ctl.c revision 268096
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 268096 2014-07-01 15:05:23Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * Size and alignment macros needed for Copan-specific HA hardware.  These
87 * can go away when the HA code is re-written, and uses busdma for any
88 * hardware.
89 */
90#define	CTL_ALIGN_8B(target, source, type)				\
91	if (((uint32_t)source & 0x7) != 0)				\
92		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
93	else								\
94		target = (type)source;
95
96#define	CTL_SIZE_8B(target, size)					\
97	if ((size & 0x7) != 0)						\
98		target = size + (0x8 - (size & 0x7));			\
99	else								\
100		target = size;
101
102#define CTL_ALIGN_8B_MARGIN	16
103
104/*
105 * Template mode pages.
106 */
107
108/*
109 * Note that these are default values only.  The actual values will be
110 * filled in when the user does a mode sense.
111 */
112static struct copan_power_subpage power_page_default = {
113	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
114	/*subpage*/ PWR_SUBPAGE_CODE,
115	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
116			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
117	/*page_version*/ PWR_VERSION,
118	/* total_luns */ 26,
119	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
120	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
121		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
122		      0, 0, 0, 0, 0, 0}
123};
124
125static struct copan_power_subpage power_page_changeable = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ 0,
131	/* total_luns */ 0,
132	/* max_active_luns*/ 0,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_aps_subpage aps_page_default = {
139	APS_PAGE_CODE | SMPH_SPF, //page_code
140	APS_SUBPAGE_CODE, //subpage
141	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
142	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
143	APS_VERSION, //page_version
144	0, //lock_active
145	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
146	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
147	0, 0, 0, 0, 0} //reserved
148};
149
150static struct copan_aps_subpage aps_page_changeable = {
151	APS_PAGE_CODE | SMPH_SPF, //page_code
152	APS_SUBPAGE_CODE, //subpage
153	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
154	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
155	0, //page_version
156	0, //lock_active
157	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
158	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0} //reserved
160};
161
162static struct copan_debugconf_subpage debugconf_page_default = {
163	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
164	DBGCNF_SUBPAGE_CODE,		/* subpage */
165	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
166	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
167	DBGCNF_VERSION,			/* page_version */
168	{CTL_TIME_IO_DEFAULT_SECS>>8,
169	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
170};
171
172static struct copan_debugconf_subpage debugconf_page_changeable = {
173	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
174	DBGCNF_SUBPAGE_CODE,		/* subpage */
175	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
176	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
177	0,				/* page_version */
178	{0xff,0xff},			/* ctl_time_io_secs */
179};
180
181static struct scsi_format_page format_page_default = {
182	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
183	/*page_length*/sizeof(struct scsi_format_page) - 2,
184	/*tracks_per_zone*/ {0, 0},
185	/*alt_sectors_per_zone*/ {0, 0},
186	/*alt_tracks_per_zone*/ {0, 0},
187	/*alt_tracks_per_lun*/ {0, 0},
188	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
189			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
190	/*bytes_per_sector*/ {0, 0},
191	/*interleave*/ {0, 0},
192	/*track_skew*/ {0, 0},
193	/*cylinder_skew*/ {0, 0},
194	/*flags*/ SFP_HSEC,
195	/*reserved*/ {0, 0, 0}
196};
197
198static struct scsi_format_page format_page_changeable = {
199	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
200	/*page_length*/sizeof(struct scsi_format_page) - 2,
201	/*tracks_per_zone*/ {0, 0},
202	/*alt_sectors_per_zone*/ {0, 0},
203	/*alt_tracks_per_zone*/ {0, 0},
204	/*alt_tracks_per_lun*/ {0, 0},
205	/*sectors_per_track*/ {0, 0},
206	/*bytes_per_sector*/ {0, 0},
207	/*interleave*/ {0, 0},
208	/*track_skew*/ {0, 0},
209	/*cylinder_skew*/ {0, 0},
210	/*flags*/ 0,
211	/*reserved*/ {0, 0, 0}
212};
213
214static struct scsi_rigid_disk_page rigid_disk_page_default = {
215	/*page_code*/SMS_RIGID_DISK_PAGE,
216	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
217	/*cylinders*/ {0, 0, 0},
218	/*heads*/ CTL_DEFAULT_HEADS,
219	/*start_write_precomp*/ {0, 0, 0},
220	/*start_reduced_current*/ {0, 0, 0},
221	/*step_rate*/ {0, 0},
222	/*landing_zone_cylinder*/ {0, 0, 0},
223	/*rpl*/ SRDP_RPL_DISABLED,
224	/*rotational_offset*/ 0,
225	/*reserved1*/ 0,
226	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
227			   CTL_DEFAULT_ROTATION_RATE & 0xff},
228	/*reserved2*/ {0, 0}
229};
230
231static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
232	/*page_code*/SMS_RIGID_DISK_PAGE,
233	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
234	/*cylinders*/ {0, 0, 0},
235	/*heads*/ 0,
236	/*start_write_precomp*/ {0, 0, 0},
237	/*start_reduced_current*/ {0, 0, 0},
238	/*step_rate*/ {0, 0},
239	/*landing_zone_cylinder*/ {0, 0, 0},
240	/*rpl*/ 0,
241	/*rotational_offset*/ 0,
242	/*reserved1*/ 0,
243	/*rotation_rate*/ {0, 0},
244	/*reserved2*/ {0, 0}
245};
246
247static struct scsi_caching_page caching_page_default = {
248	/*page_code*/SMS_CACHING_PAGE,
249	/*page_length*/sizeof(struct scsi_caching_page) - 2,
250	/*flags1*/ SCP_DISC | SCP_WCE,
251	/*ret_priority*/ 0,
252	/*disable_pf_transfer_len*/ {0xff, 0xff},
253	/*min_prefetch*/ {0, 0},
254	/*max_prefetch*/ {0xff, 0xff},
255	/*max_pf_ceiling*/ {0xff, 0xff},
256	/*flags2*/ 0,
257	/*cache_segments*/ 0,
258	/*cache_seg_size*/ {0, 0},
259	/*reserved*/ 0,
260	/*non_cache_seg_size*/ {0, 0, 0}
261};
262
263static struct scsi_caching_page caching_page_changeable = {
264	/*page_code*/SMS_CACHING_PAGE,
265	/*page_length*/sizeof(struct scsi_caching_page) - 2,
266	/*flags1*/ 0,
267	/*ret_priority*/ 0,
268	/*disable_pf_transfer_len*/ {0, 0},
269	/*min_prefetch*/ {0, 0},
270	/*max_prefetch*/ {0, 0},
271	/*max_pf_ceiling*/ {0, 0},
272	/*flags2*/ 0,
273	/*cache_segments*/ 0,
274	/*cache_seg_size*/ {0, 0},
275	/*reserved*/ 0,
276	/*non_cache_seg_size*/ {0, 0, 0}
277};
278
279static struct scsi_control_page control_page_default = {
280	/*page_code*/SMS_CONTROL_MODE_PAGE,
281	/*page_length*/sizeof(struct scsi_control_page) - 2,
282	/*rlec*/0,
283	/*queue_flags*/0,
284	/*eca_and_aen*/0,
285	/*reserved*/0,
286	/*aen_holdoff_period*/{0, 0}
287};
288
289static struct scsi_control_page control_page_changeable = {
290	/*page_code*/SMS_CONTROL_MODE_PAGE,
291	/*page_length*/sizeof(struct scsi_control_page) - 2,
292	/*rlec*/SCP_DSENSE,
293	/*queue_flags*/0,
294	/*eca_and_aen*/0,
295	/*reserved*/0,
296	/*aen_holdoff_period*/{0, 0}
297};
298
299
300/*
301 * XXX KDM move these into the softc.
302 */
303static int rcv_sync_msg;
304static int persis_offset;
305static uint8_t ctl_pause_rtr;
306static int     ctl_is_single = 1;
307static int     index_to_aps_page;
308
309SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
310static int worker_threads = -1;
311SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
312    &worker_threads, 1, "Number of worker threads");
313static int verbose = 0;
314SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
315    &verbose, 0, "Show SCSI errors returned to initiator");
316
317/*
318 * Serial number (0x80), device id (0x83), supported pages (0x00),
319 * Block limits (0xB0) and Logical Block Provisioning (0xB2)
320 */
321#define SCSI_EVPD_NUM_SUPPORTED_PAGES	5
322
323static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
324				  int param);
325static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
326static int ctl_init(void);
327void ctl_shutdown(void);
328static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
329static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
330static void ctl_ioctl_online(void *arg);
331static void ctl_ioctl_offline(void *arg);
332static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
333static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
334static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
335static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
336static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
337static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
338static int ctl_ioctl_submit_wait(union ctl_io *io);
339static void ctl_ioctl_datamove(union ctl_io *io);
340static void ctl_ioctl_done(union ctl_io *io);
341static void ctl_ioctl_hard_startstop_callback(void *arg,
342					      struct cfi_metatask *metatask);
343static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
344static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
345			      struct ctl_ooa *ooa_hdr,
346			      struct ctl_ooa_entry *kern_entries);
347static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
348		     struct thread *td);
349uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
350uint32_t ctl_port_idx(int port_num);
351#ifdef unused
352static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
353				   uint32_t targ_target, uint32_t targ_lun,
354				   int can_wait);
355static void ctl_kfree_io(union ctl_io *io);
356#endif /* unused */
357static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
358			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
359static int ctl_free_lun(struct ctl_lun *lun);
360static void ctl_create_lun(struct ctl_be_lun *be_lun);
361/**
362static void ctl_failover_change_pages(struct ctl_softc *softc,
363				      struct ctl_scsiio *ctsio, int master);
364**/
365
366static int ctl_do_mode_select(union ctl_io *io);
367static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
368			   uint64_t res_key, uint64_t sa_res_key,
369			   uint8_t type, uint32_t residx,
370			   struct ctl_scsiio *ctsio,
371			   struct scsi_per_res_out *cdb,
372			   struct scsi_per_res_out_parms* param);
373static void ctl_pro_preempt_other(struct ctl_lun *lun,
374				  union ctl_ha_msg *msg);
375static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
376static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
377static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
378static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
379static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
380					 int alloc_len);
381static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
382static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
383static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
384static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
385static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
386static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
387					 union ctl_io *ooa_io);
388static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
389				union ctl_io *starting_io);
390static int ctl_check_blocked(struct ctl_lun *lun);
391static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
392				struct ctl_lun *lun,
393				const struct ctl_cmd_entry *entry,
394				struct ctl_scsiio *ctsio);
395//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
396static void ctl_failover(void);
397static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
398			       struct ctl_scsiio *ctsio);
399static int ctl_scsiio(struct ctl_scsiio *ctsio);
400
401static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
402static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
403			    ctl_ua_type ua_type);
404static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
405			 ctl_ua_type ua_type);
406static int ctl_abort_task(union ctl_io *io);
407static void ctl_run_task(union ctl_io *io);
408#ifdef CTL_IO_DELAY
409static void ctl_datamove_timer_wakeup(void *arg);
410static void ctl_done_timer_wakeup(void *arg);
411#endif /* CTL_IO_DELAY */
412
413static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
414static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
415static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
416static void ctl_datamove_remote_write(union ctl_io *io);
417static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
418static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
419static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
420static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
421				    ctl_ha_dt_cb callback);
422static void ctl_datamove_remote_read(union ctl_io *io);
423static void ctl_datamove_remote(union ctl_io *io);
424static int ctl_process_done(union ctl_io *io);
425static void ctl_lun_thread(void *arg);
426static void ctl_work_thread(void *arg);
427static void ctl_enqueue_incoming(union ctl_io *io);
428static void ctl_enqueue_rtr(union ctl_io *io);
429static void ctl_enqueue_done(union ctl_io *io);
430static void ctl_enqueue_isc(union ctl_io *io);
431static const struct ctl_cmd_entry *
432    ctl_get_cmd_entry(struct ctl_scsiio *ctsio);
433static const struct ctl_cmd_entry *
434    ctl_validate_command(struct ctl_scsiio *ctsio);
435static int ctl_cmd_applicable(uint8_t lun_type,
436    const struct ctl_cmd_entry *entry);
437
438/*
439 * Load the serialization table.  This isn't very pretty, but is probably
440 * the easiest way to do it.
441 */
442#include "ctl_ser_table.c"
443
444/*
445 * We only need to define open, close and ioctl routines for this driver.
446 */
447static struct cdevsw ctl_cdevsw = {
448	.d_version =	D_VERSION,
449	.d_flags =	0,
450	.d_open =	ctl_open,
451	.d_close =	ctl_close,
452	.d_ioctl =	ctl_ioctl,
453	.d_name =	"ctl",
454};
455
456
457MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
458
459static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
460
461static moduledata_t ctl_moduledata = {
462	"ctl",
463	ctl_module_event_handler,
464	NULL
465};
466
467DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
468MODULE_VERSION(ctl, 1);
469
470static void
471ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
472			    union ctl_ha_msg *msg_info)
473{
474	struct ctl_scsiio *ctsio;
475
476	if (msg_info->hdr.original_sc == NULL) {
477		printf("%s: original_sc == NULL!\n", __func__);
478		/* XXX KDM now what? */
479		return;
480	}
481
482	ctsio = &msg_info->hdr.original_sc->scsiio;
483	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
484	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
485	ctsio->io_hdr.status = msg_info->hdr.status;
486	ctsio->scsi_status = msg_info->scsi.scsi_status;
487	ctsio->sense_len = msg_info->scsi.sense_len;
488	ctsio->sense_residual = msg_info->scsi.sense_residual;
489	ctsio->residual = msg_info->scsi.residual;
490	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
491	       sizeof(ctsio->sense_data));
492	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
493	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
494	ctl_enqueue_isc((union ctl_io *)ctsio);
495}
496
497static void
498ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
499				union ctl_ha_msg *msg_info)
500{
501	struct ctl_scsiio *ctsio;
502
503	if (msg_info->hdr.serializing_sc == NULL) {
504		printf("%s: serializing_sc == NULL!\n", __func__);
505		/* XXX KDM now what? */
506		return;
507	}
508
509	ctsio = &msg_info->hdr.serializing_sc->scsiio;
510#if 0
511	/*
512	 * Attempt to catch the situation where an I/O has
513	 * been freed, and we're using it again.
514	 */
515	if (ctsio->io_hdr.io_type == 0xff) {
516		union ctl_io *tmp_io;
517		tmp_io = (union ctl_io *)ctsio;
518		printf("%s: %p use after free!\n", __func__,
519		       ctsio);
520		printf("%s: type %d msg %d cdb %x iptl: "
521		       "%d:%d:%d:%d tag 0x%04x "
522		       "flag %#x status %x\n",
523			__func__,
524			tmp_io->io_hdr.io_type,
525			tmp_io->io_hdr.msg_type,
526			tmp_io->scsiio.cdb[0],
527			tmp_io->io_hdr.nexus.initid.id,
528			tmp_io->io_hdr.nexus.targ_port,
529			tmp_io->io_hdr.nexus.targ_target.id,
530			tmp_io->io_hdr.nexus.targ_lun,
531			(tmp_io->io_hdr.io_type ==
532			CTL_IO_TASK) ?
533			tmp_io->taskio.tag_num :
534			tmp_io->scsiio.tag_num,
535		        tmp_io->io_hdr.flags,
536			tmp_io->io_hdr.status);
537	}
538#endif
539	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
540	ctl_enqueue_isc((union ctl_io *)ctsio);
541}
542
543/*
544 * ISC (Inter Shelf Communication) event handler.  Events from the HA
545 * subsystem come in here.
546 */
547static void
548ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
549{
550	struct ctl_softc *ctl_softc;
551	union ctl_io *io;
552	struct ctl_prio *presio;
553	ctl_ha_status isc_status;
554
555	ctl_softc = control_softc;
556	io = NULL;
557
558
559#if 0
560	printf("CTL: Isc Msg event %d\n", event);
561#endif
562	if (event == CTL_HA_EVT_MSG_RECV) {
563		union ctl_ha_msg msg_info;
564
565		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
566					     sizeof(msg_info), /*wait*/ 0);
567#if 0
568		printf("CTL: msg_type %d\n", msg_info.msg_type);
569#endif
570		if (isc_status != 0) {
571			printf("Error receiving message, status = %d\n",
572			       isc_status);
573			return;
574		}
575
576		switch (msg_info.hdr.msg_type) {
577		case CTL_MSG_SERIALIZE:
578#if 0
579			printf("Serialize\n");
580#endif
581			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
582			if (io == NULL) {
583				printf("ctl_isc_event_handler: can't allocate "
584				       "ctl_io!\n");
585				/* Bad Juju */
586				/* Need to set busy and send msg back */
587				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
588				msg_info.hdr.status = CTL_SCSI_ERROR;
589				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
590				msg_info.scsi.sense_len = 0;
591			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
592				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
593				}
594				goto bailout;
595			}
596			ctl_zero_io(io);
597			// populate ctsio from msg_info
598			io->io_hdr.io_type = CTL_IO_SCSI;
599			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
600			io->io_hdr.original_sc = msg_info.hdr.original_sc;
601#if 0
602			printf("pOrig %x\n", (int)msg_info.original_sc);
603#endif
604			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
605					    CTL_FLAG_IO_ACTIVE;
606			/*
607			 * If we're in serialization-only mode, we don't
608			 * want to go through full done processing.  Thus
609			 * the COPY flag.
610			 *
611			 * XXX KDM add another flag that is more specific.
612			 */
613			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
614				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
615			io->io_hdr.nexus = msg_info.hdr.nexus;
616#if 0
617			printf("targ %d, port %d, iid %d, lun %d\n",
618			       io->io_hdr.nexus.targ_target.id,
619			       io->io_hdr.nexus.targ_port,
620			       io->io_hdr.nexus.initid.id,
621			       io->io_hdr.nexus.targ_lun);
622#endif
623			io->scsiio.tag_num = msg_info.scsi.tag_num;
624			io->scsiio.tag_type = msg_info.scsi.tag_type;
625			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
626			       CTL_MAX_CDBLEN);
627			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
628				const struct ctl_cmd_entry *entry;
629
630				entry = ctl_get_cmd_entry(&io->scsiio);
631				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
632				io->io_hdr.flags |=
633					entry->flags & CTL_FLAG_DATA_MASK;
634			}
635			ctl_enqueue_isc(io);
636			break;
637
638		/* Performed on the Originating SC, XFER mode only */
639		case CTL_MSG_DATAMOVE: {
640			struct ctl_sg_entry *sgl;
641			int i, j;
642
643			io = msg_info.hdr.original_sc;
644			if (io == NULL) {
645				printf("%s: original_sc == NULL!\n", __func__);
646				/* XXX KDM do something here */
647				break;
648			}
649			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
650			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
651			/*
652			 * Keep track of this, we need to send it back over
653			 * when the datamove is complete.
654			 */
655			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
656
657			if (msg_info.dt.sg_sequence == 0) {
658				/*
659				 * XXX KDM we use the preallocated S/G list
660				 * here, but we'll need to change this to
661				 * dynamic allocation if we need larger S/G
662				 * lists.
663				 */
664				if (msg_info.dt.kern_sg_entries >
665				    sizeof(io->io_hdr.remote_sglist) /
666				    sizeof(io->io_hdr.remote_sglist[0])) {
667					printf("%s: number of S/G entries "
668					    "needed %u > allocated num %zd\n",
669					    __func__,
670					    msg_info.dt.kern_sg_entries,
671					    sizeof(io->io_hdr.remote_sglist)/
672					    sizeof(io->io_hdr.remote_sglist[0]));
673
674					/*
675					 * XXX KDM send a message back to
676					 * the other side to shut down the
677					 * DMA.  The error will come back
678					 * through via the normal channel.
679					 */
680					break;
681				}
682				sgl = io->io_hdr.remote_sglist;
683				memset(sgl, 0,
684				       sizeof(io->io_hdr.remote_sglist));
685
686				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
687
688				io->scsiio.kern_sg_entries =
689					msg_info.dt.kern_sg_entries;
690				io->scsiio.rem_sg_entries =
691					msg_info.dt.kern_sg_entries;
692				io->scsiio.kern_data_len =
693					msg_info.dt.kern_data_len;
694				io->scsiio.kern_total_len =
695					msg_info.dt.kern_total_len;
696				io->scsiio.kern_data_resid =
697					msg_info.dt.kern_data_resid;
698				io->scsiio.kern_rel_offset =
699					msg_info.dt.kern_rel_offset;
700				/*
701				 * Clear out per-DMA flags.
702				 */
703				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
704				/*
705				 * Add per-DMA flags that are set for this
706				 * particular DMA request.
707				 */
708				io->io_hdr.flags |= msg_info.dt.flags &
709						    CTL_FLAG_RDMA_MASK;
710			} else
711				sgl = (struct ctl_sg_entry *)
712					io->scsiio.kern_data_ptr;
713
714			for (i = msg_info.dt.sent_sg_entries, j = 0;
715			     i < (msg_info.dt.sent_sg_entries +
716			     msg_info.dt.cur_sg_entries); i++, j++) {
717				sgl[i].addr = msg_info.dt.sg_list[j].addr;
718				sgl[i].len = msg_info.dt.sg_list[j].len;
719
720#if 0
721				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
722				       __func__,
723				       msg_info.dt.sg_list[j].addr,
724				       msg_info.dt.sg_list[j].len,
725				       sgl[i].addr, sgl[i].len, j, i);
726#endif
727			}
728#if 0
729			memcpy(&sgl[msg_info.dt.sent_sg_entries],
730			       msg_info.dt.sg_list,
731			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
732#endif
733
734			/*
735			 * If this is the last piece of the I/O, we've got
736			 * the full S/G list.  Queue processing in the thread.
737			 * Otherwise wait for the next piece.
738			 */
739			if (msg_info.dt.sg_last != 0)
740				ctl_enqueue_isc(io);
741			break;
742		}
743		/* Performed on the Serializing (primary) SC, XFER mode only */
744		case CTL_MSG_DATAMOVE_DONE: {
745			if (msg_info.hdr.serializing_sc == NULL) {
746				printf("%s: serializing_sc == NULL!\n",
747				       __func__);
748				/* XXX KDM now what? */
749				break;
750			}
751			/*
752			 * We grab the sense information here in case
753			 * there was a failure, so we can return status
754			 * back to the initiator.
755			 */
756			io = msg_info.hdr.serializing_sc;
757			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
758			io->io_hdr.status = msg_info.hdr.status;
759			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
760			io->scsiio.sense_len = msg_info.scsi.sense_len;
761			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
762			io->io_hdr.port_status = msg_info.scsi.fetd_status;
763			io->scsiio.residual = msg_info.scsi.residual;
764			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
765			       sizeof(io->scsiio.sense_data));
766			ctl_enqueue_isc(io);
767			break;
768		}
769
770		/* Preformed on Originating SC, SER_ONLY mode */
771		case CTL_MSG_R2R:
772			io = msg_info.hdr.original_sc;
773			if (io == NULL) {
774				printf("%s: Major Bummer\n", __func__);
775				return;
776			} else {
777#if 0
778				printf("pOrig %x\n",(int) ctsio);
779#endif
780			}
781			io->io_hdr.msg_type = CTL_MSG_R2R;
782			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
783			ctl_enqueue_isc(io);
784			break;
785
786		/*
787		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
788		 * mode.
789		 * Performed on the Originating (i.e. secondary) SC in XFER
790		 * mode
791		 */
792		case CTL_MSG_FINISH_IO:
793			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
794				ctl_isc_handler_finish_xfer(ctl_softc,
795							    &msg_info);
796			else
797				ctl_isc_handler_finish_ser_only(ctl_softc,
798								&msg_info);
799			break;
800
801		/* Preformed on Originating SC */
802		case CTL_MSG_BAD_JUJU:
803			io = msg_info.hdr.original_sc;
804			if (io == NULL) {
805				printf("%s: Bad JUJU!, original_sc is NULL!\n",
806				       __func__);
807				break;
808			}
809			ctl_copy_sense_data(&msg_info, io);
810			/*
811			 * IO should have already been cleaned up on other
812			 * SC so clear this flag so we won't send a message
813			 * back to finish the IO there.
814			 */
815			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
816			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
817
818			/* io = msg_info.hdr.serializing_sc; */
819			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
820			ctl_enqueue_isc(io);
821			break;
822
823		/* Handle resets sent from the other side */
824		case CTL_MSG_MANAGE_TASKS: {
825			struct ctl_taskio *taskio;
826			taskio = (struct ctl_taskio *)ctl_alloc_io(
827				(void *)ctl_softc->othersc_pool);
828			if (taskio == NULL) {
829				printf("ctl_isc_event_handler: can't allocate "
830				       "ctl_io!\n");
831				/* Bad Juju */
832				/* should I just call the proper reset func
833				   here??? */
834				goto bailout;
835			}
836			ctl_zero_io((union ctl_io *)taskio);
837			taskio->io_hdr.io_type = CTL_IO_TASK;
838			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
839			taskio->io_hdr.nexus = msg_info.hdr.nexus;
840			taskio->task_action = msg_info.task.task_action;
841			taskio->tag_num = msg_info.task.tag_num;
842			taskio->tag_type = msg_info.task.tag_type;
843#ifdef CTL_TIME_IO
844			taskio->io_hdr.start_time = time_uptime;
845			getbintime(&taskio->io_hdr.start_bt);
846#if 0
847			cs_prof_gettime(&taskio->io_hdr.start_ticks);
848#endif
849#endif /* CTL_TIME_IO */
850			ctl_run_task((union ctl_io *)taskio);
851			break;
852		}
853		/* Persistent Reserve action which needs attention */
854		case CTL_MSG_PERS_ACTION:
855			presio = (struct ctl_prio *)ctl_alloc_io(
856				(void *)ctl_softc->othersc_pool);
857			if (presio == NULL) {
858				printf("ctl_isc_event_handler: can't allocate "
859				       "ctl_io!\n");
860				/* Bad Juju */
861				/* Need to set busy and send msg back */
862				goto bailout;
863			}
864			ctl_zero_io((union ctl_io *)presio);
865			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
866			presio->pr_msg = msg_info.pr;
867			ctl_enqueue_isc((union ctl_io *)presio);
868			break;
869		case CTL_MSG_SYNC_FE:
870			rcv_sync_msg = 1;
871			break;
872		case CTL_MSG_APS_LOCK: {
873			// It's quicker to execute this then to
874			// queue it.
875			struct ctl_lun *lun;
876			struct ctl_page_index *page_index;
877			struct copan_aps_subpage *current_sp;
878			uint32_t targ_lun;
879
880			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
881			lun = ctl_softc->ctl_luns[targ_lun];
882			mtx_lock(&lun->lun_lock);
883			page_index = &lun->mode_pages.index[index_to_aps_page];
884			current_sp = (struct copan_aps_subpage *)
885				     (page_index->page_data +
886				     (page_index->page_len * CTL_PAGE_CURRENT));
887
888			current_sp->lock_active = msg_info.aps.lock_flag;
889			mtx_unlock(&lun->lun_lock);
890		        break;
891		}
892		default:
893		        printf("How did I get here?\n");
894		}
895	} else if (event == CTL_HA_EVT_MSG_SENT) {
896		if (param != CTL_HA_STATUS_SUCCESS) {
897			printf("Bad status from ctl_ha_msg_send status %d\n",
898			       param);
899		}
900		return;
901	} else if (event == CTL_HA_EVT_DISCONNECT) {
902		printf("CTL: Got a disconnect from Isc\n");
903		return;
904	} else {
905		printf("ctl_isc_event_handler: Unknown event %d\n", event);
906		return;
907	}
908
909bailout:
910	return;
911}
912
913static void
914ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
915{
916	struct scsi_sense_data *sense;
917
918	sense = &dest->scsiio.sense_data;
919	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
920	dest->scsiio.scsi_status = src->scsi.scsi_status;
921	dest->scsiio.sense_len = src->scsi.sense_len;
922	dest->io_hdr.status = src->hdr.status;
923}
924
925static int
926ctl_init(void)
927{
928	struct ctl_softc *softc;
929	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
930	struct ctl_frontend *fe;
931        uint8_t sc_id =0;
932	int i, error, retval;
933	//int isc_retval;
934
935	retval = 0;
936	ctl_pause_rtr = 0;
937        rcv_sync_msg = 0;
938
939	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
940			       M_WAITOK | M_ZERO);
941	softc = control_softc;
942
943	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
944			      "cam/ctl");
945
946	softc->dev->si_drv1 = softc;
947
948	/*
949	 * By default, return a "bad LUN" peripheral qualifier for unknown
950	 * LUNs.  The user can override this default using the tunable or
951	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
952	 */
953	softc->inquiry_pq_no_lun = 1;
954	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
955			  &softc->inquiry_pq_no_lun);
956	sysctl_ctx_init(&softc->sysctl_ctx);
957	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
958		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
959		CTLFLAG_RD, 0, "CAM Target Layer");
960
961	if (softc->sysctl_tree == NULL) {
962		printf("%s: unable to allocate sysctl tree\n", __func__);
963		destroy_dev(softc->dev);
964		free(control_softc, M_DEVBUF);
965		control_softc = NULL;
966		return (ENOMEM);
967	}
968
969	SYSCTL_ADD_INT(&softc->sysctl_ctx,
970		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
971		       "inquiry_pq_no_lun", CTLFLAG_RW,
972		       &softc->inquiry_pq_no_lun, 0,
973		       "Report no lun possible for invalid LUNs");
974
975	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
976	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
977	softc->open_count = 0;
978
979	/*
980	 * Default to actually sending a SYNCHRONIZE CACHE command down to
981	 * the drive.
982	 */
983	softc->flags = CTL_FLAG_REAL_SYNC;
984
985	/*
986	 * In Copan's HA scheme, the "master" and "slave" roles are
987	 * figured out through the slot the controller is in.  Although it
988	 * is an active/active system, someone has to be in charge.
989 	 */
990#ifdef NEEDTOPORT
991        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
992#endif
993
994        if (sc_id == 0) {
995		softc->flags |= CTL_FLAG_MASTER_SHELF;
996		persis_offset = 0;
997	} else
998		persis_offset = CTL_MAX_INITIATORS;
999
1000	/*
1001	 * XXX KDM need to figure out where we want to get our target ID
1002	 * and WWID.  Is it different on each port?
1003	 */
1004	softc->target.id = 0;
1005	softc->target.wwid[0] = 0x12345678;
1006	softc->target.wwid[1] = 0x87654321;
1007	STAILQ_INIT(&softc->lun_list);
1008	STAILQ_INIT(&softc->pending_lun_queue);
1009	STAILQ_INIT(&softc->fe_list);
1010	STAILQ_INIT(&softc->be_list);
1011	STAILQ_INIT(&softc->io_pools);
1012
1013	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1014			    &internal_pool)!= 0){
1015		printf("ctl: can't allocate %d entry internal pool, "
1016		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1017		return (ENOMEM);
1018	}
1019
1020	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1021			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1022		printf("ctl: can't allocate %d entry emergency pool, "
1023		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1024		ctl_pool_free(internal_pool);
1025		return (ENOMEM);
1026	}
1027
1028	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1029	                    &other_pool) != 0)
1030	{
1031		printf("ctl: can't allocate %d entry other SC pool, "
1032		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1033		ctl_pool_free(internal_pool);
1034		ctl_pool_free(emergency_pool);
1035		return (ENOMEM);
1036	}
1037
1038	softc->internal_pool = internal_pool;
1039	softc->emergency_pool = emergency_pool;
1040	softc->othersc_pool = other_pool;
1041
1042	if (worker_threads <= 0)
1043		worker_threads = max(1, mp_ncpus / 4);
1044	if (worker_threads > CTL_MAX_THREADS)
1045		worker_threads = CTL_MAX_THREADS;
1046
1047	for (i = 0; i < worker_threads; i++) {
1048		struct ctl_thread *thr = &softc->threads[i];
1049
1050		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1051		thr->ctl_softc = softc;
1052		STAILQ_INIT(&thr->incoming_queue);
1053		STAILQ_INIT(&thr->rtr_queue);
1054		STAILQ_INIT(&thr->done_queue);
1055		STAILQ_INIT(&thr->isc_queue);
1056
1057		error = kproc_kthread_add(ctl_work_thread, thr,
1058		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1059		if (error != 0) {
1060			printf("error creating CTL work thread!\n");
1061			ctl_pool_free(internal_pool);
1062			ctl_pool_free(emergency_pool);
1063			ctl_pool_free(other_pool);
1064			return (error);
1065		}
1066	}
1067	error = kproc_kthread_add(ctl_lun_thread, softc,
1068	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1069	if (error != 0) {
1070		printf("error creating CTL lun thread!\n");
1071		ctl_pool_free(internal_pool);
1072		ctl_pool_free(emergency_pool);
1073		ctl_pool_free(other_pool);
1074		return (error);
1075	}
1076	if (bootverbose)
1077		printf("ctl: CAM Target Layer loaded\n");
1078
1079	/*
1080	 * Initialize the initiator and portname mappings
1081	 */
1082	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1083
1084	/*
1085	 * Initialize the ioctl front end.
1086	 */
1087	fe = &softc->ioctl_info.fe;
1088	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1089	fe->port_type = CTL_PORT_IOCTL;
1090	fe->num_requested_ctl_io = 100;
1091	fe->port_name = softc->ioctl_info.port_name;
1092	fe->port_online = ctl_ioctl_online;
1093	fe->port_offline = ctl_ioctl_offline;
1094	fe->onoff_arg = &softc->ioctl_info;
1095	fe->targ_enable = ctl_ioctl_targ_enable;
1096	fe->targ_disable = ctl_ioctl_targ_disable;
1097	fe->lun_enable = ctl_ioctl_lun_enable;
1098	fe->lun_disable = ctl_ioctl_lun_disable;
1099	fe->targ_lun_arg = &softc->ioctl_info;
1100	fe->fe_datamove = ctl_ioctl_datamove;
1101	fe->fe_done = ctl_ioctl_done;
1102	fe->max_targets = 15;
1103	fe->max_target_id = 15;
1104
1105	if (ctl_frontend_register(&softc->ioctl_info.fe,
1106	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1107		printf("ctl: ioctl front end registration failed, will "
1108		       "continue anyway\n");
1109	}
1110
1111#ifdef CTL_IO_DELAY
1112	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1113		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1114		       sizeof(struct callout), CTL_TIMER_BYTES);
1115		return (EINVAL);
1116	}
1117#endif /* CTL_IO_DELAY */
1118
1119	return (0);
1120}
1121
1122void
1123ctl_shutdown(void)
1124{
1125	struct ctl_softc *softc;
1126	struct ctl_lun *lun, *next_lun;
1127	struct ctl_io_pool *pool;
1128
1129	softc = (struct ctl_softc *)control_softc;
1130
1131	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1132		printf("ctl: ioctl front end deregistration failed\n");
1133
1134	mtx_lock(&softc->ctl_lock);
1135
1136	/*
1137	 * Free up each LUN.
1138	 */
1139	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1140		next_lun = STAILQ_NEXT(lun, links);
1141		ctl_free_lun(lun);
1142	}
1143
1144	mtx_unlock(&softc->ctl_lock);
1145
1146	/*
1147	 * This will rip the rug out from under any FETDs or anyone else
1148	 * that has a pool allocated.  Since we increment our module
1149	 * refcount any time someone outside the main CTL module allocates
1150	 * a pool, we shouldn't have any problems here.  The user won't be
1151	 * able to unload the CTL module until client modules have
1152	 * successfully unloaded.
1153	 */
1154	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1155		ctl_pool_free(pool);
1156
1157#if 0
1158	ctl_shutdown_thread(softc->work_thread);
1159	mtx_destroy(&softc->queue_lock);
1160#endif
1161
1162	mtx_destroy(&softc->pool_lock);
1163	mtx_destroy(&softc->ctl_lock);
1164
1165	destroy_dev(softc->dev);
1166
1167	sysctl_ctx_free(&softc->sysctl_ctx);
1168
1169	free(control_softc, M_DEVBUF);
1170	control_softc = NULL;
1171
1172	if (bootverbose)
1173		printf("ctl: CAM Target Layer unloaded\n");
1174}
1175
1176static int
1177ctl_module_event_handler(module_t mod, int what, void *arg)
1178{
1179
1180	switch (what) {
1181	case MOD_LOAD:
1182		return (ctl_init());
1183	case MOD_UNLOAD:
1184		return (EBUSY);
1185	default:
1186		return (EOPNOTSUPP);
1187	}
1188}
1189
1190/*
1191 * XXX KDM should we do some access checks here?  Bump a reference count to
1192 * prevent a CTL module from being unloaded while someone has it open?
1193 */
1194static int
1195ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1196{
1197	return (0);
1198}
1199
1200static int
1201ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1202{
1203	return (0);
1204}
1205
1206int
1207ctl_port_enable(ctl_port_type port_type)
1208{
1209	struct ctl_softc *softc;
1210	struct ctl_frontend *fe;
1211
1212	if (ctl_is_single == 0) {
1213		union ctl_ha_msg msg_info;
1214		int isc_retval;
1215
1216#if 0
1217		printf("%s: HA mode, synchronizing frontend enable\n",
1218		        __func__);
1219#endif
1220		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1221	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1222		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1223			printf("Sync msg send error retval %d\n", isc_retval);
1224		}
1225		if (!rcv_sync_msg) {
1226			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1227			        sizeof(msg_info), 1);
1228		}
1229#if 0
1230        	printf("CTL:Frontend Enable\n");
1231	} else {
1232		printf("%s: single mode, skipping frontend synchronization\n",
1233		        __func__);
1234#endif
1235	}
1236
1237	softc = control_softc;
1238
1239	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1240		if (port_type & fe->port_type)
1241		{
1242#if 0
1243			printf("port %d\n", fe->targ_port);
1244#endif
1245			ctl_frontend_online(fe);
1246		}
1247	}
1248
1249	return (0);
1250}
1251
1252int
1253ctl_port_disable(ctl_port_type port_type)
1254{
1255	struct ctl_softc *softc;
1256	struct ctl_frontend *fe;
1257
1258	softc = control_softc;
1259
1260	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1261		if (port_type & fe->port_type)
1262			ctl_frontend_offline(fe);
1263	}
1264
1265	return (0);
1266}
1267
1268/*
1269 * Returns 0 for success, 1 for failure.
1270 * Currently the only failure mode is if there aren't enough entries
1271 * allocated.  So, in case of a failure, look at num_entries_dropped,
1272 * reallocate and try again.
1273 */
1274int
1275ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1276	      int *num_entries_filled, int *num_entries_dropped,
1277	      ctl_port_type port_type, int no_virtual)
1278{
1279	struct ctl_softc *softc;
1280	struct ctl_frontend *fe;
1281	int entries_dropped, entries_filled;
1282	int retval;
1283	int i;
1284
1285	softc = control_softc;
1286
1287	retval = 0;
1288	entries_filled = 0;
1289	entries_dropped = 0;
1290
1291	i = 0;
1292	mtx_lock(&softc->ctl_lock);
1293	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1294		struct ctl_port_entry *entry;
1295
1296		if ((fe->port_type & port_type) == 0)
1297			continue;
1298
1299		if ((no_virtual != 0)
1300		 && (fe->virtual_port != 0))
1301			continue;
1302
1303		if (entries_filled >= num_entries_alloced) {
1304			entries_dropped++;
1305			continue;
1306		}
1307		entry = &entries[i];
1308
1309		entry->port_type = fe->port_type;
1310		strlcpy(entry->port_name, fe->port_name,
1311			sizeof(entry->port_name));
1312		entry->physical_port = fe->physical_port;
1313		entry->virtual_port = fe->virtual_port;
1314		entry->wwnn = fe->wwnn;
1315		entry->wwpn = fe->wwpn;
1316
1317		i++;
1318		entries_filled++;
1319	}
1320
1321	mtx_unlock(&softc->ctl_lock);
1322
1323	if (entries_dropped > 0)
1324		retval = 1;
1325
1326	*num_entries_dropped = entries_dropped;
1327	*num_entries_filled = entries_filled;
1328
1329	return (retval);
1330}
1331
1332static void
1333ctl_ioctl_online(void *arg)
1334{
1335	struct ctl_ioctl_info *ioctl_info;
1336
1337	ioctl_info = (struct ctl_ioctl_info *)arg;
1338
1339	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1340}
1341
1342static void
1343ctl_ioctl_offline(void *arg)
1344{
1345	struct ctl_ioctl_info *ioctl_info;
1346
1347	ioctl_info = (struct ctl_ioctl_info *)arg;
1348
1349	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1350}
1351
1352/*
1353 * Remove an initiator by port number and initiator ID.
1354 * Returns 0 for success, 1 for failure.
1355 */
1356int
1357ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1358{
1359	struct ctl_softc *softc;
1360
1361	softc = control_softc;
1362
1363	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1364
1365	if ((targ_port < 0)
1366	 || (targ_port > CTL_MAX_PORTS)) {
1367		printf("%s: invalid port number %d\n", __func__, targ_port);
1368		return (1);
1369	}
1370	if (iid > CTL_MAX_INIT_PER_PORT) {
1371		printf("%s: initiator ID %u > maximun %u!\n",
1372		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1373		return (1);
1374	}
1375
1376	mtx_lock(&softc->ctl_lock);
1377
1378	softc->wwpn_iid[targ_port][iid].in_use = 0;
1379
1380	mtx_unlock(&softc->ctl_lock);
1381
1382	return (0);
1383}
1384
1385/*
1386 * Add an initiator to the initiator map.
1387 * Returns 0 for success, 1 for failure.
1388 */
1389int
1390ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1391{
1392	struct ctl_softc *softc;
1393	int retval;
1394
1395	softc = control_softc;
1396
1397	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1398
1399	retval = 0;
1400
1401	if ((targ_port < 0)
1402	 || (targ_port > CTL_MAX_PORTS)) {
1403		printf("%s: invalid port number %d\n", __func__, targ_port);
1404		return (1);
1405	}
1406	if (iid > CTL_MAX_INIT_PER_PORT) {
1407		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1408		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1409		return (1);
1410	}
1411
1412	mtx_lock(&softc->ctl_lock);
1413
1414	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1415		/*
1416		 * We don't treat this as an error.
1417		 */
1418		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1419			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1420			       __func__, targ_port, iid, (uintmax_t)wwpn);
1421			goto bailout;
1422		}
1423
1424		/*
1425		 * This is an error, but what do we do about it?  The
1426		 * driver is telling us we have a new WWPN for this
1427		 * initiator ID, so we pretty much need to use it.
1428		 */
1429		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1430		       "still at that address\n", __func__, targ_port, iid,
1431		       (uintmax_t)wwpn,
1432		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1433
1434		/*
1435		 * XXX KDM clear have_ca and ua_pending on each LUN for
1436		 * this initiator.
1437		 */
1438	}
1439	softc->wwpn_iid[targ_port][iid].in_use = 1;
1440	softc->wwpn_iid[targ_port][iid].iid = iid;
1441	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1442	softc->wwpn_iid[targ_port][iid].port = targ_port;
1443
1444bailout:
1445
1446	mtx_unlock(&softc->ctl_lock);
1447
1448	return (retval);
1449}
1450
1451/*
1452 * XXX KDM should we pretend to do something in the target/lun
1453 * enable/disable functions?
1454 */
1455static int
1456ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1457{
1458	return (0);
1459}
1460
1461static int
1462ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1463{
1464	return (0);
1465}
1466
1467static int
1468ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1469{
1470	return (0);
1471}
1472
1473static int
1474ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1475{
1476	return (0);
1477}
1478
1479/*
1480 * Data movement routine for the CTL ioctl frontend port.
1481 */
1482static int
1483ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1484{
1485	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1486	struct ctl_sg_entry ext_entry, kern_entry;
1487	int ext_sglen, ext_sg_entries, kern_sg_entries;
1488	int ext_sg_start, ext_offset;
1489	int len_to_copy, len_copied;
1490	int kern_watermark, ext_watermark;
1491	int ext_sglist_malloced;
1492	int i, j;
1493
1494	ext_sglist_malloced = 0;
1495	ext_sg_start = 0;
1496	ext_offset = 0;
1497
1498	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1499
1500	/*
1501	 * If this flag is set, fake the data transfer.
1502	 */
1503	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1504		ctsio->ext_data_filled = ctsio->ext_data_len;
1505		goto bailout;
1506	}
1507
1508	/*
1509	 * To simplify things here, if we have a single buffer, stick it in
1510	 * a S/G entry and just make it a single entry S/G list.
1511	 */
1512	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1513		int len_seen;
1514
1515		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1516
1517		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1518							   M_WAITOK);
1519		ext_sglist_malloced = 1;
1520		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1521				   ext_sglen) != 0) {
1522			ctl_set_internal_failure(ctsio,
1523						 /*sks_valid*/ 0,
1524						 /*retry_count*/ 0);
1525			goto bailout;
1526		}
1527		ext_sg_entries = ctsio->ext_sg_entries;
1528		len_seen = 0;
1529		for (i = 0; i < ext_sg_entries; i++) {
1530			if ((len_seen + ext_sglist[i].len) >=
1531			     ctsio->ext_data_filled) {
1532				ext_sg_start = i;
1533				ext_offset = ctsio->ext_data_filled - len_seen;
1534				break;
1535			}
1536			len_seen += ext_sglist[i].len;
1537		}
1538	} else {
1539		ext_sglist = &ext_entry;
1540		ext_sglist->addr = ctsio->ext_data_ptr;
1541		ext_sglist->len = ctsio->ext_data_len;
1542		ext_sg_entries = 1;
1543		ext_sg_start = 0;
1544		ext_offset = ctsio->ext_data_filled;
1545	}
1546
1547	if (ctsio->kern_sg_entries > 0) {
1548		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1549		kern_sg_entries = ctsio->kern_sg_entries;
1550	} else {
1551		kern_sglist = &kern_entry;
1552		kern_sglist->addr = ctsio->kern_data_ptr;
1553		kern_sglist->len = ctsio->kern_data_len;
1554		kern_sg_entries = 1;
1555	}
1556
1557
1558	kern_watermark = 0;
1559	ext_watermark = ext_offset;
1560	len_copied = 0;
1561	for (i = ext_sg_start, j = 0;
1562	     i < ext_sg_entries && j < kern_sg_entries;) {
1563		uint8_t *ext_ptr, *kern_ptr;
1564
1565		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1566				      kern_sglist[j].len - kern_watermark);
1567
1568		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1569		ext_ptr = ext_ptr + ext_watermark;
1570		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1571			/*
1572			 * XXX KDM fix this!
1573			 */
1574			panic("need to implement bus address support");
1575#if 0
1576			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1577#endif
1578		} else
1579			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1580		kern_ptr = kern_ptr + kern_watermark;
1581
1582		kern_watermark += len_to_copy;
1583		ext_watermark += len_to_copy;
1584
1585		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1586		     CTL_FLAG_DATA_IN) {
1587			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1588					 "bytes to user\n", len_to_copy));
1589			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1590					 "to %p\n", kern_ptr, ext_ptr));
1591			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1592				ctl_set_internal_failure(ctsio,
1593							 /*sks_valid*/ 0,
1594							 /*retry_count*/ 0);
1595				goto bailout;
1596			}
1597		} else {
1598			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1599					 "bytes from user\n", len_to_copy));
1600			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1601					 "to %p\n", ext_ptr, kern_ptr));
1602			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1603				ctl_set_internal_failure(ctsio,
1604							 /*sks_valid*/ 0,
1605							 /*retry_count*/0);
1606				goto bailout;
1607			}
1608		}
1609
1610		len_copied += len_to_copy;
1611
1612		if (ext_sglist[i].len == ext_watermark) {
1613			i++;
1614			ext_watermark = 0;
1615		}
1616
1617		if (kern_sglist[j].len == kern_watermark) {
1618			j++;
1619			kern_watermark = 0;
1620		}
1621	}
1622
1623	ctsio->ext_data_filled += len_copied;
1624
1625	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1626			 "kern_sg_entries: %d\n", ext_sg_entries,
1627			 kern_sg_entries));
1628	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1629			 "kern_data_len = %d\n", ctsio->ext_data_len,
1630			 ctsio->kern_data_len));
1631
1632
1633	/* XXX KDM set residual?? */
1634bailout:
1635
1636	if (ext_sglist_malloced != 0)
1637		free(ext_sglist, M_CTL);
1638
1639	return (CTL_RETVAL_COMPLETE);
1640}
1641
1642/*
1643 * Serialize a command that went down the "wrong" side, and so was sent to
1644 * this controller for execution.  The logic is a little different than the
1645 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1646 * sent back to the other side, but in the success case, we execute the
1647 * command on this side (XFER mode) or tell the other side to execute it
1648 * (SER_ONLY mode).
1649 */
1650static int
1651ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1652{
1653	struct ctl_softc *ctl_softc;
1654	union ctl_ha_msg msg_info;
1655	struct ctl_lun *lun;
1656	int retval = 0;
1657	uint32_t targ_lun;
1658
1659	ctl_softc = control_softc;
1660
1661	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1662	lun = ctl_softc->ctl_luns[targ_lun];
1663	if (lun==NULL)
1664	{
1665		/*
1666		 * Why isn't LUN defined? The other side wouldn't
1667		 * send a cmd if the LUN is undefined.
1668		 */
1669		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1670
1671		/* "Logical unit not supported" */
1672		ctl_set_sense_data(&msg_info.scsi.sense_data,
1673				   lun,
1674				   /*sense_format*/SSD_TYPE_NONE,
1675				   /*current_error*/ 1,
1676				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1677				   /*asc*/ 0x25,
1678				   /*ascq*/ 0x00,
1679				   SSD_ELEM_NONE);
1680
1681		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1682		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1683		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1684		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1685		msg_info.hdr.serializing_sc = NULL;
1686		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1687	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1688				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1689		}
1690		return(1);
1691
1692	}
1693
1694	mtx_lock(&lun->lun_lock);
1695    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1696
1697	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1698		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1699		 ooa_links))) {
1700	case CTL_ACTION_BLOCK:
1701		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1702		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1703				  blocked_links);
1704		break;
1705	case CTL_ACTION_PASS:
1706	case CTL_ACTION_SKIP:
1707		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1708			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1709			ctl_enqueue_rtr((union ctl_io *)ctsio);
1710		} else {
1711
1712			/* send msg back to other side */
1713			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1714			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1715			msg_info.hdr.msg_type = CTL_MSG_R2R;
1716#if 0
1717			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1718#endif
1719		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1720			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1721			}
1722		}
1723		break;
1724	case CTL_ACTION_OVERLAP:
1725		/* OVERLAPPED COMMANDS ATTEMPTED */
1726		ctl_set_sense_data(&msg_info.scsi.sense_data,
1727				   lun,
1728				   /*sense_format*/SSD_TYPE_NONE,
1729				   /*current_error*/ 1,
1730				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1731				   /*asc*/ 0x4E,
1732				   /*ascq*/ 0x00,
1733				   SSD_ELEM_NONE);
1734
1735		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1736		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1737		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1738		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1739		msg_info.hdr.serializing_sc = NULL;
1740		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1741#if 0
1742		printf("BAD JUJU:Major Bummer Overlap\n");
1743#endif
1744		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1745		retval = 1;
1746		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1747		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1748		}
1749		break;
1750	case CTL_ACTION_OVERLAP_TAG:
1751		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1752		ctl_set_sense_data(&msg_info.scsi.sense_data,
1753				   lun,
1754				   /*sense_format*/SSD_TYPE_NONE,
1755				   /*current_error*/ 1,
1756				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1757				   /*asc*/ 0x4D,
1758				   /*ascq*/ ctsio->tag_num & 0xff,
1759				   SSD_ELEM_NONE);
1760
1761		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1762		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1763		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1764		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1765		msg_info.hdr.serializing_sc = NULL;
1766		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1767#if 0
1768		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1769#endif
1770		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1771		retval = 1;
1772		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1773		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1774		}
1775		break;
1776	case CTL_ACTION_ERROR:
1777	default:
1778		/* "Internal target failure" */
1779		ctl_set_sense_data(&msg_info.scsi.sense_data,
1780				   lun,
1781				   /*sense_format*/SSD_TYPE_NONE,
1782				   /*current_error*/ 1,
1783				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1784				   /*asc*/ 0x44,
1785				   /*ascq*/ 0x00,
1786				   SSD_ELEM_NONE);
1787
1788		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1789		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1790		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1791		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1792		msg_info.hdr.serializing_sc = NULL;
1793		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1794#if 0
1795		printf("BAD JUJU:Major Bummer HW Error\n");
1796#endif
1797		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1798		retval = 1;
1799		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1800		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1801		}
1802		break;
1803	}
1804	mtx_unlock(&lun->lun_lock);
1805	return (retval);
1806}
1807
1808static int
1809ctl_ioctl_submit_wait(union ctl_io *io)
1810{
1811	struct ctl_fe_ioctl_params params;
1812	ctl_fe_ioctl_state last_state;
1813	int done, retval;
1814
1815	retval = 0;
1816
1817	bzero(&params, sizeof(params));
1818
1819	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1820	cv_init(&params.sem, "ctlioccv");
1821	params.state = CTL_IOCTL_INPROG;
1822	last_state = params.state;
1823
1824	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1825
1826	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1827
1828	/* This shouldn't happen */
1829	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1830		return (retval);
1831
1832	done = 0;
1833
1834	do {
1835		mtx_lock(&params.ioctl_mtx);
1836		/*
1837		 * Check the state here, and don't sleep if the state has
1838		 * already changed (i.e. wakeup has already occured, but we
1839		 * weren't waiting yet).
