ctl.c revision 267992
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 267992 2014-06-28 03:56:17Z hselasky $");
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				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);
431
432/*
433 * Load the serialization table.  This isn't very pretty, but is probably
434 * the easiest way to do it.
435 */
436#include "ctl_ser_table.c"
437
438/*
439 * We only need to define open, close and ioctl routines for this driver.
440 */
441static struct cdevsw ctl_cdevsw = {
442	.d_version =	D_VERSION,
443	.d_flags =	0,
444	.d_open =	ctl_open,
445	.d_close =	ctl_close,
446	.d_ioctl =	ctl_ioctl,
447	.d_name =	"ctl",
448};
449
450
451MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
452
453static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
454
455static moduledata_t ctl_moduledata = {
456	"ctl",
457	ctl_module_event_handler,
458	NULL
459};
460
461DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
462MODULE_VERSION(ctl, 1);
463
464static void
465ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
466			    union ctl_ha_msg *msg_info)
467{
468	struct ctl_scsiio *ctsio;
469
470	if (msg_info->hdr.original_sc == NULL) {
471		printf("%s: original_sc == NULL!\n", __func__);
472		/* XXX KDM now what? */
473		return;
474	}
475
476	ctsio = &msg_info->hdr.original_sc->scsiio;
477	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
478	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
479	ctsio->io_hdr.status = msg_info->hdr.status;
480	ctsio->scsi_status = msg_info->scsi.scsi_status;
481	ctsio->sense_len = msg_info->scsi.sense_len;
482	ctsio->sense_residual = msg_info->scsi.sense_residual;
483	ctsio->residual = msg_info->scsi.residual;
484	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
485	       sizeof(ctsio->sense_data));
486	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
487	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
488	ctl_enqueue_isc((union ctl_io *)ctsio);
489}
490
491static void
492ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
493				union ctl_ha_msg *msg_info)
494{
495	struct ctl_scsiio *ctsio;
496
497	if (msg_info->hdr.serializing_sc == NULL) {
498		printf("%s: serializing_sc == NULL!\n", __func__);
499		/* XXX KDM now what? */
500		return;
501	}
502
503	ctsio = &msg_info->hdr.serializing_sc->scsiio;
504#if 0
505	/*
506	 * Attempt to catch the situation where an I/O has
507	 * been freed, and we're using it again.
508	 */
509	if (ctsio->io_hdr.io_type == 0xff) {
510		union ctl_io *tmp_io;
511		tmp_io = (union ctl_io *)ctsio;
512		printf("%s: %p use after free!\n", __func__,
513		       ctsio);
514		printf("%s: type %d msg %d cdb %x iptl: "
515		       "%d:%d:%d:%d tag 0x%04x "
516		       "flag %#x status %x\n",
517			__func__,
518			tmp_io->io_hdr.io_type,
519			tmp_io->io_hdr.msg_type,
520			tmp_io->scsiio.cdb[0],
521			tmp_io->io_hdr.nexus.initid.id,
522			tmp_io->io_hdr.nexus.targ_port,
523			tmp_io->io_hdr.nexus.targ_target.id,
524			tmp_io->io_hdr.nexus.targ_lun,
525			(tmp_io->io_hdr.io_type ==
526			CTL_IO_TASK) ?
527			tmp_io->taskio.tag_num :
528			tmp_io->scsiio.tag_num,
529		        tmp_io->io_hdr.flags,
530			tmp_io->io_hdr.status);
531	}
532#endif
533	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
534	ctl_enqueue_isc((union ctl_io *)ctsio);
535}
536
537/*
538 * ISC (Inter Shelf Communication) event handler.  Events from the HA
539 * subsystem come in here.
540 */
541static void
542ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
543{
544	struct ctl_softc *ctl_softc;
545	union ctl_io *io;
546	struct ctl_prio *presio;
547	ctl_ha_status isc_status;
548
549	ctl_softc = control_softc;
550	io = NULL;
551
552
553#if 0
554	printf("CTL: Isc Msg event %d\n", event);
555#endif
556	if (event == CTL_HA_EVT_MSG_RECV) {
557		union ctl_ha_msg msg_info;
558
559		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
560					     sizeof(msg_info), /*wait*/ 0);
561#if 0
562		printf("CTL: msg_type %d\n", msg_info.msg_type);
563#endif
564		if (isc_status != 0) {
565			printf("Error receiving message, status = %d\n",
566			       isc_status);
567			return;
568		}
569
570		switch (msg_info.hdr.msg_type) {
571		case CTL_MSG_SERIALIZE:
572#if 0
573			printf("Serialize\n");
574#endif
575			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
576			if (io == NULL) {
577				printf("ctl_isc_event_handler: can't allocate "
578				       "ctl_io!\n");
579				/* Bad Juju */
580				/* Need to set busy and send msg back */
581				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
582				msg_info.hdr.status = CTL_SCSI_ERROR;
583				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
584				msg_info.scsi.sense_len = 0;
585			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
586				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
587				}
588				goto bailout;
589			}
590			ctl_zero_io(io);
591			// populate ctsio from msg_info
592			io->io_hdr.io_type = CTL_IO_SCSI;
593			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
594			io->io_hdr.original_sc = msg_info.hdr.original_sc;
595#if 0
596			printf("pOrig %x\n", (int)msg_info.original_sc);
597#endif
598			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
599					    CTL_FLAG_IO_ACTIVE;
600			/*
601			 * If we're in serialization-only mode, we don't
602			 * want to go through full done processing.  Thus
603			 * the COPY flag.
604			 *
605			 * XXX KDM add another flag that is more specific.
606			 */
607			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
608				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
609			io->io_hdr.nexus = msg_info.hdr.nexus;
610#if 0
611			printf("targ %d, port %d, iid %d, lun %d\n",
612			       io->io_hdr.nexus.targ_target.id,
613			       io->io_hdr.nexus.targ_port,
614			       io->io_hdr.nexus.initid.id,
615			       io->io_hdr.nexus.targ_lun);
616#endif
617			io->scsiio.tag_num = msg_info.scsi.tag_num;
618			io->scsiio.tag_type = msg_info.scsi.tag_type;
619			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
620			       CTL_MAX_CDBLEN);
621			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
622				struct ctl_cmd_entry *entry;
623				uint8_t opcode;
624
625				opcode = io->scsiio.cdb[0];
626				entry = &ctl_cmd_table[opcode];
627				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
628				io->io_hdr.flags |=
629					entry->flags & CTL_FLAG_DATA_MASK;
630			}
631			ctl_enqueue_isc(io);
632			break;
633
634		/* Performed on the Originating SC, XFER mode only */
635		case CTL_MSG_DATAMOVE: {
636			struct ctl_sg_entry *sgl;
637			int i, j;
638
639			io = msg_info.hdr.original_sc;
640			if (io == NULL) {
641				printf("%s: original_sc == NULL!\n", __func__);
642				/* XXX KDM do something here */
643				break;
644			}
645			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
646			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
647			/*
648			 * Keep track of this, we need to send it back over
649			 * when the datamove is complete.
650			 */
651			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
652
653			if (msg_info.dt.sg_sequence == 0) {
654				/*
655				 * XXX KDM we use the preallocated S/G list
656				 * here, but we'll need to change this to
657				 * dynamic allocation if we need larger S/G
658				 * lists.
659				 */
660				if (msg_info.dt.kern_sg_entries >
661				    sizeof(io->io_hdr.remote_sglist) /
662				    sizeof(io->io_hdr.remote_sglist[0])) {
663					printf("%s: number of S/G entries "
664					    "needed %u > allocated num %zd\n",
665					    __func__,
666					    msg_info.dt.kern_sg_entries,
667					    sizeof(io->io_hdr.remote_sglist)/
668					    sizeof(io->io_hdr.remote_sglist[0]));
669
670					/*
671					 * XXX KDM send a message back to
672					 * the other side to shut down the
673					 * DMA.  The error will come back
674					 * through via the normal channel.
675					 */
676					break;
677				}
678				sgl = io->io_hdr.remote_sglist;
679				memset(sgl, 0,
680				       sizeof(io->io_hdr.remote_sglist));
681
682				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
683
684				io->scsiio.kern_sg_entries =
685					msg_info.dt.kern_sg_entries;
686				io->scsiio.rem_sg_entries =
687					msg_info.dt.kern_sg_entries;
688				io->scsiio.kern_data_len =
689					msg_info.dt.kern_data_len;
690				io->scsiio.kern_total_len =
691					msg_info.dt.kern_total_len;
692				io->scsiio.kern_data_resid =
693					msg_info.dt.kern_data_resid;
694				io->scsiio.kern_rel_offset =
695					msg_info.dt.kern_rel_offset;
696				/*
697				 * Clear out per-DMA flags.
698				 */
699				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
700				/*
701				 * Add per-DMA flags that are set for this
702				 * particular DMA request.
703				 */
704				io->io_hdr.flags |= msg_info.dt.flags &
705						    CTL_FLAG_RDMA_MASK;
706			} else
707				sgl = (struct ctl_sg_entry *)
708					io->scsiio.kern_data_ptr;
709
710			for (i = msg_info.dt.sent_sg_entries, j = 0;
711			     i < (msg_info.dt.sent_sg_entries +
712			     msg_info.dt.cur_sg_entries); i++, j++) {
713				sgl[i].addr = msg_info.dt.sg_list[j].addr;
714				sgl[i].len = msg_info.dt.sg_list[j].len;
715
716#if 0
717				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
718				       __func__,
719				       msg_info.dt.sg_list[j].addr,
720				       msg_info.dt.sg_list[j].len,
721				       sgl[i].addr, sgl[i].len, j, i);
722#endif
723			}
724#if 0
725			memcpy(&sgl[msg_info.dt.sent_sg_entries],
726			       msg_info.dt.sg_list,
727			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
728#endif
729
730			/*
731			 * If this is the last piece of the I/O, we've got
732			 * the full S/G list.  Queue processing in the thread.
733			 * Otherwise wait for the next piece.
734			 */
735			if (msg_info.dt.sg_last != 0)
736				ctl_enqueue_isc(io);
737			break;
738		}
739		/* Performed on the Serializing (primary) SC, XFER mode only */
740		case CTL_MSG_DATAMOVE_DONE: {
741			if (msg_info.hdr.serializing_sc == NULL) {
742				printf("%s: serializing_sc == NULL!\n",
743				       __func__);
744				/* XXX KDM now what? */
745				break;
746			}
747			/*
748			 * We grab the sense information here in case
749			 * there was a failure, so we can return status
750			 * back to the initiator.
751			 */
752			io = msg_info.hdr.serializing_sc;
753			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
754			io->io_hdr.status = msg_info.hdr.status;
755			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
756			io->scsiio.sense_len = msg_info.scsi.sense_len;
757			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
758			io->io_hdr.port_status = msg_info.scsi.fetd_status;
759			io->scsiio.residual = msg_info.scsi.residual;
760			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
761			       sizeof(io->scsiio.sense_data));
762			ctl_enqueue_isc(io);
763			break;
764		}
765
766		/* Preformed on Originating SC, SER_ONLY mode */
767		case CTL_MSG_R2R:
768			io = msg_info.hdr.original_sc;
769			if (io == NULL) {
770				printf("%s: Major Bummer\n", __func__);
771				return;
772			} else {
773#if 0
774				printf("pOrig %x\n",(int) ctsio);
775#endif
776			}
777			io->io_hdr.msg_type = CTL_MSG_R2R;
778			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
779			ctl_enqueue_isc(io);
780			break;
781
782		/*
783		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
784		 * mode.
785		 * Performed on the Originating (i.e. secondary) SC in XFER
786		 * mode
787		 */
788		case CTL_MSG_FINISH_IO:
789			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
790				ctl_isc_handler_finish_xfer(ctl_softc,
791							    &msg_info);
792			else
793				ctl_isc_handler_finish_ser_only(ctl_softc,
794								&msg_info);
795			break;
796
797		/* Preformed on Originating SC */
798		case CTL_MSG_BAD_JUJU:
799			io = msg_info.hdr.original_sc;
800			if (io == NULL) {
801				printf("%s: Bad JUJU!, original_sc is NULL!\n",
802				       __func__);
803				break;
804			}
805			ctl_copy_sense_data(&msg_info, io);
806			/*
807			 * IO should have already been cleaned up on other
808			 * SC so clear this flag so we won't send a message
809			 * back to finish the IO there.
810			 */
811			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
812			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
813
814			/* io = msg_info.hdr.serializing_sc; */
815			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
816			ctl_enqueue_isc(io);
817			break;
818
819		/* Handle resets sent from the other side */
820		case CTL_MSG_MANAGE_TASKS: {
821			struct ctl_taskio *taskio;
822			taskio = (struct ctl_taskio *)ctl_alloc_io(
823				(void *)ctl_softc->othersc_pool);
824			if (taskio == NULL) {
825				printf("ctl_isc_event_handler: can't allocate "
826				       "ctl_io!\n");
827				/* Bad Juju */
828				/* should I just call the proper reset func
829				   here??? */
830				goto bailout;
831			}
832			ctl_zero_io((union ctl_io *)taskio);
833			taskio->io_hdr.io_type = CTL_IO_TASK;
834			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
835			taskio->io_hdr.nexus = msg_info.hdr.nexus;
836			taskio->task_action = msg_info.task.task_action;
837			taskio->tag_num = msg_info.task.tag_num;
838			taskio->tag_type = msg_info.task.tag_type;
839#ifdef CTL_TIME_IO
840			taskio->io_hdr.start_time = time_uptime;
841			getbintime(&taskio->io_hdr.start_bt);
842#if 0
843			cs_prof_gettime(&taskio->io_hdr.start_ticks);
844#endif
845#endif /* CTL_TIME_IO */
846			ctl_run_task((union ctl_io *)taskio);
847			break;
848		}
849		/* Persistent Reserve action which needs attention */
850		case CTL_MSG_PERS_ACTION:
851			presio = (struct ctl_prio *)ctl_alloc_io(
852				(void *)ctl_softc->othersc_pool);
853			if (presio == NULL) {
854				printf("ctl_isc_event_handler: can't allocate "
855				       "ctl_io!\n");
856				/* Bad Juju */
857				/* Need to set busy and send msg back */
858				goto bailout;
859			}
860			ctl_zero_io((union ctl_io *)presio);
861			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
862			presio->pr_msg = msg_info.pr;
863			ctl_enqueue_isc((union ctl_io *)presio);
864			break;
865		case CTL_MSG_SYNC_FE:
866			rcv_sync_msg = 1;
867			break;
868		case CTL_MSG_APS_LOCK: {
869			// It's quicker to execute this then to
870			// queue it.
871			struct ctl_lun *lun;
872			struct ctl_page_index *page_index;
873			struct copan_aps_subpage *current_sp;
874			uint32_t targ_lun;
875
876			targ_lun = msg_info.hdr.nexus.targ_mapped_lun;
877			lun = ctl_softc->ctl_luns[targ_lun];
878			mtx_lock(&lun->lun_lock);
879			page_index = &lun->mode_pages.index[index_to_aps_page];
880			current_sp = (struct copan_aps_subpage *)
881				     (page_index->page_data +
882				     (page_index->page_len * CTL_PAGE_CURRENT));
883
884			current_sp->lock_active = msg_info.aps.lock_flag;
885			mtx_unlock(&lun->lun_lock);
886		        break;
887		}
888		default:
889		        printf("How did I get here?\n");
890		}
891	} else if (event == CTL_HA_EVT_MSG_SENT) {
892		if (param != CTL_HA_STATUS_SUCCESS) {
893			printf("Bad status from ctl_ha_msg_send status %d\n",
894			       param);
895		}
896		return;
897	} else if (event == CTL_HA_EVT_DISCONNECT) {
898		printf("CTL: Got a disconnect from Isc\n");
899		return;
900	} else {
901		printf("ctl_isc_event_handler: Unknown event %d\n", event);
902		return;
903	}
904
905bailout:
906	return;
907}
908
909static void
910ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
911{
912	struct scsi_sense_data *sense;
913
914	sense = &dest->scsiio.sense_data;
915	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
916	dest->scsiio.scsi_status = src->scsi.scsi_status;
917	dest->scsiio.sense_len = src->scsi.sense_len;
918	dest->io_hdr.status = src->hdr.status;
919}
920
921static int
922ctl_init(void)
923{
924	struct ctl_softc *softc;
925	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
926	struct ctl_frontend *fe;
927        uint8_t sc_id =0;
928	int i, error, retval;
929	//int isc_retval;
930
931	retval = 0;
932	ctl_pause_rtr = 0;
933        rcv_sync_msg = 0;
934
935	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
936			       M_WAITOK | M_ZERO);
937	softc = control_softc;
938
939	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
940			      "cam/ctl");
941
942	softc->dev->si_drv1 = softc;
943
944	/*
945	 * By default, return a "bad LUN" peripheral qualifier for unknown
946	 * LUNs.  The user can override this default using the tunable or
947	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
948	 */
949	softc->inquiry_pq_no_lun = 1;
950	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
951			  &softc->inquiry_pq_no_lun);
952	sysctl_ctx_init(&softc->sysctl_ctx);
953	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
954		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
955		CTLFLAG_RD, 0, "CAM Target Layer");
956
957	if (softc->sysctl_tree == NULL) {
958		printf("%s: unable to allocate sysctl tree\n", __func__);
959		destroy_dev(softc->dev);
960		free(control_softc, M_DEVBUF);
961		control_softc = NULL;
962		return (ENOMEM);
963	}
964
965	SYSCTL_ADD_INT(&softc->sysctl_ctx,
966		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
967		       "inquiry_pq_no_lun", CTLFLAG_RW,
968		       &softc->inquiry_pq_no_lun, 0,
969		       "Report no lun possible for invalid LUNs");
970
971	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
972	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
973	softc->open_count = 0;
974
975	/*
976	 * Default to actually sending a SYNCHRONIZE CACHE command down to
977	 * the drive.
978	 */
979	softc->flags = CTL_FLAG_REAL_SYNC;
980
981	/*
982	 * In Copan's HA scheme, the "master" and "slave" roles are
983	 * figured out through the slot the controller is in.  Although it
984	 * is an active/active system, someone has to be in charge.
985 	 */
986#ifdef NEEDTOPORT
987        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
988#endif
989
990        if (sc_id == 0) {
991		softc->flags |= CTL_FLAG_MASTER_SHELF;
992		persis_offset = 0;
993	} else
994		persis_offset = CTL_MAX_INITIATORS;
995
996	/*
997	 * XXX KDM need to figure out where we want to get our target ID
998	 * and WWID.  Is it different on each port?
999	 */
1000	softc->target.id = 0;
1001	softc->target.wwid[0] = 0x12345678;
1002	softc->target.wwid[1] = 0x87654321;
1003	STAILQ_INIT(&softc->lun_list);
1004	STAILQ_INIT(&softc->pending_lun_queue);
1005	STAILQ_INIT(&softc->fe_list);
1006	STAILQ_INIT(&softc->be_list);
1007	STAILQ_INIT(&softc->io_pools);
1008
1009	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1010			    &internal_pool)!= 0){
1011		printf("ctl: can't allocate %d entry internal pool, "
1012		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1013		return (ENOMEM);
1014	}
1015
1016	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1017			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1018		printf("ctl: can't allocate %d entry emergency pool, "
1019		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1020		ctl_pool_free(internal_pool);
1021		return (ENOMEM);
1022	}
1023
1024	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1025	                    &other_pool) != 0)
1026	{
1027		printf("ctl: can't allocate %d entry other SC pool, "
1028		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1029		ctl_pool_free(internal_pool);
1030		ctl_pool_free(emergency_pool);
1031		return (ENOMEM);
1032	}
1033
1034	softc->internal_pool = internal_pool;
1035	softc->emergency_pool = emergency_pool;
1036	softc->othersc_pool = other_pool;
1037
1038	if (worker_threads <= 0)
1039		worker_threads = max(1, mp_ncpus / 4);
1040	if (worker_threads > CTL_MAX_THREADS)
1041		worker_threads = CTL_MAX_THREADS;
1042
1043	for (i = 0; i < worker_threads; i++) {
1044		struct ctl_thread *thr = &softc->threads[i];
1045
1046		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1047		thr->ctl_softc = softc;
1048		STAILQ_INIT(&thr->incoming_queue);
1049		STAILQ_INIT(&thr->rtr_queue);
1050		STAILQ_INIT(&thr->done_queue);
1051		STAILQ_INIT(&thr->isc_queue);
1052
1053		error = kproc_kthread_add(ctl_work_thread, thr,
1054		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1055		if (error != 0) {
1056			printf("error creating CTL work thread!\n");
1057			ctl_pool_free(internal_pool);
1058			ctl_pool_free(emergency_pool);
1059			ctl_pool_free(other_pool);
1060			return (error);
1061		}
1062	}
1063	error = kproc_kthread_add(ctl_lun_thread, softc,
1064	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1065	if (error != 0) {
1066		printf("error creating CTL lun thread!\n");
1067		ctl_pool_free(internal_pool);
1068		ctl_pool_free(emergency_pool);
1069		ctl_pool_free(other_pool);
1070		return (error);
1071	}
1072	if (bootverbose)
1073		printf("ctl: CAM Target Layer loaded\n");
1074
1075	/*
1076	 * Initialize the initiator and portname mappings
1077	 */
1078	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1079
1080	/*
1081	 * Initialize the ioctl front end.
1082	 */
1083	fe = &softc->ioctl_info.fe;
1084	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1085	fe->port_type = CTL_PORT_IOCTL;
1086	fe->num_requested_ctl_io = 100;
1087	fe->port_name = softc->ioctl_info.port_name;
1088	fe->port_online = ctl_ioctl_online;
1089	fe->port_offline = ctl_ioctl_offline;
1090	fe->onoff_arg = &softc->ioctl_info;
1091	fe->targ_enable = ctl_ioctl_targ_enable;
1092	fe->targ_disable = ctl_ioctl_targ_disable;
1093	fe->lun_enable = ctl_ioctl_lun_enable;
1094	fe->lun_disable = ctl_ioctl_lun_disable;
1095	fe->targ_lun_arg = &softc->ioctl_info;
1096	fe->fe_datamove = ctl_ioctl_datamove;
1097	fe->fe_done = ctl_ioctl_done;
1098	fe->max_targets = 15;
1099	fe->max_target_id = 15;
1100
1101	if (ctl_frontend_register(&softc->ioctl_info.fe,
1102	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1103		printf("ctl: ioctl front end registration failed, will "
1104		       "continue anyway\n");
1105	}
1106
1107#ifdef CTL_IO_DELAY
1108	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1109		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1110		       sizeof(struct callout), CTL_TIMER_BYTES);
1111		return (EINVAL);
1112	}
1113#endif /* CTL_IO_DELAY */
1114
1115	return (0);
1116}
1117
1118void
1119ctl_shutdown(void)
1120{
1121	struct ctl_softc *softc;
1122	struct ctl_lun *lun, *next_lun;
1123	struct ctl_io_pool *pool;
1124
1125	softc = (struct ctl_softc *)control_softc;
1126
1127	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1128		printf("ctl: ioctl front end deregistration failed\n");
1129
1130	mtx_lock(&softc->ctl_lock);
1131
1132	/*
1133	 * Free up each LUN.
1134	 */
1135	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1136		next_lun = STAILQ_NEXT(lun, links);
1137		ctl_free_lun(lun);
1138	}
1139
1140	mtx_unlock(&softc->ctl_lock);
1141
1142	/*
1143	 * This will rip the rug out from under any FETDs or anyone else
1144	 * that has a pool allocated.  Since we increment our module
1145	 * refcount any time someone outside the main CTL module allocates
1146	 * a pool, we shouldn't have any problems here.  The user won't be
1147	 * able to unload the CTL module until client modules have
1148	 * successfully unloaded.
1149	 */
1150	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1151		ctl_pool_free(pool);
1152
1153#if 0
1154	ctl_shutdown_thread(softc->work_thread);
1155	mtx_destroy(&softc->queue_lock);
1156#endif
1157
1158	mtx_destroy(&softc->pool_lock);
1159	mtx_destroy(&softc->ctl_lock);
1160
1161	destroy_dev(softc->dev);
1162
1163	sysctl_ctx_free(&softc->sysctl_ctx);
1164
1165	free(control_softc, M_DEVBUF);
1166	control_softc = NULL;
1167
1168	if (bootverbose)
1169		printf("ctl: CAM Target Layer unloaded\n");
1170}
1171
1172static int
1173ctl_module_event_handler(module_t mod, int what, void *arg)
1174{
1175
1176	switch (what) {
1177	case MOD_LOAD:
1178		return (ctl_init());
1179	case MOD_UNLOAD:
1180		return (EBUSY);
1181	default:
1182		return (EOPNOTSUPP);
1183	}
1184}
1185
1186/*
1187 * XXX KDM should we do some access checks here?  Bump a reference count to
1188 * prevent a CTL module from being unloaded while someone has it open?
1189 */
1190static int
1191ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1192{
1193	return (0);
1194}
1195
1196static int
1197ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1198{
1199	return (0);
1200}
1201
1202int
1203ctl_port_enable(ctl_port_type port_type)
1204{
1205	struct ctl_softc *softc;
1206	struct ctl_frontend *fe;
1207
1208	if (ctl_is_single == 0) {
1209		union ctl_ha_msg msg_info;
1210		int isc_retval;
1211
1212#if 0
1213		printf("%s: HA mode, synchronizing frontend enable\n",
1214		        __func__);
1215#endif
1216		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1217	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1218		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1219			printf("Sync msg send error retval %d\n", isc_retval);
1220		}
1221		if (!rcv_sync_msg) {
1222			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1223			        sizeof(msg_info), 1);
1224		}
1225#if 0
1226        	printf("CTL:Frontend Enable\n");
1227	} else {
1228		printf("%s: single mode, skipping frontend synchronization\n",
1229		        __func__);
1230#endif
1231	}
1232
1233	softc = control_softc;
1234
1235	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1236		if (port_type & fe->port_type)
1237		{
1238#if 0
1239			printf("port %d\n", fe->targ_port);
1240#endif
1241			ctl_frontend_online(fe);
1242		}
1243	}
1244
1245	return (0);
1246}
1247
1248int
1249ctl_port_disable(ctl_port_type port_type)
1250{
1251	struct ctl_softc *softc;
1252	struct ctl_frontend *fe;
1253
1254	softc = control_softc;
1255
1256	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1257		if (port_type & fe->port_type)
1258			ctl_frontend_offline(fe);
1259	}
1260
1261	return (0);
1262}
1263
1264/*
1265 * Returns 0 for success, 1 for failure.
1266 * Currently the only failure mode is if there aren't enough entries
1267 * allocated.  So, in case of a failure, look at num_entries_dropped,
1268 * reallocate and try again.
1269 */
1270int
1271ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1272	      int *num_entries_filled, int *num_entries_dropped,
1273	      ctl_port_type port_type, int no_virtual)
1274{
1275	struct ctl_softc *softc;
1276	struct ctl_frontend *fe;
1277	int entries_dropped, entries_filled;
1278	int retval;
1279	int i;
1280
1281	softc = control_softc;
1282
1283	retval = 0;
1284	entries_filled = 0;
1285	entries_dropped = 0;
1286
1287	i = 0;
1288	mtx_lock(&softc->ctl_lock);
1289	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1290		struct ctl_port_entry *entry;
1291
1292		if ((fe->port_type & port_type) == 0)
1293			continue;
1294
1295		if ((no_virtual != 0)
1296		 && (fe->virtual_port != 0))
1297			continue;
1298
1299		if (entries_filled >= num_entries_alloced) {
1300			entries_dropped++;
1301			continue;
1302		}
1303		entry = &entries[i];
1304
1305		entry->port_type = fe->port_type;
1306		strlcpy(entry->port_name, fe->port_name,
1307			sizeof(entry->port_name));
1308		entry->physical_port = fe->physical_port;
1309		entry->virtual_port = fe->virtual_port;
1310		entry->wwnn = fe->wwnn;
1311		entry->wwpn = fe->wwpn;
1312
1313		i++;
1314		entries_filled++;
1315	}
1316
1317	mtx_unlock(&softc->ctl_lock);
1318
1319	if (entries_dropped > 0)
1320		retval = 1;
1321
1322	*num_entries_dropped = entries_dropped;
1323	*num_entries_filled = entries_filled;
1324
1325	return (retval);
1326}
1327
1328static void
1329ctl_ioctl_online(void *arg)
1330{
1331	struct ctl_ioctl_info *ioctl_info;
1332
1333	ioctl_info = (struct ctl_ioctl_info *)arg;
1334
1335	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1336}
1337
1338static void
1339ctl_ioctl_offline(void *arg)
1340{
1341	struct ctl_ioctl_info *ioctl_info;
1342
1343	ioctl_info = (struct ctl_ioctl_info *)arg;
1344
1345	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1346}
1347
1348/*
1349 * Remove an initiator by port number and initiator ID.
1350 * Returns 0 for success, 1 for failure.
1351 */
1352int
1353ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1354{
1355	struct ctl_softc *softc;
1356
1357	softc = control_softc;
1358
1359	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1360
1361	if ((targ_port < 0)
1362	 || (targ_port > CTL_MAX_PORTS)) {
1363		printf("%s: invalid port number %d\n", __func__, targ_port);
1364		return (1);
1365	}
1366	if (iid > CTL_MAX_INIT_PER_PORT) {
1367		printf("%s: initiator ID %u > maximun %u!\n",
1368		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1369		return (1);
1370	}
1371
1372	mtx_lock(&softc->ctl_lock);
1373
1374	softc->wwpn_iid[targ_port][iid].in_use = 0;
1375
1376	mtx_unlock(&softc->ctl_lock);
1377
1378	return (0);
1379}
1380
1381/*
1382 * Add an initiator to the initiator map.
1383 * Returns 0 for success, 1 for failure.
1384 */
1385int
1386ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1387{
1388	struct ctl_softc *softc;
1389	int retval;
1390
1391	softc = control_softc;
1392
1393	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1394
1395	retval = 0;
1396
1397	if ((targ_port < 0)
1398	 || (targ_port > CTL_MAX_PORTS)) {
1399		printf("%s: invalid port number %d\n", __func__, targ_port);
1400		return (1);
1401	}
1402	if (iid > CTL_MAX_INIT_PER_PORT) {
1403		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1404		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1405		return (1);
1406	}
1407
1408	mtx_lock(&softc->ctl_lock);
1409
1410	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1411		/*
1412		 * We don't treat this as an error.
1413		 */
1414		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1415			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1416			       __func__, targ_port, iid, (uintmax_t)wwpn);
1417			goto bailout;
1418		}
1419
1420		/*
1421		 * This is an error, but what do we do about it?  The
1422		 * driver is telling us we have a new WWPN for this
1423		 * initiator ID, so we pretty much need to use it.
1424		 */
1425		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1426		       "still at that address\n", __func__, targ_port, iid,
1427		       (uintmax_t)wwpn,
1428		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1429
1430		/*
1431		 * XXX KDM clear have_ca and ua_pending on each LUN for
1432		 * this initiator.
1433		 */
1434	}
1435	softc->wwpn_iid[targ_port][iid].in_use = 1;
1436	softc->wwpn_iid[targ_port][iid].iid = iid;
1437	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1438	softc->wwpn_iid[targ_port][iid].port = targ_port;
1439
1440bailout:
1441
1442	mtx_unlock(&softc->ctl_lock);
1443
1444	return (retval);
1445}
1446
1447/*
1448 * XXX KDM should we pretend to do something in the target/lun
1449 * enable/disable functions?
1450 */
1451static int
1452ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1453{
1454	return (0);
1455}
1456
1457static int
1458ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1459{
1460	return (0);
1461}
1462
1463static int
1464ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1465{
1466	return (0);
1467}
1468
1469static int
1470ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1471{
1472	return (0);
1473}
1474
1475/*
1476 * Data movement routine for the CTL ioctl frontend port.
1477 */
1478static int
1479ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1480{
1481	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1482	struct ctl_sg_entry ext_entry, kern_entry;
1483	int ext_sglen, ext_sg_entries, kern_sg_entries;
1484	int ext_sg_start, ext_offset;
1485	int len_to_copy, len_copied;
1486	int kern_watermark, ext_watermark;
1487	int ext_sglist_malloced;
1488	int i, j;
1489
1490	ext_sglist_malloced = 0;
1491	ext_sg_start = 0;
1492	ext_offset = 0;
1493
1494	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1495
1496	/*
1497	 * If this flag is set, fake the data transfer.
1498	 */
1499	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1500		ctsio->ext_data_filled = ctsio->ext_data_len;
1501		goto bailout;
1502	}
1503
1504	/*
1505	 * To simplify things here, if we have a single buffer, stick it in
1506	 * a S/G entry and just make it a single entry S/G list.
1507	 */
1508	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1509		int len_seen;
1510
1511		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1512
1513		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1514							   M_WAITOK);
1515		ext_sglist_malloced = 1;
1516		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1517				   ext_sglen) != 0) {
1518			ctl_set_internal_failure(ctsio,
1519						 /*sks_valid*/ 0,
1520						 /*retry_count*/ 0);
1521			goto bailout;
1522		}
1523		ext_sg_entries = ctsio->ext_sg_entries;
1524		len_seen = 0;
1525		for (i = 0; i < ext_sg_entries; i++) {
1526			if ((len_seen + ext_sglist[i].len) >=
1527			     ctsio->ext_data_filled) {
1528				ext_sg_start = i;
1529				ext_offset = ctsio->ext_data_filled - len_seen;
1530				break;
1531			}
1532			len_seen += ext_sglist[i].len;
1533		}
1534	} else {
1535		ext_sglist = &ext_entry;
1536		ext_sglist->addr = ctsio->ext_data_ptr;
1537		ext_sglist->len = ctsio->ext_data_len;
1538		ext_sg_entries = 1;
1539		ext_sg_start = 0;
1540		ext_offset = ctsio->ext_data_filled;
1541	}
1542
1543	if (ctsio->kern_sg_entries > 0) {
1544		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1545		kern_sg_entries = ctsio->kern_sg_entries;
1546	} else {
1547		kern_sglist = &kern_entry;
1548		kern_sglist->addr = ctsio->kern_data_ptr;
1549		kern_sglist->len = ctsio->kern_data_len;
1550		kern_sg_entries = 1;
1551	}
1552
1553
1554	kern_watermark = 0;
1555	ext_watermark = ext_offset;
1556	len_copied = 0;
1557	for (i = ext_sg_start, j = 0;
1558	     i < ext_sg_entries && j < kern_sg_entries;) {
1559		uint8_t *ext_ptr, *kern_ptr;
1560
1561		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1562				      kern_sglist[j].len - kern_watermark);
1563
1564		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1565		ext_ptr = ext_ptr + ext_watermark;
1566		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1567			/*
1568			 * XXX KDM fix this!
1569			 */
1570			panic("need to implement bus address support");
1571#if 0
1572			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1573#endif
1574		} else
1575			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1576		kern_ptr = kern_ptr + kern_watermark;
1577
1578		kern_watermark += len_to_copy;
1579		ext_watermark += len_to_copy;
1580
1581		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1582		     CTL_FLAG_DATA_IN) {
1583			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1584					 "bytes to user\n", len_to_copy));
1585			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1586					 "to %p\n", kern_ptr, ext_ptr));
1587			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1588				ctl_set_internal_failure(ctsio,
1589							 /*sks_valid*/ 0,
1590							 /*retry_count*/ 0);
1591				goto bailout;
1592			}
1593		} else {
1594			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1595					 "bytes from user\n", len_to_copy));
1596			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1597					 "to %p\n", ext_ptr, kern_ptr));
1598			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1599				ctl_set_internal_failure(ctsio,
1600							 /*sks_valid*/ 0,
1601							 /*retry_count*/0);
1602				goto bailout;
1603			}
1604		}
1605
1606		len_copied += len_to_copy;
1607
1608		if (ext_sglist[i].len == ext_watermark) {
1609			i++;
1610			ext_watermark = 0;
1611		}
1612
1613		if (kern_sglist[j].len == kern_watermark) {
1614			j++;
1615			kern_watermark = 0;
1616		}
1617	}
1618
1619	ctsio->ext_data_filled += len_copied;
1620
1621	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1622			 "kern_sg_entries: %d\n", ext_sg_entries,
1623			 kern_sg_entries));
1624	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1625			 "kern_data_len = %d\n", ctsio->ext_data_len,
1626			 ctsio->kern_data_len));
1627
1628
1629	/* XXX KDM set residual?? */
1630bailout:
1631
1632	if (ext_sglist_malloced != 0)
1633		free(ext_sglist, M_CTL);
1634
1635	return (CTL_RETVAL_COMPLETE);
1636}
1637
1638/*
1639 * Serialize a command that went down the "wrong" side, and so was sent to
1640 * this controller for execution.  The logic is a little different than the
1641 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1642 * sent back to the other side, but in the success case, we execute the
1643 * command on this side (XFER mode) or tell the other side to execute it
1644 * (SER_ONLY mode).
1645 */
1646static int
1647ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1648{
1649	struct ctl_softc *ctl_softc;
1650	union ctl_ha_msg msg_info;
1651	struct ctl_lun *lun;
1652	int retval = 0;
1653	uint32_t targ_lun;
1654
1655	ctl_softc = control_softc;
1656
1657	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1658	lun = ctl_softc->ctl_luns[targ_lun];
1659	if (lun==NULL)
1660	{
1661		/*
1662		 * Why isn't LUN defined? The other side wouldn't
1663		 * send a cmd if the LUN is undefined.
