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2595					ooa_hdr, entries);
2596				if (retval != 0)
2597					break;
2598			}
2599			if (retval != 0) {
2600				mtx_unlock(&softc->ctl_lock);
2601				free(entries, M_CTL);
2602				break;
2603			}
2604		} else {
2605			lun = softc->ctl_luns[ooa_hdr->lun_num];
2606
2607			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2608						    entries);
2609		}
2610		mtx_unlock(&softc->ctl_lock);
2611
2612		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2613		ooa_hdr->fill_len = ooa_hdr->fill_num *
2614			sizeof(struct ctl_ooa_entry);
2615		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2616		if (retval != 0) {
2617			printf("%s: error copying out %d bytes for OOA dump\n",
2618			       __func__, ooa_hdr->fill_len);
2619		}
2620
2621		getbintime(&ooa_hdr->cur_bt);
2622
2623		if (cur_fill_num > ooa_hdr->alloc_num) {
2624			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2625			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2626		} else {
2627			ooa_hdr->dropped_num = 0;
2628			ooa_hdr->status = CTL_OOA_OK;
2629		}
2630
2631		free(entries, M_CTL);
2632		break;
2633	}
2634	case CTL_CHECK_OOA: {
2635		union ctl_io *io;
2636		struct ctl_lun *lun;
2637		struct ctl_ooa_info *ooa_info;
2638
2639
2640		ooa_info = (struct ctl_ooa_info *)addr;
2641
2642		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2643			ooa_info->status = CTL_OOA_INVALID_LUN;
2644			break;
2645		}
2646		mtx_lock(&softc->ctl_lock);
2647		lun = softc->ctl_luns[ooa_info->lun_id];
2648		if (lun == NULL) {
2649			mtx_unlock(&softc->ctl_lock);
2650			ooa_info->status = CTL_OOA_INVALID_LUN;
2651			break;
2652		}
2653		mtx_lock(&lun->lun_lock);
2654		mtx_unlock(&softc->ctl_lock);
2655		ooa_info->num_entries = 0;
2656		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2657		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2658		     &io->io_hdr, ooa_links)) {
2659			ooa_info->num_entries++;
2660		}
2661		mtx_unlock(&lun->lun_lock);
2662
2663		ooa_info->status = CTL_OOA_SUCCESS;
2664
2665		break;
2666	}
2667	case CTL_HARD_START:
2668	case CTL_HARD_STOP: {
2669		struct ctl_fe_ioctl_startstop_info ss_info;
2670		struct cfi_metatask *metatask;
2671		struct mtx hs_mtx;
2672
2673		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2674
2675		cv_init(&ss_info.sem, "hard start/stop cv" );
2676
2677		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2678		if (metatask == NULL) {
2679			retval = ENOMEM;
2680			mtx_destroy(&hs_mtx);
2681			break;
2682		}
2683
2684		if (cmd == CTL_HARD_START)
2685			metatask->tasktype = CFI_TASK_STARTUP;
2686		else
2687			metatask->tasktype = CFI_TASK_SHUTDOWN;
2688
2689		metatask->callback = ctl_ioctl_hard_startstop_callback;
2690		metatask->callback_arg = &ss_info;
2691
2692		cfi_action(metatask);
2693
2694		/* Wait for the callback */
2695		mtx_lock(&hs_mtx);
2696		cv_wait_sig(&ss_info.sem, &hs_mtx);
2697		mtx_unlock(&hs_mtx);
2698
2699		/*
2700		 * All information has been copied from the metatask by the
2701		 * time cv_broadcast() is called, so we free the metatask here.
2702		 */
2703		cfi_free_metatask(metatask);
2704
2705		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2706
2707		mtx_destroy(&hs_mtx);
2708		break;
2709	}
2710	case CTL_BBRREAD: {
2711		struct ctl_bbrread_info *bbr_info;
2712		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2713		struct mtx bbr_mtx;
2714		struct cfi_metatask *metatask;
2715
2716		bbr_info = (struct ctl_bbrread_info *)addr;
2717
2718		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2719
2720		bzero(&bbr_mtx, sizeof(bbr_mtx));
2721		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2722
2723		fe_bbr_info.bbr_info = bbr_info;
2724		fe_bbr_info.lock = &bbr_mtx;
2725
2726		cv_init(&fe_bbr_info.sem, "BBR read cv");
2727		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2728
2729		if (metatask == NULL) {
2730			mtx_destroy(&bbr_mtx);
2731			cv_destroy(&fe_bbr_info.sem);
2732			retval = ENOMEM;
2733			break;
2734		}
2735		metatask->tasktype = CFI_TASK_BBRREAD;
2736		metatask->callback = ctl_ioctl_bbrread_callback;
2737		metatask->callback_arg = &fe_bbr_info;
2738		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2739		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2740		metatask->taskinfo.bbrread.len = bbr_info->len;
2741
2742		cfi_action(metatask);
2743
2744		mtx_lock(&bbr_mtx);
2745		while (fe_bbr_info.wakeup_done == 0)
2746			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2747		mtx_unlock(&bbr_mtx);
2748
2749		bbr_info->status = metatask->status;
2750		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2751		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2752		memcpy(&bbr_info->sense_data,
2753		       &metatask->taskinfo.bbrread.sense_data,
2754		       ctl_min(sizeof(bbr_info->sense_data),
2755			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2756
2757		cfi_free_metatask(metatask);
2758
2759		mtx_destroy(&bbr_mtx);
2760		cv_destroy(&fe_bbr_info.sem);
2761
2762		break;
2763	}
2764	case CTL_DELAY_IO: {
2765		struct ctl_io_delay_info *delay_info;
2766#ifdef CTL_IO_DELAY
2767		struct ctl_lun *lun;
2768#endif /* CTL_IO_DELAY */
2769
2770		delay_info = (struct ctl_io_delay_info *)addr;
2771
2772#ifdef CTL_IO_DELAY
2773		mtx_lock(&softc->ctl_lock);
2774
2775		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2776		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2777			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2778		} else {
2779			lun = softc->ctl_luns[delay_info->lun_id];
2780			mtx_lock(&lun->lun_lock);
2781
2782			delay_info->status = CTL_DELAY_STATUS_OK;
2783
2784			switch (delay_info->delay_type) {
2785			case CTL_DELAY_TYPE_CONT:
2786				break;
2787			case CTL_DELAY_TYPE_ONESHOT:
2788				break;
2789			default:
2790				delay_info->status =
2791					CTL_DELAY_STATUS_INVALID_TYPE;
2792				break;
2793			}
2794
2795			switch (delay_info->delay_loc) {
2796			case CTL_DELAY_LOC_DATAMOVE:
2797				lun->delay_info.datamove_type =
2798					delay_info->delay_type;
2799				lun->delay_info.datamove_delay =
2800					delay_info->delay_secs;
2801				break;
2802			case CTL_DELAY_LOC_DONE:
2803				lun->delay_info.done_type =
2804					delay_info->delay_type;
2805				lun->delay_info.done_delay =
2806					delay_info->delay_secs;
2807				break;
2808			default:
2809				delay_info->status =
2810					CTL_DELAY_STATUS_INVALID_LOC;
2811				break;
2812			}
2813			mtx_unlock(&lun->lun_lock);
2814		}
2815
2816		mtx_unlock(&softc->ctl_lock);
2817#else
2818		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2819#endif /* CTL_IO_DELAY */
2820		break;
2821	}
2822	case CTL_REALSYNC_SET: {
2823		int *syncstate;
2824
2825		syncstate = (int *)addr;
2826
2827		mtx_lock(&softc->ctl_lock);
2828		switch (*syncstate) {
2829		case 0:
2830			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2831			break;
2832		case 1:
2833			softc->flags |= CTL_FLAG_REAL_SYNC;
2834			break;
2835		default:
2836			retval = EINVAL;
2837			break;
2838		}
2839		mtx_unlock(&softc->ctl_lock);
2840		break;
2841	}
2842	case CTL_REALSYNC_GET: {
2843		int *syncstate;
2844
2845		syncstate = (int*)addr;
2846
2847		mtx_lock(&softc->ctl_lock);
2848		if (softc->flags & CTL_FLAG_REAL_SYNC)
2849			*syncstate = 1;
2850		else
2851			*syncstate = 0;
2852		mtx_unlock(&softc->ctl_lock);
2853
2854		break;
2855	}
2856	case CTL_SETSYNC:
2857	case CTL_GETSYNC: {
2858		struct ctl_sync_info *sync_info;
2859		struct ctl_lun *lun;
2860
2861		sync_info = (struct ctl_sync_info *)addr;
2862
2863		mtx_lock(&softc->ctl_lock);
2864		lun = softc->ctl_luns[sync_info->lun_id];
2865		if (lun == NULL) {
2866			mtx_unlock(&softc->ctl_lock);
2867			sync_info->status = CTL_GS_SYNC_NO_LUN;
2868		}
2869		/*
2870		 * Get or set the sync interval.  We're not bounds checking
2871		 * in the set case, hopefully the user won't do something
2872		 * silly.
2873		 */
2874		mtx_lock(&lun->lun_lock);
2875		mtx_unlock(&softc->ctl_lock);
2876		if (cmd == CTL_GETSYNC)
2877			sync_info->sync_interval = lun->sync_interval;
2878		else
2879			lun->sync_interval = sync_info->sync_interval;
2880		mtx_unlock(&lun->lun_lock);
2881
2882		sync_info->status = CTL_GS_SYNC_OK;
2883
2884		break;
2885	}
2886	case CTL_GETSTATS: {
2887		struct ctl_stats *stats;
2888		struct ctl_lun *lun;
2889		int i;
2890
2891		stats = (struct ctl_stats *)addr;
2892
2893		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2894		     stats->alloc_len) {
2895			stats->status = CTL_SS_NEED_MORE_SPACE;
2896			stats->num_luns = softc->num_luns;
2897			break;
2898		}
2899		/*
2900		 * XXX KDM no locking here.  If the LUN list changes,
2901		 * things can blow up.
2902		 */
2903		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2904		     i++, lun = STAILQ_NEXT(lun, links)) {
2905			retval = copyout(&lun->stats, &stats->lun_stats[i],
2906					 sizeof(lun->stats));
2907			if (retval != 0)
2908				break;
2909		}
2910		stats->num_luns = softc->num_luns;
2911		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2912				 softc->num_luns;
2913		stats->status = CTL_SS_OK;
2914#ifdef CTL_TIME_IO
2915		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2916#else
2917		stats->flags = CTL_STATS_FLAG_NONE;
2918#endif
2919		getnanouptime(&stats->timestamp);
2920		break;
2921	}
2922	case CTL_ERROR_INJECT: {
2923		struct ctl_error_desc *err_desc, *new_err_desc;
2924		struct ctl_lun *lun;
2925
2926		err_desc = (struct ctl_error_desc *)addr;
2927
2928		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2929				      M_WAITOK | M_ZERO);
2930		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2931
2932		mtx_lock(&softc->ctl_lock);
2933		lun = softc->ctl_luns[err_desc->lun_id];
2934		if (lun == NULL) {
2935			mtx_unlock(&softc->ctl_lock);
2936			free(new_err_desc, M_CTL);
2937			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2938			       __func__, (uintmax_t)err_desc->lun_id);
2939			retval = EINVAL;
2940			break;
2941		}
2942		mtx_lock(&lun->lun_lock);
2943		mtx_unlock(&softc->ctl_lock);
2944
2945		/*
2946		 * We could do some checking here to verify the validity
2947		 * of the request, but given the complexity of error
2948		 * injection requests, the checking logic would be fairly
2949		 * complex.
2950		 *
2951		 * For now, if the request is invalid, it just won't get
2952		 * executed and might get deleted.
2953		 */
2954		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2955
2956		/*
2957		 * XXX KDM check to make sure the serial number is unique,
2958		 * in case we somehow manage to wrap.  That shouldn't
2959		 * happen for a very long time, but it's the right thing to
2960		 * do.
2961		 */
2962		new_err_desc->serial = lun->error_serial;
2963		err_desc->serial = lun->error_serial;
2964		lun->error_serial++;
2965
2966		mtx_unlock(&lun->lun_lock);
2967		break;
2968	}
2969	case CTL_ERROR_INJECT_DELETE: {
2970		struct ctl_error_desc *delete_desc, *desc, *desc2;
2971		struct ctl_lun *lun;
2972		int delete_done;
2973
2974		delete_desc = (struct ctl_error_desc *)addr;
2975		delete_done = 0;
2976
2977		mtx_lock(&softc->ctl_lock);
2978		lun = softc->ctl_luns[delete_desc->lun_id];
2979		if (lun == NULL) {
2980			mtx_unlock(&softc->ctl_lock);
2981			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2982			       __func__, (uintmax_t)delete_desc->lun_id);
2983			retval = EINVAL;
2984			break;
2985		}
2986		mtx_lock(&lun->lun_lock);
2987		mtx_unlock(&softc->ctl_lock);
2988		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2989			if (desc->serial != delete_desc->serial)
2990				continue;
2991
2992			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2993				      links);
2994			free(desc, M_CTL);
2995			delete_done = 1;
2996		}
2997		mtx_unlock(&lun->lun_lock);
2998		if (delete_done == 0) {
2999			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3000			       "error serial %ju on LUN %u\n", __func__,
3001			       delete_desc->serial, delete_desc->lun_id);
3002			retval = EINVAL;
3003			break;
3004		}
3005		break;
3006	}
3007	case CTL_DUMP_STRUCTS: {
3008		int i, j, k, idx;
3009		struct ctl_port *port;
3010		struct ctl_frontend *fe;
3011
3012		mtx_lock(&softc->ctl_lock);
3013		printf("CTL Persistent Reservation information start:\n");
3014		for (i = 0; i < CTL_MAX_LUNS; i++) {
3015			struct ctl_lun *lun;
3016
3017			lun = softc->ctl_luns[i];
3018
3019			if ((lun == NULL)
3020			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3021				continue;
3022
3023			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3024				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3025					idx = j * CTL_MAX_INIT_PER_PORT + k;
3026					if (lun->pr_keys[idx] == 0)
3027						continue;
3028					printf("  LUN %d port %d iid %d key "
3029					       "%#jx\n", i, j, k,
3030					       (uintmax_t)lun->pr_keys[idx]);
3031				}
3032			}
3033		}
3034		printf("CTL Persistent Reservation information end\n");
3035		printf("CTL Ports:\n");
3036		STAILQ_FOREACH(port, &softc->port_list, links) {
3037			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3038			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3039			       port->frontend->name, port->port_type,
3040			       port->physical_port, port->virtual_port,
3041			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3042			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3043				if (port->wwpn_iid[j].in_use == 0 &&
3044				    port->wwpn_iid[j].wwpn == 0 &&
3045				    port->wwpn_iid[j].name == NULL)
3046					continue;
3047
3048				printf("    iid %u use %d WWPN %#jx '%s'\n",
3049				    j, port->wwpn_iid[j].in_use,
3050				    (uintmax_t)port->wwpn_iid[j].wwpn,
3051				    port->wwpn_iid[j].name);
3052			}
3053		}
3054		printf("CTL Port information end\n");
3055		mtx_unlock(&softc->ctl_lock);
3056		/*
3057		 * XXX KDM calling this without a lock.  We'd likely want
3058		 * to drop the lock before calling the frontend's dump
3059		 * routine anyway.
3060		 */
3061		printf("CTL Frontends:\n");
3062		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3063			printf("  Frontend '%s'\n", fe->name);
3064			if (fe->fe_dump != NULL)
3065				fe->fe_dump();
3066		}
3067		printf("CTL Frontend information end\n");
3068		break;
3069	}
3070	case CTL_LUN_REQ: {
3071		struct ctl_lun_req *lun_req;
3072		struct ctl_backend_driver *backend;
3073
3074		lun_req = (struct ctl_lun_req *)addr;
3075
3076		backend = ctl_backend_find(lun_req->backend);
3077		if (backend == NULL) {
3078			lun_req->status = CTL_LUN_ERROR;
3079			snprintf(lun_req->error_str,
3080				 sizeof(lun_req->error_str),
3081				 "Backend \"%s\" not found.",
3082				 lun_req->backend);
3083			break;
3084		}
3085		if (lun_req->num_be_args > 0) {
3086			lun_req->kern_be_args = ctl_copyin_args(
3087				lun_req->num_be_args,
3088				lun_req->be_args,
3089				lun_req->error_str,
3090				sizeof(lun_req->error_str));
3091			if (lun_req->kern_be_args == NULL) {
3092				lun_req->status = CTL_LUN_ERROR;
3093				break;
3094			}
3095		}
3096
3097		retval = backend->ioctl(dev, cmd, addr, flag, td);
3098
3099		if (lun_req->num_be_args > 0) {
3100			ctl_copyout_args(lun_req->num_be_args,
3101				      lun_req->kern_be_args);
3102			ctl_free_args(lun_req->num_be_args,
3103				      lun_req->kern_be_args);
3104		}
3105		break;
3106	}
3107	case CTL_LUN_LIST: {
3108		struct sbuf *sb;
3109		struct ctl_lun *lun;
3110		struct ctl_lun_list *list;
3111		struct ctl_option *opt;
3112
3113		list = (struct ctl_lun_list *)addr;
3114
3115		/*
3116		 * Allocate a fixed length sbuf here, based on the length
3117		 * of the user's buffer.  We could allocate an auto-extending
3118		 * buffer, and then tell the user how much larger our
3119		 * amount of data is than his buffer, but that presents
3120		 * some problems:
3121		 *
3122		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3123		 *     we can't hold a lock while calling them with an
3124		 *     auto-extending buffer.
3125 		 *
3126		 * 2.  There is not currently a LUN reference counting
3127		 *     mechanism, outside of outstanding transactions on
3128		 *     the LUN's OOA queue.  So a LUN could go away on us
3129		 *     while we're getting the LUN number, backend-specific
3130		 *     information, etc.  Thus, given the way things
3131		 *     currently work, we need to hold the CTL lock while
3132		 *     grabbing LUN information.
3133		 *
3134		 * So, from the user's standpoint, the best thing to do is
3135		 * allocate what he thinks is a reasonable buffer length,
3136		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3137		 * double the buffer length and try again.  (And repeat
3138		 * that until he succeeds.)
3139		 */
3140		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3141		if (sb == NULL) {
3142			list->status = CTL_LUN_LIST_ERROR;
3143			snprintf(list->error_str, sizeof(list->error_str),
3144				 "Unable to allocate %d bytes for LUN list",
3145				 list->alloc_len);
3146			break;
3147		}
3148
3149		sbuf_printf(sb, "<ctllunlist>\n");
3150
3151		mtx_lock(&softc->ctl_lock);
3152		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3153			mtx_lock(&lun->lun_lock);
3154			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3155					     (uintmax_t)lun->lun);
3156
3157			/*
3158			 * Bail out as soon as we see that we've overfilled
3159			 * the buffer.
3160			 */
3161			if (retval != 0)
3162				break;
3163
3164			retval = sbuf_printf(sb, "\t<backend_type>%s"
3165					     "</backend_type>\n",
3166					     (lun->backend == NULL) ?  "none" :
3167					     lun->backend->name);
3168
3169			if (retval != 0)
3170				break;
3171
3172			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3173					     lun->be_lun->lun_type);
3174
3175			if (retval != 0)
3176				break;
3177
3178			if (lun->backend == NULL) {
3179				retval = sbuf_printf(sb, "</lun>\n");
3180				if (retval != 0)
3181					break;
3182				continue;
3183			}
3184
3185			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3186					     (lun->be_lun->maxlba > 0) ?
3187					     lun->be_lun->maxlba + 1 : 0);
3188
3189			if (retval != 0)
3190				break;
3191
3192			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3193					     lun->be_lun->blocksize);
3194
3195			if (retval != 0)
3196				break;
3197
3198			retval = sbuf_printf(sb, "\t<serial_number>");
3199
3200			if (retval != 0)
3201				break;
3202
3203			retval = ctl_sbuf_printf_esc(sb,
3204			    lun->be_lun->serial_num,
3205			    sizeof(lun->be_lun->serial_num));
3206
3207			if (retval != 0)
3208				break;
3209
3210			retval = sbuf_printf(sb, "</serial_number>\n");
3211
3212			if (retval != 0)
3213				break;
3214
3215			retval = sbuf_printf(sb, "\t<device_id>");
3216
3217			if (retval != 0)
3218				break;
3219
3220			retval = ctl_sbuf_printf_esc(sb,
3221			    lun->be_lun->device_id,
3222			    sizeof(lun->be_lun->device_id));
3223
3224			if (retval != 0)
3225				break;
3226
3227			retval = sbuf_printf(sb, "</device_id>\n");
3228
3229			if (retval != 0)
3230				break;
3231
3232			if (lun->backend->lun_info != NULL) {
3233				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3234				if (retval != 0)
3235					break;
3236			}
3237			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3238				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3239				    opt->name, opt->value, opt->name);
3240				if (retval != 0)
3241					break;
3242			}
3243
3244			retval = sbuf_printf(sb, "</lun>\n");
3245
3246			if (retval != 0)
3247				break;
3248			mtx_unlock(&lun->lun_lock);
3249		}
3250		if (lun != NULL)
3251			mtx_unlock(&lun->lun_lock);
3252		mtx_unlock(&softc->ctl_lock);
3253
3254		if ((retval != 0)
3255		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3256			retval = 0;
3257			sbuf_delete(sb);
3258			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3259			snprintf(list->error_str, sizeof(list->error_str),
3260				 "Out of space, %d bytes is too small",
3261				 list->alloc_len);
3262			break;
3263		}
3264
3265		sbuf_finish(sb);
3266
3267		retval = copyout(sbuf_data(sb), list->lun_xml,
3268				 sbuf_len(sb) + 1);
3269
3270		list->fill_len = sbuf_len(sb) + 1;
3271		list->status = CTL_LUN_LIST_OK;
3272		sbuf_delete(sb);
3273		break;
3274	}
3275	case CTL_ISCSI: {
3276		struct ctl_iscsi *ci;
3277		struct ctl_frontend *fe;
3278
3279		ci = (struct ctl_iscsi *)addr;
3280
3281		fe = ctl_frontend_find("iscsi");
3282		if (fe == NULL) {
3283			ci->status = CTL_ISCSI_ERROR;
3284			snprintf(ci->error_str, sizeof(ci->error_str),
3285			    "Frontend \"iscsi\" not found.");
3286			break;
3287		}
3288
3289		retval = fe->ioctl(dev, cmd, addr, flag, td);
3290		break;
3291	}
3292	case CTL_PORT_REQ: {
3293		struct ctl_req *req;
3294		struct ctl_frontend *fe;
3295
3296		req = (struct ctl_req *)addr;
3297
3298		fe = ctl_frontend_find(req->driver);
3299		if (fe == NULL) {
3300			req->status = CTL_LUN_ERROR;
3301			snprintf(req->error_str, sizeof(req->error_str),
3302			    "Frontend \"%s\" not found.", req->driver);
3303			break;
3304		}
3305		if (req->num_args > 0) {
3306			req->kern_args = ctl_copyin_args(req->num_args,
3307			    req->args, req->error_str, sizeof(req->error_str));
3308			if (req->kern_args == NULL) {
3309				req->status = CTL_LUN_ERROR;
3310				break;
3311			}
3312		}
3313
3314		retval = fe->ioctl(dev, cmd, addr, flag, td);
3315
3316		if (req->num_args > 0) {
3317			ctl_copyout_args(req->num_args, req->kern_args);
3318			ctl_free_args(req->num_args, req->kern_args);
3319		}
3320		break;
3321	}
3322	case CTL_PORT_LIST: {
3323		struct sbuf *sb;
3324		struct ctl_port *port;
3325		struct ctl_lun_list *list;
3326		struct ctl_option *opt;
3327		int j;
3328
3329		list = (struct ctl_lun_list *)addr;
3330
3331		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3332		if (sb == NULL) {
3333			list->status = CTL_LUN_LIST_ERROR;
3334			snprintf(list->error_str, sizeof(list->error_str),
3335				 "Unable to allocate %d bytes for LUN list",
3336				 list->alloc_len);
3337			break;
3338		}
3339
3340		sbuf_printf(sb, "<ctlportlist>\n");
3341
3342		mtx_lock(&softc->ctl_lock);
3343		STAILQ_FOREACH(port, &softc->port_list, links) {
3344			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3345					     (uintmax_t)port->targ_port);
3346
3347			/*
3348			 * Bail out as soon as we see that we've overfilled
3349			 * the buffer.
3350			 */
3351			if (retval != 0)
3352				break;
3353
3354			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3355			    "</frontend_type>\n", port->frontend->name);
3356			if (retval != 0)
3357				break;
3358
3359			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3360					     port->port_type);
3361			if (retval != 0)
3362				break;
3363
3364			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3365			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3366			if (retval != 0)
3367				break;
3368
3369			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3370			    port->port_name);
3371			if (retval != 0)
3372				break;
3373
3374			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3375			    port->physical_port);
3376			if (retval != 0)
3377				break;
3378
3379			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3380			    port->virtual_port);
3381			if (retval != 0)
3382				break;
3383
3384			if (port->target_devid != NULL) {
3385				sbuf_printf(sb, "\t<target>");
3386				ctl_id_sbuf(port->target_devid, sb);
3387				sbuf_printf(sb, "</target>\n");
3388			}
3389
3390			if (port->port_devid != NULL) {
3391				sbuf_printf(sb, "\t<port>");
3392				ctl_id_sbuf(port->port_devid, sb);
3393				sbuf_printf(sb, "</port>\n");
3394			}
3395
3396			if (port->port_info != NULL) {
3397				retval = port->port_info(port->onoff_arg, sb);
3398				if (retval != 0)
3399					break;
3400			}
3401			STAILQ_FOREACH(opt, &port->options, links) {
3402				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3403				    opt->name, opt->value, opt->name);
3404				if (retval != 0)
3405					break;
3406			}
3407
3408			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3409				if (port->wwpn_iid[j].in_use == 0 ||
3410				    (port->wwpn_iid[j].wwpn == 0 &&
3411				     port->wwpn_iid[j].name == NULL))
3412					continue;
3413
3414				if (port->wwpn_iid[j].name != NULL)
3415					retval = sbuf_printf(sb,
3416					    "\t<initiator>%u %s</initiator>\n",
3417					    j, port->wwpn_iid[j].name);
3418				else
3419					retval = sbuf_printf(sb,
3420					    "\t<initiator>%u naa.%08jx</initiator>\n",
3421					    j, port->wwpn_iid[j].wwpn);
3422				if (retval != 0)
3423					break;
3424			}
3425			if (retval != 0)
3426				break;
3427
3428			retval = sbuf_printf(sb, "</targ_port>\n");
3429			if (retval != 0)
3430				break;
3431		}
3432		mtx_unlock(&softc->ctl_lock);
3433
3434		if ((retval != 0)
3435		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3436			retval = 0;
3437			sbuf_delete(sb);
3438			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3439			snprintf(list->error_str, sizeof(list->error_str),
3440				 "Out of space, %d bytes is too small",
3441				 list->alloc_len);
3442			break;
3443		}
3444
3445		sbuf_finish(sb);
3446
3447		retval = copyout(sbuf_data(sb), list->lun_xml,
3448				 sbuf_len(sb) + 1);
3449
3450		list->fill_len = sbuf_len(sb) + 1;
3451		list->status = CTL_LUN_LIST_OK;
3452		sbuf_delete(sb);
3453		break;
3454	}
3455	default: {
3456		/* XXX KDM should we fix this? */
3457#if 0
3458		struct ctl_backend_driver *backend;
3459		unsigned int type;
3460		int found;
3461
3462		found = 0;
3463
3464		/*
3465		 * We encode the backend type as the ioctl type for backend
3466		 * ioctls.  So parse it out here, and then search for a
3467		 * backend of this type.
3468		 */
3469		type = _IOC_TYPE(cmd);
3470
3471		STAILQ_FOREACH(backend, &softc->be_list, links) {
3472			if (backend->type == type) {
3473				found = 1;
3474				break;
3475			}
3476		}
3477		if (found == 0) {
3478			printf("ctl: unknown ioctl command %#lx or backend "
3479			       "%d\n", cmd, type);
3480			retval = EINVAL;
3481			break;
3482		}
3483		retval = backend->ioctl(dev, cmd, addr, flag, td);
3484#endif
3485		retval = ENOTTY;
3486		break;
3487	}
3488	}
3489	return (retval);
3490}
3491
3492uint32_t
3493ctl_get_initindex(struct ctl_nexus *nexus)
3494{
3495	if (nexus->targ_port < CTL_MAX_PORTS)
3496		return (nexus->initid.id +
3497			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3498	else
3499		return (nexus->initid.id +
3500		       ((nexus->targ_port - CTL_MAX_PORTS) *
3501			CTL_MAX_INIT_PER_PORT));
3502}
3503
3504uint32_t
3505ctl_get_resindex(struct ctl_nexus *nexus)
3506{
3507	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3508}
3509
3510uint32_t
3511ctl_port_idx(int port_num)
3512{
3513	if (port_num < CTL_MAX_PORTS)
3514		return(port_num);
3515	else
3516		return(port_num - CTL_MAX_PORTS);
3517}
3518
3519static uint32_t
3520ctl_map_lun(int port_num, uint32_t lun_id)
3521{
3522	struct ctl_port *port;
3523
3524	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3525	if (port == NULL)
3526		return (UINT32_MAX);
3527	if (port->lun_map == NULL)
3528		return (lun_id);
3529	return (port->lun_map(port->targ_lun_arg, lun_id));
3530}
3531
3532static uint32_t
3533ctl_map_lun_back(int port_num, uint32_t lun_id)
3534{
3535	struct ctl_port *port;
3536	uint32_t i;
3537
3538	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3539	if (port->lun_map == NULL)
3540		return (lun_id);
3541	for (i = 0; i < CTL_MAX_LUNS; i++) {
3542		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3543			return (i);
3544	}
3545	return (UINT32_MAX);
3546}
3547
3548/*
3549 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3550 * that are a power of 2.
3551 */
3552int
3553ctl_ffz(uint32_t *mask, uint32_t size)
3554{
3555	uint32_t num_chunks, num_pieces;
3556	int i, j;
3557
3558	num_chunks = (size >> 5);
3559	if (num_chunks == 0)
3560		num_chunks++;
3561	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3562
3563	for (i = 0; i < num_chunks; i++) {
3564		for (j = 0; j < num_pieces; j++) {
3565			if ((mask[i] & (1 << j)) == 0)
3566				return ((i << 5) + j);
3567		}
3568	}
3569
3570	return (-1);
3571}
3572
3573int
3574ctl_set_mask(uint32_t *mask, uint32_t bit)
3575{
3576	uint32_t chunk, piece;
3577
3578	chunk = bit >> 5;
3579	piece = bit % (sizeof(uint32_t) * 8);
3580
3581	if ((mask[chunk] & (1 << piece)) != 0)
3582		return (-1);
3583	else
3584		mask[chunk] |= (1 << piece);
3585
3586	return (0);
3587}
3588
3589int
3590ctl_clear_mask(uint32_t *mask, uint32_t bit)
3591{
3592	uint32_t chunk, piece;
3593
3594	chunk = bit >> 5;
3595	piece = bit % (sizeof(uint32_t) * 8);
3596
3597	if ((mask[chunk] & (1 << piece)) == 0)
3598		return (-1);
3599	else
3600		mask[chunk] &= ~(1 << piece);
3601
3602	return (0);
3603}
3604
3605int
3606ctl_is_set(uint32_t *mask, uint32_t bit)
3607{
3608	uint32_t chunk, piece;
3609
3610	chunk = bit >> 5;
3611	piece = bit % (sizeof(uint32_t) * 8);
3612
3613	if ((mask[chunk] & (1 << piece)) == 0)
3614		return (0);
3615	else
3616		return (1);
3617}
3618
3619#ifdef unused
3620/*
3621 * The bus, target and lun are optional, they can be filled in later.
3622 * can_wait is used to determine whether we can wait on the malloc or not.
3623 */
3624union ctl_io*
3625ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3626	      uint32_t targ_lun, int can_wait)
3627{
3628	union ctl_io *io;
3629
3630	if (can_wait)
3631		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3632	else
3633		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3634
3635	if (io != NULL) {
3636		io->io_hdr.io_type = io_type;
3637		io->io_hdr.targ_port = targ_port;
3638		/*
3639		 * XXX KDM this needs to change/go away.  We need to move
3640		 * to a preallocated pool of ctl_scsiio structures.
3641		 */
3642		io->io_hdr.nexus.targ_target.id = targ_target;
3643		io->io_hdr.nexus.targ_lun = targ_lun;
3644	}
3645
3646	return (io);
3647}
3648
3649void
3650ctl_kfree_io(union ctl_io *io)
3651{
3652	free(io, M_CTL);
3653}
3654#endif /* unused */
3655
3656/*
3657 * ctl_softc, pool_type, total_ctl_io are passed in.
3658 * npool is passed out.
3659 */
3660int
3661ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3662		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3663{
3664	uint32_t i;
3665	union ctl_io *cur_io, *next_io;
3666	struct ctl_io_pool *pool;
3667	int retval;
3668
3669	retval = 0;
3670
3671	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3672					    M_NOWAIT | M_ZERO);
3673	if (pool == NULL) {
3674		retval = ENOMEM;
3675		goto bailout;
3676	}
3677
3678	pool->type = pool_type;
3679	pool->ctl_softc = ctl_softc;
3680
3681	mtx_lock(&ctl_softc->pool_lock);
3682	pool->id = ctl_softc->cur_pool_id++;
3683	mtx_unlock(&ctl_softc->pool_lock);
3684
3685	pool->flags = CTL_POOL_FLAG_NONE;
3686	pool->refcount = 1;		/* Reference for validity. */
3687	STAILQ_INIT(&pool->free_queue);
3688
3689	/*
3690	 * XXX KDM other options here:
3691	 * - allocate a page at a time
3692	 * - allocate one big chunk of memory.
3693	 * Page allocation might work well, but would take a little more
3694	 * tracking.
3695	 */
3696	for (i = 0; i < total_ctl_io; i++) {
3697		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3698						M_NOWAIT);
3699		if (cur_io == NULL) {
3700			retval = ENOMEM;
3701			break;
3702		}
3703		cur_io->io_hdr.pool = pool;
3704		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3705		pool->total_ctl_io++;
3706		pool->free_ctl_io++;
3707	}
3708
3709	if (retval != 0) {
3710		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3711		     cur_io != NULL; cur_io = next_io) {
3712			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3713							      links);
3714			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3715				      ctl_io_hdr, links);
3716			free(cur_io, M_CTLIO);
3717		}
3718
3719		free(pool, M_CTL);
3720		goto bailout;
3721	}
3722	mtx_lock(&ctl_softc->pool_lock);
3723	ctl_softc->num_pools++;
3724	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3725	/*
3726	 * Increment our usage count if this is an external consumer, so we
3727	 * can't get unloaded until the external consumer (most likely a
3728	 * FETD) unloads and frees his pool.
3729	 *
3730	 * XXX KDM will this increment the caller's module use count, or
3731	 * mine?
3732	 */
3733#if 0
3734	if ((pool_type != CTL_POOL_EMERGENCY)
3735	 && (pool_type != CTL_POOL_INTERNAL)
3736	 && (pool_type != CTL_POOL_4OTHERSC))
3737		MOD_INC_USE_COUNT;
3738#endif
3739
3740	mtx_unlock(&ctl_softc->pool_lock);
3741
3742	*npool = pool;
3743
3744bailout:
3745
3746	return (retval);
3747}
3748
3749static int
3750ctl_pool_acquire(struct ctl_io_pool *pool)
3751{
3752
3753	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3754
3755	if (pool->flags & CTL_POOL_FLAG_INVALID)
3756		return (EINVAL);
3757
3758	pool->refcount++;
3759
3760	return (0);
3761}
3762
3763static void
3764ctl_pool_release(struct ctl_io_pool *pool)
3765{
3766	struct ctl_softc *ctl_softc = pool->ctl_softc;
3767	union ctl_io *io;
3768
3769	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3770
3771	if (--pool->refcount != 0)
3772		return;
3773
3774	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3775		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3776			      links);
3777		free(io, M_CTLIO);
3778	}
3779
3780	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3781	ctl_softc->num_pools--;
3782
3783	/*
3784	 * XXX KDM will this decrement the caller's usage count or mine?
3785	 */
3786#if 0
3787	if ((pool->type != CTL_POOL_EMERGENCY)
3788	 && (pool->type != CTL_POOL_INTERNAL)
3789	 && (pool->type != CTL_POOL_4OTHERSC))
3790		MOD_DEC_USE_COUNT;
3791#endif
3792
3793	free(pool, M_CTL);
3794}
3795
3796void
3797ctl_pool_free(struct ctl_io_pool *pool)
3798{
3799	struct ctl_softc *ctl_softc;
3800
3801	if (pool == NULL)
3802		return;
3803
3804	ctl_softc = pool->ctl_softc;
3805	mtx_lock(&ctl_softc->pool_lock);
3806	pool->flags |= CTL_POOL_FLAG_INVALID;
3807	ctl_pool_release(pool);
3808	mtx_unlock(&ctl_softc->pool_lock);
3809}
3810
3811/*
3812 * This routine does not block (except for spinlocks of course).
3813 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3814 * possible.
3815 */
3816union ctl_io *
3817ctl_alloc_io(void *pool_ref)
3818{
3819	union ctl_io *io;
3820	struct ctl_softc *ctl_softc;
3821	struct ctl_io_pool *pool, *npool;
3822	struct ctl_io_pool *emergency_pool;
3823
3824	pool = (struct ctl_io_pool *)pool_ref;
3825
3826	if (pool == NULL) {
3827		printf("%s: pool is NULL\n", __func__);
3828		return (NULL);
3829	}
3830
3831	emergency_pool = NULL;
3832
3833	ctl_softc = pool->ctl_softc;
3834
3835	mtx_lock(&ctl_softc->pool_lock);
3836	/*
3837	 * First, try to get the io structure from the user's pool.
3838	 */
3839	if (ctl_pool_acquire(pool) == 0) {
3840		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3841		if (io != NULL) {
3842			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3843			pool->total_allocated++;
3844			pool->free_ctl_io--;
3845			mtx_unlock(&ctl_softc->pool_lock);
3846			return (io);
3847		} else
3848			ctl_pool_release(pool);
3849	}
3850	/*
3851	 * If he doesn't have any io structures left, search for an
3852	 * emergency pool and grab one from there.
3853	 */
3854	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3855		if (npool->type != CTL_POOL_EMERGENCY)
3856			continue;
3857
3858		if (ctl_pool_acquire(npool) != 0)
3859			continue;
3860
3861		emergency_pool = npool;
3862
3863		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3864		if (io != NULL) {
3865			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3866			npool->total_allocated++;
3867			npool->free_ctl_io--;
3868			mtx_unlock(&ctl_softc->pool_lock);
3869			return (io);
3870		} else
3871			ctl_pool_release(npool);
3872	}
3873
3874	/* Drop the spinlock before we malloc */
3875	mtx_unlock(&ctl_softc->pool_lock);
3876
3877	/*
3878	 * The emergency pool (if it exists) didn't have one, so try an
3879	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3880	 */
3881	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3882	if (io != NULL) {
3883		/*
3884		 * If the emergency pool exists but is empty, add this
3885		 * ctl_io to its list when it gets freed.
3886		 */
3887		if (emergency_pool != NULL) {
3888			mtx_lock(&ctl_softc->pool_lock);
3889			if (ctl_pool_acquire(emergency_pool) == 0) {
3890				io->io_hdr.pool = emergency_pool;
3891				emergency_pool->total_ctl_io++;
3892				/*
3893				 * Need to bump this, otherwise
3894				 * total_allocated and total_freed won't
3895				 * match when we no longer have anything
3896				 * outstanding.
3897				 */
3898				emergency_pool->total_allocated++;
3899			}
3900			mtx_unlock(&ctl_softc->pool_lock);
3901		} else
3902			io->io_hdr.pool = NULL;
3903	}
3904
3905	return (io);
3906}
3907
3908void
3909ctl_free_io(union ctl_io *io)
3910{
3911	if (io == NULL)
3912		return;
3913
3914	/*
3915	 * If this ctl_io has a pool, return it to that pool.
3916	 */
3917	if (io->io_hdr.pool != NULL) {
3918		struct ctl_io_pool *pool;
3919
3920		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3921		mtx_lock(&pool->ctl_softc->pool_lock);
3922		io->io_hdr.io_type = 0xff;
3923		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3924		pool->total_freed++;
3925		pool->free_ctl_io++;
3926		ctl_pool_release(pool);
3927		mtx_unlock(&pool->ctl_softc->pool_lock);
3928	} else {
3929		/*
3930		 * Otherwise, just free it.  We probably malloced it and
3931		 * the emergency pool wasn't available.
3932		 */
3933		free(io, M_CTLIO);
3934	}
3935
3936}
3937
3938void
3939ctl_zero_io(union ctl_io *io)
3940{
3941	void *pool_ref;
3942
3943	if (io == NULL)
3944		return;
3945
3946	/*
3947	 * May need to preserve linked list pointers at some point too.
3948	 */
3949	pool_ref = io->io_hdr.pool;
3950
3951	memset(io, 0, sizeof(*io));
3952
3953	io->io_hdr.pool = pool_ref;
3954}
3955
3956/*
3957 * This routine is currently used for internal copies of ctl_ios that need
3958 * to persist for some reason after we've already returned status to the
3959 * FETD.  (Thus the flag set.)
3960 *
3961 * XXX XXX
3962 * Note that this makes a blind copy of all fields in the ctl_io, except
3963 * for the pool reference.  This includes any memory that has been
3964 * allocated!  That memory will no longer be valid after done has been
3965 * called, so this would be VERY DANGEROUS for command that actually does
3966 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3967 * start and stop commands, which don't transfer any data, so this is not a
3968 * problem.  If it is used for anything else, the caller would also need to
3969 * allocate data buffer space and this routine would need to be modified to
3970 * copy the data buffer(s) as well.
3971 */
3972void
3973ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3974{
3975	void *pool_ref;
3976
3977	if ((src == NULL)
3978	 || (dest == NULL))
3979		return;
3980
3981	/*
3982	 * May need to preserve linked list pointers at some point too.
3983	 */
3984	pool_ref = dest->io_hdr.pool;
3985
3986	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3987
3988	dest->io_hdr.pool = pool_ref;
3989	/*
3990	 * We need to know that this is an internal copy, and doesn't need
3991	 * to get passed back to the FETD that allocated it.
3992	 */
3993	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3994}
3995
3996/*
3997 * This routine could be used in the future to load default and/or saved
3998 * mode page parameters for a particuar lun.
3999 */
4000static int
4001ctl_init_page_index(struct ctl_lun *lun)
4002{
4003	int i;
4004	struct ctl_page_index *page_index;
4005	const char *value;
4006
4007	memcpy(&lun->mode_pages.index, page_index_template,
4008	       sizeof(page_index_template));
4009
4010	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4011
4012		page_index = &lun->mode_pages.index[i];
4013		/*
4014		 * If this is a disk-only mode page, there's no point in
4015		 * setting it up.  For some pages, we have to have some
4016		 * basic information about the disk in order to calculate the
4017		 * mode page data.
4018		 */
4019		if ((lun->be_lun->lun_type != T_DIRECT)
4020		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4021			continue;
4022
4023		switch (page_index->page_code & SMPH_PC_MASK) {
4024		case SMS_RW_ERROR_RECOVERY_PAGE: {
4025			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4026				panic("subpage is incorrect!");
4027			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4028			       &rw_er_page_default,
4029			       sizeof(rw_er_page_default));
4030			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4031			       &rw_er_page_changeable,
4032			       sizeof(rw_er_page_changeable));
4033			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4034			       &rw_er_page_default,
4035			       sizeof(rw_er_page_default));
4036			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4037			       &rw_er_page_default,
4038			       sizeof(rw_er_page_default));
4039			page_index->page_data =
4040				(uint8_t *)lun->mode_pages.rw_er_page;
4041			break;
4042		}
4043		case SMS_FORMAT_DEVICE_PAGE: {
4044			struct scsi_format_page *format_page;
4045
4046			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4047				panic("subpage is incorrect!");
4048
4049			/*
4050			 * Sectors per track are set above.  Bytes per
4051			 * sector need to be set here on a per-LUN basis.