1840		 */
1841		if (params.state == last_state) {
1842			/* XXX KDM cv_wait_sig instead? */
1843			cv_wait(&params.sem, &params.ioctl_mtx);
1844		}
1845		last_state = params.state;
1846
1847		switch (params.state) {
1848		case CTL_IOCTL_INPROG:
1849			/* Why did we wake up? */
1850			/* XXX KDM error here? */
1851			mtx_unlock(&params.ioctl_mtx);
1852			break;
1853		case CTL_IOCTL_DATAMOVE:
1854			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1855
1856			/*
1857			 * change last_state back to INPROG to avoid
1858			 * deadlock on subsequent data moves.
1859			 */
1860			params.state = last_state = CTL_IOCTL_INPROG;
1861
1862			mtx_unlock(&params.ioctl_mtx);
1863			ctl_ioctl_do_datamove(&io->scsiio);
1864			/*
1865			 * Note that in some cases, most notably writes,
1866			 * this will queue the I/O and call us back later.
1867			 * In other cases, generally reads, this routine
1868			 * will immediately call back and wake us up,
1869			 * probably using our own context.
1870			 */
1871			io->scsiio.be_move_done(io);
1872			break;
1873		case CTL_IOCTL_DONE:
1874			mtx_unlock(&params.ioctl_mtx);
1875			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1876			done = 1;
1877			break;
1878		default:
1879			mtx_unlock(&params.ioctl_mtx);
1880			/* XXX KDM error here? */
1881			break;
1882		}
1883	} while (done == 0);
1884
1885	mtx_destroy(&params.ioctl_mtx);
1886	cv_destroy(&params.sem);
1887
1888	return (CTL_RETVAL_COMPLETE);
1889}
1890
1891static void
1892ctl_ioctl_datamove(union ctl_io *io)
1893{
1894	struct ctl_fe_ioctl_params *params;
1895
1896	params = (struct ctl_fe_ioctl_params *)
1897		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1898
1899	mtx_lock(&params->ioctl_mtx);
1900	params->state = CTL_IOCTL_DATAMOVE;
1901	cv_broadcast(&params->sem);
1902	mtx_unlock(&params->ioctl_mtx);
1903}
1904
1905static void
1906ctl_ioctl_done(union ctl_io *io)
1907{
1908	struct ctl_fe_ioctl_params *params;
1909
1910	params = (struct ctl_fe_ioctl_params *)
1911		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1912
1913	mtx_lock(&params->ioctl_mtx);
1914	params->state = CTL_IOCTL_DONE;
1915	cv_broadcast(&params->sem);
1916	mtx_unlock(&params->ioctl_mtx);
1917}
1918
1919static void
1920ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1921{
1922	struct ctl_fe_ioctl_startstop_info *sd_info;
1923
1924	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1925
1926	sd_info->hs_info.status = metatask->status;
1927	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1928	sd_info->hs_info.luns_complete =
1929		metatask->taskinfo.startstop.luns_complete;
1930	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1931
1932	cv_broadcast(&sd_info->sem);
1933}
1934
1935static void
1936ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1937{
1938	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1939
1940	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1941
1942	mtx_lock(fe_bbr_info->lock);
1943	fe_bbr_info->bbr_info->status = metatask->status;
1944	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1945	fe_bbr_info->wakeup_done = 1;
1946	mtx_unlock(fe_bbr_info->lock);
1947
1948	cv_broadcast(&fe_bbr_info->sem);
1949}
1950
1951/*
1952 * Returns 0 for success, errno for failure.
1953 */
1954static int
1955ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1956		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1957{
1958	union ctl_io *io;
1959	int retval;
1960
1961	retval = 0;
1962
1963	mtx_lock(&lun->lun_lock);
1964	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1965	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1966	     ooa_links)) {
1967		struct ctl_ooa_entry *entry;
1968
1969		/*
1970		 * If we've got more than we can fit, just count the
1971		 * remaining entries.
1972		 */
1973		if (*cur_fill_num >= ooa_hdr->alloc_num)
1974			continue;
1975
1976		entry = &kern_entries[*cur_fill_num];
1977
1978		entry->tag_num = io->scsiio.tag_num;
1979		entry->lun_num = lun->lun;
1980#ifdef CTL_TIME_IO
1981		entry->start_bt = io->io_hdr.start_bt;
1982#endif
1983		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1984		entry->cdb_len = io->scsiio.cdb_len;
1985		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1986			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1987
1988		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1989			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1990
1991		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1992			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1993
1994		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1995			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1996
1997		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1998			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1999	}
2000	mtx_unlock(&lun->lun_lock);
2001
2002	return (retval);
2003}
2004
2005static void *
2006ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2007		 size_t error_str_len)
2008{
2009	void *kptr;
2010
2011	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2012
2013	if (copyin(user_addr, kptr, len) != 0) {
2014		snprintf(error_str, error_str_len, "Error copying %d bytes "
2015			 "from user address %p to kernel address %p", len,
2016			 user_addr, kptr);
2017		free(kptr, M_CTL);
2018		return (NULL);
2019	}
2020
2021	return (kptr);
2022}
2023
2024static void
2025ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2026{
2027	int i;
2028
2029	if (be_args == NULL)
2030		return;
2031
2032	for (i = 0; i < num_be_args; i++) {
2033		free(be_args[i].kname, M_CTL);
2034		free(be_args[i].kvalue, M_CTL);
2035	}
2036
2037	free(be_args, M_CTL);
2038}
2039
2040static struct ctl_be_arg *
2041ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2042		char *error_str, size_t error_str_len)
2043{
2044	struct ctl_be_arg *args;
2045	int i;
2046
2047	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2048				error_str, error_str_len);
2049
2050	if (args == NULL)
2051		goto bailout;
2052
2053	for (i = 0; i < num_be_args; i++) {
2054		args[i].kname = NULL;
2055		args[i].kvalue = NULL;
2056	}
2057
2058	for (i = 0; i < num_be_args; i++) {
2059		uint8_t *tmpptr;
2060
2061		args[i].kname = ctl_copyin_alloc(args[i].name,
2062			args[i].namelen, error_str, error_str_len);
2063		if (args[i].kname == NULL)
2064			goto bailout;
2065
2066		if (args[i].kname[args[i].namelen - 1] != '\0') {
2067			snprintf(error_str, error_str_len, "Argument %d "
2068				 "name is not NUL-terminated", i);
2069			goto bailout;
2070		}
2071
2072		args[i].kvalue = NULL;
2073
2074		tmpptr = ctl_copyin_alloc(args[i].value,
2075			args[i].vallen, error_str, error_str_len);
2076		if (tmpptr == NULL)
2077			goto bailout;
2078
2079		args[i].kvalue = tmpptr;
2080
2081		if ((args[i].flags & CTL_BEARG_ASCII)
2082		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2083			snprintf(error_str, error_str_len, "Argument %d "
2084				 "value is not NUL-terminated", i);
2085			goto bailout;
2086		}
2087	}
2088
2089	return (args);
2090bailout:
2091
2092	ctl_free_args(num_be_args, args);
2093
2094	return (NULL);
2095}
2096
2097/*
2098 * Escape characters that are illegal or not recommended in XML.
2099 */
2100int
2101ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2102{
2103	int retval;
2104
2105	retval = 0;
2106
2107	for (; *str; str++) {
2108		switch (*str) {
2109		case '&':
2110			retval = sbuf_printf(sb, "&amp;");
2111			break;
2112		case '>':
2113			retval = sbuf_printf(sb, "&gt;");
2114			break;
2115		case '<':
2116			retval = sbuf_printf(sb, "&lt;");
2117			break;
2118		default:
2119			retval = sbuf_putc(sb, *str);
2120			break;
2121		}
2122
2123		if (retval != 0)
2124			break;
2125
2126	}
2127
2128	return (retval);
2129}
2130
2131static int
2132ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2133	  struct thread *td)
2134{
2135	struct ctl_softc *softc;
2136	int retval;
2137
2138	softc = control_softc;
2139
2140	retval = 0;
2141
2142	switch (cmd) {
2143	case CTL_IO: {
2144		union ctl_io *io;
2145		void *pool_tmp;
2146
2147		/*
2148		 * If we haven't been "enabled", don't allow any SCSI I/O
2149		 * to this FETD.
2150		 */
2151		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2152			retval = EPERM;
2153			break;
2154		}
2155
2156		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2157		if (io == NULL) {
2158			printf("ctl_ioctl: can't allocate ctl_io!\n");
2159			retval = ENOSPC;
2160			break;
2161		}
2162
2163		/*
2164		 * Need to save the pool reference so it doesn't get
2165		 * spammed by the user's ctl_io.
2166		 */
2167		pool_tmp = io->io_hdr.pool;
2168
2169		memcpy(io, (void *)addr, sizeof(*io));
2170
2171		io->io_hdr.pool = pool_tmp;
2172		/*
2173		 * No status yet, so make sure the status is set properly.
2174		 */
2175		io->io_hdr.status = CTL_STATUS_NONE;
2176
2177		/*
2178		 * The user sets the initiator ID, target and LUN IDs.
2179		 */
2180		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2181		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2182		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2183		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2184			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2185
2186		retval = ctl_ioctl_submit_wait(io);
2187
2188		if (retval != 0) {
2189			ctl_free_io(io);
2190			break;
2191		}
2192
2193		memcpy((void *)addr, io, sizeof(*io));
2194
2195		/* return this to our pool */
2196		ctl_free_io(io);
2197
2198		break;
2199	}
2200	case CTL_ENABLE_PORT:
2201	case CTL_DISABLE_PORT:
2202	case CTL_SET_PORT_WWNS: {
2203		struct ctl_frontend *fe;
2204		struct ctl_port_entry *entry;
2205
2206		entry = (struct ctl_port_entry *)addr;
2207
2208		mtx_lock(&softc->ctl_lock);
2209		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2210			int action, done;
2211
2212			action = 0;
2213			done = 0;
2214
2215			if ((entry->port_type == CTL_PORT_NONE)
2216			 && (entry->targ_port == fe->targ_port)) {
2217				/*
2218				 * If the user only wants to enable or
2219				 * disable or set WWNs on a specific port,
2220				 * do the operation and we're done.
2221				 */
2222				action = 1;
2223				done = 1;
2224			} else if (entry->port_type & fe->port_type) {
2225				/*
2226				 * Compare the user's type mask with the
2227				 * particular frontend type to see if we
2228				 * have a match.
2229				 */
2230				action = 1;
2231				done = 0;
2232
2233				/*
2234				 * Make sure the user isn't trying to set
2235				 * WWNs on multiple ports at the same time.
2236				 */
2237				if (cmd == CTL_SET_PORT_WWNS) {
2238					printf("%s: Can't set WWNs on "
2239					       "multiple ports\n", __func__);
2240					retval = EINVAL;
2241					break;
2242				}
2243			}
2244			if (action != 0) {
2245				/*
2246				 * XXX KDM we have to drop the lock here,
2247				 * because the online/offline operations
2248				 * can potentially block.  We need to
2249				 * reference count the frontends so they
2250				 * can't go away,
2251				 */
2252				mtx_unlock(&softc->ctl_lock);
2253
2254				if (cmd == CTL_ENABLE_PORT) {
2255					struct ctl_lun *lun;
2256
2257					STAILQ_FOREACH(lun, &softc->lun_list,
2258						       links) {
2259						fe->lun_enable(fe->targ_lun_arg,
2260						    lun->target,
2261						    lun->lun);
2262					}
2263
2264					ctl_frontend_online(fe);
2265				} else if (cmd == CTL_DISABLE_PORT) {
2266					struct ctl_lun *lun;
2267
2268					ctl_frontend_offline(fe);
2269
2270					STAILQ_FOREACH(lun, &softc->lun_list,
2271						       links) {
2272						fe->lun_disable(
2273						    fe->targ_lun_arg,
2274						    lun->target,
2275						    lun->lun);
2276					}
2277				}
2278
2279				mtx_lock(&softc->ctl_lock);
2280
2281				if (cmd == CTL_SET_PORT_WWNS)
2282					ctl_frontend_set_wwns(fe,
2283					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2284					    1 : 0, entry->wwnn,
2285					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2286					    1 : 0, entry->wwpn);
2287			}
2288			if (done != 0)
2289				break;
2290		}
2291		mtx_unlock(&softc->ctl_lock);
2292		break;
2293	}
2294	case CTL_GET_PORT_LIST: {
2295		struct ctl_frontend *fe;
2296		struct ctl_port_list *list;
2297		int i;
2298
2299		list = (struct ctl_port_list *)addr;
2300
2301		if (list->alloc_len != (list->alloc_num *
2302		    sizeof(struct ctl_port_entry))) {
2303			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2304			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2305			       "%zu\n", __func__, list->alloc_len,
2306			       list->alloc_num, sizeof(struct ctl_port_entry));
2307			retval = EINVAL;
2308			break;
2309		}
2310		list->fill_len = 0;
2311		list->fill_num = 0;
2312		list->dropped_num = 0;
2313		i = 0;
2314		mtx_lock(&softc->ctl_lock);
2315		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2316			struct ctl_port_entry entry, *list_entry;
2317
2318			if (list->fill_num >= list->alloc_num) {
2319				list->dropped_num++;
2320				continue;
2321			}
2322
2323			entry.port_type = fe->port_type;
2324			strlcpy(entry.port_name, fe->port_name,
2325				sizeof(entry.port_name));
2326			entry.targ_port = fe->targ_port;
2327			entry.physical_port = fe->physical_port;
2328			entry.virtual_port = fe->virtual_port;
2329			entry.wwnn = fe->wwnn;
2330			entry.wwpn = fe->wwpn;
2331			if (fe->status & CTL_PORT_STATUS_ONLINE)
2332				entry.online = 1;
2333			else
2334				entry.online = 0;
2335
2336			list_entry = &list->entries[i];
2337
2338			retval = copyout(&entry, list_entry, sizeof(entry));
2339			if (retval != 0) {
2340				printf("%s: CTL_GET_PORT_LIST: copyout "
2341				       "returned %d\n", __func__, retval);
2342				break;
2343			}
2344			i++;
2345			list->fill_num++;
2346			list->fill_len += sizeof(entry);
2347		}
2348		mtx_unlock(&softc->ctl_lock);
2349
2350		/*
2351		 * If this is non-zero, we had a copyout fault, so there's
2352		 * probably no point in attempting to set the status inside
2353		 * the structure.
2354		 */
2355		if (retval != 0)
2356			break;
2357
2358		if (list->dropped_num > 0)
2359			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2360		else
2361			list->status = CTL_PORT_LIST_OK;
2362		break;
2363	}
2364	case CTL_DUMP_OOA: {
2365		struct ctl_lun *lun;
2366		union ctl_io *io;
2367		char printbuf[128];
2368		struct sbuf sb;
2369
2370		mtx_lock(&softc->ctl_lock);
2371		printf("Dumping OOA queues:\n");
2372		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2373			mtx_lock(&lun->lun_lock);
2374			for (io = (union ctl_io *)TAILQ_FIRST(
2375			     &lun->ooa_queue); io != NULL;
2376			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2377			     ooa_links)) {
2378				sbuf_new(&sb, printbuf, sizeof(printbuf),
2379					 SBUF_FIXEDLEN);
2380				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2381					    (intmax_t)lun->lun,
2382					    io->scsiio.tag_num,
2383					    (io->io_hdr.flags &
2384					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2385					    (io->io_hdr.flags &
2386					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2387					    (io->io_hdr.flags &
2388					    CTL_FLAG_ABORT) ? " ABORT" : "",
2389			                    (io->io_hdr.flags &
2390		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2391				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2392				sbuf_finish(&sb);
2393				printf("%s\n", sbuf_data(&sb));
2394			}
2395			mtx_unlock(&lun->lun_lock);
2396		}
2397		printf("OOA queues dump done\n");
2398		mtx_unlock(&softc->ctl_lock);
2399		break;
2400	}
2401	case CTL_GET_OOA: {
2402		struct ctl_lun *lun;
2403		struct ctl_ooa *ooa_hdr;
2404		struct ctl_ooa_entry *entries;
2405		uint32_t cur_fill_num;
2406
2407		ooa_hdr = (struct ctl_ooa *)addr;
2408
2409		if ((ooa_hdr->alloc_len == 0)
2410		 || (ooa_hdr->alloc_num == 0)) {
2411			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2412			       "must be non-zero\n", __func__,
2413			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2414			retval = EINVAL;
2415			break;
2416		}
2417
2418		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2419		    sizeof(struct ctl_ooa_entry))) {
2420			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2421			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2422			       __func__, ooa_hdr->alloc_len,
2423			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2424			retval = EINVAL;
2425			break;
2426		}
2427
2428		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2429		if (entries == NULL) {
2430			printf("%s: could not allocate %d bytes for OOA "
2431			       "dump\n", __func__, ooa_hdr->alloc_len);
2432			retval = ENOMEM;
2433			break;
2434		}
2435
2436		mtx_lock(&softc->ctl_lock);
2437		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2438		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2439		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2440			mtx_unlock(&softc->ctl_lock);
2441			free(entries, M_CTL);
2442			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2443			       __func__, (uintmax_t)ooa_hdr->lun_num);
2444			retval = EINVAL;
2445			break;
2446		}
2447
2448		cur_fill_num = 0;
2449
2450		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2451			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2452				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2453					ooa_hdr, entries);
2454				if (retval != 0)
2455					break;
2456			}
2457			if (retval != 0) {
2458				mtx_unlock(&softc->ctl_lock);
2459				free(entries, M_CTL);
2460				break;
2461			}
2462		} else {
2463			lun = softc->ctl_luns[ooa_hdr->lun_num];
2464
2465			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2466						    entries);
2467		}
2468		mtx_unlock(&softc->ctl_lock);
2469
2470		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2471		ooa_hdr->fill_len = ooa_hdr->fill_num *
2472			sizeof(struct ctl_ooa_entry);
2473		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2474		if (retval != 0) {
2475			printf("%s: error copying out %d bytes for OOA dump\n",
2476			       __func__, ooa_hdr->fill_len);
2477		}
2478
2479		getbintime(&ooa_hdr->cur_bt);
2480
2481		if (cur_fill_num > ooa_hdr->alloc_num) {
2482			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2483			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2484		} else {
2485			ooa_hdr->dropped_num = 0;
2486			ooa_hdr->status = CTL_OOA_OK;
2487		}
2488
2489		free(entries, M_CTL);
2490		break;
2491	}
2492	case CTL_CHECK_OOA: {
2493		union ctl_io *io;
2494		struct ctl_lun *lun;
2495		struct ctl_ooa_info *ooa_info;
2496
2497
2498		ooa_info = (struct ctl_ooa_info *)addr;
2499
2500		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2501			ooa_info->status = CTL_OOA_INVALID_LUN;
2502			break;
2503		}
2504		mtx_lock(&softc->ctl_lock);
2505		lun = softc->ctl_luns[ooa_info->lun_id];
2506		if (lun == NULL) {
2507			mtx_unlock(&softc->ctl_lock);
2508			ooa_info->status = CTL_OOA_INVALID_LUN;
2509			break;
2510		}
2511		mtx_lock(&lun->lun_lock);
2512		mtx_unlock(&softc->ctl_lock);
2513		ooa_info->num_entries = 0;
2514		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2515		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2516		     &io->io_hdr, ooa_links)) {
2517			ooa_info->num_entries++;
2518		}
2519		mtx_unlock(&lun->lun_lock);
2520
2521		ooa_info->status = CTL_OOA_SUCCESS;
2522
2523		break;
2524	}
2525	case CTL_HARD_START:
2526	case CTL_HARD_STOP: {
2527		struct ctl_fe_ioctl_startstop_info ss_info;
2528		struct cfi_metatask *metatask;
2529		struct mtx hs_mtx;
2530
2531		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2532
2533		cv_init(&ss_info.sem, "hard start/stop cv" );
2534
2535		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2536		if (metatask == NULL) {
2537			retval = ENOMEM;
2538			mtx_destroy(&hs_mtx);
2539			break;
2540		}
2541
2542		if (cmd == CTL_HARD_START)
2543			metatask->tasktype = CFI_TASK_STARTUP;
2544		else
2545			metatask->tasktype = CFI_TASK_SHUTDOWN;
2546
2547		metatask->callback = ctl_ioctl_hard_startstop_callback;
2548		metatask->callback_arg = &ss_info;
2549
2550		cfi_action(metatask);
2551
2552		/* Wait for the callback */
2553		mtx_lock(&hs_mtx);
2554		cv_wait_sig(&ss_info.sem, &hs_mtx);
2555		mtx_unlock(&hs_mtx);
2556
2557		/*
2558		 * All information has been copied from the metatask by the
2559		 * time cv_broadcast() is called, so we free the metatask here.
2560		 */
2561		cfi_free_metatask(metatask);
2562
2563		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2564
2565		mtx_destroy(&hs_mtx);
2566		break;
2567	}
2568	case CTL_BBRREAD: {
2569		struct ctl_bbrread_info *bbr_info;
2570		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2571		struct mtx bbr_mtx;
2572		struct cfi_metatask *metatask;
2573
2574		bbr_info = (struct ctl_bbrread_info *)addr;
2575
2576		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2577
2578		bzero(&bbr_mtx, sizeof(bbr_mtx));
2579		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2580
2581		fe_bbr_info.bbr_info = bbr_info;
2582		fe_bbr_info.lock = &bbr_mtx;
2583
2584		cv_init(&fe_bbr_info.sem, "BBR read cv");
2585		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2586
2587		if (metatask == NULL) {
2588			mtx_destroy(&bbr_mtx);
2589			cv_destroy(&fe_bbr_info.sem);
2590			retval = ENOMEM;
2591			break;
2592		}
2593		metatask->tasktype = CFI_TASK_BBRREAD;
2594		metatask->callback = ctl_ioctl_bbrread_callback;
2595		metatask->callback_arg = &fe_bbr_info;
2596		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2597		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2598		metatask->taskinfo.bbrread.len = bbr_info->len;
2599
2600		cfi_action(metatask);
2601
2602		mtx_lock(&bbr_mtx);
2603		while (fe_bbr_info.wakeup_done == 0)
2604			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2605		mtx_unlock(&bbr_mtx);
2606
2607		bbr_info->status = metatask->status;
2608		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2609		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2610		memcpy(&bbr_info->sense_data,
2611		       &metatask->taskinfo.bbrread.sense_data,
2612		       ctl_min(sizeof(bbr_info->sense_data),
2613			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2614
2615		cfi_free_metatask(metatask);
2616
2617		mtx_destroy(&bbr_mtx);
2618		cv_destroy(&fe_bbr_info.sem);
2619
2620		break;
2621	}
2622	case CTL_DELAY_IO: {
2623		struct ctl_io_delay_info *delay_info;
2624#ifdef CTL_IO_DELAY
2625		struct ctl_lun *lun;
2626#endif /* CTL_IO_DELAY */
2627
2628		delay_info = (struct ctl_io_delay_info *)addr;
2629
2630#ifdef CTL_IO_DELAY
2631		mtx_lock(&softc->ctl_lock);
2632
2633		if ((delay_info->lun_id > CTL_MAX_LUNS)
2634		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2635			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2636		} else {
2637			lun = softc->ctl_luns[delay_info->lun_id];
2638			mtx_lock(&lun->lun_lock);
2639
2640			delay_info->status = CTL_DELAY_STATUS_OK;
2641
2642			switch (delay_info->delay_type) {
2643			case CTL_DELAY_TYPE_CONT:
2644				break;
2645			case CTL_DELAY_TYPE_ONESHOT:
2646				break;
2647			default:
2648				delay_info->status =
2649					CTL_DELAY_STATUS_INVALID_TYPE;
2650				break;
2651			}
2652
2653			switch (delay_info->delay_loc) {
2654			case CTL_DELAY_LOC_DATAMOVE:
2655				lun->delay_info.datamove_type =
2656					delay_info->delay_type;
2657				lun->delay_info.datamove_delay =
2658					delay_info->delay_secs;
2659				break;
2660			case CTL_DELAY_LOC_DONE:
2661				lun->delay_info.done_type =
2662					delay_info->delay_type;
2663				lun->delay_info.done_delay =
2664					delay_info->delay_secs;
2665				break;
2666			default:
2667				delay_info->status =
2668					CTL_DELAY_STATUS_INVALID_LOC;
2669				break;
2670			}
2671			mtx_unlock(&lun->lun_lock);
2672		}
2673
2674		mtx_unlock(&softc->ctl_lock);
2675#else
2676		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2677#endif /* CTL_IO_DELAY */
2678		break;
2679	}
2680	case CTL_REALSYNC_SET: {
2681		int *syncstate;
2682
2683		syncstate = (int *)addr;
2684
2685		mtx_lock(&softc->ctl_lock);
2686		switch (*syncstate) {
2687		case 0:
2688			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2689			break;
2690		case 1:
2691			softc->flags |= CTL_FLAG_REAL_SYNC;
2692			break;
2693		default:
2694			retval = EINVAL;
2695			break;
2696		}
2697		mtx_unlock(&softc->ctl_lock);
2698		break;
2699	}
2700	case CTL_REALSYNC_GET: {
2701		int *syncstate;
2702
2703		syncstate = (int*)addr;
2704
2705		mtx_lock(&softc->ctl_lock);
2706		if (softc->flags & CTL_FLAG_REAL_SYNC)
2707			*syncstate = 1;
2708		else
2709			*syncstate = 0;
2710		mtx_unlock(&softc->ctl_lock);
2711
2712		break;
2713	}
2714	case CTL_SETSYNC:
2715	case CTL_GETSYNC: {
2716		struct ctl_sync_info *sync_info;
2717		struct ctl_lun *lun;
2718
2719		sync_info = (struct ctl_sync_info *)addr;
2720
2721		mtx_lock(&softc->ctl_lock);
2722		lun = softc->ctl_luns[sync_info->lun_id];
2723		if (lun == NULL) {
2724			mtx_unlock(&softc->ctl_lock);
2725			sync_info->status = CTL_GS_SYNC_NO_LUN;
2726		}
2727		/*
2728		 * Get or set the sync interval.  We're not bounds checking
2729		 * in the set case, hopefully the user won't do something
2730		 * silly.
2731		 */
2732		mtx_lock(&lun->lun_lock);
2733		mtx_unlock(&softc->ctl_lock);
2734		if (cmd == CTL_GETSYNC)
2735			sync_info->sync_interval = lun->sync_interval;
2736		else
2737			lun->sync_interval = sync_info->sync_interval;
2738		mtx_unlock(&lun->lun_lock);
2739
2740		sync_info->status = CTL_GS_SYNC_OK;
2741
2742		break;
2743	}
2744	case CTL_GETSTATS: {
2745		struct ctl_stats *stats;
2746		struct ctl_lun *lun;
2747		int i;
2748
2749		stats = (struct ctl_stats *)addr;
2750
2751		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2752		     stats->alloc_len) {
2753			stats->status = CTL_SS_NEED_MORE_SPACE;
2754			stats->num_luns = softc->num_luns;
2755			break;
2756		}
2757		/*
2758		 * XXX KDM no locking here.  If the LUN list changes,
2759		 * things can blow up.
2760		 */
2761		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2762		     i++, lun = STAILQ_NEXT(lun, links)) {
2763			retval = copyout(&lun->stats, &stats->lun_stats[i],
2764					 sizeof(lun->stats));
2765			if (retval != 0)
2766				break;
2767		}
2768		stats->num_luns = softc->num_luns;
2769		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2770				 softc->num_luns;
2771		stats->status = CTL_SS_OK;
2772#ifdef CTL_TIME_IO
2773		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2774#else
2775		stats->flags = CTL_STATS_FLAG_NONE;
2776#endif
2777		getnanouptime(&stats->timestamp);
2778		break;
2779	}
2780	case CTL_ERROR_INJECT: {
2781		struct ctl_error_desc *err_desc, *new_err_desc;
2782		struct ctl_lun *lun;
2783
2784		err_desc = (struct ctl_error_desc *)addr;
2785
2786		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2787				      M_WAITOK | M_ZERO);
2788		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2789
2790		mtx_lock(&softc->ctl_lock);
2791		lun = softc->ctl_luns[err_desc->lun_id];
2792		if (lun == NULL) {
2793			mtx_unlock(&softc->ctl_lock);
2794			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2795			       __func__, (uintmax_t)err_desc->lun_id);
2796			retval = EINVAL;
2797			break;
2798		}
2799		mtx_lock(&lun->lun_lock);
2800		mtx_unlock(&softc->ctl_lock);
2801
2802		/*
2803		 * We could do some checking here to verify the validity
2804		 * of the request, but given the complexity of error
2805		 * injection requests, the checking logic would be fairly
2806		 * complex.
2807		 *
2808		 * For now, if the request is invalid, it just won't get
2809		 * executed and might get deleted.
2810		 */
2811		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2812
2813		/*
2814		 * XXX KDM check to make sure the serial number is unique,
2815		 * in case we somehow manage to wrap.  That shouldn't
2816		 * happen for a very long time, but it's the right thing to
2817		 * do.
2818		 */
2819		new_err_desc->serial = lun->error_serial;
2820		err_desc->serial = lun->error_serial;
2821		lun->error_serial++;
2822
2823		mtx_unlock(&lun->lun_lock);
2824		break;
2825	}
2826	case CTL_ERROR_INJECT_DELETE: {
2827		struct ctl_error_desc *delete_desc, *desc, *desc2;
2828		struct ctl_lun *lun;
2829		int delete_done;
2830
2831		delete_desc = (struct ctl_error_desc *)addr;
2832		delete_done = 0;
2833
2834		mtx_lock(&softc->ctl_lock);
2835		lun = softc->ctl_luns[delete_desc->lun_id];
2836		if (lun == NULL) {
2837			mtx_unlock(&softc->ctl_lock);
2838			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2839			       __func__, (uintmax_t)delete_desc->lun_id);
2840			retval = EINVAL;
2841			break;
2842		}
2843		mtx_lock(&lun->lun_lock);
2844		mtx_unlock(&softc->ctl_lock);
2845		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2846			if (desc->serial != delete_desc->serial)
2847				continue;
2848
2849			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2850				      links);
2851			free(desc, M_CTL);
2852			delete_done = 1;
2853		}
2854		mtx_unlock(&lun->lun_lock);
2855		if (delete_done == 0) {
2856			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2857			       "error serial %ju on LUN %u\n", __func__,
2858			       delete_desc->serial, delete_desc->lun_id);
2859			retval = EINVAL;
2860			break;
2861		}
2862		break;
2863	}
2864	case CTL_DUMP_STRUCTS: {
2865		int i, j, k;
2866		struct ctl_frontend *fe;
2867
2868		printf("CTL IID to WWPN map start:\n");
2869		for (i = 0; i < CTL_MAX_PORTS; i++) {
2870			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2871				if (softc->wwpn_iid[i][j].in_use == 0)
2872					continue;
2873
2874				printf("port %d iid %u WWPN %#jx\n",
2875				       softc->wwpn_iid[i][j].port,
2876				       softc->wwpn_iid[i][j].iid,
2877				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2878			}
2879		}
2880		printf("CTL IID to WWPN map end\n");
2881		printf("CTL Persistent Reservation information start:\n");
2882		for (i = 0; i < CTL_MAX_LUNS; i++) {
2883			struct ctl_lun *lun;
2884
2885			lun = softc->ctl_luns[i];
2886
2887			if ((lun == NULL)
2888			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2889				continue;
2890
2891			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2892				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2893					if (lun->per_res[j+k].registered == 0)
2894						continue;
2895					printf("LUN %d port %d iid %d key "
2896					       "%#jx\n", i, j, k,
2897					       (uintmax_t)scsi_8btou64(
2898					       lun->per_res[j+k].res_key.key));
2899				}
2900			}
2901		}
2902		printf("CTL Persistent Reservation information end\n");
2903		printf("CTL Frontends:\n");
2904		/*
2905		 * XXX KDM calling this without a lock.  We'd likely want
2906		 * to drop the lock before calling the frontend's dump
2907		 * routine anyway.
2908		 */
2909		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2910			printf("Frontend %s Type %u pport %d vport %d WWNN "
2911			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2912			       fe->physical_port, fe->virtual_port,
2913			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2914
2915			/*
2916			 * Frontends are not required to support the dump
2917			 * routine.
2918			 */
2919			if (fe->fe_dump == NULL)
2920				continue;
2921
2922			fe->fe_dump();
2923		}
2924		printf("CTL Frontend information end\n");
2925		break;
2926	}
2927	case CTL_LUN_REQ: {
2928		struct ctl_lun_req *lun_req;
2929		struct ctl_backend_driver *backend;
2930
2931		lun_req = (struct ctl_lun_req *)addr;
2932
2933		backend = ctl_backend_find(lun_req->backend);
2934		if (backend == NULL) {
2935			lun_req->status = CTL_LUN_ERROR;
2936			snprintf(lun_req->error_str,
2937				 sizeof(lun_req->error_str),
2938				 "Backend \"%s\" not found.",
2939				 lun_req->backend);
2940			break;
2941		}
2942		if (lun_req->num_be_args > 0) {
2943			lun_req->kern_be_args = ctl_copyin_args(
2944				lun_req->num_be_args,
2945				lun_req->be_args,
2946				lun_req->error_str,
2947				sizeof(lun_req->error_str));
2948			if (lun_req->kern_be_args == NULL) {
2949				lun_req->status = CTL_LUN_ERROR;
2950				break;
2951			}
2952		}
2953
2954		retval = backend->ioctl(dev, cmd, addr, flag, td);
2955
2956		if (lun_req->num_be_args > 0) {
2957			ctl_free_args(lun_req->num_be_args,
2958				      lun_req->kern_be_args);
2959		}
2960		break;
2961	}
2962	case CTL_LUN_LIST: {
2963		struct sbuf *sb;
2964		struct ctl_lun *lun;
2965		struct ctl_lun_list *list;
2966		struct ctl_be_lun_option *opt;
2967
2968		list = (struct ctl_lun_list *)addr;
2969
2970		/*
2971		 * Allocate a fixed length sbuf here, based on the length
2972		 * of the user's buffer.  We could allocate an auto-extending
2973		 * buffer, and then tell the user how much larger our
2974		 * amount of data is than his buffer, but that presents
2975		 * some problems:
2976		 *
2977		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2978		 *     we can't hold a lock while calling them with an
2979		 *     auto-extending buffer.
2980 		 *
2981		 * 2.  There is not currently a LUN reference counting
2982		 *     mechanism, outside of outstanding transactions on
2983		 *     the LUN's OOA queue.  So a LUN could go away on us
2984		 *     while we're getting the LUN number, backend-specific
2985		 *     information, etc.  Thus, given the way things
2986		 *     currently work, we need to hold the CTL lock while
2987		 *     grabbing LUN information.
2988		 *
2989		 * So, from the user's standpoint, the best thing to do is
2990		 * allocate what he thinks is a reasonable buffer length,
2991		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2992		 * double the buffer length and try again.  (And repeat
2993		 * that until he succeeds.)
2994		 */
2995		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2996		if (sb == NULL) {
2997			list->status = CTL_LUN_LIST_ERROR;
2998			snprintf(list->error_str, sizeof(list->error_str),
2999				 "Unable to allocate %d bytes for LUN list",
3000				 list->alloc_len);
3001			break;
3002		}
3003
3004		sbuf_printf(sb, "<ctllunlist>\n");
3005
3006		mtx_lock(&softc->ctl_lock);
3007		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3008			mtx_lock(&lun->lun_lock);
3009			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3010					     (uintmax_t)lun->lun);
3011
3012			/*
3013			 * Bail out as soon as we see that we've overfilled
3014			 * the buffer.
3015			 */
3016			if (retval != 0)
3017				break;
3018
3019			retval = sbuf_printf(sb, "<backend_type>%s"
3020					     "</backend_type>\n",
3021					     (lun->backend == NULL) ?  "none" :
3022					     lun->backend->name);
3023
3024			if (retval != 0)
3025				break;
3026
3027			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3028					     lun->be_lun->lun_type);
3029
3030			if (retval != 0)
3031				break;
3032
3033			if (lun->backend == NULL) {
3034				retval = sbuf_printf(sb, "</lun>\n");
3035				if (retval != 0)
3036					break;
3037				continue;
3038			}
3039
3040			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3041					     (lun->be_lun->maxlba > 0) ?
3042					     lun->be_lun->maxlba + 1 : 0);
3043
3044			if (retval != 0)
3045				break;
3046
3047			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3048					     lun->be_lun->blocksize);
3049
3050			if (retval != 0)
3051				break;
3052
3053			retval = sbuf_printf(sb, "<serial_number>");
3054
3055			if (retval != 0)
3056				break;
3057
3058			retval = ctl_sbuf_printf_esc(sb,
3059						     lun->be_lun->serial_num);
3060
3061			if (retval != 0)
3062				break;
3063
3064			retval = sbuf_printf(sb, "</serial_number>\n");
3065
3066			if (retval != 0)
3067				break;
3068
3069			retval = sbuf_printf(sb, "<device_id>");
3070
3071			if (retval != 0)
3072				break;
3073
3074			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3075
3076			if (retval != 0)
3077				break;
3078
3079			retval = sbuf_printf(sb, "</device_id>\n");
3080
3081			if (retval != 0)
3082				break;
3083
3084			if (lun->backend->lun_info != NULL) {
3085				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3086				if (retval != 0)
3087					break;
3088			}
3089			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3090				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3091				if (retval != 0)
3092					break;
3093			}
3094
3095			retval = sbuf_printf(sb, "</lun>\n");
3096
3097			if (retval != 0)
3098				break;
3099			mtx_unlock(&lun->lun_lock);
3100		}
3101		if (lun != NULL)
3102			mtx_unlock(&lun->lun_lock);
3103		mtx_unlock(&softc->ctl_lock);
3104
3105		if ((retval != 0)
3106		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3107			retval = 0;
3108			sbuf_delete(sb);
3109			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3110			snprintf(list->error_str, sizeof(list->error_str),
3111				 "Out of space, %d bytes is too small",
3112				 list->alloc_len);
3113			break;
3114		}
3115
3116		sbuf_finish(sb);
3117
3118		retval = copyout(sbuf_data(sb), list->lun_xml,
3119				 sbuf_len(sb) + 1);
3120
3121		list->fill_len = sbuf_len(sb) + 1;
3122		list->status = CTL_LUN_LIST_OK;
3123		sbuf_delete(sb);
3124		break;
3125	}
3126	case CTL_ISCSI: {
3127		struct ctl_iscsi *ci;
3128		struct ctl_frontend *fe;
3129
3130		ci = (struct ctl_iscsi *)addr;
3131
3132		mtx_lock(&softc->ctl_lock);
3133		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3134			if (strcmp(fe->port_name, "iscsi") == 0)
3135				break;
3136		}
3137		mtx_unlock(&softc->ctl_lock);
3138
3139		if (fe == NULL) {
3140			ci->status = CTL_ISCSI_ERROR;
3141			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3142			break;
3143		}
3144
3145		retval = fe->ioctl(dev, cmd, addr, flag, td);
3146		break;
3147	}
3148	default: {
3149		/* XXX KDM should we fix this? */
3150#if 0
3151		struct ctl_backend_driver *backend;
3152		unsigned int type;
3153		int found;
3154
3155		found = 0;
3156
3157		/*
3158		 * We encode the backend type as the ioctl type for backend
3159		 * ioctls.  So parse it out here, and then search for a
3160		 * backend of this type.
3161		 */
3162		type = _IOC_TYPE(cmd);
3163
3164		STAILQ_FOREACH(backend, &softc->be_list, links) {
3165			if (backend->type == type) {
3166				found = 1;
3167				break;
3168			}
3169		}
3170		if (found == 0) {
3171			printf("ctl: unknown ioctl command %#lx or backend "
3172			       "%d\n", cmd, type);
3173			retval = EINVAL;
3174			break;
3175		}
3176		retval = backend->ioctl(dev, cmd, addr, flag, td);
3177#endif
3178		retval = ENOTTY;
3179		break;
3180	}
3181	}
3182	return (retval);
3183}
3184
3185uint32_t
3186ctl_get_initindex(struct ctl_nexus *nexus)
3187{
3188	if (nexus->targ_port < CTL_MAX_PORTS)
3189		return (nexus->initid.id +
3190			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3191	else
3192		return (nexus->initid.id +
3193		       ((nexus->targ_port - CTL_MAX_PORTS) *
3194			CTL_MAX_INIT_PER_PORT));
3195}
3196
3197uint32_t
3198ctl_get_resindex(struct ctl_nexus *nexus)
3199{
3200	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3201}
3202
3203uint32_t
3204ctl_port_idx(int port_num)
3205{
3206	if (port_num < CTL_MAX_PORTS)
3207		return(port_num);
3208	else
3209		return(port_num - CTL_MAX_PORTS);
3210}
3211
3212/*
3213 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3214 * that are a power of 2.
3215 */
3216int
3217ctl_ffz(uint32_t *mask, uint32_t size)
3218{
3219	uint32_t num_chunks, num_pieces;
3220	int i, j;
3221
3222	num_chunks = (size >> 5);
3223	if (num_chunks == 0)
3224		num_chunks++;
3225	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3226
3227	for (i = 0; i < num_chunks; i++) {
3228		for (j = 0; j < num_pieces; j++) {
3229			if ((mask[i] & (1 << j)) == 0)
3230				return ((i << 5) + j);
3231		}
3232	}
3233
3234	return (-1);
3235}
3236
3237int
3238ctl_set_mask(uint32_t *mask, uint32_t bit)
3239{
3240	uint32_t chunk, piece;
3241
3242	chunk = bit >> 5;
3243	piece = bit % (sizeof(uint32_t) * 8);
3244
3245	if ((mask[chunk] & (1 << piece)) != 0)
3246		return (-1);
3247	else
3248		mask[chunk] |= (1 << piece);
3249
3250	return (0);
3251}
3252
3253int
3254ctl_clear_mask(uint32_t *mask, uint32_t bit)
3255{
3256	uint32_t chunk, piece;
3257
3258	chunk = bit >> 5;
3259	piece = bit % (sizeof(uint32_t) * 8);
3260
3261	if ((mask[chunk] & (1 << piece)) == 0)
3262		return (-1);
3263	else
3264		mask[chunk] &= ~(1 << piece);
3265
3266	return (0);
3267}
3268
3269int
3270ctl_is_set(uint32_t *mask, uint32_t bit)
3271{
3272	uint32_t chunk, piece;
3273
3274	chunk = bit >> 5;
3275	piece = bit % (sizeof(uint32_t) * 8);
3276
3277	if ((mask[chunk] & (1 << piece)) == 0)
3278		return (0);
3279	else
3280		return (1);
3281}
3282
3283#ifdef unused
3284/*
3285 * The bus, target and lun are optional, they can be filled in later.
3286 * can_wait is used to determine whether we can wait on the malloc or not.
3287 */
3288union ctl_io*
3289ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3290	      uint32_t targ_lun, int can_wait)
3291{
3292	union ctl_io *io;
3293
3294	if (can_wait)
3295		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3296	else
3297		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3298
3299	if (io != NULL) {
3300		io->io_hdr.io_type = io_type;
3301		io->io_hdr.targ_port = targ_port;
3302		/*
3303		 * XXX KDM this needs to change/go away.  We need to move
3304		 * to a preallocated pool of ctl_scsiio structures.
3305		 */
3306		io->io_hdr.nexus.targ_target.id = targ_target;
3307		io->io_hdr.nexus.targ_lun = targ_lun;
3308	}
3309
3310	return (io);
3311}
3312
3313void
3314ctl_kfree_io(union ctl_io *io)
3315{
3316	free(io, M_CTL);
3317}
3318#endif /* unused */
3319
3320/*
3321 * ctl_softc, pool_type, total_ctl_io are passed in.
3322 * npool is passed out.
3323 */
3324int
3325ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3326		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3327{
3328	uint32_t i;
3329	union ctl_io *cur_io, *next_io;
3330	struct ctl_io_pool *pool;
3331	int retval;
3332
3333	retval = 0;
3334
3335	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3336					    M_NOWAIT | M_ZERO);
3337	if (pool == NULL) {
3338		retval = ENOMEM;
3339		goto bailout;
3340	}
3341
3342	pool->type = pool_type;
3343	pool->ctl_softc = ctl_softc;
3344
3345	mtx_lock(&ctl_softc->pool_lock);
3346	pool->id = ctl_softc->cur_pool_id++;
3347	mtx_unlock(&ctl_softc->pool_lock);
3348
3349	pool->flags = CTL_POOL_FLAG_NONE;
3350	pool->refcount = 1;		/* Reference for validity. */
3351	STAILQ_INIT(&pool->free_queue);
3352
3353	/*
3354	 * XXX KDM other options here:
3355	 * - allocate a page at a time
3356	 * - allocate one big chunk of memory.
3357	 * Page allocation might work well, but would take a little more
3358	 * tracking.
3359	 */
3360	for (i = 0; i < total_ctl_io; i++) {
3361		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3362						M_NOWAIT);
3363		if (cur_io == NULL) {
3364			retval = ENOMEM;
3365			break;
3366		}
3367		cur_io->io_hdr.pool = pool;
3368		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3369		pool->total_ctl_io++;
3370		pool->free_ctl_io++;
3371	}
3372
3373	if (retval != 0) {
3374		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3375		     cur_io != NULL; cur_io = next_io) {
3376			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3377							      links);
3378			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3379				      ctl_io_hdr, links);
3380			free(cur_io, M_CTL);
3381		}
3382
3383		free(pool, M_CTL);
3384		goto bailout;
3385	}
3386	mtx_lock(&ctl_softc->pool_lock);
3387	ctl_softc->num_pools++;
3388	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3389	/*
3390	 * Increment our usage count if this is an external consumer, so we
3391	 * can't get unloaded until the external consumer (most likely a
3392	 * FETD) unloads and frees his pool.
3393	 *
3394	 * XXX KDM will this increment the caller's module use count, or
3395	 * mine?
3396	 */
3397#if 0
3398	if ((pool_type != CTL_POOL_EMERGENCY)
3399	 && (pool_type != CTL_POOL_INTERNAL)
3400	 && (pool_type != CTL_POOL_IOCTL)
3401	 && (pool_type != CTL_POOL_4OTHERSC))
3402		MOD_INC_USE_COUNT;
3403#endif
3404
3405	mtx_unlock(&ctl_softc->pool_lock);
3406
3407	*npool = pool;
3408
3409bailout:
3410
3411	return (retval);
3412}
3413
3414static int
3415ctl_pool_acquire(struct ctl_io_pool *pool)
3416{
3417
3418	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3419
3420	if (pool->flags & CTL_POOL_FLAG_INVALID)
3421		return (EINVAL);
3422
3423	pool->refcount++;
3424
3425	return (0);
3426}
3427
3428static void
3429ctl_pool_release(struct ctl_io_pool *pool)
3430{
3431	struct ctl_softc *ctl_softc = pool->ctl_softc;
3432	union ctl_io *io;
3433
3434	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3435
3436	if (--pool->refcount != 0)
3437		return;
3438
3439	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3440		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3441			      links);
3442		free(io, M_CTL);
3443	}
3444
3445	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3446	ctl_softc->num_pools--;
3447
3448	/*
3449	 * XXX KDM will this decrement the caller's usage count or mine?
3450	 */
3451#if 0
3452	if ((pool->type != CTL_POOL_EMERGENCY)
3453	 && (pool->type != CTL_POOL_INTERNAL)
3454	 && (pool->type != CTL_POOL_IOCTL))
3455		MOD_DEC_USE_COUNT;
3456#endif
3457
3458	free(pool, M_CTL);
3459}
3460
3461void
3462ctl_pool_free(struct ctl_io_pool *pool)
3463{
3464	struct ctl_softc *ctl_softc;
3465
3466	if (pool == NULL)
3467		return;
3468
3469	ctl_softc = pool->ctl_softc;
3470	mtx_lock(&ctl_softc->pool_lock);
3471	pool->flags |= CTL_POOL_FLAG_INVALID;
3472	ctl_pool_release(pool);
3473	mtx_unlock(&ctl_softc->pool_lock);
3474}
3475
3476/*
3477 * This routine does not block (except for spinlocks of course).
3478 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3479 * possible.
3480 */
3481union ctl_io *
3482ctl_alloc_io(void *pool_ref)
3483{
3484	union ctl_io *io;
3485	struct ctl_softc *ctl_softc;
3486	struct ctl_io_pool *pool, *npool;
3487	struct ctl_io_pool *emergency_pool;
3488
3489	pool = (struct ctl_io_pool *)pool_ref;
3490
3491	if (pool == NULL) {
3492		printf("%s: pool is NULL\n", __func__);
3493		return (NULL);
3494	}
3495
3496	emergency_pool = NULL;
3497
3498	ctl_softc = pool->ctl_softc;
3499
3500	mtx_lock(&ctl_softc->pool_lock);
3501	/*
3502	 * First, try to get the io structure from the user's pool.
3503	 */
3504	if (ctl_pool_acquire(pool) == 0) {
3505		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3506		if (io != NULL) {
3507			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3508			pool->total_allocated++;
3509			pool->free_ctl_io--;
3510			mtx_unlock(&ctl_softc->pool_lock);
3511			return (io);
3512		} else
3513			ctl_pool_release(pool);
3514	}
3515	/*
3516	 * If he doesn't have any io structures left, search for an
3517	 * emergency pool and grab one from there.
3518	 */
3519	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3520		if (npool->type != CTL_POOL_EMERGENCY)
3521			continue;
3522
3523		if (ctl_pool_acquire(npool) != 0)
3524			continue;
3525
3526		emergency_pool = npool;
3527
3528		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3529		if (io != NULL) {
3530			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3531			npool->total_allocated++;
3532			npool->free_ctl_io--;
3533			mtx_unlock(&ctl_softc->pool_lock);
3534			return (io);
3535		} else
3536			ctl_pool_release(npool);
3537	}
3538
3539	/* Drop the spinlock before we malloc */
3540	mtx_unlock(&ctl_softc->pool_lock);
3541
3542	/*
3543	 * The emergency pool (if it exists) didn't have one, so try an
3544	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3545	 */
3546	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3547	if (io != NULL) {
3548		/*
3549		 * If the emergency pool exists but is empty, add this
3550		 * ctl_io to its list when it gets freed.
