ctl.c revision 272748
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl.c 272748 2014-10-08 12:24:24Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ SCP_WCE | SCP_RCD,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
284	/*eca_and_aen*/0,
285	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/SCP_QUEUE_ALG_MASK,
296	/*eca_and_aen*/SCP_SWP,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
316    &worker_threads, 1, "Number of worker threads");
317static int verbose = 0;
318SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
319    &verbose, 0, "Show SCSI errors returned to initiator");
320
321/*
322 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
323 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
324 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
325 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
326 */
327#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
328
329static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
330				  int param);
331static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
332static int ctl_init(void);
333void ctl_shutdown(void);
334static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
335static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
336static void ctl_ioctl_online(void *arg);
337static void ctl_ioctl_offline(void *arg);
338static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
339static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
340static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
341static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
342static int ctl_ioctl_submit_wait(union ctl_io *io);
343static void ctl_ioctl_datamove(union ctl_io *io);
344static void ctl_ioctl_done(union ctl_io *io);
345static void ctl_ioctl_hard_startstop_callback(void *arg,
346					      struct cfi_metatask *metatask);
347static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
348static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
349			      struct ctl_ooa *ooa_hdr,
350			      struct ctl_ooa_entry *kern_entries);
351static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
352		     struct thread *td);
353static uint32_t ctl_map_lun(int port_num, uint32_t lun);
354static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
355#ifdef unused
356static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
357				   uint32_t targ_target, uint32_t targ_lun,
358				   int can_wait);
359static void ctl_kfree_io(union ctl_io *io);
360#endif /* unused */
361static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
362			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
363static int ctl_free_lun(struct ctl_lun *lun);
364static void ctl_create_lun(struct ctl_be_lun *be_lun);
365/**
366static void ctl_failover_change_pages(struct ctl_softc *softc,
367				      struct ctl_scsiio *ctsio, int master);
368**/
369
370static int ctl_do_mode_select(union ctl_io *io);
371static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
372			   uint64_t res_key, uint64_t sa_res_key,
373			   uint8_t type, uint32_t residx,
374			   struct ctl_scsiio *ctsio,
375			   struct scsi_per_res_out *cdb,
376			   struct scsi_per_res_out_parms* param);
377static void ctl_pro_preempt_other(struct ctl_lun *lun,
378				  union ctl_ha_msg *msg);
379static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
380static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
381static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
382static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
383static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
385static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
386					 int alloc_len);
387static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
388					 int alloc_len);
389static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
390static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
391static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
392static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
393static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
394static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
395static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
396    union ctl_io *pending_io, union ctl_io *ooa_io);
397static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
398				union ctl_io *starting_io);
399static int ctl_check_blocked(struct ctl_lun *lun);
400static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
401				struct ctl_lun *lun,
402				const struct ctl_cmd_entry *entry,
403				struct ctl_scsiio *ctsio);
404//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
405static void ctl_failover(void);
406static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
407			       struct ctl_scsiio *ctsio);
408static int ctl_scsiio(struct ctl_scsiio *ctsio);
409
410static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
411static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
412			    ctl_ua_type ua_type);
413static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
414			 ctl_ua_type ua_type);
415static int ctl_abort_task(union ctl_io *io);
416static int ctl_abort_task_set(union ctl_io *io);
417static int ctl_i_t_nexus_reset(union ctl_io *io);
418static void ctl_run_task(union ctl_io *io);
419#ifdef CTL_IO_DELAY
420static void ctl_datamove_timer_wakeup(void *arg);
421static void ctl_done_timer_wakeup(void *arg);
422#endif /* CTL_IO_DELAY */
423
424static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
425static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
426static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
427static void ctl_datamove_remote_write(union ctl_io *io);
428static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
429static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
430static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
431static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
432				    ctl_ha_dt_cb callback);
433static void ctl_datamove_remote_read(union ctl_io *io);
434static void ctl_datamove_remote(union ctl_io *io);
435static int ctl_process_done(union ctl_io *io);
436static void ctl_lun_thread(void *arg);
437static void ctl_work_thread(void *arg);
438static void ctl_enqueue_incoming(union ctl_io *io);
439static void ctl_enqueue_rtr(union ctl_io *io);
440static void ctl_enqueue_done(union ctl_io *io);
441static void ctl_enqueue_isc(union ctl_io *io);
442static const struct ctl_cmd_entry *
443    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
444static const struct ctl_cmd_entry *
445    ctl_validate_command(struct ctl_scsiio *ctsio);
446static int ctl_cmd_applicable(uint8_t lun_type,
447    const struct ctl_cmd_entry *entry);
448
449/*
450 * Load the serialization table.  This isn't very pretty, but is probably
451 * the easiest way to do it.
452 */
453#include "ctl_ser_table.c"
454
455/*
456 * We only need to define open, close and ioctl routines for this driver.
457 */
458static struct cdevsw ctl_cdevsw = {
459	.d_version =	D_VERSION,
460	.d_flags =	0,
461	.d_open =	ctl_open,
462	.d_close =	ctl_close,
463	.d_ioctl =	ctl_ioctl,
464	.d_name =	"ctl",
465};
466
467
468MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
469MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
470
471static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
472
473static moduledata_t ctl_moduledata = {
474	"ctl",
475	ctl_module_event_handler,
476	NULL
477};
478
479DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
480MODULE_VERSION(ctl, 1);
481
482static struct ctl_frontend ioctl_frontend =
483{
484	.name = "ioctl",
485};
486
487static void
488ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
489			    union ctl_ha_msg *msg_info)
490{
491	struct ctl_scsiio *ctsio;
492
493	if (msg_info->hdr.original_sc == NULL) {
494		printf("%s: original_sc == NULL!\n", __func__);
495		/* XXX KDM now what? */
496		return;
497	}
498
499	ctsio = &msg_info->hdr.original_sc->scsiio;
500	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
501	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
502	ctsio->io_hdr.status = msg_info->hdr.status;
503	ctsio->scsi_status = msg_info->scsi.scsi_status;
504	ctsio->sense_len = msg_info->scsi.sense_len;
505	ctsio->sense_residual = msg_info->scsi.sense_residual;
506	ctsio->residual = msg_info->scsi.residual;
507	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
508	       sizeof(ctsio->sense_data));
509	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
510	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
511	ctl_enqueue_isc((union ctl_io *)ctsio);
512}
513
514static void
515ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
516				union ctl_ha_msg *msg_info)
517{
518	struct ctl_scsiio *ctsio;
519
520	if (msg_info->hdr.serializing_sc == NULL) {
521		printf("%s: serializing_sc == NULL!\n", __func__);
522		/* XXX KDM now what? */
523		return;
524	}
525
526	ctsio = &msg_info->hdr.serializing_sc->scsiio;
527#if 0
528	/*
529	 * Attempt to catch the situation where an I/O has
530	 * been freed, and we're using it again.
531	 */
532	if (ctsio->io_hdr.io_type == 0xff) {
533		union ctl_io *tmp_io;
534		tmp_io = (union ctl_io *)ctsio;
535		printf("%s: %p use after free!\n", __func__,
536		       ctsio);
537		printf("%s: type %d msg %d cdb %x iptl: "
538		       "%d:%d:%d:%d tag 0x%04x "
539		       "flag %#x status %x\n",
540			__func__,
541			tmp_io->io_hdr.io_type,
542			tmp_io->io_hdr.msg_type,
543			tmp_io->scsiio.cdb[0],
544			tmp_io->io_hdr.nexus.initid.id,
545			tmp_io->io_hdr.nexus.targ_port,
546			tmp_io->io_hdr.nexus.targ_target.id,
547			tmp_io->io_hdr.nexus.targ_lun,
548			(tmp_io->io_hdr.io_type ==
549			CTL_IO_TASK) ?
550			tmp_io->taskio.tag_num :
551			tmp_io->scsiio.tag_num,
552		        tmp_io->io_hdr.flags,
553			tmp_io->io_hdr.status);
554	}
555#endif
556	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
557	ctl_enqueue_isc((union ctl_io *)ctsio);
558}
559
560/*
561 * ISC (Inter Shelf Communication) event handler.  Events from the HA
562 * subsystem come in here.
563 */
564static void
565ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
566{
567	struct ctl_softc *ctl_softc;
568	union ctl_io *io;
569	struct ctl_prio *presio;
570	ctl_ha_status isc_status;
571
572	ctl_softc = control_softc;
573	io = NULL;
574
575
576#if 0
577	printf("CTL: Isc Msg event %d\n", event);
578#endif
579	if (event == CTL_HA_EVT_MSG_RECV) {
580		union ctl_ha_msg msg_info;
581
582		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
583					     sizeof(msg_info), /*wait*/ 0);
584#if 0
585		printf("CTL: msg_type %d\n", msg_info.msg_type);
586#endif
587		if (isc_status != 0) {
588			printf("Error receiving message, status = %d\n",
589			       isc_status);
590			return;
591		}
592
593		switch (msg_info.hdr.msg_type) {
594		case CTL_MSG_SERIALIZE:
595#if 0
596			printf("Serialize\n");
597#endif
598			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
599			if (io == NULL) {
600				printf("ctl_isc_event_handler: can't allocate "
601				       "ctl_io!\n");
602				/* Bad Juju */
603				/* Need to set busy and send msg back */
604				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
605				msg_info.hdr.status = CTL_SCSI_ERROR;
606				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
607				msg_info.scsi.sense_len = 0;
608			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
609				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
610				}
611				goto bailout;
612			}
613			ctl_zero_io(io);
614			// populate ctsio from msg_info
615			io->io_hdr.io_type = CTL_IO_SCSI;
616			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
617			io->io_hdr.original_sc = msg_info.hdr.original_sc;
618#if 0
619			printf("pOrig %x\n", (int)msg_info.original_sc);
620#endif
621			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
622					    CTL_FLAG_IO_ACTIVE;
623			/*
624			 * If we're in serialization-only mode, we don't
625			 * want to go through full done processing.  Thus
626			 * the COPY flag.
627			 *
628			 * XXX KDM add another flag that is more specific.
629			 */
630			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
631				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
632			io->io_hdr.nexus = msg_info.hdr.nexus;
633#if 0
634			printf("targ %d, port %d, iid %d, lun %d\n",
635			       io->io_hdr.nexus.targ_target.id,
636			       io->io_hdr.nexus.targ_port,
637			       io->io_hdr.nexus.initid.id,
638			       io->io_hdr.nexus.targ_lun);
639#endif
640			io->scsiio.tag_num = msg_info.scsi.tag_num;
641			io->scsiio.tag_type = msg_info.scsi.tag_type;
642			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
643			       CTL_MAX_CDBLEN);
644			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
645				const struct ctl_cmd_entry *entry;
646
647				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
648				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
649				io->io_hdr.flags |=
650					entry->flags & CTL_FLAG_DATA_MASK;
651			}
652			ctl_enqueue_isc(io);
653			break;
654
655		/* Performed on the Originating SC, XFER mode only */
656		case CTL_MSG_DATAMOVE: {
657			struct ctl_sg_entry *sgl;
658			int i, j;
659
660			io = msg_info.hdr.original_sc;
661			if (io == NULL) {
662				printf("%s: original_sc == NULL!\n", __func__);
663				/* XXX KDM do something here */
664				break;
665			}
666			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
667			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
668			/*
669			 * Keep track of this, we need to send it back over
670			 * when the datamove is complete.
671			 */
672			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
673
674			if (msg_info.dt.sg_sequence == 0) {
675				/*
676				 * XXX KDM we use the preallocated S/G list
677				 * here, but we'll need to change this to
678				 * dynamic allocation if we need larger S/G
679				 * lists.
680				 */
681				if (msg_info.dt.kern_sg_entries >
682				    sizeof(io->io_hdr.remote_sglist) /
683				    sizeof(io->io_hdr.remote_sglist[0])) {
684					printf("%s: number of S/G entries "
685					    "needed %u > allocated num %zd\n",
686					    __func__,
687					    msg_info.dt.kern_sg_entries,
688					    sizeof(io->io_hdr.remote_sglist)/
689					    sizeof(io->io_hdr.remote_sglist[0]));
690
691					/*
692					 * XXX KDM send a message back to
693					 * the other side to shut down the
694					 * DMA.  The error will come back
695					 * through via the normal channel.
696					 */
697					break;
698				}
699				sgl = io->io_hdr.remote_sglist;
700				memset(sgl, 0,
701				       sizeof(io->io_hdr.remote_sglist));
702
703				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
704
705				io->scsiio.kern_sg_entries =
706					msg_info.dt.kern_sg_entries;
707				io->scsiio.rem_sg_entries =
708					msg_info.dt.kern_sg_entries;
709				io->scsiio.kern_data_len =
710					msg_info.dt.kern_data_len;
711				io->scsiio.kern_total_len =
712					msg_info.dt.kern_total_len;
713				io->scsiio.kern_data_resid =
714					msg_info.dt.kern_data_resid;
715				io->scsiio.kern_rel_offset =
716					msg_info.dt.kern_rel_offset;
717				/*
718				 * Clear out per-DMA flags.
719				 */
720				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
721				/*
722				 * Add per-DMA flags that are set for this
723				 * particular DMA request.
724				 */
725				io->io_hdr.flags |= msg_info.dt.flags &
726						    CTL_FLAG_RDMA_MASK;
727			} else
728				sgl = (struct ctl_sg_entry *)
729					io->scsiio.kern_data_ptr;
730
731			for (i = msg_info.dt.sent_sg_entries, j = 0;
732			     i < (msg_info.dt.sent_sg_entries +
733			     msg_info.dt.cur_sg_entries); i++, j++) {
734				sgl[i].addr = msg_info.dt.sg_list[j].addr;
735				sgl[i].len = msg_info.dt.sg_list[j].len;
736
737#if 0
738				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
739				       __func__,
740				       msg_info.dt.sg_list[j].addr,
741				       msg_info.dt.sg_list[j].len,
742				       sgl[i].addr, sgl[i].len, j, i);
743#endif
744			}
745#if 0
746			memcpy(&sgl[msg_info.dt.sent_sg_entries],
747			       msg_info.dt.sg_list,
748			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
749#endif
750
751			/*
752			 * If this is the last piece of the I/O, we've got
753			 * the full S/G list.  Queue processing in the thread.
754			 * Otherwise wait for the next piece.
755			 */
756			if (msg_info.dt.sg_last != 0)
757				ctl_enqueue_isc(io);
758			break;
759		}
760		/* Performed on the Serializing (primary) SC, XFER mode only */
761		case CTL_MSG_DATAMOVE_DONE: {
762			if (msg_info.hdr.serializing_sc == NULL) {
763				printf("%s: serializing_sc == NULL!\n",
764				       __func__);
765				/* XXX KDM now what? */
766				break;
767			}
768			/*
769			 * We grab the sense information here in case
770			 * there was a failure, so we can return status
771			 * back to the initiator.
772			 */
773			io = msg_info.hdr.serializing_sc;
774			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
775			io->io_hdr.status = msg_info.hdr.status;
776			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
777			io->scsiio.sense_len = msg_info.scsi.sense_len;
778			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
779			io->io_hdr.port_status = msg_info.scsi.fetd_status;
780			io->scsiio.residual = msg_info.scsi.residual;
781			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
782			       sizeof(io->scsiio.sense_data));
783			ctl_enqueue_isc(io);
784			break;
785		}
786
787		/* Preformed on Originating SC, SER_ONLY mode */
788		case CTL_MSG_R2R:
789			io = msg_info.hdr.original_sc;
790			if (io == NULL) {
791				printf("%s: Major Bummer\n", __func__);
792				return;
793			} else {
794#if 0
795				printf("pOrig %x\n",(int) ctsio);
796#endif
797			}
798			io->io_hdr.msg_type = CTL_MSG_R2R;
799			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
800			ctl_enqueue_isc(io);
801			break;
802
803		/*
804		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
805		 * mode.
806		 * Performed on the Originating (i.e. secondary) SC in XFER
807		 * mode
808		 */
809		case CTL_MSG_FINISH_IO:
810			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
811				ctl_isc_handler_finish_xfer(ctl_softc,
812							    &msg_info);
813			else
814				ctl_isc_handler_finish_ser_only(ctl_softc,
815								&msg_info);
816			break;
817
818		/* Preformed on Originating SC */
819		case CTL_MSG_BAD_JUJU:
820			io = msg_info.hdr.original_sc;
821			if (io == NULL) {
822				printf("%s: Bad JUJU!, original_sc is NULL!\n",
823				       __func__);
824				break;
825			}
826			ctl_copy_sense_data(&msg_info, io);
827			/*
828			 * IO should have already been cleaned up on other
829			 * SC so clear this flag so we won't send a message
830			 * back to finish the IO there.
831			 */
832			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
833			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
834
835			/* io = msg_info.hdr.serializing_sc; */
836			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
837			ctl_enqueue_isc(io);
838			break;
839
840		/* Handle resets sent from the other side */
841		case CTL_MSG_MANAGE_TASKS: {
842			struct ctl_taskio *taskio;
843			taskio = (struct ctl_taskio *)ctl_alloc_io(
844				(void *)ctl_softc->othersc_pool);
845			if (taskio == NULL) {
846				printf("ctl_isc_event_handler: can't allocate "
847				       "ctl_io!\n");
848				/* Bad Juju */
849				/* should I just call the proper reset func
850				   here??? */
851				goto bailout;
852			}
853			ctl_zero_io((union ctl_io *)taskio);
854			taskio->io_hdr.io_type = CTL_IO_TASK;
855			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
856			taskio->io_hdr.nexus = msg_info.hdr.nexus;
857			taskio->task_action = msg_info.task.task_action;
858			taskio->tag_num = msg_info.task.tag_num;
859			taskio->tag_type = msg_info.task.tag_type;
860#ifdef CTL_TIME_IO
861			taskio->io_hdr.start_time = time_uptime;
862			getbintime(&taskio->io_hdr.start_bt);
863#if 0
864			cs_prof_gettime(&taskio->io_hdr.start_ticks);
865#endif
866#endif /* CTL_TIME_IO */
867			ctl_run_task((union ctl_io *)taskio);
868			break;
869		}
870		/* Persistent Reserve action which needs attention */
871		case CTL_MSG_PERS_ACTION:
872			presio = (struct ctl_prio *)ctl_alloc_io(
873				(void *)ctl_softc->othersc_pool);
874			if (presio == NULL) {
875				printf("ctl_isc_event_handler: can't allocate "
876				       "ctl_io!\n");
877				/* Bad Juju */
878				/* Need to set busy and send msg back */
879				goto bailout;
880			}
881			ctl_zero_io((union ctl_io *)presio);
882			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
883			presio->pr_msg = msg_info.pr;
884			ctl_enqueue_isc((union ctl_io *)presio);
885			break;
886		case CTL_MSG_SYNC_FE:
887			rcv_sync_msg = 1;
888			break;
889		case CTL_MSG_APS_LOCK: {
890			// It's quicker to execute this then to
891			// queue it.
892			struct ctl_lun *lun;
893			struct ctl_page_index *page_index;
894			struct copan_aps_subpage *current_sp;
895			uint32_t targ_lun;
896
897			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
898			lun = ctl_softc->ctl_luns[targ_lun];
899			mtx_lock(&lun->lun_lock);
900			page_index = &lun->mode_pages.index[index_to_aps_page];
901			current_sp = (struct copan_aps_subpage *)
902				     (page_index->page_data +
903				     (page_index->page_len * CTL_PAGE_CURRENT));
904
905			current_sp->lock_active = msg_info.aps.lock_flag;
906			mtx_unlock(&lun->lun_lock);
907		        break;
908		}
909		default:
910		        printf("How did I get here?\n");
911		}
912	} else if (event == CTL_HA_EVT_MSG_SENT) {
913		if (param != CTL_HA_STATUS_SUCCESS) {
914			printf("Bad status from ctl_ha_msg_send status %d\n",
915			       param);
916		}
917		return;
918	} else if (event == CTL_HA_EVT_DISCONNECT) {
919		printf("CTL: Got a disconnect from Isc\n");
920		return;
921	} else {
922		printf("ctl_isc_event_handler: Unknown event %d\n", event);
923		return;
924	}
925
926bailout:
927	return;
928}
929
930static void
931ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
932{
933	struct scsi_sense_data *sense;
934
935	sense = &dest->scsiio.sense_data;
936	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
937	dest->scsiio.scsi_status = src->scsi.scsi_status;
938	dest->scsiio.sense_len = src->scsi.sense_len;
939	dest->io_hdr.status = src->hdr.status;
940}
941
942static int
943ctl_init(void)
944{
945	struct ctl_softc *softc;
946	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
947	struct ctl_port *port;
948        uint8_t sc_id =0;
949	int i, error, retval;
950	//int isc_retval;
951
952	retval = 0;
953	ctl_pause_rtr = 0;
954        rcv_sync_msg = 0;
955
956	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
957			       M_WAITOK | M_ZERO);
958	softc = control_softc;
959
960	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
961			      "cam/ctl");
962
963	softc->dev->si_drv1 = softc;
964
965	/*
966	 * By default, return a "bad LUN" peripheral qualifier for unknown
967	 * LUNs.  The user can override this default using the tunable or
968	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
969	 */
970	softc->inquiry_pq_no_lun = 1;
971	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
972			  &softc->inquiry_pq_no_lun);
973	sysctl_ctx_init(&softc->sysctl_ctx);
974	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
975		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
976		CTLFLAG_RD, 0, "CAM Target Layer");
977
978	if (softc->sysctl_tree == NULL) {
979		printf("%s: unable to allocate sysctl tree\n", __func__);
980		destroy_dev(softc->dev);
981		free(control_softc, M_DEVBUF);
982		control_softc = NULL;
983		return (ENOMEM);
984	}
985
986	SYSCTL_ADD_INT(&softc->sysctl_ctx,
987		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
988		       "inquiry_pq_no_lun", CTLFLAG_RW,
989		       &softc->inquiry_pq_no_lun, 0,
990		       "Report no lun possible for invalid LUNs");
991
992	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
993	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
994	softc->open_count = 0;
995
996	/*
997	 * Default to actually sending a SYNCHRONIZE CACHE command down to
998	 * the drive.
999	 */
1000	softc->flags = CTL_FLAG_REAL_SYNC;
1001
1002	/*
1003	 * In Copan's HA scheme, the "master" and "slave" roles are
1004	 * figured out through the slot the controller is in.  Although it
1005	 * is an active/active system, someone has to be in charge.
1006 	 */
1007#ifdef NEEDTOPORT
1008        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1009#endif
1010
1011        if (sc_id == 0) {
1012		softc->flags |= CTL_FLAG_MASTER_SHELF;
1013		persis_offset = 0;
1014	} else
1015		persis_offset = CTL_MAX_INITIATORS;
1016
1017	/*
1018	 * XXX KDM need to figure out where we want to get our target ID
1019	 * and WWID.  Is it different on each port?
1020	 */
1021	softc->target.id = 0;
1022	softc->target.wwid[0] = 0x12345678;
1023	softc->target.wwid[1] = 0x87654321;
1024	STAILQ_INIT(&softc->lun_list);
1025	STAILQ_INIT(&softc->pending_lun_queue);
1026	STAILQ_INIT(&softc->fe_list);
1027	STAILQ_INIT(&softc->port_list);
1028	STAILQ_INIT(&softc->be_list);
1029	STAILQ_INIT(&softc->io_pools);
1030	ctl_tpc_init(softc);
1031
1032	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1033			    &internal_pool)!= 0){
1034		printf("ctl: can't allocate %d entry internal pool, "
1035		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1036		return (ENOMEM);
1037	}
1038
1039	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1040			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1041		printf("ctl: can't allocate %d entry emergency pool, "
1042		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1043		ctl_pool_free(internal_pool);
1044		return (ENOMEM);
1045	}
1046
1047	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1048	                    &other_pool) != 0)
1049	{
1050		printf("ctl: can't allocate %d entry other SC pool, "
1051		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1052		ctl_pool_free(internal_pool);
1053		ctl_pool_free(emergency_pool);
1054		return (ENOMEM);
1055	}
1056
1057	softc->internal_pool = internal_pool;
1058	softc->emergency_pool = emergency_pool;
1059	softc->othersc_pool = other_pool;
1060
1061	if (worker_threads <= 0)
1062		worker_threads = max(1, mp_ncpus / 4);
1063	if (worker_threads > CTL_MAX_THREADS)
1064		worker_threads = CTL_MAX_THREADS;
1065
1066	for (i = 0; i < worker_threads; i++) {
1067		struct ctl_thread *thr = &softc->threads[i];
1068
1069		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1070		thr->ctl_softc = softc;
1071		STAILQ_INIT(&thr->incoming_queue);
1072		STAILQ_INIT(&thr->rtr_queue);
1073		STAILQ_INIT(&thr->done_queue);
1074		STAILQ_INIT(&thr->isc_queue);
1075
1076		error = kproc_kthread_add(ctl_work_thread, thr,
1077		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1078		if (error != 0) {
1079			printf("error creating CTL work thread!\n");
1080			ctl_pool_free(internal_pool);
1081			ctl_pool_free(emergency_pool);
1082			ctl_pool_free(other_pool);
1083			return (error);
1084		}
1085	}
1086	error = kproc_kthread_add(ctl_lun_thread, softc,
1087	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1088	if (error != 0) {
1089		printf("error creating CTL lun thread!\n");
1090		ctl_pool_free(internal_pool);
1091		ctl_pool_free(emergency_pool);
1092		ctl_pool_free(other_pool);
1093		return (error);
1094	}
1095	if (bootverbose)
1096		printf("ctl: CAM Target Layer loaded\n");
1097
1098	/*
1099	 * Initialize the ioctl front end.
1100	 */
1101	ctl_frontend_register(&ioctl_frontend);
1102	port = &softc->ioctl_info.port;
1103	port->frontend = &ioctl_frontend;
1104	sprintf(softc->ioctl_info.port_name, "ioctl");
1105	port->port_type = CTL_PORT_IOCTL;
1106	port->num_requested_ctl_io = 100;
1107	port->port_name = softc->ioctl_info.port_name;
1108	port->port_online = ctl_ioctl_online;
1109	port->port_offline = ctl_ioctl_offline;
1110	port->onoff_arg = &softc->ioctl_info;
1111	port->lun_enable = ctl_ioctl_lun_enable;
1112	port->lun_disable = ctl_ioctl_lun_disable;
1113	port->targ_lun_arg = &softc->ioctl_info;
1114	port->fe_datamove = ctl_ioctl_datamove;
1115	port->fe_done = ctl_ioctl_done;
1116	port->max_targets = 15;
1117	port->max_target_id = 15;
1118
1119	if (ctl_port_register(&softc->ioctl_info.port,
1120	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1121		printf("ctl: ioctl front end registration failed, will "
1122		       "continue anyway\n");
1123	}
1124
1125#ifdef CTL_IO_DELAY
1126	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1127		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1128		       sizeof(struct callout), CTL_TIMER_BYTES);
1129		return (EINVAL);
1130	}
1131#endif /* CTL_IO_DELAY */
1132
1133	return (0);
1134}
1135
1136void
1137ctl_shutdown(void)
1138{
1139	struct ctl_softc *softc;
1140	struct ctl_lun *lun, *next_lun;
1141	struct ctl_io_pool *pool;
1142
1143	softc = (struct ctl_softc *)control_softc;
1144
1145	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1146		printf("ctl: ioctl front end deregistration failed\n");
1147
1148	mtx_lock(&softc->ctl_lock);
1149
1150	/*
1151	 * Free up each LUN.
1152	 */
1153	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1154		next_lun = STAILQ_NEXT(lun, links);
1155		ctl_free_lun(lun);
1156	}
1157
1158	mtx_unlock(&softc->ctl_lock);
1159
1160	ctl_frontend_deregister(&ioctl_frontend);
1161
1162	/*
1163	 * This will rip the rug out from under any FETDs or anyone else
1164	 * that has a pool allocated.  Since we increment our module
1165	 * refcount any time someone outside the main CTL module allocates
1166	 * a pool, we shouldn't have any problems here.  The user won't be
1167	 * able to unload the CTL module until client modules have
1168	 * successfully unloaded.
1169	 */
1170	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1171		ctl_pool_free(pool);
1172
1173#if 0
1174	ctl_shutdown_thread(softc->work_thread);
1175	mtx_destroy(&softc->queue_lock);
1176#endif
1177
1178	ctl_tpc_shutdown(softc);
1179	mtx_destroy(&softc->pool_lock);
1180	mtx_destroy(&softc->ctl_lock);
1181
1182	destroy_dev(softc->dev);
1183
1184	sysctl_ctx_free(&softc->sysctl_ctx);
1185
1186	free(control_softc, M_DEVBUF);
1187	control_softc = NULL;
1188
1189	if (bootverbose)
1190		printf("ctl: CAM Target Layer unloaded\n");
1191}
1192
1193static int
1194ctl_module_event_handler(module_t mod, int what, void *arg)
1195{
1196
1197	switch (what) {
1198	case MOD_LOAD:
1199		return (ctl_init());
1200	case MOD_UNLOAD:
1201		return (EBUSY);
1202	default:
1203		return (EOPNOTSUPP);
1204	}
1205}
1206
1207/*
1208 * XXX KDM should we do some access checks here?  Bump a reference count to
1209 * prevent a CTL module from being unloaded while someone has it open?
1210 */
1211static int
1212ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1213{
1214	return (0);
1215}
1216
1217static int
1218ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1219{
1220	return (0);
1221}
1222
1223int
1224ctl_port_enable(ctl_port_type port_type)
1225{
1226	struct ctl_softc *softc;
1227	struct ctl_port *port;
1228
1229	if (ctl_is_single == 0) {
1230		union ctl_ha_msg msg_info;
1231		int isc_retval;
1232
1233#if 0
1234		printf("%s: HA mode, synchronizing frontend enable\n",
1235		        __func__);
1236#endif
1237		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1238	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1239		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1240			printf("Sync msg send error retval %d\n", isc_retval);
1241		}
1242		if (!rcv_sync_msg) {
1243			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1244			        sizeof(msg_info), 1);
1245		}
1246#if 0
1247        	printf("CTL:Frontend Enable\n");
1248	} else {
1249		printf("%s: single mode, skipping frontend synchronization\n",
1250		        __func__);
1251#endif
1252	}
1253
1254	softc = control_softc;
1255
1256	STAILQ_FOREACH(port, &softc->port_list, links) {
1257		if (port_type & port->port_type)
1258		{
1259#if 0
1260			printf("port %d\n", port->targ_port);
1261#endif
1262			ctl_port_online(port);
1263		}
1264	}
1265
1266	return (0);
1267}
1268
1269int
1270ctl_port_disable(ctl_port_type port_type)
1271{
1272	struct ctl_softc *softc;
1273	struct ctl_port *port;
1274
1275	softc = control_softc;
1276
1277	STAILQ_FOREACH(port, &softc->port_list, links) {
1278		if (port_type & port->port_type)
1279			ctl_port_offline(port);
1280	}
1281
1282	return (0);
1283}
1284
1285/*
1286 * Returns 0 for success, 1 for failure.
1287 * Currently the only failure mode is if there aren't enough entries
1288 * allocated.  So, in case of a failure, look at num_entries_dropped,
1289 * reallocate and try again.
1290 */
1291int
1292ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1293	      int *num_entries_filled, int *num_entries_dropped,
1294	      ctl_port_type port_type, int no_virtual)
1295{
1296	struct ctl_softc *softc;
1297	struct ctl_port *port;
1298	int entries_dropped, entries_filled;
1299	int retval;
1300	int i;
1301
1302	softc = control_softc;
1303
1304	retval = 0;
1305	entries_filled = 0;
1306	entries_dropped = 0;
1307
1308	i = 0;
1309	mtx_lock(&softc->ctl_lock);
1310	STAILQ_FOREACH(port, &softc->port_list, links) {
1311		struct ctl_port_entry *entry;
1312
1313		if ((port->port_type & port_type) == 0)
1314			continue;
1315
1316		if ((no_virtual != 0)
1317		 && (port->virtual_port != 0))
1318			continue;
1319
1320		if (entries_filled >= num_entries_alloced) {
1321			entries_dropped++;
1322			continue;
1323		}
1324		entry = &entries[i];
1325
1326		entry->port_type = port->port_type;
1327		strlcpy(entry->port_name, port->port_name,
1328			sizeof(entry->port_name));
1329		entry->physical_port = port->physical_port;
1330		entry->virtual_port = port->virtual_port;
1331		entry->wwnn = port->wwnn;
1332		entry->wwpn = port->wwpn;
1333
1334		i++;
1335		entries_filled++;
1336	}
1337
1338	mtx_unlock(&softc->ctl_lock);
1339
1340	if (entries_dropped > 0)
1341		retval = 1;
1342
1343	*num_entries_dropped = entries_dropped;
1344	*num_entries_filled = entries_filled;
1345
1346	return (retval);
1347}
1348
1349static void
1350ctl_ioctl_online(void *arg)
1351{
1352	struct ctl_ioctl_info *ioctl_info;
1353
1354	ioctl_info = (struct ctl_ioctl_info *)arg;
1355
1356	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1357}
1358
1359static void
1360ctl_ioctl_offline(void *arg)
1361{
1362	struct ctl_ioctl_info *ioctl_info;
1363
1364	ioctl_info = (struct ctl_ioctl_info *)arg;
1365
1366	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1367}
1368
1369/*
1370 * Remove an initiator by port number and initiator ID.
1371 * Returns 0 for success, -1 for failure.
1372 */
1373int
1374ctl_remove_initiator(struct ctl_port *port, int iid)
1375{
1376	struct ctl_softc *softc = control_softc;
1377
1378	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1379
1380	if (iid > CTL_MAX_INIT_PER_PORT) {
1381		printf("%s: initiator ID %u > maximun %u!\n",
1382		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1383		return (-1);
1384	}
1385
1386	mtx_lock(&softc->ctl_lock);
1387	port->wwpn_iid[iid].in_use--;
1388	port->wwpn_iid[iid].last_use = time_uptime;
1389	mtx_unlock(&softc->ctl_lock);
1390
1391	return (0);
1392}
1393
1394/*
1395 * Add an initiator to the initiator map.
1396 * Returns iid for success, < 0 for failure.
1397 */
1398int
1399ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1400{
1401	struct ctl_softc *softc = control_softc;
1402	time_t best_time;
1403	int i, best;
1404
1405	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1406
1407	if (iid >= CTL_MAX_INIT_PER_PORT) {
1408		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1409		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1410		free(name, M_CTL);
1411		return (-1);
1412	}
1413
1414	mtx_lock(&softc->ctl_lock);
1415
1416	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1417		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1418			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1419				iid = i;
1420				break;
1421			}
1422			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1423			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1424				iid = i;
1425				break;
1426			}
1427		}
1428	}
1429
1430	if (iid < 0) {
1431		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1432			if (port->wwpn_iid[i].in_use == 0 &&
1433			    port->wwpn_iid[i].wwpn == 0 &&
1434			    port->wwpn_iid[i].name == NULL) {
1435				iid = i;
1436				break;
1437			}
1438		}
1439	}
1440
1441	if (iid < 0) {
1442		best = -1;
1443		best_time = INT32_MAX;
1444		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1445			if (port->wwpn_iid[i].in_use == 0) {
1446				if (port->wwpn_iid[i].last_use < best_time) {
1447					best = i;
1448					best_time = port->wwpn_iid[i].last_use;
1449				}
1450			}
1451		}
1452		iid = best;
1453	}
1454
1455	if (iid < 0) {
1456		mtx_unlock(&softc->ctl_lock);
1457		free(name, M_CTL);
1458		return (-2);
1459	}
1460
1461	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1462		/*
1463		 * This is not an error yet.
1464		 */
1465		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1466#if 0
1467			printf("%s: port %d iid %u WWPN %#jx arrived"
1468			    " again\n", __func__, port->targ_port,
1469			    iid, (uintmax_t)wwpn);
1470#endif
1471			goto take;
1472		}
1473		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1474		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1475#if 0
1476			printf("%s: port %d iid %u name '%s' arrived"
1477			    " again\n", __func__, port->targ_port,
1478			    iid, name);
1479#endif
1480			goto take;
1481		}
1482
1483		/*
1484		 * This is an error, but what do we do about it?  The
1485		 * driver is telling us we have a new WWPN for this
1486		 * initiator ID, so we pretty much need to use it.
1487		 */
1488		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1489		    " but WWPN %#jx '%s' is still at that address\n",
1490		    __func__, port->targ_port, iid, wwpn, name,
1491		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1492		    port->wwpn_iid[iid].name);
1493
1494		/*
1495		 * XXX KDM clear have_ca and ua_pending on each LUN for
1496		 * this initiator.
1497		 */
1498	}
1499take:
1500	free(port->wwpn_iid[iid].name, M_CTL);
1501	port->wwpn_iid[iid].name = name;
1502	port->wwpn_iid[iid].wwpn = wwpn;
1503	port->wwpn_iid[iid].in_use++;
1504	mtx_unlock(&softc->ctl_lock);
1505
1506	return (iid);
1507}
1508
1509static int
1510ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1511{
1512	int len;
1513
1514	switch (port->port_type) {
1515	case CTL_PORT_FC:
1516	{
1517		struct scsi_transportid_fcp *id =
1518		    (struct scsi_transportid_fcp *)buf;
1519		if (port->wwpn_iid[iid].wwpn == 0)
1520			return (0);
1521		memset(id, 0, sizeof(*id));
1522		id->format_protocol = SCSI_PROTO_FC;
1523		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1524		return (sizeof(*id));
1525	}
1526	case CTL_PORT_ISCSI:
1527	{
1528		struct scsi_transportid_iscsi_port *id =
1529		    (struct scsi_transportid_iscsi_port *)buf;
1530		if (port->wwpn_iid[iid].name == NULL)
1531			return (0);
1532		memset(id, 0, 256);
1533		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1534		    SCSI_PROTO_ISCSI;
1535		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1536		len = roundup2(min(len, 252), 4);
1537		scsi_ulto2b(len, id->additional_length);
1538		return (sizeof(*id) + len);
1539	}
1540	case CTL_PORT_SAS:
1541	{
1542		struct scsi_transportid_sas *id =
1543		    (struct scsi_transportid_sas *)buf;
1544		if (port->wwpn_iid[iid].wwpn == 0)
1545			return (0);
1546		memset(id, 0, sizeof(*id));
1547		id->format_protocol = SCSI_PROTO_SAS;
1548		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1549		return (sizeof(*id));
1550	}
1551	default:
1552	{
1553		struct scsi_transportid_spi *id =
1554		    (struct scsi_transportid_spi *)buf;
1555		memset(id, 0, sizeof(*id));
1556		id->format_protocol = SCSI_PROTO_SPI;
1557		scsi_ulto2b(iid, id->scsi_addr);
1558		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1559		return (sizeof(*id));
1560	}
1561	}
1562}
1563
1564static int
1565ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1566{
1567	return (0);
1568}
1569
1570static int
1571ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1572{
1573	return (0);
1574}
1575
1576/*
1577 * Data movement routine for the CTL ioctl frontend port.
1578 */
1579static int
1580ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1581{
1582	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1583	struct ctl_sg_entry ext_entry, kern_entry;
1584	int ext_sglen, ext_sg_entries, kern_sg_entries;
1585	int ext_sg_start, ext_offset;
1586	int len_to_copy, len_copied;
1587	int kern_watermark, ext_watermark;
1588	int ext_sglist_malloced;
1589	int i, j;
1590
1591	ext_sglist_malloced = 0;
1592	ext_sg_start = 0;
1593	ext_offset = 0;
1594
1595	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1596
1597	/*
1598	 * If this flag is set, fake the data transfer.
1599	 */
1600	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1601		ctsio->ext_data_filled = ctsio->ext_data_len;
1602		goto bailout;
1603	}
1604
1605	/*
1606	 * To simplify things here, if we have a single buffer, stick it in
1607	 * a S/G entry and just make it a single entry S/G list.
1608	 */
1609	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1610		int len_seen;
1611
1612		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1613
1614		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1615							   M_WAITOK);
1616		ext_sglist_malloced = 1;
1617		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1618				   ext_sglen) != 0) {
1619			ctl_set_internal_failure(ctsio,
1620						 /*sks_valid*/ 0,
1621						 /*retry_count*/ 0);
1622			goto bailout;
1623		}
1624		ext_sg_entries = ctsio->ext_sg_entries;
1625		len_seen = 0;
1626		for (i = 0; i < ext_sg_entries; i++) {
1627			if ((len_seen + ext_sglist[i].len) >=
1628			     ctsio->ext_data_filled) {
1629				ext_sg_start = i;
1630				ext_offset = ctsio->ext_data_filled - len_seen;
1631				break;
1632			}
1633			len_seen += ext_sglist[i].len;
1634		}
1635	} else {
1636		ext_sglist = &ext_entry;
1637		ext_sglist->addr = ctsio->ext_data_ptr;
1638		ext_sglist->len = ctsio->ext_data_len;
1639		ext_sg_entries = 1;
1640		ext_sg_start = 0;
1641		ext_offset = ctsio->ext_data_filled;
1642	}
1643
1644	if (ctsio->kern_sg_entries > 0) {
1645		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1646		kern_sg_entries = ctsio->kern_sg_entries;
1647	} else {
1648		kern_sglist = &kern_entry;
1649		kern_sglist->addr = ctsio->kern_data_ptr;
1650		kern_sglist->len = ctsio->kern_data_len;
1651		kern_sg_entries = 1;
1652	}
1653
1654
1655	kern_watermark = 0;
1656	ext_watermark = ext_offset;
1657	len_copied = 0;
1658	for (i = ext_sg_start, j = 0;
1659	     i < ext_sg_entries && j < kern_sg_entries;) {
1660		uint8_t *ext_ptr, *kern_ptr;
1661
1662		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1663				      kern_sglist[j].len - kern_watermark);
1664
1665		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1666		ext_ptr = ext_ptr + ext_watermark;
1667		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1668			/*
1669			 * XXX KDM fix this!
1670			 */
1671			panic("need to implement bus address support");
1672#if 0
1673			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1674#endif
1675		} else
1676			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1677		kern_ptr = kern_ptr + kern_watermark;
1678
1679		kern_watermark += len_to_copy;
1680		ext_watermark += len_to_copy;
1681
1682		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1683		     CTL_FLAG_DATA_IN) {
1684			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1685					 "bytes to user\n", len_to_copy));
1686			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1687					 "to %p\n", kern_ptr, ext_ptr));
1688			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1689				ctl_set_internal_failure(ctsio,
1690							 /*sks_valid*/ 0,
1691							 /*retry_count*/ 0);
1692				goto bailout;
1693			}
1694		} else {
1695			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1696					 "bytes from user\n", len_to_copy));
1697			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1698					 "to %p\n", ext_ptr, kern_ptr));
1699			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1700				ctl_set_internal_failure(ctsio,
1701							 /*sks_valid*/ 0,
1702							 /*retry_count*/0);
1703				goto bailout;
1704			}
1705		}
1706
1707		len_copied += len_to_copy;
1708
1709		if (ext_sglist[i].len == ext_watermark) {
1710			i++;
1711			ext_watermark = 0;
1712		}
1713
1714		if (kern_sglist[j].len == kern_watermark) {
1715			j++;
1716			kern_watermark = 0;
1717		}
1718	}
1719
1720	ctsio->ext_data_filled += len_copied;
1721
1722	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1723			 "kern_sg_entries: %d\n", ext_sg_entries,
1724			 kern_sg_entries));
1725	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1726			 "kern_data_len = %d\n", ctsio->ext_data_len,
1727			 ctsio->kern_data_len));
1728
1729
1730	/* XXX KDM set residual?? */
1731bailout:
1732
1733	if (ext_sglist_malloced != 0)
1734		free(ext_sglist, M_CTL);
1735
1736	return (CTL_RETVAL_COMPLETE);
1737}
1738
1739/*
1740 * Serialize a command that went down the "wrong" side, and so was sent to
1741 * this controller for execution.  The logic is a little different than the
1742 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1743 * sent back to the other side, but in the success case, we execute the
1744 * command on this side (XFER mode) or tell the other side to execute it
1745 * (SER_ONLY mode).
1746 */
1747static int
1748ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1749{
1750	struct ctl_softc *ctl_softc;
1751	union ctl_ha_msg msg_info;
1752	struct ctl_lun *lun;
1753	int retval = 0;
1754	uint32_t targ_lun;
1755
1756	ctl_softc = control_softc;
1757
1758	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1759	lun = ctl_softc->ctl_luns[targ_lun];
1760	if (lun==NULL)
1761	{
1762		/*
1763		 * Why isn't LUN defined? The other side wouldn't
1764		 * send a cmd if the LUN is undefined.
1765		 */
1766		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1767
1768		/* "Logical unit not supported" */
1769		ctl_set_sense_data(&msg_info.scsi.sense_data,
1770				   lun,
1771				   /*sense_format*/SSD_TYPE_NONE,
1772				   /*current_error*/ 1,
1773				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1774				   /*asc*/ 0x25,
1775				   /*ascq*/ 0x00,
1776				   SSD_ELEM_NONE);
1777
1778		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1779		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1780		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1781		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1782		msg_info.hdr.serializing_sc = NULL;
1783		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1784	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1785				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1786		}
1787		return(1);
1788
1789	}
1790
1791	mtx_lock(&lun->lun_lock);
1792    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1793
1794	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1795		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1796		 ooa_links))) {
1797	case CTL_ACTION_BLOCK:
1798		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1799		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1800				  blocked_links);
1801		break;
1802	case CTL_ACTION_PASS:
1803	case CTL_ACTION_SKIP:
1804		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1805			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1806			ctl_enqueue_rtr((union ctl_io *)ctsio);
1807		} else {
1808
1809			/* send msg back to other side */
1810			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1811			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1812			msg_info.hdr.msg_type = CTL_MSG_R2R;
1813#if 0
1814			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1815#endif
1816		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1817			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1818			}
1819		}
1820		break;
1821	case CTL_ACTION_OVERLAP:
1822		/* OVERLAPPED COMMANDS ATTEMPTED */
1823		ctl_set_sense_data(&msg_info.scsi.sense_data,
1824				   lun,
1825				   /*sense_format*/SSD_TYPE_NONE,
1826				   /*current_error*/ 1,
1827				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1828				   /*asc*/ 0x4E,
1829				   /*ascq*/ 0x00,
1830				   SSD_ELEM_NONE);
1831
1832		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1833		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1834		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1835		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1836		msg_info.hdr.serializing_sc = NULL;
1837		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1838#if 0
1839		printf("BAD JUJU:Major Bummer Overlap\n");
1840#endif
1841		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1842		retval = 1;
1843		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1844		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1845		}
1846		break;
1847	case CTL_ACTION_OVERLAP_TAG:
1848		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1849		ctl_set_sense_data(&msg_info.scsi.sense_data,
1850				   lun,
1851				   /*sense_format*/SSD_TYPE_NONE,
1852				   /*current_error*/ 1,
1853				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1854				   /*asc*/ 0x4D,
1855				   /*ascq*/ ctsio->tag_num & 0xff,
1856				   SSD_ELEM_NONE);
1857
1858		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1859		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1860		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1861		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1862		msg_info.hdr.serializing_sc = NULL;
1863		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1864#if 0
1865		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1866#endif
1867		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1868		retval = 1;
1869		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1870		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1871		}
1872		break;
1873	case CTL_ACTION_ERROR:
1874	default:
1875		/* "Internal target failure" */
1876		ctl_set_sense_data(&msg_info.scsi.sense_data,
1877				   lun,
1878				   /*sense_format*/SSD_TYPE_NONE,
1879				   /*current_error*/ 1,
1880				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1881				   /*asc*/ 0x44,
1882				   /*ascq*/ 0x00,
1883				   SSD_ELEM_NONE);
1884
1885		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1886		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1887		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1888		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1889		msg_info.hdr.serializing_sc = NULL;
1890		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1891#if 0
1892		printf("BAD JUJU:Major Bummer HW Error\n");
1893#endif
1894		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1895		retval = 1;
1896		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1897		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1898		}
1899		break;
1900	}
1901	mtx_unlock(&lun->lun_lock);
1902	return (retval);
1903}
1904
1905static int
1906ctl_ioctl_submit_wait(union ctl_io *io)
1907{
1908	struct ctl_fe_ioctl_params params;
1909	ctl_fe_ioctl_state last_state;
1910	int done, retval;
1911
1912	retval = 0;
1913
1914	bzero(&params, sizeof(params));
1915
1916	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1917	cv_init(&params.sem, "ctlioccv");
1918	params.state = CTL_IOCTL_INPROG;
1919	last_state = params.state;
1920
1921	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1922
1923	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1924
1925	/* This shouldn't happen */
1926	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1927		return (retval);
1928
1929	done = 0;
1930
1931	do {
1932		mtx_lock(&params.ioctl_mtx);
1933		/*
1934		 * Check the state here, and don't sleep if the state has
1935		 * already changed (i.e. wakeup has already occured, but we
1936		 * weren't waiting yet).
1937		 */
1938		if (params.state == last_state) {
1939			/* XXX KDM cv_wait_sig instead? */
1940			cv_wait(&params.sem, &params.ioctl_mtx);
1941		}
1942		last_state = params.state;
1943
1944		switch (params.state) {
1945		case CTL_IOCTL_INPROG:
1946			/* Why did we wake up? */
1947			/* XXX KDM error here? */
1948			mtx_unlock(&params.ioctl_mtx);
1949			break;
1950		case CTL_IOCTL_DATAMOVE:
1951			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1952
1953			/*
1954			 * change last_state back to INPROG to avoid
1955			 * deadlock on subsequent data moves.
1956			 */
1957			params.state = last_state = CTL_IOCTL_INPROG;
1958
1959			mtx_unlock(&params.ioctl_mtx);
1960			ctl_ioctl_do_datamove(&io->scsiio);
1961			/*
1962			 * Note that in some cases, most notably writes,
1963			 * this will queue the I/O and call us back later.
1964			 * In other cases, generally reads, this routine
1965			 * will immediately call back and wake us up,
1966			 * probably using our own context.
1967			 */
1968			io->scsiio.be_move_done(io);
1969			break;
1970		case CTL_IOCTL_DONE:
1971			mtx_unlock(&params.ioctl_mtx);
1972			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1973			done = 1;
1974			break;
1975		default:
1976			mtx_unlock(&params.ioctl_mtx);
1977			/* XXX KDM error here? */
1978			break;
1979		}
1980	} while (done == 0);
1981
1982	mtx_destroy(&params.ioctl_mtx);
1983	cv_destroy(&params.sem);
1984
1985	return (CTL_RETVAL_COMPLETE);
1986}
1987
1988static void
1989ctl_ioctl_datamove(union ctl_io *io)
1990{
1991	struct ctl_fe_ioctl_params *params;
1992
1993	params = (struct ctl_fe_ioctl_params *)
1994		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1995
1996	mtx_lock(&params->ioctl_mtx);
1997	params->state = CTL_IOCTL_DATAMOVE;
1998	cv_broadcast(&params->sem);
1999	mtx_unlock(&params->ioctl_mtx);
2000}
2001
2002static void
2003ctl_ioctl_done(union ctl_io *io)
2004{
2005	struct ctl_fe_ioctl_params *params;
2006
2007	params = (struct ctl_fe_ioctl_params *)
2008		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2009
2010	mtx_lock(&params->ioctl_mtx);
2011	params->state = CTL_IOCTL_DONE;
2012	cv_broadcast(&params->sem);
2013	mtx_unlock(&params->ioctl_mtx);
2014}
2015
2016static void
2017ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2018{
2019	struct ctl_fe_ioctl_startstop_info *sd_info;
2020
2021	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2022
2023	sd_info->hs_info.status = metatask->status;
2024	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2025	sd_info->hs_info.luns_complete =
2026		metatask->taskinfo.startstop.luns_complete;
2027	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2028
2029	cv_broadcast(&sd_info->sem);
2030}
2031
2032static void
2033ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2034{
2035	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2036
2037	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2038
2039	mtx_lock(fe_bbr_info->lock);
2040	fe_bbr_info->bbr_info->status = metatask->status;
2041	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2042	fe_bbr_info->wakeup_done = 1;
2043	mtx_unlock(fe_bbr_info->lock);
2044
2045	cv_broadcast(&fe_bbr_info->sem);
2046}
2047
2048/*
2049 * Returns 0 for success, errno for failure.
2050 */
2051static int
2052ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2053		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2054{
2055	union ctl_io *io;
2056	int retval;
2057
2058	retval = 0;
2059
2060	mtx_lock(&lun->lun_lock);
2061	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2062	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2063	     ooa_links)) {
2064		struct ctl_ooa_entry *entry;
2065
2066		/*
2067		 * If we've got more than we can fit, just count the
2068		 * remaining entries.