4052			 */
4053			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4054			       &format_page_default,
4055			       sizeof(format_page_default));
4056			memcpy(&lun->mode_pages.format_page[
4057			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4058			       sizeof(format_page_changeable));
4059			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4060			       &format_page_default,
4061			       sizeof(format_page_default));
4062			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4063			       &format_page_default,
4064			       sizeof(format_page_default));
4065
4066			format_page = &lun->mode_pages.format_page[
4067				CTL_PAGE_CURRENT];
4068			scsi_ulto2b(lun->be_lun->blocksize,
4069				    format_page->bytes_per_sector);
4070
4071			format_page = &lun->mode_pages.format_page[
4072				CTL_PAGE_DEFAULT];
4073			scsi_ulto2b(lun->be_lun->blocksize,
4074				    format_page->bytes_per_sector);
4075
4076			format_page = &lun->mode_pages.format_page[
4077				CTL_PAGE_SAVED];
4078			scsi_ulto2b(lun->be_lun->blocksize,
4079				    format_page->bytes_per_sector);
4080
4081			page_index->page_data =
4082				(uint8_t *)lun->mode_pages.format_page;
4083			break;
4084		}
4085		case SMS_RIGID_DISK_PAGE: {
4086			struct scsi_rigid_disk_page *rigid_disk_page;
4087			uint32_t sectors_per_cylinder;
4088			uint64_t cylinders;
4089#ifndef	__XSCALE__
4090			int shift;
4091#endif /* !__XSCALE__ */
4092
4093			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4094				panic("invalid subpage value %d",
4095				      page_index->subpage);
4096
4097			/*
4098			 * Rotation rate and sectors per track are set
4099			 * above.  We calculate the cylinders here based on
4100			 * capacity.  Due to the number of heads and
4101			 * sectors per track we're using, smaller arrays
4102			 * may turn out to have 0 cylinders.  Linux and
4103			 * FreeBSD don't pay attention to these mode pages
4104			 * to figure out capacity, but Solaris does.  It
4105			 * seems to deal with 0 cylinders just fine, and
4106			 * works out a fake geometry based on the capacity.
4107			 */
4108			memcpy(&lun->mode_pages.rigid_disk_page[
4109			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4110			       sizeof(rigid_disk_page_default));
4111			memcpy(&lun->mode_pages.rigid_disk_page[
4112			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4113			       sizeof(rigid_disk_page_changeable));
4114			memcpy(&lun->mode_pages.rigid_disk_page[
4115			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4116			       sizeof(rigid_disk_page_default));
4117			memcpy(&lun->mode_pages.rigid_disk_page[
4118			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4119			       sizeof(rigid_disk_page_default));
4120
4121			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4122				CTL_DEFAULT_HEADS;
4123
4124			/*
4125			 * The divide method here will be more accurate,
4126			 * probably, but results in floating point being
4127			 * used in the kernel on i386 (__udivdi3()).  On the
4128			 * XScale, though, __udivdi3() is implemented in
4129			 * software.
4130			 *
4131			 * The shift method for cylinder calculation is
4132			 * accurate if sectors_per_cylinder is a power of
4133			 * 2.  Otherwise it might be slightly off -- you
4134			 * might have a bit of a truncation problem.
4135			 */
4136#ifdef	__XSCALE__
4137			cylinders = (lun->be_lun->maxlba + 1) /
4138				sectors_per_cylinder;
4139#else
4140			for (shift = 31; shift > 0; shift--) {
4141				if (sectors_per_cylinder & (1 << shift))
4142					break;
4143			}
4144			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4145#endif
4146
4147			/*
4148			 * We've basically got 3 bytes, or 24 bits for the
4149			 * cylinder size in the mode page.  If we're over,
4150			 * just round down to 2^24.
4151			 */
4152			if (cylinders > 0xffffff)
4153				cylinders = 0xffffff;
4154
4155			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4156				CTL_PAGE_CURRENT];
4157			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4158
4159			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4160				CTL_PAGE_DEFAULT];
4161			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4162
4163			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4164				CTL_PAGE_SAVED];
4165			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4166
4167			page_index->page_data =
4168				(uint8_t *)lun->mode_pages.rigid_disk_page;
4169			break;
4170		}
4171		case SMS_CACHING_PAGE: {
4172			struct scsi_caching_page *caching_page;
4173
4174			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4175				panic("invalid subpage value %d",
4176				      page_index->subpage);
4177			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4178			       &caching_page_default,
4179			       sizeof(caching_page_default));
4180			memcpy(&lun->mode_pages.caching_page[
4181			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4182			       sizeof(caching_page_changeable));
4183			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4184			       &caching_page_default,
4185			       sizeof(caching_page_default));
4186			caching_page = &lun->mode_pages.caching_page[
4187			    CTL_PAGE_SAVED];
4188			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4189			if (value != NULL && strcmp(value, "off") == 0)
4190				caching_page->flags1 &= ~SCP_WCE;
4191			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4192			if (value != NULL && strcmp(value, "off") == 0)
4193				caching_page->flags1 |= SCP_RCD;
4194			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4195			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4196			       sizeof(caching_page_default));
4197			page_index->page_data =
4198				(uint8_t *)lun->mode_pages.caching_page;
4199			break;
4200		}
4201		case SMS_CONTROL_MODE_PAGE: {
4202			struct scsi_control_page *control_page;
4203
4204			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4205				panic("invalid subpage value %d",
4206				      page_index->subpage);
4207
4208			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4209			       &control_page_default,
4210			       sizeof(control_page_default));
4211			memcpy(&lun->mode_pages.control_page[
4212			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4213			       sizeof(control_page_changeable));
4214			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4215			       &control_page_default,
4216			       sizeof(control_page_default));
4217			control_page = &lun->mode_pages.control_page[
4218			    CTL_PAGE_SAVED];
4219			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4220			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4221				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4222				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4223			}
4224			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4225			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4226			       sizeof(control_page_default));
4227			page_index->page_data =
4228				(uint8_t *)lun->mode_pages.control_page;
4229			break;
4230
4231		}
4232		case SMS_INFO_EXCEPTIONS_PAGE: {
4233			switch (page_index->subpage) {
4234			case SMS_SUBPAGE_PAGE_0:
4235				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4236				       &ie_page_default,
4237				       sizeof(ie_page_default));
4238				memcpy(&lun->mode_pages.ie_page[
4239				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4240				       sizeof(ie_page_changeable));
4241				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4242				       &ie_page_default,
4243				       sizeof(ie_page_default));
4244				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4245				       &ie_page_default,
4246				       sizeof(ie_page_default));
4247				page_index->page_data =
4248					(uint8_t *)lun->mode_pages.ie_page;
4249				break;
4250			case 0x02:
4251				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4252				       &lbp_page_default,
4253				       sizeof(lbp_page_default));
4254				memcpy(&lun->mode_pages.lbp_page[
4255				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4256				       sizeof(lbp_page_changeable));
4257				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4258				       &lbp_page_default,
4259				       sizeof(lbp_page_default));
4260				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4261				       &lbp_page_default,
4262				       sizeof(lbp_page_default));
4263				page_index->page_data =
4264					(uint8_t *)lun->mode_pages.lbp_page;
4265			}
4266			break;
4267		}
4268		case SMS_VENDOR_SPECIFIC_PAGE:{
4269			switch (page_index->subpage) {
4270			case DBGCNF_SUBPAGE_CODE: {
4271				struct copan_debugconf_subpage *current_page,
4272							       *saved_page;
4273
4274				memcpy(&lun->mode_pages.debugconf_subpage[
4275				       CTL_PAGE_CURRENT],
4276				       &debugconf_page_default,
4277				       sizeof(debugconf_page_default));
4278				memcpy(&lun->mode_pages.debugconf_subpage[
4279				       CTL_PAGE_CHANGEABLE],
4280				       &debugconf_page_changeable,
4281				       sizeof(debugconf_page_changeable));
4282				memcpy(&lun->mode_pages.debugconf_subpage[
4283				       CTL_PAGE_DEFAULT],
4284				       &debugconf_page_default,
4285				       sizeof(debugconf_page_default));
4286				memcpy(&lun->mode_pages.debugconf_subpage[
4287				       CTL_PAGE_SAVED],
4288				       &debugconf_page_default,
4289				       sizeof(debugconf_page_default));
4290				page_index->page_data =
4291					(uint8_t *)lun->mode_pages.debugconf_subpage;
4292
4293				current_page = (struct copan_debugconf_subpage *)
4294					(page_index->page_data +
4295					 (page_index->page_len *
4296					  CTL_PAGE_CURRENT));
4297				saved_page = (struct copan_debugconf_subpage *)
4298					(page_index->page_data +
4299					 (page_index->page_len *
4300					  CTL_PAGE_SAVED));
4301				break;
4302			}
4303			default:
4304				panic("invalid subpage value %d",
4305				      page_index->subpage);
4306				break;
4307			}
4308   			break;
4309		}
4310		default:
4311			panic("invalid page value %d",
4312			      page_index->page_code & SMPH_PC_MASK);
4313			break;
4314    	}
4315	}
4316
4317	return (CTL_RETVAL_COMPLETE);
4318}
4319
4320static int
4321ctl_init_log_page_index(struct ctl_lun *lun)
4322{
4323	struct ctl_page_index *page_index;
4324	int i, j, prev;
4325
4326	memcpy(&lun->log_pages.index, log_page_index_template,
4327	       sizeof(log_page_index_template));
4328
4329	prev = -1;
4330	for (i = 0, j = 0; i < CTL_NUM_LOG_PAGES; i++) {
4331
4332		page_index = &lun->log_pages.index[i];
4333		/*
4334		 * If this is a disk-only mode page, there's no point in
4335		 * setting it up.  For some pages, we have to have some
4336		 * basic information about the disk in order to calculate the
4337		 * mode page data.
4338		 */
4339		if ((lun->be_lun->lun_type != T_DIRECT)
4340		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4341			continue;
4342
4343		if (page_index->page_code != prev) {
4344			lun->log_pages.pages_page[j] = page_index->page_code;
4345			prev = page_index->page_code;
4346			j++;
4347		}
4348		lun->log_pages.subpages_page[i*2] = page_index->page_code;
4349		lun->log_pages.subpages_page[i*2+1] = page_index->subpage;
4350	}
4351	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4352	lun->log_pages.index[0].page_len = j;
4353	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4354	lun->log_pages.index[1].page_len = i * 2;
4355
4356	return (CTL_RETVAL_COMPLETE);
4357}
4358
4359static int
4360hex2bin(const char *str, uint8_t *buf, int buf_size)
4361{
4362	int i;
4363	u_char c;
4364
4365	memset(buf, 0, buf_size);
4366	while (isspace(str[0]))
4367		str++;
4368	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4369		str += 2;
4370	buf_size *= 2;
4371	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4372		c = str[i];
4373		if (isdigit(c))
4374			c -= '0';
4375		else if (isalpha(c))
4376			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4377		else
4378			break;
4379		if (c >= 16)
4380			break;
4381		if ((i & 1) == 0)
4382			buf[i / 2] |= (c << 4);
4383		else
4384			buf[i / 2] |= c;
4385	}
4386	return ((i + 1) / 2);
4387}
4388
4389/*
4390 * LUN allocation.
4391 *
4392 * Requirements:
4393 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4394 *   wants us to allocate the LUN and he can block.
4395 * - ctl_softc is always set
4396 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4397 *
4398 * Returns 0 for success, non-zero (errno) for failure.
4399 */
4400static int
4401ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4402	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4403{
4404	struct ctl_lun *nlun, *lun;
4405	struct ctl_port *port;
4406	struct scsi_vpd_id_descriptor *desc;
4407	struct scsi_vpd_id_t10 *t10id;
4408	const char *eui, *naa, *scsiname, *vendor, *value;
4409	int lun_number, i, lun_malloced;
4410	int devidlen, idlen1, idlen2 = 0, len;
4411
4412	if (be_lun == NULL)
4413		return (EINVAL);
4414
4415	/*
4416	 * We currently only support Direct Access or Processor LUN types.
4417	 */
4418	switch (be_lun->lun_type) {
4419	case T_DIRECT:
4420		break;
4421	case T_PROCESSOR:
4422		break;
4423	case T_SEQUENTIAL:
4424	case T_CHANGER:
4425	default:
4426		be_lun->lun_config_status(be_lun->be_lun,
4427					  CTL_LUN_CONFIG_FAILURE);
4428		break;
4429	}
4430	if (ctl_lun == NULL) {
4431		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4432		lun_malloced = 1;
4433	} else {
4434		lun_malloced = 0;
4435		lun = ctl_lun;
4436	}
4437
4438	memset(lun, 0, sizeof(*lun));
4439	if (lun_malloced)
4440		lun->flags = CTL_LUN_MALLOCED;
4441
4442	/* Generate LUN ID. */
4443	devidlen = max(CTL_DEVID_MIN_LEN,
4444	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4445	idlen1 = sizeof(*t10id) + devidlen;
4446	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4447	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4448	if (scsiname != NULL) {
4449		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4450		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4451	}
4452	eui = ctl_get_opt(&be_lun->options, "eui");
4453	if (eui != NULL) {
4454		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4455	}
4456	naa = ctl_get_opt(&be_lun->options, "naa");
4457	if (naa != NULL) {
4458		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4459	}
4460	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4461	    M_CTL, M_WAITOK | M_ZERO);
4462	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4463	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4464	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4465	desc->length = idlen1;
4466	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4467	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4468	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4469		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4470	} else {
4471		strncpy(t10id->vendor, vendor,
4472		    min(sizeof(t10id->vendor), strlen(vendor)));
4473	}
4474	strncpy((char *)t10id->vendor_spec_id,
4475	    (char *)be_lun->device_id, devidlen);
4476	if (scsiname != NULL) {
4477		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4478		    desc->length);
4479		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4480		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4481		    SVPD_ID_TYPE_SCSI_NAME;
4482		desc->length = idlen2;
4483		strlcpy(desc->identifier, scsiname, idlen2);
4484	}
4485	if (eui != NULL) {
4486		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4487		    desc->length);
4488		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4489		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4490		    SVPD_ID_TYPE_EUI64;
4491		desc->length = hex2bin(eui, desc->identifier, 16);
4492		desc->length = desc->length > 12 ? 16 :
4493		    (desc->length > 8 ? 12 : 8);
4494		len -= 16 - desc->length;
4495	}
4496	if (naa != NULL) {
4497		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4498		    desc->length);
4499		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4500		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4501		    SVPD_ID_TYPE_NAA;
4502		desc->length = hex2bin(naa, desc->identifier, 16);
4503		desc->length = desc->length > 8 ? 16 : 8;
4504		len -= 16 - desc->length;
4505	}
4506	lun->lun_devid->len = len;
4507
4508	mtx_lock(&ctl_softc->ctl_lock);
4509	/*
4510	 * See if the caller requested a particular LUN number.  If so, see
4511	 * if it is available.  Otherwise, allocate the first available LUN.
4512	 */
4513	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4514		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4515		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4516			mtx_unlock(&ctl_softc->ctl_lock);
4517			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4518				printf("ctl: requested LUN ID %d is higher "
4519				       "than CTL_MAX_LUNS - 1 (%d)\n",
4520				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4521			} else {
4522				/*
4523				 * XXX KDM return an error, or just assign
4524				 * another LUN ID in this case??
4525				 */
4526				printf("ctl: requested LUN ID %d is already "
4527				       "in use\n", be_lun->req_lun_id);
4528			}
4529			if (lun->flags & CTL_LUN_MALLOCED)
4530				free(lun, M_CTL);
4531			be_lun->lun_config_status(be_lun->be_lun,
4532						  CTL_LUN_CONFIG_FAILURE);
4533			return (ENOSPC);
4534		}
4535		lun_number = be_lun->req_lun_id;
4536	} else {
4537		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4538		if (lun_number == -1) {
4539			mtx_unlock(&ctl_softc->ctl_lock);
4540			printf("ctl: can't allocate LUN on target %ju, out of "
4541			       "LUNs\n", (uintmax_t)target_id.id);
4542			if (lun->flags & CTL_LUN_MALLOCED)
4543				free(lun, M_CTL);
4544			be_lun->lun_config_status(be_lun->be_lun,
4545						  CTL_LUN_CONFIG_FAILURE);
4546			return (ENOSPC);
4547		}
4548	}
4549	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4550
4551	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4552	lun->target = target_id;
4553	lun->lun = lun_number;
4554	lun->be_lun = be_lun;
4555	/*
4556	 * The processor LUN is always enabled.  Disk LUNs come on line
4557	 * disabled, and must be enabled by the backend.
4558	 */
4559	lun->flags |= CTL_LUN_DISABLED;
4560	lun->backend = be_lun->be;
4561	be_lun->ctl_lun = lun;
4562	be_lun->lun_id = lun_number;
4563	atomic_add_int(&be_lun->be->num_luns, 1);
4564	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4565		lun->flags |= CTL_LUN_OFFLINE;
4566
4567	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4568		lun->flags |= CTL_LUN_STOPPED;
4569
4570	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4571		lun->flags |= CTL_LUN_INOPERABLE;
4572
4573	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4574		lun->flags |= CTL_LUN_PRIMARY_SC;
4575
4576	value = ctl_get_opt(&be_lun->options, "readonly");
4577	if (value != NULL && strcmp(value, "on") == 0)
4578		lun->flags |= CTL_LUN_READONLY;
4579
4580	lun->ctl_softc = ctl_softc;
4581	TAILQ_INIT(&lun->ooa_queue);
4582	TAILQ_INIT(&lun->blocked_queue);
4583	STAILQ_INIT(&lun->error_list);
4584	ctl_tpc_lun_init(lun);
4585
4586	/*
4587	 * Initialize the mode and log page index.
4588	 */
4589	ctl_init_page_index(lun);
4590	ctl_init_log_page_index(lun);
4591
4592	/*
4593	 * Set the poweron UA for all initiators on this LUN only.
4594	 */
4595	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4596		lun->pending_ua[i] = CTL_UA_POWERON;
4597
4598	/*
4599	 * Now, before we insert this lun on the lun list, set the lun
4600	 * inventory changed UA for all other luns.
4601	 */
4602	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4603		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4604			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4605		}
4606	}
4607
4608	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4609
4610	ctl_softc->ctl_luns[lun_number] = lun;
4611
4612	ctl_softc->num_luns++;
4613
4614	/* Setup statistics gathering */
4615	lun->stats.device_type = be_lun->lun_type;
4616	lun->stats.lun_number = lun_number;
4617	if (lun->stats.device_type == T_DIRECT)
4618		lun->stats.blocksize = be_lun->blocksize;
4619	else
4620		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4621	for (i = 0;i < CTL_MAX_PORTS;i++)
4622		lun->stats.ports[i].targ_port = i;
4623
4624	mtx_unlock(&ctl_softc->ctl_lock);
4625
4626	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4627
4628	/*
4629	 * Run through each registered FETD and bring it online if it isn't
4630	 * already.  Enable the target ID if it hasn't been enabled, and
4631	 * enable this particular LUN.
4632	 */
4633	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4634		int retval;
4635
4636		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4637		if (retval != 0) {
4638			printf("ctl_alloc_lun: FETD %s port %d returned error "
4639			       "%d for lun_enable on target %ju lun %d\n",
4640			       port->port_name, port->targ_port, retval,
4641			       (uintmax_t)target_id.id, lun_number);
4642		} else
4643			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4644	}
4645	return (0);
4646}
4647
4648/*
4649 * Delete a LUN.
4650 * Assumptions:
4651 * - LUN has already been marked invalid and any pending I/O has been taken
4652 *   care of.
4653 */
4654static int
4655ctl_free_lun(struct ctl_lun *lun)
4656{
4657	struct ctl_softc *softc;
4658#if 0
4659	struct ctl_port *port;
4660#endif
4661	struct ctl_lun *nlun;
4662	int i;
4663
4664	softc = lun->ctl_softc;
4665
4666	mtx_assert(&softc->ctl_lock, MA_OWNED);
4667
4668	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4669
4670	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4671
4672	softc->ctl_luns[lun->lun] = NULL;
4673
4674	if (!TAILQ_EMPTY(&lun->ooa_queue))
4675		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4676
4677	softc->num_luns--;
4678
4679	/*
4680	 * XXX KDM this scheme only works for a single target/multiple LUN
4681	 * setup.  It needs to be revamped for a multiple target scheme.
4682	 *
4683	 * XXX KDM this results in port->lun_disable() getting called twice,
4684	 * once when ctl_disable_lun() is called, and a second time here.
4685	 * We really need to re-think the LUN disable semantics.  There
4686	 * should probably be several steps/levels to LUN removal:
4687	 *  - disable
4688	 *  - invalidate
4689	 *  - free
4690 	 *
4691	 * Right now we only have a disable method when communicating to
4692	 * the front end ports, at least for individual LUNs.
4693	 */
4694#if 0
4695	STAILQ_FOREACH(port, &softc->port_list, links) {
4696		int retval;
4697
4698		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4699					 lun->lun);
4700		if (retval != 0) {
4701			printf("ctl_free_lun: FETD %s port %d returned error "
4702			       "%d for lun_disable on target %ju lun %jd\n",
4703			       port->port_name, port->targ_port, retval,
4704			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4705		}
4706
4707		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4708			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4709
4710			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4711			if (retval != 0) {
4712				printf("ctl_free_lun: FETD %s port %d "
4713				       "returned error %d for targ_disable on "
4714				       "target %ju\n", port->port_name,
4715				       port->targ_port, retval,
4716				       (uintmax_t)lun->target.id);
4717			} else
4718				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4719
4720			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4721				continue;
4722
4723#if 0
4724			port->port_offline(port->onoff_arg);
4725			port->status &= ~CTL_PORT_STATUS_ONLINE;
4726#endif
4727		}
4728	}
4729#endif
4730
4731	/*
4732	 * Tell the backend to free resources, if this LUN has a backend.
4733	 */
4734	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4735	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4736
4737	ctl_tpc_lun_shutdown(lun);
4738	mtx_destroy(&lun->lun_lock);
4739	free(lun->lun_devid, M_CTL);
4740	if (lun->flags & CTL_LUN_MALLOCED)
4741		free(lun, M_CTL);
4742
4743	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4744		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4745			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4746		}
4747	}
4748
4749	return (0);
4750}
4751
4752static void
4753ctl_create_lun(struct ctl_be_lun *be_lun)
4754{
4755	struct ctl_softc *ctl_softc;
4756
4757	ctl_softc = control_softc;
4758
4759	/*
4760	 * ctl_alloc_lun() should handle all potential failure cases.
4761	 */
4762	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4763}
4764
4765int
4766ctl_add_lun(struct ctl_be_lun *be_lun)
4767{
4768	struct ctl_softc *ctl_softc = control_softc;
4769
4770	mtx_lock(&ctl_softc->ctl_lock);
4771	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4772	mtx_unlock(&ctl_softc->ctl_lock);
4773	wakeup(&ctl_softc->pending_lun_queue);
4774
4775	return (0);
4776}
4777
4778int
4779ctl_enable_lun(struct ctl_be_lun *be_lun)
4780{
4781	struct ctl_softc *ctl_softc;
4782	struct ctl_port *port, *nport;
4783	struct ctl_lun *lun;
4784	int retval;
4785
4786	ctl_softc = control_softc;
4787
4788	lun = (struct ctl_lun *)be_lun->ctl_lun;
4789
4790	mtx_lock(&ctl_softc->ctl_lock);
4791	mtx_lock(&lun->lun_lock);
4792	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4793		/*
4794		 * eh?  Why did we get called if the LUN is already
4795		 * enabled?
4796		 */
4797		mtx_unlock(&lun->lun_lock);
4798		mtx_unlock(&ctl_softc->ctl_lock);
4799		return (0);
4800	}
4801	lun->flags &= ~CTL_LUN_DISABLED;
4802	mtx_unlock(&lun->lun_lock);
4803
4804	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4805		nport = STAILQ_NEXT(port, links);
4806
4807		/*
4808		 * Drop the lock while we call the FETD's enable routine.
4809		 * This can lead to a callback into CTL (at least in the
4810		 * case of the internal initiator frontend.
4811		 */
4812		mtx_unlock(&ctl_softc->ctl_lock);
4813		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4814		mtx_lock(&ctl_softc->ctl_lock);
4815		if (retval != 0) {
4816			printf("%s: FETD %s port %d returned error "
4817			       "%d for lun_enable on target %ju lun %jd\n",
4818			       __func__, port->port_name, port->targ_port, retval,
4819			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4820		}
4821#if 0
4822		 else {
4823            /* NOTE:  TODO:  why does lun enable affect port status? */
4824			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4825		}
4826#endif
4827	}
4828
4829	mtx_unlock(&ctl_softc->ctl_lock);
4830
4831	return (0);
4832}
4833
4834int
4835ctl_disable_lun(struct ctl_be_lun *be_lun)
4836{
4837	struct ctl_softc *ctl_softc;
4838	struct ctl_port *port;
4839	struct ctl_lun *lun;
4840	int retval;
4841
4842	ctl_softc = control_softc;
4843
4844	lun = (struct ctl_lun *)be_lun->ctl_lun;
4845
4846	mtx_lock(&ctl_softc->ctl_lock);
4847	mtx_lock(&lun->lun_lock);
4848	if (lun->flags & CTL_LUN_DISABLED) {
4849		mtx_unlock(&lun->lun_lock);
4850		mtx_unlock(&ctl_softc->ctl_lock);
4851		return (0);
4852	}
4853	lun->flags |= CTL_LUN_DISABLED;
4854	mtx_unlock(&lun->lun_lock);
4855
4856	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4857		mtx_unlock(&ctl_softc->ctl_lock);
4858		/*
4859		 * Drop the lock before we call the frontend's disable
4860		 * routine, to avoid lock order reversals.
4861		 *
4862		 * XXX KDM what happens if the frontend list changes while
4863		 * we're traversing it?  It's unlikely, but should be handled.
4864		 */
4865		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4866					 lun->lun);
4867		mtx_lock(&ctl_softc->ctl_lock);
4868		if (retval != 0) {
4869			printf("ctl_alloc_lun: FETD %s port %d returned error "
4870			       "%d for lun_disable on target %ju lun %jd\n",
4871			       port->port_name, port->targ_port, retval,
4872			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4873		}
4874	}
4875
4876	mtx_unlock(&ctl_softc->ctl_lock);
4877
4878	return (0);
4879}
4880
4881int
4882ctl_start_lun(struct ctl_be_lun *be_lun)
4883{
4884	struct ctl_softc *ctl_softc;
4885	struct ctl_lun *lun;
4886
4887	ctl_softc = control_softc;
4888
4889	lun = (struct ctl_lun *)be_lun->ctl_lun;
4890
4891	mtx_lock(&lun->lun_lock);
4892	lun->flags &= ~CTL_LUN_STOPPED;
4893	mtx_unlock(&lun->lun_lock);
4894
4895	return (0);
4896}
4897
4898int
4899ctl_stop_lun(struct ctl_be_lun *be_lun)
4900{
4901	struct ctl_softc *ctl_softc;
4902	struct ctl_lun *lun;
4903
4904	ctl_softc = control_softc;
4905
4906	lun = (struct ctl_lun *)be_lun->ctl_lun;
4907
4908	mtx_lock(&lun->lun_lock);
4909	lun->flags |= CTL_LUN_STOPPED;
4910	mtx_unlock(&lun->lun_lock);
4911
4912	return (0);
4913}
4914
4915int
4916ctl_lun_offline(struct ctl_be_lun *be_lun)
4917{
4918	struct ctl_softc *ctl_softc;
4919	struct ctl_lun *lun;
4920
4921	ctl_softc = control_softc;
4922
4923	lun = (struct ctl_lun *)be_lun->ctl_lun;
4924
4925	mtx_lock(&lun->lun_lock);
4926	lun->flags |= CTL_LUN_OFFLINE;
4927	mtx_unlock(&lun->lun_lock);
4928
4929	return (0);
4930}
4931
4932int
4933ctl_lun_online(struct ctl_be_lun *be_lun)
4934{
4935	struct ctl_softc *ctl_softc;
4936	struct ctl_lun *lun;
4937
4938	ctl_softc = control_softc;
4939
4940	lun = (struct ctl_lun *)be_lun->ctl_lun;
4941
4942	mtx_lock(&lun->lun_lock);
4943	lun->flags &= ~CTL_LUN_OFFLINE;
4944	mtx_unlock(&lun->lun_lock);
4945
4946	return (0);
4947}
4948
4949int
4950ctl_invalidate_lun(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
4957	lun = (struct ctl_lun *)be_lun->ctl_lun;
4958
4959	mtx_lock(&lun->lun_lock);
4960
4961	/*
4962	 * The LUN needs to be disabled before it can be marked invalid.
4963	 */
4964	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4965		mtx_unlock(&lun->lun_lock);
4966		return (-1);
4967	}
4968	/*
4969	 * Mark the LUN invalid.
4970	 */
4971	lun->flags |= CTL_LUN_INVALID;
4972
4973	/*
4974	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4975	 * If we have something in the OOA queue, we'll free it when the
4976	 * last I/O completes.
4977	 */
4978	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4979		mtx_unlock(&lun->lun_lock);
4980		mtx_lock(&ctl_softc->ctl_lock);
4981		ctl_free_lun(lun);
4982		mtx_unlock(&ctl_softc->ctl_lock);
4983	} else
4984		mtx_unlock(&lun->lun_lock);
4985
4986	return (0);
4987}
4988
4989int
4990ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4991{
4992	struct ctl_softc *ctl_softc;
4993	struct ctl_lun *lun;
4994
4995	ctl_softc = control_softc;
4996	lun = (struct ctl_lun *)be_lun->ctl_lun;
4997
4998	mtx_lock(&lun->lun_lock);
4999	lun->flags |= CTL_LUN_INOPERABLE;
5000	mtx_unlock(&lun->lun_lock);
5001
5002	return (0);
5003}
5004
5005int
5006ctl_lun_operable(struct ctl_be_lun *be_lun)
5007{
5008	struct ctl_softc *ctl_softc;
5009	struct ctl_lun *lun;
5010
5011	ctl_softc = control_softc;
5012	lun = (struct ctl_lun *)be_lun->ctl_lun;
5013
5014	mtx_lock(&lun->lun_lock);
5015	lun->flags &= ~CTL_LUN_INOPERABLE;
5016	mtx_unlock(&lun->lun_lock);
5017
5018	return (0);
5019}
5020
5021void
5022ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5023{
5024	struct ctl_lun *lun;
5025	struct ctl_softc *softc;
5026	int i;
5027
5028	softc = control_softc;
5029
5030	lun = (struct ctl_lun *)be_lun->ctl_lun;
5031
5032	mtx_lock(&lun->lun_lock);
5033
5034	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5035		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5036
5037	mtx_unlock(&lun->lun_lock);
5038}
5039
5040/*
5041 * Backend "memory move is complete" callback for requests that never
5042 * make it down to say RAIDCore's configuration code.
5043 */
5044int
5045ctl_config_move_done(union ctl_io *io)
5046{
5047	int retval;
5048
5049	retval = CTL_RETVAL_COMPLETE;
5050
5051
5052	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5053	/*
5054	 * XXX KDM this shouldn't happen, but what if it does?
5055	 */
5056	if (io->io_hdr.io_type != CTL_IO_SCSI)
5057		panic("I/O type isn't CTL_IO_SCSI!");
5058
5059	if ((io->io_hdr.port_status == 0)
5060	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5061	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5062		io->io_hdr.status = CTL_SUCCESS;
5063	else if ((io->io_hdr.port_status != 0)
5064	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5065	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5066		/*
5067		 * For hardware error sense keys, the sense key
5068		 * specific value is defined to be a retry count,
5069		 * but we use it to pass back an internal FETD
5070		 * error code.  XXX KDM  Hopefully the FETD is only
5071		 * using 16 bits for an error code, since that's
5072		 * all the space we have in the sks field.
5073		 */
5074		ctl_set_internal_failure(&io->scsiio,
5075					 /*sks_valid*/ 1,
5076					 /*retry_count*/
5077					 io->io_hdr.port_status);
5078		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5079			free(io->scsiio.kern_data_ptr, M_CTL);
5080		ctl_done(io);
5081		goto bailout;
5082	}
5083
5084	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5085	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5086	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5087		/*
5088		 * XXX KDM just assuming a single pointer here, and not a
5089		 * S/G list.  If we start using S/G lists for config data,
5090		 * we'll need to know how to clean them up here as well.
5091		 */
5092		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5093			free(io->scsiio.kern_data_ptr, M_CTL);
5094		/* Hopefully the user has already set the status... */
5095		ctl_done(io);
5096	} else {
5097		/*
5098		 * XXX KDM now we need to continue data movement.  Some
5099		 * options:
5100		 * - call ctl_scsiio() again?  We don't do this for data
5101		 *   writes, because for those at least we know ahead of
5102		 *   time where the write will go and how long it is.  For
5103		 *   config writes, though, that information is largely
5104		 *   contained within the write itself, thus we need to
5105		 *   parse out the data again.
5106		 *
5107		 * - Call some other function once the data is in?
5108		 */
5109		if (ctl_debug & CTL_DEBUG_CDB_DATA)
5110			ctl_data_print(io);
5111
5112		/*
5113		 * XXX KDM call ctl_scsiio() again for now, and check flag
5114		 * bits to see whether we're allocated or not.
5115		 */
5116		retval = ctl_scsiio(&io->scsiio);
5117	}
5118bailout:
5119	return (retval);
5120}
5121
5122/*
5123 * This gets called by a backend driver when it is done with a
5124 * data_submit method.
5125 */
5126void
5127ctl_data_submit_done(union ctl_io *io)
5128{
5129	/*
5130	 * If the IO_CONT flag is set, we need to call the supplied
5131	 * function to continue processing the I/O, instead of completing
5132	 * the I/O just yet.
5133	 *
5134	 * If there is an error, though, we don't want to keep processing.
5135	 * Instead, just send status back to the initiator.
5136	 */
5137	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5138	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5139	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5140	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5141		io->scsiio.io_cont(io);
5142		return;
5143	}
5144	ctl_done(io);
5145}
5146
5147/*
5148 * This gets called by a backend driver when it is done with a
5149 * configuration write.
5150 */
5151void
5152ctl_config_write_done(union ctl_io *io)
5153{
5154	uint8_t *buf;
5155
5156	/*
5157	 * If the IO_CONT flag is set, we need to call the supplied
5158	 * function to continue processing the I/O, instead of completing
5159	 * the I/O just yet.
5160	 *
5161	 * If there is an error, though, we don't want to keep processing.
5162	 * Instead, just send status back to the initiator.
5163	 */
5164	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5165	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5166	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5167	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5168		io->scsiio.io_cont(io);
5169		return;
5170	}
5171	/*
5172	 * Since a configuration write can be done for commands that actually
5173	 * have data allocated, like write buffer, and commands that have
5174	 * no data, like start/stop unit, we need to check here.
5175	 */
5176	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5177		buf = io->scsiio.kern_data_ptr;
5178	else
5179		buf = NULL;
5180	ctl_done(io);
5181	if (buf)
5182		free(buf, M_CTL);
5183}
5184
5185/*
5186 * SCSI release command.
5187 */
5188int
5189ctl_scsi_release(struct ctl_scsiio *ctsio)
5190{
5191	int length, longid, thirdparty_id, resv_id;
5192	struct ctl_softc *ctl_softc;
5193	struct ctl_lun *lun;
5194	uint32_t residx;
5195
5196	length = 0;
5197	resv_id = 0;
5198
5199	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5200
5201	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5202	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5203	ctl_softc = control_softc;
5204
5205	switch (ctsio->cdb[0]) {
5206	case RELEASE_10: {
5207		struct scsi_release_10 *cdb;
5208
5209		cdb = (struct scsi_release_10 *)ctsio->cdb;
5210
5211		if (cdb->byte2 & SR10_LONGID)
5212			longid = 1;
5213		else
5214			thirdparty_id = cdb->thirdparty_id;
5215
5216		resv_id = cdb->resv_id;
5217		length = scsi_2btoul(cdb->length);
5218		break;
5219	}
5220	}
5221
5222
5223	/*
5224	 * XXX KDM right now, we only support LUN reservation.  We don't
5225	 * support 3rd party reservations, or extent reservations, which
5226	 * might actually need the parameter list.  If we've gotten this
5227	 * far, we've got a LUN reservation.  Anything else got kicked out
5228	 * above.  So, according to SPC, ignore the length.
5229	 */
5230	length = 0;
5231
5232	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5233	 && (length > 0)) {
5234		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5235		ctsio->kern_data_len = length;
5236		ctsio->kern_total_len = length;
5237		ctsio->kern_data_resid = 0;
5238		ctsio->kern_rel_offset = 0;
5239		ctsio->kern_sg_entries = 0;
5240		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5241		ctsio->be_move_done = ctl_config_move_done;
5242		ctl_datamove((union ctl_io *)ctsio);
5243
5244		return (CTL_RETVAL_COMPLETE);
5245	}
5246
5247	if (length > 0)
5248		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5249
5250	mtx_lock(&lun->lun_lock);
5251
5252	/*
5253	 * According to SPC, it is not an error for an intiator to attempt
5254	 * to release a reservation on a LUN that isn't reserved, or that
5255	 * is reserved by another initiator.  The reservation can only be
5256	 * released, though, by the initiator who made it or by one of
5257	 * several reset type events.
5258	 */
5259	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5260			lun->flags &= ~CTL_LUN_RESERVED;
5261
5262	mtx_unlock(&lun->lun_lock);
5263
5264	ctsio->scsi_status = SCSI_STATUS_OK;
5265	ctsio->io_hdr.status = CTL_SUCCESS;
5266
5267	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5268		free(ctsio->kern_data_ptr, M_CTL);
5269		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5270	}
5271
5272	ctl_done((union ctl_io *)ctsio);
5273	return (CTL_RETVAL_COMPLETE);
5274}
5275
5276int
5277ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5278{
5279	int extent, thirdparty, longid;
5280	int resv_id, length;
5281	uint64_t thirdparty_id;
5282	struct ctl_softc *ctl_softc;
5283	struct ctl_lun *lun;
5284	uint32_t residx;
5285
5286	extent = 0;
5287	thirdparty = 0;
5288	longid = 0;
5289	resv_id = 0;
5290	length = 0;
5291	thirdparty_id = 0;
5292
5293	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5294
5295	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5296	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5297	ctl_softc = control_softc;
5298
5299	switch (ctsio->cdb[0]) {
5300	case RESERVE_10: {
5301		struct scsi_reserve_10 *cdb;
5302
5303		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5304
5305		if (cdb->byte2 & SR10_LONGID)
5306			longid = 1;
5307		else
5308			thirdparty_id = cdb->thirdparty_id;
5309
5310		resv_id = cdb->resv_id;
5311		length = scsi_2btoul(cdb->length);
5312		break;
5313	}
5314	}
5315
5316	/*
5317	 * XXX KDM right now, we only support LUN reservation.  We don't
5318	 * support 3rd party reservations, or extent reservations, which
5319	 * might actually need the parameter list.  If we've gotten this
5320	 * far, we've got a LUN reservation.  Anything else got kicked out
5321	 * above.  So, according to SPC, ignore the length.
5322	 */
5323	length = 0;
5324
5325	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5326	 && (length > 0)) {
5327		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5328		ctsio->kern_data_len = length;
5329		ctsio->kern_total_len = length;
5330		ctsio->kern_data_resid = 0;
5331		ctsio->kern_rel_offset = 0;
5332		ctsio->kern_sg_entries = 0;
5333		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5334		ctsio->be_move_done = ctl_config_move_done;
5335		ctl_datamove((union ctl_io *)ctsio);
5336
5337		return (CTL_RETVAL_COMPLETE);
5338	}
5339
5340	if (length > 0)
5341		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5342
5343	mtx_lock(&lun->lun_lock);
5344	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5345		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5346		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5347		goto bailout;
5348	}
5349
5350	lun->flags |= CTL_LUN_RESERVED;
5351	lun->res_idx = residx;
5352
5353	ctsio->scsi_status = SCSI_STATUS_OK;
5354	ctsio->io_hdr.status = CTL_SUCCESS;
5355
5356bailout:
5357	mtx_unlock(&lun->lun_lock);
5358
5359	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5360		free(ctsio->kern_data_ptr, M_CTL);
5361		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5362	}
5363
5364	ctl_done((union ctl_io *)ctsio);
5365	return (CTL_RETVAL_COMPLETE);
5366}
5367
5368int
5369ctl_start_stop(struct ctl_scsiio *ctsio)
5370{
5371	struct scsi_start_stop_unit *cdb;
5372	struct ctl_lun *lun;
5373	struct ctl_softc *ctl_softc;
5374	int retval;
5375
5376	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5377
5378	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5379	ctl_softc = control_softc;
5380	retval = 0;
5381
5382	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5383
5384	/*
5385	 * XXX KDM
5386	 * We don't support the immediate bit on a stop unit.  In order to
5387	 * do that, we would need to code up a way to know that a stop is
5388	 * pending, and hold off any new commands until it completes, one
5389	 * way or another.  Then we could accept or reject those commands
5390	 * depending on its status.  We would almost need to do the reverse
5391	 * of what we do below for an immediate start -- return the copy of
5392	 * the ctl_io to the FETD with status to send to the host (and to
5393	 * free the copy!) and then free the original I/O once the stop
5394	 * actually completes.  That way, the OOA queue mechanism can work
5395	 * to block commands that shouldn't proceed.  Another alternative
5396	 * would be to put the copy in the queue in place of the original,
5397	 * and return the original back to the caller.  That could be
5398	 * slightly safer..
5399	 */
5400	if ((cdb->byte2 & SSS_IMMED)
5401	 && ((cdb->how & SSS_START) == 0)) {
5402		ctl_set_invalid_field(ctsio,
5403				      /*sks_valid*/ 1,
5404				      /*command*/ 1,
5405				      /*field*/ 1,
5406				      /*bit_valid*/ 1,
5407				      /*bit*/ 0);
5408		ctl_done((union ctl_io *)ctsio);
5409		return (CTL_RETVAL_COMPLETE);
5410	}
5411
5412	if ((lun->flags & CTL_LUN_PR_RESERVED)
5413	 && ((cdb->how & SSS_START)==0)) {
5414		uint32_t residx;
5415
5416		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5417		if (lun->pr_keys[residx] == 0
5418		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5419
5420			ctl_set_reservation_conflict(ctsio);
5421			ctl_done((union ctl_io *)ctsio);
5422			return (CTL_RETVAL_COMPLETE);
5423		}
5424	}
5425
5426	/*
5427	 * If there is no backend on this device, we can't start or stop
5428	 * it.  In theory we shouldn't get any start/stop commands in the
5429	 * first place at this level if the LUN doesn't have a backend.
5430	 * That should get stopped by the command decode code.
5431	 */
5432	if (lun->backend == NULL) {
5433		ctl_set_invalid_opcode(ctsio);
5434		ctl_done((union ctl_io *)ctsio);
5435		return (CTL_RETVAL_COMPLETE);
5436	}
5437
5438	/*
5439	 * XXX KDM Copan-specific offline behavior.
5440	 * Figure out a reasonable way to port this?
5441	 */
5442#ifdef NEEDTOPORT
5443	mtx_lock(&lun->lun_lock);
5444
5445	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5446	 && (lun->flags & CTL_LUN_OFFLINE)) {
5447		/*
5448		 * If the LUN is offline, and the on/offline bit isn't set,
5449		 * reject the start or stop.  Otherwise, let it through.
5450		 */
5451		mtx_unlock(&lun->lun_lock);
5452		ctl_set_lun_not_ready(ctsio);
5453		ctl_done((union ctl_io *)ctsio);
5454	} else {
5455		mtx_unlock(&lun->lun_lock);
5456#endif /* NEEDTOPORT */
5457		/*
5458		 * This could be a start or a stop when we're online,
5459		 * or a stop/offline or start/online.  A start or stop when
5460		 * we're offline is covered in the case above.
5461		 */
5462		/*
5463		 * In the non-immediate case, we send the request to
5464		 * the backend and return status to the user when
5465		 * it is done.
5466		 *
5467		 * In the immediate case, we allocate a new ctl_io
5468		 * to hold a copy of the request, and send that to
5469		 * the backend.  We then set good status on the
5470		 * user's request and return it immediately.
5471		 */
5472		if (cdb->byte2 & SSS_IMMED) {
5473			union ctl_io *new_io;
5474
5475			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5476			if (new_io == NULL) {
5477				ctl_set_busy(ctsio);
5478				ctl_done((union ctl_io *)ctsio);
5479			} else {
5480				ctl_copy_io((union ctl_io *)ctsio,
5481					    new_io);
5482				retval = lun->backend->config_write(new_io);
5483				ctl_set_success(ctsio);
5484				ctl_done((union ctl_io *)ctsio);
5485			}
5486		} else {
5487			retval = lun->backend->config_write(
5488				(union ctl_io *)ctsio);
5489		}
5490#ifdef NEEDTOPORT
5491	}
5492#endif
5493	return (retval);
5494}
5495
5496/*
5497 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5498 * we don't really do anything with the LBA and length fields if the user
5499 * passes them in.  Instead we'll just flush out the cache for the entire
5500 * LUN.
5501 */
5502int
5503ctl_sync_cache(struct ctl_scsiio *ctsio)
5504{
5505	struct ctl_lun *lun;
5506	struct ctl_softc *ctl_softc;
5507	uint64_t starting_lba;
5508	uint32_t block_count;
5509	int retval;
5510
5511	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5512
5513	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5514	ctl_softc = control_softc;
5515	retval = 0;
5516
5517	switch (ctsio->cdb[0]) {
5518	case SYNCHRONIZE_CACHE: {
5519		struct scsi_sync_cache *cdb;
5520		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5521
5522		starting_lba = scsi_4btoul(cdb->begin_lba);
5523		block_count = scsi_2btoul(cdb->lb_count);
5524		break;
5525	}
5526	case SYNCHRONIZE_CACHE_16: {
5527		struct scsi_sync_cache_16 *cdb;
5528		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5529
5530		starting_lba = scsi_8btou64(cdb->begin_lba);
5531		block_count = scsi_4btoul(cdb->lb_count);
5532		break;
5533	}
5534	default:
5535		ctl_set_invalid_opcode(ctsio);
5536		ctl_done((union ctl_io *)ctsio);
5537		goto bailout;
5538		break; /* NOTREACHED */
5539	}
5540
5541	/*
5542	 * We check the LBA and length, but don't do anything with them.
