ctl.c revision 271354
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 271354 2014-09-10 06:35:00Z 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*/0,
284	/*eca_and_aen*/0,
285	/*flags4*/SCP_TAS,
286	/*aen_holdoff_period*/{0, 0},
287	/*busy_timeout_period*/{0, 0},
288	/*extended_selftest_completion_time*/{0, 0}
289};
290
291static struct scsi_control_page control_page_changeable = {
292	/*page_code*/SMS_CONTROL_MODE_PAGE,
293	/*page_length*/sizeof(struct scsi_control_page) - 2,
294	/*rlec*/SCP_DSENSE,
295	/*queue_flags*/0,
296	/*eca_and_aen*/0,
297	/*flags4*/0,
298	/*aen_holdoff_period*/{0, 0},
299	/*busy_timeout_period*/{0, 0},
300	/*extended_selftest_completion_time*/{0, 0}
301};
302
303
304/*
305 * XXX KDM move these into the softc.
306 */
307static int rcv_sync_msg;
308static int persis_offset;
309static uint8_t ctl_pause_rtr;
310static int     ctl_is_single = 1;
311static int     index_to_aps_page;
312
313SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
314static int worker_threads = -1;
315SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
316    &worker_threads, 1, "Number of worker threads");
317static int verbose = 0;
318SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
319    &verbose, 0, "Show SCSI errors returned to initiator");
320
321/*
322 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
323 * 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	9
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_mpp(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
385					 int alloc_len);
386static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
387					 int alloc_len);
388static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
389static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
390static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
391static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
392static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
393static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
394static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
395					 union ctl_io *ooa_io);
396static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
397				union ctl_io *starting_io);
398static int ctl_check_blocked(struct ctl_lun *lun);
399static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
400				struct ctl_lun *lun,
401				const struct ctl_cmd_entry *entry,
402				struct ctl_scsiio *ctsio);
403//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
404static void ctl_failover(void);
405static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
406			       struct ctl_scsiio *ctsio);
407static int ctl_scsiio(struct ctl_scsiio *ctsio);
408
409static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
410static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
411			    ctl_ua_type ua_type);
412static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
413			 ctl_ua_type ua_type);
414static int ctl_abort_task(union ctl_io *io);
415static int ctl_abort_task_set(union ctl_io *io);
416static int ctl_i_t_nexus_reset(union ctl_io *io);
417static void ctl_run_task(union ctl_io *io);
418#ifdef CTL_IO_DELAY
419static void ctl_datamove_timer_wakeup(void *arg);
420static void ctl_done_timer_wakeup(void *arg);
421#endif /* CTL_IO_DELAY */
422
423static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
424static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
425static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
426static void ctl_datamove_remote_write(union ctl_io *io);
427static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
428static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
429static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
430static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
431				    ctl_ha_dt_cb callback);
432static void ctl_datamove_remote_read(union ctl_io *io);
433static void ctl_datamove_remote(union ctl_io *io);
434static int ctl_process_done(union ctl_io *io);
435static void ctl_lun_thread(void *arg);
436static void ctl_work_thread(void *arg);
437static void ctl_enqueue_incoming(union ctl_io *io);
438static void ctl_enqueue_rtr(union ctl_io *io);
439static void ctl_enqueue_done(union ctl_io *io);
440static void ctl_enqueue_isc(union ctl_io *io);
441static const struct ctl_cmd_entry *
442    ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
443static const struct ctl_cmd_entry *
444    ctl_validate_command(struct ctl_scsiio *ctsio);
445static int ctl_cmd_applicable(uint8_t lun_type,
446    const struct ctl_cmd_entry *entry);
447
448/*
449 * Load the serialization table.  This isn't very pretty, but is probably
450 * the easiest way to do it.
451 */
452#include "ctl_ser_table.c"
453
454/*
455 * We only need to define open, close and ioctl routines for this driver.
456 */
457static struct cdevsw ctl_cdevsw = {
458	.d_version =	D_VERSION,
459	.d_flags =	0,
460	.d_open =	ctl_open,
461	.d_close =	ctl_close,
462	.d_ioctl =	ctl_ioctl,
463	.d_name =	"ctl",
464};
465
466
467MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
468MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
469
470static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
471
472static moduledata_t ctl_moduledata = {
473	"ctl",
474	ctl_module_event_handler,
475	NULL
476};
477
478DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
479MODULE_VERSION(ctl, 1);
480
481static struct ctl_frontend ioctl_frontend =
482{
483	.name = "ioctl",
484};
485
486static void
487ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
488			    union ctl_ha_msg *msg_info)
489{
490	struct ctl_scsiio *ctsio;
491
492	if (msg_info->hdr.original_sc == NULL) {
493		printf("%s: original_sc == NULL!\n", __func__);
494		/* XXX KDM now what? */
495		return;
496	}
497
498	ctsio = &msg_info->hdr.original_sc->scsiio;
499	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
500	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
501	ctsio->io_hdr.status = msg_info->hdr.status;
502	ctsio->scsi_status = msg_info->scsi.scsi_status;
503	ctsio->sense_len = msg_info->scsi.sense_len;
504	ctsio->sense_residual = msg_info->scsi.sense_residual;
505	ctsio->residual = msg_info->scsi.residual;
506	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
507	       sizeof(ctsio->sense_data));
508	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
509	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
510	ctl_enqueue_isc((union ctl_io *)ctsio);
511}
512
513static void
514ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
515				union ctl_ha_msg *msg_info)
516{
517	struct ctl_scsiio *ctsio;
518
519	if (msg_info->hdr.serializing_sc == NULL) {
520		printf("%s: serializing_sc == NULL!\n", __func__);
521		/* XXX KDM now what? */
522		return;
523	}
524
525	ctsio = &msg_info->hdr.serializing_sc->scsiio;
526#if 0
527	/*
528	 * Attempt to catch the situation where an I/O has
529	 * been freed, and we're using it again.
530	 */
531	if (ctsio->io_hdr.io_type == 0xff) {
532		union ctl_io *tmp_io;
533		tmp_io = (union ctl_io *)ctsio;
534		printf("%s: %p use after free!\n", __func__,
535		       ctsio);
536		printf("%s: type %d msg %d cdb %x iptl: "
537		       "%d:%d:%d:%d tag 0x%04x "
538		       "flag %#x status %x\n",
539			__func__,
540			tmp_io->io_hdr.io_type,
541			tmp_io->io_hdr.msg_type,
542			tmp_io->scsiio.cdb[0],
543			tmp_io->io_hdr.nexus.initid.id,
544			tmp_io->io_hdr.nexus.targ_port,
545			tmp_io->io_hdr.nexus.targ_target.id,
546			tmp_io->io_hdr.nexus.targ_lun,
547			(tmp_io->io_hdr.io_type ==
548			CTL_IO_TASK) ?
549			tmp_io->taskio.tag_num :
550			tmp_io->scsiio.tag_num,
551		        tmp_io->io_hdr.flags,
552			tmp_io->io_hdr.status);
553	}
554#endif
555	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
556	ctl_enqueue_isc((union ctl_io *)ctsio);
557}
558
559/*
560 * ISC (Inter Shelf Communication) event handler.  Events from the HA
561 * subsystem come in here.
562 */
563static void
564ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
565{
566	struct ctl_softc *ctl_softc;
567	union ctl_io *io;
568	struct ctl_prio *presio;
569	ctl_ha_status isc_status;
570
571	ctl_softc = control_softc;
572	io = NULL;
573
574
575#if 0
576	printf("CTL: Isc Msg event %d\n", event);
577#endif
578	if (event == CTL_HA_EVT_MSG_RECV) {
579		union ctl_ha_msg msg_info;
580
581		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
582					     sizeof(msg_info), /*wait*/ 0);
583#if 0
584		printf("CTL: msg_type %d\n", msg_info.msg_type);
585#endif
586		if (isc_status != 0) {
587			printf("Error receiving message, status = %d\n",
588			       isc_status);
589			return;
590		}
591
592		switch (msg_info.hdr.msg_type) {
593		case CTL_MSG_SERIALIZE:
594#if 0
595			printf("Serialize\n");
596#endif
597			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
598			if (io == NULL) {
599				printf("ctl_isc_event_handler: can't allocate "
600				       "ctl_io!\n");
601				/* Bad Juju */
602				/* Need to set busy and send msg back */
603				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
604				msg_info.hdr.status = CTL_SCSI_ERROR;
605				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
606				msg_info.scsi.sense_len = 0;
607			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
608				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
609				}
610				goto bailout;
611			}
612			ctl_zero_io(io);
613			// populate ctsio from msg_info
614			io->io_hdr.io_type = CTL_IO_SCSI;
615			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
616			io->io_hdr.original_sc = msg_info.hdr.original_sc;
617#if 0
618			printf("pOrig %x\n", (int)msg_info.original_sc);
619#endif
620			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
621					    CTL_FLAG_IO_ACTIVE;
622			/*
623			 * If we're in serialization-only mode, we don't
624			 * want to go through full done processing.  Thus
625			 * the COPY flag.
626			 *
627			 * XXX KDM add another flag that is more specific.
628			 */
629			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
630				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
631			io->io_hdr.nexus = msg_info.hdr.nexus;
632#if 0
633			printf("targ %d, port %d, iid %d, lun %d\n",
634			       io->io_hdr.nexus.targ_target.id,
635			       io->io_hdr.nexus.targ_port,
636			       io->io_hdr.nexus.initid.id,
637			       io->io_hdr.nexus.targ_lun);
638#endif
639			io->scsiio.tag_num = msg_info.scsi.tag_num;
640			io->scsiio.tag_type = msg_info.scsi.tag_type;
641			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
642			       CTL_MAX_CDBLEN);
643			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
644				const struct ctl_cmd_entry *entry;
645
646				entry = ctl_get_cmd_entry(&io->scsiio);
647				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
648				io->io_hdr.flags |=
649					entry->flags & CTL_FLAG_DATA_MASK;
650			}
651			ctl_enqueue_isc(io);
652			break;
653
654		/* Performed on the Originating SC, XFER mode only */
655		case CTL_MSG_DATAMOVE: {
656			struct ctl_sg_entry *sgl;
657			int i, j;
658
659			io = msg_info.hdr.original_sc;
660			if (io == NULL) {
661				printf("%s: original_sc == NULL!\n", __func__);
662				/* XXX KDM do something here */
663				break;
664			}
665			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
666			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
667			/*
668			 * Keep track of this, we need to send it back over
669			 * when the datamove is complete.
670			 */
671			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
672
673			if (msg_info.dt.sg_sequence == 0) {
674				/*
675				 * XXX KDM we use the preallocated S/G list
676				 * here, but we'll need to change this to
677				 * dynamic allocation if we need larger S/G
678				 * lists.
679				 */
680				if (msg_info.dt.kern_sg_entries >
681				    sizeof(io->io_hdr.remote_sglist) /
682				    sizeof(io->io_hdr.remote_sglist[0])) {
683					printf("%s: number of S/G entries "
684					    "needed %u > allocated num %zd\n",
685					    __func__,
686					    msg_info.dt.kern_sg_entries,
687					    sizeof(io->io_hdr.remote_sglist)/
688					    sizeof(io->io_hdr.remote_sglist[0]));
689
690					/*
691					 * XXX KDM send a message back to
692					 * the other side to shut down the
693					 * DMA.  The error will come back
694					 * through via the normal channel.
695					 */
696					break;
697				}
698				sgl = io->io_hdr.remote_sglist;
699				memset(sgl, 0,
700				       sizeof(io->io_hdr.remote_sglist));
701
702				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
703
704				io->scsiio.kern_sg_entries =
705					msg_info.dt.kern_sg_entries;
706				io->scsiio.rem_sg_entries =
707					msg_info.dt.kern_sg_entries;
708				io->scsiio.kern_data_len =
709					msg_info.dt.kern_data_len;
710				io->scsiio.kern_total_len =
711					msg_info.dt.kern_total_len;
712				io->scsiio.kern_data_resid =
713					msg_info.dt.kern_data_resid;
714				io->scsiio.kern_rel_offset =
715					msg_info.dt.kern_rel_offset;
716				/*
717				 * Clear out per-DMA flags.
718				 */
719				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
720				/*
721				 * Add per-DMA flags that are set for this
722				 * particular DMA request.
723				 */
724				io->io_hdr.flags |= msg_info.dt.flags &
725						    CTL_FLAG_RDMA_MASK;
726			} else
727				sgl = (struct ctl_sg_entry *)
728					io->scsiio.kern_data_ptr;
729
730			for (i = msg_info.dt.sent_sg_entries, j = 0;
731			     i < (msg_info.dt.sent_sg_entries +
732			     msg_info.dt.cur_sg_entries); i++, j++) {
733				sgl[i].addr = msg_info.dt.sg_list[j].addr;
734				sgl[i].len = msg_info.dt.sg_list[j].len;
735
736#if 0
737				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
738				       __func__,
739				       msg_info.dt.sg_list[j].addr,
740				       msg_info.dt.sg_list[j].len,
741				       sgl[i].addr, sgl[i].len, j, i);
742#endif
743			}
744#if 0
745			memcpy(&sgl[msg_info.dt.sent_sg_entries],
746			       msg_info.dt.sg_list,
747			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
748#endif
749
750			/*
751			 * If this is the last piece of the I/O, we've got
752			 * the full S/G list.  Queue processing in the thread.
753			 * Otherwise wait for the next piece.
754			 */
755			if (msg_info.dt.sg_last != 0)
756				ctl_enqueue_isc(io);
757			break;
758		}
759		/* Performed on the Serializing (primary) SC, XFER mode only */
760		case CTL_MSG_DATAMOVE_DONE: {
761			if (msg_info.hdr.serializing_sc == NULL) {
762				printf("%s: serializing_sc == NULL!\n",
763				       __func__);
764				/* XXX KDM now what? */
765				break;
766			}
767			/*
768			 * We grab the sense information here in case
769			 * there was a failure, so we can return status
770			 * back to the initiator.
771			 */
772			io = msg_info.hdr.serializing_sc;
773			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
774			io->io_hdr.status = msg_info.hdr.status;
775			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
776			io->scsiio.sense_len = msg_info.scsi.sense_len;
777			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
778			io->io_hdr.port_status = msg_info.scsi.fetd_status;
779			io->scsiio.residual = msg_info.scsi.residual;
780			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
781			       sizeof(io->scsiio.sense_data));
782			ctl_enqueue_isc(io);
783			break;
784		}
785
786		/* Preformed on Originating SC, SER_ONLY mode */
787		case CTL_MSG_R2R:
788			io = msg_info.hdr.original_sc;
789			if (io == NULL) {
790				printf("%s: Major Bummer\n", __func__);
791				return;
792			} else {
793#if 0
794				printf("pOrig %x\n",(int) ctsio);
795#endif
796			}
797			io->io_hdr.msg_type = CTL_MSG_R2R;
798			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
799			ctl_enqueue_isc(io);
800			break;
801
802		/*
803		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
804		 * mode.
805		 * Performed on the Originating (i.e. secondary) SC in XFER
806		 * mode
807		 */
808		case CTL_MSG_FINISH_IO:
809			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
810				ctl_isc_handler_finish_xfer(ctl_softc,
811							    &msg_info);
812			else
813				ctl_isc_handler_finish_ser_only(ctl_softc,
814								&msg_info);
815			break;
816
817		/* Preformed on Originating SC */
818		case CTL_MSG_BAD_JUJU:
819			io = msg_info.hdr.original_sc;
820			if (io == NULL) {
821				printf("%s: Bad JUJU!, original_sc is NULL!\n",
822				       __func__);
823				break;
824			}
825			ctl_copy_sense_data(&msg_info, io);
826			/*
827			 * IO should have already been cleaned up on other
828			 * SC so clear this flag so we won't send a message
829			 * back to finish the IO there.
830			 */
831			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
832			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
833
834			/* io = msg_info.hdr.serializing_sc; */
835			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
836			ctl_enqueue_isc(io);
837			break;
838
839		/* Handle resets sent from the other side */
840		case CTL_MSG_MANAGE_TASKS: {
841			struct ctl_taskio *taskio;
842			taskio = (struct ctl_taskio *)ctl_alloc_io(
843				(void *)ctl_softc->othersc_pool);
844			if (taskio == NULL) {
845				printf("ctl_isc_event_handler: can't allocate "
846				       "ctl_io!\n");
847				/* Bad Juju */
848				/* should I just call the proper reset func
849				   here??? */
850				goto bailout;
851			}
852			ctl_zero_io((union ctl_io *)taskio);
853			taskio->io_hdr.io_type = CTL_IO_TASK;
854			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
855			taskio->io_hdr.nexus = msg_info.hdr.nexus;
856			taskio->task_action = msg_info.task.task_action;
857			taskio->tag_num = msg_info.task.tag_num;
858			taskio->tag_type = msg_info.task.tag_type;
859#ifdef CTL_TIME_IO
860			taskio->io_hdr.start_time = time_uptime;
861			getbintime(&taskio->io_hdr.start_bt);
862#if 0
863			cs_prof_gettime(&taskio->io_hdr.start_ticks);
864#endif
865#endif /* CTL_TIME_IO */
866			ctl_run_task((union ctl_io *)taskio);
867			break;
868		}
869		/* Persistent Reserve action which needs attention */
870		case CTL_MSG_PERS_ACTION:
871			presio = (struct ctl_prio *)ctl_alloc_io(
872				(void *)ctl_softc->othersc_pool);
873			if (presio == NULL) {
874				printf("ctl_isc_event_handler: can't allocate "
875				       "ctl_io!\n");
876				/* Bad Juju */
877				/* Need to set busy and send msg back */
878				goto bailout;
879			}
880			ctl_zero_io((union ctl_io *)presio);
881			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
882			presio->pr_msg = msg_info.pr;
883			ctl_enqueue_isc((union ctl_io *)presio);
884			break;
885		case CTL_MSG_SYNC_FE:
886			rcv_sync_msg = 1;
887			break;
888		case CTL_MSG_APS_LOCK: {
889			// It's quicker to execute this then to
890			// queue it.
891			struct ctl_lun *lun;
892			struct ctl_page_index *page_index;
893			struct copan_aps_subpage *current_sp;
894			uint32_t targ_lun;
895
896			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
897			lun = ctl_softc->ctl_luns[targ_lun];
898			mtx_lock(&lun->lun_lock);
899			page_index = &lun->mode_pages.index[index_to_aps_page];
900			current_sp = (struct copan_aps_subpage *)
901				     (page_index->page_data +
902				     (page_index->page_len * CTL_PAGE_CURRENT));
903
904			current_sp->lock_active = msg_info.aps.lock_flag;
905			mtx_unlock(&lun->lun_lock);
906		        break;
907		}
908		default:
909		        printf("How did I get here?\n");
910		}
911	} else if (event == CTL_HA_EVT_MSG_SENT) {
912		if (param != CTL_HA_STATUS_SUCCESS) {
913			printf("Bad status from ctl_ha_msg_send status %d\n",
914			       param);
915		}
916		return;
917	} else if (event == CTL_HA_EVT_DISCONNECT) {
918		printf("CTL: Got a disconnect from Isc\n");
919		return;
920	} else {
921		printf("ctl_isc_event_handler: Unknown event %d\n", event);
922		return;
923	}
924
925bailout:
926	return;
927}
928
929static void
930ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
931{
932	struct scsi_sense_data *sense;
933
934	sense = &dest->scsiio.sense_data;
935	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
936	dest->scsiio.scsi_status = src->scsi.scsi_status;
937	dest->scsiio.sense_len = src->scsi.sense_len;
938	dest->io_hdr.status = src->hdr.status;
939}
940
941static int
942ctl_init(void)
943{
944	struct ctl_softc *softc;
945	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
946	struct ctl_port *port;
947        uint8_t sc_id =0;
948	int i, error, retval;
949	//int isc_retval;
950
951	retval = 0;
952	ctl_pause_rtr = 0;
953        rcv_sync_msg = 0;
954
955	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
956			       M_WAITOK | M_ZERO);
957	softc = control_softc;
958
959	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
960			      "cam/ctl");
961
962	softc->dev->si_drv1 = softc;
963
964	/*
965	 * By default, return a "bad LUN" peripheral qualifier for unknown
966	 * LUNs.  The user can override this default using the tunable or
967	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
968	 */
969	softc->inquiry_pq_no_lun = 1;
970	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
971			  &softc->inquiry_pq_no_lun);
972	sysctl_ctx_init(&softc->sysctl_ctx);
973	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
974		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
975		CTLFLAG_RD, 0, "CAM Target Layer");
976
977	if (softc->sysctl_tree == NULL) {
978		printf("%s: unable to allocate sysctl tree\n", __func__);
979		destroy_dev(softc->dev);
980		free(control_softc, M_DEVBUF);
981		control_softc = NULL;
982		return (ENOMEM);
983	}
984
985	SYSCTL_ADD_INT(&softc->sysctl_ctx,
986		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
987		       "inquiry_pq_no_lun", CTLFLAG_RW,
988		       &softc->inquiry_pq_no_lun, 0,
989		       "Report no lun possible for invalid LUNs");
990
991	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
992	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
993	softc->open_count = 0;
994
995	/*
996	 * Default to actually sending a SYNCHRONIZE CACHE command down to
997	 * the drive.
998	 */
999	softc->flags = CTL_FLAG_REAL_SYNC;
1000
1001	/*
1002	 * In Copan's HA scheme, the "master" and "slave" roles are
1003	 * figured out through the slot the controller is in.  Although it
1004	 * is an active/active system, someone has to be in charge.
1005 	 */
1006#ifdef NEEDTOPORT
1007        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1008#endif
1009
1010        if (sc_id == 0) {
1011		softc->flags |= CTL_FLAG_MASTER_SHELF;
1012		persis_offset = 0;
1013	} else
1014		persis_offset = CTL_MAX_INITIATORS;
1015
1016	/*
1017	 * XXX KDM need to figure out where we want to get our target ID
1018	 * and WWID.  Is it different on each port?
1019	 */
1020	softc->target.id = 0;
1021	softc->target.wwid[0] = 0x12345678;
1022	softc->target.wwid[1] = 0x87654321;
1023	STAILQ_INIT(&softc->lun_list);
1024	STAILQ_INIT(&softc->pending_lun_queue);
1025	STAILQ_INIT(&softc->fe_list);
1026	STAILQ_INIT(&softc->port_list);
1027	STAILQ_INIT(&softc->be_list);
1028	STAILQ_INIT(&softc->io_pools);
1029	ctl_tpc_init(softc);
1030
1031	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1032			    &internal_pool)!= 0){
1033		printf("ctl: can't allocate %d entry internal pool, "
1034		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1035		return (ENOMEM);
1036	}
1037
1038	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1039			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1040		printf("ctl: can't allocate %d entry emergency pool, "
1041		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1042		ctl_pool_free(internal_pool);
1043		return (ENOMEM);
1044	}
1045
1046	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1047	                    &other_pool) != 0)
1048	{
1049		printf("ctl: can't allocate %d entry other SC pool, "
1050		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1051		ctl_pool_free(internal_pool);
1052		ctl_pool_free(emergency_pool);
1053		return (ENOMEM);
1054	}
1055
1056	softc->internal_pool = internal_pool;
1057	softc->emergency_pool = emergency_pool;
1058	softc->othersc_pool = other_pool;
1059
1060	if (worker_threads <= 0)
1061		worker_threads = max(1, mp_ncpus / 4);
1062	if (worker_threads > CTL_MAX_THREADS)
1063		worker_threads = CTL_MAX_THREADS;
1064
1065	for (i = 0; i < worker_threads; i++) {
1066		struct ctl_thread *thr = &softc->threads[i];
1067
1068		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1069		thr->ctl_softc = softc;
1070		STAILQ_INIT(&thr->incoming_queue);
1071		STAILQ_INIT(&thr->rtr_queue);
1072		STAILQ_INIT(&thr->done_queue);
1073		STAILQ_INIT(&thr->isc_queue);
1074
1075		error = kproc_kthread_add(ctl_work_thread, thr,
1076		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1077		if (error != 0) {
1078			printf("error creating CTL work thread!\n");
1079			ctl_pool_free(internal_pool);
1080			ctl_pool_free(emergency_pool);
1081			ctl_pool_free(other_pool);
1082			return (error);
1083		}
1084	}
1085	error = kproc_kthread_add(ctl_lun_thread, softc,
1086	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1087	if (error != 0) {
1088		printf("error creating CTL lun thread!\n");
1089		ctl_pool_free(internal_pool);
1090		ctl_pool_free(emergency_pool);
1091		ctl_pool_free(other_pool);
1092		return (error);
1093	}
1094	if (bootverbose)
1095		printf("ctl: CAM Target Layer loaded\n");
1096
1097	/*
1098	 * Initialize the ioctl front end.
1099	 */
1100	ctl_frontend_register(&ioctl_frontend);
1101	port = &softc->ioctl_info.port;
1102	port->frontend = &ioctl_frontend;
1103	sprintf(softc->ioctl_info.port_name, "ioctl");
1104	port->port_type = CTL_PORT_IOCTL;
1105	port->num_requested_ctl_io = 100;
1106	port->port_name = softc->ioctl_info.port_name;
1107	port->port_online = ctl_ioctl_online;
1108	port->port_offline = ctl_ioctl_offline;
1109	port->onoff_arg = &softc->ioctl_info;
1110	port->lun_enable = ctl_ioctl_lun_enable;
1111	port->lun_disable = ctl_ioctl_lun_disable;
1112	port->targ_lun_arg = &softc->ioctl_info;
1113	port->fe_datamove = ctl_ioctl_datamove;
1114	port->fe_done = ctl_ioctl_done;
1115	port->max_targets = 15;
1116	port->max_target_id = 15;
1117
1118	if (ctl_port_register(&softc->ioctl_info.port,
1119	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1120		printf("ctl: ioctl front end registration failed, will "
1121		       "continue anyway\n");
1122	}
1123
1124#ifdef CTL_IO_DELAY
1125	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1126		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1127		       sizeof(struct callout), CTL_TIMER_BYTES);
1128		return (EINVAL);
1129	}
1130#endif /* CTL_IO_DELAY */
1131
1132	return (0);
1133}
1134
1135void
1136ctl_shutdown(void)
1137{
1138	struct ctl_softc *softc;
1139	struct ctl_lun *lun, *next_lun;
1140	struct ctl_io_pool *pool;
1141
1142	softc = (struct ctl_softc *)control_softc;
1143
1144	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1145		printf("ctl: ioctl front end deregistration failed\n");
1146
1147	mtx_lock(&softc->ctl_lock);
1148
1149	/*
1150	 * Free up each LUN.
1151	 */
1152	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1153		next_lun = STAILQ_NEXT(lun, links);
1154		ctl_free_lun(lun);
1155	}
1156
1157	mtx_unlock(&softc->ctl_lock);
1158
1159	ctl_frontend_deregister(&ioctl_frontend);
1160
1161	/*
1162	 * This will rip the rug out from under any FETDs or anyone else
1163	 * that has a pool allocated.  Since we increment our module
1164	 * refcount any time someone outside the main CTL module allocates
1165	 * a pool, we shouldn't have any problems here.  The user won't be
1166	 * able to unload the CTL module until client modules have
1167	 * successfully unloaded.
1168	 */
1169	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1170		ctl_pool_free(pool);
1171
1172#if 0
1173	ctl_shutdown_thread(softc->work_thread);
1174	mtx_destroy(&softc->queue_lock);
1175#endif
1176
1177	ctl_tpc_shutdown(softc);
1178	mtx_destroy(&softc->pool_lock);
1179	mtx_destroy(&softc->ctl_lock);
1180
1181	destroy_dev(softc->dev);
1182
1183	sysctl_ctx_free(&softc->sysctl_ctx);
1184
1185	free(control_softc, M_DEVBUF);
1186	control_softc = NULL;
1187
1188	if (bootverbose)
1189		printf("ctl: CAM Target Layer unloaded\n");
1190}
1191
1192static int
1193ctl_module_event_handler(module_t mod, int what, void *arg)
1194{
1195
1196	switch (what) {
1197	case MOD_LOAD:
1198		return (ctl_init());
1199	case MOD_UNLOAD:
1200		return (EBUSY);
1201	default:
1202		return (EOPNOTSUPP);
1203	}
1204}
1205
1206/*
1207 * XXX KDM should we do some access checks here?  Bump a reference count to
1208 * prevent a CTL module from being unloaded while someone has it open?
1209 */
1210static int
1211ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1212{
1213	return (0);
1214}
1215
1216static int
1217ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1218{
1219	return (0);
1220}
1221
1222int
1223ctl_port_enable(ctl_port_type port_type)
1224{
1225	struct ctl_softc *softc;
1226	struct ctl_port *port;
1227
1228	if (ctl_is_single == 0) {
1229		union ctl_ha_msg msg_info;
1230		int isc_retval;
1231
1232#if 0
1233		printf("%s: HA mode, synchronizing frontend enable\n",
1234		        __func__);
1235#endif
1236		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1237	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1238		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1239			printf("Sync msg send error retval %d\n", isc_retval);
1240		}
1241		if (!rcv_sync_msg) {
1242			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1243			        sizeof(msg_info), 1);
1244		}
1245#if 0
1246        	printf("CTL:Frontend Enable\n");
1247	} else {
1248		printf("%s: single mode, skipping frontend synchronization\n",
1249		        __func__);
1250#endif
1251	}
1252
1253	softc = control_softc;
1254
1255	STAILQ_FOREACH(port, &softc->port_list, links) {
1256		if (port_type & port->port_type)
1257		{
1258#if 0
1259			printf("port %d\n", port->targ_port);
1260#endif
1261			ctl_port_online(port);
1262		}
1263	}
1264
1265	return (0);
1266}
1267
1268int
1269ctl_port_disable(ctl_port_type port_type)
1270{
1271	struct ctl_softc *softc;
1272	struct ctl_port *port;
1273
1274	softc = control_softc;
1275
1276	STAILQ_FOREACH(port, &softc->port_list, links) {
1277		if (port_type & port->port_type)
1278			ctl_port_offline(port);
1279	}
1280
1281	return (0);
1282}
1283
1284/*
1285 * Returns 0 for success, 1 for failure.
1286 * Currently the only failure mode is if there aren't enough entries
1287 * allocated.  So, in case of a failure, look at num_entries_dropped,
1288 * reallocate and try again.
1289 */
1290int
1291ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1292	      int *num_entries_filled, int *num_entries_dropped,
1293	      ctl_port_type port_type, int no_virtual)
1294{
1295	struct ctl_softc *softc;
1296	struct ctl_port *port;
1297	int entries_dropped, entries_filled;
1298	int retval;
1299	int i;
1300
1301	softc = control_softc;
1302
1303	retval = 0;
1304	entries_filled = 0;
1305	entries_dropped = 0;
1306
1307	i = 0;
1308	mtx_lock(&softc->ctl_lock);
1309	STAILQ_FOREACH(port, &softc->port_list, links) {
1310		struct ctl_port_entry *entry;
1311
1312		if ((port->port_type & port_type) == 0)
1313			continue;
1314
1315		if ((no_virtual != 0)
1316		 && (port->virtual_port != 0))
1317			continue;
1318
1319		if (entries_filled >= num_entries_alloced) {
1320			entries_dropped++;
1321			continue;
1322		}
1323		entry = &entries[i];
1324
1325		entry->port_type = port->port_type;
1326		strlcpy(entry->port_name, port->port_name,
1327			sizeof(entry->port_name));
1328		entry->physical_port = port->physical_port;
1329		entry->virtual_port = port->virtual_port;
1330		entry->wwnn = port->wwnn;
1331		entry->wwpn = port->wwpn;
1332
1333		i++;
1334		entries_filled++;
1335	}
1336
1337	mtx_unlock(&softc->ctl_lock);
1338
1339	if (entries_dropped > 0)
1340		retval = 1;
1341
1342	*num_entries_dropped = entries_dropped;
1343	*num_entries_filled = entries_filled;
1344
1345	return (retval);
1346}
1347
1348static void
1349ctl_ioctl_online(void *arg)
1350{
1351	struct ctl_ioctl_info *ioctl_info;
1352
1353	ioctl_info = (struct ctl_ioctl_info *)arg;
1354
1355	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1356}
1357
1358static void
1359ctl_ioctl_offline(void *arg)
1360{
1361	struct ctl_ioctl_info *ioctl_info;
1362
1363	ioctl_info = (struct ctl_ioctl_info *)arg;
1364
1365	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1366}
1367
1368/*
1369 * Remove an initiator by port number and initiator ID.
1370 * Returns 0 for success, -1 for failure.
1371 */
1372int
1373ctl_remove_initiator(struct ctl_port *port, int iid)
1374{
1375	struct ctl_softc *softc = control_softc;
1376
1377	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1378
1379	if (iid > CTL_MAX_INIT_PER_PORT) {
1380		printf("%s: initiator ID %u > maximun %u!\n",
1381		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1382		return (-1);
1383	}
1384
1385	mtx_lock(&softc->ctl_lock);
1386	port->wwpn_iid[iid].in_use--;
1387	port->wwpn_iid[iid].last_use = time_uptime;
1388	mtx_unlock(&softc->ctl_lock);
1389
1390	return (0);
1391}
1392
1393/*
1394 * Add an initiator to the initiator map.
1395 * Returns iid for success, < 0 for failure.
1396 */
1397int
1398ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1399{
1400	struct ctl_softc *softc = control_softc;
1401	time_t best_time;
1402	int i, best;
1403
1404	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1405
1406	if (iid >= CTL_MAX_INIT_PER_PORT) {
1407		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1408		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1409		free(name, M_CTL);
1410		return (-1);
1411	}
1412
1413	mtx_lock(&softc->ctl_lock);
1414
1415	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1416		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1417			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1418				iid = i;
1419				break;
1420			}
1421			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1422			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1423				iid = i;
1424				break;
1425			}
1426		}
1427	}
1428
1429	if (iid < 0) {
1430		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1431			if (port->wwpn_iid[i].in_use == 0 &&
1432			    port->wwpn_iid[i].wwpn == 0 &&
1433			    port->wwpn_iid[i].name == NULL) {
1434				iid = i;
1435				break;
1436			}
1437		}
1438	}
1439
1440	if (iid < 0) {
1441		best = -1;
1442		best_time = INT32_MAX;
1443		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1444			if (port->wwpn_iid[i].in_use == 0) {
1445				if (port->wwpn_iid[i].last_use < best_time) {
1446					best = i;
1447					best_time = port->wwpn_iid[i].last_use;
1448				}
1449			}
1450		}
1451		iid = best;
1452	}
1453
1454	if (iid < 0) {
1455		mtx_unlock(&softc->ctl_lock);
1456		free(name, M_CTL);
1457		return (-2);
1458	}
1459
1460	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1461		/*
1462		 * This is not an error yet.
1463		 */
1464		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1465#if 0
1466			printf("%s: port %d iid %u WWPN %#jx arrived"
1467			    " again\n", __func__, port->targ_port,
1468			    iid, (uintmax_t)wwpn);
1469#endif
1470			goto take;
1471		}
1472		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1473		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1474#if 0
1475			printf("%s: port %d iid %u name '%s' arrived"
1476			    " again\n", __func__, port->targ_port,
1477			    iid, name);
1478#endif
1479			goto take;
1480		}
1481
1482		/*
1483		 * This is an error, but what do we do about it?  The
1484		 * driver is telling us we have a new WWPN for this
1485		 * initiator ID, so we pretty much need to use it.
1486		 */
1487		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1488		    " but WWPN %#jx '%s' is still at that address\n",
1489		    __func__, port->targ_port, iid, wwpn, name,
1490		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1491		    port->wwpn_iid[iid].name);
1492
1493		/*
1494		 * XXX KDM clear have_ca and ua_pending on each LUN for
1495		 * this initiator.
1496		 */
1497	}
1498take:
1499	free(port->wwpn_iid[iid].name, M_CTL);
1500	port->wwpn_iid[iid].name = name;
1501	port->wwpn_iid[iid].wwpn = wwpn;
1502	port->wwpn_iid[iid].in_use++;
1503	mtx_unlock(&softc->ctl_lock);
1504
1505	return (iid);
1506}
1507
1508static int
1509ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1510{
1511	int len;
1512
1513	switch (port->port_type) {
1514	case CTL_PORT_FC:
1515	{
1516		struct scsi_transportid_fcp *id =
1517		    (struct scsi_transportid_fcp *)buf;
1518		if (port->wwpn_iid[iid].wwpn == 0)
1519			return (0);
1520		memset(id, 0, sizeof(*id));
1521		id->format_protocol = SCSI_PROTO_FC;
1522		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1523		return (sizeof(*id));
1524	}
1525	case CTL_PORT_ISCSI:
1526	{
1527		struct scsi_transportid_iscsi_port *id =
1528		    (struct scsi_transportid_iscsi_port *)buf;
1529		if (port->wwpn_iid[iid].name == NULL)
1530			return (0);
1531		memset(id, 0, 256);
1532		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1533		    SCSI_PROTO_ISCSI;
1534		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1535		len = roundup2(min(len, 252), 4);
1536		scsi_ulto2b(len, id->additional_length);
1537		return (sizeof(*id) + len);
1538	}
1539	case CTL_PORT_SAS:
1540	{
1541		struct scsi_transportid_sas *id =
1542		    (struct scsi_transportid_sas *)buf;
1543		if (port->wwpn_iid[iid].wwpn == 0)
1544			return (0);
1545		memset(id, 0, sizeof(*id));
1546		id->format_protocol = SCSI_PROTO_SAS;
1547		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1548		return (sizeof(*id));
1549	}
1550	default:
1551	{
1552		struct scsi_transportid_spi *id =
1553		    (struct scsi_transportid_spi *)buf;
1554		memset(id, 0, sizeof(*id));
1555		id->format_protocol = SCSI_PROTO_SPI;
1556		scsi_ulto2b(iid, id->scsi_addr);
1557		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1558		return (sizeof(*id));
1559	}
1560	}
1561}
1562
1563static int
1564ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1565{
1566	return (0);
1567}
1568
1569static int
1570ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1571{
1572	return (0);
1573}
1574
1575/*
1576 * Data movement routine for the CTL ioctl frontend port.
1577 */
1578static int
1579ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1580{
1581	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1582	struct ctl_sg_entry ext_entry, kern_entry;
1583	int ext_sglen, ext_sg_entries, kern_sg_entries;
1584	int ext_sg_start, ext_offset;
1585	int len_to_copy, len_copied;
1586	int kern_watermark, ext_watermark;
1587	int ext_sglist_malloced;
1588	int i, j;
1589
1590	ext_sglist_malloced = 0;
1591	ext_sg_start = 0;
1592	ext_offset = 0;
1593
1594	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1595
1596	/*
1597	 * If this flag is set, fake the data transfer.
1598	 */
1599	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1600		ctsio->ext_data_filled = ctsio->ext_data_len;
1601		goto bailout;
1602	}
1603
1604	/*
1605	 * To simplify things here, if we have a single buffer, stick it in
1606	 * a S/G entry and just make it a single entry S/G list.
1607	 */
1608	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1609		int len_seen;
1610
1611		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1612
1613		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1614							   M_WAITOK);
1615		ext_sglist_malloced = 1;
1616		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1617				   ext_sglen) != 0) {
1618			ctl_set_internal_failure(ctsio,
1619						 /*sks_valid*/ 0,
1620						 /*retry_count*/ 0);
1621			goto bailout;
1622		}
1623		ext_sg_entries = ctsio->ext_sg_entries;
1624		len_seen = 0;
1625		for (i = 0; i < ext_sg_entries; i++) {
1626			if ((len_seen + ext_sglist[i].len) >=
1627			     ctsio->ext_data_filled) {
1628				ext_sg_start = i;
1629				ext_offset = ctsio->ext_data_filled - len_seen;
1630				break;
1631			}
1632			len_seen += ext_sglist[i].len;
1633		}
1634	} else {
1635		ext_sglist = &ext_entry;
1636		ext_sglist->addr = ctsio->ext_data_ptr;
1637		ext_sglist->len = ctsio->ext_data_len;
1638		ext_sg_entries = 1;
1639		ext_sg_start = 0;
1640		ext_offset = ctsio->ext_data_filled;
1641	}
1642
1643	if (ctsio->kern_sg_entries > 0) {
1644		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1645		kern_sg_entries = ctsio->kern_sg_entries;
1646	} else {
1647		kern_sglist = &kern_entry;
1648		kern_sglist->addr = ctsio->kern_data_ptr;
1649		kern_sglist->len = ctsio->kern_data_len;
1650		kern_sg_entries = 1;
1651	}
1652
1653
1654	kern_watermark = 0;
1655	ext_watermark = ext_offset;
1656	len_copied = 0;
1657	for (i = ext_sg_start, j = 0;
1658	     i < ext_sg_entries && j < kern_sg_entries;) {
1659		uint8_t *ext_ptr, *kern_ptr;
1660
1661		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1662				      kern_sglist[j].len - kern_watermark);
1663
1664		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1665		ext_ptr = ext_ptr + ext_watermark;
1666		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1667			/*
1668			 * XXX KDM fix this!
1669			 */
1670			panic("need to implement bus address support");
1671#if 0
1672			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1673#endif
1674		} else
1675			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1676		kern_ptr = kern_ptr + kern_watermark;
1677
1678		kern_watermark += len_to_copy;
1679		ext_watermark += len_to_copy;
1680
1681		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1682		     CTL_FLAG_DATA_IN) {
1683			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1684					 "bytes to user\n", len_to_copy));
1685			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1686					 "to %p\n", kern_ptr, ext_ptr));
1687			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1688				ctl_set_internal_failure(ctsio,
1689							 /*sks_valid*/ 0,
1690							 /*retry_count*/ 0);
1691				goto bailout;
1692			}
1693		} else {
1694			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1695					 "bytes from user\n", len_to_copy));
1696			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1697					 "to %p\n", ext_ptr, kern_ptr));
1698			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1699				ctl_set_internal_failure(ctsio,
1700							 /*sks_valid*/ 0,
1701							 /*retry_count*/0);
1702				goto bailout;
1703			}
1704		}
1705
1706		len_copied += len_to_copy;
1707
1708		if (ext_sglist[i].len == ext_watermark) {
1709			i++;
1710			ext_watermark = 0;
1711		}
1712
1713		if (kern_sglist[j].len == kern_watermark) {
1714			j++;
1715			kern_watermark = 0;
1716		}
1717	}
1718
1719	ctsio->ext_data_filled += len_copied;
1720
1721	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1722			 "kern_sg_entries: %d\n", ext_sg_entries,
1723			 kern_sg_entries));
1724	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1725			 "kern_data_len = %d\n", ctsio->ext_data_len,
1726			 ctsio->kern_data_len));
1727
1728
1729	/* XXX KDM set residual?? */
1730bailout:
1731
1732	if (ext_sglist_malloced != 0)
1733		free(ext_sglist, M_CTL);
1734
1735	return (CTL_RETVAL_COMPLETE);
1736}
1737
1738/*
1739 * Serialize a command that went down the "wrong" side, and so was sent to
1740 * this controller for execution.  The logic is a little different than the
1741 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1742 * sent back to the other side, but in the success case, we execute the
1743 * command on this side (XFER mode) or tell the other side to execute it
1744 * (SER_ONLY mode).
1745 */
1746static int
1747ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1748{
1749	struct ctl_softc *ctl_softc;
1750	union ctl_ha_msg msg_info;
1751	struct ctl_lun *lun;
1752	int retval = 0;
1753	uint32_t targ_lun;
1754
1755	ctl_softc = control_softc;
1756
1757	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1758	lun = ctl_softc->ctl_luns[targ_lun];
1759	if (lun==NULL)
1760	{
1761		/*
1762		 * Why isn't LUN defined? The other side wouldn't
1763		 * send a cmd if the LUN is undefined.
1764		 */
1765		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1766
1767		/* "Logical unit not supported" */
1768		ctl_set_sense_data(&msg_info.scsi.sense_data,
1769				   lun,
1770				   /*sense_format*/SSD_TYPE_NONE,
1771				   /*current_error*/ 1,
1772				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1773				   /*asc*/ 0x25,
1774				   /*ascq*/ 0x00,
1775				   SSD_ELEM_NONE);
1776
1777		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1778		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1779		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1780		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1781		msg_info.hdr.serializing_sc = NULL;
1782		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1783	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1784				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1785		}
1786		return(1);
1787
1788	}
1789
1790	mtx_lock(&lun->lun_lock);
1791    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1792
1793	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1794		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1795		 ooa_links))) {
1796	case CTL_ACTION_BLOCK:
1797		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1798		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1799				  blocked_links);
1800		break;
1801	case CTL_ACTION_PASS:
1802	case CTL_ACTION_SKIP:
1803		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1804			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1805			ctl_enqueue_rtr((union ctl_io *)ctsio);
1806		} else {
1807
1808			/* send msg back to other side */
1809			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1810			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1811			msg_info.hdr.msg_type = CTL_MSG_R2R;
1812#if 0
1813			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1814#endif
1815		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1816			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1817			}
1818		}
1819		break;
1820	case CTL_ACTION_OVERLAP:
1821		/* OVERLAPPED COMMANDS ATTEMPTED */
1822		ctl_set_sense_data(&msg_info.scsi.sense_data,
1823				   lun,
1824				   /*sense_format*/SSD_TYPE_NONE,
1825				   /*current_error*/ 1,
1826				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1827				   /*asc*/ 0x4E,
1828				   /*ascq*/ 0x00,
1829				   SSD_ELEM_NONE);
1830
1831		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1832		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1833		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1834		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1835		msg_info.hdr.serializing_sc = NULL;
1836		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1837#if 0
1838		printf("BAD JUJU:Major Bummer Overlap\n");
1839#endif
1840		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1841		retval = 1;
1842		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1843		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1844		}
1845		break;
1846	case CTL_ACTION_OVERLAP_TAG:
1847		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1848		ctl_set_sense_data(&msg_info.scsi.sense_data,
1849				   lun,
1850				   /*sense_format*/SSD_TYPE_NONE,
1851				   /*current_error*/ 1,
1852				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1853				   /*asc*/ 0x4D,
1854				   /*ascq*/ ctsio->tag_num & 0xff,
1855				   SSD_ELEM_NONE);
1856
1857		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1858		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1859		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1860		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1861		msg_info.hdr.serializing_sc = NULL;
1862		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1863#if 0
1864		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1865#endif
1866		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1867		retval = 1;
1868		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1869		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1870		}
1871		break;
1872	case CTL_ACTION_ERROR:
1873	default:
1874		/* "Internal target failure" */
1875		ctl_set_sense_data(&msg_info.scsi.sense_data,
1876				   lun,
1877				   /*sense_format*/SSD_TYPE_NONE,
1878				   /*current_error*/ 1,
1879				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1880				   /*asc*/ 0x44,
1881				   /*ascq*/ 0x00,
1882				   SSD_ELEM_NONE);
1883
1884		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1885		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1886		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1887		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1888		msg_info.hdr.serializing_sc = NULL;
1889		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1890#if 0
1891		printf("BAD JUJU:Major Bummer HW Error\n");
1892#endif
1893		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1894		retval = 1;
1895		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1896		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1897		}
1898		break;
1899	}
1900	mtx_unlock(&lun->lun_lock);
1901	return (retval);
1902}
1903
1904static int
1905ctl_ioctl_submit_wait(union ctl_io *io)
1906{
1907	struct ctl_fe_ioctl_params params;
1908	ctl_fe_ioctl_state last_state;
1909	int done, retval;
1910
1911	retval = 0;
1912
1913	bzero(&params, sizeof(params));
1914
1915	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1916	cv_init(&params.sem, "ctlioccv");
1917	params.state = CTL_IOCTL_INPROG;
1918	last_state = params.state;
1919
1920	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1921
1922	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1923
1924	/* This shouldn't happen */
1925	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1926		return (retval);
1927
1928	done = 0;
1929
1930	do {
1931		mtx_lock(&params.ioctl_mtx);
1932		/*
1933		 * Check the state here, and don't sleep if the state has
1934		 * already changed (i.e. wakeup has already occured, but we
1935		 * weren't waiting yet).
1936		 */
1937		if (params.state == last_state) {
1938			/* XXX KDM cv_wait_sig instead? */
1939			cv_wait(&params.sem, &params.ioctl_mtx);
1940		}
1941		last_state = params.state;
1942
1943		switch (params.state) {
1944		case CTL_IOCTL_INPROG:
1945			/* Why did we wake up? */
1946			/* XXX KDM error here? */
1947			mtx_unlock(&params.ioctl_mtx);
1948			break;
1949		case CTL_IOCTL_DATAMOVE:
1950			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1951
1952			/*
1953			 * change last_state back to INPROG to avoid
1954			 * deadlock on subsequent data moves.
1955			 */
1956			params.state = last_state = CTL_IOCTL_INPROG;
1957
1958			mtx_unlock(&params.ioctl_mtx);
1959			ctl_ioctl_do_datamove(&io->scsiio);
1960			/*
1961			 * Note that in some cases, most notably writes,
1962			 * this will queue the I/O and call us back later.
1963			 * In other cases, generally reads, this routine
1964			 * will immediately call back and wake us up,
1965			 * probably using our own context.
1966			 */
1967			io->scsiio.be_move_done(io);
1968			break;
1969		case CTL_IOCTL_DONE:
1970			mtx_unlock(&params.ioctl_mtx);
1971			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1972			done = 1;
1973			break;
1974		default:
1975			mtx_unlock(&params.ioctl_mtx);
1976			/* XXX KDM error here? */
1977			break;
1978		}
1979	} while (done == 0);
1980
1981	mtx_destroy(&params.ioctl_mtx);
1982	cv_destroy(&params.sem);
1983
1984	return (CTL_RETVAL_COMPLETE);
1985}
1986
1987static void
1988ctl_ioctl_datamove(union ctl_io *io)
1989{
1990	struct ctl_fe_ioctl_params *params;
1991
1992	params = (struct ctl_fe_ioctl_params *)
1993		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1994
1995	mtx_lock(&params->ioctl_mtx);
1996	params->state = CTL_IOCTL_DATAMOVE;
1997	cv_broadcast(&params->sem);
1998	mtx_unlock(&params->ioctl_mtx);
1999}
2000
2001static void
2002ctl_ioctl_done(union ctl_io *io)
2003{
2004	struct ctl_fe_ioctl_params *params;
2005
2006	params = (struct ctl_fe_ioctl_params *)
2007		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2008
2009	mtx_lock(&params->ioctl_mtx);
2010	params->state = CTL_IOCTL_DONE;
2011	cv_broadcast(&params->sem);
2012	mtx_unlock(&params->ioctl_mtx);
2013}
2014
2015static void
2016ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2017{
2018	struct ctl_fe_ioctl_startstop_info *sd_info;
2019
2020	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2021
2022	sd_info->hs_info.status = metatask->status;
2023	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2024	sd_info->hs_info.luns_complete =
2025		metatask->taskinfo.startstop.luns_complete;
2026	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2027
2028	cv_broadcast(&sd_info->sem);
2029}
2030
2031static void
2032ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2033{
2034	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2035
2036	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2037
2038	mtx_lock(fe_bbr_info->lock);
2039	fe_bbr_info->bbr_info->status = metatask->status;
2040	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2041	fe_bbr_info->wakeup_done = 1;
2042	mtx_unlock(fe_bbr_info->lock);
2043
2044	cv_broadcast(&fe_bbr_info->sem);
2045}
2046
2047/*
2048 * Returns 0 for success, errno for failure.
2049 */
2050static int
2051ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2052		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2053{
2054	union ctl_io *io;
2055	int retval;
2056
2057	retval = 0;
2058
2059	mtx_lock(&lun->lun_lock);
2060	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2061	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2062	     ooa_links)) {
2063		struct ctl_ooa_entry *entry;
2064
2065		/*
2066		 * If we've got more than we can fit, just count the
2067		 * remaining entries.
2068		 */
2069		if (*cur_fill_num >= ooa_hdr->alloc_num)
2070			continue;
2071
2072		entry = &kern_entries[*cur_fill_num];
2073
2074		entry->tag_num = io->scsiio.tag_num;
2075		entry->lun_num = lun->lun;
2076#ifdef CTL_TIME_IO
2077		entry->start_bt = io->io_hdr.start_bt;
2078#endif
2079		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2080		entry->cdb_len = io->scsiio.cdb_len;
2081		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2082			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2083
2084		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2085			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2086
2087		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2088			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2089
2090		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2091			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2092
2093		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2094			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2095	}
2096	mtx_unlock(&lun->lun_lock);
2097
2098	return (retval);
2099}
2100
2101static void *
2102ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2103		 size_t error_str_len)
2104{
2105	void *kptr;
2106
2107	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2108
2109	if (copyin(user_addr, kptr, len) != 0) {
2110		snprintf(error_str, error_str_len, "Error copying %d bytes "
2111			 "from user address %p to kernel address %p", len,
2112			 user_addr, kptr);
2113		free(kptr, M_CTL);
2114		return (NULL);
2115	}
2116
2117	return (kptr);
2118}
2119
2120static void
2121ctl_free_args(int num_args, struct ctl_be_arg *args)
2122{
2123	int i;
2124
2125	if (args == NULL)
2126		return;
2127
2128	for (i = 0; i < num_args; i++) {
2129		free(args[i].kname, M_CTL);
2130		free(args[i].kvalue, M_CTL);
2131	}
2132
2133	free(args, M_CTL);
2134}
2135
2136static struct ctl_be_arg *
2137ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2138		char *error_str, size_t error_str_len)
2139{
2140	struct ctl_be_arg *args;
2141	int i;
2142
2143	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2144				error_str, error_str_len);
2145
2146	if (args == NULL)
2147		goto bailout;
2148
2149	for (i = 0; i < num_args; i++) {
2150		args[i].kname = NULL;
2151		args[i].kvalue = NULL;
2152	}
2153
2154	for (i = 0; i < num_args; i++) {
2155		uint8_t *tmpptr;
2156
2157		args[i].kname = ctl_copyin_alloc(args[i].name,
2158			args[i].namelen, error_str, error_str_len);
2159		if (args[i].kname == NULL)
2160			goto bailout;
2161
2162		if (args[i].kname[args[i].namelen - 1] != '\0') {
2163			snprintf(error_str, error_str_len, "Argument %d "
2164				 "name is not NUL-terminated", i);
2165			goto bailout;
2166		}
2167
2168		if (args[i].flags & CTL_BEARG_RD) {
2169			tmpptr = ctl_copyin_alloc(args[i].value,
2170				args[i].vallen, error_str, error_str_len);
2171			if (tmpptr == NULL)
2172				goto bailout;
2173			if ((args[i].flags & CTL_BEARG_ASCII)
2174			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2175				snprintf(error_str, error_str_len, "Argument "
2176				    "%d value is not NUL-terminated", i);
2177				goto bailout;
2178			}
2179			args[i].kvalue = tmpptr;
2180		} else {
2181			args[i].kvalue = malloc(args[i].vallen,
2182			    M_CTL, M_WAITOK | M_ZERO);
2183		}
2184	}
2185
2186	return (args);
2187bailout:
2188
2189	ctl_free_args(num_args, args);
2190
2191	return (NULL);
2192}
2193
2194static void
2195ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2196{
2197	int i;
2198
2199	for (i = 0; i < num_args; i++) {
2200		if (args[i].flags & CTL_BEARG_WR)
2201			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2202	}
2203}
2204
2205/*
2206 * Escape characters that are illegal or not recommended in XML.