2069		 */
2070		if (*cur_fill_num >= ooa_hdr->alloc_num)
2071			continue;
2072
2073		entry = &kern_entries[*cur_fill_num];
2074
2075		entry->tag_num = io->scsiio.tag_num;
2076		entry->lun_num = lun->lun;
2077#ifdef CTL_TIME_IO
2078		entry->start_bt = io->io_hdr.start_bt;
2079#endif
2080		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2081		entry->cdb_len = io->scsiio.cdb_len;
2082		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2083			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2084
2085		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2086			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2087
2088		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2089			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2090
2091		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2092			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2093
2094		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2095			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2096	}
2097	mtx_unlock(&lun->lun_lock);
2098
2099	return (retval);
2100}
2101
2102static void *
2103ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2104		 size_t error_str_len)
2105{
2106	void *kptr;
2107
2108	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2109
2110	if (copyin(user_addr, kptr, len) != 0) {
2111		snprintf(error_str, error_str_len, "Error copying %d bytes "
2112			 "from user address %p to kernel address %p", len,
2113			 user_addr, kptr);
2114		free(kptr, M_CTL);
2115		return (NULL);
2116	}
2117
2118	return (kptr);
2119}
2120
2121static void
2122ctl_free_args(int num_args, struct ctl_be_arg *args)
2123{
2124	int i;
2125
2126	if (args == NULL)
2127		return;
2128
2129	for (i = 0; i < num_args; i++) {
2130		free(args[i].kname, M_CTL);
2131		free(args[i].kvalue, M_CTL);
2132	}
2133
2134	free(args, M_CTL);
2135}
2136
2137static struct ctl_be_arg *
2138ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2139		char *error_str, size_t error_str_len)
2140{
2141	struct ctl_be_arg *args;
2142	int i;
2143
2144	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2145				error_str, error_str_len);
2146
2147	if (args == NULL)
2148		goto bailout;
2149
2150	for (i = 0; i < num_args; i++) {
2151		args[i].kname = NULL;
2152		args[i].kvalue = NULL;
2153	}
2154
2155	for (i = 0; i < num_args; i++) {
2156		uint8_t *tmpptr;
2157
2158		args[i].kname = ctl_copyin_alloc(args[i].name,
2159			args[i].namelen, error_str, error_str_len);
2160		if (args[i].kname == NULL)
2161			goto bailout;
2162
2163		if (args[i].kname[args[i].namelen - 1] != '\0') {
2164			snprintf(error_str, error_str_len, "Argument %d "
2165				 "name is not NUL-terminated", i);
2166			goto bailout;
2167		}
2168
2169		if (args[i].flags & CTL_BEARG_RD) {
2170			tmpptr = ctl_copyin_alloc(args[i].value,
2171				args[i].vallen, error_str, error_str_len);
2172			if (tmpptr == NULL)
2173				goto bailout;
2174			if ((args[i].flags & CTL_BEARG_ASCII)
2175			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2176				snprintf(error_str, error_str_len, "Argument "
2177				    "%d value is not NUL-terminated", i);
2178				goto bailout;
2179			}
2180			args[i].kvalue = tmpptr;
2181		} else {
2182			args[i].kvalue = malloc(args[i].vallen,
2183			    M_CTL, M_WAITOK | M_ZERO);
2184		}
2185	}
2186
2187	return (args);
2188bailout:
2189
2190	ctl_free_args(num_args, args);
2191
2192	return (NULL);
2193}
2194
2195static void
2196ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2197{
2198	int i;
2199
2200	for (i = 0; i < num_args; i++) {
2201		if (args[i].flags & CTL_BEARG_WR)
2202			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2203	}
2204}
2205
2206/*
2207 * Escape characters that are illegal or not recommended in XML.
2208 */
2209int
2210ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2211{
2212	int retval;
2213
2214	retval = 0;
2215
2216	for (; *str; str++) {
2217		switch (*str) {
2218		case '&':
2219			retval = sbuf_printf(sb, "&amp;");
2220			break;
2221		case '>':
2222			retval = sbuf_printf(sb, "&gt;");
2223			break;
2224		case '<':
2225			retval = sbuf_printf(sb, "&lt;");
2226			break;
2227		default:
2228			retval = sbuf_putc(sb, *str);
2229			break;
2230		}
2231
2232		if (retval != 0)
2233			break;
2234
2235	}
2236
2237	return (retval);
2238}
2239
2240static int
2241ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2242	  struct thread *td)
2243{
2244	struct ctl_softc *softc;
2245	int retval;
2246
2247	softc = control_softc;
2248
2249	retval = 0;
2250
2251	switch (cmd) {
2252	case CTL_IO: {
2253		union ctl_io *io;
2254		void *pool_tmp;
2255
2256		/*
2257		 * If we haven't been "enabled", don't allow any SCSI I/O
2258		 * to this FETD.
2259		 */
2260		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2261			retval = EPERM;
2262			break;
2263		}
2264
2265		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2266		if (io == NULL) {
2267			printf("ctl_ioctl: can't allocate ctl_io!\n");
2268			retval = ENOSPC;
2269			break;
2270		}
2271
2272		/*
2273		 * Need to save the pool reference so it doesn't get
2274		 * spammed by the user's ctl_io.
2275		 */
2276		pool_tmp = io->io_hdr.pool;
2277
2278		memcpy(io, (void *)addr, sizeof(*io));
2279
2280		io->io_hdr.pool = pool_tmp;
2281		/*
2282		 * No status yet, so make sure the status is set properly.
2283		 */
2284		io->io_hdr.status = CTL_STATUS_NONE;
2285
2286		/*
2287		 * The user sets the initiator ID, target and LUN IDs.
2288		 */
2289		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2290		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2291		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2292		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2293			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2294
2295		retval = ctl_ioctl_submit_wait(io);
2296
2297		if (retval != 0) {
2298			ctl_free_io(io);
2299			break;
2300		}
2301
2302		memcpy((void *)addr, io, sizeof(*io));
2303
2304		/* return this to our pool */
2305		ctl_free_io(io);
2306
2307		break;
2308	}
2309	case CTL_ENABLE_PORT:
2310	case CTL_DISABLE_PORT:
2311	case CTL_SET_PORT_WWNS: {
2312		struct ctl_port *port;
2313		struct ctl_port_entry *entry;
2314
2315		entry = (struct ctl_port_entry *)addr;
2316
2317		mtx_lock(&softc->ctl_lock);
2318		STAILQ_FOREACH(port, &softc->port_list, links) {
2319			int action, done;
2320
2321			action = 0;
2322			done = 0;
2323
2324			if ((entry->port_type == CTL_PORT_NONE)
2325			 && (entry->targ_port == port->targ_port)) {
2326				/*
2327				 * If the user only wants to enable or
2328				 * disable or set WWNs on a specific port,
2329				 * do the operation and we're done.
2330				 */
2331				action = 1;
2332				done = 1;
2333			} else if (entry->port_type & port->port_type) {
2334				/*
2335				 * Compare the user's type mask with the
2336				 * particular frontend type to see if we
2337				 * have a match.
2338				 */
2339				action = 1;
2340				done = 0;
2341
2342				/*
2343				 * Make sure the user isn't trying to set
2344				 * WWNs on multiple ports at the same time.
2345				 */
2346				if (cmd == CTL_SET_PORT_WWNS) {
2347					printf("%s: Can't set WWNs on "
2348					       "multiple ports\n", __func__);
2349					retval = EINVAL;
2350					break;
2351				}
2352			}
2353			if (action != 0) {
2354				/*
2355				 * XXX KDM we have to drop the lock here,
2356				 * because the online/offline operations
2357				 * can potentially block.  We need to
2358				 * reference count the frontends so they
2359				 * can't go away,
2360				 */
2361				mtx_unlock(&softc->ctl_lock);
2362
2363				if (cmd == CTL_ENABLE_PORT) {
2364					struct ctl_lun *lun;
2365
2366					STAILQ_FOREACH(lun, &softc->lun_list,
2367						       links) {
2368						port->lun_enable(port->targ_lun_arg,
2369						    lun->target,
2370						    lun->lun);
2371					}
2372
2373					ctl_port_online(port);
2374				} else if (cmd == CTL_DISABLE_PORT) {
2375					struct ctl_lun *lun;
2376
2377					ctl_port_offline(port);
2378
2379					STAILQ_FOREACH(lun, &softc->lun_list,
2380						       links) {
2381						port->lun_disable(
2382						    port->targ_lun_arg,
2383						    lun->target,
2384						    lun->lun);
2385					}
2386				}
2387
2388				mtx_lock(&softc->ctl_lock);
2389
2390				if (cmd == CTL_SET_PORT_WWNS)
2391					ctl_port_set_wwns(port,
2392					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2393					    1 : 0, entry->wwnn,
2394					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2395					    1 : 0, entry->wwpn);
2396			}
2397			if (done != 0)
2398				break;
2399		}
2400		mtx_unlock(&softc->ctl_lock);
2401		break;
2402	}
2403	case CTL_GET_PORT_LIST: {
2404		struct ctl_port *port;
2405		struct ctl_port_list *list;
2406		int i;
2407
2408		list = (struct ctl_port_list *)addr;
2409
2410		if (list->alloc_len != (list->alloc_num *
2411		    sizeof(struct ctl_port_entry))) {
2412			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2413			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2414			       "%zu\n", __func__, list->alloc_len,
2415			       list->alloc_num, sizeof(struct ctl_port_entry));
2416			retval = EINVAL;
2417			break;
2418		}
2419		list->fill_len = 0;
2420		list->fill_num = 0;
2421		list->dropped_num = 0;
2422		i = 0;
2423		mtx_lock(&softc->ctl_lock);
2424		STAILQ_FOREACH(port, &softc->port_list, links) {
2425			struct ctl_port_entry entry, *list_entry;
2426
2427			if (list->fill_num >= list->alloc_num) {
2428				list->dropped_num++;
2429				continue;
2430			}
2431
2432			entry.port_type = port->port_type;
2433			strlcpy(entry.port_name, port->port_name,
2434				sizeof(entry.port_name));
2435			entry.targ_port = port->targ_port;
2436			entry.physical_port = port->physical_port;
2437			entry.virtual_port = port->virtual_port;
2438			entry.wwnn = port->wwnn;
2439			entry.wwpn = port->wwpn;
2440			if (port->status & CTL_PORT_STATUS_ONLINE)
2441				entry.online = 1;
2442			else
2443				entry.online = 0;
2444
2445			list_entry = &list->entries[i];
2446
2447			retval = copyout(&entry, list_entry, sizeof(entry));
2448			if (retval != 0) {
2449				printf("%s: CTL_GET_PORT_LIST: copyout "
2450				       "returned %d\n", __func__, retval);
2451				break;
2452			}
2453			i++;
2454			list->fill_num++;
2455			list->fill_len += sizeof(entry);
2456		}
2457		mtx_unlock(&softc->ctl_lock);
2458
2459		/*
2460		 * If this is non-zero, we had a copyout fault, so there's
2461		 * probably no point in attempting to set the status inside
2462		 * the structure.
2463		 */
2464		if (retval != 0)
2465			break;
2466
2467		if (list->dropped_num > 0)
2468			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2469		else
2470			list->status = CTL_PORT_LIST_OK;
2471		break;
2472	}
2473	case CTL_DUMP_OOA: {
2474		struct ctl_lun *lun;
2475		union ctl_io *io;
2476		char printbuf[128];
2477		struct sbuf sb;
2478
2479		mtx_lock(&softc->ctl_lock);
2480		printf("Dumping OOA queues:\n");
2481		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2482			mtx_lock(&lun->lun_lock);
2483			for (io = (union ctl_io *)TAILQ_FIRST(
2484			     &lun->ooa_queue); io != NULL;
2485			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2486			     ooa_links)) {
2487				sbuf_new(&sb, printbuf, sizeof(printbuf),
2488					 SBUF_FIXEDLEN);
2489				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2490					    (intmax_t)lun->lun,
2491					    io->scsiio.tag_num,
2492					    (io->io_hdr.flags &
2493					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2494					    (io->io_hdr.flags &
2495					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2496					    (io->io_hdr.flags &
2497					    CTL_FLAG_ABORT) ? " ABORT" : "",
2498			                    (io->io_hdr.flags &
2499		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2500				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2501				sbuf_finish(&sb);
2502				printf("%s\n", sbuf_data(&sb));
2503			}
2504			mtx_unlock(&lun->lun_lock);
2505		}
2506		printf("OOA queues dump done\n");
2507		mtx_unlock(&softc->ctl_lock);
2508		break;
2509	}
2510	case CTL_GET_OOA: {
2511		struct ctl_lun *lun;
2512		struct ctl_ooa *ooa_hdr;
2513		struct ctl_ooa_entry *entries;
2514		uint32_t cur_fill_num;
2515
2516		ooa_hdr = (struct ctl_ooa *)addr;
2517
2518		if ((ooa_hdr->alloc_len == 0)
2519		 || (ooa_hdr->alloc_num == 0)) {
2520			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2521			       "must be non-zero\n", __func__,
2522			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2523			retval = EINVAL;
2524			break;
2525		}
2526
2527		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2528		    sizeof(struct ctl_ooa_entry))) {
2529			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2530			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2531			       __func__, ooa_hdr->alloc_len,
2532			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2533			retval = EINVAL;
2534			break;
2535		}
2536
2537		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2538		if (entries == NULL) {
2539			printf("%s: could not allocate %d bytes for OOA "
2540			       "dump\n", __func__, ooa_hdr->alloc_len);
2541			retval = ENOMEM;
2542			break;
2543		}
2544
2545		mtx_lock(&softc->ctl_lock);
2546		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2547		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2548		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2549			mtx_unlock(&softc->ctl_lock);
2550			free(entries, M_CTL);
2551			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2552			       __func__, (uintmax_t)ooa_hdr->lun_num);
2553			retval = EINVAL;
2554			break;
2555		}
2556
2557		cur_fill_num = 0;
2558
2559		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2560			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2561				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2562					ooa_hdr, entries);
2563				if (retval != 0)
2564					break;
2565			}
2566			if (retval != 0) {
2567				mtx_unlock(&softc->ctl_lock);
2568				free(entries, M_CTL);
2569				break;
2570			}
2571		} else {
2572			lun = softc->ctl_luns[ooa_hdr->lun_num];
2573
2574			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2575						    entries);
2576		}
2577		mtx_unlock(&softc->ctl_lock);
2578
2579		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2580		ooa_hdr->fill_len = ooa_hdr->fill_num *
2581			sizeof(struct ctl_ooa_entry);
2582		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2583		if (retval != 0) {
2584			printf("%s: error copying out %d bytes for OOA dump\n",
2585			       __func__, ooa_hdr->fill_len);
2586		}
2587
2588		getbintime(&ooa_hdr->cur_bt);
2589
2590		if (cur_fill_num > ooa_hdr->alloc_num) {
2591			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2592			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2593		} else {
2594			ooa_hdr->dropped_num = 0;
2595			ooa_hdr->status = CTL_OOA_OK;
2596		}
2597
2598		free(entries, M_CTL);
2599		break;
2600	}
2601	case CTL_CHECK_OOA: {
2602		union ctl_io *io;
2603		struct ctl_lun *lun;
2604		struct ctl_ooa_info *ooa_info;
2605
2606
2607		ooa_info = (struct ctl_ooa_info *)addr;
2608
2609		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2610			ooa_info->status = CTL_OOA_INVALID_LUN;
2611			break;
2612		}
2613		mtx_lock(&softc->ctl_lock);
2614		lun = softc->ctl_luns[ooa_info->lun_id];
2615		if (lun == NULL) {
2616			mtx_unlock(&softc->ctl_lock);
2617			ooa_info->status = CTL_OOA_INVALID_LUN;
2618			break;
2619		}
2620		mtx_lock(&lun->lun_lock);
2621		mtx_unlock(&softc->ctl_lock);
2622		ooa_info->num_entries = 0;
2623		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2624		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2625		     &io->io_hdr, ooa_links)) {
2626			ooa_info->num_entries++;
2627		}
2628		mtx_unlock(&lun->lun_lock);
2629
2630		ooa_info->status = CTL_OOA_SUCCESS;
2631
2632		break;
2633	}
2634	case CTL_HARD_START:
2635	case CTL_HARD_STOP: {
2636		struct ctl_fe_ioctl_startstop_info ss_info;
2637		struct cfi_metatask *metatask;
2638		struct mtx hs_mtx;
2639
2640		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2641
2642		cv_init(&ss_info.sem, "hard start/stop cv" );
2643
2644		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2645		if (metatask == NULL) {
2646			retval = ENOMEM;
2647			mtx_destroy(&hs_mtx);
2648			break;
2649		}
2650
2651		if (cmd == CTL_HARD_START)
2652			metatask->tasktype = CFI_TASK_STARTUP;
2653		else
2654			metatask->tasktype = CFI_TASK_SHUTDOWN;
2655
2656		metatask->callback = ctl_ioctl_hard_startstop_callback;
2657		metatask->callback_arg = &ss_info;
2658
2659		cfi_action(metatask);
2660
2661		/* Wait for the callback */
2662		mtx_lock(&hs_mtx);
2663		cv_wait_sig(&ss_info.sem, &hs_mtx);
2664		mtx_unlock(&hs_mtx);
2665
2666		/*
2667		 * All information has been copied from the metatask by the
2668		 * time cv_broadcast() is called, so we free the metatask here.
2669		 */
2670		cfi_free_metatask(metatask);
2671
2672		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2673
2674		mtx_destroy(&hs_mtx);
2675		break;
2676	}
2677	case CTL_BBRREAD: {
2678		struct ctl_bbrread_info *bbr_info;
2679		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2680		struct mtx bbr_mtx;
2681		struct cfi_metatask *metatask;
2682
2683		bbr_info = (struct ctl_bbrread_info *)addr;
2684
2685		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2686
2687		bzero(&bbr_mtx, sizeof(bbr_mtx));
2688		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2689
2690		fe_bbr_info.bbr_info = bbr_info;
2691		fe_bbr_info.lock = &bbr_mtx;
2692
2693		cv_init(&fe_bbr_info.sem, "BBR read cv");
2694		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2695
2696		if (metatask == NULL) {
2697			mtx_destroy(&bbr_mtx);
2698			cv_destroy(&fe_bbr_info.sem);
2699			retval = ENOMEM;
2700			break;
2701		}
2702		metatask->tasktype = CFI_TASK_BBRREAD;
2703		metatask->callback = ctl_ioctl_bbrread_callback;
2704		metatask->callback_arg = &fe_bbr_info;
2705		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2706		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2707		metatask->taskinfo.bbrread.len = bbr_info->len;
2708
2709		cfi_action(metatask);
2710
2711		mtx_lock(&bbr_mtx);
2712		while (fe_bbr_info.wakeup_done == 0)
2713			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2714		mtx_unlock(&bbr_mtx);
2715
2716		bbr_info->status = metatask->status;
2717		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2718		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2719		memcpy(&bbr_info->sense_data,
2720		       &metatask->taskinfo.bbrread.sense_data,
2721		       ctl_min(sizeof(bbr_info->sense_data),
2722			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2723
2724		cfi_free_metatask(metatask);
2725
2726		mtx_destroy(&bbr_mtx);
2727		cv_destroy(&fe_bbr_info.sem);
2728
2729		break;
2730	}
2731	case CTL_DELAY_IO: {
2732		struct ctl_io_delay_info *delay_info;
2733#ifdef CTL_IO_DELAY
2734		struct ctl_lun *lun;
2735#endif /* CTL_IO_DELAY */
2736
2737		delay_info = (struct ctl_io_delay_info *)addr;
2738
2739#ifdef CTL_IO_DELAY
2740		mtx_lock(&softc->ctl_lock);
2741
2742		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2743		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2744			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2745		} else {
2746			lun = softc->ctl_luns[delay_info->lun_id];
2747			mtx_lock(&lun->lun_lock);
2748
2749			delay_info->status = CTL_DELAY_STATUS_OK;
2750
2751			switch (delay_info->delay_type) {
2752			case CTL_DELAY_TYPE_CONT:
2753				break;
2754			case CTL_DELAY_TYPE_ONESHOT:
2755				break;
2756			default:
2757				delay_info->status =
2758					CTL_DELAY_STATUS_INVALID_TYPE;
2759				break;
2760			}
2761
2762			switch (delay_info->delay_loc) {
2763			case CTL_DELAY_LOC_DATAMOVE:
2764				lun->delay_info.datamove_type =
2765					delay_info->delay_type;
2766				lun->delay_info.datamove_delay =
2767					delay_info->delay_secs;
2768				break;
2769			case CTL_DELAY_LOC_DONE:
2770				lun->delay_info.done_type =
2771					delay_info->delay_type;
2772				lun->delay_info.done_delay =
2773					delay_info->delay_secs;
2774				break;
2775			default:
2776				delay_info->status =
2777					CTL_DELAY_STATUS_INVALID_LOC;
2778				break;
2779			}
2780			mtx_unlock(&lun->lun_lock);
2781		}
2782
2783		mtx_unlock(&softc->ctl_lock);
2784#else
2785		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2786#endif /* CTL_IO_DELAY */
2787		break;
2788	}
2789	case CTL_REALSYNC_SET: {
2790		int *syncstate;
2791
2792		syncstate = (int *)addr;
2793
2794		mtx_lock(&softc->ctl_lock);
2795		switch (*syncstate) {
2796		case 0:
2797			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2798			break;
2799		case 1:
2800			softc->flags |= CTL_FLAG_REAL_SYNC;
2801			break;
2802		default:
2803			retval = EINVAL;
2804			break;
2805		}
2806		mtx_unlock(&softc->ctl_lock);
2807		break;
2808	}
2809	case CTL_REALSYNC_GET: {
2810		int *syncstate;
2811
2812		syncstate = (int*)addr;
2813
2814		mtx_lock(&softc->ctl_lock);
2815		if (softc->flags & CTL_FLAG_REAL_SYNC)
2816			*syncstate = 1;
2817		else
2818			*syncstate = 0;
2819		mtx_unlock(&softc->ctl_lock);
2820
2821		break;
2822	}
2823	case CTL_SETSYNC:
2824	case CTL_GETSYNC: {
2825		struct ctl_sync_info *sync_info;
2826		struct ctl_lun *lun;
2827
2828		sync_info = (struct ctl_sync_info *)addr;
2829
2830		mtx_lock(&softc->ctl_lock);
2831		lun = softc->ctl_luns[sync_info->lun_id];
2832		if (lun == NULL) {
2833			mtx_unlock(&softc->ctl_lock);
2834			sync_info->status = CTL_GS_SYNC_NO_LUN;
2835		}
2836		/*
2837		 * Get or set the sync interval.  We're not bounds checking
2838		 * in the set case, hopefully the user won't do something
2839		 * silly.
2840		 */
2841		mtx_lock(&lun->lun_lock);
2842		mtx_unlock(&softc->ctl_lock);
2843		if (cmd == CTL_GETSYNC)
2844			sync_info->sync_interval = lun->sync_interval;
2845		else
2846			lun->sync_interval = sync_info->sync_interval;
2847		mtx_unlock(&lun->lun_lock);
2848
2849		sync_info->status = CTL_GS_SYNC_OK;
2850
2851		break;
2852	}
2853	case CTL_GETSTATS: {
2854		struct ctl_stats *stats;
2855		struct ctl_lun *lun;
2856		int i;
2857
2858		stats = (struct ctl_stats *)addr;
2859
2860		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2861		     stats->alloc_len) {
2862			stats->status = CTL_SS_NEED_MORE_SPACE;
2863			stats->num_luns = softc->num_luns;
2864			break;
2865		}
2866		/*
2867		 * XXX KDM no locking here.  If the LUN list changes,
2868		 * things can blow up.
2869		 */
2870		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2871		     i++, lun = STAILQ_NEXT(lun, links)) {
2872			retval = copyout(&lun->stats, &stats->lun_stats[i],
2873					 sizeof(lun->stats));
2874			if (retval != 0)
2875				break;
2876		}
2877		stats->num_luns = softc->num_luns;
2878		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2879				 softc->num_luns;
2880		stats->status = CTL_SS_OK;
2881#ifdef CTL_TIME_IO
2882		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2883#else
2884		stats->flags = CTL_STATS_FLAG_NONE;
2885#endif
2886		getnanouptime(&stats->timestamp);
2887		break;
2888	}
2889	case CTL_ERROR_INJECT: {
2890		struct ctl_error_desc *err_desc, *new_err_desc;
2891		struct ctl_lun *lun;
2892
2893		err_desc = (struct ctl_error_desc *)addr;
2894
2895		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2896				      M_WAITOK | M_ZERO);
2897		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2898
2899		mtx_lock(&softc->ctl_lock);
2900		lun = softc->ctl_luns[err_desc->lun_id];
2901		if (lun == NULL) {
2902			mtx_unlock(&softc->ctl_lock);
2903			free(new_err_desc, M_CTL);
2904			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2905			       __func__, (uintmax_t)err_desc->lun_id);
2906			retval = EINVAL;
2907			break;
2908		}
2909		mtx_lock(&lun->lun_lock);
2910		mtx_unlock(&softc->ctl_lock);
2911
2912		/*
2913		 * We could do some checking here to verify the validity
2914		 * of the request, but given the complexity of error
2915		 * injection requests, the checking logic would be fairly
2916		 * complex.
2917		 *
2918		 * For now, if the request is invalid, it just won't get
2919		 * executed and might get deleted.
2920		 */
2921		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2922
2923		/*
2924		 * XXX KDM check to make sure the serial number is unique,
2925		 * in case we somehow manage to wrap.  That shouldn't
2926		 * happen for a very long time, but it's the right thing to
2927		 * do.
2928		 */
2929		new_err_desc->serial = lun->error_serial;
2930		err_desc->serial = lun->error_serial;
2931		lun->error_serial++;
2932
2933		mtx_unlock(&lun->lun_lock);
2934		break;
2935	}
2936	case CTL_ERROR_INJECT_DELETE: {
2937		struct ctl_error_desc *delete_desc, *desc, *desc2;
2938		struct ctl_lun *lun;
2939		int delete_done;
2940
2941		delete_desc = (struct ctl_error_desc *)addr;
2942		delete_done = 0;
2943
2944		mtx_lock(&softc->ctl_lock);
2945		lun = softc->ctl_luns[delete_desc->lun_id];
2946		if (lun == NULL) {
2947			mtx_unlock(&softc->ctl_lock);
2948			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2949			       __func__, (uintmax_t)delete_desc->lun_id);
2950			retval = EINVAL;
2951			break;
2952		}
2953		mtx_lock(&lun->lun_lock);
2954		mtx_unlock(&softc->ctl_lock);
2955		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2956			if (desc->serial != delete_desc->serial)
2957				continue;
2958
2959			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2960				      links);
2961			free(desc, M_CTL);
2962			delete_done = 1;
2963		}
2964		mtx_unlock(&lun->lun_lock);
2965		if (delete_done == 0) {
2966			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2967			       "error serial %ju on LUN %u\n", __func__,
2968			       delete_desc->serial, delete_desc->lun_id);
2969			retval = EINVAL;
2970			break;
2971		}
2972		break;
2973	}
2974	case CTL_DUMP_STRUCTS: {
2975		int i, j, k, idx;
2976		struct ctl_port *port;
2977		struct ctl_frontend *fe;
2978
2979		mtx_lock(&softc->ctl_lock);
2980		printf("CTL Persistent Reservation information start:\n");
2981		for (i = 0; i < CTL_MAX_LUNS; i++) {
2982			struct ctl_lun *lun;
2983
2984			lun = softc->ctl_luns[i];
2985
2986			if ((lun == NULL)
2987			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2988				continue;
2989
2990			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2991				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2992					idx = j * CTL_MAX_INIT_PER_PORT + k;
2993					if (lun->per_res[idx].registered == 0)
2994						continue;
2995					printf("  LUN %d port %d iid %d key "
2996					       "%#jx\n", i, j, k,
2997					       (uintmax_t)scsi_8btou64(
2998					       lun->per_res[idx].res_key.key));
2999				}
3000			}
3001		}
3002		printf("CTL Persistent Reservation information end\n");
3003		printf("CTL Ports:\n");
3004		STAILQ_FOREACH(port, &softc->port_list, links) {
3005			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3006			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3007			       port->frontend->name, port->port_type,
3008			       port->physical_port, port->virtual_port,
3009			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3010			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3011				if (port->wwpn_iid[j].in_use == 0 &&
3012				    port->wwpn_iid[j].wwpn == 0 &&
3013				    port->wwpn_iid[j].name == NULL)
3014					continue;
3015
3016				printf("    iid %u use %d WWPN %#jx '%s'\n",
3017				    j, port->wwpn_iid[j].in_use,
3018				    (uintmax_t)port->wwpn_iid[j].wwpn,
3019				    port->wwpn_iid[j].name);
3020			}
3021		}
3022		printf("CTL Port information end\n");
3023		mtx_unlock(&softc->ctl_lock);
3024		/*
3025		 * XXX KDM calling this without a lock.  We'd likely want
3026		 * to drop the lock before calling the frontend's dump
3027		 * routine anyway.
3028		 */
3029		printf("CTL Frontends:\n");
3030		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3031			printf("  Frontend '%s'\n", fe->name);
3032			if (fe->fe_dump != NULL)
3033				fe->fe_dump();
3034		}
3035		printf("CTL Frontend information end\n");
3036		break;
3037	}
3038	case CTL_LUN_REQ: {
3039		struct ctl_lun_req *lun_req;
3040		struct ctl_backend_driver *backend;
3041
3042		lun_req = (struct ctl_lun_req *)addr;
3043
3044		backend = ctl_backend_find(lun_req->backend);
3045		if (backend == NULL) {
3046			lun_req->status = CTL_LUN_ERROR;
3047			snprintf(lun_req->error_str,
3048				 sizeof(lun_req->error_str),
3049				 "Backend \"%s\" not found.",
3050				 lun_req->backend);
3051			break;
3052		}
3053		if (lun_req->num_be_args > 0) {
3054			lun_req->kern_be_args = ctl_copyin_args(
3055				lun_req->num_be_args,
3056				lun_req->be_args,
3057				lun_req->error_str,
3058				sizeof(lun_req->error_str));
3059			if (lun_req->kern_be_args == NULL) {
3060				lun_req->status = CTL_LUN_ERROR;
3061				break;
3062			}
3063		}
3064
3065		retval = backend->ioctl(dev, cmd, addr, flag, td);
3066
3067		if (lun_req->num_be_args > 0) {
3068			ctl_copyout_args(lun_req->num_be_args,
3069				      lun_req->kern_be_args);
3070			ctl_free_args(lun_req->num_be_args,
3071				      lun_req->kern_be_args);
3072		}
3073		break;
3074	}
3075	case CTL_LUN_LIST: {
3076		struct sbuf *sb;
3077		struct ctl_lun *lun;
3078		struct ctl_lun_list *list;
3079		struct ctl_option *opt;
3080
3081		list = (struct ctl_lun_list *)addr;
3082
3083		/*
3084		 * Allocate a fixed length sbuf here, based on the length
3085		 * of the user's buffer.  We could allocate an auto-extending
3086		 * buffer, and then tell the user how much larger our
3087		 * amount of data is than his buffer, but that presents
3088		 * some problems:
3089		 *
3090		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3091		 *     we can't hold a lock while calling them with an
3092		 *     auto-extending buffer.
3093 		 *
3094		 * 2.  There is not currently a LUN reference counting
3095		 *     mechanism, outside of outstanding transactions on
3096		 *     the LUN's OOA queue.  So a LUN could go away on us
3097		 *     while we're getting the LUN number, backend-specific
3098		 *     information, etc.  Thus, given the way things
3099		 *     currently work, we need to hold the CTL lock while
3100		 *     grabbing LUN information.
3101		 *
3102		 * So, from the user's standpoint, the best thing to do is
3103		 * allocate what he thinks is a reasonable buffer length,
3104		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3105		 * double the buffer length and try again.  (And repeat
3106		 * that until he succeeds.)
3107		 */
3108		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3109		if (sb == NULL) {
3110			list->status = CTL_LUN_LIST_ERROR;
3111			snprintf(list->error_str, sizeof(list->error_str),
3112				 "Unable to allocate %d bytes for LUN list",
3113				 list->alloc_len);
3114			break;
3115		}
3116
3117		sbuf_printf(sb, "<ctllunlist>\n");
3118
3119		mtx_lock(&softc->ctl_lock);
3120		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3121			mtx_lock(&lun->lun_lock);
3122			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3123					     (uintmax_t)lun->lun);
3124
3125			/*
3126			 * Bail out as soon as we see that we've overfilled
3127			 * the buffer.
3128			 */
3129			if (retval != 0)
3130				break;
3131
3132			retval = sbuf_printf(sb, "\t<backend_type>%s"
3133					     "</backend_type>\n",
3134					     (lun->backend == NULL) ?  "none" :
3135					     lun->backend->name);
3136
3137			if (retval != 0)
3138				break;
3139
3140			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3141					     lun->be_lun->lun_type);
3142
3143			if (retval != 0)
3144				break;
3145
3146			if (lun->backend == NULL) {
3147				retval = sbuf_printf(sb, "</lun>\n");
3148				if (retval != 0)
3149					break;
3150				continue;
3151			}
3152
3153			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3154					     (lun->be_lun->maxlba > 0) ?
3155					     lun->be_lun->maxlba + 1 : 0);
3156
3157			if (retval != 0)
3158				break;
3159
3160			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3161					     lun->be_lun->blocksize);
3162
3163			if (retval != 0)
3164				break;
3165
3166			retval = sbuf_printf(sb, "\t<serial_number>");
3167
3168			if (retval != 0)
3169				break;
3170
3171			retval = ctl_sbuf_printf_esc(sb,
3172						     lun->be_lun->serial_num);
3173
3174			if (retval != 0)
3175				break;
3176
3177			retval = sbuf_printf(sb, "</serial_number>\n");
3178
3179			if (retval != 0)
3180				break;
3181
3182			retval = sbuf_printf(sb, "\t<device_id>");
3183
3184			if (retval != 0)
3185				break;
3186
3187			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3188
3189			if (retval != 0)
3190				break;
3191
3192			retval = sbuf_printf(sb, "</device_id>\n");
3193
3194			if (retval != 0)
3195				break;
3196
3197			if (lun->backend->lun_info != NULL) {
3198				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3199				if (retval != 0)
3200					break;
3201			}
3202			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3203				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3204				    opt->name, opt->value, opt->name);
3205				if (retval != 0)
3206					break;
3207			}
3208
3209			retval = sbuf_printf(sb, "</lun>\n");
3210
3211			if (retval != 0)
3212				break;
3213			mtx_unlock(&lun->lun_lock);
3214		}
3215		if (lun != NULL)
3216			mtx_unlock(&lun->lun_lock);
3217		mtx_unlock(&softc->ctl_lock);
3218
3219		if ((retval != 0)
3220		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3221			retval = 0;
3222			sbuf_delete(sb);
3223			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3224			snprintf(list->error_str, sizeof(list->error_str),
3225				 "Out of space, %d bytes is too small",
3226				 list->alloc_len);
3227			break;
3228		}
3229
3230		sbuf_finish(sb);
3231
3232		retval = copyout(sbuf_data(sb), list->lun_xml,
3233				 sbuf_len(sb) + 1);
3234
3235		list->fill_len = sbuf_len(sb) + 1;
3236		list->status = CTL_LUN_LIST_OK;
3237		sbuf_delete(sb);
3238		break;
3239	}
3240	case CTL_ISCSI: {
3241		struct ctl_iscsi *ci;
3242		struct ctl_frontend *fe;
3243
3244		ci = (struct ctl_iscsi *)addr;
3245
3246		fe = ctl_frontend_find("iscsi");
3247		if (fe == NULL) {
3248			ci->status = CTL_ISCSI_ERROR;
3249			snprintf(ci->error_str, sizeof(ci->error_str),
3250			    "Frontend \"iscsi\" not found.");
3251			break;
3252		}
3253
3254		retval = fe->ioctl(dev, cmd, addr, flag, td);
3255		break;
3256	}
3257	case CTL_PORT_REQ: {
3258		struct ctl_req *req;
3259		struct ctl_frontend *fe;
3260
3261		req = (struct ctl_req *)addr;
3262
3263		fe = ctl_frontend_find(req->driver);
3264		if (fe == NULL) {
3265			req->status = CTL_LUN_ERROR;
3266			snprintf(req->error_str, sizeof(req->error_str),
3267			    "Frontend \"%s\" not found.", req->driver);
3268			break;
3269		}
3270		if (req->num_args > 0) {
3271			req->kern_args = ctl_copyin_args(req->num_args,
3272			    req->args, req->error_str, sizeof(req->error_str));
3273			if (req->kern_args == NULL) {
3274				req->status = CTL_LUN_ERROR;
3275				break;
3276			}
3277		}
3278
3279		retval = fe->ioctl(dev, cmd, addr, flag, td);
3280
3281		if (req->num_args > 0) {
3282			ctl_copyout_args(req->num_args, req->kern_args);
3283			ctl_free_args(req->num_args, req->kern_args);
3284		}
3285		break;
3286	}
3287	case CTL_PORT_LIST: {
3288		struct sbuf *sb;
3289		struct ctl_port *port;
3290		struct ctl_lun_list *list;
3291		struct ctl_option *opt;
3292
3293		list = (struct ctl_lun_list *)addr;
3294
3295		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3296		if (sb == NULL) {
3297			list->status = CTL_LUN_LIST_ERROR;
3298			snprintf(list->error_str, sizeof(list->error_str),
3299				 "Unable to allocate %d bytes for LUN list",
3300				 list->alloc_len);
3301			break;
3302		}
3303
3304		sbuf_printf(sb, "<ctlportlist>\n");
3305
3306		mtx_lock(&softc->ctl_lock);
3307		STAILQ_FOREACH(port, &softc->port_list, links) {
3308			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3309					     (uintmax_t)port->targ_port);
3310
3311			/*
3312			 * Bail out as soon as we see that we've overfilled
3313			 * the buffer.
3314			 */
3315			if (retval != 0)
3316				break;
3317
3318			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3319			    "</frontend_type>\n", port->frontend->name);
3320			if (retval != 0)
3321				break;
3322
3323			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3324					     port->port_type);
3325			if (retval != 0)
3326				break;
3327
3328			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3329			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3330			if (retval != 0)
3331				break;
3332
3333			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3334			    port->port_name);
3335			if (retval != 0)
3336				break;
3337
3338			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3339			    port->physical_port);
3340			if (retval != 0)
3341				break;
3342
3343			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3344			    port->virtual_port);
3345			if (retval != 0)
3346				break;
3347
3348			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3349			    (uintmax_t)port->wwnn);
3350			if (retval != 0)
3351				break;
3352
3353			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3354			    (uintmax_t)port->wwpn);
3355			if (retval != 0)
3356				break;
3357
3358			if (port->port_info != NULL) {
3359				retval = port->port_info(port->onoff_arg, sb);
3360				if (retval != 0)
3361					break;
3362			}
3363			STAILQ_FOREACH(opt, &port->options, links) {
3364				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3365				    opt->name, opt->value, opt->name);
3366				if (retval != 0)
3367					break;
3368			}
3369
3370			retval = sbuf_printf(sb, "</targ_port>\n");
3371			if (retval != 0)
3372				break;
3373		}
3374		mtx_unlock(&softc->ctl_lock);
3375
3376		if ((retval != 0)
3377		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3378			retval = 0;
3379			sbuf_delete(sb);
3380			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3381			snprintf(list->error_str, sizeof(list->error_str),
3382				 "Out of space, %d bytes is too small",
3383				 list->alloc_len);
3384			break;
3385		}
3386
3387		sbuf_finish(sb);
3388
3389		retval = copyout(sbuf_data(sb), list->lun_xml,
3390				 sbuf_len(sb) + 1);
3391
3392		list->fill_len = sbuf_len(sb) + 1;
3393		list->status = CTL_LUN_LIST_OK;
3394		sbuf_delete(sb);
3395		break;
3396	}
3397	default: {
3398		/* XXX KDM should we fix this? */
3399#if 0
3400		struct ctl_backend_driver *backend;
3401		unsigned int type;
3402		int found;
3403
3404		found = 0;
3405
3406		/*
3407		 * We encode the backend type as the ioctl type for backend
3408		 * ioctls.  So parse it out here, and then search for a
3409		 * backend of this type.
3410		 */
3411		type = _IOC_TYPE(cmd);
3412
3413		STAILQ_FOREACH(backend, &softc->be_list, links) {
3414			if (backend->type == type) {
3415				found = 1;
3416				break;
3417			}
3418		}
3419		if (found == 0) {
3420			printf("ctl: unknown ioctl command %#lx or backend "
3421			       "%d\n", cmd, type);
3422			retval = EINVAL;
3423			break;
3424		}
3425		retval = backend->ioctl(dev, cmd, addr, flag, td);
3426#endif
3427		retval = ENOTTY;
3428		break;
3429	}
3430	}
3431	return (retval);
3432}
3433
3434uint32_t
3435ctl_get_initindex(struct ctl_nexus *nexus)
3436{
3437	if (nexus->targ_port < CTL_MAX_PORTS)
3438		return (nexus->initid.id +
3439			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3440	else
3441		return (nexus->initid.id +
3442		       ((nexus->targ_port - CTL_MAX_PORTS) *
3443			CTL_MAX_INIT_PER_PORT));
3444}
3445
3446uint32_t
3447ctl_get_resindex(struct ctl_nexus *nexus)
3448{
3449	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3450}
3451
3452uint32_t
3453ctl_port_idx(int port_num)
3454{
3455	if (port_num < CTL_MAX_PORTS)
3456		return(port_num);
3457	else
3458		return(port_num - CTL_MAX_PORTS);
3459}
3460
3461static uint32_t
3462ctl_map_lun(int port_num, uint32_t lun_id)
3463{
3464	struct ctl_port *port;
3465
3466	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3467	if (port == NULL)
3468		return (UINT32_MAX);
3469	if (port->lun_map == NULL)
3470		return (lun_id);
3471	return (port->lun_map(port->targ_lun_arg, lun_id));
3472}
3473
3474static uint32_t
3475ctl_map_lun_back(int port_num, uint32_t lun_id)
3476{
3477	struct ctl_port *port;
3478	uint32_t i;
3479
3480	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3481	if (port->lun_map == NULL)
3482		return (lun_id);
3483	for (i = 0; i < CTL_MAX_LUNS; i++) {
3484		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3485			return (i);
3486	}
3487	return (UINT32_MAX);
3488}
3489
3490/*
3491 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3492 * that are a power of 2.
3493 */
3494int
3495ctl_ffz(uint32_t *mask, uint32_t size)
3496{
3497	uint32_t num_chunks, num_pieces;
3498	int i, j;
3499
3500	num_chunks = (size >> 5);
3501	if (num_chunks == 0)
3502		num_chunks++;
3503	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3504
3505	for (i = 0; i < num_chunks; i++) {
3506		for (j = 0; j < num_pieces; j++) {
3507			if ((mask[i] & (1 << j)) == 0)
3508				return ((i << 5) + j);
3509		}
3510	}
3511
3512	return (-1);
3513}
3514
3515int
3516ctl_set_mask(uint32_t *mask, uint32_t bit)
3517{
3518	uint32_t chunk, piece;
3519
3520	chunk = bit >> 5;
3521	piece = bit % (sizeof(uint32_t) * 8);
3522
3523	if ((mask[chunk] & (1 << piece)) != 0)
3524		return (-1);
3525	else
3526		mask[chunk] |= (1 << piece);
3527
3528	return (0);
3529}
3530
3531int
3532ctl_clear_mask(uint32_t *mask, uint32_t bit)
3533{
3534	uint32_t chunk, piece;
3535
3536	chunk = bit >> 5;
3537	piece = bit % (sizeof(uint32_t) * 8);
3538
3539	if ((mask[chunk] & (1 << piece)) == 0)
3540		return (-1);
3541	else
3542		mask[chunk] &= ~(1 << piece);
3543
3544	return (0);
3545}
3546
3547int
3548ctl_is_set(uint32_t *mask, uint32_t bit)
3549{
3550	uint32_t chunk, piece;
3551
3552	chunk = bit >> 5;
3553	piece = bit % (sizeof(uint32_t) * 8);
3554
3555	if ((mask[chunk] & (1 << piece)) == 0)
3556		return (0);
3557	else
3558		return (1);
3559}
3560
3561#ifdef unused
3562/*
3563 * The bus, target and lun are optional, they can be filled in later.
3564 * can_wait is used to determine whether we can wait on the malloc or not.
3565 */
3566union ctl_io*
3567ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3568	      uint32_t targ_lun, int can_wait)
3569{
3570	union ctl_io *io;
3571
3572	if (can_wait)
3573		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3574	else
3575		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3576
3577	if (io != NULL) {
3578		io->io_hdr.io_type = io_type;
3579		io->io_hdr.targ_port = targ_port;
3580		/*
3581		 * XXX KDM this needs to change/go away.  We need to move
3582		 * to a preallocated pool of ctl_scsiio structures.
3583		 */
3584		io->io_hdr.nexus.targ_target.id = targ_target;
3585		io->io_hdr.nexus.targ_lun = targ_lun;
3586	}
3587
3588	return (io);
3589}
3590
3591void
3592ctl_kfree_io(union ctl_io *io)
3593{
3594	free(io, M_CTL);
3595}
3596#endif /* unused */
3597
3598/*
3599 * ctl_softc, pool_type, total_ctl_io are passed in.
3600 * npool is passed out.
3601 */
3602int
3603ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3604		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3605{
3606	uint32_t i;
3607	union ctl_io *cur_io, *next_io;
3608	struct ctl_io_pool *pool;
3609	int retval;
3610
3611	retval = 0;
3612
3613	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3614					    M_NOWAIT | M_ZERO);
3615	if (pool == NULL) {
3616		retval = ENOMEM;
3617		goto bailout;
3618	}
3619
3620	pool->type = pool_type;
3621	pool->ctl_softc = ctl_softc;
3622
3623	mtx_lock(&ctl_softc->pool_lock);
3624	pool->id = ctl_softc->cur_pool_id++;
3625	mtx_unlock(&ctl_softc->pool_lock);
3626
3627	pool->flags = CTL_POOL_FLAG_NONE;
3628	pool->refcount = 1;		/* Reference for validity. */
3629	STAILQ_INIT(&pool->free_queue);
3630
3631	/*
3632	 * XXX KDM other options here:
3633	 * - allocate a page at a time
3634	 * - allocate one big chunk of memory.
3635	 * Page allocation might work well, but would take a little more
3636	 * tracking.
3637	 */
3638	for (i = 0; i < total_ctl_io; i++) {
3639		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3640						M_NOWAIT);
3641		if (cur_io == NULL) {
3642			retval = ENOMEM;
3643			break;
3644		}
3645		cur_io->io_hdr.pool = pool;
3646		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3647		pool->total_ctl_io++;
3648		pool->free_ctl_io++;
3649	}
3650
3651	if (retval != 0) {
3652		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3653		     cur_io != NULL; cur_io = next_io) {
3654			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3655							      links);
3656			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3657				      ctl_io_hdr, links);
3658			free(cur_io, M_CTLIO);
3659		}
3660
3661		free(pool, M_CTL);
3662		goto bailout;
3663	}
3664	mtx_lock(&ctl_softc->pool_lock);
3665	ctl_softc->num_pools++;
3666	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3667	/*
3668	 * Increment our usage count if this is an external consumer, so we
3669	 * can't get unloaded until the external consumer (most likely a
3670	 * FETD) unloads and frees his pool.
3671	 *
3672	 * XXX KDM will this increment the caller's module use count, or
3673	 * mine?
3674	 */
3675#if 0
3676	if ((pool_type != CTL_POOL_EMERGENCY)
3677	 && (pool_type != CTL_POOL_INTERNAL)
3678	 && (pool_type != CTL_POOL_4OTHERSC))
3679		MOD_INC_USE_COUNT;
3680#endif
3681
3682	mtx_unlock(&ctl_softc->pool_lock);
3683
3684	*npool = pool;
3685
3686bailout:
3687
3688	return (retval);
3689}
3690
3691static int
3692ctl_pool_acquire(struct ctl_io_pool *pool)
3693{
3694
3695	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3696
3697	if (pool->flags & CTL_POOL_FLAG_INVALID)
3698		return (EINVAL);
3699
3700	pool->refcount++;
3701
3702	return (0);
3703}
3704
3705static void
3706ctl_pool_release(struct ctl_io_pool *pool)
3707{
3708	struct ctl_softc *ctl_softc = pool->ctl_softc;
3709	union ctl_io *io;
3710
3711	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3712
3713	if (--pool->refcount != 0)
3714		return;
3715
3716	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3717		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3718			      links);
3719		free(io, M_CTLIO);
3720	}
3721
3722	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3723	ctl_softc->num_pools--;
3724
3725	/*
3726	 * XXX KDM will this decrement the caller's usage count or mine?
3727	 */
3728#if 0
3729	if ((pool->type != CTL_POOL_EMERGENCY)
3730	 && (pool->type != CTL_POOL_INTERNAL)
3731	 && (pool->type != CTL_POOL_4OTHERSC))
3732		MOD_DEC_USE_COUNT;
3733#endif
3734
3735	free(pool, M_CTL);
3736}
3737
3738void
3739ctl_pool_free(struct ctl_io_pool *pool)
3740{
3741	struct ctl_softc *ctl_softc;
3742
3743	if (pool == NULL)
3744		return;
3745
3746	ctl_softc = pool->ctl_softc;
3747	mtx_lock(&ctl_softc->pool_lock);
3748	pool->flags |= CTL_POOL_FLAG_INVALID;
3749	ctl_pool_release(pool);
3750	mtx_unlock(&ctl_softc->pool_lock);
3751}
3752
3753/*
3754 * This routine does not block (except for spinlocks of course).
3755 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3756 * possible.
3757 */
3758union ctl_io *
3759ctl_alloc_io(void *pool_ref)
3760{
3761	union ctl_io *io;
3762	struct ctl_softc *ctl_softc;
3763	struct ctl_io_pool *pool, *npool;
3764	struct ctl_io_pool *emergency_pool;
3765
3766	pool = (struct ctl_io_pool *)pool_ref;
3767
3768	if (pool == NULL) {
3769		printf("%s: pool is NULL\n", __func__);
3770		return (NULL);
3771	}
3772
3773	emergency_pool = NULL;
3774
3775	ctl_softc = pool->ctl_softc;
3776
3777	mtx_lock(&ctl_softc->pool_lock);
3778	/*
3779	 * First, try to get the io structure from the user's pool.
3780	 */
3781	if (ctl_pool_acquire(pool) == 0) {
3782		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3783		if (io != NULL) {
3784			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3785			pool->total_allocated++;
3786			pool->free_ctl_io--;
3787			mtx_unlock(&ctl_softc->pool_lock);
3788			return (io);
3789		} else
3790			ctl_pool_release(pool);
3791	}
3792	/*
3793	 * If he doesn't have any io structures left, search for an
3794	 * emergency pool and grab one from there.
3795	 */
3796	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3797		if (npool->type != CTL_POOL_EMERGENCY)
3798			continue;
3799
3800		if (ctl_pool_acquire(npool) != 0)
3801			continue;
3802
3803		emergency_pool = npool;
3804
3805		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3806		if (io != NULL) {
3807			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3808			npool->total_allocated++;
3809			npool->free_ctl_io--;
3810			mtx_unlock(&ctl_softc->pool_lock);
3811			return (io);
3812		} else
3813			ctl_pool_release(npool);
3814	}
3815
3816	/* Drop the spinlock before we malloc */
3817	mtx_unlock(&ctl_softc->pool_lock);
3818
3819	/*
3820	 * The emergency pool (if it exists) didn't have one, so try an
3821	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3822	 */
3823	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3824	if (io != NULL) {
3825		/*
3826		 * If the emergency pool exists but is empty, add this
3827		 * ctl_io to its list when it gets freed.
3828		 */
3829		if (emergency_pool != NULL) {
3830			mtx_lock(&ctl_softc->pool_lock);
3831			if (ctl_pool_acquire(emergency_pool) == 0) {
3832				io->io_hdr.pool = emergency_pool;
3833				emergency_pool->total_ctl_io++;
3834				/*
3835				 * Need to bump this, otherwise
3836				 * total_allocated and total_freed won't
3837				 * match when we no longer have anything
3838				 * outstanding.
3839				 */
3840				emergency_pool->total_allocated++;
3841			}
3842			mtx_unlock(&ctl_softc->pool_lock);
3843		} else
3844			io->io_hdr.pool = NULL;
3845	}
3846
3847	return (io);
3848}
3849
3850void
3851ctl_free_io(union ctl_io *io)
3852{
3853	if (io == NULL)
3854		return;
3855
3856	/*
3857	 * If this ctl_io has a pool, return it to that pool.
3858	 */
3859	if (io->io_hdr.pool != NULL) {
3860		struct ctl_io_pool *pool;
3861
3862		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3863		mtx_lock(&pool->ctl_softc->pool_lock);
3864		io->io_hdr.io_type = 0xff;
3865		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3866		pool->total_freed++;
3867		pool->free_ctl_io++;
3868		ctl_pool_release(pool);
3869		mtx_unlock(&pool->ctl_softc->pool_lock);
3870	} else {
3871		/*
3872		 * Otherwise, just free it.  We probably malloced it and
3873		 * the emergency pool wasn't available.
3874		 */
3875		free(io, M_CTLIO);
3876	}
3877
3878}
3879
3880void
3881ctl_zero_io(union ctl_io *io)
3882{
3883	void *pool_ref;
3884
3885	if (io == NULL)
3886		return;
3887
3888	/*
3889	 * May need to preserve linked list pointers at some point too.
3890	 */
3891	pool_ref = io->io_hdr.pool;
3892
3893	memset(io, 0, sizeof(*io));
3894
3895	io->io_hdr.pool = pool_ref;
3896}
3897
3898/*
3899 * This routine is currently used for internal copies of ctl_ios that need
3900 * to persist for some reason after we've already returned status to the
3901 * FETD.  (Thus the flag set.)
3902 *
3903 * XXX XXX
3904 * Note that this makes a blind copy of all fields in the ctl_io, except
3905 * for the pool reference.  This includes any memory that has been
3906 * allocated!  That memory will no longer be valid after done has been
3907 * called, so this would be VERY DANGEROUS for command that actually does
3908 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3909 * start and stop commands, which don't transfer any data, so this is not a
3910 * problem.  If it is used for anything else, the caller would also need to
3911 * allocate data buffer space and this routine would need to be modified to
3912 * copy the data buffer(s) as well.
3913 */
3914void
3915ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3916{
3917	void *pool_ref;
3918
3919	if ((src == NULL)
3920	 || (dest == NULL))
3921		return;
3922
3923	/*
3924	 * May need to preserve linked list pointers at some point too.
3925	 */
3926	pool_ref = dest->io_hdr.pool;
3927
3928	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3929
3930	dest->io_hdr.pool = pool_ref;
3931	/*
3932	 * We need to know that this is an internal copy, and doesn't need
3933	 * to get passed back to the FETD that allocated it.
3934	 */
3935	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3936}
3937
3938#ifdef NEEDTOPORT
3939static void
3940ctl_update_power_subpage(struct copan_power_subpage *page)
3941{
3942	int num_luns, num_partitions, config_type;
3943	struct ctl_softc *softc;
3944	cs_BOOL_t aor_present, shelf_50pct_power;
3945	cs_raidset_personality_t rs_type;
3946	int max_active_luns;
3947
3948	softc = control_softc;
3949
3950	/* subtract out the processor LUN */
3951	num_luns = softc->num_luns - 1;
3952	/*
3953	 * Default to 7 LUNs active, which was the only number we allowed
3954	 * in the past.
3955	 */
3956	max_active_luns = 7;
3957
3958	num_partitions = config_GetRsPartitionInfo();
3959	config_type = config_GetConfigType();
3960	shelf_50pct_power = config_GetShelfPowerMode();
3961	aor_present = config_IsAorRsPresent();
3962
3963	rs_type = ddb_GetRsRaidType(1);
3964	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3965	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3966		EPRINT(0, "Unsupported RS type %d!", rs_type);
3967	}
3968
3969
3970	page->total_luns = num_luns;
3971
3972	switch (config_type) {
3973	case 40:
3974		/*
3975		 * In a 40 drive configuration, it doesn't matter what DC
3976		 * cards we have, whether we have AOR enabled or not,
3977		 * partitioning or not, or what type of RAIDset we have.