5543	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5544	 * get flushed.  This check will just help satisfy anyone who wants
5545	 * to see an error for an out of range LBA.
5546	 */
5547	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5548		ctl_set_lba_out_of_range(ctsio);
5549		ctl_done((union ctl_io *)ctsio);
5550		goto bailout;
5551	}
5552
5553	/*
5554	 * If this LUN has no backend, we can't flush the cache anyway.
5555	 */
5556	if (lun->backend == NULL) {
5557		ctl_set_invalid_opcode(ctsio);
5558		ctl_done((union ctl_io *)ctsio);
5559		goto bailout;
5560	}
5561
5562	/*
5563	 * Check to see whether we're configured to send the SYNCHRONIZE
5564	 * CACHE command directly to the back end.
5565	 */
5566	mtx_lock(&lun->lun_lock);
5567	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5568	 && (++(lun->sync_count) >= lun->sync_interval)) {
5569		lun->sync_count = 0;
5570		mtx_unlock(&lun->lun_lock);
5571		retval = lun->backend->config_write((union ctl_io *)ctsio);
5572	} else {
5573		mtx_unlock(&lun->lun_lock);
5574		ctl_set_success(ctsio);
5575		ctl_done((union ctl_io *)ctsio);
5576	}
5577
5578bailout:
5579
5580	return (retval);
5581}
5582
5583int
5584ctl_format(struct ctl_scsiio *ctsio)
5585{
5586	struct scsi_format *cdb;
5587	struct ctl_lun *lun;
5588	struct ctl_softc *ctl_softc;
5589	int length, defect_list_len;
5590
5591	CTL_DEBUG_PRINT(("ctl_format\n"));
5592
5593	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5594	ctl_softc = control_softc;
5595
5596	cdb = (struct scsi_format *)ctsio->cdb;
5597
5598	length = 0;
5599	if (cdb->byte2 & SF_FMTDATA) {
5600		if (cdb->byte2 & SF_LONGLIST)
5601			length = sizeof(struct scsi_format_header_long);
5602		else
5603			length = sizeof(struct scsi_format_header_short);
5604	}
5605
5606	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5607	 && (length > 0)) {
5608		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5609		ctsio->kern_data_len = length;
5610		ctsio->kern_total_len = length;
5611		ctsio->kern_data_resid = 0;
5612		ctsio->kern_rel_offset = 0;
5613		ctsio->kern_sg_entries = 0;
5614		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5615		ctsio->be_move_done = ctl_config_move_done;
5616		ctl_datamove((union ctl_io *)ctsio);
5617
5618		return (CTL_RETVAL_COMPLETE);
5619	}
5620
5621	defect_list_len = 0;
5622
5623	if (cdb->byte2 & SF_FMTDATA) {
5624		if (cdb->byte2 & SF_LONGLIST) {
5625			struct scsi_format_header_long *header;
5626
5627			header = (struct scsi_format_header_long *)
5628				ctsio->kern_data_ptr;
5629
5630			defect_list_len = scsi_4btoul(header->defect_list_len);
5631			if (defect_list_len != 0) {
5632				ctl_set_invalid_field(ctsio,
5633						      /*sks_valid*/ 1,
5634						      /*command*/ 0,
5635						      /*field*/ 2,
5636						      /*bit_valid*/ 0,
5637						      /*bit*/ 0);
5638				goto bailout;
5639			}
5640		} else {
5641			struct scsi_format_header_short *header;
5642
5643			header = (struct scsi_format_header_short *)
5644				ctsio->kern_data_ptr;
5645
5646			defect_list_len = scsi_2btoul(header->defect_list_len);
5647			if (defect_list_len != 0) {
5648				ctl_set_invalid_field(ctsio,
5649						      /*sks_valid*/ 1,
5650						      /*command*/ 0,
5651						      /*field*/ 2,
5652						      /*bit_valid*/ 0,
5653						      /*bit*/ 0);
5654				goto bailout;
5655			}
5656		}
5657	}
5658
5659	/*
5660	 * The format command will clear out the "Medium format corrupted"
5661	 * status if set by the configuration code.  That status is really
5662	 * just a way to notify the host that we have lost the media, and
5663	 * get them to issue a command that will basically make them think
5664	 * they're blowing away the media.
5665	 */
5666	mtx_lock(&lun->lun_lock);
5667	lun->flags &= ~CTL_LUN_INOPERABLE;
5668	mtx_unlock(&lun->lun_lock);
5669
5670	ctsio->scsi_status = SCSI_STATUS_OK;
5671	ctsio->io_hdr.status = CTL_SUCCESS;
5672bailout:
5673
5674	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5675		free(ctsio->kern_data_ptr, M_CTL);
5676		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5677	}
5678
5679	ctl_done((union ctl_io *)ctsio);
5680	return (CTL_RETVAL_COMPLETE);
5681}
5682
5683int
5684ctl_read_buffer(struct ctl_scsiio *ctsio)
5685{
5686	struct scsi_read_buffer *cdb;
5687	struct ctl_lun *lun;
5688	int buffer_offset, len;
5689	static uint8_t descr[4];
5690	static uint8_t echo_descr[4] = { 0 };
5691
5692	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5693
5694	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5695	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5696
5697	if (lun->flags & CTL_LUN_PR_RESERVED) {
5698		uint32_t residx;
5699
5700		/*
5701		 * XXX KDM need a lock here.
5702		 */
5703		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5704		if ((lun->res_type == SPR_TYPE_EX_AC
5705		  && residx != lun->pr_res_idx)
5706		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5707		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5708		  && lun->pr_keys[residx] == 0)) {
5709			ctl_set_reservation_conflict(ctsio);
5710			ctl_done((union ctl_io *)ctsio);
5711			return (CTL_RETVAL_COMPLETE);
5712	        }
5713	}
5714
5715	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5716	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5717	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5718		ctl_set_invalid_field(ctsio,
5719				      /*sks_valid*/ 1,
5720				      /*command*/ 1,
5721				      /*field*/ 1,
5722				      /*bit_valid*/ 1,
5723				      /*bit*/ 4);
5724		ctl_done((union ctl_io *)ctsio);
5725		return (CTL_RETVAL_COMPLETE);
5726	}
5727
5728	len = scsi_3btoul(cdb->length);
5729	buffer_offset = scsi_3btoul(cdb->offset);
5730
5731	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5732		ctl_set_invalid_field(ctsio,
5733				      /*sks_valid*/ 1,
5734				      /*command*/ 1,
5735				      /*field*/ 6,
5736				      /*bit_valid*/ 0,
5737				      /*bit*/ 0);
5738		ctl_done((union ctl_io *)ctsio);
5739		return (CTL_RETVAL_COMPLETE);
5740	}
5741
5742	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5743		descr[0] = 0;
5744		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5745		ctsio->kern_data_ptr = descr;
5746		len = min(len, sizeof(descr));
5747	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5748		ctsio->kern_data_ptr = echo_descr;
5749		len = min(len, sizeof(echo_descr));
5750	} else
5751		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5752	ctsio->kern_data_len = len;
5753	ctsio->kern_total_len = len;
5754	ctsio->kern_data_resid = 0;
5755	ctsio->kern_rel_offset = 0;
5756	ctsio->kern_sg_entries = 0;
5757	ctsio->be_move_done = ctl_config_move_done;
5758	ctl_datamove((union ctl_io *)ctsio);
5759
5760	return (CTL_RETVAL_COMPLETE);
5761}
5762
5763int
5764ctl_write_buffer(struct ctl_scsiio *ctsio)
5765{
5766	struct scsi_write_buffer *cdb;
5767	struct ctl_lun *lun;
5768	int buffer_offset, len;
5769
5770	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5771
5772	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5773	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5774
5775	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5776		ctl_set_invalid_field(ctsio,
5777				      /*sks_valid*/ 1,
5778				      /*command*/ 1,
5779				      /*field*/ 1,
5780				      /*bit_valid*/ 1,
5781				      /*bit*/ 4);
5782		ctl_done((union ctl_io *)ctsio);
5783		return (CTL_RETVAL_COMPLETE);
5784	}
5785
5786	len = scsi_3btoul(cdb->length);
5787	buffer_offset = scsi_3btoul(cdb->offset);
5788
5789	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5790		ctl_set_invalid_field(ctsio,
5791				      /*sks_valid*/ 1,
5792				      /*command*/ 1,
5793				      /*field*/ 6,
5794				      /*bit_valid*/ 0,
5795				      /*bit*/ 0);
5796		ctl_done((union ctl_io *)ctsio);
5797		return (CTL_RETVAL_COMPLETE);
5798	}
5799
5800	/*
5801	 * If we've got a kernel request that hasn't been malloced yet,
5802	 * malloc it and tell the caller the data buffer is here.
5803	 */
5804	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5805		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5806		ctsio->kern_data_len = len;
5807		ctsio->kern_total_len = len;
5808		ctsio->kern_data_resid = 0;
5809		ctsio->kern_rel_offset = 0;
5810		ctsio->kern_sg_entries = 0;
5811		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5812		ctsio->be_move_done = ctl_config_move_done;
5813		ctl_datamove((union ctl_io *)ctsio);
5814
5815		return (CTL_RETVAL_COMPLETE);
5816	}
5817
5818	ctl_done((union ctl_io *)ctsio);
5819
5820	return (CTL_RETVAL_COMPLETE);
5821}
5822
5823int
5824ctl_write_same(struct ctl_scsiio *ctsio)
5825{
5826	struct ctl_lun *lun;
5827	struct ctl_lba_len_flags *lbalen;
5828	uint64_t lba;
5829	uint32_t num_blocks;
5830	int len, retval;
5831	uint8_t byte2;
5832
5833	retval = CTL_RETVAL_COMPLETE;
5834
5835	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5836
5837	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5838
5839	switch (ctsio->cdb[0]) {
5840	case WRITE_SAME_10: {
5841		struct scsi_write_same_10 *cdb;
5842
5843		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5844
5845		lba = scsi_4btoul(cdb->addr);
5846		num_blocks = scsi_2btoul(cdb->length);
5847		byte2 = cdb->byte2;
5848		break;
5849	}
5850	case WRITE_SAME_16: {
5851		struct scsi_write_same_16 *cdb;
5852
5853		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5854
5855		lba = scsi_8btou64(cdb->addr);
5856		num_blocks = scsi_4btoul(cdb->length);
5857		byte2 = cdb->byte2;
5858		break;
5859	}
5860	default:
5861		/*
5862		 * We got a command we don't support.  This shouldn't
5863		 * happen, commands should be filtered out above us.
5864		 */
5865		ctl_set_invalid_opcode(ctsio);
5866		ctl_done((union ctl_io *)ctsio);
5867
5868		return (CTL_RETVAL_COMPLETE);
5869		break; /* NOTREACHED */
5870	}
5871
5872	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5873	if ((byte2 & SWS_UNMAP) == 0 &&
5874	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5875		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5876		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5877		ctl_done((union ctl_io *)ctsio);
5878		return (CTL_RETVAL_COMPLETE);
5879	}
5880
5881	/*
5882	 * The first check is to make sure we're in bounds, the second
5883	 * check is to catch wrap-around problems.  If the lba + num blocks
5884	 * is less than the lba, then we've wrapped around and the block
5885	 * range is invalid anyway.
5886	 */
5887	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5888	 || ((lba + num_blocks) < lba)) {
5889		ctl_set_lba_out_of_range(ctsio);
5890		ctl_done((union ctl_io *)ctsio);
5891		return (CTL_RETVAL_COMPLETE);
5892	}
5893
5894	/* Zero number of blocks means "to the last logical block" */
5895	if (num_blocks == 0) {
5896		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5897			ctl_set_invalid_field(ctsio,
5898					      /*sks_valid*/ 0,
5899					      /*command*/ 1,
5900					      /*field*/ 0,
5901					      /*bit_valid*/ 0,
5902					      /*bit*/ 0);
5903			ctl_done((union ctl_io *)ctsio);
5904			return (CTL_RETVAL_COMPLETE);
5905		}
5906		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5907	}
5908
5909	len = lun->be_lun->blocksize;
5910
5911	/*
5912	 * If we've got a kernel request that hasn't been malloced yet,
5913	 * malloc it and tell the caller the data buffer is here.
5914	 */
5915	if ((byte2 & SWS_NDOB) == 0 &&
5916	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5917		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5918		ctsio->kern_data_len = len;
5919		ctsio->kern_total_len = len;
5920		ctsio->kern_data_resid = 0;
5921		ctsio->kern_rel_offset = 0;
5922		ctsio->kern_sg_entries = 0;
5923		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5924		ctsio->be_move_done = ctl_config_move_done;
5925		ctl_datamove((union ctl_io *)ctsio);
5926
5927		return (CTL_RETVAL_COMPLETE);
5928	}
5929
5930	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5931	lbalen->lba = lba;
5932	lbalen->len = num_blocks;
5933	lbalen->flags = byte2;
5934	retval = lun->backend->config_write((union ctl_io *)ctsio);
5935
5936	return (retval);
5937}
5938
5939int
5940ctl_unmap(struct ctl_scsiio *ctsio)
5941{
5942	struct ctl_lun *lun;
5943	struct scsi_unmap *cdb;
5944	struct ctl_ptr_len_flags *ptrlen;
5945	struct scsi_unmap_header *hdr;
5946	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5947	uint64_t lba;
5948	uint32_t num_blocks;
5949	int len, retval;
5950	uint8_t byte2;
5951
5952	retval = CTL_RETVAL_COMPLETE;
5953
5954	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5955
5956	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5957	cdb = (struct scsi_unmap *)ctsio->cdb;
5958
5959	len = scsi_2btoul(cdb->length);
5960	byte2 = cdb->byte2;
5961
5962	/*
5963	 * If we've got a kernel request that hasn't been malloced yet,
5964	 * malloc it and tell the caller the data buffer is here.
5965	 */
5966	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5967		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5968		ctsio->kern_data_len = len;
5969		ctsio->kern_total_len = len;
5970		ctsio->kern_data_resid = 0;
5971		ctsio->kern_rel_offset = 0;
5972		ctsio->kern_sg_entries = 0;
5973		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5974		ctsio->be_move_done = ctl_config_move_done;
5975		ctl_datamove((union ctl_io *)ctsio);
5976
5977		return (CTL_RETVAL_COMPLETE);
5978	}
5979
5980	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5981	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5982	if (len < sizeof (*hdr) ||
5983	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5984	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5985	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5986		ctl_set_invalid_field(ctsio,
5987				      /*sks_valid*/ 0,
5988				      /*command*/ 0,
5989				      /*field*/ 0,
5990				      /*bit_valid*/ 0,
5991				      /*bit*/ 0);
5992		ctl_done((union ctl_io *)ctsio);
5993		return (CTL_RETVAL_COMPLETE);
5994	}
5995	len = scsi_2btoul(hdr->desc_length);
5996	buf = (struct scsi_unmap_desc *)(hdr + 1);
5997	end = buf + len / sizeof(*buf);
5998
5999	endnz = buf;
6000	for (range = buf; range < end; range++) {
6001		lba = scsi_8btou64(range->lba);
6002		num_blocks = scsi_4btoul(range->length);
6003		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6004		 || ((lba + num_blocks) < lba)) {
6005			ctl_set_lba_out_of_range(ctsio);
6006			ctl_done((union ctl_io *)ctsio);
6007			return (CTL_RETVAL_COMPLETE);
6008		}
6009		if (num_blocks != 0)
6010			endnz = range + 1;
6011	}
6012
6013	/*
6014	 * Block backend can not handle zero last range.
6015	 * Filter it out and return if there is nothing left.
6016	 */
6017	len = (uint8_t *)endnz - (uint8_t *)buf;
6018	if (len == 0) {
6019		ctl_set_success(ctsio);
6020		ctl_done((union ctl_io *)ctsio);
6021		return (CTL_RETVAL_COMPLETE);
6022	}
6023
6024	mtx_lock(&lun->lun_lock);
6025	ptrlen = (struct ctl_ptr_len_flags *)
6026	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6027	ptrlen->ptr = (void *)buf;
6028	ptrlen->len = len;
6029	ptrlen->flags = byte2;
6030	ctl_check_blocked(lun);
6031	mtx_unlock(&lun->lun_lock);
6032
6033	retval = lun->backend->config_write((union ctl_io *)ctsio);
6034	return (retval);
6035}
6036
6037/*
6038 * Note that this function currently doesn't actually do anything inside
6039 * CTL to enforce things if the DQue bit is turned on.
6040 *
6041 * Also note that this function can't be used in the default case, because
6042 * the DQue bit isn't set in the changeable mask for the control mode page
6043 * anyway.  This is just here as an example for how to implement a page
6044 * handler, and a placeholder in case we want to allow the user to turn
6045 * tagged queueing on and off.
6046 *
6047 * The D_SENSE bit handling is functional, however, and will turn
6048 * descriptor sense on and off for a given LUN.
6049 */
6050int
6051ctl_control_page_handler(struct ctl_scsiio *ctsio,
6052			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6053{
6054	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6055	struct ctl_lun *lun;
6056	struct ctl_softc *softc;
6057	int set_ua;
6058	uint32_t initidx;
6059
6060	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6061	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6062	set_ua = 0;
6063
6064	user_cp = (struct scsi_control_page *)page_ptr;
6065	current_cp = (struct scsi_control_page *)
6066		(page_index->page_data + (page_index->page_len *
6067		CTL_PAGE_CURRENT));
6068	saved_cp = (struct scsi_control_page *)
6069		(page_index->page_data + (page_index->page_len *
6070		CTL_PAGE_SAVED));
6071
6072	softc = control_softc;
6073
6074	mtx_lock(&lun->lun_lock);
6075	if (((current_cp->rlec & SCP_DSENSE) == 0)
6076	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6077		/*
6078		 * Descriptor sense is currently turned off and the user
6079		 * wants to turn it on.
6080		 */
6081		current_cp->rlec |= SCP_DSENSE;
6082		saved_cp->rlec |= SCP_DSENSE;
6083		lun->flags |= CTL_LUN_SENSE_DESC;
6084		set_ua = 1;
6085	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6086		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6087		/*
6088		 * Descriptor sense is currently turned on, and the user
6089		 * wants to turn it off.
6090		 */
6091		current_cp->rlec &= ~SCP_DSENSE;
6092		saved_cp->rlec &= ~SCP_DSENSE;
6093		lun->flags &= ~CTL_LUN_SENSE_DESC;
6094		set_ua = 1;
6095	}
6096	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6097	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6098		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6099		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6100		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6101		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6102		set_ua = 1;
6103	}
6104	if ((current_cp->eca_and_aen & SCP_SWP) !=
6105	    (user_cp->eca_and_aen & SCP_SWP)) {
6106		current_cp->eca_and_aen &= ~SCP_SWP;
6107		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6108		saved_cp->eca_and_aen &= ~SCP_SWP;
6109		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6110		set_ua = 1;
6111	}
6112	if (set_ua != 0) {
6113		int i;
6114		/*
6115		 * Let other initiators know that the mode
6116		 * parameters for this LUN have changed.
6117		 */
6118		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6119			if (i == initidx)
6120				continue;
6121
6122			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6123		}
6124	}
6125	mtx_unlock(&lun->lun_lock);
6126
6127	return (0);
6128}
6129
6130int
6131ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6132		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6133{
6134	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6135	struct ctl_lun *lun;
6136	int set_ua;
6137	uint32_t initidx;
6138
6139	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6140	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6141	set_ua = 0;
6142
6143	user_cp = (struct scsi_caching_page *)page_ptr;
6144	current_cp = (struct scsi_caching_page *)
6145		(page_index->page_data + (page_index->page_len *
6146		CTL_PAGE_CURRENT));
6147	saved_cp = (struct scsi_caching_page *)
6148		(page_index->page_data + (page_index->page_len *
6149		CTL_PAGE_SAVED));
6150
6151	mtx_lock(&lun->lun_lock);
6152	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6153	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6154		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6155		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6156		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6157		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6158		set_ua = 1;
6159	}
6160	if (set_ua != 0) {
6161		int i;
6162		/*
6163		 * Let other initiators know that the mode
6164		 * parameters for this LUN have changed.
6165		 */
6166		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6167			if (i == initidx)
6168				continue;
6169
6170			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6171		}
6172	}
6173	mtx_unlock(&lun->lun_lock);
6174
6175	return (0);
6176}
6177
6178int
6179ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6180				struct ctl_page_index *page_index,
6181				uint8_t *page_ptr)
6182{
6183	uint8_t *c;
6184	int i;
6185
6186	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6187	ctl_time_io_secs =
6188		(c[0] << 8) |
6189		(c[1] << 0) |
6190		0;
6191	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6192	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6193	printf("page data:");
6194	for (i=0; i<8; i++)
6195		printf(" %.2x",page_ptr[i]);
6196	printf("\n");
6197	return (0);
6198}
6199
6200int
6201ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6202			       struct ctl_page_index *page_index,
6203			       int pc)
6204{
6205	struct copan_debugconf_subpage *page;
6206
6207	page = (struct copan_debugconf_subpage *)page_index->page_data +
6208		(page_index->page_len * pc);
6209
6210	switch (pc) {
6211	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6212	case SMS_PAGE_CTRL_DEFAULT >> 6:
6213	case SMS_PAGE_CTRL_SAVED >> 6:
6214		/*
6215		 * We don't update the changable or default bits for this page.
6216		 */
6217		break;
6218	case SMS_PAGE_CTRL_CURRENT >> 6:
6219		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6220		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6221		break;
6222	default:
6223#ifdef NEEDTOPORT
6224		EPRINT(0, "Invalid PC %d!!", pc);
6225#endif /* NEEDTOPORT */
6226		break;
6227	}
6228	return (0);
6229}
6230
6231
6232static int
6233ctl_do_mode_select(union ctl_io *io)
6234{
6235	struct scsi_mode_page_header *page_header;
6236	struct ctl_page_index *page_index;
6237	struct ctl_scsiio *ctsio;
6238	int control_dev, page_len;
6239	int page_len_offset, page_len_size;
6240	union ctl_modepage_info *modepage_info;
6241	struct ctl_lun *lun;
6242	int *len_left, *len_used;
6243	int retval, i;
6244
6245	ctsio = &io->scsiio;
6246	page_index = NULL;
6247	page_len = 0;
6248	retval = CTL_RETVAL_COMPLETE;
6249
6250	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6251
6252	if (lun->be_lun->lun_type != T_DIRECT)
6253		control_dev = 1;
6254	else
6255		control_dev = 0;
6256
6257	modepage_info = (union ctl_modepage_info *)
6258		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6259	len_left = &modepage_info->header.len_left;
6260	len_used = &modepage_info->header.len_used;
6261
6262do_next_page:
6263
6264	page_header = (struct scsi_mode_page_header *)
6265		(ctsio->kern_data_ptr + *len_used);
6266
6267	if (*len_left == 0) {
6268		free(ctsio->kern_data_ptr, M_CTL);
6269		ctl_set_success(ctsio);
6270		ctl_done((union ctl_io *)ctsio);
6271		return (CTL_RETVAL_COMPLETE);
6272	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6273
6274		free(ctsio->kern_data_ptr, M_CTL);
6275		ctl_set_param_len_error(ctsio);
6276		ctl_done((union ctl_io *)ctsio);
6277		return (CTL_RETVAL_COMPLETE);
6278
6279	} else if ((page_header->page_code & SMPH_SPF)
6280		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6281
6282		free(ctsio->kern_data_ptr, M_CTL);
6283		ctl_set_param_len_error(ctsio);
6284		ctl_done((union ctl_io *)ctsio);
6285		return (CTL_RETVAL_COMPLETE);
6286	}
6287
6288
6289	/*
6290	 * XXX KDM should we do something with the block descriptor?
6291	 */
6292	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6293
6294		if ((control_dev != 0)
6295		 && (lun->mode_pages.index[i].page_flags &
6296		     CTL_PAGE_FLAG_DISK_ONLY))
6297			continue;
6298
6299		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6300		    (page_header->page_code & SMPH_PC_MASK))
6301			continue;
6302
6303		/*
6304		 * If neither page has a subpage code, then we've got a
6305		 * match.
6306		 */
6307		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6308		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6309			page_index = &lun->mode_pages.index[i];
6310			page_len = page_header->page_length;
6311			break;
6312		}
6313
6314		/*
6315		 * If both pages have subpages, then the subpage numbers
6316		 * have to match.
6317		 */
6318		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6319		  && (page_header->page_code & SMPH_SPF)) {
6320			struct scsi_mode_page_header_sp *sph;
6321
6322			sph = (struct scsi_mode_page_header_sp *)page_header;
6323
6324			if (lun->mode_pages.index[i].subpage ==
6325			    sph->subpage) {
6326				page_index = &lun->mode_pages.index[i];
6327				page_len = scsi_2btoul(sph->page_length);
6328				break;
6329			}
6330		}
6331	}
6332
6333	/*
6334	 * If we couldn't find the page, or if we don't have a mode select
6335	 * handler for it, send back an error to the user.
6336	 */
6337	if ((page_index == NULL)
6338	 || (page_index->select_handler == NULL)) {
6339		ctl_set_invalid_field(ctsio,
6340				      /*sks_valid*/ 1,
6341				      /*command*/ 0,
6342				      /*field*/ *len_used,
6343				      /*bit_valid*/ 0,
6344				      /*bit*/ 0);
6345		free(ctsio->kern_data_ptr, M_CTL);
6346		ctl_done((union ctl_io *)ctsio);
6347		return (CTL_RETVAL_COMPLETE);
6348	}
6349
6350	if (page_index->page_code & SMPH_SPF) {
6351		page_len_offset = 2;
6352		page_len_size = 2;
6353	} else {
6354		page_len_size = 1;
6355		page_len_offset = 1;
6356	}
6357
6358	/*
6359	 * If the length the initiator gives us isn't the one we specify in
6360	 * the mode page header, or if they didn't specify enough data in
6361	 * the CDB to avoid truncating this page, kick out the request.
6362	 */
6363	if ((page_len != (page_index->page_len - page_len_offset -
6364			  page_len_size))
6365	 || (*len_left < page_index->page_len)) {
6366
6367
6368		ctl_set_invalid_field(ctsio,
6369				      /*sks_valid*/ 1,
6370				      /*command*/ 0,
6371				      /*field*/ *len_used + page_len_offset,
6372				      /*bit_valid*/ 0,
6373				      /*bit*/ 0);
6374		free(ctsio->kern_data_ptr, M_CTL);
6375		ctl_done((union ctl_io *)ctsio);
6376		return (CTL_RETVAL_COMPLETE);
6377	}
6378
6379	/*
6380	 * Run through the mode page, checking to make sure that the bits
6381	 * the user changed are actually legal for him to change.
6382	 */
6383	for (i = 0; i < page_index->page_len; i++) {
6384		uint8_t *user_byte, *change_mask, *current_byte;
6385		int bad_bit;
6386		int j;
6387
6388		user_byte = (uint8_t *)page_header + i;
6389		change_mask = page_index->page_data +
6390			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6391		current_byte = page_index->page_data +
6392			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6393
6394		/*
6395		 * Check to see whether the user set any bits in this byte
6396		 * that he is not allowed to set.
6397		 */
6398		if ((*user_byte & ~(*change_mask)) ==
6399		    (*current_byte & ~(*change_mask)))
6400			continue;
6401
6402		/*
6403		 * Go through bit by bit to determine which one is illegal.
6404		 */
6405		bad_bit = 0;
6406		for (j = 7; j >= 0; j--) {
6407			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6408			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6409				bad_bit = i;
6410				break;
6411			}
6412		}
6413		ctl_set_invalid_field(ctsio,
6414				      /*sks_valid*/ 1,
6415				      /*command*/ 0,
6416				      /*field*/ *len_used + i,
6417				      /*bit_valid*/ 1,
6418				      /*bit*/ bad_bit);
6419		free(ctsio->kern_data_ptr, M_CTL);
6420		ctl_done((union ctl_io *)ctsio);
6421		return (CTL_RETVAL_COMPLETE);
6422	}
6423
6424	/*
6425	 * Decrement these before we call the page handler, since we may
6426	 * end up getting called back one way or another before the handler
6427	 * returns to this context.
6428	 */
6429	*len_left -= page_index->page_len;
6430	*len_used += page_index->page_len;
6431
6432	retval = page_index->select_handler(ctsio, page_index,
6433					    (uint8_t *)page_header);
6434
6435	/*
6436	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6437	 * wait until this queued command completes to finish processing
6438	 * the mode page.  If it returns anything other than
6439	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6440	 * already set the sense information, freed the data pointer, and
6441	 * completed the io for us.
6442	 */
6443	if (retval != CTL_RETVAL_COMPLETE)
6444		goto bailout_no_done;
6445
6446	/*
6447	 * If the initiator sent us more than one page, parse the next one.
6448	 */
6449	if (*len_left > 0)
6450		goto do_next_page;
6451
6452	ctl_set_success(ctsio);
6453	free(ctsio->kern_data_ptr, M_CTL);
6454	ctl_done((union ctl_io *)ctsio);
6455
6456bailout_no_done:
6457
6458	return (CTL_RETVAL_COMPLETE);
6459
6460}
6461
6462int
6463ctl_mode_select(struct ctl_scsiio *ctsio)
6464{
6465	int param_len, pf, sp;
6466	int header_size, bd_len;
6467	int len_left, len_used;
6468	struct ctl_page_index *page_index;
6469	struct ctl_lun *lun;
6470	int control_dev, page_len;
6471	union ctl_modepage_info *modepage_info;
6472	int retval;
6473
6474	pf = 0;
6475	sp = 0;
6476	page_len = 0;
6477	len_used = 0;
6478	len_left = 0;
6479	retval = 0;
6480	bd_len = 0;
6481	page_index = NULL;
6482
6483	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6484
6485	if (lun->be_lun->lun_type != T_DIRECT)
6486		control_dev = 1;
6487	else
6488		control_dev = 0;
6489
6490	switch (ctsio->cdb[0]) {
6491	case MODE_SELECT_6: {
6492		struct scsi_mode_select_6 *cdb;
6493
6494		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6495
6496		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6497		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6498
6499		param_len = cdb->length;
6500		header_size = sizeof(struct scsi_mode_header_6);
6501		break;
6502	}
6503	case MODE_SELECT_10: {
6504		struct scsi_mode_select_10 *cdb;
6505
6506		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6507
6508		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6509		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6510
6511		param_len = scsi_2btoul(cdb->length);
6512		header_size = sizeof(struct scsi_mode_header_10);
6513		break;
6514	}
6515	default:
6516		ctl_set_invalid_opcode(ctsio);
6517		ctl_done((union ctl_io *)ctsio);
6518		return (CTL_RETVAL_COMPLETE);
6519		break; /* NOTREACHED */
6520	}
6521
6522	/*
6523	 * From SPC-3:
6524	 * "A parameter list length of zero indicates that the Data-Out Buffer
6525	 * shall be empty. This condition shall not be considered as an error."
6526	 */
6527	if (param_len == 0) {
6528		ctl_set_success(ctsio);
6529		ctl_done((union ctl_io *)ctsio);
6530		return (CTL_RETVAL_COMPLETE);
6531	}
6532
6533	/*
6534	 * Since we'll hit this the first time through, prior to
6535	 * allocation, we don't need to free a data buffer here.
6536	 */
6537	if (param_len < header_size) {
6538		ctl_set_param_len_error(ctsio);
6539		ctl_done((union ctl_io *)ctsio);
6540		return (CTL_RETVAL_COMPLETE);
6541	}
6542
6543	/*
6544	 * Allocate the data buffer and grab the user's data.  In theory,
6545	 * we shouldn't have to sanity check the parameter list length here
6546	 * because the maximum size is 64K.  We should be able to malloc
6547	 * that much without too many problems.
6548	 */
6549	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6550		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6551		ctsio->kern_data_len = param_len;
6552		ctsio->kern_total_len = param_len;
6553		ctsio->kern_data_resid = 0;
6554		ctsio->kern_rel_offset = 0;
6555		ctsio->kern_sg_entries = 0;
6556		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6557		ctsio->be_move_done = ctl_config_move_done;
6558		ctl_datamove((union ctl_io *)ctsio);
6559
6560		return (CTL_RETVAL_COMPLETE);
6561	}
6562
6563	switch (ctsio->cdb[0]) {
6564	case MODE_SELECT_6: {
6565		struct scsi_mode_header_6 *mh6;
6566
6567		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6568		bd_len = mh6->blk_desc_len;
6569		break;
6570	}
6571	case MODE_SELECT_10: {
6572		struct scsi_mode_header_10 *mh10;
6573
6574		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6575		bd_len = scsi_2btoul(mh10->blk_desc_len);
6576		break;
6577	}
6578	default:
6579		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6580		break;
6581	}
6582
6583	if (param_len < (header_size + bd_len)) {
6584		free(ctsio->kern_data_ptr, M_CTL);
6585		ctl_set_param_len_error(ctsio);
6586		ctl_done((union ctl_io *)ctsio);
6587		return (CTL_RETVAL_COMPLETE);
6588	}
6589
6590	/*
6591	 * Set the IO_CONT flag, so that if this I/O gets passed to
6592	 * ctl_config_write_done(), it'll get passed back to
6593	 * ctl_do_mode_select() for further processing, or completion if
6594	 * we're all done.
6595	 */
6596	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6597	ctsio->io_cont = ctl_do_mode_select;
6598
6599	modepage_info = (union ctl_modepage_info *)
6600		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6601
6602	memset(modepage_info, 0, sizeof(*modepage_info));
6603
6604	len_left = param_len - header_size - bd_len;
6605	len_used = header_size + bd_len;
6606
6607	modepage_info->header.len_left = len_left;
6608	modepage_info->header.len_used = len_used;
6609
6610	return (ctl_do_mode_select((union ctl_io *)ctsio));
6611}
6612
6613int
6614ctl_mode_sense(struct ctl_scsiio *ctsio)
6615{
6616	struct ctl_lun *lun;
6617	int pc, page_code, dbd, llba, subpage;
6618	int alloc_len, page_len, header_len, total_len;
6619	struct scsi_mode_block_descr *block_desc;
6620	struct ctl_page_index *page_index;
6621	int control_dev;
6622
6623	dbd = 0;
6624	llba = 0;
6625	block_desc = NULL;
6626	page_index = NULL;
6627
6628	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6629
6630	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6631
6632	if (lun->be_lun->lun_type != T_DIRECT)
6633		control_dev = 1;
6634	else
6635		control_dev = 0;
6636
6637	if (lun->flags & CTL_LUN_PR_RESERVED) {
6638		uint32_t residx;
6639
6640		/*
6641		 * XXX KDM need a lock here.
6642		 */
6643		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6644		if ((lun->res_type == SPR_TYPE_EX_AC
6645		  && residx != lun->pr_res_idx)
6646		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6647		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6648		  && lun->pr_keys[residx] == 0)) {
6649			ctl_set_reservation_conflict(ctsio);
6650			ctl_done((union ctl_io *)ctsio);
6651			return (CTL_RETVAL_COMPLETE);
6652		}
6653	}
6654
6655	switch (ctsio->cdb[0]) {
6656	case MODE_SENSE_6: {
6657		struct scsi_mode_sense_6 *cdb;
6658
6659		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6660
6661		header_len = sizeof(struct scsi_mode_hdr_6);
6662		if (cdb->byte2 & SMS_DBD)
6663			dbd = 1;
6664		else
6665			header_len += sizeof(struct scsi_mode_block_descr);
6666
6667		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6668		page_code = cdb->page & SMS_PAGE_CODE;
6669		subpage = cdb->subpage;
6670		alloc_len = cdb->length;
6671		break;
6672	}
6673	case MODE_SENSE_10: {
6674		struct scsi_mode_sense_10 *cdb;
6675
6676		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6677
6678		header_len = sizeof(struct scsi_mode_hdr_10);
6679
6680		if (cdb->byte2 & SMS_DBD)
6681			dbd = 1;
6682		else
6683			header_len += sizeof(struct scsi_mode_block_descr);
6684		if (cdb->byte2 & SMS10_LLBAA)
6685			llba = 1;
6686		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6687		page_code = cdb->page & SMS_PAGE_CODE;
6688		subpage = cdb->subpage;
6689		alloc_len = scsi_2btoul(cdb->length);
6690		break;
6691	}
6692	default:
6693		ctl_set_invalid_opcode(ctsio);
6694		ctl_done((union ctl_io *)ctsio);
6695		return (CTL_RETVAL_COMPLETE);
6696		break; /* NOTREACHED */
6697	}
6698
6699	/*
6700	 * We have to make a first pass through to calculate the size of
6701	 * the pages that match the user's query.  Then we allocate enough
6702	 * memory to hold it, and actually copy the data into the buffer.
6703	 */
6704	switch (page_code) {
6705	case SMS_ALL_PAGES_PAGE: {
6706		int i;
6707
6708		page_len = 0;
6709
6710		/*
6711		 * At the moment, values other than 0 and 0xff here are
6712		 * reserved according to SPC-3.
6713		 */
6714		if ((subpage != SMS_SUBPAGE_PAGE_0)
6715		 && (subpage != SMS_SUBPAGE_ALL)) {
6716			ctl_set_invalid_field(ctsio,
6717					      /*sks_valid*/ 1,
6718					      /*command*/ 1,
6719					      /*field*/ 3,
6720					      /*bit_valid*/ 0,
6721					      /*bit*/ 0);
6722			ctl_done((union ctl_io *)ctsio);
6723			return (CTL_RETVAL_COMPLETE);
6724		}
6725
6726		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6727			if ((control_dev != 0)
6728			 && (lun->mode_pages.index[i].page_flags &
6729			     CTL_PAGE_FLAG_DISK_ONLY))
6730				continue;
6731
6732			/*
6733			 * We don't use this subpage if the user didn't
6734			 * request all subpages.
6735			 */
6736			if ((lun->mode_pages.index[i].subpage != 0)
6737			 && (subpage == SMS_SUBPAGE_PAGE_0))
6738				continue;
6739
6740#if 0
6741			printf("found page %#x len %d\n",
6742			       lun->mode_pages.index[i].page_code &
6743			       SMPH_PC_MASK,
6744			       lun->mode_pages.index[i].page_len);
6745#endif
6746			page_len += lun->mode_pages.index[i].page_len;
6747		}
6748		break;
6749	}
6750	default: {
6751		int i;
6752
6753		page_len = 0;
6754
6755		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6756			/* Look for the right page code */
6757			if ((lun->mode_pages.index[i].page_code &
6758			     SMPH_PC_MASK) != page_code)
6759				continue;
6760
6761			/* Look for the right subpage or the subpage wildcard*/
6762			if ((lun->mode_pages.index[i].subpage != subpage)
6763			 && (subpage != SMS_SUBPAGE_ALL))
6764				continue;
6765
6766			/* Make sure the page is supported for this dev type */
6767			if ((control_dev != 0)
6768			 && (lun->mode_pages.index[i].page_flags &
6769			     CTL_PAGE_FLAG_DISK_ONLY))
6770				continue;
6771
6772#if 0
6773			printf("found page %#x len %d\n",
6774			       lun->mode_pages.index[i].page_code &
6775			       SMPH_PC_MASK,
6776			       lun->mode_pages.index[i].page_len);
6777#endif
6778
6779			page_len += lun->mode_pages.index[i].page_len;
6780		}
6781
6782		if (page_len == 0) {
6783			ctl_set_invalid_field(ctsio,
6784					      /*sks_valid*/ 1,
6785					      /*command*/ 1,
6786					      /*field*/ 2,
6787					      /*bit_valid*/ 1,
6788					      /*bit*/ 5);
6789			ctl_done((union ctl_io *)ctsio);
6790			return (CTL_RETVAL_COMPLETE);
6791		}
6792		break;
6793	}
6794	}
6795
6796	total_len = header_len + page_len;
6797#if 0
6798	printf("header_len = %d, page_len = %d, total_len = %d\n",
6799	       header_len, page_len, total_len);
6800#endif
6801
6802	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6803	ctsio->kern_sg_entries = 0;
6804	ctsio->kern_data_resid = 0;
6805	ctsio->kern_rel_offset = 0;
6806	if (total_len < alloc_len) {
6807		ctsio->residual = alloc_len - total_len;
6808		ctsio->kern_data_len = total_len;
6809		ctsio->kern_total_len = total_len;
6810	} else {
6811		ctsio->residual = 0;
6812		ctsio->kern_data_len = alloc_len;
6813		ctsio->kern_total_len = alloc_len;
6814	}
6815
6816	switch (ctsio->cdb[0]) {
6817	case MODE_SENSE_6: {
6818		struct scsi_mode_hdr_6 *header;
6819
6820		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6821
6822		header->datalen = ctl_min(total_len - 1, 254);
6823		if (control_dev == 0) {
6824			header->dev_specific = 0x10; /* DPOFUA */
6825			if ((lun->flags & CTL_LUN_READONLY) ||
6826			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6827			    .eca_and_aen & SCP_SWP) != 0)
6828				    header->dev_specific |= 0x80; /* WP */
6829		}
6830		if (dbd)
6831			header->block_descr_len = 0;
6832		else
6833			header->block_descr_len =
6834				sizeof(struct scsi_mode_block_descr);
6835		block_desc = (struct scsi_mode_block_descr *)&header[1];
6836		break;
6837	}
6838	case MODE_SENSE_10: {
6839		struct scsi_mode_hdr_10 *header;
6840		int datalen;
6841
6842		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6843
6844		datalen = ctl_min(total_len - 2, 65533);
6845		scsi_ulto2b(datalen, header->datalen);
6846		if (control_dev == 0) {
6847			header->dev_specific = 0x10; /* DPOFUA */
6848			if ((lun->flags & CTL_LUN_READONLY) ||
6849			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6850			    .eca_and_aen & SCP_SWP) != 0)
6851				    header->dev_specific |= 0x80; /* WP */
6852		}
6853		if (dbd)
6854			scsi_ulto2b(0, header->block_descr_len);
6855		else
6856			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6857				    header->block_descr_len);
6858		block_desc = (struct scsi_mode_block_descr *)&header[1];
6859		break;
6860	}
6861	default:
6862		panic("invalid CDB type %#x", ctsio->cdb[0]);
6863		break; /* NOTREACHED */
6864	}
6865
6866	/*
6867	 * If we've got a disk, use its blocksize in the block
6868	 * descriptor.  Otherwise, just set it to 0.
6869	 */
6870	if (dbd == 0) {
6871		if (control_dev == 0)
6872			scsi_ulto3b(lun->be_lun->blocksize,
6873				    block_desc->block_len);
6874		else
6875			scsi_ulto3b(0, block_desc->block_len);
6876	}
6877
6878	switch (page_code) {
6879	case SMS_ALL_PAGES_PAGE: {
6880		int i, data_used;
6881
6882		data_used = header_len;
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			if ((control_dev != 0)
6889			 && (page_index->page_flags &
6890			    CTL_PAGE_FLAG_DISK_ONLY))
6891				continue;
6892
6893			/*
6894			 * We don't use this subpage if the user didn't
6895			 * request all subpages.  We already checked (above)
6896			 * to make sure the user only specified a subpage
6897			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6898			 */
6899			if ((page_index->subpage != 0)
6900			 && (subpage == SMS_SUBPAGE_PAGE_0))
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	default: {
6919		int i, data_used;
6920
6921		data_used = header_len;
6922
6923		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6924			struct ctl_page_index *page_index;
6925
6926			page_index = &lun->mode_pages.index[i];
6927
6928			/* Look for the right page code */
6929			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6930				continue;
6931
6932			/* Look for the right subpage or the subpage wildcard*/
6933			if ((page_index->subpage != subpage)
6934			 && (subpage != SMS_SUBPAGE_ALL))
6935				continue;
6936
6937			/* Make sure the page is supported for this dev type */
6938			if ((control_dev != 0)
6939			 && (page_index->page_flags &
6940			     CTL_PAGE_FLAG_DISK_ONLY))
6941				continue;
6942
6943			/*
6944			 * Call the handler, if it exists, to update the
6945			 * page to the latest values.