1664		 */
1665		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1666
1667		/* "Logical unit not supported" */
1668		ctl_set_sense_data(&msg_info.scsi.sense_data,
1669				   lun,
1670				   /*sense_format*/SSD_TYPE_NONE,
1671				   /*current_error*/ 1,
1672				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1673				   /*asc*/ 0x25,
1674				   /*ascq*/ 0x00,
1675				   SSD_ELEM_NONE);
1676
1677		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1678		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1679		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1680		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1681		msg_info.hdr.serializing_sc = NULL;
1682		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1683	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1684				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1685		}
1686		return(1);
1687
1688	}
1689
1690	mtx_lock(&lun->lun_lock);
1691    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1692
1693	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1694		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1695		 ooa_links))) {
1696	case CTL_ACTION_BLOCK:
1697		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1698		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1699				  blocked_links);
1700		break;
1701	case CTL_ACTION_PASS:
1702	case CTL_ACTION_SKIP:
1703		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1704			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1705			ctl_enqueue_rtr((union ctl_io *)ctsio);
1706		} else {
1707
1708			/* send msg back to other side */
1709			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1710			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1711			msg_info.hdr.msg_type = CTL_MSG_R2R;
1712#if 0
1713			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1714#endif
1715		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1716			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1717			}
1718		}
1719		break;
1720	case CTL_ACTION_OVERLAP:
1721		/* OVERLAPPED COMMANDS ATTEMPTED */
1722		ctl_set_sense_data(&msg_info.scsi.sense_data,
1723				   lun,
1724				   /*sense_format*/SSD_TYPE_NONE,
1725				   /*current_error*/ 1,
1726				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1727				   /*asc*/ 0x4E,
1728				   /*ascq*/ 0x00,
1729				   SSD_ELEM_NONE);
1730
1731		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1732		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1733		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1734		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1735		msg_info.hdr.serializing_sc = NULL;
1736		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1737#if 0
1738		printf("BAD JUJU:Major Bummer Overlap\n");
1739#endif
1740		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1741		retval = 1;
1742		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1743		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1744		}
1745		break;
1746	case CTL_ACTION_OVERLAP_TAG:
1747		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1748		ctl_set_sense_data(&msg_info.scsi.sense_data,
1749				   lun,
1750				   /*sense_format*/SSD_TYPE_NONE,
1751				   /*current_error*/ 1,
1752				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1753				   /*asc*/ 0x4D,
1754				   /*ascq*/ ctsio->tag_num & 0xff,
1755				   SSD_ELEM_NONE);
1756
1757		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1758		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1759		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1760		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1761		msg_info.hdr.serializing_sc = NULL;
1762		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1763#if 0
1764		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1765#endif
1766		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1767		retval = 1;
1768		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1769		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1770		}
1771		break;
1772	case CTL_ACTION_ERROR:
1773	default:
1774		/* "Internal target failure" */
1775		ctl_set_sense_data(&msg_info.scsi.sense_data,
1776				   lun,
1777				   /*sense_format*/SSD_TYPE_NONE,
1778				   /*current_error*/ 1,
1779				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1780				   /*asc*/ 0x44,
1781				   /*ascq*/ 0x00,
1782				   SSD_ELEM_NONE);
1783
1784		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1785		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1786		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1787		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1788		msg_info.hdr.serializing_sc = NULL;
1789		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1790#if 0
1791		printf("BAD JUJU:Major Bummer HW Error\n");
1792#endif
1793		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1794		retval = 1;
1795		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1796		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1797		}
1798		break;
1799	}
1800	mtx_unlock(&lun->lun_lock);
1801	return (retval);
1802}
1803
1804static int
1805ctl_ioctl_submit_wait(union ctl_io *io)
1806{
1807	struct ctl_fe_ioctl_params params;
1808	ctl_fe_ioctl_state last_state;
1809	int done, retval;
1810
1811	retval = 0;
1812
1813	bzero(&params, sizeof(params));
1814
1815	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1816	cv_init(&params.sem, "ctlioccv");
1817	params.state = CTL_IOCTL_INPROG;
1818	last_state = params.state;
1819
1820	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1821
1822	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1823
1824	/* This shouldn't happen */
1825	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1826		return (retval);
1827
1828	done = 0;
1829
1830	do {
1831		mtx_lock(&params.ioctl_mtx);
1832		/*
1833		 * Check the state here, and don't sleep if the state has
1834		 * already changed (i.e. wakeup has already occured, but we
1835		 * weren't waiting yet).
1836		 */
1837		if (params.state == last_state) {
1838			/* XXX KDM cv_wait_sig instead? */
1839			cv_wait(&params.sem, &params.ioctl_mtx);
1840		}
1841		last_state = params.state;
1842
1843		switch (params.state) {
1844		case CTL_IOCTL_INPROG:
1845			/* Why did we wake up? */
1846			/* XXX KDM error here? */
1847			mtx_unlock(&params.ioctl_mtx);
1848			break;
1849		case CTL_IOCTL_DATAMOVE:
1850			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1851
1852			/*
1853			 * change last_state back to INPROG to avoid
1854			 * deadlock on subsequent data moves.
1855			 */
1856			params.state = last_state = CTL_IOCTL_INPROG;
1857
1858			mtx_unlock(&params.ioctl_mtx);
1859			ctl_ioctl_do_datamove(&io->scsiio);
1860			/*
1861			 * Note that in some cases, most notably writes,
1862			 * this will queue the I/O and call us back later.
1863			 * In other cases, generally reads, this routine
1864			 * will immediately call back and wake us up,
1865			 * probably using our own context.
1866			 */
1867			io->scsiio.be_move_done(io);
1868			break;
1869		case CTL_IOCTL_DONE:
1870			mtx_unlock(&params.ioctl_mtx);
1871			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1872			done = 1;
1873			break;
1874		default:
1875			mtx_unlock(&params.ioctl_mtx);
1876			/* XXX KDM error here? */
1877			break;
1878		}
1879	} while (done == 0);
1880
1881	mtx_destroy(&params.ioctl_mtx);
1882	cv_destroy(&params.sem);
1883
1884	return (CTL_RETVAL_COMPLETE);
1885}
1886
1887static void
1888ctl_ioctl_datamove(union ctl_io *io)
1889{
1890	struct ctl_fe_ioctl_params *params;
1891
1892	params = (struct ctl_fe_ioctl_params *)
1893		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1894
1895	mtx_lock(&params->ioctl_mtx);
1896	params->state = CTL_IOCTL_DATAMOVE;
1897	cv_broadcast(&params->sem);
1898	mtx_unlock(&params->ioctl_mtx);
1899}
1900
1901static void
1902ctl_ioctl_done(union ctl_io *io)
1903{
1904	struct ctl_fe_ioctl_params *params;
1905
1906	params = (struct ctl_fe_ioctl_params *)
1907		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1908
1909	mtx_lock(&params->ioctl_mtx);
1910	params->state = CTL_IOCTL_DONE;
1911	cv_broadcast(&params->sem);
1912	mtx_unlock(&params->ioctl_mtx);
1913}
1914
1915static void
1916ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1917{
1918	struct ctl_fe_ioctl_startstop_info *sd_info;
1919
1920	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1921
1922	sd_info->hs_info.status = metatask->status;
1923	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1924	sd_info->hs_info.luns_complete =
1925		metatask->taskinfo.startstop.luns_complete;
1926	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1927
1928	cv_broadcast(&sd_info->sem);
1929}
1930
1931static void
1932ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1933{
1934	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1935
1936	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1937
1938	mtx_lock(fe_bbr_info->lock);
1939	fe_bbr_info->bbr_info->status = metatask->status;
1940	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1941	fe_bbr_info->wakeup_done = 1;
1942	mtx_unlock(fe_bbr_info->lock);
1943
1944	cv_broadcast(&fe_bbr_info->sem);
1945}
1946
1947/*
1948 * Returns 0 for success, errno for failure.
1949 */
1950static int
1951ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1952		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1953{
1954	union ctl_io *io;
1955	int retval;
1956
1957	retval = 0;
1958
1959	mtx_lock(&lun->lun_lock);
1960	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1961	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1962	     ooa_links)) {
1963		struct ctl_ooa_entry *entry;
1964
1965		/*
1966		 * If we've got more than we can fit, just count the
1967		 * remaining entries.
1968		 */
1969		if (*cur_fill_num >= ooa_hdr->alloc_num)
1970			continue;
1971
1972		entry = &kern_entries[*cur_fill_num];
1973
1974		entry->tag_num = io->scsiio.tag_num;
1975		entry->lun_num = lun->lun;
1976#ifdef CTL_TIME_IO
1977		entry->start_bt = io->io_hdr.start_bt;
1978#endif
1979		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1980		entry->cdb_len = io->scsiio.cdb_len;
1981		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1982			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1983
1984		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1985			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1986
1987		if (io->io_hdr.flags & CTL_FLAG_ABORT)
1988			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1989
1990		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1991			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1992
1993		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1994			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1995	}
1996	mtx_unlock(&lun->lun_lock);
1997
1998	return (retval);
1999}
2000
2001static void *
2002ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2003		 size_t error_str_len)
2004{
2005	void *kptr;
2006
2007	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2008
2009	if (copyin(user_addr, kptr, len) != 0) {
2010		snprintf(error_str, error_str_len, "Error copying %d bytes "
2011			 "from user address %p to kernel address %p", len,
2012			 user_addr, kptr);
2013		free(kptr, M_CTL);
2014		return (NULL);
2015	}
2016
2017	return (kptr);
2018}
2019
2020static void
2021ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2022{
2023	int i;
2024
2025	if (be_args == NULL)
2026		return;
2027
2028	for (i = 0; i < num_be_args; i++) {
2029		free(be_args[i].kname, M_CTL);
2030		free(be_args[i].kvalue, M_CTL);
2031	}
2032
2033	free(be_args, M_CTL);
2034}
2035
2036static struct ctl_be_arg *
2037ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2038		char *error_str, size_t error_str_len)
2039{
2040	struct ctl_be_arg *args;
2041	int i;
2042
2043	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2044				error_str, error_str_len);
2045
2046	if (args == NULL)
2047		goto bailout;
2048
2049	for (i = 0; i < num_be_args; i++) {
2050		args[i].kname = NULL;
2051		args[i].kvalue = NULL;
2052	}
2053
2054	for (i = 0; i < num_be_args; i++) {
2055		uint8_t *tmpptr;
2056
2057		args[i].kname = ctl_copyin_alloc(args[i].name,
2058			args[i].namelen, error_str, error_str_len);
2059		if (args[i].kname == NULL)
2060			goto bailout;
2061
2062		if (args[i].kname[args[i].namelen - 1] != '\0') {
2063			snprintf(error_str, error_str_len, "Argument %d "
2064				 "name is not NUL-terminated", i);
2065			goto bailout;
2066		}
2067
2068		args[i].kvalue = NULL;
2069
2070		tmpptr = ctl_copyin_alloc(args[i].value,
2071			args[i].vallen, error_str, error_str_len);
2072		if (tmpptr == NULL)
2073			goto bailout;
2074
2075		args[i].kvalue = tmpptr;
2076
2077		if ((args[i].flags & CTL_BEARG_ASCII)
2078		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2079			snprintf(error_str, error_str_len, "Argument %d "
2080				 "value is not NUL-terminated", i);
2081			goto bailout;
2082		}
2083	}
2084
2085	return (args);
2086bailout:
2087
2088	ctl_free_args(num_be_args, args);
2089
2090	return (NULL);
2091}
2092
2093/*
2094 * Escape characters that are illegal or not recommended in XML.
2095 */
2096int
2097ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2098{
2099	int retval;
2100
2101	retval = 0;
2102
2103	for (; *str; str++) {
2104		switch (*str) {
2105		case '&':
2106			retval = sbuf_printf(sb, "&amp;");
2107			break;
2108		case '>':
2109			retval = sbuf_printf(sb, "&gt;");
2110			break;
2111		case '<':
2112			retval = sbuf_printf(sb, "&lt;");
2113			break;
2114		default:
2115			retval = sbuf_putc(sb, *str);
2116			break;
2117		}
2118
2119		if (retval != 0)
2120			break;
2121
2122	}
2123
2124	return (retval);
2125}
2126
2127static int
2128ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2129	  struct thread *td)
2130{
2131	struct ctl_softc *softc;
2132	int retval;
2133
2134	softc = control_softc;
2135
2136	retval = 0;
2137
2138	switch (cmd) {
2139	case CTL_IO: {
2140		union ctl_io *io;
2141		void *pool_tmp;
2142
2143		/*
2144		 * If we haven't been "enabled", don't allow any SCSI I/O
2145		 * to this FETD.
2146		 */
2147		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2148			retval = EPERM;
2149			break;
2150		}
2151
2152		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2153		if (io == NULL) {
2154			printf("ctl_ioctl: can't allocate ctl_io!\n");
2155			retval = ENOSPC;
2156			break;
2157		}
2158
2159		/*
2160		 * Need to save the pool reference so it doesn't get
2161		 * spammed by the user's ctl_io.
2162		 */
2163		pool_tmp = io->io_hdr.pool;
2164
2165		memcpy(io, (void *)addr, sizeof(*io));
2166
2167		io->io_hdr.pool = pool_tmp;
2168		/*
2169		 * No status yet, so make sure the status is set properly.
2170		 */
2171		io->io_hdr.status = CTL_STATUS_NONE;
2172
2173		/*
2174		 * The user sets the initiator ID, target and LUN IDs.
2175		 */
2176		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2177		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2178		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2179		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2180			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2181
2182		retval = ctl_ioctl_submit_wait(io);
2183
2184		if (retval != 0) {
2185			ctl_free_io(io);
2186			break;
2187		}
2188
2189		memcpy((void *)addr, io, sizeof(*io));
2190
2191		/* return this to our pool */
2192		ctl_free_io(io);
2193
2194		break;
2195	}
2196	case CTL_ENABLE_PORT:
2197	case CTL_DISABLE_PORT:
2198	case CTL_SET_PORT_WWNS: {
2199		struct ctl_frontend *fe;
2200		struct ctl_port_entry *entry;
2201
2202		entry = (struct ctl_port_entry *)addr;
2203
2204		mtx_lock(&softc->ctl_lock);
2205		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2206			int action, done;
2207
2208			action = 0;
2209			done = 0;
2210
2211			if ((entry->port_type == CTL_PORT_NONE)
2212			 && (entry->targ_port == fe->targ_port)) {
2213				/*
2214				 * If the user only wants to enable or
2215				 * disable or set WWNs on a specific port,
2216				 * do the operation and we're done.
2217				 */
2218				action = 1;
2219				done = 1;
2220			} else if (entry->port_type & fe->port_type) {
2221				/*
2222				 * Compare the user's type mask with the
2223				 * particular frontend type to see if we
2224				 * have a match.
2225				 */
2226				action = 1;
2227				done = 0;
2228
2229				/*
2230				 * Make sure the user isn't trying to set
2231				 * WWNs on multiple ports at the same time.
2232				 */
2233				if (cmd == CTL_SET_PORT_WWNS) {
2234					printf("%s: Can't set WWNs on "
2235					       "multiple ports\n", __func__);
2236					retval = EINVAL;
2237					break;
2238				}
2239			}
2240			if (action != 0) {
2241				/*
2242				 * XXX KDM we have to drop the lock here,
2243				 * because the online/offline operations
2244				 * can potentially block.  We need to
2245				 * reference count the frontends so they
2246				 * can't go away,
2247				 */
2248				mtx_unlock(&softc->ctl_lock);
2249
2250				if (cmd == CTL_ENABLE_PORT) {
2251					struct ctl_lun *lun;
2252
2253					STAILQ_FOREACH(lun, &softc->lun_list,
2254						       links) {
2255						fe->lun_enable(fe->targ_lun_arg,
2256						    lun->target,
2257						    lun->lun);
2258					}
2259
2260					ctl_frontend_online(fe);
2261				} else if (cmd == CTL_DISABLE_PORT) {
2262					struct ctl_lun *lun;
2263
2264					ctl_frontend_offline(fe);
2265
2266					STAILQ_FOREACH(lun, &softc->lun_list,
2267						       links) {
2268						fe->lun_disable(
2269						    fe->targ_lun_arg,
2270						    lun->target,
2271						    lun->lun);
2272					}
2273				}
2274
2275				mtx_lock(&softc->ctl_lock);
2276
2277				if (cmd == CTL_SET_PORT_WWNS)
2278					ctl_frontend_set_wwns(fe,
2279					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2280					    1 : 0, entry->wwnn,
2281					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2282					    1 : 0, entry->wwpn);
2283			}
2284			if (done != 0)
2285				break;
2286		}
2287		mtx_unlock(&softc->ctl_lock);
2288		break;
2289	}
2290	case CTL_GET_PORT_LIST: {
2291		struct ctl_frontend *fe;
2292		struct ctl_port_list *list;
2293		int i;
2294
2295		list = (struct ctl_port_list *)addr;
2296
2297		if (list->alloc_len != (list->alloc_num *
2298		    sizeof(struct ctl_port_entry))) {
2299			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2300			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2301			       "%zu\n", __func__, list->alloc_len,
2302			       list->alloc_num, sizeof(struct ctl_port_entry));
2303			retval = EINVAL;
2304			break;
2305		}
2306		list->fill_len = 0;
2307		list->fill_num = 0;
2308		list->dropped_num = 0;
2309		i = 0;
2310		mtx_lock(&softc->ctl_lock);
2311		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2312			struct ctl_port_entry entry, *list_entry;
2313
2314			if (list->fill_num >= list->alloc_num) {
2315				list->dropped_num++;
2316				continue;
2317			}
2318
2319			entry.port_type = fe->port_type;
2320			strlcpy(entry.port_name, fe->port_name,
2321				sizeof(entry.port_name));
2322			entry.targ_port = fe->targ_port;
2323			entry.physical_port = fe->physical_port;
2324			entry.virtual_port = fe->virtual_port;
2325			entry.wwnn = fe->wwnn;
2326			entry.wwpn = fe->wwpn;
2327			if (fe->status & CTL_PORT_STATUS_ONLINE)
2328				entry.online = 1;
2329			else
2330				entry.online = 0;
2331
2332			list_entry = &list->entries[i];
2333
2334			retval = copyout(&entry, list_entry, sizeof(entry));
2335			if (retval != 0) {
2336				printf("%s: CTL_GET_PORT_LIST: copyout "
2337				       "returned %d\n", __func__, retval);
2338				break;
2339			}
2340			i++;
2341			list->fill_num++;
2342			list->fill_len += sizeof(entry);
2343		}
2344		mtx_unlock(&softc->ctl_lock);
2345
2346		/*
2347		 * If this is non-zero, we had a copyout fault, so there's
2348		 * probably no point in attempting to set the status inside
2349		 * the structure.
2350		 */
2351		if (retval != 0)
2352			break;
2353
2354		if (list->dropped_num > 0)
2355			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2356		else
2357			list->status = CTL_PORT_LIST_OK;
2358		break;
2359	}
2360	case CTL_DUMP_OOA: {
2361		struct ctl_lun *lun;
2362		union ctl_io *io;
2363		char printbuf[128];
2364		struct sbuf sb;
2365
2366		mtx_lock(&softc->ctl_lock);
2367		printf("Dumping OOA queues:\n");
2368		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2369			mtx_lock(&lun->lun_lock);
2370			for (io = (union ctl_io *)TAILQ_FIRST(
2371			     &lun->ooa_queue); io != NULL;
2372			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2373			     ooa_links)) {
2374				sbuf_new(&sb, printbuf, sizeof(printbuf),
2375					 SBUF_FIXEDLEN);
2376				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2377					    (intmax_t)lun->lun,
2378					    io->scsiio.tag_num,
2379					    (io->io_hdr.flags &
2380					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2381					    (io->io_hdr.flags &
2382					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2383					    (io->io_hdr.flags &
2384					    CTL_FLAG_ABORT) ? " ABORT" : "",
2385			                    (io->io_hdr.flags &
2386		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2387				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2388				sbuf_finish(&sb);
2389				printf("%s\n", sbuf_data(&sb));
2390			}
2391			mtx_unlock(&lun->lun_lock);
2392		}
2393		printf("OOA queues dump done\n");
2394		mtx_unlock(&softc->ctl_lock);
2395		break;
2396	}
2397	case CTL_GET_OOA: {
2398		struct ctl_lun *lun;
2399		struct ctl_ooa *ooa_hdr;
2400		struct ctl_ooa_entry *entries;
2401		uint32_t cur_fill_num;
2402
2403		ooa_hdr = (struct ctl_ooa *)addr;
2404
2405		if ((ooa_hdr->alloc_len == 0)
2406		 || (ooa_hdr->alloc_num == 0)) {
2407			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2408			       "must be non-zero\n", __func__,
2409			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2410			retval = EINVAL;
2411			break;
2412		}
2413
2414		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2415		    sizeof(struct ctl_ooa_entry))) {
2416			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2417			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2418			       __func__, ooa_hdr->alloc_len,
2419			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2420			retval = EINVAL;
2421			break;
2422		}
2423
2424		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2425		if (entries == NULL) {
2426			printf("%s: could not allocate %d bytes for OOA "
2427			       "dump\n", __func__, ooa_hdr->alloc_len);
2428			retval = ENOMEM;
2429			break;
2430		}
2431
2432		mtx_lock(&softc->ctl_lock);
2433		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2434		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2435		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2436			mtx_unlock(&softc->ctl_lock);
2437			free(entries, M_CTL);
2438			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2439			       __func__, (uintmax_t)ooa_hdr->lun_num);
2440			retval = EINVAL;
2441			break;
2442		}
2443
2444		cur_fill_num = 0;
2445
2446		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2447			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2448				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2449					ooa_hdr, entries);
2450				if (retval != 0)
2451					break;
2452			}
2453			if (retval != 0) {
2454				mtx_unlock(&softc->ctl_lock);
2455				free(entries, M_CTL);
2456				break;
2457			}
2458		} else {
2459			lun = softc->ctl_luns[ooa_hdr->lun_num];
2460
2461			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2462						    entries);
2463		}
2464		mtx_unlock(&softc->ctl_lock);
2465
2466		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2467		ooa_hdr->fill_len = ooa_hdr->fill_num *
2468			sizeof(struct ctl_ooa_entry);
2469		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2470		if (retval != 0) {
2471			printf("%s: error copying out %d bytes for OOA dump\n",
2472			       __func__, ooa_hdr->fill_len);
2473		}
2474
2475		getbintime(&ooa_hdr->cur_bt);
2476
2477		if (cur_fill_num > ooa_hdr->alloc_num) {
2478			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2479			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2480		} else {
2481			ooa_hdr->dropped_num = 0;
2482			ooa_hdr->status = CTL_OOA_OK;
2483		}
2484
2485		free(entries, M_CTL);
2486		break;
2487	}
2488	case CTL_CHECK_OOA: {
2489		union ctl_io *io;
2490		struct ctl_lun *lun;
2491		struct ctl_ooa_info *ooa_info;
2492
2493
2494		ooa_info = (struct ctl_ooa_info *)addr;
2495
2496		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2497			ooa_info->status = CTL_OOA_INVALID_LUN;
2498			break;
2499		}
2500		mtx_lock(&softc->ctl_lock);
2501		lun = softc->ctl_luns[ooa_info->lun_id];
2502		if (lun == NULL) {
2503			mtx_unlock(&softc->ctl_lock);
2504			ooa_info->status = CTL_OOA_INVALID_LUN;
2505			break;
2506		}
2507		mtx_lock(&lun->lun_lock);
2508		mtx_unlock(&softc->ctl_lock);
2509		ooa_info->num_entries = 0;
2510		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2511		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2512		     &io->io_hdr, ooa_links)) {
2513			ooa_info->num_entries++;
2514		}
2515		mtx_unlock(&lun->lun_lock);
2516
2517		ooa_info->status = CTL_OOA_SUCCESS;
2518
2519		break;
2520	}
2521	case CTL_HARD_START:
2522	case CTL_HARD_STOP: {
2523		struct ctl_fe_ioctl_startstop_info ss_info;
2524		struct cfi_metatask *metatask;
2525		struct mtx hs_mtx;
2526
2527		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2528
2529		cv_init(&ss_info.sem, "hard start/stop cv" );
2530
2531		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2532		if (metatask == NULL) {
2533			retval = ENOMEM;
2534			mtx_destroy(&hs_mtx);
2535			break;
2536		}
2537
2538		if (cmd == CTL_HARD_START)
2539			metatask->tasktype = CFI_TASK_STARTUP;
2540		else
2541			metatask->tasktype = CFI_TASK_SHUTDOWN;
2542
2543		metatask->callback = ctl_ioctl_hard_startstop_callback;
2544		metatask->callback_arg = &ss_info;
2545
2546		cfi_action(metatask);
2547
2548		/* Wait for the callback */
2549		mtx_lock(&hs_mtx);
2550		cv_wait_sig(&ss_info.sem, &hs_mtx);
2551		mtx_unlock(&hs_mtx);
2552
2553		/*
2554		 * All information has been copied from the metatask by the
2555		 * time cv_broadcast() is called, so we free the metatask here.
2556		 */
2557		cfi_free_metatask(metatask);
2558
2559		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2560
2561		mtx_destroy(&hs_mtx);
2562		break;
2563	}
2564	case CTL_BBRREAD: {
2565		struct ctl_bbrread_info *bbr_info;
2566		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2567		struct mtx bbr_mtx;
2568		struct cfi_metatask *metatask;
2569
2570		bbr_info = (struct ctl_bbrread_info *)addr;
2571
2572		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2573
2574		bzero(&bbr_mtx, sizeof(bbr_mtx));
2575		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2576
2577		fe_bbr_info.bbr_info = bbr_info;
2578		fe_bbr_info.lock = &bbr_mtx;
2579
2580		cv_init(&fe_bbr_info.sem, "BBR read cv");
2581		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2582
2583		if (metatask == NULL) {
2584			mtx_destroy(&bbr_mtx);
2585			cv_destroy(&fe_bbr_info.sem);
2586			retval = ENOMEM;
2587			break;
2588		}
2589		metatask->tasktype = CFI_TASK_BBRREAD;
2590		metatask->callback = ctl_ioctl_bbrread_callback;
2591		metatask->callback_arg = &fe_bbr_info;
2592		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2593		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2594		metatask->taskinfo.bbrread.len = bbr_info->len;
2595
2596		cfi_action(metatask);
2597
2598		mtx_lock(&bbr_mtx);
2599		while (fe_bbr_info.wakeup_done == 0)
2600			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2601		mtx_unlock(&bbr_mtx);
2602
2603		bbr_info->status = metatask->status;
2604		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2605		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2606		memcpy(&bbr_info->sense_data,
2607		       &metatask->taskinfo.bbrread.sense_data,
2608		       ctl_min(sizeof(bbr_info->sense_data),
2609			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2610
2611		cfi_free_metatask(metatask);
2612
2613		mtx_destroy(&bbr_mtx);
2614		cv_destroy(&fe_bbr_info.sem);
2615
2616		break;
2617	}
2618	case CTL_DELAY_IO: {
2619		struct ctl_io_delay_info *delay_info;
2620#ifdef CTL_IO_DELAY
2621		struct ctl_lun *lun;
2622#endif /* CTL_IO_DELAY */
2623
2624		delay_info = (struct ctl_io_delay_info *)addr;
2625
2626#ifdef CTL_IO_DELAY
2627		mtx_lock(&softc->ctl_lock);
2628
2629		if ((delay_info->lun_id > CTL_MAX_LUNS)
2630		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2631			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2632		} else {
2633			lun = softc->ctl_luns[delay_info->lun_id];
2634			mtx_lock(&lun->lun_lock);
2635
2636			delay_info->status = CTL_DELAY_STATUS_OK;
2637
2638			switch (delay_info->delay_type) {
2639			case CTL_DELAY_TYPE_CONT:
2640				break;
2641			case CTL_DELAY_TYPE_ONESHOT:
2642				break;
2643			default:
2644				delay_info->status =
2645					CTL_DELAY_STATUS_INVALID_TYPE;
2646				break;
2647			}
2648
2649			switch (delay_info->delay_loc) {
2650			case CTL_DELAY_LOC_DATAMOVE:
2651				lun->delay_info.datamove_type =
2652					delay_info->delay_type;
2653				lun->delay_info.datamove_delay =
2654					delay_info->delay_secs;
2655				break;
2656			case CTL_DELAY_LOC_DONE:
2657				lun->delay_info.done_type =
2658					delay_info->delay_type;
2659				lun->delay_info.done_delay =
2660					delay_info->delay_secs;
2661				break;
2662			default:
2663				delay_info->status =
2664					CTL_DELAY_STATUS_INVALID_LOC;
2665				break;
2666			}
2667			mtx_unlock(&lun->lun_lock);
2668		}
2669
2670		mtx_unlock(&softc->ctl_lock);
2671#else
2672		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2673#endif /* CTL_IO_DELAY */
2674		break;
2675	}
2676	case CTL_REALSYNC_SET: {
2677		int *syncstate;
2678
2679		syncstate = (int *)addr;
2680
2681		mtx_lock(&softc->ctl_lock);
2682		switch (*syncstate) {
2683		case 0:
2684			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2685			break;
2686		case 1:
2687			softc->flags |= CTL_FLAG_REAL_SYNC;
2688			break;
2689		default:
2690			retval = EINVAL;
2691			break;
2692		}
2693		mtx_unlock(&softc->ctl_lock);
2694		break;
2695	}
2696	case CTL_REALSYNC_GET: {
2697		int *syncstate;
2698
2699		syncstate = (int*)addr;
2700
2701		mtx_lock(&softc->ctl_lock);
2702		if (softc->flags & CTL_FLAG_REAL_SYNC)
2703			*syncstate = 1;
2704		else
2705			*syncstate = 0;
2706		mtx_unlock(&softc->ctl_lock);
2707
2708		break;
2709	}
2710	case CTL_SETSYNC:
2711	case CTL_GETSYNC: {
2712		struct ctl_sync_info *sync_info;
2713		struct ctl_lun *lun;
2714
2715		sync_info = (struct ctl_sync_info *)addr;
2716
2717		mtx_lock(&softc->ctl_lock);
2718		lun = softc->ctl_luns[sync_info->lun_id];
2719		if (lun == NULL) {
2720			mtx_unlock(&softc->ctl_lock);
2721			sync_info->status = CTL_GS_SYNC_NO_LUN;
2722		}
2723		/*
2724		 * Get or set the sync interval.  We're not bounds checking
2725		 * in the set case, hopefully the user won't do something
2726		 * silly.
2727		 */
2728		mtx_lock(&lun->lun_lock);
2729		mtx_unlock(&softc->ctl_lock);
2730		if (cmd == CTL_GETSYNC)
2731			sync_info->sync_interval = lun->sync_interval;
2732		else
2733			lun->sync_interval = sync_info->sync_interval;
2734		mtx_unlock(&lun->lun_lock);
2735
2736		sync_info->status = CTL_GS_SYNC_OK;
2737
2738		break;
2739	}
2740	case CTL_GETSTATS: {
2741		struct ctl_stats *stats;
2742		struct ctl_lun *lun;
2743		int i;
2744
2745		stats = (struct ctl_stats *)addr;
2746
2747		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2748		     stats->alloc_len) {
2749			stats->status = CTL_SS_NEED_MORE_SPACE;
2750			stats->num_luns = softc->num_luns;
2751			break;
2752		}
2753		/*
2754		 * XXX KDM no locking here.  If the LUN list changes,
2755		 * things can blow up.
2756		 */
2757		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2758		     i++, lun = STAILQ_NEXT(lun, links)) {
2759			retval = copyout(&lun->stats, &stats->lun_stats[i],
2760					 sizeof(lun->stats));
2761			if (retval != 0)
2762				break;
2763		}
2764		stats->num_luns = softc->num_luns;
2765		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2766				 softc->num_luns;
2767		stats->status = CTL_SS_OK;
2768#ifdef CTL_TIME_IO
2769		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2770#else
2771		stats->flags = CTL_STATS_FLAG_NONE;
2772#endif
2773		getnanouptime(&stats->timestamp);
2774		break;
2775	}
2776	case CTL_ERROR_INJECT: {
2777		struct ctl_error_desc *err_desc, *new_err_desc;
2778		struct ctl_lun *lun;
2779
2780		err_desc = (struct ctl_error_desc *)addr;
2781
2782		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2783				      M_WAITOK | M_ZERO);
2784		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2785
2786		mtx_lock(&softc->ctl_lock);
2787		lun = softc->ctl_luns[err_desc->lun_id];
2788		if (lun == NULL) {
2789			mtx_unlock(&softc->ctl_lock);
2790			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2791			       __func__, (uintmax_t)err_desc->lun_id);
2792			retval = EINVAL;
2793			break;
2794		}
2795		mtx_lock(&lun->lun_lock);
2796		mtx_unlock(&softc->ctl_lock);
2797
2798		/*
2799		 * We could do some checking here to verify the validity
2800		 * of the request, but given the complexity of error
2801		 * injection requests, the checking logic would be fairly
2802		 * complex.
2803		 *
2804		 * For now, if the request is invalid, it just won't get
2805		 * executed and might get deleted.
2806		 */
2807		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2808
2809		/*
2810		 * XXX KDM check to make sure the serial number is unique,
2811		 * in case we somehow manage to wrap.  That shouldn't
2812		 * happen for a very long time, but it's the right thing to
2813		 * do.
2814		 */
2815		new_err_desc->serial = lun->error_serial;
2816		err_desc->serial = lun->error_serial;
2817		lun->error_serial++;
2818
2819		mtx_unlock(&lun->lun_lock);
2820		break;
2821	}
2822	case CTL_ERROR_INJECT_DELETE: {
2823		struct ctl_error_desc *delete_desc, *desc, *desc2;
2824		struct ctl_lun *lun;
2825		int delete_done;
2826
2827		delete_desc = (struct ctl_error_desc *)addr;
2828		delete_done = 0;
2829
2830		mtx_lock(&softc->ctl_lock);
2831		lun = softc->ctl_luns[delete_desc->lun_id];
2832		if (lun == NULL) {
2833			mtx_unlock(&softc->ctl_lock);
2834			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2835			       __func__, (uintmax_t)delete_desc->lun_id);
2836			retval = EINVAL;
2837			break;
2838		}
2839		mtx_lock(&lun->lun_lock);
2840		mtx_unlock(&softc->ctl_lock);
2841		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2842			if (desc->serial != delete_desc->serial)
2843				continue;
2844
2845			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2846				      links);
2847			free(desc, M_CTL);
2848			delete_done = 1;
2849		}
2850		mtx_unlock(&lun->lun_lock);
2851		if (delete_done == 0) {
2852			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2853			       "error serial %ju on LUN %u\n", __func__,
2854			       delete_desc->serial, delete_desc->lun_id);
2855			retval = EINVAL;
2856			break;
2857		}
2858		break;
2859	}
2860	case CTL_DUMP_STRUCTS: {
2861		int i, j, k;
2862		struct ctl_frontend *fe;
2863
2864		printf("CTL IID to WWPN map start:\n");
2865		for (i = 0; i < CTL_MAX_PORTS; i++) {
2866			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2867				if (softc->wwpn_iid[i][j].in_use == 0)
2868					continue;
2869
2870				printf("port %d iid %u WWPN %#jx\n",
2871				       softc->wwpn_iid[i][j].port,
2872				       softc->wwpn_iid[i][j].iid,
2873				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2874			}
2875		}
2876		printf("CTL IID to WWPN map end\n");
2877		printf("CTL Persistent Reservation information start:\n");
2878		for (i = 0; i < CTL_MAX_LUNS; i++) {
2879			struct ctl_lun *lun;
2880
2881			lun = softc->ctl_luns[i];
2882
2883			if ((lun == NULL)
2884			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2885				continue;
2886
2887			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2888				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2889					if (lun->per_res[j+k].registered == 0)
2890						continue;
2891					printf("LUN %d port %d iid %d key "
2892					       "%#jx\n", i, j, k,
2893					       (uintmax_t)scsi_8btou64(
2894					       lun->per_res[j+k].res_key.key));
2895				}
2896			}
2897		}
2898		printf("CTL Persistent Reservation information end\n");
2899		printf("CTL Frontends:\n");
2900		/*
2901		 * XXX KDM calling this without a lock.  We'd likely want
2902		 * to drop the lock before calling the frontend's dump
2903		 * routine anyway.
2904		 */
2905		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2906			printf("Frontend %s Type %u pport %d vport %d WWNN "
2907			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2908			       fe->physical_port, fe->virtual_port,
2909			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2910
2911			/*
2912			 * Frontends are not required to support the dump
2913			 * routine.
2914			 */
2915			if (fe->fe_dump == NULL)
2916				continue;
2917
2918			fe->fe_dump();
2919		}
2920		printf("CTL Frontend information end\n");
2921		break;
2922	}
2923	case CTL_LUN_REQ: {
2924		struct ctl_lun_req *lun_req;
2925		struct ctl_backend_driver *backend;
2926
2927		lun_req = (struct ctl_lun_req *)addr;
2928
2929		backend = ctl_backend_find(lun_req->backend);
2930		if (backend == NULL) {
2931			lun_req->status = CTL_LUN_ERROR;
2932			snprintf(lun_req->error_str,
2933				 sizeof(lun_req->error_str),
2934				 "Backend \"%s\" not found.",
2935				 lun_req->backend);
2936			break;
2937		}
2938		if (lun_req->num_be_args > 0) {
2939			lun_req->kern_be_args = ctl_copyin_args(
2940				lun_req->num_be_args,
2941				lun_req->be_args,
2942				lun_req->error_str,
2943				sizeof(lun_req->error_str));
2944			if (lun_req->kern_be_args == NULL) {
2945				lun_req->status = CTL_LUN_ERROR;
2946				break;
2947			}
2948		}
2949
2950		retval = backend->ioctl(dev, cmd, addr, flag, td);
2951
2952		if (lun_req->num_be_args > 0) {
2953			ctl_free_args(lun_req->num_be_args,
2954				      lun_req->kern_be_args);
2955		}
2956		break;
2957	}
2958	case CTL_LUN_LIST: {
2959		struct sbuf *sb;
2960		struct ctl_lun *lun;
2961		struct ctl_lun_list *list;
2962		struct ctl_be_lun_option *opt;
2963
2964		list = (struct ctl_lun_list *)addr;
2965
2966		/*
2967		 * Allocate a fixed length sbuf here, based on the length
2968		 * of the user's buffer.  We could allocate an auto-extending
2969		 * buffer, and then tell the user how much larger our
2970		 * amount of data is than his buffer, but that presents
2971		 * some problems:
2972		 *
2973		 * 1.  The sbuf(9) routines use a blocking malloc, and so
2974		 *     we can't hold a lock while calling them with an
2975		 *     auto-extending buffer.
2976 		 *
2977		 * 2.  There is not currently a LUN reference counting
2978		 *     mechanism, outside of outstanding transactions on
2979		 *     the LUN's OOA queue.  So a LUN could go away on us
2980		 *     while we're getting the LUN number, backend-specific
2981		 *     information, etc.  Thus, given the way things
2982		 *     currently work, we need to hold the CTL lock while
2983		 *     grabbing LUN information.
2984		 *
2985		 * So, from the user's standpoint, the best thing to do is
2986		 * allocate what he thinks is a reasonable buffer length,
2987		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2988		 * double the buffer length and try again.  (And repeat
2989		 * that until he succeeds.)
2990		 */
2991		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2992		if (sb == NULL) {
2993			list->status = CTL_LUN_LIST_ERROR;
2994			snprintf(list->error_str, sizeof(list->error_str),
2995				 "Unable to allocate %d bytes for LUN list",
2996				 list->alloc_len);
2997			break;
2998		}
2999
3000		sbuf_printf(sb, "<ctllunlist>\n");
3001
3002		mtx_lock(&softc->ctl_lock);
3003		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3004			mtx_lock(&lun->lun_lock);
3005			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3006					     (uintmax_t)lun->lun);
3007
3008			/*
3009			 * Bail out as soon as we see that we've overfilled
3010			 * the buffer.
3011			 */
3012			if (retval != 0)
3013				break;
3014
3015			retval = sbuf_printf(sb, "<backend_type>%s"
3016					     "</backend_type>\n",
3017					     (lun->backend == NULL) ?  "none" :
3018					     lun->backend->name);
3019
3020			if (retval != 0)
3021				break;
3022
3023			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3024					     lun->be_lun->lun_type);
3025
3026			if (retval != 0)
3027				break;
3028
3029			if (lun->backend == NULL) {
3030				retval = sbuf_printf(sb, "</lun>\n");
3031				if (retval != 0)
3032					break;
3033				continue;
3034			}
3035
3036			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3037					     (lun->be_lun->maxlba > 0) ?