3551		 */
3552		if (emergency_pool != NULL) {
3553			mtx_lock(&ctl_softc->pool_lock);
3554			if (ctl_pool_acquire(emergency_pool) == 0) {
3555				io->io_hdr.pool = emergency_pool;
3556				emergency_pool->total_ctl_io++;
3557				/*
3558				 * Need to bump this, otherwise
3559				 * total_allocated and total_freed won't
3560				 * match when we no longer have anything
3561				 * outstanding.
3562				 */
3563				emergency_pool->total_allocated++;
3564			}
3565			mtx_unlock(&ctl_softc->pool_lock);
3566		} else
3567			io->io_hdr.pool = NULL;
3568	}
3569
3570	return (io);
3571}
3572
3573void
3574ctl_free_io(union ctl_io *io)
3575{
3576	if (io == NULL)
3577		return;
3578
3579	/*
3580	 * If this ctl_io has a pool, return it to that pool.
3581	 */
3582	if (io->io_hdr.pool != NULL) {
3583		struct ctl_io_pool *pool;
3584
3585		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3586		mtx_lock(&pool->ctl_softc->pool_lock);
3587		io->io_hdr.io_type = 0xff;
3588		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3589		pool->total_freed++;
3590		pool->free_ctl_io++;
3591		ctl_pool_release(pool);
3592		mtx_unlock(&pool->ctl_softc->pool_lock);
3593	} else {
3594		/*
3595		 * Otherwise, just free it.  We probably malloced it and
3596		 * the emergency pool wasn't available.
3597		 */
3598		free(io, M_CTL);
3599	}
3600
3601}
3602
3603void
3604ctl_zero_io(union ctl_io *io)
3605{
3606	void *pool_ref;
3607
3608	if (io == NULL)
3609		return;
3610
3611	/*
3612	 * May need to preserve linked list pointers at some point too.
3613	 */
3614	pool_ref = io->io_hdr.pool;
3615
3616	memset(io, 0, sizeof(*io));
3617
3618	io->io_hdr.pool = pool_ref;
3619}
3620
3621/*
3622 * This routine is currently used for internal copies of ctl_ios that need
3623 * to persist for some reason after we've already returned status to the
3624 * FETD.  (Thus the flag set.)
3625 *
3626 * XXX XXX
3627 * Note that this makes a blind copy of all fields in the ctl_io, except
3628 * for the pool reference.  This includes any memory that has been
3629 * allocated!  That memory will no longer be valid after done has been
3630 * called, so this would be VERY DANGEROUS for command that actually does
3631 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3632 * start and stop commands, which don't transfer any data, so this is not a
3633 * problem.  If it is used for anything else, the caller would also need to
3634 * allocate data buffer space and this routine would need to be modified to
3635 * copy the data buffer(s) as well.
3636 */
3637void
3638ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3639{
3640	void *pool_ref;
3641
3642	if ((src == NULL)
3643	 || (dest == NULL))
3644		return;
3645
3646	/*
3647	 * May need to preserve linked list pointers at some point too.
3648	 */
3649	pool_ref = dest->io_hdr.pool;
3650
3651	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3652
3653	dest->io_hdr.pool = pool_ref;
3654	/*
3655	 * We need to know that this is an internal copy, and doesn't need
3656	 * to get passed back to the FETD that allocated it.
3657	 */
3658	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3659}
3660
3661#ifdef NEEDTOPORT
3662static void
3663ctl_update_power_subpage(struct copan_power_subpage *page)
3664{
3665	int num_luns, num_partitions, config_type;
3666	struct ctl_softc *softc;
3667	cs_BOOL_t aor_present, shelf_50pct_power;
3668	cs_raidset_personality_t rs_type;
3669	int max_active_luns;
3670
3671	softc = control_softc;
3672
3673	/* subtract out the processor LUN */
3674	num_luns = softc->num_luns - 1;
3675	/*
3676	 * Default to 7 LUNs active, which was the only number we allowed
3677	 * in the past.
3678	 */
3679	max_active_luns = 7;
3680
3681	num_partitions = config_GetRsPartitionInfo();
3682	config_type = config_GetConfigType();
3683	shelf_50pct_power = config_GetShelfPowerMode();
3684	aor_present = config_IsAorRsPresent();
3685
3686	rs_type = ddb_GetRsRaidType(1);
3687	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3688	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3689		EPRINT(0, "Unsupported RS type %d!", rs_type);
3690	}
3691
3692
3693	page->total_luns = num_luns;
3694
3695	switch (config_type) {
3696	case 40:
3697		/*
3698		 * In a 40 drive configuration, it doesn't matter what DC
3699		 * cards we have, whether we have AOR enabled or not,
3700		 * partitioning or not, or what type of RAIDset we have.
3701		 * In that scenario, we can power up every LUN we present
3702		 * to the user.
3703		 */
3704		max_active_luns = num_luns;
3705
3706		break;
3707	case 64:
3708		if (shelf_50pct_power == CS_FALSE) {
3709			/* 25% power */
3710			if (aor_present == CS_TRUE) {
3711				if (rs_type ==
3712				     CS_RAIDSET_PERSONALITY_RAID5) {
3713					max_active_luns = 7;
3714				} else if (rs_type ==
3715					 CS_RAIDSET_PERSONALITY_RAID1){
3716					max_active_luns = 14;
3717				} else {
3718					/* XXX KDM now what?? */
3719				}
3720			} else {
3721				if (rs_type ==
3722				     CS_RAIDSET_PERSONALITY_RAID5) {
3723					max_active_luns = 8;
3724				} else if (rs_type ==
3725					 CS_RAIDSET_PERSONALITY_RAID1){
3726					max_active_luns = 16;
3727				} else {
3728					/* XXX KDM now what?? */
3729				}
3730			}
3731		} else {
3732			/* 50% power */
3733			/*
3734			 * With 50% power in a 64 drive configuration, we
3735			 * can power all LUNs we present.
3736			 */
3737			max_active_luns = num_luns;
3738		}
3739		break;
3740	case 112:
3741		if (shelf_50pct_power == CS_FALSE) {
3742			/* 25% power */
3743			if (aor_present == CS_TRUE) {
3744				if (rs_type ==
3745				     CS_RAIDSET_PERSONALITY_RAID5) {
3746					max_active_luns = 7;
3747				} else if (rs_type ==
3748					 CS_RAIDSET_PERSONALITY_RAID1){
3749					max_active_luns = 14;
3750				} else {
3751					/* XXX KDM now what?? */
3752				}
3753			} else {
3754				if (rs_type ==
3755				     CS_RAIDSET_PERSONALITY_RAID5) {
3756					max_active_luns = 8;
3757				} else if (rs_type ==
3758					 CS_RAIDSET_PERSONALITY_RAID1){
3759					max_active_luns = 16;
3760				} else {
3761					/* XXX KDM now what?? */
3762				}
3763			}
3764		} else {
3765			/* 50% power */
3766			if (aor_present == CS_TRUE) {
3767				if (rs_type ==
3768				     CS_RAIDSET_PERSONALITY_RAID5) {
3769					max_active_luns = 14;
3770				} else if (rs_type ==
3771					 CS_RAIDSET_PERSONALITY_RAID1){
3772					/*
3773					 * We're assuming here that disk
3774					 * caching is enabled, and so we're
3775					 * able to power up half of each
3776					 * LUN, and cache all writes.
3777					 */
3778					max_active_luns = num_luns;
3779				} else {
3780					/* XXX KDM now what?? */
3781				}
3782			} else {
3783				if (rs_type ==
3784				     CS_RAIDSET_PERSONALITY_RAID5) {
3785					max_active_luns = 15;
3786				} else if (rs_type ==
3787					 CS_RAIDSET_PERSONALITY_RAID1){
3788					max_active_luns = 30;
3789				} else {
3790					/* XXX KDM now what?? */
3791				}
3792			}
3793		}
3794		break;
3795	default:
3796		/*
3797		 * In this case, we have an unknown configuration, so we
3798		 * just use the default from above.
3799		 */
3800		break;
3801	}
3802
3803	page->max_active_luns = max_active_luns;
3804#if 0
3805	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3806	       page->total_luns, page->max_active_luns);
3807#endif
3808}
3809#endif /* NEEDTOPORT */
3810
3811/*
3812 * This routine could be used in the future to load default and/or saved
3813 * mode page parameters for a particuar lun.
3814 */
3815static int
3816ctl_init_page_index(struct ctl_lun *lun)
3817{
3818	int i;
3819	struct ctl_page_index *page_index;
3820	struct ctl_softc *softc;
3821
3822	memcpy(&lun->mode_pages.index, page_index_template,
3823	       sizeof(page_index_template));
3824
3825	softc = lun->ctl_softc;
3826
3827	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3828
3829		page_index = &lun->mode_pages.index[i];
3830		/*
3831		 * If this is a disk-only mode page, there's no point in
3832		 * setting it up.  For some pages, we have to have some
3833		 * basic information about the disk in order to calculate the
3834		 * mode page data.
3835		 */
3836		if ((lun->be_lun->lun_type != T_DIRECT)
3837		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3838			continue;
3839
3840		switch (page_index->page_code & SMPH_PC_MASK) {
3841		case SMS_FORMAT_DEVICE_PAGE: {
3842			struct scsi_format_page *format_page;
3843
3844			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3845				panic("subpage is incorrect!");
3846
3847			/*
3848			 * Sectors per track are set above.  Bytes per
3849			 * sector need to be set here on a per-LUN basis.
3850			 */
3851			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3852			       &format_page_default,
3853			       sizeof(format_page_default));
3854			memcpy(&lun->mode_pages.format_page[
3855			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3856			       sizeof(format_page_changeable));
3857			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3858			       &format_page_default,
3859			       sizeof(format_page_default));
3860			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3861			       &format_page_default,
3862			       sizeof(format_page_default));
3863
3864			format_page = &lun->mode_pages.format_page[
3865				CTL_PAGE_CURRENT];
3866			scsi_ulto2b(lun->be_lun->blocksize,
3867				    format_page->bytes_per_sector);
3868
3869			format_page = &lun->mode_pages.format_page[
3870				CTL_PAGE_DEFAULT];
3871			scsi_ulto2b(lun->be_lun->blocksize,
3872				    format_page->bytes_per_sector);
3873
3874			format_page = &lun->mode_pages.format_page[
3875				CTL_PAGE_SAVED];
3876			scsi_ulto2b(lun->be_lun->blocksize,
3877				    format_page->bytes_per_sector);
3878
3879			page_index->page_data =
3880				(uint8_t *)lun->mode_pages.format_page;
3881			break;
3882		}
3883		case SMS_RIGID_DISK_PAGE: {
3884			struct scsi_rigid_disk_page *rigid_disk_page;
3885			uint32_t sectors_per_cylinder;
3886			uint64_t cylinders;
3887#ifndef	__XSCALE__
3888			int shift;
3889#endif /* !__XSCALE__ */
3890
3891			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3892				panic("invalid subpage value %d",
3893				      page_index->subpage);
3894
3895			/*
3896			 * Rotation rate and sectors per track are set
3897			 * above.  We calculate the cylinders here based on
3898			 * capacity.  Due to the number of heads and
3899			 * sectors per track we're using, smaller arrays
3900			 * may turn out to have 0 cylinders.  Linux and
3901			 * FreeBSD don't pay attention to these mode pages
3902			 * to figure out capacity, but Solaris does.  It
3903			 * seems to deal with 0 cylinders just fine, and
3904			 * works out a fake geometry based on the capacity.
3905			 */
3906			memcpy(&lun->mode_pages.rigid_disk_page[
3907			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3908			       sizeof(rigid_disk_page_default));
3909			memcpy(&lun->mode_pages.rigid_disk_page[
3910			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3911			       sizeof(rigid_disk_page_changeable));
3912			memcpy(&lun->mode_pages.rigid_disk_page[
3913			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3914			       sizeof(rigid_disk_page_default));
3915			memcpy(&lun->mode_pages.rigid_disk_page[
3916			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3917			       sizeof(rigid_disk_page_default));
3918
3919			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3920				CTL_DEFAULT_HEADS;
3921
3922			/*
3923			 * The divide method here will be more accurate,
3924			 * probably, but results in floating point being
3925			 * used in the kernel on i386 (__udivdi3()).  On the
3926			 * XScale, though, __udivdi3() is implemented in
3927			 * software.
3928			 *
3929			 * The shift method for cylinder calculation is
3930			 * accurate if sectors_per_cylinder is a power of
3931			 * 2.  Otherwise it might be slightly off -- you
3932			 * might have a bit of a truncation problem.
3933			 */
3934#ifdef	__XSCALE__
3935			cylinders = (lun->be_lun->maxlba + 1) /
3936				sectors_per_cylinder;
3937#else
3938			for (shift = 31; shift > 0; shift--) {
3939				if (sectors_per_cylinder & (1 << shift))
3940					break;
3941			}
3942			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3943#endif
3944
3945			/*
3946			 * We've basically got 3 bytes, or 24 bits for the
3947			 * cylinder size in the mode page.  If we're over,
3948			 * just round down to 2^24.
3949			 */
3950			if (cylinders > 0xffffff)
3951				cylinders = 0xffffff;
3952
3953			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3954				CTL_PAGE_CURRENT];
3955			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3956
3957			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3958				CTL_PAGE_DEFAULT];
3959			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3960
3961			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3962				CTL_PAGE_SAVED];
3963			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3964
3965			page_index->page_data =
3966				(uint8_t *)lun->mode_pages.rigid_disk_page;
3967			break;
3968		}
3969		case SMS_CACHING_PAGE: {
3970
3971			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3972				panic("invalid subpage value %d",
3973				      page_index->subpage);
3974			/*
3975			 * Defaults should be okay here, no calculations
3976			 * needed.
3977			 */
3978			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3979			       &caching_page_default,
3980			       sizeof(caching_page_default));
3981			memcpy(&lun->mode_pages.caching_page[
3982			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3983			       sizeof(caching_page_changeable));
3984			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3985			       &caching_page_default,
3986			       sizeof(caching_page_default));
3987			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3988			       &caching_page_default,
3989			       sizeof(caching_page_default));
3990			page_index->page_data =
3991				(uint8_t *)lun->mode_pages.caching_page;
3992			break;
3993		}
3994		case SMS_CONTROL_MODE_PAGE: {
3995
3996			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3997				panic("invalid subpage value %d",
3998				      page_index->subpage);
3999
4000			/*
4001			 * Defaults should be okay here, no calculations
4002			 * needed.
4003			 */
4004			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4005			       &control_page_default,
4006			       sizeof(control_page_default));
4007			memcpy(&lun->mode_pages.control_page[
4008			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4009			       sizeof(control_page_changeable));
4010			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4011			       &control_page_default,
4012			       sizeof(control_page_default));
4013			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4014			       &control_page_default,
4015			       sizeof(control_page_default));
4016			page_index->page_data =
4017				(uint8_t *)lun->mode_pages.control_page;
4018			break;
4019
4020		}
4021		case SMS_VENDOR_SPECIFIC_PAGE:{
4022			switch (page_index->subpage) {
4023			case PWR_SUBPAGE_CODE: {
4024				struct copan_power_subpage *current_page,
4025							   *saved_page;
4026
4027				memcpy(&lun->mode_pages.power_subpage[
4028				       CTL_PAGE_CURRENT],
4029				       &power_page_default,
4030				       sizeof(power_page_default));
4031				memcpy(&lun->mode_pages.power_subpage[
4032				       CTL_PAGE_CHANGEABLE],
4033				       &power_page_changeable,
4034				       sizeof(power_page_changeable));
4035				memcpy(&lun->mode_pages.power_subpage[
4036				       CTL_PAGE_DEFAULT],
4037				       &power_page_default,
4038				       sizeof(power_page_default));
4039				memcpy(&lun->mode_pages.power_subpage[
4040				       CTL_PAGE_SAVED],
4041				       &power_page_default,
4042				       sizeof(power_page_default));
4043				page_index->page_data =
4044				    (uint8_t *)lun->mode_pages.power_subpage;
4045
4046				current_page = (struct copan_power_subpage *)
4047					(page_index->page_data +
4048					 (page_index->page_len *
4049					  CTL_PAGE_CURRENT));
4050			        saved_page = (struct copan_power_subpage *)
4051				        (page_index->page_data +
4052					 (page_index->page_len *
4053					  CTL_PAGE_SAVED));
4054				break;
4055			}
4056			case APS_SUBPAGE_CODE: {
4057				struct copan_aps_subpage *current_page,
4058							 *saved_page;
4059
4060				// This gets set multiple times but
4061				// it should always be the same. It's
4062				// only done during init so who cares.
4063				index_to_aps_page = i;
4064
4065				memcpy(&lun->mode_pages.aps_subpage[
4066				       CTL_PAGE_CURRENT],
4067				       &aps_page_default,
4068				       sizeof(aps_page_default));
4069				memcpy(&lun->mode_pages.aps_subpage[
4070				       CTL_PAGE_CHANGEABLE],
4071				       &aps_page_changeable,
4072				       sizeof(aps_page_changeable));
4073				memcpy(&lun->mode_pages.aps_subpage[
4074				       CTL_PAGE_DEFAULT],
4075				       &aps_page_default,
4076				       sizeof(aps_page_default));
4077				memcpy(&lun->mode_pages.aps_subpage[
4078				       CTL_PAGE_SAVED],
4079				       &aps_page_default,
4080				       sizeof(aps_page_default));
4081				page_index->page_data =
4082					(uint8_t *)lun->mode_pages.aps_subpage;
4083
4084				current_page = (struct copan_aps_subpage *)
4085					(page_index->page_data +
4086					 (page_index->page_len *
4087					  CTL_PAGE_CURRENT));
4088				saved_page = (struct copan_aps_subpage *)
4089					(page_index->page_data +
4090					 (page_index->page_len *
4091					  CTL_PAGE_SAVED));
4092				break;
4093			}
4094			case DBGCNF_SUBPAGE_CODE: {
4095				struct copan_debugconf_subpage *current_page,
4096							       *saved_page;
4097
4098				memcpy(&lun->mode_pages.debugconf_subpage[
4099				       CTL_PAGE_CURRENT],
4100				       &debugconf_page_default,
4101				       sizeof(debugconf_page_default));
4102				memcpy(&lun->mode_pages.debugconf_subpage[
4103				       CTL_PAGE_CHANGEABLE],
4104				       &debugconf_page_changeable,
4105				       sizeof(debugconf_page_changeable));
4106				memcpy(&lun->mode_pages.debugconf_subpage[
4107				       CTL_PAGE_DEFAULT],
4108				       &debugconf_page_default,
4109				       sizeof(debugconf_page_default));
4110				memcpy(&lun->mode_pages.debugconf_subpage[
4111				       CTL_PAGE_SAVED],
4112				       &debugconf_page_default,
4113				       sizeof(debugconf_page_default));
4114				page_index->page_data =
4115					(uint8_t *)lun->mode_pages.debugconf_subpage;
4116
4117				current_page = (struct copan_debugconf_subpage *)
4118					(page_index->page_data +
4119					 (page_index->page_len *
4120					  CTL_PAGE_CURRENT));
4121				saved_page = (struct copan_debugconf_subpage *)
4122					(page_index->page_data +
4123					 (page_index->page_len *
4124					  CTL_PAGE_SAVED));
4125				break;
4126			}
4127			default:
4128				panic("invalid subpage value %d",
4129				      page_index->subpage);
4130				break;
4131			}
4132   			break;
4133		}
4134		default:
4135			panic("invalid page value %d",
4136			      page_index->page_code & SMPH_PC_MASK);
4137			break;
4138    	}
4139	}
4140
4141	return (CTL_RETVAL_COMPLETE);
4142}
4143
4144/*
4145 * LUN allocation.
4146 *
4147 * Requirements:
4148 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4149 *   wants us to allocate the LUN and he can block.
4150 * - ctl_softc is always set
4151 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4152 *
4153 * Returns 0 for success, non-zero (errno) for failure.
4154 */
4155static int
4156ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4157	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4158{
4159	struct ctl_lun *nlun, *lun;
4160	struct ctl_frontend *fe;
4161	int lun_number, i, lun_malloced;
4162
4163	if (be_lun == NULL)
4164		return (EINVAL);
4165
4166	/*
4167	 * We currently only support Direct Access or Processor LUN types.
4168	 */
4169	switch (be_lun->lun_type) {
4170	case T_DIRECT:
4171		break;
4172	case T_PROCESSOR:
4173		break;
4174	case T_SEQUENTIAL:
4175	case T_CHANGER:
4176	default:
4177		be_lun->lun_config_status(be_lun->be_lun,
4178					  CTL_LUN_CONFIG_FAILURE);
4179		break;
4180	}
4181	if (ctl_lun == NULL) {
4182		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4183		lun_malloced = 1;
4184	} else {
4185		lun_malloced = 0;
4186		lun = ctl_lun;
4187	}
4188
4189	memset(lun, 0, sizeof(*lun));
4190	if (lun_malloced)
4191		lun->flags = CTL_LUN_MALLOCED;
4192
4193	mtx_lock(&ctl_softc->ctl_lock);
4194	/*
4195	 * See if the caller requested a particular LUN number.  If so, see
4196	 * if it is available.  Otherwise, allocate the first available LUN.
4197	 */
4198	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4199		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4200		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4201			mtx_unlock(&ctl_softc->ctl_lock);
4202			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4203				printf("ctl: requested LUN ID %d is higher "
4204				       "than CTL_MAX_LUNS - 1 (%d)\n",
4205				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4206			} else {
4207				/*
4208				 * XXX KDM return an error, or just assign
4209				 * another LUN ID in this case??
4210				 */
4211				printf("ctl: requested LUN ID %d is already "
4212				       "in use\n", be_lun->req_lun_id);
4213			}
4214			if (lun->flags & CTL_LUN_MALLOCED)
4215				free(lun, M_CTL);
4216			be_lun->lun_config_status(be_lun->be_lun,
4217						  CTL_LUN_CONFIG_FAILURE);
4218			return (ENOSPC);
4219		}
4220		lun_number = be_lun->req_lun_id;
4221	} else {
4222		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4223		if (lun_number == -1) {
4224			mtx_unlock(&ctl_softc->ctl_lock);
4225			printf("ctl: can't allocate LUN on target %ju, out of "
4226			       "LUNs\n", (uintmax_t)target_id.id);
4227			if (lun->flags & CTL_LUN_MALLOCED)
4228				free(lun, M_CTL);
4229			be_lun->lun_config_status(be_lun->be_lun,
4230						  CTL_LUN_CONFIG_FAILURE);
4231			return (ENOSPC);
4232		}
4233	}
4234	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4235
4236	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4237	lun->target = target_id;
4238	lun->lun = lun_number;
4239	lun->be_lun = be_lun;
4240	/*
4241	 * The processor LUN is always enabled.  Disk LUNs come on line
4242	 * disabled, and must be enabled by the backend.
4243	 */
4244	lun->flags |= CTL_LUN_DISABLED;
4245	lun->backend = be_lun->be;
4246	be_lun->ctl_lun = lun;
4247	be_lun->lun_id = lun_number;
4248	atomic_add_int(&be_lun->be->num_luns, 1);
4249	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4250		lun->flags |= CTL_LUN_STOPPED;
4251
4252	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4253		lun->flags |= CTL_LUN_INOPERABLE;
4254
4255	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4256		lun->flags |= CTL_LUN_PRIMARY_SC;
4257
4258	lun->ctl_softc = ctl_softc;
4259	TAILQ_INIT(&lun->ooa_queue);
4260	TAILQ_INIT(&lun->blocked_queue);
4261	STAILQ_INIT(&lun->error_list);
4262
4263	/*
4264	 * Initialize the mode page index.
4265	 */
4266	ctl_init_page_index(lun);
4267
4268	/*
4269	 * Set the poweron UA for all initiators on this LUN only.
4270	 */
4271	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4272		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4273
4274	/*
4275	 * Now, before we insert this lun on the lun list, set the lun
4276	 * inventory changed UA for all other luns.
4277	 */
4278	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4279		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4280			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4281		}
4282	}
4283
4284	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4285
4286	ctl_softc->ctl_luns[lun_number] = lun;
4287
4288	ctl_softc->num_luns++;
4289
4290	/* Setup statistics gathering */
4291	lun->stats.device_type = be_lun->lun_type;
4292	lun->stats.lun_number = lun_number;
4293	if (lun->stats.device_type == T_DIRECT)
4294		lun->stats.blocksize = be_lun->blocksize;
4295	else
4296		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4297	for (i = 0;i < CTL_MAX_PORTS;i++)
4298		lun->stats.ports[i].targ_port = i;
4299
4300	mtx_unlock(&ctl_softc->ctl_lock);
4301
4302	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4303
4304	/*
4305	 * Run through each registered FETD and bring it online if it isn't
4306	 * already.  Enable the target ID if it hasn't been enabled, and
4307	 * enable this particular LUN.
4308	 */
4309	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4310		int retval;
4311
4312		/*
4313		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4314		 * to do things differently if we go to a multiple target
4315		 * ID scheme.
4316		 */
4317		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4318
4319			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4320			if (retval != 0) {
4321				printf("ctl_alloc_lun: FETD %s port %d "
4322				       "returned error %d for targ_enable on "
4323				       "target %ju\n", fe->port_name,
4324				       fe->targ_port, retval,
4325				       (uintmax_t)target_id.id);
4326			} else
4327				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4328		}
4329
4330		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4331		if (retval != 0) {
4332			printf("ctl_alloc_lun: FETD %s port %d returned error "
4333			       "%d for lun_enable on target %ju lun %d\n",
4334			       fe->port_name, fe->targ_port, retval,
4335			       (uintmax_t)target_id.id, lun_number);
4336		} else
4337			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4338	}
4339	return (0);
4340}
4341
4342/*
4343 * Delete a LUN.
4344 * Assumptions:
4345 * - LUN has already been marked invalid and any pending I/O has been taken
4346 *   care of.
4347 */
4348static int
4349ctl_free_lun(struct ctl_lun *lun)
4350{
4351	struct ctl_softc *softc;
4352#if 0
4353	struct ctl_frontend *fe;
4354#endif
4355	struct ctl_lun *nlun;
4356	int i;
4357
4358	softc = lun->ctl_softc;
4359
4360	mtx_assert(&softc->ctl_lock, MA_OWNED);
4361
4362	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4363
4364	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4365
4366	softc->ctl_luns[lun->lun] = NULL;
4367
4368	if (!TAILQ_EMPTY(&lun->ooa_queue))
4369		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4370
4371	softc->num_luns--;
4372
4373	/*
4374	 * XXX KDM this scheme only works for a single target/multiple LUN
4375	 * setup.  It needs to be revamped for a multiple target scheme.
4376	 *
4377	 * XXX KDM this results in fe->lun_disable() getting called twice,
4378	 * once when ctl_disable_lun() is called, and a second time here.
4379	 * We really need to re-think the LUN disable semantics.  There
4380	 * should probably be several steps/levels to LUN removal:
4381	 *  - disable
4382	 *  - invalidate
4383	 *  - free
4384 	 *
4385	 * Right now we only have a disable method when communicating to
4386	 * the front end ports, at least for individual LUNs.
4387	 */
4388#if 0
4389	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4390		int retval;
4391
4392		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4393					 lun->lun);
4394		if (retval != 0) {
4395			printf("ctl_free_lun: FETD %s port %d returned error "
4396			       "%d for lun_disable on target %ju lun %jd\n",
4397			       fe->port_name, fe->targ_port, retval,
4398			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4399		}
4400
4401		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4402			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4403
4404			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4405			if (retval != 0) {
4406				printf("ctl_free_lun: FETD %s port %d "
4407				       "returned error %d for targ_disable on "
4408				       "target %ju\n", fe->port_name,
4409				       fe->targ_port, retval,
4410				       (uintmax_t)lun->target.id);
4411			} else
4412				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4413
4414			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4415				continue;
4416
4417#if 0
4418			fe->port_offline(fe->onoff_arg);
4419			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4420#endif
4421		}
4422	}
4423#endif
4424
4425	/*
4426	 * Tell the backend to free resources, if this LUN has a backend.
4427	 */
4428	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4429	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4430
4431	mtx_destroy(&lun->lun_lock);
4432	if (lun->flags & CTL_LUN_MALLOCED)
4433		free(lun, M_CTL);
4434
4435	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4436		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4437			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4438		}
4439	}
4440
4441	return (0);
4442}
4443
4444static void
4445ctl_create_lun(struct ctl_be_lun *be_lun)
4446{
4447	struct ctl_softc *ctl_softc;
4448
4449	ctl_softc = control_softc;
4450
4451	/*
4452	 * ctl_alloc_lun() should handle all potential failure cases.
4453	 */
4454	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4455}
4456
4457int
4458ctl_add_lun(struct ctl_be_lun *be_lun)
4459{
4460	struct ctl_softc *ctl_softc = control_softc;
4461
4462	mtx_lock(&ctl_softc->ctl_lock);
4463	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4464	mtx_unlock(&ctl_softc->ctl_lock);
4465	wakeup(&ctl_softc->pending_lun_queue);
4466
4467	return (0);
4468}
4469
4470int
4471ctl_enable_lun(struct ctl_be_lun *be_lun)
4472{
4473	struct ctl_softc *ctl_softc;
4474	struct ctl_frontend *fe, *nfe;
4475	struct ctl_lun *lun;
4476	int retval;
4477
4478	ctl_softc = control_softc;
4479
4480	lun = (struct ctl_lun *)be_lun->ctl_lun;
4481
4482	mtx_lock(&ctl_softc->ctl_lock);
4483	mtx_lock(&lun->lun_lock);
4484	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4485		/*
4486		 * eh?  Why did we get called if the LUN is already
4487		 * enabled?
4488		 */
4489		mtx_unlock(&lun->lun_lock);
4490		mtx_unlock(&ctl_softc->ctl_lock);
4491		return (0);
4492	}
4493	lun->flags &= ~CTL_LUN_DISABLED;
4494	mtx_unlock(&lun->lun_lock);
4495
4496	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4497		nfe = STAILQ_NEXT(fe, links);
4498
4499		/*
4500		 * Drop the lock while we call the FETD's enable routine.
4501		 * This can lead to a callback into CTL (at least in the
4502		 * case of the internal initiator frontend.
4503		 */
4504		mtx_unlock(&ctl_softc->ctl_lock);
4505		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4506		mtx_lock(&ctl_softc->ctl_lock);
4507		if (retval != 0) {
4508			printf("%s: FETD %s port %d returned error "
4509			       "%d for lun_enable on target %ju lun %jd\n",
4510			       __func__, fe->port_name, fe->targ_port, retval,
4511			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4512		}
4513#if 0
4514		 else {
4515            /* NOTE:  TODO:  why does lun enable affect port status? */
4516			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4517		}
4518#endif
4519	}
4520
4521	mtx_unlock(&ctl_softc->ctl_lock);
4522
4523	return (0);
4524}
4525
4526int
4527ctl_disable_lun(struct ctl_be_lun *be_lun)
4528{
4529	struct ctl_softc *ctl_softc;
4530	struct ctl_frontend *fe;
4531	struct ctl_lun *lun;
4532	int retval;
4533
4534	ctl_softc = control_softc;
4535
4536	lun = (struct ctl_lun *)be_lun->ctl_lun;
4537
4538	mtx_lock(&ctl_softc->ctl_lock);
4539	mtx_lock(&lun->lun_lock);
4540	if (lun->flags & CTL_LUN_DISABLED) {
4541		mtx_unlock(&lun->lun_lock);
4542		mtx_unlock(&ctl_softc->ctl_lock);
4543		return (0);
4544	}
4545	lun->flags |= CTL_LUN_DISABLED;
4546	mtx_unlock(&lun->lun_lock);
4547
4548	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4549		mtx_unlock(&ctl_softc->ctl_lock);
4550		/*
4551		 * Drop the lock before we call the frontend's disable
4552		 * routine, to avoid lock order reversals.
4553		 *
4554		 * XXX KDM what happens if the frontend list changes while
4555		 * we're traversing it?  It's unlikely, but should be handled.
4556		 */
4557		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4558					 lun->lun);
4559		mtx_lock(&ctl_softc->ctl_lock);
4560		if (retval != 0) {
4561			printf("ctl_alloc_lun: FETD %s port %d returned error "
4562			       "%d for lun_disable on target %ju lun %jd\n",
4563			       fe->port_name, fe->targ_port, retval,
4564			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4565		}
4566	}
4567
4568	mtx_unlock(&ctl_softc->ctl_lock);
4569
4570	return (0);
4571}
4572
4573int
4574ctl_start_lun(struct ctl_be_lun *be_lun)
4575{
4576	struct ctl_softc *ctl_softc;
4577	struct ctl_lun *lun;
4578
4579	ctl_softc = control_softc;
4580
4581	lun = (struct ctl_lun *)be_lun->ctl_lun;
4582
4583	mtx_lock(&lun->lun_lock);
4584	lun->flags &= ~CTL_LUN_STOPPED;
4585	mtx_unlock(&lun->lun_lock);
4586
4587	return (0);
4588}
4589
4590int
4591ctl_stop_lun(struct ctl_be_lun *be_lun)
4592{
4593	struct ctl_softc *ctl_softc;
4594	struct ctl_lun *lun;
4595
4596	ctl_softc = control_softc;
4597
4598	lun = (struct ctl_lun *)be_lun->ctl_lun;
4599
4600	mtx_lock(&lun->lun_lock);
4601	lun->flags |= CTL_LUN_STOPPED;
4602	mtx_unlock(&lun->lun_lock);
4603
4604	return (0);
4605}
4606
4607int
4608ctl_lun_offline(struct ctl_be_lun *be_lun)
4609{
4610	struct ctl_softc *ctl_softc;
4611	struct ctl_lun *lun;
4612
4613	ctl_softc = control_softc;
4614
4615	lun = (struct ctl_lun *)be_lun->ctl_lun;
4616
4617	mtx_lock(&lun->lun_lock);
4618	lun->flags |= CTL_LUN_OFFLINE;
4619	mtx_unlock(&lun->lun_lock);
4620
4621	return (0);
4622}
4623
4624int
4625ctl_lun_online(struct ctl_be_lun *be_lun)
4626{
4627	struct ctl_softc *ctl_softc;
4628	struct ctl_lun *lun;
4629
4630	ctl_softc = control_softc;
4631
4632	lun = (struct ctl_lun *)be_lun->ctl_lun;
4633
4634	mtx_lock(&lun->lun_lock);
4635	lun->flags &= ~CTL_LUN_OFFLINE;
4636	mtx_unlock(&lun->lun_lock);
4637
4638	return (0);
4639}
4640
4641int
4642ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4643{
4644	struct ctl_softc *ctl_softc;
4645	struct ctl_lun *lun;
4646
4647	ctl_softc = control_softc;
4648
4649	lun = (struct ctl_lun *)be_lun->ctl_lun;
4650
4651	mtx_lock(&lun->lun_lock);
4652
4653	/*
4654	 * The LUN needs to be disabled before it can be marked invalid.
4655	 */
4656	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4657		mtx_unlock(&lun->lun_lock);
4658		return (-1);
4659	}
4660	/*
4661	 * Mark the LUN invalid.
4662	 */
4663	lun->flags |= CTL_LUN_INVALID;
4664
4665	/*
4666	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4667	 * If we have something in the OOA queue, we'll free it when the
4668	 * last I/O completes.
4669	 */
4670	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4671		mtx_unlock(&lun->lun_lock);
4672		mtx_lock(&ctl_softc->ctl_lock);
4673		ctl_free_lun(lun);
4674		mtx_unlock(&ctl_softc->ctl_lock);
4675	} else
4676		mtx_unlock(&lun->lun_lock);
4677
4678	return (0);
4679}
4680
4681int
4682ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4683{
4684	struct ctl_softc *ctl_softc;
4685	struct ctl_lun *lun;
4686
4687	ctl_softc = control_softc;
4688	lun = (struct ctl_lun *)be_lun->ctl_lun;
4689
4690	mtx_lock(&lun->lun_lock);
4691	lun->flags |= CTL_LUN_INOPERABLE;
4692	mtx_unlock(&lun->lun_lock);
4693
4694	return (0);
4695}
4696
4697int
4698ctl_lun_operable(struct ctl_be_lun *be_lun)
4699{
4700	struct ctl_softc *ctl_softc;
4701	struct ctl_lun *lun;
4702
4703	ctl_softc = control_softc;
4704	lun = (struct ctl_lun *)be_lun->ctl_lun;
4705
4706	mtx_lock(&lun->lun_lock);
4707	lun->flags &= ~CTL_LUN_INOPERABLE;
4708	mtx_unlock(&lun->lun_lock);
4709
4710	return (0);
4711}
4712
4713int
4714ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4715		   int lock)
4716{
4717	struct ctl_softc *softc;
4718	struct ctl_lun *lun;
4719	struct copan_aps_subpage *current_sp;
4720	struct ctl_page_index *page_index;
4721	int i;
4722
4723	softc = control_softc;
4724
4725	mtx_lock(&softc->ctl_lock);
4726
4727	lun = (struct ctl_lun *)be_lun->ctl_lun;
4728	mtx_lock(&lun->lun_lock);
4729
4730	page_index = NULL;
4731	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4732		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4733		     APS_PAGE_CODE)
4734			continue;
4735
4736		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4737			continue;
4738		page_index = &lun->mode_pages.index[i];
4739	}
4740
4741	if (page_index == NULL) {
4742		mtx_unlock(&lun->lun_lock);
4743		mtx_unlock(&softc->ctl_lock);
4744		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4745		       (uintmax_t)lun->lun);
4746		return (1);
4747	}
4748#if 0
4749	if ((softc->aps_locked_lun != 0)
4750	 && (softc->aps_locked_lun != lun->lun)) {
4751		printf("%s: attempt to lock LUN %llu when %llu is already "
4752		       "locked\n");
4753		mtx_unlock(&lun->lun_lock);
4754		mtx_unlock(&softc->ctl_lock);
4755		return (1);
4756	}
4757#endif
4758
4759	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4760		(page_index->page_len * CTL_PAGE_CURRENT));
4761
4762	if (lock != 0) {
4763		current_sp->lock_active = APS_LOCK_ACTIVE;
4764		softc->aps_locked_lun = lun->lun;
4765	} else {
4766		current_sp->lock_active = 0;
4767		softc->aps_locked_lun = 0;
4768	}
4769
4770
4771	/*
4772	 * If we're in HA mode, try to send the lock message to the other
4773	 * side.
4774	 */
4775	if (ctl_is_single == 0) {
4776		int isc_retval;
4777		union ctl_ha_msg lock_msg;
4778
4779		lock_msg.hdr.nexus = *nexus;
4780		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4781		if (lock != 0)
4782			lock_msg.aps.lock_flag = 1;
4783		else
4784			lock_msg.aps.lock_flag = 0;
4785		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4786					 sizeof(lock_msg), 0);
4787		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4788			printf("%s: APS (lock=%d) error returned from "
4789			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4790			mtx_unlock(&lun->lun_lock);
4791			mtx_unlock(&softc->ctl_lock);
4792			return (1);
4793		}
4794	}
4795
4796	mtx_unlock(&lun->lun_lock);
4797	mtx_unlock(&softc->ctl_lock);
4798
4799	return (0);
4800}
4801
4802void
4803ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4804{
4805	struct ctl_lun *lun;
4806	struct ctl_softc *softc;
4807	int i;
4808
4809	softc = control_softc;
4810
4811	lun = (struct ctl_lun *)be_lun->ctl_lun;
4812
4813	mtx_lock(&lun->lun_lock);
4814
4815	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4816		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4817
4818	mtx_unlock(&lun->lun_lock);
4819}
4820
4821/*
4822 * Backend "memory move is complete" callback for requests that never
4823 * make it down to say RAIDCore's configuration code.
4824 */
4825int
4826ctl_config_move_done(union ctl_io *io)
4827{
4828	int retval;
4829
4830	retval = CTL_RETVAL_COMPLETE;
4831
4832
4833	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4834	/*
4835	 * XXX KDM this shouldn't happen, but what if it does?
4836	 */
4837	if (io->io_hdr.io_type != CTL_IO_SCSI)
4838		panic("I/O type isn't CTL_IO_SCSI!");
4839
4840	if ((io->io_hdr.port_status == 0)
4841	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4842	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4843		io->io_hdr.status = CTL_SUCCESS;
4844	else if ((io->io_hdr.port_status != 0)
4845	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4846	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4847		/*
4848		 * For hardware error sense keys, the sense key
4849		 * specific value is defined to be a retry count,
4850		 * but we use it to pass back an internal FETD
4851		 * error code.  XXX KDM  Hopefully the FETD is only
4852		 * using 16 bits for an error code, since that's
4853		 * all the space we have in the sks field.
4854		 */
4855		ctl_set_internal_failure(&io->scsiio,
4856					 /*sks_valid*/ 1,
4857					 /*retry_count*/
4858					 io->io_hdr.port_status);
4859		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4860			free(io->scsiio.kern_data_ptr, M_CTL);
4861		ctl_done(io);
4862		goto bailout;
4863	}
4864
4865	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4866	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4867	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4868		/*
4869		 * XXX KDM just assuming a single pointer here, and not a
4870		 * S/G list.  If we start using S/G lists for config data,
4871		 * we'll need to know how to clean them up here as well.
4872		 */
4873		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4874			free(io->scsiio.kern_data_ptr, M_CTL);
4875		/* Hopefully the user has already set the status... */
4876		ctl_done(io);
4877	} else {
4878		/*
4879		 * XXX KDM now we need to continue data movement.  Some
4880		 * options:
4881		 * - call ctl_scsiio() again?  We don't do this for data
4882		 *   writes, because for those at least we know ahead of
4883		 *   time where the write will go and how long it is.  For
4884		 *   config writes, though, that information is largely
4885		 *   contained within the write itself, thus we need to
4886		 *   parse out the data again.
4887		 *
4888		 * - Call some other function once the data is in?
4889		 */
4890
4891		/*
4892		 * XXX KDM call ctl_scsiio() again for now, and check flag
4893		 * bits to see whether we're allocated or not.
4894		 */
4895		retval = ctl_scsiio(&io->scsiio);
4896	}
4897bailout:
4898	return (retval);
4899}
4900
4901/*
4902 * This gets called by a backend driver when it is done with a
4903 * data_submit method.
4904 */
4905void
4906ctl_data_submit_done(union ctl_io *io)
4907{
4908	/*
4909	 * If the IO_CONT flag is set, we need to call the supplied
4910	 * function to continue processing the I/O, instead of completing
4911	 * the I/O just yet.
4912	 *
4913	 * If there is an error, though, we don't want to keep processing.
4914	 * Instead, just send status back to the initiator.
4915	 */
4916	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4917	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4918	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4919	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4920		io->scsiio.io_cont(io);
4921		return;
4922	}
4923	ctl_done(io);
4924}
4925
4926/*
4927 * This gets called by a backend driver when it is done with a
4928 * configuration write.
4929 */
4930void
4931ctl_config_write_done(union ctl_io *io)
4932{
4933	/*
4934	 * If the IO_CONT flag is set, we need to call the supplied
4935	 * function to continue processing the I/O, instead of completing
4936	 * the I/O just yet.
4937	 *
4938	 * If there is an error, though, we don't want to keep processing.
4939	 * Instead, just send status back to the initiator.
4940	 */
4941	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4942	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4943	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4944		io->scsiio.io_cont(io);
4945		return;
4946	}
4947	/*
4948	 * Since a configuration write can be done for commands that actually
4949	 * have data allocated, like write buffer, and commands that have
4950	 * no data, like start/stop unit, we need to check here.
4951	 */
4952	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4953		free(io->scsiio.kern_data_ptr, M_CTL);
4954	ctl_done(io);
4955}
4956
4957/*
4958 * SCSI release command.
4959 */
4960int
4961ctl_scsi_release(struct ctl_scsiio *ctsio)
4962{
4963	int length, longid, thirdparty_id, resv_id;
4964	struct ctl_softc *ctl_softc;
4965	struct ctl_lun *lun;
4966
4967	length = 0;
4968	resv_id = 0;
4969
4970	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4971
4972	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4973	ctl_softc = control_softc;
4974
4975	switch (ctsio->cdb[0]) {
4976	case RELEASE_10: {
4977		struct scsi_release_10 *cdb;
4978
4979		cdb = (struct scsi_release_10 *)ctsio->cdb;
4980
4981		if (cdb->byte2 & SR10_LONGID)
4982			longid = 1;
4983		else
4984			thirdparty_id = cdb->thirdparty_id;
4985
4986		resv_id = cdb->resv_id;
4987		length = scsi_2btoul(cdb->length);
4988		break;
4989	}
4990	}
4991
4992
4993	/*
4994	 * XXX KDM right now, we only support LUN reservation.  We don't
4995	 * support 3rd party reservations, or extent reservations, which
4996	 * might actually need the parameter list.  If we've gotten this
4997	 * far, we've got a LUN reservation.  Anything else got kicked out
4998	 * above.  So, according to SPC, ignore the length.
4999	 */
5000	length = 0;
5001
5002	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5003	 && (length > 0)) {
5004		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5005		ctsio->kern_data_len = length;
5006		ctsio->kern_total_len = length;
5007		ctsio->kern_data_resid = 0;
5008		ctsio->kern_rel_offset = 0;
5009		ctsio->kern_sg_entries = 0;
5010		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5011		ctsio->be_move_done = ctl_config_move_done;
5012		ctl_datamove((union ctl_io *)ctsio);
5013
5014		return (CTL_RETVAL_COMPLETE);
5015	}
5016
5017	if (length > 0)
5018		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5019
5020	mtx_lock(&lun->lun_lock);
5021
5022	/*
5023	 * According to SPC, it is not an error for an intiator to attempt
5024	 * to release a reservation on a LUN that isn't reserved, or that
5025	 * is reserved by another initiator.  The reservation can only be
5026	 * released, though, by the initiator who made it or by one of
5027	 * several reset type events.
5028	 */
5029	if (lun->flags & CTL_LUN_RESERVED) {
5030		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5031		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5032		 && (ctsio->io_hdr.nexus.targ_target.id ==
5033		     lun->rsv_nexus.targ_target.id)) {
5034			lun->flags &= ~CTL_LUN_RESERVED;
5035		}
5036	}
5037
5038	mtx_unlock(&lun->lun_lock);
5039
5040	ctsio->scsi_status = SCSI_STATUS_OK;
5041	ctsio->io_hdr.status = CTL_SUCCESS;
5042
5043	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5044		free(ctsio->kern_data_ptr, M_CTL);
5045		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5046	}
5047
5048	ctl_done((union ctl_io *)ctsio);
5049	return (CTL_RETVAL_COMPLETE);
5050}
5051
5052int
5053ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5054{
5055	int extent, thirdparty, longid;
5056	int resv_id, length;
5057	uint64_t thirdparty_id;
5058	struct ctl_softc *ctl_softc;
5059	struct ctl_lun *lun;
5060
5061	extent = 0;
5062	thirdparty = 0;
5063	longid = 0;
5064	resv_id = 0;
5065	length = 0;
5066	thirdparty_id = 0;
5067
5068	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5069
5070	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5071	ctl_softc = control_softc;
5072
5073	switch (ctsio->cdb[0]) {
5074	case RESERVE_10: {
5075		struct scsi_reserve_10 *cdb;
5076
5077		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5078
5079		if (cdb->byte2 & SR10_LONGID)
5080			longid = 1;
5081		else
5082			thirdparty_id = cdb->thirdparty_id;
5083
5084		resv_id = cdb->resv_id;
5085		length = scsi_2btoul(cdb->length);
5086		break;
5087	}
5088	}
5089
5090	/*
5091	 * XXX KDM right now, we only support LUN reservation.  We don't
5092	 * support 3rd party reservations, or extent reservations, which
5093	 * might actually need the parameter list.  If we've gotten this
5094	 * far, we've got a LUN reservation.  Anything else got kicked out
5095	 * above.  So, according to SPC, ignore the length.
5096	 */
5097	length = 0;
5098
5099	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5100	 && (length > 0)) {
5101		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5102		ctsio->kern_data_len = length;
5103		ctsio->kern_total_len = length;
5104		ctsio->kern_data_resid = 0;
5105		ctsio->kern_rel_offset = 0;
5106		ctsio->kern_sg_entries = 0;
5107		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5108		ctsio->be_move_done = ctl_config_move_done;
5109		ctl_datamove((union ctl_io *)ctsio);
5110
5111		return (CTL_RETVAL_COMPLETE);
5112	}
5113
5114	if (length > 0)
5115		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5116
5117	mtx_lock(&lun->lun_lock);
5118	if (lun->flags & CTL_LUN_RESERVED) {
5119		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5120		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5121		 || (ctsio->io_hdr.nexus.targ_target.id !=
5122		     lun->rsv_nexus.targ_target.id)) {
5123			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5124			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5125			goto bailout;
5126		}
5127	}
5128
5129	lun->flags |= CTL_LUN_RESERVED;
5130	lun->rsv_nexus = ctsio->io_hdr.nexus;
5131
5132	ctsio->scsi_status = SCSI_STATUS_OK;
5133	ctsio->io_hdr.status = CTL_SUCCESS;
5134
5135bailout:
5136	mtx_unlock(&lun->lun_lock);
5137
5138	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5139		free(ctsio->kern_data_ptr, M_CTL);
5140		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5141	}
5142
5143	ctl_done((union ctl_io *)ctsio);
5144	return (CTL_RETVAL_COMPLETE);
5145}
5146
5147int
5148ctl_start_stop(struct ctl_scsiio *ctsio)
5149{
5150	struct scsi_start_stop_unit *cdb;
5151	struct ctl_lun *lun;
5152	struct ctl_softc *ctl_softc;
5153	int retval;
5154
5155	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5156
5157	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5158	ctl_softc = control_softc;
5159	retval = 0;
5160
5161	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5162
5163	/*
5164	 * XXX KDM
5165	 * We don't support the immediate bit on a stop unit.  In order to
5166	 * do that, we would need to code up a way to know that a stop is
5167	 * pending, and hold off any new commands until it completes, one
5168	 * way or another.  Then we could accept or reject those commands
5169	 * depending on its status.  We would almost need to do the reverse
5170	 * of what we do below for an immediate start -- return the copy of
5171	 * the ctl_io to the FETD with status to send to the host (and to
5172	 * free the copy!) and then free the original I/O once the stop
5173	 * actually completes.  That way, the OOA queue mechanism can work
5174	 * to block commands that shouldn't proceed.  Another alternative
5175	 * would be to put the copy in the queue in place of the original,
5176	 * and return the original back to the caller.  That could be
5177	 * slightly safer..
5178	 */
5179	if ((cdb->byte2 & SSS_IMMED)
5180	 && ((cdb->how & SSS_START) == 0)) {
5181		ctl_set_invalid_field(ctsio,
5182				      /*sks_valid*/ 1,
5183				      /*command*/ 1,
5184				      /*field*/ 1,
5185				      /*bit_valid*/ 1,
5186				      /*bit*/ 0);
5187		ctl_done((union ctl_io *)ctsio);
5188		return (CTL_RETVAL_COMPLETE);
5189	}
5190
5191	if ((lun->flags & CTL_LUN_PR_RESERVED)
5192	 && ((cdb->how & SSS_START)==0)) {
5193		uint32_t residx;
5194
5195		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5196		if (!lun->per_res[residx].registered
5197		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5198
5199			ctl_set_reservation_conflict(ctsio);
5200			ctl_done((union ctl_io *)ctsio);
5201			return (CTL_RETVAL_COMPLETE);
5202		}
5203	}
5204
5205	/*
5206	 * If there is no backend on this device, we can't start or stop
5207	 * it.  In theory we shouldn't get any start/stop commands in the
5208	 * first place at this level if the LUN doesn't have a backend.
5209	 * That should get stopped by the command decode code.