6946			 */
6947			if (page_index->sense_handler != NULL)
6948				page_index->sense_handler(ctsio, page_index,pc);
6949
6950			memcpy(ctsio->kern_data_ptr + data_used,
6951			       page_index->page_data +
6952			       (page_index->page_len * pc),
6953			       page_index->page_len);
6954			data_used += page_index->page_len;
6955		}
6956		break;
6957	}
6958	}
6959
6960	ctsio->scsi_status = SCSI_STATUS_OK;
6961
6962	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6963	ctsio->be_move_done = ctl_config_move_done;
6964	ctl_datamove((union ctl_io *)ctsio);
6965
6966	return (CTL_RETVAL_COMPLETE);
6967}
6968
6969int
6970ctl_log_sense(struct ctl_scsiio *ctsio)
6971{
6972	struct ctl_lun *lun;
6973	int i, pc, page_code, subpage;
6974	int alloc_len, total_len;
6975	struct ctl_page_index *page_index;
6976	struct scsi_log_sense *cdb;
6977	struct scsi_log_header *header;
6978
6979	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6980
6981	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6982	cdb = (struct scsi_log_sense *)ctsio->cdb;
6983	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6984	page_code = cdb->page & SLS_PAGE_CODE;
6985	subpage = cdb->subpage;
6986	alloc_len = scsi_2btoul(cdb->length);
6987
6988	page_index = NULL;
6989	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6990		page_index = &lun->log_pages.index[i];
6991
6992		/* Look for the right page code */
6993		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6994			continue;
6995
6996		/* Look for the right subpage or the subpage wildcard*/
6997		if (page_index->subpage != subpage)
6998			continue;
6999
7000		break;
7001	}
7002	if (i >= CTL_NUM_LOG_PAGES) {
7003		ctl_set_invalid_field(ctsio,
7004				      /*sks_valid*/ 1,
7005				      /*command*/ 1,
7006				      /*field*/ 2,
7007				      /*bit_valid*/ 0,
7008				      /*bit*/ 0);
7009		ctl_done((union ctl_io *)ctsio);
7010		return (CTL_RETVAL_COMPLETE);
7011	}
7012
7013	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
7014
7015	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7016	ctsio->kern_sg_entries = 0;
7017	ctsio->kern_data_resid = 0;
7018	ctsio->kern_rel_offset = 0;
7019	if (total_len < alloc_len) {
7020		ctsio->residual = alloc_len - total_len;
7021		ctsio->kern_data_len = total_len;
7022		ctsio->kern_total_len = total_len;
7023	} else {
7024		ctsio->residual = 0;
7025		ctsio->kern_data_len = alloc_len;
7026		ctsio->kern_total_len = alloc_len;
7027	}
7028
7029	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
7030	header->page = page_index->page_code;
7031	if (page_index->subpage) {
7032		header->page |= SL_SPF;
7033		header->subpage = page_index->subpage;
7034	}
7035	scsi_ulto2b(page_index->page_len, header->datalen);
7036
7037	/*
7038	 * Call the handler, if it exists, to update the
7039	 * page to the latest values.
7040	 */
7041	if (page_index->sense_handler != NULL)
7042		page_index->sense_handler(ctsio, page_index, pc);
7043
7044	memcpy(header + 1, page_index->page_data, page_index->page_len);
7045
7046	ctsio->scsi_status = SCSI_STATUS_OK;
7047	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7048	ctsio->be_move_done = ctl_config_move_done;
7049	ctl_datamove((union ctl_io *)ctsio);
7050
7051	return (CTL_RETVAL_COMPLETE);
7052}
7053
7054int
7055ctl_read_capacity(struct ctl_scsiio *ctsio)
7056{
7057	struct scsi_read_capacity *cdb;
7058	struct scsi_read_capacity_data *data;
7059	struct ctl_lun *lun;
7060	uint32_t lba;
7061
7062	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7063
7064	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7065
7066	lba = scsi_4btoul(cdb->addr);
7067	if (((cdb->pmi & SRC_PMI) == 0)
7068	 && (lba != 0)) {
7069		ctl_set_invalid_field(/*ctsio*/ ctsio,
7070				      /*sks_valid*/ 1,
7071				      /*command*/ 1,
7072				      /*field*/ 2,
7073				      /*bit_valid*/ 0,
7074				      /*bit*/ 0);
7075		ctl_done((union ctl_io *)ctsio);
7076		return (CTL_RETVAL_COMPLETE);
7077	}
7078
7079	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7080
7081	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7082	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7083	ctsio->residual = 0;
7084	ctsio->kern_data_len = sizeof(*data);
7085	ctsio->kern_total_len = sizeof(*data);
7086	ctsio->kern_data_resid = 0;
7087	ctsio->kern_rel_offset = 0;
7088	ctsio->kern_sg_entries = 0;
7089
7090	/*
7091	 * If the maximum LBA is greater than 0xfffffffe, the user must
7092	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7093	 * serivce action set.
7094	 */
7095	if (lun->be_lun->maxlba > 0xfffffffe)
7096		scsi_ulto4b(0xffffffff, data->addr);
7097	else
7098		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7099
7100	/*
7101	 * XXX KDM this may not be 512 bytes...
7102	 */
7103	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7104
7105	ctsio->scsi_status = SCSI_STATUS_OK;
7106
7107	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7108	ctsio->be_move_done = ctl_config_move_done;
7109	ctl_datamove((union ctl_io *)ctsio);
7110
7111	return (CTL_RETVAL_COMPLETE);
7112}
7113
7114int
7115ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7116{
7117	struct scsi_read_capacity_16 *cdb;
7118	struct scsi_read_capacity_data_long *data;
7119	struct ctl_lun *lun;
7120	uint64_t lba;
7121	uint32_t alloc_len;
7122
7123	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7124
7125	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7126
7127	alloc_len = scsi_4btoul(cdb->alloc_len);
7128	lba = scsi_8btou64(cdb->addr);
7129
7130	if ((cdb->reladr & SRC16_PMI)
7131	 && (lba != 0)) {
7132		ctl_set_invalid_field(/*ctsio*/ ctsio,
7133				      /*sks_valid*/ 1,
7134				      /*command*/ 1,
7135				      /*field*/ 2,
7136				      /*bit_valid*/ 0,
7137				      /*bit*/ 0);
7138		ctl_done((union ctl_io *)ctsio);
7139		return (CTL_RETVAL_COMPLETE);
7140	}
7141
7142	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7143
7144	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7145	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7146
7147	if (sizeof(*data) < alloc_len) {
7148		ctsio->residual = alloc_len - sizeof(*data);
7149		ctsio->kern_data_len = sizeof(*data);
7150		ctsio->kern_total_len = sizeof(*data);
7151	} else {
7152		ctsio->residual = 0;
7153		ctsio->kern_data_len = alloc_len;
7154		ctsio->kern_total_len = alloc_len;
7155	}
7156	ctsio->kern_data_resid = 0;
7157	ctsio->kern_rel_offset = 0;
7158	ctsio->kern_sg_entries = 0;
7159
7160	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7161	/* XXX KDM this may not be 512 bytes... */
7162	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7163	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7164	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7165	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7166		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7167
7168	ctsio->scsi_status = SCSI_STATUS_OK;
7169
7170	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7171	ctsio->be_move_done = ctl_config_move_done;
7172	ctl_datamove((union ctl_io *)ctsio);
7173
7174	return (CTL_RETVAL_COMPLETE);
7175}
7176
7177int
7178ctl_read_defect(struct ctl_scsiio *ctsio)
7179{
7180	struct scsi_read_defect_data_10 *ccb10;
7181	struct scsi_read_defect_data_12 *ccb12;
7182	struct scsi_read_defect_data_hdr_10 *data10;
7183	struct scsi_read_defect_data_hdr_12 *data12;
7184	struct ctl_lun *lun;
7185	uint32_t alloc_len, data_len;
7186	uint8_t format;
7187
7188	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7189
7190	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7191	if (lun->flags & CTL_LUN_PR_RESERVED) {
7192		uint32_t residx;
7193
7194		/*
7195		 * XXX KDM need a lock here.
7196		 */
7197		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7198		if ((lun->res_type == SPR_TYPE_EX_AC
7199		  && residx != lun->pr_res_idx)
7200		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
7201		   || lun->res_type == SPR_TYPE_EX_AC_AR)
7202		  && lun->pr_keys[residx] == 0)) {
7203			ctl_set_reservation_conflict(ctsio);
7204			ctl_done((union ctl_io *)ctsio);
7205			return (CTL_RETVAL_COMPLETE);
7206	        }
7207	}
7208
7209	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7210		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7211		format = ccb10->format;
7212		alloc_len = scsi_2btoul(ccb10->alloc_length);
7213		data_len = sizeof(*data10);
7214	} else {
7215		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7216		format = ccb12->format;
7217		alloc_len = scsi_4btoul(ccb12->alloc_length);
7218		data_len = sizeof(*data12);
7219	}
7220	if (alloc_len == 0) {
7221		ctl_set_success(ctsio);
7222		ctl_done((union ctl_io *)ctsio);
7223		return (CTL_RETVAL_COMPLETE);
7224	}
7225
7226	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7227	if (data_len < alloc_len) {
7228		ctsio->residual = alloc_len - data_len;
7229		ctsio->kern_data_len = data_len;
7230		ctsio->kern_total_len = data_len;
7231	} else {
7232		ctsio->residual = 0;
7233		ctsio->kern_data_len = alloc_len;
7234		ctsio->kern_total_len = alloc_len;
7235	}
7236	ctsio->kern_data_resid = 0;
7237	ctsio->kern_rel_offset = 0;
7238	ctsio->kern_sg_entries = 0;
7239
7240	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7241		data10 = (struct scsi_read_defect_data_hdr_10 *)
7242		    ctsio->kern_data_ptr;
7243		data10->format = format;
7244		scsi_ulto2b(0, data10->length);
7245	} else {
7246		data12 = (struct scsi_read_defect_data_hdr_12 *)
7247		    ctsio->kern_data_ptr;
7248		data12->format = format;
7249		scsi_ulto2b(0, data12->generation);
7250		scsi_ulto4b(0, data12->length);
7251	}
7252
7253	ctsio->scsi_status = SCSI_STATUS_OK;
7254	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7255	ctsio->be_move_done = ctl_config_move_done;
7256	ctl_datamove((union ctl_io *)ctsio);
7257	return (CTL_RETVAL_COMPLETE);
7258}
7259
7260int
7261ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7262{
7263	struct scsi_maintenance_in *cdb;
7264	int retval;
7265	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7266	int num_target_port_groups, num_target_ports, single;
7267	struct ctl_lun *lun;
7268	struct ctl_softc *softc;
7269	struct ctl_port *port;
7270	struct scsi_target_group_data *rtg_ptr;
7271	struct scsi_target_group_data_extended *rtg_ext_ptr;
7272	struct scsi_target_port_group_descriptor *tpg_desc;
7273
7274	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7275
7276	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7277	softc = control_softc;
7278	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7279
7280	retval = CTL_RETVAL_COMPLETE;
7281
7282	switch (cdb->byte2 & STG_PDF_MASK) {
7283	case STG_PDF_LENGTH:
7284		ext = 0;
7285		break;
7286	case STG_PDF_EXTENDED:
7287		ext = 1;
7288		break;
7289	default:
7290		ctl_set_invalid_field(/*ctsio*/ ctsio,
7291				      /*sks_valid*/ 1,
7292				      /*command*/ 1,
7293				      /*field*/ 2,
7294				      /*bit_valid*/ 1,
7295				      /*bit*/ 5);
7296		ctl_done((union ctl_io *)ctsio);
7297		return(retval);
7298	}
7299
7300	single = ctl_is_single;
7301	if (single)
7302		num_target_port_groups = 1;
7303	else
7304		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7305	num_target_ports = 0;
7306	mtx_lock(&softc->ctl_lock);
7307	STAILQ_FOREACH(port, &softc->port_list, links) {
7308		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7309			continue;
7310		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7311			continue;
7312		num_target_ports++;
7313	}
7314	mtx_unlock(&softc->ctl_lock);
7315
7316	if (ext)
7317		total_len = sizeof(struct scsi_target_group_data_extended);
7318	else
7319		total_len = sizeof(struct scsi_target_group_data);
7320	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7321		num_target_port_groups +
7322	    sizeof(struct scsi_target_port_descriptor) *
7323		num_target_ports * num_target_port_groups;
7324
7325	alloc_len = scsi_4btoul(cdb->length);
7326
7327	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7328
7329	ctsio->kern_sg_entries = 0;
7330
7331	if (total_len < alloc_len) {
7332		ctsio->residual = alloc_len - total_len;
7333		ctsio->kern_data_len = total_len;
7334		ctsio->kern_total_len = total_len;
7335	} else {
7336		ctsio->residual = 0;
7337		ctsio->kern_data_len = alloc_len;
7338		ctsio->kern_total_len = alloc_len;
7339	}
7340	ctsio->kern_data_resid = 0;
7341	ctsio->kern_rel_offset = 0;
7342
7343	if (ext) {
7344		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7345		    ctsio->kern_data_ptr;
7346		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7347		rtg_ext_ptr->format_type = 0x10;
7348		rtg_ext_ptr->implicit_transition_time = 0;
7349		tpg_desc = &rtg_ext_ptr->groups[0];
7350	} else {
7351		rtg_ptr = (struct scsi_target_group_data *)
7352		    ctsio->kern_data_ptr;
7353		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7354		tpg_desc = &rtg_ptr->groups[0];
7355	}
7356
7357	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7358	mtx_lock(&softc->ctl_lock);
7359	for (g = 0; g < num_target_port_groups; g++) {
7360		if (g == pg)
7361			tpg_desc->pref_state = TPG_PRIMARY |
7362			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7363		else
7364			tpg_desc->pref_state =
7365			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7366		tpg_desc->support = TPG_AO_SUP;
7367		if (!single)
7368			tpg_desc->support |= TPG_AN_SUP;
7369		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7370		tpg_desc->status = TPG_IMPLICIT;
7371		pc = 0;
7372		STAILQ_FOREACH(port, &softc->port_list, links) {
7373			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7374				continue;
7375			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7376			    CTL_MAX_LUNS)
7377				continue;
7378			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7379			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7380			    relative_target_port_identifier);
7381			pc++;
7382		}
7383		tpg_desc->target_port_count = pc;
7384		tpg_desc = (struct scsi_target_port_group_descriptor *)
7385		    &tpg_desc->descriptors[pc];
7386	}
7387	mtx_unlock(&softc->ctl_lock);
7388
7389	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7390	ctsio->be_move_done = ctl_config_move_done;
7391
7392	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7393			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7394			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7395			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7396			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7397
7398	ctl_datamove((union ctl_io *)ctsio);
7399	return(retval);
7400}
7401
7402int
7403ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7404{
7405	struct ctl_lun *lun;
7406	struct scsi_report_supported_opcodes *cdb;
7407	const struct ctl_cmd_entry *entry, *sentry;
7408	struct scsi_report_supported_opcodes_all *all;
7409	struct scsi_report_supported_opcodes_descr *descr;
7410	struct scsi_report_supported_opcodes_one *one;
7411	int retval;
7412	int alloc_len, total_len;
7413	int opcode, service_action, i, j, num;
7414
7415	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7416
7417	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7418	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7419
7420	retval = CTL_RETVAL_COMPLETE;
7421
7422	opcode = cdb->requested_opcode;
7423	service_action = scsi_2btoul(cdb->requested_service_action);
7424	switch (cdb->options & RSO_OPTIONS_MASK) {
7425	case RSO_OPTIONS_ALL:
7426		num = 0;
7427		for (i = 0; i < 256; i++) {
7428			entry = &ctl_cmd_table[i];
7429			if (entry->flags & CTL_CMD_FLAG_SA5) {
7430				for (j = 0; j < 32; j++) {
7431					sentry = &((const struct ctl_cmd_entry *)
7432					    entry->execute)[j];
7433					if (ctl_cmd_applicable(
7434					    lun->be_lun->lun_type, sentry))
7435						num++;
7436				}
7437			} else {
7438				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7439				    entry))
7440					num++;
7441			}
7442		}
7443		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7444		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7445		break;
7446	case RSO_OPTIONS_OC:
7447		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7448			ctl_set_invalid_field(/*ctsio*/ ctsio,
7449					      /*sks_valid*/ 1,
7450					      /*command*/ 1,
7451					      /*field*/ 2,
7452					      /*bit_valid*/ 1,
7453					      /*bit*/ 2);
7454			ctl_done((union ctl_io *)ctsio);
7455			return (CTL_RETVAL_COMPLETE);
7456		}
7457		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7458		break;
7459	case RSO_OPTIONS_OC_SA:
7460		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7461		    service_action >= 32) {
7462			ctl_set_invalid_field(/*ctsio*/ ctsio,
7463					      /*sks_valid*/ 1,
7464					      /*command*/ 1,
7465					      /*field*/ 2,
7466					      /*bit_valid*/ 1,
7467					      /*bit*/ 2);
7468			ctl_done((union ctl_io *)ctsio);
7469			return (CTL_RETVAL_COMPLETE);
7470		}
7471		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7472		break;
7473	default:
7474		ctl_set_invalid_field(/*ctsio*/ ctsio,
7475				      /*sks_valid*/ 1,
7476				      /*command*/ 1,
7477				      /*field*/ 2,
7478				      /*bit_valid*/ 1,
7479				      /*bit*/ 2);
7480		ctl_done((union ctl_io *)ctsio);
7481		return (CTL_RETVAL_COMPLETE);
7482	}
7483
7484	alloc_len = scsi_4btoul(cdb->length);
7485
7486	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7487
7488	ctsio->kern_sg_entries = 0;
7489
7490	if (total_len < alloc_len) {
7491		ctsio->residual = alloc_len - total_len;
7492		ctsio->kern_data_len = total_len;
7493		ctsio->kern_total_len = total_len;
7494	} else {
7495		ctsio->residual = 0;
7496		ctsio->kern_data_len = alloc_len;
7497		ctsio->kern_total_len = alloc_len;
7498	}
7499	ctsio->kern_data_resid = 0;
7500	ctsio->kern_rel_offset = 0;
7501
7502	switch (cdb->options & RSO_OPTIONS_MASK) {
7503	case RSO_OPTIONS_ALL:
7504		all = (struct scsi_report_supported_opcodes_all *)
7505		    ctsio->kern_data_ptr;
7506		num = 0;
7507		for (i = 0; i < 256; i++) {
7508			entry = &ctl_cmd_table[i];
7509			if (entry->flags & CTL_CMD_FLAG_SA5) {
7510				for (j = 0; j < 32; j++) {
7511					sentry = &((const struct ctl_cmd_entry *)
7512					    entry->execute)[j];
7513					if (!ctl_cmd_applicable(
7514					    lun->be_lun->lun_type, sentry))
7515						continue;
7516					descr = &all->descr[num++];
7517					descr->opcode = i;
7518					scsi_ulto2b(j, descr->service_action);
7519					descr->flags = RSO_SERVACTV;
7520					scsi_ulto2b(sentry->length,
7521					    descr->cdb_length);
7522				}
7523			} else {
7524				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7525				    entry))
7526					continue;
7527				descr = &all->descr[num++];
7528				descr->opcode = i;
7529				scsi_ulto2b(0, descr->service_action);
7530				descr->flags = 0;
7531				scsi_ulto2b(entry->length, descr->cdb_length);
7532			}
7533		}
7534		scsi_ulto4b(
7535		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7536		    all->length);
7537		break;
7538	case RSO_OPTIONS_OC:
7539		one = (struct scsi_report_supported_opcodes_one *)
7540		    ctsio->kern_data_ptr;
7541		entry = &ctl_cmd_table[opcode];
7542		goto fill_one;
7543	case RSO_OPTIONS_OC_SA:
7544		one = (struct scsi_report_supported_opcodes_one *)
7545		    ctsio->kern_data_ptr;
7546		entry = &ctl_cmd_table[opcode];
7547		entry = &((const struct ctl_cmd_entry *)
7548		    entry->execute)[service_action];
7549fill_one:
7550		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7551			one->support = 3;
7552			scsi_ulto2b(entry->length, one->cdb_length);
7553			one->cdb_usage[0] = opcode;
7554			memcpy(&one->cdb_usage[1], entry->usage,
7555			    entry->length - 1);
7556		} else
7557			one->support = 1;
7558		break;
7559	}
7560
7561	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7562	ctsio->be_move_done = ctl_config_move_done;
7563
7564	ctl_datamove((union ctl_io *)ctsio);
7565	return(retval);
7566}
7567
7568int
7569ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7570{
7571	struct ctl_lun *lun;
7572	struct scsi_report_supported_tmf *cdb;
7573	struct scsi_report_supported_tmf_data *data;
7574	int retval;
7575	int alloc_len, total_len;
7576
7577	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7578
7579	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7580	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7581
7582	retval = CTL_RETVAL_COMPLETE;
7583
7584	total_len = sizeof(struct scsi_report_supported_tmf_data);
7585	alloc_len = scsi_4btoul(cdb->length);
7586
7587	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7588
7589	ctsio->kern_sg_entries = 0;
7590
7591	if (total_len < alloc_len) {
7592		ctsio->residual = alloc_len - total_len;
7593		ctsio->kern_data_len = total_len;
7594		ctsio->kern_total_len = total_len;
7595	} else {
7596		ctsio->residual = 0;
7597		ctsio->kern_data_len = alloc_len;
7598		ctsio->kern_total_len = alloc_len;
7599	}
7600	ctsio->kern_data_resid = 0;
7601	ctsio->kern_rel_offset = 0;
7602
7603	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7604	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7605	data->byte2 |= RST_ITNRS;
7606
7607	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7608	ctsio->be_move_done = ctl_config_move_done;
7609
7610	ctl_datamove((union ctl_io *)ctsio);
7611	return (retval);
7612}
7613
7614int
7615ctl_report_timestamp(struct ctl_scsiio *ctsio)
7616{
7617	struct ctl_lun *lun;
7618	struct scsi_report_timestamp *cdb;
7619	struct scsi_report_timestamp_data *data;
7620	struct timeval tv;
7621	int64_t timestamp;
7622	int retval;
7623	int alloc_len, total_len;
7624
7625	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7626
7627	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7628	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7629
7630	retval = CTL_RETVAL_COMPLETE;
7631
7632	total_len = sizeof(struct scsi_report_timestamp_data);
7633	alloc_len = scsi_4btoul(cdb->length);
7634
7635	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7636
7637	ctsio->kern_sg_entries = 0;
7638
7639	if (total_len < alloc_len) {
7640		ctsio->residual = alloc_len - total_len;
7641		ctsio->kern_data_len = total_len;
7642		ctsio->kern_total_len = total_len;
7643	} else {
7644		ctsio->residual = 0;
7645		ctsio->kern_data_len = alloc_len;
7646		ctsio->kern_total_len = alloc_len;
7647	}
7648	ctsio->kern_data_resid = 0;
7649	ctsio->kern_rel_offset = 0;
7650
7651	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7652	scsi_ulto2b(sizeof(*data) - 2, data->length);
7653	data->origin = RTS_ORIG_OUTSIDE;
7654	getmicrotime(&tv);
7655	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7656	scsi_ulto4b(timestamp >> 16, data->timestamp);
7657	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7658
7659	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7660	ctsio->be_move_done = ctl_config_move_done;
7661
7662	ctl_datamove((union ctl_io *)ctsio);
7663	return (retval);
7664}
7665
7666int
7667ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7668{
7669	struct scsi_per_res_in *cdb;
7670	int alloc_len, total_len = 0;
7671	/* struct scsi_per_res_in_rsrv in_data; */
7672	struct ctl_lun *lun;
7673	struct ctl_softc *softc;
7674
7675	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7676
7677	softc = control_softc;
7678
7679	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7680
7681	alloc_len = scsi_2btoul(cdb->length);
7682
7683	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7684
7685retry:
7686	mtx_lock(&lun->lun_lock);
7687	switch (cdb->action) {
7688	case SPRI_RK: /* read keys */
7689		total_len = sizeof(struct scsi_per_res_in_keys) +
7690			lun->pr_key_count *
7691			sizeof(struct scsi_per_res_key);
7692		break;
7693	case SPRI_RR: /* read reservation */
7694		if (lun->flags & CTL_LUN_PR_RESERVED)
7695			total_len = sizeof(struct scsi_per_res_in_rsrv);
7696		else
7697			total_len = sizeof(struct scsi_per_res_in_header);
7698		break;
7699	case SPRI_RC: /* report capabilities */
7700		total_len = sizeof(struct scsi_per_res_cap);
7701		break;
7702	case SPRI_RS: /* read full status */
7703		total_len = sizeof(struct scsi_per_res_in_header) +
7704		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7705		    lun->pr_key_count;
7706		break;
7707	default:
7708		panic("Invalid PR type %x", cdb->action);
7709	}
7710	mtx_unlock(&lun->lun_lock);
7711
7712	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7713
7714	if (total_len < alloc_len) {
7715		ctsio->residual = alloc_len - total_len;
7716		ctsio->kern_data_len = total_len;
7717		ctsio->kern_total_len = total_len;
7718	} else {
7719		ctsio->residual = 0;
7720		ctsio->kern_data_len = alloc_len;
7721		ctsio->kern_total_len = alloc_len;
7722	}
7723
7724	ctsio->kern_data_resid = 0;
7725	ctsio->kern_rel_offset = 0;
7726	ctsio->kern_sg_entries = 0;
7727
7728	mtx_lock(&lun->lun_lock);
7729	switch (cdb->action) {
7730	case SPRI_RK: { // read keys
7731        struct scsi_per_res_in_keys *res_keys;
7732		int i, key_count;
7733
7734		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7735
7736		/*
7737		 * We had to drop the lock to allocate our buffer, which
7738		 * leaves time for someone to come in with another
7739		 * persistent reservation.  (That is unlikely, though,
7740		 * since this should be the only persistent reservation
7741		 * command active right now.)
7742		 */
7743		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7744		    (lun->pr_key_count *
7745		     sizeof(struct scsi_per_res_key)))){
7746			mtx_unlock(&lun->lun_lock);
7747			free(ctsio->kern_data_ptr, M_CTL);
7748			printf("%s: reservation length changed, retrying\n",
7749			       __func__);
7750			goto retry;
7751		}
7752
7753		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7754
7755		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7756			     lun->pr_key_count, res_keys->header.length);
7757
7758		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7759			if (lun->pr_keys[i] == 0)
7760				continue;
7761
7762			/*
7763			 * We used lun->pr_key_count to calculate the
7764			 * size to allocate.  If it turns out the number of
7765			 * initiators with the registered flag set is
7766			 * larger than that (i.e. they haven't been kept in
7767			 * sync), we've got a problem.
7768			 */
7769			if (key_count >= lun->pr_key_count) {
7770#ifdef NEEDTOPORT
7771				csevent_log(CSC_CTL | CSC_SHELF_SW |
7772					    CTL_PR_ERROR,
7773					    csevent_LogType_Fault,
7774					    csevent_AlertLevel_Yellow,
7775					    csevent_FRU_ShelfController,
7776					    csevent_FRU_Firmware,
7777				        csevent_FRU_Unknown,
7778					    "registered keys %d >= key "
7779					    "count %d", key_count,
7780					    lun->pr_key_count);
7781#endif
7782				key_count++;
7783				continue;
7784			}
7785			scsi_u64to8b(lun->pr_keys[i],
7786			    res_keys->keys[key_count].key);
7787			key_count++;
7788		}
7789		break;
7790	}
7791	case SPRI_RR: { // read reservation
7792		struct scsi_per_res_in_rsrv *res;
7793		int tmp_len, header_only;
7794
7795		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7796
7797		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7798
7799		if (lun->flags & CTL_LUN_PR_RESERVED)
7800		{
7801			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7802			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7803				    res->header.length);
7804			header_only = 0;
7805		} else {
7806			tmp_len = sizeof(struct scsi_per_res_in_header);
7807			scsi_ulto4b(0, res->header.length);
7808			header_only = 1;
7809		}
7810
7811		/*
7812		 * We had to drop the lock to allocate our buffer, which
7813		 * leaves time for someone to come in with another
7814		 * persistent reservation.  (That is unlikely, though,
7815		 * since this should be the only persistent reservation
7816		 * command active right now.)
7817		 */
7818		if (tmp_len != total_len) {
7819			mtx_unlock(&lun->lun_lock);
7820			free(ctsio->kern_data_ptr, M_CTL);
7821			printf("%s: reservation status changed, retrying\n",
7822			       __func__);
7823			goto retry;
7824		}
7825
7826		/*
7827		 * No reservation held, so we're done.
7828		 */
7829		if (header_only != 0)
7830			break;
7831
7832		/*
7833		 * If the registration is an All Registrants type, the key
7834		 * is 0, since it doesn't really matter.
7835		 */
7836		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7837			scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7838			    res->data.reservation);
7839		}
7840		res->data.scopetype = lun->res_type;
7841		break;
7842	}
7843	case SPRI_RC:     //report capabilities
7844	{
7845		struct scsi_per_res_cap *res_cap;
7846		uint16_t type_mask;
7847
7848		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7849		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7850		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7851		type_mask = SPRI_TM_WR_EX_AR |
7852			    SPRI_TM_EX_AC_RO |
7853			    SPRI_TM_WR_EX_RO |
7854			    SPRI_TM_EX_AC |
7855			    SPRI_TM_WR_EX |
7856			    SPRI_TM_EX_AC_AR;
7857		scsi_ulto2b(type_mask, res_cap->type_mask);
7858		break;
7859	}
7860	case SPRI_RS: { // read full status
7861		struct scsi_per_res_in_full *res_status;
7862		struct scsi_per_res_in_full_desc *res_desc;
7863		struct ctl_port *port;
7864		int i, len;
7865
7866		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7867
7868		/*
7869		 * We had to drop the lock to allocate our buffer, which
7870		 * leaves time for someone to come in with another
7871		 * persistent reservation.  (That is unlikely, though,
7872		 * since this should be the only persistent reservation
7873		 * command active right now.)
7874		 */
7875		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7876		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7877		     lun->pr_key_count)){
7878			mtx_unlock(&lun->lun_lock);
7879			free(ctsio->kern_data_ptr, M_CTL);
7880			printf("%s: reservation length changed, retrying\n",
7881			       __func__);
7882			goto retry;
7883		}
7884
7885		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7886
7887		res_desc = &res_status->desc[0];
7888		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7889			if (lun->pr_keys[i] == 0)
7890				continue;
7891
7892			scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
7893			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7894			    (lun->pr_res_idx == i ||
7895			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7896				res_desc->flags = SPRI_FULL_R_HOLDER;
7897				res_desc->scopetype = lun->res_type;
7898			}
7899			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7900			    res_desc->rel_trgt_port_id);
7901			len = 0;
7902			port = softc->ctl_ports[
7903			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7904			if (port != NULL)
7905				len = ctl_create_iid(port,
7906				    i % CTL_MAX_INIT_PER_PORT,
7907				    res_desc->transport_id);
7908			scsi_ulto4b(len, res_desc->additional_length);
7909			res_desc = (struct scsi_per_res_in_full_desc *)
7910			    &res_desc->transport_id[len];
7911		}
7912		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7913		    res_status->header.length);
7914		break;
7915	}
7916	default:
7917		/*
7918		 * This is a bug, because we just checked for this above,
7919		 * and should have returned an error.
7920		 */
7921		panic("Invalid PR type %x", cdb->action);
7922		break; /* NOTREACHED */
7923	}
7924	mtx_unlock(&lun->lun_lock);
7925
7926	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7927	ctsio->be_move_done = ctl_config_move_done;
7928
7929	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7930			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7931			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7932			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7933			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7934
7935	ctl_datamove((union ctl_io *)ctsio);
7936
7937	return (CTL_RETVAL_COMPLETE);
7938}
7939
7940/*
7941 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7942 * it should return.
7943 */
7944static int
7945ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7946		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7947		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7948		struct scsi_per_res_out_parms* param)
7949{
7950	union ctl_ha_msg persis_io;
7951	int retval, i;
7952	int isc_retval;
7953
7954	retval = 0;
7955
7956	mtx_lock(&lun->lun_lock);
7957	if (sa_res_key == 0) {
7958		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7959			/* validate scope and type */
7960			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7961			     SPR_LU_SCOPE) {
7962				mtx_unlock(&lun->lun_lock);
7963				ctl_set_invalid_field(/*ctsio*/ ctsio,
7964						      /*sks_valid*/ 1,
7965						      /*command*/ 1,
7966						      /*field*/ 2,
7967						      /*bit_valid*/ 1,
7968						      /*bit*/ 4);
7969				ctl_done((union ctl_io *)ctsio);
7970				return (1);
7971			}
7972
7973		        if (type>8 || type==2 || type==4 || type==0) {
7974				mtx_unlock(&lun->lun_lock);
7975				ctl_set_invalid_field(/*ctsio*/ ctsio,
7976       	           				      /*sks_valid*/ 1,
7977						      /*command*/ 1,
7978						      /*field*/ 2,
7979						      /*bit_valid*/ 1,
7980						      /*bit*/ 0);
7981				ctl_done((union ctl_io *)ctsio);
7982				return (1);
7983		        }
7984
7985			/*
7986			 * Unregister everybody else and build UA for
7987			 * them
7988			 */
7989			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7990				if (i == residx || lun->pr_keys[i] == 0)
7991					continue;
7992
7993				if (!persis_offset
7994				 && i <CTL_MAX_INITIATORS)
7995					lun->pending_ua[i] |=
7996						CTL_UA_REG_PREEMPT;
7997				else if (persis_offset
7998				      && i >= persis_offset)
7999					lun->pending_ua[i-persis_offset] |=
8000						CTL_UA_REG_PREEMPT;
8001				lun->pr_keys[i] = 0;
8002			}
8003			lun->pr_key_count = 1;
8004			lun->res_type = type;
8005			if (lun->res_type != SPR_TYPE_WR_EX_AR
8006			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8007				lun->pr_res_idx = residx;
8008
8009			/* send msg to other side */
8010			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8011			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8012			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8013			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8014			persis_io.pr.pr_info.res_type = type;
8015			memcpy(persis_io.pr.pr_info.sa_res_key,
8016			       param->serv_act_res_key,
8017			       sizeof(param->serv_act_res_key));
8018			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8019			     &persis_io, sizeof(persis_io), 0)) >
8020			     CTL_HA_STATUS_SUCCESS) {
8021				printf("CTL:Persis Out error returned "
8022				       "from ctl_ha_msg_send %d\n",
8023				       isc_retval);
8024			}
8025		} else {
8026			/* not all registrants */
8027			mtx_unlock(&lun->lun_lock);
8028			free(ctsio->kern_data_ptr, M_CTL);
8029			ctl_set_invalid_field(ctsio,
8030					      /*sks_valid*/ 1,
8031					      /*command*/ 0,
8032					      /*field*/ 8,
8033					      /*bit_valid*/ 0,
8034					      /*bit*/ 0);
8035			ctl_done((union ctl_io *)ctsio);
8036			return (1);
8037		}
8038	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8039		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8040		int found = 0;
8041
8042		if (res_key == sa_res_key) {
8043			/* special case */
8044			/*
8045			 * The spec implies this is not good but doesn't
8046			 * say what to do. There are two choices either
8047			 * generate a res conflict or check condition
8048			 * with illegal field in parameter data. Since
8049			 * that is what is done when the sa_res_key is
8050			 * zero I'll take that approach since this has
8051			 * to do with the sa_res_key.
8052			 */
8053			mtx_unlock(&lun->lun_lock);
8054			free(ctsio->kern_data_ptr, M_CTL);
8055			ctl_set_invalid_field(ctsio,
8056					      /*sks_valid*/ 1,
8057					      /*command*/ 0,
8058					      /*field*/ 8,
8059					      /*bit_valid*/ 0,
8060					      /*bit*/ 0);
8061			ctl_done((union ctl_io *)ctsio);
8062			return (1);
8063		}
8064
8065		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8066			if (lun->pr_keys[i] != sa_res_key)
8067				continue;
8068
8069			found = 1;
8070			lun->pr_keys[i] = 0;
8071			lun->pr_key_count--;
8072
8073			if (!persis_offset && i < CTL_MAX_INITIATORS)
8074				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8075			else if (persis_offset && i >= persis_offset)
8076				lun->pending_ua[i-persis_offset] |=
8077					CTL_UA_REG_PREEMPT;
8078		}
8079		if (!found) {
8080			mtx_unlock(&lun->lun_lock);
8081			free(ctsio->kern_data_ptr, M_CTL);
8082			ctl_set_reservation_conflict(ctsio);
8083			ctl_done((union ctl_io *)ctsio);
8084			return (CTL_RETVAL_COMPLETE);
8085		}
8086		/* send msg to other side */
8087		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8088		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8089		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8090		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8091		persis_io.pr.pr_info.res_type = type;
8092		memcpy(persis_io.pr.pr_info.sa_res_key,
8093		       param->serv_act_res_key,
8094		       sizeof(param->serv_act_res_key));
8095		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8096		     &persis_io, sizeof(persis_io), 0)) >
8097		     CTL_HA_STATUS_SUCCESS) {
8098			printf("CTL:Persis Out error returned from "
8099			       "ctl_ha_msg_send %d\n", isc_retval);
8100		}
8101	} else {
8102		/* Reserved but not all registrants */
8103		/* sa_res_key is res holder */
8104		if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8105			/* validate scope and type */
8106			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8107			     SPR_LU_SCOPE) {
8108				mtx_unlock(&lun->lun_lock);
8109				ctl_set_invalid_field(/*ctsio*/ ctsio,
8110						      /*sks_valid*/ 1,
8111						      /*command*/ 1,
8112						      /*field*/ 2,
8113						      /*bit_valid*/ 1,
8114						      /*bit*/ 4);
8115				ctl_done((union ctl_io *)ctsio);
8116				return (1);
8117			}
8118
8119			if (type>8 || type==2 || type==4 || type==0) {
8120				mtx_unlock(&lun->lun_lock);
8121				ctl_set_invalid_field(/*ctsio*/ ctsio,
8122						      /*sks_valid*/ 1,
8123						      /*command*/ 1,
8124						      /*field*/ 2,
8125						      /*bit_valid*/ 1,
8126						      /*bit*/ 0);
8127				ctl_done((union ctl_io *)ctsio);
8128				return (1);
8129			}
8130
8131			/*
8132			 * Do the following:
8133			 * if sa_res_key != res_key remove all
8134			 * registrants w/sa_res_key and generate UA
8135			 * for these registrants(Registrations
8136			 * Preempted) if it wasn't an exclusive
8137			 * reservation generate UA(Reservations
8138			 * Preempted) for all other registered nexuses
8139			 * if the type has changed. Establish the new
8140			 * reservation and holder. If res_key and
8141			 * sa_res_key are the same do the above
8142			 * except don't unregister the res holder.
8143			 */
8144
8145			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8146				if (i == residx || lun->pr_keys[i] == 0)
8147					continue;
8148
8149				if (sa_res_key == lun->pr_keys[i]) {
8150					lun->pr_keys[i] = 0;
8151					lun->pr_key_count--;
8152
8153					if (!persis_offset
8154					 && i < CTL_MAX_INITIATORS)
8155						lun->pending_ua[i] |=
8156							CTL_UA_REG_PREEMPT;
8157					else if (persis_offset
8158					      && i >= persis_offset)
8159						lun->pending_ua[i-persis_offset] |=
8160						  CTL_UA_REG_PREEMPT;
8161				} else if (type != lun->res_type
8162					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8163					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8164						if (!persis_offset
8165						 && i < CTL_MAX_INITIATORS)
8166							lun->pending_ua[i] |=
8167							CTL_UA_RES_RELEASE;
8168						else if (persis_offset
8169						      && i >= persis_offset)
8170							lun->pending_ua[
8171							i-persis_offset] |=
8172							CTL_UA_RES_RELEASE;
8173				}
8174			}
8175			lun->res_type = type;
8176			if (lun->res_type != SPR_TYPE_WR_EX_AR
8177			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8178				lun->pr_res_idx = residx;
8179			else
8180				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8181
8182			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8183			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8184			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8185			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8186			persis_io.pr.pr_info.res_type = type;
8187			memcpy(persis_io.pr.pr_info.sa_res_key,
8188			       param->serv_act_res_key,
8189			       sizeof(param->serv_act_res_key));
8190			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8191			     &persis_io, sizeof(persis_io), 0)) >
8192			     CTL_HA_STATUS_SUCCESS) {
8193				printf("CTL:Persis Out error returned "
8194				       "from ctl_ha_msg_send %d\n",
8195				       isc_retval);
8196			}
8197		} else {
8198			/*
8199			 * sa_res_key is not the res holder just
8200			 * remove registrants
8201			 */
8202			int found=0;
8203
8204			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8205				if (sa_res_key != lun->pr_keys[i])
8206					continue;
8207
8208				found = 1;
8209				lun->pr_keys[i] = 0;
8210				lun->pr_key_count--;
8211
8212				if (!persis_offset
8213				 && i < CTL_MAX_INITIATORS)
8214					lun->pending_ua[i] |=
8215						CTL_UA_REG_PREEMPT;
8216				else if (persis_offset
8217				      && i >= persis_offset)
8218					lun->pending_ua[i-persis_offset] |=
8219						CTL_UA_REG_PREEMPT;
8220			}
8221
8222			if (!found) {
8223				mtx_unlock(&lun->lun_lock);
8224				free(ctsio->kern_data_ptr, M_CTL);
8225				ctl_set_reservation_conflict(ctsio);
8226				ctl_done((union ctl_io *)ctsio);
8227		        	return (1);
8228			}
8229			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8230			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8231			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8232			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8233			persis_io.pr.pr_info.res_type = type;
8234			memcpy(persis_io.pr.pr_info.sa_res_key,
8235			       param->serv_act_res_key,
8236			       sizeof(param->serv_act_res_key));
8237			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8238			     &persis_io, sizeof(persis_io), 0)) >
8239			     CTL_HA_STATUS_SUCCESS) {
8240				printf("CTL:Persis Out error returned "
8241				       "from ctl_ha_msg_send %d\n",
8242				isc_retval);
8243			}
8244		}
8245	}
8246
8247	lun->PRGeneration++;
8248	mtx_unlock(&lun->lun_lock);
8249
8250	return (retval);
8251}
8252
8253static void
8254ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8255{
8256	uint64_t sa_res_key;
8257	int i;
8258
8259	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8260
8261	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8262	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8263	 || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8264		if (sa_res_key == 0) {
8265			/*
8266			 * Unregister everybody else and build UA for
8267			 * them
8268			 */
8269			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8270				if (i == msg->pr.pr_info.residx ||
8271				    lun->pr_keys[i] == 0)
8272					continue;
8273
8274				if (!persis_offset
8275				 && i < CTL_MAX_INITIATORS)
8276					lun->pending_ua[i] |=
8277						CTL_UA_REG_PREEMPT;
8278				else if (persis_offset && i >= persis_offset)
8279					lun->pending_ua[i - persis_offset] |=
8280						CTL_UA_REG_PREEMPT;
8281				lun->pr_keys[i] = 0;
8282			}
8283
8284			lun->pr_key_count = 1;
8285			lun->res_type = msg->pr.pr_info.res_type;
8286			if (lun->res_type != SPR_TYPE_WR_EX_AR
8287			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8288				lun->pr_res_idx = msg->pr.pr_info.residx;
8289		} else {
8290		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8291				if (sa_res_key == lun->pr_keys[i])
8292					continue;
8293
8294				lun->pr_keys[i] = 0;
8295				lun->pr_key_count--;
8296
8297				if (!persis_offset
8298				 && i < persis_offset)
8299					lun->pending_ua[i] |=
8300						CTL_UA_REG_PREEMPT;
8301				else if (persis_offset
8302				      && i >= persis_offset)
8303					lun->pending_ua[i - persis_offset] |=
8304						CTL_UA_REG_PREEMPT;
8305			}
8306		}
8307	} else {
8308		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8309			if (i == msg->pr.pr_info.residx ||
8310			    lun->pr_keys[i] == 0)
8311				continue;
8312
8313			if (sa_res_key == lun->pr_keys[i]) {
8314				lun->pr_keys[i] = 0;
8315				lun->pr_key_count--;
8316				if (!persis_offset
8317				 && i < CTL_MAX_INITIATORS)
8318					lun->pending_ua[i] |=
8319						CTL_UA_REG_PREEMPT;
8320				else if (persis_offset
8321				      && i >= persis_offset)
8322					lun->pending_ua[i - persis_offset] |=
8323						CTL_UA_REG_PREEMPT;
8324			} else if (msg->pr.pr_info.res_type != lun->res_type
8325				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8326				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8327					if (!persis_offset
8328					 && i < persis_offset)
8329						lun->pending_ua[i] |=
8330							CTL_UA_RES_RELEASE;
8331					else if (persis_offset
8332					      && i >= persis_offset)
8333					lun->pending_ua[i - persis_offset] |=
8334						CTL_UA_RES_RELEASE;
8335			}
8336		}
8337		lun->res_type = msg->pr.pr_info.res_type;
8338		if (lun->res_type != SPR_TYPE_WR_EX_AR
8339		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8340			lun->pr_res_idx = msg->pr.pr_info.residx;
8341		else
8342			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8343	}
8344	lun->PRGeneration++;
8345
8346}
8347
8348
8349int
8350ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8351{
8352	int retval;
8353	int isc_retval;
8354	u_int32_t param_len;
8355	struct scsi_per_res_out *cdb;
8356	struct ctl_lun *lun;
8357	struct scsi_per_res_out_parms* param;
8358	struct ctl_softc *softc;
8359	uint32_t residx;
8360	uint64_t res_key, sa_res_key;
8361	uint8_t type;
8362	union ctl_ha_msg persis_io;
8363	int    i;
8364
8365	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8366
8367	retval = CTL_RETVAL_COMPLETE;
8368
8369	softc = control_softc;
8370
8371	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8372	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8373
8374	/*
8375	 * We only support whole-LUN scope.  The scope & type are ignored for
8376	 * register, register and ignore existing key and clear.