3038					     lun->be_lun->maxlba + 1 : 0);
3039
3040			if (retval != 0)
3041				break;
3042
3043			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3044					     lun->be_lun->blocksize);
3045
3046			if (retval != 0)
3047				break;
3048
3049			retval = sbuf_printf(sb, "<serial_number>");
3050
3051			if (retval != 0)
3052				break;
3053
3054			retval = ctl_sbuf_printf_esc(sb,
3055						     lun->be_lun->serial_num);
3056
3057			if (retval != 0)
3058				break;
3059
3060			retval = sbuf_printf(sb, "</serial_number>\n");
3061
3062			if (retval != 0)
3063				break;
3064
3065			retval = sbuf_printf(sb, "<device_id>");
3066
3067			if (retval != 0)
3068				break;
3069
3070			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3071
3072			if (retval != 0)
3073				break;
3074
3075			retval = sbuf_printf(sb, "</device_id>\n");
3076
3077			if (retval != 0)
3078				break;
3079
3080			if (lun->backend->lun_info != NULL) {
3081				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3082				if (retval != 0)
3083					break;
3084			}
3085			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3086				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3087				if (retval != 0)
3088					break;
3089			}
3090
3091			retval = sbuf_printf(sb, "</lun>\n");
3092
3093			if (retval != 0)
3094				break;
3095			mtx_unlock(&lun->lun_lock);
3096		}
3097		if (lun != NULL)
3098			mtx_unlock(&lun->lun_lock);
3099		mtx_unlock(&softc->ctl_lock);
3100
3101		if ((retval != 0)
3102		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3103			retval = 0;
3104			sbuf_delete(sb);
3105			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3106			snprintf(list->error_str, sizeof(list->error_str),
3107				 "Out of space, %d bytes is too small",
3108				 list->alloc_len);
3109			break;
3110		}
3111
3112		sbuf_finish(sb);
3113
3114		retval = copyout(sbuf_data(sb), list->lun_xml,
3115				 sbuf_len(sb) + 1);
3116
3117		list->fill_len = sbuf_len(sb) + 1;
3118		list->status = CTL_LUN_LIST_OK;
3119		sbuf_delete(sb);
3120		break;
3121	}
3122	case CTL_ISCSI: {
3123		struct ctl_iscsi *ci;
3124		struct ctl_frontend *fe;
3125
3126		ci = (struct ctl_iscsi *)addr;
3127
3128		mtx_lock(&softc->ctl_lock);
3129		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3130			if (strcmp(fe->port_name, "iscsi") == 0)
3131				break;
3132		}
3133		mtx_unlock(&softc->ctl_lock);
3134
3135		if (fe == NULL) {
3136			ci->status = CTL_ISCSI_ERROR;
3137			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3138			break;
3139		}
3140
3141		retval = fe->ioctl(dev, cmd, addr, flag, td);
3142		break;
3143	}
3144	default: {
3145		/* XXX KDM should we fix this? */
3146#if 0
3147		struct ctl_backend_driver *backend;
3148		unsigned int type;
3149		int found;
3150
3151		found = 0;
3152
3153		/*
3154		 * We encode the backend type as the ioctl type for backend
3155		 * ioctls.  So parse it out here, and then search for a
3156		 * backend of this type.
3157		 */
3158		type = _IOC_TYPE(cmd);
3159
3160		STAILQ_FOREACH(backend, &softc->be_list, links) {
3161			if (backend->type == type) {
3162				found = 1;
3163				break;
3164			}
3165		}
3166		if (found == 0) {
3167			printf("ctl: unknown ioctl command %#lx or backend "
3168			       "%d\n", cmd, type);
3169			retval = EINVAL;
3170			break;
3171		}
3172		retval = backend->ioctl(dev, cmd, addr, flag, td);
3173#endif
3174		retval = ENOTTY;
3175		break;
3176	}
3177	}
3178	return (retval);
3179}
3180
3181uint32_t
3182ctl_get_initindex(struct ctl_nexus *nexus)
3183{
3184	if (nexus->targ_port < CTL_MAX_PORTS)
3185		return (nexus->initid.id +
3186			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3187	else
3188		return (nexus->initid.id +
3189		       ((nexus->targ_port - CTL_MAX_PORTS) *
3190			CTL_MAX_INIT_PER_PORT));
3191}
3192
3193uint32_t
3194ctl_get_resindex(struct ctl_nexus *nexus)
3195{
3196	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3197}
3198
3199uint32_t
3200ctl_port_idx(int port_num)
3201{
3202	if (port_num < CTL_MAX_PORTS)
3203		return(port_num);
3204	else
3205		return(port_num - CTL_MAX_PORTS);
3206}
3207
3208/*
3209 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3210 * that are a power of 2.
3211 */
3212int
3213ctl_ffz(uint32_t *mask, uint32_t size)
3214{
3215	uint32_t num_chunks, num_pieces;
3216	int i, j;
3217
3218	num_chunks = (size >> 5);
3219	if (num_chunks == 0)
3220		num_chunks++;
3221	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3222
3223	for (i = 0; i < num_chunks; i++) {
3224		for (j = 0; j < num_pieces; j++) {
3225			if ((mask[i] & (1 << j)) == 0)
3226				return ((i << 5) + j);
3227		}
3228	}
3229
3230	return (-1);
3231}
3232
3233int
3234ctl_set_mask(uint32_t *mask, uint32_t bit)
3235{
3236	uint32_t chunk, piece;
3237
3238	chunk = bit >> 5;
3239	piece = bit % (sizeof(uint32_t) * 8);
3240
3241	if ((mask[chunk] & (1 << piece)) != 0)
3242		return (-1);
3243	else
3244		mask[chunk] |= (1 << piece);
3245
3246	return (0);
3247}
3248
3249int
3250ctl_clear_mask(uint32_t *mask, uint32_t bit)
3251{
3252	uint32_t chunk, piece;
3253
3254	chunk = bit >> 5;
3255	piece = bit % (sizeof(uint32_t) * 8);
3256
3257	if ((mask[chunk] & (1 << piece)) == 0)
3258		return (-1);
3259	else
3260		mask[chunk] &= ~(1 << piece);
3261
3262	return (0);
3263}
3264
3265int
3266ctl_is_set(uint32_t *mask, uint32_t bit)
3267{
3268	uint32_t chunk, piece;
3269
3270	chunk = bit >> 5;
3271	piece = bit % (sizeof(uint32_t) * 8);
3272
3273	if ((mask[chunk] & (1 << piece)) == 0)
3274		return (0);
3275	else
3276		return (1);
3277}
3278
3279#ifdef unused
3280/*
3281 * The bus, target and lun are optional, they can be filled in later.
3282 * can_wait is used to determine whether we can wait on the malloc or not.
3283 */
3284union ctl_io*
3285ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3286	      uint32_t targ_lun, int can_wait)
3287{
3288	union ctl_io *io;
3289
3290	if (can_wait)
3291		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3292	else
3293		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3294
3295	if (io != NULL) {
3296		io->io_hdr.io_type = io_type;
3297		io->io_hdr.targ_port = targ_port;
3298		/*
3299		 * XXX KDM this needs to change/go away.  We need to move
3300		 * to a preallocated pool of ctl_scsiio structures.
3301		 */
3302		io->io_hdr.nexus.targ_target.id = targ_target;
3303		io->io_hdr.nexus.targ_lun = targ_lun;
3304	}
3305
3306	return (io);
3307}
3308
3309void
3310ctl_kfree_io(union ctl_io *io)
3311{
3312	free(io, M_CTL);
3313}
3314#endif /* unused */
3315
3316/*
3317 * ctl_softc, pool_type, total_ctl_io are passed in.
3318 * npool is passed out.
3319 */
3320int
3321ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3322		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3323{
3324	uint32_t i;
3325	union ctl_io *cur_io, *next_io;
3326	struct ctl_io_pool *pool;
3327	int retval;
3328
3329	retval = 0;
3330
3331	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3332					    M_NOWAIT | M_ZERO);
3333	if (pool == NULL) {
3334		retval = ENOMEM;
3335		goto bailout;
3336	}
3337
3338	pool->type = pool_type;
3339	pool->ctl_softc = ctl_softc;
3340
3341	mtx_lock(&ctl_softc->pool_lock);
3342	pool->id = ctl_softc->cur_pool_id++;
3343	mtx_unlock(&ctl_softc->pool_lock);
3344
3345	pool->flags = CTL_POOL_FLAG_NONE;
3346	pool->refcount = 1;		/* Reference for validity. */
3347	STAILQ_INIT(&pool->free_queue);
3348
3349	/*
3350	 * XXX KDM other options here:
3351	 * - allocate a page at a time
3352	 * - allocate one big chunk of memory.
3353	 * Page allocation might work well, but would take a little more
3354	 * tracking.
3355	 */
3356	for (i = 0; i < total_ctl_io; i++) {
3357		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3358						M_NOWAIT);
3359		if (cur_io == NULL) {
3360			retval = ENOMEM;
3361			break;
3362		}
3363		cur_io->io_hdr.pool = pool;
3364		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3365		pool->total_ctl_io++;
3366		pool->free_ctl_io++;
3367	}
3368
3369	if (retval != 0) {
3370		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3371		     cur_io != NULL; cur_io = next_io) {
3372			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3373							      links);
3374			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3375				      ctl_io_hdr, links);
3376			free(cur_io, M_CTL);
3377		}
3378
3379		free(pool, M_CTL);
3380		goto bailout;
3381	}
3382	mtx_lock(&ctl_softc->pool_lock);
3383	ctl_softc->num_pools++;
3384	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3385	/*
3386	 * Increment our usage count if this is an external consumer, so we
3387	 * can't get unloaded until the external consumer (most likely a
3388	 * FETD) unloads and frees his pool.
3389	 *
3390	 * XXX KDM will this increment the caller's module use count, or
3391	 * mine?
3392	 */
3393#if 0
3394	if ((pool_type != CTL_POOL_EMERGENCY)
3395	 && (pool_type != CTL_POOL_INTERNAL)
3396	 && (pool_type != CTL_POOL_IOCTL)
3397	 && (pool_type != CTL_POOL_4OTHERSC))
3398		MOD_INC_USE_COUNT;
3399#endif
3400
3401	mtx_unlock(&ctl_softc->pool_lock);
3402
3403	*npool = pool;
3404
3405bailout:
3406
3407	return (retval);
3408}
3409
3410static int
3411ctl_pool_acquire(struct ctl_io_pool *pool)
3412{
3413
3414	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3415
3416	if (pool->flags & CTL_POOL_FLAG_INVALID)
3417		return (EINVAL);
3418
3419	pool->refcount++;
3420
3421	return (0);
3422}
3423
3424static void
3425ctl_pool_release(struct ctl_io_pool *pool)
3426{
3427	struct ctl_softc *ctl_softc = pool->ctl_softc;
3428	union ctl_io *io;
3429
3430	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3431
3432	if (--pool->refcount != 0)
3433		return;
3434
3435	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3436		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3437			      links);
3438		free(io, M_CTL);
3439	}
3440
3441	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3442	ctl_softc->num_pools--;
3443
3444	/*
3445	 * XXX KDM will this decrement the caller's usage count or mine?
3446	 */
3447#if 0
3448	if ((pool->type != CTL_POOL_EMERGENCY)
3449	 && (pool->type != CTL_POOL_INTERNAL)
3450	 && (pool->type != CTL_POOL_IOCTL))
3451		MOD_DEC_USE_COUNT;
3452#endif
3453
3454	free(pool, M_CTL);
3455}
3456
3457void
3458ctl_pool_free(struct ctl_io_pool *pool)
3459{
3460	struct ctl_softc *ctl_softc;
3461
3462	if (pool == NULL)
3463		return;
3464
3465	ctl_softc = pool->ctl_softc;
3466	mtx_lock(&ctl_softc->pool_lock);
3467	pool->flags |= CTL_POOL_FLAG_INVALID;
3468	ctl_pool_release(pool);
3469	mtx_unlock(&ctl_softc->pool_lock);
3470}
3471
3472/*
3473 * This routine does not block (except for spinlocks of course).
3474 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3475 * possible.
3476 */
3477union ctl_io *
3478ctl_alloc_io(void *pool_ref)
3479{
3480	union ctl_io *io;
3481	struct ctl_softc *ctl_softc;
3482	struct ctl_io_pool *pool, *npool;
3483	struct ctl_io_pool *emergency_pool;
3484
3485	pool = (struct ctl_io_pool *)pool_ref;
3486
3487	if (pool == NULL) {
3488		printf("%s: pool is NULL\n", __func__);
3489		return (NULL);
3490	}
3491
3492	emergency_pool = NULL;
3493
3494	ctl_softc = pool->ctl_softc;
3495
3496	mtx_lock(&ctl_softc->pool_lock);
3497	/*
3498	 * First, try to get the io structure from the user's pool.
3499	 */
3500	if (ctl_pool_acquire(pool) == 0) {
3501		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3502		if (io != NULL) {
3503			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3504			pool->total_allocated++;
3505			pool->free_ctl_io--;
3506			mtx_unlock(&ctl_softc->pool_lock);
3507			return (io);
3508		} else
3509			ctl_pool_release(pool);
3510	}
3511	/*
3512	 * If he doesn't have any io structures left, search for an
3513	 * emergency pool and grab one from there.
3514	 */
3515	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3516		if (npool->type != CTL_POOL_EMERGENCY)
3517			continue;
3518
3519		if (ctl_pool_acquire(npool) != 0)
3520			continue;
3521
3522		emergency_pool = npool;
3523
3524		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3525		if (io != NULL) {
3526			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3527			npool->total_allocated++;
3528			npool->free_ctl_io--;
3529			mtx_unlock(&ctl_softc->pool_lock);
3530			return (io);
3531		} else
3532			ctl_pool_release(npool);
3533	}
3534
3535	/* Drop the spinlock before we malloc */
3536	mtx_unlock(&ctl_softc->pool_lock);
3537
3538	/*
3539	 * The emergency pool (if it exists) didn't have one, so try an
3540	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3541	 */
3542	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3543	if (io != NULL) {
3544		/*
3545		 * If the emergency pool exists but is empty, add this
3546		 * ctl_io to its list when it gets freed.
3547		 */
3548		if (emergency_pool != NULL) {
3549			mtx_lock(&ctl_softc->pool_lock);
3550			if (ctl_pool_acquire(emergency_pool) == 0) {
3551				io->io_hdr.pool = emergency_pool;
3552				emergency_pool->total_ctl_io++;
3553				/*
3554				 * Need to bump this, otherwise
3555				 * total_allocated and total_freed won't
3556				 * match when we no longer have anything
3557				 * outstanding.
3558				 */
3559				emergency_pool->total_allocated++;
3560			}
3561			mtx_unlock(&ctl_softc->pool_lock);
3562		} else
3563			io->io_hdr.pool = NULL;
3564	}
3565
3566	return (io);
3567}
3568
3569void
3570ctl_free_io(union ctl_io *io)
3571{
3572	if (io == NULL)
3573		return;
3574
3575	/*
3576	 * If this ctl_io has a pool, return it to that pool.
3577	 */
3578	if (io->io_hdr.pool != NULL) {
3579		struct ctl_io_pool *pool;
3580
3581		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3582		mtx_lock(&pool->ctl_softc->pool_lock);
3583		io->io_hdr.io_type = 0xff;
3584		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3585		pool->total_freed++;
3586		pool->free_ctl_io++;
3587		ctl_pool_release(pool);
3588		mtx_unlock(&pool->ctl_softc->pool_lock);
3589	} else {
3590		/*
3591		 * Otherwise, just free it.  We probably malloced it and
3592		 * the emergency pool wasn't available.
3593		 */
3594		free(io, M_CTL);
3595	}
3596
3597}
3598
3599void
3600ctl_zero_io(union ctl_io *io)
3601{
3602	void *pool_ref;
3603
3604	if (io == NULL)
3605		return;
3606
3607	/*
3608	 * May need to preserve linked list pointers at some point too.
3609	 */
3610	pool_ref = io->io_hdr.pool;
3611
3612	memset(io, 0, sizeof(*io));
3613
3614	io->io_hdr.pool = pool_ref;
3615}
3616
3617/*
3618 * This routine is currently used for internal copies of ctl_ios that need
3619 * to persist for some reason after we've already returned status to the
3620 * FETD.  (Thus the flag set.)
3621 *
3622 * XXX XXX
3623 * Note that this makes a blind copy of all fields in the ctl_io, except
3624 * for the pool reference.  This includes any memory that has been
3625 * allocated!  That memory will no longer be valid after done has been
3626 * called, so this would be VERY DANGEROUS for command that actually does
3627 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3628 * start and stop commands, which don't transfer any data, so this is not a
3629 * problem.  If it is used for anything else, the caller would also need to
3630 * allocate data buffer space and this routine would need to be modified to
3631 * copy the data buffer(s) as well.
3632 */
3633void
3634ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3635{
3636	void *pool_ref;
3637
3638	if ((src == NULL)
3639	 || (dest == NULL))
3640		return;
3641
3642	/*
3643	 * May need to preserve linked list pointers at some point too.
3644	 */
3645	pool_ref = dest->io_hdr.pool;
3646
3647	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3648
3649	dest->io_hdr.pool = pool_ref;
3650	/*
3651	 * We need to know that this is an internal copy, and doesn't need
3652	 * to get passed back to the FETD that allocated it.
3653	 */
3654	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3655}
3656
3657#ifdef NEEDTOPORT
3658static void
3659ctl_update_power_subpage(struct copan_power_subpage *page)
3660{
3661	int num_luns, num_partitions, config_type;
3662	struct ctl_softc *softc;
3663	cs_BOOL_t aor_present, shelf_50pct_power;
3664	cs_raidset_personality_t rs_type;
3665	int max_active_luns;
3666
3667	softc = control_softc;
3668
3669	/* subtract out the processor LUN */
3670	num_luns = softc->num_luns - 1;
3671	/*
3672	 * Default to 7 LUNs active, which was the only number we allowed
3673	 * in the past.
3674	 */
3675	max_active_luns = 7;
3676
3677	num_partitions = config_GetRsPartitionInfo();
3678	config_type = config_GetConfigType();
3679	shelf_50pct_power = config_GetShelfPowerMode();
3680	aor_present = config_IsAorRsPresent();
3681
3682	rs_type = ddb_GetRsRaidType(1);
3683	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3684	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3685		EPRINT(0, "Unsupported RS type %d!", rs_type);
3686	}
3687
3688
3689	page->total_luns = num_luns;
3690
3691	switch (config_type) {
3692	case 40:
3693		/*
3694		 * In a 40 drive configuration, it doesn't matter what DC
3695		 * cards we have, whether we have AOR enabled or not,
3696		 * partitioning or not, or what type of RAIDset we have.
3697		 * In that scenario, we can power up every LUN we present
3698		 * to the user.
3699		 */
3700		max_active_luns = num_luns;
3701
3702		break;
3703	case 64:
3704		if (shelf_50pct_power == CS_FALSE) {
3705			/* 25% power */
3706			if (aor_present == CS_TRUE) {
3707				if (rs_type ==
3708				     CS_RAIDSET_PERSONALITY_RAID5) {
3709					max_active_luns = 7;
3710				} else if (rs_type ==
3711					 CS_RAIDSET_PERSONALITY_RAID1){
3712					max_active_luns = 14;
3713				} else {
3714					/* XXX KDM now what?? */
3715				}
3716			} else {
3717				if (rs_type ==
3718				     CS_RAIDSET_PERSONALITY_RAID5) {
3719					max_active_luns = 8;
3720				} else if (rs_type ==
3721					 CS_RAIDSET_PERSONALITY_RAID1){
3722					max_active_luns = 16;
3723				} else {
3724					/* XXX KDM now what?? */
3725				}
3726			}
3727		} else {
3728			/* 50% power */
3729			/*
3730			 * With 50% power in a 64 drive configuration, we
3731			 * can power all LUNs we present.
3732			 */
3733			max_active_luns = num_luns;
3734		}
3735		break;
3736	case 112:
3737		if (shelf_50pct_power == CS_FALSE) {
3738			/* 25% power */
3739			if (aor_present == CS_TRUE) {
3740				if (rs_type ==
3741				     CS_RAIDSET_PERSONALITY_RAID5) {
3742					max_active_luns = 7;
3743				} else if (rs_type ==
3744					 CS_RAIDSET_PERSONALITY_RAID1){
3745					max_active_luns = 14;
3746				} else {
3747					/* XXX KDM now what?? */
3748				}
3749			} else {
3750				if (rs_type ==
3751				     CS_RAIDSET_PERSONALITY_RAID5) {
3752					max_active_luns = 8;
3753				} else if (rs_type ==
3754					 CS_RAIDSET_PERSONALITY_RAID1){
3755					max_active_luns = 16;
3756				} else {
3757					/* XXX KDM now what?? */
3758				}
3759			}
3760		} else {
3761			/* 50% power */
3762			if (aor_present == CS_TRUE) {
3763				if (rs_type ==
3764				     CS_RAIDSET_PERSONALITY_RAID5) {
3765					max_active_luns = 14;
3766				} else if (rs_type ==
3767					 CS_RAIDSET_PERSONALITY_RAID1){
3768					/*
3769					 * We're assuming here that disk
3770					 * caching is enabled, and so we're
3771					 * able to power up half of each
3772					 * LUN, and cache all writes.
3773					 */
3774					max_active_luns = num_luns;
3775				} else {
3776					/* XXX KDM now what?? */
3777				}
3778			} else {
3779				if (rs_type ==
3780				     CS_RAIDSET_PERSONALITY_RAID5) {
3781					max_active_luns = 15;
3782				} else if (rs_type ==
3783					 CS_RAIDSET_PERSONALITY_RAID1){
3784					max_active_luns = 30;
3785				} else {
3786					/* XXX KDM now what?? */
3787				}
3788			}
3789		}
3790		break;
3791	default:
3792		/*
3793		 * In this case, we have an unknown configuration, so we
3794		 * just use the default from above.
3795		 */
3796		break;
3797	}
3798
3799	page->max_active_luns = max_active_luns;
3800#if 0
3801	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3802	       page->total_luns, page->max_active_luns);
3803#endif
3804}
3805#endif /* NEEDTOPORT */
3806
3807/*
3808 * This routine could be used in the future to load default and/or saved
3809 * mode page parameters for a particuar lun.
3810 */
3811static int
3812ctl_init_page_index(struct ctl_lun *lun)
3813{
3814	int i;
3815	struct ctl_page_index *page_index;
3816	struct ctl_softc *softc;
3817
3818	memcpy(&lun->mode_pages.index, page_index_template,
3819	       sizeof(page_index_template));
3820
3821	softc = lun->ctl_softc;
3822
3823	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3824
3825		page_index = &lun->mode_pages.index[i];
3826		/*
3827		 * If this is a disk-only mode page, there's no point in
3828		 * setting it up.  For some pages, we have to have some
3829		 * basic information about the disk in order to calculate the
3830		 * mode page data.
3831		 */
3832		if ((lun->be_lun->lun_type != T_DIRECT)
3833		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3834			continue;
3835
3836		switch (page_index->page_code & SMPH_PC_MASK) {
3837		case SMS_FORMAT_DEVICE_PAGE: {
3838			struct scsi_format_page *format_page;
3839
3840			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3841				panic("subpage is incorrect!");
3842
3843			/*
3844			 * Sectors per track are set above.  Bytes per
3845			 * sector need to be set here on a per-LUN basis.
3846			 */
3847			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3848			       &format_page_default,
3849			       sizeof(format_page_default));
3850			memcpy(&lun->mode_pages.format_page[
3851			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3852			       sizeof(format_page_changeable));
3853			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3854			       &format_page_default,
3855			       sizeof(format_page_default));
3856			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3857			       &format_page_default,
3858			       sizeof(format_page_default));
3859
3860			format_page = &lun->mode_pages.format_page[
3861				CTL_PAGE_CURRENT];
3862			scsi_ulto2b(lun->be_lun->blocksize,
3863				    format_page->bytes_per_sector);
3864
3865			format_page = &lun->mode_pages.format_page[
3866				CTL_PAGE_DEFAULT];
3867			scsi_ulto2b(lun->be_lun->blocksize,
3868				    format_page->bytes_per_sector);
3869
3870			format_page = &lun->mode_pages.format_page[
3871				CTL_PAGE_SAVED];
3872			scsi_ulto2b(lun->be_lun->blocksize,
3873				    format_page->bytes_per_sector);
3874
3875			page_index->page_data =
3876				(uint8_t *)lun->mode_pages.format_page;
3877			break;
3878		}
3879		case SMS_RIGID_DISK_PAGE: {
3880			struct scsi_rigid_disk_page *rigid_disk_page;
3881			uint32_t sectors_per_cylinder;
3882			uint64_t cylinders;
3883#ifndef	__XSCALE__
3884			int shift;
3885#endif /* !__XSCALE__ */
3886
3887			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3888				panic("invalid subpage value %d",
3889				      page_index->subpage);
3890
3891			/*
3892			 * Rotation rate and sectors per track are set
3893			 * above.  We calculate the cylinders here based on
3894			 * capacity.  Due to the number of heads and
3895			 * sectors per track we're using, smaller arrays
3896			 * may turn out to have 0 cylinders.  Linux and
3897			 * FreeBSD don't pay attention to these mode pages
3898			 * to figure out capacity, but Solaris does.  It
3899			 * seems to deal with 0 cylinders just fine, and
3900			 * works out a fake geometry based on the capacity.
3901			 */
3902			memcpy(&lun->mode_pages.rigid_disk_page[
3903			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3904			       sizeof(rigid_disk_page_default));
3905			memcpy(&lun->mode_pages.rigid_disk_page[
3906			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3907			       sizeof(rigid_disk_page_changeable));
3908			memcpy(&lun->mode_pages.rigid_disk_page[
3909			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3910			       sizeof(rigid_disk_page_default));
3911			memcpy(&lun->mode_pages.rigid_disk_page[
3912			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3913			       sizeof(rigid_disk_page_default));
3914
3915			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3916				CTL_DEFAULT_HEADS;
3917
3918			/*
3919			 * The divide method here will be more accurate,
3920			 * probably, but results in floating point being
3921			 * used in the kernel on i386 (__udivdi3()).  On the
3922			 * XScale, though, __udivdi3() is implemented in
3923			 * software.
3924			 *
3925			 * The shift method for cylinder calculation is
3926			 * accurate if sectors_per_cylinder is a power of
3927			 * 2.  Otherwise it might be slightly off -- you
3928			 * might have a bit of a truncation problem.
3929			 */
3930#ifdef	__XSCALE__
3931			cylinders = (lun->be_lun->maxlba + 1) /
3932				sectors_per_cylinder;
3933#else
3934			for (shift = 31; shift > 0; shift--) {
3935				if (sectors_per_cylinder & (1 << shift))
3936					break;
3937			}
3938			cylinders = (lun->be_lun->maxlba + 1) >> shift;
3939#endif
3940
3941			/*
3942			 * We've basically got 3 bytes, or 24 bits for the
3943			 * cylinder size in the mode page.  If we're over,
3944			 * just round down to 2^24.
3945			 */
3946			if (cylinders > 0xffffff)
3947				cylinders = 0xffffff;
3948
3949			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3950				CTL_PAGE_CURRENT];
3951			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3952
3953			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3954				CTL_PAGE_DEFAULT];
3955			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3956
3957			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3958				CTL_PAGE_SAVED];
3959			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3960
3961			page_index->page_data =
3962				(uint8_t *)lun->mode_pages.rigid_disk_page;
3963			break;
3964		}
3965		case SMS_CACHING_PAGE: {
3966
3967			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3968				panic("invalid subpage value %d",
3969				      page_index->subpage);
3970			/*
3971			 * Defaults should be okay here, no calculations
3972			 * needed.
3973			 */
3974			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3975			       &caching_page_default,
3976			       sizeof(caching_page_default));
3977			memcpy(&lun->mode_pages.caching_page[
3978			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3979			       sizeof(caching_page_changeable));
3980			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3981			       &caching_page_default,
3982			       sizeof(caching_page_default));
3983			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3984			       &caching_page_default,
3985			       sizeof(caching_page_default));
3986			page_index->page_data =
3987				(uint8_t *)lun->mode_pages.caching_page;
3988			break;
3989		}
3990		case SMS_CONTROL_MODE_PAGE: {
3991
3992			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3993				panic("invalid subpage value %d",
3994				      page_index->subpage);
3995
3996			/*
3997			 * Defaults should be okay here, no calculations
3998			 * needed.
3999			 */
4000			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4001			       &control_page_default,
4002			       sizeof(control_page_default));
4003			memcpy(&lun->mode_pages.control_page[
4004			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4005			       sizeof(control_page_changeable));
4006			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4007			       &control_page_default,
4008			       sizeof(control_page_default));
4009			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4010			       &control_page_default,
4011			       sizeof(control_page_default));
4012			page_index->page_data =
4013				(uint8_t *)lun->mode_pages.control_page;
4014			break;
4015
4016		}
4017		case SMS_VENDOR_SPECIFIC_PAGE:{
4018			switch (page_index->subpage) {
4019			case PWR_SUBPAGE_CODE: {
4020				struct copan_power_subpage *current_page,
4021							   *saved_page;
4022
4023				memcpy(&lun->mode_pages.power_subpage[
4024				       CTL_PAGE_CURRENT],
4025				       &power_page_default,
4026				       sizeof(power_page_default));
4027				memcpy(&lun->mode_pages.power_subpage[
4028				       CTL_PAGE_CHANGEABLE],
4029				       &power_page_changeable,
4030				       sizeof(power_page_changeable));
4031				memcpy(&lun->mode_pages.power_subpage[
4032				       CTL_PAGE_DEFAULT],
4033				       &power_page_default,
4034				       sizeof(power_page_default));
4035				memcpy(&lun->mode_pages.power_subpage[
4036				       CTL_PAGE_SAVED],
4037				       &power_page_default,
4038				       sizeof(power_page_default));
4039				page_index->page_data =
4040				    (uint8_t *)lun->mode_pages.power_subpage;
4041
4042				current_page = (struct copan_power_subpage *)
4043					(page_index->page_data +
4044					 (page_index->page_len *
4045					  CTL_PAGE_CURRENT));
4046			        saved_page = (struct copan_power_subpage *)
4047				        (page_index->page_data +
4048					 (page_index->page_len *
4049					  CTL_PAGE_SAVED));
4050				break;
4051			}
4052			case APS_SUBPAGE_CODE: {
4053				struct copan_aps_subpage *current_page,
4054							 *saved_page;
4055
4056				// This gets set multiple times but
4057				// it should always be the same. It's
4058				// only done during init so who cares.
4059				index_to_aps_page = i;
4060
4061				memcpy(&lun->mode_pages.aps_subpage[
4062				       CTL_PAGE_CURRENT],
4063				       &aps_page_default,
4064				       sizeof(aps_page_default));
4065				memcpy(&lun->mode_pages.aps_subpage[
4066				       CTL_PAGE_CHANGEABLE],
4067				       &aps_page_changeable,
4068				       sizeof(aps_page_changeable));
4069				memcpy(&lun->mode_pages.aps_subpage[
4070				       CTL_PAGE_DEFAULT],
4071				       &aps_page_default,
4072				       sizeof(aps_page_default));
4073				memcpy(&lun->mode_pages.aps_subpage[
4074				       CTL_PAGE_SAVED],
4075				       &aps_page_default,
4076				       sizeof(aps_page_default));
4077				page_index->page_data =
4078					(uint8_t *)lun->mode_pages.aps_subpage;
4079
4080				current_page = (struct copan_aps_subpage *)
4081					(page_index->page_data +
4082					 (page_index->page_len *
4083					  CTL_PAGE_CURRENT));
4084				saved_page = (struct copan_aps_subpage *)
4085					(page_index->page_data +
4086					 (page_index->page_len *
4087					  CTL_PAGE_SAVED));
4088				break;
4089			}
4090			case DBGCNF_SUBPAGE_CODE: {
4091				struct copan_debugconf_subpage *current_page,
4092							       *saved_page;
4093
4094				memcpy(&lun->mode_pages.debugconf_subpage[
4095				       CTL_PAGE_CURRENT],
4096				       &debugconf_page_default,
4097				       sizeof(debugconf_page_default));
4098				memcpy(&lun->mode_pages.debugconf_subpage[
4099				       CTL_PAGE_CHANGEABLE],
4100				       &debugconf_page_changeable,
4101				       sizeof(debugconf_page_changeable));
4102				memcpy(&lun->mode_pages.debugconf_subpage[
4103				       CTL_PAGE_DEFAULT],
4104				       &debugconf_page_default,
4105				       sizeof(debugconf_page_default));
4106				memcpy(&lun->mode_pages.debugconf_subpage[
4107				       CTL_PAGE_SAVED],
4108				       &debugconf_page_default,
4109				       sizeof(debugconf_page_default));
4110				page_index->page_data =
4111					(uint8_t *)lun->mode_pages.debugconf_subpage;
4112
4113				current_page = (struct copan_debugconf_subpage *)
4114					(page_index->page_data +
4115					 (page_index->page_len *
4116					  CTL_PAGE_CURRENT));
4117				saved_page = (struct copan_debugconf_subpage *)
4118					(page_index->page_data +
4119					 (page_index->page_len *
4120					  CTL_PAGE_SAVED));
4121				break;
4122			}
4123			default:
4124				panic("invalid subpage value %d",
4125				      page_index->subpage);
4126				break;
4127			}
4128   			break;
4129		}
4130		default:
4131			panic("invalid page value %d",
4132			      page_index->page_code & SMPH_PC_MASK);
4133			break;
4134    	}
4135	}
4136
4137	return (CTL_RETVAL_COMPLETE);
4138}
4139
4140/*
4141 * LUN allocation.
4142 *
4143 * Requirements:
4144 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4145 *   wants us to allocate the LUN and he can block.
4146 * - ctl_softc is always set
4147 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4148 *
4149 * Returns 0 for success, non-zero (errno) for failure.
4150 */
4151static int
4152ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4153	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4154{
4155	struct ctl_lun *nlun, *lun;
4156	struct ctl_frontend *fe;
4157	int lun_number, i, lun_malloced;
4158
4159	if (be_lun == NULL)
4160		return (EINVAL);
4161
4162	/*
4163	 * We currently only support Direct Access or Processor LUN types.
4164	 */
4165	switch (be_lun->lun_type) {
4166	case T_DIRECT:
4167		break;
4168	case T_PROCESSOR:
4169		break;
4170	case T_SEQUENTIAL:
4171	case T_CHANGER:
4172	default:
4173		be_lun->lun_config_status(be_lun->be_lun,
4174					  CTL_LUN_CONFIG_FAILURE);
4175		break;
4176	}
4177	if (ctl_lun == NULL) {
4178		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4179		lun_malloced = 1;
4180	} else {
4181		lun_malloced = 0;
4182		lun = ctl_lun;
4183	}
4184
4185	memset(lun, 0, sizeof(*lun));
4186	if (lun_malloced)
4187		lun->flags = CTL_LUN_MALLOCED;
4188
4189	mtx_lock(&ctl_softc->ctl_lock);
4190	/*
4191	 * See if the caller requested a particular LUN number.  If so, see
4192	 * if it is available.  Otherwise, allocate the first available LUN.
4193	 */
4194	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4195		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4196		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4197			mtx_unlock(&ctl_softc->ctl_lock);
4198			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4199				printf("ctl: requested LUN ID %d is higher "
4200				       "than CTL_MAX_LUNS - 1 (%d)\n",
4201				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4202			} else {
4203				/*
4204				 * XXX KDM return an error, or just assign
4205				 * another LUN ID in this case??
4206				 */
4207				printf("ctl: requested LUN ID %d is already "
4208				       "in use\n", be_lun->req_lun_id);
4209			}
4210			if (lun->flags & CTL_LUN_MALLOCED)
4211				free(lun, M_CTL);
4212			be_lun->lun_config_status(be_lun->be_lun,
4213						  CTL_LUN_CONFIG_FAILURE);
4214			return (ENOSPC);
4215		}
4216		lun_number = be_lun->req_lun_id;
4217	} else {
4218		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4219		if (lun_number == -1) {
4220			mtx_unlock(&ctl_softc->ctl_lock);
4221			printf("ctl: can't allocate LUN on target %ju, out of "
4222			       "LUNs\n", (uintmax_t)target_id.id);
4223			if (lun->flags & CTL_LUN_MALLOCED)
4224				free(lun, M_CTL);
4225			be_lun->lun_config_status(be_lun->be_lun,
4226						  CTL_LUN_CONFIG_FAILURE);
4227			return (ENOSPC);
4228		}
4229	}
4230	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4231
4232	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4233	lun->target = target_id;
4234	lun->lun = lun_number;
4235	lun->be_lun = be_lun;
4236	/*
4237	 * The processor LUN is always enabled.  Disk LUNs come on line
4238	 * disabled, and must be enabled by the backend.
4239	 */
4240	lun->flags |= CTL_LUN_DISABLED;
4241	lun->backend = be_lun->be;
4242	be_lun->ctl_lun = lun;
4243	be_lun->lun_id = lun_number;
4244	atomic_add_int(&be_lun->be->num_luns, 1);
4245	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4246		lun->flags |= CTL_LUN_STOPPED;
4247
4248	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4249		lun->flags |= CTL_LUN_INOPERABLE;
4250
4251	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4252		lun->flags |= CTL_LUN_PRIMARY_SC;
4253
4254	lun->ctl_softc = ctl_softc;
4255	TAILQ_INIT(&lun->ooa_queue);
4256	TAILQ_INIT(&lun->blocked_queue);
4257	STAILQ_INIT(&lun->error_list);
4258
4259	/*
4260	 * Initialize the mode page index.
4261	 */
4262	ctl_init_page_index(lun);
4263
4264	/*
4265	 * Set the poweron UA for all initiators on this LUN only.
4266	 */
4267	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4268		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4269
4270	/*
4271	 * Now, before we insert this lun on the lun list, set the lun
4272	 * inventory changed UA for all other luns.
4273	 */
4274	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4275		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4276			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4277		}
4278	}
4279
4280	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4281
4282	ctl_softc->ctl_luns[lun_number] = lun;
4283
4284	ctl_softc->num_luns++;
4285
4286	/* Setup statistics gathering */
4287	lun->stats.device_type = be_lun->lun_type;
4288	lun->stats.lun_number = lun_number;
4289	if (lun->stats.device_type == T_DIRECT)
4290		lun->stats.blocksize = be_lun->blocksize;
4291	else
4292		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4293	for (i = 0;i < CTL_MAX_PORTS;i++)
4294		lun->stats.ports[i].targ_port = i;
4295
4296	mtx_unlock(&ctl_softc->ctl_lock);
4297
4298	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4299
4300	/*
4301	 * Run through each registered FETD and bring it online if it isn't
4302	 * already.  Enable the target ID if it hasn't been enabled, and
4303	 * enable this particular LUN.
4304	 */
4305	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4306		int retval;
4307
4308		/*
4309		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4310		 * to do things differently if we go to a multiple target
4311		 * ID scheme.