3978		 * In that scenario, we can power up every LUN we present
3979		 * to the user.
3980		 */
3981		max_active_luns = num_luns;
3982
3983		break;
3984	case 64:
3985		if (shelf_50pct_power == CS_FALSE) {
3986			/* 25% power */
3987			if (aor_present == CS_TRUE) {
3988				if (rs_type ==
3989				     CS_RAIDSET_PERSONALITY_RAID5) {
3990					max_active_luns = 7;
3991				} else if (rs_type ==
3992					 CS_RAIDSET_PERSONALITY_RAID1){
3993					max_active_luns = 14;
3994				} else {
3995					/* XXX KDM now what?? */
3996				}
3997			} else {
3998				if (rs_type ==
3999				     CS_RAIDSET_PERSONALITY_RAID5) {
4000					max_active_luns = 8;
4001				} else if (rs_type ==
4002					 CS_RAIDSET_PERSONALITY_RAID1){
4003					max_active_luns = 16;
4004				} else {
4005					/* XXX KDM now what?? */
4006				}
4007			}
4008		} else {
4009			/* 50% power */
4010			/*
4011			 * With 50% power in a 64 drive configuration, we
4012			 * can power all LUNs we present.
4013			 */
4014			max_active_luns = num_luns;
4015		}
4016		break;
4017	case 112:
4018		if (shelf_50pct_power == CS_FALSE) {
4019			/* 25% power */
4020			if (aor_present == CS_TRUE) {
4021				if (rs_type ==
4022				     CS_RAIDSET_PERSONALITY_RAID5) {
4023					max_active_luns = 7;
4024				} else if (rs_type ==
4025					 CS_RAIDSET_PERSONALITY_RAID1){
4026					max_active_luns = 14;
4027				} else {
4028					/* XXX KDM now what?? */
4029				}
4030			} else {
4031				if (rs_type ==
4032				     CS_RAIDSET_PERSONALITY_RAID5) {
4033					max_active_luns = 8;
4034				} else if (rs_type ==
4035					 CS_RAIDSET_PERSONALITY_RAID1){
4036					max_active_luns = 16;
4037				} else {
4038					/* XXX KDM now what?? */
4039				}
4040			}
4041		} else {
4042			/* 50% power */
4043			if (aor_present == CS_TRUE) {
4044				if (rs_type ==
4045				     CS_RAIDSET_PERSONALITY_RAID5) {
4046					max_active_luns = 14;
4047				} else if (rs_type ==
4048					 CS_RAIDSET_PERSONALITY_RAID1){
4049					/*
4050					 * We're assuming here that disk
4051					 * caching is enabled, and so we're
4052					 * able to power up half of each
4053					 * LUN, and cache all writes.
4054					 */
4055					max_active_luns = num_luns;
4056				} else {
4057					/* XXX KDM now what?? */
4058				}
4059			} else {
4060				if (rs_type ==
4061				     CS_RAIDSET_PERSONALITY_RAID5) {
4062					max_active_luns = 15;
4063				} else if (rs_type ==
4064					 CS_RAIDSET_PERSONALITY_RAID1){
4065					max_active_luns = 30;
4066				} else {
4067					/* XXX KDM now what?? */
4068				}
4069			}
4070		}
4071		break;
4072	default:
4073		/*
4074		 * In this case, we have an unknown configuration, so we
4075		 * just use the default from above.
4076		 */
4077		break;
4078	}
4079
4080	page->max_active_luns = max_active_luns;
4081#if 0
4082	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4083	       page->total_luns, page->max_active_luns);
4084#endif
4085}
4086#endif /* NEEDTOPORT */
4087
4088/*
4089 * This routine could be used in the future to load default and/or saved
4090 * mode page parameters for a particuar lun.
4091 */
4092static int
4093ctl_init_page_index(struct ctl_lun *lun)
4094{
4095	int i;
4096	struct ctl_page_index *page_index;
4097	struct ctl_softc *softc;
4098	const char *value;
4099
4100	memcpy(&lun->mode_pages.index, page_index_template,
4101	       sizeof(page_index_template));
4102
4103	softc = lun->ctl_softc;
4104
4105	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4106
4107		page_index = &lun->mode_pages.index[i];
4108		/*
4109		 * If this is a disk-only mode page, there's no point in
4110		 * setting it up.  For some pages, we have to have some
4111		 * basic information about the disk in order to calculate the
4112		 * mode page data.
4113		 */
4114		if ((lun->be_lun->lun_type != T_DIRECT)
4115		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4116			continue;
4117
4118		switch (page_index->page_code & SMPH_PC_MASK) {
4119		case SMS_FORMAT_DEVICE_PAGE: {
4120			struct scsi_format_page *format_page;
4121
4122			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4123				panic("subpage is incorrect!");
4124
4125			/*
4126			 * Sectors per track are set above.  Bytes per
4127			 * sector need to be set here on a per-LUN basis.
4128			 */
4129			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4130			       &format_page_default,
4131			       sizeof(format_page_default));
4132			memcpy(&lun->mode_pages.format_page[
4133			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4134			       sizeof(format_page_changeable));
4135			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4136			       &format_page_default,
4137			       sizeof(format_page_default));
4138			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4139			       &format_page_default,
4140			       sizeof(format_page_default));
4141
4142			format_page = &lun->mode_pages.format_page[
4143				CTL_PAGE_CURRENT];
4144			scsi_ulto2b(lun->be_lun->blocksize,
4145				    format_page->bytes_per_sector);
4146
4147			format_page = &lun->mode_pages.format_page[
4148				CTL_PAGE_DEFAULT];
4149			scsi_ulto2b(lun->be_lun->blocksize,
4150				    format_page->bytes_per_sector);
4151
4152			format_page = &lun->mode_pages.format_page[
4153				CTL_PAGE_SAVED];
4154			scsi_ulto2b(lun->be_lun->blocksize,
4155				    format_page->bytes_per_sector);
4156
4157			page_index->page_data =
4158				(uint8_t *)lun->mode_pages.format_page;
4159			break;
4160		}
4161		case SMS_RIGID_DISK_PAGE: {
4162			struct scsi_rigid_disk_page *rigid_disk_page;
4163			uint32_t sectors_per_cylinder;
4164			uint64_t cylinders;
4165#ifndef	__XSCALE__
4166			int shift;
4167#endif /* !__XSCALE__ */
4168
4169			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4170				panic("invalid subpage value %d",
4171				      page_index->subpage);
4172
4173			/*
4174			 * Rotation rate and sectors per track are set
4175			 * above.  We calculate the cylinders here based on
4176			 * capacity.  Due to the number of heads and
4177			 * sectors per track we're using, smaller arrays
4178			 * may turn out to have 0 cylinders.  Linux and
4179			 * FreeBSD don't pay attention to these mode pages
4180			 * to figure out capacity, but Solaris does.  It
4181			 * seems to deal with 0 cylinders just fine, and
4182			 * works out a fake geometry based on the capacity.
4183			 */
4184			memcpy(&lun->mode_pages.rigid_disk_page[
4185			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4186			       sizeof(rigid_disk_page_default));
4187			memcpy(&lun->mode_pages.rigid_disk_page[
4188			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4189			       sizeof(rigid_disk_page_changeable));
4190			memcpy(&lun->mode_pages.rigid_disk_page[
4191			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4192			       sizeof(rigid_disk_page_default));
4193			memcpy(&lun->mode_pages.rigid_disk_page[
4194			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4195			       sizeof(rigid_disk_page_default));
4196
4197			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4198				CTL_DEFAULT_HEADS;
4199
4200			/*
4201			 * The divide method here will be more accurate,
4202			 * probably, but results in floating point being
4203			 * used in the kernel on i386 (__udivdi3()).  On the
4204			 * XScale, though, __udivdi3() is implemented in
4205			 * software.
4206			 *
4207			 * The shift method for cylinder calculation is
4208			 * accurate if sectors_per_cylinder is a power of
4209			 * 2.  Otherwise it might be slightly off -- you
4210			 * might have a bit of a truncation problem.
4211			 */
4212#ifdef	__XSCALE__
4213			cylinders = (lun->be_lun->maxlba + 1) /
4214				sectors_per_cylinder;
4215#else
4216			for (shift = 31; shift > 0; shift--) {
4217				if (sectors_per_cylinder & (1 << shift))
4218					break;
4219			}
4220			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4221#endif
4222
4223			/*
4224			 * We've basically got 3 bytes, or 24 bits for the
4225			 * cylinder size in the mode page.  If we're over,
4226			 * just round down to 2^24.
4227			 */
4228			if (cylinders > 0xffffff)
4229				cylinders = 0xffffff;
4230
4231			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4232				CTL_PAGE_CURRENT];
4233			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4234
4235			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4236				CTL_PAGE_DEFAULT];
4237			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4238
4239			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4240				CTL_PAGE_SAVED];
4241			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4242
4243			page_index->page_data =
4244				(uint8_t *)lun->mode_pages.rigid_disk_page;
4245			break;
4246		}
4247		case SMS_CACHING_PAGE: {
4248			struct scsi_caching_page *caching_page;
4249
4250			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4251				panic("invalid subpage value %d",
4252				      page_index->subpage);
4253			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4254			       &caching_page_default,
4255			       sizeof(caching_page_default));
4256			memcpy(&lun->mode_pages.caching_page[
4257			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4258			       sizeof(caching_page_changeable));
4259			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4260			       &caching_page_default,
4261			       sizeof(caching_page_default));
4262			caching_page = &lun->mode_pages.caching_page[
4263			    CTL_PAGE_SAVED];
4264			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4265			if (value != NULL && strcmp(value, "off") == 0)
4266				caching_page->flags1 &= ~SCP_WCE;
4267			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4268			if (value != NULL && strcmp(value, "off") == 0)
4269				caching_page->flags1 |= SCP_RCD;
4270			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4271			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4272			       sizeof(caching_page_default));
4273			page_index->page_data =
4274				(uint8_t *)lun->mode_pages.caching_page;
4275			break;
4276		}
4277		case SMS_CONTROL_MODE_PAGE: {
4278			struct scsi_control_page *control_page;
4279
4280			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4281				panic("invalid subpage value %d",
4282				      page_index->subpage);
4283
4284			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4285			       &control_page_default,
4286			       sizeof(control_page_default));
4287			memcpy(&lun->mode_pages.control_page[
4288			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4289			       sizeof(control_page_changeable));
4290			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4291			       &control_page_default,
4292			       sizeof(control_page_default));
4293			control_page = &lun->mode_pages.control_page[
4294			    CTL_PAGE_SAVED];
4295			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4296			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4297				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4298				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4299			}
4300			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4301			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4302			       sizeof(control_page_default));
4303			page_index->page_data =
4304				(uint8_t *)lun->mode_pages.control_page;
4305			break;
4306
4307		}
4308		case SMS_VENDOR_SPECIFIC_PAGE:{
4309			switch (page_index->subpage) {
4310			case PWR_SUBPAGE_CODE: {
4311				struct copan_power_subpage *current_page,
4312							   *saved_page;
4313
4314				memcpy(&lun->mode_pages.power_subpage[
4315				       CTL_PAGE_CURRENT],
4316				       &power_page_default,
4317				       sizeof(power_page_default));
4318				memcpy(&lun->mode_pages.power_subpage[
4319				       CTL_PAGE_CHANGEABLE],
4320				       &power_page_changeable,
4321				       sizeof(power_page_changeable));
4322				memcpy(&lun->mode_pages.power_subpage[
4323				       CTL_PAGE_DEFAULT],
4324				       &power_page_default,
4325				       sizeof(power_page_default));
4326				memcpy(&lun->mode_pages.power_subpage[
4327				       CTL_PAGE_SAVED],
4328				       &power_page_default,
4329				       sizeof(power_page_default));
4330				page_index->page_data =
4331				    (uint8_t *)lun->mode_pages.power_subpage;
4332
4333				current_page = (struct copan_power_subpage *)
4334					(page_index->page_data +
4335					 (page_index->page_len *
4336					  CTL_PAGE_CURRENT));
4337			        saved_page = (struct copan_power_subpage *)
4338				        (page_index->page_data +
4339					 (page_index->page_len *
4340					  CTL_PAGE_SAVED));
4341				break;
4342			}
4343			case APS_SUBPAGE_CODE: {
4344				struct copan_aps_subpage *current_page,
4345							 *saved_page;
4346
4347				// This gets set multiple times but
4348				// it should always be the same. It's
4349				// only done during init so who cares.
4350				index_to_aps_page = i;
4351
4352				memcpy(&lun->mode_pages.aps_subpage[
4353				       CTL_PAGE_CURRENT],
4354				       &aps_page_default,
4355				       sizeof(aps_page_default));
4356				memcpy(&lun->mode_pages.aps_subpage[
4357				       CTL_PAGE_CHANGEABLE],
4358				       &aps_page_changeable,
4359				       sizeof(aps_page_changeable));
4360				memcpy(&lun->mode_pages.aps_subpage[
4361				       CTL_PAGE_DEFAULT],
4362				       &aps_page_default,
4363				       sizeof(aps_page_default));
4364				memcpy(&lun->mode_pages.aps_subpage[
4365				       CTL_PAGE_SAVED],
4366				       &aps_page_default,
4367				       sizeof(aps_page_default));
4368				page_index->page_data =
4369					(uint8_t *)lun->mode_pages.aps_subpage;
4370
4371				current_page = (struct copan_aps_subpage *)
4372					(page_index->page_data +
4373					 (page_index->page_len *
4374					  CTL_PAGE_CURRENT));
4375				saved_page = (struct copan_aps_subpage *)
4376					(page_index->page_data +
4377					 (page_index->page_len *
4378					  CTL_PAGE_SAVED));
4379				break;
4380			}
4381			case DBGCNF_SUBPAGE_CODE: {
4382				struct copan_debugconf_subpage *current_page,
4383							       *saved_page;
4384
4385				memcpy(&lun->mode_pages.debugconf_subpage[
4386				       CTL_PAGE_CURRENT],
4387				       &debugconf_page_default,
4388				       sizeof(debugconf_page_default));
4389				memcpy(&lun->mode_pages.debugconf_subpage[
4390				       CTL_PAGE_CHANGEABLE],
4391				       &debugconf_page_changeable,
4392				       sizeof(debugconf_page_changeable));
4393				memcpy(&lun->mode_pages.debugconf_subpage[
4394				       CTL_PAGE_DEFAULT],
4395				       &debugconf_page_default,
4396				       sizeof(debugconf_page_default));
4397				memcpy(&lun->mode_pages.debugconf_subpage[
4398				       CTL_PAGE_SAVED],
4399				       &debugconf_page_default,
4400				       sizeof(debugconf_page_default));
4401				page_index->page_data =
4402					(uint8_t *)lun->mode_pages.debugconf_subpage;
4403
4404				current_page = (struct copan_debugconf_subpage *)
4405					(page_index->page_data +
4406					 (page_index->page_len *
4407					  CTL_PAGE_CURRENT));
4408				saved_page = (struct copan_debugconf_subpage *)
4409					(page_index->page_data +
4410					 (page_index->page_len *
4411					  CTL_PAGE_SAVED));
4412				break;
4413			}
4414			default:
4415				panic("invalid subpage value %d",
4416				      page_index->subpage);
4417				break;
4418			}
4419   			break;
4420		}
4421		default:
4422			panic("invalid page value %d",
4423			      page_index->page_code & SMPH_PC_MASK);
4424			break;
4425    	}
4426	}
4427
4428	return (CTL_RETVAL_COMPLETE);
4429}
4430
4431/*
4432 * LUN allocation.
4433 *
4434 * Requirements:
4435 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4436 *   wants us to allocate the LUN and he can block.
4437 * - ctl_softc is always set
4438 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4439 *
4440 * Returns 0 for success, non-zero (errno) for failure.
4441 */
4442static int
4443ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4444	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4445{
4446	struct ctl_lun *nlun, *lun;
4447	struct ctl_port *port;
4448	struct scsi_vpd_id_descriptor *desc;
4449	struct scsi_vpd_id_t10 *t10id;
4450	const char *eui, *naa, *scsiname, *vendor, *value;
4451	int lun_number, i, lun_malloced;
4452	int devidlen, idlen1, idlen2 = 0, len;
4453
4454	if (be_lun == NULL)
4455		return (EINVAL);
4456
4457	/*
4458	 * We currently only support Direct Access or Processor LUN types.
4459	 */
4460	switch (be_lun->lun_type) {
4461	case T_DIRECT:
4462		break;
4463	case T_PROCESSOR:
4464		break;
4465	case T_SEQUENTIAL:
4466	case T_CHANGER:
4467	default:
4468		be_lun->lun_config_status(be_lun->be_lun,
4469					  CTL_LUN_CONFIG_FAILURE);
4470		break;
4471	}
4472	if (ctl_lun == NULL) {
4473		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4474		lun_malloced = 1;
4475	} else {
4476		lun_malloced = 0;
4477		lun = ctl_lun;
4478	}
4479
4480	memset(lun, 0, sizeof(*lun));
4481	if (lun_malloced)
4482		lun->flags = CTL_LUN_MALLOCED;
4483
4484	/* Generate LUN ID. */
4485	devidlen = max(CTL_DEVID_MIN_LEN,
4486	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4487	idlen1 = sizeof(*t10id) + devidlen;
4488	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4489	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4490	if (scsiname != NULL) {
4491		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4492		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4493	}
4494	eui = ctl_get_opt(&be_lun->options, "eui");
4495	if (eui != NULL) {
4496		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4497	}
4498	naa = ctl_get_opt(&be_lun->options, "naa");
4499	if (naa != NULL) {
4500		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4501	}
4502	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4503	    M_CTL, M_WAITOK | M_ZERO);
4504	lun->lun_devid->len = len;
4505	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4506	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4507	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4508	desc->length = idlen1;
4509	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4510	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4511	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4512		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4513	} else {
4514		strncpy(t10id->vendor, vendor,
4515		    min(sizeof(t10id->vendor), strlen(vendor)));
4516	}
4517	strncpy((char *)t10id->vendor_spec_id,
4518	    (char *)be_lun->device_id, devidlen);
4519	if (scsiname != NULL) {
4520		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4521		    desc->length);
4522		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4523		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4524		    SVPD_ID_TYPE_SCSI_NAME;
4525		desc->length = idlen2;
4526		strlcpy(desc->identifier, scsiname, idlen2);
4527	}
4528	if (eui != NULL) {
4529		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4530		    desc->length);
4531		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4532		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4533		    SVPD_ID_TYPE_EUI64;
4534		desc->length = 8;
4535		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4536	}
4537	if (naa != NULL) {
4538		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4539		    desc->length);
4540		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4541		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4542		    SVPD_ID_TYPE_NAA;
4543		desc->length = 8;
4544		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4545	}
4546
4547	mtx_lock(&ctl_softc->ctl_lock);
4548	/*
4549	 * See if the caller requested a particular LUN number.  If so, see
4550	 * if it is available.  Otherwise, allocate the first available LUN.
4551	 */
4552	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4553		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4554		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4555			mtx_unlock(&ctl_softc->ctl_lock);
4556			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4557				printf("ctl: requested LUN ID %d is higher "
4558				       "than CTL_MAX_LUNS - 1 (%d)\n",
4559				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4560			} else {
4561				/*
4562				 * XXX KDM return an error, or just assign
4563				 * another LUN ID in this case??
4564				 */
4565				printf("ctl: requested LUN ID %d is already "
4566				       "in use\n", be_lun->req_lun_id);
4567			}
4568			if (lun->flags & CTL_LUN_MALLOCED)
4569				free(lun, M_CTL);
4570			be_lun->lun_config_status(be_lun->be_lun,
4571						  CTL_LUN_CONFIG_FAILURE);
4572			return (ENOSPC);
4573		}
4574		lun_number = be_lun->req_lun_id;
4575	} else {
4576		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4577		if (lun_number == -1) {
4578			mtx_unlock(&ctl_softc->ctl_lock);
4579			printf("ctl: can't allocate LUN on target %ju, out of "
4580			       "LUNs\n", (uintmax_t)target_id.id);
4581			if (lun->flags & CTL_LUN_MALLOCED)
4582				free(lun, M_CTL);
4583			be_lun->lun_config_status(be_lun->be_lun,
4584						  CTL_LUN_CONFIG_FAILURE);
4585			return (ENOSPC);
4586		}
4587	}
4588	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4589
4590	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4591	lun->target = target_id;
4592	lun->lun = lun_number;
4593	lun->be_lun = be_lun;
4594	/*
4595	 * The processor LUN is always enabled.  Disk LUNs come on line
4596	 * disabled, and must be enabled by the backend.
4597	 */
4598	lun->flags |= CTL_LUN_DISABLED;
4599	lun->backend = be_lun->be;
4600	be_lun->ctl_lun = lun;
4601	be_lun->lun_id = lun_number;
4602	atomic_add_int(&be_lun->be->num_luns, 1);
4603	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4604		lun->flags |= CTL_LUN_STOPPED;
4605
4606	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4607		lun->flags |= CTL_LUN_INOPERABLE;
4608
4609	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4610		lun->flags |= CTL_LUN_PRIMARY_SC;
4611
4612	value = ctl_get_opt(&be_lun->options, "readonly");
4613	if (value != NULL && strcmp(value, "on") == 0)
4614		lun->flags |= CTL_LUN_READONLY;
4615
4616	lun->ctl_softc = ctl_softc;
4617	TAILQ_INIT(&lun->ooa_queue);
4618	TAILQ_INIT(&lun->blocked_queue);
4619	STAILQ_INIT(&lun->error_list);
4620	ctl_tpc_lun_init(lun);
4621
4622	/*
4623	 * Initialize the mode page index.
4624	 */
4625	ctl_init_page_index(lun);
4626
4627	/*
4628	 * Set the poweron UA for all initiators on this LUN only.
4629	 */
4630	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4631		lun->pending_ua[i] = CTL_UA_POWERON;
4632
4633	/*
4634	 * Now, before we insert this lun on the lun list, set the lun
4635	 * inventory changed UA for all other luns.
4636	 */
4637	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4638		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4639			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4640		}
4641	}
4642
4643	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4644
4645	ctl_softc->ctl_luns[lun_number] = lun;
4646
4647	ctl_softc->num_luns++;
4648
4649	/* Setup statistics gathering */
4650	lun->stats.device_type = be_lun->lun_type;
4651	lun->stats.lun_number = lun_number;
4652	if (lun->stats.device_type == T_DIRECT)
4653		lun->stats.blocksize = be_lun->blocksize;
4654	else
4655		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4656	for (i = 0;i < CTL_MAX_PORTS;i++)
4657		lun->stats.ports[i].targ_port = i;
4658
4659	mtx_unlock(&ctl_softc->ctl_lock);
4660
4661	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4662
4663	/*
4664	 * Run through each registered FETD and bring it online if it isn't
4665	 * already.  Enable the target ID if it hasn't been enabled, and
4666	 * enable this particular LUN.
4667	 */
4668	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4669		int retval;
4670
4671		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4672		if (retval != 0) {
4673			printf("ctl_alloc_lun: FETD %s port %d returned error "
4674			       "%d for lun_enable on target %ju lun %d\n",
4675			       port->port_name, port->targ_port, retval,
4676			       (uintmax_t)target_id.id, lun_number);
4677		} else
4678			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4679	}
4680	return (0);
4681}
4682
4683/*
4684 * Delete a LUN.
4685 * Assumptions:
4686 * - LUN has already been marked invalid and any pending I/O has been taken
4687 *   care of.
4688 */
4689static int
4690ctl_free_lun(struct ctl_lun *lun)
4691{
4692	struct ctl_softc *softc;
4693#if 0
4694	struct ctl_port *port;
4695#endif
4696	struct ctl_lun *nlun;
4697	int i;
4698
4699	softc = lun->ctl_softc;
4700
4701	mtx_assert(&softc->ctl_lock, MA_OWNED);
4702
4703	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4704
4705	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4706
4707	softc->ctl_luns[lun->lun] = NULL;
4708
4709	if (!TAILQ_EMPTY(&lun->ooa_queue))
4710		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4711
4712	softc->num_luns--;
4713
4714	/*
4715	 * XXX KDM this scheme only works for a single target/multiple LUN
4716	 * setup.  It needs to be revamped for a multiple target scheme.
4717	 *
4718	 * XXX KDM this results in port->lun_disable() getting called twice,
4719	 * once when ctl_disable_lun() is called, and a second time here.
4720	 * We really need to re-think the LUN disable semantics.  There
4721	 * should probably be several steps/levels to LUN removal:
4722	 *  - disable
4723	 *  - invalidate
4724	 *  - free
4725 	 *
4726	 * Right now we only have a disable method when communicating to
4727	 * the front end ports, at least for individual LUNs.
4728	 */
4729#if 0
4730	STAILQ_FOREACH(port, &softc->port_list, links) {
4731		int retval;
4732
4733		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4734					 lun->lun);
4735		if (retval != 0) {
4736			printf("ctl_free_lun: FETD %s port %d returned error "
4737			       "%d for lun_disable on target %ju lun %jd\n",
4738			       port->port_name, port->targ_port, retval,
4739			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4740		}
4741
4742		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4743			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4744
4745			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4746			if (retval != 0) {
4747				printf("ctl_free_lun: FETD %s port %d "
4748				       "returned error %d for targ_disable on "
4749				       "target %ju\n", port->port_name,
4750				       port->targ_port, retval,
4751				       (uintmax_t)lun->target.id);
4752			} else
4753				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4754
4755			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4756				continue;
4757
4758#if 0
4759			port->port_offline(port->onoff_arg);
4760			port->status &= ~CTL_PORT_STATUS_ONLINE;
4761#endif
4762		}
4763	}
4764#endif
4765
4766	/*
4767	 * Tell the backend to free resources, if this LUN has a backend.
4768	 */
4769	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4770	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4771
4772	ctl_tpc_lun_shutdown(lun);
4773	mtx_destroy(&lun->lun_lock);
4774	free(lun->lun_devid, M_CTL);
4775	if (lun->flags & CTL_LUN_MALLOCED)
4776		free(lun, M_CTL);
4777
4778	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4779		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4780			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4781		}
4782	}
4783
4784	return (0);
4785}
4786
4787static void
4788ctl_create_lun(struct ctl_be_lun *be_lun)
4789{
4790	struct ctl_softc *ctl_softc;
4791
4792	ctl_softc = control_softc;
4793
4794	/*
4795	 * ctl_alloc_lun() should handle all potential failure cases.
4796	 */
4797	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4798}
4799
4800int
4801ctl_add_lun(struct ctl_be_lun *be_lun)
4802{
4803	struct ctl_softc *ctl_softc = control_softc;
4804
4805	mtx_lock(&ctl_softc->ctl_lock);
4806	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4807	mtx_unlock(&ctl_softc->ctl_lock);
4808	wakeup(&ctl_softc->pending_lun_queue);
4809
4810	return (0);
4811}
4812
4813int
4814ctl_enable_lun(struct ctl_be_lun *be_lun)
4815{
4816	struct ctl_softc *ctl_softc;
4817	struct ctl_port *port, *nport;
4818	struct ctl_lun *lun;
4819	int retval;
4820
4821	ctl_softc = control_softc;
4822
4823	lun = (struct ctl_lun *)be_lun->ctl_lun;
4824
4825	mtx_lock(&ctl_softc->ctl_lock);
4826	mtx_lock(&lun->lun_lock);
4827	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4828		/*
4829		 * eh?  Why did we get called if the LUN is already
4830		 * enabled?
4831		 */
4832		mtx_unlock(&lun->lun_lock);
4833		mtx_unlock(&ctl_softc->ctl_lock);
4834		return (0);
4835	}
4836	lun->flags &= ~CTL_LUN_DISABLED;
4837	mtx_unlock(&lun->lun_lock);
4838
4839	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4840		nport = STAILQ_NEXT(port, links);
4841
4842		/*
4843		 * Drop the lock while we call the FETD's enable routine.
4844		 * This can lead to a callback into CTL (at least in the
4845		 * case of the internal initiator frontend.
4846		 */
4847		mtx_unlock(&ctl_softc->ctl_lock);
4848		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4849		mtx_lock(&ctl_softc->ctl_lock);
4850		if (retval != 0) {
4851			printf("%s: FETD %s port %d returned error "
4852			       "%d for lun_enable on target %ju lun %jd\n",
4853			       __func__, port->port_name, port->targ_port, retval,
4854			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4855		}
4856#if 0
4857		 else {
4858            /* NOTE:  TODO:  why does lun enable affect port status? */
4859			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4860		}
4861#endif
4862	}
4863
4864	mtx_unlock(&ctl_softc->ctl_lock);
4865
4866	return (0);
4867}
4868
4869int
4870ctl_disable_lun(struct ctl_be_lun *be_lun)
4871{
4872	struct ctl_softc *ctl_softc;
4873	struct ctl_port *port;
4874	struct ctl_lun *lun;
4875	int retval;
4876
4877	ctl_softc = control_softc;
4878
4879	lun = (struct ctl_lun *)be_lun->ctl_lun;
4880
4881	mtx_lock(&ctl_softc->ctl_lock);
4882	mtx_lock(&lun->lun_lock);
4883	if (lun->flags & CTL_LUN_DISABLED) {
4884		mtx_unlock(&lun->lun_lock);
4885		mtx_unlock(&ctl_softc->ctl_lock);
4886		return (0);
4887	}
4888	lun->flags |= CTL_LUN_DISABLED;
4889	mtx_unlock(&lun->lun_lock);
4890
4891	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4892		mtx_unlock(&ctl_softc->ctl_lock);
4893		/*
4894		 * Drop the lock before we call the frontend's disable
4895		 * routine, to avoid lock order reversals.
4896		 *
4897		 * XXX KDM what happens if the frontend list changes while
4898		 * we're traversing it?  It's unlikely, but should be handled.
4899		 */
4900		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4901					 lun->lun);
4902		mtx_lock(&ctl_softc->ctl_lock);
4903		if (retval != 0) {
4904			printf("ctl_alloc_lun: FETD %s port %d returned error "
4905			       "%d for lun_disable on target %ju lun %jd\n",
4906			       port->port_name, port->targ_port, retval,
4907			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4908		}
4909	}
4910
4911	mtx_unlock(&ctl_softc->ctl_lock);
4912
4913	return (0);
4914}
4915
4916int
4917ctl_start_lun(struct ctl_be_lun *be_lun)
4918{
4919	struct ctl_softc *ctl_softc;
4920	struct ctl_lun *lun;
4921
4922	ctl_softc = control_softc;
4923
4924	lun = (struct ctl_lun *)be_lun->ctl_lun;
4925
4926	mtx_lock(&lun->lun_lock);
4927	lun->flags &= ~CTL_LUN_STOPPED;
4928	mtx_unlock(&lun->lun_lock);
4929
4930	return (0);
4931}
4932
4933int
4934ctl_stop_lun(struct ctl_be_lun *be_lun)
4935{
4936	struct ctl_softc *ctl_softc;
4937	struct ctl_lun *lun;
4938
4939	ctl_softc = control_softc;
4940
4941	lun = (struct ctl_lun *)be_lun->ctl_lun;
4942
4943	mtx_lock(&lun->lun_lock);
4944	lun->flags |= CTL_LUN_STOPPED;
4945	mtx_unlock(&lun->lun_lock);
4946
4947	return (0);
4948}
4949
4950int
4951ctl_lun_offline(struct ctl_be_lun *be_lun)
4952{
4953	struct ctl_softc *ctl_softc;
4954	struct ctl_lun *lun;
4955
4956	ctl_softc = control_softc;
4957
4958	lun = (struct ctl_lun *)be_lun->ctl_lun;
4959
4960	mtx_lock(&lun->lun_lock);
4961	lun->flags |= CTL_LUN_OFFLINE;
4962	mtx_unlock(&lun->lun_lock);
4963
4964	return (0);
4965}
4966
4967int
4968ctl_lun_online(struct ctl_be_lun *be_lun)
4969{
4970	struct ctl_softc *ctl_softc;
4971	struct ctl_lun *lun;
4972
4973	ctl_softc = control_softc;
4974
4975	lun = (struct ctl_lun *)be_lun->ctl_lun;
4976
4977	mtx_lock(&lun->lun_lock);
4978	lun->flags &= ~CTL_LUN_OFFLINE;
4979	mtx_unlock(&lun->lun_lock);
4980
4981	return (0);
4982}
4983
4984int
4985ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4986{
4987	struct ctl_softc *ctl_softc;
4988	struct ctl_lun *lun;
4989
4990	ctl_softc = control_softc;
4991
4992	lun = (struct ctl_lun *)be_lun->ctl_lun;
4993
4994	mtx_lock(&lun->lun_lock);
4995
4996	/*
4997	 * The LUN needs to be disabled before it can be marked invalid.
4998	 */
4999	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5000		mtx_unlock(&lun->lun_lock);
5001		return (-1);
5002	}
5003	/*
5004	 * Mark the LUN invalid.
5005	 */
5006	lun->flags |= CTL_LUN_INVALID;
5007
5008	/*
5009	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5010	 * If we have something in the OOA queue, we'll free it when the
5011	 * last I/O completes.
5012	 */
5013	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5014		mtx_unlock(&lun->lun_lock);
5015		mtx_lock(&ctl_softc->ctl_lock);
5016		ctl_free_lun(lun);
5017		mtx_unlock(&ctl_softc->ctl_lock);
5018	} else
5019		mtx_unlock(&lun->lun_lock);
5020
5021	return (0);
5022}
5023
5024int
5025ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5026{
5027	struct ctl_softc *ctl_softc;
5028	struct ctl_lun *lun;
5029
5030	ctl_softc = control_softc;
5031	lun = (struct ctl_lun *)be_lun->ctl_lun;
5032
5033	mtx_lock(&lun->lun_lock);
5034	lun->flags |= CTL_LUN_INOPERABLE;
5035	mtx_unlock(&lun->lun_lock);
5036
5037	return (0);
5038}
5039
5040int
5041ctl_lun_operable(struct ctl_be_lun *be_lun)
5042{
5043	struct ctl_softc *ctl_softc;
5044	struct ctl_lun *lun;
5045
5046	ctl_softc = control_softc;
5047	lun = (struct ctl_lun *)be_lun->ctl_lun;
5048
5049	mtx_lock(&lun->lun_lock);
5050	lun->flags &= ~CTL_LUN_INOPERABLE;
5051	mtx_unlock(&lun->lun_lock);
5052
5053	return (0);
5054}
5055
5056int
5057ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5058		   int lock)
5059{
5060	struct ctl_softc *softc;
5061	struct ctl_lun *lun;
5062	struct copan_aps_subpage *current_sp;
5063	struct ctl_page_index *page_index;
5064	int i;
5065
5066	softc = control_softc;
5067
5068	mtx_lock(&softc->ctl_lock);
5069
5070	lun = (struct ctl_lun *)be_lun->ctl_lun;
5071	mtx_lock(&lun->lun_lock);
5072
5073	page_index = NULL;
5074	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5075		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5076		     APS_PAGE_CODE)
5077			continue;
5078
5079		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5080			continue;
5081		page_index = &lun->mode_pages.index[i];
5082	}
5083
5084	if (page_index == NULL) {
5085		mtx_unlock(&lun->lun_lock);
5086		mtx_unlock(&softc->ctl_lock);
5087		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5088		       (uintmax_t)lun->lun);
5089		return (1);
5090	}
5091#if 0
5092	if ((softc->aps_locked_lun != 0)
5093	 && (softc->aps_locked_lun != lun->lun)) {
5094		printf("%s: attempt to lock LUN %llu when %llu is already "
5095		       "locked\n");
5096		mtx_unlock(&lun->lun_lock);
5097		mtx_unlock(&softc->ctl_lock);
5098		return (1);
5099	}
5100#endif
5101
5102	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5103		(page_index->page_len * CTL_PAGE_CURRENT));
5104
5105	if (lock != 0) {
5106		current_sp->lock_active = APS_LOCK_ACTIVE;
5107		softc->aps_locked_lun = lun->lun;
5108	} else {
5109		current_sp->lock_active = 0;
5110		softc->aps_locked_lun = 0;
5111	}
5112
5113
5114	/*
5115	 * If we're in HA mode, try to send the lock message to the other
5116	 * side.
5117	 */
5118	if (ctl_is_single == 0) {
5119		int isc_retval;
5120		union ctl_ha_msg lock_msg;
5121
5122		lock_msg.hdr.nexus = *nexus;
5123		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5124		if (lock != 0)
5125			lock_msg.aps.lock_flag = 1;
5126		else
5127			lock_msg.aps.lock_flag = 0;
5128		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5129					 sizeof(lock_msg), 0);
5130		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5131			printf("%s: APS (lock=%d) error returned from "
5132			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5133			mtx_unlock(&lun->lun_lock);
5134			mtx_unlock(&softc->ctl_lock);
5135			return (1);
5136		}
5137	}
5138
5139	mtx_unlock(&lun->lun_lock);
5140	mtx_unlock(&softc->ctl_lock);
5141
5142	return (0);
5143}
5144
5145void
5146ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5147{
5148	struct ctl_lun *lun;
5149	struct ctl_softc *softc;
5150	int i;
5151
5152	softc = control_softc;
5153
5154	lun = (struct ctl_lun *)be_lun->ctl_lun;
5155
5156	mtx_lock(&lun->lun_lock);
5157
5158	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5159		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5160
5161	mtx_unlock(&lun->lun_lock);
5162}
5163
5164/*
5165 * Backend "memory move is complete" callback for requests that never
5166 * make it down to say RAIDCore's configuration code.
5167 */
5168int
5169ctl_config_move_done(union ctl_io *io)
5170{
5171	int retval;
5172
5173	retval = CTL_RETVAL_COMPLETE;
5174
5175
5176	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5177	/*
5178	 * XXX KDM this shouldn't happen, but what if it does?
5179	 */
5180	if (io->io_hdr.io_type != CTL_IO_SCSI)
5181		panic("I/O type isn't CTL_IO_SCSI!");
5182
5183	if ((io->io_hdr.port_status == 0)
5184	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5185	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5186		io->io_hdr.status = CTL_SUCCESS;
5187	else if ((io->io_hdr.port_status != 0)
5188	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5189	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5190		/*
5191		 * For hardware error sense keys, the sense key
5192		 * specific value is defined to be a retry count,
5193		 * but we use it to pass back an internal FETD
5194		 * error code.  XXX KDM  Hopefully the FETD is only
5195		 * using 16 bits for an error code, since that's
5196		 * all the space we have in the sks field.
5197		 */
5198		ctl_set_internal_failure(&io->scsiio,
5199					 /*sks_valid*/ 1,
5200					 /*retry_count*/
5201					 io->io_hdr.port_status);
5202		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5203			free(io->scsiio.kern_data_ptr, M_CTL);
5204		ctl_done(io);
5205		goto bailout;
5206	}
5207
5208	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5209	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5210	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5211		/*
5212		 * XXX KDM just assuming a single pointer here, and not a
5213		 * S/G list.  If we start using S/G lists for config data,
5214		 * we'll need to know how to clean them up here as well.
5215		 */
5216		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5217			free(io->scsiio.kern_data_ptr, M_CTL);
5218		/* Hopefully the user has already set the status... */
5219		ctl_done(io);
5220	} else {
5221		/*
5222		 * XXX KDM now we need to continue data movement.  Some
5223		 * options:
5224		 * - call ctl_scsiio() again?  We don't do this for data
5225		 *   writes, because for those at least we know ahead of
5226		 *   time where the write will go and how long it is.  For
5227		 *   config writes, though, that information is largely
5228		 *   contained within the write itself, thus we need to
5229		 *   parse out the data again.
5230		 *
5231		 * - Call some other function once the data is in?
5232		 */
5233
5234		/*
5235		 * XXX KDM call ctl_scsiio() again for now, and check flag
5236		 * bits to see whether we're allocated or not.
5237		 */
5238		retval = ctl_scsiio(&io->scsiio);
5239	}
5240bailout:
5241	return (retval);
5242}
5243
5244/*
5245 * This gets called by a backend driver when it is done with a
5246 * data_submit method.
5247 */
5248void
5249ctl_data_submit_done(union ctl_io *io)
5250{
5251	/*
5252	 * If the IO_CONT flag is set, we need to call the supplied
5253	 * function to continue processing the I/O, instead of completing
5254	 * the I/O just yet.
5255	 *
5256	 * If there is an error, though, we don't want to keep processing.
5257	 * Instead, just send status back to the initiator.
5258	 */
5259	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5260	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5261	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5262	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5263		io->scsiio.io_cont(io);
5264		return;
5265	}
5266	ctl_done(io);
5267}
5268
5269/*
5270 * This gets called by a backend driver when it is done with a
5271 * configuration write.
5272 */
5273void
5274ctl_config_write_done(union ctl_io *io)
5275{
5276	uint8_t *buf;
5277
5278	/*
5279	 * If the IO_CONT flag is set, we need to call the supplied
5280	 * function to continue processing the I/O, instead of completing
5281	 * the I/O just yet.
5282	 *
5283	 * If there is an error, though, we don't want to keep processing.
5284	 * Instead, just send status back to the initiator.
5285	 */
5286	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5287	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5288	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5289	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5290		io->scsiio.io_cont(io);
5291		return;
5292	}
5293	/*
5294	 * Since a configuration write can be done for commands that actually
5295	 * have data allocated, like write buffer, and commands that have
5296	 * no data, like start/stop unit, we need to check here.
5297	 */
5298	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5299		buf = io->scsiio.kern_data_ptr;
5300	else
5301		buf = NULL;
5302	ctl_done(io);
5303	if (buf)
5304		free(buf, M_CTL);
5305}
5306
5307/*
5308 * SCSI release command.
5309 */
5310int
5311ctl_scsi_release(struct ctl_scsiio *ctsio)
5312{
5313	int length, longid, thirdparty_id, resv_id;
5314	struct ctl_softc *ctl_softc;
5315	struct ctl_lun *lun;
5316	uint32_t residx;
5317
5318	length = 0;
5319	resv_id = 0;
5320
5321	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5322
5323	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5324	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5325	ctl_softc = control_softc;
5326
5327	switch (ctsio->cdb[0]) {
5328	case RELEASE_10: {
5329		struct scsi_release_10 *cdb;
5330
5331		cdb = (struct scsi_release_10 *)ctsio->cdb;
5332
5333		if (cdb->byte2 & SR10_LONGID)
5334			longid = 1;
5335		else
5336			thirdparty_id = cdb->thirdparty_id;
5337
5338		resv_id = cdb->resv_id;
5339		length = scsi_2btoul(cdb->length);
5340		break;
5341	}
5342	}
5343
5344
5345	/*
5346	 * XXX KDM right now, we only support LUN reservation.  We don't
5347	 * support 3rd party reservations, or extent reservations, which
5348	 * might actually need the parameter list.  If we've gotten this
5349	 * far, we've got a LUN reservation.  Anything else got kicked out
5350	 * above.  So, according to SPC, ignore the length.
5351	 */
5352	length = 0;
5353
5354	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5355	 && (length > 0)) {
5356		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5357		ctsio->kern_data_len = length;
5358		ctsio->kern_total_len = length;
5359		ctsio->kern_data_resid = 0;
5360		ctsio->kern_rel_offset = 0;
5361		ctsio->kern_sg_entries = 0;
5362		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5363		ctsio->be_move_done = ctl_config_move_done;
5364		ctl_datamove((union ctl_io *)ctsio);
5365
5366		return (CTL_RETVAL_COMPLETE);
5367	}
5368
5369	if (length > 0)
5370		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5371
5372	mtx_lock(&lun->lun_lock);
5373
5374	/*
5375	 * According to SPC, it is not an error for an intiator to attempt
5376	 * to release a reservation on a LUN that isn't reserved, or that
5377	 * is reserved by another initiator.  The reservation can only be
5378	 * released, though, by the initiator who made it or by one of
5379	 * several reset type events.
5380	 */
5381	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5382			lun->flags &= ~CTL_LUN_RESERVED;
5383
5384	mtx_unlock(&lun->lun_lock);
5385
5386	ctsio->scsi_status = SCSI_STATUS_OK;
5387	ctsio->io_hdr.status = CTL_SUCCESS;
5388
5389	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5390		free(ctsio->kern_data_ptr, M_CTL);
5391		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5392	}
5393
5394	ctl_done((union ctl_io *)ctsio);
5395	return (CTL_RETVAL_COMPLETE);
5396}
5397
5398int
5399ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5400{
5401	int extent, thirdparty, longid;
5402	int resv_id, length;
5403	uint64_t thirdparty_id;
5404	struct ctl_softc *ctl_softc;
5405	struct ctl_lun *lun;
5406	uint32_t residx;
5407
5408	extent = 0;
5409	thirdparty = 0;
5410	longid = 0;
5411	resv_id = 0;
5412	length = 0;
5413	thirdparty_id = 0;
5414
5415	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5416
5417	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5418	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5419	ctl_softc = control_softc;
5420
5421	switch (ctsio->cdb[0]) {
5422	case RESERVE_10: {
5423		struct scsi_reserve_10 *cdb;
5424
5425		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5426
5427		if (cdb->byte2 & SR10_LONGID)
5428			longid = 1;
5429		else
5430			thirdparty_id = cdb->thirdparty_id;
5431
5432		resv_id = cdb->resv_id;
5433		length = scsi_2btoul(cdb->length);
5434		break;
5435	}
5436	}
5437
5438	/*
5439	 * XXX KDM right now, we only support LUN reservation.  We don't
5440	 * support 3rd party reservations, or extent reservations, which
5441	 * might actually need the parameter list.  If we've gotten this
5442	 * far, we've got a LUN reservation.  Anything else got kicked out
5443	 * above.  So, according to SPC, ignore the length.
5444	 */
5445	length = 0;
5446
5447	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5448	 && (length > 0)) {
5449		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5450		ctsio->kern_data_len = length;
5451		ctsio->kern_total_len = length;
5452		ctsio->kern_data_resid = 0;
5453		ctsio->kern_rel_offset = 0;
5454		ctsio->kern_sg_entries = 0;
5455		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5456		ctsio->be_move_done = ctl_config_move_done;
5457		ctl_datamove((union ctl_io *)ctsio);
5458
5459		return (CTL_RETVAL_COMPLETE);
5460	}
5461
5462	if (length > 0)
5463		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5464
5465	mtx_lock(&lun->lun_lock);
5466	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5467		ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5468		ctsio->io_hdr.status = CTL_SCSI_ERROR;
5469		goto bailout;
5470	}
5471
5472	lun->flags |= CTL_LUN_RESERVED;
5473	lun->res_idx = residx;
5474
5475	ctsio->scsi_status = SCSI_STATUS_OK;
5476	ctsio->io_hdr.status = CTL_SUCCESS;
5477
5478bailout:
5479	mtx_unlock(&lun->lun_lock);
5480
5481	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5482		free(ctsio->kern_data_ptr, M_CTL);
5483		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5484	}
5485
5486	ctl_done((union ctl_io *)ctsio);
5487	return (CTL_RETVAL_COMPLETE);
5488}
5489
5490int
5491ctl_start_stop(struct ctl_scsiio *ctsio)
5492{
5493	struct scsi_start_stop_unit *cdb;
5494	struct ctl_lun *lun;
5495	struct ctl_softc *ctl_softc;
5496	int retval;
5497
5498	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5499
5500	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5501	ctl_softc = control_softc;
5502	retval = 0;
5503
5504	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5505
5506	/*
5507	 * XXX KDM
5508	 * We don't support the immediate bit on a stop unit.  In order to
5509	 * do that, we would need to code up a way to know that a stop is
5510	 * pending, and hold off any new commands until it completes, one
5511	 * way or another.  Then we could accept or reject those commands
5512	 * depending on its status.  We would almost need to do the reverse
5513	 * of what we do below for an immediate start -- return the copy of
5514	 * the ctl_io to the FETD with status to send to the host (and to
5515	 * free the copy!) and then free the original I/O once the stop
5516	 * actually completes.  That way, the OOA queue mechanism can work
5517	 * to block commands that shouldn't proceed.  Another alternative
5518	 * would be to put the copy in the queue in place of the original,
5519	 * and return the original back to the caller.  That could be
5520	 * slightly safer..
5521	 */
5522	if ((cdb->byte2 & SSS_IMMED)
5523	 && ((cdb->how & SSS_START) == 0)) {
5524		ctl_set_invalid_field(ctsio,
5525				      /*sks_valid*/ 1,
5526				      /*command*/ 1,
5527				      /*field*/ 1,
5528				      /*bit_valid*/ 1,
5529				      /*bit*/ 0);
5530		ctl_done((union ctl_io *)ctsio);
5531		return (CTL_RETVAL_COMPLETE);
5532	}
5533
5534	if ((lun->flags & CTL_LUN_PR_RESERVED)
5535	 && ((cdb->how & SSS_START)==0)) {
5536		uint32_t residx;
5537
5538		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5539		if (!lun->per_res[residx].registered
5540		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5541
5542			ctl_set_reservation_conflict(ctsio);
5543			ctl_done((union ctl_io *)ctsio);
5544			return (CTL_RETVAL_COMPLETE);
5545		}
5546	}
5547
5548	/*
5549	 * If there is no backend on this device, we can't start or stop
5550	 * it.  In theory we shouldn't get any start/stop commands in the
5551	 * first place at this level if the LUN doesn't have a backend.
5552	 * That should get stopped by the command decode code.
5553	 */
5554	if (lun->backend == NULL) {
5555		ctl_set_invalid_opcode(ctsio);
5556		ctl_done((union ctl_io *)ctsio);
5557		return (CTL_RETVAL_COMPLETE);
5558	}
5559
5560	/*
5561	 * XXX KDM Copan-specific offline behavior.
5562	 * Figure out a reasonable way to port this?
5563	 */
5564#ifdef NEEDTOPORT
5565	mtx_lock(&lun->lun_lock);
5566
5567	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5568	 && (lun->flags & CTL_LUN_OFFLINE)) {
5569		/*
5570		 * If the LUN is offline, and the on/offline bit isn't set,
5571		 * reject the start or stop.  Otherwise, let it through.
5572		 */
5573		mtx_unlock(&lun->lun_lock);
5574		ctl_set_lun_not_ready(ctsio);
5575		ctl_done((union ctl_io *)ctsio);
5576	} else {
5577		mtx_unlock(&lun->lun_lock);
5578#endif /* NEEDTOPORT */
5579		/*
5580		 * This could be a start or a stop when we're online,
5581		 * or a stop/offline or start/online.  A start or stop when
5582		 * we're offline is covered in the case above.
5583		 */
5584		/*
5585		 * In the non-immediate case, we send the request to
5586		 * the backend and return status to the user when
5587		 * it is done.
5588		 *
5589		 * In the immediate case, we allocate a new ctl_io
5590		 * to hold a copy of the request, and send that to
5591		 * the backend.  We then set good status on the
5592		 * user's request and return it immediately.
5593		 */
5594		if (cdb->byte2 & SSS_IMMED) {
5595			union ctl_io *new_io;
5596
5597			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5598			if (new_io == NULL) {
5599				ctl_set_busy(ctsio);
5600				ctl_done((union ctl_io *)ctsio);
5601			} else {
5602				ctl_copy_io((union ctl_io *)ctsio,
5603					    new_io);
5604				retval = lun->backend->config_write(new_io);
5605				ctl_set_success(ctsio);
5606				ctl_done((union ctl_io *)ctsio);
5607			}
5608		} else {
5609			retval = lun->backend->config_write(
5610				(union ctl_io *)ctsio);
5611		}
5612#ifdef NEEDTOPORT
5613	}
5614#endif
5615	return (retval);
5616}
5617
5618/*
5619 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5620 * we don't really do anything with the LBA and length fields if the user
5621 * passes them in.  Instead we'll just flush out the cache for the entire
5622 * LUN.
5623 */
5624int
5625ctl_sync_cache(struct ctl_scsiio *ctsio)
5626{
5627	struct ctl_lun *lun;
5628	struct ctl_softc *ctl_softc;
5629	uint64_t starting_lba;
5630	uint32_t block_count;
5631	int retval;
5632
5633	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5634
5635	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5636	ctl_softc = control_softc;
5637	retval = 0;
5638
5639	switch (ctsio->cdb[0]) {
5640	case SYNCHRONIZE_CACHE: {
5641		struct scsi_sync_cache *cdb;
5642		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5643
5644		starting_lba = scsi_4btoul(cdb->begin_lba);
5645		block_count = scsi_2btoul(cdb->lb_count);
5646		break;
5647	}
5648	case SYNCHRONIZE_CACHE_16: {
5649		struct scsi_sync_cache_16 *cdb;
5650		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5651
5652		starting_lba = scsi_8btou64(cdb->begin_lba);
5653		block_count = scsi_4btoul(cdb->lb_count);
5654		break;
5655	}
5656	default:
5657		ctl_set_invalid_opcode(ctsio);
5658		ctl_done((union ctl_io *)ctsio);
5659		goto bailout;
5660		break; /* NOTREACHED */
5661	}
5662
5663	/*
5664	 * We check the LBA and length, but don't do anything with them.
5665	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5666	 * get flushed.  This check will just help satisfy anyone who wants
5667	 * to see an error for an out of range LBA.
5668	 */
5669	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5670		ctl_set_lba_out_of_range(ctsio);
5671		ctl_done((union ctl_io *)ctsio);
5672		goto bailout;
5673	}
5674
5675	/*
5676	 * If this LUN has no backend, we can't flush the cache anyway.
5677	 */
5678	if (lun->backend == NULL) {
5679		ctl_set_invalid_opcode(ctsio);
5680		ctl_done((union ctl_io *)ctsio);
5681		goto bailout;
5682	}
5683
5684	/*
5685	 * Check to see whether we're configured to send the SYNCHRONIZE
5686	 * CACHE command directly to the back end.