8377	 * We sometimes ignore scope and type on preempts too!!
8378	 * Verify reservation type here as well.
8379	 */
8380	type = cdb->scope_type & SPR_TYPE_MASK;
8381	if ((cdb->action == SPRO_RESERVE)
8382	 || (cdb->action == SPRO_RELEASE)) {
8383		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8384			ctl_set_invalid_field(/*ctsio*/ ctsio,
8385					      /*sks_valid*/ 1,
8386					      /*command*/ 1,
8387					      /*field*/ 2,
8388					      /*bit_valid*/ 1,
8389					      /*bit*/ 4);
8390			ctl_done((union ctl_io *)ctsio);
8391			return (CTL_RETVAL_COMPLETE);
8392		}
8393
8394		if (type>8 || type==2 || type==4 || type==0) {
8395			ctl_set_invalid_field(/*ctsio*/ ctsio,
8396					      /*sks_valid*/ 1,
8397					      /*command*/ 1,
8398					      /*field*/ 2,
8399					      /*bit_valid*/ 1,
8400					      /*bit*/ 0);
8401			ctl_done((union ctl_io *)ctsio);
8402			return (CTL_RETVAL_COMPLETE);
8403		}
8404	}
8405
8406	param_len = scsi_4btoul(cdb->length);
8407
8408	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8409		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8410		ctsio->kern_data_len = param_len;
8411		ctsio->kern_total_len = param_len;
8412		ctsio->kern_data_resid = 0;
8413		ctsio->kern_rel_offset = 0;
8414		ctsio->kern_sg_entries = 0;
8415		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8416		ctsio->be_move_done = ctl_config_move_done;
8417		ctl_datamove((union ctl_io *)ctsio);
8418
8419		return (CTL_RETVAL_COMPLETE);
8420	}
8421
8422	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8423
8424	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8425	res_key = scsi_8btou64(param->res_key.key);
8426	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8427
8428	/*
8429	 * Validate the reservation key here except for SPRO_REG_IGNO
8430	 * This must be done for all other service actions
8431	 */
8432	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8433		mtx_lock(&lun->lun_lock);
8434		if (lun->pr_keys[residx] != 0) {
8435		    if (res_key != lun->pr_keys[residx]) {
8436				/*
8437				 * The current key passed in doesn't match
8438				 * the one the initiator previously
8439				 * registered.
8440				 */
8441				mtx_unlock(&lun->lun_lock);
8442				free(ctsio->kern_data_ptr, M_CTL);
8443				ctl_set_reservation_conflict(ctsio);
8444				ctl_done((union ctl_io *)ctsio);
8445				return (CTL_RETVAL_COMPLETE);
8446			}
8447		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8448			/*
8449			 * We are not registered
8450			 */
8451			mtx_unlock(&lun->lun_lock);
8452			free(ctsio->kern_data_ptr, M_CTL);
8453			ctl_set_reservation_conflict(ctsio);
8454			ctl_done((union ctl_io *)ctsio);
8455			return (CTL_RETVAL_COMPLETE);
8456		} else if (res_key != 0) {
8457			/*
8458			 * We are not registered and trying to register but
8459			 * the register key isn't zero.
8460			 */
8461			mtx_unlock(&lun->lun_lock);
8462			free(ctsio->kern_data_ptr, M_CTL);
8463			ctl_set_reservation_conflict(ctsio);
8464			ctl_done((union ctl_io *)ctsio);
8465			return (CTL_RETVAL_COMPLETE);
8466		}
8467		mtx_unlock(&lun->lun_lock);
8468	}
8469
8470	switch (cdb->action & SPRO_ACTION_MASK) {
8471	case SPRO_REGISTER:
8472	case SPRO_REG_IGNO: {
8473
8474#if 0
8475		printf("Registration received\n");
8476#endif
8477
8478		/*
8479		 * We don't support any of these options, as we report in
8480		 * the read capabilities request (see
8481		 * ctl_persistent_reserve_in(), above).
8482		 */
8483		if ((param->flags & SPR_SPEC_I_PT)
8484		 || (param->flags & SPR_ALL_TG_PT)
8485		 || (param->flags & SPR_APTPL)) {
8486			int bit_ptr;
8487
8488			if (param->flags & SPR_APTPL)
8489				bit_ptr = 0;
8490			else if (param->flags & SPR_ALL_TG_PT)
8491				bit_ptr = 2;
8492			else /* SPR_SPEC_I_PT */
8493				bit_ptr = 3;
8494
8495			free(ctsio->kern_data_ptr, M_CTL);
8496			ctl_set_invalid_field(ctsio,
8497					      /*sks_valid*/ 1,
8498					      /*command*/ 0,
8499					      /*field*/ 20,
8500					      /*bit_valid*/ 1,
8501					      /*bit*/ bit_ptr);
8502			ctl_done((union ctl_io *)ctsio);
8503			return (CTL_RETVAL_COMPLETE);
8504		}
8505
8506		mtx_lock(&lun->lun_lock);
8507
8508		/*
8509		 * The initiator wants to clear the
8510		 * key/unregister.
8511		 */
8512		if (sa_res_key == 0) {
8513			if ((res_key == 0
8514			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8515			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8516			  && lun->pr_keys[residx] == 0)) {
8517				mtx_unlock(&lun->lun_lock);
8518				goto done;
8519			}
8520
8521			lun->pr_keys[residx] = 0;
8522			lun->pr_key_count--;
8523
8524			if (residx == lun->pr_res_idx) {
8525				lun->flags &= ~CTL_LUN_PR_RESERVED;
8526				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8527
8528				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8529				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8530				 && lun->pr_key_count) {
8531					/*
8532					 * If the reservation is a registrants
8533					 * only type we need to generate a UA
8534					 * for other registered inits.  The
8535					 * sense code should be RESERVATIONS
8536					 * RELEASED
8537					 */
8538
8539					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8540						if (lun->pr_keys[
8541						    i + persis_offset] == 0)
8542							continue;
8543						lun->pending_ua[i] |=
8544							CTL_UA_RES_RELEASE;
8545					}
8546				}
8547				lun->res_type = 0;
8548			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8549				if (lun->pr_key_count==0) {
8550					lun->flags &= ~CTL_LUN_PR_RESERVED;
8551					lun->res_type = 0;
8552					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8553				}
8554			}
8555			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8556			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8557			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8558			persis_io.pr.pr_info.residx = residx;
8559			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8560			     &persis_io, sizeof(persis_io), 0 )) >
8561			     CTL_HA_STATUS_SUCCESS) {
8562				printf("CTL:Persis Out error returned from "
8563				       "ctl_ha_msg_send %d\n", isc_retval);
8564			}
8565		} else /* sa_res_key != 0 */ {
8566
8567			/*
8568			 * If we aren't registered currently then increment
8569			 * the key count and set the registered flag.
8570			 */
8571			if (lun->pr_keys[residx] == 0)
8572				lun->pr_key_count++;
8573			lun->pr_keys[residx] = sa_res_key;
8574
8575			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8576			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8577			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8578			persis_io.pr.pr_info.residx = residx;
8579			memcpy(persis_io.pr.pr_info.sa_res_key,
8580			       param->serv_act_res_key,
8581			       sizeof(param->serv_act_res_key));
8582			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8583			     &persis_io, sizeof(persis_io), 0)) >
8584			     CTL_HA_STATUS_SUCCESS) {
8585				printf("CTL:Persis Out error returned from "
8586				       "ctl_ha_msg_send %d\n", isc_retval);
8587			}
8588		}
8589		lun->PRGeneration++;
8590		mtx_unlock(&lun->lun_lock);
8591
8592		break;
8593	}
8594	case SPRO_RESERVE:
8595#if 0
8596                printf("Reserve executed type %d\n", type);
8597#endif
8598		mtx_lock(&lun->lun_lock);
8599		if (lun->flags & CTL_LUN_PR_RESERVED) {
8600			/*
8601			 * if this isn't the reservation holder and it's
8602			 * not a "all registrants" type or if the type is
8603			 * different then we have a conflict
8604			 */
8605			if ((lun->pr_res_idx != residx
8606			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8607			 || lun->res_type != type) {
8608				mtx_unlock(&lun->lun_lock);
8609				free(ctsio->kern_data_ptr, M_CTL);
8610				ctl_set_reservation_conflict(ctsio);
8611				ctl_done((union ctl_io *)ctsio);
8612				return (CTL_RETVAL_COMPLETE);
8613			}
8614			mtx_unlock(&lun->lun_lock);
8615		} else /* create a reservation */ {
8616			/*
8617			 * If it's not an "all registrants" type record
8618			 * reservation holder
8619			 */
8620			if (type != SPR_TYPE_WR_EX_AR
8621			 && type != SPR_TYPE_EX_AC_AR)
8622				lun->pr_res_idx = residx; /* Res holder */
8623			else
8624				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8625
8626			lun->flags |= CTL_LUN_PR_RESERVED;
8627			lun->res_type = type;
8628
8629			mtx_unlock(&lun->lun_lock);
8630
8631			/* send msg to other side */
8632			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8633			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8634			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8635			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8636			persis_io.pr.pr_info.res_type = type;
8637			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8638			     &persis_io, sizeof(persis_io), 0)) >
8639			     CTL_HA_STATUS_SUCCESS) {
8640				printf("CTL:Persis Out error returned from "
8641				       "ctl_ha_msg_send %d\n", isc_retval);
8642			}
8643		}
8644		break;
8645
8646	case SPRO_RELEASE:
8647		mtx_lock(&lun->lun_lock);
8648		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8649			/* No reservation exists return good status */
8650			mtx_unlock(&lun->lun_lock);
8651			goto done;
8652		}
8653		/*
8654		 * Is this nexus a reservation holder?
8655		 */
8656		if (lun->pr_res_idx != residx
8657		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8658			/*
8659			 * not a res holder return good status but
8660			 * do nothing
8661			 */
8662			mtx_unlock(&lun->lun_lock);
8663			goto done;
8664		}
8665
8666		if (lun->res_type != type) {
8667			mtx_unlock(&lun->lun_lock);
8668			free(ctsio->kern_data_ptr, M_CTL);
8669			ctl_set_illegal_pr_release(ctsio);
8670			ctl_done((union ctl_io *)ctsio);
8671			return (CTL_RETVAL_COMPLETE);
8672		}
8673
8674		/* okay to release */
8675		lun->flags &= ~CTL_LUN_PR_RESERVED;
8676		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8677		lun->res_type = 0;
8678
8679		/*
8680		 * if this isn't an exclusive access
8681		 * res generate UA for all other
8682		 * registrants.
8683		 */
8684		if (type != SPR_TYPE_EX_AC
8685		 && type != SPR_TYPE_WR_EX) {
8686			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8687				if (i == residx ||
8688				    lun->pr_keys[i + persis_offset] == 0)
8689					continue;
8690				lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8691			}
8692		}
8693		mtx_unlock(&lun->lun_lock);
8694		/* Send msg to other side */
8695		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8696		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8697		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8698		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8699		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8700			printf("CTL:Persis Out error returned from "
8701			       "ctl_ha_msg_send %d\n", isc_retval);
8702		}
8703		break;
8704
8705	case SPRO_CLEAR:
8706		/* send msg to other side */
8707
8708		mtx_lock(&lun->lun_lock);
8709		lun->flags &= ~CTL_LUN_PR_RESERVED;
8710		lun->res_type = 0;
8711		lun->pr_key_count = 0;
8712		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8713
8714		lun->pr_keys[residx] = 0;
8715
8716		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8717			if (lun->pr_keys[i] != 0) {
8718				if (!persis_offset && i < CTL_MAX_INITIATORS)
8719					lun->pending_ua[i] |=
8720						CTL_UA_RES_PREEMPT;
8721				else if (persis_offset && i >= persis_offset)
8722					lun->pending_ua[i-persis_offset] |=
8723					    CTL_UA_RES_PREEMPT;
8724
8725				lun->pr_keys[i] = 0;
8726			}
8727		lun->PRGeneration++;
8728		mtx_unlock(&lun->lun_lock);
8729		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8730		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8731		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8732		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8733		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8734			printf("CTL:Persis Out error returned from "
8735			       "ctl_ha_msg_send %d\n", isc_retval);
8736		}
8737		break;
8738
8739	case SPRO_PREEMPT: {
8740		int nretval;
8741
8742		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8743					  residx, ctsio, cdb, param);
8744		if (nretval != 0)
8745			return (CTL_RETVAL_COMPLETE);
8746		break;
8747	}
8748	default:
8749		panic("Invalid PR type %x", cdb->action);
8750	}
8751
8752done:
8753	free(ctsio->kern_data_ptr, M_CTL);
8754	ctl_set_success(ctsio);
8755	ctl_done((union ctl_io *)ctsio);
8756
8757	return (retval);
8758}
8759
8760/*
8761 * This routine is for handling a message from the other SC pertaining to
8762 * persistent reserve out. All the error checking will have been done
8763 * so only perorming the action need be done here to keep the two
8764 * in sync.
8765 */
8766static void
8767ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8768{
8769	struct ctl_lun *lun;
8770	struct ctl_softc *softc;
8771	int i;
8772	uint32_t targ_lun;
8773
8774	softc = control_softc;
8775
8776	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8777	lun = softc->ctl_luns[targ_lun];
8778	mtx_lock(&lun->lun_lock);
8779	switch(msg->pr.pr_info.action) {
8780	case CTL_PR_REG_KEY:
8781		if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8782			lun->pr_key_count++;
8783		lun->pr_keys[msg->pr.pr_info.residx] =
8784		    scsi_8btou64(msg->pr.pr_info.sa_res_key);
8785		lun->PRGeneration++;
8786		break;
8787
8788	case CTL_PR_UNREG_KEY:
8789		lun->pr_keys[msg->pr.pr_info.residx] = 0;
8790		lun->pr_key_count--;
8791
8792		/* XXX Need to see if the reservation has been released */
8793		/* if so do we need to generate UA? */
8794		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8795			lun->flags &= ~CTL_LUN_PR_RESERVED;
8796			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8797
8798			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8799			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8800			 && lun->pr_key_count) {
8801				/*
8802				 * If the reservation is a registrants
8803				 * only type we need to generate a UA
8804				 * for other registered inits.  The
8805				 * sense code should be RESERVATIONS
8806				 * RELEASED
8807				 */
8808
8809				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8810					if (lun->pr_keys[i+
8811					    persis_offset] == 0)
8812						continue;
8813
8814					lun->pending_ua[i] |=
8815						CTL_UA_RES_RELEASE;
8816				}
8817			}
8818			lun->res_type = 0;
8819		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8820			if (lun->pr_key_count==0) {
8821				lun->flags &= ~CTL_LUN_PR_RESERVED;
8822				lun->res_type = 0;
8823				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8824			}
8825		}
8826		lun->PRGeneration++;
8827		break;
8828
8829	case CTL_PR_RESERVE:
8830		lun->flags |= CTL_LUN_PR_RESERVED;
8831		lun->res_type = msg->pr.pr_info.res_type;
8832		lun->pr_res_idx = msg->pr.pr_info.residx;
8833
8834		break;
8835
8836	case CTL_PR_RELEASE:
8837		/*
8838		 * if this isn't an exclusive access res generate UA for all
8839		 * other registrants.
8840		 */
8841		if (lun->res_type != SPR_TYPE_EX_AC
8842		 && lun->res_type != SPR_TYPE_WR_EX) {
8843			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8844				if (lun->pr_keys[i+persis_offset] != 0)
8845					lun->pending_ua[i] |=
8846						CTL_UA_RES_RELEASE;
8847		}
8848
8849		lun->flags &= ~CTL_LUN_PR_RESERVED;
8850		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8851		lun->res_type = 0;
8852		break;
8853
8854	case CTL_PR_PREEMPT:
8855		ctl_pro_preempt_other(lun, msg);
8856		break;
8857	case CTL_PR_CLEAR:
8858		lun->flags &= ~CTL_LUN_PR_RESERVED;
8859		lun->res_type = 0;
8860		lun->pr_key_count = 0;
8861		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8862
8863		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8864			if (lun->pr_keys[i] == 0)
8865				continue;
8866			if (!persis_offset
8867			 && i < CTL_MAX_INITIATORS)
8868				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8869			else if (persis_offset
8870			      && i >= persis_offset)
8871				lun->pending_ua[i-persis_offset] |=
8872					CTL_UA_RES_PREEMPT;
8873			lun->pr_keys[i] = 0;
8874		}
8875		lun->PRGeneration++;
8876		break;
8877	}
8878
8879	mtx_unlock(&lun->lun_lock);
8880}
8881
8882int
8883ctl_read_write(struct ctl_scsiio *ctsio)
8884{
8885	struct ctl_lun *lun;
8886	struct ctl_lba_len_flags *lbalen;
8887	uint64_t lba;
8888	uint32_t num_blocks;
8889	int flags, retval;
8890	int isread;
8891
8892	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8893
8894	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8895
8896	flags = 0;
8897	retval = CTL_RETVAL_COMPLETE;
8898
8899	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8900	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8901	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8902		uint32_t residx;
8903
8904		/*
8905		 * XXX KDM need a lock here.
8906		 */
8907		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8908		if ((lun->res_type == SPR_TYPE_EX_AC
8909		  && residx != lun->pr_res_idx)
8910		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8911		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8912		  && lun->pr_keys[residx] == 0)) {
8913			ctl_set_reservation_conflict(ctsio);
8914			ctl_done((union ctl_io *)ctsio);
8915			return (CTL_RETVAL_COMPLETE);
8916	        }
8917	}
8918
8919	switch (ctsio->cdb[0]) {
8920	case READ_6:
8921	case WRITE_6: {
8922		struct scsi_rw_6 *cdb;
8923
8924		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8925
8926		lba = scsi_3btoul(cdb->addr);
8927		/* only 5 bits are valid in the most significant address byte */
8928		lba &= 0x1fffff;
8929		num_blocks = cdb->length;
8930		/*
8931		 * This is correct according to SBC-2.
8932		 */
8933		if (num_blocks == 0)
8934			num_blocks = 256;
8935		break;
8936	}
8937	case READ_10:
8938	case WRITE_10: {
8939		struct scsi_rw_10 *cdb;
8940
8941		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8942		if (cdb->byte2 & SRW10_FUA)
8943			flags |= CTL_LLF_FUA;
8944		if (cdb->byte2 & SRW10_DPO)
8945			flags |= CTL_LLF_DPO;
8946		lba = scsi_4btoul(cdb->addr);
8947		num_blocks = scsi_2btoul(cdb->length);
8948		break;
8949	}
8950	case WRITE_VERIFY_10: {
8951		struct scsi_write_verify_10 *cdb;
8952
8953		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8954		flags |= CTL_LLF_FUA;
8955		if (cdb->byte2 & SWV_DPO)
8956			flags |= CTL_LLF_DPO;
8957		lba = scsi_4btoul(cdb->addr);
8958		num_blocks = scsi_2btoul(cdb->length);
8959		break;
8960	}
8961	case READ_12:
8962	case WRITE_12: {
8963		struct scsi_rw_12 *cdb;
8964
8965		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8966		if (cdb->byte2 & SRW12_FUA)
8967			flags |= CTL_LLF_FUA;
8968		if (cdb->byte2 & SRW12_DPO)
8969			flags |= CTL_LLF_DPO;
8970		lba = scsi_4btoul(cdb->addr);
8971		num_blocks = scsi_4btoul(cdb->length);
8972		break;
8973	}
8974	case WRITE_VERIFY_12: {
8975		struct scsi_write_verify_12 *cdb;
8976
8977		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8978		flags |= CTL_LLF_FUA;
8979		if (cdb->byte2 & SWV_DPO)
8980			flags |= CTL_LLF_DPO;
8981		lba = scsi_4btoul(cdb->addr);
8982		num_blocks = scsi_4btoul(cdb->length);
8983		break;
8984	}
8985	case READ_16:
8986	case WRITE_16: {
8987		struct scsi_rw_16 *cdb;
8988
8989		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8990		if (cdb->byte2 & SRW12_FUA)
8991			flags |= CTL_LLF_FUA;
8992		if (cdb->byte2 & SRW12_DPO)
8993			flags |= CTL_LLF_DPO;
8994		lba = scsi_8btou64(cdb->addr);
8995		num_blocks = scsi_4btoul(cdb->length);
8996		break;
8997	}
8998	case WRITE_ATOMIC_16: {
8999		struct scsi_rw_16 *cdb;
9000
9001		if (lun->be_lun->atomicblock == 0) {
9002			ctl_set_invalid_opcode(ctsio);
9003			ctl_done((union ctl_io *)ctsio);
9004			return (CTL_RETVAL_COMPLETE);
9005		}
9006
9007		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9008		if (cdb->byte2 & SRW12_FUA)
9009			flags |= CTL_LLF_FUA;
9010		if (cdb->byte2 & SRW12_DPO)
9011			flags |= CTL_LLF_DPO;
9012		lba = scsi_8btou64(cdb->addr);
9013		num_blocks = scsi_4btoul(cdb->length);
9014		if (num_blocks > lun->be_lun->atomicblock) {
9015			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9016			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9017			    /*bit*/ 0);
9018			ctl_done((union ctl_io *)ctsio);
9019			return (CTL_RETVAL_COMPLETE);
9020		}
9021		break;
9022	}
9023	case WRITE_VERIFY_16: {
9024		struct scsi_write_verify_16 *cdb;
9025
9026		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9027		flags |= CTL_LLF_FUA;
9028		if (cdb->byte2 & SWV_DPO)
9029			flags |= CTL_LLF_DPO;
9030		lba = scsi_8btou64(cdb->addr);
9031		num_blocks = scsi_4btoul(cdb->length);
9032		break;
9033	}
9034	default:
9035		/*
9036		 * We got a command we don't support.  This shouldn't
9037		 * happen, commands should be filtered out above us.
9038		 */
9039		ctl_set_invalid_opcode(ctsio);
9040		ctl_done((union ctl_io *)ctsio);
9041
9042		return (CTL_RETVAL_COMPLETE);
9043		break; /* NOTREACHED */
9044	}
9045
9046	/*
9047	 * The first check is to make sure we're in bounds, the second
9048	 * check is to catch wrap-around problems.  If the lba + num blocks
9049	 * is less than the lba, then we've wrapped around and the block
9050	 * range is invalid anyway.
9051	 */
9052	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9053	 || ((lba + num_blocks) < lba)) {
9054		ctl_set_lba_out_of_range(ctsio);
9055		ctl_done((union ctl_io *)ctsio);
9056		return (CTL_RETVAL_COMPLETE);
9057	}
9058
9059	/*
9060	 * According to SBC-3, a transfer length of 0 is not an error.
9061	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9062	 * translates to 256 blocks for those commands.
9063	 */
9064	if (num_blocks == 0) {
9065		ctl_set_success(ctsio);
9066		ctl_done((union ctl_io *)ctsio);
9067		return (CTL_RETVAL_COMPLETE);
9068	}
9069
9070	/* Set FUA and/or DPO if caches are disabled. */
9071	if (isread) {
9072		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9073		    SCP_RCD) != 0)
9074			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9075	} else {
9076		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9077		    SCP_WCE) == 0)
9078			flags |= CTL_LLF_FUA;
9079	}
9080
9081	lbalen = (struct ctl_lba_len_flags *)
9082	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9083	lbalen->lba = lba;
9084	lbalen->len = num_blocks;
9085	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9086
9087	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9088	ctsio->kern_rel_offset = 0;
9089
9090	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9091
9092	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9093
9094	return (retval);
9095}
9096
9097static int
9098ctl_cnw_cont(union ctl_io *io)
9099{
9100	struct ctl_scsiio *ctsio;
9101	struct ctl_lun *lun;
9102	struct ctl_lba_len_flags *lbalen;
9103	int retval;
9104
9105	ctsio = &io->scsiio;
9106	ctsio->io_hdr.status = CTL_STATUS_NONE;
9107	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9108	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9109	lbalen = (struct ctl_lba_len_flags *)
9110	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9111	lbalen->flags &= ~CTL_LLF_COMPARE;
9112	lbalen->flags |= CTL_LLF_WRITE;
9113
9114	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9115	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9116	return (retval);
9117}
9118
9119int
9120ctl_cnw(struct ctl_scsiio *ctsio)
9121{
9122	struct ctl_lun *lun;
9123	struct ctl_lba_len_flags *lbalen;
9124	uint64_t lba;
9125	uint32_t num_blocks;
9126	int flags, retval;
9127
9128	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9129
9130	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9131
9132	flags = 0;
9133	retval = CTL_RETVAL_COMPLETE;
9134
9135	switch (ctsio->cdb[0]) {
9136	case COMPARE_AND_WRITE: {
9137		struct scsi_compare_and_write *cdb;
9138
9139		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9140		if (cdb->byte2 & SRW10_FUA)
9141			flags |= CTL_LLF_FUA;
9142		if (cdb->byte2 & SRW10_DPO)
9143			flags |= CTL_LLF_DPO;
9144		lba = scsi_8btou64(cdb->addr);
9145		num_blocks = cdb->length;
9146		break;
9147	}
9148	default:
9149		/*
9150		 * We got a command we don't support.  This shouldn't
9151		 * happen, commands should be filtered out above us.
9152		 */
9153		ctl_set_invalid_opcode(ctsio);
9154		ctl_done((union ctl_io *)ctsio);
9155
9156		return (CTL_RETVAL_COMPLETE);
9157		break; /* NOTREACHED */
9158	}
9159
9160	/*
9161	 * The first check is to make sure we're in bounds, the second
9162	 * check is to catch wrap-around problems.  If the lba + num blocks
9163	 * is less than the lba, then we've wrapped around and the block
9164	 * range is invalid anyway.
9165	 */
9166	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9167	 || ((lba + num_blocks) < lba)) {
9168		ctl_set_lba_out_of_range(ctsio);
9169		ctl_done((union ctl_io *)ctsio);
9170		return (CTL_RETVAL_COMPLETE);
9171	}
9172
9173	/*
9174	 * According to SBC-3, a transfer length of 0 is not an error.
9175	 */
9176	if (num_blocks == 0) {
9177		ctl_set_success(ctsio);
9178		ctl_done((union ctl_io *)ctsio);
9179		return (CTL_RETVAL_COMPLETE);
9180	}
9181
9182	/* Set FUA if write cache is disabled. */
9183	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9184	    SCP_WCE) == 0)
9185		flags |= CTL_LLF_FUA;
9186
9187	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9188	ctsio->kern_rel_offset = 0;
9189
9190	/*
9191	 * Set the IO_CONT flag, so that if this I/O gets passed to
9192	 * ctl_data_submit_done(), it'll get passed back to
9193	 * ctl_ctl_cnw_cont() for further processing.
9194	 */
9195	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9196	ctsio->io_cont = ctl_cnw_cont;
9197
9198	lbalen = (struct ctl_lba_len_flags *)
9199	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9200	lbalen->lba = lba;
9201	lbalen->len = num_blocks;
9202	lbalen->flags = CTL_LLF_COMPARE | flags;
9203
9204	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9205	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9206	return (retval);
9207}
9208
9209int
9210ctl_verify(struct ctl_scsiio *ctsio)
9211{
9212	struct ctl_lun *lun;
9213	struct ctl_lba_len_flags *lbalen;
9214	uint64_t lba;
9215	uint32_t num_blocks;
9216	int bytchk, flags;
9217	int retval;
9218
9219	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9220
9221	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9222
9223	bytchk = 0;
9224	flags = CTL_LLF_FUA;
9225	retval = CTL_RETVAL_COMPLETE;
9226
9227	switch (ctsio->cdb[0]) {
9228	case VERIFY_10: {
9229		struct scsi_verify_10 *cdb;
9230
9231		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9232		if (cdb->byte2 & SVFY_BYTCHK)
9233			bytchk = 1;
9234		if (cdb->byte2 & SVFY_DPO)
9235			flags |= CTL_LLF_DPO;
9236		lba = scsi_4btoul(cdb->addr);
9237		num_blocks = scsi_2btoul(cdb->length);
9238		break;
9239	}
9240	case VERIFY_12: {
9241		struct scsi_verify_12 *cdb;
9242
9243		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9244		if (cdb->byte2 & SVFY_BYTCHK)
9245			bytchk = 1;
9246		if (cdb->byte2 & SVFY_DPO)
9247			flags |= CTL_LLF_DPO;
9248		lba = scsi_4btoul(cdb->addr);
9249		num_blocks = scsi_4btoul(cdb->length);
9250		break;
9251	}
9252	case VERIFY_16: {
9253		struct scsi_rw_16 *cdb;
9254
9255		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9256		if (cdb->byte2 & SVFY_BYTCHK)
9257			bytchk = 1;
9258		if (cdb->byte2 & SVFY_DPO)
9259			flags |= CTL_LLF_DPO;
9260		lba = scsi_8btou64(cdb->addr);
9261		num_blocks = scsi_4btoul(cdb->length);
9262		break;
9263	}
9264	default:
9265		/*
9266		 * We got a command we don't support.  This shouldn't
9267		 * happen, commands should be filtered out above us.
9268		 */
9269		ctl_set_invalid_opcode(ctsio);
9270		ctl_done((union ctl_io *)ctsio);
9271		return (CTL_RETVAL_COMPLETE);
9272	}
9273
9274	/*
9275	 * The first check is to make sure we're in bounds, the second
9276	 * check is to catch wrap-around problems.  If the lba + num blocks
9277	 * is less than the lba, then we've wrapped around and the block
9278	 * range is invalid anyway.
9279	 */
9280	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9281	 || ((lba + num_blocks) < lba)) {
9282		ctl_set_lba_out_of_range(ctsio);
9283		ctl_done((union ctl_io *)ctsio);
9284		return (CTL_RETVAL_COMPLETE);
9285	}
9286
9287	/*
9288	 * According to SBC-3, a transfer length of 0 is not an error.
9289	 */
9290	if (num_blocks == 0) {
9291		ctl_set_success(ctsio);
9292		ctl_done((union ctl_io *)ctsio);
9293		return (CTL_RETVAL_COMPLETE);
9294	}
9295
9296	lbalen = (struct ctl_lba_len_flags *)
9297	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9298	lbalen->lba = lba;
9299	lbalen->len = num_blocks;
9300	if (bytchk) {
9301		lbalen->flags = CTL_LLF_COMPARE | flags;
9302		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9303	} else {
9304		lbalen->flags = CTL_LLF_VERIFY | flags;
9305		ctsio->kern_total_len = 0;
9306	}
9307	ctsio->kern_rel_offset = 0;
9308
9309	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9310	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9311	return (retval);
9312}
9313
9314int
9315ctl_report_luns(struct ctl_scsiio *ctsio)
9316{
9317	struct scsi_report_luns *cdb;
9318	struct scsi_report_luns_data *lun_data;
9319	struct ctl_lun *lun, *request_lun;
9320	int num_luns, retval;
9321	uint32_t alloc_len, lun_datalen;
9322	int num_filled, well_known;
9323	uint32_t initidx, targ_lun_id, lun_id;
9324
9325	retval = CTL_RETVAL_COMPLETE;
9326	well_known = 0;
9327
9328	cdb = (struct scsi_report_luns *)ctsio->cdb;
9329
9330	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9331
9332	mtx_lock(&control_softc->ctl_lock);
9333	num_luns = control_softc->num_luns;
9334	mtx_unlock(&control_softc->ctl_lock);
9335
9336	switch (cdb->select_report) {
9337	case RPL_REPORT_DEFAULT:
9338	case RPL_REPORT_ALL:
9339		break;
9340	case RPL_REPORT_WELLKNOWN:
9341		well_known = 1;
9342		num_luns = 0;
9343		break;
9344	default:
9345		ctl_set_invalid_field(ctsio,
9346				      /*sks_valid*/ 1,
9347				      /*command*/ 1,
9348				      /*field*/ 2,
9349				      /*bit_valid*/ 0,
9350				      /*bit*/ 0);
9351		ctl_done((union ctl_io *)ctsio);
9352		return (retval);
9353		break; /* NOTREACHED */
9354	}
9355
9356	alloc_len = scsi_4btoul(cdb->length);
9357	/*
9358	 * The initiator has to allocate at least 16 bytes for this request,
9359	 * so he can at least get the header and the first LUN.  Otherwise
9360	 * we reject the request (per SPC-3 rev 14, section 6.21).
9361	 */
9362	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9363	    sizeof(struct scsi_report_luns_lundata))) {
9364		ctl_set_invalid_field(ctsio,
9365				      /*sks_valid*/ 1,
9366				      /*command*/ 1,
9367				      /*field*/ 6,
9368				      /*bit_valid*/ 0,
9369				      /*bit*/ 0);
9370		ctl_done((union ctl_io *)ctsio);
9371		return (retval);
9372	}
9373
9374	request_lun = (struct ctl_lun *)
9375		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9376
9377	lun_datalen = sizeof(*lun_data) +
9378		(num_luns * sizeof(struct scsi_report_luns_lundata));
9379
9380	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9381	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9382	ctsio->kern_sg_entries = 0;
9383
9384	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9385
9386	mtx_lock(&control_softc->ctl_lock);
9387	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9388		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9389		if (lun_id >= CTL_MAX_LUNS)
9390			continue;
9391		lun = control_softc->ctl_luns[lun_id];
9392		if (lun == NULL)
9393			continue;
9394
9395		if (targ_lun_id <= 0xff) {
9396			/*
9397			 * Peripheral addressing method, bus number 0.
9398			 */
9399			lun_data->luns[num_filled].lundata[0] =
9400				RPL_LUNDATA_ATYP_PERIPH;
9401			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9402			num_filled++;
9403		} else if (targ_lun_id <= 0x3fff) {
9404			/*
9405			 * Flat addressing method.
9406			 */
9407			lun_data->luns[num_filled].lundata[0] =
9408				RPL_LUNDATA_ATYP_FLAT |
9409				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9410#ifdef OLDCTLHEADERS
9411				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9412				(targ_lun_id & SRLD_BUS_LUN_MASK);
9413#endif
9414			lun_data->luns[num_filled].lundata[1] =
9415#ifdef OLDCTLHEADERS
9416				targ_lun_id >> SRLD_BUS_LUN_BITS;
9417#endif
9418				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9419			num_filled++;
9420		} else {
9421			printf("ctl_report_luns: bogus LUN number %jd, "
9422			       "skipping\n", (intmax_t)targ_lun_id);
9423		}
9424		/*
9425		 * According to SPC-3, rev 14 section 6.21:
9426		 *
9427		 * "The execution of a REPORT LUNS command to any valid and
9428		 * installed logical unit shall clear the REPORTED LUNS DATA
9429		 * HAS CHANGED unit attention condition for all logical
9430		 * units of that target with respect to the requesting
9431		 * initiator. A valid and installed logical unit is one
9432		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9433		 * INQUIRY data (see 6.4.2)."
9434		 *
9435		 * If request_lun is NULL, the LUN this report luns command
9436		 * was issued to is either disabled or doesn't exist. In that
9437		 * case, we shouldn't clear any pending lun change unit
9438		 * attention.
9439		 */
9440		if (request_lun != NULL) {
9441			mtx_lock(&lun->lun_lock);
9442			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9443			mtx_unlock(&lun->lun_lock);
9444		}
9445	}
9446	mtx_unlock(&control_softc->ctl_lock);
9447
9448	/*
9449	 * It's quite possible that we've returned fewer LUNs than we allocated
9450	 * space for.  Trim it.
9451	 */
9452	lun_datalen = sizeof(*lun_data) +
9453		(num_filled * sizeof(struct scsi_report_luns_lundata));
9454
9455	if (lun_datalen < alloc_len) {
9456		ctsio->residual = alloc_len - lun_datalen;
9457		ctsio->kern_data_len = lun_datalen;
9458		ctsio->kern_total_len = lun_datalen;
9459	} else {
9460		ctsio->residual = 0;
9461		ctsio->kern_data_len = alloc_len;
9462		ctsio->kern_total_len = alloc_len;
9463	}
9464	ctsio->kern_data_resid = 0;
9465	ctsio->kern_rel_offset = 0;
9466	ctsio->kern_sg_entries = 0;
9467
9468	/*
9469	 * We set this to the actual data length, regardless of how much
9470	 * space we actually have to return results.  If the user looks at
9471	 * this value, he'll know whether or not he allocated enough space
9472	 * and reissue the command if necessary.  We don't support well
9473	 * known logical units, so if the user asks for that, return none.
9474	 */
9475	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9476
9477	/*
9478	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9479	 * this request.
9480	 */
9481	ctsio->scsi_status = SCSI_STATUS_OK;
9482
9483	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9484	ctsio->be_move_done = ctl_config_move_done;
9485	ctl_datamove((union ctl_io *)ctsio);
9486
9487	return (retval);
9488}
9489
9490int
9491ctl_request_sense(struct ctl_scsiio *ctsio)
9492{
9493	struct scsi_request_sense *cdb;
9494	struct scsi_sense_data *sense_ptr;
9495	struct ctl_lun *lun;
9496	uint32_t initidx;
9497	int have_error;
9498	scsi_sense_data_type sense_format;
9499
9500	cdb = (struct scsi_request_sense *)ctsio->cdb;
9501
9502	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9503
9504	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9505
9506	/*
9507	 * Determine which sense format the user wants.
9508	 */
9509	if (cdb->byte2 & SRS_DESC)
9510		sense_format = SSD_TYPE_DESC;
9511	else
9512		sense_format = SSD_TYPE_FIXED;
9513
9514	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9515	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9516	ctsio->kern_sg_entries = 0;
9517
9518	/*
9519	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9520	 * larger than the largest allowed value for the length field in the
9521	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9522	 */
9523	ctsio->residual = 0;
9524	ctsio->kern_data_len = cdb->length;
9525	ctsio->kern_total_len = cdb->length;
9526
9527	ctsio->kern_data_resid = 0;
9528	ctsio->kern_rel_offset = 0;
9529	ctsio->kern_sg_entries = 0;
9530
9531	/*
9532	 * If we don't have a LUN, we don't have any pending sense.
9533	 */
9534	if (lun == NULL)
9535		goto no_sense;
9536
9537	have_error = 0;
9538	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9539	/*
9540	 * Check for pending sense, and then for pending unit attentions.
9541	 * Pending sense gets returned first, then pending unit attentions.
9542	 */
9543	mtx_lock(&lun->lun_lock);
9544#ifdef CTL_WITH_CA
9545	if (ctl_is_set(lun->have_ca, initidx)) {
9546		scsi_sense_data_type stored_format;
9547
9548		/*
9549		 * Check to see which sense format was used for the stored
9550		 * sense data.
9551		 */
9552		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9553
9554		/*
9555		 * If the user requested a different sense format than the
9556		 * one we stored, then we need to convert it to the other
9557		 * format.  If we're going from descriptor to fixed format
9558		 * sense data, we may lose things in translation, depending
9559		 * on what options were used.
9560		 *
9561		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9562		 * for some reason we'll just copy it out as-is.
9563		 */
9564		if ((stored_format == SSD_TYPE_FIXED)
9565		 && (sense_format == SSD_TYPE_DESC))
9566			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9567			    &lun->pending_sense[initidx],
9568			    (struct scsi_sense_data_desc *)sense_ptr);
9569		else if ((stored_format == SSD_TYPE_DESC)
9570		      && (sense_format == SSD_TYPE_FIXED))
9571			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9572			    &lun->pending_sense[initidx],
9573			    (struct scsi_sense_data_fixed *)sense_ptr);
9574		else
9575			memcpy(sense_ptr, &lun->pending_sense[initidx],
9576			       ctl_min(sizeof(*sense_ptr),
9577			       sizeof(lun->pending_sense[initidx])));
9578
9579		ctl_clear_mask(lun->have_ca, initidx);
9580		have_error = 1;
9581	} else
9582#endif
9583	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9584		ctl_ua_type ua_type;
9585
9586		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9587				       sense_ptr, sense_format);
9588		if (ua_type != CTL_UA_NONE)
9589			have_error = 1;
9590	}
9591	mtx_unlock(&lun->lun_lock);
9592
9593	/*
9594	 * We already have a pending error, return it.
9595	 */
9596	if (have_error != 0) {
9597		/*
9598		 * We report the SCSI status as OK, since the status of the
9599		 * request sense command itself is OK.
9600		 */
9601		ctsio->scsi_status = SCSI_STATUS_OK;
9602
9603		/*
9604		 * We report 0 for the sense length, because we aren't doing
9605		 * autosense in this case.  We're reporting sense as
9606		 * parameter data.
9607		 */
9608		ctsio->sense_len = 0;
9609		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9610		ctsio->be_move_done = ctl_config_move_done;
9611		ctl_datamove((union ctl_io *)ctsio);
9612
9613		return (CTL_RETVAL_COMPLETE);
9614	}
9615
9616no_sense:
9617
9618	/*
9619	 * No sense information to report, so we report that everything is
9620	 * okay.
9621	 */
9622	ctl_set_sense_data(sense_ptr,
9623			   lun,
9624			   sense_format,
9625			   /*current_error*/ 1,
9626			   /*sense_key*/ SSD_KEY_NO_SENSE,
9627			   /*asc*/ 0x00,
9628			   /*ascq*/ 0x00,
9629			   SSD_ELEM_NONE);
9630
9631	ctsio->scsi_status = SCSI_STATUS_OK;
9632
9633	/*
9634	 * We report 0 for the sense length, because we aren't doing
9635	 * autosense in this case.  We're reporting sense as parameter data.
9636	 */
9637	ctsio->sense_len = 0;
9638	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9639	ctsio->be_move_done = ctl_config_move_done;
9640	ctl_datamove((union ctl_io *)ctsio);
9641
9642	return (CTL_RETVAL_COMPLETE);
9643}
9644
9645int
9646ctl_tur(struct ctl_scsiio *ctsio)
9647{
9648	struct ctl_lun *lun;
9649
9650	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9651
9652	CTL_DEBUG_PRINT(("ctl_tur\n"));
9653
9654	if (lun == NULL)
9655		return (EINVAL);
9656
9657	ctsio->scsi_status = SCSI_STATUS_OK;
9658	ctsio->io_hdr.status = CTL_SUCCESS;
9659
9660	ctl_done((union ctl_io *)ctsio);
9661
9662	return (CTL_RETVAL_COMPLETE);
9663}
9664
9665#ifdef notyet
9666static int
9667ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9668{
9669
9670}
9671#endif
9672
9673static int
9674ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9675{
9676	struct scsi_vpd_supported_pages *pages;
9677	int sup_page_size;
9678	struct ctl_lun *lun;
9679
9680	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9681
9682	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9683	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9684	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9685	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9686	ctsio->kern_sg_entries = 0;
9687
9688	if (sup_page_size < alloc_len) {
9689		ctsio->residual = alloc_len - sup_page_size;
9690		ctsio->kern_data_len = sup_page_size;
9691		ctsio->kern_total_len = sup_page_size;
9692	} else {
9693		ctsio->residual = 0;
9694		ctsio->kern_data_len = alloc_len;
9695		ctsio->kern_total_len = alloc_len;
9696	}
9697	ctsio->kern_data_resid = 0;
9698	ctsio->kern_rel_offset = 0;
9699	ctsio->kern_sg_entries = 0;
9700
9701	/*
9702	 * The control device is always connected.  The disk device, on the
9703	 * other hand, may not be online all the time.  Need to change this
9704	 * to figure out whether the disk device is actually online or not.
9705	 */
9706	if (lun != NULL)
9707		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9708				lun->be_lun->lun_type;
9709	else
9710		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9711
9712	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9713	/* Supported VPD pages */
9714	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9715	/* Serial Number */
9716	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9717	/* Device Identification */
9718	pages->page_list[2] = SVPD_DEVICE_ID;
9719	/* Extended INQUIRY Data */
9720	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9721	/* Mode Page Policy */
9722	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9723	/* SCSI Ports */
9724	pages->page_list[5] = SVPD_SCSI_PORTS;
9725	/* Third-party Copy */
9726	pages->page_list[6] = SVPD_SCSI_TPC;
9727	/* Block limits */
9728	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9729	/* Block Device Characteristics */
9730	pages->page_list[8] = SVPD_BDC;
9731	/* Logical Block Provisioning */
9732	pages->page_list[9] = SVPD_LBP;
9733
9734	ctsio->scsi_status = SCSI_STATUS_OK;
9735
9736	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9737	ctsio->be_move_done = ctl_config_move_done;
9738	ctl_datamove((union ctl_io *)ctsio);
9739
9740	return (CTL_RETVAL_COMPLETE);
9741}
9742
9743static int
9744ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9745{
9746	struct scsi_vpd_unit_serial_number *sn_ptr;
9747	struct ctl_lun *lun;
9748	int data_len;
9749
9750	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9751
9752	data_len = 4 + CTL_SN_LEN;
9753	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9754	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9755	if (data_len < alloc_len) {
9756		ctsio->residual = alloc_len - data_len;
9757		ctsio->kern_data_len = data_len;
9758		ctsio->kern_total_len = data_len;
9759	} else {
9760		ctsio->residual = 0;
9761		ctsio->kern_data_len = alloc_len;
9762		ctsio->kern_total_len = alloc_len;
9763	}
9764	ctsio->kern_data_resid = 0;
9765	ctsio->kern_rel_offset = 0;
9766	ctsio->kern_sg_entries = 0;
9767
9768	/*
9769	 * The control device is always connected.  The disk device, on the
9770	 * other hand, may not be online all the time.  Need to change this
9771	 * to figure out whether the disk device is actually online or not.