4312		 */
4313		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4314
4315			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4316			if (retval != 0) {
4317				printf("ctl_alloc_lun: FETD %s port %d "
4318				       "returned error %d for targ_enable on "
4319				       "target %ju\n", fe->port_name,
4320				       fe->targ_port, retval,
4321				       (uintmax_t)target_id.id);
4322			} else
4323				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4324		}
4325
4326		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4327		if (retval != 0) {
4328			printf("ctl_alloc_lun: FETD %s port %d returned error "
4329			       "%d for lun_enable on target %ju lun %d\n",
4330			       fe->port_name, fe->targ_port, retval,
4331			       (uintmax_t)target_id.id, lun_number);
4332		} else
4333			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4334	}
4335	return (0);
4336}
4337
4338/*
4339 * Delete a LUN.
4340 * Assumptions:
4341 * - LUN has already been marked invalid and any pending I/O has been taken
4342 *   care of.
4343 */
4344static int
4345ctl_free_lun(struct ctl_lun *lun)
4346{
4347	struct ctl_softc *softc;
4348#if 0
4349	struct ctl_frontend *fe;
4350#endif
4351	struct ctl_lun *nlun;
4352	int i;
4353
4354	softc = lun->ctl_softc;
4355
4356	mtx_assert(&softc->ctl_lock, MA_OWNED);
4357
4358	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4359
4360	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4361
4362	softc->ctl_luns[lun->lun] = NULL;
4363
4364	if (!TAILQ_EMPTY(&lun->ooa_queue))
4365		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4366
4367	softc->num_luns--;
4368
4369	/*
4370	 * XXX KDM this scheme only works for a single target/multiple LUN
4371	 * setup.  It needs to be revamped for a multiple target scheme.
4372	 *
4373	 * XXX KDM this results in fe->lun_disable() getting called twice,
4374	 * once when ctl_disable_lun() is called, and a second time here.
4375	 * We really need to re-think the LUN disable semantics.  There
4376	 * should probably be several steps/levels to LUN removal:
4377	 *  - disable
4378	 *  - invalidate
4379	 *  - free
4380 	 *
4381	 * Right now we only have a disable method when communicating to
4382	 * the front end ports, at least for individual LUNs.
4383	 */
4384#if 0
4385	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4386		int retval;
4387
4388		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4389					 lun->lun);
4390		if (retval != 0) {
4391			printf("ctl_free_lun: FETD %s port %d returned error "
4392			       "%d for lun_disable on target %ju lun %jd\n",
4393			       fe->port_name, fe->targ_port, retval,
4394			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4395		}
4396
4397		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4398			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4399
4400			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4401			if (retval != 0) {
4402				printf("ctl_free_lun: FETD %s port %d "
4403				       "returned error %d for targ_disable on "
4404				       "target %ju\n", fe->port_name,
4405				       fe->targ_port, retval,
4406				       (uintmax_t)lun->target.id);
4407			} else
4408				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4409
4410			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4411				continue;
4412
4413#if 0
4414			fe->port_offline(fe->onoff_arg);
4415			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4416#endif
4417		}
4418	}
4419#endif
4420
4421	/*
4422	 * Tell the backend to free resources, if this LUN has a backend.
4423	 */
4424	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4425	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4426
4427	mtx_destroy(&lun->lun_lock);
4428	if (lun->flags & CTL_LUN_MALLOCED)
4429		free(lun, M_CTL);
4430
4431	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4432		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4433			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4434		}
4435	}
4436
4437	return (0);
4438}
4439
4440static void
4441ctl_create_lun(struct ctl_be_lun *be_lun)
4442{
4443	struct ctl_softc *ctl_softc;
4444
4445	ctl_softc = control_softc;
4446
4447	/*
4448	 * ctl_alloc_lun() should handle all potential failure cases.
4449	 */
4450	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4451}
4452
4453int
4454ctl_add_lun(struct ctl_be_lun *be_lun)
4455{
4456	struct ctl_softc *ctl_softc = control_softc;
4457
4458	mtx_lock(&ctl_softc->ctl_lock);
4459	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4460	mtx_unlock(&ctl_softc->ctl_lock);
4461	wakeup(&ctl_softc->pending_lun_queue);
4462
4463	return (0);
4464}
4465
4466int
4467ctl_enable_lun(struct ctl_be_lun *be_lun)
4468{
4469	struct ctl_softc *ctl_softc;
4470	struct ctl_frontend *fe, *nfe;
4471	struct ctl_lun *lun;
4472	int retval;
4473
4474	ctl_softc = control_softc;
4475
4476	lun = (struct ctl_lun *)be_lun->ctl_lun;
4477
4478	mtx_lock(&ctl_softc->ctl_lock);
4479	mtx_lock(&lun->lun_lock);
4480	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4481		/*
4482		 * eh?  Why did we get called if the LUN is already
4483		 * enabled?
4484		 */
4485		mtx_unlock(&lun->lun_lock);
4486		mtx_unlock(&ctl_softc->ctl_lock);
4487		return (0);
4488	}
4489	lun->flags &= ~CTL_LUN_DISABLED;
4490	mtx_unlock(&lun->lun_lock);
4491
4492	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4493		nfe = STAILQ_NEXT(fe, links);
4494
4495		/*
4496		 * Drop the lock while we call the FETD's enable routine.
4497		 * This can lead to a callback into CTL (at least in the
4498		 * case of the internal initiator frontend.
4499		 */
4500		mtx_unlock(&ctl_softc->ctl_lock);
4501		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4502		mtx_lock(&ctl_softc->ctl_lock);
4503		if (retval != 0) {
4504			printf("%s: FETD %s port %d returned error "
4505			       "%d for lun_enable on target %ju lun %jd\n",
4506			       __func__, fe->port_name, fe->targ_port, retval,
4507			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4508		}
4509#if 0
4510		 else {
4511            /* NOTE:  TODO:  why does lun enable affect port status? */
4512			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4513		}
4514#endif
4515	}
4516
4517	mtx_unlock(&ctl_softc->ctl_lock);
4518
4519	return (0);
4520}
4521
4522int
4523ctl_disable_lun(struct ctl_be_lun *be_lun)
4524{
4525	struct ctl_softc *ctl_softc;
4526	struct ctl_frontend *fe;
4527	struct ctl_lun *lun;
4528	int retval;
4529
4530	ctl_softc = control_softc;
4531
4532	lun = (struct ctl_lun *)be_lun->ctl_lun;
4533
4534	mtx_lock(&ctl_softc->ctl_lock);
4535	mtx_lock(&lun->lun_lock);
4536	if (lun->flags & CTL_LUN_DISABLED) {
4537		mtx_unlock(&lun->lun_lock);
4538		mtx_unlock(&ctl_softc->ctl_lock);
4539		return (0);
4540	}
4541	lun->flags |= CTL_LUN_DISABLED;
4542	mtx_unlock(&lun->lun_lock);
4543
4544	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4545		mtx_unlock(&ctl_softc->ctl_lock);
4546		/*
4547		 * Drop the lock before we call the frontend's disable
4548		 * routine, to avoid lock order reversals.
4549		 *
4550		 * XXX KDM what happens if the frontend list changes while
4551		 * we're traversing it?  It's unlikely, but should be handled.
4552		 */
4553		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4554					 lun->lun);
4555		mtx_lock(&ctl_softc->ctl_lock);
4556		if (retval != 0) {
4557			printf("ctl_alloc_lun: FETD %s port %d returned error "
4558			       "%d for lun_disable on target %ju lun %jd\n",
4559			       fe->port_name, fe->targ_port, retval,
4560			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4561		}
4562	}
4563
4564	mtx_unlock(&ctl_softc->ctl_lock);
4565
4566	return (0);
4567}
4568
4569int
4570ctl_start_lun(struct ctl_be_lun *be_lun)
4571{
4572	struct ctl_softc *ctl_softc;
4573	struct ctl_lun *lun;
4574
4575	ctl_softc = control_softc;
4576
4577	lun = (struct ctl_lun *)be_lun->ctl_lun;
4578
4579	mtx_lock(&lun->lun_lock);
4580	lun->flags &= ~CTL_LUN_STOPPED;
4581	mtx_unlock(&lun->lun_lock);
4582
4583	return (0);
4584}
4585
4586int
4587ctl_stop_lun(struct ctl_be_lun *be_lun)
4588{
4589	struct ctl_softc *ctl_softc;
4590	struct ctl_lun *lun;
4591
4592	ctl_softc = control_softc;
4593
4594	lun = (struct ctl_lun *)be_lun->ctl_lun;
4595
4596	mtx_lock(&lun->lun_lock);
4597	lun->flags |= CTL_LUN_STOPPED;
4598	mtx_unlock(&lun->lun_lock);
4599
4600	return (0);
4601}
4602
4603int
4604ctl_lun_offline(struct ctl_be_lun *be_lun)
4605{
4606	struct ctl_softc *ctl_softc;
4607	struct ctl_lun *lun;
4608
4609	ctl_softc = control_softc;
4610
4611	lun = (struct ctl_lun *)be_lun->ctl_lun;
4612
4613	mtx_lock(&lun->lun_lock);
4614	lun->flags |= CTL_LUN_OFFLINE;
4615	mtx_unlock(&lun->lun_lock);
4616
4617	return (0);
4618}
4619
4620int
4621ctl_lun_online(struct ctl_be_lun *be_lun)
4622{
4623	struct ctl_softc *ctl_softc;
4624	struct ctl_lun *lun;
4625
4626	ctl_softc = control_softc;
4627
4628	lun = (struct ctl_lun *)be_lun->ctl_lun;
4629
4630	mtx_lock(&lun->lun_lock);
4631	lun->flags &= ~CTL_LUN_OFFLINE;
4632	mtx_unlock(&lun->lun_lock);
4633
4634	return (0);
4635}
4636
4637int
4638ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4639{
4640	struct ctl_softc *ctl_softc;
4641	struct ctl_lun *lun;
4642
4643	ctl_softc = control_softc;
4644
4645	lun = (struct ctl_lun *)be_lun->ctl_lun;
4646
4647	mtx_lock(&lun->lun_lock);
4648
4649	/*
4650	 * The LUN needs to be disabled before it can be marked invalid.
4651	 */
4652	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4653		mtx_unlock(&lun->lun_lock);
4654		return (-1);
4655	}
4656	/*
4657	 * Mark the LUN invalid.
4658	 */
4659	lun->flags |= CTL_LUN_INVALID;
4660
4661	/*
4662	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4663	 * If we have something in the OOA queue, we'll free it when the
4664	 * last I/O completes.
4665	 */
4666	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4667		mtx_unlock(&lun->lun_lock);
4668		mtx_lock(&ctl_softc->ctl_lock);
4669		ctl_free_lun(lun);
4670		mtx_unlock(&ctl_softc->ctl_lock);
4671	} else
4672		mtx_unlock(&lun->lun_lock);
4673
4674	return (0);
4675}
4676
4677int
4678ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4679{
4680	struct ctl_softc *ctl_softc;
4681	struct ctl_lun *lun;
4682
4683	ctl_softc = control_softc;
4684	lun = (struct ctl_lun *)be_lun->ctl_lun;
4685
4686	mtx_lock(&lun->lun_lock);
4687	lun->flags |= CTL_LUN_INOPERABLE;
4688	mtx_unlock(&lun->lun_lock);
4689
4690	return (0);
4691}
4692
4693int
4694ctl_lun_operable(struct ctl_be_lun *be_lun)
4695{
4696	struct ctl_softc *ctl_softc;
4697	struct ctl_lun *lun;
4698
4699	ctl_softc = control_softc;
4700	lun = (struct ctl_lun *)be_lun->ctl_lun;
4701
4702	mtx_lock(&lun->lun_lock);
4703	lun->flags &= ~CTL_LUN_INOPERABLE;
4704	mtx_unlock(&lun->lun_lock);
4705
4706	return (0);
4707}
4708
4709int
4710ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4711		   int lock)
4712{
4713	struct ctl_softc *softc;
4714	struct ctl_lun *lun;
4715	struct copan_aps_subpage *current_sp;
4716	struct ctl_page_index *page_index;
4717	int i;
4718
4719	softc = control_softc;
4720
4721	mtx_lock(&softc->ctl_lock);
4722
4723	lun = (struct ctl_lun *)be_lun->ctl_lun;
4724	mtx_lock(&lun->lun_lock);
4725
4726	page_index = NULL;
4727	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4728		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4729		     APS_PAGE_CODE)
4730			continue;
4731
4732		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4733			continue;
4734		page_index = &lun->mode_pages.index[i];
4735	}
4736
4737	if (page_index == NULL) {
4738		mtx_unlock(&lun->lun_lock);
4739		mtx_unlock(&softc->ctl_lock);
4740		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4741		       (uintmax_t)lun->lun);
4742		return (1);
4743	}
4744#if 0
4745	if ((softc->aps_locked_lun != 0)
4746	 && (softc->aps_locked_lun != lun->lun)) {
4747		printf("%s: attempt to lock LUN %llu when %llu is already "
4748		       "locked\n");
4749		mtx_unlock(&lun->lun_lock);
4750		mtx_unlock(&softc->ctl_lock);
4751		return (1);
4752	}
4753#endif
4754
4755	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4756		(page_index->page_len * CTL_PAGE_CURRENT));
4757
4758	if (lock != 0) {
4759		current_sp->lock_active = APS_LOCK_ACTIVE;
4760		softc->aps_locked_lun = lun->lun;
4761	} else {
4762		current_sp->lock_active = 0;
4763		softc->aps_locked_lun = 0;
4764	}
4765
4766
4767	/*
4768	 * If we're in HA mode, try to send the lock message to the other
4769	 * side.
4770	 */
4771	if (ctl_is_single == 0) {
4772		int isc_retval;
4773		union ctl_ha_msg lock_msg;
4774
4775		lock_msg.hdr.nexus = *nexus;
4776		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4777		if (lock != 0)
4778			lock_msg.aps.lock_flag = 1;
4779		else
4780			lock_msg.aps.lock_flag = 0;
4781		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4782					 sizeof(lock_msg), 0);
4783		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4784			printf("%s: APS (lock=%d) error returned from "
4785			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4786			mtx_unlock(&lun->lun_lock);
4787			mtx_unlock(&softc->ctl_lock);
4788			return (1);
4789		}
4790	}
4791
4792	mtx_unlock(&lun->lun_lock);
4793	mtx_unlock(&softc->ctl_lock);
4794
4795	return (0);
4796}
4797
4798void
4799ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4800{
4801	struct ctl_lun *lun;
4802	struct ctl_softc *softc;
4803	int i;
4804
4805	softc = control_softc;
4806
4807	lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809	mtx_lock(&lun->lun_lock);
4810
4811	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4812		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4813
4814	mtx_unlock(&lun->lun_lock);
4815}
4816
4817/*
4818 * Backend "memory move is complete" callback for requests that never
4819 * make it down to say RAIDCore's configuration code.
4820 */
4821int
4822ctl_config_move_done(union ctl_io *io)
4823{
4824	int retval;
4825
4826	retval = CTL_RETVAL_COMPLETE;
4827
4828
4829	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4830	/*
4831	 * XXX KDM this shouldn't happen, but what if it does?
4832	 */
4833	if (io->io_hdr.io_type != CTL_IO_SCSI)
4834		panic("I/O type isn't CTL_IO_SCSI!");
4835
4836	if ((io->io_hdr.port_status == 0)
4837	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4838	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4839		io->io_hdr.status = CTL_SUCCESS;
4840	else 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		/*
4844		 * For hardware error sense keys, the sense key
4845		 * specific value is defined to be a retry count,
4846		 * but we use it to pass back an internal FETD
4847		 * error code.  XXX KDM  Hopefully the FETD is only
4848		 * using 16 bits for an error code, since that's
4849		 * all the space we have in the sks field.
4850		 */
4851		ctl_set_internal_failure(&io->scsiio,
4852					 /*sks_valid*/ 1,
4853					 /*retry_count*/
4854					 io->io_hdr.port_status);
4855		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4856			free(io->scsiio.kern_data_ptr, M_CTL);
4857		ctl_done(io);
4858		goto bailout;
4859	}
4860
4861	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4862	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4863	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4864		/*
4865		 * XXX KDM just assuming a single pointer here, and not a
4866		 * S/G list.  If we start using S/G lists for config data,
4867		 * we'll need to know how to clean them up here as well.
4868		 */
4869		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4870			free(io->scsiio.kern_data_ptr, M_CTL);
4871		/* Hopefully the user has already set the status... */
4872		ctl_done(io);
4873	} else {
4874		/*
4875		 * XXX KDM now we need to continue data movement.  Some
4876		 * options:
4877		 * - call ctl_scsiio() again?  We don't do this for data
4878		 *   writes, because for those at least we know ahead of
4879		 *   time where the write will go and how long it is.  For
4880		 *   config writes, though, that information is largely
4881		 *   contained within the write itself, thus we need to
4882		 *   parse out the data again.
4883		 *
4884		 * - Call some other function once the data is in?
4885		 */
4886
4887		/*
4888		 * XXX KDM call ctl_scsiio() again for now, and check flag
4889		 * bits to see whether we're allocated or not.
4890		 */
4891		retval = ctl_scsiio(&io->scsiio);
4892	}
4893bailout:
4894	return (retval);
4895}
4896
4897/*
4898 * This gets called by a backend driver when it is done with a
4899 * data_submit method.
4900 */
4901void
4902ctl_data_submit_done(union ctl_io *io)
4903{
4904	/*
4905	 * If the IO_CONT flag is set, we need to call the supplied
4906	 * function to continue processing the I/O, instead of completing
4907	 * the I/O just yet.
4908	 *
4909	 * If there is an error, though, we don't want to keep processing.
4910	 * Instead, just send status back to the initiator.
4911	 */
4912	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4913	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4914	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4915	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4916		io->scsiio.io_cont(io);
4917		return;
4918	}
4919	ctl_done(io);
4920}
4921
4922/*
4923 * This gets called by a backend driver when it is done with a
4924 * configuration write.
4925 */
4926void
4927ctl_config_write_done(union ctl_io *io)
4928{
4929	/*
4930	 * If the IO_CONT flag is set, we need to call the supplied
4931	 * function to continue processing the I/O, instead of completing
4932	 * the I/O just yet.
4933	 *
4934	 * If there is an error, though, we don't want to keep processing.
4935	 * Instead, just send status back to the initiator.
4936	 */
4937	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
4938	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
4939	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
4940		io->scsiio.io_cont(io);
4941		return;
4942	}
4943	/*
4944	 * Since a configuration write can be done for commands that actually
4945	 * have data allocated, like write buffer, and commands that have
4946	 * no data, like start/stop unit, we need to check here.
4947	 */
4948	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
4949		free(io->scsiio.kern_data_ptr, M_CTL);
4950	ctl_done(io);
4951}
4952
4953/*
4954 * SCSI release command.
4955 */
4956int
4957ctl_scsi_release(struct ctl_scsiio *ctsio)
4958{
4959	int length, longid, thirdparty_id, resv_id;
4960	struct ctl_softc *ctl_softc;
4961	struct ctl_lun *lun;
4962
4963	length = 0;
4964	resv_id = 0;
4965
4966	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4967
4968	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4969	ctl_softc = control_softc;
4970
4971	switch (ctsio->cdb[0]) {
4972	case RELEASE: {
4973		struct scsi_release *cdb;
4974
4975		cdb = (struct scsi_release *)ctsio->cdb;
4976		if ((cdb->byte2 & 0x1f) != 0) {
4977			ctl_set_invalid_field(ctsio,
4978					      /*sks_valid*/ 1,
4979					      /*command*/ 1,
4980					      /*field*/ 1,
4981					      /*bit_valid*/ 0,
4982					      /*bit*/ 0);
4983			ctl_done((union ctl_io *)ctsio);
4984			return (CTL_RETVAL_COMPLETE);
4985		}
4986		break;
4987	}
4988	case RELEASE_10: {
4989		struct scsi_release_10 *cdb;
4990
4991		cdb = (struct scsi_release_10 *)ctsio->cdb;
4992
4993		if ((cdb->byte2 & SR10_EXTENT) != 0) {
4994			ctl_set_invalid_field(ctsio,
4995					      /*sks_valid*/ 1,
4996					      /*command*/ 1,
4997					      /*field*/ 1,
4998					      /*bit_valid*/ 1,
4999					      /*bit*/ 0);
5000			ctl_done((union ctl_io *)ctsio);
5001			return (CTL_RETVAL_COMPLETE);
5002
5003		}
5004
5005		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5006			ctl_set_invalid_field(ctsio,
5007					      /*sks_valid*/ 1,
5008					      /*command*/ 1,
5009					      /*field*/ 1,
5010					      /*bit_valid*/ 1,
5011					      /*bit*/ 4);
5012			ctl_done((union ctl_io *)ctsio);
5013			return (CTL_RETVAL_COMPLETE);
5014		}
5015
5016		if (cdb->byte2 & SR10_LONGID)
5017			longid = 1;
5018		else
5019			thirdparty_id = cdb->thirdparty_id;
5020
5021		resv_id = cdb->resv_id;
5022		length = scsi_2btoul(cdb->length);
5023		break;
5024	}
5025	}
5026
5027
5028	/*
5029	 * XXX KDM right now, we only support LUN reservation.  We don't
5030	 * support 3rd party reservations, or extent reservations, which
5031	 * might actually need the parameter list.  If we've gotten this
5032	 * far, we've got a LUN reservation.  Anything else got kicked out
5033	 * above.  So, according to SPC, ignore the length.
5034	 */
5035	length = 0;
5036
5037	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5038	 && (length > 0)) {
5039		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5040		ctsio->kern_data_len = length;
5041		ctsio->kern_total_len = length;
5042		ctsio->kern_data_resid = 0;
5043		ctsio->kern_rel_offset = 0;
5044		ctsio->kern_sg_entries = 0;
5045		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5046		ctsio->be_move_done = ctl_config_move_done;
5047		ctl_datamove((union ctl_io *)ctsio);
5048
5049		return (CTL_RETVAL_COMPLETE);
5050	}
5051
5052	if (length > 0)
5053		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5054
5055	mtx_lock(&lun->lun_lock);
5056
5057	/*
5058	 * According to SPC, it is not an error for an intiator to attempt
5059	 * to release a reservation on a LUN that isn't reserved, or that
5060	 * is reserved by another initiator.  The reservation can only be
5061	 * released, though, by the initiator who made it or by one of
5062	 * several reset type events.
5063	 */
5064	if (lun->flags & CTL_LUN_RESERVED) {
5065		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5066		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5067		 && (ctsio->io_hdr.nexus.targ_target.id ==
5068		     lun->rsv_nexus.targ_target.id)) {
5069			lun->flags &= ~CTL_LUN_RESERVED;
5070		}
5071	}
5072
5073	mtx_unlock(&lun->lun_lock);
5074
5075	ctsio->scsi_status = SCSI_STATUS_OK;
5076	ctsio->io_hdr.status = CTL_SUCCESS;
5077
5078	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5079		free(ctsio->kern_data_ptr, M_CTL);
5080		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5081	}
5082
5083	ctl_done((union ctl_io *)ctsio);
5084	return (CTL_RETVAL_COMPLETE);
5085}
5086
5087int
5088ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5089{
5090	int extent, thirdparty, longid;
5091	int resv_id, length;
5092	uint64_t thirdparty_id;
5093	struct ctl_softc *ctl_softc;
5094	struct ctl_lun *lun;
5095
5096	extent = 0;
5097	thirdparty = 0;
5098	longid = 0;
5099	resv_id = 0;
5100	length = 0;
5101	thirdparty_id = 0;
5102
5103	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5104
5105	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5106	ctl_softc = control_softc;
5107
5108	switch (ctsio->cdb[0]) {
5109	case RESERVE: {
5110		struct scsi_reserve *cdb;
5111
5112		cdb = (struct scsi_reserve *)ctsio->cdb;
5113		if ((cdb->byte2 & 0x1f) != 0) {
5114			ctl_set_invalid_field(ctsio,
5115					      /*sks_valid*/ 1,
5116					      /*command*/ 1,
5117					      /*field*/ 1,
5118					      /*bit_valid*/ 0,
5119					      /*bit*/ 0);
5120			ctl_done((union ctl_io *)ctsio);
5121			return (CTL_RETVAL_COMPLETE);
5122		}
5123		resv_id = cdb->resv_id;
5124		length = scsi_2btoul(cdb->length);
5125		break;
5126	}
5127	case RESERVE_10: {
5128		struct scsi_reserve_10 *cdb;
5129
5130		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5131
5132		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5133			ctl_set_invalid_field(ctsio,
5134					      /*sks_valid*/ 1,
5135					      /*command*/ 1,
5136					      /*field*/ 1,
5137					      /*bit_valid*/ 1,
5138					      /*bit*/ 0);
5139			ctl_done((union ctl_io *)ctsio);
5140			return (CTL_RETVAL_COMPLETE);
5141		}
5142		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5143			ctl_set_invalid_field(ctsio,
5144					      /*sks_valid*/ 1,
5145					      /*command*/ 1,
5146					      /*field*/ 1,
5147					      /*bit_valid*/ 1,
5148					      /*bit*/ 4);
5149			ctl_done((union ctl_io *)ctsio);
5150			return (CTL_RETVAL_COMPLETE);
5151		}
5152		if (cdb->byte2 & SR10_LONGID)
5153			longid = 1;
5154		else
5155			thirdparty_id = cdb->thirdparty_id;
5156
5157		resv_id = cdb->resv_id;
5158		length = scsi_2btoul(cdb->length);
5159		break;
5160	}
5161	}
5162
5163	/*
5164	 * XXX KDM right now, we only support LUN reservation.  We don't
5165	 * support 3rd party reservations, or extent reservations, which
5166	 * might actually need the parameter list.  If we've gotten this
5167	 * far, we've got a LUN reservation.  Anything else got kicked out
5168	 * above.  So, according to SPC, ignore the length.
5169	 */
5170	length = 0;
5171
5172	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5173	 && (length > 0)) {
5174		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5175		ctsio->kern_data_len = length;
5176		ctsio->kern_total_len = length;
5177		ctsio->kern_data_resid = 0;
5178		ctsio->kern_rel_offset = 0;
5179		ctsio->kern_sg_entries = 0;
5180		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5181		ctsio->be_move_done = ctl_config_move_done;
5182		ctl_datamove((union ctl_io *)ctsio);
5183
5184		return (CTL_RETVAL_COMPLETE);
5185	}
5186
5187	if (length > 0)
5188		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5189
5190	mtx_lock(&lun->lun_lock);
5191	if (lun->flags & CTL_LUN_RESERVED) {
5192		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5193		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5194		 || (ctsio->io_hdr.nexus.targ_target.id !=
5195		     lun->rsv_nexus.targ_target.id)) {
5196			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5197			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5198			goto bailout;
5199		}
5200	}
5201
5202	lun->flags |= CTL_LUN_RESERVED;
5203	lun->rsv_nexus = ctsio->io_hdr.nexus;
5204
5205	ctsio->scsi_status = SCSI_STATUS_OK;
5206	ctsio->io_hdr.status = CTL_SUCCESS;
5207
5208bailout:
5209	mtx_unlock(&lun->lun_lock);
5210
5211	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5212		free(ctsio->kern_data_ptr, M_CTL);
5213		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5214	}
5215
5216	ctl_done((union ctl_io *)ctsio);
5217	return (CTL_RETVAL_COMPLETE);
5218}
5219
5220int
5221ctl_start_stop(struct ctl_scsiio *ctsio)
5222{
5223	struct scsi_start_stop_unit *cdb;
5224	struct ctl_lun *lun;
5225	struct ctl_softc *ctl_softc;
5226	int retval;
5227
5228	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5229
5230	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5231	ctl_softc = control_softc;
5232	retval = 0;
5233
5234	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5235
5236	/*
5237	 * XXX KDM
5238	 * We don't support the immediate bit on a stop unit.  In order to
5239	 * do that, we would need to code up a way to know that a stop is
5240	 * pending, and hold off any new commands until it completes, one
5241	 * way or another.  Then we could accept or reject those commands
5242	 * depending on its status.  We would almost need to do the reverse
5243	 * of what we do below for an immediate start -- return the copy of
5244	 * the ctl_io to the FETD with status to send to the host (and to
5245	 * free the copy!) and then free the original I/O once the stop
5246	 * actually completes.  That way, the OOA queue mechanism can work
5247	 * to block commands that shouldn't proceed.  Another alternative
5248	 * would be to put the copy in the queue in place of the original,
5249	 * and return the original back to the caller.  That could be
5250	 * slightly safer..
5251	 */
5252	if ((cdb->byte2 & SSS_IMMED)
5253	 && ((cdb->how & SSS_START) == 0)) {
5254		ctl_set_invalid_field(ctsio,
5255				      /*sks_valid*/ 1,
5256				      /*command*/ 1,
5257				      /*field*/ 1,
5258				      /*bit_valid*/ 1,
5259				      /*bit*/ 0);
5260		ctl_done((union ctl_io *)ctsio);
5261		return (CTL_RETVAL_COMPLETE);
5262	}
5263
5264	/*
5265	 * We don't support the power conditions field.  We need to check
5266	 * this prior to checking the load/eject and start/stop bits.
5267	 */
5268	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5269		ctl_set_invalid_field(ctsio,
5270				      /*sks_valid*/ 1,
5271				      /*command*/ 1,
5272				      /*field*/ 4,
5273				      /*bit_valid*/ 1,
5274				      /*bit*/ 4);
5275		ctl_done((union ctl_io *)ctsio);
5276		return (CTL_RETVAL_COMPLETE);
5277	}
5278
5279	/*
5280	 * Media isn't removable, so we can't load or eject it.
5281	 */
5282	if ((cdb->how & SSS_LOEJ) != 0) {
5283		ctl_set_invalid_field(ctsio,
5284				      /*sks_valid*/ 1,
5285				      /*command*/ 1,
5286				      /*field*/ 4,
5287				      /*bit_valid*/ 1,
5288				      /*bit*/ 1);
5289		ctl_done((union ctl_io *)ctsio);
5290		return (CTL_RETVAL_COMPLETE);
5291	}
5292
5293	if ((lun->flags & CTL_LUN_PR_RESERVED)
5294	 && ((cdb->how & SSS_START)==0)) {
5295		uint32_t residx;
5296
5297		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5298		if (!lun->per_res[residx].registered
5299		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5300
5301			ctl_set_reservation_conflict(ctsio);
5302			ctl_done((union ctl_io *)ctsio);
5303			return (CTL_RETVAL_COMPLETE);
5304		}
5305	}
5306
5307	/*
5308	 * If there is no backend on this device, we can't start or stop
5309	 * it.  In theory we shouldn't get any start/stop commands in the
5310	 * first place at this level if the LUN doesn't have a backend.
5311	 * That should get stopped by the command decode code.
5312	 */
5313	if (lun->backend == NULL) {
5314		ctl_set_invalid_opcode(ctsio);
5315		ctl_done((union ctl_io *)ctsio);
5316		return (CTL_RETVAL_COMPLETE);
5317	}
5318
5319	/*
5320	 * XXX KDM Copan-specific offline behavior.
5321	 * Figure out a reasonable way to port this?
5322	 */
5323#ifdef NEEDTOPORT
5324	mtx_lock(&lun->lun_lock);
5325
5326	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5327	 && (lun->flags & CTL_LUN_OFFLINE)) {
5328		/*
5329		 * If the LUN is offline, and the on/offline bit isn't set,
5330		 * reject the start or stop.  Otherwise, let it through.
5331		 */
5332		mtx_unlock(&lun->lun_lock);
5333		ctl_set_lun_not_ready(ctsio);
5334		ctl_done((union ctl_io *)ctsio);
5335	} else {
5336		mtx_unlock(&lun->lun_lock);
5337#endif /* NEEDTOPORT */
5338		/*
5339		 * This could be a start or a stop when we're online,
5340		 * or a stop/offline or start/online.  A start or stop when
5341		 * we're offline is covered in the case above.
5342		 */
5343		/*
5344		 * In the non-immediate case, we send the request to
5345		 * the backend and return status to the user when
5346		 * it is done.
5347		 *
5348		 * In the immediate case, we allocate a new ctl_io
5349		 * to hold a copy of the request, and send that to
5350		 * the backend.  We then set good status on the
5351		 * user's request and return it immediately.
5352		 */
5353		if (cdb->byte2 & SSS_IMMED) {
5354			union ctl_io *new_io;
5355
5356			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5357			if (new_io == NULL) {
5358				ctl_set_busy(ctsio);
5359				ctl_done((union ctl_io *)ctsio);
5360			} else {
5361				ctl_copy_io((union ctl_io *)ctsio,
5362					    new_io);
5363				retval = lun->backend->config_write(new_io);
5364				ctl_set_success(ctsio);
5365				ctl_done((union ctl_io *)ctsio);
5366			}
5367		} else {
5368			retval = lun->backend->config_write(
5369				(union ctl_io *)ctsio);
5370		}
5371#ifdef NEEDTOPORT
5372	}
5373#endif
5374	return (retval);
5375}
5376
5377/*
5378 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5379 * we don't really do anything with the LBA and length fields if the user
5380 * passes them in.  Instead we'll just flush out the cache for the entire
5381 * LUN.
5382 */
5383int
5384ctl_sync_cache(struct ctl_scsiio *ctsio)
5385{
5386	struct ctl_lun *lun;
5387	struct ctl_softc *ctl_softc;
5388	uint64_t starting_lba;
5389	uint32_t block_count;
5390	int reladr, immed;
5391	int retval;
5392
5393	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5394
5395	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5396	ctl_softc = control_softc;
5397	retval = 0;
5398	reladr = 0;
5399	immed = 0;
5400
5401	switch (ctsio->cdb[0]) {
5402	case SYNCHRONIZE_CACHE: {
5403		struct scsi_sync_cache *cdb;
5404		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5405
5406		if (cdb->byte2 & SSC_RELADR)
5407			reladr = 1;
5408
5409		if (cdb->byte2 & SSC_IMMED)
5410			immed = 1;
5411
5412		starting_lba = scsi_4btoul(cdb->begin_lba);
5413		block_count = scsi_2btoul(cdb->lb_count);
5414		break;
5415	}
5416	case SYNCHRONIZE_CACHE_16: {
5417		struct scsi_sync_cache_16 *cdb;
5418		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5419
5420		if (cdb->byte2 & SSC_RELADR)
5421			reladr = 1;
5422
5423		if (cdb->byte2 & SSC_IMMED)
5424			immed = 1;
5425
5426		starting_lba = scsi_8btou64(cdb->begin_lba);
5427		block_count = scsi_4btoul(cdb->lb_count);
5428		break;
5429	}
5430	default:
5431		ctl_set_invalid_opcode(ctsio);
5432		ctl_done((union ctl_io *)ctsio);
5433		goto bailout;
5434		break; /* NOTREACHED */
5435	}
5436
5437	if (immed) {
5438		/*
5439		 * We don't support the immediate bit.  Since it's in the
5440		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5441		 * commands, we can just return the same error in either
5442		 * case.
5443		 */
5444		ctl_set_invalid_field(ctsio,
5445				      /*sks_valid*/ 1,
5446				      /*command*/ 1,
5447				      /*field*/ 1,
5448				      /*bit_valid*/ 1,
5449				      /*bit*/ 1);
5450		ctl_done((union ctl_io *)ctsio);
5451		goto bailout;
5452	}
5453
5454	if (reladr) {
5455		/*
5456		 * We don't support the reladr bit either.  It can only be
5457		 * used with linked commands, and we don't support linked
5458		 * commands.  Since the bit is in the same place for the
5459		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5460		 * just return the same error in either case.
5461		 */
5462		ctl_set_invalid_field(ctsio,
5463				      /*sks_valid*/ 1,
5464				      /*command*/ 1,
5465				      /*field*/ 1,
5466				      /*bit_valid*/ 1,
5467				      /*bit*/ 0);
5468		ctl_done((union ctl_io *)ctsio);
5469		goto bailout;
5470	}
5471
5472	/*
5473	 * We check the LBA and length, but don't do anything with them.
5474	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5475	 * get flushed.  This check will just help satisfy anyone who wants
5476	 * to see an error for an out of range LBA.
5477	 */
5478	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5479		ctl_set_lba_out_of_range(ctsio);
5480		ctl_done((union ctl_io *)ctsio);
5481		goto bailout;
5482	}
5483
5484	/*
5485	 * If this LUN has no backend, we can't flush the cache anyway.
5486	 */
5487	if (lun->backend == NULL) {
5488		ctl_set_invalid_opcode(ctsio);
5489		ctl_done((union ctl_io *)ctsio);
5490		goto bailout;
5491	}
5492
5493	/*
5494	 * Check to see whether we're configured to send the SYNCHRONIZE
5495	 * CACHE command directly to the back end.
5496	 */
5497	mtx_lock(&lun->lun_lock);
5498	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5499	 && (++(lun->sync_count) >= lun->sync_interval)) {
5500		lun->sync_count = 0;
5501		mtx_unlock(&lun->lun_lock);
5502		retval = lun->backend->config_write((union ctl_io *)ctsio);
5503	} else {
5504		mtx_unlock(&lun->lun_lock);
5505		ctl_set_success(ctsio);
5506		ctl_done((union ctl_io *)ctsio);
5507	}
5508
5509bailout:
5510
5511	return (retval);
5512}
5513
5514int
5515ctl_format(struct ctl_scsiio *ctsio)
5516{
5517	struct scsi_format *cdb;
5518	struct ctl_lun *lun;
5519	struct ctl_softc *ctl_softc;
5520	int length, defect_list_len;
5521
5522	CTL_DEBUG_PRINT(("ctl_format\n"));
5523
5524	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5525	ctl_softc = control_softc;
5526
5527	cdb = (struct scsi_format *)ctsio->cdb;
5528
5529	length = 0;
5530	if (cdb->byte2 & SF_FMTDATA) {
5531		if (cdb->byte2 & SF_LONGLIST)
5532			length = sizeof(struct scsi_format_header_long);
5533		else
5534			length = sizeof(struct scsi_format_header_short);
5535	}
5536
5537	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5538	 && (length > 0)) {
5539		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5540		ctsio->kern_data_len = length;
5541		ctsio->kern_total_len = length;
5542		ctsio->kern_data_resid = 0;
5543		ctsio->kern_rel_offset = 0;
5544		ctsio->kern_sg_entries = 0;
5545		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5546		ctsio->be_move_done = ctl_config_move_done;
5547		ctl_datamove((union ctl_io *)ctsio);
5548
5549		return (CTL_RETVAL_COMPLETE);
5550	}
5551
5552	defect_list_len = 0;
5553
5554	if (cdb->byte2 & SF_FMTDATA) {
5555		if (cdb->byte2 & SF_LONGLIST) {
5556			struct scsi_format_header_long *header;
5557
5558			header = (struct scsi_format_header_long *)
5559				ctsio->kern_data_ptr;
5560
5561			defect_list_len = scsi_4btoul(header->defect_list_len);
5562			if (defect_list_len != 0) {
5563				ctl_set_invalid_field(ctsio,
5564						      /*sks_valid*/ 1,
5565						      /*command*/ 0,
5566						      /*field*/ 2,
5567						      /*bit_valid*/ 0,
5568						      /*bit*/ 0);
5569				goto bailout;
5570			}
5571		} else {
5572			struct scsi_format_header_short *header;
5573
5574			header = (struct scsi_format_header_short *)
5575				ctsio->kern_data_ptr;
5576
5577			defect_list_len = scsi_2btoul(header->defect_list_len);
5578			if (defect_list_len != 0) {
5579				ctl_set_invalid_field(ctsio,
5580						      /*sks_valid*/ 1,
5581						      /*command*/ 0,
5582						      /*field*/ 2,
5583						      /*bit_valid*/ 0,
5584						      /*bit*/ 0);
5585				goto bailout;
5586			}
5587		}
5588	}
5589
5590	/*
5591	 * The format command will clear out the "Medium format corrupted"
5592	 * status if set by the configuration code.  That status is really
5593	 * just a way to notify the host that we have lost the media, and
5594	 * get them to issue a command that will basically make them think
5595	 * they're blowing away the media.