5687	 */
5688	mtx_lock(&lun->lun_lock);
5689	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5690	 && (++(lun->sync_count) >= lun->sync_interval)) {
5691		lun->sync_count = 0;
5692		mtx_unlock(&lun->lun_lock);
5693		retval = lun->backend->config_write((union ctl_io *)ctsio);
5694	} else {
5695		mtx_unlock(&lun->lun_lock);
5696		ctl_set_success(ctsio);
5697		ctl_done((union ctl_io *)ctsio);
5698	}
5699
5700bailout:
5701
5702	return (retval);
5703}
5704
5705int
5706ctl_format(struct ctl_scsiio *ctsio)
5707{
5708	struct scsi_format *cdb;
5709	struct ctl_lun *lun;
5710	struct ctl_softc *ctl_softc;
5711	int length, defect_list_len;
5712
5713	CTL_DEBUG_PRINT(("ctl_format\n"));
5714
5715	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5716	ctl_softc = control_softc;
5717
5718	cdb = (struct scsi_format *)ctsio->cdb;
5719
5720	length = 0;
5721	if (cdb->byte2 & SF_FMTDATA) {
5722		if (cdb->byte2 & SF_LONGLIST)
5723			length = sizeof(struct scsi_format_header_long);
5724		else
5725			length = sizeof(struct scsi_format_header_short);
5726	}
5727
5728	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5729	 && (length > 0)) {
5730		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5731		ctsio->kern_data_len = length;
5732		ctsio->kern_total_len = length;
5733		ctsio->kern_data_resid = 0;
5734		ctsio->kern_rel_offset = 0;
5735		ctsio->kern_sg_entries = 0;
5736		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5737		ctsio->be_move_done = ctl_config_move_done;
5738		ctl_datamove((union ctl_io *)ctsio);
5739
5740		return (CTL_RETVAL_COMPLETE);
5741	}
5742
5743	defect_list_len = 0;
5744
5745	if (cdb->byte2 & SF_FMTDATA) {
5746		if (cdb->byte2 & SF_LONGLIST) {
5747			struct scsi_format_header_long *header;
5748
5749			header = (struct scsi_format_header_long *)
5750				ctsio->kern_data_ptr;
5751
5752			defect_list_len = scsi_4btoul(header->defect_list_len);
5753			if (defect_list_len != 0) {
5754				ctl_set_invalid_field(ctsio,
5755						      /*sks_valid*/ 1,
5756						      /*command*/ 0,
5757						      /*field*/ 2,
5758						      /*bit_valid*/ 0,
5759						      /*bit*/ 0);
5760				goto bailout;
5761			}
5762		} else {
5763			struct scsi_format_header_short *header;
5764
5765			header = (struct scsi_format_header_short *)
5766				ctsio->kern_data_ptr;
5767
5768			defect_list_len = scsi_2btoul(header->defect_list_len);
5769			if (defect_list_len != 0) {
5770				ctl_set_invalid_field(ctsio,
5771						      /*sks_valid*/ 1,
5772						      /*command*/ 0,
5773						      /*field*/ 2,
5774						      /*bit_valid*/ 0,
5775						      /*bit*/ 0);
5776				goto bailout;
5777			}
5778		}
5779	}
5780
5781	/*
5782	 * The format command will clear out the "Medium format corrupted"
5783	 * status if set by the configuration code.  That status is really
5784	 * just a way to notify the host that we have lost the media, and
5785	 * get them to issue a command that will basically make them think
5786	 * they're blowing away the media.
5787	 */
5788	mtx_lock(&lun->lun_lock);
5789	lun->flags &= ~CTL_LUN_INOPERABLE;
5790	mtx_unlock(&lun->lun_lock);
5791
5792	ctsio->scsi_status = SCSI_STATUS_OK;
5793	ctsio->io_hdr.status = CTL_SUCCESS;
5794bailout:
5795
5796	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5797		free(ctsio->kern_data_ptr, M_CTL);
5798		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5799	}
5800
5801	ctl_done((union ctl_io *)ctsio);
5802	return (CTL_RETVAL_COMPLETE);
5803}
5804
5805int
5806ctl_read_buffer(struct ctl_scsiio *ctsio)
5807{
5808	struct scsi_read_buffer *cdb;
5809	struct ctl_lun *lun;
5810	int buffer_offset, len;
5811	static uint8_t descr[4];
5812	static uint8_t echo_descr[4] = { 0 };
5813
5814	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5815
5816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5817	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5818
5819	if (lun->flags & CTL_LUN_PR_RESERVED) {
5820		uint32_t residx;
5821
5822		/*
5823		 * XXX KDM need a lock here.
5824		 */
5825		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5826		if ((lun->res_type == SPR_TYPE_EX_AC
5827		  && residx != lun->pr_res_idx)
5828		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5829		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5830		  && !lun->per_res[residx].registered)) {
5831			ctl_set_reservation_conflict(ctsio);
5832			ctl_done((union ctl_io *)ctsio);
5833			return (CTL_RETVAL_COMPLETE);
5834	        }
5835	}
5836
5837	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5838	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5839	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5840		ctl_set_invalid_field(ctsio,
5841				      /*sks_valid*/ 1,
5842				      /*command*/ 1,
5843				      /*field*/ 1,
5844				      /*bit_valid*/ 1,
5845				      /*bit*/ 4);
5846		ctl_done((union ctl_io *)ctsio);
5847		return (CTL_RETVAL_COMPLETE);
5848	}
5849
5850	len = scsi_3btoul(cdb->length);
5851	buffer_offset = scsi_3btoul(cdb->offset);
5852
5853	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5854		ctl_set_invalid_field(ctsio,
5855				      /*sks_valid*/ 1,
5856				      /*command*/ 1,
5857				      /*field*/ 6,
5858				      /*bit_valid*/ 0,
5859				      /*bit*/ 0);
5860		ctl_done((union ctl_io *)ctsio);
5861		return (CTL_RETVAL_COMPLETE);
5862	}
5863
5864	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5865		descr[0] = 0;
5866		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5867		ctsio->kern_data_ptr = descr;
5868		len = min(len, sizeof(descr));
5869	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5870		ctsio->kern_data_ptr = echo_descr;
5871		len = min(len, sizeof(echo_descr));
5872	} else
5873		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5874	ctsio->kern_data_len = len;
5875	ctsio->kern_total_len = len;
5876	ctsio->kern_data_resid = 0;
5877	ctsio->kern_rel_offset = 0;
5878	ctsio->kern_sg_entries = 0;
5879	ctsio->be_move_done = ctl_config_move_done;
5880	ctl_datamove((union ctl_io *)ctsio);
5881
5882	return (CTL_RETVAL_COMPLETE);
5883}
5884
5885int
5886ctl_write_buffer(struct ctl_scsiio *ctsio)
5887{
5888	struct scsi_write_buffer *cdb;
5889	struct ctl_lun *lun;
5890	int buffer_offset, len;
5891
5892	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5893
5894	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5895	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5896
5897	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5898		ctl_set_invalid_field(ctsio,
5899				      /*sks_valid*/ 1,
5900				      /*command*/ 1,
5901				      /*field*/ 1,
5902				      /*bit_valid*/ 1,
5903				      /*bit*/ 4);
5904		ctl_done((union ctl_io *)ctsio);
5905		return (CTL_RETVAL_COMPLETE);
5906	}
5907
5908	len = scsi_3btoul(cdb->length);
5909	buffer_offset = scsi_3btoul(cdb->offset);
5910
5911	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5912		ctl_set_invalid_field(ctsio,
5913				      /*sks_valid*/ 1,
5914				      /*command*/ 1,
5915				      /*field*/ 6,
5916				      /*bit_valid*/ 0,
5917				      /*bit*/ 0);
5918		ctl_done((union ctl_io *)ctsio);
5919		return (CTL_RETVAL_COMPLETE);
5920	}
5921
5922	/*
5923	 * If we've got a kernel request that hasn't been malloced yet,
5924	 * malloc it and tell the caller the data buffer is here.
5925	 */
5926	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5927		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5928		ctsio->kern_data_len = len;
5929		ctsio->kern_total_len = len;
5930		ctsio->kern_data_resid = 0;
5931		ctsio->kern_rel_offset = 0;
5932		ctsio->kern_sg_entries = 0;
5933		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5934		ctsio->be_move_done = ctl_config_move_done;
5935		ctl_datamove((union ctl_io *)ctsio);
5936
5937		return (CTL_RETVAL_COMPLETE);
5938	}
5939
5940	ctl_done((union ctl_io *)ctsio);
5941
5942	return (CTL_RETVAL_COMPLETE);
5943}
5944
5945int
5946ctl_write_same(struct ctl_scsiio *ctsio)
5947{
5948	struct ctl_lun *lun;
5949	struct ctl_lba_len_flags *lbalen;
5950	uint64_t lba;
5951	uint32_t num_blocks;
5952	int len, retval;
5953	uint8_t byte2;
5954
5955	retval = CTL_RETVAL_COMPLETE;
5956
5957	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5958
5959	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5960
5961	switch (ctsio->cdb[0]) {
5962	case WRITE_SAME_10: {
5963		struct scsi_write_same_10 *cdb;
5964
5965		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5966
5967		lba = scsi_4btoul(cdb->addr);
5968		num_blocks = scsi_2btoul(cdb->length);
5969		byte2 = cdb->byte2;
5970		break;
5971	}
5972	case WRITE_SAME_16: {
5973		struct scsi_write_same_16 *cdb;
5974
5975		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5976
5977		lba = scsi_8btou64(cdb->addr);
5978		num_blocks = scsi_4btoul(cdb->length);
5979		byte2 = cdb->byte2;
5980		break;
5981	}
5982	default:
5983		/*
5984		 * We got a command we don't support.  This shouldn't
5985		 * happen, commands should be filtered out above us.
5986		 */
5987		ctl_set_invalid_opcode(ctsio);
5988		ctl_done((union ctl_io *)ctsio);
5989
5990		return (CTL_RETVAL_COMPLETE);
5991		break; /* NOTREACHED */
5992	}
5993
5994	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5995	if ((byte2 & SWS_UNMAP) == 0 &&
5996	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5997		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5998		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5999		ctl_done((union ctl_io *)ctsio);
6000		return (CTL_RETVAL_COMPLETE);
6001	}
6002
6003	/*
6004	 * The first check is to make sure we're in bounds, the second
6005	 * check is to catch wrap-around problems.  If the lba + num blocks
6006	 * is less than the lba, then we've wrapped around and the block
6007	 * range is invalid anyway.
6008	 */
6009	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6010	 || ((lba + num_blocks) < lba)) {
6011		ctl_set_lba_out_of_range(ctsio);
6012		ctl_done((union ctl_io *)ctsio);
6013		return (CTL_RETVAL_COMPLETE);
6014	}
6015
6016	/* Zero number of blocks means "to the last logical block" */
6017	if (num_blocks == 0) {
6018		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
6019			ctl_set_invalid_field(ctsio,
6020					      /*sks_valid*/ 0,
6021					      /*command*/ 1,
6022					      /*field*/ 0,
6023					      /*bit_valid*/ 0,
6024					      /*bit*/ 0);
6025			ctl_done((union ctl_io *)ctsio);
6026			return (CTL_RETVAL_COMPLETE);
6027		}
6028		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6029	}
6030
6031	len = lun->be_lun->blocksize;
6032
6033	/*
6034	 * If we've got a kernel request that hasn't been malloced yet,
6035	 * malloc it and tell the caller the data buffer is here.
6036	 */
6037	if ((byte2 & SWS_NDOB) == 0 &&
6038	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6039		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6040		ctsio->kern_data_len = len;
6041		ctsio->kern_total_len = len;
6042		ctsio->kern_data_resid = 0;
6043		ctsio->kern_rel_offset = 0;
6044		ctsio->kern_sg_entries = 0;
6045		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6046		ctsio->be_move_done = ctl_config_move_done;
6047		ctl_datamove((union ctl_io *)ctsio);
6048
6049		return (CTL_RETVAL_COMPLETE);
6050	}
6051
6052	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6053	lbalen->lba = lba;
6054	lbalen->len = num_blocks;
6055	lbalen->flags = byte2;
6056	retval = lun->backend->config_write((union ctl_io *)ctsio);
6057
6058	return (retval);
6059}
6060
6061int
6062ctl_unmap(struct ctl_scsiio *ctsio)
6063{
6064	struct ctl_lun *lun;
6065	struct scsi_unmap *cdb;
6066	struct ctl_ptr_len_flags *ptrlen;
6067	struct scsi_unmap_header *hdr;
6068	struct scsi_unmap_desc *buf, *end, *endnz, *range;
6069	uint64_t lba;
6070	uint32_t num_blocks;
6071	int len, retval;
6072	uint8_t byte2;
6073
6074	retval = CTL_RETVAL_COMPLETE;
6075
6076	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6077
6078	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6079	cdb = (struct scsi_unmap *)ctsio->cdb;
6080
6081	len = scsi_2btoul(cdb->length);
6082	byte2 = cdb->byte2;
6083
6084	/*
6085	 * If we've got a kernel request that hasn't been malloced yet,
6086	 * malloc it and tell the caller the data buffer is here.
6087	 */
6088	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6089		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6090		ctsio->kern_data_len = len;
6091		ctsio->kern_total_len = len;
6092		ctsio->kern_data_resid = 0;
6093		ctsio->kern_rel_offset = 0;
6094		ctsio->kern_sg_entries = 0;
6095		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6096		ctsio->be_move_done = ctl_config_move_done;
6097		ctl_datamove((union ctl_io *)ctsio);
6098
6099		return (CTL_RETVAL_COMPLETE);
6100	}
6101
6102	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6103	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6104	if (len < sizeof (*hdr) ||
6105	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6106	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6107	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6108		ctl_set_invalid_field(ctsio,
6109				      /*sks_valid*/ 0,
6110				      /*command*/ 0,
6111				      /*field*/ 0,
6112				      /*bit_valid*/ 0,
6113				      /*bit*/ 0);
6114		ctl_done((union ctl_io *)ctsio);
6115		return (CTL_RETVAL_COMPLETE);
6116	}
6117	len = scsi_2btoul(hdr->desc_length);
6118	buf = (struct scsi_unmap_desc *)(hdr + 1);
6119	end = buf + len / sizeof(*buf);
6120
6121	endnz = buf;
6122	for (range = buf; range < end; range++) {
6123		lba = scsi_8btou64(range->lba);
6124		num_blocks = scsi_4btoul(range->length);
6125		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6126		 || ((lba + num_blocks) < lba)) {
6127			ctl_set_lba_out_of_range(ctsio);
6128			ctl_done((union ctl_io *)ctsio);
6129			return (CTL_RETVAL_COMPLETE);
6130		}
6131		if (num_blocks != 0)
6132			endnz = range + 1;
6133	}
6134
6135	/*
6136	 * Block backend can not handle zero last range.
6137	 * Filter it out and return if there is nothing left.
6138	 */
6139	len = (uint8_t *)endnz - (uint8_t *)buf;
6140	if (len == 0) {
6141		ctl_set_success(ctsio);
6142		ctl_done((union ctl_io *)ctsio);
6143		return (CTL_RETVAL_COMPLETE);
6144	}
6145
6146	mtx_lock(&lun->lun_lock);
6147	ptrlen = (struct ctl_ptr_len_flags *)
6148	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6149	ptrlen->ptr = (void *)buf;
6150	ptrlen->len = len;
6151	ptrlen->flags = byte2;
6152	ctl_check_blocked(lun);
6153	mtx_unlock(&lun->lun_lock);
6154
6155	retval = lun->backend->config_write((union ctl_io *)ctsio);
6156	return (retval);
6157}
6158
6159/*
6160 * Note that this function currently doesn't actually do anything inside
6161 * CTL to enforce things if the DQue bit is turned on.
6162 *
6163 * Also note that this function can't be used in the default case, because
6164 * the DQue bit isn't set in the changeable mask for the control mode page
6165 * anyway.  This is just here as an example for how to implement a page
6166 * handler, and a placeholder in case we want to allow the user to turn
6167 * tagged queueing on and off.
6168 *
6169 * The D_SENSE bit handling is functional, however, and will turn
6170 * descriptor sense on and off for a given LUN.
6171 */
6172int
6173ctl_control_page_handler(struct ctl_scsiio *ctsio,
6174			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6175{
6176	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6177	struct ctl_lun *lun;
6178	struct ctl_softc *softc;
6179	int set_ua;
6180	uint32_t initidx;
6181
6182	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6183	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6184	set_ua = 0;
6185
6186	user_cp = (struct scsi_control_page *)page_ptr;
6187	current_cp = (struct scsi_control_page *)
6188		(page_index->page_data + (page_index->page_len *
6189		CTL_PAGE_CURRENT));
6190	saved_cp = (struct scsi_control_page *)
6191		(page_index->page_data + (page_index->page_len *
6192		CTL_PAGE_SAVED));
6193
6194	softc = control_softc;
6195
6196	mtx_lock(&lun->lun_lock);
6197	if (((current_cp->rlec & SCP_DSENSE) == 0)
6198	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6199		/*
6200		 * Descriptor sense is currently turned off and the user
6201		 * wants to turn it on.
6202		 */
6203		current_cp->rlec |= SCP_DSENSE;
6204		saved_cp->rlec |= SCP_DSENSE;
6205		lun->flags |= CTL_LUN_SENSE_DESC;
6206		set_ua = 1;
6207	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6208		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6209		/*
6210		 * Descriptor sense is currently turned on, and the user
6211		 * wants to turn it off.
6212		 */
6213		current_cp->rlec &= ~SCP_DSENSE;
6214		saved_cp->rlec &= ~SCP_DSENSE;
6215		lun->flags &= ~CTL_LUN_SENSE_DESC;
6216		set_ua = 1;
6217	}
6218	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6219	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6220		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6221		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6222		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6223		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6224		set_ua = 1;
6225	}
6226	if ((current_cp->eca_and_aen & SCP_SWP) !=
6227	    (user_cp->eca_and_aen & SCP_SWP)) {
6228		current_cp->eca_and_aen &= ~SCP_SWP;
6229		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6230		saved_cp->eca_and_aen &= ~SCP_SWP;
6231		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6232		set_ua = 1;
6233	}
6234	if (set_ua != 0) {
6235		int i;
6236		/*
6237		 * Let other initiators know that the mode
6238		 * parameters for this LUN have changed.
6239		 */
6240		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6241			if (i == initidx)
6242				continue;
6243
6244			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6245		}
6246	}
6247	mtx_unlock(&lun->lun_lock);
6248
6249	return (0);
6250}
6251
6252int
6253ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6254		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6255{
6256	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6257	struct ctl_lun *lun;
6258	int set_ua;
6259	uint32_t initidx;
6260
6261	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6262	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6263	set_ua = 0;
6264
6265	user_cp = (struct scsi_caching_page *)page_ptr;
6266	current_cp = (struct scsi_caching_page *)
6267		(page_index->page_data + (page_index->page_len *
6268		CTL_PAGE_CURRENT));
6269	saved_cp = (struct scsi_caching_page *)
6270		(page_index->page_data + (page_index->page_len *
6271		CTL_PAGE_SAVED));
6272
6273	mtx_lock(&lun->lun_lock);
6274	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6275	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6276		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6277		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6278		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6279		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6280		set_ua = 1;
6281	}
6282	if (set_ua != 0) {
6283		int i;
6284		/*
6285		 * Let other initiators know that the mode
6286		 * parameters for this LUN have changed.
6287		 */
6288		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6289			if (i == initidx)
6290				continue;
6291
6292			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6293		}
6294	}
6295	mtx_unlock(&lun->lun_lock);
6296
6297	return (0);
6298}
6299
6300int
6301ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6302		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6303{
6304	return (0);
6305}
6306
6307int
6308ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6309			   struct ctl_page_index *page_index, int pc)
6310{
6311	struct copan_power_subpage *page;
6312
6313	page = (struct copan_power_subpage *)page_index->page_data +
6314		(page_index->page_len * pc);
6315
6316	switch (pc) {
6317	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6318		/*
6319		 * We don't update the changable bits for this page.
6320		 */
6321		break;
6322	case SMS_PAGE_CTRL_CURRENT >> 6:
6323	case SMS_PAGE_CTRL_DEFAULT >> 6:
6324	case SMS_PAGE_CTRL_SAVED >> 6:
6325#ifdef NEEDTOPORT
6326		ctl_update_power_subpage(page);
6327#endif
6328		break;
6329	default:
6330#ifdef NEEDTOPORT
6331		EPRINT(0, "Invalid PC %d!!", pc);
6332#endif
6333		break;
6334	}
6335	return (0);
6336}
6337
6338
6339int
6340ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6341		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6342{
6343	struct copan_aps_subpage *user_sp;
6344	struct copan_aps_subpage *current_sp;
6345	union ctl_modepage_info *modepage_info;
6346	struct ctl_softc *softc;
6347	struct ctl_lun *lun;
6348	int retval;
6349
6350	retval = CTL_RETVAL_COMPLETE;
6351	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6352		     (page_index->page_len * CTL_PAGE_CURRENT));
6353	softc = control_softc;
6354	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6355
6356	user_sp = (struct copan_aps_subpage *)page_ptr;
6357
6358	modepage_info = (union ctl_modepage_info *)
6359		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6360
6361	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6362	modepage_info->header.subpage = page_index->subpage;
6363	modepage_info->aps.lock_active = user_sp->lock_active;
6364
6365	mtx_lock(&softc->ctl_lock);
6366
6367	/*
6368	 * If there is a request to lock the LUN and another LUN is locked
6369	 * this is an error. If the requested LUN is already locked ignore
6370	 * the request. If no LUN is locked attempt to lock it.
6371	 * if there is a request to unlock the LUN and the LUN is currently
6372	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6373	 * if another LUN is locked or no LUN is locked.
6374	 */
6375	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6376		if (softc->aps_locked_lun == lun->lun) {
6377			/*
6378			 * This LUN is already locked, so we're done.
6379			 */
6380			retval = CTL_RETVAL_COMPLETE;
6381		} else if (softc->aps_locked_lun == 0) {
6382			/*
6383			 * No one has the lock, pass the request to the
6384			 * backend.
6385			 */
6386			retval = lun->backend->config_write(
6387				(union ctl_io *)ctsio);
6388		} else {
6389			/*
6390			 * Someone else has the lock, throw out the request.
6391			 */
6392			ctl_set_already_locked(ctsio);
6393			free(ctsio->kern_data_ptr, M_CTL);
6394			ctl_done((union ctl_io *)ctsio);
6395
6396			/*
6397			 * Set the return value so that ctl_do_mode_select()
6398			 * won't try to complete the command.  We already
6399			 * completed it here.
6400			 */
6401			retval = CTL_RETVAL_ERROR;
6402		}
6403	} else if (softc->aps_locked_lun == lun->lun) {
6404		/*
6405		 * This LUN is locked, so pass the unlock request to the
6406		 * backend.
6407		 */
6408		retval = lun->backend->config_write((union ctl_io *)ctsio);
6409	}
6410	mtx_unlock(&softc->ctl_lock);
6411
6412	return (retval);
6413}
6414
6415int
6416ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6417				struct ctl_page_index *page_index,
6418				uint8_t *page_ptr)
6419{
6420	uint8_t *c;
6421	int i;
6422
6423	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6424	ctl_time_io_secs =
6425		(c[0] << 8) |
6426		(c[1] << 0) |
6427		0;
6428	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6429	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6430	printf("page data:");
6431	for (i=0; i<8; i++)
6432		printf(" %.2x",page_ptr[i]);
6433	printf("\n");
6434	return (0);
6435}
6436
6437int
6438ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6439			       struct ctl_page_index *page_index,
6440			       int pc)
6441{
6442	struct copan_debugconf_subpage *page;
6443
6444	page = (struct copan_debugconf_subpage *)page_index->page_data +
6445		(page_index->page_len * pc);
6446
6447	switch (pc) {
6448	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6449	case SMS_PAGE_CTRL_DEFAULT >> 6:
6450	case SMS_PAGE_CTRL_SAVED >> 6:
6451		/*
6452		 * We don't update the changable or default bits for this page.
6453		 */
6454		break;
6455	case SMS_PAGE_CTRL_CURRENT >> 6:
6456		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6457		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6458		break;
6459	default:
6460#ifdef NEEDTOPORT
6461		EPRINT(0, "Invalid PC %d!!", pc);
6462#endif /* NEEDTOPORT */
6463		break;
6464	}
6465	return (0);
6466}
6467
6468
6469static int
6470ctl_do_mode_select(union ctl_io *io)
6471{
6472	struct scsi_mode_page_header *page_header;
6473	struct ctl_page_index *page_index;
6474	struct ctl_scsiio *ctsio;
6475	int control_dev, page_len;
6476	int page_len_offset, page_len_size;
6477	union ctl_modepage_info *modepage_info;
6478	struct ctl_lun *lun;
6479	int *len_left, *len_used;
6480	int retval, i;
6481
6482	ctsio = &io->scsiio;
6483	page_index = NULL;
6484	page_len = 0;
6485	retval = CTL_RETVAL_COMPLETE;
6486
6487	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6488
6489	if (lun->be_lun->lun_type != T_DIRECT)
6490		control_dev = 1;
6491	else
6492		control_dev = 0;
6493
6494	modepage_info = (union ctl_modepage_info *)
6495		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6496	len_left = &modepage_info->header.len_left;
6497	len_used = &modepage_info->header.len_used;
6498
6499do_next_page:
6500
6501	page_header = (struct scsi_mode_page_header *)
6502		(ctsio->kern_data_ptr + *len_used);
6503
6504	if (*len_left == 0) {
6505		free(ctsio->kern_data_ptr, M_CTL);
6506		ctl_set_success(ctsio);
6507		ctl_done((union ctl_io *)ctsio);
6508		return (CTL_RETVAL_COMPLETE);
6509	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6510
6511		free(ctsio->kern_data_ptr, M_CTL);
6512		ctl_set_param_len_error(ctsio);
6513		ctl_done((union ctl_io *)ctsio);
6514		return (CTL_RETVAL_COMPLETE);
6515
6516	} else if ((page_header->page_code & SMPH_SPF)
6517		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6518
6519		free(ctsio->kern_data_ptr, M_CTL);
6520		ctl_set_param_len_error(ctsio);
6521		ctl_done((union ctl_io *)ctsio);
6522		return (CTL_RETVAL_COMPLETE);
6523	}
6524
6525
6526	/*
6527	 * XXX KDM should we do something with the block descriptor?
6528	 */
6529	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6530
6531		if ((control_dev != 0)
6532		 && (lun->mode_pages.index[i].page_flags &
6533		     CTL_PAGE_FLAG_DISK_ONLY))
6534			continue;
6535
6536		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6537		    (page_header->page_code & SMPH_PC_MASK))
6538			continue;
6539
6540		/*
6541		 * If neither page has a subpage code, then we've got a
6542		 * match.
6543		 */
6544		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6545		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6546			page_index = &lun->mode_pages.index[i];
6547			page_len = page_header->page_length;
6548			break;
6549		}
6550
6551		/*
6552		 * If both pages have subpages, then the subpage numbers
6553		 * have to match.
6554		 */
6555		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6556		  && (page_header->page_code & SMPH_SPF)) {
6557			struct scsi_mode_page_header_sp *sph;
6558
6559			sph = (struct scsi_mode_page_header_sp *)page_header;
6560
6561			if (lun->mode_pages.index[i].subpage ==
6562			    sph->subpage) {
6563				page_index = &lun->mode_pages.index[i];
6564				page_len = scsi_2btoul(sph->page_length);
6565				break;
6566			}
6567		}
6568	}
6569
6570	/*
6571	 * If we couldn't find the page, or if we don't have a mode select
6572	 * handler for it, send back an error to the user.
6573	 */
6574	if ((page_index == NULL)
6575	 || (page_index->select_handler == NULL)) {
6576		ctl_set_invalid_field(ctsio,
6577				      /*sks_valid*/ 1,
6578				      /*command*/ 0,
6579				      /*field*/ *len_used,
6580				      /*bit_valid*/ 0,
6581				      /*bit*/ 0);
6582		free(ctsio->kern_data_ptr, M_CTL);
6583		ctl_done((union ctl_io *)ctsio);
6584		return (CTL_RETVAL_COMPLETE);
6585	}
6586
6587	if (page_index->page_code & SMPH_SPF) {
6588		page_len_offset = 2;
6589		page_len_size = 2;
6590	} else {
6591		page_len_size = 1;
6592		page_len_offset = 1;
6593	}
6594
6595	/*
6596	 * If the length the initiator gives us isn't the one we specify in
6597	 * the mode page header, or if they didn't specify enough data in
6598	 * the CDB to avoid truncating this page, kick out the request.
6599	 */
6600	if ((page_len != (page_index->page_len - page_len_offset -
6601			  page_len_size))
6602	 || (*len_left < page_index->page_len)) {
6603
6604
6605		ctl_set_invalid_field(ctsio,
6606				      /*sks_valid*/ 1,
6607				      /*command*/ 0,
6608				      /*field*/ *len_used + page_len_offset,
6609				      /*bit_valid*/ 0,
6610				      /*bit*/ 0);
6611		free(ctsio->kern_data_ptr, M_CTL);
6612		ctl_done((union ctl_io *)ctsio);
6613		return (CTL_RETVAL_COMPLETE);
6614	}
6615
6616	/*
6617	 * Run through the mode page, checking to make sure that the bits
6618	 * the user changed are actually legal for him to change.
6619	 */
6620	for (i = 0; i < page_index->page_len; i++) {
6621		uint8_t *user_byte, *change_mask, *current_byte;
6622		int bad_bit;
6623		int j;
6624
6625		user_byte = (uint8_t *)page_header + i;
6626		change_mask = page_index->page_data +
6627			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6628		current_byte = page_index->page_data +
6629			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6630
6631		/*
6632		 * Check to see whether the user set any bits in this byte
6633		 * that he is not allowed to set.
6634		 */
6635		if ((*user_byte & ~(*change_mask)) ==
6636		    (*current_byte & ~(*change_mask)))
6637			continue;
6638
6639		/*
6640		 * Go through bit by bit to determine which one is illegal.
6641		 */
6642		bad_bit = 0;
6643		for (j = 7; j >= 0; j--) {
6644			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6645			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6646				bad_bit = i;
6647				break;
6648			}
6649		}
6650		ctl_set_invalid_field(ctsio,
6651				      /*sks_valid*/ 1,
6652				      /*command*/ 0,
6653				      /*field*/ *len_used + i,
6654				      /*bit_valid*/ 1,
6655				      /*bit*/ bad_bit);
6656		free(ctsio->kern_data_ptr, M_CTL);
6657		ctl_done((union ctl_io *)ctsio);
6658		return (CTL_RETVAL_COMPLETE);
6659	}
6660
6661	/*
6662	 * Decrement these before we call the page handler, since we may
6663	 * end up getting called back one way or another before the handler
6664	 * returns to this context.
6665	 */
6666	*len_left -= page_index->page_len;
6667	*len_used += page_index->page_len;
6668
6669	retval = page_index->select_handler(ctsio, page_index,
6670					    (uint8_t *)page_header);
6671
6672	/*
6673	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6674	 * wait until this queued command completes to finish processing
6675	 * the mode page.  If it returns anything other than
6676	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6677	 * already set the sense information, freed the data pointer, and
6678	 * completed the io for us.
6679	 */
6680	if (retval != CTL_RETVAL_COMPLETE)
6681		goto bailout_no_done;
6682
6683	/*
6684	 * If the initiator sent us more than one page, parse the next one.
6685	 */
6686	if (*len_left > 0)
6687		goto do_next_page;
6688
6689	ctl_set_success(ctsio);
6690	free(ctsio->kern_data_ptr, M_CTL);
6691	ctl_done((union ctl_io *)ctsio);
6692
6693bailout_no_done:
6694
6695	return (CTL_RETVAL_COMPLETE);
6696
6697}
6698
6699int
6700ctl_mode_select(struct ctl_scsiio *ctsio)
6701{
6702	int param_len, pf, sp;
6703	int header_size, bd_len;
6704	int len_left, len_used;
6705	struct ctl_page_index *page_index;
6706	struct ctl_lun *lun;
6707	int control_dev, page_len;
6708	union ctl_modepage_info *modepage_info;
6709	int retval;
6710
6711	pf = 0;
6712	sp = 0;
6713	page_len = 0;
6714	len_used = 0;
6715	len_left = 0;
6716	retval = 0;
6717	bd_len = 0;
6718	page_index = NULL;
6719
6720	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6721
6722	if (lun->be_lun->lun_type != T_DIRECT)
6723		control_dev = 1;
6724	else
6725		control_dev = 0;
6726
6727	switch (ctsio->cdb[0]) {
6728	case MODE_SELECT_6: {
6729		struct scsi_mode_select_6 *cdb;
6730
6731		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6732
6733		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6734		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6735
6736		param_len = cdb->length;
6737		header_size = sizeof(struct scsi_mode_header_6);
6738		break;
6739	}
6740	case MODE_SELECT_10: {
6741		struct scsi_mode_select_10 *cdb;
6742
6743		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6744
6745		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6746		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6747
6748		param_len = scsi_2btoul(cdb->length);
6749		header_size = sizeof(struct scsi_mode_header_10);
6750		break;
6751	}
6752	default:
6753		ctl_set_invalid_opcode(ctsio);
6754		ctl_done((union ctl_io *)ctsio);
6755		return (CTL_RETVAL_COMPLETE);
6756		break; /* NOTREACHED */
6757	}
6758
6759	/*
6760	 * From SPC-3:
6761	 * "A parameter list length of zero indicates that the Data-Out Buffer
6762	 * shall be empty. This condition shall not be considered as an error."
6763	 */
6764	if (param_len == 0) {
6765		ctl_set_success(ctsio);
6766		ctl_done((union ctl_io *)ctsio);
6767		return (CTL_RETVAL_COMPLETE);
6768	}
6769
6770	/*
6771	 * Since we'll hit this the first time through, prior to
6772	 * allocation, we don't need to free a data buffer here.
6773	 */
6774	if (param_len < header_size) {
6775		ctl_set_param_len_error(ctsio);
6776		ctl_done((union ctl_io *)ctsio);
6777		return (CTL_RETVAL_COMPLETE);
6778	}
6779
6780	/*
6781	 * Allocate the data buffer and grab the user's data.  In theory,
6782	 * we shouldn't have to sanity check the parameter list length here
6783	 * because the maximum size is 64K.  We should be able to malloc
6784	 * that much without too many problems.
6785	 */
6786	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6787		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6788		ctsio->kern_data_len = param_len;
6789		ctsio->kern_total_len = param_len;
6790		ctsio->kern_data_resid = 0;
6791		ctsio->kern_rel_offset = 0;
6792		ctsio->kern_sg_entries = 0;
6793		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6794		ctsio->be_move_done = ctl_config_move_done;
6795		ctl_datamove((union ctl_io *)ctsio);
6796
6797		return (CTL_RETVAL_COMPLETE);
6798	}
6799
6800	switch (ctsio->cdb[0]) {
6801	case MODE_SELECT_6: {
6802		struct scsi_mode_header_6 *mh6;
6803
6804		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6805		bd_len = mh6->blk_desc_len;
6806		break;
6807	}
6808	case MODE_SELECT_10: {
6809		struct scsi_mode_header_10 *mh10;
6810
6811		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6812		bd_len = scsi_2btoul(mh10->blk_desc_len);
6813		break;
6814	}
6815	default:
6816		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6817		break;
6818	}
6819
6820	if (param_len < (header_size + bd_len)) {
6821		free(ctsio->kern_data_ptr, M_CTL);
6822		ctl_set_param_len_error(ctsio);
6823		ctl_done((union ctl_io *)ctsio);
6824		return (CTL_RETVAL_COMPLETE);
6825	}
6826
6827	/*
6828	 * Set the IO_CONT flag, so that if this I/O gets passed to
6829	 * ctl_config_write_done(), it'll get passed back to
6830	 * ctl_do_mode_select() for further processing, or completion if
6831	 * we're all done.
6832	 */
6833	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6834	ctsio->io_cont = ctl_do_mode_select;
6835
6836	modepage_info = (union ctl_modepage_info *)
6837		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6838
6839	memset(modepage_info, 0, sizeof(*modepage_info));
6840
6841	len_left = param_len - header_size - bd_len;
6842	len_used = header_size + bd_len;
6843
6844	modepage_info->header.len_left = len_left;
6845	modepage_info->header.len_used = len_used;
6846
6847	return (ctl_do_mode_select((union ctl_io *)ctsio));
6848}
6849
6850int
6851ctl_mode_sense(struct ctl_scsiio *ctsio)
6852{
6853	struct ctl_lun *lun;
6854	int pc, page_code, dbd, llba, subpage;
6855	int alloc_len, page_len, header_len, total_len;
6856	struct scsi_mode_block_descr *block_desc;
6857	struct ctl_page_index *page_index;
6858	int control_dev;
6859
6860	dbd = 0;
6861	llba = 0;
6862	block_desc = NULL;
6863	page_index = NULL;
6864
6865	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6866
6867	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6868
6869	if (lun->be_lun->lun_type != T_DIRECT)
6870		control_dev = 1;
6871	else
6872		control_dev = 0;
6873
6874	if (lun->flags & CTL_LUN_PR_RESERVED) {
6875		uint32_t residx;
6876
6877		/*
6878		 * XXX KDM need a lock here.
6879		 */
6880		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6881		if ((lun->res_type == SPR_TYPE_EX_AC
6882		  && residx != lun->pr_res_idx)
6883		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6884		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6885		  && !lun->per_res[residx].registered)) {
6886			ctl_set_reservation_conflict(ctsio);
6887			ctl_done((union ctl_io *)ctsio);
6888			return (CTL_RETVAL_COMPLETE);
6889		}
6890	}
6891
6892	switch (ctsio->cdb[0]) {
6893	case MODE_SENSE_6: {
6894		struct scsi_mode_sense_6 *cdb;
6895
6896		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6897
6898		header_len = sizeof(struct scsi_mode_hdr_6);
6899		if (cdb->byte2 & SMS_DBD)
6900			dbd = 1;
6901		else
6902			header_len += sizeof(struct scsi_mode_block_descr);
6903
6904		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6905		page_code = cdb->page & SMS_PAGE_CODE;
6906		subpage = cdb->subpage;
6907		alloc_len = cdb->length;
6908		break;
6909	}
6910	case MODE_SENSE_10: {
6911		struct scsi_mode_sense_10 *cdb;
6912
6913		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6914
6915		header_len = sizeof(struct scsi_mode_hdr_10);
6916
6917		if (cdb->byte2 & SMS_DBD)
6918			dbd = 1;
6919		else
6920			header_len += sizeof(struct scsi_mode_block_descr);
6921		if (cdb->byte2 & SMS10_LLBAA)
6922			llba = 1;
6923		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6924		page_code = cdb->page & SMS_PAGE_CODE;
6925		subpage = cdb->subpage;
6926		alloc_len = scsi_2btoul(cdb->length);
6927		break;
6928	}
6929	default:
6930		ctl_set_invalid_opcode(ctsio);
6931		ctl_done((union ctl_io *)ctsio);
6932		return (CTL_RETVAL_COMPLETE);
6933		break; /* NOTREACHED */
6934	}
6935
6936	/*
6937	 * We have to make a first pass through to calculate the size of
6938	 * the pages that match the user's query.  Then we allocate enough
6939	 * memory to hold it, and actually copy the data into the buffer.
6940	 */
6941	switch (page_code) {
6942	case SMS_ALL_PAGES_PAGE: {
6943		int i;
6944
6945		page_len = 0;
6946
6947		/*
6948		 * At the moment, values other than 0 and 0xff here are
6949		 * reserved according to SPC-3.
6950		 */
6951		if ((subpage != SMS_SUBPAGE_PAGE_0)
6952		 && (subpage != SMS_SUBPAGE_ALL)) {
6953			ctl_set_invalid_field(ctsio,
6954					      /*sks_valid*/ 1,
6955					      /*command*/ 1,
6956					      /*field*/ 3,
6957					      /*bit_valid*/ 0,
6958					      /*bit*/ 0);
6959			ctl_done((union ctl_io *)ctsio);
6960			return (CTL_RETVAL_COMPLETE);
6961		}
6962
6963		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6964			if ((control_dev != 0)
6965			 && (lun->mode_pages.index[i].page_flags &
6966			     CTL_PAGE_FLAG_DISK_ONLY))
6967				continue;
6968
6969			/*
6970			 * We don't use this subpage if the user didn't
6971			 * request all subpages.
6972			 */
6973			if ((lun->mode_pages.index[i].subpage != 0)
6974			 && (subpage == SMS_SUBPAGE_PAGE_0))
6975				continue;
6976
6977#if 0
6978			printf("found page %#x len %d\n",
6979			       lun->mode_pages.index[i].page_code &
6980			       SMPH_PC_MASK,
6981			       lun->mode_pages.index[i].page_len);
6982#endif
6983			page_len += lun->mode_pages.index[i].page_len;
6984		}
6985		break;
6986	}
6987	default: {
6988		int i;
6989
6990		page_len = 0;
6991
6992		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6993			/* Look for the right page code */
6994			if ((lun->mode_pages.index[i].page_code &
6995			     SMPH_PC_MASK) != page_code)
6996				continue;
6997
6998			/* Look for the right subpage or the subpage wildcard*/
6999			if ((lun->mode_pages.index[i].subpage != subpage)
7000			 && (subpage != SMS_SUBPAGE_ALL))
7001				continue;
7002
7003			/* Make sure the page is supported for this dev type */
7004			if ((control_dev != 0)
7005			 && (lun->mode_pages.index[i].page_flags &
7006			     CTL_PAGE_FLAG_DISK_ONLY))
7007				continue;
7008
7009#if 0
7010			printf("found page %#x len %d\n",
7011			       lun->mode_pages.index[i].page_code &
7012			       SMPH_PC_MASK,
7013			       lun->mode_pages.index[i].page_len);
7014#endif
7015
7016			page_len += lun->mode_pages.index[i].page_len;
7017		}
7018
7019		if (page_len == 0) {
7020			ctl_set_invalid_field(ctsio,
7021					      /*sks_valid*/ 1,
7022					      /*command*/ 1,
7023					      /*field*/ 2,
7024					      /*bit_valid*/ 1,
7025					      /*bit*/ 5);
7026			ctl_done((union ctl_io *)ctsio);
7027			return (CTL_RETVAL_COMPLETE);
7028		}
7029		break;
7030	}
7031	}
7032
7033	total_len = header_len + page_len;
7034#if 0
7035	printf("header_len = %d, page_len = %d, total_len = %d\n",
7036	       header_len, page_len, total_len);
7037#endif
7038
7039	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7040	ctsio->kern_sg_entries = 0;
7041	ctsio->kern_data_resid = 0;
7042	ctsio->kern_rel_offset = 0;
7043	if (total_len < alloc_len) {
7044		ctsio->residual = alloc_len - total_len;
7045		ctsio->kern_data_len = total_len;
7046		ctsio->kern_total_len = total_len;
7047	} else {
7048		ctsio->residual = 0;
7049		ctsio->kern_data_len = alloc_len;
7050		ctsio->kern_total_len = alloc_len;
7051	}
7052
7053	switch (ctsio->cdb[0]) {
7054	case MODE_SENSE_6: {
7055		struct scsi_mode_hdr_6 *header;
7056
7057		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7058
7059		header->datalen = ctl_min(total_len - 1, 254);
7060		if (control_dev == 0) {
7061			header->dev_specific = 0x10; /* DPOFUA */
7062			if ((lun->flags & CTL_LUN_READONLY) ||
7063			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7064			    .eca_and_aen & SCP_SWP) != 0)
7065				    header->dev_specific |= 0x80; /* WP */
7066		}
7067		if (dbd)
7068			header->block_descr_len = 0;
7069		else
7070			header->block_descr_len =
7071				sizeof(struct scsi_mode_block_descr);
7072		block_desc = (struct scsi_mode_block_descr *)&header[1];
7073		break;
7074	}
7075	case MODE_SENSE_10: {
7076		struct scsi_mode_hdr_10 *header;
7077		int datalen;
7078
7079		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7080
7081		datalen = ctl_min(total_len - 2, 65533);
7082		scsi_ulto2b(datalen, header->datalen);
7083		if (control_dev == 0) {
7084			header->dev_specific = 0x10; /* DPOFUA */
7085			if ((lun->flags & CTL_LUN_READONLY) ||
7086			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
7087			    .eca_and_aen & SCP_SWP) != 0)
7088				    header->dev_specific |= 0x80; /* WP */
7089		}
7090		if (dbd)
7091			scsi_ulto2b(0, header->block_descr_len);
7092		else
7093			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7094				    header->block_descr_len);
7095		block_desc = (struct scsi_mode_block_descr *)&header[1];
7096		break;
7097	}
7098	default:
7099		panic("invalid CDB type %#x", ctsio->cdb[0]);
7100		break; /* NOTREACHED */
7101	}
7102
7103	/*
7104	 * If we've got a disk, use its blocksize in the block
7105	 * descriptor.  Otherwise, just set it to 0.
7106	 */
7107	if (dbd == 0) {
7108		if (control_dev == 0)
7109			scsi_ulto3b(lun->be_lun->blocksize,
7110				    block_desc->block_len);
7111		else
7112			scsi_ulto3b(0, block_desc->block_len);
7113	}
7114
7115	switch (page_code) {
7116	case SMS_ALL_PAGES_PAGE: {
7117		int i, data_used;
7118
7119		data_used = header_len;
7120		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7121			struct ctl_page_index *page_index;
7122
7123			page_index = &lun->mode_pages.index[i];
7124
7125			if ((control_dev != 0)
7126			 && (page_index->page_flags &
7127			    CTL_PAGE_FLAG_DISK_ONLY))
7128				continue;
7129
7130			/*
7131			 * We don't use this subpage if the user didn't
7132			 * request all subpages.  We already checked (above)
7133			 * to make sure the user only specified a subpage
7134			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7135			 */
7136			if ((page_index->subpage != 0)
7137			 && (subpage == SMS_SUBPAGE_PAGE_0))
7138				continue;
7139
7140			/*
7141			 * Call the handler, if it exists, to update the
7142			 * page to the latest values.
7143			 */
7144			if (page_index->sense_handler != NULL)
7145				page_index->sense_handler(ctsio, page_index,pc);
7146
7147			memcpy(ctsio->kern_data_ptr + data_used,
7148			       page_index->page_data +
7149			       (page_index->page_len * pc),
7150			       page_index->page_len);
7151			data_used += page_index->page_len;
7152		}
7153		break;
7154	}
7155	default: {
7156		int i, data_used;
7157
7158		data_used = header_len;
7159
7160		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7161			struct ctl_page_index *page_index;
7162
7163			page_index = &lun->mode_pages.index[i];
7164
7165			/* Look for the right page code */
7166			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7167				continue;
7168
7169			/* Look for the right subpage or the subpage wildcard*/
7170			if ((page_index->subpage != subpage)
7171			 && (subpage != SMS_SUBPAGE_ALL))
7172				continue;
7173
7174			/* Make sure the page is supported for this dev type */
7175			if ((control_dev != 0)
7176			 && (page_index->page_flags &
7177			     CTL_PAGE_FLAG_DISK_ONLY))
7178				continue;
7179
7180			/*
7181			 * Call the handler, if it exists, to update the
7182			 * page to the latest values.
7183			 */
7184			if (page_index->sense_handler != NULL)
7185				page_index->sense_handler(ctsio, page_index,pc);
7186
7187			memcpy(ctsio->kern_data_ptr + data_used,
7188			       page_index->page_data +
7189			       (page_index->page_len * pc),
7190			       page_index->page_len);
7191			data_used += page_index->page_len;
7192		}
7193		break;
7194	}
7195	}
7196
7197	ctsio->scsi_status = SCSI_STATUS_OK;
7198
7199	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7200	ctsio->be_move_done = ctl_config_move_done;
7201	ctl_datamove((union ctl_io *)ctsio);
7202
7203	return (CTL_RETVAL_COMPLETE);
7204}
7205
7206int
7207ctl_read_capacity(struct ctl_scsiio *ctsio)
7208{
7209	struct scsi_read_capacity *cdb;
7210	struct scsi_read_capacity_data *data;
7211	struct ctl_lun *lun;
7212	uint32_t lba;
7213
7214	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7215
7216	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7217
7218	lba = scsi_4btoul(cdb->addr);
7219	if (((cdb->pmi & SRC_PMI) == 0)
7220	 && (lba != 0)) {
7221		ctl_set_invalid_field(/*ctsio*/ ctsio,
7222				      /*sks_valid*/ 1,
7223				      /*command*/ 1,
7224				      /*field*/ 2,
7225				      /*bit_valid*/ 0,
7226				      /*bit*/ 0);
7227		ctl_done((union ctl_io *)ctsio);
7228		return (CTL_RETVAL_COMPLETE);
7229	}
7230
7231	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7232
7233	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7234	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7235	ctsio->residual = 0;
7236	ctsio->kern_data_len = sizeof(*data);
7237	ctsio->kern_total_len = sizeof(*data);
7238	ctsio->kern_data_resid = 0;
7239	ctsio->kern_rel_offset = 0;
7240	ctsio->kern_sg_entries = 0;
7241
7242	/*
7243	 * If the maximum LBA is greater than 0xfffffffe, the user must
7244	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7245	 * serivce action set.
7246	 */
7247	if (lun->be_lun->maxlba > 0xfffffffe)
7248		scsi_ulto4b(0xffffffff, data->addr);
7249	else
7250		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7251
7252	/*
7253	 * XXX KDM this may not be 512 bytes...