2207 */
2208int
2209ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2210{
2211	int retval;
2212
2213	retval = 0;
2214
2215	for (; *str; str++) {
2216		switch (*str) {
2217		case '&':
2218			retval = sbuf_printf(sb, "&amp;");
2219			break;
2220		case '>':
2221			retval = sbuf_printf(sb, "&gt;");
2222			break;
2223		case '<':
2224			retval = sbuf_printf(sb, "&lt;");
2225			break;
2226		default:
2227			retval = sbuf_putc(sb, *str);
2228			break;
2229		}
2230
2231		if (retval != 0)
2232			break;
2233
2234	}
2235
2236	return (retval);
2237}
2238
2239static int
2240ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2241	  struct thread *td)
2242{
2243	struct ctl_softc *softc;
2244	int retval;
2245
2246	softc = control_softc;
2247
2248	retval = 0;
2249
2250	switch (cmd) {
2251	case CTL_IO: {
2252		union ctl_io *io;
2253		void *pool_tmp;
2254
2255		/*
2256		 * If we haven't been "enabled", don't allow any SCSI I/O
2257		 * to this FETD.
2258		 */
2259		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2260			retval = EPERM;
2261			break;
2262		}
2263
2264		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2265		if (io == NULL) {
2266			printf("ctl_ioctl: can't allocate ctl_io!\n");
2267			retval = ENOSPC;
2268			break;
2269		}
2270
2271		/*
2272		 * Need to save the pool reference so it doesn't get
2273		 * spammed by the user's ctl_io.
2274		 */
2275		pool_tmp = io->io_hdr.pool;
2276
2277		memcpy(io, (void *)addr, sizeof(*io));
2278
2279		io->io_hdr.pool = pool_tmp;
2280		/*
2281		 * No status yet, so make sure the status is set properly.
2282		 */
2283		io->io_hdr.status = CTL_STATUS_NONE;
2284
2285		/*
2286		 * The user sets the initiator ID, target and LUN IDs.
2287		 */
2288		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2289		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2290		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2291		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2292			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2293
2294		retval = ctl_ioctl_submit_wait(io);
2295
2296		if (retval != 0) {
2297			ctl_free_io(io);
2298			break;
2299		}
2300
2301		memcpy((void *)addr, io, sizeof(*io));
2302
2303		/* return this to our pool */
2304		ctl_free_io(io);
2305
2306		break;
2307	}
2308	case CTL_ENABLE_PORT:
2309	case CTL_DISABLE_PORT:
2310	case CTL_SET_PORT_WWNS: {
2311		struct ctl_port *port;
2312		struct ctl_port_entry *entry;
2313
2314		entry = (struct ctl_port_entry *)addr;
2315
2316		mtx_lock(&softc->ctl_lock);
2317		STAILQ_FOREACH(port, &softc->port_list, links) {
2318			int action, done;
2319
2320			action = 0;
2321			done = 0;
2322
2323			if ((entry->port_type == CTL_PORT_NONE)
2324			 && (entry->targ_port == port->targ_port)) {
2325				/*
2326				 * If the user only wants to enable or
2327				 * disable or set WWNs on a specific port,
2328				 * do the operation and we're done.
2329				 */
2330				action = 1;
2331				done = 1;
2332			} else if (entry->port_type & port->port_type) {
2333				/*
2334				 * Compare the user's type mask with the
2335				 * particular frontend type to see if we
2336				 * have a match.
2337				 */
2338				action = 1;
2339				done = 0;
2340
2341				/*
2342				 * Make sure the user isn't trying to set
2343				 * WWNs on multiple ports at the same time.
2344				 */
2345				if (cmd == CTL_SET_PORT_WWNS) {
2346					printf("%s: Can't set WWNs on "
2347					       "multiple ports\n", __func__);
2348					retval = EINVAL;
2349					break;
2350				}
2351			}
2352			if (action != 0) {
2353				/*
2354				 * XXX KDM we have to drop the lock here,
2355				 * because the online/offline operations
2356				 * can potentially block.  We need to
2357				 * reference count the frontends so they
2358				 * can't go away,
2359				 */
2360				mtx_unlock(&softc->ctl_lock);
2361
2362				if (cmd == CTL_ENABLE_PORT) {
2363					struct ctl_lun *lun;
2364
2365					STAILQ_FOREACH(lun, &softc->lun_list,
2366						       links) {
2367						port->lun_enable(port->targ_lun_arg,
2368						    lun->target,
2369						    lun->lun);
2370					}
2371
2372					ctl_port_online(port);
2373				} else if (cmd == CTL_DISABLE_PORT) {
2374					struct ctl_lun *lun;
2375
2376					ctl_port_offline(port);
2377
2378					STAILQ_FOREACH(lun, &softc->lun_list,
2379						       links) {
2380						port->lun_disable(
2381						    port->targ_lun_arg,
2382						    lun->target,
2383						    lun->lun);
2384					}
2385				}
2386
2387				mtx_lock(&softc->ctl_lock);
2388
2389				if (cmd == CTL_SET_PORT_WWNS)
2390					ctl_port_set_wwns(port,
2391					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2392					    1 : 0, entry->wwnn,
2393					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2394					    1 : 0, entry->wwpn);
2395			}
2396			if (done != 0)
2397				break;
2398		}
2399		mtx_unlock(&softc->ctl_lock);
2400		break;
2401	}
2402	case CTL_GET_PORT_LIST: {
2403		struct ctl_port *port;
2404		struct ctl_port_list *list;
2405		int i;
2406
2407		list = (struct ctl_port_list *)addr;
2408
2409		if (list->alloc_len != (list->alloc_num *
2410		    sizeof(struct ctl_port_entry))) {
2411			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2412			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2413			       "%zu\n", __func__, list->alloc_len,
2414			       list->alloc_num, sizeof(struct ctl_port_entry));
2415			retval = EINVAL;
2416			break;
2417		}
2418		list->fill_len = 0;
2419		list->fill_num = 0;
2420		list->dropped_num = 0;
2421		i = 0;
2422		mtx_lock(&softc->ctl_lock);
2423		STAILQ_FOREACH(port, &softc->port_list, links) {
2424			struct ctl_port_entry entry, *list_entry;
2425
2426			if (list->fill_num >= list->alloc_num) {
2427				list->dropped_num++;
2428				continue;
2429			}
2430
2431			entry.port_type = port->port_type;
2432			strlcpy(entry.port_name, port->port_name,
2433				sizeof(entry.port_name));
2434			entry.targ_port = port->targ_port;
2435			entry.physical_port = port->physical_port;
2436			entry.virtual_port = port->virtual_port;
2437			entry.wwnn = port->wwnn;
2438			entry.wwpn = port->wwpn;
2439			if (port->status & CTL_PORT_STATUS_ONLINE)
2440				entry.online = 1;
2441			else
2442				entry.online = 0;
2443
2444			list_entry = &list->entries[i];
2445
2446			retval = copyout(&entry, list_entry, sizeof(entry));
2447			if (retval != 0) {
2448				printf("%s: CTL_GET_PORT_LIST: copyout "
2449				       "returned %d\n", __func__, retval);
2450				break;
2451			}
2452			i++;
2453			list->fill_num++;
2454			list->fill_len += sizeof(entry);
2455		}
2456		mtx_unlock(&softc->ctl_lock);
2457
2458		/*
2459		 * If this is non-zero, we had a copyout fault, so there's
2460		 * probably no point in attempting to set the status inside
2461		 * the structure.
2462		 */
2463		if (retval != 0)
2464			break;
2465
2466		if (list->dropped_num > 0)
2467			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2468		else
2469			list->status = CTL_PORT_LIST_OK;
2470		break;
2471	}
2472	case CTL_DUMP_OOA: {
2473		struct ctl_lun *lun;
2474		union ctl_io *io;
2475		char printbuf[128];
2476		struct sbuf sb;
2477
2478		mtx_lock(&softc->ctl_lock);
2479		printf("Dumping OOA queues:\n");
2480		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2481			mtx_lock(&lun->lun_lock);
2482			for (io = (union ctl_io *)TAILQ_FIRST(
2483			     &lun->ooa_queue); io != NULL;
2484			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2485			     ooa_links)) {
2486				sbuf_new(&sb, printbuf, sizeof(printbuf),
2487					 SBUF_FIXEDLEN);
2488				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2489					    (intmax_t)lun->lun,
2490					    io->scsiio.tag_num,
2491					    (io->io_hdr.flags &
2492					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2493					    (io->io_hdr.flags &
2494					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2495					    (io->io_hdr.flags &
2496					    CTL_FLAG_ABORT) ? " ABORT" : "",
2497			                    (io->io_hdr.flags &
2498		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2499				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2500				sbuf_finish(&sb);
2501				printf("%s\n", sbuf_data(&sb));
2502			}
2503			mtx_unlock(&lun->lun_lock);
2504		}
2505		printf("OOA queues dump done\n");
2506		mtx_unlock(&softc->ctl_lock);
2507		break;
2508	}
2509	case CTL_GET_OOA: {
2510		struct ctl_lun *lun;
2511		struct ctl_ooa *ooa_hdr;
2512		struct ctl_ooa_entry *entries;
2513		uint32_t cur_fill_num;
2514
2515		ooa_hdr = (struct ctl_ooa *)addr;
2516
2517		if ((ooa_hdr->alloc_len == 0)
2518		 || (ooa_hdr->alloc_num == 0)) {
2519			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2520			       "must be non-zero\n", __func__,
2521			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2522			retval = EINVAL;
2523			break;
2524		}
2525
2526		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2527		    sizeof(struct ctl_ooa_entry))) {
2528			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2529			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2530			       __func__, ooa_hdr->alloc_len,
2531			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2532			retval = EINVAL;
2533			break;
2534		}
2535
2536		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2537		if (entries == NULL) {
2538			printf("%s: could not allocate %d bytes for OOA "
2539			       "dump\n", __func__, ooa_hdr->alloc_len);
2540			retval = ENOMEM;
2541			break;
2542		}
2543
2544		mtx_lock(&softc->ctl_lock);
2545		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2546		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2547		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2548			mtx_unlock(&softc->ctl_lock);
2549			free(entries, M_CTL);
2550			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2551			       __func__, (uintmax_t)ooa_hdr->lun_num);
2552			retval = EINVAL;
2553			break;
2554		}
2555
2556		cur_fill_num = 0;
2557
2558		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2559			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2560				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2561					ooa_hdr, entries);
2562				if (retval != 0)
2563					break;
2564			}
2565			if (retval != 0) {
2566				mtx_unlock(&softc->ctl_lock);
2567				free(entries, M_CTL);
2568				break;
2569			}
2570		} else {
2571			lun = softc->ctl_luns[ooa_hdr->lun_num];
2572
2573			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2574						    entries);
2575		}
2576		mtx_unlock(&softc->ctl_lock);
2577
2578		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2579		ooa_hdr->fill_len = ooa_hdr->fill_num *
2580			sizeof(struct ctl_ooa_entry);
2581		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2582		if (retval != 0) {
2583			printf("%s: error copying out %d bytes for OOA dump\n",
2584			       __func__, ooa_hdr->fill_len);
2585		}
2586
2587		getbintime(&ooa_hdr->cur_bt);
2588
2589		if (cur_fill_num > ooa_hdr->alloc_num) {
2590			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2591			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2592		} else {
2593			ooa_hdr->dropped_num = 0;
2594			ooa_hdr->status = CTL_OOA_OK;
2595		}
2596
2597		free(entries, M_CTL);
2598		break;
2599	}
2600	case CTL_CHECK_OOA: {
2601		union ctl_io *io;
2602		struct ctl_lun *lun;
2603		struct ctl_ooa_info *ooa_info;
2604
2605
2606		ooa_info = (struct ctl_ooa_info *)addr;
2607
2608		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2609			ooa_info->status = CTL_OOA_INVALID_LUN;
2610			break;
2611		}
2612		mtx_lock(&softc->ctl_lock);
2613		lun = softc->ctl_luns[ooa_info->lun_id];
2614		if (lun == NULL) {
2615			mtx_unlock(&softc->ctl_lock);
2616			ooa_info->status = CTL_OOA_INVALID_LUN;
2617			break;
2618		}
2619		mtx_lock(&lun->lun_lock);
2620		mtx_unlock(&softc->ctl_lock);
2621		ooa_info->num_entries = 0;
2622		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2623		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2624		     &io->io_hdr, ooa_links)) {
2625			ooa_info->num_entries++;
2626		}
2627		mtx_unlock(&lun->lun_lock);
2628
2629		ooa_info->status = CTL_OOA_SUCCESS;
2630
2631		break;
2632	}
2633	case CTL_HARD_START:
2634	case CTL_HARD_STOP: {
2635		struct ctl_fe_ioctl_startstop_info ss_info;
2636		struct cfi_metatask *metatask;
2637		struct mtx hs_mtx;
2638
2639		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2640
2641		cv_init(&ss_info.sem, "hard start/stop cv" );
2642
2643		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2644		if (metatask == NULL) {
2645			retval = ENOMEM;
2646			mtx_destroy(&hs_mtx);
2647			break;
2648		}
2649
2650		if (cmd == CTL_HARD_START)
2651			metatask->tasktype = CFI_TASK_STARTUP;
2652		else
2653			metatask->tasktype = CFI_TASK_SHUTDOWN;
2654
2655		metatask->callback = ctl_ioctl_hard_startstop_callback;
2656		metatask->callback_arg = &ss_info;
2657
2658		cfi_action(metatask);
2659
2660		/* Wait for the callback */
2661		mtx_lock(&hs_mtx);
2662		cv_wait_sig(&ss_info.sem, &hs_mtx);
2663		mtx_unlock(&hs_mtx);
2664
2665		/*
2666		 * All information has been copied from the metatask by the
2667		 * time cv_broadcast() is called, so we free the metatask here.
2668		 */
2669		cfi_free_metatask(metatask);
2670
2671		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2672
2673		mtx_destroy(&hs_mtx);
2674		break;
2675	}
2676	case CTL_BBRREAD: {
2677		struct ctl_bbrread_info *bbr_info;
2678		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2679		struct mtx bbr_mtx;
2680		struct cfi_metatask *metatask;
2681
2682		bbr_info = (struct ctl_bbrread_info *)addr;
2683
2684		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2685
2686		bzero(&bbr_mtx, sizeof(bbr_mtx));
2687		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2688
2689		fe_bbr_info.bbr_info = bbr_info;
2690		fe_bbr_info.lock = &bbr_mtx;
2691
2692		cv_init(&fe_bbr_info.sem, "BBR read cv");
2693		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2694
2695		if (metatask == NULL) {
2696			mtx_destroy(&bbr_mtx);
2697			cv_destroy(&fe_bbr_info.sem);
2698			retval = ENOMEM;
2699			break;
2700		}
2701		metatask->tasktype = CFI_TASK_BBRREAD;
2702		metatask->callback = ctl_ioctl_bbrread_callback;
2703		metatask->callback_arg = &fe_bbr_info;
2704		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2705		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2706		metatask->taskinfo.bbrread.len = bbr_info->len;
2707
2708		cfi_action(metatask);
2709
2710		mtx_lock(&bbr_mtx);
2711		while (fe_bbr_info.wakeup_done == 0)
2712			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2713		mtx_unlock(&bbr_mtx);
2714
2715		bbr_info->status = metatask->status;
2716		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2717		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2718		memcpy(&bbr_info->sense_data,
2719		       &metatask->taskinfo.bbrread.sense_data,
2720		       ctl_min(sizeof(bbr_info->sense_data),
2721			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2722
2723		cfi_free_metatask(metatask);
2724
2725		mtx_destroy(&bbr_mtx);
2726		cv_destroy(&fe_bbr_info.sem);
2727
2728		break;
2729	}
2730	case CTL_DELAY_IO: {
2731		struct ctl_io_delay_info *delay_info;
2732#ifdef CTL_IO_DELAY
2733		struct ctl_lun *lun;
2734#endif /* CTL_IO_DELAY */
2735
2736		delay_info = (struct ctl_io_delay_info *)addr;
2737
2738#ifdef CTL_IO_DELAY
2739		mtx_lock(&softc->ctl_lock);
2740
2741		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2742		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2743			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2744		} else {
2745			lun = softc->ctl_luns[delay_info->lun_id];
2746			mtx_lock(&lun->lun_lock);
2747
2748			delay_info->status = CTL_DELAY_STATUS_OK;
2749
2750			switch (delay_info->delay_type) {
2751			case CTL_DELAY_TYPE_CONT:
2752				break;
2753			case CTL_DELAY_TYPE_ONESHOT:
2754				break;
2755			default:
2756				delay_info->status =
2757					CTL_DELAY_STATUS_INVALID_TYPE;
2758				break;
2759			}
2760
2761			switch (delay_info->delay_loc) {
2762			case CTL_DELAY_LOC_DATAMOVE:
2763				lun->delay_info.datamove_type =
2764					delay_info->delay_type;
2765				lun->delay_info.datamove_delay =
2766					delay_info->delay_secs;
2767				break;
2768			case CTL_DELAY_LOC_DONE:
2769				lun->delay_info.done_type =
2770					delay_info->delay_type;
2771				lun->delay_info.done_delay =
2772					delay_info->delay_secs;
2773				break;
2774			default:
2775				delay_info->status =
2776					CTL_DELAY_STATUS_INVALID_LOC;
2777				break;
2778			}
2779			mtx_unlock(&lun->lun_lock);
2780		}
2781
2782		mtx_unlock(&softc->ctl_lock);
2783#else
2784		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2785#endif /* CTL_IO_DELAY */
2786		break;
2787	}
2788	case CTL_REALSYNC_SET: {
2789		int *syncstate;
2790
2791		syncstate = (int *)addr;
2792
2793		mtx_lock(&softc->ctl_lock);
2794		switch (*syncstate) {
2795		case 0:
2796			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2797			break;
2798		case 1:
2799			softc->flags |= CTL_FLAG_REAL_SYNC;
2800			break;
2801		default:
2802			retval = EINVAL;
2803			break;
2804		}
2805		mtx_unlock(&softc->ctl_lock);
2806		break;
2807	}
2808	case CTL_REALSYNC_GET: {
2809		int *syncstate;
2810
2811		syncstate = (int*)addr;
2812
2813		mtx_lock(&softc->ctl_lock);
2814		if (softc->flags & CTL_FLAG_REAL_SYNC)
2815			*syncstate = 1;
2816		else
2817			*syncstate = 0;
2818		mtx_unlock(&softc->ctl_lock);
2819
2820		break;
2821	}
2822	case CTL_SETSYNC:
2823	case CTL_GETSYNC: {
2824		struct ctl_sync_info *sync_info;
2825		struct ctl_lun *lun;
2826
2827		sync_info = (struct ctl_sync_info *)addr;
2828
2829		mtx_lock(&softc->ctl_lock);
2830		lun = softc->ctl_luns[sync_info->lun_id];
2831		if (lun == NULL) {
2832			mtx_unlock(&softc->ctl_lock);
2833			sync_info->status = CTL_GS_SYNC_NO_LUN;
2834		}
2835		/*
2836		 * Get or set the sync interval.  We're not bounds checking
2837		 * in the set case, hopefully the user won't do something
2838		 * silly.
2839		 */
2840		mtx_lock(&lun->lun_lock);
2841		mtx_unlock(&softc->ctl_lock);
2842		if (cmd == CTL_GETSYNC)
2843			sync_info->sync_interval = lun->sync_interval;
2844		else
2845			lun->sync_interval = sync_info->sync_interval;
2846		mtx_unlock(&lun->lun_lock);
2847
2848		sync_info->status = CTL_GS_SYNC_OK;
2849
2850		break;
2851	}
2852	case CTL_GETSTATS: {
2853		struct ctl_stats *stats;
2854		struct ctl_lun *lun;
2855		int i;
2856
2857		stats = (struct ctl_stats *)addr;
2858
2859		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2860		     stats->alloc_len) {
2861			stats->status = CTL_SS_NEED_MORE_SPACE;
2862			stats->num_luns = softc->num_luns;
2863			break;
2864		}
2865		/*
2866		 * XXX KDM no locking here.  If the LUN list changes,
2867		 * things can blow up.
2868		 */
2869		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2870		     i++, lun = STAILQ_NEXT(lun, links)) {
2871			retval = copyout(&lun->stats, &stats->lun_stats[i],
2872					 sizeof(lun->stats));
2873			if (retval != 0)
2874				break;
2875		}
2876		stats->num_luns = softc->num_luns;
2877		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2878				 softc->num_luns;
2879		stats->status = CTL_SS_OK;
2880#ifdef CTL_TIME_IO
2881		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2882#else
2883		stats->flags = CTL_STATS_FLAG_NONE;
2884#endif
2885		getnanouptime(&stats->timestamp);
2886		break;
2887	}
2888	case CTL_ERROR_INJECT: {
2889		struct ctl_error_desc *err_desc, *new_err_desc;
2890		struct ctl_lun *lun;
2891
2892		err_desc = (struct ctl_error_desc *)addr;
2893
2894		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2895				      M_WAITOK | M_ZERO);
2896		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2897
2898		mtx_lock(&softc->ctl_lock);
2899		lun = softc->ctl_luns[err_desc->lun_id];
2900		if (lun == NULL) {
2901			mtx_unlock(&softc->ctl_lock);
2902			free(new_err_desc, M_CTL);
2903			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2904			       __func__, (uintmax_t)err_desc->lun_id);
2905			retval = EINVAL;
2906			break;
2907		}
2908		mtx_lock(&lun->lun_lock);
2909		mtx_unlock(&softc->ctl_lock);
2910
2911		/*
2912		 * We could do some checking here to verify the validity
2913		 * of the request, but given the complexity of error
2914		 * injection requests, the checking logic would be fairly
2915		 * complex.
2916		 *
2917		 * For now, if the request is invalid, it just won't get
2918		 * executed and might get deleted.
2919		 */
2920		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2921
2922		/*
2923		 * XXX KDM check to make sure the serial number is unique,
2924		 * in case we somehow manage to wrap.  That shouldn't
2925		 * happen for a very long time, but it's the right thing to
2926		 * do.
2927		 */
2928		new_err_desc->serial = lun->error_serial;
2929		err_desc->serial = lun->error_serial;
2930		lun->error_serial++;
2931
2932		mtx_unlock(&lun->lun_lock);
2933		break;
2934	}
2935	case CTL_ERROR_INJECT_DELETE: {
2936		struct ctl_error_desc *delete_desc, *desc, *desc2;
2937		struct ctl_lun *lun;
2938		int delete_done;
2939
2940		delete_desc = (struct ctl_error_desc *)addr;
2941		delete_done = 0;
2942
2943		mtx_lock(&softc->ctl_lock);
2944		lun = softc->ctl_luns[delete_desc->lun_id];
2945		if (lun == NULL) {
2946			mtx_unlock(&softc->ctl_lock);
2947			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2948			       __func__, (uintmax_t)delete_desc->lun_id);
2949			retval = EINVAL;
2950			break;
2951		}
2952		mtx_lock(&lun->lun_lock);
2953		mtx_unlock(&softc->ctl_lock);
2954		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2955			if (desc->serial != delete_desc->serial)
2956				continue;
2957
2958			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2959				      links);
2960			free(desc, M_CTL);
2961			delete_done = 1;
2962		}
2963		mtx_unlock(&lun->lun_lock);
2964		if (delete_done == 0) {
2965			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2966			       "error serial %ju on LUN %u\n", __func__,
2967			       delete_desc->serial, delete_desc->lun_id);
2968			retval = EINVAL;
2969			break;
2970		}
2971		break;
2972	}
2973	case CTL_DUMP_STRUCTS: {
2974		int i, j, k, idx;
2975		struct ctl_port *port;
2976		struct ctl_frontend *fe;
2977
2978		mtx_lock(&softc->ctl_lock);
2979		printf("CTL Persistent Reservation information start:\n");
2980		for (i = 0; i < CTL_MAX_LUNS; i++) {
2981			struct ctl_lun *lun;
2982
2983			lun = softc->ctl_luns[i];
2984
2985			if ((lun == NULL)
2986			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2987				continue;
2988
2989			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2990				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2991					idx = j * CTL_MAX_INIT_PER_PORT + k;
2992					if (lun->per_res[idx].registered == 0)
2993						continue;
2994					printf("  LUN %d port %d iid %d key "
2995					       "%#jx\n", i, j, k,
2996					       (uintmax_t)scsi_8btou64(
2997					       lun->per_res[idx].res_key.key));
2998				}
2999			}
3000		}
3001		printf("CTL Persistent Reservation information end\n");
3002		printf("CTL Ports:\n");
3003		STAILQ_FOREACH(port, &softc->port_list, links) {
3004			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3005			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3006			       port->frontend->name, port->port_type,
3007			       port->physical_port, port->virtual_port,
3008			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3009			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3010				if (port->wwpn_iid[j].in_use == 0 &&
3011				    port->wwpn_iid[j].wwpn == 0 &&
3012				    port->wwpn_iid[j].name == NULL)
3013					continue;
3014
3015				printf("    iid %u use %d WWPN %#jx '%s'\n",
3016				    j, port->wwpn_iid[j].in_use,
3017				    (uintmax_t)port->wwpn_iid[j].wwpn,
3018				    port->wwpn_iid[j].name);
3019			}
3020		}
3021		printf("CTL Port information end\n");
3022		mtx_unlock(&softc->ctl_lock);
3023		/*
3024		 * XXX KDM calling this without a lock.  We'd likely want
3025		 * to drop the lock before calling the frontend's dump
3026		 * routine anyway.
3027		 */
3028		printf("CTL Frontends:\n");
3029		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3030			printf("  Frontend '%s'\n", fe->name);
3031			if (fe->fe_dump != NULL)
3032				fe->fe_dump();
3033		}
3034		printf("CTL Frontend information end\n");
3035		break;
3036	}
3037	case CTL_LUN_REQ: {
3038		struct ctl_lun_req *lun_req;
3039		struct ctl_backend_driver *backend;
3040
3041		lun_req = (struct ctl_lun_req *)addr;
3042
3043		backend = ctl_backend_find(lun_req->backend);
3044		if (backend == NULL) {
3045			lun_req->status = CTL_LUN_ERROR;
3046			snprintf(lun_req->error_str,
3047				 sizeof(lun_req->error_str),
3048				 "Backend \"%s\" not found.",
3049				 lun_req->backend);
3050			break;
3051		}
3052		if (lun_req->num_be_args > 0) {
3053			lun_req->kern_be_args = ctl_copyin_args(
3054				lun_req->num_be_args,
3055				lun_req->be_args,
3056				lun_req->error_str,
3057				sizeof(lun_req->error_str));
3058			if (lun_req->kern_be_args == NULL) {
3059				lun_req->status = CTL_LUN_ERROR;
3060				break;
3061			}
3062		}
3063
3064		retval = backend->ioctl(dev, cmd, addr, flag, td);
3065
3066		if (lun_req->num_be_args > 0) {
3067			ctl_copyout_args(lun_req->num_be_args,
3068				      lun_req->kern_be_args);
3069			ctl_free_args(lun_req->num_be_args,
3070				      lun_req->kern_be_args);
3071		}
3072		break;
3073	}
3074	case CTL_LUN_LIST: {
3075		struct sbuf *sb;
3076		struct ctl_lun *lun;
3077		struct ctl_lun_list *list;
3078		struct ctl_option *opt;
3079
3080		list = (struct ctl_lun_list *)addr;
3081
3082		/*
3083		 * Allocate a fixed length sbuf here, based on the length
3084		 * of the user's buffer.  We could allocate an auto-extending
3085		 * buffer, and then tell the user how much larger our
3086		 * amount of data is than his buffer, but that presents
3087		 * some problems:
3088		 *
3089		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3090		 *     we can't hold a lock while calling them with an
3091		 *     auto-extending buffer.
3092 		 *
3093		 * 2.  There is not currently a LUN reference counting
3094		 *     mechanism, outside of outstanding transactions on
3095		 *     the LUN's OOA queue.  So a LUN could go away on us
3096		 *     while we're getting the LUN number, backend-specific
3097		 *     information, etc.  Thus, given the way things
3098		 *     currently work, we need to hold the CTL lock while
3099		 *     grabbing LUN information.
3100		 *
3101		 * So, from the user's standpoint, the best thing to do is
3102		 * allocate what he thinks is a reasonable buffer length,
3103		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3104		 * double the buffer length and try again.  (And repeat
3105		 * that until he succeeds.)
3106		 */
3107		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3108		if (sb == NULL) {
3109			list->status = CTL_LUN_LIST_ERROR;
3110			snprintf(list->error_str, sizeof(list->error_str),
3111				 "Unable to allocate %d bytes for LUN list",
3112				 list->alloc_len);
3113			break;
3114		}
3115
3116		sbuf_printf(sb, "<ctllunlist>\n");
3117
3118		mtx_lock(&softc->ctl_lock);
3119		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3120			mtx_lock(&lun->lun_lock);
3121			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3122					     (uintmax_t)lun->lun);
3123
3124			/*
3125			 * Bail out as soon as we see that we've overfilled
3126			 * the buffer.
3127			 */
3128			if (retval != 0)
3129				break;
3130
3131			retval = sbuf_printf(sb, "\t<backend_type>%s"
3132					     "</backend_type>\n",
3133					     (lun->backend == NULL) ?  "none" :
3134					     lun->backend->name);
3135
3136			if (retval != 0)
3137				break;
3138
3139			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3140					     lun->be_lun->lun_type);
3141
3142			if (retval != 0)
3143				break;
3144
3145			if (lun->backend == NULL) {
3146				retval = sbuf_printf(sb, "</lun>\n");
3147				if (retval != 0)
3148					break;
3149				continue;
3150			}
3151
3152			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3153					     (lun->be_lun->maxlba > 0) ?
3154					     lun->be_lun->maxlba + 1 : 0);
3155
3156			if (retval != 0)
3157				break;
3158
3159			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3160					     lun->be_lun->blocksize);
3161
3162			if (retval != 0)
3163				break;
3164
3165			retval = sbuf_printf(sb, "\t<serial_number>");
3166
3167			if (retval != 0)
3168				break;
3169
3170			retval = ctl_sbuf_printf_esc(sb,
3171						     lun->be_lun->serial_num);
3172
3173			if (retval != 0)
3174				break;
3175
3176			retval = sbuf_printf(sb, "</serial_number>\n");
3177
3178			if (retval != 0)
3179				break;
3180
3181			retval = sbuf_printf(sb, "\t<device_id>");
3182
3183			if (retval != 0)
3184				break;
3185
3186			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3187
3188			if (retval != 0)
3189				break;
3190
3191			retval = sbuf_printf(sb, "</device_id>\n");
3192
3193			if (retval != 0)
3194				break;
3195
3196			if (lun->backend->lun_info != NULL) {
3197				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3198				if (retval != 0)
3199					break;
3200			}
3201			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3202				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3203				    opt->name, opt->value, opt->name);
3204				if (retval != 0)
3205					break;
3206			}
3207
3208			retval = sbuf_printf(sb, "</lun>\n");
3209
3210			if (retval != 0)
3211				break;
3212			mtx_unlock(&lun->lun_lock);
3213		}
3214		if (lun != NULL)
3215			mtx_unlock(&lun->lun_lock);
3216		mtx_unlock(&softc->ctl_lock);
3217
3218		if ((retval != 0)
3219		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3220			retval = 0;
3221			sbuf_delete(sb);
3222			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3223			snprintf(list->error_str, sizeof(list->error_str),
3224				 "Out of space, %d bytes is too small",
3225				 list->alloc_len);
3226			break;
3227		}
3228
3229		sbuf_finish(sb);
3230
3231		retval = copyout(sbuf_data(sb), list->lun_xml,
3232				 sbuf_len(sb) + 1);
3233
3234		list->fill_len = sbuf_len(sb) + 1;
3235		list->status = CTL_LUN_LIST_OK;
3236		sbuf_delete(sb);
3237		break;
3238	}
3239	case CTL_ISCSI: {
3240		struct ctl_iscsi *ci;
3241		struct ctl_frontend *fe;
3242
3243		ci = (struct ctl_iscsi *)addr;
3244
3245		fe = ctl_frontend_find("iscsi");
3246		if (fe == NULL) {
3247			ci->status = CTL_ISCSI_ERROR;
3248			snprintf(ci->error_str, sizeof(ci->error_str),
3249			    "Frontend \"iscsi\" not found.");
3250			break;
3251		}
3252
3253		retval = fe->ioctl(dev, cmd, addr, flag, td);
3254		break;
3255	}
3256	case CTL_PORT_REQ: {
3257		struct ctl_req *req;
3258		struct ctl_frontend *fe;
3259
3260		req = (struct ctl_req *)addr;
3261
3262		fe = ctl_frontend_find(req->driver);
3263		if (fe == NULL) {
3264			req->status = CTL_LUN_ERROR;
3265			snprintf(req->error_str, sizeof(req->error_str),
3266			    "Frontend \"%s\" not found.", req->driver);
3267			break;
3268		}
3269		if (req->num_args > 0) {
3270			req->kern_args = ctl_copyin_args(req->num_args,
3271			    req->args, req->error_str, sizeof(req->error_str));
3272			if (req->kern_args == NULL) {
3273				req->status = CTL_LUN_ERROR;
3274				break;
3275			}
3276		}
3277
3278		retval = fe->ioctl(dev, cmd, addr, flag, td);
3279
3280		if (req->num_args > 0) {
3281			ctl_copyout_args(req->num_args, req->kern_args);
3282			ctl_free_args(req->num_args, req->kern_args);
3283		}
3284		break;
3285	}
3286	case CTL_PORT_LIST: {
3287		struct sbuf *sb;
3288		struct ctl_port *port;
3289		struct ctl_lun_list *list;
3290		struct ctl_option *opt;
3291
3292		list = (struct ctl_lun_list *)addr;
3293
3294		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3295		if (sb == NULL) {
3296			list->status = CTL_LUN_LIST_ERROR;
3297			snprintf(list->error_str, sizeof(list->error_str),
3298				 "Unable to allocate %d bytes for LUN list",
3299				 list->alloc_len);
3300			break;
3301		}
3302
3303		sbuf_printf(sb, "<ctlportlist>\n");
3304
3305		mtx_lock(&softc->ctl_lock);
3306		STAILQ_FOREACH(port, &softc->port_list, links) {
3307			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3308					     (uintmax_t)port->targ_port);
3309
3310			/*
3311			 * Bail out as soon as we see that we've overfilled
3312			 * the buffer.
3313			 */
3314			if (retval != 0)
3315				break;
3316
3317			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3318			    "</frontend_type>\n", port->frontend->name);
3319			if (retval != 0)
3320				break;
3321
3322			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3323					     port->port_type);
3324			if (retval != 0)
3325				break;
3326
3327			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3328			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3329			if (retval != 0)
3330				break;
3331
3332			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3333			    port->port_name);
3334			if (retval != 0)
3335				break;
3336
3337			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3338			    port->physical_port);
3339			if (retval != 0)
3340				break;
3341
3342			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3343			    port->virtual_port);
3344			if (retval != 0)
3345				break;
3346
3347			retval = sbuf_printf(sb, "\t<wwnn>%#jx</wwnn>\n",
3348			    (uintmax_t)port->wwnn);
3349			if (retval != 0)
3350				break;
3351
3352			retval = sbuf_printf(sb, "\t<wwpn>%#jx</wwpn>\n",
3353			    (uintmax_t)port->wwpn);
3354			if (retval != 0)
3355				break;
3356
3357			if (port->port_info != NULL) {
3358				retval = port->port_info(port->onoff_arg, sb);
3359				if (retval != 0)
3360					break;
3361			}
3362			STAILQ_FOREACH(opt, &port->options, links) {
3363				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3364				    opt->name, opt->value, opt->name);
3365				if (retval != 0)
3366					break;
3367			}
3368
3369			retval = sbuf_printf(sb, "</targ_port>\n");
3370			if (retval != 0)
3371				break;
3372		}
3373		mtx_unlock(&softc->ctl_lock);
3374
3375		if ((retval != 0)
3376		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3377			retval = 0;
3378			sbuf_delete(sb);
3379			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3380			snprintf(list->error_str, sizeof(list->error_str),
3381				 "Out of space, %d bytes is too small",
3382				 list->alloc_len);
3383			break;
3384		}
3385
3386		sbuf_finish(sb);
3387
3388		retval = copyout(sbuf_data(sb), list->lun_xml,
3389				 sbuf_len(sb) + 1);
3390
3391		list->fill_len = sbuf_len(sb) + 1;
3392		list->status = CTL_LUN_LIST_OK;
3393		sbuf_delete(sb);
3394		break;
3395	}
3396	default: {
3397		/* XXX KDM should we fix this? */
3398#if 0
3399		struct ctl_backend_driver *backend;
3400		unsigned int type;
3401		int found;
3402
3403		found = 0;
3404
3405		/*
3406		 * We encode the backend type as the ioctl type for backend
3407		 * ioctls.  So parse it out here, and then search for a
3408		 * backend of this type.
3409		 */
3410		type = _IOC_TYPE(cmd);
3411
3412		STAILQ_FOREACH(backend, &softc->be_list, links) {
3413			if (backend->type == type) {
3414				found = 1;
3415				break;
3416			}
3417		}
3418		if (found == 0) {
3419			printf("ctl: unknown ioctl command %#lx or backend "
3420			       "%d\n", cmd, type);
3421			retval = EINVAL;
3422			break;
3423		}
3424		retval = backend->ioctl(dev, cmd, addr, flag, td);
3425#endif
3426		retval = ENOTTY;
3427		break;
3428	}
3429	}
3430	return (retval);
3431}
3432
3433uint32_t
3434ctl_get_initindex(struct ctl_nexus *nexus)
3435{
3436	if (nexus->targ_port < CTL_MAX_PORTS)
3437		return (nexus->initid.id +
3438			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3439	else
3440		return (nexus->initid.id +
3441		       ((nexus->targ_port - CTL_MAX_PORTS) *
3442			CTL_MAX_INIT_PER_PORT));
3443}
3444
3445uint32_t
3446ctl_get_resindex(struct ctl_nexus *nexus)
3447{
3448	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3449}
3450
3451uint32_t
3452ctl_port_idx(int port_num)
3453{
3454	if (port_num < CTL_MAX_PORTS)
3455		return(port_num);
3456	else
3457		return(port_num - CTL_MAX_PORTS);
3458}
3459
3460static uint32_t
3461ctl_map_lun(int port_num, uint32_t lun_id)
3462{
3463	struct ctl_port *port;
3464
3465	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3466	if (port == NULL)
3467		return (UINT32_MAX);
3468	if (port->lun_map == NULL)
3469		return (lun_id);
3470	return (port->lun_map(port->targ_lun_arg, lun_id));
3471}
3472
3473static uint32_t
3474ctl_map_lun_back(int port_num, uint32_t lun_id)
3475{
3476	struct ctl_port *port;
3477	uint32_t i;
3478
3479	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3480	if (port->lun_map == NULL)
3481		return (lun_id);
3482	for (i = 0; i < CTL_MAX_LUNS; i++) {
3483		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3484			return (i);
3485	}
3486	return (UINT32_MAX);
3487}
3488
3489/*
3490 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3491 * that are a power of 2.
3492 */
3493int
3494ctl_ffz(uint32_t *mask, uint32_t size)
3495{
3496	uint32_t num_chunks, num_pieces;
3497	int i, j;
3498
3499	num_chunks = (size >> 5);
3500	if (num_chunks == 0)
3501		num_chunks++;
3502	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3503
3504	for (i = 0; i < num_chunks; i++) {
3505		for (j = 0; j < num_pieces; j++) {
3506			if ((mask[i] & (1 << j)) == 0)
3507				return ((i << 5) + j);
3508		}
3509	}
3510
3511	return (-1);
3512}
3513
3514int
3515ctl_set_mask(uint32_t *mask, uint32_t bit)
3516{
3517	uint32_t chunk, piece;
3518
3519	chunk = bit >> 5;
3520	piece = bit % (sizeof(uint32_t) * 8);
3521
3522	if ((mask[chunk] & (1 << piece)) != 0)
3523		return (-1);
3524	else
3525		mask[chunk] |= (1 << piece);
3526
3527	return (0);
3528}
3529
3530int
3531ctl_clear_mask(uint32_t *mask, uint32_t bit)
3532{
3533	uint32_t chunk, piece;
3534
3535	chunk = bit >> 5;
3536	piece = bit % (sizeof(uint32_t) * 8);
3537
3538	if ((mask[chunk] & (1 << piece)) == 0)
3539		return (-1);
3540	else
3541		mask[chunk] &= ~(1 << piece);
3542
3543	return (0);
3544}
3545
3546int
3547ctl_is_set(uint32_t *mask, uint32_t bit)
3548{
3549	uint32_t chunk, piece;
3550
3551	chunk = bit >> 5;
3552	piece = bit % (sizeof(uint32_t) * 8);
3553
3554	if ((mask[chunk] & (1 << piece)) == 0)
3555		return (0);
3556	else
3557		return (1);
3558}
3559
3560#ifdef unused
3561/*
3562 * The bus, target and lun are optional, they can be filled in later.
3563 * can_wait is used to determine whether we can wait on the malloc or not.
3564 */
3565union ctl_io*
3566ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3567	      uint32_t targ_lun, int can_wait)
3568{
3569	union ctl_io *io;
3570
3571	if (can_wait)
3572		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3573	else
3574		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3575
3576	if (io != NULL) {
3577		io->io_hdr.io_type = io_type;
3578		io->io_hdr.targ_port = targ_port;
3579		/*
3580		 * XXX KDM this needs to change/go away.  We need to move
3581		 * to a preallocated pool of ctl_scsiio structures.
3582		 */
3583		io->io_hdr.nexus.targ_target.id = targ_target;
3584		io->io_hdr.nexus.targ_lun = targ_lun;
3585	}
3586
3587	return (io);
3588}
3589
3590void
3591ctl_kfree_io(union ctl_io *io)
3592{
3593	free(io, M_CTL);
3594}
3595#endif /* unused */
3596
3597/*
3598 * ctl_softc, pool_type, total_ctl_io are passed in.
3599 * npool is passed out.
3600 */
3601int
3602ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3603		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3604{
3605	uint32_t i;
3606	union ctl_io *cur_io, *next_io;
3607	struct ctl_io_pool *pool;
3608	int retval;
3609
3610	retval = 0;
3611
3612	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3613					    M_NOWAIT | M_ZERO);
3614	if (pool == NULL) {
3615		retval = ENOMEM;
3616		goto bailout;
3617	}
3618
3619	pool->type = pool_type;
3620	pool->ctl_softc = ctl_softc;
3621
3622	mtx_lock(&ctl_softc->pool_lock);
3623	pool->id = ctl_softc->cur_pool_id++;
3624	mtx_unlock(&ctl_softc->pool_lock);
3625
3626	pool->flags = CTL_POOL_FLAG_NONE;
3627	pool->refcount = 1;		/* Reference for validity. */
3628	STAILQ_INIT(&pool->free_queue);
3629
3630	/*
3631	 * XXX KDM other options here:
3632	 * - allocate a page at a time
3633	 * - allocate one big chunk of memory.
3634	 * Page allocation might work well, but would take a little more
3635	 * tracking.
3636	 */
3637	for (i = 0; i < total_ctl_io; i++) {
3638		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO,
3639						M_NOWAIT);
3640		if (cur_io == NULL) {
3641			retval = ENOMEM;
3642			break;
3643		}
3644		cur_io->io_hdr.pool = pool;
3645		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3646		pool->total_ctl_io++;
3647		pool->free_ctl_io++;
3648	}
3649
3650	if (retval != 0) {
3651		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3652		     cur_io != NULL; cur_io = next_io) {
3653			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3654							      links);
3655			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3656				      ctl_io_hdr, links);
3657			free(cur_io, M_CTLIO);
3658		}
3659
3660		free(pool, M_CTL);
3661		goto bailout;
3662	}
3663	mtx_lock(&ctl_softc->pool_lock);
3664	ctl_softc->num_pools++;
3665	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3666	/*
3667	 * Increment our usage count if this is an external consumer, so we
3668	 * can't get unloaded until the external consumer (most likely a
3669	 * FETD) unloads and frees his pool.
3670	 *
3671	 * XXX KDM will this increment the caller's module use count, or
3672	 * mine?
3673	 */
3674#if 0
3675	if ((pool_type != CTL_POOL_EMERGENCY)
3676	 && (pool_type != CTL_POOL_INTERNAL)
3677	 && (pool_type != CTL_POOL_4OTHERSC))
3678		MOD_INC_USE_COUNT;
3679#endif
3680
3681	mtx_unlock(&ctl_softc->pool_lock);
3682
3683	*npool = pool;
3684
3685bailout:
3686
3687	return (retval);
3688}
3689
3690static int
3691ctl_pool_acquire(struct ctl_io_pool *pool)
3692{
3693
3694	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3695
3696	if (pool->flags & CTL_POOL_FLAG_INVALID)
3697		return (EINVAL);
3698
3699	pool->refcount++;
3700
3701	return (0);
3702}
3703
3704static void
3705ctl_pool_release(struct ctl_io_pool *pool)
3706{
3707	struct ctl_softc *ctl_softc = pool->ctl_softc;
3708	union ctl_io *io;
3709
3710	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3711
3712	if (--pool->refcount != 0)
3713		return;
3714
3715	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3716		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3717			      links);
3718		free(io, M_CTLIO);
3719	}
3720
3721	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3722	ctl_softc->num_pools--;
3723
3724	/*
3725	 * XXX KDM will this decrement the caller's usage count or mine?
3726	 */
3727#if 0
3728	if ((pool->type != CTL_POOL_EMERGENCY)
3729	 && (pool->type != CTL_POOL_INTERNAL)
3730	 && (pool->type != CTL_POOL_4OTHERSC))
3731		MOD_DEC_USE_COUNT;
3732#endif
3733
3734	free(pool, M_CTL);
3735}
3736
3737void
3738ctl_pool_free(struct ctl_io_pool *pool)
3739{
3740	struct ctl_softc *ctl_softc;
3741
3742	if (pool == NULL)
3743		return;
3744
3745	ctl_softc = pool->ctl_softc;
3746	mtx_lock(&ctl_softc->pool_lock);
3747	pool->flags |= CTL_POOL_FLAG_INVALID;
3748	ctl_pool_release(pool);
3749	mtx_unlock(&ctl_softc->pool_lock);
3750}
3751
3752/*
3753 * This routine does not block (except for spinlocks of course).
3754 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3755 * possible.
3756 */
3757union ctl_io *
3758ctl_alloc_io(void *pool_ref)
3759{
3760	union ctl_io *io;
3761	struct ctl_softc *ctl_softc;
3762	struct ctl_io_pool *pool, *npool;
3763	struct ctl_io_pool *emergency_pool;
3764
3765	pool = (struct ctl_io_pool *)pool_ref;
3766
3767	if (pool == NULL) {
3768		printf("%s: pool is NULL\n", __func__);
3769		return (NULL);
3770	}
3771
3772	emergency_pool = NULL;
3773
3774	ctl_softc = pool->ctl_softc;
3775
3776	mtx_lock(&ctl_softc->pool_lock);
3777	/*
3778	 * First, try to get the io structure from the user's pool.
3779	 */
3780	if (ctl_pool_acquire(pool) == 0) {
3781		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3782		if (io != NULL) {
3783			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3784			pool->total_allocated++;
3785			pool->free_ctl_io--;
3786			mtx_unlock(&ctl_softc->pool_lock);
3787			return (io);
3788		} else
3789			ctl_pool_release(pool);
3790	}
3791	/*
3792	 * If he doesn't have any io structures left, search for an
3793	 * emergency pool and grab one from there.
3794	 */
3795	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3796		if (npool->type != CTL_POOL_EMERGENCY)
3797			continue;
3798
3799		if (ctl_pool_acquire(npool) != 0)
3800			continue;
3801
3802		emergency_pool = npool;
3803
3804		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3805		if (io != NULL) {
3806			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3807			npool->total_allocated++;
3808			npool->free_ctl_io--;
3809			mtx_unlock(&ctl_softc->pool_lock);
3810			return (io);
3811		} else
3812			ctl_pool_release(npool);
3813	}
3814
3815	/* Drop the spinlock before we malloc */
3816	mtx_unlock(&ctl_softc->pool_lock);
3817
3818	/*
3819	 * The emergency pool (if it exists) didn't have one, so try an
3820	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3821	 */
3822	io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT);
3823	if (io != NULL) {
3824		/*
3825		 * If the emergency pool exists but is empty, add this
3826		 * ctl_io to its list when it gets freed.
3827		 */
3828		if (emergency_pool != NULL) {
3829			mtx_lock(&ctl_softc->pool_lock);
3830			if (ctl_pool_acquire(emergency_pool) == 0) {
3831				io->io_hdr.pool = emergency_pool;
3832				emergency_pool->total_ctl_io++;
3833				/*
3834				 * Need to bump this, otherwise
3835				 * total_allocated and total_freed won't
3836				 * match when we no longer have anything
3837				 * outstanding.
3838				 */
3839				emergency_pool->total_allocated++;
3840			}
3841			mtx_unlock(&ctl_softc->pool_lock);
3842		} else
3843			io->io_hdr.pool = NULL;
3844	}
3845
3846	return (io);
3847}
3848
3849void
3850ctl_free_io(union ctl_io *io)
3851{
3852	if (io == NULL)
3853		return;
3854
3855	/*
3856	 * If this ctl_io has a pool, return it to that pool.
3857	 */
3858	if (io->io_hdr.pool != NULL) {
3859		struct ctl_io_pool *pool;
3860
3861		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3862		mtx_lock(&pool->ctl_softc->pool_lock);
3863		io->io_hdr.io_type = 0xff;
3864		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3865		pool->total_freed++;
3866		pool->free_ctl_io++;
3867		ctl_pool_release(pool);
3868		mtx_unlock(&pool->ctl_softc->pool_lock);
3869	} else {
3870		/*
3871		 * Otherwise, just free it.  We probably malloced it and
3872		 * the emergency pool wasn't available.
3873		 */
3874		free(io, M_CTLIO);
3875	}
3876
3877}
3878
3879void
3880ctl_zero_io(union ctl_io *io)
3881{
3882	void *pool_ref;
3883
3884	if (io == NULL)
3885		return;
3886
3887	/*
3888	 * May need to preserve linked list pointers at some point too.
3889	 */
3890	pool_ref = io->io_hdr.pool;
3891
3892	memset(io, 0, sizeof(*io));
3893
3894	io->io_hdr.pool = pool_ref;
3895}
3896
3897/*
3898 * This routine is currently used for internal copies of ctl_ios that need
3899 * to persist for some reason after we've already returned status to the
3900 * FETD.  (Thus the flag set.)
3901 *
3902 * XXX XXX
3903 * Note that this makes a blind copy of all fields in the ctl_io, except
3904 * for the pool reference.  This includes any memory that has been
3905 * allocated!  That memory will no longer be valid after done has been
3906 * called, so this would be VERY DANGEROUS for command that actually does
3907 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3908 * start and stop commands, which don't transfer any data, so this is not a
3909 * problem.  If it is used for anything else, the caller would also need to
3910 * allocate data buffer space and this routine would need to be modified to
3911 * copy the data buffer(s) as well.
3912 */
3913void
3914ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3915{
3916	void *pool_ref;
3917
3918	if ((src == NULL)
3919	 || (dest == NULL))
3920		return;
3921
3922	/*
3923	 * May need to preserve linked list pointers at some point too.
3924	 */
3925	pool_ref = dest->io_hdr.pool;
3926
3927	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3928
3929	dest->io_hdr.pool = pool_ref;
3930	/*
3931	 * We need to know that this is an internal copy, and doesn't need
3932	 * to get passed back to the FETD that allocated it.
3933	 */
3934	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3935}
3936
3937#ifdef NEEDTOPORT
3938static void
3939ctl_update_power_subpage(struct copan_power_subpage *page)
3940{
3941	int num_luns, num_partitions, config_type;
3942	struct ctl_softc *softc;
3943	cs_BOOL_t aor_present, shelf_50pct_power;
3944	cs_raidset_personality_t rs_type;
3945	int max_active_luns;
3946
3947	softc = control_softc;
3948
3949	/* subtract out the processor LUN */
3950	num_luns = softc->num_luns - 1;
3951	/*
3952	 * Default to 7 LUNs active, which was the only number we allowed
3953	 * in the past.