5210	 */
5211	if (lun->backend == NULL) {
5212		ctl_set_invalid_opcode(ctsio);
5213		ctl_done((union ctl_io *)ctsio);
5214		return (CTL_RETVAL_COMPLETE);
5215	}
5216
5217	/*
5218	 * XXX KDM Copan-specific offline behavior.
5219	 * Figure out a reasonable way to port this?
5220	 */
5221#ifdef NEEDTOPORT
5222	mtx_lock(&lun->lun_lock);
5223
5224	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5225	 && (lun->flags & CTL_LUN_OFFLINE)) {
5226		/*
5227		 * If the LUN is offline, and the on/offline bit isn't set,
5228		 * reject the start or stop.  Otherwise, let it through.
5229		 */
5230		mtx_unlock(&lun->lun_lock);
5231		ctl_set_lun_not_ready(ctsio);
5232		ctl_done((union ctl_io *)ctsio);
5233	} else {
5234		mtx_unlock(&lun->lun_lock);
5235#endif /* NEEDTOPORT */
5236		/*
5237		 * This could be a start or a stop when we're online,
5238		 * or a stop/offline or start/online.  A start or stop when
5239		 * we're offline is covered in the case above.
5240		 */
5241		/*
5242		 * In the non-immediate case, we send the request to
5243		 * the backend and return status to the user when
5244		 * it is done.
5245		 *
5246		 * In the immediate case, we allocate a new ctl_io
5247		 * to hold a copy of the request, and send that to
5248		 * the backend.  We then set good status on the
5249		 * user's request and return it immediately.
5250		 */
5251		if (cdb->byte2 & SSS_IMMED) {
5252			union ctl_io *new_io;
5253
5254			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5255			if (new_io == NULL) {
5256				ctl_set_busy(ctsio);
5257				ctl_done((union ctl_io *)ctsio);
5258			} else {
5259				ctl_copy_io((union ctl_io *)ctsio,
5260					    new_io);
5261				retval = lun->backend->config_write(new_io);
5262				ctl_set_success(ctsio);
5263				ctl_done((union ctl_io *)ctsio);
5264			}
5265		} else {
5266			retval = lun->backend->config_write(
5267				(union ctl_io *)ctsio);
5268		}
5269#ifdef NEEDTOPORT
5270	}
5271#endif
5272	return (retval);
5273}
5274
5275/*
5276 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5277 * we don't really do anything with the LBA and length fields if the user
5278 * passes them in.  Instead we'll just flush out the cache for the entire
5279 * LUN.
5280 */
5281int
5282ctl_sync_cache(struct ctl_scsiio *ctsio)
5283{
5284	struct ctl_lun *lun;
5285	struct ctl_softc *ctl_softc;
5286	uint64_t starting_lba;
5287	uint32_t block_count;
5288	int retval;
5289
5290	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5291
5292	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5293	ctl_softc = control_softc;
5294	retval = 0;
5295
5296	switch (ctsio->cdb[0]) {
5297	case SYNCHRONIZE_CACHE: {
5298		struct scsi_sync_cache *cdb;
5299		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5300
5301		starting_lba = scsi_4btoul(cdb->begin_lba);
5302		block_count = scsi_2btoul(cdb->lb_count);
5303		break;
5304	}
5305	case SYNCHRONIZE_CACHE_16: {
5306		struct scsi_sync_cache_16 *cdb;
5307		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5308
5309		starting_lba = scsi_8btou64(cdb->begin_lba);
5310		block_count = scsi_4btoul(cdb->lb_count);
5311		break;
5312	}
5313	default:
5314		ctl_set_invalid_opcode(ctsio);
5315		ctl_done((union ctl_io *)ctsio);
5316		goto bailout;
5317		break; /* NOTREACHED */
5318	}
5319
5320	/*
5321	 * We check the LBA and length, but don't do anything with them.
5322	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5323	 * get flushed.  This check will just help satisfy anyone who wants
5324	 * to see an error for an out of range LBA.
5325	 */
5326	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5327		ctl_set_lba_out_of_range(ctsio);
5328		ctl_done((union ctl_io *)ctsio);
5329		goto bailout;
5330	}
5331
5332	/*
5333	 * If this LUN has no backend, we can't flush the cache anyway.
5334	 */
5335	if (lun->backend == NULL) {
5336		ctl_set_invalid_opcode(ctsio);
5337		ctl_done((union ctl_io *)ctsio);
5338		goto bailout;
5339	}
5340
5341	/*
5342	 * Check to see whether we're configured to send the SYNCHRONIZE
5343	 * CACHE command directly to the back end.
5344	 */
5345	mtx_lock(&lun->lun_lock);
5346	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5347	 && (++(lun->sync_count) >= lun->sync_interval)) {
5348		lun->sync_count = 0;
5349		mtx_unlock(&lun->lun_lock);
5350		retval = lun->backend->config_write((union ctl_io *)ctsio);
5351	} else {
5352		mtx_unlock(&lun->lun_lock);
5353		ctl_set_success(ctsio);
5354		ctl_done((union ctl_io *)ctsio);
5355	}
5356
5357bailout:
5358
5359	return (retval);
5360}
5361
5362int
5363ctl_format(struct ctl_scsiio *ctsio)
5364{
5365	struct scsi_format *cdb;
5366	struct ctl_lun *lun;
5367	struct ctl_softc *ctl_softc;
5368	int length, defect_list_len;
5369
5370	CTL_DEBUG_PRINT(("ctl_format\n"));
5371
5372	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5373	ctl_softc = control_softc;
5374
5375	cdb = (struct scsi_format *)ctsio->cdb;
5376
5377	length = 0;
5378	if (cdb->byte2 & SF_FMTDATA) {
5379		if (cdb->byte2 & SF_LONGLIST)
5380			length = sizeof(struct scsi_format_header_long);
5381		else
5382			length = sizeof(struct scsi_format_header_short);
5383	}
5384
5385	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5386	 && (length > 0)) {
5387		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5388		ctsio->kern_data_len = length;
5389		ctsio->kern_total_len = length;
5390		ctsio->kern_data_resid = 0;
5391		ctsio->kern_rel_offset = 0;
5392		ctsio->kern_sg_entries = 0;
5393		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5394		ctsio->be_move_done = ctl_config_move_done;
5395		ctl_datamove((union ctl_io *)ctsio);
5396
5397		return (CTL_RETVAL_COMPLETE);
5398	}
5399
5400	defect_list_len = 0;
5401
5402	if (cdb->byte2 & SF_FMTDATA) {
5403		if (cdb->byte2 & SF_LONGLIST) {
5404			struct scsi_format_header_long *header;
5405
5406			header = (struct scsi_format_header_long *)
5407				ctsio->kern_data_ptr;
5408
5409			defect_list_len = scsi_4btoul(header->defect_list_len);
5410			if (defect_list_len != 0) {
5411				ctl_set_invalid_field(ctsio,
5412						      /*sks_valid*/ 1,
5413						      /*command*/ 0,
5414						      /*field*/ 2,
5415						      /*bit_valid*/ 0,
5416						      /*bit*/ 0);
5417				goto bailout;
5418			}
5419		} else {
5420			struct scsi_format_header_short *header;
5421
5422			header = (struct scsi_format_header_short *)
5423				ctsio->kern_data_ptr;
5424
5425			defect_list_len = scsi_2btoul(header->defect_list_len);
5426			if (defect_list_len != 0) {
5427				ctl_set_invalid_field(ctsio,
5428						      /*sks_valid*/ 1,
5429						      /*command*/ 0,
5430						      /*field*/ 2,
5431						      /*bit_valid*/ 0,
5432						      /*bit*/ 0);
5433				goto bailout;
5434			}
5435		}
5436	}
5437
5438	/*
5439	 * The format command will clear out the "Medium format corrupted"
5440	 * status if set by the configuration code.  That status is really
5441	 * just a way to notify the host that we have lost the media, and
5442	 * get them to issue a command that will basically make them think
5443	 * they're blowing away the media.
5444	 */
5445	mtx_lock(&lun->lun_lock);
5446	lun->flags &= ~CTL_LUN_INOPERABLE;
5447	mtx_unlock(&lun->lun_lock);
5448
5449	ctsio->scsi_status = SCSI_STATUS_OK;
5450	ctsio->io_hdr.status = CTL_SUCCESS;
5451bailout:
5452
5453	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5454		free(ctsio->kern_data_ptr, M_CTL);
5455		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5456	}
5457
5458	ctl_done((union ctl_io *)ctsio);
5459	return (CTL_RETVAL_COMPLETE);
5460}
5461
5462int
5463ctl_read_buffer(struct ctl_scsiio *ctsio)
5464{
5465	struct scsi_read_buffer *cdb;
5466	struct ctl_lun *lun;
5467	int buffer_offset, len;
5468	static uint8_t descr[4];
5469	static uint8_t echo_descr[4] = { 0 };
5470
5471	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5472
5473	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5474	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5475
5476	if (lun->flags & CTL_LUN_PR_RESERVED) {
5477		uint32_t residx;
5478
5479		/*
5480		 * XXX KDM need a lock here.
5481		 */
5482		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5483		if ((lun->res_type == SPR_TYPE_EX_AC
5484		  && residx != lun->pr_res_idx)
5485		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5486		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5487		  && !lun->per_res[residx].registered)) {
5488			ctl_set_reservation_conflict(ctsio);
5489			ctl_done((union ctl_io *)ctsio);
5490			return (CTL_RETVAL_COMPLETE);
5491	        }
5492	}
5493
5494	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5495	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5496	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5497		ctl_set_invalid_field(ctsio,
5498				      /*sks_valid*/ 1,
5499				      /*command*/ 1,
5500				      /*field*/ 1,
5501				      /*bit_valid*/ 1,
5502				      /*bit*/ 4);
5503		ctl_done((union ctl_io *)ctsio);
5504		return (CTL_RETVAL_COMPLETE);
5505	}
5506
5507	len = scsi_3btoul(cdb->length);
5508	buffer_offset = scsi_3btoul(cdb->offset);
5509
5510	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5511		ctl_set_invalid_field(ctsio,
5512				      /*sks_valid*/ 1,
5513				      /*command*/ 1,
5514				      /*field*/ 6,
5515				      /*bit_valid*/ 0,
5516				      /*bit*/ 0);
5517		ctl_done((union ctl_io *)ctsio);
5518		return (CTL_RETVAL_COMPLETE);
5519	}
5520
5521	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5522		descr[0] = 0;
5523		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5524		ctsio->kern_data_ptr = descr;
5525		len = min(len, sizeof(descr));
5526	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5527		ctsio->kern_data_ptr = echo_descr;
5528		len = min(len, sizeof(echo_descr));
5529	} else
5530		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5531	ctsio->kern_data_len = len;
5532	ctsio->kern_total_len = len;
5533	ctsio->kern_data_resid = 0;
5534	ctsio->kern_rel_offset = 0;
5535	ctsio->kern_sg_entries = 0;
5536	ctsio->be_move_done = ctl_config_move_done;
5537	ctl_datamove((union ctl_io *)ctsio);
5538
5539	return (CTL_RETVAL_COMPLETE);
5540}
5541
5542int
5543ctl_write_buffer(struct ctl_scsiio *ctsio)
5544{
5545	struct scsi_write_buffer *cdb;
5546	struct ctl_lun *lun;
5547	int buffer_offset, len;
5548
5549	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5550
5551	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5552	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5553
5554	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5555		ctl_set_invalid_field(ctsio,
5556				      /*sks_valid*/ 1,
5557				      /*command*/ 1,
5558				      /*field*/ 1,
5559				      /*bit_valid*/ 1,
5560				      /*bit*/ 4);
5561		ctl_done((union ctl_io *)ctsio);
5562		return (CTL_RETVAL_COMPLETE);
5563	}
5564
5565	len = scsi_3btoul(cdb->length);
5566	buffer_offset = scsi_3btoul(cdb->offset);
5567
5568	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5569		ctl_set_invalid_field(ctsio,
5570				      /*sks_valid*/ 1,
5571				      /*command*/ 1,
5572				      /*field*/ 6,
5573				      /*bit_valid*/ 0,
5574				      /*bit*/ 0);
5575		ctl_done((union ctl_io *)ctsio);
5576		return (CTL_RETVAL_COMPLETE);
5577	}
5578
5579	/*
5580	 * If we've got a kernel request that hasn't been malloced yet,
5581	 * malloc it and tell the caller the data buffer is here.
5582	 */
5583	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5584		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5585		ctsio->kern_data_len = len;
5586		ctsio->kern_total_len = len;
5587		ctsio->kern_data_resid = 0;
5588		ctsio->kern_rel_offset = 0;
5589		ctsio->kern_sg_entries = 0;
5590		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5591		ctsio->be_move_done = ctl_config_move_done;
5592		ctl_datamove((union ctl_io *)ctsio);
5593
5594		return (CTL_RETVAL_COMPLETE);
5595	}
5596
5597	ctl_done((union ctl_io *)ctsio);
5598
5599	return (CTL_RETVAL_COMPLETE);
5600}
5601
5602int
5603ctl_write_same(struct ctl_scsiio *ctsio)
5604{
5605	struct ctl_lun *lun;
5606	struct ctl_lba_len_flags *lbalen;
5607	uint64_t lba;
5608	uint32_t num_blocks;
5609	int len, retval;
5610	uint8_t byte2;
5611
5612	retval = CTL_RETVAL_COMPLETE;
5613
5614	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5615
5616	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5617
5618	switch (ctsio->cdb[0]) {
5619	case WRITE_SAME_10: {
5620		struct scsi_write_same_10 *cdb;
5621
5622		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5623
5624		lba = scsi_4btoul(cdb->addr);
5625		num_blocks = scsi_2btoul(cdb->length);
5626		byte2 = cdb->byte2;
5627		break;
5628	}
5629	case WRITE_SAME_16: {
5630		struct scsi_write_same_16 *cdb;
5631
5632		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5633
5634		lba = scsi_8btou64(cdb->addr);
5635		num_blocks = scsi_4btoul(cdb->length);
5636		byte2 = cdb->byte2;
5637		break;
5638	}
5639	default:
5640		/*
5641		 * We got a command we don't support.  This shouldn't
5642		 * happen, commands should be filtered out above us.
5643		 */
5644		ctl_set_invalid_opcode(ctsio);
5645		ctl_done((union ctl_io *)ctsio);
5646
5647		return (CTL_RETVAL_COMPLETE);
5648		break; /* NOTREACHED */
5649	}
5650
5651	/*
5652	 * The first check is to make sure we're in bounds, the second
5653	 * check is to catch wrap-around problems.  If the lba + num blocks
5654	 * is less than the lba, then we've wrapped around and the block
5655	 * range is invalid anyway.
5656	 */
5657	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5658	 || ((lba + num_blocks) < lba)) {
5659		ctl_set_lba_out_of_range(ctsio);
5660		ctl_done((union ctl_io *)ctsio);
5661		return (CTL_RETVAL_COMPLETE);
5662	}
5663
5664	/* Zero number of blocks means "to the last logical block" */
5665	if (num_blocks == 0) {
5666		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5667			ctl_set_invalid_field(ctsio,
5668					      /*sks_valid*/ 0,
5669					      /*command*/ 1,
5670					      /*field*/ 0,
5671					      /*bit_valid*/ 0,
5672					      /*bit*/ 0);
5673			ctl_done((union ctl_io *)ctsio);
5674			return (CTL_RETVAL_COMPLETE);
5675		}
5676		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5677	}
5678
5679	len = lun->be_lun->blocksize;
5680
5681	/*
5682	 * If we've got a kernel request that hasn't been malloced yet,
5683	 * malloc it and tell the caller the data buffer is here.
5684	 */
5685	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5686		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5687		ctsio->kern_data_len = len;
5688		ctsio->kern_total_len = len;
5689		ctsio->kern_data_resid = 0;
5690		ctsio->kern_rel_offset = 0;
5691		ctsio->kern_sg_entries = 0;
5692		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5693		ctsio->be_move_done = ctl_config_move_done;
5694		ctl_datamove((union ctl_io *)ctsio);
5695
5696		return (CTL_RETVAL_COMPLETE);
5697	}
5698
5699	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5700	lbalen->lba = lba;
5701	lbalen->len = num_blocks;
5702	lbalen->flags = byte2;
5703	retval = lun->backend->config_write((union ctl_io *)ctsio);
5704
5705	return (retval);
5706}
5707
5708int
5709ctl_unmap(struct ctl_scsiio *ctsio)
5710{
5711	struct ctl_lun *lun;
5712	struct scsi_unmap *cdb;
5713	struct ctl_ptr_len_flags *ptrlen;
5714	struct scsi_unmap_header *hdr;
5715	struct scsi_unmap_desc *buf, *end;
5716	uint64_t lba;
5717	uint32_t num_blocks;
5718	int len, retval;
5719	uint8_t byte2;
5720
5721	retval = CTL_RETVAL_COMPLETE;
5722
5723	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5724
5725	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5726	cdb = (struct scsi_unmap *)ctsio->cdb;
5727
5728	len = scsi_2btoul(cdb->length);
5729	byte2 = cdb->byte2;
5730
5731	/*
5732	 * If we've got a kernel request that hasn't been malloced yet,
5733	 * malloc it and tell the caller the data buffer is here.
5734	 */
5735	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5736		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5737		ctsio->kern_data_len = len;
5738		ctsio->kern_total_len = len;
5739		ctsio->kern_data_resid = 0;
5740		ctsio->kern_rel_offset = 0;
5741		ctsio->kern_sg_entries = 0;
5742		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5743		ctsio->be_move_done = ctl_config_move_done;
5744		ctl_datamove((union ctl_io *)ctsio);
5745
5746		return (CTL_RETVAL_COMPLETE);
5747	}
5748
5749	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5750	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5751	if (len < sizeof (*hdr) ||
5752	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5753	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5754	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5755		ctl_set_invalid_field(ctsio,
5756				      /*sks_valid*/ 0,
5757				      /*command*/ 0,
5758				      /*field*/ 0,
5759				      /*bit_valid*/ 0,
5760				      /*bit*/ 0);
5761		ctl_done((union ctl_io *)ctsio);
5762		return (CTL_RETVAL_COMPLETE);
5763	}
5764	len = scsi_2btoul(hdr->desc_length);
5765	buf = (struct scsi_unmap_desc *)(hdr + 1);
5766	end = buf + len / sizeof(*buf);
5767
5768	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5769	ptrlen->ptr = (void *)buf;
5770	ptrlen->len = len;
5771	ptrlen->flags = byte2;
5772
5773	for (; buf < end; buf++) {
5774		lba = scsi_8btou64(buf->lba);
5775		num_blocks = scsi_4btoul(buf->length);
5776		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5777		 || ((lba + num_blocks) < lba)) {
5778			ctl_set_lba_out_of_range(ctsio);
5779			ctl_done((union ctl_io *)ctsio);
5780			return (CTL_RETVAL_COMPLETE);
5781		}
5782	}
5783
5784	retval = lun->backend->config_write((union ctl_io *)ctsio);
5785
5786	return (retval);
5787}
5788
5789/*
5790 * Note that this function currently doesn't actually do anything inside
5791 * CTL to enforce things if the DQue bit is turned on.
5792 *
5793 * Also note that this function can't be used in the default case, because
5794 * the DQue bit isn't set in the changeable mask for the control mode page
5795 * anyway.  This is just here as an example for how to implement a page
5796 * handler, and a placeholder in case we want to allow the user to turn
5797 * tagged queueing on and off.
5798 *
5799 * The D_SENSE bit handling is functional, however, and will turn
5800 * descriptor sense on and off for a given LUN.
5801 */
5802int
5803ctl_control_page_handler(struct ctl_scsiio *ctsio,
5804			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5805{
5806	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5807	struct ctl_lun *lun;
5808	struct ctl_softc *softc;
5809	int set_ua;
5810	uint32_t initidx;
5811
5812	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5813	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5814	set_ua = 0;
5815
5816	user_cp = (struct scsi_control_page *)page_ptr;
5817	current_cp = (struct scsi_control_page *)
5818		(page_index->page_data + (page_index->page_len *
5819		CTL_PAGE_CURRENT));
5820	saved_cp = (struct scsi_control_page *)
5821		(page_index->page_data + (page_index->page_len *
5822		CTL_PAGE_SAVED));
5823
5824	softc = control_softc;
5825
5826	mtx_lock(&lun->lun_lock);
5827	if (((current_cp->rlec & SCP_DSENSE) == 0)
5828	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5829		/*
5830		 * Descriptor sense is currently turned off and the user
5831		 * wants to turn it on.
5832		 */
5833		current_cp->rlec |= SCP_DSENSE;
5834		saved_cp->rlec |= SCP_DSENSE;
5835		lun->flags |= CTL_LUN_SENSE_DESC;
5836		set_ua = 1;
5837	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5838		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5839		/*
5840		 * Descriptor sense is currently turned on, and the user
5841		 * wants to turn it off.
5842		 */
5843		current_cp->rlec &= ~SCP_DSENSE;
5844		saved_cp->rlec &= ~SCP_DSENSE;
5845		lun->flags &= ~CTL_LUN_SENSE_DESC;
5846		set_ua = 1;
5847	}
5848	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5849		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5850#ifdef NEEDTOPORT
5851			csevent_log(CSC_CTL | CSC_SHELF_SW |
5852				    CTL_UNTAG_TO_UNTAG,
5853				    csevent_LogType_Trace,
5854				    csevent_Severity_Information,
5855				    csevent_AlertLevel_Green,
5856				    csevent_FRU_Firmware,
5857				    csevent_FRU_Unknown,
5858				    "Received untagged to untagged transition");
5859#endif /* NEEDTOPORT */
5860		} else {
5861#ifdef NEEDTOPORT
5862			csevent_log(CSC_CTL | CSC_SHELF_SW |
5863				    CTL_UNTAG_TO_TAG,
5864				    csevent_LogType_ConfigChange,
5865				    csevent_Severity_Information,
5866				    csevent_AlertLevel_Green,
5867				    csevent_FRU_Firmware,
5868				    csevent_FRU_Unknown,
5869				    "Received untagged to tagged "
5870				    "queueing transition");
5871#endif /* NEEDTOPORT */
5872
5873			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5874			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5875			set_ua = 1;
5876		}
5877	} else {
5878		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5879#ifdef NEEDTOPORT
5880			csevent_log(CSC_CTL | CSC_SHELF_SW |
5881				    CTL_TAG_TO_UNTAG,
5882				    csevent_LogType_ConfigChange,
5883				    csevent_Severity_Warning,
5884				    csevent_AlertLevel_Yellow,
5885				    csevent_FRU_Firmware,
5886				    csevent_FRU_Unknown,
5887				    "Received tagged queueing to untagged "
5888				    "transition");
5889#endif /* NEEDTOPORT */
5890
5891			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5892			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5893			set_ua = 1;
5894		} else {
5895#ifdef NEEDTOPORT
5896			csevent_log(CSC_CTL | CSC_SHELF_SW |
5897				    CTL_TAG_TO_TAG,
5898				    csevent_LogType_Trace,
5899				    csevent_Severity_Information,
5900				    csevent_AlertLevel_Green,
5901				    csevent_FRU_Firmware,
5902				    csevent_FRU_Unknown,
5903				    "Received tagged queueing to tagged "
5904				    "queueing transition");
5905#endif /* NEEDTOPORT */
5906		}
5907	}
5908	if (set_ua != 0) {
5909		int i;
5910		/*
5911		 * Let other initiators know that the mode
5912		 * parameters for this LUN have changed.
5913		 */
5914		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5915			if (i == initidx)
5916				continue;
5917
5918			lun->pending_sense[i].ua_pending |=
5919				CTL_UA_MODE_CHANGE;
5920		}
5921	}
5922	mtx_unlock(&lun->lun_lock);
5923
5924	return (0);
5925}
5926
5927int
5928ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5929		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5930{
5931	return (0);
5932}
5933
5934int
5935ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5936			   struct ctl_page_index *page_index, int pc)
5937{
5938	struct copan_power_subpage *page;
5939
5940	page = (struct copan_power_subpage *)page_index->page_data +
5941		(page_index->page_len * pc);
5942
5943	switch (pc) {
5944	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5945		/*
5946		 * We don't update the changable bits for this page.
5947		 */
5948		break;
5949	case SMS_PAGE_CTRL_CURRENT >> 6:
5950	case SMS_PAGE_CTRL_DEFAULT >> 6:
5951	case SMS_PAGE_CTRL_SAVED >> 6:
5952#ifdef NEEDTOPORT
5953		ctl_update_power_subpage(page);
5954#endif
5955		break;
5956	default:
5957#ifdef NEEDTOPORT
5958		EPRINT(0, "Invalid PC %d!!", pc);
5959#endif
5960		break;
5961	}
5962	return (0);
5963}
5964
5965
5966int
5967ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5968		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5969{
5970	struct copan_aps_subpage *user_sp;
5971	struct copan_aps_subpage *current_sp;
5972	union ctl_modepage_info *modepage_info;
5973	struct ctl_softc *softc;
5974	struct ctl_lun *lun;
5975	int retval;
5976
5977	retval = CTL_RETVAL_COMPLETE;
5978	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5979		     (page_index->page_len * CTL_PAGE_CURRENT));
5980	softc = control_softc;
5981	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5982
5983	user_sp = (struct copan_aps_subpage *)page_ptr;
5984
5985	modepage_info = (union ctl_modepage_info *)
5986		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5987
5988	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5989	modepage_info->header.subpage = page_index->subpage;
5990	modepage_info->aps.lock_active = user_sp->lock_active;
5991
5992	mtx_lock(&softc->ctl_lock);
5993
5994	/*
5995	 * If there is a request to lock the LUN and another LUN is locked
5996	 * this is an error. If the requested LUN is already locked ignore
5997	 * the request. If no LUN is locked attempt to lock it.
5998	 * if there is a request to unlock the LUN and the LUN is currently
5999	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6000	 * if another LUN is locked or no LUN is locked.
6001	 */
6002	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6003		if (softc->aps_locked_lun == lun->lun) {
6004			/*
6005			 * This LUN is already locked, so we're done.
6006			 */
6007			retval = CTL_RETVAL_COMPLETE;
6008		} else if (softc->aps_locked_lun == 0) {
6009			/*
6010			 * No one has the lock, pass the request to the
6011			 * backend.
6012			 */
6013			retval = lun->backend->config_write(
6014				(union ctl_io *)ctsio);
6015		} else {
6016			/*
6017			 * Someone else has the lock, throw out the request.
6018			 */
6019			ctl_set_already_locked(ctsio);
6020			free(ctsio->kern_data_ptr, M_CTL);
6021			ctl_done((union ctl_io *)ctsio);
6022
6023			/*
6024			 * Set the return value so that ctl_do_mode_select()
6025			 * won't try to complete the command.  We already
6026			 * completed it here.
6027			 */
6028			retval = CTL_RETVAL_ERROR;
6029		}
6030	} else if (softc->aps_locked_lun == lun->lun) {
6031		/*
6032		 * This LUN is locked, so pass the unlock request to the
6033		 * backend.
6034		 */
6035		retval = lun->backend->config_write((union ctl_io *)ctsio);
6036	}
6037	mtx_unlock(&softc->ctl_lock);
6038
6039	return (retval);
6040}
6041
6042int
6043ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6044				struct ctl_page_index *page_index,
6045				uint8_t *page_ptr)
6046{
6047	uint8_t *c;
6048	int i;
6049
6050	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6051	ctl_time_io_secs =
6052		(c[0] << 8) |
6053		(c[1] << 0) |
6054		0;
6055	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6056	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6057	printf("page data:");
6058	for (i=0; i<8; i++)
6059		printf(" %.2x",page_ptr[i]);
6060	printf("\n");
6061	return (0);
6062}
6063
6064int
6065ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6066			       struct ctl_page_index *page_index,
6067			       int pc)
6068{
6069	struct copan_debugconf_subpage *page;
6070
6071	page = (struct copan_debugconf_subpage *)page_index->page_data +
6072		(page_index->page_len * pc);
6073
6074	switch (pc) {
6075	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6076	case SMS_PAGE_CTRL_DEFAULT >> 6:
6077	case SMS_PAGE_CTRL_SAVED >> 6:
6078		/*
6079		 * We don't update the changable or default bits for this page.
6080		 */
6081		break;
6082	case SMS_PAGE_CTRL_CURRENT >> 6:
6083		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6084		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6085		break;
6086	default:
6087#ifdef NEEDTOPORT
6088		EPRINT(0, "Invalid PC %d!!", pc);
6089#endif /* NEEDTOPORT */
6090		break;
6091	}
6092	return (0);
6093}
6094
6095
6096static int
6097ctl_do_mode_select(union ctl_io *io)
6098{
6099	struct scsi_mode_page_header *page_header;
6100	struct ctl_page_index *page_index;
6101	struct ctl_scsiio *ctsio;
6102	int control_dev, page_len;
6103	int page_len_offset, page_len_size;
6104	union ctl_modepage_info *modepage_info;
6105	struct ctl_lun *lun;
6106	int *len_left, *len_used;
6107	int retval, i;
6108
6109	ctsio = &io->scsiio;
6110	page_index = NULL;
6111	page_len = 0;
6112	retval = CTL_RETVAL_COMPLETE;
6113
6114	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6115
6116	if (lun->be_lun->lun_type != T_DIRECT)
6117		control_dev = 1;
6118	else
6119		control_dev = 0;
6120
6121	modepage_info = (union ctl_modepage_info *)
6122		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6123	len_left = &modepage_info->header.len_left;
6124	len_used = &modepage_info->header.len_used;
6125
6126do_next_page:
6127
6128	page_header = (struct scsi_mode_page_header *)
6129		(ctsio->kern_data_ptr + *len_used);
6130
6131	if (*len_left == 0) {
6132		free(ctsio->kern_data_ptr, M_CTL);
6133		ctl_set_success(ctsio);
6134		ctl_done((union ctl_io *)ctsio);
6135		return (CTL_RETVAL_COMPLETE);
6136	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6137
6138		free(ctsio->kern_data_ptr, M_CTL);
6139		ctl_set_param_len_error(ctsio);
6140		ctl_done((union ctl_io *)ctsio);
6141		return (CTL_RETVAL_COMPLETE);
6142
6143	} else if ((page_header->page_code & SMPH_SPF)
6144		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6145
6146		free(ctsio->kern_data_ptr, M_CTL);
6147		ctl_set_param_len_error(ctsio);
6148		ctl_done((union ctl_io *)ctsio);
6149		return (CTL_RETVAL_COMPLETE);
6150	}
6151
6152
6153	/*
6154	 * XXX KDM should we do something with the block descriptor?
6155	 */
6156	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6157
6158		if ((control_dev != 0)
6159		 && (lun->mode_pages.index[i].page_flags &
6160		     CTL_PAGE_FLAG_DISK_ONLY))
6161			continue;
6162
6163		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6164		    (page_header->page_code & SMPH_PC_MASK))
6165			continue;
6166
6167		/*
6168		 * If neither page has a subpage code, then we've got a
6169		 * match.
6170		 */
6171		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6172		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6173			page_index = &lun->mode_pages.index[i];
6174			page_len = page_header->page_length;
6175			break;
6176		}
6177
6178		/*
6179		 * If both pages have subpages, then the subpage numbers
6180		 * have to match.
6181		 */
6182		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6183		  && (page_header->page_code & SMPH_SPF)) {
6184			struct scsi_mode_page_header_sp *sph;
6185
6186			sph = (struct scsi_mode_page_header_sp *)page_header;
6187
6188			if (lun->mode_pages.index[i].subpage ==
6189			    sph->subpage) {
6190				page_index = &lun->mode_pages.index[i];
6191				page_len = scsi_2btoul(sph->page_length);
6192				break;
6193			}
6194		}
6195	}
6196
6197	/*
6198	 * If we couldn't find the page, or if we don't have a mode select
6199	 * handler for it, send back an error to the user.
6200	 */
6201	if ((page_index == NULL)
6202	 || (page_index->select_handler == NULL)) {
6203		ctl_set_invalid_field(ctsio,
6204				      /*sks_valid*/ 1,
6205				      /*command*/ 0,
6206				      /*field*/ *len_used,
6207				      /*bit_valid*/ 0,
6208				      /*bit*/ 0);
6209		free(ctsio->kern_data_ptr, M_CTL);
6210		ctl_done((union ctl_io *)ctsio);
6211		return (CTL_RETVAL_COMPLETE);
6212	}
6213
6214	if (page_index->page_code & SMPH_SPF) {
6215		page_len_offset = 2;
6216		page_len_size = 2;
6217	} else {
6218		page_len_size = 1;
6219		page_len_offset = 1;
6220	}
6221
6222	/*
6223	 * If the length the initiator gives us isn't the one we specify in
6224	 * the mode page header, or if they didn't specify enough data in
6225	 * the CDB to avoid truncating this page, kick out the request.
6226	 */
6227	if ((page_len != (page_index->page_len - page_len_offset -
6228			  page_len_size))
6229	 || (*len_left < page_index->page_len)) {
6230
6231
6232		ctl_set_invalid_field(ctsio,
6233				      /*sks_valid*/ 1,
6234				      /*command*/ 0,
6235				      /*field*/ *len_used + page_len_offset,
6236				      /*bit_valid*/ 0,
6237				      /*bit*/ 0);
6238		free(ctsio->kern_data_ptr, M_CTL);
6239		ctl_done((union ctl_io *)ctsio);
6240		return (CTL_RETVAL_COMPLETE);
6241	}
6242
6243	/*
6244	 * Run through the mode page, checking to make sure that the bits
6245	 * the user changed are actually legal for him to change.
6246	 */
6247	for (i = 0; i < page_index->page_len; i++) {
6248		uint8_t *user_byte, *change_mask, *current_byte;
6249		int bad_bit;
6250		int j;
6251
6252		user_byte = (uint8_t *)page_header + i;
6253		change_mask = page_index->page_data +
6254			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6255		current_byte = page_index->page_data +
6256			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6257
6258		/*
6259		 * Check to see whether the user set any bits in this byte
6260		 * that he is not allowed to set.
6261		 */
6262		if ((*user_byte & ~(*change_mask)) ==
6263		    (*current_byte & ~(*change_mask)))
6264			continue;
6265
6266		/*
6267		 * Go through bit by bit to determine which one is illegal.
6268		 */
6269		bad_bit = 0;
6270		for (j = 7; j >= 0; j--) {
6271			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6272			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6273				bad_bit = i;
6274				break;
6275			}
6276		}
6277		ctl_set_invalid_field(ctsio,
6278				      /*sks_valid*/ 1,
6279				      /*command*/ 0,
6280				      /*field*/ *len_used + i,
6281				      /*bit_valid*/ 1,
6282				      /*bit*/ bad_bit);
6283		free(ctsio->kern_data_ptr, M_CTL);
6284		ctl_done((union ctl_io *)ctsio);
6285		return (CTL_RETVAL_COMPLETE);
6286	}
6287
6288	/*
6289	 * Decrement these before we call the page handler, since we may
6290	 * end up getting called back one way or another before the handler
6291	 * returns to this context.
6292	 */
6293	*len_left -= page_index->page_len;
6294	*len_used += page_index->page_len;
6295
6296	retval = page_index->select_handler(ctsio, page_index,
6297					    (uint8_t *)page_header);
6298
6299	/*
6300	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6301	 * wait until this queued command completes to finish processing
6302	 * the mode page.  If it returns anything other than
6303	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6304	 * already set the sense information, freed the data pointer, and
6305	 * completed the io for us.
6306	 */
6307	if (retval != CTL_RETVAL_COMPLETE)
6308		goto bailout_no_done;
6309
6310	/*
6311	 * If the initiator sent us more than one page, parse the next one.
6312	 */
6313	if (*len_left > 0)
6314		goto do_next_page;
6315
6316	ctl_set_success(ctsio);
6317	free(ctsio->kern_data_ptr, M_CTL);
6318	ctl_done((union ctl_io *)ctsio);
6319
6320bailout_no_done:
6321
6322	return (CTL_RETVAL_COMPLETE);
6323
6324}
6325
6326int
6327ctl_mode_select(struct ctl_scsiio *ctsio)
6328{
6329	int param_len, pf, sp;
6330	int header_size, bd_len;
6331	int len_left, len_used;
6332	struct ctl_page_index *page_index;
6333	struct ctl_lun *lun;
6334	int control_dev, page_len;
6335	union ctl_modepage_info *modepage_info;
6336	int retval;
6337
6338	pf = 0;
6339	sp = 0;
6340	page_len = 0;
6341	len_used = 0;
6342	len_left = 0;
6343	retval = 0;
6344	bd_len = 0;
6345	page_index = NULL;
6346
6347	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6348
6349	if (lun->be_lun->lun_type != T_DIRECT)
6350		control_dev = 1;
6351	else
6352		control_dev = 0;
6353
6354	switch (ctsio->cdb[0]) {
6355	case MODE_SELECT_6: {
6356		struct scsi_mode_select_6 *cdb;
6357
6358		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6359
6360		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6361		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6362
6363		param_len = cdb->length;
6364		header_size = sizeof(struct scsi_mode_header_6);
6365		break;
6366	}
6367	case MODE_SELECT_10: {
6368		struct scsi_mode_select_10 *cdb;
6369
6370		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6371
6372		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6373		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6374
6375		param_len = scsi_2btoul(cdb->length);
6376		header_size = sizeof(struct scsi_mode_header_10);
6377		break;
6378	}
6379	default:
6380		ctl_set_invalid_opcode(ctsio);
6381		ctl_done((union ctl_io *)ctsio);
6382		return (CTL_RETVAL_COMPLETE);
6383		break; /* NOTREACHED */
6384	}
6385
6386	/*
6387	 * From SPC-3:
6388	 * "A parameter list length of zero indicates that the Data-Out Buffer
6389	 * shall be empty. This condition shall not be considered as an error."
6390	 */
6391	if (param_len == 0) {
6392		ctl_set_success(ctsio);
6393		ctl_done((union ctl_io *)ctsio);
6394		return (CTL_RETVAL_COMPLETE);
6395	}
6396
6397	/*
6398	 * Since we'll hit this the first time through, prior to
6399	 * allocation, we don't need to free a data buffer here.
6400	 */
6401	if (param_len < header_size) {
6402		ctl_set_param_len_error(ctsio);
6403		ctl_done((union ctl_io *)ctsio);
6404		return (CTL_RETVAL_COMPLETE);
6405	}
6406
6407	/*
6408	 * Allocate the data buffer and grab the user's data.  In theory,
6409	 * we shouldn't have to sanity check the parameter list length here
6410	 * because the maximum size is 64K.  We should be able to malloc
6411	 * that much without too many problems.
6412	 */
6413	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6414		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6415		ctsio->kern_data_len = param_len;
6416		ctsio->kern_total_len = param_len;
6417		ctsio->kern_data_resid = 0;
6418		ctsio->kern_rel_offset = 0;
6419		ctsio->kern_sg_entries = 0;
6420		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6421		ctsio->be_move_done = ctl_config_move_done;
6422		ctl_datamove((union ctl_io *)ctsio);
6423
6424		return (CTL_RETVAL_COMPLETE);
6425	}
6426
6427	switch (ctsio->cdb[0]) {
6428	case MODE_SELECT_6: {
6429		struct scsi_mode_header_6 *mh6;
6430
6431		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6432		bd_len = mh6->blk_desc_len;
6433		break;
6434	}
6435	case MODE_SELECT_10: {
6436		struct scsi_mode_header_10 *mh10;
6437
6438		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6439		bd_len = scsi_2btoul(mh10->blk_desc_len);
6440		break;
6441	}
6442	default:
6443		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6444		break;
6445	}
6446
6447	if (param_len < (header_size + bd_len)) {
6448		free(ctsio->kern_data_ptr, M_CTL);
6449		ctl_set_param_len_error(ctsio);
6450		ctl_done((union ctl_io *)ctsio);
6451		return (CTL_RETVAL_COMPLETE);
6452	}
6453
6454	/*
6455	 * Set the IO_CONT flag, so that if this I/O gets passed to
6456	 * ctl_config_write_done(), it'll get passed back to
6457	 * ctl_do_mode_select() for further processing, or completion if
6458	 * we're all done.
6459	 */
6460	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6461	ctsio->io_cont = ctl_do_mode_select;
6462
6463	modepage_info = (union ctl_modepage_info *)
6464		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6465
6466	memset(modepage_info, 0, sizeof(*modepage_info));
6467
6468	len_left = param_len - header_size - bd_len;
6469	len_used = header_size + bd_len;
6470
6471	modepage_info->header.len_left = len_left;
6472	modepage_info->header.len_used = len_used;
6473
6474	return (ctl_do_mode_select((union ctl_io *)ctsio));
6475}
6476
6477int
6478ctl_mode_sense(struct ctl_scsiio *ctsio)
6479{
6480	struct ctl_lun *lun;
6481	int pc, page_code, dbd, llba, subpage;
6482	int alloc_len, page_len, header_len, total_len;
6483	struct scsi_mode_block_descr *block_desc;
6484	struct ctl_page_index *page_index;
6485	int control_dev;
6486
6487	dbd = 0;
6488	llba = 0;
6489	block_desc = NULL;
6490	page_index = NULL;
6491
6492	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6493
6494	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6495
6496	if (lun->be_lun->lun_type != T_DIRECT)
6497		control_dev = 1;
6498	else
6499		control_dev = 0;
6500
6501	if (lun->flags & CTL_LUN_PR_RESERVED) {
6502		uint32_t residx;
6503
6504		/*
6505		 * XXX KDM need a lock here.
6506		 */
6507		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6508		if ((lun->res_type == SPR_TYPE_EX_AC
6509		  && residx != lun->pr_res_idx)
6510		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6511		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6512		  && !lun->per_res[residx].registered)) {
6513			ctl_set_reservation_conflict(ctsio);
6514			ctl_done((union ctl_io *)ctsio);
6515			return (CTL_RETVAL_COMPLETE);
6516		}
6517	}
6518
6519	switch (ctsio->cdb[0]) {
6520	case MODE_SENSE_6: {
6521		struct scsi_mode_sense_6 *cdb;
6522
6523		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6524
6525		header_len = sizeof(struct scsi_mode_hdr_6);
6526		if (cdb->byte2 & SMS_DBD)
6527			dbd = 1;
6528		else
6529			header_len += sizeof(struct scsi_mode_block_descr);
6530
6531		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6532		page_code = cdb->page & SMS_PAGE_CODE;
6533		subpage = cdb->subpage;
6534		alloc_len = cdb->length;
6535		break;
6536	}
6537	case MODE_SENSE_10: {
6538		struct scsi_mode_sense_10 *cdb;
6539
6540		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6541
6542		header_len = sizeof(struct scsi_mode_hdr_10);
6543
6544		if (cdb->byte2 & SMS_DBD)
6545			dbd = 1;
6546		else
6547			header_len += sizeof(struct scsi_mode_block_descr);
6548		if (cdb->byte2 & SMS10_LLBAA)
6549			llba = 1;
6550		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6551		page_code = cdb->page & SMS_PAGE_CODE;
6552		subpage = cdb->subpage;
6553		alloc_len = scsi_2btoul(cdb->length);
6554		break;
6555	}
6556	default:
6557		ctl_set_invalid_opcode(ctsio);
6558		ctl_done((union ctl_io *)ctsio);
6559		return (CTL_RETVAL_COMPLETE);
6560		break; /* NOTREACHED */
6561	}
6562
6563	/*
6564	 * We have to make a first pass through to calculate the size of
6565	 * the pages that match the user's query.  Then we allocate enough
6566	 * memory to hold it, and actually copy the data into the buffer.
6567	 */
6568	switch (page_code) {
6569	case SMS_ALL_PAGES_PAGE: {
6570		int i;
6571
6572		page_len = 0;
6573
6574		/*
6575		 * At the moment, values other than 0 and 0xff here are
6576		 * reserved according to SPC-3.
6577		 */
6578		if ((subpage != SMS_SUBPAGE_PAGE_0)
6579		 && (subpage != SMS_SUBPAGE_ALL)) {
6580			ctl_set_invalid_field(ctsio,
6581					      /*sks_valid*/ 1,
6582					      /*command*/ 1,
6583					      /*field*/ 3,
6584					      /*bit_valid*/ 0,
6585					      /*bit*/ 0);
6586			ctl_done((union ctl_io *)ctsio);
6587			return (CTL_RETVAL_COMPLETE);
6588		}
6589
6590		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6591			if ((control_dev != 0)
6592			 && (lun->mode_pages.index[i].page_flags &
6593			     CTL_PAGE_FLAG_DISK_ONLY))
6594				continue;
6595
6596			/*
6597			 * We don't use this subpage if the user didn't
6598			 * request all subpages.
6599			 */
6600			if ((lun->mode_pages.index[i].subpage != 0)
6601			 && (subpage == SMS_SUBPAGE_PAGE_0))
6602				continue;
6603
6604#if 0
6605			printf("found page %#x len %d\n",
6606			       lun->mode_pages.index[i].page_code &
6607			       SMPH_PC_MASK,
6608			       lun->mode_pages.index[i].page_len);
6609#endif
6610			page_len += lun->mode_pages.index[i].page_len;
6611		}
6612		break;
6613	}
6614	default: {
6615		int i;
6616
6617		page_len = 0;
6618
6619		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6620			/* Look for the right page code */
6621			if ((lun->mode_pages.index[i].page_code &
6622			     SMPH_PC_MASK) != page_code)
6623				continue;
6624
6625			/* Look for the right subpage or the subpage wildcard*/
6626			if ((lun->mode_pages.index[i].subpage != subpage)
6627			 && (subpage != SMS_SUBPAGE_ALL))
6628				continue;
6629
6630			/* Make sure the page is supported for this dev type */
6631			if ((control_dev != 0)
6632			 && (lun->mode_pages.index[i].page_flags &
6633			     CTL_PAGE_FLAG_DISK_ONLY))
6634				continue;
6635
6636#if 0
6637			printf("found page %#x len %d\n",
6638			       lun->mode_pages.index[i].page_code &
6639			       SMPH_PC_MASK,
6640			       lun->mode_pages.index[i].page_len);
6641#endif
6642
6643			page_len += lun->mode_pages.index[i].page_len;
6644		}
6645
6646		if (page_len == 0) {
6647			ctl_set_invalid_field(ctsio,
6648					      /*sks_valid*/ 1,
6649					      /*command*/ 1,
6650					      /*field*/ 2,
6651					      /*bit_valid*/ 1,
6652					      /*bit*/ 5);
6653			ctl_done((union ctl_io *)ctsio);
6654			return (CTL_RETVAL_COMPLETE);
6655		}
6656		break;
6657	}
6658	}
6659
6660	total_len = header_len + page_len;
6661#if 0
6662	printf("header_len = %d, page_len = %d, total_len = %d\n",
6663	       header_len, page_len, total_len);
6664#endif
6665
6666	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6667	ctsio->kern_sg_entries = 0;
6668	ctsio->kern_data_resid = 0;
6669	ctsio->kern_rel_offset = 0;
6670	if (total_len < alloc_len) {
6671		ctsio->residual = alloc_len - total_len;
6672		ctsio->kern_data_len = total_len;
6673		ctsio->kern_total_len = total_len;
6674	} else {
6675		ctsio->residual = 0;
6676		ctsio->kern_data_len = alloc_len;
6677		ctsio->kern_total_len = alloc_len;
6678	}
6679
6680	switch (ctsio->cdb[0]) {
6681	case MODE_SENSE_6: {
6682		struct scsi_mode_hdr_6 *header;
6683
6684		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6685
6686		header->datalen = ctl_min(total_len - 1, 254);
6687
6688		if (dbd)
6689			header->block_descr_len = 0;
6690		else
6691			header->block_descr_len =
6692				sizeof(struct scsi_mode_block_descr);
6693		block_desc = (struct scsi_mode_block_descr *)&header[1];
6694		break;
6695	}
6696	case MODE_SENSE_10: {
6697		struct scsi_mode_hdr_10 *header;
6698		int datalen;
6699
6700		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6701
6702		datalen = ctl_min(total_len - 2, 65533);
6703		scsi_ulto2b(datalen, header->datalen);
6704		if (dbd)
6705			scsi_ulto2b(0, header->block_descr_len);
6706		else
6707			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6708				    header->block_descr_len);
6709		block_desc = (struct scsi_mode_block_descr *)&header[1];
6710		break;
6711	}
6712	default:
6713		panic("invalid CDB type %#x", ctsio->cdb[0]);
6714		break; /* NOTREACHED */
6715	}
6716
6717	/*
6718	 * If we've got a disk, use its blocksize in the block
6719	 * descriptor.  Otherwise, just set it to 0.
6720	 */
6721	if (dbd == 0) {
6722		if (control_dev != 0)
6723			scsi_ulto3b(lun->be_lun->blocksize,
6724				    block_desc->block_len);
6725		else
6726			scsi_ulto3b(0, block_desc->block_len);
6727	}
6728
6729	switch (page_code) {
6730	case SMS_ALL_PAGES_PAGE: {
6731		int i, data_used;
6732
6733		data_used = header_len;
6734		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6735			struct ctl_page_index *page_index;
6736
6737			page_index = &lun->mode_pages.index[i];
6738
6739			if ((control_dev != 0)
6740			 && (page_index->page_flags &
6741			    CTL_PAGE_FLAG_DISK_ONLY))
6742				continue;
6743
6744			/*
6745			 * We don't use this subpage if the user didn't
6746			 * request all subpages.  We already checked (above)
6747			 * to make sure the user only specified a subpage
6748			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6749			 */
6750			if ((page_index->subpage != 0)
6751			 && (subpage == SMS_SUBPAGE_PAGE_0))
6752				continue;
6753
6754			/*
6755			 * Call the handler, if it exists, to update the
6756			 * page to the latest values.
6757			 */
6758			if (page_index->sense_handler != NULL)
6759				page_index->sense_handler(ctsio, page_index,pc);
6760
6761			memcpy(ctsio->kern_data_ptr + data_used,
6762			       page_index->page_data +
6763			       (page_index->page_len * pc),
6764			       page_index->page_len);
6765			data_used += page_index->page_len;
6766		}
6767		break;
6768	}
6769	default: {
6770		int i, data_used;
6771
6772		data_used = header_len;
6773
6774		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6775			struct ctl_page_index *page_index;
6776
6777			page_index = &lun->mode_pages.index[i];
6778
6779			/* Look for the right page code */
6780			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6781				continue;
6782
6783			/* Look for the right subpage or the subpage wildcard*/
6784			if ((page_index->subpage != subpage)
6785			 && (subpage != SMS_SUBPAGE_ALL))
6786				continue;
6787
6788			/* Make sure the page is supported for this dev type */
6789			if ((control_dev != 0)
6790			 && (page_index->page_flags &
6791			     CTL_PAGE_FLAG_DISK_ONLY))
6792				continue;
6793
6794			/*
6795			 * Call the handler, if it exists, to update the
6796			 * page to the latest values.
6797			 */
6798			if (page_index->sense_handler != NULL)
6799				page_index->sense_handler(ctsio, page_index,pc);
6800
6801			memcpy(ctsio->kern_data_ptr + data_used,
6802			       page_index->page_data +
6803			       (page_index->page_len * pc),
6804			       page_index->page_len);
6805			data_used += page_index->page_len;
6806		}
6807		break;
6808	}
6809	}
6810
6811	ctsio->scsi_status = SCSI_STATUS_OK;
6812
6813	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6814	ctsio->be_move_done = ctl_config_move_done;
6815	ctl_datamove((union ctl_io *)ctsio);
6816
6817	return (CTL_RETVAL_COMPLETE);
6818}
6819
6820int
6821ctl_read_capacity(struct ctl_scsiio *ctsio)
6822{
6823	struct scsi_read_capacity *cdb;
6824	struct scsi_read_capacity_data *data;
6825	struct ctl_lun *lun;
6826	uint32_t lba;
6827
6828	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6829
6830	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6831
6832	lba = scsi_4btoul(cdb->addr);
6833	if (((cdb->pmi & SRC_PMI) == 0)
6834	 && (lba != 0)) {
6835		ctl_set_invalid_field(/*ctsio*/ ctsio,
6836				      /*sks_valid*/ 1,
6837				      /*command*/ 1,
6838				      /*field*/ 2,
6839				      /*bit_valid*/ 0,
6840				      /*bit*/ 0);
6841		ctl_done((union ctl_io *)ctsio);
6842		return (CTL_RETVAL_COMPLETE);
6843	}
6844
6845	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6846
6847	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6848	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6849	ctsio->residual = 0;
6850	ctsio->kern_data_len = sizeof(*data);
6851	ctsio->kern_total_len = sizeof(*data);
6852	ctsio->kern_data_resid = 0;
6853	ctsio->kern_rel_offset = 0;
6854	ctsio->kern_sg_entries = 0;
6855
6856	/*
6857	 * If the maximum LBA is greater than 0xfffffffe, the user must
6858	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6859	 * serivce action set.