7254	 */
7255	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7256
7257	ctsio->scsi_status = SCSI_STATUS_OK;
7258
7259	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7260	ctsio->be_move_done = ctl_config_move_done;
7261	ctl_datamove((union ctl_io *)ctsio);
7262
7263	return (CTL_RETVAL_COMPLETE);
7264}
7265
7266int
7267ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7268{
7269	struct scsi_read_capacity_16 *cdb;
7270	struct scsi_read_capacity_data_long *data;
7271	struct ctl_lun *lun;
7272	uint64_t lba;
7273	uint32_t alloc_len;
7274
7275	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7276
7277	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7278
7279	alloc_len = scsi_4btoul(cdb->alloc_len);
7280	lba = scsi_8btou64(cdb->addr);
7281
7282	if ((cdb->reladr & SRC16_PMI)
7283	 && (lba != 0)) {
7284		ctl_set_invalid_field(/*ctsio*/ ctsio,
7285				      /*sks_valid*/ 1,
7286				      /*command*/ 1,
7287				      /*field*/ 2,
7288				      /*bit_valid*/ 0,
7289				      /*bit*/ 0);
7290		ctl_done((union ctl_io *)ctsio);
7291		return (CTL_RETVAL_COMPLETE);
7292	}
7293
7294	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7295
7296	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7297	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7298
7299	if (sizeof(*data) < alloc_len) {
7300		ctsio->residual = alloc_len - sizeof(*data);
7301		ctsio->kern_data_len = sizeof(*data);
7302		ctsio->kern_total_len = sizeof(*data);
7303	} else {
7304		ctsio->residual = 0;
7305		ctsio->kern_data_len = alloc_len;
7306		ctsio->kern_total_len = alloc_len;
7307	}
7308	ctsio->kern_data_resid = 0;
7309	ctsio->kern_rel_offset = 0;
7310	ctsio->kern_sg_entries = 0;
7311
7312	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7313	/* XXX KDM this may not be 512 bytes... */
7314	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7315	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7316	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7317	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7318		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7319
7320	ctsio->scsi_status = SCSI_STATUS_OK;
7321
7322	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7323	ctsio->be_move_done = ctl_config_move_done;
7324	ctl_datamove((union ctl_io *)ctsio);
7325
7326	return (CTL_RETVAL_COMPLETE);
7327}
7328
7329int
7330ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7331{
7332	struct scsi_maintenance_in *cdb;
7333	int retval;
7334	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7335	int num_target_port_groups, num_target_ports, single;
7336	struct ctl_lun *lun;
7337	struct ctl_softc *softc;
7338	struct ctl_port *port;
7339	struct scsi_target_group_data *rtg_ptr;
7340	struct scsi_target_group_data_extended *rtg_ext_ptr;
7341	struct scsi_target_port_group_descriptor *tpg_desc;
7342
7343	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7344
7345	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7346	softc = control_softc;
7347	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7348
7349	retval = CTL_RETVAL_COMPLETE;
7350
7351	switch (cdb->byte2 & STG_PDF_MASK) {
7352	case STG_PDF_LENGTH:
7353		ext = 0;
7354		break;
7355	case STG_PDF_EXTENDED:
7356		ext = 1;
7357		break;
7358	default:
7359		ctl_set_invalid_field(/*ctsio*/ ctsio,
7360				      /*sks_valid*/ 1,
7361				      /*command*/ 1,
7362				      /*field*/ 2,
7363				      /*bit_valid*/ 1,
7364				      /*bit*/ 5);
7365		ctl_done((union ctl_io *)ctsio);
7366		return(retval);
7367	}
7368
7369	single = ctl_is_single;
7370	if (single)
7371		num_target_port_groups = 1;
7372	else
7373		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7374	num_target_ports = 0;
7375	mtx_lock(&softc->ctl_lock);
7376	STAILQ_FOREACH(port, &softc->port_list, links) {
7377		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7378			continue;
7379		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7380			continue;
7381		num_target_ports++;
7382	}
7383	mtx_unlock(&softc->ctl_lock);
7384
7385	if (ext)
7386		total_len = sizeof(struct scsi_target_group_data_extended);
7387	else
7388		total_len = sizeof(struct scsi_target_group_data);
7389	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7390		num_target_port_groups +
7391	    sizeof(struct scsi_target_port_descriptor) *
7392		num_target_ports * num_target_port_groups;
7393
7394	alloc_len = scsi_4btoul(cdb->length);
7395
7396	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7397
7398	ctsio->kern_sg_entries = 0;
7399
7400	if (total_len < alloc_len) {
7401		ctsio->residual = alloc_len - total_len;
7402		ctsio->kern_data_len = total_len;
7403		ctsio->kern_total_len = total_len;
7404	} else {
7405		ctsio->residual = 0;
7406		ctsio->kern_data_len = alloc_len;
7407		ctsio->kern_total_len = alloc_len;
7408	}
7409	ctsio->kern_data_resid = 0;
7410	ctsio->kern_rel_offset = 0;
7411
7412	if (ext) {
7413		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7414		    ctsio->kern_data_ptr;
7415		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7416		rtg_ext_ptr->format_type = 0x10;
7417		rtg_ext_ptr->implicit_transition_time = 0;
7418		tpg_desc = &rtg_ext_ptr->groups[0];
7419	} else {
7420		rtg_ptr = (struct scsi_target_group_data *)
7421		    ctsio->kern_data_ptr;
7422		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7423		tpg_desc = &rtg_ptr->groups[0];
7424	}
7425
7426	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7427	mtx_lock(&softc->ctl_lock);
7428	for (g = 0; g < num_target_port_groups; g++) {
7429		if (g == pg)
7430			tpg_desc->pref_state = TPG_PRIMARY |
7431			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7432		else
7433			tpg_desc->pref_state =
7434			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7435		tpg_desc->support = TPG_AO_SUP;
7436		if (!single)
7437			tpg_desc->support |= TPG_AN_SUP;
7438		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7439		tpg_desc->status = TPG_IMPLICIT;
7440		pc = 0;
7441		STAILQ_FOREACH(port, &softc->port_list, links) {
7442			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7443				continue;
7444			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7445			    CTL_MAX_LUNS)
7446				continue;
7447			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7448			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7449			    relative_target_port_identifier);
7450			pc++;
7451		}
7452		tpg_desc->target_port_count = pc;
7453		tpg_desc = (struct scsi_target_port_group_descriptor *)
7454		    &tpg_desc->descriptors[pc];
7455	}
7456	mtx_unlock(&softc->ctl_lock);
7457
7458	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7459	ctsio->be_move_done = ctl_config_move_done;
7460
7461	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7462			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7463			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7464			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7465			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7466
7467	ctl_datamove((union ctl_io *)ctsio);
7468	return(retval);
7469}
7470
7471int
7472ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7473{
7474	struct ctl_lun *lun;
7475	struct scsi_report_supported_opcodes *cdb;
7476	const struct ctl_cmd_entry *entry, *sentry;
7477	struct scsi_report_supported_opcodes_all *all;
7478	struct scsi_report_supported_opcodes_descr *descr;
7479	struct scsi_report_supported_opcodes_one *one;
7480	int retval;
7481	int alloc_len, total_len;
7482	int opcode, service_action, i, j, num;
7483
7484	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7485
7486	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7487	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7488
7489	retval = CTL_RETVAL_COMPLETE;
7490
7491	opcode = cdb->requested_opcode;
7492	service_action = scsi_2btoul(cdb->requested_service_action);
7493	switch (cdb->options & RSO_OPTIONS_MASK) {
7494	case RSO_OPTIONS_ALL:
7495		num = 0;
7496		for (i = 0; i < 256; i++) {
7497			entry = &ctl_cmd_table[i];
7498			if (entry->flags & CTL_CMD_FLAG_SA5) {
7499				for (j = 0; j < 32; j++) {
7500					sentry = &((const struct ctl_cmd_entry *)
7501					    entry->execute)[j];
7502					if (ctl_cmd_applicable(
7503					    lun->be_lun->lun_type, sentry))
7504						num++;
7505				}
7506			} else {
7507				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7508				    entry))
7509					num++;
7510			}
7511		}
7512		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7513		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7514		break;
7515	case RSO_OPTIONS_OC:
7516		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7517			ctl_set_invalid_field(/*ctsio*/ ctsio,
7518					      /*sks_valid*/ 1,
7519					      /*command*/ 1,
7520					      /*field*/ 2,
7521					      /*bit_valid*/ 1,
7522					      /*bit*/ 2);
7523			ctl_done((union ctl_io *)ctsio);
7524			return (CTL_RETVAL_COMPLETE);
7525		}
7526		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7527		break;
7528	case RSO_OPTIONS_OC_SA:
7529		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7530		    service_action >= 32) {
7531			ctl_set_invalid_field(/*ctsio*/ ctsio,
7532					      /*sks_valid*/ 1,
7533					      /*command*/ 1,
7534					      /*field*/ 2,
7535					      /*bit_valid*/ 1,
7536					      /*bit*/ 2);
7537			ctl_done((union ctl_io *)ctsio);
7538			return (CTL_RETVAL_COMPLETE);
7539		}
7540		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7541		break;
7542	default:
7543		ctl_set_invalid_field(/*ctsio*/ ctsio,
7544				      /*sks_valid*/ 1,
7545				      /*command*/ 1,
7546				      /*field*/ 2,
7547				      /*bit_valid*/ 1,
7548				      /*bit*/ 2);
7549		ctl_done((union ctl_io *)ctsio);
7550		return (CTL_RETVAL_COMPLETE);
7551	}
7552
7553	alloc_len = scsi_4btoul(cdb->length);
7554
7555	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7556
7557	ctsio->kern_sg_entries = 0;
7558
7559	if (total_len < alloc_len) {
7560		ctsio->residual = alloc_len - total_len;
7561		ctsio->kern_data_len = total_len;
7562		ctsio->kern_total_len = total_len;
7563	} else {
7564		ctsio->residual = 0;
7565		ctsio->kern_data_len = alloc_len;
7566		ctsio->kern_total_len = alloc_len;
7567	}
7568	ctsio->kern_data_resid = 0;
7569	ctsio->kern_rel_offset = 0;
7570
7571	switch (cdb->options & RSO_OPTIONS_MASK) {
7572	case RSO_OPTIONS_ALL:
7573		all = (struct scsi_report_supported_opcodes_all *)
7574		    ctsio->kern_data_ptr;
7575		num = 0;
7576		for (i = 0; i < 256; i++) {
7577			entry = &ctl_cmd_table[i];
7578			if (entry->flags & CTL_CMD_FLAG_SA5) {
7579				for (j = 0; j < 32; j++) {
7580					sentry = &((const struct ctl_cmd_entry *)
7581					    entry->execute)[j];
7582					if (!ctl_cmd_applicable(
7583					    lun->be_lun->lun_type, sentry))
7584						continue;
7585					descr = &all->descr[num++];
7586					descr->opcode = i;
7587					scsi_ulto2b(j, descr->service_action);
7588					descr->flags = RSO_SERVACTV;
7589					scsi_ulto2b(sentry->length,
7590					    descr->cdb_length);
7591				}
7592			} else {
7593				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7594				    entry))
7595					continue;
7596				descr = &all->descr[num++];
7597				descr->opcode = i;
7598				scsi_ulto2b(0, descr->service_action);
7599				descr->flags = 0;
7600				scsi_ulto2b(entry->length, descr->cdb_length);
7601			}
7602		}
7603		scsi_ulto4b(
7604		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7605		    all->length);
7606		break;
7607	case RSO_OPTIONS_OC:
7608		one = (struct scsi_report_supported_opcodes_one *)
7609		    ctsio->kern_data_ptr;
7610		entry = &ctl_cmd_table[opcode];
7611		goto fill_one;
7612	case RSO_OPTIONS_OC_SA:
7613		one = (struct scsi_report_supported_opcodes_one *)
7614		    ctsio->kern_data_ptr;
7615		entry = &ctl_cmd_table[opcode];
7616		entry = &((const struct ctl_cmd_entry *)
7617		    entry->execute)[service_action];
7618fill_one:
7619		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7620			one->support = 3;
7621			scsi_ulto2b(entry->length, one->cdb_length);
7622			one->cdb_usage[0] = opcode;
7623			memcpy(&one->cdb_usage[1], entry->usage,
7624			    entry->length - 1);
7625		} else
7626			one->support = 1;
7627		break;
7628	}
7629
7630	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7631	ctsio->be_move_done = ctl_config_move_done;
7632
7633	ctl_datamove((union ctl_io *)ctsio);
7634	return(retval);
7635}
7636
7637int
7638ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7639{
7640	struct ctl_lun *lun;
7641	struct scsi_report_supported_tmf *cdb;
7642	struct scsi_report_supported_tmf_data *data;
7643	int retval;
7644	int alloc_len, total_len;
7645
7646	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7647
7648	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7649	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7650
7651	retval = CTL_RETVAL_COMPLETE;
7652
7653	total_len = sizeof(struct scsi_report_supported_tmf_data);
7654	alloc_len = scsi_4btoul(cdb->length);
7655
7656	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7657
7658	ctsio->kern_sg_entries = 0;
7659
7660	if (total_len < alloc_len) {
7661		ctsio->residual = alloc_len - total_len;
7662		ctsio->kern_data_len = total_len;
7663		ctsio->kern_total_len = total_len;
7664	} else {
7665		ctsio->residual = 0;
7666		ctsio->kern_data_len = alloc_len;
7667		ctsio->kern_total_len = alloc_len;
7668	}
7669	ctsio->kern_data_resid = 0;
7670	ctsio->kern_rel_offset = 0;
7671
7672	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7673	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7674	data->byte2 |= RST_ITNRS;
7675
7676	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7677	ctsio->be_move_done = ctl_config_move_done;
7678
7679	ctl_datamove((union ctl_io *)ctsio);
7680	return (retval);
7681}
7682
7683int
7684ctl_report_timestamp(struct ctl_scsiio *ctsio)
7685{
7686	struct ctl_lun *lun;
7687	struct scsi_report_timestamp *cdb;
7688	struct scsi_report_timestamp_data *data;
7689	struct timeval tv;
7690	int64_t timestamp;
7691	int retval;
7692	int alloc_len, total_len;
7693
7694	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7695
7696	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7697	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7698
7699	retval = CTL_RETVAL_COMPLETE;
7700
7701	total_len = sizeof(struct scsi_report_timestamp_data);
7702	alloc_len = scsi_4btoul(cdb->length);
7703
7704	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7705
7706	ctsio->kern_sg_entries = 0;
7707
7708	if (total_len < alloc_len) {
7709		ctsio->residual = alloc_len - total_len;
7710		ctsio->kern_data_len = total_len;
7711		ctsio->kern_total_len = total_len;
7712	} else {
7713		ctsio->residual = 0;
7714		ctsio->kern_data_len = alloc_len;
7715		ctsio->kern_total_len = alloc_len;
7716	}
7717	ctsio->kern_data_resid = 0;
7718	ctsio->kern_rel_offset = 0;
7719
7720	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7721	scsi_ulto2b(sizeof(*data) - 2, data->length);
7722	data->origin = RTS_ORIG_OUTSIDE;
7723	getmicrotime(&tv);
7724	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7725	scsi_ulto4b(timestamp >> 16, data->timestamp);
7726	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7727
7728	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7729	ctsio->be_move_done = ctl_config_move_done;
7730
7731	ctl_datamove((union ctl_io *)ctsio);
7732	return (retval);
7733}
7734
7735int
7736ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7737{
7738	struct scsi_per_res_in *cdb;
7739	int alloc_len, total_len = 0;
7740	/* struct scsi_per_res_in_rsrv in_data; */
7741	struct ctl_lun *lun;
7742	struct ctl_softc *softc;
7743
7744	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7745
7746	softc = control_softc;
7747
7748	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7749
7750	alloc_len = scsi_2btoul(cdb->length);
7751
7752	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7753
7754retry:
7755	mtx_lock(&lun->lun_lock);
7756	switch (cdb->action) {
7757	case SPRI_RK: /* read keys */
7758		total_len = sizeof(struct scsi_per_res_in_keys) +
7759			lun->pr_key_count *
7760			sizeof(struct scsi_per_res_key);
7761		break;
7762	case SPRI_RR: /* read reservation */
7763		if (lun->flags & CTL_LUN_PR_RESERVED)
7764			total_len = sizeof(struct scsi_per_res_in_rsrv);
7765		else
7766			total_len = sizeof(struct scsi_per_res_in_header);
7767		break;
7768	case SPRI_RC: /* report capabilities */
7769		total_len = sizeof(struct scsi_per_res_cap);
7770		break;
7771	case SPRI_RS: /* read full status */
7772		total_len = sizeof(struct scsi_per_res_in_header) +
7773		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7774		    lun->pr_key_count;
7775		break;
7776	default:
7777		panic("Invalid PR type %x", cdb->action);
7778	}
7779	mtx_unlock(&lun->lun_lock);
7780
7781	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7782
7783	if (total_len < alloc_len) {
7784		ctsio->residual = alloc_len - total_len;
7785		ctsio->kern_data_len = total_len;
7786		ctsio->kern_total_len = total_len;
7787	} else {
7788		ctsio->residual = 0;
7789		ctsio->kern_data_len = alloc_len;
7790		ctsio->kern_total_len = alloc_len;
7791	}
7792
7793	ctsio->kern_data_resid = 0;
7794	ctsio->kern_rel_offset = 0;
7795	ctsio->kern_sg_entries = 0;
7796
7797	mtx_lock(&lun->lun_lock);
7798	switch (cdb->action) {
7799	case SPRI_RK: { // read keys
7800        struct scsi_per_res_in_keys *res_keys;
7801		int i, key_count;
7802
7803		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7804
7805		/*
7806		 * We had to drop the lock to allocate our buffer, which
7807		 * leaves time for someone to come in with another
7808		 * persistent reservation.  (That is unlikely, though,
7809		 * since this should be the only persistent reservation
7810		 * command active right now.)
7811		 */
7812		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7813		    (lun->pr_key_count *
7814		     sizeof(struct scsi_per_res_key)))){
7815			mtx_unlock(&lun->lun_lock);
7816			free(ctsio->kern_data_ptr, M_CTL);
7817			printf("%s: reservation length changed, retrying\n",
7818			       __func__);
7819			goto retry;
7820		}
7821
7822		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7823
7824		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7825			     lun->pr_key_count, res_keys->header.length);
7826
7827		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7828			if (!lun->per_res[i].registered)
7829				continue;
7830
7831			/*
7832			 * We used lun->pr_key_count to calculate the
7833			 * size to allocate.  If it turns out the number of
7834			 * initiators with the registered flag set is
7835			 * larger than that (i.e. they haven't been kept in
7836			 * sync), we've got a problem.
7837			 */
7838			if (key_count >= lun->pr_key_count) {
7839#ifdef NEEDTOPORT
7840				csevent_log(CSC_CTL | CSC_SHELF_SW |
7841					    CTL_PR_ERROR,
7842					    csevent_LogType_Fault,
7843					    csevent_AlertLevel_Yellow,
7844					    csevent_FRU_ShelfController,
7845					    csevent_FRU_Firmware,
7846				        csevent_FRU_Unknown,
7847					    "registered keys %d >= key "
7848					    "count %d", key_count,
7849					    lun->pr_key_count);
7850#endif
7851				key_count++;
7852				continue;
7853			}
7854			memcpy(res_keys->keys[key_count].key,
7855			       lun->per_res[i].res_key.key,
7856			       ctl_min(sizeof(res_keys->keys[key_count].key),
7857			       sizeof(lun->per_res[i].res_key)));
7858			key_count++;
7859		}
7860		break;
7861	}
7862	case SPRI_RR: { // read reservation
7863		struct scsi_per_res_in_rsrv *res;
7864		int tmp_len, header_only;
7865
7866		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7867
7868		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7869
7870		if (lun->flags & CTL_LUN_PR_RESERVED)
7871		{
7872			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7873			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7874				    res->header.length);
7875			header_only = 0;
7876		} else {
7877			tmp_len = sizeof(struct scsi_per_res_in_header);
7878			scsi_ulto4b(0, res->header.length);
7879			header_only = 1;
7880		}
7881
7882		/*
7883		 * We had to drop the lock to allocate our buffer, which
7884		 * leaves time for someone to come in with another
7885		 * persistent reservation.  (That is unlikely, though,
7886		 * since this should be the only persistent reservation
7887		 * command active right now.)
7888		 */
7889		if (tmp_len != total_len) {
7890			mtx_unlock(&lun->lun_lock);
7891			free(ctsio->kern_data_ptr, M_CTL);
7892			printf("%s: reservation status changed, retrying\n",
7893			       __func__);
7894			goto retry;
7895		}
7896
7897		/*
7898		 * No reservation held, so we're done.
7899		 */
7900		if (header_only != 0)
7901			break;
7902
7903		/*
7904		 * If the registration is an All Registrants type, the key
7905		 * is 0, since it doesn't really matter.
7906		 */
7907		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7908			memcpy(res->data.reservation,
7909			       &lun->per_res[lun->pr_res_idx].res_key,
7910			       sizeof(struct scsi_per_res_key));
7911		}
7912		res->data.scopetype = lun->res_type;
7913		break;
7914	}
7915	case SPRI_RC:     //report capabilities
7916	{
7917		struct scsi_per_res_cap *res_cap;
7918		uint16_t type_mask;
7919
7920		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7921		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7922		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7923		type_mask = SPRI_TM_WR_EX_AR |
7924			    SPRI_TM_EX_AC_RO |
7925			    SPRI_TM_WR_EX_RO |
7926			    SPRI_TM_EX_AC |
7927			    SPRI_TM_WR_EX |
7928			    SPRI_TM_EX_AC_AR;
7929		scsi_ulto2b(type_mask, res_cap->type_mask);
7930		break;
7931	}
7932	case SPRI_RS: { // read full status
7933		struct scsi_per_res_in_full *res_status;
7934		struct scsi_per_res_in_full_desc *res_desc;
7935		struct ctl_port *port;
7936		int i, len;
7937
7938		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7939
7940		/*
7941		 * We had to drop the lock to allocate our buffer, which
7942		 * leaves time for someone to come in with another
7943		 * persistent reservation.  (That is unlikely, though,
7944		 * since this should be the only persistent reservation
7945		 * command active right now.)
7946		 */
7947		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7948		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7949		     lun->pr_key_count)){
7950			mtx_unlock(&lun->lun_lock);
7951			free(ctsio->kern_data_ptr, M_CTL);
7952			printf("%s: reservation length changed, retrying\n",
7953			       __func__);
7954			goto retry;
7955		}
7956
7957		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7958
7959		res_desc = &res_status->desc[0];
7960		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7961			if (!lun->per_res[i].registered)
7962				continue;
7963
7964			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7965			    sizeof(res_desc->res_key));
7966			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7967			    (lun->pr_res_idx == i ||
7968			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7969				res_desc->flags = SPRI_FULL_R_HOLDER;
7970				res_desc->scopetype = lun->res_type;
7971			}
7972			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7973			    res_desc->rel_trgt_port_id);
7974			len = 0;
7975			port = softc->ctl_ports[
7976			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7977			if (port != NULL)
7978				len = ctl_create_iid(port,
7979				    i % CTL_MAX_INIT_PER_PORT,
7980				    res_desc->transport_id);
7981			scsi_ulto4b(len, res_desc->additional_length);
7982			res_desc = (struct scsi_per_res_in_full_desc *)
7983			    &res_desc->transport_id[len];
7984		}
7985		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7986		    res_status->header.length);
7987		break;
7988	}
7989	default:
7990		/*
7991		 * This is a bug, because we just checked for this above,
7992		 * and should have returned an error.
7993		 */
7994		panic("Invalid PR type %x", cdb->action);
7995		break; /* NOTREACHED */
7996	}
7997	mtx_unlock(&lun->lun_lock);
7998
7999	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8000	ctsio->be_move_done = ctl_config_move_done;
8001
8002	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
8003			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
8004			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
8005			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
8006			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
8007
8008	ctl_datamove((union ctl_io *)ctsio);
8009
8010	return (CTL_RETVAL_COMPLETE);
8011}
8012
8013/*
8014 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8015 * it should return.
8016 */
8017static int
8018ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8019		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8020		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8021		struct scsi_per_res_out_parms* param)
8022{
8023	union ctl_ha_msg persis_io;
8024	int retval, i;
8025	int isc_retval;
8026
8027	retval = 0;
8028
8029	mtx_lock(&lun->lun_lock);
8030	if (sa_res_key == 0) {
8031		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8032			/* validate scope and type */
8033			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8034			     SPR_LU_SCOPE) {
8035				mtx_unlock(&lun->lun_lock);
8036				ctl_set_invalid_field(/*ctsio*/ ctsio,
8037						      /*sks_valid*/ 1,
8038						      /*command*/ 1,
8039						      /*field*/ 2,
8040						      /*bit_valid*/ 1,
8041						      /*bit*/ 4);
8042				ctl_done((union ctl_io *)ctsio);
8043				return (1);
8044			}
8045
8046		        if (type>8 || type==2 || type==4 || type==0) {
8047				mtx_unlock(&lun->lun_lock);
8048				ctl_set_invalid_field(/*ctsio*/ ctsio,
8049       	           				      /*sks_valid*/ 1,
8050						      /*command*/ 1,
8051						      /*field*/ 2,
8052						      /*bit_valid*/ 1,
8053						      /*bit*/ 0);
8054				ctl_done((union ctl_io *)ctsio);
8055				return (1);
8056		        }
8057
8058			/* temporarily unregister this nexus */
8059			lun->per_res[residx].registered = 0;
8060
8061			/*
8062			 * Unregister everybody else and build UA for
8063			 * them
8064			 */
8065			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8066				if (lun->per_res[i].registered == 0)
8067					continue;
8068
8069				if (!persis_offset
8070				 && i <CTL_MAX_INITIATORS)
8071					lun->pending_ua[i] |=
8072						CTL_UA_REG_PREEMPT;
8073				else if (persis_offset
8074				      && i >= persis_offset)
8075					lun->pending_ua[i-persis_offset] |=
8076						CTL_UA_REG_PREEMPT;
8077				lun->per_res[i].registered = 0;
8078				memset(&lun->per_res[i].res_key, 0,
8079				       sizeof(struct scsi_per_res_key));
8080			}
8081			lun->per_res[residx].registered = 1;
8082			lun->pr_key_count = 1;
8083			lun->res_type = type;
8084			if (lun->res_type != SPR_TYPE_WR_EX_AR
8085			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8086				lun->pr_res_idx = residx;
8087
8088			/* send msg to other side */
8089			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8090			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8091			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8092			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8093			persis_io.pr.pr_info.res_type = type;
8094			memcpy(persis_io.pr.pr_info.sa_res_key,
8095			       param->serv_act_res_key,
8096			       sizeof(param->serv_act_res_key));
8097			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8098			     &persis_io, sizeof(persis_io), 0)) >
8099			     CTL_HA_STATUS_SUCCESS) {
8100				printf("CTL:Persis Out error returned "
8101				       "from ctl_ha_msg_send %d\n",
8102				       isc_retval);
8103			}
8104		} else {
8105			/* not all registrants */
8106			mtx_unlock(&lun->lun_lock);
8107			free(ctsio->kern_data_ptr, M_CTL);
8108			ctl_set_invalid_field(ctsio,
8109					      /*sks_valid*/ 1,
8110					      /*command*/ 0,
8111					      /*field*/ 8,
8112					      /*bit_valid*/ 0,
8113					      /*bit*/ 0);
8114			ctl_done((union ctl_io *)ctsio);
8115			return (1);
8116		}
8117	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8118		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8119		int found = 0;
8120
8121		if (res_key == sa_res_key) {
8122			/* special case */
8123			/*
8124			 * The spec implies this is not good but doesn't
8125			 * say what to do. There are two choices either
8126			 * generate a res conflict or check condition
8127			 * with illegal field in parameter data. Since
8128			 * that is what is done when the sa_res_key is
8129			 * zero I'll take that approach since this has
8130			 * to do with the sa_res_key.
8131			 */
8132			mtx_unlock(&lun->lun_lock);
8133			free(ctsio->kern_data_ptr, M_CTL);
8134			ctl_set_invalid_field(ctsio,
8135					      /*sks_valid*/ 1,
8136					      /*command*/ 0,
8137					      /*field*/ 8,
8138					      /*bit_valid*/ 0,
8139					      /*bit*/ 0);
8140			ctl_done((union ctl_io *)ctsio);
8141			return (1);
8142		}
8143
8144		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8145			if (lun->per_res[i].registered
8146			 && memcmp(param->serv_act_res_key,
8147			    lun->per_res[i].res_key.key,
8148			    sizeof(struct scsi_per_res_key)) != 0)
8149				continue;
8150
8151			found = 1;
8152			lun->per_res[i].registered = 0;
8153			memset(&lun->per_res[i].res_key, 0,
8154			       sizeof(struct scsi_per_res_key));
8155			lun->pr_key_count--;
8156
8157			if (!persis_offset && i < CTL_MAX_INITIATORS)
8158				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8159			else if (persis_offset && i >= persis_offset)
8160				lun->pending_ua[i-persis_offset] |=
8161					CTL_UA_REG_PREEMPT;
8162		}
8163		if (!found) {
8164			mtx_unlock(&lun->lun_lock);
8165			free(ctsio->kern_data_ptr, M_CTL);
8166			ctl_set_reservation_conflict(ctsio);
8167			ctl_done((union ctl_io *)ctsio);
8168			return (CTL_RETVAL_COMPLETE);
8169		}
8170		/* send msg to other side */
8171		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8172		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8173		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8174		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8175		persis_io.pr.pr_info.res_type = type;
8176		memcpy(persis_io.pr.pr_info.sa_res_key,
8177		       param->serv_act_res_key,
8178		       sizeof(param->serv_act_res_key));
8179		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8180		     &persis_io, sizeof(persis_io), 0)) >
8181		     CTL_HA_STATUS_SUCCESS) {
8182			printf("CTL:Persis Out error returned from "
8183			       "ctl_ha_msg_send %d\n", isc_retval);
8184		}
8185	} else {
8186		/* Reserved but not all registrants */
8187		/* sa_res_key is res holder */
8188		if (memcmp(param->serv_act_res_key,
8189                   lun->per_res[lun->pr_res_idx].res_key.key,
8190                   sizeof(struct scsi_per_res_key)) == 0) {
8191			/* validate scope and type */
8192			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8193			     SPR_LU_SCOPE) {
8194				mtx_unlock(&lun->lun_lock);
8195				ctl_set_invalid_field(/*ctsio*/ ctsio,
8196						      /*sks_valid*/ 1,
8197						      /*command*/ 1,
8198						      /*field*/ 2,
8199						      /*bit_valid*/ 1,
8200						      /*bit*/ 4);
8201				ctl_done((union ctl_io *)ctsio);
8202				return (1);
8203			}
8204
8205			if (type>8 || type==2 || type==4 || type==0) {
8206				mtx_unlock(&lun->lun_lock);
8207				ctl_set_invalid_field(/*ctsio*/ ctsio,
8208						      /*sks_valid*/ 1,
8209						      /*command*/ 1,
8210						      /*field*/ 2,
8211						      /*bit_valid*/ 1,
8212						      /*bit*/ 0);
8213				ctl_done((union ctl_io *)ctsio);
8214				return (1);
8215			}
8216
8217			/*
8218			 * Do the following:
8219			 * if sa_res_key != res_key remove all
8220			 * registrants w/sa_res_key and generate UA
8221			 * for these registrants(Registrations
8222			 * Preempted) if it wasn't an exclusive
8223			 * reservation generate UA(Reservations
8224			 * Preempted) for all other registered nexuses
8225			 * if the type has changed. Establish the new
8226			 * reservation and holder. If res_key and
8227			 * sa_res_key are the same do the above
8228			 * except don't unregister the res holder.
8229			 */
8230
8231			/*
8232			 * Temporarily unregister so it won't get
8233			 * removed or UA generated
8234			 */
8235			lun->per_res[residx].registered = 0;
8236			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8237				if (lun->per_res[i].registered == 0)
8238					continue;
8239
8240				if (memcmp(param->serv_act_res_key,
8241				    lun->per_res[i].res_key.key,
8242				    sizeof(struct scsi_per_res_key)) == 0) {
8243					lun->per_res[i].registered = 0;
8244					memset(&lun->per_res[i].res_key,
8245					       0,
8246					       sizeof(struct scsi_per_res_key));
8247					lun->pr_key_count--;
8248
8249					if (!persis_offset
8250					 && i < CTL_MAX_INITIATORS)
8251						lun->pending_ua[i] |=
8252							CTL_UA_REG_PREEMPT;
8253					else if (persis_offset
8254					      && i >= persis_offset)
8255						lun->pending_ua[i-persis_offset] |=
8256						  CTL_UA_REG_PREEMPT;
8257				} else if (type != lun->res_type
8258					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8259					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8260						if (!persis_offset
8261						 && i < CTL_MAX_INITIATORS)
8262							lun->pending_ua[i] |=
8263							CTL_UA_RES_RELEASE;
8264						else if (persis_offset
8265						      && i >= persis_offset)
8266							lun->pending_ua[
8267							i-persis_offset] |=
8268							CTL_UA_RES_RELEASE;
8269				}
8270			}
8271			lun->per_res[residx].registered = 1;
8272			lun->res_type = type;
8273			if (lun->res_type != SPR_TYPE_WR_EX_AR
8274			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8275				lun->pr_res_idx = residx;
8276			else
8277				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8278
8279			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8280			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8281			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8282			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8283			persis_io.pr.pr_info.res_type = type;
8284			memcpy(persis_io.pr.pr_info.sa_res_key,
8285			       param->serv_act_res_key,
8286			       sizeof(param->serv_act_res_key));
8287			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8288			     &persis_io, sizeof(persis_io), 0)) >
8289			     CTL_HA_STATUS_SUCCESS) {
8290				printf("CTL:Persis Out error returned "
8291				       "from ctl_ha_msg_send %d\n",
8292				       isc_retval);
8293			}
8294		} else {
8295			/*
8296			 * sa_res_key is not the res holder just
8297			 * remove registrants
8298			 */
8299			int found=0;
8300
8301			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8302				if (memcmp(param->serv_act_res_key,
8303				    lun->per_res[i].res_key.key,
8304				    sizeof(struct scsi_per_res_key)) != 0)
8305					continue;
8306
8307				found = 1;
8308				lun->per_res[i].registered = 0;
8309				memset(&lun->per_res[i].res_key, 0,
8310				       sizeof(struct scsi_per_res_key));
8311				lun->pr_key_count--;
8312
8313				if (!persis_offset
8314				 && i < CTL_MAX_INITIATORS)
8315					lun->pending_ua[i] |=
8316						CTL_UA_REG_PREEMPT;
8317				else if (persis_offset
8318				      && i >= persis_offset)
8319					lun->pending_ua[i-persis_offset] |=
8320						CTL_UA_REG_PREEMPT;
8321			}
8322
8323			if (!found) {
8324				mtx_unlock(&lun->lun_lock);
8325				free(ctsio->kern_data_ptr, M_CTL);
8326				ctl_set_reservation_conflict(ctsio);
8327				ctl_done((union ctl_io *)ctsio);
8328		        	return (1);
8329			}
8330			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8331			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8332			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8333			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8334			persis_io.pr.pr_info.res_type = type;
8335			memcpy(persis_io.pr.pr_info.sa_res_key,
8336			       param->serv_act_res_key,
8337			       sizeof(param->serv_act_res_key));
8338			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8339			     &persis_io, sizeof(persis_io), 0)) >
8340			     CTL_HA_STATUS_SUCCESS) {
8341				printf("CTL:Persis Out error returned "
8342				       "from ctl_ha_msg_send %d\n",
8343				isc_retval);
8344			}
8345		}
8346	}
8347
8348	lun->PRGeneration++;
8349	mtx_unlock(&lun->lun_lock);
8350
8351	return (retval);
8352}
8353
8354static void
8355ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8356{
8357	int i;
8358
8359	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8360	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8361	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8362		   msg->pr.pr_info.sa_res_key,
8363		   sizeof(struct scsi_per_res_key)) != 0) {
8364		uint64_t sa_res_key;
8365		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8366
8367		if (sa_res_key == 0) {
8368			/* temporarily unregister this nexus */
8369			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8370
8371			/*
8372			 * Unregister everybody else and build UA for
8373			 * them
8374			 */
8375			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8376				if (lun->per_res[i].registered == 0)
8377					continue;
8378
8379				if (!persis_offset
8380				 && i < CTL_MAX_INITIATORS)
8381					lun->pending_ua[i] |=
8382						CTL_UA_REG_PREEMPT;
8383				else if (persis_offset && i >= persis_offset)
8384					lun->pending_ua[i - persis_offset] |=
8385						CTL_UA_REG_PREEMPT;
8386				lun->per_res[i].registered = 0;
8387				memset(&lun->per_res[i].res_key, 0,
8388				       sizeof(struct scsi_per_res_key));
8389			}
8390
8391			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8392			lun->pr_key_count = 1;
8393			lun->res_type = msg->pr.pr_info.res_type;
8394			if (lun->res_type != SPR_TYPE_WR_EX_AR
8395			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8396				lun->pr_res_idx = msg->pr.pr_info.residx;
8397		} else {
8398		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8399				if (memcmp(msg->pr.pr_info.sa_res_key,
8400		                   lun->per_res[i].res_key.key,
8401		                   sizeof(struct scsi_per_res_key)) != 0)
8402					continue;
8403
8404				lun->per_res[i].registered = 0;
8405				memset(&lun->per_res[i].res_key, 0,
8406				       sizeof(struct scsi_per_res_key));
8407				lun->pr_key_count--;
8408
8409				if (!persis_offset
8410				 && i < persis_offset)
8411					lun->pending_ua[i] |=
8412						CTL_UA_REG_PREEMPT;
8413				else if (persis_offset
8414				      && i >= persis_offset)
8415					lun->pending_ua[i - persis_offset] |=
8416						CTL_UA_REG_PREEMPT;
8417			}
8418		}
8419	} else {
8420		/*
8421		 * Temporarily unregister so it won't get removed
8422		 * or UA generated
8423		 */
8424		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8425		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8426			if (lun->per_res[i].registered == 0)
8427				continue;
8428
8429			if (memcmp(msg->pr.pr_info.sa_res_key,
8430	                   lun->per_res[i].res_key.key,
8431	                   sizeof(struct scsi_per_res_key)) == 0) {
8432				lun->per_res[i].registered = 0;
8433				memset(&lun->per_res[i].res_key, 0,
8434				       sizeof(struct scsi_per_res_key));
8435				lun->pr_key_count--;
8436				if (!persis_offset
8437				 && i < CTL_MAX_INITIATORS)
8438					lun->pending_ua[i] |=
8439						CTL_UA_REG_PREEMPT;
8440				else if (persis_offset
8441				      && i >= persis_offset)
8442					lun->pending_ua[i - persis_offset] |=
8443						CTL_UA_REG_PREEMPT;
8444			} else if (msg->pr.pr_info.res_type != lun->res_type
8445				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8446				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8447					if (!persis_offset
8448					 && i < persis_offset)
8449						lun->pending_ua[i] |=
8450							CTL_UA_RES_RELEASE;
8451					else if (persis_offset
8452					      && i >= persis_offset)
8453					lun->pending_ua[i - persis_offset] |=
8454						CTL_UA_RES_RELEASE;
8455			}
8456		}
8457		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8458		lun->res_type = msg->pr.pr_info.res_type;
8459		if (lun->res_type != SPR_TYPE_WR_EX_AR
8460		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8461			lun->pr_res_idx = msg->pr.pr_info.residx;
8462		else
8463			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8464	}
8465	lun->PRGeneration++;
8466
8467}
8468
8469
8470int
8471ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8472{
8473	int retval;
8474	int isc_retval;
8475	u_int32_t param_len;
8476	struct scsi_per_res_out *cdb;
8477	struct ctl_lun *lun;
8478	struct scsi_per_res_out_parms* param;
8479	struct ctl_softc *softc;
8480	uint32_t residx;
8481	uint64_t res_key, sa_res_key;
8482	uint8_t type;
8483	union ctl_ha_msg persis_io;
8484	int    i;
8485
8486	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8487
8488	retval = CTL_RETVAL_COMPLETE;
8489
8490	softc = control_softc;
8491
8492	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8493	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8494
8495	/*
8496	 * We only support whole-LUN scope.  The scope & type are ignored for
8497	 * register, register and ignore existing key and clear.
8498	 * We sometimes ignore scope and type on preempts too!!
8499	 * Verify reservation type here as well.
8500	 */
8501	type = cdb->scope_type & SPR_TYPE_MASK;
8502	if ((cdb->action == SPRO_RESERVE)
8503	 || (cdb->action == SPRO_RELEASE)) {
8504		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8505			ctl_set_invalid_field(/*ctsio*/ ctsio,
8506					      /*sks_valid*/ 1,
8507					      /*command*/ 1,
8508					      /*field*/ 2,
8509					      /*bit_valid*/ 1,
8510					      /*bit*/ 4);
8511			ctl_done((union ctl_io *)ctsio);
8512			return (CTL_RETVAL_COMPLETE);
8513		}
8514
8515		if (type>8 || type==2 || type==4 || type==0) {
8516			ctl_set_invalid_field(/*ctsio*/ ctsio,
8517					      /*sks_valid*/ 1,
8518					      /*command*/ 1,
8519					      /*field*/ 2,
8520					      /*bit_valid*/ 1,
8521					      /*bit*/ 0);
8522			ctl_done((union ctl_io *)ctsio);
8523			return (CTL_RETVAL_COMPLETE);
8524		}
8525	}
8526
8527	param_len = scsi_4btoul(cdb->length);
8528
8529	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8530		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8531		ctsio->kern_data_len = param_len;
8532		ctsio->kern_total_len = param_len;
8533		ctsio->kern_data_resid = 0;
8534		ctsio->kern_rel_offset = 0;
8535		ctsio->kern_sg_entries = 0;
8536		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8537		ctsio->be_move_done = ctl_config_move_done;
8538		ctl_datamove((union ctl_io *)ctsio);
8539
8540		return (CTL_RETVAL_COMPLETE);
8541	}
8542
8543	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8544
8545	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8546	res_key = scsi_8btou64(param->res_key.key);
8547	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8548
8549	/*
8550	 * Validate the reservation key here except for SPRO_REG_IGNO
8551	 * This must be done for all other service actions
8552	 */
8553	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8554		mtx_lock(&lun->lun_lock);
8555		if (lun->per_res[residx].registered) {
8556		    if (memcmp(param->res_key.key,
8557			       lun->per_res[residx].res_key.key,
8558			       ctl_min(sizeof(param->res_key),
8559			       sizeof(lun->per_res[residx].res_key))) != 0) {
8560				/*
8561				 * The current key passed in doesn't match
8562				 * the one the initiator previously
8563				 * registered.
8564				 */
8565				mtx_unlock(&lun->lun_lock);
8566				free(ctsio->kern_data_ptr, M_CTL);
8567				ctl_set_reservation_conflict(ctsio);
8568				ctl_done((union ctl_io *)ctsio);
8569				return (CTL_RETVAL_COMPLETE);
8570			}
8571		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8572			/*
8573			 * We are not registered
8574			 */
8575			mtx_unlock(&lun->lun_lock);
8576			free(ctsio->kern_data_ptr, M_CTL);
8577			ctl_set_reservation_conflict(ctsio);
8578			ctl_done((union ctl_io *)ctsio);
8579			return (CTL_RETVAL_COMPLETE);
8580		} else if (res_key != 0) {
8581			/*
8582			 * We are not registered and trying to register but
8583			 * the register key isn't zero.
8584			 */
8585			mtx_unlock(&lun->lun_lock);
8586			free(ctsio->kern_data_ptr, M_CTL);
8587			ctl_set_reservation_conflict(ctsio);
8588			ctl_done((union ctl_io *)ctsio);
8589			return (CTL_RETVAL_COMPLETE);
8590		}
8591		mtx_unlock(&lun->lun_lock);
8592	}
8593
8594	switch (cdb->action & SPRO_ACTION_MASK) {
8595	case SPRO_REGISTER:
8596	case SPRO_REG_IGNO: {
8597
8598#if 0
8599		printf("Registration received\n");
8600#endif
8601
8602		/*
8603		 * We don't support any of these options, as we report in
8604		 * the read capabilities request (see
8605		 * ctl_persistent_reserve_in(), above).
8606		 */
8607		if ((param->flags & SPR_SPEC_I_PT)
8608		 || (param->flags & SPR_ALL_TG_PT)
8609		 || (param->flags & SPR_APTPL)) {
8610			int bit_ptr;
8611
8612			if (param->flags & SPR_APTPL)
8613				bit_ptr = 0;
8614			else if (param->flags & SPR_ALL_TG_PT)
8615				bit_ptr = 2;
8616			else /* SPR_SPEC_I_PT */
8617				bit_ptr = 3;
8618
8619			free(ctsio->kern_data_ptr, M_CTL);
8620			ctl_set_invalid_field(ctsio,
8621					      /*sks_valid*/ 1,
8622					      /*command*/ 0,
8623					      /*field*/ 20,
8624					      /*bit_valid*/ 1,
8625					      /*bit*/ bit_ptr);
8626			ctl_done((union ctl_io *)ctsio);
8627			return (CTL_RETVAL_COMPLETE);
8628		}
8629
8630		mtx_lock(&lun->lun_lock);
8631
8632		/*
8633		 * The initiator wants to clear the
8634		 * key/unregister.
8635		 */
8636		if (sa_res_key == 0) {
8637			if ((res_key == 0
8638			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8639			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8640			  && !lun->per_res[residx].registered)) {
8641				mtx_unlock(&lun->lun_lock);
8642				goto done;
8643			}
8644
8645			lun->per_res[residx].registered = 0;
8646			memset(&lun->per_res[residx].res_key,
8647			       0, sizeof(lun->per_res[residx].res_key));
8648			lun->pr_key_count--;
8649
8650			if (residx == lun->pr_res_idx) {
8651				lun->flags &= ~CTL_LUN_PR_RESERVED;
8652				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8653
8654				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8655				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8656				 && lun->pr_key_count) {
8657					/*
8658					 * If the reservation is a registrants
8659					 * only type we need to generate a UA
8660					 * for other registered inits.  The
8661					 * sense code should be RESERVATIONS
8662					 * RELEASED
8663					 */
8664
8665					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8666						if (lun->per_res[
8667						    i+persis_offset].registered
8668						    == 0)
8669							continue;
8670						lun->pending_ua[i] |=
8671							CTL_UA_RES_RELEASE;
8672					}
8673				}
8674				lun->res_type = 0;
8675			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8676				if (lun->pr_key_count==0) {
8677					lun->flags &= ~CTL_LUN_PR_RESERVED;
8678					lun->res_type = 0;
8679					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8680				}
8681			}
8682			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8683			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8684			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8685			persis_io.pr.pr_info.residx = residx;
8686			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8687			     &persis_io, sizeof(persis_io), 0 )) >
8688			     CTL_HA_STATUS_SUCCESS) {
8689				printf("CTL:Persis Out error returned from "
8690				       "ctl_ha_msg_send %d\n", isc_retval);
8691			}
8692		} else /* sa_res_key != 0 */ {
8693
8694			/*
8695			 * If we aren't registered currently then increment
8696			 * the key count and set the registered flag.
8697			 */
8698			if (!lun->per_res[residx].registered) {
8699				lun->pr_key_count++;
8700				lun->per_res[residx].registered = 1;
8701			}
8702
8703			memcpy(&lun->per_res[residx].res_key,
8704			       param->serv_act_res_key,
8705			       ctl_min(sizeof(param->serv_act_res_key),
8706			       sizeof(lun->per_res[residx].res_key)));
8707
8708			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8709			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8710			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8711			persis_io.pr.pr_info.residx = residx;
8712			memcpy(persis_io.pr.pr_info.sa_res_key,
8713			       param->serv_act_res_key,
8714			       sizeof(param->serv_act_res_key));
8715			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8716			     &persis_io, sizeof(persis_io), 0)) >
8717			     CTL_HA_STATUS_SUCCESS) {
8718				printf("CTL:Persis Out error returned from "
8719				       "ctl_ha_msg_send %d\n", isc_retval);
8720			}
8721		}
8722		lun->PRGeneration++;
8723		mtx_unlock(&lun->lun_lock);
8724
8725		break;
8726	}
8727	case SPRO_RESERVE:
8728#if 0
8729                printf("Reserve executed type %d\n", type);
8730#endif
8731		mtx_lock(&lun->lun_lock);
8732		if (lun->flags & CTL_LUN_PR_RESERVED) {
8733			/*
8734			 * if this isn't the reservation holder and it's
8735			 * not a "all registrants" type or if the type is
8736			 * different then we have a conflict
8737			 */
8738			if ((lun->pr_res_idx != residx
8739			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8740			 || lun->res_type != type) {
8741				mtx_unlock(&lun->lun_lock);
8742				free(ctsio->kern_data_ptr, M_CTL);
8743				ctl_set_reservation_conflict(ctsio);
8744				ctl_done((union ctl_io *)ctsio);
8745				return (CTL_RETVAL_COMPLETE);
8746			}
8747			mtx_unlock(&lun->lun_lock);
8748		} else /* create a reservation */ {
8749			/*
8750			 * If it's not an "all registrants" type record
8751			 * reservation holder
8752			 */
8753			if (type != SPR_TYPE_WR_EX_AR
8754			 && type != SPR_TYPE_EX_AC_AR)
8755				lun->pr_res_idx = residx; /* Res holder */
8756			else
8757				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8758
8759			lun->flags |= CTL_LUN_PR_RESERVED;
8760			lun->res_type = type;
8761
8762			mtx_unlock(&lun->lun_lock);
8763
8764			/* send msg to other side */
8765			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8766			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8767			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8768			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8769			persis_io.pr.pr_info.res_type = type;
8770			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8771			     &persis_io, sizeof(persis_io), 0)) >
8772			     CTL_HA_STATUS_SUCCESS) {
8773				printf("CTL:Persis Out error returned from "
8774				       "ctl_ha_msg_send %d\n", isc_retval);
8775			}
8776		}
8777		break;
8778
8779	case SPRO_RELEASE:
8780		mtx_lock(&lun->lun_lock);
8781		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8782			/* No reservation exists return good status */
8783			mtx_unlock(&lun->lun_lock);
8784			goto done;
8785		}
8786		/*
8787		 * Is this nexus a reservation holder?
8788		 */
8789		if (lun->pr_res_idx != residx
8790		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8791			/*
8792			 * not a res holder return good status but
8793			 * do nothing
8794			 */
8795			mtx_unlock(&lun->lun_lock);
8796			goto done;
8797		}
8798
8799		if (lun->res_type != type) {
8800			mtx_unlock(&lun->lun_lock);
8801			free(ctsio->kern_data_ptr, M_CTL);
8802			ctl_set_illegal_pr_release(ctsio);
8803			ctl_done((union ctl_io *)ctsio);
8804			return (CTL_RETVAL_COMPLETE);
8805		}
8806
8807		/* okay to release */
8808		lun->flags &= ~CTL_LUN_PR_RESERVED;
8809		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8810		lun->res_type = 0;
8811
8812		/*
8813		 * if this isn't an exclusive access
8814		 * res generate UA for all other
8815		 * registrants.
8816		 */
8817		if (type != SPR_TYPE_EX_AC
8818		 && type != SPR_TYPE_WR_EX) {
8819			/*
8820			 * temporarily unregister so we don't generate UA
8821			 */
8822			lun->per_res[residx].registered = 0;
8823
8824			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8825				if (lun->per_res[i+persis_offset].registered
8826				    == 0)
8827					continue;
8828				lun->pending_ua[i] |=
8829					CTL_UA_RES_RELEASE;
8830			}
8831
8832			lun->per_res[residx].registered = 1;
8833		}
8834		mtx_unlock(&lun->lun_lock);
8835		/* Send msg to other side */
8836		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8837		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8838		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8839		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8840		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8841			printf("CTL:Persis Out error returned from "
8842			       "ctl_ha_msg_send %d\n", isc_retval);
8843		}
8844		break;
8845
8846	case SPRO_CLEAR:
8847		/* send msg to other side */
8848
8849		mtx_lock(&lun->lun_lock);
8850		lun->flags &= ~CTL_LUN_PR_RESERVED;
8851		lun->res_type = 0;
8852		lun->pr_key_count = 0;
8853		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8854
8855
8856		memset(&lun->per_res[residx].res_key,
8857		       0, sizeof(lun->per_res[residx].res_key));
8858		lun->per_res[residx].registered = 0;
8859
8860		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8861			if (lun->per_res[i].registered) {
8862				if (!persis_offset && i < CTL_MAX_INITIATORS)
8863					lun->pending_ua[i] |=
8864						CTL_UA_RES_PREEMPT;
8865				else if (persis_offset && i >= persis_offset)
8866					lun->pending_ua[i-persis_offset] |=
8867					    CTL_UA_RES_PREEMPT;
8868
8869				memset(&lun->per_res[i].res_key,
8870				       0, sizeof(struct scsi_per_res_key));
8871				lun->per_res[i].registered = 0;
8872			}
8873		lun->PRGeneration++;
8874		mtx_unlock(&lun->lun_lock);
8875		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8876		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8877		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8878		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8879		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8880			printf("CTL:Persis Out error returned from "
8881			       "ctl_ha_msg_send %d\n", isc_retval);
8882		}
8883		break;
8884
8885	case SPRO_PREEMPT: {
8886		int nretval;
8887
8888		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8889					  residx, ctsio, cdb, param);
8890		if (nretval != 0)
8891			return (CTL_RETVAL_COMPLETE);
8892		break;
8893	}
8894	default:
8895		panic("Invalid PR type %x", cdb->action);
8896	}
8897
8898done:
8899	free(ctsio->kern_data_ptr, M_CTL);
8900	ctl_set_success(ctsio);
8901	ctl_done((union ctl_io *)ctsio);
8902
8903	return (retval);
8904}
8905
8906/*
8907 * This routine is for handling a message from the other SC pertaining to
8908 * persistent reserve out. All the error checking will have been done
8909 * so only perorming the action need be done here to keep the two
8910 * in sync.
8911 */
8912static void
8913ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8914{
8915	struct ctl_lun *lun;
8916	struct ctl_softc *softc;
8917	int i;
8918	uint32_t targ_lun;
8919
8920	softc = control_softc;
8921
8922	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8923	lun = softc->ctl_luns[targ_lun];
8924	mtx_lock(&lun->lun_lock);
8925	switch(msg->pr.pr_info.action) {
8926	case CTL_PR_REG_KEY:
8927		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8928			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8929			lun->pr_key_count++;
8930		}
8931		lun->PRGeneration++;
8932		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8933		       msg->pr.pr_info.sa_res_key,
8934		       sizeof(struct scsi_per_res_key));
8935		break;
8936
8937	case CTL_PR_UNREG_KEY:
8938		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8939		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8940		       0, sizeof(struct scsi_per_res_key));
8941		lun->pr_key_count--;
8942
8943		/* XXX Need to see if the reservation has been released */
8944		/* if so do we need to generate UA? */
8945		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8946			lun->flags &= ~CTL_LUN_PR_RESERVED;
8947			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8948
8949			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8950			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8951			 && lun->pr_key_count) {
8952				/*
8953				 * If the reservation is a registrants
8954				 * only type we need to generate a UA
8955				 * for other registered inits.  The
8956				 * sense code should be RESERVATIONS
8957				 * RELEASED
8958				 */
8959
8960				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8961					if (lun->per_res[i+
8962					    persis_offset].registered == 0)
8963						continue;
8964
8965					lun->pending_ua[i] |=
8966						CTL_UA_RES_RELEASE;
8967				}
8968			}
8969			lun->res_type = 0;
8970		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8971			if (lun->pr_key_count==0) {
8972				lun->flags &= ~CTL_LUN_PR_RESERVED;
8973				lun->res_type = 0;
8974				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8975			}
8976		}
8977		lun->PRGeneration++;
8978		break;
8979
8980	case CTL_PR_RESERVE:
8981		lun->flags |= CTL_LUN_PR_RESERVED;
8982		lun->res_type = msg->pr.pr_info.res_type;
8983		lun->pr_res_idx = msg->pr.pr_info.residx;
8984
8985		break;
8986
8987	case CTL_PR_RELEASE:
8988		/*
8989		 * if this isn't an exclusive access res generate UA for all
8990		 * other registrants.
8991		 */
8992		if (lun->res_type != SPR_TYPE_EX_AC
8993		 && lun->res_type != SPR_TYPE_WR_EX) {
8994			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8995				if (lun->per_res[i+persis_offset].registered)
8996					lun->pending_ua[i] |=
8997						CTL_UA_RES_RELEASE;
8998		}
8999
9000		lun->flags &= ~CTL_LUN_PR_RESERVED;
9001		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
9002		lun->res_type = 0;
9003		break;
9004
9005	case CTL_PR_PREEMPT:
9006		ctl_pro_preempt_other(lun, msg);
9007		break;
9008	case CTL_PR_CLEAR:
9009		lun->flags &= ~CTL_LUN_PR_RESERVED;
9010		lun->res_type = 0;
9011		lun->pr_key_count = 0;
9012		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
9013
9014		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
9015			if (lun->per_res[i].registered == 0)
9016				continue;
9017			if (!persis_offset
9018			 && i < CTL_MAX_INITIATORS)
9019				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
9020			else if (persis_offset
9021			      && i >= persis_offset)
9022				lun->pending_ua[i-persis_offset] |=
9023					CTL_UA_RES_PREEMPT;
9024			memset(&lun->per_res[i].res_key, 0,
9025			       sizeof(struct scsi_per_res_key));
9026			lun->per_res[i].registered = 0;
9027		}
9028		lun->PRGeneration++;
9029		break;
9030	}
9031
9032	mtx_unlock(&lun->lun_lock);
9033}
9034
9035int
9036ctl_read_write(struct ctl_scsiio *ctsio)
9037{
9038	struct ctl_lun *lun;
9039	struct ctl_lba_len_flags *lbalen;
9040	uint64_t lba;
9041	uint32_t num_blocks;
9042	int flags, retval;
9043	int isread;
9044
9045	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9046
9047	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9048
9049	flags = 0;
9050	retval = CTL_RETVAL_COMPLETE;
9051
9052	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9053	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9054	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9055		uint32_t residx;
9056
9057		/*
9058		 * XXX KDM need a lock here.
9059		 */
9060		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9061		if ((lun->res_type == SPR_TYPE_EX_AC
9062		  && residx != lun->pr_res_idx)
9063		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9064		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9065		  && !lun->per_res[residx].registered)) {
9066			ctl_set_reservation_conflict(ctsio);
9067			ctl_done((union ctl_io *)ctsio);
9068			return (CTL_RETVAL_COMPLETE);
9069	        }
9070	}
9071
9072	switch (ctsio->cdb[0]) {
9073	case READ_6:
9074	case WRITE_6: {
9075		struct scsi_rw_6 *cdb;
9076
9077		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9078
9079		lba = scsi_3btoul(cdb->addr);
9080		/* only 5 bits are valid in the most significant address byte */
9081		lba &= 0x1fffff;
9082		num_blocks = cdb->length;
9083		/*
9084		 * This is correct according to SBC-2.