3954	 */
3955	max_active_luns = 7;
3956
3957	num_partitions = config_GetRsPartitionInfo();
3958	config_type = config_GetConfigType();
3959	shelf_50pct_power = config_GetShelfPowerMode();
3960	aor_present = config_IsAorRsPresent();
3961
3962	rs_type = ddb_GetRsRaidType(1);
3963	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3964	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3965		EPRINT(0, "Unsupported RS type %d!", rs_type);
3966	}
3967
3968
3969	page->total_luns = num_luns;
3970
3971	switch (config_type) {
3972	case 40:
3973		/*
3974		 * In a 40 drive configuration, it doesn't matter what DC
3975		 * cards we have, whether we have AOR enabled or not,
3976		 * partitioning or not, or what type of RAIDset we have.
3977		 * In that scenario, we can power up every LUN we present
3978		 * to the user.
3979		 */
3980		max_active_luns = num_luns;
3981
3982		break;
3983	case 64:
3984		if (shelf_50pct_power == CS_FALSE) {
3985			/* 25% power */
3986			if (aor_present == CS_TRUE) {
3987				if (rs_type ==
3988				     CS_RAIDSET_PERSONALITY_RAID5) {
3989					max_active_luns = 7;
3990				} else if (rs_type ==
3991					 CS_RAIDSET_PERSONALITY_RAID1){
3992					max_active_luns = 14;
3993				} else {
3994					/* XXX KDM now what?? */
3995				}
3996			} else {
3997				if (rs_type ==
3998				     CS_RAIDSET_PERSONALITY_RAID5) {
3999					max_active_luns = 8;
4000				} else if (rs_type ==
4001					 CS_RAIDSET_PERSONALITY_RAID1){
4002					max_active_luns = 16;
4003				} else {
4004					/* XXX KDM now what?? */
4005				}
4006			}
4007		} else {
4008			/* 50% power */
4009			/*
4010			 * With 50% power in a 64 drive configuration, we
4011			 * can power all LUNs we present.
4012			 */
4013			max_active_luns = num_luns;
4014		}
4015		break;
4016	case 112:
4017		if (shelf_50pct_power == CS_FALSE) {
4018			/* 25% power */
4019			if (aor_present == CS_TRUE) {
4020				if (rs_type ==
4021				     CS_RAIDSET_PERSONALITY_RAID5) {
4022					max_active_luns = 7;
4023				} else if (rs_type ==
4024					 CS_RAIDSET_PERSONALITY_RAID1){
4025					max_active_luns = 14;
4026				} else {
4027					/* XXX KDM now what?? */
4028				}
4029			} else {
4030				if (rs_type ==
4031				     CS_RAIDSET_PERSONALITY_RAID5) {
4032					max_active_luns = 8;
4033				} else if (rs_type ==
4034					 CS_RAIDSET_PERSONALITY_RAID1){
4035					max_active_luns = 16;
4036				} else {
4037					/* XXX KDM now what?? */
4038				}
4039			}
4040		} else {
4041			/* 50% power */
4042			if (aor_present == CS_TRUE) {
4043				if (rs_type ==
4044				     CS_RAIDSET_PERSONALITY_RAID5) {
4045					max_active_luns = 14;
4046				} else if (rs_type ==
4047					 CS_RAIDSET_PERSONALITY_RAID1){
4048					/*
4049					 * We're assuming here that disk
4050					 * caching is enabled, and so we're
4051					 * able to power up half of each
4052					 * LUN, and cache all writes.
4053					 */
4054					max_active_luns = num_luns;
4055				} else {
4056					/* XXX KDM now what?? */
4057				}
4058			} else {
4059				if (rs_type ==
4060				     CS_RAIDSET_PERSONALITY_RAID5) {
4061					max_active_luns = 15;
4062				} else if (rs_type ==
4063					 CS_RAIDSET_PERSONALITY_RAID1){
4064					max_active_luns = 30;
4065				} else {
4066					/* XXX KDM now what?? */
4067				}
4068			}
4069		}
4070		break;
4071	default:
4072		/*
4073		 * In this case, we have an unknown configuration, so we
4074		 * just use the default from above.
4075		 */
4076		break;
4077	}
4078
4079	page->max_active_luns = max_active_luns;
4080#if 0
4081	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
4082	       page->total_luns, page->max_active_luns);
4083#endif
4084}
4085#endif /* NEEDTOPORT */
4086
4087/*
4088 * This routine could be used in the future to load default and/or saved
4089 * mode page parameters for a particuar lun.
4090 */
4091static int
4092ctl_init_page_index(struct ctl_lun *lun)
4093{
4094	int i;
4095	struct ctl_page_index *page_index;
4096	struct ctl_softc *softc;
4097
4098	memcpy(&lun->mode_pages.index, page_index_template,
4099	       sizeof(page_index_template));
4100
4101	softc = lun->ctl_softc;
4102
4103	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4104
4105		page_index = &lun->mode_pages.index[i];
4106		/*
4107		 * If this is a disk-only mode page, there's no point in
4108		 * setting it up.  For some pages, we have to have some
4109		 * basic information about the disk in order to calculate the
4110		 * mode page data.
4111		 */
4112		if ((lun->be_lun->lun_type != T_DIRECT)
4113		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4114			continue;
4115
4116		switch (page_index->page_code & SMPH_PC_MASK) {
4117		case SMS_FORMAT_DEVICE_PAGE: {
4118			struct scsi_format_page *format_page;
4119
4120			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4121				panic("subpage is incorrect!");
4122
4123			/*
4124			 * Sectors per track are set above.  Bytes per
4125			 * sector need to be set here on a per-LUN basis.
4126			 */
4127			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4128			       &format_page_default,
4129			       sizeof(format_page_default));
4130			memcpy(&lun->mode_pages.format_page[
4131			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4132			       sizeof(format_page_changeable));
4133			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4134			       &format_page_default,
4135			       sizeof(format_page_default));
4136			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4137			       &format_page_default,
4138			       sizeof(format_page_default));
4139
4140			format_page = &lun->mode_pages.format_page[
4141				CTL_PAGE_CURRENT];
4142			scsi_ulto2b(lun->be_lun->blocksize,
4143				    format_page->bytes_per_sector);
4144
4145			format_page = &lun->mode_pages.format_page[
4146				CTL_PAGE_DEFAULT];
4147			scsi_ulto2b(lun->be_lun->blocksize,
4148				    format_page->bytes_per_sector);
4149
4150			format_page = &lun->mode_pages.format_page[
4151				CTL_PAGE_SAVED];
4152			scsi_ulto2b(lun->be_lun->blocksize,
4153				    format_page->bytes_per_sector);
4154
4155			page_index->page_data =
4156				(uint8_t *)lun->mode_pages.format_page;
4157			break;
4158		}
4159		case SMS_RIGID_DISK_PAGE: {
4160			struct scsi_rigid_disk_page *rigid_disk_page;
4161			uint32_t sectors_per_cylinder;
4162			uint64_t cylinders;
4163#ifndef	__XSCALE__
4164			int shift;
4165#endif /* !__XSCALE__ */
4166
4167			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4168				panic("invalid subpage value %d",
4169				      page_index->subpage);
4170
4171			/*
4172			 * Rotation rate and sectors per track are set
4173			 * above.  We calculate the cylinders here based on
4174			 * capacity.  Due to the number of heads and
4175			 * sectors per track we're using, smaller arrays
4176			 * may turn out to have 0 cylinders.  Linux and
4177			 * FreeBSD don't pay attention to these mode pages
4178			 * to figure out capacity, but Solaris does.  It
4179			 * seems to deal with 0 cylinders just fine, and
4180			 * works out a fake geometry based on the capacity.
4181			 */
4182			memcpy(&lun->mode_pages.rigid_disk_page[
4183			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
4184			       sizeof(rigid_disk_page_default));
4185			memcpy(&lun->mode_pages.rigid_disk_page[
4186			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4187			       sizeof(rigid_disk_page_changeable));
4188			memcpy(&lun->mode_pages.rigid_disk_page[
4189			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4190			       sizeof(rigid_disk_page_default));
4191			memcpy(&lun->mode_pages.rigid_disk_page[
4192			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4193			       sizeof(rigid_disk_page_default));
4194
4195			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4196				CTL_DEFAULT_HEADS;
4197
4198			/*
4199			 * The divide method here will be more accurate,
4200			 * probably, but results in floating point being
4201			 * used in the kernel on i386 (__udivdi3()).  On the
4202			 * XScale, though, __udivdi3() is implemented in
4203			 * software.
4204			 *
4205			 * The shift method for cylinder calculation is
4206			 * accurate if sectors_per_cylinder is a power of
4207			 * 2.  Otherwise it might be slightly off -- you
4208			 * might have a bit of a truncation problem.
4209			 */
4210#ifdef	__XSCALE__
4211			cylinders = (lun->be_lun->maxlba + 1) /
4212				sectors_per_cylinder;
4213#else
4214			for (shift = 31; shift > 0; shift--) {
4215				if (sectors_per_cylinder & (1 << shift))
4216					break;
4217			}
4218			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4219#endif
4220
4221			/*
4222			 * We've basically got 3 bytes, or 24 bits for the
4223			 * cylinder size in the mode page.  If we're over,
4224			 * just round down to 2^24.
4225			 */
4226			if (cylinders > 0xffffff)
4227				cylinders = 0xffffff;
4228
4229			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4230				CTL_PAGE_CURRENT];
4231			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4232
4233			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4234				CTL_PAGE_DEFAULT];
4235			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4236
4237			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4238				CTL_PAGE_SAVED];
4239			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4240
4241			page_index->page_data =
4242				(uint8_t *)lun->mode_pages.rigid_disk_page;
4243			break;
4244		}
4245		case SMS_CACHING_PAGE: {
4246
4247			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4248				panic("invalid subpage value %d",
4249				      page_index->subpage);
4250			/*
4251			 * Defaults should be okay here, no calculations
4252			 * needed.
4253			 */
4254			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4255			       &caching_page_default,
4256			       sizeof(caching_page_default));
4257			memcpy(&lun->mode_pages.caching_page[
4258			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4259			       sizeof(caching_page_changeable));
4260			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4261			       &caching_page_default,
4262			       sizeof(caching_page_default));
4263			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4264			       &caching_page_default,
4265			       sizeof(caching_page_default));
4266			page_index->page_data =
4267				(uint8_t *)lun->mode_pages.caching_page;
4268			break;
4269		}
4270		case SMS_CONTROL_MODE_PAGE: {
4271
4272			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4273				panic("invalid subpage value %d",
4274				      page_index->subpage);
4275
4276			/*
4277			 * Defaults should be okay here, no calculations
4278			 * needed.
4279			 */
4280			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4281			       &control_page_default,
4282			       sizeof(control_page_default));
4283			memcpy(&lun->mode_pages.control_page[
4284			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4285			       sizeof(control_page_changeable));
4286			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4287			       &control_page_default,
4288			       sizeof(control_page_default));
4289			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4290			       &control_page_default,
4291			       sizeof(control_page_default));
4292			page_index->page_data =
4293				(uint8_t *)lun->mode_pages.control_page;
4294			break;
4295
4296		}
4297		case SMS_VENDOR_SPECIFIC_PAGE:{
4298			switch (page_index->subpage) {
4299			case PWR_SUBPAGE_CODE: {
4300				struct copan_power_subpage *current_page,
4301							   *saved_page;
4302
4303				memcpy(&lun->mode_pages.power_subpage[
4304				       CTL_PAGE_CURRENT],
4305				       &power_page_default,
4306				       sizeof(power_page_default));
4307				memcpy(&lun->mode_pages.power_subpage[
4308				       CTL_PAGE_CHANGEABLE],
4309				       &power_page_changeable,
4310				       sizeof(power_page_changeable));
4311				memcpy(&lun->mode_pages.power_subpage[
4312				       CTL_PAGE_DEFAULT],
4313				       &power_page_default,
4314				       sizeof(power_page_default));
4315				memcpy(&lun->mode_pages.power_subpage[
4316				       CTL_PAGE_SAVED],
4317				       &power_page_default,
4318				       sizeof(power_page_default));
4319				page_index->page_data =
4320				    (uint8_t *)lun->mode_pages.power_subpage;
4321
4322				current_page = (struct copan_power_subpage *)
4323					(page_index->page_data +
4324					 (page_index->page_len *
4325					  CTL_PAGE_CURRENT));
4326			        saved_page = (struct copan_power_subpage *)
4327				        (page_index->page_data +
4328					 (page_index->page_len *
4329					  CTL_PAGE_SAVED));
4330				break;
4331			}
4332			case APS_SUBPAGE_CODE: {
4333				struct copan_aps_subpage *current_page,
4334							 *saved_page;
4335
4336				// This gets set multiple times but
4337				// it should always be the same. It's
4338				// only done during init so who cares.
4339				index_to_aps_page = i;
4340
4341				memcpy(&lun->mode_pages.aps_subpage[
4342				       CTL_PAGE_CURRENT],
4343				       &aps_page_default,
4344				       sizeof(aps_page_default));
4345				memcpy(&lun->mode_pages.aps_subpage[
4346				       CTL_PAGE_CHANGEABLE],
4347				       &aps_page_changeable,
4348				       sizeof(aps_page_changeable));
4349				memcpy(&lun->mode_pages.aps_subpage[
4350				       CTL_PAGE_DEFAULT],
4351				       &aps_page_default,
4352				       sizeof(aps_page_default));
4353				memcpy(&lun->mode_pages.aps_subpage[
4354				       CTL_PAGE_SAVED],
4355				       &aps_page_default,
4356				       sizeof(aps_page_default));
4357				page_index->page_data =
4358					(uint8_t *)lun->mode_pages.aps_subpage;
4359
4360				current_page = (struct copan_aps_subpage *)
4361					(page_index->page_data +
4362					 (page_index->page_len *
4363					  CTL_PAGE_CURRENT));
4364				saved_page = (struct copan_aps_subpage *)
4365					(page_index->page_data +
4366					 (page_index->page_len *
4367					  CTL_PAGE_SAVED));
4368				break;
4369			}
4370			case DBGCNF_SUBPAGE_CODE: {
4371				struct copan_debugconf_subpage *current_page,
4372							       *saved_page;
4373
4374				memcpy(&lun->mode_pages.debugconf_subpage[
4375				       CTL_PAGE_CURRENT],
4376				       &debugconf_page_default,
4377				       sizeof(debugconf_page_default));
4378				memcpy(&lun->mode_pages.debugconf_subpage[
4379				       CTL_PAGE_CHANGEABLE],
4380				       &debugconf_page_changeable,
4381				       sizeof(debugconf_page_changeable));
4382				memcpy(&lun->mode_pages.debugconf_subpage[
4383				       CTL_PAGE_DEFAULT],
4384				       &debugconf_page_default,
4385				       sizeof(debugconf_page_default));
4386				memcpy(&lun->mode_pages.debugconf_subpage[
4387				       CTL_PAGE_SAVED],
4388				       &debugconf_page_default,
4389				       sizeof(debugconf_page_default));
4390				page_index->page_data =
4391					(uint8_t *)lun->mode_pages.debugconf_subpage;
4392
4393				current_page = (struct copan_debugconf_subpage *)
4394					(page_index->page_data +
4395					 (page_index->page_len *
4396					  CTL_PAGE_CURRENT));
4397				saved_page = (struct copan_debugconf_subpage *)
4398					(page_index->page_data +
4399					 (page_index->page_len *
4400					  CTL_PAGE_SAVED));
4401				break;
4402			}
4403			default:
4404				panic("invalid subpage value %d",
4405				      page_index->subpage);
4406				break;
4407			}
4408   			break;
4409		}
4410		default:
4411			panic("invalid page value %d",
4412			      page_index->page_code & SMPH_PC_MASK);
4413			break;
4414    	}
4415	}
4416
4417	return (CTL_RETVAL_COMPLETE);
4418}
4419
4420/*
4421 * LUN allocation.
4422 *
4423 * Requirements:
4424 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4425 *   wants us to allocate the LUN and he can block.
4426 * - ctl_softc is always set
4427 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4428 *
4429 * Returns 0 for success, non-zero (errno) for failure.
4430 */
4431static int
4432ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4433	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4434{
4435	struct ctl_lun *nlun, *lun;
4436	struct ctl_port *port;
4437	struct scsi_vpd_id_descriptor *desc;
4438	struct scsi_vpd_id_t10 *t10id;
4439	const char *eui, *naa, *scsiname, *vendor;
4440	int lun_number, i, lun_malloced;
4441	int devidlen, idlen1, idlen2 = 0, len;
4442
4443	if (be_lun == NULL)
4444		return (EINVAL);
4445
4446	/*
4447	 * We currently only support Direct Access or Processor LUN types.
4448	 */
4449	switch (be_lun->lun_type) {
4450	case T_DIRECT:
4451		break;
4452	case T_PROCESSOR:
4453		break;
4454	case T_SEQUENTIAL:
4455	case T_CHANGER:
4456	default:
4457		be_lun->lun_config_status(be_lun->be_lun,
4458					  CTL_LUN_CONFIG_FAILURE);
4459		break;
4460	}
4461	if (ctl_lun == NULL) {
4462		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4463		lun_malloced = 1;
4464	} else {
4465		lun_malloced = 0;
4466		lun = ctl_lun;
4467	}
4468
4469	memset(lun, 0, sizeof(*lun));
4470	if (lun_malloced)
4471		lun->flags = CTL_LUN_MALLOCED;
4472
4473	/* Generate LUN ID. */
4474	devidlen = max(CTL_DEVID_MIN_LEN,
4475	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4476	idlen1 = sizeof(*t10id) + devidlen;
4477	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4478	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4479	if (scsiname != NULL) {
4480		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4481		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4482	}
4483	eui = ctl_get_opt(&be_lun->options, "eui");
4484	if (eui != NULL) {
4485		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4486	}
4487	naa = ctl_get_opt(&be_lun->options, "naa");
4488	if (naa != NULL) {
4489		len += sizeof(struct scsi_vpd_id_descriptor) + 8;
4490	}
4491	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4492	    M_CTL, M_WAITOK | M_ZERO);
4493	lun->lun_devid->len = len;
4494	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4495	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4496	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4497	desc->length = idlen1;
4498	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4499	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4500	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4501		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4502	} else {
4503		strncpy(t10id->vendor, vendor,
4504		    min(sizeof(t10id->vendor), strlen(vendor)));
4505	}
4506	strncpy((char *)t10id->vendor_spec_id,
4507	    (char *)be_lun->device_id, devidlen);
4508	if (scsiname != NULL) {
4509		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4510		    desc->length);
4511		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4512		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4513		    SVPD_ID_TYPE_SCSI_NAME;
4514		desc->length = idlen2;
4515		strlcpy(desc->identifier, scsiname, idlen2);
4516	}
4517	if (eui != NULL) {
4518		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4519		    desc->length);
4520		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4521		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4522		    SVPD_ID_TYPE_EUI64;
4523		desc->length = 8;
4524		scsi_u64to8b(strtouq(eui, NULL, 0), desc->identifier);
4525	}
4526	if (naa != NULL) {
4527		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4528		    desc->length);
4529		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4530		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4531		    SVPD_ID_TYPE_NAA;
4532		desc->length = 8;
4533		scsi_u64to8b(strtouq(naa, NULL, 0), desc->identifier);
4534	}
4535
4536	mtx_lock(&ctl_softc->ctl_lock);
4537	/*
4538	 * See if the caller requested a particular LUN number.  If so, see
4539	 * if it is available.  Otherwise, allocate the first available LUN.
4540	 */
4541	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4542		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4543		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4544			mtx_unlock(&ctl_softc->ctl_lock);
4545			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4546				printf("ctl: requested LUN ID %d is higher "
4547				       "than CTL_MAX_LUNS - 1 (%d)\n",
4548				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4549			} else {
4550				/*
4551				 * XXX KDM return an error, or just assign
4552				 * another LUN ID in this case??
4553				 */
4554				printf("ctl: requested LUN ID %d is already "
4555				       "in use\n", be_lun->req_lun_id);
4556			}
4557			if (lun->flags & CTL_LUN_MALLOCED)
4558				free(lun, M_CTL);
4559			be_lun->lun_config_status(be_lun->be_lun,
4560						  CTL_LUN_CONFIG_FAILURE);
4561			return (ENOSPC);
4562		}
4563		lun_number = be_lun->req_lun_id;
4564	} else {
4565		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4566		if (lun_number == -1) {
4567			mtx_unlock(&ctl_softc->ctl_lock);
4568			printf("ctl: can't allocate LUN on target %ju, out of "
4569			       "LUNs\n", (uintmax_t)target_id.id);
4570			if (lun->flags & CTL_LUN_MALLOCED)
4571				free(lun, M_CTL);
4572			be_lun->lun_config_status(be_lun->be_lun,
4573						  CTL_LUN_CONFIG_FAILURE);
4574			return (ENOSPC);
4575		}
4576	}
4577	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4578
4579	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4580	lun->target = target_id;
4581	lun->lun = lun_number;
4582	lun->be_lun = be_lun;
4583	/*
4584	 * The processor LUN is always enabled.  Disk LUNs come on line
4585	 * disabled, and must be enabled by the backend.
4586	 */
4587	lun->flags |= CTL_LUN_DISABLED;
4588	lun->backend = be_lun->be;
4589	be_lun->ctl_lun = lun;
4590	be_lun->lun_id = lun_number;
4591	atomic_add_int(&be_lun->be->num_luns, 1);
4592	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4593		lun->flags |= CTL_LUN_STOPPED;
4594
4595	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4596		lun->flags |= CTL_LUN_INOPERABLE;
4597
4598	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4599		lun->flags |= CTL_LUN_PRIMARY_SC;
4600
4601	lun->ctl_softc = ctl_softc;
4602	TAILQ_INIT(&lun->ooa_queue);
4603	TAILQ_INIT(&lun->blocked_queue);
4604	STAILQ_INIT(&lun->error_list);
4605	ctl_tpc_lun_init(lun);
4606
4607	/*
4608	 * Initialize the mode page index.
4609	 */
4610	ctl_init_page_index(lun);
4611
4612	/*
4613	 * Set the poweron UA for all initiators on this LUN only.
4614	 */
4615	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4616		lun->pending_ua[i] = CTL_UA_POWERON;
4617
4618	/*
4619	 * Now, before we insert this lun on the lun list, set the lun
4620	 * inventory changed UA for all other luns.
4621	 */
4622	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4623		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4624			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4625		}
4626	}
4627
4628	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4629
4630	ctl_softc->ctl_luns[lun_number] = lun;
4631
4632	ctl_softc->num_luns++;
4633
4634	/* Setup statistics gathering */
4635	lun->stats.device_type = be_lun->lun_type;
4636	lun->stats.lun_number = lun_number;
4637	if (lun->stats.device_type == T_DIRECT)
4638		lun->stats.blocksize = be_lun->blocksize;
4639	else
4640		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4641	for (i = 0;i < CTL_MAX_PORTS;i++)
4642		lun->stats.ports[i].targ_port = i;
4643
4644	mtx_unlock(&ctl_softc->ctl_lock);
4645
4646	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4647
4648	/*
4649	 * Run through each registered FETD and bring it online if it isn't
4650	 * already.  Enable the target ID if it hasn't been enabled, and
4651	 * enable this particular LUN.
4652	 */
4653	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4654		int retval;
4655
4656		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4657		if (retval != 0) {
4658			printf("ctl_alloc_lun: FETD %s port %d returned error "
4659			       "%d for lun_enable on target %ju lun %d\n",
4660			       port->port_name, port->targ_port, retval,
4661			       (uintmax_t)target_id.id, lun_number);
4662		} else
4663			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4664	}
4665	return (0);
4666}
4667
4668/*
4669 * Delete a LUN.
4670 * Assumptions:
4671 * - LUN has already been marked invalid and any pending I/O has been taken
4672 *   care of.
4673 */
4674static int
4675ctl_free_lun(struct ctl_lun *lun)
4676{
4677	struct ctl_softc *softc;
4678#if 0
4679	struct ctl_port *port;
4680#endif
4681	struct ctl_lun *nlun;
4682	int i;
4683
4684	softc = lun->ctl_softc;
4685
4686	mtx_assert(&softc->ctl_lock, MA_OWNED);
4687
4688	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4689
4690	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4691
4692	softc->ctl_luns[lun->lun] = NULL;
4693
4694	if (!TAILQ_EMPTY(&lun->ooa_queue))
4695		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4696
4697	softc->num_luns--;
4698
4699	/*
4700	 * XXX KDM this scheme only works for a single target/multiple LUN
4701	 * setup.  It needs to be revamped for a multiple target scheme.
4702	 *
4703	 * XXX KDM this results in port->lun_disable() getting called twice,
4704	 * once when ctl_disable_lun() is called, and a second time here.
4705	 * We really need to re-think the LUN disable semantics.  There
4706	 * should probably be several steps/levels to LUN removal:
4707	 *  - disable
4708	 *  - invalidate
4709	 *  - free
4710 	 *
4711	 * Right now we only have a disable method when communicating to
4712	 * the front end ports, at least for individual LUNs.
4713	 */
4714#if 0
4715	STAILQ_FOREACH(port, &softc->port_list, links) {
4716		int retval;
4717
4718		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4719					 lun->lun);
4720		if (retval != 0) {
4721			printf("ctl_free_lun: FETD %s port %d returned error "
4722			       "%d for lun_disable on target %ju lun %jd\n",
4723			       port->port_name, port->targ_port, retval,
4724			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4725		}
4726
4727		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4728			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4729
4730			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4731			if (retval != 0) {
4732				printf("ctl_free_lun: FETD %s port %d "
4733				       "returned error %d for targ_disable on "
4734				       "target %ju\n", port->port_name,
4735				       port->targ_port, retval,
4736				       (uintmax_t)lun->target.id);
4737			} else
4738				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4739
4740			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4741				continue;
4742
4743#if 0
4744			port->port_offline(port->onoff_arg);
4745			port->status &= ~CTL_PORT_STATUS_ONLINE;
4746#endif
4747		}
4748	}
4749#endif
4750
4751	/*
4752	 * Tell the backend to free resources, if this LUN has a backend.
4753	 */
4754	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4755	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4756
4757	ctl_tpc_lun_shutdown(lun);
4758	mtx_destroy(&lun->lun_lock);
4759	free(lun->lun_devid, M_CTL);
4760	if (lun->flags & CTL_LUN_MALLOCED)
4761		free(lun, M_CTL);
4762
4763	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4764		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4765			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4766		}
4767	}
4768
4769	return (0);
4770}
4771
4772static void
4773ctl_create_lun(struct ctl_be_lun *be_lun)
4774{
4775	struct ctl_softc *ctl_softc;
4776
4777	ctl_softc = control_softc;
4778
4779	/*
4780	 * ctl_alloc_lun() should handle all potential failure cases.
4781	 */
4782	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4783}
4784
4785int
4786ctl_add_lun(struct ctl_be_lun *be_lun)
4787{
4788	struct ctl_softc *ctl_softc = control_softc;
4789
4790	mtx_lock(&ctl_softc->ctl_lock);
4791	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4792	mtx_unlock(&ctl_softc->ctl_lock);
4793	wakeup(&ctl_softc->pending_lun_queue);
4794
4795	return (0);
4796}
4797
4798int
4799ctl_enable_lun(struct ctl_be_lun *be_lun)
4800{
4801	struct ctl_softc *ctl_softc;
4802	struct ctl_port *port, *nport;
4803	struct ctl_lun *lun;
4804	int retval;
4805
4806	ctl_softc = control_softc;
4807
4808	lun = (struct ctl_lun *)be_lun->ctl_lun;
4809
4810	mtx_lock(&ctl_softc->ctl_lock);
4811	mtx_lock(&lun->lun_lock);
4812	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4813		/*
4814		 * eh?  Why did we get called if the LUN is already
4815		 * enabled?
4816		 */
4817		mtx_unlock(&lun->lun_lock);
4818		mtx_unlock(&ctl_softc->ctl_lock);
4819		return (0);
4820	}
4821	lun->flags &= ~CTL_LUN_DISABLED;
4822	mtx_unlock(&lun->lun_lock);
4823
4824	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4825		nport = STAILQ_NEXT(port, links);
4826
4827		/*
4828		 * Drop the lock while we call the FETD's enable routine.
4829		 * This can lead to a callback into CTL (at least in the
4830		 * case of the internal initiator frontend.
4831		 */
4832		mtx_unlock(&ctl_softc->ctl_lock);
4833		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4834		mtx_lock(&ctl_softc->ctl_lock);
4835		if (retval != 0) {
4836			printf("%s: FETD %s port %d returned error "
4837			       "%d for lun_enable on target %ju lun %jd\n",
4838			       __func__, port->port_name, port->targ_port, retval,
4839			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4840		}
4841#if 0
4842		 else {
4843            /* NOTE:  TODO:  why does lun enable affect port status? */
4844			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4845		}
4846#endif
4847	}
4848
4849	mtx_unlock(&ctl_softc->ctl_lock);
4850
4851	return (0);
4852}
4853
4854int
4855ctl_disable_lun(struct ctl_be_lun *be_lun)
4856{
4857	struct ctl_softc *ctl_softc;
4858	struct ctl_port *port;
4859	struct ctl_lun *lun;
4860	int retval;
4861
4862	ctl_softc = control_softc;
4863
4864	lun = (struct ctl_lun *)be_lun->ctl_lun;
4865
4866	mtx_lock(&ctl_softc->ctl_lock);
4867	mtx_lock(&lun->lun_lock);
4868	if (lun->flags & CTL_LUN_DISABLED) {
4869		mtx_unlock(&lun->lun_lock);
4870		mtx_unlock(&ctl_softc->ctl_lock);
4871		return (0);
4872	}
4873	lun->flags |= CTL_LUN_DISABLED;
4874	mtx_unlock(&lun->lun_lock);
4875
4876	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4877		mtx_unlock(&ctl_softc->ctl_lock);
4878		/*
4879		 * Drop the lock before we call the frontend's disable
4880		 * routine, to avoid lock order reversals.
4881		 *
4882		 * XXX KDM what happens if the frontend list changes while
4883		 * we're traversing it?  It's unlikely, but should be handled.
4884		 */
4885		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4886					 lun->lun);
4887		mtx_lock(&ctl_softc->ctl_lock);
4888		if (retval != 0) {
4889			printf("ctl_alloc_lun: FETD %s port %d returned error "
4890			       "%d for lun_disable on target %ju lun %jd\n",
4891			       port->port_name, port->targ_port, retval,
4892			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4893		}
4894	}
4895
4896	mtx_unlock(&ctl_softc->ctl_lock);
4897
4898	return (0);
4899}
4900
4901int
4902ctl_start_lun(struct ctl_be_lun *be_lun)
4903{
4904	struct ctl_softc *ctl_softc;
4905	struct ctl_lun *lun;
4906
4907	ctl_softc = control_softc;
4908
4909	lun = (struct ctl_lun *)be_lun->ctl_lun;
4910
4911	mtx_lock(&lun->lun_lock);
4912	lun->flags &= ~CTL_LUN_STOPPED;
4913	mtx_unlock(&lun->lun_lock);
4914
4915	return (0);
4916}
4917
4918int
4919ctl_stop_lun(struct ctl_be_lun *be_lun)
4920{
4921	struct ctl_softc *ctl_softc;
4922	struct ctl_lun *lun;
4923
4924	ctl_softc = control_softc;
4925
4926	lun = (struct ctl_lun *)be_lun->ctl_lun;
4927
4928	mtx_lock(&lun->lun_lock);
4929	lun->flags |= CTL_LUN_STOPPED;
4930	mtx_unlock(&lun->lun_lock);
4931
4932	return (0);
4933}
4934
4935int
4936ctl_lun_offline(struct ctl_be_lun *be_lun)
4937{
4938	struct ctl_softc *ctl_softc;
4939	struct ctl_lun *lun;
4940
4941	ctl_softc = control_softc;
4942
4943	lun = (struct ctl_lun *)be_lun->ctl_lun;
4944
4945	mtx_lock(&lun->lun_lock);
4946	lun->flags |= CTL_LUN_OFFLINE;
4947	mtx_unlock(&lun->lun_lock);
4948
4949	return (0);
4950}
4951
4952int
4953ctl_lun_online(struct ctl_be_lun *be_lun)
4954{
4955	struct ctl_softc *ctl_softc;
4956	struct ctl_lun *lun;
4957
4958	ctl_softc = control_softc;
4959
4960	lun = (struct ctl_lun *)be_lun->ctl_lun;
4961
4962	mtx_lock(&lun->lun_lock);
4963	lun->flags &= ~CTL_LUN_OFFLINE;
4964	mtx_unlock(&lun->lun_lock);
4965
4966	return (0);
4967}
4968
4969int
4970ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4971{
4972	struct ctl_softc *ctl_softc;
4973	struct ctl_lun *lun;
4974
4975	ctl_softc = control_softc;
4976
4977	lun = (struct ctl_lun *)be_lun->ctl_lun;
4978
4979	mtx_lock(&lun->lun_lock);
4980
4981	/*
4982	 * The LUN needs to be disabled before it can be marked invalid.
4983	 */
4984	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4985		mtx_unlock(&lun->lun_lock);
4986		return (-1);
4987	}
4988	/*
4989	 * Mark the LUN invalid.
4990	 */
4991	lun->flags |= CTL_LUN_INVALID;
4992
4993	/*
4994	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4995	 * If we have something in the OOA queue, we'll free it when the
4996	 * last I/O completes.
4997	 */
4998	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4999		mtx_unlock(&lun->lun_lock);
5000		mtx_lock(&ctl_softc->ctl_lock);
5001		ctl_free_lun(lun);
5002		mtx_unlock(&ctl_softc->ctl_lock);
5003	} else
5004		mtx_unlock(&lun->lun_lock);
5005
5006	return (0);
5007}
5008
5009int
5010ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5011{
5012	struct ctl_softc *ctl_softc;
5013	struct ctl_lun *lun;
5014
5015	ctl_softc = control_softc;
5016	lun = (struct ctl_lun *)be_lun->ctl_lun;
5017
5018	mtx_lock(&lun->lun_lock);
5019	lun->flags |= CTL_LUN_INOPERABLE;
5020	mtx_unlock(&lun->lun_lock);
5021
5022	return (0);
5023}
5024
5025int
5026ctl_lun_operable(struct ctl_be_lun *be_lun)
5027{
5028	struct ctl_softc *ctl_softc;
5029	struct ctl_lun *lun;
5030
5031	ctl_softc = control_softc;
5032	lun = (struct ctl_lun *)be_lun->ctl_lun;
5033
5034	mtx_lock(&lun->lun_lock);
5035	lun->flags &= ~CTL_LUN_INOPERABLE;
5036	mtx_unlock(&lun->lun_lock);
5037
5038	return (0);
5039}
5040
5041int
5042ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
5043		   int lock)
5044{
5045	struct ctl_softc *softc;
5046	struct ctl_lun *lun;
5047	struct copan_aps_subpage *current_sp;
5048	struct ctl_page_index *page_index;
5049	int i;
5050
5051	softc = control_softc;
5052
5053	mtx_lock(&softc->ctl_lock);
5054
5055	lun = (struct ctl_lun *)be_lun->ctl_lun;
5056	mtx_lock(&lun->lun_lock);
5057
5058	page_index = NULL;
5059	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5060		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5061		     APS_PAGE_CODE)
5062			continue;
5063
5064		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
5065			continue;
5066		page_index = &lun->mode_pages.index[i];
5067	}
5068
5069	if (page_index == NULL) {
5070		mtx_unlock(&lun->lun_lock);
5071		mtx_unlock(&softc->ctl_lock);
5072		printf("%s: APS subpage not found for lun %ju!\n", __func__,
5073		       (uintmax_t)lun->lun);
5074		return (1);
5075	}
5076#if 0
5077	if ((softc->aps_locked_lun != 0)
5078	 && (softc->aps_locked_lun != lun->lun)) {
5079		printf("%s: attempt to lock LUN %llu when %llu is already "
5080		       "locked\n");
5081		mtx_unlock(&lun->lun_lock);
5082		mtx_unlock(&softc->ctl_lock);
5083		return (1);
5084	}
5085#endif
5086
5087	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5088		(page_index->page_len * CTL_PAGE_CURRENT));
5089
5090	if (lock != 0) {
5091		current_sp->lock_active = APS_LOCK_ACTIVE;
5092		softc->aps_locked_lun = lun->lun;
5093	} else {
5094		current_sp->lock_active = 0;
5095		softc->aps_locked_lun = 0;
5096	}
5097
5098
5099	/*
5100	 * If we're in HA mode, try to send the lock message to the other
5101	 * side.
5102	 */
5103	if (ctl_is_single == 0) {
5104		int isc_retval;
5105		union ctl_ha_msg lock_msg;
5106
5107		lock_msg.hdr.nexus = *nexus;
5108		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
5109		if (lock != 0)
5110			lock_msg.aps.lock_flag = 1;
5111		else
5112			lock_msg.aps.lock_flag = 0;
5113		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
5114					 sizeof(lock_msg), 0);
5115		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
5116			printf("%s: APS (lock=%d) error returned from "
5117			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
5118			mtx_unlock(&lun->lun_lock);
5119			mtx_unlock(&softc->ctl_lock);
5120			return (1);
5121		}
5122	}
5123
5124	mtx_unlock(&lun->lun_lock);
5125	mtx_unlock(&softc->ctl_lock);
5126
5127	return (0);
5128}
5129
5130void
5131ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5132{
5133	struct ctl_lun *lun;
5134	struct ctl_softc *softc;
5135	int i;
5136
5137	softc = control_softc;
5138
5139	lun = (struct ctl_lun *)be_lun->ctl_lun;
5140
5141	mtx_lock(&lun->lun_lock);
5142
5143	for (i = 0; i < CTL_MAX_INITIATORS; i++)
5144		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
5145
5146	mtx_unlock(&lun->lun_lock);
5147}
5148
5149/*
5150 * Backend "memory move is complete" callback for requests that never
5151 * make it down to say RAIDCore's configuration code.
5152 */
5153int
5154ctl_config_move_done(union ctl_io *io)
5155{
5156	int retval;
5157
5158	retval = CTL_RETVAL_COMPLETE;
5159
5160
5161	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5162	/*
5163	 * XXX KDM this shouldn't happen, but what if it does?
5164	 */
5165	if (io->io_hdr.io_type != CTL_IO_SCSI)
5166		panic("I/O type isn't CTL_IO_SCSI!");
5167
5168	if ((io->io_hdr.port_status == 0)
5169	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5170	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
5171		io->io_hdr.status = CTL_SUCCESS;
5172	else if ((io->io_hdr.port_status != 0)
5173	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
5174	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
5175		/*
5176		 * For hardware error sense keys, the sense key
5177		 * specific value is defined to be a retry count,
5178		 * but we use it to pass back an internal FETD
5179		 * error code.  XXX KDM  Hopefully the FETD is only
5180		 * using 16 bits for an error code, since that's
5181		 * all the space we have in the sks field.
5182		 */
5183		ctl_set_internal_failure(&io->scsiio,
5184					 /*sks_valid*/ 1,
5185					 /*retry_count*/
5186					 io->io_hdr.port_status);
5187		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5188			free(io->scsiio.kern_data_ptr, M_CTL);
5189		ctl_done(io);
5190		goto bailout;
5191	}
5192
5193	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
5194	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
5195	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5196		/*
5197		 * XXX KDM just assuming a single pointer here, and not a
5198		 * S/G list.  If we start using S/G lists for config data,
5199		 * we'll need to know how to clean them up here as well.
5200		 */
5201		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5202			free(io->scsiio.kern_data_ptr, M_CTL);
5203		/* Hopefully the user has already set the status... */
5204		ctl_done(io);
5205	} else {
5206		/*
5207		 * XXX KDM now we need to continue data movement.  Some
5208		 * options:
5209		 * - call ctl_scsiio() again?  We don't do this for data
5210		 *   writes, because for those at least we know ahead of
5211		 *   time where the write will go and how long it is.  For
5212		 *   config writes, though, that information is largely
5213		 *   contained within the write itself, thus we need to
5214		 *   parse out the data again.
5215		 *
5216		 * - Call some other function once the data is in?
5217		 */
5218
5219		/*
5220		 * XXX KDM call ctl_scsiio() again for now, and check flag
5221		 * bits to see whether we're allocated or not.
5222		 */
5223		retval = ctl_scsiio(&io->scsiio);
5224	}
5225bailout:
5226	return (retval);
5227}
5228
5229/*
5230 * This gets called by a backend driver when it is done with a
5231 * data_submit method.
5232 */
5233void
5234ctl_data_submit_done(union ctl_io *io)
5235{
5236	/*
5237	 * If the IO_CONT flag is set, we need to call the supplied
5238	 * function to continue processing the I/O, instead of completing
5239	 * the I/O just yet.
5240	 *
5241	 * If there is an error, though, we don't want to keep processing.
5242	 * Instead, just send status back to the initiator.
5243	 */
5244	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5245	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5246	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5247	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5248		io->scsiio.io_cont(io);
5249		return;
5250	}
5251	ctl_done(io);
5252}
5253
5254/*
5255 * This gets called by a backend driver when it is done with a
5256 * configuration write.
5257 */
5258void
5259ctl_config_write_done(union ctl_io *io)
5260{
5261	/*
5262	 * If the IO_CONT flag is set, we need to call the supplied
5263	 * function to continue processing the I/O, instead of completing
5264	 * the I/O just yet.
5265	 *
5266	 * If there is an error, though, we don't want to keep processing.
5267	 * Instead, just send status back to the initiator.
5268	 */
5269	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5270	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5271	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5272		io->scsiio.io_cont(io);
5273		return;
5274	}
5275	/*
5276	 * Since a configuration write can be done for commands that actually
5277	 * have data allocated, like write buffer, and commands that have
5278	 * no data, like start/stop unit, we need to check here.
5279	 */
5280	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5281		free(io->scsiio.kern_data_ptr, M_CTL);
5282	ctl_done(io);
5283}
5284
5285/*
5286 * SCSI release command.
5287 */
5288int
5289ctl_scsi_release(struct ctl_scsiio *ctsio)
5290{
5291	int length, longid, thirdparty_id, resv_id;
5292	struct ctl_softc *ctl_softc;
5293	struct ctl_lun *lun;
5294
5295	length = 0;
5296	resv_id = 0;
5297
5298	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5299
5300	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5301	ctl_softc = control_softc;
5302
5303	switch (ctsio->cdb[0]) {
5304	case RELEASE_10: {
5305		struct scsi_release_10 *cdb;
5306
5307		cdb = (struct scsi_release_10 *)ctsio->cdb;
5308
5309		if (cdb->byte2 & SR10_LONGID)
5310			longid = 1;
5311		else
5312			thirdparty_id = cdb->thirdparty_id;
5313
5314		resv_id = cdb->resv_id;
5315		length = scsi_2btoul(cdb->length);
5316		break;
5317	}
5318	}
5319
5320
5321	/*
5322	 * XXX KDM right now, we only support LUN reservation.  We don't
5323	 * support 3rd party reservations, or extent reservations, which
5324	 * might actually need the parameter list.  If we've gotten this
5325	 * far, we've got a LUN reservation.  Anything else got kicked out
5326	 * above.  So, according to SPC, ignore the length.
5327	 */
5328	length = 0;
5329
5330	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5331	 && (length > 0)) {
5332		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5333		ctsio->kern_data_len = length;
5334		ctsio->kern_total_len = length;
5335		ctsio->kern_data_resid = 0;
5336		ctsio->kern_rel_offset = 0;
5337		ctsio->kern_sg_entries = 0;
5338		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5339		ctsio->be_move_done = ctl_config_move_done;
5340		ctl_datamove((union ctl_io *)ctsio);
5341
5342		return (CTL_RETVAL_COMPLETE);
5343	}
5344
5345	if (length > 0)
5346		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5347
5348	mtx_lock(&lun->lun_lock);
5349
5350	/*
5351	 * According to SPC, it is not an error for an intiator to attempt
5352	 * to release a reservation on a LUN that isn't reserved, or that
5353	 * is reserved by another initiator.  The reservation can only be
5354	 * released, though, by the initiator who made it or by one of
5355	 * several reset type events.
5356	 */
5357	if (lun->flags & CTL_LUN_RESERVED) {
5358		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5359		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5360		 && (ctsio->io_hdr.nexus.targ_target.id ==
5361		     lun->rsv_nexus.targ_target.id)) {
5362			lun->flags &= ~CTL_LUN_RESERVED;
5363		}
5364	}
5365
5366	mtx_unlock(&lun->lun_lock);
5367
5368	ctsio->scsi_status = SCSI_STATUS_OK;
5369	ctsio->io_hdr.status = CTL_SUCCESS;
5370
5371	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5372		free(ctsio->kern_data_ptr, M_CTL);
5373		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5374	}
5375
5376	ctl_done((union ctl_io *)ctsio);
5377	return (CTL_RETVAL_COMPLETE);
5378}
5379
5380int
5381ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5382{
5383	int extent, thirdparty, longid;
5384	int resv_id, length;
5385	uint64_t thirdparty_id;
5386	struct ctl_softc *ctl_softc;
5387	struct ctl_lun *lun;
5388
5389	extent = 0;
5390	thirdparty = 0;
5391	longid = 0;
5392	resv_id = 0;
5393	length = 0;
5394	thirdparty_id = 0;
5395
5396	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5397
5398	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5399	ctl_softc = control_softc;
5400
5401	switch (ctsio->cdb[0]) {
5402	case RESERVE_10: {
5403		struct scsi_reserve_10 *cdb;
5404
5405		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5406
5407		if (cdb->byte2 & SR10_LONGID)
5408			longid = 1;
5409		else
5410			thirdparty_id = cdb->thirdparty_id;
5411
5412		resv_id = cdb->resv_id;
5413		length = scsi_2btoul(cdb->length);
5414		break;
5415	}
5416	}
5417
5418	/*
5419	 * XXX KDM right now, we only support LUN reservation.  We don't
5420	 * support 3rd party reservations, or extent reservations, which
5421	 * might actually need the parameter list.  If we've gotten this
5422	 * far, we've got a LUN reservation.  Anything else got kicked out
5423	 * above.  So, according to SPC, ignore the length.
5424	 */
5425	length = 0;
5426
5427	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5428	 && (length > 0)) {
5429		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5430		ctsio->kern_data_len = length;
5431		ctsio->kern_total_len = length;
5432		ctsio->kern_data_resid = 0;
5433		ctsio->kern_rel_offset = 0;
5434		ctsio->kern_sg_entries = 0;
5435		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5436		ctsio->be_move_done = ctl_config_move_done;
5437		ctl_datamove((union ctl_io *)ctsio);
5438
5439		return (CTL_RETVAL_COMPLETE);
5440	}
5441
5442	if (length > 0)
5443		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5444
5445	mtx_lock(&lun->lun_lock);
5446	if (lun->flags & CTL_LUN_RESERVED) {
5447		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5448		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5449		 || (ctsio->io_hdr.nexus.targ_target.id !=
5450		     lun->rsv_nexus.targ_target.id)) {
5451			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5452			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5453			goto bailout;
5454		}
5455	}
5456
5457	lun->flags |= CTL_LUN_RESERVED;
5458	lun->rsv_nexus = ctsio->io_hdr.nexus;
5459
5460	ctsio->scsi_status = SCSI_STATUS_OK;
5461	ctsio->io_hdr.status = CTL_SUCCESS;
5462
5463bailout:
5464	mtx_unlock(&lun->lun_lock);
5465
5466	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5467		free(ctsio->kern_data_ptr, M_CTL);
5468		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5469	}
5470
5471	ctl_done((union ctl_io *)ctsio);
5472	return (CTL_RETVAL_COMPLETE);
5473}
5474
5475int
5476ctl_start_stop(struct ctl_scsiio *ctsio)
5477{
5478	struct scsi_start_stop_unit *cdb;
5479	struct ctl_lun *lun;
5480	struct ctl_softc *ctl_softc;
5481	int retval;
5482
5483	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5484
5485	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5486	ctl_softc = control_softc;
5487	retval = 0;
5488
5489	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5490
5491	/*
5492	 * XXX KDM
5493	 * We don't support the immediate bit on a stop unit.  In order to
5494	 * do that, we would need to code up a way to know that a stop is
5495	 * pending, and hold off any new commands until it completes, one
5496	 * way or another.  Then we could accept or reject those commands
5497	 * depending on its status.  We would almost need to do the reverse
5498	 * of what we do below for an immediate start -- return the copy of
5499	 * the ctl_io to the FETD with status to send to the host (and to
5500	 * free the copy!) and then free the original I/O once the stop
5501	 * actually completes.  That way, the OOA queue mechanism can work
5502	 * to block commands that shouldn't proceed.  Another alternative
5503	 * would be to put the copy in the queue in place of the original,
5504	 * and return the original back to the caller.  That could be
5505	 * slightly safer..
5506	 */
5507	if ((cdb->byte2 & SSS_IMMED)
5508	 && ((cdb->how & SSS_START) == 0)) {
5509		ctl_set_invalid_field(ctsio,
5510				      /*sks_valid*/ 1,
5511				      /*command*/ 1,
5512				      /*field*/ 1,
5513				      /*bit_valid*/ 1,
5514				      /*bit*/ 0);
5515		ctl_done((union ctl_io *)ctsio);
5516		return (CTL_RETVAL_COMPLETE);
5517	}
5518
5519	if ((lun->flags & CTL_LUN_PR_RESERVED)
5520	 && ((cdb->how & SSS_START)==0)) {
5521		uint32_t residx;
5522
5523		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5524		if (!lun->per_res[residx].registered
5525		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5526
5527			ctl_set_reservation_conflict(ctsio);
5528			ctl_done((union ctl_io *)ctsio);
5529			return (CTL_RETVAL_COMPLETE);
5530		}
5531	}
5532
5533	/*
5534	 * If there is no backend on this device, we can't start or stop
5535	 * it.  In theory we shouldn't get any start/stop commands in the
5536	 * first place at this level if the LUN doesn't have a backend.
5537	 * That should get stopped by the command decode code.
5538	 */
5539	if (lun->backend == NULL) {
5540		ctl_set_invalid_opcode(ctsio);
5541		ctl_done((union ctl_io *)ctsio);
5542		return (CTL_RETVAL_COMPLETE);
5543	}
5544
5545	/*
5546	 * XXX KDM Copan-specific offline behavior.
5547	 * Figure out a reasonable way to port this?
5548	 */
5549#ifdef NEEDTOPORT
5550	mtx_lock(&lun->lun_lock);
5551
5552	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5553	 && (lun->flags & CTL_LUN_OFFLINE)) {
5554		/*
5555		 * If the LUN is offline, and the on/offline bit isn't set,
5556		 * reject the start or stop.  Otherwise, let it through.
5557		 */
5558		mtx_unlock(&lun->lun_lock);
5559		ctl_set_lun_not_ready(ctsio);
5560		ctl_done((union ctl_io *)ctsio);
5561	} else {
5562		mtx_unlock(&lun->lun_lock);
5563#endif /* NEEDTOPORT */
5564		/*
5565		 * This could be a start or a stop when we're online,
5566		 * or a stop/offline or start/online.  A start or stop when
5567		 * we're offline is covered in the case above.
5568		 */
5569		/*
5570		 * In the non-immediate case, we send the request to
5571		 * the backend and return status to the user when
5572		 * it is done.
5573		 *
5574		 * In the immediate case, we allocate a new ctl_io
5575		 * to hold a copy of the request, and send that to
5576		 * the backend.  We then set good status on the
5577		 * user's request and return it immediately.
5578		 */
5579		if (cdb->byte2 & SSS_IMMED) {
5580			union ctl_io *new_io;
5581
5582			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5583			if (new_io == NULL) {
5584				ctl_set_busy(ctsio);
5585				ctl_done((union ctl_io *)ctsio);
5586			} else {
5587				ctl_copy_io((union ctl_io *)ctsio,
5588					    new_io);
5589				retval = lun->backend->config_write(new_io);
5590				ctl_set_success(ctsio);
5591				ctl_done((union ctl_io *)ctsio);
5592			}
5593		} else {
5594			retval = lun->backend->config_write(
5595				(union ctl_io *)ctsio);
5596		}
5597#ifdef NEEDTOPORT
5598	}
5599#endif
5600	return (retval);
5601}
5602
5603/*
5604 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5605 * we don't really do anything with the LBA and length fields if the user
5606 * passes them in.  Instead we'll just flush out the cache for the entire
5607 * LUN.
5608 */
5609int
5610ctl_sync_cache(struct ctl_scsiio *ctsio)
5611{
5612	struct ctl_lun *lun;
5613	struct ctl_softc *ctl_softc;
5614	uint64_t starting_lba;
5615	uint32_t block_count;
5616	int retval;
5617
5618	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5619
5620	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5621	ctl_softc = control_softc;
5622	retval = 0;
5623
5624	switch (ctsio->cdb[0]) {
5625	case SYNCHRONIZE_CACHE: {
5626		struct scsi_sync_cache *cdb;
5627		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5628
5629		starting_lba = scsi_4btoul(cdb->begin_lba);
5630		block_count = scsi_2btoul(cdb->lb_count);
5631		break;
5632	}
5633	case SYNCHRONIZE_CACHE_16: {
5634		struct scsi_sync_cache_16 *cdb;
5635		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5636
5637		starting_lba = scsi_8btou64(cdb->begin_lba);
5638		block_count = scsi_4btoul(cdb->lb_count);
5639		break;
5640	}
5641	default:
5642		ctl_set_invalid_opcode(ctsio);
5643		ctl_done((union ctl_io *)ctsio);
5644		goto bailout;
5645		break; /* NOTREACHED */
5646	}
5647
5648	/*
5649	 * We check the LBA and length, but don't do anything with them.
5650	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5651	 * get flushed.  This check will just help satisfy anyone who wants
5652	 * to see an error for an out of range LBA.
5653	 */
5654	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5655		ctl_set_lba_out_of_range(ctsio);
5656		ctl_done((union ctl_io *)ctsio);
5657		goto bailout;
5658	}
5659
5660	/*
5661	 * If this LUN has no backend, we can't flush the cache anyway.