9772	 */
9773	if (lun != NULL)
9774		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9775				  lun->be_lun->lun_type;
9776	else
9777		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9778
9779	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9780	sn_ptr->length = CTL_SN_LEN;
9781	/*
9782	 * If we don't have a LUN, we just leave the serial number as
9783	 * all spaces.
9784	 */
9785	if (lun != NULL) {
9786		strncpy((char *)sn_ptr->serial_num,
9787			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9788	} else
9789		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9790	ctsio->scsi_status = SCSI_STATUS_OK;
9791
9792	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9793	ctsio->be_move_done = ctl_config_move_done;
9794	ctl_datamove((union ctl_io *)ctsio);
9795
9796	return (CTL_RETVAL_COMPLETE);
9797}
9798
9799
9800static int
9801ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9802{
9803	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9804	struct ctl_lun *lun;
9805	int data_len;
9806
9807	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9808
9809	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9810	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9811	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9812	ctsio->kern_sg_entries = 0;
9813
9814	if (data_len < alloc_len) {
9815		ctsio->residual = alloc_len - data_len;
9816		ctsio->kern_data_len = data_len;
9817		ctsio->kern_total_len = data_len;
9818	} else {
9819		ctsio->residual = 0;
9820		ctsio->kern_data_len = alloc_len;
9821		ctsio->kern_total_len = alloc_len;
9822	}
9823	ctsio->kern_data_resid = 0;
9824	ctsio->kern_rel_offset = 0;
9825	ctsio->kern_sg_entries = 0;
9826
9827	/*
9828	 * The control device is always connected.  The disk device, on the
9829	 * other hand, may not be online all the time.
9830	 */
9831	if (lun != NULL)
9832		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9833				     lun->be_lun->lun_type;
9834	else
9835		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9836	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9837	eid_ptr->page_length = data_len - 4;
9838	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9839	eid_ptr->flags3 = SVPD_EID_V_SUP;
9840
9841	ctsio->scsi_status = SCSI_STATUS_OK;
9842	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9843	ctsio->be_move_done = ctl_config_move_done;
9844	ctl_datamove((union ctl_io *)ctsio);
9845
9846	return (CTL_RETVAL_COMPLETE);
9847}
9848
9849static int
9850ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9851{
9852	struct scsi_vpd_mode_page_policy *mpp_ptr;
9853	struct ctl_lun *lun;
9854	int data_len;
9855
9856	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9857
9858	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9859	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9860
9861	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9862	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9863	ctsio->kern_sg_entries = 0;
9864
9865	if (data_len < alloc_len) {
9866		ctsio->residual = alloc_len - data_len;
9867		ctsio->kern_data_len = data_len;
9868		ctsio->kern_total_len = data_len;
9869	} else {
9870		ctsio->residual = 0;
9871		ctsio->kern_data_len = alloc_len;
9872		ctsio->kern_total_len = alloc_len;
9873	}
9874	ctsio->kern_data_resid = 0;
9875	ctsio->kern_rel_offset = 0;
9876	ctsio->kern_sg_entries = 0;
9877
9878	/*
9879	 * The control device is always connected.  The disk device, on the
9880	 * other hand, may not be online all the time.
9881	 */
9882	if (lun != NULL)
9883		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9884				     lun->be_lun->lun_type;
9885	else
9886		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9887	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9888	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9889	mpp_ptr->descr[0].page_code = 0x3f;
9890	mpp_ptr->descr[0].subpage_code = 0xff;
9891	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9892
9893	ctsio->scsi_status = SCSI_STATUS_OK;
9894	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9895	ctsio->be_move_done = ctl_config_move_done;
9896	ctl_datamove((union ctl_io *)ctsio);
9897
9898	return (CTL_RETVAL_COMPLETE);
9899}
9900
9901static int
9902ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9903{
9904	struct scsi_vpd_device_id *devid_ptr;
9905	struct scsi_vpd_id_descriptor *desc;
9906	struct ctl_softc *ctl_softc;
9907	struct ctl_lun *lun;
9908	struct ctl_port *port;
9909	int data_len;
9910	uint8_t proto;
9911
9912	ctl_softc = control_softc;
9913
9914	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9915	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9916
9917	data_len = sizeof(struct scsi_vpd_device_id) +
9918	    sizeof(struct scsi_vpd_id_descriptor) +
9919		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9920	    sizeof(struct scsi_vpd_id_descriptor) +
9921		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9922	if (lun && lun->lun_devid)
9923		data_len += lun->lun_devid->len;
9924	if (port->port_devid)
9925		data_len += port->port_devid->len;
9926	if (port->target_devid)
9927		data_len += port->target_devid->len;
9928
9929	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9930	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9931	ctsio->kern_sg_entries = 0;
9932
9933	if (data_len < alloc_len) {
9934		ctsio->residual = alloc_len - data_len;
9935		ctsio->kern_data_len = data_len;
9936		ctsio->kern_total_len = data_len;
9937	} else {
9938		ctsio->residual = 0;
9939		ctsio->kern_data_len = alloc_len;
9940		ctsio->kern_total_len = alloc_len;
9941	}
9942	ctsio->kern_data_resid = 0;
9943	ctsio->kern_rel_offset = 0;
9944	ctsio->kern_sg_entries = 0;
9945
9946	/*
9947	 * The control device is always connected.  The disk device, on the
9948	 * other hand, may not be online all the time.
9949	 */
9950	if (lun != NULL)
9951		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9952				     lun->be_lun->lun_type;
9953	else
9954		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9955	devid_ptr->page_code = SVPD_DEVICE_ID;
9956	scsi_ulto2b(data_len - 4, devid_ptr->length);
9957
9958	if (port->port_type == CTL_PORT_FC)
9959		proto = SCSI_PROTO_FC << 4;
9960	else if (port->port_type == CTL_PORT_ISCSI)
9961		proto = SCSI_PROTO_ISCSI << 4;
9962	else
9963		proto = SCSI_PROTO_SPI << 4;
9964	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9965
9966	/*
9967	 * We're using a LUN association here.  i.e., this device ID is a
9968	 * per-LUN identifier.
9969	 */
9970	if (lun && lun->lun_devid) {
9971		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9972		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9973		    lun->lun_devid->len);
9974	}
9975
9976	/*
9977	 * This is for the WWPN which is a port association.
9978	 */
9979	if (port->port_devid) {
9980		memcpy(desc, port->port_devid->data, port->port_devid->len);
9981		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9982		    port->port_devid->len);
9983	}
9984
9985	/*
9986	 * This is for the Relative Target Port(type 4h) identifier
9987	 */
9988	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9989	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9990	    SVPD_ID_TYPE_RELTARG;
9991	desc->length = 4;
9992	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9993	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9994	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9995
9996	/*
9997	 * This is for the Target Port Group(type 5h) identifier
9998	 */
9999	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10000	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10001	    SVPD_ID_TYPE_TPORTGRP;
10002	desc->length = 4;
10003	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10004	    &desc->identifier[2]);
10005	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10006	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10007
10008	/*
10009	 * This is for the Target identifier
10010	 */
10011	if (port->target_devid) {
10012		memcpy(desc, port->target_devid->data, port->target_devid->len);
10013	}
10014
10015	ctsio->scsi_status = SCSI_STATUS_OK;
10016	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10017	ctsio->be_move_done = ctl_config_move_done;
10018	ctl_datamove((union ctl_io *)ctsio);
10019
10020	return (CTL_RETVAL_COMPLETE);
10021}
10022
10023static int
10024ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10025{
10026	struct ctl_softc *softc = control_softc;
10027	struct scsi_vpd_scsi_ports *sp;
10028	struct scsi_vpd_port_designation *pd;
10029	struct scsi_vpd_port_designation_cont *pdc;
10030	struct ctl_lun *lun;
10031	struct ctl_port *port;
10032	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10033	int num_target_port_groups, single;
10034
10035	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10036
10037	single = ctl_is_single;
10038	if (single)
10039		num_target_port_groups = 1;
10040	else
10041		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10042	num_target_ports = 0;
10043	iid_len = 0;
10044	id_len = 0;
10045	mtx_lock(&softc->ctl_lock);
10046	STAILQ_FOREACH(port, &softc->port_list, links) {
10047		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10048			continue;
10049		if (lun != NULL &&
10050		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10051		    CTL_MAX_LUNS)
10052			continue;
10053		num_target_ports++;
10054		if (port->init_devid)
10055			iid_len += port->init_devid->len;
10056		if (port->port_devid)
10057			id_len += port->port_devid->len;
10058	}
10059	mtx_unlock(&softc->ctl_lock);
10060
10061	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10062	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10063	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10064	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10065	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10066	ctsio->kern_sg_entries = 0;
10067
10068	if (data_len < alloc_len) {
10069		ctsio->residual = alloc_len - data_len;
10070		ctsio->kern_data_len = data_len;
10071		ctsio->kern_total_len = data_len;
10072	} else {
10073		ctsio->residual = 0;
10074		ctsio->kern_data_len = alloc_len;
10075		ctsio->kern_total_len = alloc_len;
10076	}
10077	ctsio->kern_data_resid = 0;
10078	ctsio->kern_rel_offset = 0;
10079	ctsio->kern_sg_entries = 0;
10080
10081	/*
10082	 * The control device is always connected.  The disk device, on the
10083	 * other hand, may not be online all the time.  Need to change this
10084	 * to figure out whether the disk device is actually online or not.
10085	 */
10086	if (lun != NULL)
10087		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10088				  lun->be_lun->lun_type;
10089	else
10090		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10091
10092	sp->page_code = SVPD_SCSI_PORTS;
10093	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10094	    sp->page_length);
10095	pd = &sp->design[0];
10096
10097	mtx_lock(&softc->ctl_lock);
10098	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10099		pg = 0;
10100	else
10101		pg = 1;
10102	for (g = 0; g < num_target_port_groups; g++) {
10103		STAILQ_FOREACH(port, &softc->port_list, links) {
10104			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10105				continue;
10106			if (lun != NULL &&
10107			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10108			    CTL_MAX_LUNS)
10109				continue;
10110			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10111			scsi_ulto2b(p, pd->relative_port_id);
10112			if (port->init_devid && g == pg) {
10113				iid_len = port->init_devid->len;
10114				memcpy(pd->initiator_transportid,
10115				    port->init_devid->data, port->init_devid->len);
10116			} else
10117				iid_len = 0;
10118			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10119			pdc = (struct scsi_vpd_port_designation_cont *)
10120			    (&pd->initiator_transportid[iid_len]);
10121			if (port->port_devid && g == pg) {
10122				id_len = port->port_devid->len;
10123				memcpy(pdc->target_port_descriptors,
10124				    port->port_devid->data, port->port_devid->len);
10125			} else
10126				id_len = 0;
10127			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10128			pd = (struct scsi_vpd_port_designation *)
10129			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10130		}
10131	}
10132	mtx_unlock(&softc->ctl_lock);
10133
10134	ctsio->scsi_status = SCSI_STATUS_OK;
10135	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10136	ctsio->be_move_done = ctl_config_move_done;
10137	ctl_datamove((union ctl_io *)ctsio);
10138
10139	return (CTL_RETVAL_COMPLETE);
10140}
10141
10142static int
10143ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10144{
10145	struct scsi_vpd_block_limits *bl_ptr;
10146	struct ctl_lun *lun;
10147	int bs;
10148
10149	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10150
10151	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10152	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10153	ctsio->kern_sg_entries = 0;
10154
10155	if (sizeof(*bl_ptr) < alloc_len) {
10156		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10157		ctsio->kern_data_len = sizeof(*bl_ptr);
10158		ctsio->kern_total_len = sizeof(*bl_ptr);
10159	} else {
10160		ctsio->residual = 0;
10161		ctsio->kern_data_len = alloc_len;
10162		ctsio->kern_total_len = alloc_len;
10163	}
10164	ctsio->kern_data_resid = 0;
10165	ctsio->kern_rel_offset = 0;
10166	ctsio->kern_sg_entries = 0;
10167
10168	/*
10169	 * The control device is always connected.  The disk device, on the
10170	 * other hand, may not be online all the time.  Need to change this
10171	 * to figure out whether the disk device is actually online or not.
10172	 */
10173	if (lun != NULL)
10174		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10175				  lun->be_lun->lun_type;
10176	else
10177		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10178
10179	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10180	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10181	bl_ptr->max_cmp_write_len = 0xff;
10182	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10183	if (lun != NULL) {
10184		bs = lun->be_lun->blocksize;
10185		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10186		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10187			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10188			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10189			if (lun->be_lun->pblockexp != 0) {
10190				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10191				    bl_ptr->opt_unmap_grain);
10192				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10193				    bl_ptr->unmap_grain_align);
10194			}
10195		}
10196		scsi_ulto4b(lun->be_lun->atomicblock,
10197		    bl_ptr->max_atomic_transfer_length);
10198		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10199		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10200	}
10201	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10202
10203	ctsio->scsi_status = SCSI_STATUS_OK;
10204	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10205	ctsio->be_move_done = ctl_config_move_done;
10206	ctl_datamove((union ctl_io *)ctsio);
10207
10208	return (CTL_RETVAL_COMPLETE);
10209}
10210
10211static int
10212ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10213{
10214	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10215	struct ctl_lun *lun;
10216	const char *value;
10217	u_int i;
10218
10219	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10220
10221	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10222	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10223	ctsio->kern_sg_entries = 0;
10224
10225	if (sizeof(*bdc_ptr) < alloc_len) {
10226		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10227		ctsio->kern_data_len = sizeof(*bdc_ptr);
10228		ctsio->kern_total_len = sizeof(*bdc_ptr);
10229	} else {
10230		ctsio->residual = 0;
10231		ctsio->kern_data_len = alloc_len;
10232		ctsio->kern_total_len = alloc_len;
10233	}
10234	ctsio->kern_data_resid = 0;
10235	ctsio->kern_rel_offset = 0;
10236	ctsio->kern_sg_entries = 0;
10237
10238	/*
10239	 * The control device is always connected.  The disk device, on the
10240	 * other hand, may not be online all the time.  Need to change this
10241	 * to figure out whether the disk device is actually online or not.
10242	 */
10243	if (lun != NULL)
10244		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10245				  lun->be_lun->lun_type;
10246	else
10247		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10248	bdc_ptr->page_code = SVPD_BDC;
10249	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10250	if (lun != NULL &&
10251	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10252		i = strtol(value, NULL, 0);
10253	else
10254		i = SVPD_NON_ROTATING;
10255	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10256	if (lun != NULL &&
10257	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10258		i = strtol(value, NULL, 0);
10259	else
10260		i = 0;
10261	bdc_ptr->wab_wac_ff = (i & 0x0f);
10262	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10263
10264	ctsio->scsi_status = SCSI_STATUS_OK;
10265	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10266	ctsio->be_move_done = ctl_config_move_done;
10267	ctl_datamove((union ctl_io *)ctsio);
10268
10269	return (CTL_RETVAL_COMPLETE);
10270}
10271
10272static int
10273ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10274{
10275	struct scsi_vpd_logical_block_prov *lbp_ptr;
10276	struct ctl_lun *lun;
10277
10278	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10279
10280	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10281	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10282	ctsio->kern_sg_entries = 0;
10283
10284	if (sizeof(*lbp_ptr) < alloc_len) {
10285		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10286		ctsio->kern_data_len = sizeof(*lbp_ptr);
10287		ctsio->kern_total_len = sizeof(*lbp_ptr);
10288	} else {
10289		ctsio->residual = 0;
10290		ctsio->kern_data_len = alloc_len;
10291		ctsio->kern_total_len = alloc_len;
10292	}
10293	ctsio->kern_data_resid = 0;
10294	ctsio->kern_rel_offset = 0;
10295	ctsio->kern_sg_entries = 0;
10296
10297	/*
10298	 * The control device is always connected.  The disk device, on the
10299	 * other hand, may not be online all the time.  Need to change this
10300	 * to figure out whether the disk device is actually online or not.
10301	 */
10302	if (lun != NULL)
10303		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10304				  lun->be_lun->lun_type;
10305	else
10306		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10307
10308	lbp_ptr->page_code = SVPD_LBP;
10309	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10310	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10311		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10312		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10313		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10314	}
10315
10316	ctsio->scsi_status = SCSI_STATUS_OK;
10317	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10318	ctsio->be_move_done = ctl_config_move_done;
10319	ctl_datamove((union ctl_io *)ctsio);
10320
10321	return (CTL_RETVAL_COMPLETE);
10322}
10323
10324static int
10325ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10326{
10327	struct scsi_inquiry *cdb;
10328	struct ctl_lun *lun;
10329	int alloc_len, retval;
10330
10331	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10332	cdb = (struct scsi_inquiry *)ctsio->cdb;
10333
10334	retval = CTL_RETVAL_COMPLETE;
10335
10336	alloc_len = scsi_2btoul(cdb->length);
10337
10338	switch (cdb->page_code) {
10339	case SVPD_SUPPORTED_PAGES:
10340		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10341		break;
10342	case SVPD_UNIT_SERIAL_NUMBER:
10343		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10344		break;
10345	case SVPD_DEVICE_ID:
10346		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10347		break;
10348	case SVPD_EXTENDED_INQUIRY_DATA:
10349		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10350		break;
10351	case SVPD_MODE_PAGE_POLICY:
10352		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10353		break;
10354	case SVPD_SCSI_PORTS:
10355		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10356		break;
10357	case SVPD_SCSI_TPC:
10358		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10359		break;
10360	case SVPD_BLOCK_LIMITS:
10361		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10362		break;
10363	case SVPD_BDC:
10364		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10365		break;
10366	case SVPD_LBP:
10367		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10368		break;
10369	default:
10370		ctl_set_invalid_field(ctsio,
10371				      /*sks_valid*/ 1,
10372				      /*command*/ 1,
10373				      /*field*/ 2,
10374				      /*bit_valid*/ 0,
10375				      /*bit*/ 0);
10376		ctl_done((union ctl_io *)ctsio);
10377		retval = CTL_RETVAL_COMPLETE;
10378		break;
10379	}
10380
10381	return (retval);
10382}
10383
10384static int
10385ctl_inquiry_std(struct ctl_scsiio *ctsio)
10386{
10387	struct scsi_inquiry_data *inq_ptr;
10388	struct scsi_inquiry *cdb;
10389	struct ctl_softc *ctl_softc;
10390	struct ctl_lun *lun;
10391	char *val;
10392	uint32_t alloc_len, data_len;
10393	ctl_port_type port_type;
10394
10395	ctl_softc = control_softc;
10396
10397	/*
10398	 * Figure out whether we're talking to a Fibre Channel port or not.
10399	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10400	 * SCSI front ends.
10401	 */
10402	port_type = ctl_softc->ctl_ports[
10403	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10404	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10405		port_type = CTL_PORT_SCSI;
10406
10407	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10408	cdb = (struct scsi_inquiry *)ctsio->cdb;
10409	alloc_len = scsi_2btoul(cdb->length);
10410
10411	/*
10412	 * We malloc the full inquiry data size here and fill it
10413	 * in.  If the user only asks for less, we'll give him
10414	 * that much.
10415	 */
10416	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10417	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10418	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10419	ctsio->kern_sg_entries = 0;
10420	ctsio->kern_data_resid = 0;
10421	ctsio->kern_rel_offset = 0;
10422
10423	if (data_len < alloc_len) {
10424		ctsio->residual = alloc_len - data_len;
10425		ctsio->kern_data_len = data_len;
10426		ctsio->kern_total_len = data_len;
10427	} else {
10428		ctsio->residual = 0;
10429		ctsio->kern_data_len = alloc_len;
10430		ctsio->kern_total_len = alloc_len;
10431	}
10432
10433	/*
10434	 * If we have a LUN configured, report it as connected.  Otherwise,
10435	 * report that it is offline or no device is supported, depending
10436	 * on the value of inquiry_pq_no_lun.
10437	 *
10438	 * According to the spec (SPC-4 r34), the peripheral qualifier
10439	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10440	 *
10441	 * "A peripheral device having the specified peripheral device type
10442	 * is not connected to this logical unit. However, the device
10443	 * server is capable of supporting the specified peripheral device
10444	 * type on this logical unit."
10445	 *
10446	 * According to the same spec, the peripheral qualifier
10447	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10448	 *
10449	 * "The device server is not capable of supporting a peripheral
10450	 * device on this logical unit. For this peripheral qualifier the
10451	 * peripheral device type shall be set to 1Fh. All other peripheral
10452	 * device type values are reserved for this peripheral qualifier."
10453	 *
10454	 * Given the text, it would seem that we probably want to report that
10455	 * the LUN is offline here.  There is no LUN connected, but we can
10456	 * support a LUN at the given LUN number.
10457	 *
10458	 * In the real world, though, it sounds like things are a little
10459	 * different:
10460	 *
10461	 * - Linux, when presented with a LUN with the offline peripheral
10462	 *   qualifier, will create an sg driver instance for it.  So when
10463	 *   you attach it to CTL, you wind up with a ton of sg driver
10464	 *   instances.  (One for every LUN that Linux bothered to probe.)
10465	 *   Linux does this despite the fact that it issues a REPORT LUNs
10466	 *   to LUN 0 to get the inventory of supported LUNs.
10467	 *
10468	 * - There is other anecdotal evidence (from Emulex folks) about
10469	 *   arrays that use the offline peripheral qualifier for LUNs that
10470	 *   are on the "passive" path in an active/passive array.
10471	 *
10472	 * So the solution is provide a hopefully reasonable default
10473	 * (return bad/no LUN) and allow the user to change the behavior
10474	 * with a tunable/sysctl variable.
10475	 */
10476	if (lun != NULL)
10477		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10478				  lun->be_lun->lun_type;
10479	else if (ctl_softc->inquiry_pq_no_lun == 0)
10480		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10481	else
10482		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10483
10484	/* RMB in byte 2 is 0 */
10485	inq_ptr->version = SCSI_REV_SPC4;
10486
10487	/*
10488	 * According to SAM-3, even if a device only supports a single
10489	 * level of LUN addressing, it should still set the HISUP bit:
10490	 *
10491	 * 4.9.1 Logical unit numbers overview
10492	 *
10493	 * All logical unit number formats described in this standard are
10494	 * hierarchical in structure even when only a single level in that
10495	 * hierarchy is used. The HISUP bit shall be set to one in the
10496	 * standard INQUIRY data (see SPC-2) when any logical unit number
10497	 * format described in this standard is used.  Non-hierarchical
10498	 * formats are outside the scope of this standard.
10499	 *
10500	 * Therefore we set the HiSup bit here.
10501	 *
10502	 * The reponse format is 2, per SPC-3.
10503	 */
10504	inq_ptr->response_format = SID_HiSup | 2;
10505
10506	inq_ptr->additional_length = data_len -
10507	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10508	CTL_DEBUG_PRINT(("additional_length = %d\n",
10509			 inq_ptr->additional_length));
10510
10511	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10512	/* 16 bit addressing */
10513	if (port_type == CTL_PORT_SCSI)
10514		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10515	/* XXX set the SID_MultiP bit here if we're actually going to
10516	   respond on multiple ports */
10517	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10518
10519	/* 16 bit data bus, synchronous transfers */
10520	if (port_type == CTL_PORT_SCSI)
10521		inq_ptr->flags = SID_WBus16 | SID_Sync;
10522	/*
10523	 * XXX KDM do we want to support tagged queueing on the control
10524	 * device at all?
10525	 */
10526	if ((lun == NULL)
10527	 || (lun->be_lun->lun_type != T_PROCESSOR))
10528		inq_ptr->flags |= SID_CmdQue;
10529	/*
10530	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10531	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10532	 * name and 4 bytes for the revision.
10533	 */
10534	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10535	    "vendor")) == NULL) {
10536		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10537	} else {
10538		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10539		strncpy(inq_ptr->vendor, val,
10540		    min(sizeof(inq_ptr->vendor), strlen(val)));
10541	}
10542	if (lun == NULL) {
10543		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10544		    sizeof(inq_ptr->product));
10545	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10546		switch (lun->be_lun->lun_type) {
10547		case T_DIRECT:
10548			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10549			    sizeof(inq_ptr->product));
10550			break;
10551		case T_PROCESSOR:
10552			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10553			    sizeof(inq_ptr->product));
10554			break;
10555		default:
10556			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10557			    sizeof(inq_ptr->product));
10558			break;
10559		}
10560	} else {
10561		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10562		strncpy(inq_ptr->product, val,
10563		    min(sizeof(inq_ptr->product), strlen(val)));
10564	}
10565
10566	/*
10567	 * XXX make this a macro somewhere so it automatically gets
10568	 * incremented when we make changes.
10569	 */
10570	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10571	    "revision")) == NULL) {
10572		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10573	} else {
10574		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10575		strncpy(inq_ptr->revision, val,
10576		    min(sizeof(inq_ptr->revision), strlen(val)));
10577	}
10578
10579	/*
10580	 * For parallel SCSI, we support double transition and single
10581	 * transition clocking.  We also support QAS (Quick Arbitration
10582	 * and Selection) and Information Unit transfers on both the
10583	 * control and array devices.
10584	 */
10585	if (port_type == CTL_PORT_SCSI)
10586		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10587				    SID_SPI_IUS;
10588
10589	/* SAM-5 (no version claimed) */
10590	scsi_ulto2b(0x00A0, inq_ptr->version1);
10591	/* SPC-4 (no version claimed) */
10592	scsi_ulto2b(0x0460, inq_ptr->version2);
10593	if (port_type == CTL_PORT_FC) {
10594		/* FCP-2 ANSI INCITS.350:2003 */
10595		scsi_ulto2b(0x0917, inq_ptr->version3);
10596	} else if (port_type == CTL_PORT_SCSI) {
10597		/* SPI-4 ANSI INCITS.362:200x */
10598		scsi_ulto2b(0x0B56, inq_ptr->version3);
10599	} else if (port_type == CTL_PORT_ISCSI) {
10600		/* iSCSI (no version claimed) */
10601		scsi_ulto2b(0x0960, inq_ptr->version3);
10602	} else if (port_type == CTL_PORT_SAS) {
10603		/* SAS (no version claimed) */
10604		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10605	}
10606
10607	if (lun == NULL) {
10608		/* SBC-4 (no version claimed) */
10609		scsi_ulto2b(0x0600, inq_ptr->version4);
10610	} else {
10611		switch (lun->be_lun->lun_type) {
10612		case T_DIRECT:
10613			/* SBC-4 (no version claimed) */
10614			scsi_ulto2b(0x0600, inq_ptr->version4);
10615			break;
10616		case T_PROCESSOR:
10617		default:
10618			break;
10619		}
10620	}
10621
10622	ctsio->scsi_status = SCSI_STATUS_OK;
10623	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10624	ctsio->be_move_done = ctl_config_move_done;
10625	ctl_datamove((union ctl_io *)ctsio);
10626	return (CTL_RETVAL_COMPLETE);
10627}
10628
10629int
10630ctl_inquiry(struct ctl_scsiio *ctsio)
10631{
10632	struct scsi_inquiry *cdb;
10633	int retval;
10634
10635	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10636
10637	cdb = (struct scsi_inquiry *)ctsio->cdb;
10638	if (cdb->byte2 & SI_EVPD)
10639		retval = ctl_inquiry_evpd(ctsio);
10640	else if (cdb->page_code == 0)
10641		retval = ctl_inquiry_std(ctsio);
10642	else {
10643		ctl_set_invalid_field(ctsio,
10644				      /*sks_valid*/ 1,
10645				      /*command*/ 1,
10646				      /*field*/ 2,
10647				      /*bit_valid*/ 0,
10648				      /*bit*/ 0);
10649		ctl_done((union ctl_io *)ctsio);
10650		return (CTL_RETVAL_COMPLETE);
10651	}
10652
10653	return (retval);
10654}
10655
10656/*
10657 * For known CDB types, parse the LBA and length.
10658 */
10659static int
10660ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10661{
10662	if (io->io_hdr.io_type != CTL_IO_SCSI)
10663		return (1);
10664
10665	switch (io->scsiio.cdb[0]) {
10666	case COMPARE_AND_WRITE: {
10667		struct scsi_compare_and_write *cdb;
10668
10669		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10670
10671		*lba = scsi_8btou64(cdb->addr);
10672		*len = cdb->length;
10673		break;
10674	}
10675	case READ_6:
10676	case WRITE_6: {
10677		struct scsi_rw_6 *cdb;
10678
10679		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10680
10681		*lba = scsi_3btoul(cdb->addr);
10682		/* only 5 bits are valid in the most significant address byte */
10683		*lba &= 0x1fffff;
10684		*len = cdb->length;
10685		break;
10686	}
10687	case READ_10:
10688	case WRITE_10: {
10689		struct scsi_rw_10 *cdb;
10690
10691		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10692
10693		*lba = scsi_4btoul(cdb->addr);
10694		*len = scsi_2btoul(cdb->length);
10695		break;
10696	}
10697	case WRITE_VERIFY_10: {
10698		struct scsi_write_verify_10 *cdb;
10699
10700		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10701
10702		*lba = scsi_4btoul(cdb->addr);
10703		*len = scsi_2btoul(cdb->length);
10704		break;
10705	}
10706	case READ_12:
10707	case WRITE_12: {
10708		struct scsi_rw_12 *cdb;
10709
10710		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10711
10712		*lba = scsi_4btoul(cdb->addr);
10713		*len = scsi_4btoul(cdb->length);
10714		break;
10715	}
10716	case WRITE_VERIFY_12: {
10717		struct scsi_write_verify_12 *cdb;
10718
10719		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10720
10721		*lba = scsi_4btoul(cdb->addr);
10722		*len = scsi_4btoul(cdb->length);
10723		break;
10724	}
10725	case READ_16:
10726	case WRITE_16:
10727	case WRITE_ATOMIC_16: {
10728		struct scsi_rw_16 *cdb;
10729
10730		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10731
10732		*lba = scsi_8btou64(cdb->addr);
10733		*len = scsi_4btoul(cdb->length);
10734		break;
10735	}
10736	case WRITE_VERIFY_16: {
10737		struct scsi_write_verify_16 *cdb;
10738
10739		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10740
10741		*lba = scsi_8btou64(cdb->addr);
10742		*len = scsi_4btoul(cdb->length);
10743		break;
10744	}
10745	case WRITE_SAME_10: {
10746		struct scsi_write_same_10 *cdb;
10747
10748		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10749
10750		*lba = scsi_4btoul(cdb->addr);
10751		*len = scsi_2btoul(cdb->length);
10752		break;
10753	}
10754	case WRITE_SAME_16: {
10755		struct scsi_write_same_16 *cdb;
10756
10757		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10758
10759		*lba = scsi_8btou64(cdb->addr);
10760		*len = scsi_4btoul(cdb->length);
10761		break;
10762	}
10763	case VERIFY_10: {
10764		struct scsi_verify_10 *cdb;
10765
10766		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10767
10768		*lba = scsi_4btoul(cdb->addr);
10769		*len = scsi_2btoul(cdb->length);
10770		break;
10771	}
10772	case VERIFY_12: {
10773		struct scsi_verify_12 *cdb;
10774
10775		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10776
10777		*lba = scsi_4btoul(cdb->addr);
10778		*len = scsi_4btoul(cdb->length);
10779		break;
10780	}
10781	case VERIFY_16: {
10782		struct scsi_verify_16 *cdb;
10783
10784		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10785
10786		*lba = scsi_8btou64(cdb->addr);
10787		*len = scsi_4btoul(cdb->length);
10788		break;
10789	}
10790	case UNMAP: {
10791		*lba = 0;
10792		*len = UINT64_MAX;
10793		break;
10794	}
10795	default:
10796		return (1);
10797		break; /* NOTREACHED */
10798	}
10799
10800	return (0);
10801}
10802
10803static ctl_action
10804ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10805{
10806	uint64_t endlba1, endlba2;
10807
10808	endlba1 = lba1 + len1 - 1;
10809	endlba2 = lba2 + len2 - 1;
10810
10811	if ((endlba1 < lba2)
10812	 || (endlba2 < lba1))
10813		return (CTL_ACTION_PASS);
10814	else
10815		return (CTL_ACTION_BLOCK);
10816}
10817
10818static int
10819ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10820{
10821	struct ctl_ptr_len_flags *ptrlen;
10822	struct scsi_unmap_desc *buf, *end, *range;
10823	uint64_t lba;
10824	uint32_t len;
10825
10826	/* If not UNMAP -- go other way. */
10827	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10828	    io->scsiio.cdb[0] != UNMAP)
10829		return (CTL_ACTION_ERROR);
10830
10831	/* If UNMAP without data -- block and wait for data. */
10832	ptrlen = (struct ctl_ptr_len_flags *)
10833	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10834	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10835	    ptrlen->ptr == NULL)
10836		return (CTL_ACTION_BLOCK);
10837
10838	/* UNMAP with data -- check for collision. */
10839	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10840	end = buf + ptrlen->len / sizeof(*buf);
10841	for (range = buf; range < end; range++) {
10842		lba = scsi_8btou64(range->lba);
10843		len = scsi_4btoul(range->length);
10844		if ((lba < lba2 + len2) && (lba + len > lba2))
10845			return (CTL_ACTION_BLOCK);
10846	}
10847	return (CTL_ACTION_PASS);
10848}
10849
10850static ctl_action
10851ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10852{
10853	uint64_t lba1, lba2;
10854	uint64_t len1, len2;
10855	int retval;
10856
10857	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10858		return (CTL_ACTION_ERROR);
10859
10860	retval = ctl_extent_check_unmap(io2, lba1, len1);
10861	if (retval != CTL_ACTION_ERROR)
10862		return (retval);
10863
10864	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10865		return (CTL_ACTION_ERROR);
10866
10867	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10868}
10869
10870static ctl_action
10871ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10872    union ctl_io *ooa_io)
10873{
10874	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10875	ctl_serialize_action *serialize_row;
10876
10877	/*
10878	 * The initiator attempted multiple untagged commands at the same
10879	 * time.  Can't do that.
10880	 */
10881	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10882	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10883	 && ((pending_io->io_hdr.nexus.targ_port ==
10884	      ooa_io->io_hdr.nexus.targ_port)
10885	  && (pending_io->io_hdr.nexus.initid.id ==
10886	      ooa_io->io_hdr.nexus.initid.id))
10887	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10888		return (CTL_ACTION_OVERLAP);
10889
10890	/*
10891	 * The initiator attempted to send multiple tagged commands with
10892	 * the same ID.  (It's fine if different initiators have the same
10893	 * tag ID.)
10894	 *
10895	 * Even if all of those conditions are true, we don't kill the I/O
10896	 * if the command ahead of us has been aborted.  We won't end up
10897	 * sending it to the FETD, and it's perfectly legal to resend a
10898	 * command with the same tag number as long as the previous
10899	 * instance of this tag number has been aborted somehow.
10900	 */
10901	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10902	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10903	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10904	 && ((pending_io->io_hdr.nexus.targ_port ==
10905	      ooa_io->io_hdr.nexus.targ_port)
10906	  && (pending_io->io_hdr.nexus.initid.id ==
10907	      ooa_io->io_hdr.nexus.initid.id))
10908	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10909		return (CTL_ACTION_OVERLAP_TAG);
10910
10911	/*
10912	 * If we get a head of queue tag, SAM-3 says that we should
10913	 * immediately execute it.
10914	 *
10915	 * What happens if this command would normally block for some other
10916	 * reason?  e.g. a request sense with a head of queue tag
10917	 * immediately after a write.  Normally that would block, but this
10918	 * will result in its getting executed immediately...
10919	 *
10920	 * We currently return "pass" instead of "skip", so we'll end up
10921	 * going through the rest of the queue to check for overlapped tags.
10922	 *
10923	 * XXX KDM check for other types of blockage first??
10924	 */
10925	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10926		return (CTL_ACTION_PASS);
10927
10928	/*
10929	 * Ordered tags have to block until all items ahead of them
10930	 * have completed.  If we get called with an ordered tag, we always
10931	 * block, if something else is ahead of us in the queue.
10932	 */
10933	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10934		return (CTL_ACTION_BLOCK);
10935
10936	/*
10937	 * Simple tags get blocked until all head of queue and ordered tags
10938	 * ahead of them have completed.  I'm lumping untagged commands in
10939	 * with simple tags here.  XXX KDM is that the right thing to do?
10940	 */
10941	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10942	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10943	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10944	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10945		return (CTL_ACTION_BLOCK);
10946
10947	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10948	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10949
10950	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10951
10952	switch (serialize_row[pending_entry->seridx]) {
10953	case CTL_SER_BLOCK:
10954		return (CTL_ACTION_BLOCK);
10955	case CTL_SER_EXTENT:
10956		return (ctl_extent_check(pending_io, ooa_io));
10957	case CTL_SER_EXTENTOPT:
10958		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10959		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10960			return (ctl_extent_check(pending_io, ooa_io));
10961		/* FALLTHROUGH */
10962	case CTL_SER_PASS:
10963		return (CTL_ACTION_PASS);
10964	case CTL_SER_BLOCKOPT:
10965		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10966		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10967			return (CTL_ACTION_BLOCK);
10968		return (CTL_ACTION_PASS);
10969	case CTL_SER_SKIP:
10970		return (CTL_ACTION_SKIP);
10971	default:
10972		panic("invalid serialization value %d",
10973		      serialize_row[pending_entry->seridx]);
10974	}
10975
10976	return (CTL_ACTION_ERROR);
10977}
10978
10979/*
10980 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10981 * Assumptions:
10982 * - pending_io is generally either incoming, or on the blocked queue
10983 * - starting I/O is the I/O we want to start the check with.
10984 */
10985static ctl_action
10986ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10987	      union ctl_io *starting_io)
10988{
10989	union ctl_io *ooa_io;
10990	ctl_action action;
10991
10992	mtx_assert(&lun->lun_lock, MA_OWNED);
10993
10994	/*
10995	 * Run back along the OOA queue, starting with the current
10996	 * blocked I/O and going through every I/O before it on the
10997	 * queue.  If starting_io is NULL, we'll just end up returning
10998	 * CTL_ACTION_PASS.
10999	 */
11000	for (ooa_io = starting_io; ooa_io != NULL;
11001	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11002	     ooa_links)){
11003
11004		/*
11005		 * This routine just checks to see whether
11006		 * cur_blocked is blocked by ooa_io, which is ahead
11007		 * of it in the queue.  It doesn't queue/dequeue
11008		 * cur_blocked.
11009		 */
11010		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11011		switch (action) {
11012		case CTL_ACTION_BLOCK:
11013		case CTL_ACTION_OVERLAP:
11014		case CTL_ACTION_OVERLAP_TAG:
11015		case CTL_ACTION_SKIP:
11016		case CTL_ACTION_ERROR:
11017			return (action);
11018			break; /* NOTREACHED */
11019		case CTL_ACTION_PASS:
11020			break;
11021		default:
11022			panic("invalid action %d", action);
11023			break;  /* NOTREACHED */
11024		}
11025	}
11026
11027	return (CTL_ACTION_PASS);
11028}
11029
11030/*
11031 * Assumptions:
11032 * - An I/O has just completed, and has been removed from the per-LUN OOA
11033 *   queue, so some items on the blocked queue may now be unblocked.
11034 */
11035static int
11036ctl_check_blocked(struct ctl_lun *lun)
11037{
11038	union ctl_io *cur_blocked, *next_blocked;
11039
11040	mtx_assert(&lun->lun_lock, MA_OWNED);
11041
11042	/*
11043	 * Run forward from the head of the blocked queue, checking each
11044	 * entry against the I/Os prior to it on the OOA queue to see if
11045	 * there is still any blockage.
11046	 *
11047	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11048	 * with our removing a variable on it while it is traversing the
11049	 * list.
11050	 */
11051	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11052	     cur_blocked != NULL; cur_blocked = next_blocked) {
11053		union ctl_io *prev_ooa;
11054		ctl_action action;
11055
11056		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11057							  blocked_links);
11058
11059		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11060						      ctl_ooaq, ooa_links);
11061
11062		/*
11063		 * If cur_blocked happens to be the first item in the OOA
11064		 * queue now, prev_ooa will be NULL, and the action
11065		 * returned will just be CTL_ACTION_PASS.
11066		 */
11067		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11068
11069		switch (action) {
11070		case CTL_ACTION_BLOCK:
11071			/* Nothing to do here, still blocked */
11072			break;
11073		case CTL_ACTION_OVERLAP:
11074		case CTL_ACTION_OVERLAP_TAG:
11075			/*
11076			 * This shouldn't happen!  In theory we've already
11077			 * checked this command for overlap...
11078			 */
11079			break;
11080		case CTL_ACTION_PASS:
11081		case CTL_ACTION_SKIP: {
11082			struct ctl_softc *softc;
11083			const struct ctl_cmd_entry *entry;
11084			uint32_t initidx;
11085			int isc_retval;
11086
11087			/*
11088			 * The skip case shouldn't happen, this transaction
11089			 * should have never made it onto the blocked queue.
11090			 */
11091			/*
11092			 * This I/O is no longer blocked, we can remove it
11093			 * from the blocked queue.  Since this is a TAILQ
11094			 * (doubly linked list), we can do O(1) removals
11095			 * from any place on the list.
11096			 */
11097			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11098				     blocked_links);
11099			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11100
11101			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11102				/*
11103				 * Need to send IO back to original side to
11104				 * run
11105				 */
11106				union ctl_ha_msg msg_info;
11107
11108				msg_info.hdr.original_sc =
11109					cur_blocked->io_hdr.original_sc;
11110				msg_info.hdr.serializing_sc = cur_blocked;
11111				msg_info.hdr.msg_type = CTL_MSG_R2R;
11112				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11113				     &msg_info, sizeof(msg_info), 0)) >
11114				     CTL_HA_STATUS_SUCCESS) {
11115					printf("CTL:Check Blocked error from "
11116					       "ctl_ha_msg_send %d\n",
11117					       isc_retval);
11118				}
11119				break;
11120			}
11121			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11122			softc = control_softc;
11123
11124			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11125
11126			/*
11127			 * Check this I/O for LUN state changes that may
11128			 * have happened while this command was blocked.
11129			 * The LUN state may have been changed by a command
11130			 * ahead of us in the queue, so we need to re-check
11131			 * for any states that can be caused by SCSI
11132			 * commands.
11133			 */
11134			if (ctl_scsiio_lun_check(softc, lun, entry,
11135						 &cur_blocked->scsiio) == 0) {
11136				cur_blocked->io_hdr.flags |=
11137				                      CTL_FLAG_IS_WAS_ON_RTR;
11138				ctl_enqueue_rtr(cur_blocked);
11139			} else
11140				ctl_done(cur_blocked);
11141			break;
11142		}
11143		default:
11144			/*
11145			 * This probably shouldn't happen -- we shouldn't
11146			 * get CTL_ACTION_ERROR, or anything else.
11147			 */
11148			break;
11149		}
11150	}
11151
11152	return (CTL_RETVAL_COMPLETE);
11153}
11154
11155/*
11156 * This routine (with one exception) checks LUN flags that can be set by
11157 * commands ahead of us in the OOA queue.  These flags have to be checked
11158 * when a command initially comes in, and when we pull a command off the
11159 * blocked queue and are preparing to execute it.  The reason we have to
11160 * check these flags for commands on the blocked queue is that the LUN
11161 * state may have been changed by a command ahead of us while we're on the
11162 * blocked queue.
11163 *
11164 * Ordering is somewhat important with these checks, so please pay
11165 * careful attention to the placement of any new checks.
11166 */
11167static int
11168ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11169    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11170{
11171	int retval;
11172	uint32_t residx;
11173
11174	retval = 0;
11175
11176	mtx_assert(&lun->lun_lock, MA_OWNED);
11177
11178	/*
11179	 * If this shelf is a secondary shelf controller, we have to reject
11180	 * any media access commands.
11181	 */
11182#if 0
11183	/* No longer needed for HA */
11184	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11185	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11186		ctl_set_lun_standby(ctsio);
11187		retval = 1;
11188		goto bailout;
11189	}
11190#endif
11191
11192	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11193		if (lun->flags & CTL_LUN_READONLY) {
11194			ctl_set_sense(ctsio, /*current_error*/ 1,
11195			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11196			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11197			retval = 1;
11198			goto bailout;
11199		}
11200		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11201		    .eca_and_aen & SCP_SWP) != 0) {
11202			ctl_set_sense(ctsio, /*current_error*/ 1,
11203			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11204			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11205			retval = 1;
11206			goto bailout;
11207		}
11208	}
11209
11210	/*
11211	 * Check for a reservation conflict.  If this command isn't allowed
11212	 * even on reserved LUNs, and if this initiator isn't the one who
11213	 * reserved us, reject the command with a reservation conflict.