6860	 */
6861	if (lun->be_lun->maxlba > 0xfffffffe)
6862		scsi_ulto4b(0xffffffff, data->addr);
6863	else
6864		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6865
6866	/*
6867	 * XXX KDM this may not be 512 bytes...
6868	 */
6869	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6870
6871	ctsio->scsi_status = SCSI_STATUS_OK;
6872
6873	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6874	ctsio->be_move_done = ctl_config_move_done;
6875	ctl_datamove((union ctl_io *)ctsio);
6876
6877	return (CTL_RETVAL_COMPLETE);
6878}
6879
6880int
6881ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6882{
6883	struct scsi_read_capacity_16 *cdb;
6884	struct scsi_read_capacity_data_long *data;
6885	struct ctl_lun *lun;
6886	uint64_t lba;
6887	uint32_t alloc_len;
6888
6889	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6890
6891	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6892
6893	alloc_len = scsi_4btoul(cdb->alloc_len);
6894	lba = scsi_8btou64(cdb->addr);
6895
6896	if ((cdb->reladr & SRC16_PMI)
6897	 && (lba != 0)) {
6898		ctl_set_invalid_field(/*ctsio*/ ctsio,
6899				      /*sks_valid*/ 1,
6900				      /*command*/ 1,
6901				      /*field*/ 2,
6902				      /*bit_valid*/ 0,
6903				      /*bit*/ 0);
6904		ctl_done((union ctl_io *)ctsio);
6905		return (CTL_RETVAL_COMPLETE);
6906	}
6907
6908	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6909
6910	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6911	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6912
6913	if (sizeof(*data) < alloc_len) {
6914		ctsio->residual = alloc_len - sizeof(*data);
6915		ctsio->kern_data_len = sizeof(*data);
6916		ctsio->kern_total_len = sizeof(*data);
6917	} else {
6918		ctsio->residual = 0;
6919		ctsio->kern_data_len = alloc_len;
6920		ctsio->kern_total_len = alloc_len;
6921	}
6922	ctsio->kern_data_resid = 0;
6923	ctsio->kern_rel_offset = 0;
6924	ctsio->kern_sg_entries = 0;
6925
6926	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6927	/* XXX KDM this may not be 512 bytes... */
6928	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6929	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6930	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6931	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6932		data->lalba_lbp[0] |= SRC16_LBPME;
6933
6934	ctsio->scsi_status = SCSI_STATUS_OK;
6935
6936	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6937	ctsio->be_move_done = ctl_config_move_done;
6938	ctl_datamove((union ctl_io *)ctsio);
6939
6940	return (CTL_RETVAL_COMPLETE);
6941}
6942
6943int
6944ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
6945{
6946	struct scsi_maintenance_in *cdb;
6947	int retval;
6948	int alloc_len, total_len = 0;
6949	int num_target_port_groups, single;
6950	struct ctl_lun *lun;
6951	struct ctl_softc *softc;
6952	struct scsi_target_group_data *rtg_ptr;
6953	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6954	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6955	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6956
6957	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
6958
6959	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6960	softc = control_softc;
6961	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6962
6963	retval = CTL_RETVAL_COMPLETE;
6964
6965	single = ctl_is_single;
6966	if (single)
6967        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6968	else
6969        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6970
6971	total_len = sizeof(struct scsi_target_group_data) +
6972		sizeof(struct scsi_target_port_group_descriptor) *
6973		num_target_port_groups +
6974		sizeof(struct scsi_target_port_descriptor) *
6975		NUM_PORTS_PER_GRP * num_target_port_groups;
6976
6977	alloc_len = scsi_4btoul(cdb->length);
6978
6979	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6980
6981	ctsio->kern_sg_entries = 0;
6982
6983	if (total_len < alloc_len) {
6984		ctsio->residual = alloc_len - total_len;
6985		ctsio->kern_data_len = total_len;
6986		ctsio->kern_total_len = total_len;
6987	} else {
6988		ctsio->residual = 0;
6989		ctsio->kern_data_len = alloc_len;
6990		ctsio->kern_total_len = alloc_len;
6991	}
6992	ctsio->kern_data_resid = 0;
6993	ctsio->kern_rel_offset = 0;
6994
6995	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
6996
6997	tpg_desc_ptr1 = &rtg_ptr->groups[0];
6998	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
6999	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7000	        &tp_desc_ptr1_1->desc_list[0];
7001
7002	if (single == 0) {
7003		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7004	                &tp_desc_ptr1_2->desc_list[0];
7005		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7006		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7007	        	&tp_desc_ptr2_1->desc_list[0];
7008        } else {
7009		tpg_desc_ptr2 = NULL;
7010		tp_desc_ptr2_1 = NULL;
7011		tp_desc_ptr2_2 = NULL;
7012	}
7013
7014	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7015	if (single == 0) {
7016        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7017			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7018				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7019				tpg_desc_ptr2->pref_state =
7020					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7021			} else {
7022				tpg_desc_ptr1->pref_state =
7023					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7024				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7025			}
7026		} else {
7027			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7028				tpg_desc_ptr1->pref_state =
7029					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7030				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7031			} else {
7032				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7033				tpg_desc_ptr2->pref_state =
7034					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7035			}
7036		}
7037	} else {
7038		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7039	}
7040	tpg_desc_ptr1->support = 0;
7041	tpg_desc_ptr1->target_port_group[1] = 1;
7042	tpg_desc_ptr1->status = TPG_IMPLICIT;
7043	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7044
7045	if (single == 0) {
7046		tpg_desc_ptr2->support = 0;
7047		tpg_desc_ptr2->target_port_group[1] = 2;
7048		tpg_desc_ptr2->status = TPG_IMPLICIT;
7049		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7050
7051		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7052		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7053
7054		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7055		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7056	} else {
7057        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7058			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7059			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7060		} else {
7061			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7062			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7063		}
7064	}
7065
7066	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7067	ctsio->be_move_done = ctl_config_move_done;
7068
7069	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7070			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7071			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7072			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7073			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7074
7075	ctl_datamove((union ctl_io *)ctsio);
7076	return(retval);
7077}
7078
7079int
7080ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7081{
7082	struct ctl_lun *lun;
7083	struct scsi_report_supported_opcodes *cdb;
7084	const struct ctl_cmd_entry *entry, *sentry;
7085	struct scsi_report_supported_opcodes_all *all;
7086	struct scsi_report_supported_opcodes_descr *descr;
7087	struct scsi_report_supported_opcodes_one *one;
7088	int retval;
7089	int alloc_len, total_len;
7090	int opcode, service_action, i, j, num;
7091
7092	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7093
7094	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7095	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7096
7097	retval = CTL_RETVAL_COMPLETE;
7098
7099	opcode = cdb->requested_opcode;
7100	service_action = scsi_2btoul(cdb->requested_service_action);
7101	switch (cdb->options & RSO_OPTIONS_MASK) {
7102	case RSO_OPTIONS_ALL:
7103		num = 0;
7104		for (i = 0; i < 256; i++) {
7105			entry = &ctl_cmd_table[i];
7106			if (entry->flags & CTL_CMD_FLAG_SA5) {
7107				for (j = 0; j < 32; j++) {
7108					sentry = &((const struct ctl_cmd_entry *)
7109					    entry->execute)[j];
7110					if (ctl_cmd_applicable(
7111					    lun->be_lun->lun_type, sentry))
7112						num++;
7113				}
7114			} else {
7115				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7116				    entry))
7117					num++;
7118			}
7119		}
7120		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7121		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7122		break;
7123	case RSO_OPTIONS_OC:
7124		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7125			ctl_set_invalid_field(/*ctsio*/ ctsio,
7126					      /*sks_valid*/ 1,
7127					      /*command*/ 1,
7128					      /*field*/ 2,
7129					      /*bit_valid*/ 1,
7130					      /*bit*/ 2);
7131			ctl_done((union ctl_io *)ctsio);
7132			return (CTL_RETVAL_COMPLETE);
7133		}
7134		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7135		break;
7136	case RSO_OPTIONS_OC_SA:
7137		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7138		    service_action >= 32) {
7139			ctl_set_invalid_field(/*ctsio*/ ctsio,
7140					      /*sks_valid*/ 1,
7141					      /*command*/ 1,
7142					      /*field*/ 2,
7143					      /*bit_valid*/ 1,
7144					      /*bit*/ 2);
7145			ctl_done((union ctl_io *)ctsio);
7146			return (CTL_RETVAL_COMPLETE);
7147		}
7148		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7149		break;
7150	default:
7151		ctl_set_invalid_field(/*ctsio*/ ctsio,
7152				      /*sks_valid*/ 1,
7153				      /*command*/ 1,
7154				      /*field*/ 2,
7155				      /*bit_valid*/ 1,
7156				      /*bit*/ 2);
7157		ctl_done((union ctl_io *)ctsio);
7158		return (CTL_RETVAL_COMPLETE);
7159	}
7160
7161	alloc_len = scsi_4btoul(cdb->length);
7162
7163	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7164
7165	ctsio->kern_sg_entries = 0;
7166
7167	if (total_len < alloc_len) {
7168		ctsio->residual = alloc_len - total_len;
7169		ctsio->kern_data_len = total_len;
7170		ctsio->kern_total_len = total_len;
7171	} else {
7172		ctsio->residual = 0;
7173		ctsio->kern_data_len = alloc_len;
7174		ctsio->kern_total_len = alloc_len;
7175	}
7176	ctsio->kern_data_resid = 0;
7177	ctsio->kern_rel_offset = 0;
7178
7179	switch (cdb->options & RSO_OPTIONS_MASK) {
7180	case RSO_OPTIONS_ALL:
7181		all = (struct scsi_report_supported_opcodes_all *)
7182		    ctsio->kern_data_ptr;
7183		num = 0;
7184		for (i = 0; i < 256; i++) {
7185			entry = &ctl_cmd_table[i];
7186			if (entry->flags & CTL_CMD_FLAG_SA5) {
7187				for (j = 0; j < 32; j++) {
7188					sentry = &((const struct ctl_cmd_entry *)
7189					    entry->execute)[j];
7190					if (!ctl_cmd_applicable(
7191					    lun->be_lun->lun_type, sentry))
7192						continue;
7193					descr = &all->descr[num++];
7194					descr->opcode = i;
7195					scsi_ulto2b(j, descr->service_action);
7196					descr->flags = RSO_SERVACTV;
7197					scsi_ulto2b(sentry->length,
7198					    descr->cdb_length);
7199				}
7200			} else {
7201				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7202				    entry))
7203					continue;
7204				descr = &all->descr[num++];
7205				descr->opcode = i;
7206				scsi_ulto2b(0, descr->service_action);
7207				descr->flags = 0;
7208				scsi_ulto2b(entry->length, descr->cdb_length);
7209			}
7210		}
7211		scsi_ulto4b(
7212		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7213		    all->length);
7214		break;
7215	case RSO_OPTIONS_OC:
7216		one = (struct scsi_report_supported_opcodes_one *)
7217		    ctsio->kern_data_ptr;
7218		entry = &ctl_cmd_table[opcode];
7219		goto fill_one;
7220	case RSO_OPTIONS_OC_SA:
7221		one = (struct scsi_report_supported_opcodes_one *)
7222		    ctsio->kern_data_ptr;
7223		entry = &ctl_cmd_table[opcode];
7224		entry = &((const struct ctl_cmd_entry *)
7225		    entry->execute)[service_action];
7226fill_one:
7227		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7228			one->support = 3;
7229			scsi_ulto2b(entry->length, one->cdb_length);
7230			one->cdb_usage[0] = opcode;
7231			memcpy(&one->cdb_usage[1], entry->usage,
7232			    entry->length - 1);
7233		} else
7234			one->support = 1;
7235		break;
7236	}
7237
7238	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7239	ctsio->be_move_done = ctl_config_move_done;
7240
7241	ctl_datamove((union ctl_io *)ctsio);
7242	return(retval);
7243}
7244
7245int
7246ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7247{
7248	struct ctl_lun *lun;
7249	struct scsi_report_supported_tmf *cdb;
7250	struct scsi_report_supported_tmf_data *data;
7251	int retval;
7252	int alloc_len, total_len;
7253
7254	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7255
7256	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7257	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7258
7259	retval = CTL_RETVAL_COMPLETE;
7260
7261	total_len = sizeof(struct scsi_report_supported_tmf_data);
7262	alloc_len = scsi_4btoul(cdb->length);
7263
7264	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7265
7266	ctsio->kern_sg_entries = 0;
7267
7268	if (total_len < alloc_len) {
7269		ctsio->residual = alloc_len - total_len;
7270		ctsio->kern_data_len = total_len;
7271		ctsio->kern_total_len = total_len;
7272	} else {
7273		ctsio->residual = 0;
7274		ctsio->kern_data_len = alloc_len;
7275		ctsio->kern_total_len = alloc_len;
7276	}
7277	ctsio->kern_data_resid = 0;
7278	ctsio->kern_rel_offset = 0;
7279
7280	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7281	data->byte1 |= RST_ATS | RST_LURS | RST_TRS;
7282
7283	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7284	ctsio->be_move_done = ctl_config_move_done;
7285
7286	ctl_datamove((union ctl_io *)ctsio);
7287	return (retval);
7288}
7289
7290int
7291ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7292{
7293	struct scsi_per_res_in *cdb;
7294	int alloc_len, total_len = 0;
7295	/* struct scsi_per_res_in_rsrv in_data; */
7296	struct ctl_lun *lun;
7297	struct ctl_softc *softc;
7298
7299	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7300
7301	softc = control_softc;
7302
7303	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7304
7305	alloc_len = scsi_2btoul(cdb->length);
7306
7307	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7308
7309retry:
7310	mtx_lock(&lun->lun_lock);
7311	switch (cdb->action) {
7312	case SPRI_RK: /* read keys */
7313		total_len = sizeof(struct scsi_per_res_in_keys) +
7314			lun->pr_key_count *
7315			sizeof(struct scsi_per_res_key);
7316		break;
7317	case SPRI_RR: /* read reservation */
7318		if (lun->flags & CTL_LUN_PR_RESERVED)
7319			total_len = sizeof(struct scsi_per_res_in_rsrv);
7320		else
7321			total_len = sizeof(struct scsi_per_res_in_header);
7322		break;
7323	case SPRI_RC: /* report capabilities */
7324		total_len = sizeof(struct scsi_per_res_cap);
7325		break;
7326	default:
7327		panic("Invalid PR type %x", cdb->action);
7328	}
7329	mtx_unlock(&lun->lun_lock);
7330
7331	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7332
7333	if (total_len < alloc_len) {
7334		ctsio->residual = alloc_len - total_len;
7335		ctsio->kern_data_len = total_len;
7336		ctsio->kern_total_len = total_len;
7337	} else {
7338		ctsio->residual = 0;
7339		ctsio->kern_data_len = alloc_len;
7340		ctsio->kern_total_len = alloc_len;
7341	}
7342
7343	ctsio->kern_data_resid = 0;
7344	ctsio->kern_rel_offset = 0;
7345	ctsio->kern_sg_entries = 0;
7346
7347	mtx_lock(&lun->lun_lock);
7348	switch (cdb->action) {
7349	case SPRI_RK: { // read keys
7350        struct scsi_per_res_in_keys *res_keys;
7351		int i, key_count;
7352
7353		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7354
7355		/*
7356		 * We had to drop the lock to allocate our buffer, which
7357		 * leaves time for someone to come in with another
7358		 * persistent reservation.  (That is unlikely, though,
7359		 * since this should be the only persistent reservation
7360		 * command active right now.)
7361		 */
7362		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7363		    (lun->pr_key_count *
7364		     sizeof(struct scsi_per_res_key)))){
7365			mtx_unlock(&lun->lun_lock);
7366			free(ctsio->kern_data_ptr, M_CTL);
7367			printf("%s: reservation length changed, retrying\n",
7368			       __func__);
7369			goto retry;
7370		}
7371
7372		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7373
7374		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7375			     lun->pr_key_count, res_keys->header.length);
7376
7377		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7378			if (!lun->per_res[i].registered)
7379				continue;
7380
7381			/*
7382			 * We used lun->pr_key_count to calculate the
7383			 * size to allocate.  If it turns out the number of
7384			 * initiators with the registered flag set is
7385			 * larger than that (i.e. they haven't been kept in
7386			 * sync), we've got a problem.
7387			 */
7388			if (key_count >= lun->pr_key_count) {
7389#ifdef NEEDTOPORT
7390				csevent_log(CSC_CTL | CSC_SHELF_SW |
7391					    CTL_PR_ERROR,
7392					    csevent_LogType_Fault,
7393					    csevent_AlertLevel_Yellow,
7394					    csevent_FRU_ShelfController,
7395					    csevent_FRU_Firmware,
7396				        csevent_FRU_Unknown,
7397					    "registered keys %d >= key "
7398					    "count %d", key_count,
7399					    lun->pr_key_count);
7400#endif
7401				key_count++;
7402				continue;
7403			}
7404			memcpy(res_keys->keys[key_count].key,
7405			       lun->per_res[i].res_key.key,
7406			       ctl_min(sizeof(res_keys->keys[key_count].key),
7407			       sizeof(lun->per_res[i].res_key)));
7408			key_count++;
7409		}
7410		break;
7411	}
7412	case SPRI_RR: { // read reservation
7413		struct scsi_per_res_in_rsrv *res;
7414		int tmp_len, header_only;
7415
7416		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7417
7418		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7419
7420		if (lun->flags & CTL_LUN_PR_RESERVED)
7421		{
7422			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7423			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7424				    res->header.length);
7425			header_only = 0;
7426		} else {
7427			tmp_len = sizeof(struct scsi_per_res_in_header);
7428			scsi_ulto4b(0, res->header.length);
7429			header_only = 1;
7430		}
7431
7432		/*
7433		 * We had to drop the lock to allocate our buffer, which
7434		 * leaves time for someone to come in with another
7435		 * persistent reservation.  (That is unlikely, though,
7436		 * since this should be the only persistent reservation
7437		 * command active right now.)
7438		 */
7439		if (tmp_len != total_len) {
7440			mtx_unlock(&lun->lun_lock);
7441			free(ctsio->kern_data_ptr, M_CTL);
7442			printf("%s: reservation status changed, retrying\n",
7443			       __func__);
7444			goto retry;
7445		}
7446
7447		/*
7448		 * No reservation held, so we're done.
7449		 */
7450		if (header_only != 0)
7451			break;
7452
7453		/*
7454		 * If the registration is an All Registrants type, the key
7455		 * is 0, since it doesn't really matter.
7456		 */
7457		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7458			memcpy(res->data.reservation,
7459			       &lun->per_res[lun->pr_res_idx].res_key,
7460			       sizeof(struct scsi_per_res_key));
7461		}
7462		res->data.scopetype = lun->res_type;
7463		break;
7464	}
7465	case SPRI_RC:     //report capabilities
7466	{
7467		struct scsi_per_res_cap *res_cap;
7468		uint16_t type_mask;
7469
7470		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7471		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7472		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7473		type_mask = SPRI_TM_WR_EX_AR |
7474			    SPRI_TM_EX_AC_RO |
7475			    SPRI_TM_WR_EX_RO |
7476			    SPRI_TM_EX_AC |
7477			    SPRI_TM_WR_EX |
7478			    SPRI_TM_EX_AC_AR;
7479		scsi_ulto2b(type_mask, res_cap->type_mask);
7480		break;
7481	}
7482	case SPRI_RS: //read full status
7483	default:
7484		/*
7485		 * This is a bug, because we just checked for this above,
7486		 * and should have returned an error.
7487		 */
7488		panic("Invalid PR type %x", cdb->action);
7489		break; /* NOTREACHED */
7490	}
7491	mtx_unlock(&lun->lun_lock);
7492
7493	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7494	ctsio->be_move_done = ctl_config_move_done;
7495
7496	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7497			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7498			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7499			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7500			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7501
7502	ctl_datamove((union ctl_io *)ctsio);
7503
7504	return (CTL_RETVAL_COMPLETE);
7505}
7506
7507/*
7508 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7509 * it should return.
7510 */
7511static int
7512ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7513		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7514		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7515		struct scsi_per_res_out_parms* param)
7516{
7517	union ctl_ha_msg persis_io;
7518	int retval, i;
7519	int isc_retval;
7520
7521	retval = 0;
7522
7523	mtx_lock(&lun->lun_lock);
7524	if (sa_res_key == 0) {
7525		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7526			/* validate scope and type */
7527			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7528			     SPR_LU_SCOPE) {
7529				mtx_unlock(&lun->lun_lock);
7530				ctl_set_invalid_field(/*ctsio*/ ctsio,
7531						      /*sks_valid*/ 1,
7532						      /*command*/ 1,
7533						      /*field*/ 2,
7534						      /*bit_valid*/ 1,
7535						      /*bit*/ 4);
7536				ctl_done((union ctl_io *)ctsio);
7537				return (1);
7538			}
7539
7540		        if (type>8 || type==2 || type==4 || type==0) {
7541				mtx_unlock(&lun->lun_lock);
7542				ctl_set_invalid_field(/*ctsio*/ ctsio,
7543       	           				      /*sks_valid*/ 1,
7544						      /*command*/ 1,
7545						      /*field*/ 2,
7546						      /*bit_valid*/ 1,
7547						      /*bit*/ 0);
7548				ctl_done((union ctl_io *)ctsio);
7549				return (1);
7550		        }
7551
7552			/* temporarily unregister this nexus */
7553			lun->per_res[residx].registered = 0;
7554
7555			/*
7556			 * Unregister everybody else and build UA for
7557			 * them
7558			 */
7559			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7560				if (lun->per_res[i].registered == 0)
7561					continue;
7562
7563				if (!persis_offset
7564				 && i <CTL_MAX_INITIATORS)
7565					lun->pending_sense[i].ua_pending |=
7566						CTL_UA_REG_PREEMPT;
7567				else if (persis_offset
7568				      && i >= persis_offset)
7569					lun->pending_sense[i-persis_offset
7570						].ua_pending |=
7571						CTL_UA_REG_PREEMPT;
7572				lun->per_res[i].registered = 0;
7573				memset(&lun->per_res[i].res_key, 0,
7574				       sizeof(struct scsi_per_res_key));
7575			}
7576			lun->per_res[residx].registered = 1;
7577			lun->pr_key_count = 1;
7578			lun->res_type = type;
7579			if (lun->res_type != SPR_TYPE_WR_EX_AR
7580			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7581				lun->pr_res_idx = residx;
7582
7583			/* send msg to other side */
7584			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7585			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7586			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7587			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7588			persis_io.pr.pr_info.res_type = type;
7589			memcpy(persis_io.pr.pr_info.sa_res_key,
7590			       param->serv_act_res_key,
7591			       sizeof(param->serv_act_res_key));
7592			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7593			     &persis_io, sizeof(persis_io), 0)) >
7594			     CTL_HA_STATUS_SUCCESS) {
7595				printf("CTL:Persis Out error returned "
7596				       "from ctl_ha_msg_send %d\n",
7597				       isc_retval);
7598			}
7599		} else {
7600			/* not all registrants */
7601			mtx_unlock(&lun->lun_lock);
7602			free(ctsio->kern_data_ptr, M_CTL);
7603			ctl_set_invalid_field(ctsio,
7604					      /*sks_valid*/ 1,
7605					      /*command*/ 0,
7606					      /*field*/ 8,
7607					      /*bit_valid*/ 0,
7608					      /*bit*/ 0);
7609			ctl_done((union ctl_io *)ctsio);
7610			return (1);
7611		}
7612	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7613		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7614		int found = 0;
7615
7616		if (res_key == sa_res_key) {
7617			/* special case */
7618			/*
7619			 * The spec implies this is not good but doesn't
7620			 * say what to do. There are two choices either
7621			 * generate a res conflict or check condition
7622			 * with illegal field in parameter data. Since
7623			 * that is what is done when the sa_res_key is
7624			 * zero I'll take that approach since this has
7625			 * to do with the sa_res_key.
7626			 */
7627			mtx_unlock(&lun->lun_lock);
7628			free(ctsio->kern_data_ptr, M_CTL);
7629			ctl_set_invalid_field(ctsio,
7630					      /*sks_valid*/ 1,
7631					      /*command*/ 0,
7632					      /*field*/ 8,
7633					      /*bit_valid*/ 0,
7634					      /*bit*/ 0);
7635			ctl_done((union ctl_io *)ctsio);
7636			return (1);
7637		}
7638
7639		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7640			if (lun->per_res[i].registered
7641			 && memcmp(param->serv_act_res_key,
7642			    lun->per_res[i].res_key.key,
7643			    sizeof(struct scsi_per_res_key)) != 0)
7644				continue;
7645
7646			found = 1;
7647			lun->per_res[i].registered = 0;
7648			memset(&lun->per_res[i].res_key, 0,
7649			       sizeof(struct scsi_per_res_key));
7650			lun->pr_key_count--;
7651
7652			if (!persis_offset
7653			 && i < CTL_MAX_INITIATORS)
7654				lun->pending_sense[i].ua_pending |=
7655					CTL_UA_REG_PREEMPT;
7656			else if (persis_offset
7657			      && i >= persis_offset)
7658				lun->pending_sense[i-persis_offset].ua_pending|=
7659					CTL_UA_REG_PREEMPT;
7660		}
7661		if (!found) {
7662			mtx_unlock(&lun->lun_lock);
7663			free(ctsio->kern_data_ptr, M_CTL);
7664			ctl_set_reservation_conflict(ctsio);
7665			ctl_done((union ctl_io *)ctsio);
7666			return (CTL_RETVAL_COMPLETE);
7667		}
7668		/* send msg to other side */
7669		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7670		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7671		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7672		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7673		persis_io.pr.pr_info.res_type = type;
7674		memcpy(persis_io.pr.pr_info.sa_res_key,
7675		       param->serv_act_res_key,
7676		       sizeof(param->serv_act_res_key));
7677		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7678		     &persis_io, sizeof(persis_io), 0)) >
7679		     CTL_HA_STATUS_SUCCESS) {
7680			printf("CTL:Persis Out error returned from "
7681			       "ctl_ha_msg_send %d\n", isc_retval);
7682		}
7683	} else {
7684		/* Reserved but not all registrants */
7685		/* sa_res_key is res holder */
7686		if (memcmp(param->serv_act_res_key,
7687                   lun->per_res[lun->pr_res_idx].res_key.key,
7688                   sizeof(struct scsi_per_res_key)) == 0) {
7689			/* validate scope and type */
7690			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7691			     SPR_LU_SCOPE) {
7692				mtx_unlock(&lun->lun_lock);
7693				ctl_set_invalid_field(/*ctsio*/ ctsio,
7694						      /*sks_valid*/ 1,
7695						      /*command*/ 1,
7696						      /*field*/ 2,
7697						      /*bit_valid*/ 1,
7698						      /*bit*/ 4);
7699				ctl_done((union ctl_io *)ctsio);
7700				return (1);
7701			}
7702
7703			if (type>8 || type==2 || type==4 || type==0) {
7704				mtx_unlock(&lun->lun_lock);
7705				ctl_set_invalid_field(/*ctsio*/ ctsio,
7706						      /*sks_valid*/ 1,
7707						      /*command*/ 1,
7708						      /*field*/ 2,
7709						      /*bit_valid*/ 1,
7710						      /*bit*/ 0);
7711				ctl_done((union ctl_io *)ctsio);
7712				return (1);
7713			}
7714
7715			/*
7716			 * Do the following:
7717			 * if sa_res_key != res_key remove all
7718			 * registrants w/sa_res_key and generate UA
7719			 * for these registrants(Registrations
7720			 * Preempted) if it wasn't an exclusive
7721			 * reservation generate UA(Reservations
7722			 * Preempted) for all other registered nexuses
7723			 * if the type has changed. Establish the new
7724			 * reservation and holder. If res_key and
7725			 * sa_res_key are the same do the above
7726			 * except don't unregister the res holder.
7727			 */
7728
7729			/*
7730			 * Temporarily unregister so it won't get
7731			 * removed or UA generated
7732			 */
7733			lun->per_res[residx].registered = 0;
7734			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7735				if (lun->per_res[i].registered == 0)
7736					continue;
7737
7738				if (memcmp(param->serv_act_res_key,
7739				    lun->per_res[i].res_key.key,
7740				    sizeof(struct scsi_per_res_key)) == 0) {
7741					lun->per_res[i].registered = 0;
7742					memset(&lun->per_res[i].res_key,
7743					       0,
7744					       sizeof(struct scsi_per_res_key));
7745					lun->pr_key_count--;
7746
7747					if (!persis_offset
7748					 && i < CTL_MAX_INITIATORS)
7749						lun->pending_sense[i
7750							].ua_pending |=
7751							CTL_UA_REG_PREEMPT;
7752					else if (persis_offset
7753					      && i >= persis_offset)
7754						lun->pending_sense[
7755						  i-persis_offset].ua_pending |=
7756						  CTL_UA_REG_PREEMPT;
7757				} else if (type != lun->res_type
7758					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7759					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7760						if (!persis_offset
7761						 && i < CTL_MAX_INITIATORS)
7762							lun->pending_sense[i
7763							].ua_pending |=
7764							CTL_UA_RES_RELEASE;
7765						else if (persis_offset
7766						      && i >= persis_offset)
7767							lun->pending_sense[
7768							i-persis_offset
7769							].ua_pending |=
7770							CTL_UA_RES_RELEASE;
7771				}
7772			}
7773			lun->per_res[residx].registered = 1;
7774			lun->res_type = type;
7775			if (lun->res_type != SPR_TYPE_WR_EX_AR
7776			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7777				lun->pr_res_idx = residx;
7778			else
7779				lun->pr_res_idx =
7780					CTL_PR_ALL_REGISTRANTS;
7781
7782			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7783			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7784			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7785			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7786			persis_io.pr.pr_info.res_type = type;
7787			memcpy(persis_io.pr.pr_info.sa_res_key,
7788			       param->serv_act_res_key,
7789			       sizeof(param->serv_act_res_key));
7790			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7791			     &persis_io, sizeof(persis_io), 0)) >
7792			     CTL_HA_STATUS_SUCCESS) {
7793				printf("CTL:Persis Out error returned "
7794				       "from ctl_ha_msg_send %d\n",
7795				       isc_retval);
7796			}
7797		} else {
7798			/*
7799			 * sa_res_key is not the res holder just
7800			 * remove registrants
7801			 */
7802			int found=0;
7803
7804			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7805				if (memcmp(param->serv_act_res_key,
7806				    lun->per_res[i].res_key.key,
7807				    sizeof(struct scsi_per_res_key)) != 0)
7808					continue;
7809
7810				found = 1;
7811				lun->per_res[i].registered = 0;
7812				memset(&lun->per_res[i].res_key, 0,
7813				       sizeof(struct scsi_per_res_key));
7814				lun->pr_key_count--;
7815
7816				if (!persis_offset
7817				 && i < CTL_MAX_INITIATORS)
7818					lun->pending_sense[i].ua_pending |=
7819						CTL_UA_REG_PREEMPT;
7820				else if (persis_offset
7821				      && i >= persis_offset)
7822					lun->pending_sense[
7823						i-persis_offset].ua_pending |=
7824						CTL_UA_REG_PREEMPT;
7825			}
7826
7827			if (!found) {
7828				mtx_unlock(&lun->lun_lock);
7829				free(ctsio->kern_data_ptr, M_CTL);
7830				ctl_set_reservation_conflict(ctsio);
7831				ctl_done((union ctl_io *)ctsio);
7832		        	return (1);
7833			}
7834			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7835			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7836			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7837			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7838			persis_io.pr.pr_info.res_type = type;
7839			memcpy(persis_io.pr.pr_info.sa_res_key,
7840			       param->serv_act_res_key,
7841			       sizeof(param->serv_act_res_key));
7842			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7843			     &persis_io, sizeof(persis_io), 0)) >
7844			     CTL_HA_STATUS_SUCCESS) {
7845				printf("CTL:Persis Out error returned "
7846				       "from ctl_ha_msg_send %d\n",
7847				isc_retval);
7848			}
7849		}
7850	}
7851
7852	lun->PRGeneration++;
7853	mtx_unlock(&lun->lun_lock);
7854
7855	return (retval);
7856}
7857
7858static void
7859ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7860{
7861	int i;
7862
7863	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7864	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7865	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7866		   msg->pr.pr_info.sa_res_key,
7867		   sizeof(struct scsi_per_res_key)) != 0) {
7868		uint64_t sa_res_key;
7869		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7870
7871		if (sa_res_key == 0) {
7872			/* temporarily unregister this nexus */
7873			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7874
7875			/*
7876			 * Unregister everybody else and build UA for
7877			 * them
7878			 */
7879			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7880				if (lun->per_res[i].registered == 0)
7881					continue;
7882
7883				if (!persis_offset
7884				 && i < CTL_MAX_INITIATORS)
7885					lun->pending_sense[i].ua_pending |=
7886						CTL_UA_REG_PREEMPT;
7887				else if (persis_offset && i >= persis_offset)
7888					lun->pending_sense[i -
7889						persis_offset].ua_pending |=
7890						CTL_UA_REG_PREEMPT;
7891				lun->per_res[i].registered = 0;
7892				memset(&lun->per_res[i].res_key, 0,
7893				       sizeof(struct scsi_per_res_key));
7894			}
7895
7896			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7897			lun->pr_key_count = 1;
7898			lun->res_type = msg->pr.pr_info.res_type;
7899			if (lun->res_type != SPR_TYPE_WR_EX_AR
7900			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7901				lun->pr_res_idx = msg->pr.pr_info.residx;
7902		} else {
7903		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7904				if (memcmp(msg->pr.pr_info.sa_res_key,
7905		                   lun->per_res[i].res_key.key,
7906		                   sizeof(struct scsi_per_res_key)) != 0)
7907					continue;
7908
7909				lun->per_res[i].registered = 0;
7910				memset(&lun->per_res[i].res_key, 0,
7911				       sizeof(struct scsi_per_res_key));
7912				lun->pr_key_count--;
7913
7914				if (!persis_offset
7915				 && i < persis_offset)
7916					lun->pending_sense[i].ua_pending |=
7917						CTL_UA_REG_PREEMPT;
7918				else if (persis_offset
7919				      && i >= persis_offset)
7920					lun->pending_sense[i -
7921						persis_offset].ua_pending |=
7922						CTL_UA_REG_PREEMPT;
7923			}
7924		}
7925	} else {
7926		/*
7927		 * Temporarily unregister so it won't get removed
7928		 * or UA generated
7929		 */
7930		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7931		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7932			if (lun->per_res[i].registered == 0)
7933				continue;
7934
7935			if (memcmp(msg->pr.pr_info.sa_res_key,
7936	                   lun->per_res[i].res_key.key,
7937	                   sizeof(struct scsi_per_res_key)) == 0) {
7938				lun->per_res[i].registered = 0;
7939				memset(&lun->per_res[i].res_key, 0,
7940				       sizeof(struct scsi_per_res_key));
7941				lun->pr_key_count--;
7942				if (!persis_offset
7943				 && i < CTL_MAX_INITIATORS)
7944					lun->pending_sense[i].ua_pending |=
7945						CTL_UA_REG_PREEMPT;
7946				else if (persis_offset
7947				      && i >= persis_offset)
7948					lun->pending_sense[i -
7949						persis_offset].ua_pending |=
7950						CTL_UA_REG_PREEMPT;
7951			} else if (msg->pr.pr_info.res_type != lun->res_type
7952				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7953				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7954					if (!persis_offset
7955					 && i < persis_offset)
7956						lun->pending_sense[i
7957							].ua_pending |=
7958							CTL_UA_RES_RELEASE;
7959					else if (persis_offset
7960					      && i >= persis_offset)
7961					lun->pending_sense[i -
7962						persis_offset].ua_pending |=
7963						CTL_UA_RES_RELEASE;
7964			}
7965		}
7966		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7967		lun->res_type = msg->pr.pr_info.res_type;
7968		if (lun->res_type != SPR_TYPE_WR_EX_AR
7969		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7970			lun->pr_res_idx = msg->pr.pr_info.residx;
7971		else
7972			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7973	}
7974	lun->PRGeneration++;
7975
7976}
7977
7978
7979int
7980ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7981{
7982	int retval;
7983	int isc_retval;
7984	u_int32_t param_len;
7985	struct scsi_per_res_out *cdb;
7986	struct ctl_lun *lun;
7987	struct scsi_per_res_out_parms* param;
7988	struct ctl_softc *softc;
7989	uint32_t residx;
7990	uint64_t res_key, sa_res_key;
7991	uint8_t type;
7992	union ctl_ha_msg persis_io;
7993	int    i;
7994
7995	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7996
7997	retval = CTL_RETVAL_COMPLETE;
7998
7999	softc = control_softc;
8000
8001	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8002	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8003
8004	/*
8005	 * We only support whole-LUN scope.  The scope & type are ignored for
8006	 * register, register and ignore existing key and clear.
8007	 * We sometimes ignore scope and type on preempts too!!
8008	 * Verify reservation type here as well.
8009	 */
8010	type = cdb->scope_type & SPR_TYPE_MASK;
8011	if ((cdb->action == SPRO_RESERVE)
8012	 || (cdb->action == SPRO_RELEASE)) {
8013		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8014			ctl_set_invalid_field(/*ctsio*/ ctsio,
8015					      /*sks_valid*/ 1,
8016					      /*command*/ 1,
8017					      /*field*/ 2,
8018					      /*bit_valid*/ 1,
8019					      /*bit*/ 4);
8020			ctl_done((union ctl_io *)ctsio);
8021			return (CTL_RETVAL_COMPLETE);
8022		}
8023
8024		if (type>8 || type==2 || type==4 || type==0) {
8025			ctl_set_invalid_field(/*ctsio*/ ctsio,
8026					      /*sks_valid*/ 1,
8027					      /*command*/ 1,
8028					      /*field*/ 2,
8029					      /*bit_valid*/ 1,
8030					      /*bit*/ 0);
8031			ctl_done((union ctl_io *)ctsio);
8032			return (CTL_RETVAL_COMPLETE);
8033		}
8034	}
8035
8036	param_len = scsi_4btoul(cdb->length);
8037
8038	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8039		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8040		ctsio->kern_data_len = param_len;
8041		ctsio->kern_total_len = param_len;
8042		ctsio->kern_data_resid = 0;
8043		ctsio->kern_rel_offset = 0;
8044		ctsio->kern_sg_entries = 0;
8045		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8046		ctsio->be_move_done = ctl_config_move_done;
8047		ctl_datamove((union ctl_io *)ctsio);
8048
8049		return (CTL_RETVAL_COMPLETE);
8050	}
8051
8052	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8053
8054	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8055	res_key = scsi_8btou64(param->res_key.key);
8056	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8057
8058	/*
8059	 * Validate the reservation key here except for SPRO_REG_IGNO
8060	 * This must be done for all other service actions
8061	 */
8062	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8063		mtx_lock(&lun->lun_lock);
8064		if (lun->per_res[residx].registered) {
8065		    if (memcmp(param->res_key.key,
8066			       lun->per_res[residx].res_key.key,
8067			       ctl_min(sizeof(param->res_key),
8068			       sizeof(lun->per_res[residx].res_key))) != 0) {
8069				/*
8070				 * The current key passed in doesn't match
8071				 * the one the initiator previously
8072				 * registered.
8073				 */
8074				mtx_unlock(&lun->lun_lock);
8075				free(ctsio->kern_data_ptr, M_CTL);
8076				ctl_set_reservation_conflict(ctsio);
8077				ctl_done((union ctl_io *)ctsio);
8078				return (CTL_RETVAL_COMPLETE);
8079			}
8080		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8081			/*
8082			 * We are not registered
8083			 */
8084			mtx_unlock(&lun->lun_lock);
8085			free(ctsio->kern_data_ptr, M_CTL);
8086			ctl_set_reservation_conflict(ctsio);
8087			ctl_done((union ctl_io *)ctsio);
8088			return (CTL_RETVAL_COMPLETE);
8089		} else if (res_key != 0) {
8090			/*
8091			 * We are not registered and trying to register but
8092			 * the register key isn't zero.
8093			 */
8094			mtx_unlock(&lun->lun_lock);
8095			free(ctsio->kern_data_ptr, M_CTL);
8096			ctl_set_reservation_conflict(ctsio);
8097			ctl_done((union ctl_io *)ctsio);
8098			return (CTL_RETVAL_COMPLETE);
8099		}
8100		mtx_unlock(&lun->lun_lock);
8101	}
8102
8103	switch (cdb->action & SPRO_ACTION_MASK) {
8104	case SPRO_REGISTER:
8105	case SPRO_REG_IGNO: {
8106
8107#if 0
8108		printf("Registration received\n");
8109#endif
8110
8111		/*
8112		 * We don't support any of these options, as we report in
8113		 * the read capabilities request (see
8114		 * ctl_persistent_reserve_in(), above).
8115		 */
8116		if ((param->flags & SPR_SPEC_I_PT)
8117		 || (param->flags & SPR_ALL_TG_PT)
8118		 || (param->flags & SPR_APTPL)) {
8119			int bit_ptr;
8120
8121			if (param->flags & SPR_APTPL)
8122				bit_ptr = 0;
8123			else if (param->flags & SPR_ALL_TG_PT)
8124				bit_ptr = 2;
8125			else /* SPR_SPEC_I_PT */
8126				bit_ptr = 3;
8127
8128			free(ctsio->kern_data_ptr, M_CTL);
8129			ctl_set_invalid_field(ctsio,
8130					      /*sks_valid*/ 1,
8131					      /*command*/ 0,
8132					      /*field*/ 20,
8133					      /*bit_valid*/ 1,
8134					      /*bit*/ bit_ptr);
8135			ctl_done((union ctl_io *)ctsio);
8136			return (CTL_RETVAL_COMPLETE);
8137		}
8138
8139		mtx_lock(&lun->lun_lock);
8140
8141		/*
8142		 * The initiator wants to clear the
8143		 * key/unregister.
8144		 */
8145		if (sa_res_key == 0) {
8146			if ((res_key == 0
8147			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8148			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8149			  && !lun->per_res[residx].registered)) {
8150				mtx_unlock(&lun->lun_lock);
8151				goto done;
8152			}
8153
8154			lun->per_res[residx].registered = 0;
8155			memset(&lun->per_res[residx].res_key,
8156			       0, sizeof(lun->per_res[residx].res_key));
8157			lun->pr_key_count--;
8158
8159			if (residx == lun->pr_res_idx) {
8160				lun->flags &= ~CTL_LUN_PR_RESERVED;
8161				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8162
8163				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8164				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8165				 && lun->pr_key_count) {
8166					/*
8167					 * If the reservation is a registrants
8168					 * only type we need to generate a UA
8169					 * for other registered inits.  The
8170					 * sense code should be RESERVATIONS
8171					 * RELEASED
8172					 */
8173
8174					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8175						if (lun->per_res[
8176						    i+persis_offset].registered
8177						    == 0)
8178							continue;
8179						lun->pending_sense[i
8180							].ua_pending |=
8181							CTL_UA_RES_RELEASE;
8182					}
8183				}
8184				lun->res_type = 0;
8185			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8186				if (lun->pr_key_count==0) {
8187					lun->flags &= ~CTL_LUN_PR_RESERVED;
8188					lun->res_type = 0;
8189					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8190				}
8191			}
8192			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8193			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8194			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8195			persis_io.pr.pr_info.residx = residx;
8196			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8197			     &persis_io, sizeof(persis_io), 0 )) >
8198			     CTL_HA_STATUS_SUCCESS) {
8199				printf("CTL:Persis Out error returned from "
8200				       "ctl_ha_msg_send %d\n", isc_retval);
8201			}
8202		} else /* sa_res_key != 0 */ {
8203
8204			/*
8205			 * If we aren't registered currently then increment
8206			 * the key count and set the registered flag.
8207			 */
8208			if (!lun->per_res[residx].registered) {
8209				lun->pr_key_count++;
8210				lun->per_res[residx].registered = 1;
8211			}
8212
8213			memcpy(&lun->per_res[residx].res_key,
8214			       param->serv_act_res_key,
8215			       ctl_min(sizeof(param->serv_act_res_key),
8216			       sizeof(lun->per_res[residx].res_key)));
8217
8218			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8219			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8220			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8221			persis_io.pr.pr_info.residx = residx;
8222			memcpy(persis_io.pr.pr_info.sa_res_key,
8223			       param->serv_act_res_key,
8224			       sizeof(param->serv_act_res_key));
8225			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8226			     &persis_io, sizeof(persis_io), 0)) >
8227			     CTL_HA_STATUS_SUCCESS) {
8228				printf("CTL:Persis Out error returned from "
8229				       "ctl_ha_msg_send %d\n", isc_retval);
8230			}
8231		}
8232		lun->PRGeneration++;
8233		mtx_unlock(&lun->lun_lock);
8234
8235		break;
8236	}
8237	case SPRO_RESERVE:
8238#if 0
8239                printf("Reserve executed type %d\n", type);
8240#endif
8241		mtx_lock(&lun->lun_lock);
8242		if (lun->flags & CTL_LUN_PR_RESERVED) {
8243			/*
8244			 * if this isn't the reservation holder and it's
8245			 * not a "all registrants" type or if the type is
8246			 * different then we have a conflict
8247			 */
8248			if ((lun->pr_res_idx != residx
8249			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8250			 || lun->res_type != type) {
8251				mtx_unlock(&lun->lun_lock);
8252				free(ctsio->kern_data_ptr, M_CTL);
8253				ctl_set_reservation_conflict(ctsio);
8254				ctl_done((union ctl_io *)ctsio);
8255				return (CTL_RETVAL_COMPLETE);
8256			}
8257			mtx_unlock(&lun->lun_lock);
8258		} else /* create a reservation */ {
8259			/*
8260			 * If it's not an "all registrants" type record
8261			 * reservation holder
8262			 */
8263			if (type != SPR_TYPE_WR_EX_AR
8264			 && type != SPR_TYPE_EX_AC_AR)
8265				lun->pr_res_idx = residx; /* Res holder */
8266			else
8267				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8268
8269			lun->flags |= CTL_LUN_PR_RESERVED;
8270			lun->res_type = type;
8271
8272			mtx_unlock(&lun->lun_lock);
8273
8274			/* send msg to other side */
8275			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8276			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8277			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8278			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8279			persis_io.pr.pr_info.res_type = type;
8280			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8281			     &persis_io, sizeof(persis_io), 0)) >
8282			     CTL_HA_STATUS_SUCCESS) {
8283				printf("CTL:Persis Out error returned from "
8284				       "ctl_ha_msg_send %d\n", isc_retval);
8285			}
8286		}
8287		break;
8288
8289	case SPRO_RELEASE:
8290		mtx_lock(&lun->lun_lock);
8291		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8292			/* No reservation exists return good status */
8293			mtx_unlock(&lun->lun_lock);
8294			goto done;
8295		}
8296		/*
8297		 * Is this nexus a reservation holder?
8298		 */
8299		if (lun->pr_res_idx != residx
8300		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8301			/*
8302			 * not a res holder return good status but
8303			 * do nothing
8304			 */
8305			mtx_unlock(&lun->lun_lock);
8306			goto done;
8307		}
8308
8309		if (lun->res_type != type) {
8310			mtx_unlock(&lun->lun_lock);
8311			free(ctsio->kern_data_ptr, M_CTL);
8312			ctl_set_illegal_pr_release(ctsio);
8313			ctl_done((union ctl_io *)ctsio);
8314			return (CTL_RETVAL_COMPLETE);
8315		}
8316
8317		/* okay to release */
8318		lun->flags &= ~CTL_LUN_PR_RESERVED;
8319		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8320		lun->res_type = 0;
8321
8322		/*
8323		 * if this isn't an exclusive access
8324		 * res generate UA for all other
8325		 * registrants.
8326		 */
8327		if (type != SPR_TYPE_EX_AC
8328		 && type != SPR_TYPE_WR_EX) {
8329			/*
8330			 * temporarily unregister so we don't generate UA
8331			 */
8332			lun->per_res[residx].registered = 0;
8333
8334			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8335				if (lun->per_res[i+persis_offset].registered
8336				    == 0)
8337					continue;
8338				lun->pending_sense[i].ua_pending |=
8339					CTL_UA_RES_RELEASE;
8340			}
8341
8342			lun->per_res[residx].registered = 1;
8343		}
8344		mtx_unlock(&lun->lun_lock);
8345		/* Send msg to other side */
8346		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8347		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8348		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8349		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8350		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8351			printf("CTL:Persis Out error returned from "
8352			       "ctl_ha_msg_send %d\n", isc_retval);
8353		}
8354		break;
8355
8356	case SPRO_CLEAR:
8357		/* send msg to other side */
8358
8359		mtx_lock(&lun->lun_lock);
8360		lun->flags &= ~CTL_LUN_PR_RESERVED;
8361		lun->res_type = 0;
8362		lun->pr_key_count = 0;
8363		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8364
8365
8366		memset(&lun->per_res[residx].res_key,
8367		       0, sizeof(lun->per_res[residx].res_key));
8368		lun->per_res[residx].registered = 0;
8369
8370		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8371			if (lun->per_res[i].registered) {
8372				if (!persis_offset && i < CTL_MAX_INITIATORS)
8373					lun->pending_sense[i].ua_pending |=
8374						CTL_UA_RES_PREEMPT;
8375				else if (persis_offset && i >= persis_offset)
8376					lun->pending_sense[i-persis_offset
8377					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8378
8379				memset(&lun->per_res[i].res_key,
8380				       0, sizeof(struct scsi_per_res_key));
8381				lun->per_res[i].registered = 0;
8382			}
8383		lun->PRGeneration++;
8384		mtx_unlock(&lun->lun_lock);
8385		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8386		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8387		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8388		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8389		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8390			printf("CTL:Persis Out error returned from "
8391			       "ctl_ha_msg_send %d\n", isc_retval);
8392		}
8393		break;
8394
8395	case SPRO_PREEMPT: {
8396		int nretval;
8397
8398		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8399					  residx, ctsio, cdb, param);
8400		if (nretval != 0)
8401			return (CTL_RETVAL_COMPLETE);
8402		break;
8403	}
8404	default:
8405		panic("Invalid PR type %x", cdb->action);
8406	}
8407
8408done:
8409	free(ctsio->kern_data_ptr, M_CTL);
8410	ctl_set_success(ctsio);
8411	ctl_done((union ctl_io *)ctsio);
8412
8413	return (retval);
8414}
8415
8416/*
8417 * This routine is for handling a message from the other SC pertaining to
8418 * persistent reserve out. All the error checking will have been done
8419 * so only perorming the action need be done here to keep the two
8420 * in sync.