5662	 */
5663	if (lun->backend == NULL) {
5664		ctl_set_invalid_opcode(ctsio);
5665		ctl_done((union ctl_io *)ctsio);
5666		goto bailout;
5667	}
5668
5669	/*
5670	 * Check to see whether we're configured to send the SYNCHRONIZE
5671	 * CACHE command directly to the back end.
5672	 */
5673	mtx_lock(&lun->lun_lock);
5674	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5675	 && (++(lun->sync_count) >= lun->sync_interval)) {
5676		lun->sync_count = 0;
5677		mtx_unlock(&lun->lun_lock);
5678		retval = lun->backend->config_write((union ctl_io *)ctsio);
5679	} else {
5680		mtx_unlock(&lun->lun_lock);
5681		ctl_set_success(ctsio);
5682		ctl_done((union ctl_io *)ctsio);
5683	}
5684
5685bailout:
5686
5687	return (retval);
5688}
5689
5690int
5691ctl_format(struct ctl_scsiio *ctsio)
5692{
5693	struct scsi_format *cdb;
5694	struct ctl_lun *lun;
5695	struct ctl_softc *ctl_softc;
5696	int length, defect_list_len;
5697
5698	CTL_DEBUG_PRINT(("ctl_format\n"));
5699
5700	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5701	ctl_softc = control_softc;
5702
5703	cdb = (struct scsi_format *)ctsio->cdb;
5704
5705	length = 0;
5706	if (cdb->byte2 & SF_FMTDATA) {
5707		if (cdb->byte2 & SF_LONGLIST)
5708			length = sizeof(struct scsi_format_header_long);
5709		else
5710			length = sizeof(struct scsi_format_header_short);
5711	}
5712
5713	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5714	 && (length > 0)) {
5715		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5716		ctsio->kern_data_len = length;
5717		ctsio->kern_total_len = length;
5718		ctsio->kern_data_resid = 0;
5719		ctsio->kern_rel_offset = 0;
5720		ctsio->kern_sg_entries = 0;
5721		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5722		ctsio->be_move_done = ctl_config_move_done;
5723		ctl_datamove((union ctl_io *)ctsio);
5724
5725		return (CTL_RETVAL_COMPLETE);
5726	}
5727
5728	defect_list_len = 0;
5729
5730	if (cdb->byte2 & SF_FMTDATA) {
5731		if (cdb->byte2 & SF_LONGLIST) {
5732			struct scsi_format_header_long *header;
5733
5734			header = (struct scsi_format_header_long *)
5735				ctsio->kern_data_ptr;
5736
5737			defect_list_len = scsi_4btoul(header->defect_list_len);
5738			if (defect_list_len != 0) {
5739				ctl_set_invalid_field(ctsio,
5740						      /*sks_valid*/ 1,
5741						      /*command*/ 0,
5742						      /*field*/ 2,
5743						      /*bit_valid*/ 0,
5744						      /*bit*/ 0);
5745				goto bailout;
5746			}
5747		} else {
5748			struct scsi_format_header_short *header;
5749
5750			header = (struct scsi_format_header_short *)
5751				ctsio->kern_data_ptr;
5752
5753			defect_list_len = scsi_2btoul(header->defect_list_len);
5754			if (defect_list_len != 0) {
5755				ctl_set_invalid_field(ctsio,
5756						      /*sks_valid*/ 1,
5757						      /*command*/ 0,
5758						      /*field*/ 2,
5759						      /*bit_valid*/ 0,
5760						      /*bit*/ 0);
5761				goto bailout;
5762			}
5763		}
5764	}
5765
5766	/*
5767	 * The format command will clear out the "Medium format corrupted"
5768	 * status if set by the configuration code.  That status is really
5769	 * just a way to notify the host that we have lost the media, and
5770	 * get them to issue a command that will basically make them think
5771	 * they're blowing away the media.
5772	 */
5773	mtx_lock(&lun->lun_lock);
5774	lun->flags &= ~CTL_LUN_INOPERABLE;
5775	mtx_unlock(&lun->lun_lock);
5776
5777	ctsio->scsi_status = SCSI_STATUS_OK;
5778	ctsio->io_hdr.status = CTL_SUCCESS;
5779bailout:
5780
5781	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5782		free(ctsio->kern_data_ptr, M_CTL);
5783		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5784	}
5785
5786	ctl_done((union ctl_io *)ctsio);
5787	return (CTL_RETVAL_COMPLETE);
5788}
5789
5790int
5791ctl_read_buffer(struct ctl_scsiio *ctsio)
5792{
5793	struct scsi_read_buffer *cdb;
5794	struct ctl_lun *lun;
5795	int buffer_offset, len;
5796	static uint8_t descr[4];
5797	static uint8_t echo_descr[4] = { 0 };
5798
5799	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5800
5801	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5802	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5803
5804	if (lun->flags & CTL_LUN_PR_RESERVED) {
5805		uint32_t residx;
5806
5807		/*
5808		 * XXX KDM need a lock here.
5809		 */
5810		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5811		if ((lun->res_type == SPR_TYPE_EX_AC
5812		  && residx != lun->pr_res_idx)
5813		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5814		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5815		  && !lun->per_res[residx].registered)) {
5816			ctl_set_reservation_conflict(ctsio);
5817			ctl_done((union ctl_io *)ctsio);
5818			return (CTL_RETVAL_COMPLETE);
5819	        }
5820	}
5821
5822	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5823	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5824	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5825		ctl_set_invalid_field(ctsio,
5826				      /*sks_valid*/ 1,
5827				      /*command*/ 1,
5828				      /*field*/ 1,
5829				      /*bit_valid*/ 1,
5830				      /*bit*/ 4);
5831		ctl_done((union ctl_io *)ctsio);
5832		return (CTL_RETVAL_COMPLETE);
5833	}
5834
5835	len = scsi_3btoul(cdb->length);
5836	buffer_offset = scsi_3btoul(cdb->offset);
5837
5838	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5839		ctl_set_invalid_field(ctsio,
5840				      /*sks_valid*/ 1,
5841				      /*command*/ 1,
5842				      /*field*/ 6,
5843				      /*bit_valid*/ 0,
5844				      /*bit*/ 0);
5845		ctl_done((union ctl_io *)ctsio);
5846		return (CTL_RETVAL_COMPLETE);
5847	}
5848
5849	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5850		descr[0] = 0;
5851		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5852		ctsio->kern_data_ptr = descr;
5853		len = min(len, sizeof(descr));
5854	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5855		ctsio->kern_data_ptr = echo_descr;
5856		len = min(len, sizeof(echo_descr));
5857	} else
5858		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5859	ctsio->kern_data_len = len;
5860	ctsio->kern_total_len = len;
5861	ctsio->kern_data_resid = 0;
5862	ctsio->kern_rel_offset = 0;
5863	ctsio->kern_sg_entries = 0;
5864	ctsio->be_move_done = ctl_config_move_done;
5865	ctl_datamove((union ctl_io *)ctsio);
5866
5867	return (CTL_RETVAL_COMPLETE);
5868}
5869
5870int
5871ctl_write_buffer(struct ctl_scsiio *ctsio)
5872{
5873	struct scsi_write_buffer *cdb;
5874	struct ctl_lun *lun;
5875	int buffer_offset, len;
5876
5877	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5878
5879	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5880	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5881
5882	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5883		ctl_set_invalid_field(ctsio,
5884				      /*sks_valid*/ 1,
5885				      /*command*/ 1,
5886				      /*field*/ 1,
5887				      /*bit_valid*/ 1,
5888				      /*bit*/ 4);
5889		ctl_done((union ctl_io *)ctsio);
5890		return (CTL_RETVAL_COMPLETE);
5891	}
5892
5893	len = scsi_3btoul(cdb->length);
5894	buffer_offset = scsi_3btoul(cdb->offset);
5895
5896	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5897		ctl_set_invalid_field(ctsio,
5898				      /*sks_valid*/ 1,
5899				      /*command*/ 1,
5900				      /*field*/ 6,
5901				      /*bit_valid*/ 0,
5902				      /*bit*/ 0);
5903		ctl_done((union ctl_io *)ctsio);
5904		return (CTL_RETVAL_COMPLETE);
5905	}
5906
5907	/*
5908	 * If we've got a kernel request that hasn't been malloced yet,
5909	 * malloc it and tell the caller the data buffer is here.
5910	 */
5911	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5912		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5913		ctsio->kern_data_len = len;
5914		ctsio->kern_total_len = len;
5915		ctsio->kern_data_resid = 0;
5916		ctsio->kern_rel_offset = 0;
5917		ctsio->kern_sg_entries = 0;
5918		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5919		ctsio->be_move_done = ctl_config_move_done;
5920		ctl_datamove((union ctl_io *)ctsio);
5921
5922		return (CTL_RETVAL_COMPLETE);
5923	}
5924
5925	ctl_done((union ctl_io *)ctsio);
5926
5927	return (CTL_RETVAL_COMPLETE);
5928}
5929
5930int
5931ctl_write_same(struct ctl_scsiio *ctsio)
5932{
5933	struct ctl_lun *lun;
5934	struct ctl_lba_len_flags *lbalen;
5935	uint64_t lba;
5936	uint32_t num_blocks;
5937	int len, retval;
5938	uint8_t byte2;
5939
5940	retval = CTL_RETVAL_COMPLETE;
5941
5942	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5943
5944	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5945
5946	switch (ctsio->cdb[0]) {
5947	case WRITE_SAME_10: {
5948		struct scsi_write_same_10 *cdb;
5949
5950		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5951
5952		lba = scsi_4btoul(cdb->addr);
5953		num_blocks = scsi_2btoul(cdb->length);
5954		byte2 = cdb->byte2;
5955		break;
5956	}
5957	case WRITE_SAME_16: {
5958		struct scsi_write_same_16 *cdb;
5959
5960		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5961
5962		lba = scsi_8btou64(cdb->addr);
5963		num_blocks = scsi_4btoul(cdb->length);
5964		byte2 = cdb->byte2;
5965		break;
5966	}
5967	default:
5968		/*
5969		 * We got a command we don't support.  This shouldn't
5970		 * happen, commands should be filtered out above us.
5971		 */
5972		ctl_set_invalid_opcode(ctsio);
5973		ctl_done((union ctl_io *)ctsio);
5974
5975		return (CTL_RETVAL_COMPLETE);
5976		break; /* NOTREACHED */
5977	}
5978
5979	/*
5980	 * The first check is to make sure we're in bounds, the second
5981	 * check is to catch wrap-around problems.  If the lba + num blocks
5982	 * is less than the lba, then we've wrapped around and the block
5983	 * range is invalid anyway.
5984	 */
5985	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5986	 || ((lba + num_blocks) < lba)) {
5987		ctl_set_lba_out_of_range(ctsio);
5988		ctl_done((union ctl_io *)ctsio);
5989		return (CTL_RETVAL_COMPLETE);
5990	}
5991
5992	/* Zero number of blocks means "to the last logical block" */
5993	if (num_blocks == 0) {
5994		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5995			ctl_set_invalid_field(ctsio,
5996					      /*sks_valid*/ 0,
5997					      /*command*/ 1,
5998					      /*field*/ 0,
5999					      /*bit_valid*/ 0,
6000					      /*bit*/ 0);
6001			ctl_done((union ctl_io *)ctsio);
6002			return (CTL_RETVAL_COMPLETE);
6003		}
6004		num_blocks = (lun->be_lun->maxlba + 1) - lba;
6005	}
6006
6007	len = lun->be_lun->blocksize;
6008
6009	/*
6010	 * If we've got a kernel request that hasn't been malloced yet,
6011	 * malloc it and tell the caller the data buffer is here.
6012	 */
6013	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6014		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6015		ctsio->kern_data_len = len;
6016		ctsio->kern_total_len = len;
6017		ctsio->kern_data_resid = 0;
6018		ctsio->kern_rel_offset = 0;
6019		ctsio->kern_sg_entries = 0;
6020		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6021		ctsio->be_move_done = ctl_config_move_done;
6022		ctl_datamove((union ctl_io *)ctsio);
6023
6024		return (CTL_RETVAL_COMPLETE);
6025	}
6026
6027	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6028	lbalen->lba = lba;
6029	lbalen->len = num_blocks;
6030	lbalen->flags = byte2;
6031	retval = lun->backend->config_write((union ctl_io *)ctsio);
6032
6033	return (retval);
6034}
6035
6036int
6037ctl_unmap(struct ctl_scsiio *ctsio)
6038{
6039	struct ctl_lun *lun;
6040	struct scsi_unmap *cdb;
6041	struct ctl_ptr_len_flags *ptrlen;
6042	struct scsi_unmap_header *hdr;
6043	struct scsi_unmap_desc *buf, *end;
6044	uint64_t lba;
6045	uint32_t num_blocks;
6046	int len, retval;
6047	uint8_t byte2;
6048
6049	retval = CTL_RETVAL_COMPLETE;
6050
6051	CTL_DEBUG_PRINT(("ctl_unmap\n"));
6052
6053	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6054	cdb = (struct scsi_unmap *)ctsio->cdb;
6055
6056	len = scsi_2btoul(cdb->length);
6057	byte2 = cdb->byte2;
6058
6059	/*
6060	 * If we've got a kernel request that hasn't been malloced yet,
6061	 * malloc it and tell the caller the data buffer is here.
6062	 */
6063	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6064		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6065		ctsio->kern_data_len = len;
6066		ctsio->kern_total_len = len;
6067		ctsio->kern_data_resid = 0;
6068		ctsio->kern_rel_offset = 0;
6069		ctsio->kern_sg_entries = 0;
6070		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6071		ctsio->be_move_done = ctl_config_move_done;
6072		ctl_datamove((union ctl_io *)ctsio);
6073
6074		return (CTL_RETVAL_COMPLETE);
6075	}
6076
6077	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6078	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6079	if (len < sizeof (*hdr) ||
6080	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6081	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6082	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6083		ctl_set_invalid_field(ctsio,
6084				      /*sks_valid*/ 0,
6085				      /*command*/ 0,
6086				      /*field*/ 0,
6087				      /*bit_valid*/ 0,
6088				      /*bit*/ 0);
6089		ctl_done((union ctl_io *)ctsio);
6090		return (CTL_RETVAL_COMPLETE);
6091	}
6092	len = scsi_2btoul(hdr->desc_length);
6093	buf = (struct scsi_unmap_desc *)(hdr + 1);
6094	end = buf + len / sizeof(*buf);
6095
6096	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6097	ptrlen->ptr = (void *)buf;
6098	ptrlen->len = len;
6099	ptrlen->flags = byte2;
6100
6101	for (; buf < end; buf++) {
6102		lba = scsi_8btou64(buf->lba);
6103		num_blocks = scsi_4btoul(buf->length);
6104		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6105		 || ((lba + num_blocks) < lba)) {
6106			ctl_set_lba_out_of_range(ctsio);
6107			ctl_done((union ctl_io *)ctsio);
6108			return (CTL_RETVAL_COMPLETE);
6109		}
6110	}
6111
6112	retval = lun->backend->config_write((union ctl_io *)ctsio);
6113
6114	return (retval);
6115}
6116
6117/*
6118 * Note that this function currently doesn't actually do anything inside
6119 * CTL to enforce things if the DQue bit is turned on.
6120 *
6121 * Also note that this function can't be used in the default case, because
6122 * the DQue bit isn't set in the changeable mask for the control mode page
6123 * anyway.  This is just here as an example for how to implement a page
6124 * handler, and a placeholder in case we want to allow the user to turn
6125 * tagged queueing on and off.
6126 *
6127 * The D_SENSE bit handling is functional, however, and will turn
6128 * descriptor sense on and off for a given LUN.
6129 */
6130int
6131ctl_control_page_handler(struct ctl_scsiio *ctsio,
6132			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6133{
6134	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6135	struct ctl_lun *lun;
6136	struct ctl_softc *softc;
6137	int set_ua;
6138	uint32_t initidx;
6139
6140	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6141	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6142	set_ua = 0;
6143
6144	user_cp = (struct scsi_control_page *)page_ptr;
6145	current_cp = (struct scsi_control_page *)
6146		(page_index->page_data + (page_index->page_len *
6147		CTL_PAGE_CURRENT));
6148	saved_cp = (struct scsi_control_page *)
6149		(page_index->page_data + (page_index->page_len *
6150		CTL_PAGE_SAVED));
6151
6152	softc = control_softc;
6153
6154	mtx_lock(&lun->lun_lock);
6155	if (((current_cp->rlec & SCP_DSENSE) == 0)
6156	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6157		/*
6158		 * Descriptor sense is currently turned off and the user
6159		 * wants to turn it on.
6160		 */
6161		current_cp->rlec |= SCP_DSENSE;
6162		saved_cp->rlec |= SCP_DSENSE;
6163		lun->flags |= CTL_LUN_SENSE_DESC;
6164		set_ua = 1;
6165	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6166		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6167		/*
6168		 * Descriptor sense is currently turned on, and the user
6169		 * wants to turn it off.
6170		 */
6171		current_cp->rlec &= ~SCP_DSENSE;
6172		saved_cp->rlec &= ~SCP_DSENSE;
6173		lun->flags &= ~CTL_LUN_SENSE_DESC;
6174		set_ua = 1;
6175	}
6176	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6177		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6178#ifdef NEEDTOPORT
6179			csevent_log(CSC_CTL | CSC_SHELF_SW |
6180				    CTL_UNTAG_TO_UNTAG,
6181				    csevent_LogType_Trace,
6182				    csevent_Severity_Information,
6183				    csevent_AlertLevel_Green,
6184				    csevent_FRU_Firmware,
6185				    csevent_FRU_Unknown,
6186				    "Received untagged to untagged transition");
6187#endif /* NEEDTOPORT */
6188		} else {
6189#ifdef NEEDTOPORT
6190			csevent_log(CSC_CTL | CSC_SHELF_SW |
6191				    CTL_UNTAG_TO_TAG,
6192				    csevent_LogType_ConfigChange,
6193				    csevent_Severity_Information,
6194				    csevent_AlertLevel_Green,
6195				    csevent_FRU_Firmware,
6196				    csevent_FRU_Unknown,
6197				    "Received untagged to tagged "
6198				    "queueing transition");
6199#endif /* NEEDTOPORT */
6200
6201			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6202			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6203			set_ua = 1;
6204		}
6205	} else {
6206		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6207#ifdef NEEDTOPORT
6208			csevent_log(CSC_CTL | CSC_SHELF_SW |
6209				    CTL_TAG_TO_UNTAG,
6210				    csevent_LogType_ConfigChange,
6211				    csevent_Severity_Warning,
6212				    csevent_AlertLevel_Yellow,
6213				    csevent_FRU_Firmware,
6214				    csevent_FRU_Unknown,
6215				    "Received tagged queueing to untagged "
6216				    "transition");
6217#endif /* NEEDTOPORT */
6218
6219			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6220			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6221			set_ua = 1;
6222		} else {
6223#ifdef NEEDTOPORT
6224			csevent_log(CSC_CTL | CSC_SHELF_SW |
6225				    CTL_TAG_TO_TAG,
6226				    csevent_LogType_Trace,
6227				    csevent_Severity_Information,
6228				    csevent_AlertLevel_Green,
6229				    csevent_FRU_Firmware,
6230				    csevent_FRU_Unknown,
6231				    "Received tagged queueing to tagged "
6232				    "queueing transition");
6233#endif /* NEEDTOPORT */
6234		}
6235	}
6236	if (set_ua != 0) {
6237		int i;
6238		/*
6239		 * Let other initiators know that the mode
6240		 * parameters for this LUN have changed.
6241		 */
6242		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6243			if (i == initidx)
6244				continue;
6245
6246			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6247		}
6248	}
6249	mtx_unlock(&lun->lun_lock);
6250
6251	return (0);
6252}
6253
6254int
6255ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6256		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6257{
6258	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6259	struct ctl_lun *lun;
6260	int set_ua;
6261	uint32_t initidx;
6262
6263	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6264	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6265	set_ua = 0;
6266
6267	user_cp = (struct scsi_caching_page *)page_ptr;
6268	current_cp = (struct scsi_caching_page *)
6269		(page_index->page_data + (page_index->page_len *
6270		CTL_PAGE_CURRENT));
6271	saved_cp = (struct scsi_caching_page *)
6272		(page_index->page_data + (page_index->page_len *
6273		CTL_PAGE_SAVED));
6274
6275	mtx_lock(&lun->lun_lock);
6276	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6277	    (user_cp->flags1 & (SCP_WCE | SCP_RCD)))
6278		set_ua = 1;
6279	current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6280	current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6281	saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6282	saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6283	if (set_ua != 0) {
6284		int i;
6285		/*
6286		 * Let other initiators know that the mode
6287		 * parameters for this LUN have changed.
6288		 */
6289		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6290			if (i == initidx)
6291				continue;
6292
6293			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6294		}
6295	}
6296	mtx_unlock(&lun->lun_lock);
6297
6298	return (0);
6299}
6300
6301int
6302ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6303		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6304{
6305	return (0);
6306}
6307
6308int
6309ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6310			   struct ctl_page_index *page_index, int pc)
6311{
6312	struct copan_power_subpage *page;
6313
6314	page = (struct copan_power_subpage *)page_index->page_data +
6315		(page_index->page_len * pc);
6316
6317	switch (pc) {
6318	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6319		/*
6320		 * We don't update the changable bits for this page.
6321		 */
6322		break;
6323	case SMS_PAGE_CTRL_CURRENT >> 6:
6324	case SMS_PAGE_CTRL_DEFAULT >> 6:
6325	case SMS_PAGE_CTRL_SAVED >> 6:
6326#ifdef NEEDTOPORT
6327		ctl_update_power_subpage(page);
6328#endif
6329		break;
6330	default:
6331#ifdef NEEDTOPORT
6332		EPRINT(0, "Invalid PC %d!!", pc);
6333#endif
6334		break;
6335	}
6336	return (0);
6337}
6338
6339
6340int
6341ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6342		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6343{
6344	struct copan_aps_subpage *user_sp;
6345	struct copan_aps_subpage *current_sp;
6346	union ctl_modepage_info *modepage_info;
6347	struct ctl_softc *softc;
6348	struct ctl_lun *lun;
6349	int retval;
6350
6351	retval = CTL_RETVAL_COMPLETE;
6352	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6353		     (page_index->page_len * CTL_PAGE_CURRENT));
6354	softc = control_softc;
6355	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6356
6357	user_sp = (struct copan_aps_subpage *)page_ptr;
6358
6359	modepage_info = (union ctl_modepage_info *)
6360		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6361
6362	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6363	modepage_info->header.subpage = page_index->subpage;
6364	modepage_info->aps.lock_active = user_sp->lock_active;
6365
6366	mtx_lock(&softc->ctl_lock);
6367
6368	/*
6369	 * If there is a request to lock the LUN and another LUN is locked
6370	 * this is an error. If the requested LUN is already locked ignore
6371	 * the request. If no LUN is locked attempt to lock it.
6372	 * if there is a request to unlock the LUN and the LUN is currently
6373	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6374	 * if another LUN is locked or no LUN is locked.
6375	 */
6376	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6377		if (softc->aps_locked_lun == lun->lun) {
6378			/*
6379			 * This LUN is already locked, so we're done.
6380			 */
6381			retval = CTL_RETVAL_COMPLETE;
6382		} else if (softc->aps_locked_lun == 0) {
6383			/*
6384			 * No one has the lock, pass the request to the
6385			 * backend.
6386			 */
6387			retval = lun->backend->config_write(
6388				(union ctl_io *)ctsio);
6389		} else {
6390			/*
6391			 * Someone else has the lock, throw out the request.
6392			 */
6393			ctl_set_already_locked(ctsio);
6394			free(ctsio->kern_data_ptr, M_CTL);
6395			ctl_done((union ctl_io *)ctsio);
6396
6397			/*
6398			 * Set the return value so that ctl_do_mode_select()
6399			 * won't try to complete the command.  We already
6400			 * completed it here.
6401			 */
6402			retval = CTL_RETVAL_ERROR;
6403		}
6404	} else if (softc->aps_locked_lun == lun->lun) {
6405		/*
6406		 * This LUN is locked, so pass the unlock request to the
6407		 * backend.
6408		 */
6409		retval = lun->backend->config_write((union ctl_io *)ctsio);
6410	}
6411	mtx_unlock(&softc->ctl_lock);
6412
6413	return (retval);
6414}
6415
6416int
6417ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6418				struct ctl_page_index *page_index,
6419				uint8_t *page_ptr)
6420{
6421	uint8_t *c;
6422	int i;
6423
6424	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6425	ctl_time_io_secs =
6426		(c[0] << 8) |
6427		(c[1] << 0) |
6428		0;
6429	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6430	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6431	printf("page data:");
6432	for (i=0; i<8; i++)
6433		printf(" %.2x",page_ptr[i]);
6434	printf("\n");
6435	return (0);
6436}
6437
6438int
6439ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6440			       struct ctl_page_index *page_index,
6441			       int pc)
6442{
6443	struct copan_debugconf_subpage *page;
6444
6445	page = (struct copan_debugconf_subpage *)page_index->page_data +
6446		(page_index->page_len * pc);
6447
6448	switch (pc) {
6449	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6450	case SMS_PAGE_CTRL_DEFAULT >> 6:
6451	case SMS_PAGE_CTRL_SAVED >> 6:
6452		/*
6453		 * We don't update the changable or default bits for this page.
6454		 */
6455		break;
6456	case SMS_PAGE_CTRL_CURRENT >> 6:
6457		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6458		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6459		break;
6460	default:
6461#ifdef NEEDTOPORT
6462		EPRINT(0, "Invalid PC %d!!", pc);
6463#endif /* NEEDTOPORT */
6464		break;
6465	}
6466	return (0);
6467}
6468
6469
6470static int
6471ctl_do_mode_select(union ctl_io *io)
6472{
6473	struct scsi_mode_page_header *page_header;
6474	struct ctl_page_index *page_index;
6475	struct ctl_scsiio *ctsio;
6476	int control_dev, page_len;
6477	int page_len_offset, page_len_size;
6478	union ctl_modepage_info *modepage_info;
6479	struct ctl_lun *lun;
6480	int *len_left, *len_used;
6481	int retval, i;
6482
6483	ctsio = &io->scsiio;
6484	page_index = NULL;
6485	page_len = 0;
6486	retval = CTL_RETVAL_COMPLETE;
6487
6488	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6489
6490	if (lun->be_lun->lun_type != T_DIRECT)
6491		control_dev = 1;
6492	else
6493		control_dev = 0;
6494
6495	modepage_info = (union ctl_modepage_info *)
6496		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6497	len_left = &modepage_info->header.len_left;
6498	len_used = &modepage_info->header.len_used;
6499
6500do_next_page:
6501
6502	page_header = (struct scsi_mode_page_header *)
6503		(ctsio->kern_data_ptr + *len_used);
6504
6505	if (*len_left == 0) {
6506		free(ctsio->kern_data_ptr, M_CTL);
6507		ctl_set_success(ctsio);
6508		ctl_done((union ctl_io *)ctsio);
6509		return (CTL_RETVAL_COMPLETE);
6510	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6511
6512		free(ctsio->kern_data_ptr, M_CTL);
6513		ctl_set_param_len_error(ctsio);
6514		ctl_done((union ctl_io *)ctsio);
6515		return (CTL_RETVAL_COMPLETE);
6516
6517	} else if ((page_header->page_code & SMPH_SPF)
6518		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6519
6520		free(ctsio->kern_data_ptr, M_CTL);
6521		ctl_set_param_len_error(ctsio);
6522		ctl_done((union ctl_io *)ctsio);
6523		return (CTL_RETVAL_COMPLETE);
6524	}
6525
6526
6527	/*
6528	 * XXX KDM should we do something with the block descriptor?
6529	 */
6530	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6531
6532		if ((control_dev != 0)
6533		 && (lun->mode_pages.index[i].page_flags &
6534		     CTL_PAGE_FLAG_DISK_ONLY))
6535			continue;
6536
6537		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6538		    (page_header->page_code & SMPH_PC_MASK))
6539			continue;
6540
6541		/*
6542		 * If neither page has a subpage code, then we've got a
6543		 * match.
6544		 */
6545		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6546		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6547			page_index = &lun->mode_pages.index[i];
6548			page_len = page_header->page_length;
6549			break;
6550		}
6551
6552		/*
6553		 * If both pages have subpages, then the subpage numbers
6554		 * have to match.
6555		 */
6556		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6557		  && (page_header->page_code & SMPH_SPF)) {
6558			struct scsi_mode_page_header_sp *sph;
6559
6560			sph = (struct scsi_mode_page_header_sp *)page_header;
6561
6562			if (lun->mode_pages.index[i].subpage ==
6563			    sph->subpage) {
6564				page_index = &lun->mode_pages.index[i];
6565				page_len = scsi_2btoul(sph->page_length);
6566				break;
6567			}
6568		}
6569	}
6570
6571	/*
6572	 * If we couldn't find the page, or if we don't have a mode select
6573	 * handler for it, send back an error to the user.
6574	 */
6575	if ((page_index == NULL)
6576	 || (page_index->select_handler == NULL)) {
6577		ctl_set_invalid_field(ctsio,
6578				      /*sks_valid*/ 1,
6579				      /*command*/ 0,
6580				      /*field*/ *len_used,
6581				      /*bit_valid*/ 0,
6582				      /*bit*/ 0);
6583		free(ctsio->kern_data_ptr, M_CTL);
6584		ctl_done((union ctl_io *)ctsio);
6585		return (CTL_RETVAL_COMPLETE);
6586	}
6587
6588	if (page_index->page_code & SMPH_SPF) {
6589		page_len_offset = 2;
6590		page_len_size = 2;
6591	} else {
6592		page_len_size = 1;
6593		page_len_offset = 1;
6594	}
6595
6596	/*
6597	 * If the length the initiator gives us isn't the one we specify in
6598	 * the mode page header, or if they didn't specify enough data in
6599	 * the CDB to avoid truncating this page, kick out the request.
6600	 */
6601	if ((page_len != (page_index->page_len - page_len_offset -
6602			  page_len_size))
6603	 || (*len_left < page_index->page_len)) {
6604
6605
6606		ctl_set_invalid_field(ctsio,
6607				      /*sks_valid*/ 1,
6608				      /*command*/ 0,
6609				      /*field*/ *len_used + page_len_offset,
6610				      /*bit_valid*/ 0,
6611				      /*bit*/ 0);
6612		free(ctsio->kern_data_ptr, M_CTL);
6613		ctl_done((union ctl_io *)ctsio);
6614		return (CTL_RETVAL_COMPLETE);
6615	}
6616
6617	/*
6618	 * Run through the mode page, checking to make sure that the bits
6619	 * the user changed are actually legal for him to change.
6620	 */
6621	for (i = 0; i < page_index->page_len; i++) {
6622		uint8_t *user_byte, *change_mask, *current_byte;
6623		int bad_bit;
6624		int j;
6625
6626		user_byte = (uint8_t *)page_header + i;
6627		change_mask = page_index->page_data +
6628			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6629		current_byte = page_index->page_data +
6630			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6631
6632		/*
6633		 * Check to see whether the user set any bits in this byte
6634		 * that he is not allowed to set.
6635		 */
6636		if ((*user_byte & ~(*change_mask)) ==
6637		    (*current_byte & ~(*change_mask)))
6638			continue;
6639
6640		/*
6641		 * Go through bit by bit to determine which one is illegal.
6642		 */
6643		bad_bit = 0;
6644		for (j = 7; j >= 0; j--) {
6645			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6646			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6647				bad_bit = i;
6648				break;
6649			}
6650		}
6651		ctl_set_invalid_field(ctsio,
6652				      /*sks_valid*/ 1,
6653				      /*command*/ 0,
6654				      /*field*/ *len_used + i,
6655				      /*bit_valid*/ 1,
6656				      /*bit*/ bad_bit);
6657		free(ctsio->kern_data_ptr, M_CTL);
6658		ctl_done((union ctl_io *)ctsio);
6659		return (CTL_RETVAL_COMPLETE);
6660	}
6661
6662	/*
6663	 * Decrement these before we call the page handler, since we may
6664	 * end up getting called back one way or another before the handler
6665	 * returns to this context.
6666	 */
6667	*len_left -= page_index->page_len;
6668	*len_used += page_index->page_len;
6669
6670	retval = page_index->select_handler(ctsio, page_index,
6671					    (uint8_t *)page_header);
6672
6673	/*
6674	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6675	 * wait until this queued command completes to finish processing
6676	 * the mode page.  If it returns anything other than
6677	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6678	 * already set the sense information, freed the data pointer, and
6679	 * completed the io for us.
6680	 */
6681	if (retval != CTL_RETVAL_COMPLETE)
6682		goto bailout_no_done;
6683
6684	/*
6685	 * If the initiator sent us more than one page, parse the next one.
6686	 */
6687	if (*len_left > 0)
6688		goto do_next_page;
6689
6690	ctl_set_success(ctsio);
6691	free(ctsio->kern_data_ptr, M_CTL);
6692	ctl_done((union ctl_io *)ctsio);
6693
6694bailout_no_done:
6695
6696	return (CTL_RETVAL_COMPLETE);
6697
6698}
6699
6700int
6701ctl_mode_select(struct ctl_scsiio *ctsio)
6702{
6703	int param_len, pf, sp;
6704	int header_size, bd_len;
6705	int len_left, len_used;
6706	struct ctl_page_index *page_index;
6707	struct ctl_lun *lun;
6708	int control_dev, page_len;
6709	union ctl_modepage_info *modepage_info;
6710	int retval;
6711
6712	pf = 0;
6713	sp = 0;
6714	page_len = 0;
6715	len_used = 0;
6716	len_left = 0;
6717	retval = 0;
6718	bd_len = 0;
6719	page_index = NULL;
6720
6721	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6722
6723	if (lun->be_lun->lun_type != T_DIRECT)
6724		control_dev = 1;
6725	else
6726		control_dev = 0;
6727
6728	switch (ctsio->cdb[0]) {
6729	case MODE_SELECT_6: {
6730		struct scsi_mode_select_6 *cdb;
6731
6732		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6733
6734		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6735		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6736
6737		param_len = cdb->length;
6738		header_size = sizeof(struct scsi_mode_header_6);
6739		break;
6740	}
6741	case MODE_SELECT_10: {
6742		struct scsi_mode_select_10 *cdb;
6743
6744		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6745
6746		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6747		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6748
6749		param_len = scsi_2btoul(cdb->length);
6750		header_size = sizeof(struct scsi_mode_header_10);
6751		break;
6752	}
6753	default:
6754		ctl_set_invalid_opcode(ctsio);
6755		ctl_done((union ctl_io *)ctsio);
6756		return (CTL_RETVAL_COMPLETE);
6757		break; /* NOTREACHED */
6758	}
6759
6760	/*
6761	 * From SPC-3:
6762	 * "A parameter list length of zero indicates that the Data-Out Buffer
6763	 * shall be empty. This condition shall not be considered as an error."
6764	 */
6765	if (param_len == 0) {
6766		ctl_set_success(ctsio);
6767		ctl_done((union ctl_io *)ctsio);
6768		return (CTL_RETVAL_COMPLETE);
6769	}
6770
6771	/*
6772	 * Since we'll hit this the first time through, prior to
6773	 * allocation, we don't need to free a data buffer here.
6774	 */
6775	if (param_len < header_size) {
6776		ctl_set_param_len_error(ctsio);
6777		ctl_done((union ctl_io *)ctsio);
6778		return (CTL_RETVAL_COMPLETE);
6779	}
6780
6781	/*
6782	 * Allocate the data buffer and grab the user's data.  In theory,
6783	 * we shouldn't have to sanity check the parameter list length here
6784	 * because the maximum size is 64K.  We should be able to malloc
6785	 * that much without too many problems.
6786	 */
6787	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6788		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6789		ctsio->kern_data_len = param_len;
6790		ctsio->kern_total_len = param_len;
6791		ctsio->kern_data_resid = 0;
6792		ctsio->kern_rel_offset = 0;
6793		ctsio->kern_sg_entries = 0;
6794		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6795		ctsio->be_move_done = ctl_config_move_done;
6796		ctl_datamove((union ctl_io *)ctsio);
6797
6798		return (CTL_RETVAL_COMPLETE);
6799	}
6800
6801	switch (ctsio->cdb[0]) {
6802	case MODE_SELECT_6: {
6803		struct scsi_mode_header_6 *mh6;
6804
6805		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6806		bd_len = mh6->blk_desc_len;
6807		break;
6808	}
6809	case MODE_SELECT_10: {
6810		struct scsi_mode_header_10 *mh10;
6811
6812		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6813		bd_len = scsi_2btoul(mh10->blk_desc_len);
6814		break;
6815	}
6816	default:
6817		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6818		break;
6819	}
6820
6821	if (param_len < (header_size + bd_len)) {
6822		free(ctsio->kern_data_ptr, M_CTL);
6823		ctl_set_param_len_error(ctsio);
6824		ctl_done((union ctl_io *)ctsio);
6825		return (CTL_RETVAL_COMPLETE);
6826	}
6827
6828	/*
6829	 * Set the IO_CONT flag, so that if this I/O gets passed to
6830	 * ctl_config_write_done(), it'll get passed back to
6831	 * ctl_do_mode_select() for further processing, or completion if
6832	 * we're all done.
6833	 */
6834	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6835	ctsio->io_cont = ctl_do_mode_select;
6836
6837	modepage_info = (union ctl_modepage_info *)
6838		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6839
6840	memset(modepage_info, 0, sizeof(*modepage_info));
6841
6842	len_left = param_len - header_size - bd_len;
6843	len_used = header_size + bd_len;
6844
6845	modepage_info->header.len_left = len_left;
6846	modepage_info->header.len_used = len_used;
6847
6848	return (ctl_do_mode_select((union ctl_io *)ctsio));
6849}
6850
6851int
6852ctl_mode_sense(struct ctl_scsiio *ctsio)
6853{
6854	struct ctl_lun *lun;
6855	int pc, page_code, dbd, llba, subpage;
6856	int alloc_len, page_len, header_len, total_len;
6857	struct scsi_mode_block_descr *block_desc;
6858	struct ctl_page_index *page_index;
6859	int control_dev;
6860
6861	dbd = 0;
6862	llba = 0;
6863	block_desc = NULL;
6864	page_index = NULL;
6865
6866	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6867
6868	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6869
6870	if (lun->be_lun->lun_type != T_DIRECT)
6871		control_dev = 1;
6872	else
6873		control_dev = 0;
6874
6875	if (lun->flags & CTL_LUN_PR_RESERVED) {
6876		uint32_t residx;
6877
6878		/*
6879		 * XXX KDM need a lock here.
6880		 */
6881		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6882		if ((lun->res_type == SPR_TYPE_EX_AC
6883		  && residx != lun->pr_res_idx)
6884		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6885		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6886		  && !lun->per_res[residx].registered)) {
6887			ctl_set_reservation_conflict(ctsio);
6888			ctl_done((union ctl_io *)ctsio);
6889			return (CTL_RETVAL_COMPLETE);
6890		}
6891	}
6892
6893	switch (ctsio->cdb[0]) {
6894	case MODE_SENSE_6: {
6895		struct scsi_mode_sense_6 *cdb;
6896
6897		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6898
6899		header_len = sizeof(struct scsi_mode_hdr_6);
6900		if (cdb->byte2 & SMS_DBD)
6901			dbd = 1;
6902		else
6903			header_len += sizeof(struct scsi_mode_block_descr);
6904
6905		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6906		page_code = cdb->page & SMS_PAGE_CODE;
6907		subpage = cdb->subpage;
6908		alloc_len = cdb->length;
6909		break;
6910	}
6911	case MODE_SENSE_10: {
6912		struct scsi_mode_sense_10 *cdb;
6913
6914		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6915
6916		header_len = sizeof(struct scsi_mode_hdr_10);
6917
6918		if (cdb->byte2 & SMS_DBD)
6919			dbd = 1;
6920		else
6921			header_len += sizeof(struct scsi_mode_block_descr);
6922		if (cdb->byte2 & SMS10_LLBAA)
6923			llba = 1;
6924		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6925		page_code = cdb->page & SMS_PAGE_CODE;
6926		subpage = cdb->subpage;
6927		alloc_len = scsi_2btoul(cdb->length);
6928		break;
6929	}
6930	default:
6931		ctl_set_invalid_opcode(ctsio);
6932		ctl_done((union ctl_io *)ctsio);
6933		return (CTL_RETVAL_COMPLETE);
6934		break; /* NOTREACHED */
6935	}
6936
6937	/*
6938	 * We have to make a first pass through to calculate the size of
6939	 * the pages that match the user's query.  Then we allocate enough
6940	 * memory to hold it, and actually copy the data into the buffer.
6941	 */
6942	switch (page_code) {
6943	case SMS_ALL_PAGES_PAGE: {
6944		int i;
6945
6946		page_len = 0;
6947
6948		/*
6949		 * At the moment, values other than 0 and 0xff here are
6950		 * reserved according to SPC-3.
6951		 */
6952		if ((subpage != SMS_SUBPAGE_PAGE_0)
6953		 && (subpage != SMS_SUBPAGE_ALL)) {
6954			ctl_set_invalid_field(ctsio,
6955					      /*sks_valid*/ 1,
6956					      /*command*/ 1,
6957					      /*field*/ 3,
6958					      /*bit_valid*/ 0,
6959					      /*bit*/ 0);
6960			ctl_done((union ctl_io *)ctsio);
6961			return (CTL_RETVAL_COMPLETE);
6962		}
6963
6964		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6965			if ((control_dev != 0)
6966			 && (lun->mode_pages.index[i].page_flags &
6967			     CTL_PAGE_FLAG_DISK_ONLY))
6968				continue;
6969
6970			/*
6971			 * We don't use this subpage if the user didn't
6972			 * request all subpages.
6973			 */
6974			if ((lun->mode_pages.index[i].subpage != 0)
6975			 && (subpage == SMS_SUBPAGE_PAGE_0))
6976				continue;
6977
6978#if 0
6979			printf("found page %#x len %d\n",
6980			       lun->mode_pages.index[i].page_code &
6981			       SMPH_PC_MASK,
6982			       lun->mode_pages.index[i].page_len);
6983#endif
6984			page_len += lun->mode_pages.index[i].page_len;
6985		}
6986		break;
6987	}
6988	default: {
6989		int i;
6990
6991		page_len = 0;
6992
6993		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6994			/* Look for the right page code */
6995			if ((lun->mode_pages.index[i].page_code &
6996			     SMPH_PC_MASK) != page_code)
6997				continue;
6998
6999			/* Look for the right subpage or the subpage wildcard*/
7000			if ((lun->mode_pages.index[i].subpage != subpage)
7001			 && (subpage != SMS_SUBPAGE_ALL))
7002				continue;
7003
7004			/* Make sure the page is supported for this dev type */
7005			if ((control_dev != 0)
7006			 && (lun->mode_pages.index[i].page_flags &
7007			     CTL_PAGE_FLAG_DISK_ONLY))
7008				continue;
7009
7010#if 0
7011			printf("found page %#x len %d\n",
7012			       lun->mode_pages.index[i].page_code &
7013			       SMPH_PC_MASK,
7014			       lun->mode_pages.index[i].page_len);
7015#endif
7016
7017			page_len += lun->mode_pages.index[i].page_len;
7018		}
7019
7020		if (page_len == 0) {
7021			ctl_set_invalid_field(ctsio,
7022					      /*sks_valid*/ 1,
7023					      /*command*/ 1,
7024					      /*field*/ 2,
7025					      /*bit_valid*/ 1,
7026					      /*bit*/ 5);
7027			ctl_done((union ctl_io *)ctsio);
7028			return (CTL_RETVAL_COMPLETE);
7029		}
7030		break;
7031	}
7032	}
7033
7034	total_len = header_len + page_len;
7035#if 0
7036	printf("header_len = %d, page_len = %d, total_len = %d\n",
7037	       header_len, page_len, total_len);
7038#endif
7039
7040	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7041	ctsio->kern_sg_entries = 0;
7042	ctsio->kern_data_resid = 0;
7043	ctsio->kern_rel_offset = 0;
7044	if (total_len < alloc_len) {
7045		ctsio->residual = alloc_len - total_len;
7046		ctsio->kern_data_len = total_len;
7047		ctsio->kern_total_len = total_len;
7048	} else {
7049		ctsio->residual = 0;
7050		ctsio->kern_data_len = alloc_len;
7051		ctsio->kern_total_len = alloc_len;
7052	}
7053
7054	switch (ctsio->cdb[0]) {
7055	case MODE_SENSE_6: {
7056		struct scsi_mode_hdr_6 *header;
7057
7058		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
7059
7060		header->datalen = ctl_min(total_len - 1, 254);
7061		if (control_dev == 0)
7062			header->dev_specific = 0x10; /* DPOFUA */
7063		if (dbd)
7064			header->block_descr_len = 0;
7065		else
7066			header->block_descr_len =
7067				sizeof(struct scsi_mode_block_descr);
7068		block_desc = (struct scsi_mode_block_descr *)&header[1];
7069		break;
7070	}
7071	case MODE_SENSE_10: {
7072		struct scsi_mode_hdr_10 *header;
7073		int datalen;
7074
7075		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
7076
7077		datalen = ctl_min(total_len - 2, 65533);
7078		scsi_ulto2b(datalen, header->datalen);
7079		if (control_dev == 0)
7080			header->dev_specific = 0x10; /* DPOFUA */
7081		if (dbd)
7082			scsi_ulto2b(0, header->block_descr_len);
7083		else
7084			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
7085				    header->block_descr_len);
7086		block_desc = (struct scsi_mode_block_descr *)&header[1];
7087		break;
7088	}
7089	default:
7090		panic("invalid CDB type %#x", ctsio->cdb[0]);
7091		break; /* NOTREACHED */
7092	}
7093
7094	/*
7095	 * If we've got a disk, use its blocksize in the block
7096	 * descriptor.  Otherwise, just set it to 0.
7097	 */
7098	if (dbd == 0) {
7099		if (control_dev != 0)
7100			scsi_ulto3b(lun->be_lun->blocksize,
7101				    block_desc->block_len);
7102		else
7103			scsi_ulto3b(0, block_desc->block_len);
7104	}
7105
7106	switch (page_code) {
7107	case SMS_ALL_PAGES_PAGE: {
7108		int i, data_used;
7109
7110		data_used = header_len;
7111		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7112			struct ctl_page_index *page_index;
7113
7114			page_index = &lun->mode_pages.index[i];
7115
7116			if ((control_dev != 0)
7117			 && (page_index->page_flags &
7118			    CTL_PAGE_FLAG_DISK_ONLY))
7119				continue;
7120
7121			/*
7122			 * We don't use this subpage if the user didn't
7123			 * request all subpages.  We already checked (above)
7124			 * to make sure the user only specified a subpage
7125			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7126			 */
7127			if ((page_index->subpage != 0)
7128			 && (subpage == SMS_SUBPAGE_PAGE_0))
7129				continue;
7130
7131			/*
7132			 * Call the handler, if it exists, to update the
7133			 * page to the latest values.
7134			 */
7135			if (page_index->sense_handler != NULL)
7136				page_index->sense_handler(ctsio, page_index,pc);
7137
7138			memcpy(ctsio->kern_data_ptr + data_used,
7139			       page_index->page_data +
7140			       (page_index->page_len * pc),
7141			       page_index->page_len);
7142			data_used += page_index->page_len;
7143		}
7144		break;
7145	}
7146	default: {
7147		int i, data_used;
7148
7149		data_used = header_len;
7150
7151		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7152			struct ctl_page_index *page_index;
7153
7154			page_index = &lun->mode_pages.index[i];
7155
7156			/* Look for the right page code */
7157			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7158				continue;
7159
7160			/* Look for the right subpage or the subpage wildcard*/
7161			if ((page_index->subpage != subpage)
7162			 && (subpage != SMS_SUBPAGE_ALL))
7163				continue;
7164
7165			/* Make sure the page is supported for this dev type */
7166			if ((control_dev != 0)
7167			 && (page_index->page_flags &
7168			     CTL_PAGE_FLAG_DISK_ONLY))
7169				continue;
7170
7171			/*
7172			 * Call the handler, if it exists, to update the
7173			 * page to the latest values.
7174			 */
7175			if (page_index->sense_handler != NULL)
7176				page_index->sense_handler(ctsio, page_index,pc);
7177
7178			memcpy(ctsio->kern_data_ptr + data_used,
7179			       page_index->page_data +
7180			       (page_index->page_len * pc),
7181			       page_index->page_len);
7182			data_used += page_index->page_len;
7183		}
7184		break;
7185	}
7186	}
7187
7188	ctsio->scsi_status = SCSI_STATUS_OK;
7189
7190	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7191	ctsio->be_move_done = ctl_config_move_done;
7192	ctl_datamove((union ctl_io *)ctsio);
7193
7194	return (CTL_RETVAL_COMPLETE);
7195}
7196
7197int
7198ctl_read_capacity(struct ctl_scsiio *ctsio)
7199{
7200	struct scsi_read_capacity *cdb;
7201	struct scsi_read_capacity_data *data;
7202	struct ctl_lun *lun;
7203	uint32_t lba;
7204
7205	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7206
7207	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7208
7209	lba = scsi_4btoul(cdb->addr);
7210	if (((cdb->pmi & SRC_PMI) == 0)
7211	 && (lba != 0)) {
7212		ctl_set_invalid_field(/*ctsio*/ ctsio,
7213				      /*sks_valid*/ 1,
7214				      /*command*/ 1,
7215				      /*field*/ 2,
7216				      /*bit_valid*/ 0,
7217				      /*bit*/ 0);
7218		ctl_done((union ctl_io *)ctsio);
7219		return (CTL_RETVAL_COMPLETE);
7220	}
7221
7222	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7223
7224	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7225	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7226	ctsio->residual = 0;
7227	ctsio->kern_data_len = sizeof(*data);
7228	ctsio->kern_total_len = sizeof(*data);
7229	ctsio->kern_data_resid = 0;
7230	ctsio->kern_rel_offset = 0;
7231	ctsio->kern_sg_entries = 0;
7232
7233	/*
7234	 * If the maximum LBA is greater than 0xfffffffe, the user must
7235	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7236	 * serivce action set.
7237	 */
7238	if (lun->be_lun->maxlba > 0xfffffffe)
7239		scsi_ulto4b(0xffffffff, data->addr);
7240	else
7241		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7242
7243	/*
7244	 * XXX KDM this may not be 512 bytes...