11214	 */
11215	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11216	if ((lun->flags & CTL_LUN_RESERVED)
11217	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11218		if (lun->res_idx != residx) {
11219			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11220			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11221			retval = 1;
11222			goto bailout;
11223		}
11224	}
11225
11226	if ((lun->flags & CTL_LUN_PR_RESERVED)
11227	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11228		/*
11229		 * if we aren't registered or it's a res holder type
11230		 * reservation and this isn't the res holder then set a
11231		 * conflict.
11232		 * NOTE: Commands which might be allowed on write exclusive
11233		 * type reservations are checked in the particular command
11234		 * for a conflict. Read and SSU are the only ones.
11235		 */
11236		if (lun->pr_keys[residx] == 0
11237		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11238			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11239			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11240			retval = 1;
11241			goto bailout;
11242		}
11243
11244	}
11245
11246	if ((lun->flags & CTL_LUN_OFFLINE)
11247	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11248		ctl_set_lun_not_ready(ctsio);
11249		retval = 1;
11250		goto bailout;
11251	}
11252
11253	/*
11254	 * If the LUN is stopped, see if this particular command is allowed
11255	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11256	 */
11257	if ((lun->flags & CTL_LUN_STOPPED)
11258	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11259		/* "Logical unit not ready, initializing cmd. required" */
11260		ctl_set_lun_stopped(ctsio);
11261		retval = 1;
11262		goto bailout;
11263	}
11264
11265	if ((lun->flags & CTL_LUN_INOPERABLE)
11266	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11267		/* "Medium format corrupted" */
11268		ctl_set_medium_format_corrupted(ctsio);
11269		retval = 1;
11270		goto bailout;
11271	}
11272
11273bailout:
11274	return (retval);
11275
11276}
11277
11278static void
11279ctl_failover_io(union ctl_io *io, int have_lock)
11280{
11281	ctl_set_busy(&io->scsiio);
11282	ctl_done(io);
11283}
11284
11285static void
11286ctl_failover(void)
11287{
11288	struct ctl_lun *lun;
11289	struct ctl_softc *ctl_softc;
11290	union ctl_io *next_io, *pending_io;
11291	union ctl_io *io;
11292	int lun_idx;
11293	int i;
11294
11295	ctl_softc = control_softc;
11296
11297	mtx_lock(&ctl_softc->ctl_lock);
11298	/*
11299	 * Remove any cmds from the other SC from the rtr queue.  These
11300	 * will obviously only be for LUNs for which we're the primary.
11301	 * We can't send status or get/send data for these commands.
11302	 * Since they haven't been executed yet, we can just remove them.
11303	 * We'll either abort them or delete them below, depending on
11304	 * which HA mode we're in.
11305	 */
11306#ifdef notyet
11307	mtx_lock(&ctl_softc->queue_lock);
11308	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11309	     io != NULL; io = next_io) {
11310		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11311		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11312			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11313				      ctl_io_hdr, links);
11314	}
11315	mtx_unlock(&ctl_softc->queue_lock);
11316#endif
11317
11318	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11319		lun = ctl_softc->ctl_luns[lun_idx];
11320		if (lun==NULL)
11321			continue;
11322
11323		/*
11324		 * Processor LUNs are primary on both sides.
11325		 * XXX will this always be true?
11326		 */
11327		if (lun->be_lun->lun_type == T_PROCESSOR)
11328			continue;
11329
11330		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11331		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11332			printf("FAILOVER: primary lun %d\n", lun_idx);
11333		        /*
11334			 * Remove all commands from the other SC. First from the
11335			 * blocked queue then from the ooa queue. Once we have
11336			 * removed them. Call ctl_check_blocked to see if there
11337			 * is anything that can run.
11338			 */
11339			for (io = (union ctl_io *)TAILQ_FIRST(
11340			     &lun->blocked_queue); io != NULL; io = next_io) {
11341
11342		        	next_io = (union ctl_io *)TAILQ_NEXT(
11343				    &io->io_hdr, blocked_links);
11344
11345				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11346					TAILQ_REMOVE(&lun->blocked_queue,
11347						     &io->io_hdr,blocked_links);
11348					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11349					TAILQ_REMOVE(&lun->ooa_queue,
11350						     &io->io_hdr, ooa_links);
11351
11352					ctl_free_io(io);
11353				}
11354			}
11355
11356			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11357	     		     io != NULL; io = next_io) {
11358
11359		        	next_io = (union ctl_io *)TAILQ_NEXT(
11360				    &io->io_hdr, ooa_links);
11361
11362				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11363
11364					TAILQ_REMOVE(&lun->ooa_queue,
11365						&io->io_hdr,
11366					     	ooa_links);
11367
11368					ctl_free_io(io);
11369				}
11370			}
11371			ctl_check_blocked(lun);
11372		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11373			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11374
11375			printf("FAILOVER: primary lun %d\n", lun_idx);
11376			/*
11377			 * Abort all commands from the other SC.  We can't
11378			 * send status back for them now.  These should get
11379			 * cleaned up when they are completed or come out
11380			 * for a datamove operation.
11381			 */
11382			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11383	     		     io != NULL; io = next_io) {
11384		        	next_io = (union ctl_io *)TAILQ_NEXT(
11385					&io->io_hdr, ooa_links);
11386
11387				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11388					io->io_hdr.flags |= CTL_FLAG_ABORT;
11389			}
11390		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11391			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11392
11393			printf("FAILOVER: secondary lun %d\n", lun_idx);
11394
11395			lun->flags |= CTL_LUN_PRIMARY_SC;
11396
11397			/*
11398			 * We send all I/O that was sent to this controller
11399			 * and redirected to the other side back with
11400			 * busy status, and have the initiator retry it.
11401			 * Figuring out how much data has been transferred,
11402			 * etc. and picking up where we left off would be
11403			 * very tricky.
11404			 *
11405			 * XXX KDM need to remove I/O from the blocked
11406			 * queue as well!
11407			 */
11408			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11409			     &lun->ooa_queue); pending_io != NULL;
11410			     pending_io = next_io) {
11411
11412				next_io =  (union ctl_io *)TAILQ_NEXT(
11413					&pending_io->io_hdr, ooa_links);
11414
11415				pending_io->io_hdr.flags &=
11416					~CTL_FLAG_SENT_2OTHER_SC;
11417
11418				if (pending_io->io_hdr.flags &
11419				    CTL_FLAG_IO_ACTIVE) {
11420					pending_io->io_hdr.flags |=
11421						CTL_FLAG_FAILOVER;
11422				} else {
11423					ctl_set_busy(&pending_io->scsiio);
11424					ctl_done(pending_io);
11425				}
11426			}
11427
11428			/*
11429			 * Build Unit Attention
11430			 */
11431			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11432				lun->pending_ua[i] |=
11433				                     CTL_UA_ASYM_ACC_CHANGE;
11434			}
11435		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11436			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11437			printf("FAILOVER: secondary lun %d\n", lun_idx);
11438			/*
11439			 * if the first io on the OOA is not on the RtR queue
11440			 * add it.
11441			 */
11442			lun->flags |= CTL_LUN_PRIMARY_SC;
11443
11444			pending_io = (union ctl_io *)TAILQ_FIRST(
11445			    &lun->ooa_queue);
11446			if (pending_io==NULL) {
11447				printf("Nothing on OOA queue\n");
11448				continue;
11449			}
11450
11451			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11452			if ((pending_io->io_hdr.flags &
11453			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11454				pending_io->io_hdr.flags |=
11455				    CTL_FLAG_IS_WAS_ON_RTR;
11456				ctl_enqueue_rtr(pending_io);
11457			}
11458#if 0
11459			else
11460			{
11461				printf("Tag 0x%04x is running\n",
11462				      pending_io->scsiio.tag_num);
11463			}
11464#endif
11465
11466			next_io = (union ctl_io *)TAILQ_NEXT(
11467			    &pending_io->io_hdr, ooa_links);
11468			for (pending_io=next_io; pending_io != NULL;
11469			     pending_io = next_io) {
11470				pending_io->io_hdr.flags &=
11471				    ~CTL_FLAG_SENT_2OTHER_SC;
11472				next_io = (union ctl_io *)TAILQ_NEXT(
11473					&pending_io->io_hdr, ooa_links);
11474				if (pending_io->io_hdr.flags &
11475				    CTL_FLAG_IS_WAS_ON_RTR) {
11476#if 0
11477				        printf("Tag 0x%04x is running\n",
11478				      		pending_io->scsiio.tag_num);
11479#endif
11480					continue;
11481				}
11482
11483				switch (ctl_check_ooa(lun, pending_io,
11484			            (union ctl_io *)TAILQ_PREV(
11485				    &pending_io->io_hdr, ctl_ooaq,
11486				    ooa_links))) {
11487
11488				case CTL_ACTION_BLOCK:
11489					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11490							  &pending_io->io_hdr,
11491							  blocked_links);
11492					pending_io->io_hdr.flags |=
11493					    CTL_FLAG_BLOCKED;
11494					break;
11495				case CTL_ACTION_PASS:
11496				case CTL_ACTION_SKIP:
11497					pending_io->io_hdr.flags |=
11498					    CTL_FLAG_IS_WAS_ON_RTR;
11499					ctl_enqueue_rtr(pending_io);
11500					break;
11501				case CTL_ACTION_OVERLAP:
11502					ctl_set_overlapped_cmd(
11503					    (struct ctl_scsiio *)pending_io);
11504					ctl_done(pending_io);
11505					break;
11506				case CTL_ACTION_OVERLAP_TAG:
11507					ctl_set_overlapped_tag(
11508					    (struct ctl_scsiio *)pending_io,
11509					    pending_io->scsiio.tag_num & 0xff);
11510					ctl_done(pending_io);
11511					break;
11512				case CTL_ACTION_ERROR:
11513				default:
11514					ctl_set_internal_failure(
11515						(struct ctl_scsiio *)pending_io,
11516						0,  // sks_valid
11517						0); //retry count
11518					ctl_done(pending_io);
11519					break;
11520				}
11521			}
11522
11523			/*
11524			 * Build Unit Attention
11525			 */
11526			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11527				lun->pending_ua[i] |=
11528				                     CTL_UA_ASYM_ACC_CHANGE;
11529			}
11530		} else {
11531			panic("Unhandled HA mode failover, LUN flags = %#x, "
11532			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11533		}
11534	}
11535	ctl_pause_rtr = 0;
11536	mtx_unlock(&ctl_softc->ctl_lock);
11537}
11538
11539static int
11540ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11541{
11542	struct ctl_lun *lun;
11543	const struct ctl_cmd_entry *entry;
11544	uint32_t initidx, targ_lun;
11545	int retval;
11546
11547	retval = 0;
11548
11549	lun = NULL;
11550
11551	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11552	if ((targ_lun < CTL_MAX_LUNS)
11553	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11554		lun = ctl_softc->ctl_luns[targ_lun];
11555		/*
11556		 * If the LUN is invalid, pretend that it doesn't exist.
11557		 * It will go away as soon as all pending I/O has been
11558		 * completed.
11559		 */
11560		if (lun->flags & CTL_LUN_DISABLED) {
11561			lun = NULL;
11562		} else {
11563			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11564			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11565				lun->be_lun;
11566			if (lun->be_lun->lun_type == T_PROCESSOR) {
11567				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11568			}
11569
11570			/*
11571			 * Every I/O goes into the OOA queue for a
11572			 * particular LUN, and stays there until completion.
11573			 */
11574			mtx_lock(&lun->lun_lock);
11575			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11576			    ooa_links);
11577		}
11578	} else {
11579		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11580		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11581	}
11582
11583	/* Get command entry and return error if it is unsuppotyed. */
11584	entry = ctl_validate_command(ctsio);
11585	if (entry == NULL) {
11586		if (lun)
11587			mtx_unlock(&lun->lun_lock);
11588		return (retval);
11589	}
11590
11591	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11592	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11593
11594	/*
11595	 * Check to see whether we can send this command to LUNs that don't
11596	 * exist.  This should pretty much only be the case for inquiry
11597	 * and request sense.  Further checks, below, really require having
11598	 * a LUN, so we can't really check the command anymore.  Just put
11599	 * it on the rtr queue.
11600	 */
11601	if (lun == NULL) {
11602		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11603			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11604			ctl_enqueue_rtr((union ctl_io *)ctsio);
11605			return (retval);
11606		}
11607
11608		ctl_set_unsupported_lun(ctsio);
11609		ctl_done((union ctl_io *)ctsio);
11610		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11611		return (retval);
11612	} else {
11613		/*
11614		 * Make sure we support this particular command on this LUN.
11615		 * e.g., we don't support writes to the control LUN.
11616		 */
11617		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11618			mtx_unlock(&lun->lun_lock);
11619			ctl_set_invalid_opcode(ctsio);
11620			ctl_done((union ctl_io *)ctsio);
11621			return (retval);
11622		}
11623	}
11624
11625	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11626
11627#ifdef CTL_WITH_CA
11628	/*
11629	 * If we've got a request sense, it'll clear the contingent
11630	 * allegiance condition.  Otherwise, if we have a CA condition for
11631	 * this initiator, clear it, because it sent down a command other
11632	 * than request sense.
11633	 */
11634	if ((ctsio->cdb[0] != REQUEST_SENSE)
11635	 && (ctl_is_set(lun->have_ca, initidx)))
11636		ctl_clear_mask(lun->have_ca, initidx);
11637#endif
11638
11639	/*
11640	 * If the command has this flag set, it handles its own unit
11641	 * attention reporting, we shouldn't do anything.  Otherwise we
11642	 * check for any pending unit attentions, and send them back to the
11643	 * initiator.  We only do this when a command initially comes in,
11644	 * not when we pull it off the blocked queue.
11645	 *
11646	 * According to SAM-3, section 5.3.2, the order that things get
11647	 * presented back to the host is basically unit attentions caused
11648	 * by some sort of reset event, busy status, reservation conflicts
11649	 * or task set full, and finally any other status.
11650	 *
11651	 * One issue here is that some of the unit attentions we report
11652	 * don't fall into the "reset" category (e.g. "reported luns data
11653	 * has changed").  So reporting it here, before the reservation
11654	 * check, may be technically wrong.  I guess the only thing to do
11655	 * would be to check for and report the reset events here, and then
11656	 * check for the other unit attention types after we check for a
11657	 * reservation conflict.
11658	 *
11659	 * XXX KDM need to fix this
11660	 */
11661	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11662		ctl_ua_type ua_type;
11663
11664		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11665			scsi_sense_data_type sense_format;
11666
11667			if (lun != NULL)
11668				sense_format = (lun->flags &
11669				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11670				    SSD_TYPE_FIXED;
11671			else
11672				sense_format = SSD_TYPE_FIXED;
11673
11674			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11675			    &ctsio->sense_data, sense_format);
11676			if (ua_type != CTL_UA_NONE) {
11677				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11678				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11679						       CTL_AUTOSENSE;
11680				ctsio->sense_len = SSD_FULL_SIZE;
11681				mtx_unlock(&lun->lun_lock);
11682				ctl_done((union ctl_io *)ctsio);
11683				return (retval);
11684			}
11685		}
11686	}
11687
11688
11689	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11690		mtx_unlock(&lun->lun_lock);
11691		ctl_done((union ctl_io *)ctsio);
11692		return (retval);
11693	}
11694
11695	/*
11696	 * XXX CHD this is where we want to send IO to other side if
11697	 * this LUN is secondary on this SC. We will need to make a copy
11698	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11699	 * the copy we send as FROM_OTHER.
11700	 * We also need to stuff the address of the original IO so we can
11701	 * find it easily. Something similar will need be done on the other
11702	 * side so when we are done we can find the copy.
11703	 */
11704	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11705		union ctl_ha_msg msg_info;
11706		int isc_retval;
11707
11708		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11709
11710		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11711		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11712#if 0
11713		printf("1. ctsio %p\n", ctsio);
11714#endif
11715		msg_info.hdr.serializing_sc = NULL;
11716		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11717		msg_info.scsi.tag_num = ctsio->tag_num;
11718		msg_info.scsi.tag_type = ctsio->tag_type;
11719		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11720
11721		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11722
11723		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11724		    (void *)&msg_info, sizeof(msg_info), 0)) >
11725		    CTL_HA_STATUS_SUCCESS) {
11726			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11727			       isc_retval);
11728			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11729		} else {
11730#if 0
11731			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11732#endif
11733		}
11734
11735		/*
11736		 * XXX KDM this I/O is off the incoming queue, but hasn't
11737		 * been inserted on any other queue.  We may need to come
11738		 * up with a holding queue while we wait for serialization
11739		 * so that we have an idea of what we're waiting for from
11740		 * the other side.
11741		 */
11742		mtx_unlock(&lun->lun_lock);
11743		return (retval);
11744	}
11745
11746	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11747			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11748			      ctl_ooaq, ooa_links))) {
11749	case CTL_ACTION_BLOCK:
11750		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11751		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11752				  blocked_links);
11753		mtx_unlock(&lun->lun_lock);
11754		return (retval);
11755	case CTL_ACTION_PASS:
11756	case CTL_ACTION_SKIP:
11757		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11758		mtx_unlock(&lun->lun_lock);
11759		ctl_enqueue_rtr((union ctl_io *)ctsio);
11760		break;
11761	case CTL_ACTION_OVERLAP:
11762		mtx_unlock(&lun->lun_lock);
11763		ctl_set_overlapped_cmd(ctsio);
11764		ctl_done((union ctl_io *)ctsio);
11765		break;
11766	case CTL_ACTION_OVERLAP_TAG:
11767		mtx_unlock(&lun->lun_lock);
11768		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11769		ctl_done((union ctl_io *)ctsio);
11770		break;
11771	case CTL_ACTION_ERROR:
11772	default:
11773		mtx_unlock(&lun->lun_lock);
11774		ctl_set_internal_failure(ctsio,
11775					 /*sks_valid*/ 0,
11776					 /*retry_count*/ 0);
11777		ctl_done((union ctl_io *)ctsio);
11778		break;
11779	}
11780	return (retval);
11781}
11782
11783const struct ctl_cmd_entry *
11784ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11785{
11786	const struct ctl_cmd_entry *entry;
11787	int service_action;
11788
11789	entry = &ctl_cmd_table[ctsio->cdb[0]];
11790	if (sa)
11791		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11792	if (entry->flags & CTL_CMD_FLAG_SA5) {
11793		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11794		entry = &((const struct ctl_cmd_entry *)
11795		    entry->execute)[service_action];
11796	}
11797	return (entry);
11798}
11799
11800const struct ctl_cmd_entry *
11801ctl_validate_command(struct ctl_scsiio *ctsio)
11802{
11803	const struct ctl_cmd_entry *entry;
11804	int i, sa;
11805	uint8_t diff;
11806
11807	entry = ctl_get_cmd_entry(ctsio, &sa);
11808	if (entry->execute == NULL) {
11809		if (sa)
11810			ctl_set_invalid_field(ctsio,
11811					      /*sks_valid*/ 1,
11812					      /*command*/ 1,
11813					      /*field*/ 1,
11814					      /*bit_valid*/ 1,
11815					      /*bit*/ 4);
11816		else
11817			ctl_set_invalid_opcode(ctsio);
11818		ctl_done((union ctl_io *)ctsio);
11819		return (NULL);
11820	}
11821	KASSERT(entry->length > 0,
11822	    ("Not defined length for command 0x%02x/0x%02x",
11823	     ctsio->cdb[0], ctsio->cdb[1]));
11824	for (i = 1; i < entry->length; i++) {
11825		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11826		if (diff == 0)
11827			continue;
11828		ctl_set_invalid_field(ctsio,
11829				      /*sks_valid*/ 1,
11830				      /*command*/ 1,
11831				      /*field*/ i,
11832				      /*bit_valid*/ 1,
11833				      /*bit*/ fls(diff) - 1);
11834		ctl_done((union ctl_io *)ctsio);
11835		return (NULL);
11836	}
11837	return (entry);
11838}
11839
11840static int
11841ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11842{
11843
11844	switch (lun_type) {
11845	case T_PROCESSOR:
11846		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11847		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11848			return (0);
11849		break;
11850	case T_DIRECT:
11851		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11852		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11853			return (0);
11854		break;
11855	default:
11856		return (0);
11857	}
11858	return (1);
11859}
11860
11861static int
11862ctl_scsiio(struct ctl_scsiio *ctsio)
11863{
11864	int retval;
11865	const struct ctl_cmd_entry *entry;
11866
11867	retval = CTL_RETVAL_COMPLETE;
11868
11869	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11870
11871	entry = ctl_get_cmd_entry(ctsio, NULL);
11872
11873	/*
11874	 * If this I/O has been aborted, just send it straight to
11875	 * ctl_done() without executing it.
11876	 */
11877	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11878		ctl_done((union ctl_io *)ctsio);
11879		goto bailout;
11880	}
11881
11882	/*
11883	 * All the checks should have been handled by ctl_scsiio_precheck().
11884	 * We should be clear now to just execute the I/O.
11885	 */
11886	retval = entry->execute(ctsio);
11887
11888bailout:
11889	return (retval);
11890}
11891
11892/*
11893 * Since we only implement one target right now, a bus reset simply resets
11894 * our single target.
11895 */
11896static int
11897ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11898{
11899	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11900}
11901
11902static int
11903ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11904		 ctl_ua_type ua_type)
11905{
11906	struct ctl_lun *lun;
11907	int retval;
11908
11909	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11910		union ctl_ha_msg msg_info;
11911
11912		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11913		msg_info.hdr.nexus = io->io_hdr.nexus;
11914		if (ua_type==CTL_UA_TARG_RESET)
11915			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11916		else
11917			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11918		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11919		msg_info.hdr.original_sc = NULL;
11920		msg_info.hdr.serializing_sc = NULL;
11921		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11922		    (void *)&msg_info, sizeof(msg_info), 0)) {
11923		}
11924	}
11925	retval = 0;
11926
11927	mtx_lock(&ctl_softc->ctl_lock);
11928	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11929		retval += ctl_lun_reset(lun, io, ua_type);
11930	mtx_unlock(&ctl_softc->ctl_lock);
11931
11932	return (retval);
11933}
11934
11935/*
11936 * The LUN should always be set.  The I/O is optional, and is used to
11937 * distinguish between I/Os sent by this initiator, and by other
11938 * initiators.  We set unit attention for initiators other than this one.
11939 * SAM-3 is vague on this point.  It does say that a unit attention should
11940 * be established for other initiators when a LUN is reset (see section
11941 * 5.7.3), but it doesn't specifically say that the unit attention should
11942 * be established for this particular initiator when a LUN is reset.  Here
11943 * is the relevant text, from SAM-3 rev 8:
11944 *
11945 * 5.7.2 When a SCSI initiator port aborts its own tasks
11946 *
11947 * When a SCSI initiator port causes its own task(s) to be aborted, no
11948 * notification that the task(s) have been aborted shall be returned to
11949 * the SCSI initiator port other than the completion response for the
11950 * command or task management function action that caused the task(s) to
11951 * be aborted and notification(s) associated with related effects of the
11952 * action (e.g., a reset unit attention condition).
11953 *
11954 * XXX KDM for now, we're setting unit attention for all initiators.
11955 */
11956static int
11957ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11958{
11959	union ctl_io *xio;
11960#if 0
11961	uint32_t initindex;
11962#endif
11963	int i;
11964
11965	mtx_lock(&lun->lun_lock);
11966	/*
11967	 * Run through the OOA queue and abort each I/O.
11968	 */
11969#if 0
11970	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11971#endif
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		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11975	}
11976
11977	/*
11978	 * This version sets unit attention for every
11979	 */
11980#if 0
11981	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11982	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11983		if (initindex == i)
11984			continue;
11985		lun->pending_ua[i] |= ua_type;
11986	}
11987#endif
11988
11989	/*
11990	 * A reset (any kind, really) clears reservations established with
11991	 * RESERVE/RELEASE.  It does not clear reservations established
11992	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11993	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11994	 * reservations made with the RESERVE/RELEASE commands, because
11995	 * those commands are obsolete in SPC-3.
11996	 */
11997	lun->flags &= ~CTL_LUN_RESERVED;
11998
11999	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12000#ifdef CTL_WITH_CA
12001		ctl_clear_mask(lun->have_ca, i);
12002#endif
12003		lun->pending_ua[i] |= ua_type;
12004	}
12005	mtx_unlock(&lun->lun_lock);
12006
12007	return (0);
12008}
12009
12010static void
12011ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12012    int other_sc)
12013{
12014	union ctl_io *xio;
12015
12016	mtx_assert(&lun->lun_lock, MA_OWNED);
12017
12018	/*
12019	 * Run through the OOA queue and attempt to find the given I/O.
12020	 * The target port, initiator ID, tag type and tag number have to
12021	 * match the values that we got from the initiator.  If we have an
12022	 * untagged command to abort, simply abort the first untagged command
12023	 * we come to.  We only allow one untagged command at a time of course.
12024	 */
12025	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12026	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12027
12028		if ((targ_port == UINT32_MAX ||
12029		     targ_port == xio->io_hdr.nexus.targ_port) &&
12030		    (init_id == UINT32_MAX ||
12031		     init_id == xio->io_hdr.nexus.initid.id)) {
12032			if (targ_port != xio->io_hdr.nexus.targ_port ||
12033			    init_id != xio->io_hdr.nexus.initid.id)
12034				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12035			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12036			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12037				union ctl_ha_msg msg_info;
12038
12039				msg_info.hdr.nexus = xio->io_hdr.nexus;
12040				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12041				msg_info.task.tag_num = xio->scsiio.tag_num;
12042				msg_info.task.tag_type = xio->scsiio.tag_type;
12043				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12044				msg_info.hdr.original_sc = NULL;
12045				msg_info.hdr.serializing_sc = NULL;
12046				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12047				    (void *)&msg_info, sizeof(msg_info), 0);
12048			}
12049		}
12050	}
12051}
12052
12053static int
12054ctl_abort_task_set(union ctl_io *io)
12055{
12056	struct ctl_softc *softc = control_softc;
12057	struct ctl_lun *lun;
12058	uint32_t targ_lun;
12059
12060	/*
12061	 * Look up the LUN.
12062	 */
12063	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12064	mtx_lock(&softc->ctl_lock);
12065	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12066		lun = softc->ctl_luns[targ_lun];
12067	else {
12068		mtx_unlock(&softc->ctl_lock);
12069		return (1);
12070	}
12071
12072	mtx_lock(&lun->lun_lock);
12073	mtx_unlock(&softc->ctl_lock);
12074	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12075		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12076		    io->io_hdr.nexus.initid.id,
12077		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12078	} else { /* CTL_TASK_CLEAR_TASK_SET */
12079		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12080		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12081	}
12082	mtx_unlock(&lun->lun_lock);
12083	return (0);
12084}
12085
12086static int
12087ctl_i_t_nexus_reset(union ctl_io *io)
12088{
12089	struct ctl_softc *softc = control_softc;
12090	struct ctl_lun *lun;
12091	uint32_t initindex, residx;
12092
12093	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12094	residx = ctl_get_resindex(&io->io_hdr.nexus);
12095	mtx_lock(&softc->ctl_lock);
12096	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12097		mtx_lock(&lun->lun_lock);
12098		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12099		    io->io_hdr.nexus.initid.id,
12100		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12101#ifdef CTL_WITH_CA
12102		ctl_clear_mask(lun->have_ca, initindex);
12103#endif
12104		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12105			lun->flags &= ~CTL_LUN_RESERVED;
12106		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12107		mtx_unlock(&lun->lun_lock);
12108	}
12109	mtx_unlock(&softc->ctl_lock);
12110	return (0);
12111}
12112
12113static int
12114ctl_abort_task(union ctl_io *io)
12115{
12116	union ctl_io *xio;
12117	struct ctl_lun *lun;
12118	struct ctl_softc *ctl_softc;
12119#if 0
12120	struct sbuf sb;
12121	char printbuf[128];
12122#endif
12123	int found;
12124	uint32_t targ_lun;
12125
12126	ctl_softc = control_softc;
12127	found = 0;
12128
12129	/*
12130	 * Look up the LUN.
12131	 */
12132	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12133	mtx_lock(&ctl_softc->ctl_lock);
12134	if ((targ_lun < CTL_MAX_LUNS)
12135	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12136		lun = ctl_softc->ctl_luns[targ_lun];
12137	else {
12138		mtx_unlock(&ctl_softc->ctl_lock);
12139		return (1);
12140	}
12141
12142#if 0
12143	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12144	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12145#endif
12146
12147	mtx_lock(&lun->lun_lock);
12148	mtx_unlock(&ctl_softc->ctl_lock);
12149	/*
12150	 * Run through the OOA queue and attempt to find the given I/O.
12151	 * The target port, initiator ID, tag type and tag number have to
12152	 * match the values that we got from the initiator.  If we have an
12153	 * untagged command to abort, simply abort the first untagged command
12154	 * we come to.  We only allow one untagged command at a time of course.
12155	 */
12156#if 0
12157	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12158#endif
12159	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12160	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12161#if 0
12162		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12163
12164		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12165			    lun->lun, xio->scsiio.tag_num,
12166			    xio->scsiio.tag_type,
12167			    (xio->io_hdr.blocked_links.tqe_prev
12168			    == NULL) ? "" : " BLOCKED",
12169			    (xio->io_hdr.flags &
12170			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12171			    (xio->io_hdr.flags &
12172			    CTL_FLAG_ABORT) ? " ABORT" : "",
12173			    (xio->io_hdr.flags &
12174			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12175		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12176		sbuf_finish(&sb);
12177		printf("%s\n", sbuf_data(&sb));
12178#endif
12179
12180		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12181		 && (xio->io_hdr.nexus.initid.id ==
12182		     io->io_hdr.nexus.initid.id)) {
12183			/*
12184			 * If the abort says that the task is untagged, the
12185			 * task in the queue must be untagged.  Otherwise,
12186			 * we just check to see whether the tag numbers
12187			 * match.  This is because the QLogic firmware
12188			 * doesn't pass back the tag type in an abort
12189			 * request.
12190			 */
12191#if 0
12192			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12193			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12194			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12195#endif
12196			/*
12197			 * XXX KDM we've got problems with FC, because it
12198			 * doesn't send down a tag type with aborts.  So we
12199			 * can only really go by the tag number...
12200			 * This may cause problems with parallel SCSI.
12201			 * Need to figure that out!!
12202			 */
12203			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12204				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12205				found = 1;
12206				if ((io->io_hdr.flags &
12207				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12208				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12209					union ctl_ha_msg msg_info;
12210
12211					io->io_hdr.flags |=
12212					                CTL_FLAG_SENT_2OTHER_SC;
12213					msg_info.hdr.nexus = io->io_hdr.nexus;
12214					msg_info.task.task_action =
12215						CTL_TASK_ABORT_TASK;
12216					msg_info.task.tag_num =
12217						io->taskio.tag_num;
12218					msg_info.task.tag_type =
12219						io->taskio.tag_type;
12220					msg_info.hdr.msg_type =
12221						CTL_MSG_MANAGE_TASKS;
12222					msg_info.hdr.original_sc = NULL;
12223					msg_info.hdr.serializing_sc = NULL;
12224#if 0
12225					printf("Sent Abort to other side\n");
12226#endif
12227					if (CTL_HA_STATUS_SUCCESS !=
12228					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12229		    				(void *)&msg_info,
12230						sizeof(msg_info), 0)) {
12231					}
12232				}
12233#if 0
12234				printf("ctl_abort_task: found I/O to abort\n");
12235#endif
12236				break;
12237			}
12238		}
12239	}
12240	mtx_unlock(&lun->lun_lock);
12241
12242	if (found == 0) {
12243		/*
12244		 * This isn't really an error.  It's entirely possible for
12245		 * the abort and command completion to cross on the wire.
12246		 * This is more of an informative/diagnostic error.
12247		 */
12248#if 0
12249		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12250		       "%d:%d:%d:%d tag %d type %d\n",
12251		       io->io_hdr.nexus.initid.id,
12252		       io->io_hdr.nexus.targ_port,
12253		       io->io_hdr.nexus.targ_target.id,
12254		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12255		       io->taskio.tag_type);
12256#endif
12257	}
12258	return (0);
12259}
12260
12261static void
12262ctl_run_task(union ctl_io *io)
12263{
12264	struct ctl_softc *ctl_softc = control_softc;
12265	int retval = 1;
12266	const char *task_desc;
12267
12268	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12269
12270	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12271	    ("ctl_run_task: Unextected io_type %d\n",
12272	     io->io_hdr.io_type));
12273
12274	task_desc = ctl_scsi_task_string(&io->taskio);
12275	if (task_desc != NULL) {
12276#ifdef NEEDTOPORT
12277		csevent_log(CSC_CTL | CSC_SHELF_SW |
12278			    CTL_TASK_REPORT,
12279			    csevent_LogType_Trace,
12280			    csevent_Severity_Information,
12281			    csevent_AlertLevel_Green,
12282			    csevent_FRU_Firmware,
12283			    csevent_FRU_Unknown,
12284			    "CTL: received task: %s",task_desc);
12285#endif
12286	} else {
12287#ifdef NEEDTOPORT
12288		csevent_log(CSC_CTL | CSC_SHELF_SW |
12289			    CTL_TASK_REPORT,
12290			    csevent_LogType_Trace,
12291			    csevent_Severity_Information,
12292			    csevent_AlertLevel_Green,
12293			    csevent_FRU_Firmware,
12294			    csevent_FRU_Unknown,
12295			    "CTL: received unknown task "
12296			    "type: %d (%#x)",
12297			    io->taskio.task_action,
12298			    io->taskio.task_action);
12299#endif
12300	}
12301	switch (io->taskio.task_action) {
12302	case CTL_TASK_ABORT_TASK:
12303		retval = ctl_abort_task(io);
12304		break;
12305	case CTL_TASK_ABORT_TASK_SET:
12306	case CTL_TASK_CLEAR_TASK_SET:
12307		retval = ctl_abort_task_set(io);
12308		break;
12309	case CTL_TASK_CLEAR_ACA:
12310		break;
12311	case CTL_TASK_I_T_NEXUS_RESET:
12312		retval = ctl_i_t_nexus_reset(io);
12313		break;
12314	case CTL_TASK_LUN_RESET: {
12315		struct ctl_lun *lun;
12316		uint32_t targ_lun;
12317
12318		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12319		mtx_lock(&ctl_softc->ctl_lock);
12320		if ((targ_lun < CTL_MAX_LUNS)
12321		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12322			lun = ctl_softc->ctl_luns[targ_lun];
12323		else {
12324			mtx_unlock(&ctl_softc->ctl_lock);
12325			retval = 1;
12326			break;
12327		}
12328
12329		if (!(io->io_hdr.flags &
12330		    CTL_FLAG_FROM_OTHER_SC)) {
12331			union ctl_ha_msg msg_info;
12332
12333			io->io_hdr.flags |=
12334				CTL_FLAG_SENT_2OTHER_SC;
12335			msg_info.hdr.msg_type =
12336				CTL_MSG_MANAGE_TASKS;
12337			msg_info.hdr.nexus = io->io_hdr.nexus;
12338			msg_info.task.task_action =
12339				CTL_TASK_LUN_RESET;
12340			msg_info.hdr.original_sc = NULL;
12341			msg_info.hdr.serializing_sc = NULL;
12342			if (CTL_HA_STATUS_SUCCESS !=
12343			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12344			    (void *)&msg_info,
12345			    sizeof(msg_info), 0)) {
12346			}
12347		}
12348
12349		retval = ctl_lun_reset(lun, io,
12350				       CTL_UA_LUN_RESET);
12351		mtx_unlock(&ctl_softc->ctl_lock);
12352		break;
12353	}
12354	case CTL_TASK_TARGET_RESET:
12355		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12356		break;
12357	case CTL_TASK_BUS_RESET:
12358		retval = ctl_bus_reset(ctl_softc, io);
12359		break;
12360	case CTL_TASK_PORT_LOGIN:
12361		break;
12362	case CTL_TASK_PORT_LOGOUT:
12363		break;
12364	default:
12365		printf("ctl_run_task: got unknown task management event %d\n",
12366		       io->taskio.task_action);
12367		break;
12368	}
12369	if (retval == 0)
12370		io->io_hdr.status = CTL_SUCCESS;
12371	else
12372		io->io_hdr.status = CTL_ERROR;
12373	ctl_done(io);
12374}
12375
12376/*
12377 * For HA operation.  Handle commands that come in from the other
12378 * controller.
12379 */
12380static void
12381ctl_handle_isc(union ctl_io *io)
12382{
12383	int free_io;
12384	struct ctl_lun *lun;
12385	struct ctl_softc *ctl_softc;
12386	uint32_t targ_lun;
12387
12388	ctl_softc = control_softc;
12389
12390	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12391	lun = ctl_softc->ctl_luns[targ_lun];
12392
12393	switch (io->io_hdr.msg_type) {
12394	case CTL_MSG_SERIALIZE:
12395		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12396		break;
12397	case CTL_MSG_R2R: {
12398		const struct ctl_cmd_entry *entry;
12399
12400		/*
12401		 * This is only used in SER_ONLY mode.
12402		 */
12403		free_io = 0;
12404		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12405		mtx_lock(&lun->lun_lock);
12406		if (ctl_scsiio_lun_check(ctl_softc, lun,
12407		    entry, (struct ctl_scsiio *)io) != 0) {
12408			mtx_unlock(&lun->lun_lock);
12409			ctl_done(io);
12410			break;
12411		}
12412		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12413		mtx_unlock(&lun->lun_lock);
12414		ctl_enqueue_rtr(io);
12415		break;
12416	}
12417	case CTL_MSG_FINISH_IO:
12418		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12419			free_io = 0;
12420			ctl_done(io);
12421		} else {
12422			free_io = 1;
12423			mtx_lock(&lun->lun_lock);
12424			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12425				     ooa_links);
12426			ctl_check_blocked(lun);
12427			mtx_unlock(&lun->lun_lock);
12428		}
12429		break;
12430	case CTL_MSG_PERS_ACTION:
12431		ctl_hndl_per_res_out_on_other_sc(
12432			(union ctl_ha_msg *)&io->presio.pr_msg);
12433		free_io = 1;
12434		break;
12435	case CTL_MSG_BAD_JUJU:
12436		free_io = 0;
12437		ctl_done(io);
12438		break;
12439	case CTL_MSG_DATAMOVE:
12440		/* Only used in XFER mode */
12441		free_io = 0;
12442		ctl_datamove_remote(io);
12443		break;
12444	case CTL_MSG_DATAMOVE_DONE:
12445		/* Only used in XFER mode */
12446		free_io = 0;
12447		io->scsiio.be_move_done(io);
12448		break;
12449	default:
12450		free_io = 1;
12451		printf("%s: Invalid message type %d\n",
12452		       __func__, io->io_hdr.msg_type);
12453		break;
12454	}
12455	if (free_io)
12456		ctl_free_io(io);
12457
12458}
12459
12460
12461/*
12462 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12463 * there is no match.
12464 */
12465static ctl_lun_error_pattern
12466ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12467{
12468	const struct ctl_cmd_entry *entry;
12469	ctl_lun_error_pattern filtered_pattern, pattern;
12470
12471	pattern = desc->error_pattern;
12472
12473	/*
12474	 * XXX KDM we need more data passed into this function to match a
12475	 * custom pattern, and we actually need to implement custom pattern
12476	 * matching.
12477	 */
12478	if (pattern & CTL_LUN_PAT_CMD)
12479		return (CTL_LUN_PAT_CMD);
12480
12481	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12482		return (CTL_LUN_PAT_ANY);
12483
12484	entry = ctl_get_cmd_entry(ctsio, NULL);
12485
12486	filtered_pattern = entry->pattern & pattern;
12487
12488	/*
12489	 * If the user requested specific flags in the pattern (e.g.
12490	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12491	 * flags.
12492	 *
12493	 * If the user did not specify any flags, it doesn't matter whether
12494	 * or not the command supports the flags.
12495	 */
12496	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12497	     (pattern & ~CTL_LUN_PAT_MASK))
12498		return (CTL_LUN_PAT_NONE);
12499
12500	/*
12501	 * If the user asked for a range check, see if the requested LBA
12502	 * range overlaps with this command's LBA range.
12503	 */
12504	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12505		uint64_t lba1;
12506		uint64_t len1;
12507		ctl_action action;
12508		int retval;
12509
12510		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12511		if (retval != 0)
12512			return (CTL_LUN_PAT_NONE);
12513
12514		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12515					      desc->lba_range.len);
12516		/*
12517		 * A "pass" means that the LBA ranges don't overlap, so
12518		 * this doesn't match the user's range criteria.
12519		 */
12520		if (action == CTL_ACTION_PASS)
12521			return (CTL_LUN_PAT_NONE);
12522	}
12523
12524	return (filtered_pattern);
12525}
12526
12527static void
12528ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12529{
12530	struct ctl_error_desc *desc, *desc2;
12531
12532	mtx_assert(&lun->lun_lock, MA_OWNED);
12533
12534	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12535		ctl_lun_error_pattern pattern;
12536		/*
12537		 * Check to see whether this particular command matches
12538		 * the pattern in the descriptor.
12539		 */
12540		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12541		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12542			continue;
12543
12544		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12545		case CTL_LUN_INJ_ABORTED:
12546			ctl_set_aborted(&io->scsiio);
12547			break;
12548		case CTL_LUN_INJ_MEDIUM_ERR:
12549			ctl_set_medium_error(&io->scsiio);
12550			break;
12551		case CTL_LUN_INJ_UA:
12552			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12553			 * OCCURRED */
12554			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12555			break;
12556		case CTL_LUN_INJ_CUSTOM:
12557			/*
12558			 * We're assuming the user knows what he is doing.
12559			 * Just copy the sense information without doing
12560			 * checks.
12561			 */
12562			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12563			      ctl_min(sizeof(desc->custom_sense),
12564				      sizeof(io->scsiio.sense_data)));
12565			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12566			io->scsiio.sense_len = SSD_FULL_SIZE;
12567			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12568			break;
12569		case CTL_LUN_INJ_NONE:
12570		default:
12571			/*
12572			 * If this is an error injection type we don't know
12573			 * about, clear the continuous flag (if it is set)
12574			 * so it will get deleted below.
12575			 */
12576			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12577			break;
12578		}
12579		/*
12580		 * By default, each error injection action is a one-shot
12581		 */
12582		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12583			continue;
12584
12585		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12586
12587		free(desc, M_CTL);
12588	}
12589}
12590
12591#ifdef CTL_IO_DELAY
12592static void
12593ctl_datamove_timer_wakeup(void *arg)
12594{
12595	union ctl_io *io;
12596
12597	io = (union ctl_io *)arg;
12598
12599	ctl_datamove(io);
12600}
12601#endif /* CTL_IO_DELAY */
12602
12603void
12604ctl_datamove(union ctl_io *io)
12605{
12606	void (*fe_datamove)(union ctl_io *io);
12607
12608	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12609
12610	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12611
12612#ifdef CTL_TIME_IO
12613	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12614		char str[256];
12615		char path_str[64];
12616		struct sbuf sb;
12617
12618		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12619		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12620
12621		sbuf_cat(&sb, path_str);
12622		switch (io->io_hdr.io_type) {
12623		case CTL_IO_SCSI:
12624			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12625			sbuf_printf(&sb, "\n");
12626			sbuf_cat(&sb, path_str);
12627			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12628				    io->scsiio.tag_num, io->scsiio.tag_type);
12629			break;
12630		case CTL_IO_TASK:
12631			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12632				    "Tag Type: %d\n", io->taskio.task_action,
12633				    io->taskio.tag_num, io->taskio.tag_type);
12634			break;
12635		default:
12636			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12637			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12638			break;
12639		}
12640		sbuf_cat(&sb, path_str);
12641		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12642			    (intmax_t)time_uptime - io->io_hdr.start_time);
12643		sbuf_finish(&sb);
12644		printf("%s", sbuf_data(&sb));
12645	}
12646#endif /* CTL_TIME_IO */
12647
12648#ifdef CTL_IO_DELAY
12649	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12650		struct ctl_lun *lun;
12651
12652		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12653
12654		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12655	} else {
12656		struct ctl_lun *lun;
12657
12658		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12659		if ((lun != NULL)
12660		 && (lun->delay_info.datamove_delay > 0)) {
12661			struct callout *callout;
12662
12663			callout = (struct callout *)&io->io_hdr.timer_bytes;
12664			callout_init(callout, /*mpsafe*/ 1);
12665			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12666			callout_reset(callout,
12667				      lun->delay_info.datamove_delay * hz,
12668				      ctl_datamove_timer_wakeup, io);
12669			if (lun->delay_info.datamove_type ==
12670			    CTL_DELAY_TYPE_ONESHOT)
12671				lun->delay_info.datamove_delay = 0;
12672			return;
12673		}
12674	}
12675#endif
12676
12677	/*
12678	 * This command has been aborted.  Set the port status, so we fail
12679	 * the data move.