8421 */
8422static void
8423ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8424{
8425	struct ctl_lun *lun;
8426	struct ctl_softc *softc;
8427	int i;
8428	uint32_t targ_lun;
8429
8430	softc = control_softc;
8431
8432	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8433	lun = softc->ctl_luns[targ_lun];
8434	mtx_lock(&lun->lun_lock);
8435	switch(msg->pr.pr_info.action) {
8436	case CTL_PR_REG_KEY:
8437		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8438			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8439			lun->pr_key_count++;
8440		}
8441		lun->PRGeneration++;
8442		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8443		       msg->pr.pr_info.sa_res_key,
8444		       sizeof(struct scsi_per_res_key));
8445		break;
8446
8447	case CTL_PR_UNREG_KEY:
8448		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8449		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8450		       0, sizeof(struct scsi_per_res_key));
8451		lun->pr_key_count--;
8452
8453		/* XXX Need to see if the reservation has been released */
8454		/* if so do we need to generate UA? */
8455		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8456			lun->flags &= ~CTL_LUN_PR_RESERVED;
8457			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8458
8459			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8460			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8461			 && lun->pr_key_count) {
8462				/*
8463				 * If the reservation is a registrants
8464				 * only type we need to generate a UA
8465				 * for other registered inits.  The
8466				 * sense code should be RESERVATIONS
8467				 * RELEASED
8468				 */
8469
8470				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8471					if (lun->per_res[i+
8472					    persis_offset].registered == 0)
8473						continue;
8474
8475					lun->pending_sense[i
8476						].ua_pending |=
8477						CTL_UA_RES_RELEASE;
8478				}
8479			}
8480			lun->res_type = 0;
8481		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8482			if (lun->pr_key_count==0) {
8483				lun->flags &= ~CTL_LUN_PR_RESERVED;
8484				lun->res_type = 0;
8485				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8486			}
8487		}
8488		lun->PRGeneration++;
8489		break;
8490
8491	case CTL_PR_RESERVE:
8492		lun->flags |= CTL_LUN_PR_RESERVED;
8493		lun->res_type = msg->pr.pr_info.res_type;
8494		lun->pr_res_idx = msg->pr.pr_info.residx;
8495
8496		break;
8497
8498	case CTL_PR_RELEASE:
8499		/*
8500		 * if this isn't an exclusive access res generate UA for all
8501		 * other registrants.
8502		 */
8503		if (lun->res_type != SPR_TYPE_EX_AC
8504		 && lun->res_type != SPR_TYPE_WR_EX) {
8505			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8506				if (lun->per_res[i+persis_offset].registered)
8507					lun->pending_sense[i].ua_pending |=
8508						CTL_UA_RES_RELEASE;
8509		}
8510
8511		lun->flags &= ~CTL_LUN_PR_RESERVED;
8512		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8513		lun->res_type = 0;
8514		break;
8515
8516	case CTL_PR_PREEMPT:
8517		ctl_pro_preempt_other(lun, msg);
8518		break;
8519	case CTL_PR_CLEAR:
8520		lun->flags &= ~CTL_LUN_PR_RESERVED;
8521		lun->res_type = 0;
8522		lun->pr_key_count = 0;
8523		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8524
8525		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8526			if (lun->per_res[i].registered == 0)
8527				continue;
8528			if (!persis_offset
8529			 && i < CTL_MAX_INITIATORS)
8530				lun->pending_sense[i].ua_pending |=
8531					CTL_UA_RES_PREEMPT;
8532			else if (persis_offset
8533			      && i >= persis_offset)
8534   				lun->pending_sense[i-persis_offset].ua_pending|=
8535					CTL_UA_RES_PREEMPT;
8536			memset(&lun->per_res[i].res_key, 0,
8537			       sizeof(struct scsi_per_res_key));
8538			lun->per_res[i].registered = 0;
8539		}
8540		lun->PRGeneration++;
8541		break;
8542	}
8543
8544	mtx_unlock(&lun->lun_lock);
8545}
8546
8547int
8548ctl_read_write(struct ctl_scsiio *ctsio)
8549{
8550	struct ctl_lun *lun;
8551	struct ctl_lba_len_flags *lbalen;
8552	uint64_t lba;
8553	uint32_t num_blocks;
8554	int fua, dpo;
8555	int retval;
8556	int isread;
8557
8558	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8559
8560	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8561
8562	fua = 0;
8563	dpo = 0;
8564
8565	retval = CTL_RETVAL_COMPLETE;
8566
8567	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8568	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8569	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8570		uint32_t residx;
8571
8572		/*
8573		 * XXX KDM need a lock here.
8574		 */
8575		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8576		if ((lun->res_type == SPR_TYPE_EX_AC
8577		  && residx != lun->pr_res_idx)
8578		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8579		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8580		  && !lun->per_res[residx].registered)) {
8581			ctl_set_reservation_conflict(ctsio);
8582			ctl_done((union ctl_io *)ctsio);
8583			return (CTL_RETVAL_COMPLETE);
8584	        }
8585	}
8586
8587	switch (ctsio->cdb[0]) {
8588	case READ_6:
8589	case WRITE_6: {
8590		struct scsi_rw_6 *cdb;
8591
8592		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8593
8594		lba = scsi_3btoul(cdb->addr);
8595		/* only 5 bits are valid in the most significant address byte */
8596		lba &= 0x1fffff;
8597		num_blocks = cdb->length;
8598		/*
8599		 * This is correct according to SBC-2.
8600		 */
8601		if (num_blocks == 0)
8602			num_blocks = 256;
8603		break;
8604	}
8605	case READ_10:
8606	case WRITE_10: {
8607		struct scsi_rw_10 *cdb;
8608
8609		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8610
8611		if (cdb->byte2 & SRW10_FUA)
8612			fua = 1;
8613		if (cdb->byte2 & SRW10_DPO)
8614			dpo = 1;
8615
8616		lba = scsi_4btoul(cdb->addr);
8617		num_blocks = scsi_2btoul(cdb->length);
8618		break;
8619	}
8620	case WRITE_VERIFY_10: {
8621		struct scsi_write_verify_10 *cdb;
8622
8623		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8624
8625		/*
8626		 * XXX KDM we should do actual write verify support at some
8627		 * point.  This is obviously fake, we're just translating
8628		 * things to a write.  So we don't even bother checking the
8629		 * BYTCHK field, since we don't do any verification.  If
8630		 * the user asks for it, we'll just pretend we did it.
8631		 */
8632		if (cdb->byte2 & SWV_DPO)
8633			dpo = 1;
8634
8635		lba = scsi_4btoul(cdb->addr);
8636		num_blocks = scsi_2btoul(cdb->length);
8637		break;
8638	}
8639	case READ_12:
8640	case WRITE_12: {
8641		struct scsi_rw_12 *cdb;
8642
8643		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8644
8645		if (cdb->byte2 & SRW12_FUA)
8646			fua = 1;
8647		if (cdb->byte2 & SRW12_DPO)
8648			dpo = 1;
8649		lba = scsi_4btoul(cdb->addr);
8650		num_blocks = scsi_4btoul(cdb->length);
8651		break;
8652	}
8653	case WRITE_VERIFY_12: {
8654		struct scsi_write_verify_12 *cdb;
8655
8656		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8657
8658		if (cdb->byte2 & SWV_DPO)
8659			dpo = 1;
8660
8661		lba = scsi_4btoul(cdb->addr);
8662		num_blocks = scsi_4btoul(cdb->length);
8663
8664		break;
8665	}
8666	case READ_16:
8667	case WRITE_16: {
8668		struct scsi_rw_16 *cdb;
8669
8670		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8671
8672		if (cdb->byte2 & SRW12_FUA)
8673			fua = 1;
8674		if (cdb->byte2 & SRW12_DPO)
8675			dpo = 1;
8676
8677		lba = scsi_8btou64(cdb->addr);
8678		num_blocks = scsi_4btoul(cdb->length);
8679		break;
8680	}
8681	case WRITE_VERIFY_16: {
8682		struct scsi_write_verify_16 *cdb;
8683
8684		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8685
8686		if (cdb->byte2 & SWV_DPO)
8687			dpo = 1;
8688
8689		lba = scsi_8btou64(cdb->addr);
8690		num_blocks = scsi_4btoul(cdb->length);
8691		break;
8692	}
8693	default:
8694		/*
8695		 * We got a command we don't support.  This shouldn't
8696		 * happen, commands should be filtered out above us.
8697		 */
8698		ctl_set_invalid_opcode(ctsio);
8699		ctl_done((union ctl_io *)ctsio);
8700
8701		return (CTL_RETVAL_COMPLETE);
8702		break; /* NOTREACHED */
8703	}
8704
8705	/*
8706	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8707	 * interesting for us, but if RAIDCore is in write-back mode,
8708	 * getting it to do write-through for a particular transaction may
8709	 * not be possible.
8710	 */
8711
8712	/*
8713	 * The first check is to make sure we're in bounds, the second
8714	 * check is to catch wrap-around problems.  If the lba + num blocks
8715	 * is less than the lba, then we've wrapped around and the block
8716	 * range is invalid anyway.
8717	 */
8718	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8719	 || ((lba + num_blocks) < lba)) {
8720		ctl_set_lba_out_of_range(ctsio);
8721		ctl_done((union ctl_io *)ctsio);
8722		return (CTL_RETVAL_COMPLETE);
8723	}
8724
8725	/*
8726	 * According to SBC-3, a transfer length of 0 is not an error.
8727	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8728	 * translates to 256 blocks for those commands.
8729	 */
8730	if (num_blocks == 0) {
8731		ctl_set_success(ctsio);
8732		ctl_done((union ctl_io *)ctsio);
8733		return (CTL_RETVAL_COMPLETE);
8734	}
8735
8736	lbalen = (struct ctl_lba_len_flags *)
8737	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8738	lbalen->lba = lba;
8739	lbalen->len = num_blocks;
8740	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8741
8742	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8743	ctsio->kern_rel_offset = 0;
8744
8745	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8746
8747	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8748
8749	return (retval);
8750}
8751
8752static int
8753ctl_cnw_cont(union ctl_io *io)
8754{
8755	struct ctl_scsiio *ctsio;
8756	struct ctl_lun *lun;
8757	struct ctl_lba_len_flags *lbalen;
8758	int retval;
8759
8760	ctsio = &io->scsiio;
8761	ctsio->io_hdr.status = CTL_STATUS_NONE;
8762	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8763	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8764	lbalen = (struct ctl_lba_len_flags *)
8765	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8766	lbalen->flags = CTL_LLF_WRITE;
8767
8768	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8769	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8770	return (retval);
8771}
8772
8773int
8774ctl_cnw(struct ctl_scsiio *ctsio)
8775{
8776	struct ctl_lun *lun;
8777	struct ctl_lba_len_flags *lbalen;
8778	uint64_t lba;
8779	uint32_t num_blocks;
8780	int fua, dpo;
8781	int retval;
8782
8783	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8784
8785	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8786
8787	fua = 0;
8788	dpo = 0;
8789
8790	retval = CTL_RETVAL_COMPLETE;
8791
8792	switch (ctsio->cdb[0]) {
8793	case COMPARE_AND_WRITE: {
8794		struct scsi_compare_and_write *cdb;
8795
8796		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8797
8798		if (cdb->byte2 & SRW10_FUA)
8799			fua = 1;
8800		if (cdb->byte2 & SRW10_DPO)
8801			dpo = 1;
8802		lba = scsi_8btou64(cdb->addr);
8803		num_blocks = cdb->length;
8804		break;
8805	}
8806	default:
8807		/*
8808		 * We got a command we don't support.  This shouldn't
8809		 * happen, commands should be filtered out above us.
8810		 */
8811		ctl_set_invalid_opcode(ctsio);
8812		ctl_done((union ctl_io *)ctsio);
8813
8814		return (CTL_RETVAL_COMPLETE);
8815		break; /* NOTREACHED */
8816	}
8817
8818	/*
8819	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8820	 * interesting for us, but if RAIDCore is in write-back mode,
8821	 * getting it to do write-through for a particular transaction may
8822	 * not be possible.
8823	 */
8824
8825	/*
8826	 * The first check is to make sure we're in bounds, the second
8827	 * check is to catch wrap-around problems.  If the lba + num blocks
8828	 * is less than the lba, then we've wrapped around and the block
8829	 * range is invalid anyway.
8830	 */
8831	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8832	 || ((lba + num_blocks) < lba)) {
8833		ctl_set_lba_out_of_range(ctsio);
8834		ctl_done((union ctl_io *)ctsio);
8835		return (CTL_RETVAL_COMPLETE);
8836	}
8837
8838	/*
8839	 * According to SBC-3, a transfer length of 0 is not an error.
8840	 */
8841	if (num_blocks == 0) {
8842		ctl_set_success(ctsio);
8843		ctl_done((union ctl_io *)ctsio);
8844		return (CTL_RETVAL_COMPLETE);
8845	}
8846
8847	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8848	ctsio->kern_rel_offset = 0;
8849
8850	/*
8851	 * Set the IO_CONT flag, so that if this I/O gets passed to
8852	 * ctl_data_submit_done(), it'll get passed back to
8853	 * ctl_ctl_cnw_cont() for further processing.
8854	 */
8855	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8856	ctsio->io_cont = ctl_cnw_cont;
8857
8858	lbalen = (struct ctl_lba_len_flags *)
8859	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8860	lbalen->lba = lba;
8861	lbalen->len = num_blocks;
8862	lbalen->flags = CTL_LLF_COMPARE;
8863
8864	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8865	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8866	return (retval);
8867}
8868
8869int
8870ctl_verify(struct ctl_scsiio *ctsio)
8871{
8872	struct ctl_lun *lun;
8873	struct ctl_lba_len_flags *lbalen;
8874	uint64_t lba;
8875	uint32_t num_blocks;
8876	int bytchk, dpo;
8877	int retval;
8878
8879	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8880
8881	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8882
8883	bytchk = 0;
8884	dpo = 0;
8885	retval = CTL_RETVAL_COMPLETE;
8886
8887	switch (ctsio->cdb[0]) {
8888	case VERIFY_10: {
8889		struct scsi_verify_10 *cdb;
8890
8891		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8892		if (cdb->byte2 & SVFY_BYTCHK)
8893			bytchk = 1;
8894		if (cdb->byte2 & SVFY_DPO)
8895			dpo = 1;
8896		lba = scsi_4btoul(cdb->addr);
8897		num_blocks = scsi_2btoul(cdb->length);
8898		break;
8899	}
8900	case VERIFY_12: {
8901		struct scsi_verify_12 *cdb;
8902
8903		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8904		if (cdb->byte2 & SVFY_BYTCHK)
8905			bytchk = 1;
8906		if (cdb->byte2 & SVFY_DPO)
8907			dpo = 1;
8908		lba = scsi_4btoul(cdb->addr);
8909		num_blocks = scsi_4btoul(cdb->length);
8910		break;
8911	}
8912	case VERIFY_16: {
8913		struct scsi_rw_16 *cdb;
8914
8915		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8916		if (cdb->byte2 & SVFY_BYTCHK)
8917			bytchk = 1;
8918		if (cdb->byte2 & SVFY_DPO)
8919			dpo = 1;
8920		lba = scsi_8btou64(cdb->addr);
8921		num_blocks = scsi_4btoul(cdb->length);
8922		break;
8923	}
8924	default:
8925		/*
8926		 * We got a command we don't support.  This shouldn't
8927		 * happen, commands should be filtered out above us.
8928		 */
8929		ctl_set_invalid_opcode(ctsio);
8930		ctl_done((union ctl_io *)ctsio);
8931		return (CTL_RETVAL_COMPLETE);
8932	}
8933
8934	/*
8935	 * The first check is to make sure we're in bounds, the second
8936	 * check is to catch wrap-around problems.  If the lba + num blocks
8937	 * is less than the lba, then we've wrapped around and the block
8938	 * range is invalid anyway.
8939	 */
8940	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8941	 || ((lba + num_blocks) < lba)) {
8942		ctl_set_lba_out_of_range(ctsio);
8943		ctl_done((union ctl_io *)ctsio);
8944		return (CTL_RETVAL_COMPLETE);
8945	}
8946
8947	/*
8948	 * According to SBC-3, a transfer length of 0 is not an error.
8949	 */
8950	if (num_blocks == 0) {
8951		ctl_set_success(ctsio);
8952		ctl_done((union ctl_io *)ctsio);
8953		return (CTL_RETVAL_COMPLETE);
8954	}
8955
8956	lbalen = (struct ctl_lba_len_flags *)
8957	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8958	lbalen->lba = lba;
8959	lbalen->len = num_blocks;
8960	if (bytchk) {
8961		lbalen->flags = CTL_LLF_COMPARE;
8962		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8963	} else {
8964		lbalen->flags = CTL_LLF_VERIFY;
8965		ctsio->kern_total_len = 0;
8966	}
8967	ctsio->kern_rel_offset = 0;
8968
8969	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8970	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8971	return (retval);
8972}
8973
8974int
8975ctl_report_luns(struct ctl_scsiio *ctsio)
8976{
8977	struct scsi_report_luns *cdb;
8978	struct scsi_report_luns_data *lun_data;
8979	struct ctl_lun *lun, *request_lun;
8980	int num_luns, retval;
8981	uint32_t alloc_len, lun_datalen;
8982	int num_filled, well_known;
8983	uint32_t initidx, targ_lun_id, lun_id;
8984
8985	retval = CTL_RETVAL_COMPLETE;
8986	well_known = 0;
8987
8988	cdb = (struct scsi_report_luns *)ctsio->cdb;
8989
8990	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8991
8992	mtx_lock(&control_softc->ctl_lock);
8993	num_luns = control_softc->num_luns;
8994	mtx_unlock(&control_softc->ctl_lock);
8995
8996	switch (cdb->select_report) {
8997	case RPL_REPORT_DEFAULT:
8998	case RPL_REPORT_ALL:
8999		break;
9000	case RPL_REPORT_WELLKNOWN:
9001		well_known = 1;
9002		num_luns = 0;
9003		break;
9004	default:
9005		ctl_set_invalid_field(ctsio,
9006				      /*sks_valid*/ 1,
9007				      /*command*/ 1,
9008				      /*field*/ 2,
9009				      /*bit_valid*/ 0,
9010				      /*bit*/ 0);
9011		ctl_done((union ctl_io *)ctsio);
9012		return (retval);
9013		break; /* NOTREACHED */
9014	}
9015
9016	alloc_len = scsi_4btoul(cdb->length);
9017	/*
9018	 * The initiator has to allocate at least 16 bytes for this request,
9019	 * so he can at least get the header and the first LUN.  Otherwise
9020	 * we reject the request (per SPC-3 rev 14, section 6.21).
9021	 */
9022	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9023	    sizeof(struct scsi_report_luns_lundata))) {
9024		ctl_set_invalid_field(ctsio,
9025				      /*sks_valid*/ 1,
9026				      /*command*/ 1,
9027				      /*field*/ 6,
9028				      /*bit_valid*/ 0,
9029				      /*bit*/ 0);
9030		ctl_done((union ctl_io *)ctsio);
9031		return (retval);
9032	}
9033
9034	request_lun = (struct ctl_lun *)
9035		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9036
9037	lun_datalen = sizeof(*lun_data) +
9038		(num_luns * sizeof(struct scsi_report_luns_lundata));
9039
9040	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9041	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9042	ctsio->kern_sg_entries = 0;
9043
9044	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9045
9046	mtx_lock(&control_softc->ctl_lock);
9047	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9048		lun_id = targ_lun_id;
9049		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9050			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9051		if (lun_id >= CTL_MAX_LUNS)
9052			continue;
9053		lun = control_softc->ctl_luns[lun_id];
9054		if (lun == NULL)
9055			continue;
9056
9057		if (targ_lun_id <= 0xff) {
9058			/*
9059			 * Peripheral addressing method, bus number 0.
9060			 */
9061			lun_data->luns[num_filled].lundata[0] =
9062				RPL_LUNDATA_ATYP_PERIPH;
9063			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9064			num_filled++;
9065		} else if (targ_lun_id <= 0x3fff) {
9066			/*
9067			 * Flat addressing method.
9068			 */
9069			lun_data->luns[num_filled].lundata[0] =
9070				RPL_LUNDATA_ATYP_FLAT |
9071				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9072#ifdef OLDCTLHEADERS
9073				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9074				(targ_lun_id & SRLD_BUS_LUN_MASK);
9075#endif
9076			lun_data->luns[num_filled].lundata[1] =
9077#ifdef OLDCTLHEADERS
9078				targ_lun_id >> SRLD_BUS_LUN_BITS;
9079#endif
9080				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9081			num_filled++;
9082		} else {
9083			printf("ctl_report_luns: bogus LUN number %jd, "
9084			       "skipping\n", (intmax_t)targ_lun_id);
9085		}
9086		/*
9087		 * According to SPC-3, rev 14 section 6.21:
9088		 *
9089		 * "The execution of a REPORT LUNS command to any valid and
9090		 * installed logical unit shall clear the REPORTED LUNS DATA
9091		 * HAS CHANGED unit attention condition for all logical
9092		 * units of that target with respect to the requesting
9093		 * initiator. A valid and installed logical unit is one
9094		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9095		 * INQUIRY data (see 6.4.2)."
9096		 *
9097		 * If request_lun is NULL, the LUN this report luns command
9098		 * was issued to is either disabled or doesn't exist. In that
9099		 * case, we shouldn't clear any pending lun change unit
9100		 * attention.
9101		 */
9102		if (request_lun != NULL) {
9103			mtx_lock(&lun->lun_lock);
9104			lun->pending_sense[initidx].ua_pending &=
9105				~CTL_UA_LUN_CHANGE;
9106			mtx_unlock(&lun->lun_lock);
9107		}
9108	}
9109	mtx_unlock(&control_softc->ctl_lock);
9110
9111	/*
9112	 * It's quite possible that we've returned fewer LUNs than we allocated
9113	 * space for.  Trim it.
9114	 */
9115	lun_datalen = sizeof(*lun_data) +
9116		(num_filled * sizeof(struct scsi_report_luns_lundata));
9117
9118	if (lun_datalen < alloc_len) {
9119		ctsio->residual = alloc_len - lun_datalen;
9120		ctsio->kern_data_len = lun_datalen;
9121		ctsio->kern_total_len = lun_datalen;
9122	} else {
9123		ctsio->residual = 0;
9124		ctsio->kern_data_len = alloc_len;
9125		ctsio->kern_total_len = alloc_len;
9126	}
9127	ctsio->kern_data_resid = 0;
9128	ctsio->kern_rel_offset = 0;
9129	ctsio->kern_sg_entries = 0;
9130
9131	/*
9132	 * We set this to the actual data length, regardless of how much
9133	 * space we actually have to return results.  If the user looks at
9134	 * this value, he'll know whether or not he allocated enough space
9135	 * and reissue the command if necessary.  We don't support well
9136	 * known logical units, so if the user asks for that, return none.
9137	 */
9138	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9139
9140	/*
9141	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9142	 * this request.
9143	 */
9144	ctsio->scsi_status = SCSI_STATUS_OK;
9145
9146	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9147	ctsio->be_move_done = ctl_config_move_done;
9148	ctl_datamove((union ctl_io *)ctsio);
9149
9150	return (retval);
9151}
9152
9153int
9154ctl_request_sense(struct ctl_scsiio *ctsio)
9155{
9156	struct scsi_request_sense *cdb;
9157	struct scsi_sense_data *sense_ptr;
9158	struct ctl_lun *lun;
9159	uint32_t initidx;
9160	int have_error;
9161	scsi_sense_data_type sense_format;
9162
9163	cdb = (struct scsi_request_sense *)ctsio->cdb;
9164
9165	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9166
9167	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9168
9169	/*
9170	 * Determine which sense format the user wants.
9171	 */
9172	if (cdb->byte2 & SRS_DESC)
9173		sense_format = SSD_TYPE_DESC;
9174	else
9175		sense_format = SSD_TYPE_FIXED;
9176
9177	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9178	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9179	ctsio->kern_sg_entries = 0;
9180
9181	/*
9182	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9183	 * larger than the largest allowed value for the length field in the
9184	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9185	 */
9186	ctsio->residual = 0;
9187	ctsio->kern_data_len = cdb->length;
9188	ctsio->kern_total_len = cdb->length;
9189
9190	ctsio->kern_data_resid = 0;
9191	ctsio->kern_rel_offset = 0;
9192	ctsio->kern_sg_entries = 0;
9193
9194	/*
9195	 * If we don't have a LUN, we don't have any pending sense.
9196	 */
9197	if (lun == NULL)
9198		goto no_sense;
9199
9200	have_error = 0;
9201	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9202	/*
9203	 * Check for pending sense, and then for pending unit attentions.
9204	 * Pending sense gets returned first, then pending unit attentions.
9205	 */
9206	mtx_lock(&lun->lun_lock);
9207	if (ctl_is_set(lun->have_ca, initidx)) {
9208		scsi_sense_data_type stored_format;
9209
9210		/*
9211		 * Check to see which sense format was used for the stored
9212		 * sense data.
9213		 */
9214		stored_format = scsi_sense_type(
9215		    &lun->pending_sense[initidx].sense);
9216
9217		/*
9218		 * If the user requested a different sense format than the
9219		 * one we stored, then we need to convert it to the other
9220		 * format.  If we're going from descriptor to fixed format
9221		 * sense data, we may lose things in translation, depending
9222		 * on what options were used.
9223		 *
9224		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9225		 * for some reason we'll just copy it out as-is.
9226		 */
9227		if ((stored_format == SSD_TYPE_FIXED)
9228		 && (sense_format == SSD_TYPE_DESC))
9229			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9230			    &lun->pending_sense[initidx].sense,
9231			    (struct scsi_sense_data_desc *)sense_ptr);
9232		else if ((stored_format == SSD_TYPE_DESC)
9233		      && (sense_format == SSD_TYPE_FIXED))
9234			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9235			    &lun->pending_sense[initidx].sense,
9236			    (struct scsi_sense_data_fixed *)sense_ptr);
9237		else
9238			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9239			       ctl_min(sizeof(*sense_ptr),
9240			       sizeof(lun->pending_sense[initidx].sense)));
9241
9242		ctl_clear_mask(lun->have_ca, initidx);
9243		have_error = 1;
9244	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9245		ctl_ua_type ua_type;
9246
9247		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9248				       sense_ptr, sense_format);
9249		if (ua_type != CTL_UA_NONE) {
9250			have_error = 1;
9251			/* We're reporting this UA, so clear it */
9252			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9253		}
9254	}
9255	mtx_unlock(&lun->lun_lock);
9256
9257	/*
9258	 * We already have a pending error, return it.
9259	 */
9260	if (have_error != 0) {
9261		/*
9262		 * We report the SCSI status as OK, since the status of the
9263		 * request sense command itself is OK.
9264		 */
9265		ctsio->scsi_status = SCSI_STATUS_OK;
9266
9267		/*
9268		 * We report 0 for the sense length, because we aren't doing
9269		 * autosense in this case.  We're reporting sense as
9270		 * parameter data.
9271		 */
9272		ctsio->sense_len = 0;
9273		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9274		ctsio->be_move_done = ctl_config_move_done;
9275		ctl_datamove((union ctl_io *)ctsio);
9276
9277		return (CTL_RETVAL_COMPLETE);
9278	}
9279
9280no_sense:
9281
9282	/*
9283	 * No sense information to report, so we report that everything is
9284	 * okay.
9285	 */
9286	ctl_set_sense_data(sense_ptr,
9287			   lun,
9288			   sense_format,
9289			   /*current_error*/ 1,
9290			   /*sense_key*/ SSD_KEY_NO_SENSE,
9291			   /*asc*/ 0x00,
9292			   /*ascq*/ 0x00,
9293			   SSD_ELEM_NONE);
9294
9295	ctsio->scsi_status = SCSI_STATUS_OK;
9296
9297	/*
9298	 * We report 0 for the sense length, because we aren't doing
9299	 * autosense in this case.  We're reporting sense as parameter data.
9300	 */
9301	ctsio->sense_len = 0;
9302	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9303	ctsio->be_move_done = ctl_config_move_done;
9304	ctl_datamove((union ctl_io *)ctsio);
9305
9306	return (CTL_RETVAL_COMPLETE);
9307}
9308
9309int
9310ctl_tur(struct ctl_scsiio *ctsio)
9311{
9312	struct ctl_lun *lun;
9313
9314	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9315
9316	CTL_DEBUG_PRINT(("ctl_tur\n"));
9317
9318	if (lun == NULL)
9319		return (EINVAL);
9320
9321	ctsio->scsi_status = SCSI_STATUS_OK;
9322	ctsio->io_hdr.status = CTL_SUCCESS;
9323
9324	ctl_done((union ctl_io *)ctsio);
9325
9326	return (CTL_RETVAL_COMPLETE);
9327}
9328
9329#ifdef notyet
9330static int
9331ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9332{
9333
9334}
9335#endif
9336
9337static int
9338ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9339{
9340	struct scsi_vpd_supported_pages *pages;
9341	int sup_page_size;
9342	struct ctl_lun *lun;
9343
9344	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9345
9346	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9347	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9348	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9349	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9350	ctsio->kern_sg_entries = 0;
9351
9352	if (sup_page_size < alloc_len) {
9353		ctsio->residual = alloc_len - sup_page_size;
9354		ctsio->kern_data_len = sup_page_size;
9355		ctsio->kern_total_len = sup_page_size;
9356	} else {
9357		ctsio->residual = 0;
9358		ctsio->kern_data_len = alloc_len;
9359		ctsio->kern_total_len = alloc_len;
9360	}
9361	ctsio->kern_data_resid = 0;
9362	ctsio->kern_rel_offset = 0;
9363	ctsio->kern_sg_entries = 0;
9364
9365	/*
9366	 * The control device is always connected.  The disk device, on the
9367	 * other hand, may not be online all the time.  Need to change this
9368	 * to figure out whether the disk device is actually online or not.
9369	 */
9370	if (lun != NULL)
9371		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9372				lun->be_lun->lun_type;
9373	else
9374		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9375
9376	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9377	/* Supported VPD pages */
9378	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9379	/* Serial Number */
9380	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9381	/* Device Identification */
9382	pages->page_list[2] = SVPD_DEVICE_ID;
9383	/* Block limits */
9384	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9385	/* Logical Block Provisioning */
9386	pages->page_list[4] = SVPD_LBP;
9387
9388	ctsio->scsi_status = SCSI_STATUS_OK;
9389
9390	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9391	ctsio->be_move_done = ctl_config_move_done;
9392	ctl_datamove((union ctl_io *)ctsio);
9393
9394	return (CTL_RETVAL_COMPLETE);
9395}
9396
9397static int
9398ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9399{
9400	struct scsi_vpd_unit_serial_number *sn_ptr;
9401	struct ctl_lun *lun;
9402
9403	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9404
9405	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9406	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9407	ctsio->kern_sg_entries = 0;
9408
9409	if (sizeof(*sn_ptr) < alloc_len) {
9410		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9411		ctsio->kern_data_len = sizeof(*sn_ptr);
9412		ctsio->kern_total_len = sizeof(*sn_ptr);
9413	} else {
9414		ctsio->residual = 0;
9415		ctsio->kern_data_len = alloc_len;
9416		ctsio->kern_total_len = alloc_len;
9417	}
9418	ctsio->kern_data_resid = 0;
9419	ctsio->kern_rel_offset = 0;
9420	ctsio->kern_sg_entries = 0;
9421
9422	/*
9423	 * The control device is always connected.  The disk device, on the
9424	 * other hand, may not be online all the time.  Need to change this
9425	 * to figure out whether the disk device is actually online or not.
9426	 */
9427	if (lun != NULL)
9428		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9429				  lun->be_lun->lun_type;
9430	else
9431		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9432
9433	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9434	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9435	/*
9436	 * If we don't have a LUN, we just leave the serial number as
9437	 * all spaces.
9438	 */
9439	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9440	if (lun != NULL) {
9441		strncpy((char *)sn_ptr->serial_num,
9442			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9443	}
9444	ctsio->scsi_status = SCSI_STATUS_OK;
9445
9446	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9447	ctsio->be_move_done = ctl_config_move_done;
9448	ctl_datamove((union ctl_io *)ctsio);
9449
9450	return (CTL_RETVAL_COMPLETE);
9451}
9452
9453
9454static int
9455ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9456{
9457	struct scsi_vpd_device_id *devid_ptr;
9458	struct scsi_vpd_id_descriptor *desc, *desc1;
9459	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9460	struct scsi_vpd_id_t10 *t10id;
9461	struct ctl_softc *ctl_softc;
9462	struct ctl_lun *lun;
9463	struct ctl_frontend *fe;
9464	char *val;
9465	int data_len, devid_len;
9466
9467	ctl_softc = control_softc;
9468
9469	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9470
9471	if (fe->devid != NULL)
9472		return ((fe->devid)(ctsio, alloc_len));
9473
9474	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9475
9476	if (lun == NULL) {
9477		devid_len = CTL_DEVID_MIN_LEN;
9478	} else {
9479		devid_len = max(CTL_DEVID_MIN_LEN,
9480		    strnlen(lun->be_lun->device_id, CTL_DEVID_LEN));
9481	}
9482
9483	data_len = sizeof(struct scsi_vpd_device_id) +
9484		sizeof(struct scsi_vpd_id_descriptor) +
9485		sizeof(struct scsi_vpd_id_t10) + devid_len +
9486		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9487		sizeof(struct scsi_vpd_id_descriptor) +
9488		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9489		sizeof(struct scsi_vpd_id_descriptor) +
9490		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9491
9492	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9493	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9494	ctsio->kern_sg_entries = 0;
9495
9496	if (data_len < alloc_len) {
9497		ctsio->residual = alloc_len - data_len;
9498		ctsio->kern_data_len = data_len;
9499		ctsio->kern_total_len = data_len;
9500	} else {
9501		ctsio->residual = 0;
9502		ctsio->kern_data_len = alloc_len;
9503		ctsio->kern_total_len = alloc_len;
9504	}
9505	ctsio->kern_data_resid = 0;
9506	ctsio->kern_rel_offset = 0;
9507	ctsio->kern_sg_entries = 0;
9508
9509	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9510	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9511	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9512		sizeof(struct scsi_vpd_id_t10) + devid_len);
9513	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9514	          CTL_WWPN_LEN);
9515	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9516	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9517
9518	/*
9519	 * The control device is always connected.  The disk device, on the
9520	 * other hand, may not be online all the time.
9521	 */
9522	if (lun != NULL)
9523		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9524				     lun->be_lun->lun_type;
9525	else
9526		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9527
9528	devid_ptr->page_code = SVPD_DEVICE_ID;
9529
9530	scsi_ulto2b(data_len - 4, devid_ptr->length);
9531
9532	/*
9533	 * For Fibre channel,
9534	 */
9535	if (fe->port_type == CTL_PORT_FC)
9536	{
9537		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9538				      SVPD_ID_CODESET_ASCII;
9539        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9540		              SVPD_ID_CODESET_BINARY;
9541	}
9542	else
9543	{
9544		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9545				      SVPD_ID_CODESET_ASCII;
9546        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9547		              SVPD_ID_CODESET_BINARY;
9548	}
9549	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9550
9551	/*
9552	 * We're using a LUN association here.  i.e., this device ID is a
9553	 * per-LUN identifier.
9554	 */
9555	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9556	desc->length = sizeof(*t10id) + devid_len;
9557	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9558		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9559	} else {
9560		memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9561		strncpy(t10id->vendor, val,
9562		    min(sizeof(t10id->vendor), strlen(val)));
9563	}
9564
9565	/*
9566	 * desc1 is for the WWPN which is a port asscociation.
9567	 */
9568	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9569	desc1->length = CTL_WWPN_LEN;
9570	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9571	/* For testing just create the WWPN */
9572#if 0
9573	ddb_GetWWNN((char *)desc1->identifier);
9574
9575	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9576	/* This is so Copancontrol will return something sane */
9577	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9578	    ctsio->io_hdr.nexus.targ_port!=8)
9579		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9580	else
9581		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9582#endif
9583
9584	be64enc(desc1->identifier, fe->wwpn);
9585
9586	/*
9587	 * desc2 is for the Relative Target Port(type 4h) identifier
9588	 */
9589	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9590	                 | SVPD_ID_TYPE_RELTARG;
9591	desc2->length = 4;
9592//#if 0
9593	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9594	/* This is so Copancontrol will return something sane */
9595	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9596	    ctsio->io_hdr.nexus.targ_port!=8)
9597		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9598	else
9599	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9600//#endif
9601
9602	/*
9603	 * desc3 is for the Target Port Group(type 5h) identifier
9604	 */
9605	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9606	                 | SVPD_ID_TYPE_TPORTGRP;
9607	desc3->length = 4;
9608	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9609		desc3->identifier[3] = 1;
9610	else
9611		desc3->identifier[3] = 2;
9612
9613	/*
9614	 * If we've actually got a backend, copy the device id from the
9615	 * per-LUN data.  Otherwise, set it to all spaces.
9616	 */
9617	if (lun != NULL) {
9618		/*
9619		 * Copy the backend's LUN ID.
9620		 */
9621		strncpy((char *)t10id->vendor_spec_id,
9622			(char *)lun->be_lun->device_id, devid_len);
9623	} else {
9624		/*
9625		 * No backend, set this to spaces.
9626		 */
9627		memset(t10id->vendor_spec_id, 0x20, devid_len);
9628	}
9629
9630	ctsio->scsi_status = SCSI_STATUS_OK;
9631
9632	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9633	ctsio->be_move_done = ctl_config_move_done;
9634	ctl_datamove((union ctl_io *)ctsio);
9635
9636	return (CTL_RETVAL_COMPLETE);
9637}
9638
9639static int
9640ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9641{
9642	struct scsi_vpd_block_limits *bl_ptr;
9643	struct ctl_lun *lun;
9644	int bs;
9645
9646	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9647	bs = lun->be_lun->blocksize;
9648
9649	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9650	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9651	ctsio->kern_sg_entries = 0;
9652
9653	if (sizeof(*bl_ptr) < alloc_len) {
9654		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9655		ctsio->kern_data_len = sizeof(*bl_ptr);
9656		ctsio->kern_total_len = sizeof(*bl_ptr);
9657	} else {
9658		ctsio->residual = 0;
9659		ctsio->kern_data_len = alloc_len;
9660		ctsio->kern_total_len = alloc_len;
9661	}
9662	ctsio->kern_data_resid = 0;
9663	ctsio->kern_rel_offset = 0;
9664	ctsio->kern_sg_entries = 0;
9665
9666	/*
9667	 * The control device is always connected.  The disk device, on the
9668	 * other hand, may not be online all the time.  Need to change this
9669	 * to figure out whether the disk device is actually online or not.
9670	 */
9671	if (lun != NULL)
9672		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9673				  lun->be_lun->lun_type;
9674	else
9675		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9676
9677	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9678	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9679	bl_ptr->max_cmp_write_len = 0xff;
9680	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9681	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9682	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9683		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9684		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9685	}
9686	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9687
9688	ctsio->scsi_status = SCSI_STATUS_OK;
9689	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9690	ctsio->be_move_done = ctl_config_move_done;
9691	ctl_datamove((union ctl_io *)ctsio);
9692
9693	return (CTL_RETVAL_COMPLETE);
9694}
9695
9696static int
9697ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9698{
9699	struct scsi_vpd_logical_block_prov *lbp_ptr;
9700	struct ctl_lun *lun;
9701	int bs;
9702
9703	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9704	bs = lun->be_lun->blocksize;
9705
9706	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9707	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9708	ctsio->kern_sg_entries = 0;
9709
9710	if (sizeof(*lbp_ptr) < alloc_len) {
9711		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9712		ctsio->kern_data_len = sizeof(*lbp_ptr);
9713		ctsio->kern_total_len = sizeof(*lbp_ptr);
9714	} else {
9715		ctsio->residual = 0;
9716		ctsio->kern_data_len = alloc_len;
9717		ctsio->kern_total_len = alloc_len;
9718	}
9719	ctsio->kern_data_resid = 0;
9720	ctsio->kern_rel_offset = 0;
9721	ctsio->kern_sg_entries = 0;
9722
9723	/*
9724	 * The control device is always connected.  The disk device, on the
9725	 * other hand, may not be online all the time.  Need to change this
9726	 * to figure out whether the disk device is actually online or not.
9727	 */
9728	if (lun != NULL)
9729		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9730				  lun->be_lun->lun_type;
9731	else
9732		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9733
9734	lbp_ptr->page_code = SVPD_LBP;
9735	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9736		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9737
9738	ctsio->scsi_status = SCSI_STATUS_OK;
9739	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9740	ctsio->be_move_done = ctl_config_move_done;
9741	ctl_datamove((union ctl_io *)ctsio);
9742
9743	return (CTL_RETVAL_COMPLETE);
9744}
9745
9746static int
9747ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9748{
9749	struct scsi_inquiry *cdb;
9750	struct ctl_lun *lun;
9751	int alloc_len, retval;
9752
9753	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9754	cdb = (struct scsi_inquiry *)ctsio->cdb;
9755
9756	retval = CTL_RETVAL_COMPLETE;
9757
9758	alloc_len = scsi_2btoul(cdb->length);
9759
9760	switch (cdb->page_code) {
9761	case SVPD_SUPPORTED_PAGES:
9762		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9763		break;
9764	case SVPD_UNIT_SERIAL_NUMBER:
9765		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9766		break;
9767	case SVPD_DEVICE_ID:
9768		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9769		break;
9770	case SVPD_BLOCK_LIMITS:
9771		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9772		break;
9773	case SVPD_LBP:
9774		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9775		break;
9776	default:
9777		ctl_set_invalid_field(ctsio,
9778				      /*sks_valid*/ 1,
9779				      /*command*/ 1,
9780				      /*field*/ 2,
9781				      /*bit_valid*/ 0,
9782				      /*bit*/ 0);
9783		ctl_done((union ctl_io *)ctsio);
9784		retval = CTL_RETVAL_COMPLETE;
9785		break;
9786	}
9787
9788	return (retval);
9789}
9790
9791static int
9792ctl_inquiry_std(struct ctl_scsiio *ctsio)
9793{
9794	struct scsi_inquiry_data *inq_ptr;
9795	struct scsi_inquiry *cdb;
9796	struct ctl_softc *ctl_softc;
9797	struct ctl_lun *lun;
9798	char *val;
9799	uint32_t alloc_len;
9800	int is_fc;
9801
9802	ctl_softc = control_softc;
9803
9804	/*
9805	 * Figure out whether we're talking to a Fibre Channel port or not.
9806	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9807	 * SCSI front ends.
9808	 */
9809	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9810	    CTL_PORT_FC)
9811		is_fc = 0;
9812	else
9813		is_fc = 1;
9814
9815	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9816	cdb = (struct scsi_inquiry *)ctsio->cdb;
9817	alloc_len = scsi_2btoul(cdb->length);
9818
9819	/*
9820	 * We malloc the full inquiry data size here and fill it
9821	 * in.  If the user only asks for less, we'll give him
9822	 * that much.
9823	 */
9824	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9825	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9826	ctsio->kern_sg_entries = 0;
9827	ctsio->kern_data_resid = 0;
9828	ctsio->kern_rel_offset = 0;
9829
9830	if (sizeof(*inq_ptr) < alloc_len) {
9831		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9832		ctsio->kern_data_len = sizeof(*inq_ptr);
9833		ctsio->kern_total_len = sizeof(*inq_ptr);
9834	} else {
9835		ctsio->residual = 0;
9836		ctsio->kern_data_len = alloc_len;
9837		ctsio->kern_total_len = alloc_len;
9838	}
9839
9840	/*
9841	 * If we have a LUN configured, report it as connected.  Otherwise,
9842	 * report that it is offline or no device is supported, depending
9843	 * on the value of inquiry_pq_no_lun.
9844	 *
9845	 * According to the spec (SPC-4 r34), the peripheral qualifier
9846	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9847	 *
9848	 * "A peripheral device having the specified peripheral device type
9849	 * is not connected to this logical unit. However, the device
9850	 * server is capable of supporting the specified peripheral device
9851	 * type on this logical unit."
9852	 *
9853	 * According to the same spec, the peripheral qualifier
9854	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9855	 *
9856	 * "The device server is not capable of supporting a peripheral
9857	 * device on this logical unit. For this peripheral qualifier the
9858	 * peripheral device type shall be set to 1Fh. All other peripheral
9859	 * device type values are reserved for this peripheral qualifier."
9860	 *
9861	 * Given the text, it would seem that we probably want to report that
9862	 * the LUN is offline here.  There is no LUN connected, but we can
9863	 * support a LUN at the given LUN number.
9864	 *
9865	 * In the real world, though, it sounds like things are a little
9866	 * different:
9867	 *
9868	 * - Linux, when presented with a LUN with the offline peripheral
9869	 *   qualifier, will create an sg driver instance for it.  So when
9870	 *   you attach it to CTL, you wind up with a ton of sg driver
9871	 *   instances.  (One for every LUN that Linux bothered to probe.)
9872	 *   Linux does this despite the fact that it issues a REPORT LUNs
9873	 *   to LUN 0 to get the inventory of supported LUNs.
9874	 *
9875	 * - There is other anecdotal evidence (from Emulex folks) about
9876	 *   arrays that use the offline peripheral qualifier for LUNs that
9877	 *   are on the "passive" path in an active/passive array.
9878	 *
9879	 * So the solution is provide a hopefully reasonable default
9880	 * (return bad/no LUN) and allow the user to change the behavior
9881	 * with a tunable/sysctl variable.
9882	 */
9883	if (lun != NULL)
9884		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9885				  lun->be_lun->lun_type;
9886	else if (ctl_softc->inquiry_pq_no_lun == 0)
9887		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9888	else
9889		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9890
9891	/* RMB in byte 2 is 0 */
9892	inq_ptr->version = SCSI_REV_SPC3;
9893
9894	/*
9895	 * According to SAM-3, even if a device only supports a single
9896	 * level of LUN addressing, it should still set the HISUP bit:
9897	 *
9898	 * 4.9.1 Logical unit numbers overview
9899	 *
9900	 * All logical unit number formats described in this standard are
9901	 * hierarchical in structure even when only a single level in that
9902	 * hierarchy is used. The HISUP bit shall be set to one in the
9903	 * standard INQUIRY data (see SPC-2) when any logical unit number
9904	 * format described in this standard is used.  Non-hierarchical
9905	 * formats are outside the scope of this standard.
9906	 *
9907	 * Therefore we set the HiSup bit here.
9908	 *
9909	 * The reponse format is 2, per SPC-3.
9910	 */
9911	inq_ptr->response_format = SID_HiSup | 2;
9912
9913	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9914	CTL_DEBUG_PRINT(("additional_length = %d\n",
9915			 inq_ptr->additional_length));
9916
9917	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9918	/* 16 bit addressing */
9919	if (is_fc == 0)
9920		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9921	/* XXX set the SID_MultiP bit here if we're actually going to
9922	   respond on multiple ports */
9923	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9924
9925	/* 16 bit data bus, synchronous transfers */
9926	/* XXX these flags don't apply for FC */
9927	if (is_fc == 0)
9928		inq_ptr->flags = SID_WBus16 | SID_Sync;
9929	/*
9930	 * XXX KDM do we want to support tagged queueing on the control
9931	 * device at all?
9932	 */
9933	if ((lun == NULL)
9934	 || (lun->be_lun->lun_type != T_PROCESSOR))
9935		inq_ptr->flags |= SID_CmdQue;
9936	/*
9937	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9938	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9939	 * name and 4 bytes for the revision.
9940	 */
9941	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9942		strcpy(inq_ptr->vendor, CTL_VENDOR);
9943	} else {
9944		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
9945		strncpy(inq_ptr->vendor, val,
9946		    min(sizeof(inq_ptr->vendor), strlen(val)));
9947	}
9948	if (lun == NULL) {
9949		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9950	} else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
9951		switch (lun->be_lun->lun_type) {
9952		case T_DIRECT:
9953			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9954			break;
9955		case T_PROCESSOR:
9956			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9957			break;
9958		default:
9959			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9960			break;
9961		}
9962	} else {
9963		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
9964		strncpy(inq_ptr->product, val,
9965		    min(sizeof(inq_ptr->product), strlen(val)));
9966	}
9967
9968	/*
9969	 * XXX make this a macro somewhere so it automatically gets
9970	 * incremented when we make changes.
9971	 */
9972	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
9973		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9974	} else {
9975		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
9976		strncpy(inq_ptr->revision, val,
9977		    min(sizeof(inq_ptr->revision), strlen(val)));
9978	}
9979
9980	/*
9981	 * For parallel SCSI, we support double transition and single
9982	 * transition clocking.  We also support QAS (Quick Arbitration
9983	 * and Selection) and Information Unit transfers on both the
9984	 * control and array devices.
9985	 */
9986	if (is_fc == 0)
9987		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9988				    SID_SPI_IUS;
9989
9990	/* SAM-3 */
9991	scsi_ulto2b(0x0060, inq_ptr->version1);
9992	/* SPC-3 (no version claimed) XXX should we claim a version? */
9993	scsi_ulto2b(0x0300, inq_ptr->version2);
9994	if (is_fc) {
9995		/* FCP-2 ANSI INCITS.350:2003 */
9996		scsi_ulto2b(0x0917, inq_ptr->version3);
9997	} else {
9998		/* SPI-4 ANSI INCITS.362:200x */
9999		scsi_ulto2b(0x0B56, inq_ptr->version3);
10000	}
10001
10002	if (lun == NULL) {
10003		/* SBC-2 (no version claimed) XXX should we claim a version? */
10004		scsi_ulto2b(0x0320, inq_ptr->version4);
10005	} else {
10006		switch (lun->be_lun->lun_type) {
10007		case T_DIRECT:
10008			/*
10009			 * SBC-2 (no version claimed) XXX should we claim a
10010			 * version?
10011			 */
10012			scsi_ulto2b(0x0320, inq_ptr->version4);
10013			break;
10014		case T_PROCESSOR:
10015		default:
10016			break;
10017		}
10018	}
10019
10020	ctsio->scsi_status = SCSI_STATUS_OK;
10021	if (ctsio->kern_data_len > 0) {
10022		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10023		ctsio->be_move_done = ctl_config_move_done;
10024		ctl_datamove((union ctl_io *)ctsio);
10025	} else {
10026		ctsio->io_hdr.status = CTL_SUCCESS;
10027		ctl_done((union ctl_io *)ctsio);
10028	}
10029
10030	return (CTL_RETVAL_COMPLETE);
10031}
10032
10033int
10034ctl_inquiry(struct ctl_scsiio *ctsio)
10035{
10036	struct scsi_inquiry *cdb;
10037	int retval;
10038
10039	cdb = (struct scsi_inquiry *)ctsio->cdb;
10040
10041	retval = 0;
10042
10043	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10044
10045	/*
10046	 * Right now, we don't support the CmdDt inquiry information.
10047	 * This would be nice to support in the future.  When we do
10048	 * support it, we should change this test so that it checks to make
10049	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10050	 */
10051#ifdef notyet
10052	if (((cdb->byte2 & SI_EVPD)
10053	 && (cdb->byte2 & SI_CMDDT)))
10054#endif
10055	if (cdb->byte2 & SI_CMDDT) {
10056		/*
10057		 * Point to the SI_CMDDT bit.  We might change this
10058		 * when we support SI_CMDDT, but since both bits would be
10059		 * "wrong", this should probably just stay as-is then.
10060		 */
10061		ctl_set_invalid_field(ctsio,
10062				      /*sks_valid*/ 1,
10063				      /*command*/ 1,
10064				      /*field*/ 1,
10065				      /*bit_valid*/ 1,
10066				      /*bit*/ 1);
10067		ctl_done((union ctl_io *)ctsio);
10068		return (CTL_RETVAL_COMPLETE);
10069	}
10070	if (cdb->byte2 & SI_EVPD)
10071		retval = ctl_inquiry_evpd(ctsio);
10072#ifdef notyet
10073	else if (cdb->byte2 & SI_CMDDT)
10074		retval = ctl_inquiry_cmddt(ctsio);
10075#endif
10076	else
10077		retval = ctl_inquiry_std(ctsio);
10078
10079	return (retval);
10080}
10081
10082/*
10083 * For known CDB types, parse the LBA and length.