7245	 */
7246	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7247
7248	ctsio->scsi_status = SCSI_STATUS_OK;
7249
7250	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7251	ctsio->be_move_done = ctl_config_move_done;
7252	ctl_datamove((union ctl_io *)ctsio);
7253
7254	return (CTL_RETVAL_COMPLETE);
7255}
7256
7257int
7258ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7259{
7260	struct scsi_read_capacity_16 *cdb;
7261	struct scsi_read_capacity_data_long *data;
7262	struct ctl_lun *lun;
7263	uint64_t lba;
7264	uint32_t alloc_len;
7265
7266	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7267
7268	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7269
7270	alloc_len = scsi_4btoul(cdb->alloc_len);
7271	lba = scsi_8btou64(cdb->addr);
7272
7273	if ((cdb->reladr & SRC16_PMI)
7274	 && (lba != 0)) {
7275		ctl_set_invalid_field(/*ctsio*/ ctsio,
7276				      /*sks_valid*/ 1,
7277				      /*command*/ 1,
7278				      /*field*/ 2,
7279				      /*bit_valid*/ 0,
7280				      /*bit*/ 0);
7281		ctl_done((union ctl_io *)ctsio);
7282		return (CTL_RETVAL_COMPLETE);
7283	}
7284
7285	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7286
7287	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7288	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7289
7290	if (sizeof(*data) < alloc_len) {
7291		ctsio->residual = alloc_len - sizeof(*data);
7292		ctsio->kern_data_len = sizeof(*data);
7293		ctsio->kern_total_len = sizeof(*data);
7294	} else {
7295		ctsio->residual = 0;
7296		ctsio->kern_data_len = alloc_len;
7297		ctsio->kern_total_len = alloc_len;
7298	}
7299	ctsio->kern_data_resid = 0;
7300	ctsio->kern_rel_offset = 0;
7301	ctsio->kern_sg_entries = 0;
7302
7303	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7304	/* XXX KDM this may not be 512 bytes... */
7305	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7306	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7307	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7308	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7309		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7310
7311	ctsio->scsi_status = SCSI_STATUS_OK;
7312
7313	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7314	ctsio->be_move_done = ctl_config_move_done;
7315	ctl_datamove((union ctl_io *)ctsio);
7316
7317	return (CTL_RETVAL_COMPLETE);
7318}
7319
7320int
7321ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7322{
7323	struct scsi_maintenance_in *cdb;
7324	int retval;
7325	int alloc_len, ext, total_len = 0, g, p, pc, pg;
7326	int num_target_port_groups, num_target_ports, single;
7327	struct ctl_lun *lun;
7328	struct ctl_softc *softc;
7329	struct ctl_port *port;
7330	struct scsi_target_group_data *rtg_ptr;
7331	struct scsi_target_group_data_extended *rtg_ext_ptr;
7332	struct scsi_target_port_group_descriptor *tpg_desc;
7333
7334	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7335
7336	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7337	softc = control_softc;
7338	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7339
7340	retval = CTL_RETVAL_COMPLETE;
7341
7342	switch (cdb->byte2 & STG_PDF_MASK) {
7343	case STG_PDF_LENGTH:
7344		ext = 0;
7345		break;
7346	case STG_PDF_EXTENDED:
7347		ext = 1;
7348		break;
7349	default:
7350		ctl_set_invalid_field(/*ctsio*/ ctsio,
7351				      /*sks_valid*/ 1,
7352				      /*command*/ 1,
7353				      /*field*/ 2,
7354				      /*bit_valid*/ 1,
7355				      /*bit*/ 5);
7356		ctl_done((union ctl_io *)ctsio);
7357		return(retval);
7358	}
7359
7360	single = ctl_is_single;
7361	if (single)
7362		num_target_port_groups = 1;
7363	else
7364		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7365	num_target_ports = 0;
7366	mtx_lock(&softc->ctl_lock);
7367	STAILQ_FOREACH(port, &softc->port_list, links) {
7368		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7369			continue;
7370		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7371			continue;
7372		num_target_ports++;
7373	}
7374	mtx_unlock(&softc->ctl_lock);
7375
7376	if (ext)
7377		total_len = sizeof(struct scsi_target_group_data_extended);
7378	else
7379		total_len = sizeof(struct scsi_target_group_data);
7380	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7381		num_target_port_groups +
7382	    sizeof(struct scsi_target_port_descriptor) *
7383		num_target_ports * num_target_port_groups;
7384
7385	alloc_len = scsi_4btoul(cdb->length);
7386
7387	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7388
7389	ctsio->kern_sg_entries = 0;
7390
7391	if (total_len < alloc_len) {
7392		ctsio->residual = alloc_len - total_len;
7393		ctsio->kern_data_len = total_len;
7394		ctsio->kern_total_len = total_len;
7395	} else {
7396		ctsio->residual = 0;
7397		ctsio->kern_data_len = alloc_len;
7398		ctsio->kern_total_len = alloc_len;
7399	}
7400	ctsio->kern_data_resid = 0;
7401	ctsio->kern_rel_offset = 0;
7402
7403	if (ext) {
7404		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7405		    ctsio->kern_data_ptr;
7406		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7407		rtg_ext_ptr->format_type = 0x10;
7408		rtg_ext_ptr->implicit_transition_time = 0;
7409		tpg_desc = &rtg_ext_ptr->groups[0];
7410	} else {
7411		rtg_ptr = (struct scsi_target_group_data *)
7412		    ctsio->kern_data_ptr;
7413		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7414		tpg_desc = &rtg_ptr->groups[0];
7415	}
7416
7417	pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS;
7418	mtx_lock(&softc->ctl_lock);
7419	for (g = 0; g < num_target_port_groups; g++) {
7420		if (g == pg)
7421			tpg_desc->pref_state = TPG_PRIMARY |
7422			    TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7423		else
7424			tpg_desc->pref_state =
7425			    TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7426		tpg_desc->support = TPG_AO_SUP;
7427		if (!single)
7428			tpg_desc->support |= TPG_AN_SUP;
7429		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7430		tpg_desc->status = TPG_IMPLICIT;
7431		pc = 0;
7432		STAILQ_FOREACH(port, &softc->port_list, links) {
7433			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7434				continue;
7435			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7436			    CTL_MAX_LUNS)
7437				continue;
7438			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7439			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7440			    relative_target_port_identifier);
7441			pc++;
7442		}
7443		tpg_desc->target_port_count = pc;
7444		tpg_desc = (struct scsi_target_port_group_descriptor *)
7445		    &tpg_desc->descriptors[pc];
7446	}
7447	mtx_unlock(&softc->ctl_lock);
7448
7449	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7450	ctsio->be_move_done = ctl_config_move_done;
7451
7452	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7453			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7454			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7455			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7456			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7457
7458	ctl_datamove((union ctl_io *)ctsio);
7459	return(retval);
7460}
7461
7462int
7463ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7464{
7465	struct ctl_lun *lun;
7466	struct scsi_report_supported_opcodes *cdb;
7467	const struct ctl_cmd_entry *entry, *sentry;
7468	struct scsi_report_supported_opcodes_all *all;
7469	struct scsi_report_supported_opcodes_descr *descr;
7470	struct scsi_report_supported_opcodes_one *one;
7471	int retval;
7472	int alloc_len, total_len;
7473	int opcode, service_action, i, j, num;
7474
7475	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7476
7477	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7478	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7479
7480	retval = CTL_RETVAL_COMPLETE;
7481
7482	opcode = cdb->requested_opcode;
7483	service_action = scsi_2btoul(cdb->requested_service_action);
7484	switch (cdb->options & RSO_OPTIONS_MASK) {
7485	case RSO_OPTIONS_ALL:
7486		num = 0;
7487		for (i = 0; i < 256; i++) {
7488			entry = &ctl_cmd_table[i];
7489			if (entry->flags & CTL_CMD_FLAG_SA5) {
7490				for (j = 0; j < 32; j++) {
7491					sentry = &((const struct ctl_cmd_entry *)
7492					    entry->execute)[j];
7493					if (ctl_cmd_applicable(
7494					    lun->be_lun->lun_type, sentry))
7495						num++;
7496				}
7497			} else {
7498				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7499				    entry))
7500					num++;
7501			}
7502		}
7503		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7504		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7505		break;
7506	case RSO_OPTIONS_OC:
7507		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7508			ctl_set_invalid_field(/*ctsio*/ ctsio,
7509					      /*sks_valid*/ 1,
7510					      /*command*/ 1,
7511					      /*field*/ 2,
7512					      /*bit_valid*/ 1,
7513					      /*bit*/ 2);
7514			ctl_done((union ctl_io *)ctsio);
7515			return (CTL_RETVAL_COMPLETE);
7516		}
7517		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7518		break;
7519	case RSO_OPTIONS_OC_SA:
7520		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7521		    service_action >= 32) {
7522			ctl_set_invalid_field(/*ctsio*/ ctsio,
7523					      /*sks_valid*/ 1,
7524					      /*command*/ 1,
7525					      /*field*/ 2,
7526					      /*bit_valid*/ 1,
7527					      /*bit*/ 2);
7528			ctl_done((union ctl_io *)ctsio);
7529			return (CTL_RETVAL_COMPLETE);
7530		}
7531		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7532		break;
7533	default:
7534		ctl_set_invalid_field(/*ctsio*/ ctsio,
7535				      /*sks_valid*/ 1,
7536				      /*command*/ 1,
7537				      /*field*/ 2,
7538				      /*bit_valid*/ 1,
7539				      /*bit*/ 2);
7540		ctl_done((union ctl_io *)ctsio);
7541		return (CTL_RETVAL_COMPLETE);
7542	}
7543
7544	alloc_len = scsi_4btoul(cdb->length);
7545
7546	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7547
7548	ctsio->kern_sg_entries = 0;
7549
7550	if (total_len < alloc_len) {
7551		ctsio->residual = alloc_len - total_len;
7552		ctsio->kern_data_len = total_len;
7553		ctsio->kern_total_len = total_len;
7554	} else {
7555		ctsio->residual = 0;
7556		ctsio->kern_data_len = alloc_len;
7557		ctsio->kern_total_len = alloc_len;
7558	}
7559	ctsio->kern_data_resid = 0;
7560	ctsio->kern_rel_offset = 0;
7561
7562	switch (cdb->options & RSO_OPTIONS_MASK) {
7563	case RSO_OPTIONS_ALL:
7564		all = (struct scsi_report_supported_opcodes_all *)
7565		    ctsio->kern_data_ptr;
7566		num = 0;
7567		for (i = 0; i < 256; i++) {
7568			entry = &ctl_cmd_table[i];
7569			if (entry->flags & CTL_CMD_FLAG_SA5) {
7570				for (j = 0; j < 32; j++) {
7571					sentry = &((const struct ctl_cmd_entry *)
7572					    entry->execute)[j];
7573					if (!ctl_cmd_applicable(
7574					    lun->be_lun->lun_type, sentry))
7575						continue;
7576					descr = &all->descr[num++];
7577					descr->opcode = i;
7578					scsi_ulto2b(j, descr->service_action);
7579					descr->flags = RSO_SERVACTV;
7580					scsi_ulto2b(sentry->length,
7581					    descr->cdb_length);
7582				}
7583			} else {
7584				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7585				    entry))
7586					continue;
7587				descr = &all->descr[num++];
7588				descr->opcode = i;
7589				scsi_ulto2b(0, descr->service_action);
7590				descr->flags = 0;
7591				scsi_ulto2b(entry->length, descr->cdb_length);
7592			}
7593		}
7594		scsi_ulto4b(
7595		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7596		    all->length);
7597		break;
7598	case RSO_OPTIONS_OC:
7599		one = (struct scsi_report_supported_opcodes_one *)
7600		    ctsio->kern_data_ptr;
7601		entry = &ctl_cmd_table[opcode];
7602		goto fill_one;
7603	case RSO_OPTIONS_OC_SA:
7604		one = (struct scsi_report_supported_opcodes_one *)
7605		    ctsio->kern_data_ptr;
7606		entry = &ctl_cmd_table[opcode];
7607		entry = &((const struct ctl_cmd_entry *)
7608		    entry->execute)[service_action];
7609fill_one:
7610		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7611			one->support = 3;
7612			scsi_ulto2b(entry->length, one->cdb_length);
7613			one->cdb_usage[0] = opcode;
7614			memcpy(&one->cdb_usage[1], entry->usage,
7615			    entry->length - 1);
7616		} else
7617			one->support = 1;
7618		break;
7619	}
7620
7621	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7622	ctsio->be_move_done = ctl_config_move_done;
7623
7624	ctl_datamove((union ctl_io *)ctsio);
7625	return(retval);
7626}
7627
7628int
7629ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7630{
7631	struct ctl_lun *lun;
7632	struct scsi_report_supported_tmf *cdb;
7633	struct scsi_report_supported_tmf_data *data;
7634	int retval;
7635	int alloc_len, total_len;
7636
7637	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7638
7639	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7640	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7641
7642	retval = CTL_RETVAL_COMPLETE;
7643
7644	total_len = sizeof(struct scsi_report_supported_tmf_data);
7645	alloc_len = scsi_4btoul(cdb->length);
7646
7647	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7648
7649	ctsio->kern_sg_entries = 0;
7650
7651	if (total_len < alloc_len) {
7652		ctsio->residual = alloc_len - total_len;
7653		ctsio->kern_data_len = total_len;
7654		ctsio->kern_total_len = total_len;
7655	} else {
7656		ctsio->residual = 0;
7657		ctsio->kern_data_len = alloc_len;
7658		ctsio->kern_total_len = alloc_len;
7659	}
7660	ctsio->kern_data_resid = 0;
7661	ctsio->kern_rel_offset = 0;
7662
7663	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7664	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7665	data->byte2 |= RST_ITNRS;
7666
7667	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7668	ctsio->be_move_done = ctl_config_move_done;
7669
7670	ctl_datamove((union ctl_io *)ctsio);
7671	return (retval);
7672}
7673
7674int
7675ctl_report_timestamp(struct ctl_scsiio *ctsio)
7676{
7677	struct ctl_lun *lun;
7678	struct scsi_report_timestamp *cdb;
7679	struct scsi_report_timestamp_data *data;
7680	struct timeval tv;
7681	int64_t timestamp;
7682	int retval;
7683	int alloc_len, total_len;
7684
7685	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7686
7687	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7688	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7689
7690	retval = CTL_RETVAL_COMPLETE;
7691
7692	total_len = sizeof(struct scsi_report_timestamp_data);
7693	alloc_len = scsi_4btoul(cdb->length);
7694
7695	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7696
7697	ctsio->kern_sg_entries = 0;
7698
7699	if (total_len < alloc_len) {
7700		ctsio->residual = alloc_len - total_len;
7701		ctsio->kern_data_len = total_len;
7702		ctsio->kern_total_len = total_len;
7703	} else {
7704		ctsio->residual = 0;
7705		ctsio->kern_data_len = alloc_len;
7706		ctsio->kern_total_len = alloc_len;
7707	}
7708	ctsio->kern_data_resid = 0;
7709	ctsio->kern_rel_offset = 0;
7710
7711	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7712	scsi_ulto2b(sizeof(*data) - 2, data->length);
7713	data->origin = RTS_ORIG_OUTSIDE;
7714	getmicrotime(&tv);
7715	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7716	scsi_ulto4b(timestamp >> 16, data->timestamp);
7717	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7718
7719	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7720	ctsio->be_move_done = ctl_config_move_done;
7721
7722	ctl_datamove((union ctl_io *)ctsio);
7723	return (retval);
7724}
7725
7726int
7727ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7728{
7729	struct scsi_per_res_in *cdb;
7730	int alloc_len, total_len = 0;
7731	/* struct scsi_per_res_in_rsrv in_data; */
7732	struct ctl_lun *lun;
7733	struct ctl_softc *softc;
7734
7735	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7736
7737	softc = control_softc;
7738
7739	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7740
7741	alloc_len = scsi_2btoul(cdb->length);
7742
7743	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7744
7745retry:
7746	mtx_lock(&lun->lun_lock);
7747	switch (cdb->action) {
7748	case SPRI_RK: /* read keys */
7749		total_len = sizeof(struct scsi_per_res_in_keys) +
7750			lun->pr_key_count *
7751			sizeof(struct scsi_per_res_key);
7752		break;
7753	case SPRI_RR: /* read reservation */
7754		if (lun->flags & CTL_LUN_PR_RESERVED)
7755			total_len = sizeof(struct scsi_per_res_in_rsrv);
7756		else
7757			total_len = sizeof(struct scsi_per_res_in_header);
7758		break;
7759	case SPRI_RC: /* report capabilities */
7760		total_len = sizeof(struct scsi_per_res_cap);
7761		break;
7762	case SPRI_RS: /* read full status */
7763		total_len = sizeof(struct scsi_per_res_in_header) +
7764		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7765		    lun->pr_key_count;
7766		break;
7767	default:
7768		panic("Invalid PR type %x", cdb->action);
7769	}
7770	mtx_unlock(&lun->lun_lock);
7771
7772	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7773
7774	if (total_len < alloc_len) {
7775		ctsio->residual = alloc_len - total_len;
7776		ctsio->kern_data_len = total_len;
7777		ctsio->kern_total_len = total_len;
7778	} else {
7779		ctsio->residual = 0;
7780		ctsio->kern_data_len = alloc_len;
7781		ctsio->kern_total_len = alloc_len;
7782	}
7783
7784	ctsio->kern_data_resid = 0;
7785	ctsio->kern_rel_offset = 0;
7786	ctsio->kern_sg_entries = 0;
7787
7788	mtx_lock(&lun->lun_lock);
7789	switch (cdb->action) {
7790	case SPRI_RK: { // read keys
7791        struct scsi_per_res_in_keys *res_keys;
7792		int i, key_count;
7793
7794		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7795
7796		/*
7797		 * We had to drop the lock to allocate our buffer, which
7798		 * leaves time for someone to come in with another
7799		 * persistent reservation.  (That is unlikely, though,
7800		 * since this should be the only persistent reservation
7801		 * command active right now.)
7802		 */
7803		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7804		    (lun->pr_key_count *
7805		     sizeof(struct scsi_per_res_key)))){
7806			mtx_unlock(&lun->lun_lock);
7807			free(ctsio->kern_data_ptr, M_CTL);
7808			printf("%s: reservation length changed, retrying\n",
7809			       __func__);
7810			goto retry;
7811		}
7812
7813		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7814
7815		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7816			     lun->pr_key_count, res_keys->header.length);
7817
7818		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7819			if (!lun->per_res[i].registered)
7820				continue;
7821
7822			/*
7823			 * We used lun->pr_key_count to calculate the
7824			 * size to allocate.  If it turns out the number of
7825			 * initiators with the registered flag set is
7826			 * larger than that (i.e. they haven't been kept in
7827			 * sync), we've got a problem.
7828			 */
7829			if (key_count >= lun->pr_key_count) {
7830#ifdef NEEDTOPORT
7831				csevent_log(CSC_CTL | CSC_SHELF_SW |
7832					    CTL_PR_ERROR,
7833					    csevent_LogType_Fault,
7834					    csevent_AlertLevel_Yellow,
7835					    csevent_FRU_ShelfController,
7836					    csevent_FRU_Firmware,
7837				        csevent_FRU_Unknown,
7838					    "registered keys %d >= key "
7839					    "count %d", key_count,
7840					    lun->pr_key_count);
7841#endif
7842				key_count++;
7843				continue;
7844			}
7845			memcpy(res_keys->keys[key_count].key,
7846			       lun->per_res[i].res_key.key,
7847			       ctl_min(sizeof(res_keys->keys[key_count].key),
7848			       sizeof(lun->per_res[i].res_key)));
7849			key_count++;
7850		}
7851		break;
7852	}
7853	case SPRI_RR: { // read reservation
7854		struct scsi_per_res_in_rsrv *res;
7855		int tmp_len, header_only;
7856
7857		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7858
7859		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7860
7861		if (lun->flags & CTL_LUN_PR_RESERVED)
7862		{
7863			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7864			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7865				    res->header.length);
7866			header_only = 0;
7867		} else {
7868			tmp_len = sizeof(struct scsi_per_res_in_header);
7869			scsi_ulto4b(0, res->header.length);
7870			header_only = 1;
7871		}
7872
7873		/*
7874		 * We had to drop the lock to allocate our buffer, which
7875		 * leaves time for someone to come in with another
7876		 * persistent reservation.  (That is unlikely, though,
7877		 * since this should be the only persistent reservation
7878		 * command active right now.)
7879		 */
7880		if (tmp_len != total_len) {
7881			mtx_unlock(&lun->lun_lock);
7882			free(ctsio->kern_data_ptr, M_CTL);
7883			printf("%s: reservation status changed, retrying\n",
7884			       __func__);
7885			goto retry;
7886		}
7887
7888		/*
7889		 * No reservation held, so we're done.
7890		 */
7891		if (header_only != 0)
7892			break;
7893
7894		/*
7895		 * If the registration is an All Registrants type, the key
7896		 * is 0, since it doesn't really matter.
7897		 */
7898		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7899			memcpy(res->data.reservation,
7900			       &lun->per_res[lun->pr_res_idx].res_key,
7901			       sizeof(struct scsi_per_res_key));
7902		}
7903		res->data.scopetype = lun->res_type;
7904		break;
7905	}
7906	case SPRI_RC:     //report capabilities
7907	{
7908		struct scsi_per_res_cap *res_cap;
7909		uint16_t type_mask;
7910
7911		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7912		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7913		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7914		type_mask = SPRI_TM_WR_EX_AR |
7915			    SPRI_TM_EX_AC_RO |
7916			    SPRI_TM_WR_EX_RO |
7917			    SPRI_TM_EX_AC |
7918			    SPRI_TM_WR_EX |
7919			    SPRI_TM_EX_AC_AR;
7920		scsi_ulto2b(type_mask, res_cap->type_mask);
7921		break;
7922	}
7923	case SPRI_RS: { // read full status
7924		struct scsi_per_res_in_full *res_status;
7925		struct scsi_per_res_in_full_desc *res_desc;
7926		struct ctl_port *port;
7927		int i, len;
7928
7929		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7930
7931		/*
7932		 * We had to drop the lock to allocate our buffer, which
7933		 * leaves time for someone to come in with another
7934		 * persistent reservation.  (That is unlikely, though,
7935		 * since this should be the only persistent reservation
7936		 * command active right now.)
7937		 */
7938		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7939		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7940		     lun->pr_key_count)){
7941			mtx_unlock(&lun->lun_lock);
7942			free(ctsio->kern_data_ptr, M_CTL);
7943			printf("%s: reservation length changed, retrying\n",
7944			       __func__);
7945			goto retry;
7946		}
7947
7948		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7949
7950		res_desc = &res_status->desc[0];
7951		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7952			if (!lun->per_res[i].registered)
7953				continue;
7954
7955			memcpy(&res_desc->res_key, &lun->per_res[i].res_key.key,
7956			    sizeof(res_desc->res_key));
7957			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7958			    (lun->pr_res_idx == i ||
7959			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7960				res_desc->flags = SPRI_FULL_R_HOLDER;
7961				res_desc->scopetype = lun->res_type;
7962			}
7963			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7964			    res_desc->rel_trgt_port_id);
7965			len = 0;
7966			port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7967			if (port != NULL)
7968				len = ctl_create_iid(port,
7969				    i % CTL_MAX_INIT_PER_PORT,
7970				    res_desc->transport_id);
7971			scsi_ulto4b(len, res_desc->additional_length);
7972			res_desc = (struct scsi_per_res_in_full_desc *)
7973			    &res_desc->transport_id[len];
7974		}
7975		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7976		    res_status->header.length);
7977		break;
7978	}
7979	default:
7980		/*
7981		 * This is a bug, because we just checked for this above,
7982		 * and should have returned an error.
7983		 */
7984		panic("Invalid PR type %x", cdb->action);
7985		break; /* NOTREACHED */
7986	}
7987	mtx_unlock(&lun->lun_lock);
7988
7989	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7990	ctsio->be_move_done = ctl_config_move_done;
7991
7992	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7993			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7994			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7995			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7996			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7997
7998	ctl_datamove((union ctl_io *)ctsio);
7999
8000	return (CTL_RETVAL_COMPLETE);
8001}
8002
8003/*
8004 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8005 * it should return.
8006 */
8007static int
8008ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8009		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8010		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8011		struct scsi_per_res_out_parms* param)
8012{
8013	union ctl_ha_msg persis_io;
8014	int retval, i;
8015	int isc_retval;
8016
8017	retval = 0;
8018
8019	mtx_lock(&lun->lun_lock);
8020	if (sa_res_key == 0) {
8021		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8022			/* validate scope and type */
8023			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8024			     SPR_LU_SCOPE) {
8025				mtx_unlock(&lun->lun_lock);
8026				ctl_set_invalid_field(/*ctsio*/ ctsio,
8027						      /*sks_valid*/ 1,
8028						      /*command*/ 1,
8029						      /*field*/ 2,
8030						      /*bit_valid*/ 1,
8031						      /*bit*/ 4);
8032				ctl_done((union ctl_io *)ctsio);
8033				return (1);
8034			}
8035
8036		        if (type>8 || type==2 || type==4 || type==0) {
8037				mtx_unlock(&lun->lun_lock);
8038				ctl_set_invalid_field(/*ctsio*/ ctsio,
8039       	           				      /*sks_valid*/ 1,
8040						      /*command*/ 1,
8041						      /*field*/ 2,
8042						      /*bit_valid*/ 1,
8043						      /*bit*/ 0);
8044				ctl_done((union ctl_io *)ctsio);
8045				return (1);
8046		        }
8047
8048			/* temporarily unregister this nexus */
8049			lun->per_res[residx].registered = 0;
8050
8051			/*
8052			 * Unregister everybody else and build UA for
8053			 * them
8054			 */
8055			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8056				if (lun->per_res[i].registered == 0)
8057					continue;
8058
8059				if (!persis_offset
8060				 && i <CTL_MAX_INITIATORS)
8061					lun->pending_ua[i] |=
8062						CTL_UA_REG_PREEMPT;
8063				else if (persis_offset
8064				      && i >= persis_offset)
8065					lun->pending_ua[i-persis_offset] |=
8066						CTL_UA_REG_PREEMPT;
8067				lun->per_res[i].registered = 0;
8068				memset(&lun->per_res[i].res_key, 0,
8069				       sizeof(struct scsi_per_res_key));
8070			}
8071			lun->per_res[residx].registered = 1;
8072			lun->pr_key_count = 1;
8073			lun->res_type = type;
8074			if (lun->res_type != SPR_TYPE_WR_EX_AR
8075			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8076				lun->pr_res_idx = residx;
8077
8078			/* send msg to other side */
8079			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8080			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8081			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8082			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8083			persis_io.pr.pr_info.res_type = type;
8084			memcpy(persis_io.pr.pr_info.sa_res_key,
8085			       param->serv_act_res_key,
8086			       sizeof(param->serv_act_res_key));
8087			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8088			     &persis_io, sizeof(persis_io), 0)) >
8089			     CTL_HA_STATUS_SUCCESS) {
8090				printf("CTL:Persis Out error returned "
8091				       "from ctl_ha_msg_send %d\n",
8092				       isc_retval);
8093			}
8094		} else {
8095			/* not all registrants */
8096			mtx_unlock(&lun->lun_lock);
8097			free(ctsio->kern_data_ptr, M_CTL);
8098			ctl_set_invalid_field(ctsio,
8099					      /*sks_valid*/ 1,
8100					      /*command*/ 0,
8101					      /*field*/ 8,
8102					      /*bit_valid*/ 0,
8103					      /*bit*/ 0);
8104			ctl_done((union ctl_io *)ctsio);
8105			return (1);
8106		}
8107	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8108		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8109		int found = 0;
8110
8111		if (res_key == sa_res_key) {
8112			/* special case */
8113			/*
8114			 * The spec implies this is not good but doesn't
8115			 * say what to do. There are two choices either
8116			 * generate a res conflict or check condition
8117			 * with illegal field in parameter data. Since
8118			 * that is what is done when the sa_res_key is
8119			 * zero I'll take that approach since this has
8120			 * to do with the sa_res_key.
8121			 */
8122			mtx_unlock(&lun->lun_lock);
8123			free(ctsio->kern_data_ptr, M_CTL);
8124			ctl_set_invalid_field(ctsio,
8125					      /*sks_valid*/ 1,
8126					      /*command*/ 0,
8127					      /*field*/ 8,
8128					      /*bit_valid*/ 0,
8129					      /*bit*/ 0);
8130			ctl_done((union ctl_io *)ctsio);
8131			return (1);
8132		}
8133
8134		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8135			if (lun->per_res[i].registered
8136			 && memcmp(param->serv_act_res_key,
8137			    lun->per_res[i].res_key.key,
8138			    sizeof(struct scsi_per_res_key)) != 0)
8139				continue;
8140
8141			found = 1;
8142			lun->per_res[i].registered = 0;
8143			memset(&lun->per_res[i].res_key, 0,
8144			       sizeof(struct scsi_per_res_key));
8145			lun->pr_key_count--;
8146
8147			if (!persis_offset && i < CTL_MAX_INITIATORS)
8148				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8149			else if (persis_offset && i >= persis_offset)
8150				lun->pending_ua[i-persis_offset] |=
8151					CTL_UA_REG_PREEMPT;
8152		}
8153		if (!found) {
8154			mtx_unlock(&lun->lun_lock);
8155			free(ctsio->kern_data_ptr, M_CTL);
8156			ctl_set_reservation_conflict(ctsio);
8157			ctl_done((union ctl_io *)ctsio);
8158			return (CTL_RETVAL_COMPLETE);
8159		}
8160		/* send msg to other side */
8161		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8162		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8163		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8164		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8165		persis_io.pr.pr_info.res_type = type;
8166		memcpy(persis_io.pr.pr_info.sa_res_key,
8167		       param->serv_act_res_key,
8168		       sizeof(param->serv_act_res_key));
8169		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8170		     &persis_io, sizeof(persis_io), 0)) >
8171		     CTL_HA_STATUS_SUCCESS) {
8172			printf("CTL:Persis Out error returned from "
8173			       "ctl_ha_msg_send %d\n", isc_retval);
8174		}
8175	} else {
8176		/* Reserved but not all registrants */
8177		/* sa_res_key is res holder */
8178		if (memcmp(param->serv_act_res_key,
8179                   lun->per_res[lun->pr_res_idx].res_key.key,
8180                   sizeof(struct scsi_per_res_key)) == 0) {
8181			/* validate scope and type */
8182			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8183			     SPR_LU_SCOPE) {
8184				mtx_unlock(&lun->lun_lock);
8185				ctl_set_invalid_field(/*ctsio*/ ctsio,
8186						      /*sks_valid*/ 1,
8187						      /*command*/ 1,
8188						      /*field*/ 2,
8189						      /*bit_valid*/ 1,
8190						      /*bit*/ 4);
8191				ctl_done((union ctl_io *)ctsio);
8192				return (1);
8193			}
8194
8195			if (type>8 || type==2 || type==4 || type==0) {
8196				mtx_unlock(&lun->lun_lock);
8197				ctl_set_invalid_field(/*ctsio*/ ctsio,
8198						      /*sks_valid*/ 1,
8199						      /*command*/ 1,
8200						      /*field*/ 2,
8201						      /*bit_valid*/ 1,
8202						      /*bit*/ 0);
8203				ctl_done((union ctl_io *)ctsio);
8204				return (1);
8205			}
8206
8207			/*
8208			 * Do the following:
8209			 * if sa_res_key != res_key remove all
8210			 * registrants w/sa_res_key and generate UA
8211			 * for these registrants(Registrations
8212			 * Preempted) if it wasn't an exclusive
8213			 * reservation generate UA(Reservations
8214			 * Preempted) for all other registered nexuses
8215			 * if the type has changed. Establish the new
8216			 * reservation and holder. If res_key and
8217			 * sa_res_key are the same do the above
8218			 * except don't unregister the res holder.
8219			 */
8220
8221			/*
8222			 * Temporarily unregister so it won't get
8223			 * removed or UA generated
8224			 */
8225			lun->per_res[residx].registered = 0;
8226			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8227				if (lun->per_res[i].registered == 0)
8228					continue;
8229
8230				if (memcmp(param->serv_act_res_key,
8231				    lun->per_res[i].res_key.key,
8232				    sizeof(struct scsi_per_res_key)) == 0) {
8233					lun->per_res[i].registered = 0;
8234					memset(&lun->per_res[i].res_key,
8235					       0,
8236					       sizeof(struct scsi_per_res_key));
8237					lun->pr_key_count--;
8238
8239					if (!persis_offset
8240					 && i < CTL_MAX_INITIATORS)
8241						lun->pending_ua[i] |=
8242							CTL_UA_REG_PREEMPT;
8243					else if (persis_offset
8244					      && i >= persis_offset)
8245						lun->pending_ua[i-persis_offset] |=
8246						  CTL_UA_REG_PREEMPT;
8247				} else if (type != lun->res_type
8248					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8249					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8250						if (!persis_offset
8251						 && i < CTL_MAX_INITIATORS)
8252							lun->pending_ua[i] |=
8253							CTL_UA_RES_RELEASE;
8254						else if (persis_offset
8255						      && i >= persis_offset)
8256							lun->pending_ua[
8257							i-persis_offset] |=
8258							CTL_UA_RES_RELEASE;
8259				}
8260			}
8261			lun->per_res[residx].registered = 1;
8262			lun->res_type = type;
8263			if (lun->res_type != SPR_TYPE_WR_EX_AR
8264			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8265				lun->pr_res_idx = residx;
8266			else
8267				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8268
8269			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8270			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8271			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8272			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8273			persis_io.pr.pr_info.res_type = type;
8274			memcpy(persis_io.pr.pr_info.sa_res_key,
8275			       param->serv_act_res_key,
8276			       sizeof(param->serv_act_res_key));
8277			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8278			     &persis_io, sizeof(persis_io), 0)) >
8279			     CTL_HA_STATUS_SUCCESS) {
8280				printf("CTL:Persis Out error returned "
8281				       "from ctl_ha_msg_send %d\n",
8282				       isc_retval);
8283			}
8284		} else {
8285			/*
8286			 * sa_res_key is not the res holder just
8287			 * remove registrants
8288			 */
8289			int found=0;
8290
8291			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8292				if (memcmp(param->serv_act_res_key,
8293				    lun->per_res[i].res_key.key,
8294				    sizeof(struct scsi_per_res_key)) != 0)
8295					continue;
8296
8297				found = 1;
8298				lun->per_res[i].registered = 0;
8299				memset(&lun->per_res[i].res_key, 0,
8300				       sizeof(struct scsi_per_res_key));
8301				lun->pr_key_count--;
8302
8303				if (!persis_offset
8304				 && i < CTL_MAX_INITIATORS)
8305					lun->pending_ua[i] |=
8306						CTL_UA_REG_PREEMPT;
8307				else if (persis_offset
8308				      && i >= persis_offset)
8309					lun->pending_ua[i-persis_offset] |=
8310						CTL_UA_REG_PREEMPT;
8311			}
8312
8313			if (!found) {
8314				mtx_unlock(&lun->lun_lock);
8315				free(ctsio->kern_data_ptr, M_CTL);
8316				ctl_set_reservation_conflict(ctsio);
8317				ctl_done((union ctl_io *)ctsio);
8318		        	return (1);
8319			}
8320			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8321			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8322			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8323			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8324			persis_io.pr.pr_info.res_type = type;
8325			memcpy(persis_io.pr.pr_info.sa_res_key,
8326			       param->serv_act_res_key,
8327			       sizeof(param->serv_act_res_key));
8328			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8329			     &persis_io, sizeof(persis_io), 0)) >
8330			     CTL_HA_STATUS_SUCCESS) {
8331				printf("CTL:Persis Out error returned "
8332				       "from ctl_ha_msg_send %d\n",
8333				isc_retval);
8334			}
8335		}
8336	}
8337
8338	lun->PRGeneration++;
8339	mtx_unlock(&lun->lun_lock);
8340
8341	return (retval);
8342}
8343
8344static void
8345ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8346{
8347	int i;
8348
8349	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8350	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8351	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
8352		   msg->pr.pr_info.sa_res_key,
8353		   sizeof(struct scsi_per_res_key)) != 0) {
8354		uint64_t sa_res_key;
8355		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8356
8357		if (sa_res_key == 0) {
8358			/* temporarily unregister this nexus */
8359			lun->per_res[msg->pr.pr_info.residx].registered = 0;
8360
8361			/*
8362			 * Unregister everybody else and build UA for
8363			 * them
8364			 */
8365			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8366				if (lun->per_res[i].registered == 0)
8367					continue;
8368
8369				if (!persis_offset
8370				 && i < CTL_MAX_INITIATORS)
8371					lun->pending_ua[i] |=
8372						CTL_UA_REG_PREEMPT;
8373				else if (persis_offset && i >= persis_offset)
8374					lun->pending_ua[i - persis_offset] |=
8375						CTL_UA_REG_PREEMPT;
8376				lun->per_res[i].registered = 0;
8377				memset(&lun->per_res[i].res_key, 0,
8378				       sizeof(struct scsi_per_res_key));
8379			}
8380
8381			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8382			lun->pr_key_count = 1;
8383			lun->res_type = msg->pr.pr_info.res_type;
8384			if (lun->res_type != SPR_TYPE_WR_EX_AR
8385			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8386				lun->pr_res_idx = msg->pr.pr_info.residx;
8387		} else {
8388		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8389				if (memcmp(msg->pr.pr_info.sa_res_key,
8390		                   lun->per_res[i].res_key.key,
8391		                   sizeof(struct scsi_per_res_key)) != 0)
8392					continue;
8393
8394				lun->per_res[i].registered = 0;
8395				memset(&lun->per_res[i].res_key, 0,
8396				       sizeof(struct scsi_per_res_key));
8397				lun->pr_key_count--;
8398
8399				if (!persis_offset
8400				 && i < persis_offset)
8401					lun->pending_ua[i] |=
8402						CTL_UA_REG_PREEMPT;
8403				else if (persis_offset
8404				      && i >= persis_offset)
8405					lun->pending_ua[i - persis_offset] |=
8406						CTL_UA_REG_PREEMPT;
8407			}
8408		}
8409	} else {
8410		/*
8411		 * Temporarily unregister so it won't get removed
8412		 * or UA generated
8413		 */
8414		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8415		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8416			if (lun->per_res[i].registered == 0)
8417				continue;
8418
8419			if (memcmp(msg->pr.pr_info.sa_res_key,
8420	                   lun->per_res[i].res_key.key,
8421	                   sizeof(struct scsi_per_res_key)) == 0) {
8422				lun->per_res[i].registered = 0;
8423				memset(&lun->per_res[i].res_key, 0,
8424				       sizeof(struct scsi_per_res_key));
8425				lun->pr_key_count--;
8426				if (!persis_offset
8427				 && i < CTL_MAX_INITIATORS)
8428					lun->pending_ua[i] |=
8429						CTL_UA_REG_PREEMPT;
8430				else if (persis_offset
8431				      && i >= persis_offset)
8432					lun->pending_ua[i - persis_offset] |=
8433						CTL_UA_REG_PREEMPT;
8434			} else if (msg->pr.pr_info.res_type != lun->res_type
8435				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8436				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8437					if (!persis_offset
8438					 && i < persis_offset)
8439						lun->pending_ua[i] |=
8440							CTL_UA_RES_RELEASE;
8441					else if (persis_offset
8442					      && i >= persis_offset)
8443					lun->pending_ua[i - persis_offset] |=
8444						CTL_UA_RES_RELEASE;
8445			}
8446		}
8447		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8448		lun->res_type = msg->pr.pr_info.res_type;
8449		if (lun->res_type != SPR_TYPE_WR_EX_AR
8450		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8451			lun->pr_res_idx = msg->pr.pr_info.residx;
8452		else
8453			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8454	}
8455	lun->PRGeneration++;
8456
8457}
8458
8459
8460int
8461ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8462{
8463	int retval;
8464	int isc_retval;
8465	u_int32_t param_len;
8466	struct scsi_per_res_out *cdb;
8467	struct ctl_lun *lun;
8468	struct scsi_per_res_out_parms* param;
8469	struct ctl_softc *softc;
8470	uint32_t residx;
8471	uint64_t res_key, sa_res_key;
8472	uint8_t type;
8473	union ctl_ha_msg persis_io;
8474	int    i;
8475
8476	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8477
8478	retval = CTL_RETVAL_COMPLETE;
8479
8480	softc = control_softc;
8481
8482	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8483	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8484
8485	/*
8486	 * We only support whole-LUN scope.  The scope & type are ignored for
8487	 * register, register and ignore existing key and clear.
8488	 * We sometimes ignore scope and type on preempts too!!
8489	 * Verify reservation type here as well.
8490	 */
8491	type = cdb->scope_type & SPR_TYPE_MASK;
8492	if ((cdb->action == SPRO_RESERVE)
8493	 || (cdb->action == SPRO_RELEASE)) {
8494		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8495			ctl_set_invalid_field(/*ctsio*/ ctsio,
8496					      /*sks_valid*/ 1,
8497					      /*command*/ 1,
8498					      /*field*/ 2,
8499					      /*bit_valid*/ 1,
8500					      /*bit*/ 4);
8501			ctl_done((union ctl_io *)ctsio);
8502			return (CTL_RETVAL_COMPLETE);
8503		}
8504
8505		if (type>8 || type==2 || type==4 || type==0) {
8506			ctl_set_invalid_field(/*ctsio*/ ctsio,
8507					      /*sks_valid*/ 1,
8508					      /*command*/ 1,
8509					      /*field*/ 2,
8510					      /*bit_valid*/ 1,
8511					      /*bit*/ 0);
8512			ctl_done((union ctl_io *)ctsio);
8513			return (CTL_RETVAL_COMPLETE);
8514		}
8515	}
8516
8517	param_len = scsi_4btoul(cdb->length);
8518
8519	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8520		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8521		ctsio->kern_data_len = param_len;
8522		ctsio->kern_total_len = param_len;
8523		ctsio->kern_data_resid = 0;
8524		ctsio->kern_rel_offset = 0;
8525		ctsio->kern_sg_entries = 0;
8526		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8527		ctsio->be_move_done = ctl_config_move_done;
8528		ctl_datamove((union ctl_io *)ctsio);
8529
8530		return (CTL_RETVAL_COMPLETE);
8531	}
8532
8533	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8534
8535	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8536	res_key = scsi_8btou64(param->res_key.key);
8537	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8538
8539	/*
8540	 * Validate the reservation key here except for SPRO_REG_IGNO
8541	 * This must be done for all other service actions
8542	 */
8543	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8544		mtx_lock(&lun->lun_lock);
8545		if (lun->per_res[residx].registered) {
8546		    if (memcmp(param->res_key.key,
8547			       lun->per_res[residx].res_key.key,
8548			       ctl_min(sizeof(param->res_key),
8549			       sizeof(lun->per_res[residx].res_key))) != 0) {
8550				/*
8551				 * The current key passed in doesn't match
8552				 * the one the initiator previously
8553				 * registered.
8554				 */
8555				mtx_unlock(&lun->lun_lock);
8556				free(ctsio->kern_data_ptr, M_CTL);
8557				ctl_set_reservation_conflict(ctsio);
8558				ctl_done((union ctl_io *)ctsio);
8559				return (CTL_RETVAL_COMPLETE);
8560			}
8561		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8562			/*
8563			 * We are not 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		} else if (res_key != 0) {
8571			/*
8572			 * We are not registered and trying to register but
8573			 * the register key isn't zero.
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		}
8581		mtx_unlock(&lun->lun_lock);
8582	}
8583
8584	switch (cdb->action & SPRO_ACTION_MASK) {
8585	case SPRO_REGISTER:
8586	case SPRO_REG_IGNO: {
8587
8588#if 0
8589		printf("Registration received\n");
8590#endif
8591
8592		/*
8593		 * We don't support any of these options, as we report in
8594		 * the read capabilities request (see
8595		 * ctl_persistent_reserve_in(), above).
8596		 */
8597		if ((param->flags & SPR_SPEC_I_PT)
8598		 || (param->flags & SPR_ALL_TG_PT)
8599		 || (param->flags & SPR_APTPL)) {
8600			int bit_ptr;
8601
8602			if (param->flags & SPR_APTPL)
8603				bit_ptr = 0;
8604			else if (param->flags & SPR_ALL_TG_PT)
8605				bit_ptr = 2;
8606			else /* SPR_SPEC_I_PT */
8607				bit_ptr = 3;
8608
8609			free(ctsio->kern_data_ptr, M_CTL);
8610			ctl_set_invalid_field(ctsio,
8611					      /*sks_valid*/ 1,
8612					      /*command*/ 0,
8613					      /*field*/ 20,
8614					      /*bit_valid*/ 1,
8615					      /*bit*/ bit_ptr);
8616			ctl_done((union ctl_io *)ctsio);
8617			return (CTL_RETVAL_COMPLETE);
8618		}
8619
8620		mtx_lock(&lun->lun_lock);
8621
8622		/*
8623		 * The initiator wants to clear the
8624		 * key/unregister.
8625		 */
8626		if (sa_res_key == 0) {
8627			if ((res_key == 0
8628			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8629			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8630			  && !lun->per_res[residx].registered)) {
8631				mtx_unlock(&lun->lun_lock);
8632				goto done;
8633			}
8634
8635			lun->per_res[residx].registered = 0;
8636			memset(&lun->per_res[residx].res_key,
8637			       0, sizeof(lun->per_res[residx].res_key));
8638			lun->pr_key_count--;
8639
8640			if (residx == lun->pr_res_idx) {
8641				lun->flags &= ~CTL_LUN_PR_RESERVED;
8642				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8643
8644				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8645				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8646				 && lun->pr_key_count) {
8647					/*
8648					 * If the reservation is a registrants
8649					 * only type we need to generate a UA
8650					 * for other registered inits.  The
8651					 * sense code should be RESERVATIONS
8652					 * RELEASED
8653					 */
8654
8655					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8656						if (lun->per_res[
8657						    i+persis_offset].registered
8658						    == 0)
8659							continue;
8660						lun->pending_ua[i] |=
8661							CTL_UA_RES_RELEASE;
8662					}
8663				}
8664				lun->res_type = 0;
8665			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8666				if (lun->pr_key_count==0) {
8667					lun->flags &= ~CTL_LUN_PR_RESERVED;
8668					lun->res_type = 0;
8669					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8670				}
8671			}
8672			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8673			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8674			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8675			persis_io.pr.pr_info.residx = residx;
8676			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8677			     &persis_io, sizeof(persis_io), 0 )) >
8678			     CTL_HA_STATUS_SUCCESS) {
8679				printf("CTL:Persis Out error returned from "
8680				       "ctl_ha_msg_send %d\n", isc_retval);
8681			}
8682		} else /* sa_res_key != 0 */ {
8683
8684			/*
8685			 * If we aren't registered currently then increment
8686			 * the key count and set the registered flag.
8687			 */
8688			if (!lun->per_res[residx].registered) {
8689				lun->pr_key_count++;
8690				lun->per_res[residx].registered = 1;
8691			}
8692
8693			memcpy(&lun->per_res[residx].res_key,
8694			       param->serv_act_res_key,
8695			       ctl_min(sizeof(param->serv_act_res_key),
8696			       sizeof(lun->per_res[residx].res_key)));
8697
8698			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8699			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8700			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8701			persis_io.pr.pr_info.residx = residx;
8702			memcpy(persis_io.pr.pr_info.sa_res_key,
8703			       param->serv_act_res_key,
8704			       sizeof(param->serv_act_res_key));
8705			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8706			     &persis_io, sizeof(persis_io), 0)) >
8707			     CTL_HA_STATUS_SUCCESS) {
8708				printf("CTL:Persis Out error returned from "
8709				       "ctl_ha_msg_send %d\n", isc_retval);
8710			}
8711		}
8712		lun->PRGeneration++;
8713		mtx_unlock(&lun->lun_lock);
8714
8715		break;
8716	}
8717	case SPRO_RESERVE:
8718#if 0
8719                printf("Reserve executed type %d\n", type);
8720#endif
8721		mtx_lock(&lun->lun_lock);
8722		if (lun->flags & CTL_LUN_PR_RESERVED) {
8723			/*
8724			 * if this isn't the reservation holder and it's
8725			 * not a "all registrants" type or if the type is
8726			 * different then we have a conflict
8727			 */
8728			if ((lun->pr_res_idx != residx
8729			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8730			 || lun->res_type != type) {
8731				mtx_unlock(&lun->lun_lock);
8732				free(ctsio->kern_data_ptr, M_CTL);
8733				ctl_set_reservation_conflict(ctsio);
8734				ctl_done((union ctl_io *)ctsio);
8735				return (CTL_RETVAL_COMPLETE);
8736			}
8737			mtx_unlock(&lun->lun_lock);
8738		} else /* create a reservation */ {
8739			/*
8740			 * If it's not an "all registrants" type record
8741			 * reservation holder
8742			 */
8743			if (type != SPR_TYPE_WR_EX_AR
8744			 && type != SPR_TYPE_EX_AC_AR)
8745				lun->pr_res_idx = residx; /* Res holder */
8746			else
8747				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8748
8749			lun->flags |= CTL_LUN_PR_RESERVED;
8750			lun->res_type = type;
8751
8752			mtx_unlock(&lun->lun_lock);
8753
8754			/* send msg to other side */
8755			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8756			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8757			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8758			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8759			persis_io.pr.pr_info.res_type = type;
8760			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8761			     &persis_io, sizeof(persis_io), 0)) >
8762			     CTL_HA_STATUS_SUCCESS) {
8763				printf("CTL:Persis Out error returned from "
8764				       "ctl_ha_msg_send %d\n", isc_retval);
8765			}
8766		}
8767		break;
8768
8769	case SPRO_RELEASE:
8770		mtx_lock(&lun->lun_lock);
8771		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8772			/* No reservation exists return good status */
8773			mtx_unlock(&lun->lun_lock);
8774			goto done;
8775		}
8776		/*
8777		 * Is this nexus a reservation holder?
8778		 */
8779		if (lun->pr_res_idx != residx
8780		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8781			/*
8782			 * not a res holder return good status but
8783			 * do nothing
8784			 */
8785			mtx_unlock(&lun->lun_lock);
8786			goto done;
8787		}
8788
8789		if (lun->res_type != type) {
8790			mtx_unlock(&lun->lun_lock);
8791			free(ctsio->kern_data_ptr, M_CTL);
8792			ctl_set_illegal_pr_release(ctsio);
8793			ctl_done((union ctl_io *)ctsio);
8794			return (CTL_RETVAL_COMPLETE);
8795		}
8796
8797		/* okay to release */
8798		lun->flags &= ~CTL_LUN_PR_RESERVED;
8799		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8800		lun->res_type = 0;
8801
8802		/*
8803		 * if this isn't an exclusive access
8804		 * res generate UA for all other
8805		 * registrants.
8806		 */
8807		if (type != SPR_TYPE_EX_AC
8808		 && type != SPR_TYPE_WR_EX) {
8809			/*
8810			 * temporarily unregister so we don't generate UA
8811			 */
8812			lun->per_res[residx].registered = 0;
8813
8814			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8815				if (lun->per_res[i+persis_offset].registered
8816				    == 0)
8817					continue;
8818				lun->pending_ua[i] |=
8819					CTL_UA_RES_RELEASE;
8820			}
8821
8822			lun->per_res[residx].registered = 1;
8823		}
8824		mtx_unlock(&lun->lun_lock);
8825		/* Send msg to other side */
8826		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8827		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8828		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8829		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8830		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8831			printf("CTL:Persis Out error returned from "
8832			       "ctl_ha_msg_send %d\n", isc_retval);
8833		}
8834		break;
8835
8836	case SPRO_CLEAR:
8837		/* send msg to other side */
8838
8839		mtx_lock(&lun->lun_lock);
8840		lun->flags &= ~CTL_LUN_PR_RESERVED;
8841		lun->res_type = 0;
8842		lun->pr_key_count = 0;
8843		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8844
8845
8846		memset(&lun->per_res[residx].res_key,
8847		       0, sizeof(lun->per_res[residx].res_key));
8848		lun->per_res[residx].registered = 0;
8849
8850		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8851			if (lun->per_res[i].registered) {
8852				if (!persis_offset && i < CTL_MAX_INITIATORS)
8853					lun->pending_ua[i] |=
8854						CTL_UA_RES_PREEMPT;
8855				else if (persis_offset && i >= persis_offset)
8856					lun->pending_ua[i-persis_offset] |=
8857					    CTL_UA_RES_PREEMPT;
8858
8859				memset(&lun->per_res[i].res_key,
8860				       0, sizeof(struct scsi_per_res_key));
8861				lun->per_res[i].registered = 0;
8862			}
8863		lun->PRGeneration++;
8864		mtx_unlock(&lun->lun_lock);
8865		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8866		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8867		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8868		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8869		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8870			printf("CTL:Persis Out error returned from "
8871			       "ctl_ha_msg_send %d\n", isc_retval);
8872		}
8873		break;
8874
8875	case SPRO_PREEMPT: {
8876		int nretval;
8877
8878		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8879					  residx, ctsio, cdb, param);
8880		if (nretval != 0)
8881			return (CTL_RETVAL_COMPLETE);
8882		break;
8883	}
8884	default:
8885		panic("Invalid PR type %x", cdb->action);
8886	}
8887
8888done:
8889	free(ctsio->kern_data_ptr, M_CTL);
8890	ctl_set_success(ctsio);
8891	ctl_done((union ctl_io *)ctsio);
8892
8893	return (retval);
8894}
8895
8896/*
8897 * This routine is for handling a message from the other SC pertaining to
8898 * persistent reserve out. All the error checking will have been done
8899 * so only perorming the action need be done here to keep the two
8900 * in sync.
8901 */
8902static void
8903ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8904{
8905	struct ctl_lun *lun;
8906	struct ctl_softc *softc;
8907	int i;
8908	uint32_t targ_lun;
8909
8910	softc = control_softc;
8911
8912	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8913	lun = softc->ctl_luns[targ_lun];
8914	mtx_lock(&lun->lun_lock);
8915	switch(msg->pr.pr_info.action) {
8916	case CTL_PR_REG_KEY:
8917		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8918			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8919			lun->pr_key_count++;
8920		}
8921		lun->PRGeneration++;
8922		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8923		       msg->pr.pr_info.sa_res_key,
8924		       sizeof(struct scsi_per_res_key));
8925		break;
8926
8927	case CTL_PR_UNREG_KEY:
8928		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8929		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8930		       0, sizeof(struct scsi_per_res_key));
8931		lun->pr_key_count--;
8932
8933		/* XXX Need to see if the reservation has been released */
8934		/* if so do we need to generate UA? */
8935		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8936			lun->flags &= ~CTL_LUN_PR_RESERVED;
8937			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8938
8939			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8940			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8941			 && lun->pr_key_count) {
8942				/*
8943				 * If the reservation is a registrants
8944				 * only type we need to generate a UA
8945				 * for other registered inits.  The
8946				 * sense code should be RESERVATIONS
8947				 * RELEASED
8948				 */
8949
8950				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8951					if (lun->per_res[i+
8952					    persis_offset].registered == 0)
8953						continue;
8954
8955					lun->pending_ua[i] |=
8956						CTL_UA_RES_RELEASE;
8957				}
8958			}
8959			lun->res_type = 0;
8960		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8961			if (lun->pr_key_count==0) {
8962				lun->flags &= ~CTL_LUN_PR_RESERVED;
8963				lun->res_type = 0;
8964				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8965			}
8966		}
8967		lun->PRGeneration++;
8968		break;
8969
8970	case CTL_PR_RESERVE:
8971		lun->flags |= CTL_LUN_PR_RESERVED;
8972		lun->res_type = msg->pr.pr_info.res_type;
8973		lun->pr_res_idx = msg->pr.pr_info.residx;
8974
8975		break;
8976
8977	case CTL_PR_RELEASE:
8978		/*
8979		 * if this isn't an exclusive access res generate UA for all
8980		 * other registrants.
8981		 */
8982		if (lun->res_type != SPR_TYPE_EX_AC
8983		 && lun->res_type != SPR_TYPE_WR_EX) {
8984			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8985				if (lun->per_res[i+persis_offset].registered)
8986					lun->pending_ua[i] |=
8987						CTL_UA_RES_RELEASE;
8988		}
8989
8990		lun->flags &= ~CTL_LUN_PR_RESERVED;
8991		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8992		lun->res_type = 0;
8993		break;
8994
8995	case CTL_PR_PREEMPT:
8996		ctl_pro_preempt_other(lun, msg);
8997		break;
8998	case CTL_PR_CLEAR:
8999		lun->flags &= ~CTL_LUN_PR_RESERVED;
9000		lun->res_type = 0;
9001		lun->pr_key_count = 0;
9002		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
9003
9004		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
9005			if (lun->per_res[i].registered == 0)
9006				continue;
9007			if (!persis_offset
9008			 && i < CTL_MAX_INITIATORS)
9009				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
9010			else if (persis_offset
9011			      && i >= persis_offset)
9012				lun->pending_ua[i-persis_offset] |=
9013					CTL_UA_RES_PREEMPT;
9014			memset(&lun->per_res[i].res_key, 0,
9015			       sizeof(struct scsi_per_res_key));
9016			lun->per_res[i].registered = 0;
9017		}
9018		lun->PRGeneration++;
9019		break;
9020	}
9021
9022	mtx_unlock(&lun->lun_lock);
9023}
9024
9025int
9026ctl_read_write(struct ctl_scsiio *ctsio)
9027{
9028	struct ctl_lun *lun;
9029	struct ctl_lba_len_flags *lbalen;
9030	uint64_t lba;
9031	uint32_t num_blocks;
9032	int flags, retval;
9033	int isread;
9034
9035	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9036
9037	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
9038
9039	flags = 0;
9040	retval = CTL_RETVAL_COMPLETE;
9041
9042	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
9043	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
9044	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
9045		uint32_t residx;
9046
9047		/*
9048		 * XXX KDM need a lock here.