9085		 */
9086		if (num_blocks == 0)
9087			num_blocks = 256;
9088		break;
9089	}
9090	case READ_10:
9091	case WRITE_10: {
9092		struct scsi_rw_10 *cdb;
9093
9094		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9095		if (cdb->byte2 & SRW10_FUA)
9096			flags |= CTL_LLF_FUA;
9097		if (cdb->byte2 & SRW10_DPO)
9098			flags |= CTL_LLF_DPO;
9099		lba = scsi_4btoul(cdb->addr);
9100		num_blocks = scsi_2btoul(cdb->length);
9101		break;
9102	}
9103	case WRITE_VERIFY_10: {
9104		struct scsi_write_verify_10 *cdb;
9105
9106		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9107		flags |= CTL_LLF_FUA;
9108		if (cdb->byte2 & SWV_DPO)
9109			flags |= CTL_LLF_DPO;
9110		lba = scsi_4btoul(cdb->addr);
9111		num_blocks = scsi_2btoul(cdb->length);
9112		break;
9113	}
9114	case READ_12:
9115	case WRITE_12: {
9116		struct scsi_rw_12 *cdb;
9117
9118		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9119		if (cdb->byte2 & SRW12_FUA)
9120			flags |= CTL_LLF_FUA;
9121		if (cdb->byte2 & SRW12_DPO)
9122			flags |= CTL_LLF_DPO;
9123		lba = scsi_4btoul(cdb->addr);
9124		num_blocks = scsi_4btoul(cdb->length);
9125		break;
9126	}
9127	case WRITE_VERIFY_12: {
9128		struct scsi_write_verify_12 *cdb;
9129
9130		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9131		flags |= CTL_LLF_FUA;
9132		if (cdb->byte2 & SWV_DPO)
9133			flags |= CTL_LLF_DPO;
9134		lba = scsi_4btoul(cdb->addr);
9135		num_blocks = scsi_4btoul(cdb->length);
9136		break;
9137	}
9138	case READ_16:
9139	case WRITE_16: {
9140		struct scsi_rw_16 *cdb;
9141
9142		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9143		if (cdb->byte2 & SRW12_FUA)
9144			flags |= CTL_LLF_FUA;
9145		if (cdb->byte2 & SRW12_DPO)
9146			flags |= CTL_LLF_DPO;
9147		lba = scsi_8btou64(cdb->addr);
9148		num_blocks = scsi_4btoul(cdb->length);
9149		break;
9150	}
9151	case WRITE_ATOMIC_16: {
9152		struct scsi_rw_16 *cdb;
9153
9154		if (lun->be_lun->atomicblock == 0) {
9155			ctl_set_invalid_opcode(ctsio);
9156			ctl_done((union ctl_io *)ctsio);
9157			return (CTL_RETVAL_COMPLETE);
9158		}
9159
9160		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9161		if (cdb->byte2 & SRW12_FUA)
9162			flags |= CTL_LLF_FUA;
9163		if (cdb->byte2 & SRW12_DPO)
9164			flags |= CTL_LLF_DPO;
9165		lba = scsi_8btou64(cdb->addr);
9166		num_blocks = scsi_4btoul(cdb->length);
9167		if (num_blocks > lun->be_lun->atomicblock) {
9168			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9169			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9170			    /*bit*/ 0);
9171			ctl_done((union ctl_io *)ctsio);
9172			return (CTL_RETVAL_COMPLETE);
9173		}
9174		break;
9175	}
9176	case WRITE_VERIFY_16: {
9177		struct scsi_write_verify_16 *cdb;
9178
9179		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9180		flags |= CTL_LLF_FUA;
9181		if (cdb->byte2 & SWV_DPO)
9182			flags |= CTL_LLF_DPO;
9183		lba = scsi_8btou64(cdb->addr);
9184		num_blocks = scsi_4btoul(cdb->length);
9185		break;
9186	}
9187	default:
9188		/*
9189		 * We got a command we don't support.  This shouldn't
9190		 * happen, commands should be filtered out above us.
9191		 */
9192		ctl_set_invalid_opcode(ctsio);
9193		ctl_done((union ctl_io *)ctsio);
9194
9195		return (CTL_RETVAL_COMPLETE);
9196		break; /* NOTREACHED */
9197	}
9198
9199	/*
9200	 * The first check is to make sure we're in bounds, the second
9201	 * check is to catch wrap-around problems.  If the lba + num blocks
9202	 * is less than the lba, then we've wrapped around and the block
9203	 * range is invalid anyway.
9204	 */
9205	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9206	 || ((lba + num_blocks) < lba)) {
9207		ctl_set_lba_out_of_range(ctsio);
9208		ctl_done((union ctl_io *)ctsio);
9209		return (CTL_RETVAL_COMPLETE);
9210	}
9211
9212	/*
9213	 * According to SBC-3, a transfer length of 0 is not an error.
9214	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9215	 * translates to 256 blocks for those commands.
9216	 */
9217	if (num_blocks == 0) {
9218		ctl_set_success(ctsio);
9219		ctl_done((union ctl_io *)ctsio);
9220		return (CTL_RETVAL_COMPLETE);
9221	}
9222
9223	/* Set FUA and/or DPO if caches are disabled. */
9224	if (isread) {
9225		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9226		    SCP_RCD) != 0)
9227			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9228	} else {
9229		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9230		    SCP_WCE) == 0)
9231			flags |= CTL_LLF_FUA;
9232	}
9233
9234	lbalen = (struct ctl_lba_len_flags *)
9235	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9236	lbalen->lba = lba;
9237	lbalen->len = num_blocks;
9238	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9239
9240	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9241	ctsio->kern_rel_offset = 0;
9242
9243	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9244
9245	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9246
9247	return (retval);
9248}
9249
9250static int
9251ctl_cnw_cont(union ctl_io *io)
9252{
9253	struct ctl_scsiio *ctsio;
9254	struct ctl_lun *lun;
9255	struct ctl_lba_len_flags *lbalen;
9256	int retval;
9257
9258	ctsio = &io->scsiio;
9259	ctsio->io_hdr.status = CTL_STATUS_NONE;
9260	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9261	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9262	lbalen = (struct ctl_lba_len_flags *)
9263	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9264	lbalen->flags &= ~CTL_LLF_COMPARE;
9265	lbalen->flags |= CTL_LLF_WRITE;
9266
9267	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9268	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9269	return (retval);
9270}
9271
9272int
9273ctl_cnw(struct ctl_scsiio *ctsio)
9274{
9275	struct ctl_lun *lun;
9276	struct ctl_lba_len_flags *lbalen;
9277	uint64_t lba;
9278	uint32_t num_blocks;
9279	int flags, retval;
9280
9281	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9282
9283	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9284
9285	flags = 0;
9286	retval = CTL_RETVAL_COMPLETE;
9287
9288	switch (ctsio->cdb[0]) {
9289	case COMPARE_AND_WRITE: {
9290		struct scsi_compare_and_write *cdb;
9291
9292		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9293		if (cdb->byte2 & SRW10_FUA)
9294			flags |= CTL_LLF_FUA;
9295		if (cdb->byte2 & SRW10_DPO)
9296			flags |= CTL_LLF_DPO;
9297		lba = scsi_8btou64(cdb->addr);
9298		num_blocks = cdb->length;
9299		break;
9300	}
9301	default:
9302		/*
9303		 * We got a command we don't support.  This shouldn't
9304		 * happen, commands should be filtered out above us.
9305		 */
9306		ctl_set_invalid_opcode(ctsio);
9307		ctl_done((union ctl_io *)ctsio);
9308
9309		return (CTL_RETVAL_COMPLETE);
9310		break; /* NOTREACHED */
9311	}
9312
9313	/*
9314	 * The first check is to make sure we're in bounds, the second
9315	 * check is to catch wrap-around problems.  If the lba + num blocks
9316	 * is less than the lba, then we've wrapped around and the block
9317	 * range is invalid anyway.
9318	 */
9319	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9320	 || ((lba + num_blocks) < lba)) {
9321		ctl_set_lba_out_of_range(ctsio);
9322		ctl_done((union ctl_io *)ctsio);
9323		return (CTL_RETVAL_COMPLETE);
9324	}
9325
9326	/*
9327	 * According to SBC-3, a transfer length of 0 is not an error.
9328	 */
9329	if (num_blocks == 0) {
9330		ctl_set_success(ctsio);
9331		ctl_done((union ctl_io *)ctsio);
9332		return (CTL_RETVAL_COMPLETE);
9333	}
9334
9335	/* Set FUA if write cache is disabled. */
9336	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9337	    SCP_WCE) == 0)
9338		flags |= CTL_LLF_FUA;
9339
9340	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9341	ctsio->kern_rel_offset = 0;
9342
9343	/*
9344	 * Set the IO_CONT flag, so that if this I/O gets passed to
9345	 * ctl_data_submit_done(), it'll get passed back to
9346	 * ctl_ctl_cnw_cont() for further processing.
9347	 */
9348	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9349	ctsio->io_cont = ctl_cnw_cont;
9350
9351	lbalen = (struct ctl_lba_len_flags *)
9352	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9353	lbalen->lba = lba;
9354	lbalen->len = num_blocks;
9355	lbalen->flags = CTL_LLF_COMPARE | flags;
9356
9357	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9358	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9359	return (retval);
9360}
9361
9362int
9363ctl_verify(struct ctl_scsiio *ctsio)
9364{
9365	struct ctl_lun *lun;
9366	struct ctl_lba_len_flags *lbalen;
9367	uint64_t lba;
9368	uint32_t num_blocks;
9369	int bytchk, flags;
9370	int retval;
9371
9372	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9373
9374	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9375
9376	bytchk = 0;
9377	flags = CTL_LLF_FUA;
9378	retval = CTL_RETVAL_COMPLETE;
9379
9380	switch (ctsio->cdb[0]) {
9381	case VERIFY_10: {
9382		struct scsi_verify_10 *cdb;
9383
9384		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9385		if (cdb->byte2 & SVFY_BYTCHK)
9386			bytchk = 1;
9387		if (cdb->byte2 & SVFY_DPO)
9388			flags |= CTL_LLF_DPO;
9389		lba = scsi_4btoul(cdb->addr);
9390		num_blocks = scsi_2btoul(cdb->length);
9391		break;
9392	}
9393	case VERIFY_12: {
9394		struct scsi_verify_12 *cdb;
9395
9396		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9397		if (cdb->byte2 & SVFY_BYTCHK)
9398			bytchk = 1;
9399		if (cdb->byte2 & SVFY_DPO)
9400			flags |= CTL_LLF_DPO;
9401		lba = scsi_4btoul(cdb->addr);
9402		num_blocks = scsi_4btoul(cdb->length);
9403		break;
9404	}
9405	case VERIFY_16: {
9406		struct scsi_rw_16 *cdb;
9407
9408		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9409		if (cdb->byte2 & SVFY_BYTCHK)
9410			bytchk = 1;
9411		if (cdb->byte2 & SVFY_DPO)
9412			flags |= CTL_LLF_DPO;
9413		lba = scsi_8btou64(cdb->addr);
9414		num_blocks = scsi_4btoul(cdb->length);
9415		break;
9416	}
9417	default:
9418		/*
9419		 * We got a command we don't support.  This shouldn't
9420		 * happen, commands should be filtered out above us.
9421		 */
9422		ctl_set_invalid_opcode(ctsio);
9423		ctl_done((union ctl_io *)ctsio);
9424		return (CTL_RETVAL_COMPLETE);
9425	}
9426
9427	/*
9428	 * The first check is to make sure we're in bounds, the second
9429	 * check is to catch wrap-around problems.  If the lba + num blocks
9430	 * is less than the lba, then we've wrapped around and the block
9431	 * range is invalid anyway.
9432	 */
9433	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9434	 || ((lba + num_blocks) < lba)) {
9435		ctl_set_lba_out_of_range(ctsio);
9436		ctl_done((union ctl_io *)ctsio);
9437		return (CTL_RETVAL_COMPLETE);
9438	}
9439
9440	/*
9441	 * According to SBC-3, a transfer length of 0 is not an error.
9442	 */
9443	if (num_blocks == 0) {
9444		ctl_set_success(ctsio);
9445		ctl_done((union ctl_io *)ctsio);
9446		return (CTL_RETVAL_COMPLETE);
9447	}
9448
9449	lbalen = (struct ctl_lba_len_flags *)
9450	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9451	lbalen->lba = lba;
9452	lbalen->len = num_blocks;
9453	if (bytchk) {
9454		lbalen->flags = CTL_LLF_COMPARE | flags;
9455		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9456	} else {
9457		lbalen->flags = CTL_LLF_VERIFY | flags;
9458		ctsio->kern_total_len = 0;
9459	}
9460	ctsio->kern_rel_offset = 0;
9461
9462	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9463	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9464	return (retval);
9465}
9466
9467int
9468ctl_report_luns(struct ctl_scsiio *ctsio)
9469{
9470	struct scsi_report_luns *cdb;
9471	struct scsi_report_luns_data *lun_data;
9472	struct ctl_lun *lun, *request_lun;
9473	int num_luns, retval;
9474	uint32_t alloc_len, lun_datalen;
9475	int num_filled, well_known;
9476	uint32_t initidx, targ_lun_id, lun_id;
9477
9478	retval = CTL_RETVAL_COMPLETE;
9479	well_known = 0;
9480
9481	cdb = (struct scsi_report_luns *)ctsio->cdb;
9482
9483	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9484
9485	mtx_lock(&control_softc->ctl_lock);
9486	num_luns = control_softc->num_luns;
9487	mtx_unlock(&control_softc->ctl_lock);
9488
9489	switch (cdb->select_report) {
9490	case RPL_REPORT_DEFAULT:
9491	case RPL_REPORT_ALL:
9492		break;
9493	case RPL_REPORT_WELLKNOWN:
9494		well_known = 1;
9495		num_luns = 0;
9496		break;
9497	default:
9498		ctl_set_invalid_field(ctsio,
9499				      /*sks_valid*/ 1,
9500				      /*command*/ 1,
9501				      /*field*/ 2,
9502				      /*bit_valid*/ 0,
9503				      /*bit*/ 0);
9504		ctl_done((union ctl_io *)ctsio);
9505		return (retval);
9506		break; /* NOTREACHED */
9507	}
9508
9509	alloc_len = scsi_4btoul(cdb->length);
9510	/*
9511	 * The initiator has to allocate at least 16 bytes for this request,
9512	 * so he can at least get the header and the first LUN.  Otherwise
9513	 * we reject the request (per SPC-3 rev 14, section 6.21).
9514	 */
9515	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9516	    sizeof(struct scsi_report_luns_lundata))) {
9517		ctl_set_invalid_field(ctsio,
9518				      /*sks_valid*/ 1,
9519				      /*command*/ 1,
9520				      /*field*/ 6,
9521				      /*bit_valid*/ 0,
9522				      /*bit*/ 0);
9523		ctl_done((union ctl_io *)ctsio);
9524		return (retval);
9525	}
9526
9527	request_lun = (struct ctl_lun *)
9528		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9529
9530	lun_datalen = sizeof(*lun_data) +
9531		(num_luns * sizeof(struct scsi_report_luns_lundata));
9532
9533	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9534	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9535	ctsio->kern_sg_entries = 0;
9536
9537	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9538
9539	mtx_lock(&control_softc->ctl_lock);
9540	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9541		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9542		if (lun_id >= CTL_MAX_LUNS)
9543			continue;
9544		lun = control_softc->ctl_luns[lun_id];
9545		if (lun == NULL)
9546			continue;
9547
9548		if (targ_lun_id <= 0xff) {
9549			/*
9550			 * Peripheral addressing method, bus number 0.
9551			 */
9552			lun_data->luns[num_filled].lundata[0] =
9553				RPL_LUNDATA_ATYP_PERIPH;
9554			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9555			num_filled++;
9556		} else if (targ_lun_id <= 0x3fff) {
9557			/*
9558			 * Flat addressing method.
9559			 */
9560			lun_data->luns[num_filled].lundata[0] =
9561				RPL_LUNDATA_ATYP_FLAT |
9562				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9563#ifdef OLDCTLHEADERS
9564				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9565				(targ_lun_id & SRLD_BUS_LUN_MASK);
9566#endif
9567			lun_data->luns[num_filled].lundata[1] =
9568#ifdef OLDCTLHEADERS
9569				targ_lun_id >> SRLD_BUS_LUN_BITS;
9570#endif
9571				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9572			num_filled++;
9573		} else {
9574			printf("ctl_report_luns: bogus LUN number %jd, "
9575			       "skipping\n", (intmax_t)targ_lun_id);
9576		}
9577		/*
9578		 * According to SPC-3, rev 14 section 6.21:
9579		 *
9580		 * "The execution of a REPORT LUNS command to any valid and
9581		 * installed logical unit shall clear the REPORTED LUNS DATA
9582		 * HAS CHANGED unit attention condition for all logical
9583		 * units of that target with respect to the requesting
9584		 * initiator. A valid and installed logical unit is one
9585		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9586		 * INQUIRY data (see 6.4.2)."
9587		 *
9588		 * If request_lun is NULL, the LUN this report luns command
9589		 * was issued to is either disabled or doesn't exist. In that
9590		 * case, we shouldn't clear any pending lun change unit
9591		 * attention.
9592		 */
9593		if (request_lun != NULL) {
9594			mtx_lock(&lun->lun_lock);
9595			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9596			mtx_unlock(&lun->lun_lock);
9597		}
9598	}
9599	mtx_unlock(&control_softc->ctl_lock);
9600
9601	/*
9602	 * It's quite possible that we've returned fewer LUNs than we allocated
9603	 * space for.  Trim it.
9604	 */
9605	lun_datalen = sizeof(*lun_data) +
9606		(num_filled * sizeof(struct scsi_report_luns_lundata));
9607
9608	if (lun_datalen < alloc_len) {
9609		ctsio->residual = alloc_len - lun_datalen;
9610		ctsio->kern_data_len = lun_datalen;
9611		ctsio->kern_total_len = lun_datalen;
9612	} else {
9613		ctsio->residual = 0;
9614		ctsio->kern_data_len = alloc_len;
9615		ctsio->kern_total_len = alloc_len;
9616	}
9617	ctsio->kern_data_resid = 0;
9618	ctsio->kern_rel_offset = 0;
9619	ctsio->kern_sg_entries = 0;
9620
9621	/*
9622	 * We set this to the actual data length, regardless of how much
9623	 * space we actually have to return results.  If the user looks at
9624	 * this value, he'll know whether or not he allocated enough space
9625	 * and reissue the command if necessary.  We don't support well
9626	 * known logical units, so if the user asks for that, return none.
9627	 */
9628	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9629
9630	/*
9631	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9632	 * this request.
9633	 */
9634	ctsio->scsi_status = SCSI_STATUS_OK;
9635
9636	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9637	ctsio->be_move_done = ctl_config_move_done;
9638	ctl_datamove((union ctl_io *)ctsio);
9639
9640	return (retval);
9641}
9642
9643int
9644ctl_request_sense(struct ctl_scsiio *ctsio)
9645{
9646	struct scsi_request_sense *cdb;
9647	struct scsi_sense_data *sense_ptr;
9648	struct ctl_lun *lun;
9649	uint32_t initidx;
9650	int have_error;
9651	scsi_sense_data_type sense_format;
9652
9653	cdb = (struct scsi_request_sense *)ctsio->cdb;
9654
9655	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9656
9657	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9658
9659	/*
9660	 * Determine which sense format the user wants.
9661	 */
9662	if (cdb->byte2 & SRS_DESC)
9663		sense_format = SSD_TYPE_DESC;
9664	else
9665		sense_format = SSD_TYPE_FIXED;
9666
9667	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9668	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9669	ctsio->kern_sg_entries = 0;
9670
9671	/*
9672	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9673	 * larger than the largest allowed value for the length field in the
9674	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9675	 */
9676	ctsio->residual = 0;
9677	ctsio->kern_data_len = cdb->length;
9678	ctsio->kern_total_len = cdb->length;
9679
9680	ctsio->kern_data_resid = 0;
9681	ctsio->kern_rel_offset = 0;
9682	ctsio->kern_sg_entries = 0;
9683
9684	/*
9685	 * If we don't have a LUN, we don't have any pending sense.
9686	 */
9687	if (lun == NULL)
9688		goto no_sense;
9689
9690	have_error = 0;
9691	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9692	/*
9693	 * Check for pending sense, and then for pending unit attentions.
9694	 * Pending sense gets returned first, then pending unit attentions.
9695	 */
9696	mtx_lock(&lun->lun_lock);
9697#ifdef CTL_WITH_CA
9698	if (ctl_is_set(lun->have_ca, initidx)) {
9699		scsi_sense_data_type stored_format;
9700
9701		/*
9702		 * Check to see which sense format was used for the stored
9703		 * sense data.
9704		 */
9705		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9706
9707		/*
9708		 * If the user requested a different sense format than the
9709		 * one we stored, then we need to convert it to the other
9710		 * format.  If we're going from descriptor to fixed format
9711		 * sense data, we may lose things in translation, depending
9712		 * on what options were used.
9713		 *
9714		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9715		 * for some reason we'll just copy it out as-is.
9716		 */
9717		if ((stored_format == SSD_TYPE_FIXED)
9718		 && (sense_format == SSD_TYPE_DESC))
9719			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9720			    &lun->pending_sense[initidx],
9721			    (struct scsi_sense_data_desc *)sense_ptr);
9722		else if ((stored_format == SSD_TYPE_DESC)
9723		      && (sense_format == SSD_TYPE_FIXED))
9724			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9725			    &lun->pending_sense[initidx],
9726			    (struct scsi_sense_data_fixed *)sense_ptr);
9727		else
9728			memcpy(sense_ptr, &lun->pending_sense[initidx],
9729			       ctl_min(sizeof(*sense_ptr),
9730			       sizeof(lun->pending_sense[initidx])));
9731
9732		ctl_clear_mask(lun->have_ca, initidx);
9733		have_error = 1;
9734	} else
9735#endif
9736	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9737		ctl_ua_type ua_type;
9738
9739		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9740				       sense_ptr, sense_format);
9741		if (ua_type != CTL_UA_NONE)
9742			have_error = 1;
9743	}
9744	mtx_unlock(&lun->lun_lock);
9745
9746	/*
9747	 * We already have a pending error, return it.
9748	 */
9749	if (have_error != 0) {
9750		/*
9751		 * We report the SCSI status as OK, since the status of the
9752		 * request sense command itself is OK.
9753		 */
9754		ctsio->scsi_status = SCSI_STATUS_OK;
9755
9756		/*
9757		 * We report 0 for the sense length, because we aren't doing
9758		 * autosense in this case.  We're reporting sense as
9759		 * parameter data.
9760		 */
9761		ctsio->sense_len = 0;
9762		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9763		ctsio->be_move_done = ctl_config_move_done;
9764		ctl_datamove((union ctl_io *)ctsio);
9765
9766		return (CTL_RETVAL_COMPLETE);
9767	}
9768
9769no_sense:
9770
9771	/*
9772	 * No sense information to report, so we report that everything is
9773	 * okay.
9774	 */
9775	ctl_set_sense_data(sense_ptr,
9776			   lun,
9777			   sense_format,
9778			   /*current_error*/ 1,
9779			   /*sense_key*/ SSD_KEY_NO_SENSE,
9780			   /*asc*/ 0x00,
9781			   /*ascq*/ 0x00,
9782			   SSD_ELEM_NONE);
9783
9784	ctsio->scsi_status = SCSI_STATUS_OK;
9785
9786	/*
9787	 * We report 0 for the sense length, because we aren't doing
9788	 * autosense in this case.  We're reporting sense as parameter data.
9789	 */
9790	ctsio->sense_len = 0;
9791	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9792	ctsio->be_move_done = ctl_config_move_done;
9793	ctl_datamove((union ctl_io *)ctsio);
9794
9795	return (CTL_RETVAL_COMPLETE);
9796}
9797
9798int
9799ctl_tur(struct ctl_scsiio *ctsio)
9800{
9801	struct ctl_lun *lun;
9802
9803	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9804
9805	CTL_DEBUG_PRINT(("ctl_tur\n"));
9806
9807	if (lun == NULL)
9808		return (EINVAL);
9809
9810	ctsio->scsi_status = SCSI_STATUS_OK;
9811	ctsio->io_hdr.status = CTL_SUCCESS;
9812
9813	ctl_done((union ctl_io *)ctsio);
9814
9815	return (CTL_RETVAL_COMPLETE);
9816}
9817
9818#ifdef notyet
9819static int
9820ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9821{
9822
9823}
9824#endif
9825
9826static int
9827ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9828{
9829	struct scsi_vpd_supported_pages *pages;
9830	int sup_page_size;
9831	struct ctl_lun *lun;
9832
9833	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9834
9835	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9836	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9837	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9838	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9839	ctsio->kern_sg_entries = 0;
9840
9841	if (sup_page_size < alloc_len) {
9842		ctsio->residual = alloc_len - sup_page_size;
9843		ctsio->kern_data_len = sup_page_size;
9844		ctsio->kern_total_len = sup_page_size;
9845	} else {
9846		ctsio->residual = 0;
9847		ctsio->kern_data_len = alloc_len;
9848		ctsio->kern_total_len = alloc_len;
9849	}
9850	ctsio->kern_data_resid = 0;
9851	ctsio->kern_rel_offset = 0;
9852	ctsio->kern_sg_entries = 0;
9853
9854	/*
9855	 * The control device is always connected.  The disk device, on the
9856	 * other hand, may not be online all the time.  Need to change this
9857	 * to figure out whether the disk device is actually online or not.
9858	 */
9859	if (lun != NULL)
9860		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9861				lun->be_lun->lun_type;
9862	else
9863		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9864
9865	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9866	/* Supported VPD pages */
9867	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9868	/* Serial Number */
9869	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9870	/* Device Identification */
9871	pages->page_list[2] = SVPD_DEVICE_ID;
9872	/* Extended INQUIRY Data */
9873	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9874	/* Mode Page Policy */
9875	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9876	/* SCSI Ports */
9877	pages->page_list[5] = SVPD_SCSI_PORTS;
9878	/* Third-party Copy */
9879	pages->page_list[6] = SVPD_SCSI_TPC;
9880	/* Block limits */
9881	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9882	/* Block Device Characteristics */
9883	pages->page_list[8] = SVPD_BDC;
9884	/* Logical Block Provisioning */
9885	pages->page_list[9] = SVPD_LBP;
9886
9887	ctsio->scsi_status = SCSI_STATUS_OK;
9888
9889	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9890	ctsio->be_move_done = ctl_config_move_done;
9891	ctl_datamove((union ctl_io *)ctsio);
9892
9893	return (CTL_RETVAL_COMPLETE);
9894}
9895
9896static int
9897ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9898{
9899	struct scsi_vpd_unit_serial_number *sn_ptr;
9900	struct ctl_lun *lun;
9901
9902	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9903
9904	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9905	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9906	ctsio->kern_sg_entries = 0;
9907
9908	if (sizeof(*sn_ptr) < alloc_len) {
9909		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9910		ctsio->kern_data_len = sizeof(*sn_ptr);
9911		ctsio->kern_total_len = sizeof(*sn_ptr);
9912	} else {
9913		ctsio->residual = 0;
9914		ctsio->kern_data_len = alloc_len;
9915		ctsio->kern_total_len = alloc_len;
9916	}
9917	ctsio->kern_data_resid = 0;
9918	ctsio->kern_rel_offset = 0;
9919	ctsio->kern_sg_entries = 0;
9920
9921	/*
9922	 * The control device is always connected.  The disk device, on the
9923	 * other hand, may not be online all the time.  Need to change this
9924	 * to figure out whether the disk device is actually online or not.
9925	 */
9926	if (lun != NULL)
9927		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9928				  lun->be_lun->lun_type;
9929	else
9930		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9931
9932	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9933	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9934	/*
9935	 * If we don't have a LUN, we just leave the serial number as
9936	 * all spaces.
9937	 */
9938	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9939	if (lun != NULL) {
9940		strncpy((char *)sn_ptr->serial_num,
9941			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9942	}
9943	ctsio->scsi_status = SCSI_STATUS_OK;
9944
9945	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9946	ctsio->be_move_done = ctl_config_move_done;
9947	ctl_datamove((union ctl_io *)ctsio);
9948
9949	return (CTL_RETVAL_COMPLETE);
9950}
9951
9952
9953static int
9954ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9955{
9956	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9957	struct ctl_lun *lun;
9958	int data_len;
9959
9960	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9961
9962	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9963	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9964	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9965	ctsio->kern_sg_entries = 0;
9966
9967	if (data_len < alloc_len) {
9968		ctsio->residual = alloc_len - data_len;
9969		ctsio->kern_data_len = data_len;
9970		ctsio->kern_total_len = data_len;
9971	} else {
9972		ctsio->residual = 0;
9973		ctsio->kern_data_len = alloc_len;
9974		ctsio->kern_total_len = alloc_len;
9975	}
9976	ctsio->kern_data_resid = 0;
9977	ctsio->kern_rel_offset = 0;
9978	ctsio->kern_sg_entries = 0;
9979
9980	/*
9981	 * The control device is always connected.  The disk device, on the
9982	 * other hand, may not be online all the time.
9983	 */
9984	if (lun != NULL)
9985		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9986				     lun->be_lun->lun_type;
9987	else
9988		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9989	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9990	eid_ptr->page_length = data_len - 4;
9991	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9992	eid_ptr->flags3 = SVPD_EID_V_SUP;
9993
9994	ctsio->scsi_status = SCSI_STATUS_OK;
9995	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9996	ctsio->be_move_done = ctl_config_move_done;
9997	ctl_datamove((union ctl_io *)ctsio);
9998
9999	return (CTL_RETVAL_COMPLETE);
10000}
10001
10002static int
10003ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
10004{
10005	struct scsi_vpd_mode_page_policy *mpp_ptr;
10006	struct ctl_lun *lun;
10007	int data_len;
10008
10009	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10010
10011	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
10012	    sizeof(struct scsi_vpd_mode_page_policy_descr);
10013
10014	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10015	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
10016	ctsio->kern_sg_entries = 0;
10017
10018	if (data_len < alloc_len) {
10019		ctsio->residual = alloc_len - data_len;
10020		ctsio->kern_data_len = data_len;
10021		ctsio->kern_total_len = data_len;
10022	} else {
10023		ctsio->residual = 0;
10024		ctsio->kern_data_len = alloc_len;
10025		ctsio->kern_total_len = alloc_len;
10026	}
10027	ctsio->kern_data_resid = 0;
10028	ctsio->kern_rel_offset = 0;
10029	ctsio->kern_sg_entries = 0;
10030
10031	/*
10032	 * The control device is always connected.  The disk device, on the
10033	 * other hand, may not be online all the time.
10034	 */
10035	if (lun != NULL)
10036		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10037				     lun->be_lun->lun_type;
10038	else
10039		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10040	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
10041	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
10042	mpp_ptr->descr[0].page_code = 0x3f;
10043	mpp_ptr->descr[0].subpage_code = 0xff;
10044	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
10045
10046	ctsio->scsi_status = SCSI_STATUS_OK;
10047	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10048	ctsio->be_move_done = ctl_config_move_done;
10049	ctl_datamove((union ctl_io *)ctsio);
10050
10051	return (CTL_RETVAL_COMPLETE);
10052}
10053
10054static int
10055ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
10056{
10057	struct scsi_vpd_device_id *devid_ptr;
10058	struct scsi_vpd_id_descriptor *desc;
10059	struct ctl_softc *ctl_softc;
10060	struct ctl_lun *lun;
10061	struct ctl_port *port;
10062	int data_len;
10063	uint8_t proto;
10064
10065	ctl_softc = control_softc;
10066
10067	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
10068	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10069
10070	data_len = sizeof(struct scsi_vpd_device_id) +
10071	    sizeof(struct scsi_vpd_id_descriptor) +
10072		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
10073	    sizeof(struct scsi_vpd_id_descriptor) +
10074		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
10075	if (lun && lun->lun_devid)
10076		data_len += lun->lun_devid->len;
10077	if (port->port_devid)
10078		data_len += port->port_devid->len;
10079	if (port->target_devid)
10080		data_len += port->target_devid->len;
10081
10082	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10083	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10084	ctsio->kern_sg_entries = 0;
10085
10086	if (data_len < alloc_len) {
10087		ctsio->residual = alloc_len - data_len;
10088		ctsio->kern_data_len = data_len;
10089		ctsio->kern_total_len = data_len;
10090	} else {
10091		ctsio->residual = 0;
10092		ctsio->kern_data_len = alloc_len;
10093		ctsio->kern_total_len = alloc_len;
10094	}
10095	ctsio->kern_data_resid = 0;
10096	ctsio->kern_rel_offset = 0;
10097	ctsio->kern_sg_entries = 0;
10098
10099	/*
10100	 * The control device is always connected.  The disk device, on the
10101	 * other hand, may not be online all the time.
10102	 */
10103	if (lun != NULL)
10104		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10105				     lun->be_lun->lun_type;
10106	else
10107		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10108	devid_ptr->page_code = SVPD_DEVICE_ID;
10109	scsi_ulto2b(data_len - 4, devid_ptr->length);
10110
10111	if (port->port_type == CTL_PORT_FC)
10112		proto = SCSI_PROTO_FC << 4;
10113	else if (port->port_type == CTL_PORT_ISCSI)
10114		proto = SCSI_PROTO_ISCSI << 4;
10115	else
10116		proto = SCSI_PROTO_SPI << 4;
10117	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10118
10119	/*
10120	 * We're using a LUN association here.  i.e., this device ID is a
10121	 * per-LUN identifier.
10122	 */
10123	if (lun && lun->lun_devid) {
10124		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10125		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10126		    lun->lun_devid->len);
10127	}
10128
10129	/*
10130	 * This is for the WWPN which is a port association.
10131	 */
10132	if (port->port_devid) {
10133		memcpy(desc, port->port_devid->data, port->port_devid->len);
10134		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10135		    port->port_devid->len);
10136	}
10137
10138	/*
10139	 * This is for the Relative Target Port(type 4h) identifier
10140	 */
10141	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10142	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10143	    SVPD_ID_TYPE_RELTARG;
10144	desc->length = 4;
10145	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10146	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10147	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10148
10149	/*
10150	 * This is for the Target Port Group(type 5h) identifier
10151	 */
10152	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10153	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10154	    SVPD_ID_TYPE_TPORTGRP;
10155	desc->length = 4;
10156	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10157	    &desc->identifier[2]);
10158	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10159	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10160
10161	/*
10162	 * This is for the Target identifier
10163	 */
10164	if (port->target_devid) {
10165		memcpy(desc, port->target_devid->data, port->target_devid->len);
10166	}
10167
10168	ctsio->scsi_status = SCSI_STATUS_OK;
10169	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10170	ctsio->be_move_done = ctl_config_move_done;
10171	ctl_datamove((union ctl_io *)ctsio);
10172
10173	return (CTL_RETVAL_COMPLETE);
10174}
10175
10176static int
10177ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10178{
10179	struct ctl_softc *softc = control_softc;
10180	struct scsi_vpd_scsi_ports *sp;
10181	struct scsi_vpd_port_designation *pd;
10182	struct scsi_vpd_port_designation_cont *pdc;
10183	struct ctl_lun *lun;
10184	struct ctl_port *port;
10185	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10186	int num_target_port_groups, single;
10187
10188	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10189
10190	single = ctl_is_single;
10191	if (single)
10192		num_target_port_groups = 1;
10193	else
10194		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10195	num_target_ports = 0;
10196	iid_len = 0;
10197	id_len = 0;
10198	mtx_lock(&softc->ctl_lock);
10199	STAILQ_FOREACH(port, &softc->port_list, links) {
10200		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10201			continue;
10202		if (lun != NULL &&
10203		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10204		    CTL_MAX_LUNS)
10205			continue;
10206		num_target_ports++;
10207		if (port->init_devid)
10208			iid_len += port->init_devid->len;
10209		if (port->port_devid)
10210			id_len += port->port_devid->len;
10211	}
10212	mtx_unlock(&softc->ctl_lock);
10213
10214	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10215	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10216	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10217	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10218	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10219	ctsio->kern_sg_entries = 0;
10220
10221	if (data_len < alloc_len) {
10222		ctsio->residual = alloc_len - data_len;
10223		ctsio->kern_data_len = data_len;
10224		ctsio->kern_total_len = data_len;
10225	} else {
10226		ctsio->residual = 0;
10227		ctsio->kern_data_len = alloc_len;
10228		ctsio->kern_total_len = alloc_len;
10229	}
10230	ctsio->kern_data_resid = 0;
10231	ctsio->kern_rel_offset = 0;
10232	ctsio->kern_sg_entries = 0;
10233
10234	/*
10235	 * The control device is always connected.  The disk device, on the
10236	 * other hand, may not be online all the time.  Need to change this
10237	 * to figure out whether the disk device is actually online or not.
10238	 */
10239	if (lun != NULL)
10240		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10241				  lun->be_lun->lun_type;
10242	else
10243		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10244
10245	sp->page_code = SVPD_SCSI_PORTS;
10246	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10247	    sp->page_length);
10248	pd = &sp->design[0];
10249
10250	mtx_lock(&softc->ctl_lock);
10251	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10252		pg = 0;
10253	else
10254		pg = 1;
10255	for (g = 0; g < num_target_port_groups; g++) {
10256		STAILQ_FOREACH(port, &softc->port_list, links) {
10257			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10258				continue;
10259			if (lun != NULL &&
10260			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10261			    CTL_MAX_LUNS)
10262				continue;
10263			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10264			scsi_ulto2b(p, pd->relative_port_id);
10265			if (port->init_devid && g == pg) {
10266				iid_len = port->init_devid->len;
10267				memcpy(pd->initiator_transportid,
10268				    port->init_devid->data, port->init_devid->len);
10269			} else
10270				iid_len = 0;
10271			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10272			pdc = (struct scsi_vpd_port_designation_cont *)
10273			    (&pd->initiator_transportid[iid_len]);
10274			if (port->port_devid && g == pg) {
10275				id_len = port->port_devid->len;
10276				memcpy(pdc->target_port_descriptors,
10277				    port->port_devid->data, port->port_devid->len);
10278			} else
10279				id_len = 0;
10280			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10281			pd = (struct scsi_vpd_port_designation *)
10282			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10283		}
10284	}
10285	mtx_unlock(&softc->ctl_lock);
10286
10287	ctsio->scsi_status = SCSI_STATUS_OK;
10288	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10289	ctsio->be_move_done = ctl_config_move_done;
10290	ctl_datamove((union ctl_io *)ctsio);
10291
10292	return (CTL_RETVAL_COMPLETE);
10293}
10294
10295static int
10296ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10297{
10298	struct scsi_vpd_block_limits *bl_ptr;
10299	struct ctl_lun *lun;
10300	int bs;
10301
10302	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10303
10304	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10305	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10306	ctsio->kern_sg_entries = 0;
10307
10308	if (sizeof(*bl_ptr) < alloc_len) {
10309		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10310		ctsio->kern_data_len = sizeof(*bl_ptr);
10311		ctsio->kern_total_len = sizeof(*bl_ptr);
10312	} else {
10313		ctsio->residual = 0;
10314		ctsio->kern_data_len = alloc_len;
10315		ctsio->kern_total_len = alloc_len;
10316	}
10317	ctsio->kern_data_resid = 0;
10318	ctsio->kern_rel_offset = 0;
10319	ctsio->kern_sg_entries = 0;
10320
10321	/*
10322	 * The control device is always connected.  The disk device, on the
10323	 * other hand, may not be online all the time.  Need to change this
10324	 * to figure out whether the disk device is actually online or not.
10325	 */
10326	if (lun != NULL)
10327		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10328				  lun->be_lun->lun_type;
10329	else
10330		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10331
10332	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10333	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10334	bl_ptr->max_cmp_write_len = 0xff;
10335	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10336	if (lun != NULL) {
10337		bs = lun->be_lun->blocksize;
10338		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10339		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10340			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10341			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10342			if (lun->be_lun->pblockexp != 0) {
10343				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10344				    bl_ptr->opt_unmap_grain);
10345				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10346				    bl_ptr->unmap_grain_align);
10347			}
10348		}
10349		scsi_ulto4b(lun->be_lun->atomicblock,
10350		    bl_ptr->max_atomic_transfer_length);
10351		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10352		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10353	}
10354	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10355
10356	ctsio->scsi_status = SCSI_STATUS_OK;
10357	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10358	ctsio->be_move_done = ctl_config_move_done;
10359	ctl_datamove((union ctl_io *)ctsio);
10360
10361	return (CTL_RETVAL_COMPLETE);
10362}
10363
10364static int
10365ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10366{
10367	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10368	struct ctl_lun *lun;
10369
10370	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10371
10372	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10373	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10374	ctsio->kern_sg_entries = 0;
10375
10376	if (sizeof(*bdc_ptr) < alloc_len) {
10377		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10378		ctsio->kern_data_len = sizeof(*bdc_ptr);
10379		ctsio->kern_total_len = sizeof(*bdc_ptr);
10380	} else {
10381		ctsio->residual = 0;
10382		ctsio->kern_data_len = alloc_len;
10383		ctsio->kern_total_len = alloc_len;
10384	}
10385	ctsio->kern_data_resid = 0;
10386	ctsio->kern_rel_offset = 0;
10387	ctsio->kern_sg_entries = 0;
10388
10389	/*
10390	 * The control device is always connected.  The disk device, on the
10391	 * other hand, may not be online all the time.  Need to change this
10392	 * to figure out whether the disk device is actually online or not.
10393	 */
10394	if (lun != NULL)
10395		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10396				  lun->be_lun->lun_type;
10397	else
10398		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10399	bdc_ptr->page_code = SVPD_BDC;
10400	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10401	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10402	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10403
10404	ctsio->scsi_status = SCSI_STATUS_OK;
10405	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10406	ctsio->be_move_done = ctl_config_move_done;
10407	ctl_datamove((union ctl_io *)ctsio);
10408
10409	return (CTL_RETVAL_COMPLETE);
10410}
10411
10412static int
10413ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10414{
10415	struct scsi_vpd_logical_block_prov *lbp_ptr;
10416	struct ctl_lun *lun;
10417
10418	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10419
10420	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10421	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10422	ctsio->kern_sg_entries = 0;
10423
10424	if (sizeof(*lbp_ptr) < alloc_len) {
10425		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10426		ctsio->kern_data_len = sizeof(*lbp_ptr);
10427		ctsio->kern_total_len = sizeof(*lbp_ptr);
10428	} else {
10429		ctsio->residual = 0;
10430		ctsio->kern_data_len = alloc_len;
10431		ctsio->kern_total_len = alloc_len;
10432	}
10433	ctsio->kern_data_resid = 0;
10434	ctsio->kern_rel_offset = 0;
10435	ctsio->kern_sg_entries = 0;
10436
10437	/*
10438	 * The control device is always connected.  The disk device, on the
10439	 * other hand, may not be online all the time.  Need to change this
10440	 * to figure out whether the disk device is actually online or not.
10441	 */
10442	if (lun != NULL)
10443		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10444				  lun->be_lun->lun_type;
10445	else
10446		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10447
10448	lbp_ptr->page_code = SVPD_LBP;
10449	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10450	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10451		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10452		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10453		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10454	}
10455
10456	ctsio->scsi_status = SCSI_STATUS_OK;
10457	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10458	ctsio->be_move_done = ctl_config_move_done;
10459	ctl_datamove((union ctl_io *)ctsio);
10460
10461	return (CTL_RETVAL_COMPLETE);
10462}
10463
10464static int
10465ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10466{
10467	struct scsi_inquiry *cdb;
10468	struct ctl_lun *lun;
10469	int alloc_len, retval;
10470
10471	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10472	cdb = (struct scsi_inquiry *)ctsio->cdb;
10473
10474	retval = CTL_RETVAL_COMPLETE;
10475
10476	alloc_len = scsi_2btoul(cdb->length);
10477
10478	switch (cdb->page_code) {
10479	case SVPD_SUPPORTED_PAGES:
10480		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10481		break;
10482	case SVPD_UNIT_SERIAL_NUMBER:
10483		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10484		break;
10485	case SVPD_DEVICE_ID:
10486		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10487		break;
10488	case SVPD_EXTENDED_INQUIRY_DATA:
10489		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10490		break;
10491	case SVPD_MODE_PAGE_POLICY:
10492		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10493		break;
10494	case SVPD_SCSI_PORTS:
10495		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10496		break;
10497	case SVPD_SCSI_TPC:
10498		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10499		break;
10500	case SVPD_BLOCK_LIMITS:
10501		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10502		break;
10503	case SVPD_BDC:
10504		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10505		break;
10506	case SVPD_LBP:
10507		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10508		break;
10509	default:
10510		ctl_set_invalid_field(ctsio,
10511				      /*sks_valid*/ 1,
10512				      /*command*/ 1,
10513				      /*field*/ 2,
10514				      /*bit_valid*/ 0,
10515				      /*bit*/ 0);
10516		ctl_done((union ctl_io *)ctsio);
10517		retval = CTL_RETVAL_COMPLETE;
10518		break;
10519	}
10520
10521	return (retval);
10522}
10523
10524static int
10525ctl_inquiry_std(struct ctl_scsiio *ctsio)
10526{
10527	struct scsi_inquiry_data *inq_ptr;
10528	struct scsi_inquiry *cdb;
10529	struct ctl_softc *ctl_softc;
10530	struct ctl_lun *lun;
10531	char *val;
10532	uint32_t alloc_len, data_len;
10533	ctl_port_type port_type;
10534
10535	ctl_softc = control_softc;
10536
10537	/*
10538	 * Figure out whether we're talking to a Fibre Channel port or not.
10539	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10540	 * SCSI front ends.
10541	 */
10542	port_type = ctl_softc->ctl_ports[
10543	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10544	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10545		port_type = CTL_PORT_SCSI;
10546
10547	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10548	cdb = (struct scsi_inquiry *)ctsio->cdb;
10549	alloc_len = scsi_2btoul(cdb->length);
10550
10551	/*
10552	 * We malloc the full inquiry data size here and fill it
10553	 * in.  If the user only asks for less, we'll give him
10554	 * that much.
10555	 */
10556	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10557	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10558	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10559	ctsio->kern_sg_entries = 0;
10560	ctsio->kern_data_resid = 0;
10561	ctsio->kern_rel_offset = 0;
10562
10563	if (data_len < alloc_len) {
10564		ctsio->residual = alloc_len - data_len;
10565		ctsio->kern_data_len = data_len;
10566		ctsio->kern_total_len = data_len;
10567	} else {
10568		ctsio->residual = 0;
10569		ctsio->kern_data_len = alloc_len;
10570		ctsio->kern_total_len = alloc_len;
10571	}
10572
10573	/*
10574	 * If we have a LUN configured, report it as connected.  Otherwise,
10575	 * report that it is offline or no device is supported, depending
10576	 * on the value of inquiry_pq_no_lun.
10577	 *
10578	 * According to the spec (SPC-4 r34), the peripheral qualifier
10579	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10580	 *
10581	 * "A peripheral device having the specified peripheral device type
10582	 * is not connected to this logical unit. However, the device
10583	 * server is capable of supporting the specified peripheral device
10584	 * type on this logical unit."
10585	 *
10586	 * According to the same spec, the peripheral qualifier
10587	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10588	 *
10589	 * "The device server is not capable of supporting a peripheral
10590	 * device on this logical unit. For this peripheral qualifier the
10591	 * peripheral device type shall be set to 1Fh. All other peripheral
10592	 * device type values are reserved for this peripheral qualifier."
10593	 *
10594	 * Given the text, it would seem that we probably want to report that
10595	 * the LUN is offline here.  There is no LUN connected, but we can
10596	 * support a LUN at the given LUN number.
10597	 *
10598	 * In the real world, though, it sounds like things are a little
10599	 * different:
10600	 *
10601	 * - Linux, when presented with a LUN with the offline peripheral
10602	 *   qualifier, will create an sg driver instance for it.  So when
10603	 *   you attach it to CTL, you wind up with a ton of sg driver
10604	 *   instances.  (One for every LUN that Linux bothered to probe.)
10605	 *   Linux does this despite the fact that it issues a REPORT LUNs
10606	 *   to LUN 0 to get the inventory of supported LUNs.
10607	 *
10608	 * - There is other anecdotal evidence (from Emulex folks) about
10609	 *   arrays that use the offline peripheral qualifier for LUNs that
10610	 *   are on the "passive" path in an active/passive array.
10611	 *
10612	 * So the solution is provide a hopefully reasonable default
10613	 * (return bad/no LUN) and allow the user to change the behavior
10614	 * with a tunable/sysctl variable.
10615	 */
10616	if (lun != NULL)
10617		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10618				  lun->be_lun->lun_type;
10619	else if (ctl_softc->inquiry_pq_no_lun == 0)
10620		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10621	else
10622		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10623
10624	/* RMB in byte 2 is 0 */
10625	inq_ptr->version = SCSI_REV_SPC4;
10626
10627	/*
10628	 * According to SAM-3, even if a device only supports a single
10629	 * level of LUN addressing, it should still set the HISUP bit:
10630	 *
10631	 * 4.9.1 Logical unit numbers overview
10632	 *
10633	 * All logical unit number formats described in this standard are
10634	 * hierarchical in structure even when only a single level in that
10635	 * hierarchy is used. The HISUP bit shall be set to one in the
10636	 * standard INQUIRY data (see SPC-2) when any logical unit number
10637	 * format described in this standard is used.  Non-hierarchical
10638	 * formats are outside the scope of this standard.
10639	 *
10640	 * Therefore we set the HiSup bit here.
10641	 *
10642	 * The reponse format is 2, per SPC-3.
10643	 */
10644	inq_ptr->response_format = SID_HiSup | 2;
10645
10646	inq_ptr->additional_length = data_len -
10647	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10648	CTL_DEBUG_PRINT(("additional_length = %d\n",
10649			 inq_ptr->additional_length));
10650
10651	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10652	/* 16 bit addressing */
10653	if (port_type == CTL_PORT_SCSI)
10654		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10655	/* XXX set the SID_MultiP bit here if we're actually going to
10656	   respond on multiple ports */
10657	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10658
10659	/* 16 bit data bus, synchronous transfers */
10660	if (port_type == CTL_PORT_SCSI)
10661		inq_ptr->flags = SID_WBus16 | SID_Sync;
10662	/*
10663	 * XXX KDM do we want to support tagged queueing on the control
10664	 * device at all?
10665	 */
10666	if ((lun == NULL)
10667	 || (lun->be_lun->lun_type != T_PROCESSOR))
10668		inq_ptr->flags |= SID_CmdQue;
10669	/*
10670	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10671	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10672	 * name and 4 bytes for the revision.
10673	 */
10674	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10675	    "vendor")) == NULL) {
10676		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10677	} else {
10678		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10679		strncpy(inq_ptr->vendor, val,
10680		    min(sizeof(inq_ptr->vendor), strlen(val)));
10681	}
10682	if (lun == NULL) {
10683		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10684		    sizeof(inq_ptr->product));
10685	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10686		switch (lun->be_lun->lun_type) {
10687		case T_DIRECT:
10688			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10689			    sizeof(inq_ptr->product));
10690			break;
10691		case T_PROCESSOR:
10692			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10693			    sizeof(inq_ptr->product));
10694			break;
10695		default:
10696			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10697			    sizeof(inq_ptr->product));
10698			break;
10699		}
10700	} else {
10701		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10702		strncpy(inq_ptr->product, val,
10703		    min(sizeof(inq_ptr->product), strlen(val)));
10704	}
10705
10706	/*
10707	 * XXX make this a macro somewhere so it automatically gets
10708	 * incremented when we make changes.
10709	 */
10710	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10711	    "revision")) == NULL) {
10712		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10713	} else {
10714		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10715		strncpy(inq_ptr->revision, val,
10716		    min(sizeof(inq_ptr->revision), strlen(val)));
10717	}
10718
10719	/*
10720	 * For parallel SCSI, we support double transition and single
10721	 * transition clocking.  We also support QAS (Quick Arbitration
10722	 * and Selection) and Information Unit transfers on both the
10723	 * control and array devices.
10724	 */
10725	if (port_type == CTL_PORT_SCSI)
10726		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10727				    SID_SPI_IUS;
10728
10729	/* SAM-5 (no version claimed) */
10730	scsi_ulto2b(0x00A0, inq_ptr->version1);
10731	/* SPC-4 (no version claimed) */
10732	scsi_ulto2b(0x0460, inq_ptr->version2);
10733	if (port_type == CTL_PORT_FC) {
10734		/* FCP-2 ANSI INCITS.350:2003 */
10735		scsi_ulto2b(0x0917, inq_ptr->version3);
10736	} else if (port_type == CTL_PORT_SCSI) {
10737		/* SPI-4 ANSI INCITS.362:200x */
10738		scsi_ulto2b(0x0B56, inq_ptr->version3);
10739	} else if (port_type == CTL_PORT_ISCSI) {
10740		/* iSCSI (no version claimed) */
10741		scsi_ulto2b(0x0960, inq_ptr->version3);
10742	} else if (port_type == CTL_PORT_SAS) {
10743		/* SAS (no version claimed) */
10744		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10745	}
10746
10747	if (lun == NULL) {
10748		/* SBC-4 (no version claimed) */
10749		scsi_ulto2b(0x0600, inq_ptr->version4);
10750	} else {
10751		switch (lun->be_lun->lun_type) {
10752		case T_DIRECT:
10753			/* SBC-4 (no version claimed) */
10754			scsi_ulto2b(0x0600, inq_ptr->version4);
10755			break;
10756		case T_PROCESSOR:
10757		default:
10758			break;
10759		}
10760	}
10761
10762	ctsio->scsi_status = SCSI_STATUS_OK;
10763	if (ctsio->kern_data_len > 0) {
10764		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10765		ctsio->be_move_done = ctl_config_move_done;
10766		ctl_datamove((union ctl_io *)ctsio);
10767	} else {
10768		ctsio->io_hdr.status = CTL_SUCCESS;
10769		ctl_done((union ctl_io *)ctsio);
10770	}
10771
10772	return (CTL_RETVAL_COMPLETE);
10773}
10774
10775int
10776ctl_inquiry(struct ctl_scsiio *ctsio)
10777{
10778	struct scsi_inquiry *cdb;
10779	int retval;
10780
10781	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10782
10783	cdb = (struct scsi_inquiry *)ctsio->cdb;
10784	if (cdb->byte2 & SI_EVPD)
10785		retval = ctl_inquiry_evpd(ctsio);
10786	else if (cdb->page_code == 0)
10787		retval = ctl_inquiry_std(ctsio);
10788	else {
10789		ctl_set_invalid_field(ctsio,
10790				      /*sks_valid*/ 1,
10791				      /*command*/ 1,
10792				      /*field*/ 2,
10793				      /*bit_valid*/ 0,
10794				      /*bit*/ 0);
10795		ctl_done((union ctl_io *)ctsio);
10796		return (CTL_RETVAL_COMPLETE);
10797	}
10798
10799	return (retval);
10800}
10801
10802/*
10803 * For known CDB types, parse the LBA and length.