10084 */
10085static int
10086ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10087{
10088	if (io->io_hdr.io_type != CTL_IO_SCSI)
10089		return (1);
10090
10091	switch (io->scsiio.cdb[0]) {
10092	case COMPARE_AND_WRITE: {
10093		struct scsi_compare_and_write *cdb;
10094
10095		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10096
10097		*lba = scsi_8btou64(cdb->addr);
10098		*len = cdb->length;
10099		break;
10100	}
10101	case READ_6:
10102	case WRITE_6: {
10103		struct scsi_rw_6 *cdb;
10104
10105		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10106
10107		*lba = scsi_3btoul(cdb->addr);
10108		/* only 5 bits are valid in the most significant address byte */
10109		*lba &= 0x1fffff;
10110		*len = cdb->length;
10111		break;
10112	}
10113	case READ_10:
10114	case WRITE_10: {
10115		struct scsi_rw_10 *cdb;
10116
10117		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10118
10119		*lba = scsi_4btoul(cdb->addr);
10120		*len = scsi_2btoul(cdb->length);
10121		break;
10122	}
10123	case WRITE_VERIFY_10: {
10124		struct scsi_write_verify_10 *cdb;
10125
10126		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10127
10128		*lba = scsi_4btoul(cdb->addr);
10129		*len = scsi_2btoul(cdb->length);
10130		break;
10131	}
10132	case READ_12:
10133	case WRITE_12: {
10134		struct scsi_rw_12 *cdb;
10135
10136		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10137
10138		*lba = scsi_4btoul(cdb->addr);
10139		*len = scsi_4btoul(cdb->length);
10140		break;
10141	}
10142	case WRITE_VERIFY_12: {
10143		struct scsi_write_verify_12 *cdb;
10144
10145		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10146
10147		*lba = scsi_4btoul(cdb->addr);
10148		*len = scsi_4btoul(cdb->length);
10149		break;
10150	}
10151	case READ_16:
10152	case WRITE_16: {
10153		struct scsi_rw_16 *cdb;
10154
10155		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10156
10157		*lba = scsi_8btou64(cdb->addr);
10158		*len = scsi_4btoul(cdb->length);
10159		break;
10160	}
10161	case WRITE_VERIFY_16: {
10162		struct scsi_write_verify_16 *cdb;
10163
10164		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10165
10166
10167		*lba = scsi_8btou64(cdb->addr);
10168		*len = scsi_4btoul(cdb->length);
10169		break;
10170	}
10171	case WRITE_SAME_10: {
10172		struct scsi_write_same_10 *cdb;
10173
10174		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10175
10176		*lba = scsi_4btoul(cdb->addr);
10177		*len = scsi_2btoul(cdb->length);
10178		break;
10179	}
10180	case WRITE_SAME_16: {
10181		struct scsi_write_same_16 *cdb;
10182
10183		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10184
10185		*lba = scsi_8btou64(cdb->addr);
10186		*len = scsi_4btoul(cdb->length);
10187		break;
10188	}
10189	case VERIFY_10: {
10190		struct scsi_verify_10 *cdb;
10191
10192		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10193
10194		*lba = scsi_4btoul(cdb->addr);
10195		*len = scsi_2btoul(cdb->length);
10196		break;
10197	}
10198	case VERIFY_12: {
10199		struct scsi_verify_12 *cdb;
10200
10201		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10202
10203		*lba = scsi_4btoul(cdb->addr);
10204		*len = scsi_4btoul(cdb->length);
10205		break;
10206	}
10207	case VERIFY_16: {
10208		struct scsi_verify_16 *cdb;
10209
10210		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10211
10212		*lba = scsi_8btou64(cdb->addr);
10213		*len = scsi_4btoul(cdb->length);
10214		break;
10215	}
10216	default:
10217		return (1);
10218		break; /* NOTREACHED */
10219	}
10220
10221	return (0);
10222}
10223
10224static ctl_action
10225ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10226{
10227	uint64_t endlba1, endlba2;
10228
10229	endlba1 = lba1 + len1 - 1;
10230	endlba2 = lba2 + len2 - 1;
10231
10232	if ((endlba1 < lba2)
10233	 || (endlba2 < lba1))
10234		return (CTL_ACTION_PASS);
10235	else
10236		return (CTL_ACTION_BLOCK);
10237}
10238
10239static ctl_action
10240ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10241{
10242	uint64_t lba1, lba2;
10243	uint32_t len1, len2;
10244	int retval;
10245
10246	retval = ctl_get_lba_len(io1, &lba1, &len1);
10247	if (retval != 0)
10248		return (CTL_ACTION_ERROR);
10249
10250	retval = ctl_get_lba_len(io2, &lba2, &len2);
10251	if (retval != 0)
10252		return (CTL_ACTION_ERROR);
10253
10254	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10255}
10256
10257static ctl_action
10258ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10259{
10260	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10261	ctl_serialize_action *serialize_row;
10262
10263	/*
10264	 * The initiator attempted multiple untagged commands at the same
10265	 * time.  Can't do that.
10266	 */
10267	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10268	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10269	 && ((pending_io->io_hdr.nexus.targ_port ==
10270	      ooa_io->io_hdr.nexus.targ_port)
10271	  && (pending_io->io_hdr.nexus.initid.id ==
10272	      ooa_io->io_hdr.nexus.initid.id))
10273	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10274		return (CTL_ACTION_OVERLAP);
10275
10276	/*
10277	 * The initiator attempted to send multiple tagged commands with
10278	 * the same ID.  (It's fine if different initiators have the same
10279	 * tag ID.)
10280	 *
10281	 * Even if all of those conditions are true, we don't kill the I/O
10282	 * if the command ahead of us has been aborted.  We won't end up
10283	 * sending it to the FETD, and it's perfectly legal to resend a
10284	 * command with the same tag number as long as the previous
10285	 * instance of this tag number has been aborted somehow.
10286	 */
10287	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10288	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10289	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10290	 && ((pending_io->io_hdr.nexus.targ_port ==
10291	      ooa_io->io_hdr.nexus.targ_port)
10292	  && (pending_io->io_hdr.nexus.initid.id ==
10293	      ooa_io->io_hdr.nexus.initid.id))
10294	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10295		return (CTL_ACTION_OVERLAP_TAG);
10296
10297	/*
10298	 * If we get a head of queue tag, SAM-3 says that we should
10299	 * immediately execute it.
10300	 *
10301	 * What happens if this command would normally block for some other
10302	 * reason?  e.g. a request sense with a head of queue tag
10303	 * immediately after a write.  Normally that would block, but this
10304	 * will result in its getting executed immediately...
10305	 *
10306	 * We currently return "pass" instead of "skip", so we'll end up
10307	 * going through the rest of the queue to check for overlapped tags.
10308	 *
10309	 * XXX KDM check for other types of blockage first??
10310	 */
10311	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10312		return (CTL_ACTION_PASS);
10313
10314	/*
10315	 * Ordered tags have to block until all items ahead of them
10316	 * have completed.  If we get called with an ordered tag, we always
10317	 * block, if something else is ahead of us in the queue.
10318	 */
10319	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10320		return (CTL_ACTION_BLOCK);
10321
10322	/*
10323	 * Simple tags get blocked until all head of queue and ordered tags
10324	 * ahead of them have completed.  I'm lumping untagged commands in
10325	 * with simple tags here.  XXX KDM is that the right thing to do?
10326	 */
10327	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10328	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10329	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10330	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10331		return (CTL_ACTION_BLOCK);
10332
10333	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio);
10334	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio);
10335
10336	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10337
10338	switch (serialize_row[pending_entry->seridx]) {
10339	case CTL_SER_BLOCK:
10340		return (CTL_ACTION_BLOCK);
10341		break; /* NOTREACHED */
10342	case CTL_SER_EXTENT:
10343		return (ctl_extent_check(pending_io, ooa_io));
10344		break; /* NOTREACHED */
10345	case CTL_SER_PASS:
10346		return (CTL_ACTION_PASS);
10347		break; /* NOTREACHED */
10348	case CTL_SER_SKIP:
10349		return (CTL_ACTION_SKIP);
10350		break;
10351	default:
10352		panic("invalid serialization value %d",
10353		      serialize_row[pending_entry->seridx]);
10354		break; /* NOTREACHED */
10355	}
10356
10357	return (CTL_ACTION_ERROR);
10358}
10359
10360/*
10361 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10362 * Assumptions:
10363 * - pending_io is generally either incoming, or on the blocked queue
10364 * - starting I/O is the I/O we want to start the check with.
10365 */
10366static ctl_action
10367ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10368	      union ctl_io *starting_io)
10369{
10370	union ctl_io *ooa_io;
10371	ctl_action action;
10372
10373	mtx_assert(&lun->lun_lock, MA_OWNED);
10374
10375	/*
10376	 * Run back along the OOA queue, starting with the current
10377	 * blocked I/O and going through every I/O before it on the
10378	 * queue.  If starting_io is NULL, we'll just end up returning
10379	 * CTL_ACTION_PASS.
10380	 */
10381	for (ooa_io = starting_io; ooa_io != NULL;
10382	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10383	     ooa_links)){
10384
10385		/*
10386		 * This routine just checks to see whether
10387		 * cur_blocked is blocked by ooa_io, which is ahead
10388		 * of it in the queue.  It doesn't queue/dequeue
10389		 * cur_blocked.
10390		 */
10391		action = ctl_check_for_blockage(pending_io, ooa_io);
10392		switch (action) {
10393		case CTL_ACTION_BLOCK:
10394		case CTL_ACTION_OVERLAP:
10395		case CTL_ACTION_OVERLAP_TAG:
10396		case CTL_ACTION_SKIP:
10397		case CTL_ACTION_ERROR:
10398			return (action);
10399			break; /* NOTREACHED */
10400		case CTL_ACTION_PASS:
10401			break;
10402		default:
10403			panic("invalid action %d", action);
10404			break;  /* NOTREACHED */
10405		}
10406	}
10407
10408	return (CTL_ACTION_PASS);
10409}
10410
10411/*
10412 * Assumptions:
10413 * - An I/O has just completed, and has been removed from the per-LUN OOA
10414 *   queue, so some items on the blocked queue may now be unblocked.
10415 */
10416static int
10417ctl_check_blocked(struct ctl_lun *lun)
10418{
10419	union ctl_io *cur_blocked, *next_blocked;
10420
10421	mtx_assert(&lun->lun_lock, MA_OWNED);
10422
10423	/*
10424	 * Run forward from the head of the blocked queue, checking each
10425	 * entry against the I/Os prior to it on the OOA queue to see if
10426	 * there is still any blockage.
10427	 *
10428	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10429	 * with our removing a variable on it while it is traversing the
10430	 * list.
10431	 */
10432	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10433	     cur_blocked != NULL; cur_blocked = next_blocked) {
10434		union ctl_io *prev_ooa;
10435		ctl_action action;
10436
10437		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10438							  blocked_links);
10439
10440		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10441						      ctl_ooaq, ooa_links);
10442
10443		/*
10444		 * If cur_blocked happens to be the first item in the OOA
10445		 * queue now, prev_ooa will be NULL, and the action
10446		 * returned will just be CTL_ACTION_PASS.
10447		 */
10448		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10449
10450		switch (action) {
10451		case CTL_ACTION_BLOCK:
10452			/* Nothing to do here, still blocked */
10453			break;
10454		case CTL_ACTION_OVERLAP:
10455		case CTL_ACTION_OVERLAP_TAG:
10456			/*
10457			 * This shouldn't happen!  In theory we've already
10458			 * checked this command for overlap...
10459			 */
10460			break;
10461		case CTL_ACTION_PASS:
10462		case CTL_ACTION_SKIP: {
10463			struct ctl_softc *softc;
10464			const struct ctl_cmd_entry *entry;
10465			uint32_t initidx;
10466			int isc_retval;
10467
10468			/*
10469			 * The skip case shouldn't happen, this transaction
10470			 * should have never made it onto the blocked queue.
10471			 */
10472			/*
10473			 * This I/O is no longer blocked, we can remove it
10474			 * from the blocked queue.  Since this is a TAILQ
10475			 * (doubly linked list), we can do O(1) removals
10476			 * from any place on the list.
10477			 */
10478			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10479				     blocked_links);
10480			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10481
10482			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10483				/*
10484				 * Need to send IO back to original side to
10485				 * run
10486				 */
10487				union ctl_ha_msg msg_info;
10488
10489				msg_info.hdr.original_sc =
10490					cur_blocked->io_hdr.original_sc;
10491				msg_info.hdr.serializing_sc = cur_blocked;
10492				msg_info.hdr.msg_type = CTL_MSG_R2R;
10493				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10494				     &msg_info, sizeof(msg_info), 0)) >
10495				     CTL_HA_STATUS_SUCCESS) {
10496					printf("CTL:Check Blocked error from "
10497					       "ctl_ha_msg_send %d\n",
10498					       isc_retval);
10499				}
10500				break;
10501			}
10502			entry = ctl_get_cmd_entry(&cur_blocked->scsiio);
10503			softc = control_softc;
10504
10505			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10506
10507			/*
10508			 * Check this I/O for LUN state changes that may
10509			 * have happened while this command was blocked.
10510			 * The LUN state may have been changed by a command
10511			 * ahead of us in the queue, so we need to re-check
10512			 * for any states that can be caused by SCSI
10513			 * commands.
10514			 */
10515			if (ctl_scsiio_lun_check(softc, lun, entry,
10516						 &cur_blocked->scsiio) == 0) {
10517				cur_blocked->io_hdr.flags |=
10518				                      CTL_FLAG_IS_WAS_ON_RTR;
10519				ctl_enqueue_rtr(cur_blocked);
10520			} else
10521				ctl_done(cur_blocked);
10522			break;
10523		}
10524		default:
10525			/*
10526			 * This probably shouldn't happen -- we shouldn't
10527			 * get CTL_ACTION_ERROR, or anything else.
10528			 */
10529			break;
10530		}
10531	}
10532
10533	return (CTL_RETVAL_COMPLETE);
10534}
10535
10536/*
10537 * This routine (with one exception) checks LUN flags that can be set by
10538 * commands ahead of us in the OOA queue.  These flags have to be checked
10539 * when a command initially comes in, and when we pull a command off the
10540 * blocked queue and are preparing to execute it.  The reason we have to
10541 * check these flags for commands on the blocked queue is that the LUN
10542 * state may have been changed by a command ahead of us while we're on the
10543 * blocked queue.
10544 *
10545 * Ordering is somewhat important with these checks, so please pay
10546 * careful attention to the placement of any new checks.
10547 */
10548static int
10549ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10550    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10551{
10552	int retval;
10553
10554	retval = 0;
10555
10556	mtx_assert(&lun->lun_lock, MA_OWNED);
10557
10558	/*
10559	 * If this shelf is a secondary shelf controller, we have to reject
10560	 * any media access commands.
10561	 */
10562#if 0
10563	/* No longer needed for HA */
10564	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10565	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10566		ctl_set_lun_standby(ctsio);
10567		retval = 1;
10568		goto bailout;
10569	}
10570#endif
10571
10572	/*
10573	 * Check for a reservation conflict.  If this command isn't allowed
10574	 * even on reserved LUNs, and if this initiator isn't the one who
10575	 * reserved us, reject the command with a reservation conflict.
10576	 */
10577	if ((lun->flags & CTL_LUN_RESERVED)
10578	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10579		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10580		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10581		 || (ctsio->io_hdr.nexus.targ_target.id !=
10582		     lun->rsv_nexus.targ_target.id)) {
10583			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10584			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10585			retval = 1;
10586			goto bailout;
10587		}
10588	}
10589
10590	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10591	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10592		uint32_t residx;
10593
10594		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10595		/*
10596		 * if we aren't registered or it's a res holder type
10597		 * reservation and this isn't the res holder then set a
10598		 * conflict.
10599		 * NOTE: Commands which might be allowed on write exclusive
10600		 * type reservations are checked in the particular command
10601		 * for a conflict. Read and SSU are the only ones.
10602		 */
10603		if (!lun->per_res[residx].registered
10604		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10605			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10606			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10607			retval = 1;
10608			goto bailout;
10609		}
10610
10611	}
10612
10613	if ((lun->flags & CTL_LUN_OFFLINE)
10614	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10615		ctl_set_lun_not_ready(ctsio);
10616		retval = 1;
10617		goto bailout;
10618	}
10619
10620	/*
10621	 * If the LUN is stopped, see if this particular command is allowed
10622	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10623	 */
10624	if ((lun->flags & CTL_LUN_STOPPED)
10625	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10626		/* "Logical unit not ready, initializing cmd. required" */
10627		ctl_set_lun_stopped(ctsio);
10628		retval = 1;
10629		goto bailout;
10630	}
10631
10632	if ((lun->flags & CTL_LUN_INOPERABLE)
10633	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10634		/* "Medium format corrupted" */
10635		ctl_set_medium_format_corrupted(ctsio);
10636		retval = 1;
10637		goto bailout;
10638	}
10639
10640bailout:
10641	return (retval);
10642
10643}
10644
10645static void
10646ctl_failover_io(union ctl_io *io, int have_lock)
10647{
10648	ctl_set_busy(&io->scsiio);
10649	ctl_done(io);
10650}
10651
10652static void
10653ctl_failover(void)
10654{
10655	struct ctl_lun *lun;
10656	struct ctl_softc *ctl_softc;
10657	union ctl_io *next_io, *pending_io;
10658	union ctl_io *io;
10659	int lun_idx;
10660	int i;
10661
10662	ctl_softc = control_softc;
10663
10664	mtx_lock(&ctl_softc->ctl_lock);
10665	/*
10666	 * Remove any cmds from the other SC from the rtr queue.  These
10667	 * will obviously only be for LUNs for which we're the primary.
10668	 * We can't send status or get/send data for these commands.
10669	 * Since they haven't been executed yet, we can just remove them.
10670	 * We'll either abort them or delete them below, depending on
10671	 * which HA mode we're in.
10672	 */
10673#ifdef notyet
10674	mtx_lock(&ctl_softc->queue_lock);
10675	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10676	     io != NULL; io = next_io) {
10677		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10678		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10679			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10680				      ctl_io_hdr, links);
10681	}
10682	mtx_unlock(&ctl_softc->queue_lock);
10683#endif
10684
10685	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10686		lun = ctl_softc->ctl_luns[lun_idx];
10687		if (lun==NULL)
10688			continue;
10689
10690		/*
10691		 * Processor LUNs are primary on both sides.
10692		 * XXX will this always be true?
10693		 */
10694		if (lun->be_lun->lun_type == T_PROCESSOR)
10695			continue;
10696
10697		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10698		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10699			printf("FAILOVER: primary lun %d\n", lun_idx);
10700		        /*
10701			 * Remove all commands from the other SC. First from the
10702			 * blocked queue then from the ooa queue. Once we have
10703			 * removed them. Call ctl_check_blocked to see if there
10704			 * is anything that can run.
10705			 */
10706			for (io = (union ctl_io *)TAILQ_FIRST(
10707			     &lun->blocked_queue); io != NULL; io = next_io) {
10708
10709		        	next_io = (union ctl_io *)TAILQ_NEXT(
10710				    &io->io_hdr, blocked_links);
10711
10712				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10713					TAILQ_REMOVE(&lun->blocked_queue,
10714						     &io->io_hdr,blocked_links);
10715					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10716					TAILQ_REMOVE(&lun->ooa_queue,
10717						     &io->io_hdr, ooa_links);
10718
10719					ctl_free_io(io);
10720				}
10721			}
10722
10723			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10724	     		     io != NULL; io = next_io) {
10725
10726		        	next_io = (union ctl_io *)TAILQ_NEXT(
10727				    &io->io_hdr, ooa_links);
10728
10729				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10730
10731					TAILQ_REMOVE(&lun->ooa_queue,
10732						&io->io_hdr,
10733					     	ooa_links);
10734
10735					ctl_free_io(io);
10736				}
10737			}
10738			ctl_check_blocked(lun);
10739		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10740			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10741
10742			printf("FAILOVER: primary lun %d\n", lun_idx);
10743			/*
10744			 * Abort all commands from the other SC.  We can't
10745			 * send status back for them now.  These should get
10746			 * cleaned up when they are completed or come out
10747			 * for a datamove operation.
10748			 */
10749			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10750	     		     io != NULL; io = next_io) {
10751		        	next_io = (union ctl_io *)TAILQ_NEXT(
10752					&io->io_hdr, ooa_links);
10753
10754				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10755					io->io_hdr.flags |= CTL_FLAG_ABORT;
10756			}
10757		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10758			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10759
10760			printf("FAILOVER: secondary lun %d\n", lun_idx);
10761
10762			lun->flags |= CTL_LUN_PRIMARY_SC;
10763
10764			/*
10765			 * We send all I/O that was sent to this controller
10766			 * and redirected to the other side back with
10767			 * busy status, and have the initiator retry it.
10768			 * Figuring out how much data has been transferred,
10769			 * etc. and picking up where we left off would be
10770			 * very tricky.
10771			 *
10772			 * XXX KDM need to remove I/O from the blocked
10773			 * queue as well!
10774			 */
10775			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10776			     &lun->ooa_queue); pending_io != NULL;
10777			     pending_io = next_io) {
10778
10779				next_io =  (union ctl_io *)TAILQ_NEXT(
10780					&pending_io->io_hdr, ooa_links);
10781
10782				pending_io->io_hdr.flags &=
10783					~CTL_FLAG_SENT_2OTHER_SC;
10784
10785				if (pending_io->io_hdr.flags &
10786				    CTL_FLAG_IO_ACTIVE) {
10787					pending_io->io_hdr.flags |=
10788						CTL_FLAG_FAILOVER;
10789				} else {
10790					ctl_set_busy(&pending_io->scsiio);
10791					ctl_done(pending_io);
10792				}
10793			}
10794
10795			/*
10796			 * Build Unit Attention
10797			 */
10798			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10799				lun->pending_sense[i].ua_pending |=
10800				                     CTL_UA_ASYM_ACC_CHANGE;
10801			}
10802		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10803			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10804			printf("FAILOVER: secondary lun %d\n", lun_idx);
10805			/*
10806			 * if the first io on the OOA is not on the RtR queue
10807			 * add it.
10808			 */
10809			lun->flags |= CTL_LUN_PRIMARY_SC;
10810
10811			pending_io = (union ctl_io *)TAILQ_FIRST(
10812			    &lun->ooa_queue);
10813			if (pending_io==NULL) {
10814				printf("Nothing on OOA queue\n");
10815				continue;
10816			}
10817
10818			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10819			if ((pending_io->io_hdr.flags &
10820			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10821				pending_io->io_hdr.flags |=
10822				    CTL_FLAG_IS_WAS_ON_RTR;
10823				ctl_enqueue_rtr(pending_io);
10824			}
10825#if 0
10826			else
10827			{
10828				printf("Tag 0x%04x is running\n",
10829				      pending_io->scsiio.tag_num);
10830			}
10831#endif
10832
10833			next_io = (union ctl_io *)TAILQ_NEXT(
10834			    &pending_io->io_hdr, ooa_links);
10835			for (pending_io=next_io; pending_io != NULL;
10836			     pending_io = next_io) {
10837				pending_io->io_hdr.flags &=
10838				    ~CTL_FLAG_SENT_2OTHER_SC;
10839				next_io = (union ctl_io *)TAILQ_NEXT(
10840					&pending_io->io_hdr, ooa_links);
10841				if (pending_io->io_hdr.flags &
10842				    CTL_FLAG_IS_WAS_ON_RTR) {
10843#if 0
10844				        printf("Tag 0x%04x is running\n",
10845				      		pending_io->scsiio.tag_num);
10846#endif
10847					continue;
10848				}
10849
10850				switch (ctl_check_ooa(lun, pending_io,
10851			            (union ctl_io *)TAILQ_PREV(
10852				    &pending_io->io_hdr, ctl_ooaq,
10853				    ooa_links))) {
10854
10855				case CTL_ACTION_BLOCK:
10856					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10857							  &pending_io->io_hdr,
10858							  blocked_links);
10859					pending_io->io_hdr.flags |=
10860					    CTL_FLAG_BLOCKED;
10861					break;
10862				case CTL_ACTION_PASS:
10863				case CTL_ACTION_SKIP:
10864					pending_io->io_hdr.flags |=
10865					    CTL_FLAG_IS_WAS_ON_RTR;
10866					ctl_enqueue_rtr(pending_io);
10867					break;
10868				case CTL_ACTION_OVERLAP:
10869					ctl_set_overlapped_cmd(
10870					    (struct ctl_scsiio *)pending_io);
10871					ctl_done(pending_io);
10872					break;
10873				case CTL_ACTION_OVERLAP_TAG:
10874					ctl_set_overlapped_tag(
10875					    (struct ctl_scsiio *)pending_io,
10876					    pending_io->scsiio.tag_num & 0xff);
10877					ctl_done(pending_io);
10878					break;
10879				case CTL_ACTION_ERROR:
10880				default:
10881					ctl_set_internal_failure(
10882						(struct ctl_scsiio *)pending_io,
10883						0,  // sks_valid
10884						0); //retry count
10885					ctl_done(pending_io);
10886					break;
10887				}
10888			}
10889
10890			/*
10891			 * Build Unit Attention
10892			 */
10893			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10894				lun->pending_sense[i].ua_pending |=
10895				                     CTL_UA_ASYM_ACC_CHANGE;
10896			}
10897		} else {
10898			panic("Unhandled HA mode failover, LUN flags = %#x, "
10899			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10900		}
10901	}
10902	ctl_pause_rtr = 0;
10903	mtx_unlock(&ctl_softc->ctl_lock);
10904}
10905
10906static int
10907ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10908{
10909	struct ctl_lun *lun;
10910	const struct ctl_cmd_entry *entry;
10911	uint32_t initidx, targ_lun;
10912	int retval;
10913
10914	retval = 0;
10915
10916	lun = NULL;
10917
10918	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
10919	if ((targ_lun < CTL_MAX_LUNS)
10920	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
10921		lun = ctl_softc->ctl_luns[targ_lun];
10922		/*
10923		 * If the LUN is invalid, pretend that it doesn't exist.
10924		 * It will go away as soon as all pending I/O has been
10925		 * completed.
10926		 */
10927		if (lun->flags & CTL_LUN_DISABLED) {
10928			lun = NULL;
10929		} else {
10930			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10931			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10932				lun->be_lun;
10933			if (lun->be_lun->lun_type == T_PROCESSOR) {
10934				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10935			}
10936
10937			/*
10938			 * Every I/O goes into the OOA queue for a
10939			 * particular LUN, and stays there until completion.
10940			 */
10941			mtx_lock(&lun->lun_lock);
10942			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
10943			    ooa_links);
10944		}
10945	} else {
10946		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10947		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10948	}
10949
10950	/* Get command entry and return error if it is unsuppotyed. */
10951	entry = ctl_validate_command(ctsio);
10952	if (entry == NULL) {
10953		if (lun)
10954			mtx_unlock(&lun->lun_lock);
10955		return (retval);
10956	}
10957
10958	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10959	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10960
10961	/*
10962	 * Check to see whether we can send this command to LUNs that don't
10963	 * exist.  This should pretty much only be the case for inquiry
10964	 * and request sense.  Further checks, below, really require having
10965	 * a LUN, so we can't really check the command anymore.  Just put
10966	 * it on the rtr queue.
10967	 */
10968	if (lun == NULL) {
10969		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
10970			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10971			ctl_enqueue_rtr((union ctl_io *)ctsio);
10972			return (retval);
10973		}
10974
10975		ctl_set_unsupported_lun(ctsio);
10976		ctl_done((union ctl_io *)ctsio);
10977		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
10978		return (retval);
10979	} else {
10980		/*
10981		 * Make sure we support this particular command on this LUN.
10982		 * e.g., we don't support writes to the control LUN.
10983		 */
10984		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
10985			mtx_unlock(&lun->lun_lock);
10986			ctl_set_invalid_opcode(ctsio);
10987			ctl_done((union ctl_io *)ctsio);
10988			return (retval);
10989		}
10990	}
10991
10992	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10993
10994	/*
10995	 * If we've got a request sense, it'll clear the contingent
10996	 * allegiance condition.  Otherwise, if we have a CA condition for
10997	 * this initiator, clear it, because it sent down a command other
10998	 * than request sense.
10999	 */
11000	if ((ctsio->cdb[0] != REQUEST_SENSE)
11001	 && (ctl_is_set(lun->have_ca, initidx)))
11002		ctl_clear_mask(lun->have_ca, initidx);
11003
11004	/*
11005	 * If the command has this flag set, it handles its own unit
11006	 * attention reporting, we shouldn't do anything.  Otherwise we
11007	 * check for any pending unit attentions, and send them back to the
11008	 * initiator.  We only do this when a command initially comes in,
11009	 * not when we pull it off the blocked queue.
11010	 *
11011	 * According to SAM-3, section 5.3.2, the order that things get
11012	 * presented back to the host is basically unit attentions caused
11013	 * by some sort of reset event, busy status, reservation conflicts
11014	 * or task set full, and finally any other status.
11015	 *
11016	 * One issue here is that some of the unit attentions we report
11017	 * don't fall into the "reset" category (e.g. "reported luns data
11018	 * has changed").  So reporting it here, before the reservation
11019	 * check, may be technically wrong.  I guess the only thing to do
11020	 * would be to check for and report the reset events here, and then
11021	 * check for the other unit attention types after we check for a
11022	 * reservation conflict.
11023	 *
11024	 * XXX KDM need to fix this
11025	 */
11026	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11027		ctl_ua_type ua_type;
11028
11029		ua_type = lun->pending_sense[initidx].ua_pending;
11030		if (ua_type != CTL_UA_NONE) {
11031			scsi_sense_data_type sense_format;
11032
11033			if (lun != NULL)
11034				sense_format = (lun->flags &
11035				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11036				    SSD_TYPE_FIXED;
11037			else
11038				sense_format = SSD_TYPE_FIXED;
11039
11040			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11041					       sense_format);
11042			if (ua_type != CTL_UA_NONE) {
11043				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11044				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11045						       CTL_AUTOSENSE;
11046				ctsio->sense_len = SSD_FULL_SIZE;
11047				lun->pending_sense[initidx].ua_pending &=
11048					~ua_type;
11049				mtx_unlock(&lun->lun_lock);
11050				ctl_done((union ctl_io *)ctsio);
11051				return (retval);
11052			}
11053		}
11054	}
11055
11056
11057	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11058		mtx_unlock(&lun->lun_lock);
11059		ctl_done((union ctl_io *)ctsio);
11060		return (retval);
11061	}
11062
11063	/*
11064	 * XXX CHD this is where we want to send IO to other side if
11065	 * this LUN is secondary on this SC. We will need to make a copy
11066	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11067	 * the copy we send as FROM_OTHER.
11068	 * We also need to stuff the address of the original IO so we can
11069	 * find it easily. Something similar will need be done on the other
11070	 * side so when we are done we can find the copy.
11071	 */
11072	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11073		union ctl_ha_msg msg_info;
11074		int isc_retval;
11075
11076		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11077
11078		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11079		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11080#if 0
11081		printf("1. ctsio %p\n", ctsio);
11082#endif
11083		msg_info.hdr.serializing_sc = NULL;
11084		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11085		msg_info.scsi.tag_num = ctsio->tag_num;
11086		msg_info.scsi.tag_type = ctsio->tag_type;
11087		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11088
11089		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11090
11091		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11092		    (void *)&msg_info, sizeof(msg_info), 0)) >
11093		    CTL_HA_STATUS_SUCCESS) {
11094			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11095			       isc_retval);
11096			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11097		} else {
11098#if 0
11099			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11100#endif
11101		}
11102
11103		/*
11104		 * XXX KDM this I/O is off the incoming queue, but hasn't
11105		 * been inserted on any other queue.  We may need to come
11106		 * up with a holding queue while we wait for serialization
11107		 * so that we have an idea of what we're waiting for from
11108		 * the other side.
11109		 */
11110		mtx_unlock(&lun->lun_lock);
11111		return (retval);
11112	}
11113
11114	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11115			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11116			      ctl_ooaq, ooa_links))) {
11117	case CTL_ACTION_BLOCK:
11118		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11119		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11120				  blocked_links);
11121		mtx_unlock(&lun->lun_lock);
11122		return (retval);
11123	case CTL_ACTION_PASS:
11124	case CTL_ACTION_SKIP:
11125		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11126		mtx_unlock(&lun->lun_lock);
11127		ctl_enqueue_rtr((union ctl_io *)ctsio);
11128		break;
11129	case CTL_ACTION_OVERLAP:
11130		mtx_unlock(&lun->lun_lock);
11131		ctl_set_overlapped_cmd(ctsio);
11132		ctl_done((union ctl_io *)ctsio);
11133		break;
11134	case CTL_ACTION_OVERLAP_TAG:
11135		mtx_unlock(&lun->lun_lock);
11136		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11137		ctl_done((union ctl_io *)ctsio);
11138		break;
11139	case CTL_ACTION_ERROR:
11140	default:
11141		mtx_unlock(&lun->lun_lock);
11142		ctl_set_internal_failure(ctsio,
11143					 /*sks_valid*/ 0,
11144					 /*retry_count*/ 0);
11145		ctl_done((union ctl_io *)ctsio);
11146		break;
11147	}
11148	return (retval);
11149}
11150
11151const struct ctl_cmd_entry *
11152ctl_get_cmd_entry(struct ctl_scsiio *ctsio)
11153{
11154	const struct ctl_cmd_entry *entry;
11155	int service_action;
11156
11157	entry = &ctl_cmd_table[ctsio->cdb[0]];
11158	if (entry->flags & CTL_CMD_FLAG_SA5) {
11159		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11160		entry = &((const struct ctl_cmd_entry *)
11161		    entry->execute)[service_action];
11162	}
11163	return (entry);
11164}
11165
11166const struct ctl_cmd_entry *
11167ctl_validate_command(struct ctl_scsiio *ctsio)
11168{
11169	const struct ctl_cmd_entry *entry;
11170	int i;
11171	uint8_t diff;
11172
11173	entry = ctl_get_cmd_entry(ctsio);
11174	if (entry->execute == NULL) {
11175		ctl_set_invalid_opcode(ctsio);
11176		ctl_done((union ctl_io *)ctsio);
11177		return (NULL);
11178	}
11179	KASSERT(entry->length > 0,
11180	    ("Not defined length for command 0x%02x/0x%02x",
11181	     ctsio->cdb[0], ctsio->cdb[1]));
11182	for (i = 1; i < entry->length; i++) {
11183		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11184		if (diff == 0)
11185			continue;
11186		ctl_set_invalid_field(ctsio,
11187				      /*sks_valid*/ 1,
11188				      /*command*/ 1,
11189				      /*field*/ i,
11190				      /*bit_valid*/ 1,
11191				      /*bit*/ fls(diff) - 1);
11192		ctl_done((union ctl_io *)ctsio);
11193		return (NULL);
11194	}
11195	return (entry);
11196}
11197
11198static int
11199ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11200{
11201
11202	switch (lun_type) {
11203	case T_PROCESSOR:
11204		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11205		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11206			return (0);
11207		break;
11208	case T_DIRECT:
11209		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11210		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11211			return (0);
11212		break;
11213	default:
11214		return (0);
11215	}
11216	return (1);
11217}
11218
11219static int
11220ctl_scsiio(struct ctl_scsiio *ctsio)
11221{
11222	int retval;
11223	const struct ctl_cmd_entry *entry;
11224
11225	retval = CTL_RETVAL_COMPLETE;
11226
11227	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11228
11229	entry = ctl_get_cmd_entry(ctsio);
11230
11231	/*
11232	 * If this I/O has been aborted, just send it straight to
11233	 * ctl_done() without executing it.
11234	 */
11235	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11236		ctl_done((union ctl_io *)ctsio);
11237		goto bailout;
11238	}
11239
11240	/*
11241	 * All the checks should have been handled by ctl_scsiio_precheck().
11242	 * We should be clear now to just execute the I/O.
11243	 */
11244	retval = entry->execute(ctsio);
11245
11246bailout:
11247	return (retval);
11248}
11249
11250/*
11251 * Since we only implement one target right now, a bus reset simply resets
11252 * our single target.
11253 */
11254static int
11255ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11256{
11257	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11258}
11259
11260static int
11261ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11262		 ctl_ua_type ua_type)
11263{
11264	struct ctl_lun *lun;
11265	int retval;
11266
11267	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11268		union ctl_ha_msg msg_info;
11269
11270		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11271		msg_info.hdr.nexus = io->io_hdr.nexus;
11272		if (ua_type==CTL_UA_TARG_RESET)
11273			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11274		else
11275			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11276		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11277		msg_info.hdr.original_sc = NULL;
11278		msg_info.hdr.serializing_sc = NULL;
11279		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11280		    (void *)&msg_info, sizeof(msg_info), 0)) {
11281		}
11282	}
11283	retval = 0;
11284
11285	mtx_lock(&ctl_softc->ctl_lock);
11286	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11287		retval += ctl_lun_reset(lun, io, ua_type);
11288	mtx_unlock(&ctl_softc->ctl_lock);
11289
11290	return (retval);
11291}
11292
11293/*
11294 * The LUN should always be set.  The I/O is optional, and is used to
11295 * distinguish between I/Os sent by this initiator, and by other
11296 * initiators.  We set unit attention for initiators other than this one.
11297 * SAM-3 is vague on this point.  It does say that a unit attention should
11298 * be established for other initiators when a LUN is reset (see section
11299 * 5.7.3), but it doesn't specifically say that the unit attention should
11300 * be established for this particular initiator when a LUN is reset.  Here
11301 * is the relevant text, from SAM-3 rev 8:
11302 *
11303 * 5.7.2 When a SCSI initiator port aborts its own tasks
11304 *
11305 * When a SCSI initiator port causes its own task(s) to be aborted, no
11306 * notification that the task(s) have been aborted shall be returned to
11307 * the SCSI initiator port other than the completion response for the
11308 * command or task management function action that caused the task(s) to
11309 * be aborted and notification(s) associated with related effects of the
11310 * action (e.g., a reset unit attention condition).
11311 *
11312 * XXX KDM for now, we're setting unit attention for all initiators.
11313 */
11314static int
11315ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11316{
11317	union ctl_io *xio;
11318#if 0
11319	uint32_t initindex;
11320#endif
11321	int i;
11322
11323	mtx_lock(&lun->lun_lock);
11324	/*
11325	 * Run through the OOA queue and abort each I/O.
11326	 */
11327#if 0
11328	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11329#endif
11330	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11331	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11332		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11333	}
11334
11335	/*
11336	 * This version sets unit attention for every
11337	 */
11338#if 0
11339	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11340	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11341		if (initindex == i)
11342			continue;
11343		lun->pending_sense[i].ua_pending |= ua_type;
11344	}
11345#endif
11346
11347	/*
11348	 * A reset (any kind, really) clears reservations established with
11349	 * RESERVE/RELEASE.  It does not clear reservations established
11350	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11351	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11352	 * reservations made with the RESERVE/RELEASE commands, because
11353	 * those commands are obsolete in SPC-3.
11354	 */
11355	lun->flags &= ~CTL_LUN_RESERVED;
11356
11357	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11358		ctl_clear_mask(lun->have_ca, i);
11359		lun->pending_sense[i].ua_pending |= ua_type;
11360	}
11361	mtx_lock(&lun->lun_lock);
11362
11363	return (0);
11364}
11365
11366static int
11367ctl_abort_task(union ctl_io *io)
11368{
11369	union ctl_io *xio;
11370	struct ctl_lun *lun;
11371	struct ctl_softc *ctl_softc;
11372#if 0
11373	struct sbuf sb;
11374	char printbuf[128];
11375#endif
11376	int found;
11377	uint32_t targ_lun;
11378
11379	ctl_softc = control_softc;
11380	found = 0;
11381
11382	/*
11383	 * Look up the LUN.
11384	 */
11385	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11386	mtx_lock(&ctl_softc->ctl_lock);
11387	if ((targ_lun < CTL_MAX_LUNS)
11388	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11389		lun = ctl_softc->ctl_luns[targ_lun];
11390	else {
11391		mtx_unlock(&ctl_softc->ctl_lock);
11392		goto bailout;
11393	}
11394
11395#if 0
11396	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11397	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11398#endif
11399
11400	mtx_lock(&lun->lun_lock);
11401	mtx_unlock(&ctl_softc->ctl_lock);
11402	/*
11403	 * Run through the OOA queue and attempt to find the given I/O.
11404	 * The target port, initiator ID, tag type and tag number have to
11405	 * match the values that we got from the initiator.  If we have an
11406	 * untagged command to abort, simply abort the first untagged command
11407	 * we come to.  We only allow one untagged command at a time of course.
11408	 */
11409#if 0
11410	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11411#endif
11412	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11413	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11414#if 0
11415		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11416
11417		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11418			    lun->lun, xio->scsiio.tag_num,
11419			    xio->scsiio.tag_type,
11420			    (xio->io_hdr.blocked_links.tqe_prev
11421			    == NULL) ? "" : " BLOCKED",
11422			    (xio->io_hdr.flags &
11423			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11424			    (xio->io_hdr.flags &
11425			    CTL_FLAG_ABORT) ? " ABORT" : "",
11426			    (xio->io_hdr.flags &
11427			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11428		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11429		sbuf_finish(&sb);
11430		printf("%s\n", sbuf_data(&sb));
11431#endif
11432
11433		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11434		 && (xio->io_hdr.nexus.initid.id ==
11435		     io->io_hdr.nexus.initid.id)) {
11436			/*
11437			 * If the abort says that the task is untagged, the
11438			 * task in the queue must be untagged.  Otherwise,
11439			 * we just check to see whether the tag numbers
11440			 * match.  This is because the QLogic firmware
11441			 * doesn't pass back the tag type in an abort
11442			 * request.
11443			 */
11444#if 0
11445			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11446			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11447			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11448#endif
11449			/*
11450			 * XXX KDM we've got problems with FC, because it
11451			 * doesn't send down a tag type with aborts.  So we
11452			 * can only really go by the tag number...
11453			 * This may cause problems with parallel SCSI.
11454			 * Need to figure that out!!
11455			 */
11456			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11457				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11458				found = 1;
11459				if ((io->io_hdr.flags &
11460				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11461				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11462					union ctl_ha_msg msg_info;
11463
11464					io->io_hdr.flags |=
11465					                CTL_FLAG_SENT_2OTHER_SC;
11466					msg_info.hdr.nexus = io->io_hdr.nexus;
11467					msg_info.task.task_action =
11468						CTL_TASK_ABORT_TASK;
11469					msg_info.task.tag_num =
11470						io->taskio.tag_num;
11471					msg_info.task.tag_type =
11472						io->taskio.tag_type;
11473					msg_info.hdr.msg_type =
11474						CTL_MSG_MANAGE_TASKS;
11475					msg_info.hdr.original_sc = NULL;
11476					msg_info.hdr.serializing_sc = NULL;
11477#if 0
11478					printf("Sent Abort to other side\n");
11479#endif
11480					if (CTL_HA_STATUS_SUCCESS !=
11481					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11482		    				(void *)&msg_info,
11483						sizeof(msg_info), 0)) {
11484					}
11485				}
11486#if 0
11487				printf("ctl_abort_task: found I/O to abort\n");
11488#endif
11489				break;
11490			}
11491		}
11492	}
11493	mtx_unlock(&lun->lun_lock);
11494
11495bailout:
11496
11497	if (found == 0) {
11498		/*
11499		 * This isn't really an error.  It's entirely possible for
11500		 * the abort and command completion to cross on the wire.
11501		 * This is more of an informative/diagnostic error.
11502		 */
11503#if 0
11504		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11505		       "%d:%d:%d:%d tag %d type %d\n",
11506		       io->io_hdr.nexus.initid.id,
11507		       io->io_hdr.nexus.targ_port,
11508		       io->io_hdr.nexus.targ_target.id,
11509		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11510		       io->taskio.tag_type);
11511#endif
11512		return (1);
11513	} else
11514		return (0);
11515}
11516
11517/*
11518 * This routine cannot block!  It must be callable from an interrupt
11519 * handler as well as from the work thread.
11520 */
11521static void
11522ctl_run_task(union ctl_io *io)
11523{
11524	struct ctl_softc *ctl_softc;
11525	int retval;
11526	const char *task_desc;
11527
11528	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11529
11530	ctl_softc = control_softc;
11531	retval = 0;
11532
11533	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11534	    ("ctl_run_task: Unextected io_type %d\n",
11535	     io->io_hdr.io_type));
11536
11537	task_desc = ctl_scsi_task_string(&io->taskio);
11538	if (task_desc != NULL) {
11539#ifdef NEEDTOPORT
11540		csevent_log(CSC_CTL | CSC_SHELF_SW |
11541			    CTL_TASK_REPORT,
11542			    csevent_LogType_Trace,
11543			    csevent_Severity_Information,
11544			    csevent_AlertLevel_Green,
11545			    csevent_FRU_Firmware,
11546			    csevent_FRU_Unknown,
11547			    "CTL: received task: %s",task_desc);
11548#endif
11549	} else {
11550#ifdef NEEDTOPORT
11551		csevent_log(CSC_CTL | CSC_SHELF_SW |
11552			    CTL_TASK_REPORT,
11553			    csevent_LogType_Trace,
11554			    csevent_Severity_Information,
11555			    csevent_AlertLevel_Green,
11556			    csevent_FRU_Firmware,
11557			    csevent_FRU_Unknown,
11558			    "CTL: received unknown task "
11559			    "type: %d (%#x)",
11560			    io->taskio.task_action,
11561			    io->taskio.task_action);
11562#endif
11563	}
11564	switch (io->taskio.task_action) {
11565	case CTL_TASK_ABORT_TASK:
11566		retval = ctl_abort_task(io);
11567		break;
11568	case CTL_TASK_ABORT_TASK_SET:
11569		break;
11570	case CTL_TASK_CLEAR_ACA:
11571		break;
11572	case CTL_TASK_CLEAR_TASK_SET:
11573		break;
11574	case CTL_TASK_LUN_RESET: {
11575		struct ctl_lun *lun;
11576		uint32_t targ_lun;
11577		int retval;
11578
11579		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11580		mtx_lock(&ctl_softc->ctl_lock);
11581		if ((targ_lun < CTL_MAX_LUNS)
11582		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11583			lun = ctl_softc->ctl_luns[targ_lun];
11584		else {
11585			mtx_unlock(&ctl_softc->ctl_lock);
11586			retval = 1;
11587			break;
11588		}
11589
11590		if (!(io->io_hdr.flags &
11591		    CTL_FLAG_FROM_OTHER_SC)) {
11592			union ctl_ha_msg msg_info;
11593
11594			io->io_hdr.flags |=
11595				CTL_FLAG_SENT_2OTHER_SC;
11596			msg_info.hdr.msg_type =
11597				CTL_MSG_MANAGE_TASKS;
11598			msg_info.hdr.nexus = io->io_hdr.nexus;
11599			msg_info.task.task_action =
11600				CTL_TASK_LUN_RESET;
11601			msg_info.hdr.original_sc = NULL;
11602			msg_info.hdr.serializing_sc = NULL;
11603			if (CTL_HA_STATUS_SUCCESS !=
11604			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11605			    (void *)&msg_info,
11606			    sizeof(msg_info), 0)) {
11607			}
11608		}
11609
11610		retval = ctl_lun_reset(lun, io,
11611				       CTL_UA_LUN_RESET);
11612		mtx_unlock(&ctl_softc->ctl_lock);
11613		break;
11614	}
11615	case CTL_TASK_TARGET_RESET:
11616		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
11617		break;
11618	case CTL_TASK_BUS_RESET:
11619		retval = ctl_bus_reset(ctl_softc, io);
11620		break;
11621	case CTL_TASK_PORT_LOGIN:
11622		break;
11623	case CTL_TASK_PORT_LOGOUT:
11624		break;
11625	default:
11626		printf("ctl_run_task: got unknown task management event %d\n",
11627		       io->taskio.task_action);
11628		break;
11629	}
11630	if (retval == 0)
11631		io->io_hdr.status = CTL_SUCCESS;
11632	else
11633		io->io_hdr.status = CTL_ERROR;
11634
11635	/*
11636	 * This will queue this I/O to the done queue, but the
11637	 * work thread won't be able to process it until we
11638	 * return and the lock is released.
11639	 */
11640	ctl_done(io);
11641}
11642
11643/*
11644 * For HA operation.  Handle commands that come in from the other
11645 * controller.
11646 */
11647static void
11648ctl_handle_isc(union ctl_io *io)
11649{
11650	int free_io;
11651	struct ctl_lun *lun;
11652	struct ctl_softc *ctl_softc;
11653	uint32_t targ_lun;
11654
11655	ctl_softc = control_softc;
11656
11657	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11658	lun = ctl_softc->ctl_luns[targ_lun];
11659
11660	switch (io->io_hdr.msg_type) {
11661	case CTL_MSG_SERIALIZE:
11662		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11663		break;
11664	case CTL_MSG_R2R: {
11665		const struct ctl_cmd_entry *entry;
11666
11667		/*
11668		 * This is only used in SER_ONLY mode.
11669		 */
11670		free_io = 0;
11671		entry = ctl_get_cmd_entry(&io->scsiio);
11672		mtx_lock(&lun->lun_lock);
11673		if (ctl_scsiio_lun_check(ctl_softc, lun,
11674		    entry, (struct ctl_scsiio *)io) != 0) {
11675			mtx_unlock(&lun->lun_lock);
11676			ctl_done(io);
11677			break;
11678		}
11679		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11680		mtx_unlock(&lun->lun_lock);
11681		ctl_enqueue_rtr(io);
11682		break;
11683	}
11684	case CTL_MSG_FINISH_IO:
11685		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11686			free_io = 0;
11687			ctl_done(io);
11688		} else {
11689			free_io = 1;
11690			mtx_lock(&lun->lun_lock);
11691			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11692				     ooa_links);
11693			ctl_check_blocked(lun);
11694			mtx_unlock(&lun->lun_lock);
11695		}
11696		break;
11697	case CTL_MSG_PERS_ACTION:
11698		ctl_hndl_per_res_out_on_other_sc(
11699			(union ctl_ha_msg *)&io->presio.pr_msg);
11700		free_io = 1;
11701		break;
11702	case CTL_MSG_BAD_JUJU:
11703		free_io = 0;
11704		ctl_done(io);
11705		break;
11706	case CTL_MSG_DATAMOVE:
11707		/* Only used in XFER mode */
11708		free_io = 0;
11709		ctl_datamove_remote(io);
11710		break;
11711	case CTL_MSG_DATAMOVE_DONE:
11712		/* Only used in XFER mode */
11713		free_io = 0;
11714		io->scsiio.be_move_done(io);
11715		break;
11716	default:
11717		free_io = 1;
11718		printf("%s: Invalid message type %d\n",
11719		       __func__, io->io_hdr.msg_type);
11720		break;
11721	}
11722	if (free_io)
11723		ctl_free_io(io);
11724
11725}
11726
11727
11728/*
11729 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11730 * there is no match.
11731 */
11732static ctl_lun_error_pattern
11733ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11734{
11735	const struct ctl_cmd_entry *entry;
11736	ctl_lun_error_pattern filtered_pattern, pattern;
11737
11738	pattern = desc->error_pattern;
11739
11740	/*
11741	 * XXX KDM we need more data passed into this function to match a
11742	 * custom pattern, and we actually need to implement custom pattern
11743	 * matching.
11744	 */
11745	if (pattern & CTL_LUN_PAT_CMD)
11746		return (CTL_LUN_PAT_CMD);
11747
11748	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11749		return (CTL_LUN_PAT_ANY);
11750
11751	entry = ctl_get_cmd_entry(ctsio);
11752
11753	filtered_pattern = entry->pattern & pattern;
11754
11755	/*
11756	 * If the user requested specific flags in the pattern (e.g.