9049		 */
9050		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
9051		if ((lun->res_type == SPR_TYPE_EX_AC
9052		  && residx != lun->pr_res_idx)
9053		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
9054		   || lun->res_type == SPR_TYPE_EX_AC_AR)
9055		  && !lun->per_res[residx].registered)) {
9056			ctl_set_reservation_conflict(ctsio);
9057			ctl_done((union ctl_io *)ctsio);
9058			return (CTL_RETVAL_COMPLETE);
9059	        }
9060	}
9061
9062	switch (ctsio->cdb[0]) {
9063	case READ_6:
9064	case WRITE_6: {
9065		struct scsi_rw_6 *cdb;
9066
9067		cdb = (struct scsi_rw_6 *)ctsio->cdb;
9068
9069		lba = scsi_3btoul(cdb->addr);
9070		/* only 5 bits are valid in the most significant address byte */
9071		lba &= 0x1fffff;
9072		num_blocks = cdb->length;
9073		/*
9074		 * This is correct according to SBC-2.
9075		 */
9076		if (num_blocks == 0)
9077			num_blocks = 256;
9078		break;
9079	}
9080	case READ_10:
9081	case WRITE_10: {
9082		struct scsi_rw_10 *cdb;
9083
9084		cdb = (struct scsi_rw_10 *)ctsio->cdb;
9085		if (cdb->byte2 & SRW10_FUA)
9086			flags |= CTL_LLF_FUA;
9087		if (cdb->byte2 & SRW10_DPO)
9088			flags |= CTL_LLF_DPO;
9089		lba = scsi_4btoul(cdb->addr);
9090		num_blocks = scsi_2btoul(cdb->length);
9091		break;
9092	}
9093	case WRITE_VERIFY_10: {
9094		struct scsi_write_verify_10 *cdb;
9095
9096		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
9097		flags |= CTL_LLF_FUA;
9098		if (cdb->byte2 & SWV_DPO)
9099			flags |= CTL_LLF_DPO;
9100		lba = scsi_4btoul(cdb->addr);
9101		num_blocks = scsi_2btoul(cdb->length);
9102		break;
9103	}
9104	case READ_12:
9105	case WRITE_12: {
9106		struct scsi_rw_12 *cdb;
9107
9108		cdb = (struct scsi_rw_12 *)ctsio->cdb;
9109		if (cdb->byte2 & SRW12_FUA)
9110			flags |= CTL_LLF_FUA;
9111		if (cdb->byte2 & SRW12_DPO)
9112			flags |= CTL_LLF_DPO;
9113		lba = scsi_4btoul(cdb->addr);
9114		num_blocks = scsi_4btoul(cdb->length);
9115		break;
9116	}
9117	case WRITE_VERIFY_12: {
9118		struct scsi_write_verify_12 *cdb;
9119
9120		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
9121		flags |= CTL_LLF_FUA;
9122		if (cdb->byte2 & SWV_DPO)
9123			flags |= CTL_LLF_DPO;
9124		lba = scsi_4btoul(cdb->addr);
9125		num_blocks = scsi_4btoul(cdb->length);
9126		break;
9127	}
9128	case READ_16:
9129	case WRITE_16: {
9130		struct scsi_rw_16 *cdb;
9131
9132		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9133		if (cdb->byte2 & SRW12_FUA)
9134			flags |= CTL_LLF_FUA;
9135		if (cdb->byte2 & SRW12_DPO)
9136			flags |= CTL_LLF_DPO;
9137		lba = scsi_8btou64(cdb->addr);
9138		num_blocks = scsi_4btoul(cdb->length);
9139		break;
9140	}
9141	case WRITE_VERIFY_16: {
9142		struct scsi_write_verify_16 *cdb;
9143
9144		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9145		flags |= CTL_LLF_FUA;
9146		if (cdb->byte2 & SWV_DPO)
9147			flags |= CTL_LLF_DPO;
9148		lba = scsi_8btou64(cdb->addr);
9149		num_blocks = scsi_4btoul(cdb->length);
9150		break;
9151	}
9152	default:
9153		/*
9154		 * We got a command we don't support.  This shouldn't
9155		 * happen, commands should be filtered out above us.
9156		 */
9157		ctl_set_invalid_opcode(ctsio);
9158		ctl_done((union ctl_io *)ctsio);
9159
9160		return (CTL_RETVAL_COMPLETE);
9161		break; /* NOTREACHED */
9162	}
9163
9164	/*
9165	 * The first check is to make sure we're in bounds, the second
9166	 * check is to catch wrap-around problems.  If the lba + num blocks
9167	 * is less than the lba, then we've wrapped around and the block
9168	 * range is invalid anyway.
9169	 */
9170	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9171	 || ((lba + num_blocks) < lba)) {
9172		ctl_set_lba_out_of_range(ctsio);
9173		ctl_done((union ctl_io *)ctsio);
9174		return (CTL_RETVAL_COMPLETE);
9175	}
9176
9177	/*
9178	 * According to SBC-3, a transfer length of 0 is not an error.
9179	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9180	 * translates to 256 blocks for those commands.
9181	 */
9182	if (num_blocks == 0) {
9183		ctl_set_success(ctsio);
9184		ctl_done((union ctl_io *)ctsio);
9185		return (CTL_RETVAL_COMPLETE);
9186	}
9187
9188	/* Set FUA and/or DPO if caches are disabled. */
9189	if (isread) {
9190		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9191		    SCP_RCD) != 0)
9192			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9193	} else {
9194		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9195		    SCP_WCE) == 0)
9196			flags |= CTL_LLF_FUA;
9197	}
9198
9199	lbalen = (struct ctl_lba_len_flags *)
9200	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9201	lbalen->lba = lba;
9202	lbalen->len = num_blocks;
9203	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9204
9205	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9206	ctsio->kern_rel_offset = 0;
9207
9208	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9209
9210	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9211
9212	return (retval);
9213}
9214
9215static int
9216ctl_cnw_cont(union ctl_io *io)
9217{
9218	struct ctl_scsiio *ctsio;
9219	struct ctl_lun *lun;
9220	struct ctl_lba_len_flags *lbalen;
9221	int retval;
9222
9223	ctsio = &io->scsiio;
9224	ctsio->io_hdr.status = CTL_STATUS_NONE;
9225	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9226	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9227	lbalen = (struct ctl_lba_len_flags *)
9228	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9229	lbalen->flags &= ~CTL_LLF_COMPARE;
9230	lbalen->flags |= CTL_LLF_WRITE;
9231
9232	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9233	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9234	return (retval);
9235}
9236
9237int
9238ctl_cnw(struct ctl_scsiio *ctsio)
9239{
9240	struct ctl_lun *lun;
9241	struct ctl_lba_len_flags *lbalen;
9242	uint64_t lba;
9243	uint32_t num_blocks;
9244	int flags, retval;
9245
9246	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9247
9248	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9249
9250	flags = 0;
9251	retval = CTL_RETVAL_COMPLETE;
9252
9253	switch (ctsio->cdb[0]) {
9254	case COMPARE_AND_WRITE: {
9255		struct scsi_compare_and_write *cdb;
9256
9257		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9258		if (cdb->byte2 & SRW10_FUA)
9259			flags |= CTL_LLF_FUA;
9260		if (cdb->byte2 & SRW10_DPO)
9261			flags |= CTL_LLF_DPO;
9262		lba = scsi_8btou64(cdb->addr);
9263		num_blocks = cdb->length;
9264		break;
9265	}
9266	default:
9267		/*
9268		 * We got a command we don't support.  This shouldn't
9269		 * happen, commands should be filtered out above us.
9270		 */
9271		ctl_set_invalid_opcode(ctsio);
9272		ctl_done((union ctl_io *)ctsio);
9273
9274		return (CTL_RETVAL_COMPLETE);
9275		break; /* NOTREACHED */
9276	}
9277
9278	/*
9279	 * The first check is to make sure we're in bounds, the second
9280	 * check is to catch wrap-around problems.  If the lba + num blocks
9281	 * is less than the lba, then we've wrapped around and the block
9282	 * range is invalid anyway.
9283	 */
9284	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9285	 || ((lba + num_blocks) < lba)) {
9286		ctl_set_lba_out_of_range(ctsio);
9287		ctl_done((union ctl_io *)ctsio);
9288		return (CTL_RETVAL_COMPLETE);
9289	}
9290
9291	/*
9292	 * According to SBC-3, a transfer length of 0 is not an error.
9293	 */
9294	if (num_blocks == 0) {
9295		ctl_set_success(ctsio);
9296		ctl_done((union ctl_io *)ctsio);
9297		return (CTL_RETVAL_COMPLETE);
9298	}
9299
9300	/* Set FUA if write cache is disabled. */
9301	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9302	    SCP_WCE) == 0)
9303		flags |= CTL_LLF_FUA;
9304
9305	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9306	ctsio->kern_rel_offset = 0;
9307
9308	/*
9309	 * Set the IO_CONT flag, so that if this I/O gets passed to
9310	 * ctl_data_submit_done(), it'll get passed back to
9311	 * ctl_ctl_cnw_cont() for further processing.
9312	 */
9313	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9314	ctsio->io_cont = ctl_cnw_cont;
9315
9316	lbalen = (struct ctl_lba_len_flags *)
9317	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9318	lbalen->lba = lba;
9319	lbalen->len = num_blocks;
9320	lbalen->flags = CTL_LLF_COMPARE | flags;
9321
9322	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9323	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9324	return (retval);
9325}
9326
9327int
9328ctl_verify(struct ctl_scsiio *ctsio)
9329{
9330	struct ctl_lun *lun;
9331	struct ctl_lba_len_flags *lbalen;
9332	uint64_t lba;
9333	uint32_t num_blocks;
9334	int bytchk, flags;
9335	int retval;
9336
9337	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9338
9339	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9340
9341	bytchk = 0;
9342	flags = CTL_LLF_FUA;
9343	retval = CTL_RETVAL_COMPLETE;
9344
9345	switch (ctsio->cdb[0]) {
9346	case VERIFY_10: {
9347		struct scsi_verify_10 *cdb;
9348
9349		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9350		if (cdb->byte2 & SVFY_BYTCHK)
9351			bytchk = 1;
9352		if (cdb->byte2 & SVFY_DPO)
9353			flags |= CTL_LLF_DPO;
9354		lba = scsi_4btoul(cdb->addr);
9355		num_blocks = scsi_2btoul(cdb->length);
9356		break;
9357	}
9358	case VERIFY_12: {
9359		struct scsi_verify_12 *cdb;
9360
9361		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9362		if (cdb->byte2 & SVFY_BYTCHK)
9363			bytchk = 1;
9364		if (cdb->byte2 & SVFY_DPO)
9365			flags |= CTL_LLF_DPO;
9366		lba = scsi_4btoul(cdb->addr);
9367		num_blocks = scsi_4btoul(cdb->length);
9368		break;
9369	}
9370	case VERIFY_16: {
9371		struct scsi_rw_16 *cdb;
9372
9373		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9374		if (cdb->byte2 & SVFY_BYTCHK)
9375			bytchk = 1;
9376		if (cdb->byte2 & SVFY_DPO)
9377			flags |= CTL_LLF_DPO;
9378		lba = scsi_8btou64(cdb->addr);
9379		num_blocks = scsi_4btoul(cdb->length);
9380		break;
9381	}
9382	default:
9383		/*
9384		 * We got a command we don't support.  This shouldn't
9385		 * happen, commands should be filtered out above us.
9386		 */
9387		ctl_set_invalid_opcode(ctsio);
9388		ctl_done((union ctl_io *)ctsio);
9389		return (CTL_RETVAL_COMPLETE);
9390	}
9391
9392	/*
9393	 * The first check is to make sure we're in bounds, the second
9394	 * check is to catch wrap-around problems.  If the lba + num blocks
9395	 * is less than the lba, then we've wrapped around and the block
9396	 * range is invalid anyway.
9397	 */
9398	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9399	 || ((lba + num_blocks) < lba)) {
9400		ctl_set_lba_out_of_range(ctsio);
9401		ctl_done((union ctl_io *)ctsio);
9402		return (CTL_RETVAL_COMPLETE);
9403	}
9404
9405	/*
9406	 * According to SBC-3, a transfer length of 0 is not an error.
9407	 */
9408	if (num_blocks == 0) {
9409		ctl_set_success(ctsio);
9410		ctl_done((union ctl_io *)ctsio);
9411		return (CTL_RETVAL_COMPLETE);
9412	}
9413
9414	lbalen = (struct ctl_lba_len_flags *)
9415	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9416	lbalen->lba = lba;
9417	lbalen->len = num_blocks;
9418	if (bytchk) {
9419		lbalen->flags = CTL_LLF_COMPARE | flags;
9420		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9421	} else {
9422		lbalen->flags = CTL_LLF_VERIFY | flags;
9423		ctsio->kern_total_len = 0;
9424	}
9425	ctsio->kern_rel_offset = 0;
9426
9427	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9428	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9429	return (retval);
9430}
9431
9432int
9433ctl_report_luns(struct ctl_scsiio *ctsio)
9434{
9435	struct scsi_report_luns *cdb;
9436	struct scsi_report_luns_data *lun_data;
9437	struct ctl_lun *lun, *request_lun;
9438	int num_luns, retval;
9439	uint32_t alloc_len, lun_datalen;
9440	int num_filled, well_known;
9441	uint32_t initidx, targ_lun_id, lun_id;
9442
9443	retval = CTL_RETVAL_COMPLETE;
9444	well_known = 0;
9445
9446	cdb = (struct scsi_report_luns *)ctsio->cdb;
9447
9448	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9449
9450	mtx_lock(&control_softc->ctl_lock);
9451	num_luns = control_softc->num_luns;
9452	mtx_unlock(&control_softc->ctl_lock);
9453
9454	switch (cdb->select_report) {
9455	case RPL_REPORT_DEFAULT:
9456	case RPL_REPORT_ALL:
9457		break;
9458	case RPL_REPORT_WELLKNOWN:
9459		well_known = 1;
9460		num_luns = 0;
9461		break;
9462	default:
9463		ctl_set_invalid_field(ctsio,
9464				      /*sks_valid*/ 1,
9465				      /*command*/ 1,
9466				      /*field*/ 2,
9467				      /*bit_valid*/ 0,
9468				      /*bit*/ 0);
9469		ctl_done((union ctl_io *)ctsio);
9470		return (retval);
9471		break; /* NOTREACHED */
9472	}
9473
9474	alloc_len = scsi_4btoul(cdb->length);
9475	/*
9476	 * The initiator has to allocate at least 16 bytes for this request,
9477	 * so he can at least get the header and the first LUN.  Otherwise
9478	 * we reject the request (per SPC-3 rev 14, section 6.21).
9479	 */
9480	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9481	    sizeof(struct scsi_report_luns_lundata))) {
9482		ctl_set_invalid_field(ctsio,
9483				      /*sks_valid*/ 1,
9484				      /*command*/ 1,
9485				      /*field*/ 6,
9486				      /*bit_valid*/ 0,
9487				      /*bit*/ 0);
9488		ctl_done((union ctl_io *)ctsio);
9489		return (retval);
9490	}
9491
9492	request_lun = (struct ctl_lun *)
9493		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9494
9495	lun_datalen = sizeof(*lun_data) +
9496		(num_luns * sizeof(struct scsi_report_luns_lundata));
9497
9498	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9499	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9500	ctsio->kern_sg_entries = 0;
9501
9502	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9503
9504	mtx_lock(&control_softc->ctl_lock);
9505	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9506		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9507		if (lun_id >= CTL_MAX_LUNS)
9508			continue;
9509		lun = control_softc->ctl_luns[lun_id];
9510		if (lun == NULL)
9511			continue;
9512
9513		if (targ_lun_id <= 0xff) {
9514			/*
9515			 * Peripheral addressing method, bus number 0.
9516			 */
9517			lun_data->luns[num_filled].lundata[0] =
9518				RPL_LUNDATA_ATYP_PERIPH;
9519			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9520			num_filled++;
9521		} else if (targ_lun_id <= 0x3fff) {
9522			/*
9523			 * Flat addressing method.
9524			 */
9525			lun_data->luns[num_filled].lundata[0] =
9526				RPL_LUNDATA_ATYP_FLAT |
9527				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9528#ifdef OLDCTLHEADERS
9529				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9530				(targ_lun_id & SRLD_BUS_LUN_MASK);
9531#endif
9532			lun_data->luns[num_filled].lundata[1] =
9533#ifdef OLDCTLHEADERS
9534				targ_lun_id >> SRLD_BUS_LUN_BITS;
9535#endif
9536				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9537			num_filled++;
9538		} else {
9539			printf("ctl_report_luns: bogus LUN number %jd, "
9540			       "skipping\n", (intmax_t)targ_lun_id);
9541		}
9542		/*
9543		 * According to SPC-3, rev 14 section 6.21:
9544		 *
9545		 * "The execution of a REPORT LUNS command to any valid and
9546		 * installed logical unit shall clear the REPORTED LUNS DATA
9547		 * HAS CHANGED unit attention condition for all logical
9548		 * units of that target with respect to the requesting
9549		 * initiator. A valid and installed logical unit is one
9550		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9551		 * INQUIRY data (see 6.4.2)."
9552		 *
9553		 * If request_lun is NULL, the LUN this report luns command
9554		 * was issued to is either disabled or doesn't exist. In that
9555		 * case, we shouldn't clear any pending lun change unit
9556		 * attention.
9557		 */
9558		if (request_lun != NULL) {
9559			mtx_lock(&lun->lun_lock);
9560			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9561			mtx_unlock(&lun->lun_lock);
9562		}
9563	}
9564	mtx_unlock(&control_softc->ctl_lock);
9565
9566	/*
9567	 * It's quite possible that we've returned fewer LUNs than we allocated
9568	 * space for.  Trim it.
9569	 */
9570	lun_datalen = sizeof(*lun_data) +
9571		(num_filled * sizeof(struct scsi_report_luns_lundata));
9572
9573	if (lun_datalen < alloc_len) {
9574		ctsio->residual = alloc_len - lun_datalen;
9575		ctsio->kern_data_len = lun_datalen;
9576		ctsio->kern_total_len = lun_datalen;
9577	} else {
9578		ctsio->residual = 0;
9579		ctsio->kern_data_len = alloc_len;
9580		ctsio->kern_total_len = alloc_len;
9581	}
9582	ctsio->kern_data_resid = 0;
9583	ctsio->kern_rel_offset = 0;
9584	ctsio->kern_sg_entries = 0;
9585
9586	/*
9587	 * We set this to the actual data length, regardless of how much
9588	 * space we actually have to return results.  If the user looks at
9589	 * this value, he'll know whether or not he allocated enough space
9590	 * and reissue the command if necessary.  We don't support well
9591	 * known logical units, so if the user asks for that, return none.
9592	 */
9593	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9594
9595	/*
9596	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9597	 * this request.
9598	 */
9599	ctsio->scsi_status = SCSI_STATUS_OK;
9600
9601	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9602	ctsio->be_move_done = ctl_config_move_done;
9603	ctl_datamove((union ctl_io *)ctsio);
9604
9605	return (retval);
9606}
9607
9608int
9609ctl_request_sense(struct ctl_scsiio *ctsio)
9610{
9611	struct scsi_request_sense *cdb;
9612	struct scsi_sense_data *sense_ptr;
9613	struct ctl_lun *lun;
9614	uint32_t initidx;
9615	int have_error;
9616	scsi_sense_data_type sense_format;
9617
9618	cdb = (struct scsi_request_sense *)ctsio->cdb;
9619
9620	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9621
9622	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9623
9624	/*
9625	 * Determine which sense format the user wants.
9626	 */
9627	if (cdb->byte2 & SRS_DESC)
9628		sense_format = SSD_TYPE_DESC;
9629	else
9630		sense_format = SSD_TYPE_FIXED;
9631
9632	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9633	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9634	ctsio->kern_sg_entries = 0;
9635
9636	/*
9637	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9638	 * larger than the largest allowed value for the length field in the
9639	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9640	 */
9641	ctsio->residual = 0;
9642	ctsio->kern_data_len = cdb->length;
9643	ctsio->kern_total_len = cdb->length;
9644
9645	ctsio->kern_data_resid = 0;
9646	ctsio->kern_rel_offset = 0;
9647	ctsio->kern_sg_entries = 0;
9648
9649	/*
9650	 * If we don't have a LUN, we don't have any pending sense.
9651	 */
9652	if (lun == NULL)
9653		goto no_sense;
9654
9655	have_error = 0;
9656	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9657	/*
9658	 * Check for pending sense, and then for pending unit attentions.
9659	 * Pending sense gets returned first, then pending unit attentions.
9660	 */
9661	mtx_lock(&lun->lun_lock);
9662#ifdef CTL_WITH_CA
9663	if (ctl_is_set(lun->have_ca, initidx)) {
9664		scsi_sense_data_type stored_format;
9665
9666		/*
9667		 * Check to see which sense format was used for the stored
9668		 * sense data.
9669		 */
9670		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9671
9672		/*
9673		 * If the user requested a different sense format than the
9674		 * one we stored, then we need to convert it to the other
9675		 * format.  If we're going from descriptor to fixed format
9676		 * sense data, we may lose things in translation, depending
9677		 * on what options were used.
9678		 *
9679		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9680		 * for some reason we'll just copy it out as-is.
9681		 */
9682		if ((stored_format == SSD_TYPE_FIXED)
9683		 && (sense_format == SSD_TYPE_DESC))
9684			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9685			    &lun->pending_sense[initidx],
9686			    (struct scsi_sense_data_desc *)sense_ptr);
9687		else if ((stored_format == SSD_TYPE_DESC)
9688		      && (sense_format == SSD_TYPE_FIXED))
9689			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9690			    &lun->pending_sense[initidx],
9691			    (struct scsi_sense_data_fixed *)sense_ptr);
9692		else
9693			memcpy(sense_ptr, &lun->pending_sense[initidx],
9694			       ctl_min(sizeof(*sense_ptr),
9695			       sizeof(lun->pending_sense[initidx])));
9696
9697		ctl_clear_mask(lun->have_ca, initidx);
9698		have_error = 1;
9699	} else
9700#endif
9701	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9702		ctl_ua_type ua_type;
9703
9704		ua_type = ctl_build_ua(lun->pending_ua[initidx],
9705				       sense_ptr, sense_format);
9706		if (ua_type != CTL_UA_NONE) {
9707			have_error = 1;
9708			/* We're reporting this UA, so clear it */
9709			lun->pending_ua[initidx] &= ~ua_type;
9710		}
9711	}
9712	mtx_unlock(&lun->lun_lock);
9713
9714	/*
9715	 * We already have a pending error, return it.
9716	 */
9717	if (have_error != 0) {
9718		/*
9719		 * We report the SCSI status as OK, since the status of the
9720		 * request sense command itself is OK.
9721		 */
9722		ctsio->scsi_status = SCSI_STATUS_OK;
9723
9724		/*
9725		 * We report 0 for the sense length, because we aren't doing
9726		 * autosense in this case.  We're reporting sense as
9727		 * parameter data.
9728		 */
9729		ctsio->sense_len = 0;
9730		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9731		ctsio->be_move_done = ctl_config_move_done;
9732		ctl_datamove((union ctl_io *)ctsio);
9733
9734		return (CTL_RETVAL_COMPLETE);
9735	}
9736
9737no_sense:
9738
9739	/*
9740	 * No sense information to report, so we report that everything is
9741	 * okay.
9742	 */
9743	ctl_set_sense_data(sense_ptr,
9744			   lun,
9745			   sense_format,
9746			   /*current_error*/ 1,
9747			   /*sense_key*/ SSD_KEY_NO_SENSE,
9748			   /*asc*/ 0x00,
9749			   /*ascq*/ 0x00,
9750			   SSD_ELEM_NONE);
9751
9752	ctsio->scsi_status = SCSI_STATUS_OK;
9753
9754	/*
9755	 * We report 0 for the sense length, because we aren't doing
9756	 * autosense in this case.  We're reporting sense as parameter data.
9757	 */
9758	ctsio->sense_len = 0;
9759	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9760	ctsio->be_move_done = ctl_config_move_done;
9761	ctl_datamove((union ctl_io *)ctsio);
9762
9763	return (CTL_RETVAL_COMPLETE);
9764}
9765
9766int
9767ctl_tur(struct ctl_scsiio *ctsio)
9768{
9769	struct ctl_lun *lun;
9770
9771	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9772
9773	CTL_DEBUG_PRINT(("ctl_tur\n"));
9774
9775	if (lun == NULL)
9776		return (EINVAL);
9777
9778	ctsio->scsi_status = SCSI_STATUS_OK;
9779	ctsio->io_hdr.status = CTL_SUCCESS;
9780
9781	ctl_done((union ctl_io *)ctsio);
9782
9783	return (CTL_RETVAL_COMPLETE);
9784}
9785
9786#ifdef notyet
9787static int
9788ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9789{
9790
9791}
9792#endif
9793
9794static int
9795ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9796{
9797	struct scsi_vpd_supported_pages *pages;
9798	int sup_page_size;
9799	struct ctl_lun *lun;
9800
9801	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9802
9803	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9804	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9805	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9806	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9807	ctsio->kern_sg_entries = 0;
9808
9809	if (sup_page_size < alloc_len) {
9810		ctsio->residual = alloc_len - sup_page_size;
9811		ctsio->kern_data_len = sup_page_size;
9812		ctsio->kern_total_len = sup_page_size;
9813	} else {
9814		ctsio->residual = 0;
9815		ctsio->kern_data_len = alloc_len;
9816		ctsio->kern_total_len = alloc_len;
9817	}
9818	ctsio->kern_data_resid = 0;
9819	ctsio->kern_rel_offset = 0;
9820	ctsio->kern_sg_entries = 0;
9821
9822	/*
9823	 * The control device is always connected.  The disk device, on the
9824	 * other hand, may not be online all the time.  Need to change this
9825	 * to figure out whether the disk device is actually online or not.
9826	 */
9827	if (lun != NULL)
9828		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9829				lun->be_lun->lun_type;
9830	else
9831		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9832
9833	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9834	/* Supported VPD pages */
9835	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9836	/* Serial Number */
9837	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9838	/* Device Identification */
9839	pages->page_list[2] = SVPD_DEVICE_ID;
9840	/* Mode Page Policy */
9841	pages->page_list[3] = SVPD_MODE_PAGE_POLICY;
9842	/* SCSI Ports */
9843	pages->page_list[4] = SVPD_SCSI_PORTS;
9844	/* Third-party Copy */
9845	pages->page_list[5] = SVPD_SCSI_TPC;
9846	/* Block limits */
9847	pages->page_list[6] = SVPD_BLOCK_LIMITS;
9848	/* Block Device Characteristics */
9849	pages->page_list[7] = SVPD_BDC;
9850	/* Logical Block Provisioning */
9851	pages->page_list[8] = SVPD_LBP;
9852
9853	ctsio->scsi_status = SCSI_STATUS_OK;
9854
9855	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9856	ctsio->be_move_done = ctl_config_move_done;
9857	ctl_datamove((union ctl_io *)ctsio);
9858
9859	return (CTL_RETVAL_COMPLETE);
9860}
9861
9862static int
9863ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9864{
9865	struct scsi_vpd_unit_serial_number *sn_ptr;
9866	struct ctl_lun *lun;
9867
9868	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9869
9870	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9871	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9872	ctsio->kern_sg_entries = 0;
9873
9874	if (sizeof(*sn_ptr) < alloc_len) {
9875		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9876		ctsio->kern_data_len = sizeof(*sn_ptr);
9877		ctsio->kern_total_len = sizeof(*sn_ptr);
9878	} else {
9879		ctsio->residual = 0;
9880		ctsio->kern_data_len = alloc_len;
9881		ctsio->kern_total_len = alloc_len;
9882	}
9883	ctsio->kern_data_resid = 0;
9884	ctsio->kern_rel_offset = 0;
9885	ctsio->kern_sg_entries = 0;
9886
9887	/*
9888	 * The control device is always connected.  The disk device, on the
9889	 * other hand, may not be online all the time.  Need to change this
9890	 * to figure out whether the disk device is actually online or not.
9891	 */
9892	if (lun != NULL)
9893		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9894				  lun->be_lun->lun_type;
9895	else
9896		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9897
9898	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9899	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9900	/*
9901	 * If we don't have a LUN, we just leave the serial number as
9902	 * all spaces.
9903	 */
9904	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9905	if (lun != NULL) {
9906		strncpy((char *)sn_ptr->serial_num,
9907			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9908	}
9909	ctsio->scsi_status = SCSI_STATUS_OK;
9910
9911	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9912	ctsio->be_move_done = ctl_config_move_done;
9913	ctl_datamove((union ctl_io *)ctsio);
9914
9915	return (CTL_RETVAL_COMPLETE);
9916}
9917
9918
9919static int
9920ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9921{
9922	struct scsi_vpd_mode_page_policy *mpp_ptr;
9923	struct ctl_lun *lun;
9924	int data_len;
9925
9926	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9927
9928	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9929	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9930
9931	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9932	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9933	ctsio->kern_sg_entries = 0;
9934
9935	if (data_len < alloc_len) {
9936		ctsio->residual = alloc_len - data_len;
9937		ctsio->kern_data_len = data_len;
9938		ctsio->kern_total_len = data_len;
9939	} else {
9940		ctsio->residual = 0;
9941		ctsio->kern_data_len = alloc_len;
9942		ctsio->kern_total_len = alloc_len;
9943	}
9944	ctsio->kern_data_resid = 0;
9945	ctsio->kern_rel_offset = 0;
9946	ctsio->kern_sg_entries = 0;
9947
9948	/*
9949	 * The control device is always connected.  The disk device, on the
9950	 * other hand, may not be online all the time.
9951	 */
9952	if (lun != NULL)
9953		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9954				     lun->be_lun->lun_type;
9955	else
9956		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9957	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9958	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9959	mpp_ptr->descr[0].page_code = 0x3f;
9960	mpp_ptr->descr[0].subpage_code = 0xff;
9961	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9962
9963	ctsio->scsi_status = SCSI_STATUS_OK;
9964	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9965	ctsio->be_move_done = ctl_config_move_done;
9966	ctl_datamove((union ctl_io *)ctsio);
9967
9968	return (CTL_RETVAL_COMPLETE);
9969}
9970
9971static int
9972ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9973{
9974	struct scsi_vpd_device_id *devid_ptr;
9975	struct scsi_vpd_id_descriptor *desc;
9976	struct ctl_softc *ctl_softc;
9977	struct ctl_lun *lun;
9978	struct ctl_port *port;
9979	int data_len;
9980	uint8_t proto;
9981
9982	ctl_softc = control_softc;
9983
9984	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9985	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9986
9987	data_len = sizeof(struct scsi_vpd_device_id) +
9988	    sizeof(struct scsi_vpd_id_descriptor) +
9989		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9990	    sizeof(struct scsi_vpd_id_descriptor) +
9991		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9992	if (lun && lun->lun_devid)
9993		data_len += lun->lun_devid->len;
9994	if (port->port_devid)
9995		data_len += port->port_devid->len;
9996	if (port->target_devid)
9997		data_len += port->target_devid->len;
9998
9999	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10000	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
10001	ctsio->kern_sg_entries = 0;
10002
10003	if (data_len < alloc_len) {
10004		ctsio->residual = alloc_len - data_len;
10005		ctsio->kern_data_len = data_len;
10006		ctsio->kern_total_len = data_len;
10007	} else {
10008		ctsio->residual = 0;
10009		ctsio->kern_data_len = alloc_len;
10010		ctsio->kern_total_len = alloc_len;
10011	}
10012	ctsio->kern_data_resid = 0;
10013	ctsio->kern_rel_offset = 0;
10014	ctsio->kern_sg_entries = 0;
10015
10016	/*
10017	 * The control device is always connected.  The disk device, on the
10018	 * other hand, may not be online all the time.
10019	 */
10020	if (lun != NULL)
10021		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10022				     lun->be_lun->lun_type;
10023	else
10024		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10025	devid_ptr->page_code = SVPD_DEVICE_ID;
10026	scsi_ulto2b(data_len - 4, devid_ptr->length);
10027
10028	if (port->port_type == CTL_PORT_FC)
10029		proto = SCSI_PROTO_FC << 4;
10030	else if (port->port_type == CTL_PORT_ISCSI)
10031		proto = SCSI_PROTO_ISCSI << 4;
10032	else
10033		proto = SCSI_PROTO_SPI << 4;
10034	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
10035
10036	/*
10037	 * We're using a LUN association here.  i.e., this device ID is a
10038	 * per-LUN identifier.
10039	 */
10040	if (lun && lun->lun_devid) {
10041		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
10042		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10043		    lun->lun_devid->len);
10044	}
10045
10046	/*
10047	 * This is for the WWPN which is a port association.
10048	 */
10049	if (port->port_devid) {
10050		memcpy(desc, port->port_devid->data, port->port_devid->len);
10051		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
10052		    port->port_devid->len);
10053	}
10054
10055	/*
10056	 * This is for the Relative Target Port(type 4h) identifier
10057	 */
10058	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10059	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10060	    SVPD_ID_TYPE_RELTARG;
10061	desc->length = 4;
10062	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
10063	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10064	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
10065
10066	/*
10067	 * This is for the Target Port Group(type 5h) identifier
10068	 */
10069	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
10070	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
10071	    SVPD_ID_TYPE_TPORTGRP;
10072	desc->length = 4;
10073	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
10074	    &desc->identifier[2]);
10075	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
10076	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
10077
10078	/*
10079	 * This is for the Target identifier
10080	 */
10081	if (port->target_devid) {
10082		memcpy(desc, port->target_devid->data, port->target_devid->len);
10083	}
10084
10085	ctsio->scsi_status = SCSI_STATUS_OK;
10086	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10087	ctsio->be_move_done = ctl_config_move_done;
10088	ctl_datamove((union ctl_io *)ctsio);
10089
10090	return (CTL_RETVAL_COMPLETE);
10091}
10092
10093static int
10094ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
10095{
10096	struct ctl_softc *softc = control_softc;
10097	struct scsi_vpd_scsi_ports *sp;
10098	struct scsi_vpd_port_designation *pd;
10099	struct scsi_vpd_port_designation_cont *pdc;
10100	struct ctl_lun *lun;
10101	struct ctl_port *port;
10102	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
10103	int num_target_port_groups, single;
10104
10105	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10106
10107	single = ctl_is_single;
10108	if (single)
10109		num_target_port_groups = 1;
10110	else
10111		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10112	num_target_ports = 0;
10113	iid_len = 0;
10114	id_len = 0;
10115	mtx_lock(&softc->ctl_lock);
10116	STAILQ_FOREACH(port, &softc->port_list, links) {
10117		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10118			continue;
10119		if (lun != NULL &&
10120		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10121		    CTL_MAX_LUNS)
10122			continue;
10123		num_target_ports++;
10124		if (port->init_devid)
10125			iid_len += port->init_devid->len;
10126		if (port->port_devid)
10127			id_len += port->port_devid->len;
10128	}
10129	mtx_unlock(&softc->ctl_lock);
10130
10131	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10132	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10133	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10134	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10135	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10136	ctsio->kern_sg_entries = 0;
10137
10138	if (data_len < alloc_len) {
10139		ctsio->residual = alloc_len - data_len;
10140		ctsio->kern_data_len = data_len;
10141		ctsio->kern_total_len = data_len;
10142	} else {
10143		ctsio->residual = 0;
10144		ctsio->kern_data_len = alloc_len;
10145		ctsio->kern_total_len = alloc_len;
10146	}
10147	ctsio->kern_data_resid = 0;
10148	ctsio->kern_rel_offset = 0;
10149	ctsio->kern_sg_entries = 0;
10150
10151	/*
10152	 * The control device is always connected.  The disk device, on the
10153	 * other hand, may not be online all the time.  Need to change this
10154	 * to figure out whether the disk device is actually online or not.
10155	 */
10156	if (lun != NULL)
10157		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10158				  lun->be_lun->lun_type;
10159	else
10160		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10161
10162	sp->page_code = SVPD_SCSI_PORTS;
10163	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10164	    sp->page_length);
10165	pd = &sp->design[0];
10166
10167	mtx_lock(&softc->ctl_lock);
10168	if (softc->flags & CTL_FLAG_MASTER_SHELF)
10169		pg = 0;
10170	else
10171		pg = 1;
10172	for (g = 0; g < num_target_port_groups; g++) {
10173		STAILQ_FOREACH(port, &softc->port_list, links) {
10174			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10175				continue;
10176			if (lun != NULL &&
10177			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10178			    CTL_MAX_LUNS)
10179				continue;
10180			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10181			scsi_ulto2b(p, pd->relative_port_id);
10182			if (port->init_devid && g == pg) {
10183				iid_len = port->init_devid->len;
10184				memcpy(pd->initiator_transportid,
10185				    port->init_devid->data, port->init_devid->len);
10186			} else
10187				iid_len = 0;
10188			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10189			pdc = (struct scsi_vpd_port_designation_cont *)
10190			    (&pd->initiator_transportid[iid_len]);
10191			if (port->port_devid && g == pg) {
10192				id_len = port->port_devid->len;
10193				memcpy(pdc->target_port_descriptors,
10194				    port->port_devid->data, port->port_devid->len);
10195			} else
10196				id_len = 0;
10197			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10198			pd = (struct scsi_vpd_port_designation *)
10199			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10200		}
10201	}
10202	mtx_unlock(&softc->ctl_lock);
10203
10204	ctsio->scsi_status = SCSI_STATUS_OK;
10205	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10206	ctsio->be_move_done = ctl_config_move_done;
10207	ctl_datamove((union ctl_io *)ctsio);
10208
10209	return (CTL_RETVAL_COMPLETE);
10210}
10211
10212static int
10213ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10214{
10215	struct scsi_vpd_block_limits *bl_ptr;
10216	struct ctl_lun *lun;
10217	int bs;
10218
10219	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10220
10221	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10222	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10223	ctsio->kern_sg_entries = 0;
10224
10225	if (sizeof(*bl_ptr) < alloc_len) {
10226		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10227		ctsio->kern_data_len = sizeof(*bl_ptr);
10228		ctsio->kern_total_len = sizeof(*bl_ptr);
10229	} else {
10230		ctsio->residual = 0;
10231		ctsio->kern_data_len = alloc_len;
10232		ctsio->kern_total_len = alloc_len;
10233	}
10234	ctsio->kern_data_resid = 0;
10235	ctsio->kern_rel_offset = 0;
10236	ctsio->kern_sg_entries = 0;
10237
10238	/*
10239	 * The control device is always connected.  The disk device, on the
10240	 * other hand, may not be online all the time.  Need to change this
10241	 * to figure out whether the disk device is actually online or not.
10242	 */
10243	if (lun != NULL)
10244		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10245				  lun->be_lun->lun_type;
10246	else
10247		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10248
10249	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10250	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
10251	bl_ptr->max_cmp_write_len = 0xff;
10252	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10253	if (lun != NULL) {
10254		bs = lun->be_lun->blocksize;
10255		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10256		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10257			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10258			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10259			if (lun->be_lun->pblockexp != 0) {
10260				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10261				    bl_ptr->opt_unmap_grain);
10262				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10263				    bl_ptr->unmap_grain_align);
10264			}
10265		}
10266	}
10267	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10268
10269	ctsio->scsi_status = SCSI_STATUS_OK;
10270	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10271	ctsio->be_move_done = ctl_config_move_done;
10272	ctl_datamove((union ctl_io *)ctsio);
10273
10274	return (CTL_RETVAL_COMPLETE);
10275}
10276
10277static int
10278ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10279{
10280	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10281	struct ctl_lun *lun;
10282
10283	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10284
10285	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10286	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10287	ctsio->kern_sg_entries = 0;
10288
10289	if (sizeof(*bdc_ptr) < alloc_len) {
10290		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10291		ctsio->kern_data_len = sizeof(*bdc_ptr);
10292		ctsio->kern_total_len = sizeof(*bdc_ptr);
10293	} else {
10294		ctsio->residual = 0;
10295		ctsio->kern_data_len = alloc_len;
10296		ctsio->kern_total_len = alloc_len;
10297	}
10298	ctsio->kern_data_resid = 0;
10299	ctsio->kern_rel_offset = 0;
10300	ctsio->kern_sg_entries = 0;
10301
10302	/*
10303	 * The control device is always connected.  The disk device, on the
10304	 * other hand, may not be online all the time.  Need to change this
10305	 * to figure out whether the disk device is actually online or not.
10306	 */
10307	if (lun != NULL)
10308		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10309				  lun->be_lun->lun_type;
10310	else
10311		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10312	bdc_ptr->page_code = SVPD_BDC;
10313	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10314	scsi_ulto2b(SVPD_NON_ROTATING, bdc_ptr->medium_rotation_rate);
10315	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10316
10317	ctsio->scsi_status = SCSI_STATUS_OK;
10318	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10319	ctsio->be_move_done = ctl_config_move_done;
10320	ctl_datamove((union ctl_io *)ctsio);
10321
10322	return (CTL_RETVAL_COMPLETE);
10323}
10324
10325static int
10326ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10327{
10328	struct scsi_vpd_logical_block_prov *lbp_ptr;
10329	struct ctl_lun *lun;
10330
10331	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10332
10333	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10334	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10335	ctsio->kern_sg_entries = 0;
10336
10337	if (sizeof(*lbp_ptr) < alloc_len) {
10338		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10339		ctsio->kern_data_len = sizeof(*lbp_ptr);
10340		ctsio->kern_total_len = sizeof(*lbp_ptr);
10341	} else {
10342		ctsio->residual = 0;
10343		ctsio->kern_data_len = alloc_len;
10344		ctsio->kern_total_len = alloc_len;
10345	}
10346	ctsio->kern_data_resid = 0;
10347	ctsio->kern_rel_offset = 0;
10348	ctsio->kern_sg_entries = 0;
10349
10350	/*
10351	 * The control device is always connected.  The disk device, on the
10352	 * other hand, may not be online all the time.  Need to change this
10353	 * to figure out whether the disk device is actually online or not.
10354	 */
10355	if (lun != NULL)
10356		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10357				  lun->be_lun->lun_type;
10358	else
10359		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10360
10361	lbp_ptr->page_code = SVPD_LBP;
10362	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10363	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10364		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10365		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10366		lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10367	}
10368
10369	ctsio->scsi_status = SCSI_STATUS_OK;
10370	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10371	ctsio->be_move_done = ctl_config_move_done;
10372	ctl_datamove((union ctl_io *)ctsio);
10373
10374	return (CTL_RETVAL_COMPLETE);
10375}
10376
10377static int
10378ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10379{
10380	struct scsi_inquiry *cdb;
10381	struct ctl_lun *lun;
10382	int alloc_len, retval;
10383
10384	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10385	cdb = (struct scsi_inquiry *)ctsio->cdb;
10386
10387	retval = CTL_RETVAL_COMPLETE;
10388
10389	alloc_len = scsi_2btoul(cdb->length);
10390
10391	switch (cdb->page_code) {
10392	case SVPD_SUPPORTED_PAGES:
10393		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10394		break;
10395	case SVPD_UNIT_SERIAL_NUMBER:
10396		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10397		break;
10398	case SVPD_DEVICE_ID:
10399		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10400		break;
10401	case SVPD_MODE_PAGE_POLICY:
10402		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10403		break;
10404	case SVPD_SCSI_PORTS:
10405		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10406		break;
10407	case SVPD_SCSI_TPC:
10408		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10409		break;
10410	case SVPD_BLOCK_LIMITS:
10411		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10412		break;
10413	case SVPD_BDC:
10414		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10415		break;
10416	case SVPD_LBP:
10417		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10418		break;
10419	default:
10420		ctl_set_invalid_field(ctsio,
10421				      /*sks_valid*/ 1,
10422				      /*command*/ 1,
10423				      /*field*/ 2,
10424				      /*bit_valid*/ 0,
10425				      /*bit*/ 0);
10426		ctl_done((union ctl_io *)ctsio);
10427		retval = CTL_RETVAL_COMPLETE;
10428		break;
10429	}
10430
10431	return (retval);
10432}
10433
10434static int
10435ctl_inquiry_std(struct ctl_scsiio *ctsio)
10436{
10437	struct scsi_inquiry_data *inq_ptr;
10438	struct scsi_inquiry *cdb;
10439	struct ctl_softc *ctl_softc;
10440	struct ctl_lun *lun;
10441	char *val;
10442	uint32_t alloc_len;
10443	ctl_port_type port_type;
10444
10445	ctl_softc = control_softc;
10446
10447	/*
10448	 * Figure out whether we're talking to a Fibre Channel port or not.
10449	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10450	 * SCSI front ends.
10451	 */
10452	port_type = ctl_softc->ctl_ports[
10453	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10454	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10455		port_type = CTL_PORT_SCSI;
10456
10457	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10458	cdb = (struct scsi_inquiry *)ctsio->cdb;
10459	alloc_len = scsi_2btoul(cdb->length);
10460
10461	/*
10462	 * We malloc the full inquiry data size here and fill it
10463	 * in.  If the user only asks for less, we'll give him
10464	 * that much.
10465	 */
10466	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10467	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10468	ctsio->kern_sg_entries = 0;
10469	ctsio->kern_data_resid = 0;
10470	ctsio->kern_rel_offset = 0;
10471
10472	if (sizeof(*inq_ptr) < alloc_len) {
10473		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10474		ctsio->kern_data_len = sizeof(*inq_ptr);
10475		ctsio->kern_total_len = sizeof(*inq_ptr);
10476	} else {
10477		ctsio->residual = 0;
10478		ctsio->kern_data_len = alloc_len;
10479		ctsio->kern_total_len = alloc_len;
10480	}
10481
10482	/*
10483	 * If we have a LUN configured, report it as connected.  Otherwise,
10484	 * report that it is offline or no device is supported, depending
10485	 * on the value of inquiry_pq_no_lun.
10486	 *
10487	 * According to the spec (SPC-4 r34), the peripheral qualifier
10488	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10489	 *
10490	 * "A peripheral device having the specified peripheral device type
10491	 * is not connected to this logical unit. However, the device
10492	 * server is capable of supporting the specified peripheral device
10493	 * type on this logical unit."
10494	 *
10495	 * According to the same spec, the peripheral qualifier
10496	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10497	 *
10498	 * "The device server is not capable of supporting a peripheral
10499	 * device on this logical unit. For this peripheral qualifier the
10500	 * peripheral device type shall be set to 1Fh. All other peripheral
10501	 * device type values are reserved for this peripheral qualifier."
10502	 *
10503	 * Given the text, it would seem that we probably want to report that
10504	 * the LUN is offline here.  There is no LUN connected, but we can
10505	 * support a LUN at the given LUN number.
10506	 *
10507	 * In the real world, though, it sounds like things are a little
10508	 * different:
10509	 *
10510	 * - Linux, when presented with a LUN with the offline peripheral
10511	 *   qualifier, will create an sg driver instance for it.  So when
10512	 *   you attach it to CTL, you wind up with a ton of sg driver
10513	 *   instances.  (One for every LUN that Linux bothered to probe.)
10514	 *   Linux does this despite the fact that it issues a REPORT LUNs
10515	 *   to LUN 0 to get the inventory of supported LUNs.
10516	 *
10517	 * - There is other anecdotal evidence (from Emulex folks) about
10518	 *   arrays that use the offline peripheral qualifier for LUNs that
10519	 *   are on the "passive" path in an active/passive array.
10520	 *
10521	 * So the solution is provide a hopefully reasonable default
10522	 * (return bad/no LUN) and allow the user to change the behavior
10523	 * with a tunable/sysctl variable.
10524	 */
10525	if (lun != NULL)
10526		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10527				  lun->be_lun->lun_type;
10528	else if (ctl_softc->inquiry_pq_no_lun == 0)
10529		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10530	else
10531		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10532
10533	/* RMB in byte 2 is 0 */
10534	inq_ptr->version = SCSI_REV_SPC4;
10535
10536	/*
10537	 * According to SAM-3, even if a device only supports a single
10538	 * level of LUN addressing, it should still set the HISUP bit:
10539	 *
10540	 * 4.9.1 Logical unit numbers overview
10541	 *
10542	 * All logical unit number formats described in this standard are
10543	 * hierarchical in structure even when only a single level in that
10544	 * hierarchy is used. The HISUP bit shall be set to one in the
10545	 * standard INQUIRY data (see SPC-2) when any logical unit number
10546	 * format described in this standard is used.  Non-hierarchical
10547	 * formats are outside the scope of this standard.
10548	 *
10549	 * Therefore we set the HiSup bit here.
10550	 *
10551	 * The reponse format is 2, per SPC-3.
10552	 */
10553	inq_ptr->response_format = SID_HiSup | 2;
10554
10555	inq_ptr->additional_length =
10556	    offsetof(struct scsi_inquiry_data, vendor_specific1) -
10557	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10558	CTL_DEBUG_PRINT(("additional_length = %d\n",
10559			 inq_ptr->additional_length));
10560
10561	inq_ptr->spc3_flags = SPC3_SID_3PC;
10562	if (!ctl_is_single)
10563		inq_ptr->spc3_flags |= SPC3_SID_TPGS_IMPLICIT;
10564	/* 16 bit addressing */
10565	if (port_type == CTL_PORT_SCSI)
10566		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10567	/* XXX set the SID_MultiP bit here if we're actually going to
10568	   respond on multiple ports */
10569	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10570
10571	/* 16 bit data bus, synchronous transfers */
10572	if (port_type == CTL_PORT_SCSI)
10573		inq_ptr->flags = SID_WBus16 | SID_Sync;
10574	/*
10575	 * XXX KDM do we want to support tagged queueing on the control
10576	 * device at all?
10577	 */
10578	if ((lun == NULL)
10579	 || (lun->be_lun->lun_type != T_PROCESSOR))
10580		inq_ptr->flags |= SID_CmdQue;
10581	/*
10582	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10583	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10584	 * name and 4 bytes for the revision.
10585	 */
10586	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10587	    "vendor")) == NULL) {
10588		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10589	} else {
10590		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10591		strncpy(inq_ptr->vendor, val,
10592		    min(sizeof(inq_ptr->vendor), strlen(val)));
10593	}
10594	if (lun == NULL) {
10595		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10596		    sizeof(inq_ptr->product));
10597	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10598		switch (lun->be_lun->lun_type) {
10599		case T_DIRECT:
10600			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10601			    sizeof(inq_ptr->product));
10602			break;
10603		case T_PROCESSOR:
10604			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10605			    sizeof(inq_ptr->product));
10606			break;
10607		default:
10608			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10609			    sizeof(inq_ptr->product));
10610			break;
10611		}
10612	} else {
10613		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10614		strncpy(inq_ptr->product, val,
10615		    min(sizeof(inq_ptr->product), strlen(val)));
10616	}
10617
10618	/*
10619	 * XXX make this a macro somewhere so it automatically gets
10620	 * incremented when we make changes.
10621	 */
10622	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10623	    "revision")) == NULL) {
10624		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10625	} else {
10626		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10627		strncpy(inq_ptr->revision, val,
10628		    min(sizeof(inq_ptr->revision), strlen(val)));
10629	}
10630
10631	/*
10632	 * For parallel SCSI, we support double transition and single
10633	 * transition clocking.  We also support QAS (Quick Arbitration
10634	 * and Selection) and Information Unit transfers on both the
10635	 * control and array devices.
10636	 */
10637	if (port_type == CTL_PORT_SCSI)
10638		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10639				    SID_SPI_IUS;
10640
10641	/* SAM-5 (no version claimed) */
10642	scsi_ulto2b(0x00A0, inq_ptr->version1);
10643	/* SPC-4 (no version claimed) */
10644	scsi_ulto2b(0x0460, inq_ptr->version2);
10645	if (port_type == CTL_PORT_FC) {
10646		/* FCP-2 ANSI INCITS.350:2003 */
10647		scsi_ulto2b(0x0917, inq_ptr->version3);
10648	} else if (port_type == CTL_PORT_SCSI) {
10649		/* SPI-4 ANSI INCITS.362:200x */
10650		scsi_ulto2b(0x0B56, inq_ptr->version3);
10651	} else if (port_type == CTL_PORT_ISCSI) {
10652		/* iSCSI (no version claimed) */
10653		scsi_ulto2b(0x0960, inq_ptr->version3);
10654	} else if (port_type == CTL_PORT_SAS) {
10655		/* SAS (no version claimed) */
10656		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10657	}
10658
10659	if (lun == NULL) {
10660		/* SBC-3 (no version claimed) */
10661		scsi_ulto2b(0x04C0, inq_ptr->version4);
10662	} else {
10663		switch (lun->be_lun->lun_type) {
10664		case T_DIRECT:
10665			/* SBC-3 (no version claimed) */
10666			scsi_ulto2b(0x04C0, inq_ptr->version4);
10667			break;
10668		case T_PROCESSOR:
10669		default:
10670			break;
10671		}
10672	}
10673
10674	ctsio->scsi_status = SCSI_STATUS_OK;
10675	if (ctsio->kern_data_len > 0) {
10676		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10677		ctsio->be_move_done = ctl_config_move_done;
10678		ctl_datamove((union ctl_io *)ctsio);
10679	} else {
10680		ctsio->io_hdr.status = CTL_SUCCESS;
10681		ctl_done((union ctl_io *)ctsio);
10682	}
10683
10684	return (CTL_RETVAL_COMPLETE);
10685}
10686
10687int
10688ctl_inquiry(struct ctl_scsiio *ctsio)
10689{
10690	struct scsi_inquiry *cdb;
10691	int retval;
10692
10693	cdb = (struct scsi_inquiry *)ctsio->cdb;
10694
10695	retval = 0;
10696
10697	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10698
10699	/*
10700	 * Right now, we don't support the CmdDt inquiry information.