5596	 */
5597	mtx_lock(&lun->lun_lock);
5598	lun->flags &= ~CTL_LUN_INOPERABLE;
5599	mtx_unlock(&lun->lun_lock);
5600
5601	ctsio->scsi_status = SCSI_STATUS_OK;
5602	ctsio->io_hdr.status = CTL_SUCCESS;
5603bailout:
5604
5605	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5606		free(ctsio->kern_data_ptr, M_CTL);
5607		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5608	}
5609
5610	ctl_done((union ctl_io *)ctsio);
5611	return (CTL_RETVAL_COMPLETE);
5612}
5613
5614int
5615ctl_read_buffer(struct ctl_scsiio *ctsio)
5616{
5617	struct scsi_read_buffer *cdb;
5618	struct ctl_lun *lun;
5619	int buffer_offset, len;
5620	static uint8_t descr[4];
5621	static uint8_t echo_descr[4] = { 0 };
5622
5623	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5624
5625	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5626	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5627
5628	if (lun->flags & CTL_LUN_PR_RESERVED) {
5629		uint32_t residx;
5630
5631		/*
5632		 * XXX KDM need a lock here.
5633		 */
5634		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5635		if ((lun->res_type == SPR_TYPE_EX_AC
5636		  && residx != lun->pr_res_idx)
5637		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5638		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5639		  && !lun->per_res[residx].registered)) {
5640			ctl_set_reservation_conflict(ctsio);
5641			ctl_done((union ctl_io *)ctsio);
5642			return (CTL_RETVAL_COMPLETE);
5643	        }
5644	}
5645
5646	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5647	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5648	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5649		ctl_set_invalid_field(ctsio,
5650				      /*sks_valid*/ 1,
5651				      /*command*/ 1,
5652				      /*field*/ 1,
5653				      /*bit_valid*/ 1,
5654				      /*bit*/ 4);
5655		ctl_done((union ctl_io *)ctsio);
5656		return (CTL_RETVAL_COMPLETE);
5657	}
5658	if (cdb->buffer_id != 0) {
5659		ctl_set_invalid_field(ctsio,
5660				      /*sks_valid*/ 1,
5661				      /*command*/ 1,
5662				      /*field*/ 2,
5663				      /*bit_valid*/ 0,
5664				      /*bit*/ 0);
5665		ctl_done((union ctl_io *)ctsio);
5666		return (CTL_RETVAL_COMPLETE);
5667	}
5668
5669	len = scsi_3btoul(cdb->length);
5670	buffer_offset = scsi_3btoul(cdb->offset);
5671
5672	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5673		ctl_set_invalid_field(ctsio,
5674				      /*sks_valid*/ 1,
5675				      /*command*/ 1,
5676				      /*field*/ 6,
5677				      /*bit_valid*/ 0,
5678				      /*bit*/ 0);
5679		ctl_done((union ctl_io *)ctsio);
5680		return (CTL_RETVAL_COMPLETE);
5681	}
5682
5683	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5684		descr[0] = 0;
5685		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5686		ctsio->kern_data_ptr = descr;
5687		len = min(len, sizeof(descr));
5688	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5689		ctsio->kern_data_ptr = echo_descr;
5690		len = min(len, sizeof(echo_descr));
5691	} else
5692		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5693	ctsio->kern_data_len = len;
5694	ctsio->kern_total_len = len;
5695	ctsio->kern_data_resid = 0;
5696	ctsio->kern_rel_offset = 0;
5697	ctsio->kern_sg_entries = 0;
5698	ctsio->be_move_done = ctl_config_move_done;
5699	ctl_datamove((union ctl_io *)ctsio);
5700
5701	return (CTL_RETVAL_COMPLETE);
5702}
5703
5704int
5705ctl_write_buffer(struct ctl_scsiio *ctsio)
5706{
5707	struct scsi_write_buffer *cdb;
5708	struct ctl_lun *lun;
5709	int buffer_offset, len;
5710
5711	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5712
5713	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5714	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5715
5716	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5717		ctl_set_invalid_field(ctsio,
5718				      /*sks_valid*/ 1,
5719				      /*command*/ 1,
5720				      /*field*/ 1,
5721				      /*bit_valid*/ 1,
5722				      /*bit*/ 4);
5723		ctl_done((union ctl_io *)ctsio);
5724		return (CTL_RETVAL_COMPLETE);
5725	}
5726	if (cdb->buffer_id != 0) {
5727		ctl_set_invalid_field(ctsio,
5728				      /*sks_valid*/ 1,
5729				      /*command*/ 1,
5730				      /*field*/ 2,
5731				      /*bit_valid*/ 0,
5732				      /*bit*/ 0);
5733		ctl_done((union ctl_io *)ctsio);
5734		return (CTL_RETVAL_COMPLETE);
5735	}
5736
5737	len = scsi_3btoul(cdb->length);
5738	buffer_offset = scsi_3btoul(cdb->offset);
5739
5740	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5741		ctl_set_invalid_field(ctsio,
5742				      /*sks_valid*/ 1,
5743				      /*command*/ 1,
5744				      /*field*/ 6,
5745				      /*bit_valid*/ 0,
5746				      /*bit*/ 0);
5747		ctl_done((union ctl_io *)ctsio);
5748		return (CTL_RETVAL_COMPLETE);
5749	}
5750
5751	/*
5752	 * If we've got a kernel request that hasn't been malloced yet,
5753	 * malloc it and tell the caller the data buffer is here.
5754	 */
5755	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5756		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5757		ctsio->kern_data_len = len;
5758		ctsio->kern_total_len = len;
5759		ctsio->kern_data_resid = 0;
5760		ctsio->kern_rel_offset = 0;
5761		ctsio->kern_sg_entries = 0;
5762		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5763		ctsio->be_move_done = ctl_config_move_done;
5764		ctl_datamove((union ctl_io *)ctsio);
5765
5766		return (CTL_RETVAL_COMPLETE);
5767	}
5768
5769	ctl_done((union ctl_io *)ctsio);
5770
5771	return (CTL_RETVAL_COMPLETE);
5772}
5773
5774int
5775ctl_write_same(struct ctl_scsiio *ctsio)
5776{
5777	struct ctl_lun *lun;
5778	struct ctl_lba_len_flags *lbalen;
5779	uint64_t lba;
5780	uint32_t num_blocks;
5781	int len, retval;
5782	uint8_t byte2;
5783
5784	retval = CTL_RETVAL_COMPLETE;
5785
5786	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5787
5788	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5789
5790	switch (ctsio->cdb[0]) {
5791	case WRITE_SAME_10: {
5792		struct scsi_write_same_10 *cdb;
5793
5794		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5795
5796		lba = scsi_4btoul(cdb->addr);
5797		num_blocks = scsi_2btoul(cdb->length);
5798		byte2 = cdb->byte2;
5799		break;
5800	}
5801	case WRITE_SAME_16: {
5802		struct scsi_write_same_16 *cdb;
5803
5804		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5805
5806		lba = scsi_8btou64(cdb->addr);
5807		num_blocks = scsi_4btoul(cdb->length);
5808		byte2 = cdb->byte2;
5809		break;
5810	}
5811	default:
5812		/*
5813		 * We got a command we don't support.  This shouldn't
5814		 * happen, commands should be filtered out above us.
5815		 */
5816		ctl_set_invalid_opcode(ctsio);
5817		ctl_done((union ctl_io *)ctsio);
5818
5819		return (CTL_RETVAL_COMPLETE);
5820		break; /* NOTREACHED */
5821	}
5822
5823	/*
5824	 * The first check is to make sure we're in bounds, the second
5825	 * check is to catch wrap-around problems.  If the lba + num blocks
5826	 * is less than the lba, then we've wrapped around and the block
5827	 * range is invalid anyway.
5828	 */
5829	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5830	 || ((lba + num_blocks) < lba)) {
5831		ctl_set_lba_out_of_range(ctsio);
5832		ctl_done((union ctl_io *)ctsio);
5833		return (CTL_RETVAL_COMPLETE);
5834	}
5835
5836	/* Zero number of blocks means "to the last logical block" */
5837	if (num_blocks == 0) {
5838		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5839			ctl_set_invalid_field(ctsio,
5840					      /*sks_valid*/ 0,
5841					      /*command*/ 1,
5842					      /*field*/ 0,
5843					      /*bit_valid*/ 0,
5844					      /*bit*/ 0);
5845			ctl_done((union ctl_io *)ctsio);
5846			return (CTL_RETVAL_COMPLETE);
5847		}
5848		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5849	}
5850
5851	len = lun->be_lun->blocksize;
5852
5853	/*
5854	 * If we've got a kernel request that hasn't been malloced yet,
5855	 * malloc it and tell the caller the data buffer is here.
5856	 */
5857	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5858		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5859		ctsio->kern_data_len = len;
5860		ctsio->kern_total_len = len;
5861		ctsio->kern_data_resid = 0;
5862		ctsio->kern_rel_offset = 0;
5863		ctsio->kern_sg_entries = 0;
5864		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5865		ctsio->be_move_done = ctl_config_move_done;
5866		ctl_datamove((union ctl_io *)ctsio);
5867
5868		return (CTL_RETVAL_COMPLETE);
5869	}
5870
5871	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5872	lbalen->lba = lba;
5873	lbalen->len = num_blocks;
5874	lbalen->flags = byte2;
5875	retval = lun->backend->config_write((union ctl_io *)ctsio);
5876
5877	return (retval);
5878}
5879
5880int
5881ctl_unmap(struct ctl_scsiio *ctsio)
5882{
5883	struct ctl_lun *lun;
5884	struct scsi_unmap *cdb;
5885	struct ctl_ptr_len_flags *ptrlen;
5886	struct scsi_unmap_header *hdr;
5887	struct scsi_unmap_desc *buf, *end;
5888	uint64_t lba;
5889	uint32_t num_blocks;
5890	int len, retval;
5891	uint8_t byte2;
5892
5893	retval = CTL_RETVAL_COMPLETE;
5894
5895	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5896
5897	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5898	cdb = (struct scsi_unmap *)ctsio->cdb;
5899
5900	len = scsi_2btoul(cdb->length);
5901	byte2 = cdb->byte2;
5902
5903	/*
5904	 * If we've got a kernel request that hasn't been malloced yet,
5905	 * malloc it and tell the caller the data buffer is here.
5906	 */
5907	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5908		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5909		ctsio->kern_data_len = len;
5910		ctsio->kern_total_len = len;
5911		ctsio->kern_data_resid = 0;
5912		ctsio->kern_rel_offset = 0;
5913		ctsio->kern_sg_entries = 0;
5914		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5915		ctsio->be_move_done = ctl_config_move_done;
5916		ctl_datamove((union ctl_io *)ctsio);
5917
5918		return (CTL_RETVAL_COMPLETE);
5919	}
5920
5921	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5922	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5923	if (len < sizeof (*hdr) ||
5924	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5925	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5926	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5927		ctl_set_invalid_field(ctsio,
5928				      /*sks_valid*/ 0,
5929				      /*command*/ 0,
5930				      /*field*/ 0,
5931				      /*bit_valid*/ 0,
5932				      /*bit*/ 0);
5933		ctl_done((union ctl_io *)ctsio);
5934		return (CTL_RETVAL_COMPLETE);
5935	}
5936	len = scsi_2btoul(hdr->desc_length);
5937	buf = (struct scsi_unmap_desc *)(hdr + 1);
5938	end = buf + len / sizeof(*buf);
5939
5940	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5941	ptrlen->ptr = (void *)buf;
5942	ptrlen->len = len;
5943	ptrlen->flags = byte2;
5944
5945	for (; buf < end; buf++) {
5946		lba = scsi_8btou64(buf->lba);
5947		num_blocks = scsi_4btoul(buf->length);
5948		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5949		 || ((lba + num_blocks) < lba)) {
5950			ctl_set_lba_out_of_range(ctsio);
5951			ctl_done((union ctl_io *)ctsio);
5952			return (CTL_RETVAL_COMPLETE);
5953		}
5954	}
5955
5956	retval = lun->backend->config_write((union ctl_io *)ctsio);
5957
5958	return (retval);
5959}
5960
5961/*
5962 * Note that this function currently doesn't actually do anything inside
5963 * CTL to enforce things if the DQue bit is turned on.
5964 *
5965 * Also note that this function can't be used in the default case, because
5966 * the DQue bit isn't set in the changeable mask for the control mode page
5967 * anyway.  This is just here as an example for how to implement a page
5968 * handler, and a placeholder in case we want to allow the user to turn
5969 * tagged queueing on and off.
5970 *
5971 * The D_SENSE bit handling is functional, however, and will turn
5972 * descriptor sense on and off for a given LUN.
5973 */
5974int
5975ctl_control_page_handler(struct ctl_scsiio *ctsio,
5976			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5977{
5978	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5979	struct ctl_lun *lun;
5980	struct ctl_softc *softc;
5981	int set_ua;
5982	uint32_t initidx;
5983
5984	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5985	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5986	set_ua = 0;
5987
5988	user_cp = (struct scsi_control_page *)page_ptr;
5989	current_cp = (struct scsi_control_page *)
5990		(page_index->page_data + (page_index->page_len *
5991		CTL_PAGE_CURRENT));
5992	saved_cp = (struct scsi_control_page *)
5993		(page_index->page_data + (page_index->page_len *
5994		CTL_PAGE_SAVED));
5995
5996	softc = control_softc;
5997
5998	mtx_lock(&lun->lun_lock);
5999	if (((current_cp->rlec & SCP_DSENSE) == 0)
6000	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6001		/*
6002		 * Descriptor sense is currently turned off and the user
6003		 * wants to turn it on.
6004		 */
6005		current_cp->rlec |= SCP_DSENSE;
6006		saved_cp->rlec |= SCP_DSENSE;
6007		lun->flags |= CTL_LUN_SENSE_DESC;
6008		set_ua = 1;
6009	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6010		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6011		/*
6012		 * Descriptor sense is currently turned on, and the user
6013		 * wants to turn it off.
6014		 */
6015		current_cp->rlec &= ~SCP_DSENSE;
6016		saved_cp->rlec &= ~SCP_DSENSE;
6017		lun->flags &= ~CTL_LUN_SENSE_DESC;
6018		set_ua = 1;
6019	}
6020	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6021		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6022#ifdef NEEDTOPORT
6023			csevent_log(CSC_CTL | CSC_SHELF_SW |
6024				    CTL_UNTAG_TO_UNTAG,
6025				    csevent_LogType_Trace,
6026				    csevent_Severity_Information,
6027				    csevent_AlertLevel_Green,
6028				    csevent_FRU_Firmware,
6029				    csevent_FRU_Unknown,
6030				    "Received untagged to untagged transition");
6031#endif /* NEEDTOPORT */
6032		} else {
6033#ifdef NEEDTOPORT
6034			csevent_log(CSC_CTL | CSC_SHELF_SW |
6035				    CTL_UNTAG_TO_TAG,
6036				    csevent_LogType_ConfigChange,
6037				    csevent_Severity_Information,
6038				    csevent_AlertLevel_Green,
6039				    csevent_FRU_Firmware,
6040				    csevent_FRU_Unknown,
6041				    "Received untagged to tagged "
6042				    "queueing transition");
6043#endif /* NEEDTOPORT */
6044
6045			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6046			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6047			set_ua = 1;
6048		}
6049	} else {
6050		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6051#ifdef NEEDTOPORT
6052			csevent_log(CSC_CTL | CSC_SHELF_SW |
6053				    CTL_TAG_TO_UNTAG,
6054				    csevent_LogType_ConfigChange,
6055				    csevent_Severity_Warning,
6056				    csevent_AlertLevel_Yellow,
6057				    csevent_FRU_Firmware,
6058				    csevent_FRU_Unknown,
6059				    "Received tagged queueing to untagged "
6060				    "transition");
6061#endif /* NEEDTOPORT */
6062
6063			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6064			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6065			set_ua = 1;
6066		} else {
6067#ifdef NEEDTOPORT
6068			csevent_log(CSC_CTL | CSC_SHELF_SW |
6069				    CTL_TAG_TO_TAG,
6070				    csevent_LogType_Trace,
6071				    csevent_Severity_Information,
6072				    csevent_AlertLevel_Green,
6073				    csevent_FRU_Firmware,
6074				    csevent_FRU_Unknown,
6075				    "Received tagged queueing to tagged "
6076				    "queueing transition");
6077#endif /* NEEDTOPORT */
6078		}
6079	}
6080	if (set_ua != 0) {
6081		int i;
6082		/*
6083		 * Let other initiators know that the mode
6084		 * parameters for this LUN have changed.
6085		 */
6086		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6087			if (i == initidx)
6088				continue;
6089
6090			lun->pending_sense[i].ua_pending |=
6091				CTL_UA_MODE_CHANGE;
6092		}
6093	}
6094	mtx_unlock(&lun->lun_lock);
6095
6096	return (0);
6097}
6098
6099int
6100ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6101		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6102{
6103	return (0);
6104}
6105
6106int
6107ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6108			   struct ctl_page_index *page_index, int pc)
6109{
6110	struct copan_power_subpage *page;
6111
6112	page = (struct copan_power_subpage *)page_index->page_data +
6113		(page_index->page_len * pc);
6114
6115	switch (pc) {
6116	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6117		/*
6118		 * We don't update the changable bits for this page.
6119		 */
6120		break;
6121	case SMS_PAGE_CTRL_CURRENT >> 6:
6122	case SMS_PAGE_CTRL_DEFAULT >> 6:
6123	case SMS_PAGE_CTRL_SAVED >> 6:
6124#ifdef NEEDTOPORT
6125		ctl_update_power_subpage(page);
6126#endif
6127		break;
6128	default:
6129#ifdef NEEDTOPORT
6130		EPRINT(0, "Invalid PC %d!!", pc);
6131#endif
6132		break;
6133	}
6134	return (0);
6135}
6136
6137
6138int
6139ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6140		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6141{
6142	struct copan_aps_subpage *user_sp;
6143	struct copan_aps_subpage *current_sp;
6144	union ctl_modepage_info *modepage_info;
6145	struct ctl_softc *softc;
6146	struct ctl_lun *lun;
6147	int retval;
6148
6149	retval = CTL_RETVAL_COMPLETE;
6150	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6151		     (page_index->page_len * CTL_PAGE_CURRENT));
6152	softc = control_softc;
6153	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6154
6155	user_sp = (struct copan_aps_subpage *)page_ptr;
6156
6157	modepage_info = (union ctl_modepage_info *)
6158		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6159
6160	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6161	modepage_info->header.subpage = page_index->subpage;
6162	modepage_info->aps.lock_active = user_sp->lock_active;
6163
6164	mtx_lock(&softc->ctl_lock);
6165
6166	/*
6167	 * If there is a request to lock the LUN and another LUN is locked
6168	 * this is an error. If the requested LUN is already locked ignore
6169	 * the request. If no LUN is locked attempt to lock it.
6170	 * if there is a request to unlock the LUN and the LUN is currently
6171	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6172	 * if another LUN is locked or no LUN is locked.
6173	 */
6174	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6175		if (softc->aps_locked_lun == lun->lun) {
6176			/*
6177			 * This LUN is already locked, so we're done.
6178			 */
6179			retval = CTL_RETVAL_COMPLETE;
6180		} else if (softc->aps_locked_lun == 0) {
6181			/*
6182			 * No one has the lock, pass the request to the
6183			 * backend.
6184			 */
6185			retval = lun->backend->config_write(
6186				(union ctl_io *)ctsio);
6187		} else {
6188			/*
6189			 * Someone else has the lock, throw out the request.
6190			 */
6191			ctl_set_already_locked(ctsio);
6192			free(ctsio->kern_data_ptr, M_CTL);
6193			ctl_done((union ctl_io *)ctsio);
6194
6195			/*
6196			 * Set the return value so that ctl_do_mode_select()
6197			 * won't try to complete the command.  We already
6198			 * completed it here.
6199			 */
6200			retval = CTL_RETVAL_ERROR;
6201		}
6202	} else if (softc->aps_locked_lun == lun->lun) {
6203		/*
6204		 * This LUN is locked, so pass the unlock request to the
6205		 * backend.
6206		 */
6207		retval = lun->backend->config_write((union ctl_io *)ctsio);
6208	}
6209	mtx_unlock(&softc->ctl_lock);
6210
6211	return (retval);
6212}
6213
6214int
6215ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6216				struct ctl_page_index *page_index,
6217				uint8_t *page_ptr)
6218{
6219	uint8_t *c;
6220	int i;
6221
6222	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6223	ctl_time_io_secs =
6224		(c[0] << 8) |
6225		(c[1] << 0) |
6226		0;
6227	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6228	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6229	printf("page data:");
6230	for (i=0; i<8; i++)
6231		printf(" %.2x",page_ptr[i]);
6232	printf("\n");
6233	return (0);
6234}
6235
6236int
6237ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6238			       struct ctl_page_index *page_index,
6239			       int pc)
6240{
6241	struct copan_debugconf_subpage *page;
6242
6243	page = (struct copan_debugconf_subpage *)page_index->page_data +
6244		(page_index->page_len * pc);
6245
6246	switch (pc) {
6247	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6248	case SMS_PAGE_CTRL_DEFAULT >> 6:
6249	case SMS_PAGE_CTRL_SAVED >> 6:
6250		/*
6251		 * We don't update the changable or default bits for this page.
6252		 */
6253		break;
6254	case SMS_PAGE_CTRL_CURRENT >> 6:
6255		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6256		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6257		break;
6258	default:
6259#ifdef NEEDTOPORT
6260		EPRINT(0, "Invalid PC %d!!", pc);
6261#endif /* NEEDTOPORT */
6262		break;
6263	}
6264	return (0);
6265}
6266
6267
6268static int
6269ctl_do_mode_select(union ctl_io *io)
6270{
6271	struct scsi_mode_page_header *page_header;
6272	struct ctl_page_index *page_index;
6273	struct ctl_scsiio *ctsio;
6274	int control_dev, page_len;
6275	int page_len_offset, page_len_size;
6276	union ctl_modepage_info *modepage_info;
6277	struct ctl_lun *lun;
6278	int *len_left, *len_used;
6279	int retval, i;
6280
6281	ctsio = &io->scsiio;
6282	page_index = NULL;
6283	page_len = 0;
6284	retval = CTL_RETVAL_COMPLETE;
6285
6286	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6287
6288	if (lun->be_lun->lun_type != T_DIRECT)
6289		control_dev = 1;
6290	else
6291		control_dev = 0;
6292
6293	modepage_info = (union ctl_modepage_info *)
6294		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6295	len_left = &modepage_info->header.len_left;
6296	len_used = &modepage_info->header.len_used;
6297
6298do_next_page:
6299
6300	page_header = (struct scsi_mode_page_header *)
6301		(ctsio->kern_data_ptr + *len_used);
6302
6303	if (*len_left == 0) {
6304		free(ctsio->kern_data_ptr, M_CTL);
6305		ctl_set_success(ctsio);
6306		ctl_done((union ctl_io *)ctsio);
6307		return (CTL_RETVAL_COMPLETE);
6308	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6309
6310		free(ctsio->kern_data_ptr, M_CTL);
6311		ctl_set_param_len_error(ctsio);
6312		ctl_done((union ctl_io *)ctsio);
6313		return (CTL_RETVAL_COMPLETE);
6314
6315	} else if ((page_header->page_code & SMPH_SPF)
6316		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6317
6318		free(ctsio->kern_data_ptr, M_CTL);
6319		ctl_set_param_len_error(ctsio);
6320		ctl_done((union ctl_io *)ctsio);
6321		return (CTL_RETVAL_COMPLETE);
6322	}
6323
6324
6325	/*
6326	 * XXX KDM should we do something with the block descriptor?
6327	 */
6328	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6329
6330		if ((control_dev != 0)
6331		 && (lun->mode_pages.index[i].page_flags &
6332		     CTL_PAGE_FLAG_DISK_ONLY))
6333			continue;
6334
6335		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6336		    (page_header->page_code & SMPH_PC_MASK))
6337			continue;
6338
6339		/*
6340		 * If neither page has a subpage code, then we've got a
6341		 * match.
6342		 */
6343		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6344		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6345			page_index = &lun->mode_pages.index[i];
6346			page_len = page_header->page_length;
6347			break;
6348		}
6349
6350		/*
6351		 * If both pages have subpages, then the subpage numbers
6352		 * have to match.
6353		 */
6354		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6355		  && (page_header->page_code & SMPH_SPF)) {
6356			struct scsi_mode_page_header_sp *sph;
6357
6358			sph = (struct scsi_mode_page_header_sp *)page_header;
6359
6360			if (lun->mode_pages.index[i].subpage ==
6361			    sph->subpage) {
6362				page_index = &lun->mode_pages.index[i];
6363				page_len = scsi_2btoul(sph->page_length);
6364				break;
6365			}
6366		}
6367	}
6368
6369	/*
6370	 * If we couldn't find the page, or if we don't have a mode select
6371	 * handler for it, send back an error to the user.
6372	 */
6373	if ((page_index == NULL)
6374	 || (page_index->select_handler == NULL)) {
6375		ctl_set_invalid_field(ctsio,
6376				      /*sks_valid*/ 1,
6377				      /*command*/ 0,
6378				      /*field*/ *len_used,
6379				      /*bit_valid*/ 0,
6380				      /*bit*/ 0);
6381		free(ctsio->kern_data_ptr, M_CTL);
6382		ctl_done((union ctl_io *)ctsio);
6383		return (CTL_RETVAL_COMPLETE);
6384	}
6385
6386	if (page_index->page_code & SMPH_SPF) {
6387		page_len_offset = 2;
6388		page_len_size = 2;
6389	} else {
6390		page_len_size = 1;
6391		page_len_offset = 1;
6392	}
6393
6394	/*
6395	 * If the length the initiator gives us isn't the one we specify in
6396	 * the mode page header, or if they didn't specify enough data in
6397	 * the CDB to avoid truncating this page, kick out the request.
6398	 */
6399	if ((page_len != (page_index->page_len - page_len_offset -
6400			  page_len_size))
6401	 || (*len_left < page_index->page_len)) {
6402
6403
6404		ctl_set_invalid_field(ctsio,
6405				      /*sks_valid*/ 1,
6406				      /*command*/ 0,
6407				      /*field*/ *len_used + page_len_offset,
6408				      /*bit_valid*/ 0,
6409				      /*bit*/ 0);
6410		free(ctsio->kern_data_ptr, M_CTL);
6411		ctl_done((union ctl_io *)ctsio);
6412		return (CTL_RETVAL_COMPLETE);
6413	}
6414
6415	/*
6416	 * Run through the mode page, checking to make sure that the bits
6417	 * the user changed are actually legal for him to change.
6418	 */
6419	for (i = 0; i < page_index->page_len; i++) {
6420		uint8_t *user_byte, *change_mask, *current_byte;
6421		int bad_bit;
6422		int j;
6423
6424		user_byte = (uint8_t *)page_header + i;
6425		change_mask = page_index->page_data +
6426			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6427		current_byte = page_index->page_data +
6428			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6429
6430		/*
6431		 * Check to see whether the user set any bits in this byte
6432		 * that he is not allowed to set.
6433		 */
6434		if ((*user_byte & ~(*change_mask)) ==
6435		    (*current_byte & ~(*change_mask)))
6436			continue;
6437
6438		/*
6439		 * Go through bit by bit to determine which one is illegal.
6440		 */
6441		bad_bit = 0;
6442		for (j = 7; j >= 0; j--) {
6443			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6444			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6445				bad_bit = i;
6446				break;
6447			}
6448		}
6449		ctl_set_invalid_field(ctsio,
6450				      /*sks_valid*/ 1,
6451				      /*command*/ 0,
6452				      /*field*/ *len_used + i,
6453				      /*bit_valid*/ 1,
6454				      /*bit*/ bad_bit);
6455		free(ctsio->kern_data_ptr, M_CTL);
6456		ctl_done((union ctl_io *)ctsio);
6457		return (CTL_RETVAL_COMPLETE);
6458	}
6459
6460	/*
6461	 * Decrement these before we call the page handler, since we may
6462	 * end up getting called back one way or another before the handler
6463	 * returns to this context.
6464	 */
6465	*len_left -= page_index->page_len;
6466	*len_used += page_index->page_len;
6467
6468	retval = page_index->select_handler(ctsio, page_index,
6469					    (uint8_t *)page_header);
6470
6471	/*
6472	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6473	 * wait until this queued command completes to finish processing
6474	 * the mode page.  If it returns anything other than
6475	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6476	 * already set the sense information, freed the data pointer, and
6477	 * completed the io for us.
6478	 */
6479	if (retval != CTL_RETVAL_COMPLETE)
6480		goto bailout_no_done;
6481
6482	/*
6483	 * If the initiator sent us more than one page, parse the next one.
6484	 */
6485	if (*len_left > 0)
6486		goto do_next_page;
6487
6488	ctl_set_success(ctsio);
6489	free(ctsio->kern_data_ptr, M_CTL);
6490	ctl_done((union ctl_io *)ctsio);
6491
6492bailout_no_done:
6493
6494	return (CTL_RETVAL_COMPLETE);
6495
6496}
6497
6498int
6499ctl_mode_select(struct ctl_scsiio *ctsio)
6500{
6501	int param_len, pf, sp;
6502	int header_size, bd_len;
6503	int len_left, len_used;
6504	struct ctl_page_index *page_index;
6505	struct ctl_lun *lun;
6506	int control_dev, page_len;
6507	union ctl_modepage_info *modepage_info;
6508	int retval;
6509
6510	pf = 0;
6511	sp = 0;
6512	page_len = 0;
6513	len_used = 0;
6514	len_left = 0;
6515	retval = 0;
6516	bd_len = 0;
6517	page_index = NULL;
6518
6519	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6520
6521	if (lun->be_lun->lun_type != T_DIRECT)
6522		control_dev = 1;
6523	else
6524		control_dev = 0;
6525
6526	switch (ctsio->cdb[0]) {
6527	case MODE_SELECT_6: {
6528		struct scsi_mode_select_6 *cdb;
6529
6530		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6531
6532		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6533		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6534
6535		param_len = cdb->length;
6536		header_size = sizeof(struct scsi_mode_header_6);
6537		break;
6538	}
6539	case MODE_SELECT_10: {
6540		struct scsi_mode_select_10 *cdb;
6541
6542		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6543
6544		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6545		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6546
6547		param_len = scsi_2btoul(cdb->length);
6548		header_size = sizeof(struct scsi_mode_header_10);
6549		break;
6550	}
6551	default:
6552		ctl_set_invalid_opcode(ctsio);
6553		ctl_done((union ctl_io *)ctsio);
6554		return (CTL_RETVAL_COMPLETE);
6555		break; /* NOTREACHED */
6556	}
6557
6558	/*
6559	 * From SPC-3:
6560	 * "A parameter list length of zero indicates that the Data-Out Buffer
6561	 * shall be empty. This condition shall not be considered as an error."
6562	 */
6563	if (param_len == 0) {
6564		ctl_set_success(ctsio);
6565		ctl_done((union ctl_io *)ctsio);
6566		return (CTL_RETVAL_COMPLETE);
6567	}
6568
6569	/*
6570	 * Since we'll hit this the first time through, prior to
6571	 * allocation, we don't need to free a data buffer here.
6572	 */
6573	if (param_len < header_size) {
6574		ctl_set_param_len_error(ctsio);
6575		ctl_done((union ctl_io *)ctsio);
6576		return (CTL_RETVAL_COMPLETE);
6577	}
6578
6579	/*
6580	 * Allocate the data buffer and grab the user's data.  In theory,
6581	 * we shouldn't have to sanity check the parameter list length here
6582	 * because the maximum size is 64K.  We should be able to malloc
6583	 * that much without too many problems.
6584	 */
6585	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6586		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6587		ctsio->kern_data_len = param_len;
6588		ctsio->kern_total_len = param_len;
6589		ctsio->kern_data_resid = 0;
6590		ctsio->kern_rel_offset = 0;
6591		ctsio->kern_sg_entries = 0;
6592		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6593		ctsio->be_move_done = ctl_config_move_done;
6594		ctl_datamove((union ctl_io *)ctsio);
6595
6596		return (CTL_RETVAL_COMPLETE);
6597	}
6598
6599	switch (ctsio->cdb[0]) {
6600	case MODE_SELECT_6: {
6601		struct scsi_mode_header_6 *mh6;
6602
6603		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6604		bd_len = mh6->blk_desc_len;
6605		break;
6606	}
6607	case MODE_SELECT_10: {
6608		struct scsi_mode_header_10 *mh10;
6609
6610		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6611		bd_len = scsi_2btoul(mh10->blk_desc_len);
6612		break;
6613	}
6614	default:
6615		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6616		break;
6617	}
6618
6619	if (param_len < (header_size + bd_len)) {
6620		free(ctsio->kern_data_ptr, M_CTL);
6621		ctl_set_param_len_error(ctsio);
6622		ctl_done((union ctl_io *)ctsio);
6623		return (CTL_RETVAL_COMPLETE);
6624	}
6625
6626	/*
6627	 * Set the IO_CONT flag, so that if this I/O gets passed to
6628	 * ctl_config_write_done(), it'll get passed back to
6629	 * ctl_do_mode_select() for further processing, or completion if
6630	 * we're all done.
6631	 */
6632	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6633	ctsio->io_cont = ctl_do_mode_select;
6634
6635	modepage_info = (union ctl_modepage_info *)
6636		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6637
6638	memset(modepage_info, 0, sizeof(*modepage_info));
6639
6640	len_left = param_len - header_size - bd_len;
6641	len_used = header_size + bd_len;
6642
6643	modepage_info->header.len_left = len_left;
6644	modepage_info->header.len_used = len_used;
6645
6646	return (ctl_do_mode_select((union ctl_io *)ctsio));
6647}
6648
6649int
6650ctl_mode_sense(struct ctl_scsiio *ctsio)
6651{
6652	struct ctl_lun *lun;
6653	int pc, page_code, dbd, llba, subpage;
6654	int alloc_len, page_len, header_len, total_len;
6655	struct scsi_mode_block_descr *block_desc;
6656	struct ctl_page_index *page_index;
6657	int control_dev;
6658
6659	dbd = 0;
6660	llba = 0;
6661	block_desc = NULL;
6662	page_index = NULL;
6663
6664	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6665
6666	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6667
6668	if (lun->be_lun->lun_type != T_DIRECT)
6669		control_dev = 1;
6670	else
6671		control_dev = 0;
6672
6673	if (lun->flags & CTL_LUN_PR_RESERVED) {
6674		uint32_t residx;
6675
6676		/*
6677		 * XXX KDM need a lock here.
6678		 */
6679		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
6680		if ((lun->res_type == SPR_TYPE_EX_AC
6681		  && residx != lun->pr_res_idx)
6682		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
6683		   || lun->res_type == SPR_TYPE_EX_AC_AR)
6684		  && !lun->per_res[residx].registered)) {
6685			ctl_set_reservation_conflict(ctsio);
6686			ctl_done((union ctl_io *)ctsio);
6687			return (CTL_RETVAL_COMPLETE);
6688		}
6689	}
6690
6691	switch (ctsio->cdb[0]) {
6692	case MODE_SENSE_6: {
6693		struct scsi_mode_sense_6 *cdb;
6694
6695		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6696
6697		header_len = sizeof(struct scsi_mode_hdr_6);
6698		if (cdb->byte2 & SMS_DBD)
6699			dbd = 1;
6700		else
6701			header_len += sizeof(struct scsi_mode_block_descr);
6702
6703		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6704		page_code = cdb->page & SMS_PAGE_CODE;
6705		subpage = cdb->subpage;
6706		alloc_len = cdb->length;
6707		break;
6708	}
6709	case MODE_SENSE_10: {
6710		struct scsi_mode_sense_10 *cdb;
6711
6712		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6713
6714		header_len = sizeof(struct scsi_mode_hdr_10);
6715
6716		if (cdb->byte2 & SMS_DBD)
6717			dbd = 1;
6718		else
6719			header_len += sizeof(struct scsi_mode_block_descr);
6720		if (cdb->byte2 & SMS10_LLBAA)
6721			llba = 1;
6722		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6723		page_code = cdb->page & SMS_PAGE_CODE;
6724		subpage = cdb->subpage;
6725		alloc_len = scsi_2btoul(cdb->length);
6726		break;
6727	}
6728	default:
6729		ctl_set_invalid_opcode(ctsio);
6730		ctl_done((union ctl_io *)ctsio);
6731		return (CTL_RETVAL_COMPLETE);
6732		break; /* NOTREACHED */
6733	}
6734
6735	/*
6736	 * We have to make a first pass through to calculate the size of
6737	 * the pages that match the user's query.  Then we allocate enough
6738	 * memory to hold it, and actually copy the data into the buffer.
6739	 */
6740	switch (page_code) {
6741	case SMS_ALL_PAGES_PAGE: {
6742		int i;
6743
6744		page_len = 0;
6745
6746		/*
6747		 * At the moment, values other than 0 and 0xff here are
6748		 * reserved according to SPC-3.
6749		 */
6750		if ((subpage != SMS_SUBPAGE_PAGE_0)
6751		 && (subpage != SMS_SUBPAGE_ALL)) {
6752			ctl_set_invalid_field(ctsio,
6753					      /*sks_valid*/ 1,
6754					      /*command*/ 1,
6755					      /*field*/ 3,
6756					      /*bit_valid*/ 0,
6757					      /*bit*/ 0);
6758			ctl_done((union ctl_io *)ctsio);
6759			return (CTL_RETVAL_COMPLETE);
6760		}
6761
6762		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6763			if ((control_dev != 0)
6764			 && (lun->mode_pages.index[i].page_flags &
6765			     CTL_PAGE_FLAG_DISK_ONLY))
6766				continue;
6767
6768			/*
6769			 * We don't use this subpage if the user didn't
6770			 * request all subpages.