11757	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11758	 * flags.
11759	 *
11760	 * If the user did not specify any flags, it doesn't matter whether
11761	 * or not the command supports the flags.
11762	 */
11763	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11764	     (pattern & ~CTL_LUN_PAT_MASK))
11765		return (CTL_LUN_PAT_NONE);
11766
11767	/*
11768	 * If the user asked for a range check, see if the requested LBA
11769	 * range overlaps with this command's LBA range.
11770	 */
11771	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11772		uint64_t lba1;
11773		uint32_t len1;
11774		ctl_action action;
11775		int retval;
11776
11777		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11778		if (retval != 0)
11779			return (CTL_LUN_PAT_NONE);
11780
11781		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11782					      desc->lba_range.len);
11783		/*
11784		 * A "pass" means that the LBA ranges don't overlap, so
11785		 * this doesn't match the user's range criteria.
11786		 */
11787		if (action == CTL_ACTION_PASS)
11788			return (CTL_LUN_PAT_NONE);
11789	}
11790
11791	return (filtered_pattern);
11792}
11793
11794static void
11795ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11796{
11797	struct ctl_error_desc *desc, *desc2;
11798
11799	mtx_assert(&lun->lun_lock, MA_OWNED);
11800
11801	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11802		ctl_lun_error_pattern pattern;
11803		/*
11804		 * Check to see whether this particular command matches
11805		 * the pattern in the descriptor.
11806		 */
11807		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11808		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11809			continue;
11810
11811		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11812		case CTL_LUN_INJ_ABORTED:
11813			ctl_set_aborted(&io->scsiio);
11814			break;
11815		case CTL_LUN_INJ_MEDIUM_ERR:
11816			ctl_set_medium_error(&io->scsiio);
11817			break;
11818		case CTL_LUN_INJ_UA:
11819			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11820			 * OCCURRED */
11821			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11822			break;
11823		case CTL_LUN_INJ_CUSTOM:
11824			/*
11825			 * We're assuming the user knows what he is doing.
11826			 * Just copy the sense information without doing
11827			 * checks.
11828			 */
11829			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11830			      ctl_min(sizeof(desc->custom_sense),
11831				      sizeof(io->scsiio.sense_data)));
11832			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11833			io->scsiio.sense_len = SSD_FULL_SIZE;
11834			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11835			break;
11836		case CTL_LUN_INJ_NONE:
11837		default:
11838			/*
11839			 * If this is an error injection type we don't know
11840			 * about, clear the continuous flag (if it is set)
11841			 * so it will get deleted below.
11842			 */
11843			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11844			break;
11845		}
11846		/*
11847		 * By default, each error injection action is a one-shot
11848		 */
11849		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11850			continue;
11851
11852		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11853
11854		free(desc, M_CTL);
11855	}
11856}
11857
11858#ifdef CTL_IO_DELAY
11859static void
11860ctl_datamove_timer_wakeup(void *arg)
11861{
11862	union ctl_io *io;
11863
11864	io = (union ctl_io *)arg;
11865
11866	ctl_datamove(io);
11867}
11868#endif /* CTL_IO_DELAY */
11869
11870void
11871ctl_datamove(union ctl_io *io)
11872{
11873	void (*fe_datamove)(union ctl_io *io);
11874
11875	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11876
11877	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11878
11879#ifdef CTL_TIME_IO
11880	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11881		char str[256];
11882		char path_str[64];
11883		struct sbuf sb;
11884
11885		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11886		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11887
11888		sbuf_cat(&sb, path_str);
11889		switch (io->io_hdr.io_type) {
11890		case CTL_IO_SCSI:
11891			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11892			sbuf_printf(&sb, "\n");
11893			sbuf_cat(&sb, path_str);
11894			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11895				    io->scsiio.tag_num, io->scsiio.tag_type);
11896			break;
11897		case CTL_IO_TASK:
11898			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11899				    "Tag Type: %d\n", io->taskio.task_action,
11900				    io->taskio.tag_num, io->taskio.tag_type);
11901			break;
11902		default:
11903			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11904			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11905			break;
11906		}
11907		sbuf_cat(&sb, path_str);
11908		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11909			    (intmax_t)time_uptime - io->io_hdr.start_time);
11910		sbuf_finish(&sb);
11911		printf("%s", sbuf_data(&sb));
11912	}
11913#endif /* CTL_TIME_IO */
11914
11915#ifdef CTL_IO_DELAY
11916	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11917		struct ctl_lun *lun;
11918
11919		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11920
11921		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11922	} else {
11923		struct ctl_lun *lun;
11924
11925		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11926		if ((lun != NULL)
11927		 && (lun->delay_info.datamove_delay > 0)) {
11928			struct callout *callout;
11929
11930			callout = (struct callout *)&io->io_hdr.timer_bytes;
11931			callout_init(callout, /*mpsafe*/ 1);
11932			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11933			callout_reset(callout,
11934				      lun->delay_info.datamove_delay * hz,
11935				      ctl_datamove_timer_wakeup, io);
11936			if (lun->delay_info.datamove_type ==
11937			    CTL_DELAY_TYPE_ONESHOT)
11938				lun->delay_info.datamove_delay = 0;
11939			return;
11940		}
11941	}
11942#endif
11943
11944	/*
11945	 * This command has been aborted.  Set the port status, so we fail
11946	 * the data move.
11947	 */
11948	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11949		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11950		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11951		       io->io_hdr.nexus.targ_port,
11952		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11953		       io->io_hdr.nexus.targ_lun);
11954		io->io_hdr.status = CTL_CMD_ABORTED;
11955		io->io_hdr.port_status = 31337;
11956		/*
11957		 * Note that the backend, in this case, will get the
11958		 * callback in its context.  In other cases it may get
11959		 * called in the frontend's interrupt thread context.
11960		 */
11961		io->scsiio.be_move_done(io);
11962		return;
11963	}
11964
11965	/*
11966	 * If we're in XFER mode and this I/O is from the other shelf
11967	 * controller, we need to send the DMA to the other side to
11968	 * actually transfer the data to/from the host.  In serialize only
11969	 * mode the transfer happens below CTL and ctl_datamove() is only
11970	 * called on the machine that originally received the I/O.
11971	 */
11972	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11973	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11974		union ctl_ha_msg msg;
11975		uint32_t sg_entries_sent;
11976		int do_sg_copy;
11977		int i;
11978
11979		memset(&msg, 0, sizeof(msg));
11980		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11981		msg.hdr.original_sc = io->io_hdr.original_sc;
11982		msg.hdr.serializing_sc = io;
11983		msg.hdr.nexus = io->io_hdr.nexus;
11984		msg.dt.flags = io->io_hdr.flags;
11985		/*
11986		 * We convert everything into a S/G list here.  We can't
11987		 * pass by reference, only by value between controllers.
11988		 * So we can't pass a pointer to the S/G list, only as many
11989		 * S/G entries as we can fit in here.  If it's possible for
11990		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11991		 * then we need to break this up into multiple transfers.
11992		 */
11993		if (io->scsiio.kern_sg_entries == 0) {
11994			msg.dt.kern_sg_entries = 1;
11995			/*
11996			 * If this is in cached memory, flush the cache
11997			 * before we send the DMA request to the other
11998			 * controller.  We want to do this in either the
11999			 * read or the write case.  The read case is
12000			 * straightforward.  In the write case, we want to
12001			 * make sure nothing is in the local cache that
12002			 * could overwrite the DMAed data.
12003			 */
12004			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12005				/*
12006				 * XXX KDM use bus_dmamap_sync() here.
12007				 */
12008			}
12009
12010			/*
12011			 * Convert to a physical address if this is a
12012			 * virtual address.
12013			 */
12014			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12015				msg.dt.sg_list[0].addr =
12016					io->scsiio.kern_data_ptr;
12017			} else {
12018				/*
12019				 * XXX KDM use busdma here!
12020				 */
12021#if 0
12022				msg.dt.sg_list[0].addr = (void *)
12023					vtophys(io->scsiio.kern_data_ptr);
12024#endif
12025			}
12026
12027			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12028			do_sg_copy = 0;
12029		} else {
12030			struct ctl_sg_entry *sgl;
12031
12032			do_sg_copy = 1;
12033			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12034			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12035			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12036				/*
12037				 * XXX KDM use bus_dmamap_sync() here.
12038				 */
12039			}
12040		}
12041
12042		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12043		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12044		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12045		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12046		msg.dt.sg_sequence = 0;
12047
12048		/*
12049		 * Loop until we've sent all of the S/G entries.  On the
12050		 * other end, we'll recompose these S/G entries into one
12051		 * contiguous list before passing it to the
12052		 */
12053		for (sg_entries_sent = 0; sg_entries_sent <
12054		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12055			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12056				sizeof(msg.dt.sg_list[0])),
12057				msg.dt.kern_sg_entries - sg_entries_sent);
12058
12059			if (do_sg_copy != 0) {
12060				struct ctl_sg_entry *sgl;
12061				int j;
12062
12063				sgl = (struct ctl_sg_entry *)
12064					io->scsiio.kern_data_ptr;
12065				/*
12066				 * If this is in cached memory, flush the cache
12067				 * before we send the DMA request to the other
12068				 * controller.  We want to do this in either
12069				 * the * read or the write case.  The read
12070				 * case is straightforward.  In the write
12071				 * case, we want to make sure nothing is
12072				 * in the local cache that could overwrite
12073				 * the DMAed data.
12074				 */
12075
12076				for (i = sg_entries_sent, j = 0;
12077				     i < msg.dt.cur_sg_entries; i++, j++) {
12078					if ((io->io_hdr.flags &
12079					     CTL_FLAG_NO_DATASYNC) == 0) {
12080						/*
12081						 * XXX KDM use bus_dmamap_sync()
12082						 */
12083					}
12084					if ((io->io_hdr.flags &
12085					     CTL_FLAG_BUS_ADDR) == 0) {
12086						/*
12087						 * XXX KDM use busdma.
12088						 */
12089#if 0
12090						msg.dt.sg_list[j].addr =(void *)
12091						       vtophys(sgl[i].addr);
12092#endif
12093					} else {
12094						msg.dt.sg_list[j].addr =
12095							sgl[i].addr;
12096					}
12097					msg.dt.sg_list[j].len = sgl[i].len;
12098				}
12099			}
12100
12101			sg_entries_sent += msg.dt.cur_sg_entries;
12102			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12103				msg.dt.sg_last = 1;
12104			else
12105				msg.dt.sg_last = 0;
12106
12107			/*
12108			 * XXX KDM drop and reacquire the lock here?
12109			 */
12110			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12111			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12112				/*
12113				 * XXX do something here.
12114				 */
12115			}
12116
12117			msg.dt.sent_sg_entries = sg_entries_sent;
12118		}
12119		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12120		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12121			ctl_failover_io(io, /*have_lock*/ 0);
12122
12123	} else {
12124
12125		/*
12126		 * Lookup the fe_datamove() function for this particular
12127		 * front end.
12128		 */
12129		fe_datamove =
12130		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12131
12132		fe_datamove(io);
12133	}
12134}
12135
12136static void
12137ctl_send_datamove_done(union ctl_io *io, int have_lock)
12138{
12139	union ctl_ha_msg msg;
12140	int isc_status;
12141
12142	memset(&msg, 0, sizeof(msg));
12143
12144	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12145	msg.hdr.original_sc = io;
12146	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12147	msg.hdr.nexus = io->io_hdr.nexus;
12148	msg.hdr.status = io->io_hdr.status;
12149	msg.scsi.tag_num = io->scsiio.tag_num;
12150	msg.scsi.tag_type = io->scsiio.tag_type;
12151	msg.scsi.scsi_status = io->scsiio.scsi_status;
12152	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12153	       sizeof(io->scsiio.sense_data));
12154	msg.scsi.sense_len = io->scsiio.sense_len;
12155	msg.scsi.sense_residual = io->scsiio.sense_residual;
12156	msg.scsi.fetd_status = io->io_hdr.port_status;
12157	msg.scsi.residual = io->scsiio.residual;
12158	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12159
12160	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12161		ctl_failover_io(io, /*have_lock*/ have_lock);
12162		return;
12163	}
12164
12165	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12166	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12167		/* XXX do something if this fails */
12168	}
12169
12170}
12171
12172/*
12173 * The DMA to the remote side is done, now we need to tell the other side
12174 * we're done so it can continue with its data movement.
12175 */
12176static void
12177ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12178{
12179	union ctl_io *io;
12180
12181	io = rq->context;
12182
12183	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12184		printf("%s: ISC DMA write failed with error %d", __func__,
12185		       rq->ret);
12186		ctl_set_internal_failure(&io->scsiio,
12187					 /*sks_valid*/ 1,
12188					 /*retry_count*/ rq->ret);
12189	}
12190
12191	ctl_dt_req_free(rq);
12192
12193	/*
12194	 * In this case, we had to malloc the memory locally.  Free it.
12195	 */
12196	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12197		int i;
12198		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12199			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12200	}
12201	/*
12202	 * The data is in local and remote memory, so now we need to send
12203	 * status (good or back) back to the other side.
12204	 */
12205	ctl_send_datamove_done(io, /*have_lock*/ 0);
12206}
12207
12208/*
12209 * We've moved the data from the host/controller into local memory.  Now we
12210 * need to push it over to the remote controller's memory.
12211 */
12212static int
12213ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12214{
12215	int retval;
12216
12217	retval = 0;
12218
12219	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12220					  ctl_datamove_remote_write_cb);
12221
12222	return (retval);
12223}
12224
12225static void
12226ctl_datamove_remote_write(union ctl_io *io)
12227{
12228	int retval;
12229	void (*fe_datamove)(union ctl_io *io);
12230
12231	/*
12232	 * - Get the data from the host/HBA into local memory.
12233	 * - DMA memory from the local controller to the remote controller.
12234	 * - Send status back to the remote controller.
12235	 */
12236
12237	retval = ctl_datamove_remote_sgl_setup(io);
12238	if (retval != 0)
12239		return;
12240
12241	/* Switch the pointer over so the FETD knows what to do */
12242	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12243
12244	/*
12245	 * Use a custom move done callback, since we need to send completion
12246	 * back to the other controller, not to the backend on this side.
12247	 */
12248	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12249
12250	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12251
12252	fe_datamove(io);
12253
12254	return;
12255
12256}
12257
12258static int
12259ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12260{
12261#if 0
12262	char str[256];
12263	char path_str[64];
12264	struct sbuf sb;
12265#endif
12266
12267	/*
12268	 * In this case, we had to malloc the memory locally.  Free it.
12269	 */
12270	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12271		int i;
12272		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12273			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12274	}
12275
12276#if 0
12277	scsi_path_string(io, path_str, sizeof(path_str));
12278	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12279	sbuf_cat(&sb, path_str);
12280	scsi_command_string(&io->scsiio, NULL, &sb);
12281	sbuf_printf(&sb, "\n");
12282	sbuf_cat(&sb, path_str);
12283	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12284		    io->scsiio.tag_num, io->scsiio.tag_type);
12285	sbuf_cat(&sb, path_str);
12286	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12287		    io->io_hdr.flags, io->io_hdr.status);
12288	sbuf_finish(&sb);
12289	printk("%s", sbuf_data(&sb));
12290#endif
12291
12292
12293	/*
12294	 * The read is done, now we need to send status (good or bad) back
12295	 * to the other side.
12296	 */
12297	ctl_send_datamove_done(io, /*have_lock*/ 0);
12298
12299	return (0);
12300}
12301
12302static void
12303ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12304{
12305	union ctl_io *io;
12306	void (*fe_datamove)(union ctl_io *io);
12307
12308	io = rq->context;
12309
12310	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12311		printf("%s: ISC DMA read failed with error %d", __func__,
12312		       rq->ret);
12313		ctl_set_internal_failure(&io->scsiio,
12314					 /*sks_valid*/ 1,
12315					 /*retry_count*/ rq->ret);
12316	}
12317
12318	ctl_dt_req_free(rq);
12319
12320	/* Switch the pointer over so the FETD knows what to do */
12321	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12322
12323	/*
12324	 * Use a custom move done callback, since we need to send completion
12325	 * back to the other controller, not to the backend on this side.
12326	 */
12327	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12328
12329	/* XXX KDM add checks like the ones in ctl_datamove? */
12330
12331	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12332
12333	fe_datamove(io);
12334}
12335
12336static int
12337ctl_datamove_remote_sgl_setup(union ctl_io *io)
12338{
12339	struct ctl_sg_entry *local_sglist, *remote_sglist;
12340	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12341	struct ctl_softc *softc;
12342	int retval;
12343	int i;
12344
12345	retval = 0;
12346	softc = control_softc;
12347
12348	local_sglist = io->io_hdr.local_sglist;
12349	local_dma_sglist = io->io_hdr.local_dma_sglist;
12350	remote_sglist = io->io_hdr.remote_sglist;
12351	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12352
12353	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12354		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12355			local_sglist[i].len = remote_sglist[i].len;
12356
12357			/*
12358			 * XXX Detect the situation where the RS-level I/O
12359			 * redirector on the other side has already read the
12360			 * data off of the AOR RS on this side, and
12361			 * transferred it to remote (mirror) memory on the
12362			 * other side.  Since we already have the data in
12363			 * memory here, we just need to use it.
12364			 *
12365			 * XXX KDM this can probably be removed once we
12366			 * get the cache device code in and take the
12367			 * current AOR implementation out.
12368			 */
12369#ifdef NEEDTOPORT
12370			if ((remote_sglist[i].addr >=
12371			     (void *)vtophys(softc->mirr->addr))
12372			 && (remote_sglist[i].addr <
12373			     ((void *)vtophys(softc->mirr->addr) +
12374			     CacheMirrorOffset))) {
12375				local_sglist[i].addr = remote_sglist[i].addr -
12376					CacheMirrorOffset;
12377				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12378				     CTL_FLAG_DATA_IN)
12379					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12380			} else {
12381				local_sglist[i].addr = remote_sglist[i].addr +
12382					CacheMirrorOffset;
12383			}
12384#endif
12385#if 0
12386			printf("%s: local %p, remote %p, len %d\n",
12387			       __func__, local_sglist[i].addr,
12388			       remote_sglist[i].addr, local_sglist[i].len);
12389#endif
12390		}
12391	} else {
12392		uint32_t len_to_go;
12393
12394		/*
12395		 * In this case, we don't have automatically allocated
12396		 * memory for this I/O on this controller.  This typically
12397		 * happens with internal CTL I/O -- e.g. inquiry, mode
12398		 * sense, etc.  Anything coming from RAIDCore will have
12399		 * a mirror area available.
12400		 */
12401		len_to_go = io->scsiio.kern_data_len;
12402
12403		/*
12404		 * Clear the no datasync flag, we have to use malloced
12405		 * buffers.
12406		 */
12407		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12408
12409		/*
12410		 * The difficult thing here is that the size of the various
12411		 * S/G segments may be different than the size from the
12412		 * remote controller.  That'll make it harder when DMAing
12413		 * the data back to the other side.
12414		 */
12415		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12416		     sizeof(io->io_hdr.remote_sglist[0])) &&
12417		     (len_to_go > 0); i++) {
12418			local_sglist[i].len = ctl_min(len_to_go, 131072);
12419			CTL_SIZE_8B(local_dma_sglist[i].len,
12420				    local_sglist[i].len);
12421			local_sglist[i].addr =
12422				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12423
12424			local_dma_sglist[i].addr = local_sglist[i].addr;
12425
12426			if (local_sglist[i].addr == NULL) {
12427				int j;
12428
12429				printf("malloc failed for %zd bytes!",
12430				       local_dma_sglist[i].len);
12431				for (j = 0; j < i; j++) {
12432					free(local_sglist[j].addr, M_CTL);
12433				}
12434				ctl_set_internal_failure(&io->scsiio,
12435							 /*sks_valid*/ 1,
12436							 /*retry_count*/ 4857);
12437				retval = 1;
12438				goto bailout_error;
12439
12440			}
12441			/* XXX KDM do we need a sync here? */
12442
12443			len_to_go -= local_sglist[i].len;
12444		}
12445		/*
12446		 * Reset the number of S/G entries accordingly.  The
12447		 * original number of S/G entries is available in
12448		 * rem_sg_entries.
12449		 */
12450		io->scsiio.kern_sg_entries = i;
12451
12452#if 0
12453		printf("%s: kern_sg_entries = %d\n", __func__,
12454		       io->scsiio.kern_sg_entries);
12455		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12456			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12457			       local_sglist[i].addr, local_sglist[i].len,
12458			       local_dma_sglist[i].len);
12459#endif
12460	}
12461
12462
12463	return (retval);
12464
12465bailout_error:
12466
12467	ctl_send_datamove_done(io, /*have_lock*/ 0);
12468
12469	return (retval);
12470}
12471
12472static int
12473ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12474			 ctl_ha_dt_cb callback)
12475{
12476	struct ctl_ha_dt_req *rq;
12477	struct ctl_sg_entry *remote_sglist, *local_sglist;
12478	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12479	uint32_t local_used, remote_used, total_used;
12480	int retval;
12481	int i, j;
12482
12483	retval = 0;
12484
12485	rq = ctl_dt_req_alloc();
12486
12487	/*
12488	 * If we failed to allocate the request, and if the DMA didn't fail
12489	 * anyway, set busy status.  This is just a resource allocation
12490	 * failure.
12491	 */
12492	if ((rq == NULL)
12493	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12494		ctl_set_busy(&io->scsiio);
12495
12496	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12497
12498		if (rq != NULL)
12499			ctl_dt_req_free(rq);
12500
12501		/*
12502		 * The data move failed.  We need to return status back
12503		 * to the other controller.  No point in trying to DMA
12504		 * data to the remote controller.
12505		 */
12506
12507		ctl_send_datamove_done(io, /*have_lock*/ 0);
12508
12509		retval = 1;
12510
12511		goto bailout;
12512	}
12513
12514	local_sglist = io->io_hdr.local_sglist;
12515	local_dma_sglist = io->io_hdr.local_dma_sglist;
12516	remote_sglist = io->io_hdr.remote_sglist;
12517	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12518	local_used = 0;
12519	remote_used = 0;
12520	total_used = 0;
12521
12522	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12523		rq->ret = CTL_HA_STATUS_SUCCESS;
12524		rq->context = io;
12525		callback(rq);
12526		goto bailout;
12527	}
12528
12529	/*
12530	 * Pull/push the data over the wire from/to the other controller.
12531	 * This takes into account the possibility that the local and
12532	 * remote sglists may not be identical in terms of the size of
12533	 * the elements and the number of elements.
12534	 *
12535	 * One fundamental assumption here is that the length allocated for
12536	 * both the local and remote sglists is identical.  Otherwise, we've
12537	 * essentially got a coding error of some sort.
12538	 */
12539	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12540		int isc_ret;
12541		uint32_t cur_len, dma_length;
12542		uint8_t *tmp_ptr;
12543
12544		rq->id = CTL_HA_DATA_CTL;
12545		rq->command = command;
12546		rq->context = io;
12547
12548		/*
12549		 * Both pointers should be aligned.  But it is possible
12550		 * that the allocation length is not.  They should both
12551		 * also have enough slack left over at the end, though,
12552		 * to round up to the next 8 byte boundary.
12553		 */
12554		cur_len = ctl_min(local_sglist[i].len - local_used,
12555				  remote_sglist[j].len - remote_used);
12556
12557		/*
12558		 * In this case, we have a size issue and need to decrease
12559		 * the size, except in the case where we actually have less
12560		 * than 8 bytes left.  In that case, we need to increase
12561		 * the DMA length to get the last bit.
12562		 */
12563		if ((cur_len & 0x7) != 0) {
12564			if (cur_len > 0x7) {
12565				cur_len = cur_len - (cur_len & 0x7);
12566				dma_length = cur_len;
12567			} else {
12568				CTL_SIZE_8B(dma_length, cur_len);
12569			}
12570
12571		} else
12572			dma_length = cur_len;
12573
12574		/*
12575		 * If we had to allocate memory for this I/O, instead of using
12576		 * the non-cached mirror memory, we'll need to flush the cache
12577		 * before trying to DMA to the other controller.
12578		 *
12579		 * We could end up doing this multiple times for the same
12580		 * segment if we have a larger local segment than remote
12581		 * segment.  That shouldn't be an issue.
12582		 */
12583		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12584			/*
12585			 * XXX KDM use bus_dmamap_sync() here.
12586			 */
12587		}
12588
12589		rq->size = dma_length;
12590
12591		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12592		tmp_ptr += local_used;
12593
12594		/* Use physical addresses when talking to ISC hardware */
12595		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12596			/* XXX KDM use busdma */
12597#if 0
12598			rq->local = vtophys(tmp_ptr);
12599#endif
12600		} else
12601			rq->local = tmp_ptr;
12602
12603		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12604		tmp_ptr += remote_used;
12605		rq->remote = tmp_ptr;
12606
12607		rq->callback = NULL;
12608
12609		local_used += cur_len;
12610		if (local_used >= local_sglist[i].len) {
12611			i++;
12612			local_used = 0;
12613		}
12614
12615		remote_used += cur_len;
12616		if (remote_used >= remote_sglist[j].len) {
12617			j++;
12618			remote_used = 0;
12619		}
12620		total_used += cur_len;
12621
12622		if (total_used >= io->scsiio.kern_data_len)
12623			rq->callback = callback;
12624
12625		if ((rq->size & 0x7) != 0) {
12626			printf("%s: warning: size %d is not on 8b boundary\n",
12627			       __func__, rq->size);
12628		}
12629		if (((uintptr_t)rq->local & 0x7) != 0) {
12630			printf("%s: warning: local %p not on 8b boundary\n",
12631			       __func__, rq->local);
12632		}
12633		if (((uintptr_t)rq->remote & 0x7) != 0) {
12634			printf("%s: warning: remote %p not on 8b boundary\n",
12635			       __func__, rq->local);
12636		}
12637#if 0
12638		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12639		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12640		       rq->local, rq->remote, rq->size);
12641#endif
12642
12643		isc_ret = ctl_dt_single(rq);
12644		if (isc_ret == CTL_HA_STATUS_WAIT)
12645			continue;
12646
12647		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12648			rq->ret = CTL_HA_STATUS_SUCCESS;
12649		} else {
12650			rq->ret = isc_ret;
12651		}
12652		callback(rq);
12653		goto bailout;
12654	}
12655
12656bailout:
12657	return (retval);
12658
12659}
12660
12661static void
12662ctl_datamove_remote_read(union ctl_io *io)
12663{
12664	int retval;
12665	int i;
12666
12667	/*
12668	 * This will send an error to the other controller in the case of a
12669	 * failure.
12670	 */
12671	retval = ctl_datamove_remote_sgl_setup(io);
12672	if (retval != 0)
12673		return;
12674
12675	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12676					  ctl_datamove_remote_read_cb);
12677	if ((retval != 0)
12678	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12679		/*
12680		 * Make sure we free memory if there was an error..  The
12681		 * ctl_datamove_remote_xfer() function will send the
12682		 * datamove done message, or call the callback with an
12683		 * error if there is a problem.
12684		 */
12685		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12686			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12687	}
12688
12689	return;
12690}
12691
12692/*
12693 * Process a datamove request from the other controller.  This is used for
12694 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12695 * first.  Once that is complete, the data gets DMAed into the remote
12696 * controller's memory.  For reads, we DMA from the remote controller's
12697 * memory into our memory first, and then move it out to the FETD.
12698 */
12699static void
12700ctl_datamove_remote(union ctl_io *io)
12701{
12702	struct ctl_softc *softc;
12703
12704	softc = control_softc;
12705
12706	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12707
12708	/*
12709	 * Note that we look for an aborted I/O here, but don't do some of
12710	 * the other checks that ctl_datamove() normally does.  We don't
12711	 * need to run the task queue, because this I/O is on the ISC
12712	 * queue, which is executed by the work thread after the task queue.
12713	 * We don't need to run the datamove delay code, since that should
12714	 * have been done if need be on the other controller.
12715	 */
12716	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12717
12718		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12719		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12720		       io->io_hdr.nexus.targ_port,
12721		       io->io_hdr.nexus.targ_target.id,
12722		       io->io_hdr.nexus.targ_lun);
12723		io->io_hdr.status = CTL_CMD_ABORTED;
12724		io->io_hdr.port_status = 31338;
12725
12726		ctl_send_datamove_done(io, /*have_lock*/ 0);
12727
12728		return;
12729	}
12730
12731	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12732		ctl_datamove_remote_write(io);
12733	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12734		ctl_datamove_remote_read(io);
12735	} else {
12736		union ctl_ha_msg msg;
12737		struct scsi_sense_data *sense;
12738		uint8_t sks[3];
12739		int retry_count;
12740
12741		memset(&msg, 0, sizeof(msg));
12742
12743		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12744		msg.hdr.status = CTL_SCSI_ERROR;
12745		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12746
12747		retry_count = 4243;
12748
12749		sense = &msg.scsi.sense_data;
12750		sks[0] = SSD_SCS_VALID;
12751		sks[1] = (retry_count >> 8) & 0xff;
12752		sks[2] = retry_count & 0xff;
12753
12754		/* "Internal target failure" */
12755		scsi_set_sense_data(sense,
12756				    /*sense_format*/ SSD_TYPE_NONE,
12757				    /*current_error*/ 1,
12758				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12759				    /*asc*/ 0x44,
12760				    /*ascq*/ 0x00,
12761				    /*type*/ SSD_ELEM_SKS,
12762				    /*size*/ sizeof(sks),
12763				    /*data*/ sks,
12764				    SSD_ELEM_NONE);
12765
12766		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12767		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12768			ctl_failover_io(io, /*have_lock*/ 1);
12769			return;
12770		}
12771
12772		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12773		    CTL_HA_STATUS_SUCCESS) {
12774			/* XXX KDM what to do if this fails? */
12775		}
12776		return;
12777	}
12778
12779}
12780
12781static int
12782ctl_process_done(union ctl_io *io)
12783{
12784	struct ctl_lun *lun;
12785	struct ctl_softc *ctl_softc;
12786	void (*fe_done)(union ctl_io *io);
12787	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12788
12789	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12790
12791	fe_done =
12792	    control_softc->ctl_ports[targ_port]->fe_done;
12793
12794#ifdef CTL_TIME_IO
12795	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12796		char str[256];
12797		char path_str[64];
12798		struct sbuf sb;
12799
12800		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12801		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12802
12803		sbuf_cat(&sb, path_str);
12804		switch (io->io_hdr.io_type) {
12805		case CTL_IO_SCSI:
12806			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12807			sbuf_printf(&sb, "\n");
12808			sbuf_cat(&sb, path_str);
12809			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12810				    io->scsiio.tag_num, io->scsiio.tag_type);
12811			break;
12812		case CTL_IO_TASK:
12813			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12814				    "Tag Type: %d\n", io->taskio.task_action,
12815				    io->taskio.tag_num, io->taskio.tag_type);
12816			break;
12817		default:
12818			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12819			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12820			break;
12821		}
12822		sbuf_cat(&sb, path_str);
12823		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12824			    (intmax_t)time_uptime - io->io_hdr.start_time);
12825		sbuf_finish(&sb);
12826		printf("%s", sbuf_data(&sb));
12827	}
12828#endif /* CTL_TIME_IO */
12829
12830	switch (io->io_hdr.io_type) {
12831	case CTL_IO_SCSI:
12832		break;
12833	case CTL_IO_TASK:
12834		if (bootverbose || verbose > 0)
12835			ctl_io_error_print(io, NULL);
12836		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12837			ctl_free_io(io);
12838		else
12839			fe_done(io);
12840		return (CTL_RETVAL_COMPLETE);
12841		break;
12842	default:
12843		printf("ctl_process_done: invalid io type %d\n",
12844		       io->io_hdr.io_type);
12845		panic("ctl_process_done: invalid io type %d\n",
12846		      io->io_hdr.io_type);
12847		break; /* NOTREACHED */
12848	}
12849
12850	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12851	if (lun == NULL) {
12852		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12853				 io->io_hdr.nexus.targ_mapped_lun));
12854		fe_done(io);
12855		goto bailout;
12856	}
12857	ctl_softc = lun->ctl_softc;
12858
12859	mtx_lock(&lun->lun_lock);
12860
12861	/*
12862	 * Check to see if we have any errors to inject here.  We only
12863	 * inject errors for commands that don't already have errors set.
12864	 */
12865	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12866	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12867		ctl_inject_error(lun, io);
12868
12869	/*
12870	 * XXX KDM how do we treat commands that aren't completed
12871	 * successfully?
12872	 *
12873	 * XXX KDM should we also track I/O latency?
12874	 */
12875	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12876	    io->io_hdr.io_type == CTL_IO_SCSI) {
12877#ifdef CTL_TIME_IO
12878		struct bintime cur_bt;
12879#endif
12880		int type;
12881
12882		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12883		    CTL_FLAG_DATA_IN)
12884			type = CTL_STATS_READ;
12885		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12886		    CTL_FLAG_DATA_OUT)
12887			type = CTL_STATS_WRITE;
12888		else
12889			type = CTL_STATS_NO_IO;
12890
12891		lun->stats.ports[targ_port].bytes[type] +=
12892		    io->scsiio.kern_total_len;
12893		lun->stats.ports[targ_port].operations[type]++;
12894#ifdef CTL_TIME_IO
12895		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
12896		   &io->io_hdr.dma_bt);
12897		lun->stats.ports[targ_port].num_dmas[type] +=
12898		    io->io_hdr.num_dmas;
12899		getbintime(&cur_bt);
12900		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12901		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
12902#endif
12903	}
12904
12905	/*
12906	 * Remove this from the OOA queue.
12907	 */
12908	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12909
12910	/*
12911	 * Run through the blocked queue on this LUN and see if anything
12912	 * has become unblocked, now that this transaction is done.
12913	 */
12914	ctl_check_blocked(lun);
12915
12916	/*
12917	 * If the LUN has been invalidated, free it if there is nothing
12918	 * left on its OOA queue.
12919	 */
12920	if ((lun->flags & CTL_LUN_INVALID)
12921	 && TAILQ_EMPTY(&lun->ooa_queue)) {
12922		mtx_unlock(&lun->lun_lock);
12923		mtx_lock(&ctl_softc->ctl_lock);
12924		ctl_free_lun(lun);
12925		mtx_unlock(&ctl_softc->ctl_lock);
12926	} else
12927		mtx_unlock(&lun->lun_lock);
12928
12929	/*
12930	 * If this command has been aborted, make sure we set the status
12931	 * properly.  The FETD is responsible for freeing the I/O and doing
12932	 * whatever it needs to do to clean up its state.
12933	 */
12934	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12935		io->io_hdr.status = CTL_CMD_ABORTED;
12936
12937	/*
12938	 * We print out status for every task management command.  For SCSI
12939	 * commands, we filter out any unit attention errors; they happen
12940	 * on every boot, and would clutter up the log.  Note:  task
12941	 * management commands aren't printed here, they are printed above,
12942	 * since they should never even make it down here.
12943	 */
12944	switch (io->io_hdr.io_type) {
12945	case CTL_IO_SCSI: {
12946		int error_code, sense_key, asc, ascq;
12947
12948		sense_key = 0;
12949
12950		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12951		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12952			/*
12953			 * Since this is just for printing, no need to
12954			 * show errors here.
12955			 */
12956			scsi_extract_sense_len(&io->scsiio.sense_data,
12957					       io->scsiio.sense_len,
12958					       &error_code,
12959					       &sense_key,
12960					       &asc,
12961					       &ascq,
12962					       /*show_errors*/ 0);
12963		}
12964
12965		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12966		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12967		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12968		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12969
12970			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12971				ctl_softc->skipped_prints++;
12972			} else {
12973				uint32_t skipped_prints;
12974
12975				skipped_prints = ctl_softc->skipped_prints;
12976
12977				ctl_softc->skipped_prints = 0;
12978				ctl_softc->last_print_jiffies = time_uptime;
12979
12980				if (skipped_prints > 0) {
12981#ifdef NEEDTOPORT
12982					csevent_log(CSC_CTL | CSC_SHELF_SW |
12983					    CTL_ERROR_REPORT,
12984					    csevent_LogType_Trace,
12985					    csevent_Severity_Information,
12986					    csevent_AlertLevel_Green,
12987					    csevent_FRU_Firmware,
12988					    csevent_FRU_Unknown,
12989					    "High CTL error volume, %d prints "
12990					    "skipped", skipped_prints);
12991#endif
12992				}
12993				if (bootverbose || verbose > 0)
12994					ctl_io_error_print(io, NULL);
12995			}
12996		}
12997		break;
12998	}
12999	case CTL_IO_TASK:
13000		if (bootverbose || verbose > 0)
13001			ctl_io_error_print(io, NULL);
13002		break;
13003	default:
13004		break;
13005	}
13006
13007	/*
13008	 * Tell the FETD or the other shelf controller we're done with this
13009	 * command.  Note that only SCSI commands get to this point.  Task
13010	 * management commands are completed above.
13011	 *
13012	 * We only send status to the other controller if we're in XFER
13013	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13014	 * received the I/O (from CTL's perspective), and so the status is
13015	 * generated there.
13016	 *
13017	 * XXX KDM if we hold the lock here, we could cause a deadlock
13018	 * if the frontend comes back in in this context to queue
13019	 * something.
13020	 */
13021	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13022	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13023		union ctl_ha_msg msg;
13024
13025		memset(&msg, 0, sizeof(msg));
13026		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13027		msg.hdr.original_sc = io->io_hdr.original_sc;
13028		msg.hdr.nexus = io->io_hdr.nexus;
13029		msg.hdr.status = io->io_hdr.status;
13030		msg.scsi.scsi_status = io->scsiio.scsi_status;
13031		msg.scsi.tag_num = io->scsiio.tag_num;
13032		msg.scsi.tag_type = io->scsiio.tag_type;
13033		msg.scsi.sense_len = io->scsiio.sense_len;
13034		msg.scsi.sense_residual = io->scsiio.sense_residual;
13035		msg.scsi.residual = io->scsiio.residual;
13036		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13037		       sizeof(io->scsiio.sense_data));
13038		/*
13039		 * We copy this whether or not this is an I/O-related
13040		 * command.  Otherwise, we'd have to go and check to see
13041		 * whether it's a read/write command, and it really isn't
13042		 * worth it.
13043		 */
13044		memcpy(&msg.scsi.lbalen,
13045		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13046		       sizeof(msg.scsi.lbalen));
13047
13048		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13049				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13050			/* XXX do something here */
13051		}
13052
13053		ctl_free_io(io);
13054	} else
13055		fe_done(io);
13056
13057bailout:
13058
13059	return (CTL_RETVAL_COMPLETE);
13060}
13061
13062/*
13063 * Front end should call this if it doesn't do autosense.  When the request
13064 * sense comes back in from the initiator, we'll dequeue this and send it.
13065 */
13066int
13067ctl_queue_sense(union ctl_io *io)
13068{
13069	struct ctl_lun *lun;
13070	struct ctl_softc *ctl_softc;
13071	uint32_t initidx, targ_lun;
13072
13073	ctl_softc = control_softc;
13074
13075	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13076
13077	/*
13078	 * LUN lookup will likely move to the ctl_work_thread() once we
13079	 * have our new queueing infrastructure (that doesn't put things on
13080	 * a per-LUN queue initially).  That is so that we can handle
13081	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13082	 * can't deal with that right now.
13083	 */
13084	mtx_lock(&ctl_softc->ctl_lock);
13085
13086	/*
13087	 * If we don't have a LUN for this, just toss the sense
13088	 * information.
13089	 */
13090	targ_lun = io->io_hdr.nexus.targ_lun;
13091	if (io->io_hdr.nexus.lun_map_fn != NULL)
13092		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13093	if ((targ_lun < CTL_MAX_LUNS)
13094	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13095		lun = ctl_softc->ctl_luns[targ_lun];
13096	else
13097		goto bailout;
13098
13099	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13100
13101	mtx_lock(&lun->lun_lock);
13102	/*
13103	 * Already have CA set for this LUN...toss the sense information.
13104	 */
13105	if (ctl_is_set(lun->have_ca, initidx)) {
13106		mtx_unlock(&lun->lun_lock);
13107		goto bailout;
13108	}
13109
13110	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13111	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13112	       sizeof(io->scsiio.sense_data)));
13113	ctl_set_mask(lun->have_ca, initidx);
13114	mtx_unlock(&lun->lun_lock);
13115
13116bailout:
13117	mtx_unlock(&ctl_softc->ctl_lock);
13118
13119	ctl_free_io(io);
13120
13121	return (CTL_RETVAL_COMPLETE);
13122}
13123
13124/*
13125 * Primary command inlet from frontend ports.  All SCSI and task I/O
13126 * requests must go through this function.
13127 */
13128int
13129ctl_queue(union ctl_io *io)
13130{
13131	struct ctl_softc *ctl_softc;
13132
13133	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13134
13135	ctl_softc = control_softc;
13136
13137#ifdef CTL_TIME_IO
13138	io->io_hdr.start_time = time_uptime;
13139	getbintime(&io->io_hdr.start_bt);
13140#endif /* CTL_TIME_IO */
13141
13142	/* Map FE-specific LUN ID into global one. */
13143	if (io->io_hdr.nexus.lun_map_fn != NULL)
13144		io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.lun_map_fn(
13145		    io->io_hdr.nexus.lun_map_arg, io->io_hdr.nexus.targ_lun);
13146	else
13147		io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.targ_lun;
13148
13149	switch (io->io_hdr.io_type) {
13150	case CTL_IO_SCSI:
13151		ctl_enqueue_incoming(io);
13152		break;
13153	case CTL_IO_TASK:
13154		ctl_run_task(io);
13155		break;
13156	default:
13157		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13158		return (EINVAL);
13159	}
13160
13161	return (CTL_RETVAL_COMPLETE);
13162}
13163
13164#ifdef CTL_IO_DELAY
13165static void
13166ctl_done_timer_wakeup(void *arg)
13167{
13168	union ctl_io *io;
13169
13170	io = (union ctl_io *)arg;
13171	ctl_done(io);
13172}
13173#endif /* CTL_IO_DELAY */
13174
13175void
13176ctl_done(union ctl_io *io)
13177{
13178	struct ctl_softc *ctl_softc;
13179
13180	ctl_softc = control_softc;
13181
13182	/*
13183	 * Enable this to catch duplicate completion issues.
13184	 */
13185#if 0
13186	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13187		printf("%s: type %d msg %d cdb %x iptl: "
13188		       "%d:%d:%d:%d tag 0x%04x "
13189		       "flag %#x status %x\n",
13190			__func__,
13191			io->io_hdr.io_type,
13192			io->io_hdr.msg_type,
13193			io->scsiio.cdb[0],
13194			io->io_hdr.nexus.initid.id,
13195			io->io_hdr.nexus.targ_port,
13196			io->io_hdr.nexus.targ_target.id,
13197			io->io_hdr.nexus.targ_lun,
13198			(io->io_hdr.io_type ==
13199			CTL_IO_TASK) ?
13200			io->taskio.tag_num :
13201			io->scsiio.tag_num,
13202		        io->io_hdr.flags,
13203			io->io_hdr.status);
13204	} else
13205		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13206#endif
13207
13208	/*
13209	 * This is an internal copy of an I/O, and should not go through
13210	 * the normal done processing logic.
13211	 */
13212	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13213		return;
13214
13215	/*
13216	 * We need to send a msg to the serializing shelf to finish the IO
13217	 * as well.  We don't send a finish message to the other shelf if
13218	 * this is a task management command.  Task management commands
13219	 * aren't serialized in the OOA queue, but rather just executed on
13220	 * both shelf controllers for commands that originated on that
13221	 * controller.
13222	 */
13223	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13224	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13225		union ctl_ha_msg msg_io;
13226
13227		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13228		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13229		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13230		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13231		}
13232		/* continue on to finish IO */
13233	}
13234#ifdef CTL_IO_DELAY
13235	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13236		struct ctl_lun *lun;
13237
13238		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13239
13240		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13241	} else {
13242		struct ctl_lun *lun;
13243
13244		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13245
13246		if ((lun != NULL)
13247		 && (lun->delay_info.done_delay > 0)) {
13248			struct callout *callout;
13249
13250			callout = (struct callout *)&io->io_hdr.timer_bytes;
13251			callout_init(callout, /*mpsafe*/ 1);
13252			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13253			callout_reset(callout,
13254				      lun->delay_info.done_delay * hz,
13255				      ctl_done_timer_wakeup, io);
13256			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13257				lun->delay_info.done_delay = 0;
13258			return;
13259		}
13260	}
13261#endif /* CTL_IO_DELAY */
13262
13263	ctl_enqueue_done(io);
13264}
13265
13266int
13267ctl_isc(struct ctl_scsiio *ctsio)
13268{
13269	struct ctl_lun *lun;
13270	int retval;
13271
13272	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13273
13274	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13275
13276	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13277
13278	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13279
13280	return (retval);
13281}
13282
13283
13284static void
13285ctl_work_thread(void *arg)
13286{
13287	struct ctl_thread *thr = (struct ctl_thread *)arg;
13288	struct ctl_softc *softc = thr->ctl_softc;
13289	union ctl_io *io;
13290	int retval;
13291
13292	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13293
13294	for (;;) {
13295		retval = 0;
13296
13297		/*
13298		 * We handle the queues in this order:
13299		 * - ISC
13300		 * - done queue (to free up resources, unblock other commands)
13301		 * - RtR queue
13302		 * - incoming queue
13303		 *
13304		 * If those queues are empty, we break out of the loop and
13305		 * go to sleep.
13306		 */
13307		mtx_lock(&thr->queue_lock);
13308		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13309		if (io != NULL) {
13310			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13311			mtx_unlock(&thr->queue_lock);
13312			ctl_handle_isc(io);
13313			continue;
13314		}
13315		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13316		if (io != NULL) {
13317			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13318			/* clear any blocked commands, call fe_done */
13319			mtx_unlock(&thr->queue_lock);
13320			retval = ctl_process_done(io);
13321			continue;
13322		}
13323		if (!ctl_pause_rtr) {
13324			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13325			if (io != NULL) {
13326				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13327				mtx_unlock(&thr->queue_lock);
13328				retval = ctl_scsiio(&io->scsiio);
13329				if (retval != CTL_RETVAL_COMPLETE)
13330					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13331				continue;
13332			}
13333		}
13334		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13335		if (io != NULL) {
13336			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13337			mtx_unlock(&thr->queue_lock);
13338			ctl_scsiio_precheck(softc, &io->scsiio);
13339			continue;
13340		}
13341
13342		/* Sleep until we have something to do. */
13343		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13344	}
13345}
13346
13347static void
13348ctl_lun_thread(void *arg)
13349{
13350	struct ctl_softc *softc = (struct ctl_softc *)arg;
13351	struct ctl_be_lun *be_lun;
13352	int retval;
13353
13354	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13355
13356	for (;;) {
13357		retval = 0;
13358		mtx_lock(&softc->ctl_lock);
13359		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13360		if (be_lun != NULL) {
13361			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13362			mtx_unlock(&softc->ctl_lock);
13363			ctl_create_lun(be_lun);
13364			continue;
13365		}
13366
13367		/* Sleep until we have something to do. */
13368		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13369		    PDROP | PRIBIO, "-", 0);
13370	}
13371}
13372
13373static void
13374ctl_enqueue_incoming(union ctl_io *io)
13375{
13376	struct ctl_softc *softc = control_softc;
13377	struct ctl_thread *thr;
13378
13379	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13380	mtx_lock(&thr->queue_lock);
13381	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13382	mtx_unlock(&thr->queue_lock);
13383	wakeup(thr);
13384}
13385
13386static void
13387ctl_enqueue_rtr(union ctl_io *io)
13388{
13389	struct ctl_softc *softc = control_softc;
13390	struct ctl_thread *thr;
13391
13392	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13393	mtx_lock(&thr->queue_lock);
13394	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13395	mtx_unlock(&thr->queue_lock);
13396	wakeup(thr);
13397}
13398
13399static void
13400ctl_enqueue_done(union ctl_io *io)
13401{
13402	struct ctl_softc *softc = control_softc;
13403	struct ctl_thread *thr;
13404
13405	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13406	mtx_lock(&thr->queue_lock);
13407	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13408	mtx_unlock(&thr->queue_lock);
13409	wakeup(thr);
13410}
13411
13412static void
13413ctl_enqueue_isc(union ctl_io *io)
13414{
13415	struct ctl_softc *softc = control_softc;
13416	struct ctl_thread *thr;
13417
13418	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13419	mtx_lock(&thr->queue_lock);
13420	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13421	mtx_unlock(&thr->queue_lock);
13422	wakeup(thr);
13423}
13424
13425/* Initialization and failover */
13426
13427void
13428ctl_init_isc_msg(void)
13429{
13430	printf("CTL: Still calling this thing\n");
13431}
13432
13433/*
13434 * Init component
13435 * 	Initializes component into configuration defined by bootMode
13436 *	(see hasc-sv.c)
13437 *  	returns hasc_Status:
13438 * 		OK
13439 *		ERROR - fatal error
13440 */
13441static ctl_ha_comp_status
13442ctl_isc_init(struct ctl_ha_component *c)
13443{
13444	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13445
13446	c->status = ret;
13447	return ret;
13448}
13449
13450/* Start component
13451 * 	Starts component in state requested. If component starts successfully,
13452 *	it must set its own state to the requestrd state
13453 *	When requested state is HASC_STATE_HA, the component may refine it
13454 * 	by adding _SLAVE or _MASTER flags.
13455 *	Currently allowed state transitions are:
13456 *	UNKNOWN->HA		- initial startup
13457 *	UNKNOWN->SINGLE - initial startup when no parter detected
13458 *	HA->SINGLE		- failover
13459 * returns ctl_ha_comp_status:
13460 * 		OK	- component successfully started in requested state
13461 *		FAILED  - could not start the requested state, failover may
13462 * 			  be possible
13463 *		ERROR	- fatal error detected, no future startup possible
13464 */
13465static ctl_ha_comp_status
13466ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13467{
13468	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13469
13470	printf("%s: go\n", __func__);
13471
13472	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13473	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13474		ctl_is_single = 0;
13475		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13476		    != CTL_HA_STATUS_SUCCESS) {
13477			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13478			ret = CTL_HA_COMP_STATUS_ERROR;
13479		}
13480	} else if (CTL_HA_STATE_IS_HA(c->state)
13481		&& CTL_HA_STATE_IS_SINGLE(state)){
13482		// HA->SINGLE transition
13483	        ctl_failover();
13484		ctl_is_single = 1;
13485	} else {
13486		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13487		       c->state, state);
13488		ret = CTL_HA_COMP_STATUS_ERROR;
13489	}
13490	if (CTL_HA_STATE_IS_SINGLE(state))
13491		ctl_is_single = 1;
13492
13493	c->state = state;
13494	c->status = ret;
13495	return ret;
13496}
13497
13498/*
13499 * Quiesce component
13500 * The component must clear any error conditions (set status to OK) and
13501 * prepare itself to another Start call
13502 * returns ctl_ha_comp_status:
13503 * 	OK
13504 *	ERROR
13505 */
13506static ctl_ha_comp_status
13507ctl_isc_quiesce(struct ctl_ha_component *c)
13508{
13509	int ret = CTL_HA_COMP_STATUS_OK;
13510
13511	ctl_pause_rtr = 1;
13512	c->status = ret;
13513	return ret;
13514}
13515
13516struct ctl_ha_component ctl_ha_component_ctlisc =
13517{
13518	.name = "CTL ISC",
13519	.state = CTL_HA_STATE_UNKNOWN,
13520	.init = ctl_isc_init,
13521	.start = ctl_isc_start,
13522	.quiesce = ctl_isc_quiesce
13523};
13524
13525/*
13526 *  vim: ts=8
13527 */
13528