10701	 * This would be nice to support in the future.  When we do
10702	 * support it, we should change this test so that it checks to make
10703	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10704	 */
10705#ifdef notyet
10706	if (((cdb->byte2 & SI_EVPD)
10707	 && (cdb->byte2 & SI_CMDDT)))
10708#endif
10709	if (cdb->byte2 & SI_CMDDT) {
10710		/*
10711		 * Point to the SI_CMDDT bit.  We might change this
10712		 * when we support SI_CMDDT, but since both bits would be
10713		 * "wrong", this should probably just stay as-is then.
10714		 */
10715		ctl_set_invalid_field(ctsio,
10716				      /*sks_valid*/ 1,
10717				      /*command*/ 1,
10718				      /*field*/ 1,
10719				      /*bit_valid*/ 1,
10720				      /*bit*/ 1);
10721		ctl_done((union ctl_io *)ctsio);
10722		return (CTL_RETVAL_COMPLETE);
10723	}
10724	if (cdb->byte2 & SI_EVPD)
10725		retval = ctl_inquiry_evpd(ctsio);
10726#ifdef notyet
10727	else if (cdb->byte2 & SI_CMDDT)
10728		retval = ctl_inquiry_cmddt(ctsio);
10729#endif
10730	else
10731		retval = ctl_inquiry_std(ctsio);
10732
10733	return (retval);
10734}
10735
10736/*
10737 * For known CDB types, parse the LBA and length.
10738 */
10739static int
10740ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10741{
10742	if (io->io_hdr.io_type != CTL_IO_SCSI)
10743		return (1);
10744
10745	switch (io->scsiio.cdb[0]) {
10746	case COMPARE_AND_WRITE: {
10747		struct scsi_compare_and_write *cdb;
10748
10749		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10750
10751		*lba = scsi_8btou64(cdb->addr);
10752		*len = cdb->length;
10753		break;
10754	}
10755	case READ_6:
10756	case WRITE_6: {
10757		struct scsi_rw_6 *cdb;
10758
10759		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10760
10761		*lba = scsi_3btoul(cdb->addr);
10762		/* only 5 bits are valid in the most significant address byte */
10763		*lba &= 0x1fffff;
10764		*len = cdb->length;
10765		break;
10766	}
10767	case READ_10:
10768	case WRITE_10: {
10769		struct scsi_rw_10 *cdb;
10770
10771		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10772
10773		*lba = scsi_4btoul(cdb->addr);
10774		*len = scsi_2btoul(cdb->length);
10775		break;
10776	}
10777	case WRITE_VERIFY_10: {
10778		struct scsi_write_verify_10 *cdb;
10779
10780		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10781
10782		*lba = scsi_4btoul(cdb->addr);
10783		*len = scsi_2btoul(cdb->length);
10784		break;
10785	}
10786	case READ_12:
10787	case WRITE_12: {
10788		struct scsi_rw_12 *cdb;
10789
10790		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10791
10792		*lba = scsi_4btoul(cdb->addr);
10793		*len = scsi_4btoul(cdb->length);
10794		break;
10795	}
10796	case WRITE_VERIFY_12: {
10797		struct scsi_write_verify_12 *cdb;
10798
10799		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10800
10801		*lba = scsi_4btoul(cdb->addr);
10802		*len = scsi_4btoul(cdb->length);
10803		break;
10804	}
10805	case READ_16:
10806	case WRITE_16: {
10807		struct scsi_rw_16 *cdb;
10808
10809		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10810
10811		*lba = scsi_8btou64(cdb->addr);
10812		*len = scsi_4btoul(cdb->length);
10813		break;
10814	}
10815	case WRITE_VERIFY_16: {
10816		struct scsi_write_verify_16 *cdb;
10817
10818		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10819
10820
10821		*lba = scsi_8btou64(cdb->addr);
10822		*len = scsi_4btoul(cdb->length);
10823		break;
10824	}
10825	case WRITE_SAME_10: {
10826		struct scsi_write_same_10 *cdb;
10827
10828		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10829
10830		*lba = scsi_4btoul(cdb->addr);
10831		*len = scsi_2btoul(cdb->length);
10832		break;
10833	}
10834	case WRITE_SAME_16: {
10835		struct scsi_write_same_16 *cdb;
10836
10837		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10838
10839		*lba = scsi_8btou64(cdb->addr);
10840		*len = scsi_4btoul(cdb->length);
10841		break;
10842	}
10843	case VERIFY_10: {
10844		struct scsi_verify_10 *cdb;
10845
10846		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10847
10848		*lba = scsi_4btoul(cdb->addr);
10849		*len = scsi_2btoul(cdb->length);
10850		break;
10851	}
10852	case VERIFY_12: {
10853		struct scsi_verify_12 *cdb;
10854
10855		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10856
10857		*lba = scsi_4btoul(cdb->addr);
10858		*len = scsi_4btoul(cdb->length);
10859		break;
10860	}
10861	case VERIFY_16: {
10862		struct scsi_verify_16 *cdb;
10863
10864		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10865
10866		*lba = scsi_8btou64(cdb->addr);
10867		*len = scsi_4btoul(cdb->length);
10868		break;
10869	}
10870	default:
10871		return (1);
10872		break; /* NOTREACHED */
10873	}
10874
10875	return (0);
10876}
10877
10878static ctl_action
10879ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10880{
10881	uint64_t endlba1, endlba2;
10882
10883	endlba1 = lba1 + len1 - 1;
10884	endlba2 = lba2 + len2 - 1;
10885
10886	if ((endlba1 < lba2)
10887	 || (endlba2 < lba1))
10888		return (CTL_ACTION_PASS);
10889	else
10890		return (CTL_ACTION_BLOCK);
10891}
10892
10893static ctl_action
10894ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10895{
10896	uint64_t lba1, lba2;
10897	uint32_t len1, len2;
10898	int retval;
10899
10900	retval = ctl_get_lba_len(io1, &lba1, &len1);
10901	if (retval != 0)
10902		return (CTL_ACTION_ERROR);
10903
10904	retval = ctl_get_lba_len(io2, &lba2, &len2);
10905	if (retval != 0)
10906		return (CTL_ACTION_ERROR);
10907
10908	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10909}
10910
10911static ctl_action
10912ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10913{
10914	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10915	ctl_serialize_action *serialize_row;
10916
10917	/*
10918	 * The initiator attempted multiple untagged commands at the same
10919	 * time.  Can't do that.
10920	 */
10921	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10922	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10923	 && ((pending_io->io_hdr.nexus.targ_port ==
10924	      ooa_io->io_hdr.nexus.targ_port)
10925	  && (pending_io->io_hdr.nexus.initid.id ==
10926	      ooa_io->io_hdr.nexus.initid.id))
10927	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10928		return (CTL_ACTION_OVERLAP);
10929
10930	/*
10931	 * The initiator attempted to send multiple tagged commands with
10932	 * the same ID.  (It's fine if different initiators have the same
10933	 * tag ID.)
10934	 *
10935	 * Even if all of those conditions are true, we don't kill the I/O
10936	 * if the command ahead of us has been aborted.  We won't end up
10937	 * sending it to the FETD, and it's perfectly legal to resend a
10938	 * command with the same tag number as long as the previous
10939	 * instance of this tag number has been aborted somehow.
10940	 */
10941	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10942	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10943	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10944	 && ((pending_io->io_hdr.nexus.targ_port ==
10945	      ooa_io->io_hdr.nexus.targ_port)
10946	  && (pending_io->io_hdr.nexus.initid.id ==
10947	      ooa_io->io_hdr.nexus.initid.id))
10948	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10949		return (CTL_ACTION_OVERLAP_TAG);
10950
10951	/*
10952	 * If we get a head of queue tag, SAM-3 says that we should
10953	 * immediately execute it.
10954	 *
10955	 * What happens if this command would normally block for some other
10956	 * reason?  e.g. a request sense with a head of queue tag
10957	 * immediately after a write.  Normally that would block, but this
10958	 * will result in its getting executed immediately...
10959	 *
10960	 * We currently return "pass" instead of "skip", so we'll end up
10961	 * going through the rest of the queue to check for overlapped tags.
10962	 *
10963	 * XXX KDM check for other types of blockage first??
10964	 */
10965	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10966		return (CTL_ACTION_PASS);
10967
10968	/*
10969	 * Ordered tags have to block until all items ahead of them
10970	 * have completed.  If we get called with an ordered tag, we always
10971	 * block, if something else is ahead of us in the queue.
10972	 */
10973	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10974		return (CTL_ACTION_BLOCK);
10975
10976	/*
10977	 * Simple tags get blocked until all head of queue and ordered tags
10978	 * ahead of them have completed.  I'm lumping untagged commands in
10979	 * with simple tags here.  XXX KDM is that the right thing to do?
10980	 */
10981	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10982	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10983	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10984	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10985		return (CTL_ACTION_BLOCK);
10986
10987	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10988	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10989
10990	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10991
10992	switch (serialize_row[pending_entry->seridx]) {
10993	case CTL_SER_BLOCK:
10994		return (CTL_ACTION_BLOCK);
10995		break; /* NOTREACHED */
10996	case CTL_SER_EXTENT:
10997		return (ctl_extent_check(pending_io, ooa_io));
10998		break; /* NOTREACHED */
10999	case CTL_SER_PASS:
11000		return (CTL_ACTION_PASS);
11001		break; /* NOTREACHED */
11002	case CTL_SER_SKIP:
11003		return (CTL_ACTION_SKIP);
11004		break;
11005	default:
11006		panic("invalid serialization value %d",
11007		      serialize_row[pending_entry->seridx]);
11008		break; /* NOTREACHED */
11009	}
11010
11011	return (CTL_ACTION_ERROR);
11012}
11013
11014/*
11015 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11016 * Assumptions:
11017 * - pending_io is generally either incoming, or on the blocked queue
11018 * - starting I/O is the I/O we want to start the check with.
11019 */
11020static ctl_action
11021ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11022	      union ctl_io *starting_io)
11023{
11024	union ctl_io *ooa_io;
11025	ctl_action action;
11026
11027	mtx_assert(&lun->lun_lock, MA_OWNED);
11028
11029	/*
11030	 * Run back along the OOA queue, starting with the current
11031	 * blocked I/O and going through every I/O before it on the
11032	 * queue.  If starting_io is NULL, we'll just end up returning
11033	 * CTL_ACTION_PASS.
11034	 */
11035	for (ooa_io = starting_io; ooa_io != NULL;
11036	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11037	     ooa_links)){
11038
11039		/*
11040		 * This routine just checks to see whether
11041		 * cur_blocked is blocked by ooa_io, which is ahead
11042		 * of it in the queue.  It doesn't queue/dequeue
11043		 * cur_blocked.
11044		 */
11045		action = ctl_check_for_blockage(pending_io, ooa_io);
11046		switch (action) {
11047		case CTL_ACTION_BLOCK:
11048		case CTL_ACTION_OVERLAP:
11049		case CTL_ACTION_OVERLAP_TAG:
11050		case CTL_ACTION_SKIP:
11051		case CTL_ACTION_ERROR:
11052			return (action);
11053			break; /* NOTREACHED */
11054		case CTL_ACTION_PASS:
11055			break;
11056		default:
11057			panic("invalid action %d", action);
11058			break;  /* NOTREACHED */
11059		}
11060	}
11061
11062	return (CTL_ACTION_PASS);
11063}
11064
11065/*
11066 * Assumptions:
11067 * - An I/O has just completed, and has been removed from the per-LUN OOA
11068 *   queue, so some items on the blocked queue may now be unblocked.
11069 */
11070static int
11071ctl_check_blocked(struct ctl_lun *lun)
11072{
11073	union ctl_io *cur_blocked, *next_blocked;
11074
11075	mtx_assert(&lun->lun_lock, MA_OWNED);
11076
11077	/*
11078	 * Run forward from the head of the blocked queue, checking each
11079	 * entry against the I/Os prior to it on the OOA queue to see if
11080	 * there is still any blockage.
11081	 *
11082	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11083	 * with our removing a variable on it while it is traversing the
11084	 * list.
11085	 */
11086	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11087	     cur_blocked != NULL; cur_blocked = next_blocked) {
11088		union ctl_io *prev_ooa;
11089		ctl_action action;
11090
11091		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11092							  blocked_links);
11093
11094		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11095						      ctl_ooaq, ooa_links);
11096
11097		/*
11098		 * If cur_blocked happens to be the first item in the OOA
11099		 * queue now, prev_ooa will be NULL, and the action
11100		 * returned will just be CTL_ACTION_PASS.
11101		 */
11102		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11103
11104		switch (action) {
11105		case CTL_ACTION_BLOCK:
11106			/* Nothing to do here, still blocked */
11107			break;
11108		case CTL_ACTION_OVERLAP:
11109		case CTL_ACTION_OVERLAP_TAG:
11110			/*
11111			 * This shouldn't happen!  In theory we've already
11112			 * checked this command for overlap...
11113			 */
11114			break;
11115		case CTL_ACTION_PASS:
11116		case CTL_ACTION_SKIP: {
11117			struct ctl_softc *softc;
11118			const struct ctl_cmd_entry *entry;
11119			uint32_t initidx;
11120			int isc_retval;
11121
11122			/*
11123			 * The skip case shouldn't happen, this transaction
11124			 * should have never made it onto the blocked queue.
11125			 */
11126			/*
11127			 * This I/O is no longer blocked, we can remove it
11128			 * from the blocked queue.  Since this is a TAILQ
11129			 * (doubly linked list), we can do O(1) removals
11130			 * from any place on the list.
11131			 */
11132			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11133				     blocked_links);
11134			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11135
11136			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11137				/*
11138				 * Need to send IO back to original side to
11139				 * run
11140				 */
11141				union ctl_ha_msg msg_info;
11142
11143				msg_info.hdr.original_sc =
11144					cur_blocked->io_hdr.original_sc;
11145				msg_info.hdr.serializing_sc = cur_blocked;
11146				msg_info.hdr.msg_type = CTL_MSG_R2R;
11147				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11148				     &msg_info, sizeof(msg_info), 0)) >
11149				     CTL_HA_STATUS_SUCCESS) {
11150					printf("CTL:Check Blocked error from "
11151					       "ctl_ha_msg_send %d\n",
11152					       isc_retval);
11153				}
11154				break;
11155			}
11156			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
11157			softc = control_softc;
11158
11159			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11160
11161			/*
11162			 * Check this I/O for LUN state changes that may
11163			 * have happened while this command was blocked.
11164			 * The LUN state may have been changed by a command
11165			 * ahead of us in the queue, so we need to re-check
11166			 * for any states that can be caused by SCSI
11167			 * commands.
11168			 */
11169			if (ctl_scsiio_lun_check(softc, lun, entry,
11170						 &cur_blocked->scsiio) == 0) {
11171				cur_blocked->io_hdr.flags |=
11172				                      CTL_FLAG_IS_WAS_ON_RTR;
11173				ctl_enqueue_rtr(cur_blocked);
11174			} else
11175				ctl_done(cur_blocked);
11176			break;
11177		}
11178		default:
11179			/*
11180			 * This probably shouldn't happen -- we shouldn't
11181			 * get CTL_ACTION_ERROR, or anything else.
11182			 */
11183			break;
11184		}
11185	}
11186
11187	return (CTL_RETVAL_COMPLETE);
11188}
11189
11190/*
11191 * This routine (with one exception) checks LUN flags that can be set by
11192 * commands ahead of us in the OOA queue.  These flags have to be checked
11193 * when a command initially comes in, and when we pull a command off the
11194 * blocked queue and are preparing to execute it.  The reason we have to
11195 * check these flags for commands on the blocked queue is that the LUN
11196 * state may have been changed by a command ahead of us while we're on the
11197 * blocked queue.
11198 *
11199 * Ordering is somewhat important with these checks, so please pay
11200 * careful attention to the placement of any new checks.
11201 */
11202static int
11203ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11204    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11205{
11206	int retval;
11207
11208	retval = 0;
11209
11210	mtx_assert(&lun->lun_lock, MA_OWNED);
11211
11212	/*
11213	 * If this shelf is a secondary shelf controller, we have to reject
11214	 * any media access commands.
11215	 */
11216#if 0
11217	/* No longer needed for HA */
11218	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
11219	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
11220		ctl_set_lun_standby(ctsio);
11221		retval = 1;
11222		goto bailout;
11223	}
11224#endif
11225
11226	/*
11227	 * Check for a reservation conflict.  If this command isn't allowed
11228	 * even on reserved LUNs, and if this initiator isn't the one who
11229	 * reserved us, reject the command with a reservation conflict.
11230	 */
11231	if ((lun->flags & CTL_LUN_RESERVED)
11232	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11233		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
11234		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
11235		 || (ctsio->io_hdr.nexus.targ_target.id !=
11236		     lun->rsv_nexus.targ_target.id)) {
11237			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11238			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11239			retval = 1;
11240			goto bailout;
11241		}
11242	}
11243
11244	if ( (lun->flags & CTL_LUN_PR_RESERVED)
11245	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
11246		uint32_t residx;
11247
11248		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11249		/*
11250		 * if we aren't registered or it's a res holder type
11251		 * reservation and this isn't the res holder then set a
11252		 * conflict.
11253		 * NOTE: Commands which might be allowed on write exclusive
11254		 * type reservations are checked in the particular command
11255		 * for a conflict. Read and SSU are the only ones.
11256		 */
11257		if (!lun->per_res[residx].registered
11258		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11259			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
11260			ctsio->io_hdr.status = CTL_SCSI_ERROR;
11261			retval = 1;
11262			goto bailout;
11263		}
11264
11265	}
11266
11267	if ((lun->flags & CTL_LUN_OFFLINE)
11268	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11269		ctl_set_lun_not_ready(ctsio);
11270		retval = 1;
11271		goto bailout;
11272	}
11273
11274	/*
11275	 * If the LUN is stopped, see if this particular command is allowed
11276	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11277	 */
11278	if ((lun->flags & CTL_LUN_STOPPED)
11279	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11280		/* "Logical unit not ready, initializing cmd. required" */
11281		ctl_set_lun_stopped(ctsio);
11282		retval = 1;
11283		goto bailout;
11284	}
11285
11286	if ((lun->flags & CTL_LUN_INOPERABLE)
11287	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11288		/* "Medium format corrupted" */
11289		ctl_set_medium_format_corrupted(ctsio);
11290		retval = 1;
11291		goto bailout;
11292	}
11293
11294bailout:
11295	return (retval);
11296
11297}
11298
11299static void
11300ctl_failover_io(union ctl_io *io, int have_lock)
11301{
11302	ctl_set_busy(&io->scsiio);
11303	ctl_done(io);
11304}
11305
11306static void
11307ctl_failover(void)
11308{
11309	struct ctl_lun *lun;
11310	struct ctl_softc *ctl_softc;
11311	union ctl_io *next_io, *pending_io;
11312	union ctl_io *io;
11313	int lun_idx;
11314	int i;
11315
11316	ctl_softc = control_softc;
11317
11318	mtx_lock(&ctl_softc->ctl_lock);
11319	/*
11320	 * Remove any cmds from the other SC from the rtr queue.  These
11321	 * will obviously only be for LUNs for which we're the primary.
11322	 * We can't send status or get/send data for these commands.
11323	 * Since they haven't been executed yet, we can just remove them.
11324	 * We'll either abort them or delete them below, depending on
11325	 * which HA mode we're in.
11326	 */
11327#ifdef notyet
11328	mtx_lock(&ctl_softc->queue_lock);
11329	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11330	     io != NULL; io = next_io) {
11331		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11332		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11333			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11334				      ctl_io_hdr, links);
11335	}
11336	mtx_unlock(&ctl_softc->queue_lock);
11337#endif
11338
11339	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11340		lun = ctl_softc->ctl_luns[lun_idx];
11341		if (lun==NULL)
11342			continue;
11343
11344		/*
11345		 * Processor LUNs are primary on both sides.
11346		 * XXX will this always be true?
11347		 */
11348		if (lun->be_lun->lun_type == T_PROCESSOR)
11349			continue;
11350
11351		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11352		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11353			printf("FAILOVER: primary lun %d\n", lun_idx);
11354		        /*
11355			 * Remove all commands from the other SC. First from the
11356			 * blocked queue then from the ooa queue. Once we have
11357			 * removed them. Call ctl_check_blocked to see if there
11358			 * is anything that can run.
11359			 */
11360			for (io = (union ctl_io *)TAILQ_FIRST(
11361			     &lun->blocked_queue); io != NULL; io = next_io) {
11362
11363		        	next_io = (union ctl_io *)TAILQ_NEXT(
11364				    &io->io_hdr, blocked_links);
11365
11366				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11367					TAILQ_REMOVE(&lun->blocked_queue,
11368						     &io->io_hdr,blocked_links);
11369					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11370					TAILQ_REMOVE(&lun->ooa_queue,
11371						     &io->io_hdr, ooa_links);
11372
11373					ctl_free_io(io);
11374				}
11375			}
11376
11377			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11378	     		     io != NULL; io = next_io) {
11379
11380		        	next_io = (union ctl_io *)TAILQ_NEXT(
11381				    &io->io_hdr, ooa_links);
11382
11383				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11384
11385					TAILQ_REMOVE(&lun->ooa_queue,
11386						&io->io_hdr,
11387					     	ooa_links);
11388
11389					ctl_free_io(io);
11390				}
11391			}
11392			ctl_check_blocked(lun);
11393		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11394			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11395
11396			printf("FAILOVER: primary lun %d\n", lun_idx);
11397			/*
11398			 * Abort all commands from the other SC.  We can't
11399			 * send status back for them now.  These should get
11400			 * cleaned up when they are completed or come out
11401			 * for a datamove operation.
11402			 */
11403			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11404	     		     io != NULL; io = next_io) {
11405		        	next_io = (union ctl_io *)TAILQ_NEXT(
11406					&io->io_hdr, ooa_links);
11407
11408				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11409					io->io_hdr.flags |= CTL_FLAG_ABORT;
11410			}
11411		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11412			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11413
11414			printf("FAILOVER: secondary lun %d\n", lun_idx);
11415
11416			lun->flags |= CTL_LUN_PRIMARY_SC;
11417
11418			/*
11419			 * We send all I/O that was sent to this controller
11420			 * and redirected to the other side back with
11421			 * busy status, and have the initiator retry it.
11422			 * Figuring out how much data has been transferred,
11423			 * etc. and picking up where we left off would be
11424			 * very tricky.
11425			 *
11426			 * XXX KDM need to remove I/O from the blocked
11427			 * queue as well!
11428			 */
11429			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11430			     &lun->ooa_queue); pending_io != NULL;
11431			     pending_io = next_io) {
11432
11433				next_io =  (union ctl_io *)TAILQ_NEXT(
11434					&pending_io->io_hdr, ooa_links);
11435
11436				pending_io->io_hdr.flags &=
11437					~CTL_FLAG_SENT_2OTHER_SC;
11438
11439				if (pending_io->io_hdr.flags &
11440				    CTL_FLAG_IO_ACTIVE) {
11441					pending_io->io_hdr.flags |=
11442						CTL_FLAG_FAILOVER;
11443				} else {
11444					ctl_set_busy(&pending_io->scsiio);
11445					ctl_done(pending_io);
11446				}
11447			}
11448
11449			/*
11450			 * Build Unit Attention
11451			 */
11452			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11453				lun->pending_ua[i] |=
11454				                     CTL_UA_ASYM_ACC_CHANGE;
11455			}
11456		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11457			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11458			printf("FAILOVER: secondary lun %d\n", lun_idx);
11459			/*
11460			 * if the first io on the OOA is not on the RtR queue
11461			 * add it.
11462			 */
11463			lun->flags |= CTL_LUN_PRIMARY_SC;
11464
11465			pending_io = (union ctl_io *)TAILQ_FIRST(
11466			    &lun->ooa_queue);
11467			if (pending_io==NULL) {
11468				printf("Nothing on OOA queue\n");
11469				continue;
11470			}
11471
11472			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11473			if ((pending_io->io_hdr.flags &
11474			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11475				pending_io->io_hdr.flags |=
11476				    CTL_FLAG_IS_WAS_ON_RTR;
11477				ctl_enqueue_rtr(pending_io);
11478			}
11479#if 0
11480			else
11481			{
11482				printf("Tag 0x%04x is running\n",
11483				      pending_io->scsiio.tag_num);
11484			}
11485#endif
11486
11487			next_io = (union ctl_io *)TAILQ_NEXT(
11488			    &pending_io->io_hdr, ooa_links);
11489			for (pending_io=next_io; pending_io != NULL;
11490			     pending_io = next_io) {
11491				pending_io->io_hdr.flags &=
11492				    ~CTL_FLAG_SENT_2OTHER_SC;
11493				next_io = (union ctl_io *)TAILQ_NEXT(
11494					&pending_io->io_hdr, ooa_links);
11495				if (pending_io->io_hdr.flags &
11496				    CTL_FLAG_IS_WAS_ON_RTR) {
11497#if 0
11498				        printf("Tag 0x%04x is running\n",
11499				      		pending_io->scsiio.tag_num);
11500#endif
11501					continue;
11502				}
11503
11504				switch (ctl_check_ooa(lun, pending_io,
11505			            (union ctl_io *)TAILQ_PREV(
11506				    &pending_io->io_hdr, ctl_ooaq,
11507				    ooa_links))) {
11508
11509				case CTL_ACTION_BLOCK:
11510					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11511							  &pending_io->io_hdr,
11512							  blocked_links);
11513					pending_io->io_hdr.flags |=
11514					    CTL_FLAG_BLOCKED;
11515					break;
11516				case CTL_ACTION_PASS:
11517				case CTL_ACTION_SKIP:
11518					pending_io->io_hdr.flags |=
11519					    CTL_FLAG_IS_WAS_ON_RTR;
11520					ctl_enqueue_rtr(pending_io);
11521					break;
11522				case CTL_ACTION_OVERLAP:
11523					ctl_set_overlapped_cmd(
11524					    (struct ctl_scsiio *)pending_io);
11525					ctl_done(pending_io);
11526					break;
11527				case CTL_ACTION_OVERLAP_TAG:
11528					ctl_set_overlapped_tag(
11529					    (struct ctl_scsiio *)pending_io,
11530					    pending_io->scsiio.tag_num & 0xff);
11531					ctl_done(pending_io);
11532					break;
11533				case CTL_ACTION_ERROR:
11534				default:
11535					ctl_set_internal_failure(
11536						(struct ctl_scsiio *)pending_io,
11537						0,  // sks_valid
11538						0); //retry count
11539					ctl_done(pending_io);
11540					break;
11541				}
11542			}
11543
11544			/*
11545			 * Build Unit Attention
11546			 */
11547			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11548				lun->pending_ua[i] |=
11549				                     CTL_UA_ASYM_ACC_CHANGE;
11550			}
11551		} else {
11552			panic("Unhandled HA mode failover, LUN flags = %#x, "
11553			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11554		}
11555	}
11556	ctl_pause_rtr = 0;
11557	mtx_unlock(&ctl_softc->ctl_lock);
11558}
11559
11560static int
11561ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11562{
11563	struct ctl_lun *lun;
11564	const struct ctl_cmd_entry *entry;
11565	uint32_t initidx, targ_lun;
11566	int retval;
11567
11568	retval = 0;
11569
11570	lun = NULL;
11571
11572	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11573	if ((targ_lun < CTL_MAX_LUNS)
11574	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11575		lun = ctl_softc->ctl_luns[targ_lun];
11576		/*
11577		 * If the LUN is invalid, pretend that it doesn't exist.
11578		 * It will go away as soon as all pending I/O has been
11579		 * completed.
11580		 */
11581		if (lun->flags & CTL_LUN_DISABLED) {
11582			lun = NULL;
11583		} else {
11584			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11585			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11586				lun->be_lun;
11587			if (lun->be_lun->lun_type == T_PROCESSOR) {
11588				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11589			}
11590
11591			/*
11592			 * Every I/O goes into the OOA queue for a
11593			 * particular LUN, and stays there until completion.
11594			 */
11595			mtx_lock(&lun->lun_lock);
11596			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11597			    ooa_links);
11598		}
11599	} else {
11600		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11601		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11602	}
11603
11604	/* Get command entry and return error if it is unsuppotyed. */
11605	entry = ctl_validate_command(ctsio);
11606	if (entry == NULL) {
11607		if (lun)
11608			mtx_unlock(&lun->lun_lock);
11609		return (retval);
11610	}
11611
11612	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11613	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11614
11615	/*
11616	 * Check to see whether we can send this command to LUNs that don't
11617	 * exist.  This should pretty much only be the case for inquiry
11618	 * and request sense.  Further checks, below, really require having
11619	 * a LUN, so we can't really check the command anymore.  Just put
11620	 * it on the rtr queue.
11621	 */
11622	if (lun == NULL) {
11623		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11624			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11625			ctl_enqueue_rtr((union ctl_io *)ctsio);
11626			return (retval);
11627		}
11628
11629		ctl_set_unsupported_lun(ctsio);
11630		ctl_done((union ctl_io *)ctsio);
11631		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11632		return (retval);
11633	} else {
11634		/*
11635		 * Make sure we support this particular command on this LUN.
11636		 * e.g., we don't support writes to the control LUN.
11637		 */
11638		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11639			mtx_unlock(&lun->lun_lock);
11640			ctl_set_invalid_opcode(ctsio);
11641			ctl_done((union ctl_io *)ctsio);
11642			return (retval);
11643		}
11644	}
11645
11646	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11647
11648#ifdef CTL_WITH_CA
11649	/*
11650	 * If we've got a request sense, it'll clear the contingent
11651	 * allegiance condition.  Otherwise, if we have a CA condition for
11652	 * this initiator, clear it, because it sent down a command other
11653	 * than request sense.
11654	 */
11655	if ((ctsio->cdb[0] != REQUEST_SENSE)
11656	 && (ctl_is_set(lun->have_ca, initidx)))
11657		ctl_clear_mask(lun->have_ca, initidx);
11658#endif
11659
11660	/*
11661	 * If the command has this flag set, it handles its own unit
11662	 * attention reporting, we shouldn't do anything.  Otherwise we
11663	 * check for any pending unit attentions, and send them back to the
11664	 * initiator.  We only do this when a command initially comes in,
11665	 * not when we pull it off the blocked queue.
11666	 *
11667	 * According to SAM-3, section 5.3.2, the order that things get
11668	 * presented back to the host is basically unit attentions caused
11669	 * by some sort of reset event, busy status, reservation conflicts
11670	 * or task set full, and finally any other status.
11671	 *
11672	 * One issue here is that some of the unit attentions we report
11673	 * don't fall into the "reset" category (e.g. "reported luns data
11674	 * has changed").  So reporting it here, before the reservation
11675	 * check, may be technically wrong.  I guess the only thing to do
11676	 * would be to check for and report the reset events here, and then
11677	 * check for the other unit attention types after we check for a
11678	 * reservation conflict.
11679	 *
11680	 * XXX KDM need to fix this
11681	 */
11682	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11683		ctl_ua_type ua_type;
11684
11685		ua_type = lun->pending_ua[initidx];
11686		if (ua_type != CTL_UA_NONE) {
11687			scsi_sense_data_type sense_format;
11688
11689			if (lun != NULL)
11690				sense_format = (lun->flags &
11691				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11692				    SSD_TYPE_FIXED;
11693			else
11694				sense_format = SSD_TYPE_FIXED;
11695
11696			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11697					       sense_format);
11698			if (ua_type != CTL_UA_NONE) {
11699				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11700				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11701						       CTL_AUTOSENSE;
11702				ctsio->sense_len = SSD_FULL_SIZE;
11703				lun->pending_ua[initidx] &= ~ua_type;
11704				mtx_unlock(&lun->lun_lock);
11705				ctl_done((union ctl_io *)ctsio);
11706				return (retval);
11707			}
11708		}
11709	}
11710
11711
11712	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11713		mtx_unlock(&lun->lun_lock);
11714		ctl_done((union ctl_io *)ctsio);
11715		return (retval);
11716	}
11717
11718	/*
11719	 * XXX CHD this is where we want to send IO to other side if
11720	 * this LUN is secondary on this SC. We will need to make a copy
11721	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11722	 * the copy we send as FROM_OTHER.
11723	 * We also need to stuff the address of the original IO so we can
11724	 * find it easily. Something similar will need be done on the other
11725	 * side so when we are done we can find the copy.
11726	 */
11727	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11728		union ctl_ha_msg msg_info;
11729		int isc_retval;
11730
11731		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11732
11733		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11734		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11735#if 0
11736		printf("1. ctsio %p\n", ctsio);
11737#endif
11738		msg_info.hdr.serializing_sc = NULL;
11739		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11740		msg_info.scsi.tag_num = ctsio->tag_num;
11741		msg_info.scsi.tag_type = ctsio->tag_type;
11742		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11743
11744		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11745
11746		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11747		    (void *)&msg_info, sizeof(msg_info), 0)) >
11748		    CTL_HA_STATUS_SUCCESS) {
11749			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11750			       isc_retval);
11751			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11752		} else {
11753#if 0
11754			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11755#endif
11756		}
11757
11758		/*
11759		 * XXX KDM this I/O is off the incoming queue, but hasn't
11760		 * been inserted on any other queue.  We may need to come
11761		 * up with a holding queue while we wait for serialization
11762		 * so that we have an idea of what we're waiting for from
11763		 * the other side.
11764		 */
11765		mtx_unlock(&lun->lun_lock);
11766		return (retval);
11767	}
11768
11769	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11770			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11771			      ctl_ooaq, ooa_links))) {
11772	case CTL_ACTION_BLOCK:
11773		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11774		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11775				  blocked_links);
11776		mtx_unlock(&lun->lun_lock);
11777		return (retval);
11778	case CTL_ACTION_PASS:
11779	case CTL_ACTION_SKIP:
11780		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11781		mtx_unlock(&lun->lun_lock);
11782		ctl_enqueue_rtr((union ctl_io *)ctsio);
11783		break;
11784	case CTL_ACTION_OVERLAP:
11785		mtx_unlock(&lun->lun_lock);
11786		ctl_set_overlapped_cmd(ctsio);
11787		ctl_done((union ctl_io *)ctsio);
11788		break;
11789	case CTL_ACTION_OVERLAP_TAG:
11790		mtx_unlock(&lun->lun_lock);
11791		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11792		ctl_done((union ctl_io *)ctsio);
11793		break;
11794	case CTL_ACTION_ERROR:
11795	default:
11796		mtx_unlock(&lun->lun_lock);
11797		ctl_set_internal_failure(ctsio,
11798					 /*sks_valid*/ 0,
11799					 /*retry_count*/ 0);
11800		ctl_done((union ctl_io *)ctsio);
11801		break;
11802	}
11803	return (retval);
11804}
11805
11806const struct ctl_cmd_entry *
11807ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11808{
11809	const struct ctl_cmd_entry *entry;
11810	int service_action;
11811
11812	entry = &ctl_cmd_table[ctsio->cdb[0]];
11813	if (entry->flags & CTL_CMD_FLAG_SA5) {
11814		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11815		entry = &((const struct ctl_cmd_entry *)
11816		    entry->execute)[service_action];
11817	}
11818	return (entry);
11819}
11820
11821const struct ctl_cmd_entry *
11822ctl_validate_command(struct ctl_scsiio *ctsio)
11823{
11824	const struct ctl_cmd_entry *entry;
11825	int i;
11826	uint8_t diff;
11827
11828	entry = ctl_get_cmd_entry(ctsio);
11829	if (entry->execute == NULL) {
11830		ctl_set_invalid_opcode(ctsio);
11831		ctl_done((union ctl_io *)ctsio);
11832		return (NULL);
11833	}
11834	KASSERT(entry->length > 0,
11835	    ("Not defined length for command 0x%02x/0x%02x",
11836	     ctsio->cdb[0], ctsio->cdb[1]));
11837	for (i = 1; i < entry->length; i++) {
11838		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11839		if (diff == 0)
11840			continue;
11841		ctl_set_invalid_field(ctsio,
11842				      /*sks_valid*/ 1,
11843				      /*command*/ 1,
11844				      /*field*/ i,
11845				      /*bit_valid*/ 1,
11846				      /*bit*/ fls(diff) - 1);
11847		ctl_done((union ctl_io *)ctsio);
11848		return (NULL);
11849	}
11850	return (entry);
11851}
11852
11853static int
11854ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11855{
11856
11857	switch (lun_type) {
11858	case T_PROCESSOR:
11859		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11860		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11861			return (0);
11862		break;
11863	case T_DIRECT:
11864		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11865		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11866			return (0);
11867		break;
11868	default:
11869		return (0);
11870	}
11871	return (1);
11872}
11873
11874static int
11875ctl_scsiio(struct ctl_scsiio *ctsio)
11876{
11877	int retval;
11878	const struct ctl_cmd_entry *entry;
11879
11880	retval = CTL_RETVAL_COMPLETE;
11881
11882	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11883
11884	entry = ctl_get_cmd_entry(ctsio);
11885
11886	/*
11887	 * If this I/O has been aborted, just send it straight to
11888	 * ctl_done() without executing it.
11889	 */
11890	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11891		ctl_done((union ctl_io *)ctsio);
11892		goto bailout;
11893	}
11894
11895	/*
11896	 * All the checks should have been handled by ctl_scsiio_precheck().
11897	 * We should be clear now to just execute the I/O.
11898	 */
11899	retval = entry->execute(ctsio);
11900
11901bailout:
11902	return (retval);
11903}
11904
11905/*
11906 * Since we only implement one target right now, a bus reset simply resets
11907 * our single target.
11908 */
11909static int
11910ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11911{
11912	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11913}
11914
11915static int
11916ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11917		 ctl_ua_type ua_type)
11918{
11919	struct ctl_lun *lun;
11920	int retval;
11921
11922	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11923		union ctl_ha_msg msg_info;
11924
11925		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11926		msg_info.hdr.nexus = io->io_hdr.nexus;
11927		if (ua_type==CTL_UA_TARG_RESET)
11928			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11929		else
11930			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11931		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11932		msg_info.hdr.original_sc = NULL;
11933		msg_info.hdr.serializing_sc = NULL;
11934		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11935		    (void *)&msg_info, sizeof(msg_info), 0)) {
11936		}
11937	}
11938	retval = 0;
11939
11940	mtx_lock(&ctl_softc->ctl_lock);
11941	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11942		retval += ctl_lun_reset(lun, io, ua_type);
11943	mtx_unlock(&ctl_softc->ctl_lock);
11944
11945	return (retval);
11946}
11947
11948/*
11949 * The LUN should always be set.  The I/O is optional, and is used to
11950 * distinguish between I/Os sent by this initiator, and by other
11951 * initiators.  We set unit attention for initiators other than this one.
11952 * SAM-3 is vague on this point.  It does say that a unit attention should
11953 * be established for other initiators when a LUN is reset (see section
11954 * 5.7.3), but it doesn't specifically say that the unit attention should
11955 * be established for this particular initiator when a LUN is reset.  Here
11956 * is the relevant text, from SAM-3 rev 8:
11957 *
11958 * 5.7.2 When a SCSI initiator port aborts its own tasks
11959 *
11960 * When a SCSI initiator port causes its own task(s) to be aborted, no
11961 * notification that the task(s) have been aborted shall be returned to
11962 * the SCSI initiator port other than the completion response for the
11963 * command or task management function action that caused the task(s) to
11964 * be aborted and notification(s) associated with related effects of the
11965 * action (e.g., a reset unit attention condition).
11966 *
11967 * XXX KDM for now, we're setting unit attention for all initiators.
11968 */
11969static int
11970ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11971{
11972	union ctl_io *xio;
11973#if 0
11974	uint32_t initindex;
11975#endif
11976	int i;
11977
11978	mtx_lock(&lun->lun_lock);
11979	/*
11980	 * Run through the OOA queue and abort each I/O.
11981	 */
11982#if 0
11983	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11984#endif
11985	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11986	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11987		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11988	}
11989
11990	/*
11991	 * This version sets unit attention for every
11992	 */
11993#if 0
11994	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11995	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11996		if (initindex == i)
11997			continue;
11998		lun->pending_ua[i] |= ua_type;
11999	}
12000#endif
12001
12002	/*
12003	 * A reset (any kind, really) clears reservations established with
12004	 * RESERVE/RELEASE.  It does not clear reservations established
12005	 * with PERSISTENT RESERVE OUT, but we don't support that at the
12006	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12007	 * reservations made with the RESERVE/RELEASE commands, because
12008	 * those commands are obsolete in SPC-3.
12009	 */
12010	lun->flags &= ~CTL_LUN_RESERVED;
12011
12012	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
12013#ifdef CTL_WITH_CA
12014		ctl_clear_mask(lun->have_ca, i);
12015#endif
12016		lun->pending_ua[i] |= ua_type;
12017	}
12018	mtx_unlock(&lun->lun_lock);
12019
12020	return (0);
12021}
12022
12023static int
12024ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12025    int other_sc)
12026{
12027	union ctl_io *xio;
12028	int found;
12029
12030	mtx_assert(&lun->lun_lock, MA_OWNED);
12031
12032	/*
12033	 * Run through the OOA queue and attempt to find the given I/O.
12034	 * The target port, initiator ID, tag type and tag number have to
12035	 * match the values that we got from the initiator.  If we have an
12036	 * untagged command to abort, simply abort the first untagged command
12037	 * we come to.  We only allow one untagged command at a time of course.
12038	 */
12039	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12040	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12041
12042		if ((targ_port == UINT32_MAX ||
12043		     targ_port == xio->io_hdr.nexus.targ_port) &&
12044		    (init_id == UINT32_MAX ||
12045		     init_id == xio->io_hdr.nexus.initid.id)) {
12046			if (targ_port != xio->io_hdr.nexus.targ_port ||
12047			    init_id != xio->io_hdr.nexus.initid.id)
12048				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12049			xio->io_hdr.flags |= CTL_FLAG_ABORT;
12050			found = 1;
12051			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12052				union ctl_ha_msg msg_info;
12053
12054				msg_info.hdr.nexus = xio->io_hdr.nexus;
12055				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12056				msg_info.task.tag_num = xio->scsiio.tag_num;
12057				msg_info.task.tag_type = xio->scsiio.tag_type;
12058				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12059				msg_info.hdr.original_sc = NULL;
12060				msg_info.hdr.serializing_sc = NULL;
12061				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12062				    (void *)&msg_info, sizeof(msg_info), 0);
12063			}
12064		}
12065	}
12066	return (found);
12067}
12068
12069static int
12070ctl_abort_task_set(union ctl_io *io)
12071{
12072	struct ctl_softc *softc = control_softc;
12073	struct ctl_lun *lun;
12074	uint32_t targ_lun;
12075
12076	/*
12077	 * Look up the LUN.
12078	 */
12079	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12080	mtx_lock(&softc->ctl_lock);
12081	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12082		lun = softc->ctl_luns[targ_lun];
12083	else {
12084		mtx_unlock(&softc->ctl_lock);
12085		return (1);
12086	}
12087
12088	mtx_lock(&lun->lun_lock);
12089	mtx_unlock(&softc->ctl_lock);
12090	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12091		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12092		    io->io_hdr.nexus.initid.id,
12093		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12094	} else { /* CTL_TASK_CLEAR_TASK_SET */
12095		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12096		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12097	}
12098	mtx_unlock(&lun->lun_lock);
12099	return (0);
12100}
12101
12102static int
12103ctl_i_t_nexus_reset(union ctl_io *io)
12104{
12105	struct ctl_softc *softc = control_softc;
12106	struct ctl_lun *lun;
12107	uint32_t initindex;
12108
12109	initindex = ctl_get_initindex(&io->io_hdr.nexus);
12110	mtx_lock(&softc->ctl_lock);
12111	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12112		mtx_lock(&lun->lun_lock);
12113		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12114		    io->io_hdr.nexus.initid.id,
12115		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12116#ifdef CTL_WITH_CA
12117		ctl_clear_mask(lun->have_ca, initindex);
12118#endif
12119		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
12120		mtx_unlock(&lun->lun_lock);
12121	}
12122	mtx_unlock(&softc->ctl_lock);
12123	return (0);
12124}
12125
12126static int
12127ctl_abort_task(union ctl_io *io)
12128{
12129	union ctl_io *xio;
12130	struct ctl_lun *lun;
12131	struct ctl_softc *ctl_softc;
12132#if 0
12133	struct sbuf sb;
12134	char printbuf[128];
12135#endif
12136	int found;
12137	uint32_t targ_lun;
12138
12139	ctl_softc = control_softc;
12140	found = 0;
12141
12142	/*
12143	 * Look up the LUN.
12144	 */
12145	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12146	mtx_lock(&ctl_softc->ctl_lock);
12147	if ((targ_lun < CTL_MAX_LUNS)
12148	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12149		lun = ctl_softc->ctl_luns[targ_lun];
12150	else {
12151		mtx_unlock(&ctl_softc->ctl_lock);
12152		return (1);
12153	}
12154
12155#if 0
12156	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12157	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12158#endif
12159
12160	mtx_lock(&lun->lun_lock);
12161	mtx_unlock(&ctl_softc->ctl_lock);
12162	/*
12163	 * Run through the OOA queue and attempt to find the given I/O.
12164	 * The target port, initiator ID, tag type and tag number have to
12165	 * match the values that we got from the initiator.  If we have an
12166	 * untagged command to abort, simply abort the first untagged command
12167	 * we come to.  We only allow one untagged command at a time of course.
12168	 */
12169#if 0
12170	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12171#endif
12172	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12173	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12174#if 0
12175		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12176
12177		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12178			    lun->lun, xio->scsiio.tag_num,
12179			    xio->scsiio.tag_type,
12180			    (xio->io_hdr.blocked_links.tqe_prev
12181			    == NULL) ? "" : " BLOCKED",
12182			    (xio->io_hdr.flags &
12183			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12184			    (xio->io_hdr.flags &
12185			    CTL_FLAG_ABORT) ? " ABORT" : "",
12186			    (xio->io_hdr.flags &
12187			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12188		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12189		sbuf_finish(&sb);
12190		printf("%s\n", sbuf_data(&sb));
12191#endif
12192
12193		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12194		 && (xio->io_hdr.nexus.initid.id ==
12195		     io->io_hdr.nexus.initid.id)) {
12196			/*
12197			 * If the abort says that the task is untagged, the
12198			 * task in the queue must be untagged.  Otherwise,
12199			 * we just check to see whether the tag numbers
12200			 * match.  This is because the QLogic firmware
12201			 * doesn't pass back the tag type in an abort
12202			 * request.
12203			 */
12204#if 0
12205			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12206			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12207			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12208#endif
12209			/*
12210			 * XXX KDM we've got problems with FC, because it
12211			 * doesn't send down a tag type with aborts.  So we
12212			 * can only really go by the tag number...
12213			 * This may cause problems with parallel SCSI.
12214			 * Need to figure that out!!
12215			 */
12216			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12217				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12218				found = 1;
12219				if ((io->io_hdr.flags &
12220				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12221				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12222					union ctl_ha_msg msg_info;
12223
12224					io->io_hdr.flags |=
12225					                CTL_FLAG_SENT_2OTHER_SC;
12226					msg_info.hdr.nexus = io->io_hdr.nexus;
12227					msg_info.task.task_action =
12228						CTL_TASK_ABORT_TASK;
12229					msg_info.task.tag_num =
12230						io->taskio.tag_num;
12231					msg_info.task.tag_type =
12232						io->taskio.tag_type;
12233					msg_info.hdr.msg_type =
12234						CTL_MSG_MANAGE_TASKS;
12235					msg_info.hdr.original_sc = NULL;
12236					msg_info.hdr.serializing_sc = NULL;
12237#if 0
12238					printf("Sent Abort to other side\n");
12239#endif
12240					if (CTL_HA_STATUS_SUCCESS !=
12241					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12242		    				(void *)&msg_info,
12243						sizeof(msg_info), 0)) {
12244					}
12245				}
12246#if 0
12247				printf("ctl_abort_task: found I/O to abort\n");
12248#endif
12249				break;
12250			}
12251		}
12252	}
12253	mtx_unlock(&lun->lun_lock);
12254
12255	if (found == 0) {
12256		/*
12257		 * This isn't really an error.  It's entirely possible for
12258		 * the abort and command completion to cross on the wire.
12259		 * This is more of an informative/diagnostic error.
12260		 */
12261#if 0
12262		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12263		       "%d:%d:%d:%d tag %d type %d\n",
12264		       io->io_hdr.nexus.initid.id,
12265		       io->io_hdr.nexus.targ_port,
12266		       io->io_hdr.nexus.targ_target.id,
12267		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12268		       io->taskio.tag_type);
12269#endif
12270	}
12271	return (0);
12272}
12273
12274static void
12275ctl_run_task(union ctl_io *io)
12276{
12277	struct ctl_softc *ctl_softc = control_softc;
12278	int retval = 1;
12279	const char *task_desc;
12280
12281	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12282
12283	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12284	    ("ctl_run_task: Unextected io_type %d\n",
12285	     io->io_hdr.io_type));
12286
12287	task_desc = ctl_scsi_task_string(&io->taskio);
12288	if (task_desc != NULL) {
12289#ifdef NEEDTOPORT
12290		csevent_log(CSC_CTL | CSC_SHELF_SW |
12291			    CTL_TASK_REPORT,
12292			    csevent_LogType_Trace,
12293			    csevent_Severity_Information,
12294			    csevent_AlertLevel_Green,
12295			    csevent_FRU_Firmware,
12296			    csevent_FRU_Unknown,
12297			    "CTL: received task: %s",task_desc);
12298#endif
12299	} else {
12300#ifdef NEEDTOPORT
12301		csevent_log(CSC_CTL | CSC_SHELF_SW |
12302			    CTL_TASK_REPORT,
12303			    csevent_LogType_Trace,
12304			    csevent_Severity_Information,
12305			    csevent_AlertLevel_Green,
12306			    csevent_FRU_Firmware,
12307			    csevent_FRU_Unknown,
12308			    "CTL: received unknown task "
12309			    "type: %d (%#x)",
12310			    io->taskio.task_action,
12311			    io->taskio.task_action);
12312#endif
12313	}
12314	switch (io->taskio.task_action) {
12315	case CTL_TASK_ABORT_TASK:
12316		retval = ctl_abort_task(io);
12317		break;
12318	case CTL_TASK_ABORT_TASK_SET:
12319	case CTL_TASK_CLEAR_TASK_SET:
12320		retval = ctl_abort_task_set(io);
12321		break;
12322	case CTL_TASK_CLEAR_ACA:
12323		break;
12324	case CTL_TASK_I_T_NEXUS_RESET:
12325		retval = ctl_i_t_nexus_reset(io);
12326		break;
12327	case CTL_TASK_LUN_RESET: {
12328		struct ctl_lun *lun;
12329		uint32_t targ_lun;
12330
12331		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12332		mtx_lock(&ctl_softc->ctl_lock);
12333		if ((targ_lun < CTL_MAX_LUNS)
12334		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12335			lun = ctl_softc->ctl_luns[targ_lun];
12336		else {
12337			mtx_unlock(&ctl_softc->ctl_lock);
12338			retval = 1;
12339			break;
12340		}
12341
12342		if (!(io->io_hdr.flags &
12343		    CTL_FLAG_FROM_OTHER_SC)) {
12344			union ctl_ha_msg msg_info;
12345
12346			io->io_hdr.flags |=
12347				CTL_FLAG_SENT_2OTHER_SC;
12348			msg_info.hdr.msg_type =
12349				CTL_MSG_MANAGE_TASKS;
12350			msg_info.hdr.nexus = io->io_hdr.nexus;
12351			msg_info.task.task_action =
12352				CTL_TASK_LUN_RESET;
12353			msg_info.hdr.original_sc = NULL;
12354			msg_info.hdr.serializing_sc = NULL;
12355			if (CTL_HA_STATUS_SUCCESS !=
12356			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12357			    (void *)&msg_info,
12358			    sizeof(msg_info), 0)) {
12359			}
12360		}
12361
12362		retval = ctl_lun_reset(lun, io,
12363				       CTL_UA_LUN_RESET);
12364		mtx_unlock(&ctl_softc->ctl_lock);
12365		break;
12366	}
12367	case CTL_TASK_TARGET_RESET:
12368		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12369		break;
12370	case CTL_TASK_BUS_RESET:
12371		retval = ctl_bus_reset(ctl_softc, io);
12372		break;
12373	case CTL_TASK_PORT_LOGIN:
12374		break;
12375	case CTL_TASK_PORT_LOGOUT:
12376		break;
12377	default:
12378		printf("ctl_run_task: got unknown task management event %d\n",
12379		       io->taskio.task_action);
12380		break;
12381	}
12382	if (retval == 0)
12383		io->io_hdr.status = CTL_SUCCESS;
12384	else
12385		io->io_hdr.status = CTL_ERROR;
12386	ctl_done(io);
12387}
12388
12389/*
12390 * For HA operation.  Handle commands that come in from the other
12391 * controller.
12392 */
12393static void
12394ctl_handle_isc(union ctl_io *io)
12395{
12396	int free_io;
12397	struct ctl_lun *lun;
12398	struct ctl_softc *ctl_softc;
12399	uint32_t targ_lun;
12400
12401	ctl_softc = control_softc;
12402
12403	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12404	lun = ctl_softc->ctl_luns[targ_lun];
12405
12406	switch (io->io_hdr.msg_type) {
12407	case CTL_MSG_SERIALIZE:
12408		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12409		break;
12410	case CTL_MSG_R2R: {
12411		const struct ctl_cmd_entry *entry;
12412
12413		/*
12414		 * This is only used in SER_ONLY mode.