6771			 */
6772			if ((lun->mode_pages.index[i].subpage != 0)
6773			 && (subpage == SMS_SUBPAGE_PAGE_0))
6774				continue;
6775
6776#if 0
6777			printf("found page %#x len %d\n",
6778			       lun->mode_pages.index[i].page_code &
6779			       SMPH_PC_MASK,
6780			       lun->mode_pages.index[i].page_len);
6781#endif
6782			page_len += lun->mode_pages.index[i].page_len;
6783		}
6784		break;
6785	}
6786	default: {
6787		int i;
6788
6789		page_len = 0;
6790
6791		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6792			/* Look for the right page code */
6793			if ((lun->mode_pages.index[i].page_code &
6794			     SMPH_PC_MASK) != page_code)
6795				continue;
6796
6797			/* Look for the right subpage or the subpage wildcard*/
6798			if ((lun->mode_pages.index[i].subpage != subpage)
6799			 && (subpage != SMS_SUBPAGE_ALL))
6800				continue;
6801
6802			/* Make sure the page is supported for this dev type */
6803			if ((control_dev != 0)
6804			 && (lun->mode_pages.index[i].page_flags &
6805			     CTL_PAGE_FLAG_DISK_ONLY))
6806				continue;
6807
6808#if 0
6809			printf("found page %#x len %d\n",
6810			       lun->mode_pages.index[i].page_code &
6811			       SMPH_PC_MASK,
6812			       lun->mode_pages.index[i].page_len);
6813#endif
6814
6815			page_len += lun->mode_pages.index[i].page_len;
6816		}
6817
6818		if (page_len == 0) {
6819			ctl_set_invalid_field(ctsio,
6820					      /*sks_valid*/ 1,
6821					      /*command*/ 1,
6822					      /*field*/ 2,
6823					      /*bit_valid*/ 1,
6824					      /*bit*/ 5);
6825			ctl_done((union ctl_io *)ctsio);
6826			return (CTL_RETVAL_COMPLETE);
6827		}
6828		break;
6829	}
6830	}
6831
6832	total_len = header_len + page_len;
6833#if 0
6834	printf("header_len = %d, page_len = %d, total_len = %d\n",
6835	       header_len, page_len, total_len);
6836#endif
6837
6838	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6839	ctsio->kern_sg_entries = 0;
6840	ctsio->kern_data_resid = 0;
6841	ctsio->kern_rel_offset = 0;
6842	if (total_len < alloc_len) {
6843		ctsio->residual = alloc_len - total_len;
6844		ctsio->kern_data_len = total_len;
6845		ctsio->kern_total_len = total_len;
6846	} else {
6847		ctsio->residual = 0;
6848		ctsio->kern_data_len = alloc_len;
6849		ctsio->kern_total_len = alloc_len;
6850	}
6851
6852	switch (ctsio->cdb[0]) {
6853	case MODE_SENSE_6: {
6854		struct scsi_mode_hdr_6 *header;
6855
6856		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6857
6858		header->datalen = ctl_min(total_len - 1, 254);
6859
6860		if (dbd)
6861			header->block_descr_len = 0;
6862		else
6863			header->block_descr_len =
6864				sizeof(struct scsi_mode_block_descr);
6865		block_desc = (struct scsi_mode_block_descr *)&header[1];
6866		break;
6867	}
6868	case MODE_SENSE_10: {
6869		struct scsi_mode_hdr_10 *header;
6870		int datalen;
6871
6872		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6873
6874		datalen = ctl_min(total_len - 2, 65533);
6875		scsi_ulto2b(datalen, header->datalen);
6876		if (dbd)
6877			scsi_ulto2b(0, header->block_descr_len);
6878		else
6879			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6880				    header->block_descr_len);
6881		block_desc = (struct scsi_mode_block_descr *)&header[1];
6882		break;
6883	}
6884	default:
6885		panic("invalid CDB type %#x", ctsio->cdb[0]);
6886		break; /* NOTREACHED */
6887	}
6888
6889	/*
6890	 * If we've got a disk, use its blocksize in the block
6891	 * descriptor.  Otherwise, just set it to 0.
6892	 */
6893	if (dbd == 0) {
6894		if (control_dev != 0)
6895			scsi_ulto3b(lun->be_lun->blocksize,
6896				    block_desc->block_len);
6897		else
6898			scsi_ulto3b(0, block_desc->block_len);
6899	}
6900
6901	switch (page_code) {
6902	case SMS_ALL_PAGES_PAGE: {
6903		int i, data_used;
6904
6905		data_used = header_len;
6906		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6907			struct ctl_page_index *page_index;
6908
6909			page_index = &lun->mode_pages.index[i];
6910
6911			if ((control_dev != 0)
6912			 && (page_index->page_flags &
6913			    CTL_PAGE_FLAG_DISK_ONLY))
6914				continue;
6915
6916			/*
6917			 * We don't use this subpage if the user didn't
6918			 * request all subpages.  We already checked (above)
6919			 * to make sure the user only specified a subpage
6920			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6921			 */
6922			if ((page_index->subpage != 0)
6923			 && (subpage == SMS_SUBPAGE_PAGE_0))
6924				continue;
6925
6926			/*
6927			 * Call the handler, if it exists, to update the
6928			 * page to the latest values.
6929			 */
6930			if (page_index->sense_handler != NULL)
6931				page_index->sense_handler(ctsio, page_index,pc);
6932
6933			memcpy(ctsio->kern_data_ptr + data_used,
6934			       page_index->page_data +
6935			       (page_index->page_len * pc),
6936			       page_index->page_len);
6937			data_used += page_index->page_len;
6938		}
6939		break;
6940	}
6941	default: {
6942		int i, data_used;
6943
6944		data_used = header_len;
6945
6946		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6947			struct ctl_page_index *page_index;
6948
6949			page_index = &lun->mode_pages.index[i];
6950
6951			/* Look for the right page code */
6952			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6953				continue;
6954
6955			/* Look for the right subpage or the subpage wildcard*/
6956			if ((page_index->subpage != subpage)
6957			 && (subpage != SMS_SUBPAGE_ALL))
6958				continue;
6959
6960			/* Make sure the page is supported for this dev type */
6961			if ((control_dev != 0)
6962			 && (page_index->page_flags &
6963			     CTL_PAGE_FLAG_DISK_ONLY))
6964				continue;
6965
6966			/*
6967			 * Call the handler, if it exists, to update the
6968			 * page to the latest values.
6969			 */
6970			if (page_index->sense_handler != NULL)
6971				page_index->sense_handler(ctsio, page_index,pc);
6972
6973			memcpy(ctsio->kern_data_ptr + data_used,
6974			       page_index->page_data +
6975			       (page_index->page_len * pc),
6976			       page_index->page_len);
6977			data_used += page_index->page_len;
6978		}
6979		break;
6980	}
6981	}
6982
6983	ctsio->scsi_status = SCSI_STATUS_OK;
6984
6985	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6986	ctsio->be_move_done = ctl_config_move_done;
6987	ctl_datamove((union ctl_io *)ctsio);
6988
6989	return (CTL_RETVAL_COMPLETE);
6990}
6991
6992int
6993ctl_read_capacity(struct ctl_scsiio *ctsio)
6994{
6995	struct scsi_read_capacity *cdb;
6996	struct scsi_read_capacity_data *data;
6997	struct ctl_lun *lun;
6998	uint32_t lba;
6999
7000	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7001
7002	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7003
7004	lba = scsi_4btoul(cdb->addr);
7005	if (((cdb->pmi & SRC_PMI) == 0)
7006	 && (lba != 0)) {
7007		ctl_set_invalid_field(/*ctsio*/ ctsio,
7008				      /*sks_valid*/ 1,
7009				      /*command*/ 1,
7010				      /*field*/ 2,
7011				      /*bit_valid*/ 0,
7012				      /*bit*/ 0);
7013		ctl_done((union ctl_io *)ctsio);
7014		return (CTL_RETVAL_COMPLETE);
7015	}
7016
7017	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7018
7019	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7020	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7021	ctsio->residual = 0;
7022	ctsio->kern_data_len = sizeof(*data);
7023	ctsio->kern_total_len = sizeof(*data);
7024	ctsio->kern_data_resid = 0;
7025	ctsio->kern_rel_offset = 0;
7026	ctsio->kern_sg_entries = 0;
7027
7028	/*
7029	 * If the maximum LBA is greater than 0xfffffffe, the user must
7030	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7031	 * serivce action set.
7032	 */
7033	if (lun->be_lun->maxlba > 0xfffffffe)
7034		scsi_ulto4b(0xffffffff, data->addr);
7035	else
7036		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7037
7038	/*
7039	 * XXX KDM this may not be 512 bytes...
7040	 */
7041	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7042
7043	ctsio->scsi_status = SCSI_STATUS_OK;
7044
7045	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7046	ctsio->be_move_done = ctl_config_move_done;
7047	ctl_datamove((union ctl_io *)ctsio);
7048
7049	return (CTL_RETVAL_COMPLETE);
7050}
7051
7052static int
7053ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7054{
7055	struct scsi_read_capacity_16 *cdb;
7056	struct scsi_read_capacity_data_long *data;
7057	struct ctl_lun *lun;
7058	uint64_t lba;
7059	uint32_t alloc_len;
7060
7061	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7062
7063	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7064
7065	alloc_len = scsi_4btoul(cdb->alloc_len);
7066	lba = scsi_8btou64(cdb->addr);
7067
7068	if ((cdb->reladr & SRC16_PMI)
7069	 && (lba != 0)) {
7070		ctl_set_invalid_field(/*ctsio*/ ctsio,
7071				      /*sks_valid*/ 1,
7072				      /*command*/ 1,
7073				      /*field*/ 2,
7074				      /*bit_valid*/ 0,
7075				      /*bit*/ 0);
7076		ctl_done((union ctl_io *)ctsio);
7077		return (CTL_RETVAL_COMPLETE);
7078	}
7079
7080	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7081
7082	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7083	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7084
7085	if (sizeof(*data) < alloc_len) {
7086		ctsio->residual = alloc_len - sizeof(*data);
7087		ctsio->kern_data_len = sizeof(*data);
7088		ctsio->kern_total_len = sizeof(*data);
7089	} else {
7090		ctsio->residual = 0;
7091		ctsio->kern_data_len = alloc_len;
7092		ctsio->kern_total_len = alloc_len;
7093	}
7094	ctsio->kern_data_resid = 0;
7095	ctsio->kern_rel_offset = 0;
7096	ctsio->kern_sg_entries = 0;
7097
7098	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7099	/* XXX KDM this may not be 512 bytes... */
7100	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7101	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7102	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7103	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7104		data->lalba_lbp[0] |= SRC16_LBPME;
7105
7106	ctsio->scsi_status = SCSI_STATUS_OK;
7107
7108	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7109	ctsio->be_move_done = ctl_config_move_done;
7110	ctl_datamove((union ctl_io *)ctsio);
7111
7112	return (CTL_RETVAL_COMPLETE);
7113}
7114
7115int
7116ctl_service_action_in(struct ctl_scsiio *ctsio)
7117{
7118	struct scsi_service_action_in *cdb;
7119	int retval;
7120
7121	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
7122
7123	cdb = (struct scsi_service_action_in *)ctsio->cdb;
7124
7125	retval = CTL_RETVAL_COMPLETE;
7126
7127	switch (cdb->service_action) {
7128	case SRC16_SERVICE_ACTION:
7129		retval = ctl_read_capacity_16(ctsio);
7130		break;
7131	default:
7132		ctl_set_invalid_field(/*ctsio*/ ctsio,
7133				      /*sks_valid*/ 1,
7134				      /*command*/ 1,
7135				      /*field*/ 1,
7136				      /*bit_valid*/ 1,
7137				      /*bit*/ 4);
7138		ctl_done((union ctl_io *)ctsio);
7139		break;
7140	}
7141
7142	return (retval);
7143}
7144
7145int
7146ctl_maintenance_in(struct ctl_scsiio *ctsio)
7147{
7148	struct scsi_maintenance_in *cdb;
7149	int retval;
7150	int alloc_len, total_len = 0;
7151	int num_target_port_groups, single;
7152	struct ctl_lun *lun;
7153	struct ctl_softc *softc;
7154	struct scsi_target_group_data *rtg_ptr;
7155	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
7156	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
7157	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
7158
7159	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
7160
7161	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7162	softc = control_softc;
7163	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7164
7165	retval = CTL_RETVAL_COMPLETE;
7166
7167	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
7168		ctl_set_invalid_field(/*ctsio*/ ctsio,
7169				      /*sks_valid*/ 1,
7170				      /*command*/ 1,
7171				      /*field*/ 1,
7172				      /*bit_valid*/ 1,
7173				      /*bit*/ 4);
7174		ctl_done((union ctl_io *)ctsio);
7175		return(retval);
7176	}
7177
7178	single = ctl_is_single;
7179	if (single)
7180        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
7181	else
7182        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7183
7184	total_len = sizeof(struct scsi_target_group_data) +
7185		sizeof(struct scsi_target_port_group_descriptor) *
7186		num_target_port_groups +
7187		sizeof(struct scsi_target_port_descriptor) *
7188		NUM_PORTS_PER_GRP * num_target_port_groups;
7189
7190	alloc_len = scsi_4btoul(cdb->length);
7191
7192	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7193
7194	ctsio->kern_sg_entries = 0;
7195
7196	if (total_len < alloc_len) {
7197		ctsio->residual = alloc_len - total_len;
7198		ctsio->kern_data_len = total_len;
7199		ctsio->kern_total_len = total_len;
7200	} else {
7201		ctsio->residual = 0;
7202		ctsio->kern_data_len = alloc_len;
7203		ctsio->kern_total_len = alloc_len;
7204	}
7205	ctsio->kern_data_resid = 0;
7206	ctsio->kern_rel_offset = 0;
7207
7208	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7209
7210	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7211	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7212	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7213	        &tp_desc_ptr1_1->desc_list[0];
7214
7215	if (single == 0) {
7216		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7217	                &tp_desc_ptr1_2->desc_list[0];
7218		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7219		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7220	        	&tp_desc_ptr2_1->desc_list[0];
7221        } else {
7222		tpg_desc_ptr2 = NULL;
7223		tp_desc_ptr2_1 = NULL;
7224		tp_desc_ptr2_2 = NULL;
7225	}
7226
7227	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7228	if (single == 0) {
7229        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7230			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7231				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7232				tpg_desc_ptr2->pref_state =
7233					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7234			} else {
7235				tpg_desc_ptr1->pref_state =
7236					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7237				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7238			}
7239		} else {
7240			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7241				tpg_desc_ptr1->pref_state =
7242					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7243				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7244			} else {
7245				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7246				tpg_desc_ptr2->pref_state =
7247					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7248			}
7249		}
7250	} else {
7251		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7252	}
7253	tpg_desc_ptr1->support = 0;
7254	tpg_desc_ptr1->target_port_group[1] = 1;
7255	tpg_desc_ptr1->status = TPG_IMPLICIT;
7256	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7257
7258	if (single == 0) {
7259		tpg_desc_ptr2->support = 0;
7260		tpg_desc_ptr2->target_port_group[1] = 2;
7261		tpg_desc_ptr2->status = TPG_IMPLICIT;
7262		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7263
7264		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7265		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7266
7267		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7268		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7269	} else {
7270        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7271			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7272			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7273		} else {
7274			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7275			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7276		}
7277	}
7278
7279	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7280	ctsio->be_move_done = ctl_config_move_done;
7281
7282	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7283			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7284			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7285			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7286			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7287
7288	ctl_datamove((union ctl_io *)ctsio);
7289	return(retval);
7290}
7291
7292int
7293ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7294{
7295	struct scsi_per_res_in *cdb;
7296	int alloc_len, total_len = 0;
7297	/* struct scsi_per_res_in_rsrv in_data; */
7298	struct ctl_lun *lun;
7299	struct ctl_softc *softc;
7300
7301	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7302
7303	softc = control_softc;
7304
7305	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7306
7307	alloc_len = scsi_2btoul(cdb->length);
7308
7309	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7310
7311retry:
7312	mtx_lock(&lun->lun_lock);
7313	switch (cdb->action) {
7314	case SPRI_RK: /* read keys */
7315		total_len = sizeof(struct scsi_per_res_in_keys) +
7316			lun->pr_key_count *
7317			sizeof(struct scsi_per_res_key);
7318		break;
7319	case SPRI_RR: /* read reservation */
7320		if (lun->flags & CTL_LUN_PR_RESERVED)
7321			total_len = sizeof(struct scsi_per_res_in_rsrv);
7322		else
7323			total_len = sizeof(struct scsi_per_res_in_header);
7324		break;
7325	case SPRI_RC: /* report capabilities */
7326		total_len = sizeof(struct scsi_per_res_cap);
7327		break;
7328	case SPRI_RS: /* read full status */
7329	default:
7330		mtx_unlock(&lun->lun_lock);
7331		ctl_set_invalid_field(ctsio,
7332				      /*sks_valid*/ 1,
7333				      /*command*/ 1,
7334				      /*field*/ 1,
7335				      /*bit_valid*/ 1,
7336				      /*bit*/ 0);
7337		ctl_done((union ctl_io *)ctsio);
7338		return (CTL_RETVAL_COMPLETE);
7339		break; /* NOTREACHED */
7340	}
7341	mtx_unlock(&lun->lun_lock);
7342
7343	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7344
7345	if (total_len < alloc_len) {
7346		ctsio->residual = alloc_len - total_len;
7347		ctsio->kern_data_len = total_len;
7348		ctsio->kern_total_len = total_len;
7349	} else {
7350		ctsio->residual = 0;
7351		ctsio->kern_data_len = alloc_len;
7352		ctsio->kern_total_len = alloc_len;
7353	}
7354
7355	ctsio->kern_data_resid = 0;
7356	ctsio->kern_rel_offset = 0;
7357	ctsio->kern_sg_entries = 0;
7358
7359	mtx_lock(&lun->lun_lock);
7360	switch (cdb->action) {
7361	case SPRI_RK: { // read keys
7362        struct scsi_per_res_in_keys *res_keys;
7363		int i, key_count;
7364
7365		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7366
7367		/*
7368		 * We had to drop the lock to allocate our buffer, which
7369		 * leaves time for someone to come in with another
7370		 * persistent reservation.  (That is unlikely, though,
7371		 * since this should be the only persistent reservation
7372		 * command active right now.)
7373		 */
7374		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7375		    (lun->pr_key_count *
7376		     sizeof(struct scsi_per_res_key)))){
7377			mtx_unlock(&lun->lun_lock);
7378			free(ctsio->kern_data_ptr, M_CTL);
7379			printf("%s: reservation length changed, retrying\n",
7380			       __func__);
7381			goto retry;
7382		}
7383
7384		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7385
7386		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7387			     lun->pr_key_count, res_keys->header.length);
7388
7389		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7390			if (!lun->per_res[i].registered)
7391				continue;
7392
7393			/*
7394			 * We used lun->pr_key_count to calculate the
7395			 * size to allocate.  If it turns out the number of
7396			 * initiators with the registered flag set is
7397			 * larger than that (i.e. they haven't been kept in
7398			 * sync), we've got a problem.
7399			 */
7400			if (key_count >= lun->pr_key_count) {
7401#ifdef NEEDTOPORT
7402				csevent_log(CSC_CTL | CSC_SHELF_SW |
7403					    CTL_PR_ERROR,
7404					    csevent_LogType_Fault,
7405					    csevent_AlertLevel_Yellow,
7406					    csevent_FRU_ShelfController,
7407					    csevent_FRU_Firmware,
7408				        csevent_FRU_Unknown,
7409					    "registered keys %d >= key "
7410					    "count %d", key_count,
7411					    lun->pr_key_count);
7412#endif
7413				key_count++;
7414				continue;
7415			}
7416			memcpy(res_keys->keys[key_count].key,
7417			       lun->per_res[i].res_key.key,
7418			       ctl_min(sizeof(res_keys->keys[key_count].key),
7419			       sizeof(lun->per_res[i].res_key)));
7420			key_count++;
7421		}
7422		break;
7423	}
7424	case SPRI_RR: { // read reservation
7425		struct scsi_per_res_in_rsrv *res;
7426		int tmp_len, header_only;
7427
7428		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7429
7430		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7431
7432		if (lun->flags & CTL_LUN_PR_RESERVED)
7433		{
7434			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7435			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7436				    res->header.length);
7437			header_only = 0;
7438		} else {
7439			tmp_len = sizeof(struct scsi_per_res_in_header);
7440			scsi_ulto4b(0, res->header.length);
7441			header_only = 1;
7442		}
7443
7444		/*
7445		 * We had to drop the lock to allocate our buffer, which
7446		 * leaves time for someone to come in with another
7447		 * persistent reservation.  (That is unlikely, though,
7448		 * since this should be the only persistent reservation
7449		 * command active right now.)
7450		 */
7451		if (tmp_len != total_len) {
7452			mtx_unlock(&lun->lun_lock);
7453			free(ctsio->kern_data_ptr, M_CTL);
7454			printf("%s: reservation status changed, retrying\n",
7455			       __func__);
7456			goto retry;
7457		}
7458
7459		/*
7460		 * No reservation held, so we're done.
7461		 */
7462		if (header_only != 0)
7463			break;
7464
7465		/*
7466		 * If the registration is an All Registrants type, the key
7467		 * is 0, since it doesn't really matter.
7468		 */
7469		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7470			memcpy(res->data.reservation,
7471			       &lun->per_res[lun->pr_res_idx].res_key,
7472			       sizeof(struct scsi_per_res_key));
7473		}
7474		res->data.scopetype = lun->res_type;
7475		break;
7476	}
7477	case SPRI_RC:     //report capabilities
7478	{
7479		struct scsi_per_res_cap *res_cap;
7480		uint16_t type_mask;
7481
7482		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7483		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7484		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_3;
7485		type_mask = SPRI_TM_WR_EX_AR |
7486			    SPRI_TM_EX_AC_RO |
7487			    SPRI_TM_WR_EX_RO |
7488			    SPRI_TM_EX_AC |
7489			    SPRI_TM_WR_EX |
7490			    SPRI_TM_EX_AC_AR;
7491		scsi_ulto2b(type_mask, res_cap->type_mask);
7492		break;
7493	}
7494	case SPRI_RS: //read full status
7495	default:
7496		/*
7497		 * This is a bug, because we just checked for this above,
7498		 * and should have returned an error.
7499		 */
7500		panic("Invalid PR type %x", cdb->action);
7501		break; /* NOTREACHED */
7502	}
7503	mtx_unlock(&lun->lun_lock);
7504
7505	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7506	ctsio->be_move_done = ctl_config_move_done;
7507
7508	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7509			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7510			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7511			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7512			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7513
7514	ctl_datamove((union ctl_io *)ctsio);
7515
7516	return (CTL_RETVAL_COMPLETE);
7517}
7518
7519/*
7520 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7521 * it should return.
7522 */
7523static int
7524ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7525		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7526		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7527		struct scsi_per_res_out_parms* param)
7528{
7529	union ctl_ha_msg persis_io;
7530	int retval, i;
7531	int isc_retval;
7532
7533	retval = 0;
7534
7535	mtx_lock(&lun->lun_lock);
7536	if (sa_res_key == 0) {
7537		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7538			/* validate scope and type */
7539			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7540			     SPR_LU_SCOPE) {
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*/ 4);
7548				ctl_done((union ctl_io *)ctsio);
7549				return (1);
7550			}
7551
7552		        if (type>8 || type==2 || type==4 || type==0) {
7553				mtx_unlock(&lun->lun_lock);
7554				ctl_set_invalid_field(/*ctsio*/ ctsio,
7555       	           				      /*sks_valid*/ 1,
7556						      /*command*/ 1,
7557						      /*field*/ 2,
7558						      /*bit_valid*/ 1,
7559						      /*bit*/ 0);
7560				ctl_done((union ctl_io *)ctsio);
7561				return (1);
7562		        }
7563
7564			/* temporarily unregister this nexus */
7565			lun->per_res[residx].registered = 0;
7566
7567			/*
7568			 * Unregister everybody else and build UA for
7569			 * them
7570			 */
7571			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7572				if (lun->per_res[i].registered == 0)
7573					continue;
7574
7575				if (!persis_offset
7576				 && i <CTL_MAX_INITIATORS)
7577					lun->pending_sense[i].ua_pending |=
7578						CTL_UA_REG_PREEMPT;
7579				else if (persis_offset
7580				      && i >= persis_offset)
7581					lun->pending_sense[i-persis_offset
7582						].ua_pending |=
7583						CTL_UA_REG_PREEMPT;
7584				lun->per_res[i].registered = 0;
7585				memset(&lun->per_res[i].res_key, 0,
7586				       sizeof(struct scsi_per_res_key));
7587			}
7588			lun->per_res[residx].registered = 1;
7589			lun->pr_key_count = 1;
7590			lun->res_type = type;
7591			if (lun->res_type != SPR_TYPE_WR_EX_AR
7592			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7593				lun->pr_res_idx = residx;
7594
7595			/* send msg to other side */
7596			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7597			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7598			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7599			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7600			persis_io.pr.pr_info.res_type = type;
7601			memcpy(persis_io.pr.pr_info.sa_res_key,
7602			       param->serv_act_res_key,
7603			       sizeof(param->serv_act_res_key));
7604			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7605			     &persis_io, sizeof(persis_io), 0)) >
7606			     CTL_HA_STATUS_SUCCESS) {
7607				printf("CTL:Persis Out error returned "
7608				       "from ctl_ha_msg_send %d\n",
7609				       isc_retval);
7610			}
7611		} else {
7612			/* not all registrants */
7613			mtx_unlock(&lun->lun_lock);
7614			free(ctsio->kern_data_ptr, M_CTL);
7615			ctl_set_invalid_field(ctsio,
7616					      /*sks_valid*/ 1,
7617					      /*command*/ 0,
7618					      /*field*/ 8,
7619					      /*bit_valid*/ 0,
7620					      /*bit*/ 0);
7621			ctl_done((union ctl_io *)ctsio);
7622			return (1);
7623		}
7624	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7625		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7626		int found = 0;
7627
7628		if (res_key == sa_res_key) {
7629			/* special case */
7630			/*
7631			 * The spec implies this is not good but doesn't
7632			 * say what to do. There are two choices either
7633			 * generate a res conflict or check condition
7634			 * with illegal field in parameter data. Since
7635			 * that is what is done when the sa_res_key is
7636			 * zero I'll take that approach since this has
7637			 * to do with the sa_res_key.
7638			 */
7639			mtx_unlock(&lun->lun_lock);
7640			free(ctsio->kern_data_ptr, M_CTL);
7641			ctl_set_invalid_field(ctsio,
7642					      /*sks_valid*/ 1,
7643					      /*command*/ 0,
7644					      /*field*/ 8,
7645					      /*bit_valid*/ 0,
7646					      /*bit*/ 0);
7647			ctl_done((union ctl_io *)ctsio);
7648			return (1);
7649		}
7650
7651		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7652			if (lun->per_res[i].registered
7653			 && memcmp(param->serv_act_res_key,
7654			    lun->per_res[i].res_key.key,
7655			    sizeof(struct scsi_per_res_key)) != 0)
7656				continue;
7657
7658			found = 1;
7659			lun->per_res[i].registered = 0;
7660			memset(&lun->per_res[i].res_key, 0,
7661			       sizeof(struct scsi_per_res_key));
7662			lun->pr_key_count--;
7663
7664			if (!persis_offset
7665			 && i < CTL_MAX_INITIATORS)
7666				lun->pending_sense[i].ua_pending |=
7667					CTL_UA_REG_PREEMPT;
7668			else if (persis_offset
7669			      && i >= persis_offset)
7670				lun->pending_sense[i-persis_offset].ua_pending|=
7671					CTL_UA_REG_PREEMPT;
7672		}
7673		if (!found) {
7674			mtx_unlock(&lun->lun_lock);
7675			free(ctsio->kern_data_ptr, M_CTL);
7676			ctl_set_reservation_conflict(ctsio);
7677			ctl_done((union ctl_io *)ctsio);
7678			return (CTL_RETVAL_COMPLETE);
7679		}
7680		/* send msg to other side */
7681		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7682		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7683		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7684		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7685		persis_io.pr.pr_info.res_type = type;
7686		memcpy(persis_io.pr.pr_info.sa_res_key,
7687		       param->serv_act_res_key,
7688		       sizeof(param->serv_act_res_key));
7689		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7690		     &persis_io, sizeof(persis_io), 0)) >
7691		     CTL_HA_STATUS_SUCCESS) {
7692			printf("CTL:Persis Out error returned from "
7693			       "ctl_ha_msg_send %d\n", isc_retval);
7694		}
7695	} else {
7696		/* Reserved but not all registrants */
7697		/* sa_res_key is res holder */
7698		if (memcmp(param->serv_act_res_key,
7699                   lun->per_res[lun->pr_res_idx].res_key.key,
7700                   sizeof(struct scsi_per_res_key)) == 0) {
7701			/* validate scope and type */
7702			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7703			     SPR_LU_SCOPE) {
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*/ 4);
7711				ctl_done((union ctl_io *)ctsio);
7712				return (1);
7713			}
7714
7715			if (type>8 || type==2 || type==4 || type==0) {
7716				mtx_unlock(&lun->lun_lock);
7717				ctl_set_invalid_field(/*ctsio*/ ctsio,
7718						      /*sks_valid*/ 1,
7719						      /*command*/ 1,
7720						      /*field*/ 2,
7721						      /*bit_valid*/ 1,
7722						      /*bit*/ 0);
7723				ctl_done((union ctl_io *)ctsio);
7724				return (1);
7725			}
7726
7727			/*
7728			 * Do the following:
7729			 * if sa_res_key != res_key remove all
7730			 * registrants w/sa_res_key and generate UA
7731			 * for these registrants(Registrations
7732			 * Preempted) if it wasn't an exclusive
7733			 * reservation generate UA(Reservations
7734			 * Preempted) for all other registered nexuses
7735			 * if the type has changed. Establish the new
7736			 * reservation and holder. If res_key and
7737			 * sa_res_key are the same do the above
7738			 * except don't unregister the res holder.
7739			 */
7740
7741			/*
7742			 * Temporarily unregister so it won't get
7743			 * removed or UA generated
7744			 */
7745			lun->per_res[residx].registered = 0;
7746			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7747				if (lun->per_res[i].registered == 0)
7748					continue;
7749
7750				if (memcmp(param->serv_act_res_key,
7751				    lun->per_res[i].res_key.key,
7752				    sizeof(struct scsi_per_res_key)) == 0) {
7753					lun->per_res[i].registered = 0;
7754					memset(&lun->per_res[i].res_key,
7755					       0,
7756					       sizeof(struct scsi_per_res_key));
7757					lun->pr_key_count--;
7758
7759					if (!persis_offset
7760					 && i < CTL_MAX_INITIATORS)
7761						lun->pending_sense[i
7762							].ua_pending |=
7763							CTL_UA_REG_PREEMPT;
7764					else if (persis_offset
7765					      && i >= persis_offset)
7766						lun->pending_sense[
7767						  i-persis_offset].ua_pending |=
7768						  CTL_UA_REG_PREEMPT;
7769				} else if (type != lun->res_type
7770					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7771					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7772						if (!persis_offset
7773						 && i < CTL_MAX_INITIATORS)
7774							lun->pending_sense[i
7775							].ua_pending |=
7776							CTL_UA_RES_RELEASE;
7777						else if (persis_offset
7778						      && i >= persis_offset)
7779							lun->pending_sense[
7780							i-persis_offset
7781							].ua_pending |=
7782							CTL_UA_RES_RELEASE;
7783				}
7784			}
7785			lun->per_res[residx].registered = 1;
7786			lun->res_type = type;
7787			if (lun->res_type != SPR_TYPE_WR_EX_AR
7788			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7789				lun->pr_res_idx = residx;
7790			else
7791				lun->pr_res_idx =
7792					CTL_PR_ALL_REGISTRANTS;
7793
7794			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7795			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7796			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7797			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7798			persis_io.pr.pr_info.res_type = type;
7799			memcpy(persis_io.pr.pr_info.sa_res_key,
7800			       param->serv_act_res_key,
7801			       sizeof(param->serv_act_res_key));
7802			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7803			     &persis_io, sizeof(persis_io), 0)) >
7804			     CTL_HA_STATUS_SUCCESS) {
7805				printf("CTL:Persis Out error returned "
7806				       "from ctl_ha_msg_send %d\n",
7807				       isc_retval);
7808			}
7809		} else {
7810			/*
7811			 * sa_res_key is not the res holder just
7812			 * remove registrants
7813			 */
7814			int found=0;
7815
7816			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7817				if (memcmp(param->serv_act_res_key,
7818				    lun->per_res[i].res_key.key,
7819				    sizeof(struct scsi_per_res_key)) != 0)
7820					continue;
7821
7822				found = 1;
7823				lun->per_res[i].registered = 0;
7824				memset(&lun->per_res[i].res_key, 0,
7825				       sizeof(struct scsi_per_res_key));
7826				lun->pr_key_count--;
7827
7828				if (!persis_offset
7829				 && i < CTL_MAX_INITIATORS)
7830					lun->pending_sense[i].ua_pending |=
7831						CTL_UA_REG_PREEMPT;
7832				else if (persis_offset
7833				      && i >= persis_offset)
7834					lun->pending_sense[
7835						i-persis_offset].ua_pending |=
7836						CTL_UA_REG_PREEMPT;
7837			}
7838
7839			if (!found) {
7840				mtx_unlock(&lun->lun_lock);
7841				free(ctsio->kern_data_ptr, M_CTL);
7842				ctl_set_reservation_conflict(ctsio);
7843				ctl_done((union ctl_io *)ctsio);
7844		        	return (1);
7845			}
7846			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7847			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7848			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7849			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7850			persis_io.pr.pr_info.res_type = type;
7851			memcpy(persis_io.pr.pr_info.sa_res_key,
7852			       param->serv_act_res_key,
7853			       sizeof(param->serv_act_res_key));
7854			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7855			     &persis_io, sizeof(persis_io), 0)) >
7856			     CTL_HA_STATUS_SUCCESS) {
7857				printf("CTL:Persis Out error returned "
7858				       "from ctl_ha_msg_send %d\n",
7859				isc_retval);
7860			}
7861		}
7862	}
7863
7864	lun->PRGeneration++;
7865	mtx_unlock(&lun->lun_lock);
7866
7867	return (retval);
7868}
7869
7870static void
7871ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7872{
7873	int i;
7874
7875	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7876	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7877	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7878		   msg->pr.pr_info.sa_res_key,
7879		   sizeof(struct scsi_per_res_key)) != 0) {
7880		uint64_t sa_res_key;
7881		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7882
7883		if (sa_res_key == 0) {
7884			/* temporarily unregister this nexus */
7885			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7886
7887			/*
7888			 * Unregister everybody else and build UA for
7889			 * them
7890			 */
7891			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7892				if (lun->per_res[i].registered == 0)
7893					continue;
7894
7895				if (!persis_offset
7896				 && i < CTL_MAX_INITIATORS)
7897					lun->pending_sense[i].ua_pending |=
7898						CTL_UA_REG_PREEMPT;
7899				else if (persis_offset && i >= persis_offset)
7900					lun->pending_sense[i -
7901						persis_offset].ua_pending |=
7902						CTL_UA_REG_PREEMPT;
7903				lun->per_res[i].registered = 0;
7904				memset(&lun->per_res[i].res_key, 0,
7905				       sizeof(struct scsi_per_res_key));
7906			}
7907
7908			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7909			lun->pr_key_count = 1;
7910			lun->res_type = msg->pr.pr_info.res_type;
7911			if (lun->res_type != SPR_TYPE_WR_EX_AR
7912			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7913				lun->pr_res_idx = msg->pr.pr_info.residx;
7914		} else {
7915		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7916				if (memcmp(msg->pr.pr_info.sa_res_key,
7917		                   lun->per_res[i].res_key.key,
7918		                   sizeof(struct scsi_per_res_key)) != 0)
7919					continue;
7920
7921				lun->per_res[i].registered = 0;
7922				memset(&lun->per_res[i].res_key, 0,
7923				       sizeof(struct scsi_per_res_key));
7924				lun->pr_key_count--;
7925
7926				if (!persis_offset
7927				 && i < persis_offset)
7928					lun->pending_sense[i].ua_pending |=
7929						CTL_UA_REG_PREEMPT;
7930				else if (persis_offset
7931				      && i >= persis_offset)
7932					lun->pending_sense[i -
7933						persis_offset].ua_pending |=
7934						CTL_UA_REG_PREEMPT;
7935			}
7936		}
7937	} else {
7938		/*
7939		 * Temporarily unregister so it won't get removed
7940		 * or UA generated
7941		 */
7942		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7943		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7944			if (lun->per_res[i].registered == 0)
7945				continue;
7946
7947			if (memcmp(msg->pr.pr_info.sa_res_key,
7948	                   lun->per_res[i].res_key.key,
7949	                   sizeof(struct scsi_per_res_key)) == 0) {
7950				lun->per_res[i].registered = 0;
7951				memset(&lun->per_res[i].res_key, 0,
7952				       sizeof(struct scsi_per_res_key));
7953				lun->pr_key_count--;
7954				if (!persis_offset
7955				 && i < CTL_MAX_INITIATORS)
7956					lun->pending_sense[i].ua_pending |=
7957						CTL_UA_REG_PREEMPT;
7958				else if (persis_offset
7959				      && i >= persis_offset)
7960					lun->pending_sense[i -
7961						persis_offset].ua_pending |=
7962						CTL_UA_REG_PREEMPT;
7963			} else if (msg->pr.pr_info.res_type != lun->res_type
7964				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7965				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7966					if (!persis_offset
7967					 && i < persis_offset)
7968						lun->pending_sense[i
7969							].ua_pending |=
7970							CTL_UA_RES_RELEASE;
7971					else if (persis_offset
7972					      && i >= persis_offset)
7973					lun->pending_sense[i -
7974						persis_offset].ua_pending |=
7975						CTL_UA_RES_RELEASE;
7976			}
7977		}
7978		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7979		lun->res_type = msg->pr.pr_info.res_type;
7980		if (lun->res_type != SPR_TYPE_WR_EX_AR
7981		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7982			lun->pr_res_idx = msg->pr.pr_info.residx;
7983		else
7984			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7985	}
7986	lun->PRGeneration++;
7987
7988}
7989
7990
7991int
7992ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7993{
7994	int retval;
7995	int isc_retval;
7996	u_int32_t param_len;
7997	struct scsi_per_res_out *cdb;
7998	struct ctl_lun *lun;
7999	struct scsi_per_res_out_parms* param;
8000	struct ctl_softc *softc;
8001	uint32_t residx;
8002	uint64_t res_key, sa_res_key;
8003	uint8_t type;
8004	union ctl_ha_msg persis_io;
8005	int    i;
8006
8007	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8008
8009	retval = CTL_RETVAL_COMPLETE;
8010
8011	softc = control_softc;
8012
8013	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8014	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8015
8016	/*
8017	 * We only support whole-LUN scope.  The scope & type are ignored for
8018	 * register, register and ignore existing key and clear.
8019	 * We sometimes ignore scope and type on preempts too!!
8020	 * Verify reservation type here as well.