10804 */
10805static int
10806ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10807{
10808	if (io->io_hdr.io_type != CTL_IO_SCSI)
10809		return (1);
10810
10811	switch (io->scsiio.cdb[0]) {
10812	case COMPARE_AND_WRITE: {
10813		struct scsi_compare_and_write *cdb;
10814
10815		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10816
10817		*lba = scsi_8btou64(cdb->addr);
10818		*len = cdb->length;
10819		break;
10820	}
10821	case READ_6:
10822	case WRITE_6: {
10823		struct scsi_rw_6 *cdb;
10824
10825		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10826
10827		*lba = scsi_3btoul(cdb->addr);
10828		/* only 5 bits are valid in the most significant address byte */
10829		*lba &= 0x1fffff;
10830		*len = cdb->length;
10831		break;
10832	}
10833	case READ_10:
10834	case WRITE_10: {
10835		struct scsi_rw_10 *cdb;
10836
10837		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10838
10839		*lba = scsi_4btoul(cdb->addr);
10840		*len = scsi_2btoul(cdb->length);
10841		break;
10842	}
10843	case WRITE_VERIFY_10: {
10844		struct scsi_write_verify_10 *cdb;
10845
10846		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10847
10848		*lba = scsi_4btoul(cdb->addr);
10849		*len = scsi_2btoul(cdb->length);
10850		break;
10851	}
10852	case READ_12:
10853	case WRITE_12: {
10854		struct scsi_rw_12 *cdb;
10855
10856		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10857
10858		*lba = scsi_4btoul(cdb->addr);
10859		*len = scsi_4btoul(cdb->length);
10860		break;
10861	}
10862	case WRITE_VERIFY_12: {
10863		struct scsi_write_verify_12 *cdb;
10864
10865		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10866
10867		*lba = scsi_4btoul(cdb->addr);
10868		*len = scsi_4btoul(cdb->length);
10869		break;
10870	}
10871	case READ_16:
10872	case WRITE_16:
10873	case WRITE_ATOMIC_16: {
10874		struct scsi_rw_16 *cdb;
10875
10876		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10877
10878		*lba = scsi_8btou64(cdb->addr);
10879		*len = scsi_4btoul(cdb->length);
10880		break;
10881	}
10882	case WRITE_VERIFY_16: {
10883		struct scsi_write_verify_16 *cdb;
10884
10885		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10886
10887		*lba = scsi_8btou64(cdb->addr);
10888		*len = scsi_4btoul(cdb->length);
10889		break;
10890	}
10891	case WRITE_SAME_10: {
10892		struct scsi_write_same_10 *cdb;
10893
10894		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10895
10896		*lba = scsi_4btoul(cdb->addr);
10897		*len = scsi_2btoul(cdb->length);
10898		break;
10899	}
10900	case WRITE_SAME_16: {
10901		struct scsi_write_same_16 *cdb;
10902
10903		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10904
10905		*lba = scsi_8btou64(cdb->addr);
10906		*len = scsi_4btoul(cdb->length);
10907		break;
10908	}
10909	case VERIFY_10: {
10910		struct scsi_verify_10 *cdb;
10911
10912		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10913
10914		*lba = scsi_4btoul(cdb->addr);
10915		*len = scsi_2btoul(cdb->length);
10916		break;
10917	}
10918	case VERIFY_12: {
10919		struct scsi_verify_12 *cdb;
10920
10921		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10922
10923		*lba = scsi_4btoul(cdb->addr);
10924		*len = scsi_4btoul(cdb->length);
10925		break;
10926	}
10927	case VERIFY_16: {
10928		struct scsi_verify_16 *cdb;
10929
10930		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10931
10932		*lba = scsi_8btou64(cdb->addr);
10933		*len = scsi_4btoul(cdb->length);
10934		break;
10935	}
10936	case UNMAP: {
10937		*lba = 0;
10938		*len = UINT64_MAX;
10939		break;
10940	}
10941	default:
10942		return (1);
10943		break; /* NOTREACHED */
10944	}
10945
10946	return (0);
10947}
10948
10949static ctl_action
10950ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10951{
10952	uint64_t endlba1, endlba2;
10953
10954	endlba1 = lba1 + len1 - 1;
10955	endlba2 = lba2 + len2 - 1;
10956
10957	if ((endlba1 < lba2)
10958	 || (endlba2 < lba1))
10959		return (CTL_ACTION_PASS);
10960	else
10961		return (CTL_ACTION_BLOCK);
10962}
10963
10964static int
10965ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10966{
10967	struct ctl_ptr_len_flags *ptrlen;
10968	struct scsi_unmap_desc *buf, *end, *range;
10969	uint64_t lba;
10970	uint32_t len;
10971
10972	/* If not UNMAP -- go other way. */
10973	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10974	    io->scsiio.cdb[0] != UNMAP)
10975		return (CTL_ACTION_ERROR);
10976
10977	/* If UNMAP without data -- block and wait for data. */
10978	ptrlen = (struct ctl_ptr_len_flags *)
10979	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10980	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10981	    ptrlen->ptr == NULL)
10982		return (CTL_ACTION_BLOCK);
10983
10984	/* UNMAP with data -- check for collision. */
10985	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10986	end = buf + ptrlen->len / sizeof(*buf);
10987	for (range = buf; range < end; range++) {
10988		lba = scsi_8btou64(range->lba);
10989		len = scsi_4btoul(range->length);
10990		if ((lba < lba2 + len2) && (lba + len > lba2))
10991			return (CTL_ACTION_BLOCK);
10992	}
10993	return (CTL_ACTION_PASS);
10994}
10995
10996static ctl_action
10997ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10998{
10999	uint64_t lba1, lba2;
11000	uint64_t len1, len2;
11001	int retval;
11002
11003	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
11004		return (CTL_ACTION_ERROR);
11005
11006	retval = ctl_extent_check_unmap(io2, lba1, len1);
11007	if (retval != CTL_ACTION_ERROR)
11008		return (retval);
11009
11010	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
11011		return (CTL_ACTION_ERROR);
11012
11013	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
11014}
11015
11016static ctl_action
11017ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
11018    union ctl_io *ooa_io)
11019{
11020	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
11021	ctl_serialize_action *serialize_row;
11022
11023	/*
11024	 * The initiator attempted multiple untagged commands at the same
11025	 * time.  Can't do that.
11026	 */
11027	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11028	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11029	 && ((pending_io->io_hdr.nexus.targ_port ==
11030	      ooa_io->io_hdr.nexus.targ_port)
11031	  && (pending_io->io_hdr.nexus.initid.id ==
11032	      ooa_io->io_hdr.nexus.initid.id))
11033	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11034		return (CTL_ACTION_OVERLAP);
11035
11036	/*
11037	 * The initiator attempted to send multiple tagged commands with
11038	 * the same ID.  (It's fine if different initiators have the same
11039	 * tag ID.)
11040	 *
11041	 * Even if all of those conditions are true, we don't kill the I/O
11042	 * if the command ahead of us has been aborted.  We won't end up
11043	 * sending it to the FETD, and it's perfectly legal to resend a
11044	 * command with the same tag number as long as the previous
11045	 * instance of this tag number has been aborted somehow.
11046	 */
11047	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11048	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11049	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11050	 && ((pending_io->io_hdr.nexus.targ_port ==
11051	      ooa_io->io_hdr.nexus.targ_port)
11052	  && (pending_io->io_hdr.nexus.initid.id ==
11053	      ooa_io->io_hdr.nexus.initid.id))
11054	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
11055		return (CTL_ACTION_OVERLAP_TAG);
11056
11057	/*
11058	 * If we get a head of queue tag, SAM-3 says that we should
11059	 * immediately execute it.
11060	 *
11061	 * What happens if this command would normally block for some other
11062	 * reason?  e.g. a request sense with a head of queue tag
11063	 * immediately after a write.  Normally that would block, but this
11064	 * will result in its getting executed immediately...
11065	 *
11066	 * We currently return "pass" instead of "skip", so we'll end up
11067	 * going through the rest of the queue to check for overlapped tags.
11068	 *
11069	 * XXX KDM check for other types of blockage first??
11070	 */
11071	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11072		return (CTL_ACTION_PASS);
11073
11074	/*
11075	 * Ordered tags have to block until all items ahead of them
11076	 * have completed.  If we get called with an ordered tag, we always
11077	 * block, if something else is ahead of us in the queue.
11078	 */
11079	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11080		return (CTL_ACTION_BLOCK);
11081
11082	/*
11083	 * Simple tags get blocked until all head of queue and ordered tags
11084	 * ahead of them have completed.  I'm lumping untagged commands in
11085	 * with simple tags here.  XXX KDM is that the right thing to do?
11086	 */
11087	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11088	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11089	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11090	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11091		return (CTL_ACTION_BLOCK);
11092
11093	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11094	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11095
11096	serialize_row = ctl_serialize_table[ooa_entry->seridx];
11097
11098	switch (serialize_row[pending_entry->seridx]) {
11099	case CTL_SER_BLOCK:
11100		return (CTL_ACTION_BLOCK);
11101	case CTL_SER_EXTENT:
11102		return (ctl_extent_check(pending_io, ooa_io));
11103	case CTL_SER_EXTENTOPT:
11104		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11105		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11106			return (ctl_extent_check(pending_io, ooa_io));
11107		/* FALLTHROUGH */
11108	case CTL_SER_PASS:
11109		return (CTL_ACTION_PASS);
11110	case CTL_SER_BLOCKOPT:
11111		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11112		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11113			return (CTL_ACTION_BLOCK);
11114		return (CTL_ACTION_PASS);
11115	case CTL_SER_SKIP:
11116		return (CTL_ACTION_SKIP);
11117	default:
11118		panic("invalid serialization value %d",
11119		      serialize_row[pending_entry->seridx]);
11120	}
11121
11122	return (CTL_ACTION_ERROR);
11123}
11124
11125/*
11126 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11127 * Assumptions:
11128 * - pending_io is generally either incoming, or on the blocked queue
11129 * - starting I/O is the I/O we want to start the check with.
11130 */
11131static ctl_action
11132ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11133	      union ctl_io *starting_io)
11134{
11135	union ctl_io *ooa_io;
11136	ctl_action action;
11137
11138	mtx_assert(&lun->lun_lock, MA_OWNED);
11139
11140	/*
11141	 * Run back along the OOA queue, starting with the current
11142	 * blocked I/O and going through every I/O before it on the
11143	 * queue.  If starting_io is NULL, we'll just end up returning
11144	 * CTL_ACTION_PASS.
11145	 */
11146	for (ooa_io = starting_io; ooa_io != NULL;
11147	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11148	     ooa_links)){
11149
11150		/*
11151		 * This routine just checks to see whether
11152		 * cur_blocked is blocked by ooa_io, which is ahead
11153		 * of it in the queue.  It doesn't queue/dequeue
11154		 * cur_blocked.
11155		 */
11156		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11157		switch (action) {
11158		case CTL_ACTION_BLOCK:
11159		case CTL_ACTION_OVERLAP:
11160		case CTL_ACTION_OVERLAP_TAG:
11161		case CTL_ACTION_SKIP:
11162		case CTL_ACTION_ERROR:
11163			return (action);
11164			break; /* NOTREACHED */
11165		case CTL_ACTION_PASS:
11166			break;
11167		default:
11168			panic("invalid action %d", action);
11169			break;  /* NOTREACHED */
11170		}
11171	}
11172
11173	return (CTL_ACTION_PASS);
11174}
11175
11176/*
11177 * Assumptions:
11178 * - An I/O has just completed, and has been removed from the per-LUN OOA
11179 *   queue, so some items on the blocked queue may now be unblocked.
11180 */
11181static int
11182ctl_check_blocked(struct ctl_lun *lun)
11183{
11184	union ctl_io *cur_blocked, *next_blocked;
11185
11186	mtx_assert(&lun->lun_lock, MA_OWNED);
11187
11188	/*
11189	 * Run forward from the head of the blocked queue, checking each
11190	 * entry against the I/Os prior to it on the OOA queue to see if
11191	 * there is still any blockage.
11192	 *
11193	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11194	 * with our removing a variable on it while it is traversing the
11195	 * list.
11196	 */
11197	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11198	     cur_blocked != NULL; cur_blocked = next_blocked) {
11199		union ctl_io *prev_ooa;
11200		ctl_action action;
11201
11202		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11203							  blocked_links);
11204
11205		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11206						      ctl_ooaq, ooa_links);
11207
11208		/*
11209		 * If cur_blocked happens to be the first item in the OOA
11210		 * queue now, prev_ooa will be NULL, and the action
11211		 * returned will just be CTL_ACTION_PASS.
11212		 */
11213		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11214
11215		switch (action) {
11216		case CTL_ACTION_BLOCK:
11217			/* Nothing to do here, still blocked */
11218			break;
11219		case CTL_ACTION_OVERLAP:
11220		case CTL_ACTION_OVERLAP_TAG:
11221			/*
11222			 * This shouldn't happen!  In theory we've already
11223			 * checked this command for overlap...
11224			 */
11225			break;
11226		case CTL_ACTION_PASS:
11227		case CTL_ACTION_SKIP: {
11228			struct ctl_softc *softc;
11229			const struct ctl_cmd_entry *entry;
11230			uint32_t initidx;
11231			int isc_retval;
11232
11233			/*
11234			 * The skip case shouldn't happen, this transaction
11235			 * should have never made it onto the blocked queue.
11236			 */
11237			/*
11238			 * This I/O is no longer blocked, we can remove it
11239			 * from the blocked queue.  Since this is a TAILQ
11240			 * (doubly linked list), we can do O(1) removals
11241			 * from any place on the list.
11242			 */
11243			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11244				     blocked_links);
11245			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11246
11247			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11248				/*
11249				 * Need to send IO back to original side to
11250				 * run
11251				 */
11252				union ctl_ha_msg msg_info;
11253
11254				msg_info.hdr.original_sc =
11255					cur_blocked->io_hdr.original_sc;
11256				msg_info.hdr.serializing_sc = cur_blocked;
11257				msg_info.hdr.msg_type = CTL_MSG_R2R;
11258				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11259				     &msg_info, sizeof(msg_info), 0)) >
11260				     CTL_HA_STATUS_SUCCESS) {
11261					printf("CTL:Check Blocked error from "
11262					       "ctl_ha_msg_send %d\n",
11263					       isc_retval);
11264				}
11265				break;
11266			}
11267			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11268			softc = control_softc;
11269
11270			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11271
11272			/*
11273			 * Check this I/O for LUN state changes that may
11274			 * have happened while this command was blocked.
11275			 * The LUN state may have been changed by a command
11276			 * ahead of us in the queue, so we need to re-check
11277			 * for any states that can be caused by SCSI
11278			 * commands.
11279			 */
11280			if (ctl_scsiio_lun_check(softc, lun, entry,
11281						 &cur_blocked->scsiio) == 0) {
11282				cur_blocked->io_hdr.flags |=
11283				                      CTL_FLAG_IS_WAS_ON_RTR;
11284				ctl_enqueue_rtr(cur_blocked);
11285			} else
11286				ctl_done(cur_blocked);
11287			break;
11288		}
11289		default:
11290			/*
11291			 * This probably shouldn't happen -- we shouldn't
11292			 * get CTL_ACTION_ERROR, or anything else.
11293			 */
11294			break;
11295		}
11296	}
11297
11298	return (CTL_RETVAL_COMPLETE);
11299}
11300
11301/*
11302 * This routine (with one exception) checks LUN flags that can be set by
11303 * commands ahead of us in the OOA queue.  These flags have to be checked
11304 * when a command initially comes in, and when we pull a command off the
11305 * blocked queue and are preparing to execute it.  The reason we have to
11306 * check these flags for commands on the blocked queue is that the LUN
11307 * state may have been changed by a command ahead of us while we're on the
11308 * blocked queue.
11309 *
11310 * Ordering is somewhat important with these checks, so please pay
11311 * careful attention to the placement of any new checks.
11312 */
11313static int
11314ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11315    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11316{
11317	int retval;
11318	uint32_t residx;
11319
11320	retval = 0;
11321
11322	mtx_assert(&lun->lun_lock, MA_OWNED);
11323
11324	/*
11325	 * If this shelf is a secondary shelf controller, we have to reject
11326	 * any media access commands.
11327	 */
11328#if 0
11329	/* No longer needed for HA */
11330	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11331	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11332		ctl_set_lun_standby(ctsio);
11333		retval = 1;
11334		goto bailout;
11335	}
11336#endif
11337
11338	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11339		if (lun->flags & CTL_LUN_READONLY) {
11340			ctl_set_sense(ctsio, /*current_error*/ 1,
11341			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11342			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11343			retval = 1;
11344			goto bailout;
11345		}
11346		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11347		    .eca_and_aen & SCP_SWP) != 0) {
11348			ctl_set_sense(ctsio, /*current_error*/ 1,
11349			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11350			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11351			retval = 1;
11352			goto bailout;
11353		}
11354	}
11355
11356	/*
11357	 * Check for a reservation conflict.  If this command isn't allowed
11358	 * even on reserved LUNs, and if this initiator isn't the one who
11359	 * reserved us, reject the command with a reservation conflict.
11360	 */
11361	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11362	if ((lun->flags & CTL_LUN_RESERVED)
11363	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11364		if (lun->res_idx != residx) {
11365			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11366			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11367			retval = 1;
11368			goto bailout;
11369		}
11370	}
11371
11372	if ((lun->flags & CTL_LUN_PR_RESERVED)
11373	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11374		/*
11375		 * if we aren't registered or it's a res holder type
11376		 * reservation and this isn't the res holder then set a
11377		 * conflict.
11378		 * NOTE: Commands which might be allowed on write exclusive
11379		 * type reservations are checked in the particular command
11380		 * for a conflict. Read and SSU are the only ones.
11381		 */
11382		if (!lun->per_res[residx].registered
11383		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11384			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11385			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11386			retval = 1;
11387			goto bailout;
11388		}
11389
11390	}
11391
11392	if ((lun->flags & CTL_LUN_OFFLINE)
11393	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11394		ctl_set_lun_not_ready(ctsio);
11395		retval = 1;
11396		goto bailout;
11397	}
11398
11399	/*
11400	 * If the LUN is stopped, see if this particular command is allowed
11401	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11402	 */
11403	if ((lun->flags & CTL_LUN_STOPPED)
11404	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11405		/* "Logical unit not ready, initializing cmd. required" */
11406		ctl_set_lun_stopped(ctsio);
11407		retval = 1;
11408		goto bailout;
11409	}
11410
11411	if ((lun->flags & CTL_LUN_INOPERABLE)
11412	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11413		/* "Medium format corrupted" */
11414		ctl_set_medium_format_corrupted(ctsio);
11415		retval = 1;
11416		goto bailout;
11417	}
11418
11419bailout:
11420	return (retval);
11421
11422}
11423
11424static void
11425ctl_failover_io(union ctl_io *io, int have_lock)
11426{
11427	ctl_set_busy(&io->scsiio);
11428	ctl_done(io);
11429}
11430
11431static void
11432ctl_failover(void)
11433{
11434	struct ctl_lun *lun;
11435	struct ctl_softc *ctl_softc;
11436	union ctl_io *next_io, *pending_io;
11437	union ctl_io *io;
11438	int lun_idx;
11439	int i;
11440
11441	ctl_softc = control_softc;
11442
11443	mtx_lock(&ctl_softc->ctl_lock);
11444	/*
11445	 * Remove any cmds from the other SC from the rtr queue.  These
11446	 * will obviously only be for LUNs for which we're the primary.
11447	 * We can't send status or get/send data for these commands.
11448	 * Since they haven't been executed yet, we can just remove them.
11449	 * We'll either abort them or delete them below, depending on
11450	 * which HA mode we're in.
11451	 */
11452#ifdef notyet
11453	mtx_lock(&ctl_softc->queue_lock);
11454	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11455	     io != NULL; io = next_io) {
11456		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11457		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11458			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11459				      ctl_io_hdr, links);
11460	}
11461	mtx_unlock(&ctl_softc->queue_lock);
11462#endif
11463
11464	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11465		lun = ctl_softc->ctl_luns[lun_idx];
11466		if (lun==NULL)
11467			continue;
11468
11469		/*
11470		 * Processor LUNs are primary on both sides.
11471		 * XXX will this always be true?
11472		 */
11473		if (lun->be_lun->lun_type == T_PROCESSOR)
11474			continue;
11475
11476		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11477		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11478			printf("FAILOVER: primary lun %d\n", lun_idx);
11479		        /*
11480			 * Remove all commands from the other SC. First from the
11481			 * blocked queue then from the ooa queue. Once we have
11482			 * removed them. Call ctl_check_blocked to see if there
11483			 * is anything that can run.
11484			 */
11485			for (io = (union ctl_io *)TAILQ_FIRST(
11486			     &lun->blocked_queue); io != NULL; io = next_io) {
11487
11488		        	next_io = (union ctl_io *)TAILQ_NEXT(
11489				    &io->io_hdr, blocked_links);
11490
11491				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11492					TAILQ_REMOVE(&lun->blocked_queue,
11493						     &io->io_hdr,blocked_links);
11494					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11495					TAILQ_REMOVE(&lun->ooa_queue,
11496						     &io->io_hdr, ooa_links);
11497
11498					ctl_free_io(io);
11499				}
11500			}
11501
11502			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11503	     		     io != NULL; io = next_io) {
11504
11505		        	next_io = (union ctl_io *)TAILQ_NEXT(
11506				    &io->io_hdr, ooa_links);
11507
11508				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11509
11510					TAILQ_REMOVE(&lun->ooa_queue,
11511						&io->io_hdr,
11512					     	ooa_links);
11513
11514					ctl_free_io(io);
11515				}
11516			}
11517			ctl_check_blocked(lun);
11518		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11519			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11520
11521			printf("FAILOVER: primary lun %d\n", lun_idx);
11522			/*
11523			 * Abort all commands from the other SC.  We can't
11524			 * send status back for them now.  These should get
11525			 * cleaned up when they are completed or come out
11526			 * for a datamove operation.
11527			 */
11528			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11529	     		     io != NULL; io = next_io) {
11530		        	next_io = (union ctl_io *)TAILQ_NEXT(
11531					&io->io_hdr, ooa_links);
11532
11533				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11534					io->io_hdr.flags |= CTL_FLAG_ABORT;
11535			}
11536		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11537			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11538
11539			printf("FAILOVER: secondary lun %d\n", lun_idx);
11540
11541			lun->flags |= CTL_LUN_PRIMARY_SC;
11542
11543			/*
11544			 * We send all I/O that was sent to this controller
11545			 * and redirected to the other side back with
11546			 * busy status, and have the initiator retry it.
11547			 * Figuring out how much data has been transferred,
11548			 * etc. and picking up where we left off would be
11549			 * very tricky.
11550			 *
11551			 * XXX KDM need to remove I/O from the blocked
11552			 * queue as well!
11553			 */
11554			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11555			     &lun->ooa_queue); pending_io != NULL;
11556			     pending_io = next_io) {
11557
11558				next_io =  (union ctl_io *)TAILQ_NEXT(
11559					&pending_io->io_hdr, ooa_links);
11560
11561				pending_io->io_hdr.flags &=
11562					~CTL_FLAG_SENT_2OTHER_SC;
11563
11564				if (pending_io->io_hdr.flags &
11565				    CTL_FLAG_IO_ACTIVE) {
11566					pending_io->io_hdr.flags |=
11567						CTL_FLAG_FAILOVER;
11568				} else {
11569					ctl_set_busy(&pending_io->scsiio);
11570					ctl_done(pending_io);
11571				}
11572			}
11573
11574			/*
11575			 * Build Unit Attention
11576			 */
11577			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11578				lun->pending_ua[i] |=
11579				                     CTL_UA_ASYM_ACC_CHANGE;
11580			}
11581		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11582			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11583			printf("FAILOVER: secondary lun %d\n", lun_idx);
11584			/*
11585			 * if the first io on the OOA is not on the RtR queue
11586			 * add it.
11587			 */
11588			lun->flags |= CTL_LUN_PRIMARY_SC;
11589
11590			pending_io = (union ctl_io *)TAILQ_FIRST(
11591			    &lun->ooa_queue);
11592			if (pending_io==NULL) {
11593				printf("Nothing on OOA queue\n");
11594				continue;
11595			}
11596
11597			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11598			if ((pending_io->io_hdr.flags &
11599			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11600				pending_io->io_hdr.flags |=
11601				    CTL_FLAG_IS_WAS_ON_RTR;
11602				ctl_enqueue_rtr(pending_io);
11603			}
11604#if 0
11605			else
11606			{
11607				printf("Tag 0x%04x is running\n",
11608				      pending_io->scsiio.tag_num);
11609			}
11610#endif
11611
11612			next_io = (union ctl_io *)TAILQ_NEXT(
11613			    &pending_io->io_hdr, ooa_links);
11614			for (pending_io=next_io; pending_io != NULL;
11615			     pending_io = next_io) {
11616				pending_io->io_hdr.flags &=
11617				    ~CTL_FLAG_SENT_2OTHER_SC;
11618				next_io = (union ctl_io *)TAILQ_NEXT(
11619					&pending_io->io_hdr, ooa_links);
11620				if (pending_io->io_hdr.flags &
11621				    CTL_FLAG_IS_WAS_ON_RTR) {
11622#if 0
11623				        printf("Tag 0x%04x is running\n",
11624				      		pending_io->scsiio.tag_num);
11625#endif
11626					continue;
11627				}
11628
11629				switch (ctl_check_ooa(lun, pending_io,
11630			            (union ctl_io *)TAILQ_PREV(
11631				    &pending_io->io_hdr, ctl_ooaq,
11632				    ooa_links))) {
11633
11634				case CTL_ACTION_BLOCK:
11635					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11636							  &pending_io->io_hdr,
11637							  blocked_links);
11638					pending_io->io_hdr.flags |=
11639					    CTL_FLAG_BLOCKED;
11640					break;
11641				case CTL_ACTION_PASS:
11642				case CTL_ACTION_SKIP:
11643					pending_io->io_hdr.flags |=
11644					    CTL_FLAG_IS_WAS_ON_RTR;
11645					ctl_enqueue_rtr(pending_io);
11646					break;
11647				case CTL_ACTION_OVERLAP:
11648					ctl_set_overlapped_cmd(
11649					    (struct ctl_scsiio *)pending_io);
11650					ctl_done(pending_io);
11651					break;
11652				case CTL_ACTION_OVERLAP_TAG:
11653					ctl_set_overlapped_tag(
11654					    (struct ctl_scsiio *)pending_io,
11655					    pending_io->scsiio.tag_num & 0xff);
11656					ctl_done(pending_io);
11657					break;
11658				case CTL_ACTION_ERROR:
11659				default:
11660					ctl_set_internal_failure(
11661						(struct ctl_scsiio *)pending_io,
11662						0,  // sks_valid
11663						0); //retry count
11664					ctl_done(pending_io);
11665					break;
11666				}
11667			}
11668
11669			/*
11670			 * Build Unit Attention
11671			 */
11672			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11673				lun->pending_ua[i] |=
11674				                     CTL_UA_ASYM_ACC_CHANGE;
11675			}
11676		} else {
11677			panic("Unhandled HA mode failover, LUN flags = %#x, "
11678			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11679		}
11680	}
11681	ctl_pause_rtr = 0;
11682	mtx_unlock(&ctl_softc->ctl_lock);
11683}
11684
11685static int
11686ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11687{
11688	struct ctl_lun *lun;
11689	const struct ctl_cmd_entry *entry;
11690	uint32_t initidx, targ_lun;
11691	int retval;
11692
11693	retval = 0;
11694
11695	lun = NULL;
11696
11697	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11698	if ((targ_lun < CTL_MAX_LUNS)
11699	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11700		lun = ctl_softc->ctl_luns[targ_lun];
11701		/*
11702		 * If the LUN is invalid, pretend that it doesn't exist.
11703		 * It will go away as soon as all pending I/O has been
11704		 * completed.
11705		 */
11706		if (lun->flags & CTL_LUN_DISABLED) {
11707			lun = NULL;
11708		} else {
11709			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11710			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11711				lun->be_lun;
11712			if (lun->be_lun->lun_type == T_PROCESSOR) {
11713				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11714			}
11715
11716			/*
11717			 * Every I/O goes into the OOA queue for a
11718			 * particular LUN, and stays there until completion.
11719			 */
11720			mtx_lock(&lun->lun_lock);
11721			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11722			    ooa_links);
11723		}
11724	} else {
11725		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11726		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11727	}
11728
11729	/* Get command entry and return error if it is unsuppotyed. */
11730	entry = ctl_validate_command(ctsio);
11731	if (entry == NULL) {
11732		if (lun)
11733			mtx_unlock(&lun->lun_lock);
11734		return (retval);
11735	}
11736
11737	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11738	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11739
11740	/*
11741	 * Check to see whether we can send this command to LUNs that don't
11742	 * exist.  This should pretty much only be the case for inquiry
11743	 * and request sense.  Further checks, below, really require having
11744	 * a LUN, so we can't really check the command anymore.  Just put
11745	 * it on the rtr queue.
11746	 */
11747	if (lun == NULL) {
11748		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11749			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11750			ctl_enqueue_rtr((union ctl_io *)ctsio);
11751			return (retval);
11752		}
11753
11754		ctl_set_unsupported_lun(ctsio);
11755		ctl_done((union ctl_io *)ctsio);
11756		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11757		return (retval);
11758	} else {
11759		/*
11760		 * Make sure we support this particular command on this LUN.
11761		 * e.g., we don't support writes to the control LUN.
11762		 */
11763		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11764			mtx_unlock(&lun->lun_lock);
11765			ctl_set_invalid_opcode(ctsio);
11766			ctl_done((union ctl_io *)ctsio);
11767			return (retval);
11768		}
11769	}
11770
11771	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11772
11773#ifdef CTL_WITH_CA
11774	/*
11775	 * If we've got a request sense, it'll clear the contingent
11776	 * allegiance condition.  Otherwise, if we have a CA condition for
11777	 * this initiator, clear it, because it sent down a command other
11778	 * than request sense.
11779	 */
11780	if ((ctsio->cdb[0] != REQUEST_SENSE)
11781	 && (ctl_is_set(lun->have_ca, initidx)))
11782		ctl_clear_mask(lun->have_ca, initidx);
11783#endif
11784
11785	/*
11786	 * If the command has this flag set, it handles its own unit
11787	 * attention reporting, we shouldn't do anything.  Otherwise we
11788	 * check for any pending unit attentions, and send them back to the
11789	 * initiator.  We only do this when a command initially comes in,
11790	 * not when we pull it off the blocked queue.
11791	 *
11792	 * According to SAM-3, section 5.3.2, the order that things get
11793	 * presented back to the host is basically unit attentions caused
11794	 * by some sort of reset event, busy status, reservation conflicts
11795	 * or task set full, and finally any other status.
11796	 *
11797	 * One issue here is that some of the unit attentions we report
11798	 * don't fall into the "reset" category (e.g. "reported luns data
11799	 * has changed").  So reporting it here, before the reservation
11800	 * check, may be technically wrong.  I guess the only thing to do
11801	 * would be to check for and report the reset events here, and then
11802	 * check for the other unit attention types after we check for a
11803	 * reservation conflict.
11804	 *
11805	 * XXX KDM need to fix this
11806	 */
11807	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11808		ctl_ua_type ua_type;
11809
11810		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11811			scsi_sense_data_type sense_format;
11812
11813			if (lun != NULL)
11814				sense_format = (lun->flags &
11815				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11816				    SSD_TYPE_FIXED;
11817			else
11818				sense_format = SSD_TYPE_FIXED;
11819
11820			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11821			    &ctsio->sense_data, sense_format);
11822			if (ua_type != CTL_UA_NONE) {
11823				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11824				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11825						       CTL_AUTOSENSE;
11826				ctsio->sense_len = SSD_FULL_SIZE;
11827				mtx_unlock(&lun->lun_lock);
11828				ctl_done((union ctl_io *)ctsio);
11829				return (retval);
11830			}
11831		}
11832	}
11833
11834
11835	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11836		mtx_unlock(&lun->lun_lock);
11837		ctl_done((union ctl_io *)ctsio);
11838		return (retval);
11839	}
11840
11841	/*
11842	 * XXX CHD this is where we want to send IO to other side if
11843	 * this LUN is secondary on this SC. We will need to make a copy
11844	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11845	 * the copy we send as FROM_OTHER.
11846	 * We also need to stuff the address of the original IO so we can
11847	 * find it easily. Something similar will need be done on the other
11848	 * side so when we are done we can find the copy.
11849	 */
11850	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11851		union ctl_ha_msg msg_info;
11852		int isc_retval;
11853
11854		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11855
11856		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11857		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11858#if 0
11859		printf("1. ctsio %p\n", ctsio);
11860#endif
11861		msg_info.hdr.serializing_sc = NULL;
11862		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11863		msg_info.scsi.tag_num = ctsio->tag_num;
11864		msg_info.scsi.tag_type = ctsio->tag_type;
11865		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11866
11867		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11868
11869		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11870		    (void *)&msg_info, sizeof(msg_info), 0)) >
11871		    CTL_HA_STATUS_SUCCESS) {
11872			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11873			       isc_retval);
11874			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11875		} else {
11876#if 0
11877			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11878#endif
11879		}
11880
11881		/*
11882		 * XXX KDM this I/O is off the incoming queue, but hasn't
11883		 * been inserted on any other queue.  We may need to come
11884		 * up with a holding queue while we wait for serialization
11885		 * so that we have an idea of what we're waiting for from
11886		 * the other side.
11887		 */
11888		mtx_unlock(&lun->lun_lock);
11889		return (retval);
11890	}
11891
11892	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11893			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11894			      ctl_ooaq, ooa_links))) {
11895	case CTL_ACTION_BLOCK:
11896		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11897		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11898				  blocked_links);
11899		mtx_unlock(&lun->lun_lock);
11900		return (retval);
11901	case CTL_ACTION_PASS:
11902	case CTL_ACTION_SKIP:
11903		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11904		mtx_unlock(&lun->lun_lock);
11905		ctl_enqueue_rtr((union ctl_io *)ctsio);
11906		break;
11907	case CTL_ACTION_OVERLAP:
11908		mtx_unlock(&lun->lun_lock);
11909		ctl_set_overlapped_cmd(ctsio);
11910		ctl_done((union ctl_io *)ctsio);
11911		break;
11912	case CTL_ACTION_OVERLAP_TAG:
11913		mtx_unlock(&lun->lun_lock);
11914		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11915		ctl_done((union ctl_io *)ctsio);
11916		break;
11917	case CTL_ACTION_ERROR:
11918	default:
11919		mtx_unlock(&lun->lun_lock);
11920		ctl_set_internal_failure(ctsio,
11921					 /*sks_valid*/ 0,
11922					 /*retry_count*/ 0);
11923		ctl_done((union ctl_io *)ctsio);
11924		break;
11925	}
11926	return (retval);
11927}
11928
11929const struct ctl_cmd_entry *
11930ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11931{
11932	const struct ctl_cmd_entry *entry;
11933	int service_action;
11934
11935	entry = &ctl_cmd_table[ctsio->cdb[0]];
11936	if (sa)
11937		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11938	if (entry->flags & CTL_CMD_FLAG_SA5) {
11939		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11940		entry = &((const struct ctl_cmd_entry *)
11941		    entry->execute)[service_action];
11942	}
11943	return (entry);
11944}
11945
11946const struct ctl_cmd_entry *
11947ctl_validate_command(struct ctl_scsiio *ctsio)
11948{
11949	const struct ctl_cmd_entry *entry;
11950	int i, sa;
11951	uint8_t diff;
11952
11953	entry = ctl_get_cmd_entry(ctsio, &sa);
11954	if (entry->execute == NULL) {
11955		if (sa)
11956			ctl_set_invalid_field(ctsio,
11957					      /*sks_valid*/ 1,
11958					      /*command*/ 1,
11959					      /*field*/ 1,
11960					      /*bit_valid*/ 1,
11961					      /*bit*/ 4);
11962		else
11963			ctl_set_invalid_opcode(ctsio);
11964		ctl_done((union ctl_io *)ctsio);
11965		return (NULL);
11966	}
11967	KASSERT(entry->length > 0,
11968	    ("Not defined length for command 0x%02x/0x%02x",
11969	     ctsio->cdb[0], ctsio->cdb[1]));
11970	for (i = 1; i < entry->length; i++) {
11971		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11972		if (diff == 0)
11973			continue;
11974		ctl_set_invalid_field(ctsio,
11975				      /*sks_valid*/ 1,
11976				      /*command*/ 1,
11977				      /*field*/ i,
11978				      /*bit_valid*/ 1,
11979				      /*bit*/ fls(diff) - 1);
11980		ctl_done((union ctl_io *)ctsio);
11981		return (NULL);
11982	}
11983	return (entry);
11984}
11985
11986static int
11987ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11988{
11989
11990	switch (lun_type) {
11991	case T_PROCESSOR:
11992		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11993		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11994			return (0);
11995		break;
11996	case T_DIRECT:
11997		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11998		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11999			return (0);
12000		break;
12001	default:
12002		return (0);
12003	}
12004	return (1);
12005}
12006
12007static int
12008ctl_scsiio(struct ctl_scsiio *ctsio)
12009{
12010	int retval;
12011	const struct ctl_cmd_entry *entry;
12012
12013	retval = CTL_RETVAL_COMPLETE;
12014
12015	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
12016
12017	entry = ctl_get_cmd_entry(ctsio, NULL);
12018
12019	/*
12020	 * If this I/O has been aborted, just send it straight to
12021	 * ctl_done() without executing it.
12022	 */
12023	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
12024		ctl_done((union ctl_io *)ctsio);
12025		goto bailout;
12026	}
12027
12028	/*
12029	 * All the checks should have been handled by ctl_scsiio_precheck().
12030	 * We should be clear now to just execute the I/O.
12031	 */
12032	retval = entry->execute(ctsio);
12033
12034bailout:
12035	return (retval);
12036}
12037
12038/*
12039 * Since we only implement one target right now, a bus reset simply resets
12040 * our single target.
12041 */
12042static int
12043ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
12044{
12045	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
12046}
12047
12048static int
12049ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
12050		 ctl_ua_type ua_type)
12051{
12052	struct ctl_lun *lun;
12053	int retval;
12054
12055	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12056		union ctl_ha_msg msg_info;
12057
12058		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
12059		msg_info.hdr.nexus = io->io_hdr.nexus;
12060		if (ua_type==CTL_UA_TARG_RESET)
12061			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
12062		else
12063			msg_info.task.task_action = CTL_TASK_BUS_RESET;
12064		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12065		msg_info.hdr.original_sc = NULL;
12066		msg_info.hdr.serializing_sc = NULL;
12067		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12068		    (void *)&msg_info, sizeof(msg_info), 0)) {
12069		}
12070	}
12071	retval = 0;
12072
12073	mtx_lock(&ctl_softc->ctl_lock);
12074	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
12075		retval += ctl_lun_reset(lun, io, ua_type);
12076	mtx_unlock(&ctl_softc->ctl_lock);
12077
12078	return (retval);
12079}
12080
12081/*
12082 * The LUN should always be set.  The I/O is optional, and is used to
12083 * distinguish between I/Os sent by this initiator, and by other
12084 * initiators.  We set unit attention for initiators other than this one.
12085 * SAM-3 is vague on this point.  It does say that a unit attention should
12086 * be established for other initiators when a LUN is reset (see section
12087 * 5.7.3), but it doesn't specifically say that the unit attention should
12088 * be established for this particular initiator when a LUN is reset.  Here
12089 * is the relevant text, from SAM-3 rev 8:
12090 *
12091 * 5.7.2 When a SCSI initiator port aborts its own tasks
12092 *
12093 * When a SCSI initiator port causes its own task(s) to be aborted, no
12094 * notification that the task(s) have been aborted shall be returned to
12095 * the SCSI initiator port other than the completion response for the
12096 * command or task management function action that caused the task(s) to
12097 * be aborted and notification(s) associated with related effects of the
12098 * action (e.g., a reset unit attention condition).
12099 *
12100 * XXX KDM for now, we're setting unit attention for all initiators.
12101 */
12102static int
12103ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
12104{
12105	union ctl_io *xio;
12106#if 0
12107	uint32_t initindex;
12108#endif
12109	int i;
12110
12111	mtx_lock(&lun->lun_lock);
12112	/*
12113	 * Run through the OOA queue and abort each I/O.
12114	 */
12115#if 0
12116	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12117#endif
12118	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12119	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12120		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
12121	}
12122
12123	/*
12124	 * This version sets unit attention for every
12125	 */
12126#if 0
12127	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12128	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12129		if (initindex == i)
12130			continue;
12131		lun->pending_ua[i] |= ua_type;
12132	}
12133#endif
12134
12135	/*
12136	 * A reset (any kind, really) clears reservations established with
12137	 * RESERVE/RELEASE.  It does not clear reservations established
12138	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12139	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12140	 * reservations made with the RESERVE/RELEASE commands, because
12141	 * those commands are obsolete in SPC-3.
12142	 */
12143	lun->flags &= ~CTL_LUN_RESERVED;
12144
12145	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12146#ifdef CTL_WITH_CA
12147		ctl_clear_mask(lun->have_ca, i);
12148#endif
12149		lun->pending_ua[i] |= ua_type;
12150	}
12151	mtx_unlock(&lun->lun_lock);
12152
12153	return (0);
12154}
12155
12156static void
12157ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12158    int other_sc)
12159{
12160	union ctl_io *xio;
12161
12162	mtx_assert(&lun->lun_lock, MA_OWNED);
12163
12164	/*
12165	 * Run through the OOA queue and attempt to find the given I/O.
12166	 * The target port, initiator ID, tag type and tag number have to
12167	 * match the values that we got from the initiator.  If we have an
12168	 * untagged command to abort, simply abort the first untagged command
12169	 * we come to.  We only allow one untagged command at a time of course.
12170	 */
12171	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12172	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12173
12174		if ((targ_port == UINT32_MAX ||
12175		     targ_port == xio->io_hdr.nexus.targ_port) &&
12176		    (init_id == UINT32_MAX ||
12177		     init_id == xio->io_hdr.nexus.initid.id)) {
12178			if (targ_port != xio->io_hdr.nexus.targ_port ||
12179			    init_id != xio->io_hdr.nexus.initid.id)
12180				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12181			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12182			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12183				union ctl_ha_msg msg_info;
12184
12185				msg_info.hdr.nexus = xio->io_hdr.nexus;
12186				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12187				msg_info.task.tag_num = xio->scsiio.tag_num;
12188				msg_info.task.tag_type = xio->scsiio.tag_type;
12189				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12190				msg_info.hdr.original_sc = NULL;
12191				msg_info.hdr.serializing_sc = NULL;
12192				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12193				    (void *)&msg_info, sizeof(msg_info), 0);
12194			}
12195		}
12196	}
12197}
12198
12199static int
12200ctl_abort_task_set(union ctl_io *io)
12201{
12202	struct ctl_softc *softc = control_softc;
12203	struct ctl_lun *lun;
12204	uint32_t targ_lun;
12205
12206	/*
12207	 * Look up the LUN.
12208	 */
12209	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12210	mtx_lock(&softc->ctl_lock);
12211	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12212		lun = softc->ctl_luns[targ_lun];
12213	else {
12214		mtx_unlock(&softc->ctl_lock);
12215		return (1);
12216	}
12217
12218	mtx_lock(&lun->lun_lock);
12219	mtx_unlock(&softc->ctl_lock);
12220	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12221		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12222		    io->io_hdr.nexus.initid.id,
12223		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12224	} else { /* CTL_TASK_CLEAR_TASK_SET */
12225		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12226		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12227	}
12228	mtx_unlock(&lun->lun_lock);
12229	return (0);
12230}
12231
12232static int
12233ctl_i_t_nexus_reset(union ctl_io *io)
12234{
12235	struct ctl_softc *softc = control_softc;
12236	struct ctl_lun *lun;
12237	uint32_t initindex, residx;
12238
12239	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12240	residx = ctl_get_resindex(&io->io_hdr.nexus);
12241	mtx_lock(&softc->ctl_lock);
12242	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12243		mtx_lock(&lun->lun_lock);
12244		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12245		    io->io_hdr.nexus.initid.id,
12246		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12247#ifdef CTL_WITH_CA
12248		ctl_clear_mask(lun->have_ca, initindex);
12249#endif
12250		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12251			lun->flags &= ~CTL_LUN_RESERVED;
12252		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12253		mtx_unlock(&lun->lun_lock);
12254	}
12255	mtx_unlock(&softc->ctl_lock);
12256	return (0);
12257}
12258
12259static int
12260ctl_abort_task(union ctl_io *io)
12261{
12262	union ctl_io *xio;
12263	struct ctl_lun *lun;
12264	struct ctl_softc *ctl_softc;
12265#if 0
12266	struct sbuf sb;
12267	char printbuf[128];
12268#endif
12269	int found;
12270	uint32_t targ_lun;
12271
12272	ctl_softc = control_softc;
12273	found = 0;
12274
12275	/*
12276	 * Look up the LUN.
12277	 */
12278	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12279	mtx_lock(&ctl_softc->ctl_lock);
12280	if ((targ_lun < CTL_MAX_LUNS)
12281	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12282		lun = ctl_softc->ctl_luns[targ_lun];
12283	else {
12284		mtx_unlock(&ctl_softc->ctl_lock);
12285		return (1);
12286	}
12287
12288#if 0
12289	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12290	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12291#endif
12292
12293	mtx_lock(&lun->lun_lock);
12294	mtx_unlock(&ctl_softc->ctl_lock);
12295	/*
12296	 * Run through the OOA queue and attempt to find the given I/O.
12297	 * The target port, initiator ID, tag type and tag number have to
12298	 * match the values that we got from the initiator.  If we have an
12299	 * untagged command to abort, simply abort the first untagged command
12300	 * we come to.  We only allow one untagged command at a time of course.
12301	 */
12302#if 0
12303	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12304#endif
12305	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12306	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12307#if 0
12308		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12309
12310		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12311			    lun->lun, xio->scsiio.tag_num,
12312			    xio->scsiio.tag_type,
12313			    (xio->io_hdr.blocked_links.tqe_prev
12314			    == NULL) ? "" : " BLOCKED",
12315			    (xio->io_hdr.flags &
12316			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12317			    (xio->io_hdr.flags &
12318			    CTL_FLAG_ABORT) ? " ABORT" : "",
12319			    (xio->io_hdr.flags &
12320			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12321		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12322		sbuf_finish(&sb);
12323		printf("%s\n", sbuf_data(&sb));
12324#endif
12325
12326		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12327		 && (xio->io_hdr.nexus.initid.id ==
12328		     io->io_hdr.nexus.initid.id)) {
12329			/*
12330			 * If the abort says that the task is untagged, the
12331			 * task in the queue must be untagged.  Otherwise,
12332			 * we just check to see whether the tag numbers
12333			 * match.  This is because the QLogic firmware
12334			 * doesn't pass back the tag type in an abort
12335			 * request.
12336			 */
12337#if 0
12338			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12339			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12340			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12341#endif
12342			/*
12343			 * XXX KDM we've got problems with FC, because it
12344			 * doesn't send down a tag type with aborts.  So we
12345			 * can only really go by the tag number...
12346			 * This may cause problems with parallel SCSI.
12347			 * Need to figure that out!!
12348			 */
12349			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12350				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12351				found = 1;
12352				if ((io->io_hdr.flags &
12353				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12354				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12355					union ctl_ha_msg msg_info;
12356
12357					io->io_hdr.flags |=
12358					                CTL_FLAG_SENT_2OTHER_SC;
12359					msg_info.hdr.nexus = io->io_hdr.nexus;
12360					msg_info.task.task_action =
12361						CTL_TASK_ABORT_TASK;
12362					msg_info.task.tag_num =
12363						io->taskio.tag_num;
12364					msg_info.task.tag_type =
12365						io->taskio.tag_type;
12366					msg_info.hdr.msg_type =
12367						CTL_MSG_MANAGE_TASKS;
12368					msg_info.hdr.original_sc = NULL;
12369					msg_info.hdr.serializing_sc = NULL;
12370#if 0
12371					printf("Sent Abort to other side\n");
12372#endif
12373					if (CTL_HA_STATUS_SUCCESS !=
12374					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12375		    				(void *)&msg_info,
12376						sizeof(msg_info), 0)) {
12377					}
12378				}
12379#if 0
12380				printf("ctl_abort_task: found I/O to abort\n");
12381#endif
12382				break;
12383			}
12384		}
12385	}
12386	mtx_unlock(&lun->lun_lock);
12387
12388	if (found == 0) {
12389		/*
12390		 * This isn't really an error.  It's entirely possible for
12391		 * the abort and command completion to cross on the wire.
12392		 * This is more of an informative/diagnostic error.
12393		 */
12394#if 0
12395		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12396		       "%d:%d:%d:%d tag %d type %d\n",
12397		       io->io_hdr.nexus.initid.id,
12398		       io->io_hdr.nexus.targ_port,
12399		       io->io_hdr.nexus.targ_target.id,
12400		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12401		       io->taskio.tag_type);
12402#endif
12403	}
12404	return (0);
12405}
12406
12407static void
12408ctl_run_task(union ctl_io *io)
12409{
12410	struct ctl_softc *ctl_softc = control_softc;
12411	int retval = 1;
12412	const char *task_desc;
12413
12414	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12415
12416	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12417	    ("ctl_run_task: Unextected io_type %d\n",
12418	     io->io_hdr.io_type));
12419
12420	task_desc = ctl_scsi_task_string(&io->taskio);
12421	if (task_desc != NULL) {
12422#ifdef NEEDTOPORT
12423		csevent_log(CSC_CTL | CSC_SHELF_SW |
12424			    CTL_TASK_REPORT,
12425			    csevent_LogType_Trace,
12426			    csevent_Severity_Information,
12427			    csevent_AlertLevel_Green,
12428			    csevent_FRU_Firmware,
12429			    csevent_FRU_Unknown,
12430			    "CTL: received task: %s",task_desc);
12431#endif
12432	} else {
12433#ifdef NEEDTOPORT
12434		csevent_log(CSC_CTL | CSC_SHELF_SW |
12435			    CTL_TASK_REPORT,
12436			    csevent_LogType_Trace,
12437			    csevent_Severity_Information,
12438			    csevent_AlertLevel_Green,
12439			    csevent_FRU_Firmware,
12440			    csevent_FRU_Unknown,
12441			    "CTL: received unknown task "
12442			    "type: %d (%#x)",
12443			    io->taskio.task_action,
12444			    io->taskio.task_action);
12445#endif
12446	}
12447	switch (io->taskio.task_action) {
12448	case CTL_TASK_ABORT_TASK:
12449		retval = ctl_abort_task(io);
12450		break;
12451	case CTL_TASK_ABORT_TASK_SET:
12452	case CTL_TASK_CLEAR_TASK_SET:
12453		retval = ctl_abort_task_set(io);
12454		break;
12455	case CTL_TASK_CLEAR_ACA:
12456		break;
12457	case CTL_TASK_I_T_NEXUS_RESET:
12458		retval = ctl_i_t_nexus_reset(io);
12459		break;
12460	case CTL_TASK_LUN_RESET: {
12461		struct ctl_lun *lun;
12462		uint32_t targ_lun;
12463
12464		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12465		mtx_lock(&ctl_softc->ctl_lock);
12466		if ((targ_lun < CTL_MAX_LUNS)
12467		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12468			lun = ctl_softc->ctl_luns[targ_lun];
12469		else {
12470			mtx_unlock(&ctl_softc->ctl_lock);
12471			retval = 1;
12472			break;
12473		}
12474
12475		if (!(io->io_hdr.flags &
12476		    CTL_FLAG_FROM_OTHER_SC)) {
12477			union ctl_ha_msg msg_info;
12478
12479			io->io_hdr.flags |=
12480				CTL_FLAG_SENT_2OTHER_SC;
12481			msg_info.hdr.msg_type =
12482				CTL_MSG_MANAGE_TASKS;
12483			msg_info.hdr.nexus = io->io_hdr.nexus;
12484			msg_info.task.task_action =
12485				CTL_TASK_LUN_RESET;
12486			msg_info.hdr.original_sc = NULL;
12487			msg_info.hdr.serializing_sc = NULL;
12488			if (CTL_HA_STATUS_SUCCESS !=
12489			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12490			    (void *)&msg_info,
12491			    sizeof(msg_info), 0)) {
12492			}
12493		}
12494
12495		retval = ctl_lun_reset(lun, io,
12496				       CTL_UA_LUN_RESET);
12497		mtx_unlock(&ctl_softc->ctl_lock);
12498		break;
12499	}
12500	case CTL_TASK_TARGET_RESET:
12501		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12502		break;
12503	case CTL_TASK_BUS_RESET:
12504		retval = ctl_bus_reset(ctl_softc, io);
12505		break;
12506	case CTL_TASK_PORT_LOGIN:
12507		break;
12508	case CTL_TASK_PORT_LOGOUT:
12509		break;
12510	default:
12511		printf("ctl_run_task: got unknown task management event %d\n",
12512		       io->taskio.task_action);
12513		break;
12514	}
12515	if (retval == 0)
12516		io->io_hdr.status = CTL_SUCCESS;
12517	else
12518		io->io_hdr.status = CTL_ERROR;
12519	ctl_done(io);
12520}
12521
12522/*
12523 * For HA operation.  Handle commands that come in from the other
12524 * controller.
12525 */
12526static void
12527ctl_handle_isc(union ctl_io *io)
12528{
12529	int free_io;
12530	struct ctl_lun *lun;
12531	struct ctl_softc *ctl_softc;
12532	uint32_t targ_lun;
12533
12534	ctl_softc = control_softc;
12535
12536	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12537	lun = ctl_softc->ctl_luns[targ_lun];
12538
12539	switch (io->io_hdr.msg_type) {
12540	case CTL_MSG_SERIALIZE:
12541		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12542		break;
12543	case CTL_MSG_R2R: {
12544		const struct ctl_cmd_entry *entry;
12545
12546		/*
12547		 * This is only used in SER_ONLY mode.