12415		 */
12416		free_io = 0;
12417		entry = ctl_get_cmd_entry(&io->scsiio);
12418		mtx_lock(&lun->lun_lock);
12419		if (ctl_scsiio_lun_check(ctl_softc, lun,
12420		    entry, (struct ctl_scsiio *)io) != 0) {
12421			mtx_unlock(&lun->lun_lock);
12422			ctl_done(io);
12423			break;
12424		}
12425		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12426		mtx_unlock(&lun->lun_lock);
12427		ctl_enqueue_rtr(io);
12428		break;
12429	}
12430	case CTL_MSG_FINISH_IO:
12431		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12432			free_io = 0;
12433			ctl_done(io);
12434		} else {
12435			free_io = 1;
12436			mtx_lock(&lun->lun_lock);
12437			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12438				     ooa_links);
12439			ctl_check_blocked(lun);
12440			mtx_unlock(&lun->lun_lock);
12441		}
12442		break;
12443	case CTL_MSG_PERS_ACTION:
12444		ctl_hndl_per_res_out_on_other_sc(
12445			(union ctl_ha_msg *)&io->presio.pr_msg);
12446		free_io = 1;
12447		break;
12448	case CTL_MSG_BAD_JUJU:
12449		free_io = 0;
12450		ctl_done(io);
12451		break;
12452	case CTL_MSG_DATAMOVE:
12453		/* Only used in XFER mode */
12454		free_io = 0;
12455		ctl_datamove_remote(io);
12456		break;
12457	case CTL_MSG_DATAMOVE_DONE:
12458		/* Only used in XFER mode */
12459		free_io = 0;
12460		io->scsiio.be_move_done(io);
12461		break;
12462	default:
12463		free_io = 1;
12464		printf("%s: Invalid message type %d\n",
12465		       __func__, io->io_hdr.msg_type);
12466		break;
12467	}
12468	if (free_io)
12469		ctl_free_io(io);
12470
12471}
12472
12473
12474/*
12475 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12476 * there is no match.
12477 */
12478static ctl_lun_error_pattern
12479ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12480{
12481	const struct ctl_cmd_entry *entry;
12482	ctl_lun_error_pattern filtered_pattern, pattern;
12483
12484	pattern = desc->error_pattern;
12485
12486	/*
12487	 * XXX KDM we need more data passed into this function to match a
12488	 * custom pattern, and we actually need to implement custom pattern
12489	 * matching.
12490	 */
12491	if (pattern & CTL_LUN_PAT_CMD)
12492		return (CTL_LUN_PAT_CMD);
12493
12494	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12495		return (CTL_LUN_PAT_ANY);
12496
12497	entry = ctl_get_cmd_entry(ctsio);
12498
12499	filtered_pattern = entry->pattern & pattern;
12500
12501	/*
12502	 * If the user requested specific flags in the pattern (e.g.
12503	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12504	 * flags.
12505	 *
12506	 * If the user did not specify any flags, it doesn't matter whether
12507	 * or not the command supports the flags.
12508	 */
12509	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12510	     (pattern & ~CTL_LUN_PAT_MASK))
12511		return (CTL_LUN_PAT_NONE);
12512
12513	/*
12514	 * If the user asked for a range check, see if the requested LBA
12515	 * range overlaps with this command's LBA range.
12516	 */
12517	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12518		uint64_t lba1;
12519		uint32_t len1;
12520		ctl_action action;
12521		int retval;
12522
12523		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12524		if (retval != 0)
12525			return (CTL_LUN_PAT_NONE);
12526
12527		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12528					      desc->lba_range.len);
12529		/*
12530		 * A "pass" means that the LBA ranges don't overlap, so
12531		 * this doesn't match the user's range criteria.
12532		 */
12533		if (action == CTL_ACTION_PASS)
12534			return (CTL_LUN_PAT_NONE);
12535	}
12536
12537	return (filtered_pattern);
12538}
12539
12540static void
12541ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12542{
12543	struct ctl_error_desc *desc, *desc2;
12544
12545	mtx_assert(&lun->lun_lock, MA_OWNED);
12546
12547	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12548		ctl_lun_error_pattern pattern;
12549		/*
12550		 * Check to see whether this particular command matches
12551		 * the pattern in the descriptor.
12552		 */
12553		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12554		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12555			continue;
12556
12557		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12558		case CTL_LUN_INJ_ABORTED:
12559			ctl_set_aborted(&io->scsiio);
12560			break;
12561		case CTL_LUN_INJ_MEDIUM_ERR:
12562			ctl_set_medium_error(&io->scsiio);
12563			break;
12564		case CTL_LUN_INJ_UA:
12565			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12566			 * OCCURRED */
12567			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12568			break;
12569		case CTL_LUN_INJ_CUSTOM:
12570			/*
12571			 * We're assuming the user knows what he is doing.
12572			 * Just copy the sense information without doing
12573			 * checks.
12574			 */
12575			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12576			      ctl_min(sizeof(desc->custom_sense),
12577				      sizeof(io->scsiio.sense_data)));
12578			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12579			io->scsiio.sense_len = SSD_FULL_SIZE;
12580			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12581			break;
12582		case CTL_LUN_INJ_NONE:
12583		default:
12584			/*
12585			 * If this is an error injection type we don't know
12586			 * about, clear the continuous flag (if it is set)
12587			 * so it will get deleted below.
12588			 */
12589			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12590			break;
12591		}
12592		/*
12593		 * By default, each error injection action is a one-shot
12594		 */
12595		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12596			continue;
12597
12598		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12599
12600		free(desc, M_CTL);
12601	}
12602}
12603
12604#ifdef CTL_IO_DELAY
12605static void
12606ctl_datamove_timer_wakeup(void *arg)
12607{
12608	union ctl_io *io;
12609
12610	io = (union ctl_io *)arg;
12611
12612	ctl_datamove(io);
12613}
12614#endif /* CTL_IO_DELAY */
12615
12616void
12617ctl_datamove(union ctl_io *io)
12618{
12619	void (*fe_datamove)(union ctl_io *io);
12620
12621	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12622
12623	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12624
12625#ifdef CTL_TIME_IO
12626	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12627		char str[256];
12628		char path_str[64];
12629		struct sbuf sb;
12630
12631		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12632		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12633
12634		sbuf_cat(&sb, path_str);
12635		switch (io->io_hdr.io_type) {
12636		case CTL_IO_SCSI:
12637			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12638			sbuf_printf(&sb, "\n");
12639			sbuf_cat(&sb, path_str);
12640			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12641				    io->scsiio.tag_num, io->scsiio.tag_type);
12642			break;
12643		case CTL_IO_TASK:
12644			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12645				    "Tag Type: %d\n", io->taskio.task_action,
12646				    io->taskio.tag_num, io->taskio.tag_type);
12647			break;
12648		default:
12649			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12650			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12651			break;
12652		}
12653		sbuf_cat(&sb, path_str);
12654		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12655			    (intmax_t)time_uptime - io->io_hdr.start_time);
12656		sbuf_finish(&sb);
12657		printf("%s", sbuf_data(&sb));
12658	}
12659#endif /* CTL_TIME_IO */
12660
12661#ifdef CTL_IO_DELAY
12662	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12663		struct ctl_lun *lun;
12664
12665		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12666
12667		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12668	} else {
12669		struct ctl_lun *lun;
12670
12671		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12672		if ((lun != NULL)
12673		 && (lun->delay_info.datamove_delay > 0)) {
12674			struct callout *callout;
12675
12676			callout = (struct callout *)&io->io_hdr.timer_bytes;
12677			callout_init(callout, /*mpsafe*/ 1);
12678			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12679			callout_reset(callout,
12680				      lun->delay_info.datamove_delay * hz,
12681				      ctl_datamove_timer_wakeup, io);
12682			if (lun->delay_info.datamove_type ==
12683			    CTL_DELAY_TYPE_ONESHOT)
12684				lun->delay_info.datamove_delay = 0;
12685			return;
12686		}
12687	}
12688#endif
12689
12690	/*
12691	 * This command has been aborted.  Set the port status, so we fail
12692	 * the data move.
12693	 */
12694	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12695		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12696		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12697		       io->io_hdr.nexus.targ_port,
12698		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12699		       io->io_hdr.nexus.targ_lun);
12700		io->io_hdr.port_status = 31337;
12701		/*
12702		 * Note that the backend, in this case, will get the
12703		 * callback in its context.  In other cases it may get
12704		 * called in the frontend's interrupt thread context.
12705		 */
12706		io->scsiio.be_move_done(io);
12707		return;
12708	}
12709
12710	/*
12711	 * If we're in XFER mode and this I/O is from the other shelf
12712	 * controller, we need to send the DMA to the other side to
12713	 * actually transfer the data to/from the host.  In serialize only
12714	 * mode the transfer happens below CTL and ctl_datamove() is only
12715	 * called on the machine that originally received the I/O.
12716	 */
12717	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12718	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12719		union ctl_ha_msg msg;
12720		uint32_t sg_entries_sent;
12721		int do_sg_copy;
12722		int i;
12723
12724		memset(&msg, 0, sizeof(msg));
12725		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12726		msg.hdr.original_sc = io->io_hdr.original_sc;
12727		msg.hdr.serializing_sc = io;
12728		msg.hdr.nexus = io->io_hdr.nexus;
12729		msg.dt.flags = io->io_hdr.flags;
12730		/*
12731		 * We convert everything into a S/G list here.  We can't
12732		 * pass by reference, only by value between controllers.
12733		 * So we can't pass a pointer to the S/G list, only as many
12734		 * S/G entries as we can fit in here.  If it's possible for
12735		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12736		 * then we need to break this up into multiple transfers.
12737		 */
12738		if (io->scsiio.kern_sg_entries == 0) {
12739			msg.dt.kern_sg_entries = 1;
12740			/*
12741			 * If this is in cached memory, flush the cache
12742			 * before we send the DMA request to the other
12743			 * controller.  We want to do this in either the
12744			 * read or the write case.  The read case is
12745			 * straightforward.  In the write case, we want to
12746			 * make sure nothing is in the local cache that
12747			 * could overwrite the DMAed data.
12748			 */
12749			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12750				/*
12751				 * XXX KDM use bus_dmamap_sync() here.
12752				 */
12753			}
12754
12755			/*
12756			 * Convert to a physical address if this is a
12757			 * virtual address.
12758			 */
12759			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12760				msg.dt.sg_list[0].addr =
12761					io->scsiio.kern_data_ptr;
12762			} else {
12763				/*
12764				 * XXX KDM use busdma here!
12765				 */
12766#if 0
12767				msg.dt.sg_list[0].addr = (void *)
12768					vtophys(io->scsiio.kern_data_ptr);
12769#endif
12770			}
12771
12772			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12773			do_sg_copy = 0;
12774		} else {
12775			struct ctl_sg_entry *sgl;
12776
12777			do_sg_copy = 1;
12778			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12779			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12780			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12781				/*
12782				 * XXX KDM use bus_dmamap_sync() here.
12783				 */
12784			}
12785		}
12786
12787		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12788		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12789		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12790		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12791		msg.dt.sg_sequence = 0;
12792
12793		/*
12794		 * Loop until we've sent all of the S/G entries.  On the
12795		 * other end, we'll recompose these S/G entries into one
12796		 * contiguous list before passing it to the
12797		 */
12798		for (sg_entries_sent = 0; sg_entries_sent <
12799		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12800			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12801				sizeof(msg.dt.sg_list[0])),
12802				msg.dt.kern_sg_entries - sg_entries_sent);
12803
12804			if (do_sg_copy != 0) {
12805				struct ctl_sg_entry *sgl;
12806				int j;
12807
12808				sgl = (struct ctl_sg_entry *)
12809					io->scsiio.kern_data_ptr;
12810				/*
12811				 * If this is in cached memory, flush the cache
12812				 * before we send the DMA request to the other
12813				 * controller.  We want to do this in either
12814				 * the * read or the write case.  The read
12815				 * case is straightforward.  In the write
12816				 * case, we want to make sure nothing is
12817				 * in the local cache that could overwrite
12818				 * the DMAed data.
12819				 */
12820
12821				for (i = sg_entries_sent, j = 0;
12822				     i < msg.dt.cur_sg_entries; i++, j++) {
12823					if ((io->io_hdr.flags &
12824					     CTL_FLAG_NO_DATASYNC) == 0) {
12825						/*
12826						 * XXX KDM use bus_dmamap_sync()
12827						 */
12828					}
12829					if ((io->io_hdr.flags &
12830					     CTL_FLAG_BUS_ADDR) == 0) {
12831						/*
12832						 * XXX KDM use busdma.
12833						 */
12834#if 0
12835						msg.dt.sg_list[j].addr =(void *)
12836						       vtophys(sgl[i].addr);
12837#endif
12838					} else {
12839						msg.dt.sg_list[j].addr =
12840							sgl[i].addr;
12841					}
12842					msg.dt.sg_list[j].len = sgl[i].len;
12843				}
12844			}
12845
12846			sg_entries_sent += msg.dt.cur_sg_entries;
12847			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12848				msg.dt.sg_last = 1;
12849			else
12850				msg.dt.sg_last = 0;
12851
12852			/*
12853			 * XXX KDM drop and reacquire the lock here?
12854			 */
12855			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12856			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12857				/*
12858				 * XXX do something here.
12859				 */
12860			}
12861
12862			msg.dt.sent_sg_entries = sg_entries_sent;
12863		}
12864		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12865		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12866			ctl_failover_io(io, /*have_lock*/ 0);
12867
12868	} else {
12869
12870		/*
12871		 * Lookup the fe_datamove() function for this particular
12872		 * front end.
12873		 */
12874		fe_datamove =
12875		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12876
12877		fe_datamove(io);
12878	}
12879}
12880
12881static void
12882ctl_send_datamove_done(union ctl_io *io, int have_lock)
12883{
12884	union ctl_ha_msg msg;
12885	int isc_status;
12886
12887	memset(&msg, 0, sizeof(msg));
12888
12889	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12890	msg.hdr.original_sc = io;
12891	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12892	msg.hdr.nexus = io->io_hdr.nexus;
12893	msg.hdr.status = io->io_hdr.status;
12894	msg.scsi.tag_num = io->scsiio.tag_num;
12895	msg.scsi.tag_type = io->scsiio.tag_type;
12896	msg.scsi.scsi_status = io->scsiio.scsi_status;
12897	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12898	       sizeof(io->scsiio.sense_data));
12899	msg.scsi.sense_len = io->scsiio.sense_len;
12900	msg.scsi.sense_residual = io->scsiio.sense_residual;
12901	msg.scsi.fetd_status = io->io_hdr.port_status;
12902	msg.scsi.residual = io->scsiio.residual;
12903	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12904
12905	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12906		ctl_failover_io(io, /*have_lock*/ have_lock);
12907		return;
12908	}
12909
12910	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12911	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12912		/* XXX do something if this fails */
12913	}
12914
12915}
12916
12917/*
12918 * The DMA to the remote side is done, now we need to tell the other side
12919 * we're done so it can continue with its data movement.
12920 */
12921static void
12922ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12923{
12924	union ctl_io *io;
12925
12926	io = rq->context;
12927
12928	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12929		printf("%s: ISC DMA write failed with error %d", __func__,
12930		       rq->ret);
12931		ctl_set_internal_failure(&io->scsiio,
12932					 /*sks_valid*/ 1,
12933					 /*retry_count*/ rq->ret);
12934	}
12935
12936	ctl_dt_req_free(rq);
12937
12938	/*
12939	 * In this case, we had to malloc the memory locally.  Free it.
12940	 */
12941	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12942		int i;
12943		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12944			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12945	}
12946	/*
12947	 * The data is in local and remote memory, so now we need to send
12948	 * status (good or back) back to the other side.
12949	 */
12950	ctl_send_datamove_done(io, /*have_lock*/ 0);
12951}
12952
12953/*
12954 * We've moved the data from the host/controller into local memory.  Now we
12955 * need to push it over to the remote controller's memory.
12956 */
12957static int
12958ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12959{
12960	int retval;
12961
12962	retval = 0;
12963
12964	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12965					  ctl_datamove_remote_write_cb);
12966
12967	return (retval);
12968}
12969
12970static void
12971ctl_datamove_remote_write(union ctl_io *io)
12972{
12973	int retval;
12974	void (*fe_datamove)(union ctl_io *io);
12975
12976	/*
12977	 * - Get the data from the host/HBA into local memory.
12978	 * - DMA memory from the local controller to the remote controller.
12979	 * - Send status back to the remote controller.
12980	 */
12981
12982	retval = ctl_datamove_remote_sgl_setup(io);
12983	if (retval != 0)
12984		return;
12985
12986	/* Switch the pointer over so the FETD knows what to do */
12987	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12988
12989	/*
12990	 * Use a custom move done callback, since we need to send completion
12991	 * back to the other controller, not to the backend on this side.
12992	 */
12993	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12994
12995	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12996
12997	fe_datamove(io);
12998
12999	return;
13000
13001}
13002
13003static int
13004ctl_datamove_remote_dm_read_cb(union ctl_io *io)
13005{
13006#if 0
13007	char str[256];
13008	char path_str[64];
13009	struct sbuf sb;
13010#endif
13011
13012	/*
13013	 * In this case, we had to malloc the memory locally.  Free it.
13014	 */
13015	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
13016		int i;
13017		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13018			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13019	}
13020
13021#if 0
13022	scsi_path_string(io, path_str, sizeof(path_str));
13023	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13024	sbuf_cat(&sb, path_str);
13025	scsi_command_string(&io->scsiio, NULL, &sb);
13026	sbuf_printf(&sb, "\n");
13027	sbuf_cat(&sb, path_str);
13028	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13029		    io->scsiio.tag_num, io->scsiio.tag_type);
13030	sbuf_cat(&sb, path_str);
13031	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
13032		    io->io_hdr.flags, io->io_hdr.status);
13033	sbuf_finish(&sb);
13034	printk("%s", sbuf_data(&sb));
13035#endif
13036
13037
13038	/*
13039	 * The read is done, now we need to send status (good or bad) back
13040	 * to the other side.
13041	 */
13042	ctl_send_datamove_done(io, /*have_lock*/ 0);
13043
13044	return (0);
13045}
13046
13047static void
13048ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13049{
13050	union ctl_io *io;
13051	void (*fe_datamove)(union ctl_io *io);
13052
13053	io = rq->context;
13054
13055	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13056		printf("%s: ISC DMA read failed with error %d", __func__,
13057		       rq->ret);
13058		ctl_set_internal_failure(&io->scsiio,
13059					 /*sks_valid*/ 1,
13060					 /*retry_count*/ rq->ret);
13061	}
13062
13063	ctl_dt_req_free(rq);
13064
13065	/* Switch the pointer over so the FETD knows what to do */
13066	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13067
13068	/*
13069	 * Use a custom move done callback, since we need to send completion
13070	 * back to the other controller, not to the backend on this side.
13071	 */
13072	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13073
13074	/* XXX KDM add checks like the ones in ctl_datamove? */
13075
13076	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13077
13078	fe_datamove(io);
13079}
13080
13081static int
13082ctl_datamove_remote_sgl_setup(union ctl_io *io)
13083{
13084	struct ctl_sg_entry *local_sglist, *remote_sglist;
13085	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13086	struct ctl_softc *softc;
13087	int retval;
13088	int i;
13089
13090	retval = 0;
13091	softc = control_softc;
13092
13093	local_sglist = io->io_hdr.local_sglist;
13094	local_dma_sglist = io->io_hdr.local_dma_sglist;
13095	remote_sglist = io->io_hdr.remote_sglist;
13096	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13097
13098	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13099		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13100			local_sglist[i].len = remote_sglist[i].len;
13101
13102			/*
13103			 * XXX Detect the situation where the RS-level I/O
13104			 * redirector on the other side has already read the
13105			 * data off of the AOR RS on this side, and
13106			 * transferred it to remote (mirror) memory on the
13107			 * other side.  Since we already have the data in
13108			 * memory here, we just need to use it.
13109			 *
13110			 * XXX KDM this can probably be removed once we
13111			 * get the cache device code in and take the
13112			 * current AOR implementation out.
13113			 */
13114#ifdef NEEDTOPORT
13115			if ((remote_sglist[i].addr >=
13116			     (void *)vtophys(softc->mirr->addr))
13117			 && (remote_sglist[i].addr <
13118			     ((void *)vtophys(softc->mirr->addr) +
13119			     CacheMirrorOffset))) {
13120				local_sglist[i].addr = remote_sglist[i].addr -
13121					CacheMirrorOffset;
13122				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13123				     CTL_FLAG_DATA_IN)
13124					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13125			} else {
13126				local_sglist[i].addr = remote_sglist[i].addr +
13127					CacheMirrorOffset;
13128			}
13129#endif
13130#if 0
13131			printf("%s: local %p, remote %p, len %d\n",
13132			       __func__, local_sglist[i].addr,
13133			       remote_sglist[i].addr, local_sglist[i].len);
13134#endif
13135		}
13136	} else {
13137		uint32_t len_to_go;
13138
13139		/*
13140		 * In this case, we don't have automatically allocated
13141		 * memory for this I/O on this controller.  This typically
13142		 * happens with internal CTL I/O -- e.g. inquiry, mode
13143		 * sense, etc.  Anything coming from RAIDCore will have
13144		 * a mirror area available.
13145		 */
13146		len_to_go = io->scsiio.kern_data_len;
13147
13148		/*
13149		 * Clear the no datasync flag, we have to use malloced
13150		 * buffers.
13151		 */
13152		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13153
13154		/*
13155		 * The difficult thing here is that the size of the various
13156		 * S/G segments may be different than the size from the
13157		 * remote controller.  That'll make it harder when DMAing
13158		 * the data back to the other side.
13159		 */
13160		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13161		     sizeof(io->io_hdr.remote_sglist[0])) &&
13162		     (len_to_go > 0); i++) {
13163			local_sglist[i].len = ctl_min(len_to_go, 131072);
13164			CTL_SIZE_8B(local_dma_sglist[i].len,
13165				    local_sglist[i].len);
13166			local_sglist[i].addr =
13167				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13168
13169			local_dma_sglist[i].addr = local_sglist[i].addr;
13170
13171			if (local_sglist[i].addr == NULL) {
13172				int j;
13173
13174				printf("malloc failed for %zd bytes!",
13175				       local_dma_sglist[i].len);
13176				for (j = 0; j < i; j++) {
13177					free(local_sglist[j].addr, M_CTL);
13178				}
13179				ctl_set_internal_failure(&io->scsiio,
13180							 /*sks_valid*/ 1,
13181							 /*retry_count*/ 4857);
13182				retval = 1;
13183				goto bailout_error;
13184
13185			}
13186			/* XXX KDM do we need a sync here? */
13187
13188			len_to_go -= local_sglist[i].len;
13189		}
13190		/*
13191		 * Reset the number of S/G entries accordingly.  The
13192		 * original number of S/G entries is available in
13193		 * rem_sg_entries.
13194		 */
13195		io->scsiio.kern_sg_entries = i;
13196
13197#if 0
13198		printf("%s: kern_sg_entries = %d\n", __func__,
13199		       io->scsiio.kern_sg_entries);
13200		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13201			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13202			       local_sglist[i].addr, local_sglist[i].len,
13203			       local_dma_sglist[i].len);
13204#endif
13205	}
13206
13207
13208	return (retval);
13209
13210bailout_error:
13211
13212	ctl_send_datamove_done(io, /*have_lock*/ 0);
13213
13214	return (retval);
13215}
13216
13217static int
13218ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13219			 ctl_ha_dt_cb callback)
13220{
13221	struct ctl_ha_dt_req *rq;
13222	struct ctl_sg_entry *remote_sglist, *local_sglist;
13223	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13224	uint32_t local_used, remote_used, total_used;
13225	int retval;
13226	int i, j;
13227
13228	retval = 0;
13229
13230	rq = ctl_dt_req_alloc();
13231
13232	/*
13233	 * If we failed to allocate the request, and if the DMA didn't fail
13234	 * anyway, set busy status.  This is just a resource allocation
13235	 * failure.
13236	 */
13237	if ((rq == NULL)
13238	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13239		ctl_set_busy(&io->scsiio);
13240
13241	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13242
13243		if (rq != NULL)
13244			ctl_dt_req_free(rq);
13245
13246		/*
13247		 * The data move failed.  We need to return status back
13248		 * to the other controller.  No point in trying to DMA
13249		 * data to the remote controller.
13250		 */
13251
13252		ctl_send_datamove_done(io, /*have_lock*/ 0);
13253
13254		retval = 1;
13255
13256		goto bailout;
13257	}
13258
13259	local_sglist = io->io_hdr.local_sglist;
13260	local_dma_sglist = io->io_hdr.local_dma_sglist;
13261	remote_sglist = io->io_hdr.remote_sglist;
13262	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13263	local_used = 0;
13264	remote_used = 0;
13265	total_used = 0;
13266
13267	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13268		rq->ret = CTL_HA_STATUS_SUCCESS;
13269		rq->context = io;
13270		callback(rq);
13271		goto bailout;
13272	}
13273
13274	/*
13275	 * Pull/push the data over the wire from/to the other controller.
13276	 * This takes into account the possibility that the local and
13277	 * remote sglists may not be identical in terms of the size of
13278	 * the elements and the number of elements.
13279	 *
13280	 * One fundamental assumption here is that the length allocated for
13281	 * both the local and remote sglists is identical.  Otherwise, we've
13282	 * essentially got a coding error of some sort.
13283	 */
13284	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13285		int isc_ret;
13286		uint32_t cur_len, dma_length;
13287		uint8_t *tmp_ptr;
13288
13289		rq->id = CTL_HA_DATA_CTL;
13290		rq->command = command;
13291		rq->context = io;
13292
13293		/*
13294		 * Both pointers should be aligned.  But it is possible
13295		 * that the allocation length is not.  They should both
13296		 * also have enough slack left over at the end, though,
13297		 * to round up to the next 8 byte boundary.
13298		 */
13299		cur_len = ctl_min(local_sglist[i].len - local_used,
13300				  remote_sglist[j].len - remote_used);
13301
13302		/*
13303		 * In this case, we have a size issue and need to decrease
13304		 * the size, except in the case where we actually have less
13305		 * than 8 bytes left.  In that case, we need to increase
13306		 * the DMA length to get the last bit.
13307		 */
13308		if ((cur_len & 0x7) != 0) {
13309			if (cur_len > 0x7) {
13310				cur_len = cur_len - (cur_len & 0x7);
13311				dma_length = cur_len;
13312			} else {
13313				CTL_SIZE_8B(dma_length, cur_len);
13314			}
13315
13316		} else
13317			dma_length = cur_len;
13318
13319		/*
13320		 * If we had to allocate memory for this I/O, instead of using
13321		 * the non-cached mirror memory, we'll need to flush the cache
13322		 * before trying to DMA to the other controller.
13323		 *
13324		 * We could end up doing this multiple times for the same
13325		 * segment if we have a larger local segment than remote
13326		 * segment.  That shouldn't be an issue.
13327		 */
13328		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13329			/*
13330			 * XXX KDM use bus_dmamap_sync() here.
13331			 */
13332		}
13333
13334		rq->size = dma_length;
13335
13336		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13337		tmp_ptr += local_used;
13338
13339		/* Use physical addresses when talking to ISC hardware */
13340		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13341			/* XXX KDM use busdma */
13342#if 0
13343			rq->local = vtophys(tmp_ptr);
13344#endif
13345		} else
13346			rq->local = tmp_ptr;
13347
13348		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13349		tmp_ptr += remote_used;
13350		rq->remote = tmp_ptr;
13351
13352		rq->callback = NULL;
13353
13354		local_used += cur_len;
13355		if (local_used >= local_sglist[i].len) {
13356			i++;
13357			local_used = 0;
13358		}
13359
13360		remote_used += cur_len;
13361		if (remote_used >= remote_sglist[j].len) {
13362			j++;
13363			remote_used = 0;
13364		}
13365		total_used += cur_len;
13366
13367		if (total_used >= io->scsiio.kern_data_len)
13368			rq->callback = callback;
13369
13370		if ((rq->size & 0x7) != 0) {
13371			printf("%s: warning: size %d is not on 8b boundary\n",
13372			       __func__, rq->size);
13373		}
13374		if (((uintptr_t)rq->local & 0x7) != 0) {
13375			printf("%s: warning: local %p not on 8b boundary\n",
13376			       __func__, rq->local);
13377		}
13378		if (((uintptr_t)rq->remote & 0x7) != 0) {
13379			printf("%s: warning: remote %p not on 8b boundary\n",
13380			       __func__, rq->local);
13381		}
13382#if 0
13383		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13384		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13385		       rq->local, rq->remote, rq->size);
13386#endif
13387
13388		isc_ret = ctl_dt_single(rq);
13389		if (isc_ret == CTL_HA_STATUS_WAIT)
13390			continue;
13391
13392		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13393			rq->ret = CTL_HA_STATUS_SUCCESS;
13394		} else {
13395			rq->ret = isc_ret;
13396		}
13397		callback(rq);
13398		goto bailout;
13399	}
13400
13401bailout:
13402	return (retval);
13403
13404}
13405
13406static void
13407ctl_datamove_remote_read(union ctl_io *io)
13408{
13409	int retval;
13410	int i;
13411
13412	/*
13413	 * This will send an error to the other controller in the case of a
13414	 * failure.
13415	 */
13416	retval = ctl_datamove_remote_sgl_setup(io);
13417	if (retval != 0)
13418		return;
13419
13420	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13421					  ctl_datamove_remote_read_cb);
13422	if ((retval != 0)
13423	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13424		/*
13425		 * Make sure we free memory if there was an error..  The
13426		 * ctl_datamove_remote_xfer() function will send the
13427		 * datamove done message, or call the callback with an
13428		 * error if there is a problem.
13429		 */
13430		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13431			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13432	}
13433
13434	return;
13435}
13436
13437/*
13438 * Process a datamove request from the other controller.  This is used for
13439 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13440 * first.  Once that is complete, the data gets DMAed into the remote
13441 * controller's memory.  For reads, we DMA from the remote controller's
13442 * memory into our memory first, and then move it out to the FETD.
13443 */
13444static void
13445ctl_datamove_remote(union ctl_io *io)
13446{
13447	struct ctl_softc *softc;
13448
13449	softc = control_softc;
13450
13451	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13452
13453	/*
13454	 * Note that we look for an aborted I/O here, but don't do some of
13455	 * the other checks that ctl_datamove() normally does.
13456	 * We don't need to run the datamove delay code, since that should
13457	 * have been done if need be on the other controller.
13458	 */
13459	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13460		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13461		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13462		       io->io_hdr.nexus.targ_port,
13463		       io->io_hdr.nexus.targ_target.id,
13464		       io->io_hdr.nexus.targ_lun);
13465		io->io_hdr.port_status = 31338;
13466		ctl_send_datamove_done(io, /*have_lock*/ 0);
13467		return;
13468	}
13469
13470	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13471		ctl_datamove_remote_write(io);
13472	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13473		ctl_datamove_remote_read(io);
13474	} else {
13475		union ctl_ha_msg msg;
13476		struct scsi_sense_data *sense;
13477		uint8_t sks[3];
13478		int retry_count;
13479
13480		memset(&msg, 0, sizeof(msg));
13481
13482		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13483		msg.hdr.status = CTL_SCSI_ERROR;
13484		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13485
13486		retry_count = 4243;
13487
13488		sense = &msg.scsi.sense_data;
13489		sks[0] = SSD_SCS_VALID;
13490		sks[1] = (retry_count >> 8) & 0xff;
13491		sks[2] = retry_count & 0xff;
13492
13493		/* "Internal target failure" */
13494		scsi_set_sense_data(sense,
13495				    /*sense_format*/ SSD_TYPE_NONE,
13496				    /*current_error*/ 1,
13497				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13498				    /*asc*/ 0x44,
13499				    /*ascq*/ 0x00,
13500				    /*type*/ SSD_ELEM_SKS,
13501				    /*size*/ sizeof(sks),
13502				    /*data*/ sks,
13503				    SSD_ELEM_NONE);
13504
13505		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13506		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13507			ctl_failover_io(io, /*have_lock*/ 1);
13508			return;
13509		}
13510
13511		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13512		    CTL_HA_STATUS_SUCCESS) {
13513			/* XXX KDM what to do if this fails? */
13514		}
13515		return;
13516	}
13517
13518}
13519
13520static int
13521ctl_process_done(union ctl_io *io)
13522{
13523	struct ctl_lun *lun;
13524	struct ctl_softc *ctl_softc;
13525	void (*fe_done)(union ctl_io *io);
13526	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13527
13528	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13529
13530	fe_done =
13531	    control_softc->ctl_ports[targ_port]->fe_done;
13532
13533#ifdef CTL_TIME_IO
13534	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13535		char str[256];
13536		char path_str[64];
13537		struct sbuf sb;
13538
13539		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13540		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13541
13542		sbuf_cat(&sb, path_str);
13543		switch (io->io_hdr.io_type) {
13544		case CTL_IO_SCSI:
13545			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13546			sbuf_printf(&sb, "\n");
13547			sbuf_cat(&sb, path_str);
13548			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13549				    io->scsiio.tag_num, io->scsiio.tag_type);
13550			break;
13551		case CTL_IO_TASK:
13552			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13553				    "Tag Type: %d\n", io->taskio.task_action,
13554				    io->taskio.tag_num, io->taskio.tag_type);
13555			break;
13556		default:
13557			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13558			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13559			break;
13560		}
13561		sbuf_cat(&sb, path_str);
13562		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13563			    (intmax_t)time_uptime - io->io_hdr.start_time);
13564		sbuf_finish(&sb);
13565		printf("%s", sbuf_data(&sb));
13566	}
13567#endif /* CTL_TIME_IO */
13568
13569	switch (io->io_hdr.io_type) {
13570	case CTL_IO_SCSI:
13571		break;
13572	case CTL_IO_TASK:
13573		if (bootverbose || verbose > 0)
13574			ctl_io_error_print(io, NULL);
13575		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13576			ctl_free_io(io);
13577		else
13578			fe_done(io);
13579		return (CTL_RETVAL_COMPLETE);
13580		break;
13581	default:
13582		printf("ctl_process_done: invalid io type %d\n",
13583		       io->io_hdr.io_type);
13584		panic("ctl_process_done: invalid io type %d\n",
13585		      io->io_hdr.io_type);
13586		break; /* NOTREACHED */
13587	}
13588
13589	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13590	if (lun == NULL) {
13591		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13592				 io->io_hdr.nexus.targ_mapped_lun));
13593		fe_done(io);
13594		goto bailout;
13595	}
13596	ctl_softc = lun->ctl_softc;
13597
13598	mtx_lock(&lun->lun_lock);
13599
13600	/*
13601	 * Check to see if we have any errors to inject here.  We only
13602	 * inject errors for commands that don't already have errors set.
13603	 */
13604	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13605	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13606		ctl_inject_error(lun, io);
13607
13608	/*
13609	 * XXX KDM how do we treat commands that aren't completed
13610	 * successfully?
13611	 *
13612	 * XXX KDM should we also track I/O latency?
13613	 */
13614	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13615	    io->io_hdr.io_type == CTL_IO_SCSI) {
13616#ifdef CTL_TIME_IO
13617		struct bintime cur_bt;
13618#endif
13619		int type;
13620
13621		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13622		    CTL_FLAG_DATA_IN)
13623			type = CTL_STATS_READ;
13624		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13625		    CTL_FLAG_DATA_OUT)
13626			type = CTL_STATS_WRITE;
13627		else
13628			type = CTL_STATS_NO_IO;
13629
13630		lun->stats.ports[targ_port].bytes[type] +=
13631		    io->scsiio.kern_total_len;
13632		lun->stats.ports[targ_port].operations[type]++;
13633#ifdef CTL_TIME_IO
13634		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13635		   &io->io_hdr.dma_bt);
13636		lun->stats.ports[targ_port].num_dmas[type] +=
13637		    io->io_hdr.num_dmas;
13638		getbintime(&cur_bt);
13639		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13640		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13641#endif
13642	}
13643
13644	/*
13645	 * Remove this from the OOA queue.
13646	 */
13647	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13648
13649	/*
13650	 * Run through the blocked queue on this LUN and see if anything
13651	 * has become unblocked, now that this transaction is done.
13652	 */
13653	ctl_check_blocked(lun);
13654
13655	/*
13656	 * If the LUN has been invalidated, free it if there is nothing
13657	 * left on its OOA queue.
13658	 */
13659	if ((lun->flags & CTL_LUN_INVALID)
13660	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13661		mtx_unlock(&lun->lun_lock);
13662		mtx_lock(&ctl_softc->ctl_lock);
13663		ctl_free_lun(lun);
13664		mtx_unlock(&ctl_softc->ctl_lock);
13665	} else
13666		mtx_unlock(&lun->lun_lock);
13667
13668	/*
13669	 * If this command has been aborted, make sure we set the status
13670	 * properly.  The FETD is responsible for freeing the I/O and doing
13671	 * whatever it needs to do to clean up its state.
13672	 */
13673	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13674		ctl_set_task_aborted(&io->scsiio);
13675
13676	/*
13677	 * We print out status for every task management command.  For SCSI
13678	 * commands, we filter out any unit attention errors; they happen
13679	 * on every boot, and would clutter up the log.  Note:  task
13680	 * management commands aren't printed here, they are printed above,
13681	 * since they should never even make it down here.
13682	 */
13683	switch (io->io_hdr.io_type) {
13684	case CTL_IO_SCSI: {
13685		int error_code, sense_key, asc, ascq;
13686
13687		sense_key = 0;
13688
13689		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13690		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13691			/*
13692			 * Since this is just for printing, no need to
13693			 * show errors here.
13694			 */
13695			scsi_extract_sense_len(&io->scsiio.sense_data,
13696					       io->scsiio.sense_len,
13697					       &error_code,
13698					       &sense_key,
13699					       &asc,
13700					       &ascq,
13701					       /*show_errors*/ 0);
13702		}
13703
13704		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13705		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13706		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13707		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13708
13709			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13710				ctl_softc->skipped_prints++;
13711			} else {
13712				uint32_t skipped_prints;
13713
13714				skipped_prints = ctl_softc->skipped_prints;
13715
13716				ctl_softc->skipped_prints = 0;
13717				ctl_softc->last_print_jiffies = time_uptime;
13718
13719				if (skipped_prints > 0) {
13720#ifdef NEEDTOPORT
13721					csevent_log(CSC_CTL | CSC_SHELF_SW |
13722					    CTL_ERROR_REPORT,
13723					    csevent_LogType_Trace,
13724					    csevent_Severity_Information,
13725					    csevent_AlertLevel_Green,
13726					    csevent_FRU_Firmware,
13727					    csevent_FRU_Unknown,
13728					    "High CTL error volume, %d prints "
13729					    "skipped", skipped_prints);
13730#endif
13731				}
13732				if (bootverbose || verbose > 0)
13733					ctl_io_error_print(io, NULL);
13734			}
13735		}
13736		break;
13737	}
13738	case CTL_IO_TASK:
13739		if (bootverbose || verbose > 0)
13740			ctl_io_error_print(io, NULL);
13741		break;
13742	default:
13743		break;
13744	}
13745
13746	/*
13747	 * Tell the FETD or the other shelf controller we're done with this
13748	 * command.  Note that only SCSI commands get to this point.  Task
13749	 * management commands are completed above.
13750	 *
13751	 * We only send status to the other controller if we're in XFER
13752	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13753	 * received the I/O (from CTL's perspective), and so the status is
13754	 * generated there.
13755	 *
13756	 * XXX KDM if we hold the lock here, we could cause a deadlock
13757	 * if the frontend comes back in in this context to queue
13758	 * something.
13759	 */
13760	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13761	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13762		union ctl_ha_msg msg;
13763
13764		memset(&msg, 0, sizeof(msg));
13765		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13766		msg.hdr.original_sc = io->io_hdr.original_sc;
13767		msg.hdr.nexus = io->io_hdr.nexus;
13768		msg.hdr.status = io->io_hdr.status;
13769		msg.scsi.scsi_status = io->scsiio.scsi_status;
13770		msg.scsi.tag_num = io->scsiio.tag_num;
13771		msg.scsi.tag_type = io->scsiio.tag_type;
13772		msg.scsi.sense_len = io->scsiio.sense_len;
13773		msg.scsi.sense_residual = io->scsiio.sense_residual;
13774		msg.scsi.residual = io->scsiio.residual;
13775		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13776		       sizeof(io->scsiio.sense_data));
13777		/*
13778		 * We copy this whether or not this is an I/O-related
13779		 * command.  Otherwise, we'd have to go and check to see
13780		 * whether it's a read/write command, and it really isn't
13781		 * worth it.
13782		 */
13783		memcpy(&msg.scsi.lbalen,
13784		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13785		       sizeof(msg.scsi.lbalen));
13786
13787		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13788				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13789			/* XXX do something here */
13790		}
13791
13792		ctl_free_io(io);
13793	} else
13794		fe_done(io);
13795
13796bailout:
13797
13798	return (CTL_RETVAL_COMPLETE);
13799}
13800
13801#ifdef CTL_WITH_CA
13802/*
13803 * Front end should call this if it doesn't do autosense.  When the request
13804 * sense comes back in from the initiator, we'll dequeue this and send it.
13805 */
13806int
13807ctl_queue_sense(union ctl_io *io)
13808{
13809	struct ctl_lun *lun;
13810	struct ctl_softc *ctl_softc;
13811	uint32_t initidx, targ_lun;
13812
13813	ctl_softc = control_softc;
13814
13815	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13816
13817	/*
13818	 * LUN lookup will likely move to the ctl_work_thread() once we
13819	 * have our new queueing infrastructure (that doesn't put things on
13820	 * a per-LUN queue initially).  That is so that we can handle
13821	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13822	 * can't deal with that right now.
13823	 */
13824	mtx_lock(&ctl_softc->ctl_lock);
13825
13826	/*
13827	 * If we don't have a LUN for this, just toss the sense
13828	 * information.
13829	 */
13830	targ_lun = io->io_hdr.nexus.targ_lun;
13831	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13832	if ((targ_lun < CTL_MAX_LUNS)
13833	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13834		lun = ctl_softc->ctl_luns[targ_lun];
13835	else
13836		goto bailout;
13837
13838	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13839
13840	mtx_lock(&lun->lun_lock);
13841	/*
13842	 * Already have CA set for this LUN...toss the sense information.
13843	 */
13844	if (ctl_is_set(lun->have_ca, initidx)) {
13845		mtx_unlock(&lun->lun_lock);
13846		goto bailout;
13847	}
13848
13849	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13850	       ctl_min(sizeof(lun->pending_sense[initidx]),
13851	       sizeof(io->scsiio.sense_data)));
13852	ctl_set_mask(lun->have_ca, initidx);
13853	mtx_unlock(&lun->lun_lock);
13854
13855bailout:
13856	mtx_unlock(&ctl_softc->ctl_lock);
13857
13858	ctl_free_io(io);
13859
13860	return (CTL_RETVAL_COMPLETE);
13861}
13862#endif
13863
13864/*
13865 * Primary command inlet from frontend ports.  All SCSI and task I/O
13866 * requests must go through this function.
13867 */
13868int
13869ctl_queue(union ctl_io *io)
13870{
13871	struct ctl_softc *ctl_softc;
13872
13873	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13874
13875	ctl_softc = control_softc;
13876
13877#ifdef CTL_TIME_IO
13878	io->io_hdr.start_time = time_uptime;
13879	getbintime(&io->io_hdr.start_bt);
13880#endif /* CTL_TIME_IO */
13881
13882	/* Map FE-specific LUN ID into global one. */
13883	io->io_hdr.nexus.targ_mapped_lun =
13884	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13885
13886	switch (io->io_hdr.io_type) {
13887	case CTL_IO_SCSI:
13888	case CTL_IO_TASK:
13889		ctl_enqueue_incoming(io);
13890		break;
13891	default:
13892		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13893		return (EINVAL);
13894	}
13895
13896	return (CTL_RETVAL_COMPLETE);
13897}
13898
13899#ifdef CTL_IO_DELAY
13900static void
13901ctl_done_timer_wakeup(void *arg)
13902{
13903	union ctl_io *io;
13904
13905	io = (union ctl_io *)arg;
13906	ctl_done(io);
13907}
13908#endif /* CTL_IO_DELAY */
13909
13910void
13911ctl_done(union ctl_io *io)
13912{
13913	struct ctl_softc *ctl_softc;
13914
13915	ctl_softc = control_softc;
13916
13917	/*
13918	 * Enable this to catch duplicate completion issues.
13919	 */
13920#if 0
13921	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13922		printf("%s: type %d msg %d cdb %x iptl: "
13923		       "%d:%d:%d:%d tag 0x%04x "
13924		       "flag %#x status %x\n",
13925			__func__,
13926			io->io_hdr.io_type,
13927			io->io_hdr.msg_type,
13928			io->scsiio.cdb[0],
13929			io->io_hdr.nexus.initid.id,
13930			io->io_hdr.nexus.targ_port,
13931			io->io_hdr.nexus.targ_target.id,
13932			io->io_hdr.nexus.targ_lun,
13933			(io->io_hdr.io_type ==
13934			CTL_IO_TASK) ?
13935			io->taskio.tag_num :
13936			io->scsiio.tag_num,
13937		        io->io_hdr.flags,
13938			io->io_hdr.status);
13939	} else
13940		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13941#endif
13942
13943	/*
13944	 * This is an internal copy of an I/O, and should not go through
13945	 * the normal done processing logic.
13946	 */
13947	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13948		return;
13949
13950	/*
13951	 * We need to send a msg to the serializing shelf to finish the IO
13952	 * as well.  We don't send a finish message to the other shelf if
13953	 * this is a task management command.  Task management commands
13954	 * aren't serialized in the OOA queue, but rather just executed on
13955	 * both shelf controllers for commands that originated on that
13956	 * controller.
13957	 */
13958	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13959	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13960		union ctl_ha_msg msg_io;
13961
13962		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13963		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13964		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13965		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13966		}
13967		/* continue on to finish IO */
13968	}
13969#ifdef CTL_IO_DELAY
13970	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13971		struct ctl_lun *lun;
13972
13973		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13974
13975		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13976	} else {
13977		struct ctl_lun *lun;
13978
13979		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13980
13981		if ((lun != NULL)
13982		 && (lun->delay_info.done_delay > 0)) {
13983			struct callout *callout;
13984
13985			callout = (struct callout *)&io->io_hdr.timer_bytes;
13986			callout_init(callout, /*mpsafe*/ 1);
13987			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13988			callout_reset(callout,
13989				      lun->delay_info.done_delay * hz,
13990				      ctl_done_timer_wakeup, io);
13991			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13992				lun->delay_info.done_delay = 0;
13993			return;
13994		}
13995	}
13996#endif /* CTL_IO_DELAY */
13997
13998	ctl_enqueue_done(io);
13999}
14000
14001int
14002ctl_isc(struct ctl_scsiio *ctsio)
14003{
14004	struct ctl_lun *lun;
14005	int retval;
14006
14007	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
14008
14009	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
14010
14011	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
14012
14013	retval = lun->backend->data_submit((union ctl_io *)ctsio);
14014
14015	return (retval);
14016}
14017
14018
14019static void
14020ctl_work_thread(void *arg)
14021{
14022	struct ctl_thread *thr = (struct ctl_thread *)arg;
14023	struct ctl_softc *softc = thr->ctl_softc;
14024	union ctl_io *io;
14025	int retval;
14026
14027	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
14028
14029	for (;;) {
14030		retval = 0;
14031
14032		/*
14033		 * We handle the queues in this order:
14034		 * - ISC
14035		 * - done queue (to free up resources, unblock other commands)
14036		 * - RtR queue
14037		 * - incoming queue
14038		 *
14039		 * If those queues are empty, we break out of the loop and
14040		 * go to sleep.
14041		 */
14042		mtx_lock(&thr->queue_lock);
14043		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
14044		if (io != NULL) {
14045			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
14046			mtx_unlock(&thr->queue_lock);
14047			ctl_handle_isc(io);
14048			continue;
14049		}
14050		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
14051		if (io != NULL) {
14052			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
14053			/* clear any blocked commands, call fe_done */
14054			mtx_unlock(&thr->queue_lock);
14055			retval = ctl_process_done(io);
14056			continue;
14057		}
14058		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
14059		if (io != NULL) {
14060			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
14061			mtx_unlock(&thr->queue_lock);
14062			if (io->io_hdr.io_type == CTL_IO_TASK)
14063				ctl_run_task(io);
14064			else
14065				ctl_scsiio_precheck(softc, &io->scsiio);
14066			continue;
14067		}
14068		if (!ctl_pause_rtr) {
14069			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
14070			if (io != NULL) {
14071				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
14072				mtx_unlock(&thr->queue_lock);
14073				retval = ctl_scsiio(&io->scsiio);
14074				if (retval != CTL_RETVAL_COMPLETE)
14075					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
14076				continue;
14077			}
14078		}
14079
14080		/* Sleep until we have something to do. */
14081		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
14082	}
14083}
14084
14085static void
14086ctl_lun_thread(void *arg)
14087{
14088	struct ctl_softc *softc = (struct ctl_softc *)arg;
14089	struct ctl_be_lun *be_lun;
14090	int retval;
14091
14092	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14093
14094	for (;;) {
14095		retval = 0;
14096		mtx_lock(&softc->ctl_lock);
14097		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14098		if (be_lun != NULL) {
14099			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14100			mtx_unlock(&softc->ctl_lock);
14101			ctl_create_lun(be_lun);
14102			continue;
14103		}
14104
14105		/* Sleep until we have something to do. */
14106		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14107		    PDROP | PRIBIO, "-", 0);
14108	}
14109}
14110
14111static void
14112ctl_enqueue_incoming(union ctl_io *io)
14113{
14114	struct ctl_softc *softc = control_softc;
14115	struct ctl_thread *thr;
14116	u_int idx;
14117
14118	idx = (io->io_hdr.nexus.targ_port * 127 +
14119	       io->io_hdr.nexus.initid.id) % worker_threads;
14120	thr = &softc->threads[idx];
14121	mtx_lock(&thr->queue_lock);
14122	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14123	mtx_unlock(&thr->queue_lock);
14124	wakeup(thr);
14125}
14126
14127static void
14128ctl_enqueue_rtr(union ctl_io *io)
14129{
14130	struct ctl_softc *softc = control_softc;
14131	struct ctl_thread *thr;
14132
14133	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14134	mtx_lock(&thr->queue_lock);
14135	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14136	mtx_unlock(&thr->queue_lock);
14137	wakeup(thr);
14138}
14139
14140static void
14141ctl_enqueue_done(union ctl_io *io)
14142{
14143	struct ctl_softc *softc = control_softc;
14144	struct ctl_thread *thr;
14145
14146	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14147	mtx_lock(&thr->queue_lock);
14148	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14149	mtx_unlock(&thr->queue_lock);
14150	wakeup(thr);
14151}
14152
14153static void
14154ctl_enqueue_isc(union ctl_io *io)
14155{
14156	struct ctl_softc *softc = control_softc;
14157	struct ctl_thread *thr;
14158
14159	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14160	mtx_lock(&thr->queue_lock);
14161	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14162	mtx_unlock(&thr->queue_lock);
14163	wakeup(thr);
14164}
14165
14166/* Initialization and failover */
14167
14168void
14169ctl_init_isc_msg(void)
14170{
14171	printf("CTL: Still calling this thing\n");
14172}
14173
14174/*
14175 * Init component
14176 * 	Initializes component into configuration defined by bootMode
14177 *	(see hasc-sv.c)
14178 *  	returns hasc_Status:
14179 * 		OK
14180 *		ERROR - fatal error
14181 */
14182static ctl_ha_comp_status
14183ctl_isc_init(struct ctl_ha_component *c)
14184{
14185	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14186
14187	c->status = ret;
14188	return ret;
14189}
14190
14191/* Start component
14192 * 	Starts component in state requested. If component starts successfully,
14193 *	it must set its own state to the requestrd state
14194 *	When requested state is HASC_STATE_HA, the component may refine it
14195 * 	by adding _SLAVE or _MASTER flags.
14196 *	Currently allowed state transitions are:
14197 *	UNKNOWN->HA		- initial startup
14198 *	UNKNOWN->SINGLE - initial startup when no parter detected
14199 *	HA->SINGLE		- failover
14200 * returns ctl_ha_comp_status:
14201 * 		OK	- component successfully started in requested state
14202 *		FAILED  - could not start the requested state, failover may
14203 * 			  be possible
14204 *		ERROR	- fatal error detected, no future startup possible
14205 */
14206static ctl_ha_comp_status
14207ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14208{
14209	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14210
14211	printf("%s: go\n", __func__);
14212
14213	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14214	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14215		ctl_is_single = 0;
14216		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14217		    != CTL_HA_STATUS_SUCCESS) {
14218			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14219			ret = CTL_HA_COMP_STATUS_ERROR;
14220		}
14221	} else if (CTL_HA_STATE_IS_HA(c->state)
14222		&& CTL_HA_STATE_IS_SINGLE(state)){
14223		// HA->SINGLE transition
14224	        ctl_failover();
14225		ctl_is_single = 1;
14226	} else {
14227		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14228		       c->state, state);
14229		ret = CTL_HA_COMP_STATUS_ERROR;
14230	}
14231	if (CTL_HA_STATE_IS_SINGLE(state))
14232		ctl_is_single = 1;
14233
14234	c->state = state;
14235	c->status = ret;
14236	return ret;
14237}
14238
14239/*
14240 * Quiesce component
14241 * The component must clear any error conditions (set status to OK) and
14242 * prepare itself to another Start call
14243 * returns ctl_ha_comp_status:
14244 * 	OK
14245 *	ERROR
14246 */
14247static ctl_ha_comp_status
14248ctl_isc_quiesce(struct ctl_ha_component *c)
14249{
14250	int ret = CTL_HA_COMP_STATUS_OK;
14251
14252	ctl_pause_rtr = 1;
14253	c->status = ret;
14254	return ret;
14255}
14256
14257struct ctl_ha_component ctl_ha_component_ctlisc =
14258{
14259	.name = "CTL ISC",
14260	.state = CTL_HA_STATE_UNKNOWN,
14261	.init = ctl_isc_init,
14262	.start = ctl_isc_start,
14263	.quiesce = ctl_isc_quiesce
14264};
14265
14266/*
14267 *  vim: ts=8
14268 */
14269