12680	 */
12681	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12682		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12683		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12684		       io->io_hdr.nexus.targ_port,
12685		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12686		       io->io_hdr.nexus.targ_lun);
12687		io->io_hdr.port_status = 31337;
12688		/*
12689		 * Note that the backend, in this case, will get the
12690		 * callback in its context.  In other cases it may get
12691		 * called in the frontend's interrupt thread context.
12692		 */
12693		io->scsiio.be_move_done(io);
12694		return;
12695	}
12696
12697	/* Don't confuse frontend with zero length data move. */
12698	if (io->scsiio.kern_data_len == 0) {
12699		io->scsiio.be_move_done(io);
12700		return;
12701	}
12702
12703	/*
12704	 * If we're in XFER mode and this I/O is from the other shelf
12705	 * controller, we need to send the DMA to the other side to
12706	 * actually transfer the data to/from the host.  In serialize only
12707	 * mode the transfer happens below CTL and ctl_datamove() is only
12708	 * called on the machine that originally received the I/O.
12709	 */
12710	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12711	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12712		union ctl_ha_msg msg;
12713		uint32_t sg_entries_sent;
12714		int do_sg_copy;
12715		int i;
12716
12717		memset(&msg, 0, sizeof(msg));
12718		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12719		msg.hdr.original_sc = io->io_hdr.original_sc;
12720		msg.hdr.serializing_sc = io;
12721		msg.hdr.nexus = io->io_hdr.nexus;
12722		msg.dt.flags = io->io_hdr.flags;
12723		/*
12724		 * We convert everything into a S/G list here.  We can't
12725		 * pass by reference, only by value between controllers.
12726		 * So we can't pass a pointer to the S/G list, only as many
12727		 * S/G entries as we can fit in here.  If it's possible for
12728		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12729		 * then we need to break this up into multiple transfers.
12730		 */
12731		if (io->scsiio.kern_sg_entries == 0) {
12732			msg.dt.kern_sg_entries = 1;
12733			/*
12734			 * If this is in cached memory, flush the cache
12735			 * before we send the DMA request to the other
12736			 * controller.  We want to do this in either the
12737			 * read or the write case.  The read case is
12738			 * straightforward.  In the write case, we want to
12739			 * make sure nothing is in the local cache that
12740			 * could overwrite the DMAed data.
12741			 */
12742			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12743				/*
12744				 * XXX KDM use bus_dmamap_sync() here.
12745				 */
12746			}
12747
12748			/*
12749			 * Convert to a physical address if this is a
12750			 * virtual address.
12751			 */
12752			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12753				msg.dt.sg_list[0].addr =
12754					io->scsiio.kern_data_ptr;
12755			} else {
12756				/*
12757				 * XXX KDM use busdma here!
12758				 */
12759#if 0
12760				msg.dt.sg_list[0].addr = (void *)
12761					vtophys(io->scsiio.kern_data_ptr);
12762#endif
12763			}
12764
12765			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12766			do_sg_copy = 0;
12767		} else {
12768			struct ctl_sg_entry *sgl;
12769
12770			do_sg_copy = 1;
12771			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12772			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12773			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12774				/*
12775				 * XXX KDM use bus_dmamap_sync() here.
12776				 */
12777			}
12778		}
12779
12780		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12781		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12782		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12783		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12784		msg.dt.sg_sequence = 0;
12785
12786		/*
12787		 * Loop until we've sent all of the S/G entries.  On the
12788		 * other end, we'll recompose these S/G entries into one
12789		 * contiguous list before passing it to the
12790		 */
12791		for (sg_entries_sent = 0; sg_entries_sent <
12792		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12793			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12794				sizeof(msg.dt.sg_list[0])),
12795				msg.dt.kern_sg_entries - sg_entries_sent);
12796
12797			if (do_sg_copy != 0) {
12798				struct ctl_sg_entry *sgl;
12799				int j;
12800
12801				sgl = (struct ctl_sg_entry *)
12802					io->scsiio.kern_data_ptr;
12803				/*
12804				 * If this is in cached memory, flush the cache
12805				 * before we send the DMA request to the other
12806				 * controller.  We want to do this in either
12807				 * the * read or the write case.  The read
12808				 * case is straightforward.  In the write
12809				 * case, we want to make sure nothing is
12810				 * in the local cache that could overwrite
12811				 * the DMAed data.
12812				 */
12813
12814				for (i = sg_entries_sent, j = 0;
12815				     i < msg.dt.cur_sg_entries; i++, j++) {
12816					if ((io->io_hdr.flags &
12817					     CTL_FLAG_NO_DATASYNC) == 0) {
12818						/*
12819						 * XXX KDM use bus_dmamap_sync()
12820						 */
12821					}
12822					if ((io->io_hdr.flags &
12823					     CTL_FLAG_BUS_ADDR) == 0) {
12824						/*
12825						 * XXX KDM use busdma.
12826						 */
12827#if 0
12828						msg.dt.sg_list[j].addr =(void *)
12829						       vtophys(sgl[i].addr);
12830#endif
12831					} else {
12832						msg.dt.sg_list[j].addr =
12833							sgl[i].addr;
12834					}
12835					msg.dt.sg_list[j].len = sgl[i].len;
12836				}
12837			}
12838
12839			sg_entries_sent += msg.dt.cur_sg_entries;
12840			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12841				msg.dt.sg_last = 1;
12842			else
12843				msg.dt.sg_last = 0;
12844
12845			/*
12846			 * XXX KDM drop and reacquire the lock here?
12847			 */
12848			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12849			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12850				/*
12851				 * XXX do something here.
12852				 */
12853			}
12854
12855			msg.dt.sent_sg_entries = sg_entries_sent;
12856		}
12857		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12858		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12859			ctl_failover_io(io, /*have_lock*/ 0);
12860
12861	} else {
12862
12863		/*
12864		 * Lookup the fe_datamove() function for this particular
12865		 * front end.
12866		 */
12867		fe_datamove =
12868		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12869
12870		fe_datamove(io);
12871	}
12872}
12873
12874static void
12875ctl_send_datamove_done(union ctl_io *io, int have_lock)
12876{
12877	union ctl_ha_msg msg;
12878	int isc_status;
12879
12880	memset(&msg, 0, sizeof(msg));
12881
12882	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12883	msg.hdr.original_sc = io;
12884	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12885	msg.hdr.nexus = io->io_hdr.nexus;
12886	msg.hdr.status = io->io_hdr.status;
12887	msg.scsi.tag_num = io->scsiio.tag_num;
12888	msg.scsi.tag_type = io->scsiio.tag_type;
12889	msg.scsi.scsi_status = io->scsiio.scsi_status;
12890	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12891	       sizeof(io->scsiio.sense_data));
12892	msg.scsi.sense_len = io->scsiio.sense_len;
12893	msg.scsi.sense_residual = io->scsiio.sense_residual;
12894	msg.scsi.fetd_status = io->io_hdr.port_status;
12895	msg.scsi.residual = io->scsiio.residual;
12896	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12897
12898	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12899		ctl_failover_io(io, /*have_lock*/ have_lock);
12900		return;
12901	}
12902
12903	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12904	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12905		/* XXX do something if this fails */
12906	}
12907
12908}
12909
12910/*
12911 * The DMA to the remote side is done, now we need to tell the other side
12912 * we're done so it can continue with its data movement.
12913 */
12914static void
12915ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12916{
12917	union ctl_io *io;
12918
12919	io = rq->context;
12920
12921	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12922		printf("%s: ISC DMA write failed with error %d", __func__,
12923		       rq->ret);
12924		ctl_set_internal_failure(&io->scsiio,
12925					 /*sks_valid*/ 1,
12926					 /*retry_count*/ rq->ret);
12927	}
12928
12929	ctl_dt_req_free(rq);
12930
12931	/*
12932	 * In this case, we had to malloc the memory locally.  Free it.
12933	 */
12934	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12935		int i;
12936		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12937			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12938	}
12939	/*
12940	 * The data is in local and remote memory, so now we need to send
12941	 * status (good or back) back to the other side.
12942	 */
12943	ctl_send_datamove_done(io, /*have_lock*/ 0);
12944}
12945
12946/*
12947 * We've moved the data from the host/controller into local memory.  Now we
12948 * need to push it over to the remote controller's memory.
12949 */
12950static int
12951ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12952{
12953	int retval;
12954
12955	retval = 0;
12956
12957	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12958					  ctl_datamove_remote_write_cb);
12959
12960	return (retval);
12961}
12962
12963static void
12964ctl_datamove_remote_write(union ctl_io *io)
12965{
12966	int retval;
12967	void (*fe_datamove)(union ctl_io *io);
12968
12969	/*
12970	 * - Get the data from the host/HBA into local memory.
12971	 * - DMA memory from the local controller to the remote controller.
12972	 * - Send status back to the remote controller.
12973	 */
12974
12975	retval = ctl_datamove_remote_sgl_setup(io);
12976	if (retval != 0)
12977		return;
12978
12979	/* Switch the pointer over so the FETD knows what to do */
12980	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12981
12982	/*
12983	 * Use a custom move done callback, since we need to send completion
12984	 * back to the other controller, not to the backend on this side.
12985	 */
12986	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12987
12988	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12989
12990	fe_datamove(io);
12991
12992	return;
12993
12994}
12995
12996static int
12997ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12998{
12999#if 0
13000	char str[256];
13001	char path_str[64];
13002	struct sbuf sb;
13003#endif
13004
13005	/*
13006	 * In this case, we had to malloc the memory locally.  Free it.
13007	 */
13008	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13009		int i;
13010		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13011			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13012	}
13013
13014#if 0
13015	scsi_path_string(io, path_str, sizeof(path_str));
13016	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13017	sbuf_cat(&sb, path_str);
13018	scsi_command_string(&io->scsiio, NULL, &sb);
13019	sbuf_printf(&sb, "\n");
13020	sbuf_cat(&sb, path_str);
13021	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13022		    io->scsiio.tag_num, io->scsiio.tag_type);
13023	sbuf_cat(&sb, path_str);
13024	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13025		    io->io_hdr.flags, io->io_hdr.status);
13026	sbuf_finish(&sb);
13027	printk("%s", sbuf_data(&sb));
13028#endif
13029
13030
13031	/*
13032	 * The read is done, now we need to send status (good or bad) back
13033	 * to the other side.
13034	 */
13035	ctl_send_datamove_done(io, /*have_lock*/ 0);
13036
13037	return (0);
13038}
13039
13040static void
13041ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13042{
13043	union ctl_io *io;
13044	void (*fe_datamove)(union ctl_io *io);
13045
13046	io = rq->context;
13047
13048	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13049		printf("%s: ISC DMA read failed with error %d", __func__,
13050		       rq->ret);
13051		ctl_set_internal_failure(&io->scsiio,
13052					 /*sks_valid*/ 1,
13053					 /*retry_count*/ rq->ret);
13054	}
13055
13056	ctl_dt_req_free(rq);
13057
13058	/* Switch the pointer over so the FETD knows what to do */
13059	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13060
13061	/*
13062	 * Use a custom move done callback, since we need to send completion
13063	 * back to the other controller, not to the backend on this side.
13064	 */
13065	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13066
13067	/* XXX KDM add checks like the ones in ctl_datamove? */
13068
13069	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13070
13071	fe_datamove(io);
13072}
13073
13074static int
13075ctl_datamove_remote_sgl_setup(union ctl_io *io)
13076{
13077	struct ctl_sg_entry *local_sglist, *remote_sglist;
13078	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13079	struct ctl_softc *softc;
13080	int retval;
13081	int i;
13082
13083	retval = 0;
13084	softc = control_softc;
13085
13086	local_sglist = io->io_hdr.local_sglist;
13087	local_dma_sglist = io->io_hdr.local_dma_sglist;
13088	remote_sglist = io->io_hdr.remote_sglist;
13089	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13090
13091	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13092		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13093			local_sglist[i].len = remote_sglist[i].len;
13094
13095			/*
13096			 * XXX Detect the situation where the RS-level I/O
13097			 * redirector on the other side has already read the
13098			 * data off of the AOR RS on this side, and
13099			 * transferred it to remote (mirror) memory on the
13100			 * other side.  Since we already have the data in
13101			 * memory here, we just need to use it.
13102			 *
13103			 * XXX KDM this can probably be removed once we
13104			 * get the cache device code in and take the
13105			 * current AOR implementation out.
13106			 */
13107#ifdef NEEDTOPORT
13108			if ((remote_sglist[i].addr >=
13109			     (void *)vtophys(softc->mirr->addr))
13110			 && (remote_sglist[i].addr <
13111			     ((void *)vtophys(softc->mirr->addr) +
13112			     CacheMirrorOffset))) {
13113				local_sglist[i].addr = remote_sglist[i].addr -
13114					CacheMirrorOffset;
13115				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13116				     CTL_FLAG_DATA_IN)
13117					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13118			} else {
13119				local_sglist[i].addr = remote_sglist[i].addr +
13120					CacheMirrorOffset;
13121			}
13122#endif
13123#if 0
13124			printf("%s: local %p, remote %p, len %d\n",
13125			       __func__, local_sglist[i].addr,
13126			       remote_sglist[i].addr, local_sglist[i].len);
13127#endif
13128		}
13129	} else {
13130		uint32_t len_to_go;
13131
13132		/*
13133		 * In this case, we don't have automatically allocated
13134		 * memory for this I/O on this controller.  This typically
13135		 * happens with internal CTL I/O -- e.g. inquiry, mode
13136		 * sense, etc.  Anything coming from RAIDCore will have
13137		 * a mirror area available.
13138		 */
13139		len_to_go = io->scsiio.kern_data_len;
13140
13141		/*
13142		 * Clear the no datasync flag, we have to use malloced
13143		 * buffers.
13144		 */
13145		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13146
13147		/*
13148		 * The difficult thing here is that the size of the various
13149		 * S/G segments may be different than the size from the
13150		 * remote controller.  That'll make it harder when DMAing
13151		 * the data back to the other side.
13152		 */
13153		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13154		     sizeof(io->io_hdr.remote_sglist[0])) &&
13155		     (len_to_go > 0); i++) {
13156			local_sglist[i].len = ctl_min(len_to_go, 131072);
13157			CTL_SIZE_8B(local_dma_sglist[i].len,
13158				    local_sglist[i].len);
13159			local_sglist[i].addr =
13160				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13161
13162			local_dma_sglist[i].addr = local_sglist[i].addr;
13163
13164			if (local_sglist[i].addr == NULL) {
13165				int j;
13166
13167				printf("malloc failed for %zd bytes!",
13168				       local_dma_sglist[i].len);
13169				for (j = 0; j < i; j++) {
13170					free(local_sglist[j].addr, M_CTL);
13171				}
13172				ctl_set_internal_failure(&io->scsiio,
13173							 /*sks_valid*/ 1,
13174							 /*retry_count*/ 4857);
13175				retval = 1;
13176				goto bailout_error;
13177
13178			}
13179			/* XXX KDM do we need a sync here? */
13180
13181			len_to_go -= local_sglist[i].len;
13182		}
13183		/*
13184		 * Reset the number of S/G entries accordingly.  The
13185		 * original number of S/G entries is available in
13186		 * rem_sg_entries.
13187		 */
13188		io->scsiio.kern_sg_entries = i;
13189
13190#if 0
13191		printf("%s: kern_sg_entries = %d\n", __func__,
13192		       io->scsiio.kern_sg_entries);
13193		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13194			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13195			       local_sglist[i].addr, local_sglist[i].len,
13196			       local_dma_sglist[i].len);
13197#endif
13198	}
13199
13200
13201	return (retval);
13202
13203bailout_error:
13204
13205	ctl_send_datamove_done(io, /*have_lock*/ 0);
13206
13207	return (retval);
13208}
13209
13210static int
13211ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13212			 ctl_ha_dt_cb callback)
13213{
13214	struct ctl_ha_dt_req *rq;
13215	struct ctl_sg_entry *remote_sglist, *local_sglist;
13216	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13217	uint32_t local_used, remote_used, total_used;
13218	int retval;
13219	int i, j;
13220
13221	retval = 0;
13222
13223	rq = ctl_dt_req_alloc();
13224
13225	/*
13226	 * If we failed to allocate the request, and if the DMA didn't fail
13227	 * anyway, set busy status.  This is just a resource allocation
13228	 * failure.
13229	 */
13230	if ((rq == NULL)
13231	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13232		ctl_set_busy(&io->scsiio);
13233
13234	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13235
13236		if (rq != NULL)
13237			ctl_dt_req_free(rq);
13238
13239		/*
13240		 * The data move failed.  We need to return status back
13241		 * to the other controller.  No point in trying to DMA
13242		 * data to the remote controller.
13243		 */
13244
13245		ctl_send_datamove_done(io, /*have_lock*/ 0);
13246
13247		retval = 1;
13248
13249		goto bailout;
13250	}
13251
13252	local_sglist = io->io_hdr.local_sglist;
13253	local_dma_sglist = io->io_hdr.local_dma_sglist;
13254	remote_sglist = io->io_hdr.remote_sglist;
13255	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13256	local_used = 0;
13257	remote_used = 0;
13258	total_used = 0;
13259
13260	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13261		rq->ret = CTL_HA_STATUS_SUCCESS;
13262		rq->context = io;
13263		callback(rq);
13264		goto bailout;
13265	}
13266
13267	/*
13268	 * Pull/push the data over the wire from/to the other controller.
13269	 * This takes into account the possibility that the local and
13270	 * remote sglists may not be identical in terms of the size of
13271	 * the elements and the number of elements.
13272	 *
13273	 * One fundamental assumption here is that the length allocated for
13274	 * both the local and remote sglists is identical.  Otherwise, we've
13275	 * essentially got a coding error of some sort.
13276	 */
13277	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13278		int isc_ret;
13279		uint32_t cur_len, dma_length;
13280		uint8_t *tmp_ptr;
13281
13282		rq->id = CTL_HA_DATA_CTL;
13283		rq->command = command;
13284		rq->context = io;
13285
13286		/*
13287		 * Both pointers should be aligned.  But it is possible
13288		 * that the allocation length is not.  They should both
13289		 * also have enough slack left over at the end, though,
13290		 * to round up to the next 8 byte boundary.
13291		 */
13292		cur_len = ctl_min(local_sglist[i].len - local_used,
13293				  remote_sglist[j].len - remote_used);
13294
13295		/*
13296		 * In this case, we have a size issue and need to decrease
13297		 * the size, except in the case where we actually have less
13298		 * than 8 bytes left.  In that case, we need to increase
13299		 * the DMA length to get the last bit.
13300		 */
13301		if ((cur_len & 0x7) != 0) {
13302			if (cur_len > 0x7) {
13303				cur_len = cur_len - (cur_len & 0x7);
13304				dma_length = cur_len;
13305			} else {
13306				CTL_SIZE_8B(dma_length, cur_len);
13307			}
13308
13309		} else
13310			dma_length = cur_len;
13311
13312		/*
13313		 * If we had to allocate memory for this I/O, instead of using
13314		 * the non-cached mirror memory, we'll need to flush the cache
13315		 * before trying to DMA to the other controller.
13316		 *
13317		 * We could end up doing this multiple times for the same
13318		 * segment if we have a larger local segment than remote
13319		 * segment.  That shouldn't be an issue.
13320		 */
13321		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13322			/*
13323			 * XXX KDM use bus_dmamap_sync() here.
13324			 */
13325		}
13326
13327		rq->size = dma_length;
13328
13329		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13330		tmp_ptr += local_used;
13331
13332		/* Use physical addresses when talking to ISC hardware */
13333		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13334			/* XXX KDM use busdma */
13335#if 0
13336			rq->local = vtophys(tmp_ptr);
13337#endif
13338		} else
13339			rq->local = tmp_ptr;
13340
13341		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13342		tmp_ptr += remote_used;
13343		rq->remote = tmp_ptr;
13344
13345		rq->callback = NULL;
13346
13347		local_used += cur_len;
13348		if (local_used >= local_sglist[i].len) {
13349			i++;
13350			local_used = 0;
13351		}
13352
13353		remote_used += cur_len;
13354		if (remote_used >= remote_sglist[j].len) {
13355			j++;
13356			remote_used = 0;
13357		}
13358		total_used += cur_len;
13359
13360		if (total_used >= io->scsiio.kern_data_len)
13361			rq->callback = callback;
13362
13363		if ((rq->size & 0x7) != 0) {
13364			printf("%s: warning: size %d is not on 8b boundary\n",
13365			       __func__, rq->size);
13366		}
13367		if (((uintptr_t)rq->local & 0x7) != 0) {
13368			printf("%s: warning: local %p not on 8b boundary\n",
13369			       __func__, rq->local);
13370		}
13371		if (((uintptr_t)rq->remote & 0x7) != 0) {
13372			printf("%s: warning: remote %p not on 8b boundary\n",
13373			       __func__, rq->local);
13374		}
13375#if 0
13376		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13377		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13378		       rq->local, rq->remote, rq->size);
13379#endif
13380
13381		isc_ret = ctl_dt_single(rq);
13382		if (isc_ret == CTL_HA_STATUS_WAIT)
13383			continue;
13384
13385		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13386			rq->ret = CTL_HA_STATUS_SUCCESS;
13387		} else {
13388			rq->ret = isc_ret;
13389		}
13390		callback(rq);
13391		goto bailout;
13392	}
13393
13394bailout:
13395	return (retval);
13396
13397}
13398
13399static void
13400ctl_datamove_remote_read(union ctl_io *io)
13401{
13402	int retval;
13403	int i;
13404
13405	/*
13406	 * This will send an error to the other controller in the case of a
13407	 * failure.
13408	 */
13409	retval = ctl_datamove_remote_sgl_setup(io);
13410	if (retval != 0)
13411		return;
13412
13413	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13414					  ctl_datamove_remote_read_cb);
13415	if ((retval != 0)
13416	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13417		/*
13418		 * Make sure we free memory if there was an error..  The
13419		 * ctl_datamove_remote_xfer() function will send the
13420		 * datamove done message, or call the callback with an
13421		 * error if there is a problem.
13422		 */
13423		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13424			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13425	}
13426
13427	return;
13428}
13429
13430/*
13431 * Process a datamove request from the other controller.  This is used for
13432 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13433 * first.  Once that is complete, the data gets DMAed into the remote
13434 * controller's memory.  For reads, we DMA from the remote controller's
13435 * memory into our memory first, and then move it out to the FETD.
13436 */
13437static void
13438ctl_datamove_remote(union ctl_io *io)
13439{
13440	struct ctl_softc *softc;
13441
13442	softc = control_softc;
13443
13444	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13445
13446	/*
13447	 * Note that we look for an aborted I/O here, but don't do some of
13448	 * the other checks that ctl_datamove() normally does.
13449	 * We don't need to run the datamove delay code, since that should
13450	 * have been done if need be on the other controller.
13451	 */
13452	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13453		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13454		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13455		       io->io_hdr.nexus.targ_port,
13456		       io->io_hdr.nexus.targ_target.id,
13457		       io->io_hdr.nexus.targ_lun);
13458		io->io_hdr.port_status = 31338;
13459		ctl_send_datamove_done(io, /*have_lock*/ 0);
13460		return;
13461	}
13462
13463	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13464		ctl_datamove_remote_write(io);
13465	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13466		ctl_datamove_remote_read(io);
13467	} else {
13468		union ctl_ha_msg msg;
13469		struct scsi_sense_data *sense;
13470		uint8_t sks[3];
13471		int retry_count;
13472
13473		memset(&msg, 0, sizeof(msg));
13474
13475		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13476		msg.hdr.status = CTL_SCSI_ERROR;
13477		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13478
13479		retry_count = 4243;
13480
13481		sense = &msg.scsi.sense_data;
13482		sks[0] = SSD_SCS_VALID;
13483		sks[1] = (retry_count >> 8) & 0xff;
13484		sks[2] = retry_count & 0xff;
13485
13486		/* "Internal target failure" */
13487		scsi_set_sense_data(sense,
13488				    /*sense_format*/ SSD_TYPE_NONE,
13489				    /*current_error*/ 1,
13490				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13491				    /*asc*/ 0x44,
13492				    /*ascq*/ 0x00,
13493				    /*type*/ SSD_ELEM_SKS,
13494				    /*size*/ sizeof(sks),
13495				    /*data*/ sks,
13496				    SSD_ELEM_NONE);
13497
13498		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13499		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13500			ctl_failover_io(io, /*have_lock*/ 1);
13501			return;
13502		}
13503
13504		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13505		    CTL_HA_STATUS_SUCCESS) {
13506			/* XXX KDM what to do if this fails? */
13507		}
13508		return;
13509	}
13510
13511}
13512
13513static int
13514ctl_process_done(union ctl_io *io)
13515{
13516	struct ctl_lun *lun;
13517	struct ctl_softc *ctl_softc;
13518	void (*fe_done)(union ctl_io *io);
13519	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13520
13521	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13522
13523	fe_done =
13524	    control_softc->ctl_ports[targ_port]->fe_done;
13525
13526#ifdef CTL_TIME_IO
13527	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13528		char str[256];
13529		char path_str[64];
13530		struct sbuf sb;
13531
13532		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13533		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13534
13535		sbuf_cat(&sb, path_str);
13536		switch (io->io_hdr.io_type) {
13537		case CTL_IO_SCSI:
13538			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13539			sbuf_printf(&sb, "\n");
13540			sbuf_cat(&sb, path_str);
13541			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13542				    io->scsiio.tag_num, io->scsiio.tag_type);
13543			break;
13544		case CTL_IO_TASK:
13545			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13546				    "Tag Type: %d\n", io->taskio.task_action,
13547				    io->taskio.tag_num, io->taskio.tag_type);
13548			break;
13549		default:
13550			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13551			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13552			break;
13553		}
13554		sbuf_cat(&sb, path_str);
13555		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13556			    (intmax_t)time_uptime - io->io_hdr.start_time);
13557		sbuf_finish(&sb);
13558		printf("%s", sbuf_data(&sb));
13559	}
13560#endif /* CTL_TIME_IO */
13561
13562	switch (io->io_hdr.io_type) {
13563	case CTL_IO_SCSI:
13564		break;
13565	case CTL_IO_TASK:
13566		if (bootverbose || (ctl_debug & CTL_DEBUG_INFO))
13567			ctl_io_error_print(io, NULL);
13568		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13569			ctl_free_io(io);
13570		else
13571			fe_done(io);
13572		return (CTL_RETVAL_COMPLETE);
13573	default:
13574		panic("ctl_process_done: invalid io type %d\n",
13575		      io->io_hdr.io_type);
13576		break; /* NOTREACHED */
13577	}
13578
13579	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13580	if (lun == NULL) {
13581		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13582				 io->io_hdr.nexus.targ_mapped_lun));
13583		fe_done(io);
13584		goto bailout;
13585	}
13586	ctl_softc = lun->ctl_softc;
13587
13588	mtx_lock(&lun->lun_lock);
13589
13590	/*
13591	 * Check to see if we have any errors to inject here.  We only
13592	 * inject errors for commands that don't already have errors set.
13593	 */
13594	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13595	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13596		ctl_inject_error(lun, io);
13597
13598	/*
13599	 * XXX KDM how do we treat commands that aren't completed
13600	 * successfully?
13601	 *
13602	 * XXX KDM should we also track I/O latency?
13603	 */
13604	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13605	    io->io_hdr.io_type == CTL_IO_SCSI) {
13606#ifdef CTL_TIME_IO
13607		struct bintime cur_bt;
13608#endif
13609		int type;
13610
13611		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13612		    CTL_FLAG_DATA_IN)
13613			type = CTL_STATS_READ;
13614		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13615		    CTL_FLAG_DATA_OUT)
13616			type = CTL_STATS_WRITE;
13617		else
13618			type = CTL_STATS_NO_IO;
13619
13620		lun->stats.ports[targ_port].bytes[type] +=
13621		    io->scsiio.kern_total_len;
13622		lun->stats.ports[targ_port].operations[type]++;
13623#ifdef CTL_TIME_IO
13624		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13625		   &io->io_hdr.dma_bt);
13626		lun->stats.ports[targ_port].num_dmas[type] +=
13627		    io->io_hdr.num_dmas;
13628		getbintime(&cur_bt);
13629		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13630		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13631#endif
13632	}
13633
13634	/*
13635	 * Remove this from the OOA queue.
13636	 */
13637	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13638
13639	/*
13640	 * Run through the blocked queue on this LUN and see if anything
13641	 * has become unblocked, now that this transaction is done.
13642	 */
13643	ctl_check_blocked(lun);
13644
13645	/*
13646	 * If the LUN has been invalidated, free it if there is nothing
13647	 * left on its OOA queue.
13648	 */
13649	if ((lun->flags & CTL_LUN_INVALID)
13650	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13651		mtx_unlock(&lun->lun_lock);
13652		mtx_lock(&ctl_softc->ctl_lock);
13653		ctl_free_lun(lun);
13654		mtx_unlock(&ctl_softc->ctl_lock);
13655	} else
13656		mtx_unlock(&lun->lun_lock);
13657
13658	/*
13659	 * If this command has been aborted, make sure we set the status
13660	 * properly.  The FETD is responsible for freeing the I/O and doing
13661	 * whatever it needs to do to clean up its state.
13662	 */
13663	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13664		ctl_set_task_aborted(&io->scsiio);
13665
13666	/*
13667	 * If enabled, print command error status.
13668	 * We don't print UAs unless debugging was enabled explicitly.
13669	 */
13670	do {
13671		if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
13672			break;
13673		if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0)
13674			break;
13675		if ((ctl_debug & CTL_DEBUG_INFO) == 0 &&
13676		    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) &&
13677		     (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13678			int error_code, sense_key, asc, ascq;
13679
13680			scsi_extract_sense_len(&io->scsiio.sense_data,
13681			    io->scsiio.sense_len, &error_code, &sense_key,
13682			    &asc, &ascq, /*show_errors*/ 0);
13683			if (sense_key == SSD_KEY_UNIT_ATTENTION)
13684				break;
13685		}
13686
13687		ctl_io_error_print(io, NULL);
13688	} while (0);
13689
13690	/*
13691	 * Tell the FETD or the other shelf controller we're done with this
13692	 * command.  Note that only SCSI commands get to this point.  Task
13693	 * management commands are completed above.
13694	 *
13695	 * We only send status to the other controller if we're in XFER
13696	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13697	 * received the I/O (from CTL's perspective), and so the status is
13698	 * generated there.
13699	 *
13700	 * XXX KDM if we hold the lock here, we could cause a deadlock
13701	 * if the frontend comes back in in this context to queue
13702	 * something.
13703	 */
13704	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13705	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13706		union ctl_ha_msg msg;
13707
13708		memset(&msg, 0, sizeof(msg));
13709		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13710		msg.hdr.original_sc = io->io_hdr.original_sc;
13711		msg.hdr.nexus = io->io_hdr.nexus;
13712		msg.hdr.status = io->io_hdr.status;
13713		msg.scsi.scsi_status = io->scsiio.scsi_status;
13714		msg.scsi.tag_num = io->scsiio.tag_num;
13715		msg.scsi.tag_type = io->scsiio.tag_type;
13716		msg.scsi.sense_len = io->scsiio.sense_len;
13717		msg.scsi.sense_residual = io->scsiio.sense_residual;
13718		msg.scsi.residual = io->scsiio.residual;
13719		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13720		       sizeof(io->scsiio.sense_data));
13721		/*
13722		 * We copy this whether or not this is an I/O-related
13723		 * command.  Otherwise, we'd have to go and check to see
13724		 * whether it's a read/write command, and it really isn't
13725		 * worth it.
13726		 */
13727		memcpy(&msg.scsi.lbalen,
13728		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13729		       sizeof(msg.scsi.lbalen));
13730
13731		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13732				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13733			/* XXX do something here */
13734		}
13735
13736		ctl_free_io(io);
13737	} else
13738		fe_done(io);
13739
13740bailout:
13741
13742	return (CTL_RETVAL_COMPLETE);
13743}
13744
13745#ifdef CTL_WITH_CA
13746/*
13747 * Front end should call this if it doesn't do autosense.  When the request
13748 * sense comes back in from the initiator, we'll dequeue this and send it.
13749 */
13750int
13751ctl_queue_sense(union ctl_io *io)
13752{
13753	struct ctl_lun *lun;
13754	struct ctl_softc *ctl_softc;
13755	uint32_t initidx, targ_lun;
13756
13757	ctl_softc = control_softc;
13758
13759	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13760
13761	/*
13762	 * LUN lookup will likely move to the ctl_work_thread() once we
13763	 * have our new queueing infrastructure (that doesn't put things on
13764	 * a per-LUN queue initially).  That is so that we can handle
13765	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13766	 * can't deal with that right now.
13767	 */
13768	mtx_lock(&ctl_softc->ctl_lock);
13769
13770	/*
13771	 * If we don't have a LUN for this, just toss the sense
13772	 * information.
13773	 */
13774	targ_lun = io->io_hdr.nexus.targ_lun;
13775	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13776	if ((targ_lun < CTL_MAX_LUNS)
13777	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13778		lun = ctl_softc->ctl_luns[targ_lun];
13779	else
13780		goto bailout;
13781
13782	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13783
13784	mtx_lock(&lun->lun_lock);
13785	/*
13786	 * Already have CA set for this LUN...toss the sense information.
13787	 */
13788	if (ctl_is_set(lun->have_ca, initidx)) {
13789		mtx_unlock(&lun->lun_lock);
13790		goto bailout;
13791	}
13792
13793	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13794	       ctl_min(sizeof(lun->pending_sense[initidx]),
13795	       sizeof(io->scsiio.sense_data)));
13796	ctl_set_mask(lun->have_ca, initidx);
13797	mtx_unlock(&lun->lun_lock);
13798
13799bailout:
13800	mtx_unlock(&ctl_softc->ctl_lock);
13801
13802	ctl_free_io(io);
13803
13804	return (CTL_RETVAL_COMPLETE);
13805}
13806#endif
13807
13808/*
13809 * Primary command inlet from frontend ports.  All SCSI and task I/O
13810 * requests must go through this function.
13811 */
13812int
13813ctl_queue(union ctl_io *io)
13814{
13815	struct ctl_softc *ctl_softc;
13816
13817	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13818
13819	ctl_softc = control_softc;
13820
13821#ifdef CTL_TIME_IO
13822	io->io_hdr.start_time = time_uptime;
13823	getbintime(&io->io_hdr.start_bt);
13824#endif /* CTL_TIME_IO */
13825
13826	/* Map FE-specific LUN ID into global one. */
13827	io->io_hdr.nexus.targ_mapped_lun =
13828	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13829
13830	switch (io->io_hdr.io_type) {
13831	case CTL_IO_SCSI:
13832	case CTL_IO_TASK:
13833		if (ctl_debug & CTL_DEBUG_CDB)
13834			ctl_io_print(io);
13835		ctl_enqueue_incoming(io);
13836		break;
13837	default:
13838		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13839		return (EINVAL);
13840	}
13841
13842	return (CTL_RETVAL_COMPLETE);
13843}
13844
13845#ifdef CTL_IO_DELAY
13846static void
13847ctl_done_timer_wakeup(void *arg)
13848{
13849	union ctl_io *io;
13850
13851	io = (union ctl_io *)arg;
13852	ctl_done(io);
13853}
13854#endif /* CTL_IO_DELAY */
13855
13856void
13857ctl_done(union ctl_io *io)
13858{
13859	struct ctl_softc *ctl_softc;
13860
13861	ctl_softc = control_softc;
13862
13863	/*
13864	 * Enable this to catch duplicate completion issues.
13865	 */
13866#if 0
13867	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13868		printf("%s: type %d msg %d cdb %x iptl: "
13869		       "%d:%d:%d:%d tag 0x%04x "
13870		       "flag %#x status %x\n",
13871			__func__,
13872			io->io_hdr.io_type,
13873			io->io_hdr.msg_type,
13874			io->scsiio.cdb[0],
13875			io->io_hdr.nexus.initid.id,
13876			io->io_hdr.nexus.targ_port,
13877			io->io_hdr.nexus.targ_target.id,
13878			io->io_hdr.nexus.targ_lun,
13879			(io->io_hdr.io_type ==
13880			CTL_IO_TASK) ?
13881			io->taskio.tag_num :
13882			io->scsiio.tag_num,
13883		        io->io_hdr.flags,
13884			io->io_hdr.status);
13885	} else
13886		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13887#endif
13888
13889	/*
13890	 * This is an internal copy of an I/O, and should not go through
13891	 * the normal done processing logic.
13892	 */
13893	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13894		return;
13895
13896	/*
13897	 * We need to send a msg to the serializing shelf to finish the IO
13898	 * as well.  We don't send a finish message to the other shelf if
13899	 * this is a task management command.  Task management commands
13900	 * aren't serialized in the OOA queue, but rather just executed on
13901	 * both shelf controllers for commands that originated on that
13902	 * controller.
13903	 */
13904	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13905	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13906		union ctl_ha_msg msg_io;
13907
13908		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13909		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13910		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13911		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13912		}
13913		/* continue on to finish IO */
13914	}
13915#ifdef CTL_IO_DELAY
13916	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13917		struct ctl_lun *lun;
13918
13919		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13920
13921		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13922	} else {
13923		struct ctl_lun *lun;
13924
13925		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13926
13927		if ((lun != NULL)
13928		 && (lun->delay_info.done_delay > 0)) {
13929			struct callout *callout;
13930
13931			callout = (struct callout *)&io->io_hdr.timer_bytes;
13932			callout_init(callout, /*mpsafe*/ 1);
13933			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13934			callout_reset(callout,
13935				      lun->delay_info.done_delay * hz,
13936				      ctl_done_timer_wakeup, io);
13937			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13938				lun->delay_info.done_delay = 0;
13939			return;
13940		}
13941	}
13942#endif /* CTL_IO_DELAY */
13943
13944	ctl_enqueue_done(io);
13945}
13946
13947int
13948ctl_isc(struct ctl_scsiio *ctsio)
13949{
13950	struct ctl_lun *lun;
13951	int retval;
13952
13953	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13954
13955	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13956
13957	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13958
13959	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13960
13961	return (retval);
13962}
13963
13964
13965static void
13966ctl_work_thread(void *arg)
13967{
13968	struct ctl_thread *thr = (struct ctl_thread *)arg;
13969	struct ctl_softc *softc = thr->ctl_softc;
13970	union ctl_io *io;
13971	int retval;
13972
13973	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13974
13975	for (;;) {
13976		retval = 0;
13977
13978		/*
13979		 * We handle the queues in this order:
13980		 * - ISC
13981		 * - done queue (to free up resources, unblock other commands)
13982		 * - RtR queue
13983		 * - incoming queue
13984		 *
13985		 * If those queues are empty, we break out of the loop and
13986		 * go to sleep.
13987		 */
13988		mtx_lock(&thr->queue_lock);
13989		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13990		if (io != NULL) {
13991			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13992			mtx_unlock(&thr->queue_lock);
13993			ctl_handle_isc(io);
13994			continue;
13995		}
13996		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13997		if (io != NULL) {
13998			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13999			/* clear any blocked commands, call fe_done */
14000			mtx_unlock(&thr->queue_lock);
14001			retval = ctl_process_done(io);
14002			continue;
14003		}
14004		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14005		if (io != NULL) {
14006			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14007			mtx_unlock(&thr->queue_lock);
14008			if (io->io_hdr.io_type == CTL_IO_TASK)
14009				ctl_run_task(io);
14010			else
14011				ctl_scsiio_precheck(softc, &io->scsiio);
14012			continue;
14013		}
14014		if (!ctl_pause_rtr) {
14015			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14016			if (io != NULL) {
14017				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14018				mtx_unlock(&thr->queue_lock);
14019				retval = ctl_scsiio(&io->scsiio);
14020				if (retval != CTL_RETVAL_COMPLETE)
14021					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14022				continue;
14023			}
14024		}
14025
14026		/* Sleep until we have something to do. */
14027		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14028	}
14029}
14030
14031static void
14032ctl_lun_thread(void *arg)
14033{
14034	struct ctl_softc *softc = (struct ctl_softc *)arg;
14035	struct ctl_be_lun *be_lun;
14036	int retval;
14037
14038	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14039
14040	for (;;) {
14041		retval = 0;
14042		mtx_lock(&softc->ctl_lock);
14043		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14044		if (be_lun != NULL) {
14045			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14046			mtx_unlock(&softc->ctl_lock);
14047			ctl_create_lun(be_lun);
14048			continue;
14049		}
14050
14051		/* Sleep until we have something to do. */
14052		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14053		    PDROP | PRIBIO, "-", 0);
14054	}
14055}
14056
14057static void
14058ctl_enqueue_incoming(union ctl_io *io)
14059{
14060	struct ctl_softc *softc = control_softc;
14061	struct ctl_thread *thr;
14062	u_int idx;
14063
14064	idx = (io->io_hdr.nexus.targ_port * 127 +
14065	       io->io_hdr.nexus.initid.id) % worker_threads;
14066	thr = &softc->threads[idx];
14067	mtx_lock(&thr->queue_lock);
14068	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14069	mtx_unlock(&thr->queue_lock);
14070	wakeup(thr);
14071}
14072
14073static void
14074ctl_enqueue_rtr(union ctl_io *io)
14075{
14076	struct ctl_softc *softc = control_softc;
14077	struct ctl_thread *thr;
14078
14079	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14080	mtx_lock(&thr->queue_lock);
14081	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14082	mtx_unlock(&thr->queue_lock);
14083	wakeup(thr);
14084}
14085
14086static void
14087ctl_enqueue_done(union ctl_io *io)
14088{
14089	struct ctl_softc *softc = control_softc;
14090	struct ctl_thread *thr;
14091
14092	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14093	mtx_lock(&thr->queue_lock);
14094	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14095	mtx_unlock(&thr->queue_lock);
14096	wakeup(thr);
14097}
14098
14099static void
14100ctl_enqueue_isc(union ctl_io *io)
14101{
14102	struct ctl_softc *softc = control_softc;
14103	struct ctl_thread *thr;
14104
14105	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14106	mtx_lock(&thr->queue_lock);
14107	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14108	mtx_unlock(&thr->queue_lock);
14109	wakeup(thr);
14110}
14111
14112/* Initialization and failover */
14113
14114void
14115ctl_init_isc_msg(void)
14116{
14117	printf("CTL: Still calling this thing\n");
14118}
14119
14120/*
14121 * Init component
14122 * 	Initializes component into configuration defined by bootMode
14123 *	(see hasc-sv.c)
14124 *  	returns hasc_Status:
14125 * 		OK
14126 *		ERROR - fatal error
14127 */
14128static ctl_ha_comp_status
14129ctl_isc_init(struct ctl_ha_component *c)
14130{
14131	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14132
14133	c->status = ret;
14134	return ret;
14135}
14136
14137/* Start component
14138 * 	Starts component in state requested. If component starts successfully,
14139 *	it must set its own state to the requestrd state
14140 *	When requested state is HASC_STATE_HA, the component may refine it
14141 * 	by adding _SLAVE or _MASTER flags.
14142 *	Currently allowed state transitions are:
14143 *	UNKNOWN->HA		- initial startup
14144 *	UNKNOWN->SINGLE - initial startup when no parter detected
14145 *	HA->SINGLE		- failover
14146 * returns ctl_ha_comp_status:
14147 * 		OK	- component successfully started in requested state
14148 *		FAILED  - could not start the requested state, failover may
14149 * 			  be possible
14150 *		ERROR	- fatal error detected, no future startup possible
14151 */
14152static ctl_ha_comp_status
14153ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14154{
14155	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14156
14157	printf("%s: go\n", __func__);
14158
14159	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14160	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14161		ctl_is_single = 0;
14162		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14163		    != CTL_HA_STATUS_SUCCESS) {
14164			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14165			ret = CTL_HA_COMP_STATUS_ERROR;
14166		}
14167	} else if (CTL_HA_STATE_IS_HA(c->state)
14168		&& CTL_HA_STATE_IS_SINGLE(state)){
14169		// HA->SINGLE transition
14170	        ctl_failover();
14171		ctl_is_single = 1;
14172	} else {
14173		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14174		       c->state, state);
14175		ret = CTL_HA_COMP_STATUS_ERROR;
14176	}
14177	if (CTL_HA_STATE_IS_SINGLE(state))
14178		ctl_is_single = 1;
14179
14180	c->state = state;
14181	c->status = ret;
14182	return ret;
14183}
14184
14185/*
14186 * Quiesce component
14187 * The component must clear any error conditions (set status to OK) and
14188 * prepare itself to another Start call
14189 * returns ctl_ha_comp_status:
14190 * 	OK
14191 *	ERROR
14192 */
14193static ctl_ha_comp_status
14194ctl_isc_quiesce(struct ctl_ha_component *c)
14195{
14196	int ret = CTL_HA_COMP_STATUS_OK;
14197
14198	ctl_pause_rtr = 1;
14199	c->status = ret;
14200	return ret;
14201}
14202
14203struct ctl_ha_component ctl_ha_component_ctlisc =
14204{
14205	.name = "CTL ISC",
14206	.state = CTL_HA_STATE_UNKNOWN,
14207	.init = ctl_isc_init,
14208	.start = ctl_isc_start,
14209	.quiesce = ctl_isc_quiesce
14210};
14211
14212/*
14213 *  vim: ts=8
14214 */
14215