8021	 */
8022	type = cdb->scope_type & SPR_TYPE_MASK;
8023	if ((cdb->action == SPRO_RESERVE)
8024	 || (cdb->action == SPRO_RELEASE)) {
8025		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8026			ctl_set_invalid_field(/*ctsio*/ ctsio,
8027					      /*sks_valid*/ 1,
8028					      /*command*/ 1,
8029					      /*field*/ 2,
8030					      /*bit_valid*/ 1,
8031					      /*bit*/ 4);
8032			ctl_done((union ctl_io *)ctsio);
8033			return (CTL_RETVAL_COMPLETE);
8034		}
8035
8036		if (type>8 || type==2 || type==4 || type==0) {
8037			ctl_set_invalid_field(/*ctsio*/ ctsio,
8038					      /*sks_valid*/ 1,
8039					      /*command*/ 1,
8040					      /*field*/ 2,
8041					      /*bit_valid*/ 1,
8042					      /*bit*/ 0);
8043			ctl_done((union ctl_io *)ctsio);
8044			return (CTL_RETVAL_COMPLETE);
8045		}
8046	}
8047
8048	switch (cdb->action & SPRO_ACTION_MASK) {
8049	case SPRO_REGISTER:
8050	case SPRO_RESERVE:
8051	case SPRO_RELEASE:
8052	case SPRO_CLEAR:
8053	case SPRO_PREEMPT:
8054	case SPRO_REG_IGNO:
8055		break;
8056	case SPRO_REG_MOVE:
8057	case SPRO_PRE_ABO:
8058	default:
8059		ctl_set_invalid_field(/*ctsio*/ ctsio,
8060				      /*sks_valid*/ 1,
8061				      /*command*/ 1,
8062				      /*field*/ 1,
8063				      /*bit_valid*/ 1,
8064				      /*bit*/ 0);
8065		ctl_done((union ctl_io *)ctsio);
8066		return (CTL_RETVAL_COMPLETE);
8067		break; /* NOTREACHED */
8068	}
8069
8070	param_len = scsi_4btoul(cdb->length);
8071
8072	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8073		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8074		ctsio->kern_data_len = param_len;
8075		ctsio->kern_total_len = param_len;
8076		ctsio->kern_data_resid = 0;
8077		ctsio->kern_rel_offset = 0;
8078		ctsio->kern_sg_entries = 0;
8079		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8080		ctsio->be_move_done = ctl_config_move_done;
8081		ctl_datamove((union ctl_io *)ctsio);
8082
8083		return (CTL_RETVAL_COMPLETE);
8084	}
8085
8086	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8087
8088	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8089	res_key = scsi_8btou64(param->res_key.key);
8090	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8091
8092	/*
8093	 * Validate the reservation key here except for SPRO_REG_IGNO
8094	 * This must be done for all other service actions
8095	 */
8096	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8097		mtx_lock(&lun->lun_lock);
8098		if (lun->per_res[residx].registered) {
8099		    if (memcmp(param->res_key.key,
8100			       lun->per_res[residx].res_key.key,
8101			       ctl_min(sizeof(param->res_key),
8102			       sizeof(lun->per_res[residx].res_key))) != 0) {
8103				/*
8104				 * The current key passed in doesn't match
8105				 * the one the initiator previously
8106				 * registered.
8107				 */
8108				mtx_unlock(&lun->lun_lock);
8109				free(ctsio->kern_data_ptr, M_CTL);
8110				ctl_set_reservation_conflict(ctsio);
8111				ctl_done((union ctl_io *)ctsio);
8112				return (CTL_RETVAL_COMPLETE);
8113			}
8114		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8115			/*
8116			 * We are not registered
8117			 */
8118			mtx_unlock(&lun->lun_lock);
8119			free(ctsio->kern_data_ptr, M_CTL);
8120			ctl_set_reservation_conflict(ctsio);
8121			ctl_done((union ctl_io *)ctsio);
8122			return (CTL_RETVAL_COMPLETE);
8123		} else if (res_key != 0) {
8124			/*
8125			 * We are not registered and trying to register but
8126			 * the register key isn't zero.
8127			 */
8128			mtx_unlock(&lun->lun_lock);
8129			free(ctsio->kern_data_ptr, M_CTL);
8130			ctl_set_reservation_conflict(ctsio);
8131			ctl_done((union ctl_io *)ctsio);
8132			return (CTL_RETVAL_COMPLETE);
8133		}
8134		mtx_unlock(&lun->lun_lock);
8135	}
8136
8137	switch (cdb->action & SPRO_ACTION_MASK) {
8138	case SPRO_REGISTER:
8139	case SPRO_REG_IGNO: {
8140
8141#if 0
8142		printf("Registration received\n");
8143#endif
8144
8145		/*
8146		 * We don't support any of these options, as we report in
8147		 * the read capabilities request (see
8148		 * ctl_persistent_reserve_in(), above).
8149		 */
8150		if ((param->flags & SPR_SPEC_I_PT)
8151		 || (param->flags & SPR_ALL_TG_PT)
8152		 || (param->flags & SPR_APTPL)) {
8153			int bit_ptr;
8154
8155			if (param->flags & SPR_APTPL)
8156				bit_ptr = 0;
8157			else if (param->flags & SPR_ALL_TG_PT)
8158				bit_ptr = 2;
8159			else /* SPR_SPEC_I_PT */
8160				bit_ptr = 3;
8161
8162			free(ctsio->kern_data_ptr, M_CTL);
8163			ctl_set_invalid_field(ctsio,
8164					      /*sks_valid*/ 1,
8165					      /*command*/ 0,
8166					      /*field*/ 20,
8167					      /*bit_valid*/ 1,
8168					      /*bit*/ bit_ptr);
8169			ctl_done((union ctl_io *)ctsio);
8170			return (CTL_RETVAL_COMPLETE);
8171		}
8172
8173		mtx_lock(&lun->lun_lock);
8174
8175		/*
8176		 * The initiator wants to clear the
8177		 * key/unregister.
8178		 */
8179		if (sa_res_key == 0) {
8180			if ((res_key == 0
8181			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8182			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8183			  && !lun->per_res[residx].registered)) {
8184				mtx_unlock(&lun->lun_lock);
8185				goto done;
8186			}
8187
8188			lun->per_res[residx].registered = 0;
8189			memset(&lun->per_res[residx].res_key,
8190			       0, sizeof(lun->per_res[residx].res_key));
8191			lun->pr_key_count--;
8192
8193			if (residx == lun->pr_res_idx) {
8194				lun->flags &= ~CTL_LUN_PR_RESERVED;
8195				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8196
8197				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8198				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8199				 && lun->pr_key_count) {
8200					/*
8201					 * If the reservation is a registrants
8202					 * only type we need to generate a UA
8203					 * for other registered inits.  The
8204					 * sense code should be RESERVATIONS
8205					 * RELEASED
8206					 */
8207
8208					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8209						if (lun->per_res[
8210						    i+persis_offset].registered
8211						    == 0)
8212							continue;
8213						lun->pending_sense[i
8214							].ua_pending |=
8215							CTL_UA_RES_RELEASE;
8216					}
8217				}
8218				lun->res_type = 0;
8219			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8220				if (lun->pr_key_count==0) {
8221					lun->flags &= ~CTL_LUN_PR_RESERVED;
8222					lun->res_type = 0;
8223					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8224				}
8225			}
8226			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8227			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8228			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8229			persis_io.pr.pr_info.residx = residx;
8230			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8231			     &persis_io, sizeof(persis_io), 0 )) >
8232			     CTL_HA_STATUS_SUCCESS) {
8233				printf("CTL:Persis Out error returned from "
8234				       "ctl_ha_msg_send %d\n", isc_retval);
8235			}
8236		} else /* sa_res_key != 0 */ {
8237
8238			/*
8239			 * If we aren't registered currently then increment
8240			 * the key count and set the registered flag.
8241			 */
8242			if (!lun->per_res[residx].registered) {
8243				lun->pr_key_count++;
8244				lun->per_res[residx].registered = 1;
8245			}
8246
8247			memcpy(&lun->per_res[residx].res_key,
8248			       param->serv_act_res_key,
8249			       ctl_min(sizeof(param->serv_act_res_key),
8250			       sizeof(lun->per_res[residx].res_key)));
8251
8252			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8253			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8254			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8255			persis_io.pr.pr_info.residx = residx;
8256			memcpy(persis_io.pr.pr_info.sa_res_key,
8257			       param->serv_act_res_key,
8258			       sizeof(param->serv_act_res_key));
8259			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8260			     &persis_io, sizeof(persis_io), 0)) >
8261			     CTL_HA_STATUS_SUCCESS) {
8262				printf("CTL:Persis Out error returned from "
8263				       "ctl_ha_msg_send %d\n", isc_retval);
8264			}
8265		}
8266		lun->PRGeneration++;
8267		mtx_unlock(&lun->lun_lock);
8268
8269		break;
8270	}
8271	case SPRO_RESERVE:
8272#if 0
8273                printf("Reserve executed type %d\n", type);
8274#endif
8275		mtx_lock(&lun->lun_lock);
8276		if (lun->flags & CTL_LUN_PR_RESERVED) {
8277			/*
8278			 * if this isn't the reservation holder and it's
8279			 * not a "all registrants" type or if the type is
8280			 * different then we have a conflict
8281			 */
8282			if ((lun->pr_res_idx != residx
8283			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8284			 || lun->res_type != type) {
8285				mtx_unlock(&lun->lun_lock);
8286				free(ctsio->kern_data_ptr, M_CTL);
8287				ctl_set_reservation_conflict(ctsio);
8288				ctl_done((union ctl_io *)ctsio);
8289				return (CTL_RETVAL_COMPLETE);
8290			}
8291			mtx_unlock(&lun->lun_lock);
8292		} else /* create a reservation */ {
8293			/*
8294			 * If it's not an "all registrants" type record
8295			 * reservation holder
8296			 */
8297			if (type != SPR_TYPE_WR_EX_AR
8298			 && type != SPR_TYPE_EX_AC_AR)
8299				lun->pr_res_idx = residx; /* Res holder */
8300			else
8301				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8302
8303			lun->flags |= CTL_LUN_PR_RESERVED;
8304			lun->res_type = type;
8305
8306			mtx_unlock(&lun->lun_lock);
8307
8308			/* send msg to other side */
8309			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8310			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8311			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8312			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8313			persis_io.pr.pr_info.res_type = type;
8314			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8315			     &persis_io, sizeof(persis_io), 0)) >
8316			     CTL_HA_STATUS_SUCCESS) {
8317				printf("CTL:Persis Out error returned from "
8318				       "ctl_ha_msg_send %d\n", isc_retval);
8319			}
8320		}
8321		break;
8322
8323	case SPRO_RELEASE:
8324		mtx_lock(&lun->lun_lock);
8325		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8326			/* No reservation exists return good status */
8327			mtx_unlock(&lun->lun_lock);
8328			goto done;
8329		}
8330		/*
8331		 * Is this nexus a reservation holder?
8332		 */
8333		if (lun->pr_res_idx != residx
8334		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8335			/*
8336			 * not a res holder return good status but
8337			 * do nothing
8338			 */
8339			mtx_unlock(&lun->lun_lock);
8340			goto done;
8341		}
8342
8343		if (lun->res_type != type) {
8344			mtx_unlock(&lun->lun_lock);
8345			free(ctsio->kern_data_ptr, M_CTL);
8346			ctl_set_illegal_pr_release(ctsio);
8347			ctl_done((union ctl_io *)ctsio);
8348			return (CTL_RETVAL_COMPLETE);
8349		}
8350
8351		/* okay to release */
8352		lun->flags &= ~CTL_LUN_PR_RESERVED;
8353		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8354		lun->res_type = 0;
8355
8356		/*
8357		 * if this isn't an exclusive access
8358		 * res generate UA for all other
8359		 * registrants.
8360		 */
8361		if (type != SPR_TYPE_EX_AC
8362		 && type != SPR_TYPE_WR_EX) {
8363			/*
8364			 * temporarily unregister so we don't generate UA
8365			 */
8366			lun->per_res[residx].registered = 0;
8367
8368			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8369				if (lun->per_res[i+persis_offset].registered
8370				    == 0)
8371					continue;
8372				lun->pending_sense[i].ua_pending |=
8373					CTL_UA_RES_RELEASE;
8374			}
8375
8376			lun->per_res[residx].registered = 1;
8377		}
8378		mtx_unlock(&lun->lun_lock);
8379		/* Send msg to other side */
8380		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8381		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8382		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8383		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8384		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8385			printf("CTL:Persis Out error returned from "
8386			       "ctl_ha_msg_send %d\n", isc_retval);
8387		}
8388		break;
8389
8390	case SPRO_CLEAR:
8391		/* send msg to other side */
8392
8393		mtx_lock(&lun->lun_lock);
8394		lun->flags &= ~CTL_LUN_PR_RESERVED;
8395		lun->res_type = 0;
8396		lun->pr_key_count = 0;
8397		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8398
8399
8400		memset(&lun->per_res[residx].res_key,
8401		       0, sizeof(lun->per_res[residx].res_key));
8402		lun->per_res[residx].registered = 0;
8403
8404		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8405			if (lun->per_res[i].registered) {
8406				if (!persis_offset && i < CTL_MAX_INITIATORS)
8407					lun->pending_sense[i].ua_pending |=
8408						CTL_UA_RES_PREEMPT;
8409				else if (persis_offset && i >= persis_offset)
8410					lun->pending_sense[i-persis_offset
8411					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8412
8413				memset(&lun->per_res[i].res_key,
8414				       0, sizeof(struct scsi_per_res_key));
8415				lun->per_res[i].registered = 0;
8416			}
8417		lun->PRGeneration++;
8418		mtx_unlock(&lun->lun_lock);
8419		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8420		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8421		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8422		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8423		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8424			printf("CTL:Persis Out error returned from "
8425			       "ctl_ha_msg_send %d\n", isc_retval);
8426		}
8427		break;
8428
8429	case SPRO_PREEMPT: {
8430		int nretval;
8431
8432		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8433					  residx, ctsio, cdb, param);
8434		if (nretval != 0)
8435			return (CTL_RETVAL_COMPLETE);
8436		break;
8437	}
8438	case SPRO_REG_MOVE:
8439	case SPRO_PRE_ABO:
8440	default:
8441		free(ctsio->kern_data_ptr, M_CTL);
8442		ctl_set_invalid_field(/*ctsio*/ ctsio,
8443				      /*sks_valid*/ 1,
8444				      /*command*/ 1,
8445				      /*field*/ 1,
8446				      /*bit_valid*/ 1,
8447				      /*bit*/ 0);
8448		ctl_done((union ctl_io *)ctsio);
8449		return (CTL_RETVAL_COMPLETE);
8450		break; /* NOTREACHED */
8451	}
8452
8453done:
8454	free(ctsio->kern_data_ptr, M_CTL);
8455	ctl_set_success(ctsio);
8456	ctl_done((union ctl_io *)ctsio);
8457
8458	return (retval);
8459}
8460
8461/*
8462 * This routine is for handling a message from the other SC pertaining to
8463 * persistent reserve out. All the error checking will have been done
8464 * so only perorming the action need be done here to keep the two
8465 * in sync.
8466 */
8467static void
8468ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8469{
8470	struct ctl_lun *lun;
8471	struct ctl_softc *softc;
8472	int i;
8473	uint32_t targ_lun;
8474
8475	softc = control_softc;
8476
8477	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8478	lun = softc->ctl_luns[targ_lun];
8479	mtx_lock(&lun->lun_lock);
8480	switch(msg->pr.pr_info.action) {
8481	case CTL_PR_REG_KEY:
8482		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8483			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8484			lun->pr_key_count++;
8485		}
8486		lun->PRGeneration++;
8487		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8488		       msg->pr.pr_info.sa_res_key,
8489		       sizeof(struct scsi_per_res_key));
8490		break;
8491
8492	case CTL_PR_UNREG_KEY:
8493		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8494		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8495		       0, sizeof(struct scsi_per_res_key));
8496		lun->pr_key_count--;
8497
8498		/* XXX Need to see if the reservation has been released */
8499		/* if so do we need to generate UA? */
8500		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8501			lun->flags &= ~CTL_LUN_PR_RESERVED;
8502			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8503
8504			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8505			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8506			 && lun->pr_key_count) {
8507				/*
8508				 * If the reservation is a registrants
8509				 * only type we need to generate a UA
8510				 * for other registered inits.  The
8511				 * sense code should be RESERVATIONS
8512				 * RELEASED
8513				 */
8514
8515				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8516					if (lun->per_res[i+
8517					    persis_offset].registered == 0)
8518						continue;
8519
8520					lun->pending_sense[i
8521						].ua_pending |=
8522						CTL_UA_RES_RELEASE;
8523				}
8524			}
8525			lun->res_type = 0;
8526		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8527			if (lun->pr_key_count==0) {
8528				lun->flags &= ~CTL_LUN_PR_RESERVED;
8529				lun->res_type = 0;
8530				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8531			}
8532		}
8533		lun->PRGeneration++;
8534		break;
8535
8536	case CTL_PR_RESERVE:
8537		lun->flags |= CTL_LUN_PR_RESERVED;
8538		lun->res_type = msg->pr.pr_info.res_type;
8539		lun->pr_res_idx = msg->pr.pr_info.residx;
8540
8541		break;
8542
8543	case CTL_PR_RELEASE:
8544		/*
8545		 * if this isn't an exclusive access res generate UA for all
8546		 * other registrants.
8547		 */
8548		if (lun->res_type != SPR_TYPE_EX_AC
8549		 && lun->res_type != SPR_TYPE_WR_EX) {
8550			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8551				if (lun->per_res[i+persis_offset].registered)
8552					lun->pending_sense[i].ua_pending |=
8553						CTL_UA_RES_RELEASE;
8554		}
8555
8556		lun->flags &= ~CTL_LUN_PR_RESERVED;
8557		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8558		lun->res_type = 0;
8559		break;
8560
8561	case CTL_PR_PREEMPT:
8562		ctl_pro_preempt_other(lun, msg);
8563		break;
8564	case CTL_PR_CLEAR:
8565		lun->flags &= ~CTL_LUN_PR_RESERVED;
8566		lun->res_type = 0;
8567		lun->pr_key_count = 0;
8568		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8569
8570		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8571			if (lun->per_res[i].registered == 0)
8572				continue;
8573			if (!persis_offset
8574			 && i < CTL_MAX_INITIATORS)
8575				lun->pending_sense[i].ua_pending |=
8576					CTL_UA_RES_PREEMPT;
8577			else if (persis_offset
8578			      && i >= persis_offset)
8579   				lun->pending_sense[i-persis_offset].ua_pending|=
8580					CTL_UA_RES_PREEMPT;
8581			memset(&lun->per_res[i].res_key, 0,
8582			       sizeof(struct scsi_per_res_key));
8583			lun->per_res[i].registered = 0;
8584		}
8585		lun->PRGeneration++;
8586		break;
8587	}
8588
8589	mtx_unlock(&lun->lun_lock);
8590}
8591
8592int
8593ctl_read_write(struct ctl_scsiio *ctsio)
8594{
8595	struct ctl_lun *lun;
8596	struct ctl_lba_len_flags *lbalen;
8597	uint64_t lba;
8598	uint32_t num_blocks;
8599	int reladdr, fua, dpo, ebp;
8600	int retval;
8601	int isread;
8602
8603	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8604
8605	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8606
8607	reladdr = 0;
8608	fua = 0;
8609	dpo = 0;
8610	ebp = 0;
8611
8612	retval = CTL_RETVAL_COMPLETE;
8613
8614	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8615	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8616	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8617		uint32_t residx;
8618
8619		/*
8620		 * XXX KDM need a lock here.
8621		 */
8622		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8623		if ((lun->res_type == SPR_TYPE_EX_AC
8624		  && residx != lun->pr_res_idx)
8625		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8626		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8627		  && !lun->per_res[residx].registered)) {
8628			ctl_set_reservation_conflict(ctsio);
8629			ctl_done((union ctl_io *)ctsio);
8630			return (CTL_RETVAL_COMPLETE);
8631	        }
8632	}
8633
8634	switch (ctsio->cdb[0]) {
8635	case READ_6:
8636	case WRITE_6: {
8637		struct scsi_rw_6 *cdb;
8638
8639		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8640
8641		lba = scsi_3btoul(cdb->addr);
8642		/* only 5 bits are valid in the most significant address byte */
8643		lba &= 0x1fffff;
8644		num_blocks = cdb->length;
8645		/*
8646		 * This is correct according to SBC-2.
8647		 */
8648		if (num_blocks == 0)
8649			num_blocks = 256;
8650		break;
8651	}
8652	case READ_10:
8653	case WRITE_10: {
8654		struct scsi_rw_10 *cdb;
8655
8656		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8657
8658		if (cdb->byte2 & SRW10_RELADDR)
8659			reladdr = 1;
8660		if (cdb->byte2 & SRW10_FUA)
8661			fua = 1;
8662		if (cdb->byte2 & SRW10_DPO)
8663			dpo = 1;
8664
8665		if ((cdb->opcode == WRITE_10)
8666		 && (cdb->byte2 & SRW10_EBP))
8667			ebp = 1;
8668
8669		lba = scsi_4btoul(cdb->addr);
8670		num_blocks = scsi_2btoul(cdb->length);
8671		break;
8672	}
8673	case WRITE_VERIFY_10: {
8674		struct scsi_write_verify_10 *cdb;
8675
8676		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8677
8678		/*
8679		 * XXX KDM we should do actual write verify support at some
8680		 * point.  This is obviously fake, we're just translating
8681		 * things to a write.  So we don't even bother checking the
8682		 * BYTCHK field, since we don't do any verification.  If
8683		 * the user asks for it, we'll just pretend we did it.
8684		 */
8685		if (cdb->byte2 & SWV_DPO)
8686			dpo = 1;
8687
8688		lba = scsi_4btoul(cdb->addr);
8689		num_blocks = scsi_2btoul(cdb->length);
8690		break;
8691	}
8692	case READ_12:
8693	case WRITE_12: {
8694		struct scsi_rw_12 *cdb;
8695
8696		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8697
8698		if (cdb->byte2 & SRW12_RELADDR)
8699			reladdr = 1;
8700		if (cdb->byte2 & SRW12_FUA)
8701			fua = 1;
8702		if (cdb->byte2 & SRW12_DPO)
8703			dpo = 1;
8704		lba = scsi_4btoul(cdb->addr);
8705		num_blocks = scsi_4btoul(cdb->length);
8706		break;
8707	}
8708	case WRITE_VERIFY_12: {
8709		struct scsi_write_verify_12 *cdb;
8710
8711		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8712
8713		if (cdb->byte2 & SWV_DPO)
8714			dpo = 1;
8715
8716		lba = scsi_4btoul(cdb->addr);
8717		num_blocks = scsi_4btoul(cdb->length);
8718
8719		break;
8720	}
8721	case READ_16:
8722	case WRITE_16: {
8723		struct scsi_rw_16 *cdb;
8724
8725		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8726
8727		if (cdb->byte2 & SRW12_RELADDR)
8728			reladdr = 1;
8729		if (cdb->byte2 & SRW12_FUA)
8730			fua = 1;
8731		if (cdb->byte2 & SRW12_DPO)
8732			dpo = 1;
8733
8734		lba = scsi_8btou64(cdb->addr);
8735		num_blocks = scsi_4btoul(cdb->length);
8736		break;
8737	}
8738	case WRITE_VERIFY_16: {
8739		struct scsi_write_verify_16 *cdb;
8740
8741		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8742
8743		if (cdb->byte2 & SWV_DPO)
8744			dpo = 1;
8745
8746		lba = scsi_8btou64(cdb->addr);
8747		num_blocks = scsi_4btoul(cdb->length);
8748		break;
8749	}
8750	default:
8751		/*
8752		 * We got a command we don't support.  This shouldn't
8753		 * happen, commands should be filtered out above us.
8754		 */
8755		ctl_set_invalid_opcode(ctsio);
8756		ctl_done((union ctl_io *)ctsio);
8757
8758		return (CTL_RETVAL_COMPLETE);
8759		break; /* NOTREACHED */
8760	}
8761
8762	/*
8763	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8764	 * interesting for us, but if RAIDCore is in write-back mode,
8765	 * getting it to do write-through for a particular transaction may
8766	 * not be possible.
8767	 */
8768	/*
8769	 * We don't support relative addressing.  That also requires
8770	 * supporting linked commands, which we don't do.
8771	 */
8772	if (reladdr != 0) {
8773		ctl_set_invalid_field(ctsio,
8774				      /*sks_valid*/ 1,
8775				      /*command*/ 1,
8776				      /*field*/ 1,
8777				      /*bit_valid*/ 1,
8778				      /*bit*/ 0);
8779		ctl_done((union ctl_io *)ctsio);
8780		return (CTL_RETVAL_COMPLETE);
8781	}
8782
8783	/*
8784	 * The first check is to make sure we're in bounds, the second
8785	 * check is to catch wrap-around problems.  If the lba + num blocks
8786	 * is less than the lba, then we've wrapped around and the block
8787	 * range is invalid anyway.
8788	 */
8789	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8790	 || ((lba + num_blocks) < lba)) {
8791		ctl_set_lba_out_of_range(ctsio);
8792		ctl_done((union ctl_io *)ctsio);
8793		return (CTL_RETVAL_COMPLETE);
8794	}
8795
8796	/*
8797	 * According to SBC-3, a transfer length of 0 is not an error.
8798	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8799	 * translates to 256 blocks for those commands.
8800	 */
8801	if (num_blocks == 0) {
8802		ctl_set_success(ctsio);
8803		ctl_done((union ctl_io *)ctsio);
8804		return (CTL_RETVAL_COMPLETE);
8805	}
8806
8807	lbalen = (struct ctl_lba_len_flags *)
8808	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8809	lbalen->lba = lba;
8810	lbalen->len = num_blocks;
8811	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8812
8813	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8814	ctsio->kern_rel_offset = 0;
8815
8816	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8817
8818	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8819
8820	return (retval);
8821}
8822
8823static int
8824ctl_cnw_cont(union ctl_io *io)
8825{
8826	struct ctl_scsiio *ctsio;
8827	struct ctl_lun *lun;
8828	struct ctl_lba_len_flags *lbalen;
8829	int retval;
8830
8831	ctsio = &io->scsiio;
8832	ctsio->io_hdr.status = CTL_STATUS_NONE;
8833	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8834	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8835	lbalen = (struct ctl_lba_len_flags *)
8836	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8837	lbalen->flags = CTL_LLF_WRITE;
8838
8839	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8840	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8841	return (retval);
8842}
8843
8844int
8845ctl_cnw(struct ctl_scsiio *ctsio)
8846{
8847	struct ctl_lun *lun;
8848	struct ctl_lba_len_flags *lbalen;
8849	uint64_t lba;
8850	uint32_t num_blocks;
8851	int fua, dpo;
8852	int retval;
8853
8854	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8855
8856	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8857
8858	fua = 0;
8859	dpo = 0;
8860
8861	retval = CTL_RETVAL_COMPLETE;
8862
8863	switch (ctsio->cdb[0]) {
8864	case COMPARE_AND_WRITE: {
8865		struct scsi_compare_and_write *cdb;
8866
8867		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8868
8869		if (cdb->byte2 & SRW10_FUA)
8870			fua = 1;
8871		if (cdb->byte2 & SRW10_DPO)
8872			dpo = 1;
8873		lba = scsi_8btou64(cdb->addr);
8874		num_blocks = cdb->length;
8875		break;
8876	}
8877	default:
8878		/*
8879		 * We got a command we don't support.  This shouldn't
8880		 * happen, commands should be filtered out above us.
8881		 */
8882		ctl_set_invalid_opcode(ctsio);
8883		ctl_done((union ctl_io *)ctsio);
8884
8885		return (CTL_RETVAL_COMPLETE);
8886		break; /* NOTREACHED */
8887	}
8888
8889	/*
8890	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8891	 * interesting for us, but if RAIDCore is in write-back mode,
8892	 * getting it to do write-through for a particular transaction may
8893	 * not be possible.
8894	 */
8895
8896	/*
8897	 * The first check is to make sure we're in bounds, the second
8898	 * check is to catch wrap-around problems.  If the lba + num blocks
8899	 * is less than the lba, then we've wrapped around and the block
8900	 * range is invalid anyway.
8901	 */
8902	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8903	 || ((lba + num_blocks) < lba)) {
8904		ctl_set_lba_out_of_range(ctsio);
8905		ctl_done((union ctl_io *)ctsio);
8906		return (CTL_RETVAL_COMPLETE);
8907	}
8908
8909	/*
8910	 * According to SBC-3, a transfer length of 0 is not an error.
8911	 */
8912	if (num_blocks == 0) {
8913		ctl_set_success(ctsio);
8914		ctl_done((union ctl_io *)ctsio);
8915		return (CTL_RETVAL_COMPLETE);
8916	}
8917
8918	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8919	ctsio->kern_rel_offset = 0;
8920
8921	/*
8922	 * Set the IO_CONT flag, so that if this I/O gets passed to
8923	 * ctl_data_submit_done(), it'll get passed back to
8924	 * ctl_ctl_cnw_cont() for further processing.
8925	 */
8926	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8927	ctsio->io_cont = ctl_cnw_cont;
8928
8929	lbalen = (struct ctl_lba_len_flags *)
8930	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8931	lbalen->lba = lba;
8932	lbalen->len = num_blocks;
8933	lbalen->flags = CTL_LLF_COMPARE;
8934
8935	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8936	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8937	return (retval);
8938}
8939
8940int
8941ctl_verify(struct ctl_scsiio *ctsio)
8942{
8943	struct ctl_lun *lun;
8944	struct ctl_lba_len_flags *lbalen;
8945	uint64_t lba;
8946	uint32_t num_blocks;
8947	int bytchk, dpo;
8948	int retval;
8949
8950	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8951
8952	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8953
8954	bytchk = 0;
8955	dpo = 0;
8956	retval = CTL_RETVAL_COMPLETE;
8957
8958	switch (ctsio->cdb[0]) {
8959	case VERIFY_10: {
8960		struct scsi_verify_10 *cdb;
8961
8962		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8963		if (cdb->byte2 & SVFY_BYTCHK)
8964			bytchk = 1;
8965		if (cdb->byte2 & SVFY_DPO)
8966			dpo = 1;
8967		lba = scsi_4btoul(cdb->addr);
8968		num_blocks = scsi_2btoul(cdb->length);
8969		break;
8970	}
8971	case VERIFY_12: {
8972		struct scsi_verify_12 *cdb;
8973
8974		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8975		if (cdb->byte2 & SVFY_BYTCHK)
8976			bytchk = 1;
8977		if (cdb->byte2 & SVFY_DPO)
8978			dpo = 1;
8979		lba = scsi_4btoul(cdb->addr);
8980		num_blocks = scsi_4btoul(cdb->length);
8981		break;
8982	}
8983	case VERIFY_16: {
8984		struct scsi_rw_16 *cdb;
8985
8986		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8987		if (cdb->byte2 & SVFY_BYTCHK)
8988			bytchk = 1;
8989		if (cdb->byte2 & SVFY_DPO)
8990			dpo = 1;
8991		lba = scsi_8btou64(cdb->addr);
8992		num_blocks = scsi_4btoul(cdb->length);
8993		break;
8994	}
8995	default:
8996		/*
8997		 * We got a command we don't support.  This shouldn't
8998		 * happen, commands should be filtered out above us.
8999		 */
9000		ctl_set_invalid_opcode(ctsio);
9001		ctl_done((union ctl_io *)ctsio);
9002		return (CTL_RETVAL_COMPLETE);
9003	}
9004
9005	/*
9006	 * The first check is to make sure we're in bounds, the second
9007	 * check is to catch wrap-around problems.  If the lba + num blocks
9008	 * is less than the lba, then we've wrapped around and the block
9009	 * range is invalid anyway.
9010	 */
9011	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9012	 || ((lba + num_blocks) < lba)) {
9013		ctl_set_lba_out_of_range(ctsio);
9014		ctl_done((union ctl_io *)ctsio);
9015		return (CTL_RETVAL_COMPLETE);
9016	}
9017
9018	/*
9019	 * According to SBC-3, a transfer length of 0 is not an error.
9020	 */
9021	if (num_blocks == 0) {
9022		ctl_set_success(ctsio);
9023		ctl_done((union ctl_io *)ctsio);
9024		return (CTL_RETVAL_COMPLETE);
9025	}
9026
9027	lbalen = (struct ctl_lba_len_flags *)
9028	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9029	lbalen->lba = lba;
9030	lbalen->len = num_blocks;
9031	if (bytchk) {
9032		lbalen->flags = CTL_LLF_COMPARE;
9033		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9034	} else {
9035		lbalen->flags = CTL_LLF_VERIFY;
9036		ctsio->kern_total_len = 0;
9037	}
9038	ctsio->kern_rel_offset = 0;
9039
9040	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9041	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9042	return (retval);
9043}
9044
9045int
9046ctl_report_luns(struct ctl_scsiio *ctsio)
9047{
9048	struct scsi_report_luns *cdb;
9049	struct scsi_report_luns_data *lun_data;
9050	struct ctl_lun *lun, *request_lun;
9051	int num_luns, retval;
9052	uint32_t alloc_len, lun_datalen;
9053	int num_filled, well_known;
9054	uint32_t initidx, targ_lun_id, lun_id;
9055
9056	retval = CTL_RETVAL_COMPLETE;
9057	well_known = 0;
9058
9059	cdb = (struct scsi_report_luns *)ctsio->cdb;
9060
9061	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9062
9063	mtx_lock(&control_softc->ctl_lock);
9064	num_luns = control_softc->num_luns;
9065	mtx_unlock(&control_softc->ctl_lock);
9066
9067	switch (cdb->select_report) {
9068	case RPL_REPORT_DEFAULT:
9069	case RPL_REPORT_ALL:
9070		break;
9071	case RPL_REPORT_WELLKNOWN:
9072		well_known = 1;
9073		num_luns = 0;
9074		break;
9075	default:
9076		ctl_set_invalid_field(ctsio,
9077				      /*sks_valid*/ 1,
9078				      /*command*/ 1,
9079				      /*field*/ 2,
9080				      /*bit_valid*/ 0,
9081				      /*bit*/ 0);
9082		ctl_done((union ctl_io *)ctsio);
9083		return (retval);
9084		break; /* NOTREACHED */
9085	}
9086
9087	alloc_len = scsi_4btoul(cdb->length);
9088	/*
9089	 * The initiator has to allocate at least 16 bytes for this request,
9090	 * so he can at least get the header and the first LUN.  Otherwise
9091	 * we reject the request (per SPC-3 rev 14, section 6.21).
9092	 */
9093	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9094	    sizeof(struct scsi_report_luns_lundata))) {
9095		ctl_set_invalid_field(ctsio,
9096				      /*sks_valid*/ 1,
9097				      /*command*/ 1,
9098				      /*field*/ 6,
9099				      /*bit_valid*/ 0,
9100				      /*bit*/ 0);
9101		ctl_done((union ctl_io *)ctsio);
9102		return (retval);
9103	}
9104
9105	request_lun = (struct ctl_lun *)
9106		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9107
9108	lun_datalen = sizeof(*lun_data) +
9109		(num_luns * sizeof(struct scsi_report_luns_lundata));
9110
9111	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9112	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9113	ctsio->kern_sg_entries = 0;
9114
9115	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9116
9117	mtx_lock(&control_softc->ctl_lock);
9118	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9119		lun_id = targ_lun_id;
9120		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9121			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9122		if (lun_id >= CTL_MAX_LUNS)
9123			continue;
9124		lun = control_softc->ctl_luns[lun_id];
9125		if (lun == NULL)
9126			continue;
9127
9128		if (targ_lun_id <= 0xff) {
9129			/*
9130			 * Peripheral addressing method, bus number 0.
9131			 */
9132			lun_data->luns[num_filled].lundata[0] =
9133				RPL_LUNDATA_ATYP_PERIPH;
9134			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9135			num_filled++;
9136		} else if (targ_lun_id <= 0x3fff) {
9137			/*
9138			 * Flat addressing method.
9139			 */
9140			lun_data->luns[num_filled].lundata[0] =
9141				RPL_LUNDATA_ATYP_FLAT |
9142				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9143#ifdef OLDCTLHEADERS
9144				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9145				(targ_lun_id & SRLD_BUS_LUN_MASK);
9146#endif
9147			lun_data->luns[num_filled].lundata[1] =
9148#ifdef OLDCTLHEADERS
9149				targ_lun_id >> SRLD_BUS_LUN_BITS;
9150#endif
9151				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9152			num_filled++;
9153		} else {
9154			printf("ctl_report_luns: bogus LUN number %jd, "
9155			       "skipping\n", (intmax_t)targ_lun_id);
9156		}
9157		/*
9158		 * According to SPC-3, rev 14 section 6.21:
9159		 *
9160		 * "The execution of a REPORT LUNS command to any valid and
9161		 * installed logical unit shall clear the REPORTED LUNS DATA
9162		 * HAS CHANGED unit attention condition for all logical
9163		 * units of that target with respect to the requesting
9164		 * initiator. A valid and installed logical unit is one
9165		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9166		 * INQUIRY data (see 6.4.2)."
9167		 *
9168		 * If request_lun is NULL, the LUN this report luns command
9169		 * was issued to is either disabled or doesn't exist. In that
9170		 * case, we shouldn't clear any pending lun change unit
9171		 * attention.
9172		 */
9173		if (request_lun != NULL) {
9174			mtx_lock(&lun->lun_lock);
9175			lun->pending_sense[initidx].ua_pending &=
9176				~CTL_UA_LUN_CHANGE;
9177			mtx_unlock(&lun->lun_lock);
9178		}
9179	}
9180	mtx_unlock(&control_softc->ctl_lock);
9181
9182	/*
9183	 * It's quite possible that we've returned fewer LUNs than we allocated
9184	 * space for.  Trim it.
9185	 */
9186	lun_datalen = sizeof(*lun_data) +
9187		(num_filled * sizeof(struct scsi_report_luns_lundata));
9188
9189	if (lun_datalen < alloc_len) {
9190		ctsio->residual = alloc_len - lun_datalen;
9191		ctsio->kern_data_len = lun_datalen;
9192		ctsio->kern_total_len = lun_datalen;
9193	} else {
9194		ctsio->residual = 0;
9195		ctsio->kern_data_len = alloc_len;
9196		ctsio->kern_total_len = alloc_len;
9197	}
9198	ctsio->kern_data_resid = 0;
9199	ctsio->kern_rel_offset = 0;
9200	ctsio->kern_sg_entries = 0;
9201
9202	/*
9203	 * We set this to the actual data length, regardless of how much
9204	 * space we actually have to return results.  If the user looks at
9205	 * this value, he'll know whether or not he allocated enough space
9206	 * and reissue the command if necessary.  We don't support well
9207	 * known logical units, so if the user asks for that, return none.
9208	 */
9209	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9210
9211	/*
9212	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9213	 * this request.
9214	 */
9215	ctsio->scsi_status = SCSI_STATUS_OK;
9216
9217	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9218	ctsio->be_move_done = ctl_config_move_done;
9219	ctl_datamove((union ctl_io *)ctsio);
9220
9221	return (retval);
9222}
9223
9224int
9225ctl_request_sense(struct ctl_scsiio *ctsio)
9226{
9227	struct scsi_request_sense *cdb;
9228	struct scsi_sense_data *sense_ptr;
9229	struct ctl_lun *lun;
9230	uint32_t initidx;
9231	int have_error;
9232	scsi_sense_data_type sense_format;
9233
9234	cdb = (struct scsi_request_sense *)ctsio->cdb;
9235
9236	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9237
9238	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9239
9240	/*
9241	 * Determine which sense format the user wants.