12548		 */
12549		free_io = 0;
12550		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12551		mtx_lock(&lun->lun_lock);
12552		if (ctl_scsiio_lun_check(ctl_softc, lun,
12553		    entry, (struct ctl_scsiio *)io) != 0) {
12554			mtx_unlock(&lun->lun_lock);
12555			ctl_done(io);
12556			break;
12557		}
12558		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12559		mtx_unlock(&lun->lun_lock);
12560		ctl_enqueue_rtr(io);
12561		break;
12562	}
12563	case CTL_MSG_FINISH_IO:
12564		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12565			free_io = 0;
12566			ctl_done(io);
12567		} else {
12568			free_io = 1;
12569			mtx_lock(&lun->lun_lock);
12570			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12571				     ooa_links);
12572			ctl_check_blocked(lun);
12573			mtx_unlock(&lun->lun_lock);
12574		}
12575		break;
12576	case CTL_MSG_PERS_ACTION:
12577		ctl_hndl_per_res_out_on_other_sc(
12578			(union ctl_ha_msg *)&io->presio.pr_msg);
12579		free_io = 1;
12580		break;
12581	case CTL_MSG_BAD_JUJU:
12582		free_io = 0;
12583		ctl_done(io);
12584		break;
12585	case CTL_MSG_DATAMOVE:
12586		/* Only used in XFER mode */
12587		free_io = 0;
12588		ctl_datamove_remote(io);
12589		break;
12590	case CTL_MSG_DATAMOVE_DONE:
12591		/* Only used in XFER mode */
12592		free_io = 0;
12593		io->scsiio.be_move_done(io);
12594		break;
12595	default:
12596		free_io = 1;
12597		printf("%s: Invalid message type %d\n",
12598		       __func__, io->io_hdr.msg_type);
12599		break;
12600	}
12601	if (free_io)
12602		ctl_free_io(io);
12603
12604}
12605
12606
12607/*
12608 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12609 * there is no match.
12610 */
12611static ctl_lun_error_pattern
12612ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12613{
12614	const struct ctl_cmd_entry *entry;
12615	ctl_lun_error_pattern filtered_pattern, pattern;
12616
12617	pattern = desc->error_pattern;
12618
12619	/*
12620	 * XXX KDM we need more data passed into this function to match a
12621	 * custom pattern, and we actually need to implement custom pattern
12622	 * matching.
12623	 */
12624	if (pattern & CTL_LUN_PAT_CMD)
12625		return (CTL_LUN_PAT_CMD);
12626
12627	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12628		return (CTL_LUN_PAT_ANY);
12629
12630	entry = ctl_get_cmd_entry(ctsio, NULL);
12631
12632	filtered_pattern = entry->pattern & pattern;
12633
12634	/*
12635	 * If the user requested specific flags in the pattern (e.g.
12636	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12637	 * flags.
12638	 *
12639	 * If the user did not specify any flags, it doesn't matter whether
12640	 * or not the command supports the flags.
12641	 */
12642	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12643	     (pattern & ~CTL_LUN_PAT_MASK))
12644		return (CTL_LUN_PAT_NONE);
12645
12646	/*
12647	 * If the user asked for a range check, see if the requested LBA
12648	 * range overlaps with this command's LBA range.
12649	 */
12650	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12651		uint64_t lba1;
12652		uint64_t len1;
12653		ctl_action action;
12654		int retval;
12655
12656		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12657		if (retval != 0)
12658			return (CTL_LUN_PAT_NONE);
12659
12660		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12661					      desc->lba_range.len);
12662		/*
12663		 * A "pass" means that the LBA ranges don't overlap, so
12664		 * this doesn't match the user's range criteria.
12665		 */
12666		if (action == CTL_ACTION_PASS)
12667			return (CTL_LUN_PAT_NONE);
12668	}
12669
12670	return (filtered_pattern);
12671}
12672
12673static void
12674ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12675{
12676	struct ctl_error_desc *desc, *desc2;
12677
12678	mtx_assert(&lun->lun_lock, MA_OWNED);
12679
12680	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12681		ctl_lun_error_pattern pattern;
12682		/*
12683		 * Check to see whether this particular command matches
12684		 * the pattern in the descriptor.
12685		 */
12686		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12687		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12688			continue;
12689
12690		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12691		case CTL_LUN_INJ_ABORTED:
12692			ctl_set_aborted(&io->scsiio);
12693			break;
12694		case CTL_LUN_INJ_MEDIUM_ERR:
12695			ctl_set_medium_error(&io->scsiio);
12696			break;
12697		case CTL_LUN_INJ_UA:
12698			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12699			 * OCCURRED */
12700			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12701			break;
12702		case CTL_LUN_INJ_CUSTOM:
12703			/*
12704			 * We're assuming the user knows what he is doing.
12705			 * Just copy the sense information without doing
12706			 * checks.
12707			 */
12708			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12709			      ctl_min(sizeof(desc->custom_sense),
12710				      sizeof(io->scsiio.sense_data)));
12711			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12712			io->scsiio.sense_len = SSD_FULL_SIZE;
12713			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12714			break;
12715		case CTL_LUN_INJ_NONE:
12716		default:
12717			/*
12718			 * If this is an error injection type we don't know
12719			 * about, clear the continuous flag (if it is set)
12720			 * so it will get deleted below.
12721			 */
12722			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12723			break;
12724		}
12725		/*
12726		 * By default, each error injection action is a one-shot
12727		 */
12728		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12729			continue;
12730
12731		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12732
12733		free(desc, M_CTL);
12734	}
12735}
12736
12737#ifdef CTL_IO_DELAY
12738static void
12739ctl_datamove_timer_wakeup(void *arg)
12740{
12741	union ctl_io *io;
12742
12743	io = (union ctl_io *)arg;
12744
12745	ctl_datamove(io);
12746}
12747#endif /* CTL_IO_DELAY */
12748
12749void
12750ctl_datamove(union ctl_io *io)
12751{
12752	void (*fe_datamove)(union ctl_io *io);
12753
12754	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12755
12756	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12757
12758#ifdef CTL_TIME_IO
12759	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12760		char str[256];
12761		char path_str[64];
12762		struct sbuf sb;
12763
12764		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12765		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12766
12767		sbuf_cat(&sb, path_str);
12768		switch (io->io_hdr.io_type) {
12769		case CTL_IO_SCSI:
12770			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12771			sbuf_printf(&sb, "\n");
12772			sbuf_cat(&sb, path_str);
12773			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12774				    io->scsiio.tag_num, io->scsiio.tag_type);
12775			break;
12776		case CTL_IO_TASK:
12777			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12778				    "Tag Type: %d\n", io->taskio.task_action,
12779				    io->taskio.tag_num, io->taskio.tag_type);
12780			break;
12781		default:
12782			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12783			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12784			break;
12785		}
12786		sbuf_cat(&sb, path_str);
12787		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12788			    (intmax_t)time_uptime - io->io_hdr.start_time);
12789		sbuf_finish(&sb);
12790		printf("%s", sbuf_data(&sb));
12791	}
12792#endif /* CTL_TIME_IO */
12793
12794#ifdef CTL_IO_DELAY
12795	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12796		struct ctl_lun *lun;
12797
12798		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12799
12800		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12801	} else {
12802		struct ctl_lun *lun;
12803
12804		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12805		if ((lun != NULL)
12806		 && (lun->delay_info.datamove_delay > 0)) {
12807			struct callout *callout;
12808
12809			callout = (struct callout *)&io->io_hdr.timer_bytes;
12810			callout_init(callout, /*mpsafe*/ 1);
12811			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12812			callout_reset(callout,
12813				      lun->delay_info.datamove_delay * hz,
12814				      ctl_datamove_timer_wakeup, io);
12815			if (lun->delay_info.datamove_type ==
12816			    CTL_DELAY_TYPE_ONESHOT)
12817				lun->delay_info.datamove_delay = 0;
12818			return;
12819		}
12820	}
12821#endif
12822
12823	/*
12824	 * This command has been aborted.  Set the port status, so we fail
12825	 * the data move.
12826	 */
12827	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12828		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12829		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12830		       io->io_hdr.nexus.targ_port,
12831		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12832		       io->io_hdr.nexus.targ_lun);
12833		io->io_hdr.port_status = 31337;
12834		/*
12835		 * Note that the backend, in this case, will get the
12836		 * callback in its context.  In other cases it may get
12837		 * called in the frontend's interrupt thread context.
12838		 */
12839		io->scsiio.be_move_done(io);
12840		return;
12841	}
12842
12843	/*
12844	 * If we're in XFER mode and this I/O is from the other shelf
12845	 * controller, we need to send the DMA to the other side to
12846	 * actually transfer the data to/from the host.  In serialize only
12847	 * mode the transfer happens below CTL and ctl_datamove() is only
12848	 * called on the machine that originally received the I/O.
12849	 */
12850	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12851	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12852		union ctl_ha_msg msg;
12853		uint32_t sg_entries_sent;
12854		int do_sg_copy;
12855		int i;
12856
12857		memset(&msg, 0, sizeof(msg));
12858		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12859		msg.hdr.original_sc = io->io_hdr.original_sc;
12860		msg.hdr.serializing_sc = io;
12861		msg.hdr.nexus = io->io_hdr.nexus;
12862		msg.dt.flags = io->io_hdr.flags;
12863		/*
12864		 * We convert everything into a S/G list here.  We can't
12865		 * pass by reference, only by value between controllers.
12866		 * So we can't pass a pointer to the S/G list, only as many
12867		 * S/G entries as we can fit in here.  If it's possible for
12868		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12869		 * then we need to break this up into multiple transfers.
12870		 */
12871		if (io->scsiio.kern_sg_entries == 0) {
12872			msg.dt.kern_sg_entries = 1;
12873			/*
12874			 * If this is in cached memory, flush the cache
12875			 * before we send the DMA request to the other
12876			 * controller.  We want to do this in either the
12877			 * read or the write case.  The read case is
12878			 * straightforward.  In the write case, we want to
12879			 * make sure nothing is in the local cache that
12880			 * could overwrite the DMAed data.
12881			 */
12882			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12883				/*
12884				 * XXX KDM use bus_dmamap_sync() here.
12885				 */
12886			}
12887
12888			/*
12889			 * Convert to a physical address if this is a
12890			 * virtual address.
12891			 */
12892			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12893				msg.dt.sg_list[0].addr =
12894					io->scsiio.kern_data_ptr;
12895			} else {
12896				/*
12897				 * XXX KDM use busdma here!
12898				 */
12899#if 0
12900				msg.dt.sg_list[0].addr = (void *)
12901					vtophys(io->scsiio.kern_data_ptr);
12902#endif
12903			}
12904
12905			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12906			do_sg_copy = 0;
12907		} else {
12908			struct ctl_sg_entry *sgl;
12909
12910			do_sg_copy = 1;
12911			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12912			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12913			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12914				/*
12915				 * XXX KDM use bus_dmamap_sync() here.
12916				 */
12917			}
12918		}
12919
12920		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12921		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12922		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12923		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12924		msg.dt.sg_sequence = 0;
12925
12926		/*
12927		 * Loop until we've sent all of the S/G entries.  On the
12928		 * other end, we'll recompose these S/G entries into one
12929		 * contiguous list before passing it to the
12930		 */
12931		for (sg_entries_sent = 0; sg_entries_sent <
12932		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12933			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12934				sizeof(msg.dt.sg_list[0])),
12935				msg.dt.kern_sg_entries - sg_entries_sent);
12936
12937			if (do_sg_copy != 0) {
12938				struct ctl_sg_entry *sgl;
12939				int j;
12940
12941				sgl = (struct ctl_sg_entry *)
12942					io->scsiio.kern_data_ptr;
12943				/*
12944				 * If this is in cached memory, flush the cache
12945				 * before we send the DMA request to the other
12946				 * controller.  We want to do this in either
12947				 * the * read or the write case.  The read
12948				 * case is straightforward.  In the write
12949				 * case, we want to make sure nothing is
12950				 * in the local cache that could overwrite
12951				 * the DMAed data.
12952				 */
12953
12954				for (i = sg_entries_sent, j = 0;
12955				     i < msg.dt.cur_sg_entries; i++, j++) {
12956					if ((io->io_hdr.flags &
12957					     CTL_FLAG_NO_DATASYNC) == 0) {
12958						/*
12959						 * XXX KDM use bus_dmamap_sync()
12960						 */
12961					}
12962					if ((io->io_hdr.flags &
12963					     CTL_FLAG_BUS_ADDR) == 0) {
12964						/*
12965						 * XXX KDM use busdma.
12966						 */
12967#if 0
12968						msg.dt.sg_list[j].addr =(void *)
12969						       vtophys(sgl[i].addr);
12970#endif
12971					} else {
12972						msg.dt.sg_list[j].addr =
12973							sgl[i].addr;
12974					}
12975					msg.dt.sg_list[j].len = sgl[i].len;
12976				}
12977			}
12978
12979			sg_entries_sent += msg.dt.cur_sg_entries;
12980			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12981				msg.dt.sg_last = 1;
12982			else
12983				msg.dt.sg_last = 0;
12984
12985			/*
12986			 * XXX KDM drop and reacquire the lock here?
12987			 */
12988			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12989			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12990				/*
12991				 * XXX do something here.
12992				 */
12993			}
12994
12995			msg.dt.sent_sg_entries = sg_entries_sent;
12996		}
12997		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12998		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12999			ctl_failover_io(io, /*have_lock*/ 0);
13000
13001	} else {
13002
13003		/*
13004		 * Lookup the fe_datamove() function for this particular
13005		 * front end.
13006		 */
13007		fe_datamove =
13008		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13009
13010		fe_datamove(io);
13011	}
13012}
13013
13014static void
13015ctl_send_datamove_done(union ctl_io *io, int have_lock)
13016{
13017	union ctl_ha_msg msg;
13018	int isc_status;
13019
13020	memset(&msg, 0, sizeof(msg));
13021
13022	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
13023	msg.hdr.original_sc = io;
13024	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13025	msg.hdr.nexus = io->io_hdr.nexus;
13026	msg.hdr.status = io->io_hdr.status;
13027	msg.scsi.tag_num = io->scsiio.tag_num;
13028	msg.scsi.tag_type = io->scsiio.tag_type;
13029	msg.scsi.scsi_status = io->scsiio.scsi_status;
13030	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13031	       sizeof(io->scsiio.sense_data));
13032	msg.scsi.sense_len = io->scsiio.sense_len;
13033	msg.scsi.sense_residual = io->scsiio.sense_residual;
13034	msg.scsi.fetd_status = io->io_hdr.port_status;
13035	msg.scsi.residual = io->scsiio.residual;
13036	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13037
13038	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13039		ctl_failover_io(io, /*have_lock*/ have_lock);
13040		return;
13041	}
13042
13043	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
13044	if (isc_status > CTL_HA_STATUS_SUCCESS) {
13045		/* XXX do something if this fails */
13046	}
13047
13048}
13049
13050/*
13051 * The DMA to the remote side is done, now we need to tell the other side
13052 * we're done so it can continue with its data movement.
13053 */
13054static void
13055ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
13056{
13057	union ctl_io *io;
13058
13059	io = rq->context;
13060
13061	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13062		printf("%s: ISC DMA write failed with error %d", __func__,
13063		       rq->ret);
13064		ctl_set_internal_failure(&io->scsiio,
13065					 /*sks_valid*/ 1,
13066					 /*retry_count*/ rq->ret);
13067	}
13068
13069	ctl_dt_req_free(rq);
13070
13071	/*
13072	 * In this case, we had to malloc the memory locally.  Free it.
13073	 */
13074	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13075		int i;
13076		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13077			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13078	}
13079	/*
13080	 * The data is in local and remote memory, so now we need to send
13081	 * status (good or back) back to the other side.
13082	 */
13083	ctl_send_datamove_done(io, /*have_lock*/ 0);
13084}
13085
13086/*
13087 * We've moved the data from the host/controller into local memory.  Now we
13088 * need to push it over to the remote controller's memory.
13089 */
13090static int
13091ctl_datamove_remote_dm_write_cb(union ctl_io *io)
13092{
13093	int retval;
13094
13095	retval = 0;
13096
13097	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
13098					  ctl_datamove_remote_write_cb);
13099
13100	return (retval);
13101}
13102
13103static void
13104ctl_datamove_remote_write(union ctl_io *io)
13105{
13106	int retval;
13107	void (*fe_datamove)(union ctl_io *io);
13108
13109	/*
13110	 * - Get the data from the host/HBA into local memory.
13111	 * - DMA memory from the local controller to the remote controller.
13112	 * - Send status back to the remote controller.
13113	 */
13114
13115	retval = ctl_datamove_remote_sgl_setup(io);
13116	if (retval != 0)
13117		return;
13118
13119	/* Switch the pointer over so the FETD knows what to do */
13120	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13121
13122	/*
13123	 * Use a custom move done callback, since we need to send completion
13124	 * back to the other controller, not to the backend on this side.
13125	 */
13126	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
13127
13128	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13129
13130	fe_datamove(io);
13131
13132	return;
13133
13134}
13135
13136static int
13137ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13138{
13139#if 0
13140	char str[256];
13141	char path_str[64];
13142	struct sbuf sb;
13143#endif
13144
13145	/*
13146	 * In this case, we had to malloc the memory locally.  Free it.
13147	 */
13148	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13149		int i;
13150		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13151			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13152	}
13153
13154#if 0
13155	scsi_path_string(io, path_str, sizeof(path_str));
13156	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13157	sbuf_cat(&sb, path_str);
13158	scsi_command_string(&io->scsiio, NULL, &sb);
13159	sbuf_printf(&sb, "\n");
13160	sbuf_cat(&sb, path_str);
13161	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13162		    io->scsiio.tag_num, io->scsiio.tag_type);
13163	sbuf_cat(&sb, path_str);
13164	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13165		    io->io_hdr.flags, io->io_hdr.status);
13166	sbuf_finish(&sb);
13167	printk("%s", sbuf_data(&sb));
13168#endif
13169
13170
13171	/*
13172	 * The read is done, now we need to send status (good or bad) back
13173	 * to the other side.
13174	 */
13175	ctl_send_datamove_done(io, /*have_lock*/ 0);
13176
13177	return (0);
13178}
13179
13180static void
13181ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13182{
13183	union ctl_io *io;
13184	void (*fe_datamove)(union ctl_io *io);
13185
13186	io = rq->context;
13187
13188	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13189		printf("%s: ISC DMA read failed with error %d", __func__,
13190		       rq->ret);
13191		ctl_set_internal_failure(&io->scsiio,
13192					 /*sks_valid*/ 1,
13193					 /*retry_count*/ rq->ret);
13194	}
13195
13196	ctl_dt_req_free(rq);
13197
13198	/* Switch the pointer over so the FETD knows what to do */
13199	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13200
13201	/*
13202	 * Use a custom move done callback, since we need to send completion
13203	 * back to the other controller, not to the backend on this side.
13204	 */
13205	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13206
13207	/* XXX KDM add checks like the ones in ctl_datamove? */
13208
13209	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13210
13211	fe_datamove(io);
13212}
13213
13214static int
13215ctl_datamove_remote_sgl_setup(union ctl_io *io)
13216{
13217	struct ctl_sg_entry *local_sglist, *remote_sglist;
13218	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13219	struct ctl_softc *softc;
13220	int retval;
13221	int i;
13222
13223	retval = 0;
13224	softc = control_softc;
13225
13226	local_sglist = io->io_hdr.local_sglist;
13227	local_dma_sglist = io->io_hdr.local_dma_sglist;
13228	remote_sglist = io->io_hdr.remote_sglist;
13229	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13230
13231	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13232		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13233			local_sglist[i].len = remote_sglist[i].len;
13234
13235			/*
13236			 * XXX Detect the situation where the RS-level I/O
13237			 * redirector on the other side has already read the
13238			 * data off of the AOR RS on this side, and
13239			 * transferred it to remote (mirror) memory on the
13240			 * other side.  Since we already have the data in
13241			 * memory here, we just need to use it.
13242			 *
13243			 * XXX KDM this can probably be removed once we
13244			 * get the cache device code in and take the
13245			 * current AOR implementation out.
13246			 */
13247#ifdef NEEDTOPORT
13248			if ((remote_sglist[i].addr >=
13249			     (void *)vtophys(softc->mirr->addr))
13250			 && (remote_sglist[i].addr <
13251			     ((void *)vtophys(softc->mirr->addr) +
13252			     CacheMirrorOffset))) {
13253				local_sglist[i].addr = remote_sglist[i].addr -
13254					CacheMirrorOffset;
13255				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13256				     CTL_FLAG_DATA_IN)
13257					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13258			} else {
13259				local_sglist[i].addr = remote_sglist[i].addr +
13260					CacheMirrorOffset;
13261			}
13262#endif
13263#if 0
13264			printf("%s: local %p, remote %p, len %d\n",
13265			       __func__, local_sglist[i].addr,
13266			       remote_sglist[i].addr, local_sglist[i].len);
13267#endif
13268		}
13269	} else {
13270		uint32_t len_to_go;
13271
13272		/*
13273		 * In this case, we don't have automatically allocated
13274		 * memory for this I/O on this controller.  This typically
13275		 * happens with internal CTL I/O -- e.g. inquiry, mode
13276		 * sense, etc.  Anything coming from RAIDCore will have
13277		 * a mirror area available.
13278		 */
13279		len_to_go = io->scsiio.kern_data_len;
13280
13281		/*
13282		 * Clear the no datasync flag, we have to use malloced
13283		 * buffers.
13284		 */
13285		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13286
13287		/*
13288		 * The difficult thing here is that the size of the various
13289		 * S/G segments may be different than the size from the
13290		 * remote controller.  That'll make it harder when DMAing
13291		 * the data back to the other side.
13292		 */
13293		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13294		     sizeof(io->io_hdr.remote_sglist[0])) &&
13295		     (len_to_go > 0); i++) {
13296			local_sglist[i].len = ctl_min(len_to_go, 131072);
13297			CTL_SIZE_8B(local_dma_sglist[i].len,
13298				    local_sglist[i].len);
13299			local_sglist[i].addr =
13300				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13301
13302			local_dma_sglist[i].addr = local_sglist[i].addr;
13303
13304			if (local_sglist[i].addr == NULL) {
13305				int j;
13306
13307				printf("malloc failed for %zd bytes!",
13308				       local_dma_sglist[i].len);
13309				for (j = 0; j < i; j++) {
13310					free(local_sglist[j].addr, M_CTL);
13311				}
13312				ctl_set_internal_failure(&io->scsiio,
13313							 /*sks_valid*/ 1,
13314							 /*retry_count*/ 4857);
13315				retval = 1;
13316				goto bailout_error;
13317
13318			}
13319			/* XXX KDM do we need a sync here? */
13320
13321			len_to_go -= local_sglist[i].len;
13322		}
13323		/*
13324		 * Reset the number of S/G entries accordingly.  The
13325		 * original number of S/G entries is available in
13326		 * rem_sg_entries.
13327		 */
13328		io->scsiio.kern_sg_entries = i;
13329
13330#if 0
13331		printf("%s: kern_sg_entries = %d\n", __func__,
13332		       io->scsiio.kern_sg_entries);
13333		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13334			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13335			       local_sglist[i].addr, local_sglist[i].len,
13336			       local_dma_sglist[i].len);
13337#endif
13338	}
13339
13340
13341	return (retval);
13342
13343bailout_error:
13344
13345	ctl_send_datamove_done(io, /*have_lock*/ 0);
13346
13347	return (retval);
13348}
13349
13350static int
13351ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13352			 ctl_ha_dt_cb callback)
13353{
13354	struct ctl_ha_dt_req *rq;
13355	struct ctl_sg_entry *remote_sglist, *local_sglist;
13356	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13357	uint32_t local_used, remote_used, total_used;
13358	int retval;
13359	int i, j;
13360
13361	retval = 0;
13362
13363	rq = ctl_dt_req_alloc();
13364
13365	/*
13366	 * If we failed to allocate the request, and if the DMA didn't fail
13367	 * anyway, set busy status.  This is just a resource allocation
13368	 * failure.
13369	 */
13370	if ((rq == NULL)
13371	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13372		ctl_set_busy(&io->scsiio);
13373
13374	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13375
13376		if (rq != NULL)
13377			ctl_dt_req_free(rq);
13378
13379		/*
13380		 * The data move failed.  We need to return status back
13381		 * to the other controller.  No point in trying to DMA
13382		 * data to the remote controller.
13383		 */
13384
13385		ctl_send_datamove_done(io, /*have_lock*/ 0);
13386
13387		retval = 1;
13388
13389		goto bailout;
13390	}
13391
13392	local_sglist = io->io_hdr.local_sglist;
13393	local_dma_sglist = io->io_hdr.local_dma_sglist;
13394	remote_sglist = io->io_hdr.remote_sglist;
13395	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13396	local_used = 0;
13397	remote_used = 0;
13398	total_used = 0;
13399
13400	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13401		rq->ret = CTL_HA_STATUS_SUCCESS;
13402		rq->context = io;
13403		callback(rq);
13404		goto bailout;
13405	}
13406
13407	/*
13408	 * Pull/push the data over the wire from/to the other controller.
13409	 * This takes into account the possibility that the local and
13410	 * remote sglists may not be identical in terms of the size of
13411	 * the elements and the number of elements.
13412	 *
13413	 * One fundamental assumption here is that the length allocated for
13414	 * both the local and remote sglists is identical.  Otherwise, we've
13415	 * essentially got a coding error of some sort.
13416	 */
13417	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13418		int isc_ret;
13419		uint32_t cur_len, dma_length;
13420		uint8_t *tmp_ptr;
13421
13422		rq->id = CTL_HA_DATA_CTL;
13423		rq->command = command;
13424		rq->context = io;
13425
13426		/*
13427		 * Both pointers should be aligned.  But it is possible
13428		 * that the allocation length is not.  They should both
13429		 * also have enough slack left over at the end, though,
13430		 * to round up to the next 8 byte boundary.
13431		 */
13432		cur_len = ctl_min(local_sglist[i].len - local_used,
13433				  remote_sglist[j].len - remote_used);
13434
13435		/*
13436		 * In this case, we have a size issue and need to decrease
13437		 * the size, except in the case where we actually have less
13438		 * than 8 bytes left.  In that case, we need to increase
13439		 * the DMA length to get the last bit.
13440		 */
13441		if ((cur_len & 0x7) != 0) {
13442			if (cur_len > 0x7) {
13443				cur_len = cur_len - (cur_len & 0x7);
13444				dma_length = cur_len;
13445			} else {
13446				CTL_SIZE_8B(dma_length, cur_len);
13447			}
13448
13449		} else
13450			dma_length = cur_len;
13451
13452		/*
13453		 * If we had to allocate memory for this I/O, instead of using
13454		 * the non-cached mirror memory, we'll need to flush the cache
13455		 * before trying to DMA to the other controller.
13456		 *
13457		 * We could end up doing this multiple times for the same
13458		 * segment if we have a larger local segment than remote
13459		 * segment.  That shouldn't be an issue.
13460		 */
13461		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13462			/*
13463			 * XXX KDM use bus_dmamap_sync() here.
13464			 */
13465		}
13466
13467		rq->size = dma_length;
13468
13469		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13470		tmp_ptr += local_used;
13471
13472		/* Use physical addresses when talking to ISC hardware */
13473		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13474			/* XXX KDM use busdma */
13475#if 0
13476			rq->local = vtophys(tmp_ptr);
13477#endif
13478		} else
13479			rq->local = tmp_ptr;
13480
13481		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13482		tmp_ptr += remote_used;
13483		rq->remote = tmp_ptr;
13484
13485		rq->callback = NULL;
13486
13487		local_used += cur_len;
13488		if (local_used >= local_sglist[i].len) {
13489			i++;
13490			local_used = 0;
13491		}
13492
13493		remote_used += cur_len;
13494		if (remote_used >= remote_sglist[j].len) {
13495			j++;
13496			remote_used = 0;
13497		}
13498		total_used += cur_len;
13499
13500		if (total_used >= io->scsiio.kern_data_len)
13501			rq->callback = callback;
13502
13503		if ((rq->size & 0x7) != 0) {
13504			printf("%s: warning: size %d is not on 8b boundary\n",
13505			       __func__, rq->size);
13506		}
13507		if (((uintptr_t)rq->local & 0x7) != 0) {
13508			printf("%s: warning: local %p not on 8b boundary\n",
13509			       __func__, rq->local);
13510		}
13511		if (((uintptr_t)rq->remote & 0x7) != 0) {
13512			printf("%s: warning: remote %p not on 8b boundary\n",
13513			       __func__, rq->local);
13514		}
13515#if 0
13516		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13517		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13518		       rq->local, rq->remote, rq->size);
13519#endif
13520
13521		isc_ret = ctl_dt_single(rq);
13522		if (isc_ret == CTL_HA_STATUS_WAIT)
13523			continue;
13524
13525		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13526			rq->ret = CTL_HA_STATUS_SUCCESS;
13527		} else {
13528			rq->ret = isc_ret;
13529		}
13530		callback(rq);
13531		goto bailout;
13532	}
13533
13534bailout:
13535	return (retval);
13536
13537}
13538
13539static void
13540ctl_datamove_remote_read(union ctl_io *io)
13541{
13542	int retval;
13543	int i;
13544
13545	/*
13546	 * This will send an error to the other controller in the case of a
13547	 * failure.
13548	 */
13549	retval = ctl_datamove_remote_sgl_setup(io);
13550	if (retval != 0)
13551		return;
13552
13553	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13554					  ctl_datamove_remote_read_cb);
13555	if ((retval != 0)
13556	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13557		/*
13558		 * Make sure we free memory if there was an error..  The
13559		 * ctl_datamove_remote_xfer() function will send the
13560		 * datamove done message, or call the callback with an
13561		 * error if there is a problem.
13562		 */
13563		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13564			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13565	}
13566
13567	return;
13568}
13569
13570/*
13571 * Process a datamove request from the other controller.  This is used for
13572 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13573 * first.  Once that is complete, the data gets DMAed into the remote
13574 * controller's memory.  For reads, we DMA from the remote controller's
13575 * memory into our memory first, and then move it out to the FETD.
13576 */
13577static void
13578ctl_datamove_remote(union ctl_io *io)
13579{
13580	struct ctl_softc *softc;
13581
13582	softc = control_softc;
13583
13584	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13585
13586	/*
13587	 * Note that we look for an aborted I/O here, but don't do some of
13588	 * the other checks that ctl_datamove() normally does.
13589	 * We don't need to run the datamove delay code, since that should
13590	 * have been done if need be on the other controller.
13591	 */
13592	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13593		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13594		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13595		       io->io_hdr.nexus.targ_port,
13596		       io->io_hdr.nexus.targ_target.id,
13597		       io->io_hdr.nexus.targ_lun);
13598		io->io_hdr.port_status = 31338;
13599		ctl_send_datamove_done(io, /*have_lock*/ 0);
13600		return;
13601	}
13602
13603	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13604		ctl_datamove_remote_write(io);
13605	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13606		ctl_datamove_remote_read(io);
13607	} else {
13608		union ctl_ha_msg msg;
13609		struct scsi_sense_data *sense;
13610		uint8_t sks[3];
13611		int retry_count;
13612
13613		memset(&msg, 0, sizeof(msg));
13614
13615		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13616		msg.hdr.status = CTL_SCSI_ERROR;
13617		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13618
13619		retry_count = 4243;
13620
13621		sense = &msg.scsi.sense_data;
13622		sks[0] = SSD_SCS_VALID;
13623		sks[1] = (retry_count >> 8) & 0xff;
13624		sks[2] = retry_count & 0xff;
13625
13626		/* "Internal target failure" */
13627		scsi_set_sense_data(sense,
13628				    /*sense_format*/ SSD_TYPE_NONE,
13629				    /*current_error*/ 1,
13630				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13631				    /*asc*/ 0x44,
13632				    /*ascq*/ 0x00,
13633				    /*type*/ SSD_ELEM_SKS,
13634				    /*size*/ sizeof(sks),
13635				    /*data*/ sks,
13636				    SSD_ELEM_NONE);
13637
13638		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13639		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13640			ctl_failover_io(io, /*have_lock*/ 1);
13641			return;
13642		}
13643
13644		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13645		    CTL_HA_STATUS_SUCCESS) {
13646			/* XXX KDM what to do if this fails? */
13647		}
13648		return;
13649	}
13650
13651}
13652
13653static int
13654ctl_process_done(union ctl_io *io)
13655{
13656	struct ctl_lun *lun;
13657	struct ctl_softc *ctl_softc;
13658	void (*fe_done)(union ctl_io *io);
13659	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13660
13661	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13662
13663	fe_done =
13664	    control_softc->ctl_ports[targ_port]->fe_done;
13665
13666#ifdef CTL_TIME_IO
13667	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13668		char str[256];
13669		char path_str[64];
13670		struct sbuf sb;
13671
13672		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13673		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13674
13675		sbuf_cat(&sb, path_str);
13676		switch (io->io_hdr.io_type) {
13677		case CTL_IO_SCSI:
13678			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13679			sbuf_printf(&sb, "\n");
13680			sbuf_cat(&sb, path_str);
13681			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13682				    io->scsiio.tag_num, io->scsiio.tag_type);
13683			break;
13684		case CTL_IO_TASK:
13685			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13686				    "Tag Type: %d\n", io->taskio.task_action,
13687				    io->taskio.tag_num, io->taskio.tag_type);
13688			break;
13689		default:
13690			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13691			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13692			break;
13693		}
13694		sbuf_cat(&sb, path_str);
13695		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13696			    (intmax_t)time_uptime - io->io_hdr.start_time);
13697		sbuf_finish(&sb);
13698		printf("%s", sbuf_data(&sb));
13699	}
13700#endif /* CTL_TIME_IO */
13701
13702	switch (io->io_hdr.io_type) {
13703	case CTL_IO_SCSI:
13704		break;
13705	case CTL_IO_TASK:
13706		if (bootverbose || verbose > 0)
13707			ctl_io_error_print(io, NULL);
13708		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13709			ctl_free_io(io);
13710		else
13711			fe_done(io);
13712		return (CTL_RETVAL_COMPLETE);
13713		break;
13714	default:
13715		printf("ctl_process_done: invalid io type %d\n",
13716		       io->io_hdr.io_type);
13717		panic("ctl_process_done: invalid io type %d\n",
13718		      io->io_hdr.io_type);
13719		break; /* NOTREACHED */
13720	}
13721
13722	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13723	if (lun == NULL) {
13724		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13725				 io->io_hdr.nexus.targ_mapped_lun));
13726		fe_done(io);
13727		goto bailout;
13728	}
13729	ctl_softc = lun->ctl_softc;
13730
13731	mtx_lock(&lun->lun_lock);
13732
13733	/*
13734	 * Check to see if we have any errors to inject here.  We only
13735	 * inject errors for commands that don't already have errors set.
13736	 */
13737	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13738	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13739		ctl_inject_error(lun, io);
13740
13741	/*
13742	 * XXX KDM how do we treat commands that aren't completed
13743	 * successfully?
13744	 *
13745	 * XXX KDM should we also track I/O latency?
13746	 */
13747	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13748	    io->io_hdr.io_type == CTL_IO_SCSI) {
13749#ifdef CTL_TIME_IO
13750		struct bintime cur_bt;
13751#endif
13752		int type;
13753
13754		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13755		    CTL_FLAG_DATA_IN)
13756			type = CTL_STATS_READ;
13757		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13758		    CTL_FLAG_DATA_OUT)
13759			type = CTL_STATS_WRITE;
13760		else
13761			type = CTL_STATS_NO_IO;
13762
13763		lun->stats.ports[targ_port].bytes[type] +=
13764		    io->scsiio.kern_total_len;
13765		lun->stats.ports[targ_port].operations[type]++;
13766#ifdef CTL_TIME_IO
13767		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13768		   &io->io_hdr.dma_bt);
13769		lun->stats.ports[targ_port].num_dmas[type] +=
13770		    io->io_hdr.num_dmas;
13771		getbintime(&cur_bt);
13772		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13773		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13774#endif
13775	}
13776
13777	/*
13778	 * Remove this from the OOA queue.
13779	 */
13780	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13781
13782	/*
13783	 * Run through the blocked queue on this LUN and see if anything
13784	 * has become unblocked, now that this transaction is done.
13785	 */
13786	ctl_check_blocked(lun);
13787
13788	/*
13789	 * If the LUN has been invalidated, free it if there is nothing
13790	 * left on its OOA queue.
13791	 */
13792	if ((lun->flags & CTL_LUN_INVALID)
13793	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13794		mtx_unlock(&lun->lun_lock);
13795		mtx_lock(&ctl_softc->ctl_lock);
13796		ctl_free_lun(lun);
13797		mtx_unlock(&ctl_softc->ctl_lock);
13798	} else
13799		mtx_unlock(&lun->lun_lock);
13800
13801	/*
13802	 * If this command has been aborted, make sure we set the status
13803	 * properly.  The FETD is responsible for freeing the I/O and doing
13804	 * whatever it needs to do to clean up its state.
13805	 */
13806	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13807		ctl_set_task_aborted(&io->scsiio);
13808
13809	/*
13810	 * We print out status for every task management command.  For SCSI
13811	 * commands, we filter out any unit attention errors; they happen
13812	 * on every boot, and would clutter up the log.  Note:  task
13813	 * management commands aren't printed here, they are printed above,
13814	 * since they should never even make it down here.
13815	 */
13816	switch (io->io_hdr.io_type) {
13817	case CTL_IO_SCSI: {
13818		int error_code, sense_key, asc, ascq;
13819
13820		sense_key = 0;
13821
13822		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13823		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13824			/*
13825			 * Since this is just for printing, no need to
13826			 * show errors here.
13827			 */
13828			scsi_extract_sense_len(&io->scsiio.sense_data,
13829					       io->scsiio.sense_len,
13830					       &error_code,
13831					       &sense_key,
13832					       &asc,
13833					       &ascq,
13834					       /*show_errors*/ 0);
13835		}
13836
13837		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13838		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13839		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13840		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13841
13842			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13843				ctl_softc->skipped_prints++;
13844			} else {
13845				uint32_t skipped_prints;
13846
13847				skipped_prints = ctl_softc->skipped_prints;
13848
13849				ctl_softc->skipped_prints = 0;
13850				ctl_softc->last_print_jiffies = time_uptime;
13851
13852				if (skipped_prints > 0) {
13853#ifdef NEEDTOPORT
13854					csevent_log(CSC_CTL | CSC_SHELF_SW |
13855					    CTL_ERROR_REPORT,
13856					    csevent_LogType_Trace,
13857					    csevent_Severity_Information,
13858					    csevent_AlertLevel_Green,
13859					    csevent_FRU_Firmware,
13860					    csevent_FRU_Unknown,
13861					    "High CTL error volume, %d prints "
13862					    "skipped", skipped_prints);
13863#endif
13864				}
13865				if (bootverbose || verbose > 0)
13866					ctl_io_error_print(io, NULL);
13867			}
13868		}
13869		break;
13870	}
13871	case CTL_IO_TASK:
13872		if (bootverbose || verbose > 0)
13873			ctl_io_error_print(io, NULL);
13874		break;
13875	default:
13876		break;
13877	}
13878
13879	/*
13880	 * Tell the FETD or the other shelf controller we're done with this
13881	 * command.  Note that only SCSI commands get to this point.  Task
13882	 * management commands are completed above.
13883	 *
13884	 * We only send status to the other controller if we're in XFER
13885	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13886	 * received the I/O (from CTL's perspective), and so the status is
13887	 * generated there.
13888	 *
13889	 * XXX KDM if we hold the lock here, we could cause a deadlock
13890	 * if the frontend comes back in in this context to queue
13891	 * something.
13892	 */
13893	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13894	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13895		union ctl_ha_msg msg;
13896
13897		memset(&msg, 0, sizeof(msg));
13898		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13899		msg.hdr.original_sc = io->io_hdr.original_sc;
13900		msg.hdr.nexus = io->io_hdr.nexus;
13901		msg.hdr.status = io->io_hdr.status;
13902		msg.scsi.scsi_status = io->scsiio.scsi_status;
13903		msg.scsi.tag_num = io->scsiio.tag_num;
13904		msg.scsi.tag_type = io->scsiio.tag_type;
13905		msg.scsi.sense_len = io->scsiio.sense_len;
13906		msg.scsi.sense_residual = io->scsiio.sense_residual;
13907		msg.scsi.residual = io->scsiio.residual;
13908		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13909		       sizeof(io->scsiio.sense_data));
13910		/*
13911		 * We copy this whether or not this is an I/O-related
13912		 * command.  Otherwise, we'd have to go and check to see
13913		 * whether it's a read/write command, and it really isn't
13914		 * worth it.
13915		 */
13916		memcpy(&msg.scsi.lbalen,
13917		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13918		       sizeof(msg.scsi.lbalen));
13919
13920		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13921				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13922			/* XXX do something here */
13923		}
13924
13925		ctl_free_io(io);
13926	} else
13927		fe_done(io);
13928
13929bailout:
13930
13931	return (CTL_RETVAL_COMPLETE);
13932}
13933
13934#ifdef CTL_WITH_CA
13935/*
13936 * Front end should call this if it doesn't do autosense.  When the request
13937 * sense comes back in from the initiator, we'll dequeue this and send it.
13938 */
13939int
13940ctl_queue_sense(union ctl_io *io)
13941{
13942	struct ctl_lun *lun;
13943	struct ctl_softc *ctl_softc;
13944	uint32_t initidx, targ_lun;
13945
13946	ctl_softc = control_softc;
13947
13948	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13949
13950	/*
13951	 * LUN lookup will likely move to the ctl_work_thread() once we
13952	 * have our new queueing infrastructure (that doesn't put things on
13953	 * a per-LUN queue initially).  That is so that we can handle
13954	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13955	 * can't deal with that right now.
13956	 */
13957	mtx_lock(&ctl_softc->ctl_lock);
13958
13959	/*
13960	 * If we don't have a LUN for this, just toss the sense
13961	 * information.
13962	 */
13963	targ_lun = io->io_hdr.nexus.targ_lun;
13964	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13965	if ((targ_lun < CTL_MAX_LUNS)
13966	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13967		lun = ctl_softc->ctl_luns[targ_lun];
13968	else
13969		goto bailout;
13970
13971	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13972
13973	mtx_lock(&lun->lun_lock);
13974	/*
13975	 * Already have CA set for this LUN...toss the sense information.
13976	 */
13977	if (ctl_is_set(lun->have_ca, initidx)) {
13978		mtx_unlock(&lun->lun_lock);
13979		goto bailout;
13980	}
13981
13982	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13983	       ctl_min(sizeof(lun->pending_sense[initidx]),
13984	       sizeof(io->scsiio.sense_data)));
13985	ctl_set_mask(lun->have_ca, initidx);
13986	mtx_unlock(&lun->lun_lock);
13987
13988bailout:
13989	mtx_unlock(&ctl_softc->ctl_lock);
13990
13991	ctl_free_io(io);
13992
13993	return (CTL_RETVAL_COMPLETE);
13994}
13995#endif
13996
13997/*
13998 * Primary command inlet from frontend ports.  All SCSI and task I/O
13999 * requests must go through this function.
14000 */
14001int
14002ctl_queue(union ctl_io *io)
14003{
14004	struct ctl_softc *ctl_softc;
14005
14006	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
14007
14008	ctl_softc = control_softc;
14009
14010#ifdef CTL_TIME_IO
14011	io->io_hdr.start_time = time_uptime;
14012	getbintime(&io->io_hdr.start_bt);
14013#endif /* CTL_TIME_IO */
14014
14015	/* Map FE-specific LUN ID into global one. */
14016	io->io_hdr.nexus.targ_mapped_lun =
14017	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
14018
14019	switch (io->io_hdr.io_type) {
14020	case CTL_IO_SCSI:
14021	case CTL_IO_TASK:
14022		ctl_enqueue_incoming(io);
14023		break;
14024	default:
14025		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
14026		return (EINVAL);
14027	}
14028
14029	return (CTL_RETVAL_COMPLETE);
14030}
14031
14032#ifdef CTL_IO_DELAY
14033static void
14034ctl_done_timer_wakeup(void *arg)
14035{
14036	union ctl_io *io;
14037
14038	io = (union ctl_io *)arg;
14039	ctl_done(io);
14040}
14041#endif /* CTL_IO_DELAY */
14042
14043void
14044ctl_done(union ctl_io *io)
14045{
14046	struct ctl_softc *ctl_softc;
14047
14048	ctl_softc = control_softc;
14049
14050	/*
14051	 * Enable this to catch duplicate completion issues.
14052	 */
14053#if 0
14054	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
14055		printf("%s: type %d msg %d cdb %x iptl: "
14056		       "%d:%d:%d:%d tag 0x%04x "
14057		       "flag %#x status %x\n",
14058			__func__,
14059			io->io_hdr.io_type,
14060			io->io_hdr.msg_type,
14061			io->scsiio.cdb[0],
14062			io->io_hdr.nexus.initid.id,
14063			io->io_hdr.nexus.targ_port,
14064			io->io_hdr.nexus.targ_target.id,
14065			io->io_hdr.nexus.targ_lun,
14066			(io->io_hdr.io_type ==
14067			CTL_IO_TASK) ?
14068			io->taskio.tag_num :
14069			io->scsiio.tag_num,
14070		        io->io_hdr.flags,
14071			io->io_hdr.status);
14072	} else
14073		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
14074#endif
14075
14076	/*
14077	 * This is an internal copy of an I/O, and should not go through
14078	 * the normal done processing logic.
14079	 */
14080	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
14081		return;
14082
14083	/*
14084	 * We need to send a msg to the serializing shelf to finish the IO
14085	 * as well.  We don't send a finish message to the other shelf if
14086	 * this is a task management command.  Task management commands
14087	 * aren't serialized in the OOA queue, but rather just executed on
14088	 * both shelf controllers for commands that originated on that
14089	 * controller.
14090	 */
14091	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
14092	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
14093		union ctl_ha_msg msg_io;
14094
14095		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
14096		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
14097		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
14098		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
14099		}
14100		/* continue on to finish IO */
14101	}
14102#ifdef CTL_IO_DELAY
14103	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
14104		struct ctl_lun *lun;
14105
14106		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14107
14108		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
14109	} else {
14110		struct ctl_lun *lun;
14111
14112		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14113
14114		if ((lun != NULL)
14115		 && (lun->delay_info.done_delay > 0)) {
14116			struct callout *callout;
14117
14118			callout = (struct callout *)&io->io_hdr.timer_bytes;
14119			callout_init(callout, /*mpsafe*/ 1);
14120			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
14121			callout_reset(callout,
14122				      lun->delay_info.done_delay * hz,
14123				      ctl_done_timer_wakeup, io);
14124			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
14125				lun->delay_info.done_delay = 0;
14126			return;
14127		}
14128	}
14129#endif /* CTL_IO_DELAY */
14130
14131	ctl_enqueue_done(io);
14132}
14133
14134int
14135ctl_isc(struct ctl_scsiio *ctsio)
14136{
14137	struct ctl_lun *lun;
14138	int retval;
14139
14140	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14141
14142	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14143
14144	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14145
14146	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14147
14148	return (retval);
14149}
14150
14151
14152static void
14153ctl_work_thread(void *arg)
14154{
14155	struct ctl_thread *thr = (struct ctl_thread *)arg;
14156	struct ctl_softc *softc = thr->ctl_softc;
14157	union ctl_io *io;
14158	int retval;
14159
14160	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14161
14162	for (;;) {
14163		retval = 0;
14164
14165		/*
14166		 * We handle the queues in this order:
14167		 * - ISC
14168		 * - done queue (to free up resources, unblock other commands)
14169		 * - RtR queue
14170		 * - incoming queue
14171		 *
14172		 * If those queues are empty, we break out of the loop and
14173		 * go to sleep.
14174		 */
14175		mtx_lock(&thr->queue_lock);
14176		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14177		if (io != NULL) {
14178			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14179			mtx_unlock(&thr->queue_lock);
14180			ctl_handle_isc(io);
14181			continue;
14182		}
14183		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14184		if (io != NULL) {
14185			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14186			/* clear any blocked commands, call fe_done */
14187			mtx_unlock(&thr->queue_lock);
14188			retval = ctl_process_done(io);
14189			continue;
14190		}
14191		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14192		if (io != NULL) {
14193			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14194			mtx_unlock(&thr->queue_lock);
14195			if (io->io_hdr.io_type == CTL_IO_TASK)
14196				ctl_run_task(io);
14197			else
14198				ctl_scsiio_precheck(softc, &io->scsiio);
14199			continue;
14200		}
14201		if (!ctl_pause_rtr) {
14202			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14203			if (io != NULL) {
14204				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14205				mtx_unlock(&thr->queue_lock);
14206				retval = ctl_scsiio(&io->scsiio);
14207				if (retval != CTL_RETVAL_COMPLETE)
14208					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14209				continue;
14210			}
14211		}
14212
14213		/* Sleep until we have something to do. */
14214		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14215	}
14216}
14217
14218static void
14219ctl_lun_thread(void *arg)
14220{
14221	struct ctl_softc *softc = (struct ctl_softc *)arg;
14222	struct ctl_be_lun *be_lun;
14223	int retval;
14224
14225	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14226
14227	for (;;) {
14228		retval = 0;
14229		mtx_lock(&softc->ctl_lock);
14230		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14231		if (be_lun != NULL) {
14232			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14233			mtx_unlock(&softc->ctl_lock);
14234			ctl_create_lun(be_lun);
14235			continue;
14236		}
14237
14238		/* Sleep until we have something to do. */
14239		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14240		    PDROP | PRIBIO, "-", 0);
14241	}
14242}
14243
14244static void
14245ctl_enqueue_incoming(union ctl_io *io)
14246{
14247	struct ctl_softc *softc = control_softc;
14248	struct ctl_thread *thr;
14249	u_int idx;
14250
14251	idx = (io->io_hdr.nexus.targ_port * 127 +
14252	       io->io_hdr.nexus.initid.id) % worker_threads;
14253	thr = &softc->threads[idx];
14254	mtx_lock(&thr->queue_lock);
14255	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14256	mtx_unlock(&thr->queue_lock);
14257	wakeup(thr);
14258}
14259
14260static void
14261ctl_enqueue_rtr(union ctl_io *io)
14262{
14263	struct ctl_softc *softc = control_softc;
14264	struct ctl_thread *thr;
14265
14266	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14267	mtx_lock(&thr->queue_lock);
14268	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14269	mtx_unlock(&thr->queue_lock);
14270	wakeup(thr);
14271}
14272
14273static void
14274ctl_enqueue_done(union ctl_io *io)
14275{
14276	struct ctl_softc *softc = control_softc;
14277	struct ctl_thread *thr;
14278
14279	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14280	mtx_lock(&thr->queue_lock);
14281	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14282	mtx_unlock(&thr->queue_lock);
14283	wakeup(thr);
14284}
14285
14286static void
14287ctl_enqueue_isc(union ctl_io *io)
14288{
14289	struct ctl_softc *softc = control_softc;
14290	struct ctl_thread *thr;
14291
14292	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14293	mtx_lock(&thr->queue_lock);
14294	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14295	mtx_unlock(&thr->queue_lock);
14296	wakeup(thr);
14297}
14298
14299/* Initialization and failover */
14300
14301void
14302ctl_init_isc_msg(void)
14303{
14304	printf("CTL: Still calling this thing\n");
14305}
14306
14307/*
14308 * Init component
14309 * 	Initializes component into configuration defined by bootMode
14310 *	(see hasc-sv.c)
14311 *  	returns hasc_Status:
14312 * 		OK
14313 *		ERROR - fatal error
14314 */
14315static ctl_ha_comp_status
14316ctl_isc_init(struct ctl_ha_component *c)
14317{
14318	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14319
14320	c->status = ret;
14321	return ret;
14322}
14323
14324/* Start component
14325 * 	Starts component in state requested. If component starts successfully,
14326 *	it must set its own state to the requestrd state
14327 *	When requested state is HASC_STATE_HA, the component may refine it
14328 * 	by adding _SLAVE or _MASTER flags.
14329 *	Currently allowed state transitions are:
14330 *	UNKNOWN->HA		- initial startup
14331 *	UNKNOWN->SINGLE - initial startup when no parter detected
14332 *	HA->SINGLE		- failover
14333 * returns ctl_ha_comp_status:
14334 * 		OK	- component successfully started in requested state
14335 *		FAILED  - could not start the requested state, failover may
14336 * 			  be possible
14337 *		ERROR	- fatal error detected, no future startup possible
14338 */
14339static ctl_ha_comp_status
14340ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14341{
14342	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14343
14344	printf("%s: go\n", __func__);
14345
14346	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14347	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14348		ctl_is_single = 0;
14349		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14350		    != CTL_HA_STATUS_SUCCESS) {
14351			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14352			ret = CTL_HA_COMP_STATUS_ERROR;
14353		}
14354	} else if (CTL_HA_STATE_IS_HA(c->state)
14355		&& CTL_HA_STATE_IS_SINGLE(state)){
14356		// HA->SINGLE transition
14357	        ctl_failover();
14358		ctl_is_single = 1;
14359	} else {
14360		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14361		       c->state, state);
14362		ret = CTL_HA_COMP_STATUS_ERROR;
14363	}
14364	if (CTL_HA_STATE_IS_SINGLE(state))
14365		ctl_is_single = 1;
14366
14367	c->state = state;
14368	c->status = ret;
14369	return ret;
14370}
14371
14372/*
14373 * Quiesce component
14374 * The component must clear any error conditions (set status to OK) and
14375 * prepare itself to another Start call
14376 * returns ctl_ha_comp_status:
14377 * 	OK
14378 *	ERROR
14379 */
14380static ctl_ha_comp_status
14381ctl_isc_quiesce(struct ctl_ha_component *c)
14382{
14383	int ret = CTL_HA_COMP_STATUS_OK;
14384
14385	ctl_pause_rtr = 1;
14386	c->status = ret;
14387	return ret;
14388}
14389
14390struct ctl_ha_component ctl_ha_component_ctlisc =
14391{
14392	.name = "CTL ISC",
14393	.state = CTL_HA_STATE_UNKNOWN,
14394	.init = ctl_isc_init,
14395	.start = ctl_isc_start,
14396	.quiesce = ctl_isc_quiesce
14397};
14398
14399/*
14400 *  vim: ts=8
14401 */
14402