9242	 */
9243	if (cdb->byte2 & SRS_DESC)
9244		sense_format = SSD_TYPE_DESC;
9245	else
9246		sense_format = SSD_TYPE_FIXED;
9247
9248	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9249	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9250	ctsio->kern_sg_entries = 0;
9251
9252	/*
9253	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9254	 * larger than the largest allowed value for the length field in the
9255	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9256	 */
9257	ctsio->residual = 0;
9258	ctsio->kern_data_len = cdb->length;
9259	ctsio->kern_total_len = cdb->length;
9260
9261	ctsio->kern_data_resid = 0;
9262	ctsio->kern_rel_offset = 0;
9263	ctsio->kern_sg_entries = 0;
9264
9265	/*
9266	 * If we don't have a LUN, we don't have any pending sense.
9267	 */
9268	if (lun == NULL)
9269		goto no_sense;
9270
9271	have_error = 0;
9272	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9273	/*
9274	 * Check for pending sense, and then for pending unit attentions.
9275	 * Pending sense gets returned first, then pending unit attentions.
9276	 */
9277	mtx_lock(&lun->lun_lock);
9278	if (ctl_is_set(lun->have_ca, initidx)) {
9279		scsi_sense_data_type stored_format;
9280
9281		/*
9282		 * Check to see which sense format was used for the stored
9283		 * sense data.
9284		 */
9285		stored_format = scsi_sense_type(
9286		    &lun->pending_sense[initidx].sense);
9287
9288		/*
9289		 * If the user requested a different sense format than the
9290		 * one we stored, then we need to convert it to the other
9291		 * format.  If we're going from descriptor to fixed format
9292		 * sense data, we may lose things in translation, depending
9293		 * on what options were used.
9294		 *
9295		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9296		 * for some reason we'll just copy it out as-is.
9297		 */
9298		if ((stored_format == SSD_TYPE_FIXED)
9299		 && (sense_format == SSD_TYPE_DESC))
9300			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9301			    &lun->pending_sense[initidx].sense,
9302			    (struct scsi_sense_data_desc *)sense_ptr);
9303		else if ((stored_format == SSD_TYPE_DESC)
9304		      && (sense_format == SSD_TYPE_FIXED))
9305			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9306			    &lun->pending_sense[initidx].sense,
9307			    (struct scsi_sense_data_fixed *)sense_ptr);
9308		else
9309			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9310			       ctl_min(sizeof(*sense_ptr),
9311			       sizeof(lun->pending_sense[initidx].sense)));
9312
9313		ctl_clear_mask(lun->have_ca, initidx);
9314		have_error = 1;
9315	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9316		ctl_ua_type ua_type;
9317
9318		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9319				       sense_ptr, sense_format);
9320		if (ua_type != CTL_UA_NONE) {
9321			have_error = 1;
9322			/* We're reporting this UA, so clear it */
9323			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9324		}
9325	}
9326	mtx_unlock(&lun->lun_lock);
9327
9328	/*
9329	 * We already have a pending error, return it.
9330	 */
9331	if (have_error != 0) {
9332		/*
9333		 * We report the SCSI status as OK, since the status of the
9334		 * request sense command itself is OK.
9335		 */
9336		ctsio->scsi_status = SCSI_STATUS_OK;
9337
9338		/*
9339		 * We report 0 for the sense length, because we aren't doing
9340		 * autosense in this case.  We're reporting sense as
9341		 * parameter data.
9342		 */
9343		ctsio->sense_len = 0;
9344		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9345		ctsio->be_move_done = ctl_config_move_done;
9346		ctl_datamove((union ctl_io *)ctsio);
9347
9348		return (CTL_RETVAL_COMPLETE);
9349	}
9350
9351no_sense:
9352
9353	/*
9354	 * No sense information to report, so we report that everything is
9355	 * okay.
9356	 */
9357	ctl_set_sense_data(sense_ptr,
9358			   lun,
9359			   sense_format,
9360			   /*current_error*/ 1,
9361			   /*sense_key*/ SSD_KEY_NO_SENSE,
9362			   /*asc*/ 0x00,
9363			   /*ascq*/ 0x00,
9364			   SSD_ELEM_NONE);
9365
9366	ctsio->scsi_status = SCSI_STATUS_OK;
9367
9368	/*
9369	 * We report 0 for the sense length, because we aren't doing
9370	 * autosense in this case.  We're reporting sense as parameter data.
9371	 */
9372	ctsio->sense_len = 0;
9373	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9374	ctsio->be_move_done = ctl_config_move_done;
9375	ctl_datamove((union ctl_io *)ctsio);
9376
9377	return (CTL_RETVAL_COMPLETE);
9378}
9379
9380int
9381ctl_tur(struct ctl_scsiio *ctsio)
9382{
9383	struct ctl_lun *lun;
9384
9385	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9386
9387	CTL_DEBUG_PRINT(("ctl_tur\n"));
9388
9389	if (lun == NULL)
9390		return (EINVAL);
9391
9392	ctsio->scsi_status = SCSI_STATUS_OK;
9393	ctsio->io_hdr.status = CTL_SUCCESS;
9394
9395	ctl_done((union ctl_io *)ctsio);
9396
9397	return (CTL_RETVAL_COMPLETE);
9398}
9399
9400#ifdef notyet
9401static int
9402ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9403{
9404
9405}
9406#endif
9407
9408static int
9409ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9410{
9411	struct scsi_vpd_supported_pages *pages;
9412	int sup_page_size;
9413	struct ctl_lun *lun;
9414
9415	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9416
9417	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9418	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9419	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9420	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9421	ctsio->kern_sg_entries = 0;
9422
9423	if (sup_page_size < alloc_len) {
9424		ctsio->residual = alloc_len - sup_page_size;
9425		ctsio->kern_data_len = sup_page_size;
9426		ctsio->kern_total_len = sup_page_size;
9427	} else {
9428		ctsio->residual = 0;
9429		ctsio->kern_data_len = alloc_len;
9430		ctsio->kern_total_len = alloc_len;
9431	}
9432	ctsio->kern_data_resid = 0;
9433	ctsio->kern_rel_offset = 0;
9434	ctsio->kern_sg_entries = 0;
9435
9436	/*
9437	 * The control device is always connected.  The disk device, on the
9438	 * other hand, may not be online all the time.  Need to change this
9439	 * to figure out whether the disk device is actually online or not.
9440	 */
9441	if (lun != NULL)
9442		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9443				lun->be_lun->lun_type;
9444	else
9445		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9446
9447	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9448	/* Supported VPD pages */
9449	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9450	/* Serial Number */
9451	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9452	/* Device Identification */
9453	pages->page_list[2] = SVPD_DEVICE_ID;
9454	/* Block limits */
9455	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9456	/* Logical Block Provisioning */
9457	pages->page_list[4] = SVPD_LBP;
9458
9459	ctsio->scsi_status = SCSI_STATUS_OK;
9460
9461	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9462	ctsio->be_move_done = ctl_config_move_done;
9463	ctl_datamove((union ctl_io *)ctsio);
9464
9465	return (CTL_RETVAL_COMPLETE);
9466}
9467
9468static int
9469ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9470{
9471	struct scsi_vpd_unit_serial_number *sn_ptr;
9472	struct ctl_lun *lun;
9473
9474	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9475
9476	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9477	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9478	ctsio->kern_sg_entries = 0;
9479
9480	if (sizeof(*sn_ptr) < alloc_len) {
9481		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9482		ctsio->kern_data_len = sizeof(*sn_ptr);
9483		ctsio->kern_total_len = sizeof(*sn_ptr);
9484	} else {
9485		ctsio->residual = 0;
9486		ctsio->kern_data_len = alloc_len;
9487		ctsio->kern_total_len = alloc_len;
9488	}
9489	ctsio->kern_data_resid = 0;
9490	ctsio->kern_rel_offset = 0;
9491	ctsio->kern_sg_entries = 0;
9492
9493	/*
9494	 * The control device is always connected.  The disk device, on the
9495	 * other hand, may not be online all the time.  Need to change this
9496	 * to figure out whether the disk device is actually online or not.
9497	 */
9498	if (lun != NULL)
9499		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9500				  lun->be_lun->lun_type;
9501	else
9502		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9503
9504	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9505	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9506	/*
9507	 * If we don't have a LUN, we just leave the serial number as
9508	 * all spaces.
9509	 */
9510	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9511	if (lun != NULL) {
9512		strncpy((char *)sn_ptr->serial_num,
9513			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9514	}
9515	ctsio->scsi_status = SCSI_STATUS_OK;
9516
9517	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9518	ctsio->be_move_done = ctl_config_move_done;
9519	ctl_datamove((union ctl_io *)ctsio);
9520
9521	return (CTL_RETVAL_COMPLETE);
9522}
9523
9524
9525static int
9526ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9527{
9528	struct scsi_vpd_device_id *devid_ptr;
9529	struct scsi_vpd_id_descriptor *desc, *desc1;
9530	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9531	struct scsi_vpd_id_t10 *t10id;
9532	struct ctl_softc *ctl_softc;
9533	struct ctl_lun *lun;
9534	struct ctl_frontend *fe;
9535	char *val;
9536	int data_len, devid_len;
9537
9538	ctl_softc = control_softc;
9539
9540	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9541
9542	if (fe->devid != NULL)
9543		return ((fe->devid)(ctsio, alloc_len));
9544
9545	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9546
9547	if (lun == NULL) {
9548		devid_len = CTL_DEVID_MIN_LEN;
9549	} else {
9550		devid_len = max(CTL_DEVID_MIN_LEN,
9551		    strnlen(lun->be_lun->device_id, CTL_DEVID_LEN));
9552	}
9553
9554	data_len = sizeof(struct scsi_vpd_device_id) +
9555		sizeof(struct scsi_vpd_id_descriptor) +
9556		sizeof(struct scsi_vpd_id_t10) + devid_len +
9557		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9558		sizeof(struct scsi_vpd_id_descriptor) +
9559		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9560		sizeof(struct scsi_vpd_id_descriptor) +
9561		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9562
9563	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9564	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9565	ctsio->kern_sg_entries = 0;
9566
9567	if (data_len < alloc_len) {
9568		ctsio->residual = alloc_len - data_len;
9569		ctsio->kern_data_len = data_len;
9570		ctsio->kern_total_len = data_len;
9571	} else {
9572		ctsio->residual = 0;
9573		ctsio->kern_data_len = alloc_len;
9574		ctsio->kern_total_len = alloc_len;
9575	}
9576	ctsio->kern_data_resid = 0;
9577	ctsio->kern_rel_offset = 0;
9578	ctsio->kern_sg_entries = 0;
9579
9580	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9581	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9582	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9583		sizeof(struct scsi_vpd_id_t10) + devid_len);
9584	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9585	          CTL_WWPN_LEN);
9586	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9587	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9588
9589	/*
9590	 * The control device is always connected.  The disk device, on the
9591	 * other hand, may not be online all the time.
9592	 */
9593	if (lun != NULL)
9594		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9595				     lun->be_lun->lun_type;
9596	else
9597		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9598
9599	devid_ptr->page_code = SVPD_DEVICE_ID;
9600
9601	scsi_ulto2b(data_len - 4, devid_ptr->length);
9602
9603	/*
9604	 * For Fibre channel,
9605	 */
9606	if (fe->port_type == CTL_PORT_FC)
9607	{
9608		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9609				      SVPD_ID_CODESET_ASCII;
9610        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9611		              SVPD_ID_CODESET_BINARY;
9612	}
9613	else
9614	{
9615		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9616				      SVPD_ID_CODESET_ASCII;
9617        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9618		              SVPD_ID_CODESET_BINARY;
9619	}
9620	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9621
9622	/*
9623	 * We're using a LUN association here.  i.e., this device ID is a
9624	 * per-LUN identifier.
9625	 */
9626	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9627	desc->length = sizeof(*t10id) + devid_len;
9628	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9629		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9630	} else {
9631		memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9632		strncpy(t10id->vendor, val,
9633		    min(sizeof(t10id->vendor), strlen(val)));
9634	}
9635
9636	/*
9637	 * desc1 is for the WWPN which is a port asscociation.
9638	 */
9639	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9640	desc1->length = CTL_WWPN_LEN;
9641	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9642	/* For testing just create the WWPN */
9643#if 0
9644	ddb_GetWWNN((char *)desc1->identifier);
9645
9646	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9647	/* This is so Copancontrol will return something sane */
9648	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9649	    ctsio->io_hdr.nexus.targ_port!=8)
9650		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9651	else
9652		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9653#endif
9654
9655	be64enc(desc1->identifier, fe->wwpn);
9656
9657	/*
9658	 * desc2 is for the Relative Target Port(type 4h) identifier
9659	 */
9660	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9661	                 | SVPD_ID_TYPE_RELTARG;
9662	desc2->length = 4;
9663//#if 0
9664	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9665	/* This is so Copancontrol will return something sane */
9666	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9667	    ctsio->io_hdr.nexus.targ_port!=8)
9668		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9669	else
9670	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9671//#endif
9672
9673	/*
9674	 * desc3 is for the Target Port Group(type 5h) identifier
9675	 */
9676	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9677	                 | SVPD_ID_TYPE_TPORTGRP;
9678	desc3->length = 4;
9679	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9680		desc3->identifier[3] = 1;
9681	else
9682		desc3->identifier[3] = 2;
9683
9684	/*
9685	 * If we've actually got a backend, copy the device id from the
9686	 * per-LUN data.  Otherwise, set it to all spaces.
9687	 */
9688	if (lun != NULL) {
9689		/*
9690		 * Copy the backend's LUN ID.
9691		 */
9692		strncpy((char *)t10id->vendor_spec_id,
9693			(char *)lun->be_lun->device_id, devid_len);
9694	} else {
9695		/*
9696		 * No backend, set this to spaces.
9697		 */
9698		memset(t10id->vendor_spec_id, 0x20, devid_len);
9699	}
9700
9701	ctsio->scsi_status = SCSI_STATUS_OK;
9702
9703	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9704	ctsio->be_move_done = ctl_config_move_done;
9705	ctl_datamove((union ctl_io *)ctsio);
9706
9707	return (CTL_RETVAL_COMPLETE);
9708}
9709
9710static int
9711ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9712{
9713	struct scsi_vpd_block_limits *bl_ptr;
9714	struct ctl_lun *lun;
9715	int bs;
9716
9717	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9718	bs = lun->be_lun->blocksize;
9719
9720	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9721	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9722	ctsio->kern_sg_entries = 0;
9723
9724	if (sizeof(*bl_ptr) < alloc_len) {
9725		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9726		ctsio->kern_data_len = sizeof(*bl_ptr);
9727		ctsio->kern_total_len = sizeof(*bl_ptr);
9728	} else {
9729		ctsio->residual = 0;
9730		ctsio->kern_data_len = alloc_len;
9731		ctsio->kern_total_len = alloc_len;
9732	}
9733	ctsio->kern_data_resid = 0;
9734	ctsio->kern_rel_offset = 0;
9735	ctsio->kern_sg_entries = 0;
9736
9737	/*
9738	 * The control device is always connected.  The disk device, on the
9739	 * other hand, may not be online all the time.  Need to change this
9740	 * to figure out whether the disk device is actually online or not.
9741	 */
9742	if (lun != NULL)
9743		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9744				  lun->be_lun->lun_type;
9745	else
9746		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9747
9748	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9749	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9750	bl_ptr->max_cmp_write_len = 0xff;
9751	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9752	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9753	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9754		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9755		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9756	}
9757	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9758
9759	ctsio->scsi_status = SCSI_STATUS_OK;
9760	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9761	ctsio->be_move_done = ctl_config_move_done;
9762	ctl_datamove((union ctl_io *)ctsio);
9763
9764	return (CTL_RETVAL_COMPLETE);
9765}
9766
9767static int
9768ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9769{
9770	struct scsi_vpd_logical_block_prov *lbp_ptr;
9771	struct ctl_lun *lun;
9772	int bs;
9773
9774	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9775	bs = lun->be_lun->blocksize;
9776
9777	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9778	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9779	ctsio->kern_sg_entries = 0;
9780
9781	if (sizeof(*lbp_ptr) < alloc_len) {
9782		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9783		ctsio->kern_data_len = sizeof(*lbp_ptr);
9784		ctsio->kern_total_len = sizeof(*lbp_ptr);
9785	} else {
9786		ctsio->residual = 0;
9787		ctsio->kern_data_len = alloc_len;
9788		ctsio->kern_total_len = alloc_len;
9789	}
9790	ctsio->kern_data_resid = 0;
9791	ctsio->kern_rel_offset = 0;
9792	ctsio->kern_sg_entries = 0;
9793
9794	/*
9795	 * The control device is always connected.  The disk device, on the
9796	 * other hand, may not be online all the time.  Need to change this
9797	 * to figure out whether the disk device is actually online or not.
9798	 */
9799	if (lun != NULL)
9800		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9801				  lun->be_lun->lun_type;
9802	else
9803		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9804
9805	lbp_ptr->page_code = SVPD_LBP;
9806	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9807		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9808
9809	ctsio->scsi_status = SCSI_STATUS_OK;
9810	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9811	ctsio->be_move_done = ctl_config_move_done;
9812	ctl_datamove((union ctl_io *)ctsio);
9813
9814	return (CTL_RETVAL_COMPLETE);
9815}
9816
9817static int
9818ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9819{
9820	struct scsi_inquiry *cdb;
9821	struct ctl_lun *lun;
9822	int alloc_len, retval;
9823
9824	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9825	cdb = (struct scsi_inquiry *)ctsio->cdb;
9826
9827	retval = CTL_RETVAL_COMPLETE;
9828
9829	alloc_len = scsi_2btoul(cdb->length);
9830
9831	switch (cdb->page_code) {
9832	case SVPD_SUPPORTED_PAGES:
9833		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9834		break;
9835	case SVPD_UNIT_SERIAL_NUMBER:
9836		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9837		break;
9838	case SVPD_DEVICE_ID:
9839		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9840		break;
9841	case SVPD_BLOCK_LIMITS:
9842		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9843		break;
9844	case SVPD_LBP:
9845		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9846		break;
9847	default:
9848		ctl_set_invalid_field(ctsio,
9849				      /*sks_valid*/ 1,
9850				      /*command*/ 1,
9851				      /*field*/ 2,
9852				      /*bit_valid*/ 0,
9853				      /*bit*/ 0);
9854		ctl_done((union ctl_io *)ctsio);
9855		retval = CTL_RETVAL_COMPLETE;
9856		break;
9857	}
9858
9859	return (retval);
9860}
9861
9862static int
9863ctl_inquiry_std(struct ctl_scsiio *ctsio)
9864{
9865	struct scsi_inquiry_data *inq_ptr;
9866	struct scsi_inquiry *cdb;
9867	struct ctl_softc *ctl_softc;
9868	struct ctl_lun *lun;
9869	char *val;
9870	uint32_t alloc_len;
9871	int is_fc;
9872
9873	ctl_softc = control_softc;
9874
9875	/*
9876	 * Figure out whether we're talking to a Fibre Channel port or not.
9877	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9878	 * SCSI front ends.
9879	 */
9880	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9881	    CTL_PORT_FC)
9882		is_fc = 0;
9883	else
9884		is_fc = 1;
9885
9886	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9887	cdb = (struct scsi_inquiry *)ctsio->cdb;
9888	alloc_len = scsi_2btoul(cdb->length);
9889
9890	/*
9891	 * We malloc the full inquiry data size here and fill it
9892	 * in.  If the user only asks for less, we'll give him
9893	 * that much.
9894	 */
9895	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9896	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9897	ctsio->kern_sg_entries = 0;
9898	ctsio->kern_data_resid = 0;
9899	ctsio->kern_rel_offset = 0;
9900
9901	if (sizeof(*inq_ptr) < alloc_len) {
9902		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9903		ctsio->kern_data_len = sizeof(*inq_ptr);
9904		ctsio->kern_total_len = sizeof(*inq_ptr);
9905	} else {
9906		ctsio->residual = 0;
9907		ctsio->kern_data_len = alloc_len;
9908		ctsio->kern_total_len = alloc_len;
9909	}
9910
9911	/*
9912	 * If we have a LUN configured, report it as connected.  Otherwise,
9913	 * report that it is offline or no device is supported, depending
9914	 * on the value of inquiry_pq_no_lun.
9915	 *
9916	 * According to the spec (SPC-4 r34), the peripheral qualifier
9917	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9918	 *
9919	 * "A peripheral device having the specified peripheral device type
9920	 * is not connected to this logical unit. However, the device
9921	 * server is capable of supporting the specified peripheral device
9922	 * type on this logical unit."
9923	 *
9924	 * According to the same spec, the peripheral qualifier
9925	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9926	 *
9927	 * "The device server is not capable of supporting a peripheral
9928	 * device on this logical unit. For this peripheral qualifier the
9929	 * peripheral device type shall be set to 1Fh. All other peripheral
9930	 * device type values are reserved for this peripheral qualifier."
9931	 *
9932	 * Given the text, it would seem that we probably want to report that
9933	 * the LUN is offline here.  There is no LUN connected, but we can
9934	 * support a LUN at the given LUN number.
9935	 *
9936	 * In the real world, though, it sounds like things are a little
9937	 * different:
9938	 *
9939	 * - Linux, when presented with a LUN with the offline peripheral
9940	 *   qualifier, will create an sg driver instance for it.  So when
9941	 *   you attach it to CTL, you wind up with a ton of sg driver
9942	 *   instances.  (One for every LUN that Linux bothered to probe.)
9943	 *   Linux does this despite the fact that it issues a REPORT LUNs
9944	 *   to LUN 0 to get the inventory of supported LUNs.
9945	 *
9946	 * - There is other anecdotal evidence (from Emulex folks) about
9947	 *   arrays that use the offline peripheral qualifier for LUNs that
9948	 *   are on the "passive" path in an active/passive array.
9949	 *
9950	 * So the solution is provide a hopefully reasonable default
9951	 * (return bad/no LUN) and allow the user to change the behavior
9952	 * with a tunable/sysctl variable.
9953	 */
9954	if (lun != NULL)
9955		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9956				  lun->be_lun->lun_type;
9957	else if (ctl_softc->inquiry_pq_no_lun == 0)
9958		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9959	else
9960		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9961
9962	/* RMB in byte 2 is 0 */
9963	inq_ptr->version = SCSI_REV_SPC3;
9964
9965	/*
9966	 * According to SAM-3, even if a device only supports a single
9967	 * level of LUN addressing, it should still set the HISUP bit:
9968	 *
9969	 * 4.9.1 Logical unit numbers overview
9970	 *
9971	 * All logical unit number formats described in this standard are
9972	 * hierarchical in structure even when only a single level in that
9973	 * hierarchy is used. The HISUP bit shall be set to one in the
9974	 * standard INQUIRY data (see SPC-2) when any logical unit number
9975	 * format described in this standard is used.  Non-hierarchical
9976	 * formats are outside the scope of this standard.
9977	 *
9978	 * Therefore we set the HiSup bit here.
9979	 *
9980	 * The reponse format is 2, per SPC-3.
9981	 */
9982	inq_ptr->response_format = SID_HiSup | 2;
9983
9984	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9985	CTL_DEBUG_PRINT(("additional_length = %d\n",
9986			 inq_ptr->additional_length));
9987
9988	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9989	/* 16 bit addressing */
9990	if (is_fc == 0)
9991		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9992	/* XXX set the SID_MultiP bit here if we're actually going to
9993	   respond on multiple ports */
9994	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9995
9996	/* 16 bit data bus, synchronous transfers */
9997	/* XXX these flags don't apply for FC */
9998	if (is_fc == 0)
9999		inq_ptr->flags = SID_WBus16 | SID_Sync;
10000	/*
10001	 * XXX KDM do we want to support tagged queueing on the control
10002	 * device at all?
10003	 */
10004	if ((lun == NULL)
10005	 || (lun->be_lun->lun_type != T_PROCESSOR))
10006		inq_ptr->flags |= SID_CmdQue;
10007	/*
10008	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10009	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10010	 * name and 4 bytes for the revision.
10011	 */
10012	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
10013		strcpy(inq_ptr->vendor, CTL_VENDOR);
10014	} else {
10015		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10016		strncpy(inq_ptr->vendor, val,
10017		    min(sizeof(inq_ptr->vendor), strlen(val)));
10018	}
10019	if (lun == NULL) {
10020		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10021	} else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
10022		switch (lun->be_lun->lun_type) {
10023		case T_DIRECT:
10024			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10025			break;
10026		case T_PROCESSOR:
10027			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10028			break;
10029		default:
10030			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10031			break;
10032		}
10033	} else {
10034		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10035		strncpy(inq_ptr->product, val,
10036		    min(sizeof(inq_ptr->product), strlen(val)));
10037	}
10038
10039	/*
10040	 * XXX make this a macro somewhere so it automatically gets
10041	 * incremented when we make changes.
10042	 */
10043	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
10044		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10045	} else {
10046		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10047		strncpy(inq_ptr->revision, val,
10048		    min(sizeof(inq_ptr->revision), strlen(val)));
10049	}
10050
10051	/*
10052	 * For parallel SCSI, we support double transition and single
10053	 * transition clocking.  We also support QAS (Quick Arbitration
10054	 * and Selection) and Information Unit transfers on both the
10055	 * control and array devices.
10056	 */
10057	if (is_fc == 0)
10058		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10059				    SID_SPI_IUS;
10060
10061	/* SAM-3 */
10062	scsi_ulto2b(0x0060, inq_ptr->version1);
10063	/* SPC-3 (no version claimed) XXX should we claim a version? */
10064	scsi_ulto2b(0x0300, inq_ptr->version2);
10065	if (is_fc) {
10066		/* FCP-2 ANSI INCITS.350:2003 */
10067		scsi_ulto2b(0x0917, inq_ptr->version3);
10068	} else {
10069		/* SPI-4 ANSI INCITS.362:200x */
10070		scsi_ulto2b(0x0B56, inq_ptr->version3);
10071	}
10072
10073	if (lun == NULL) {
10074		/* SBC-2 (no version claimed) XXX should we claim a version? */
10075		scsi_ulto2b(0x0320, inq_ptr->version4);
10076	} else {
10077		switch (lun->be_lun->lun_type) {
10078		case T_DIRECT:
10079			/*
10080			 * SBC-2 (no version claimed) XXX should we claim a
10081			 * version?
10082			 */
10083			scsi_ulto2b(0x0320, inq_ptr->version4);
10084			break;
10085		case T_PROCESSOR:
10086		default:
10087			break;
10088		}
10089	}
10090
10091	ctsio->scsi_status = SCSI_STATUS_OK;
10092	if (ctsio->kern_data_len > 0) {
10093		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10094		ctsio->be_move_done = ctl_config_move_done;
10095		ctl_datamove((union ctl_io *)ctsio);
10096	} else {
10097		ctsio->io_hdr.status = CTL_SUCCESS;
10098		ctl_done((union ctl_io *)ctsio);
10099	}
10100
10101	return (CTL_RETVAL_COMPLETE);
10102}
10103
10104int
10105ctl_inquiry(struct ctl_scsiio *ctsio)
10106{
10107	struct scsi_inquiry *cdb;
10108	int retval;
10109
10110	cdb = (struct scsi_inquiry *)ctsio->cdb;
10111
10112	retval = 0;
10113
10114	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10115
10116	/*
10117	 * Right now, we don't support the CmdDt inquiry information.
10118	 * This would be nice to support in the future.  When we do
10119	 * support it, we should change this test so that it checks to make
10120	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10121	 */
10122#ifdef notyet
10123	if (((cdb->byte2 & SI_EVPD)
10124	 && (cdb->byte2 & SI_CMDDT)))
10125#endif
10126	if (cdb->byte2 & SI_CMDDT) {
10127		/*
10128		 * Point to the SI_CMDDT bit.  We might change this
10129		 * when we support SI_CMDDT, but since both bits would be
10130		 * "wrong", this should probably just stay as-is then.
10131		 */
10132		ctl_set_invalid_field(ctsio,
10133				      /*sks_valid*/ 1,
10134				      /*command*/ 1,
10135				      /*field*/ 1,
10136				      /*bit_valid*/ 1,
10137				      /*bit*/ 1);
10138		ctl_done((union ctl_io *)ctsio);
10139		return (CTL_RETVAL_COMPLETE);
10140	}
10141	if (cdb->byte2 & SI_EVPD)
10142		retval = ctl_inquiry_evpd(ctsio);
10143#ifdef notyet
10144	else if (cdb->byte2 & SI_CMDDT)
10145		retval = ctl_inquiry_cmddt(ctsio);
10146#endif
10147	else
10148		retval = ctl_inquiry_std(ctsio);
10149
10150	return (retval);
10151}
10152
10153/*
10154 * For known CDB types, parse the LBA and length.
10155 */
10156static int
10157ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10158{
10159	if (io->io_hdr.io_type != CTL_IO_SCSI)
10160		return (1);
10161
10162	switch (io->scsiio.cdb[0]) {
10163	case COMPARE_AND_WRITE: {
10164		struct scsi_compare_and_write *cdb;
10165
10166		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10167
10168		*lba = scsi_8btou64(cdb->addr);
10169		*len = cdb->length;
10170		break;
10171	}
10172	case READ_6:
10173	case WRITE_6: {
10174		struct scsi_rw_6 *cdb;
10175
10176		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10177
10178		*lba = scsi_3btoul(cdb->addr);
10179		/* only 5 bits are valid in the most significant address byte */
10180		*lba &= 0x1fffff;
10181		*len = cdb->length;
10182		break;
10183	}
10184	case READ_10:
10185	case WRITE_10: {
10186		struct scsi_rw_10 *cdb;
10187
10188		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10189
10190		*lba = scsi_4btoul(cdb->addr);
10191		*len = scsi_2btoul(cdb->length);
10192		break;
10193	}
10194	case WRITE_VERIFY_10: {
10195		struct scsi_write_verify_10 *cdb;
10196
10197		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10198
10199		*lba = scsi_4btoul(cdb->addr);
10200		*len = scsi_2btoul(cdb->length);
10201		break;
10202	}
10203	case READ_12:
10204	case WRITE_12: {
10205		struct scsi_rw_12 *cdb;
10206
10207		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10208
10209		*lba = scsi_4btoul(cdb->addr);
10210		*len = scsi_4btoul(cdb->length);
10211		break;
10212	}
10213	case WRITE_VERIFY_12: {
10214		struct scsi_write_verify_12 *cdb;
10215
10216		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10217
10218		*lba = scsi_4btoul(cdb->addr);
10219		*len = scsi_4btoul(cdb->length);
10220		break;
10221	}
10222	case READ_16:
10223	case WRITE_16: {
10224		struct scsi_rw_16 *cdb;
10225
10226		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10227
10228		*lba = scsi_8btou64(cdb->addr);
10229		*len = scsi_4btoul(cdb->length);
10230		break;
10231	}
10232	case WRITE_VERIFY_16: {
10233		struct scsi_write_verify_16 *cdb;
10234
10235		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10236
10237
10238		*lba = scsi_8btou64(cdb->addr);
10239		*len = scsi_4btoul(cdb->length);
10240		break;
10241	}
10242	case WRITE_SAME_10: {
10243		struct scsi_write_same_10 *cdb;
10244
10245		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10246
10247		*lba = scsi_4btoul(cdb->addr);
10248		*len = scsi_2btoul(cdb->length);
10249		break;
10250	}
10251	case WRITE_SAME_16: {
10252		struct scsi_write_same_16 *cdb;
10253
10254		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10255
10256		*lba = scsi_8btou64(cdb->addr);
10257		*len = scsi_4btoul(cdb->length);
10258		break;
10259	}
10260	case VERIFY_10: {
10261		struct scsi_verify_10 *cdb;
10262
10263		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10264
10265		*lba = scsi_4btoul(cdb->addr);
10266		*len = scsi_2btoul(cdb->length);
10267		break;
10268	}
10269	case VERIFY_12: {
10270		struct scsi_verify_12 *cdb;
10271
10272		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10273
10274		*lba = scsi_4btoul(cdb->addr);
10275		*len = scsi_4btoul(cdb->length);
10276		break;
10277	}
10278	case VERIFY_16: {
10279		struct scsi_verify_16 *cdb;
10280
10281		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10282
10283		*lba = scsi_8btou64(cdb->addr);
10284		*len = scsi_4btoul(cdb->length);
10285		break;
10286	}
10287	default:
10288		return (1);
10289		break; /* NOTREACHED */
10290	}
10291
10292	return (0);
10293}
10294
10295static ctl_action
10296ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10297{
10298	uint64_t endlba1, endlba2;
10299
10300	endlba1 = lba1 + len1 - 1;
10301	endlba2 = lba2 + len2 - 1;
10302
10303	if ((endlba1 < lba2)
10304	 || (endlba2 < lba1))
10305		return (CTL_ACTION_PASS);
10306	else
10307		return (CTL_ACTION_BLOCK);
10308}
10309
10310static ctl_action
10311ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10312{
10313	uint64_t lba1, lba2;
10314	uint32_t len1, len2;
10315	int retval;
10316
10317	retval = ctl_get_lba_len(io1, &lba1, &len1);
10318	if (retval != 0)
10319		return (CTL_ACTION_ERROR);
10320
10321	retval = ctl_get_lba_len(io2, &lba2, &len2);
10322	if (retval != 0)
10323		return (CTL_ACTION_ERROR);
10324
10325	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10326}
10327
10328static ctl_action
10329ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10330{
10331	struct ctl_cmd_entry *pending_entry, *ooa_entry;
10332	ctl_serialize_action *serialize_row;
10333
10334	/*
10335	 * The initiator attempted multiple untagged commands at the same
10336	 * time.  Can't do that.
10337	 */
10338	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10339	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10340	 && ((pending_io->io_hdr.nexus.targ_port ==
10341	      ooa_io->io_hdr.nexus.targ_port)
10342	  && (pending_io->io_hdr.nexus.initid.id ==
10343	      ooa_io->io_hdr.nexus.initid.id))
10344	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10345		return (CTL_ACTION_OVERLAP);
10346
10347	/*
10348	 * The initiator attempted to send multiple tagged commands with
10349	 * the same ID.  (It's fine if different initiators have the same
10350	 * tag ID.)
10351	 *
10352	 * Even if all of those conditions are true, we don't kill the I/O
10353	 * if the command ahead of us has been aborted.  We won't end up
10354	 * sending it to the FETD, and it's perfectly legal to resend a
10355	 * command with the same tag number as long as the previous
10356	 * instance of this tag number has been aborted somehow.
10357	 */
10358	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10359	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10360	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10361	 && ((pending_io->io_hdr.nexus.targ_port ==
10362	      ooa_io->io_hdr.nexus.targ_port)
10363	  && (pending_io->io_hdr.nexus.initid.id ==
10364	      ooa_io->io_hdr.nexus.initid.id))
10365	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10366		return (CTL_ACTION_OVERLAP_TAG);
10367
10368	/*
10369	 * If we get a head of queue tag, SAM-3 says that we should
10370	 * immediately execute it.
10371	 *
10372	 * What happens if this command would normally block for some other
10373	 * reason?  e.g. a request sense with a head of queue tag
10374	 * immediately after a write.  Normally that would block, but this
10375	 * will result in its getting executed immediately...
10376	 *
10377	 * We currently return "pass" instead of "skip", so we'll end up
10378	 * going through the rest of the queue to check for overlapped tags.
10379	 *
10380	 * XXX KDM check for other types of blockage first??
10381	 */
10382	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10383		return (CTL_ACTION_PASS);
10384
10385	/*
10386	 * Ordered tags have to block until all items ahead of them
10387	 * have completed.  If we get called with an ordered tag, we always
10388	 * block, if something else is ahead of us in the queue.
10389	 */
10390	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10391		return (CTL_ACTION_BLOCK);
10392
10393	/*
10394	 * Simple tags get blocked until all head of queue and ordered tags
10395	 * ahead of them have completed.  I'm lumping untagged commands in
10396	 * with simple tags here.  XXX KDM is that the right thing to do?
10397	 */
10398	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10399	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10400	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10401	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10402		return (CTL_ACTION_BLOCK);
10403
10404	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
10405	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
10406
10407	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10408
10409	switch (serialize_row[pending_entry->seridx]) {
10410	case CTL_SER_BLOCK:
10411		return (CTL_ACTION_BLOCK);
10412		break; /* NOTREACHED */
10413	case CTL_SER_EXTENT:
10414		return (ctl_extent_check(pending_io, ooa_io));
10415		break; /* NOTREACHED */
10416	case CTL_SER_PASS:
10417		return (CTL_ACTION_PASS);
10418		break; /* NOTREACHED */
10419	case CTL_SER_SKIP:
10420		return (CTL_ACTION_SKIP);
10421		break;
10422	default:
10423		panic("invalid serialization value %d",
10424		      serialize_row[pending_entry->seridx]);
10425		break; /* NOTREACHED */
10426	}
10427
10428	return (CTL_ACTION_ERROR);
10429}
10430
10431/*
10432 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10433 * Assumptions:
10434 * - pending_io is generally either incoming, or on the blocked queue
10435 * - starting I/O is the I/O we want to start the check with.
10436 */
10437static ctl_action
10438ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10439	      union ctl_io *starting_io)
10440{
10441	union ctl_io *ooa_io;
10442	ctl_action action;
10443
10444	mtx_assert(&lun->lun_lock, MA_OWNED);
10445
10446	/*
10447	 * Run back along the OOA queue, starting with the current
10448	 * blocked I/O and going through every I/O before it on the
10449	 * queue.  If starting_io is NULL, we'll just end up returning
10450	 * CTL_ACTION_PASS.
10451	 */
10452	for (ooa_io = starting_io; ooa_io != NULL;
10453	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10454	     ooa_links)){
10455
10456		/*
10457		 * This routine just checks to see whether
10458		 * cur_blocked is blocked by ooa_io, which is ahead
10459		 * of it in the queue.  It doesn't queue/dequeue
10460		 * cur_blocked.
10461		 */
10462		action = ctl_check_for_blockage(pending_io, ooa_io);
10463		switch (action) {
10464		case CTL_ACTION_BLOCK:
10465		case CTL_ACTION_OVERLAP:
10466		case CTL_ACTION_OVERLAP_TAG:
10467		case CTL_ACTION_SKIP:
10468		case CTL_ACTION_ERROR:
10469			return (action);
10470			break; /* NOTREACHED */
10471		case CTL_ACTION_PASS:
10472			break;
10473		default:
10474			panic("invalid action %d", action);
10475			break;  /* NOTREACHED */
10476		}
10477	}
10478
10479	return (CTL_ACTION_PASS);
10480}
10481
10482/*
10483 * Assumptions:
10484 * - An I/O has just completed, and has been removed from the per-LUN OOA
10485 *   queue, so some items on the blocked queue may now be unblocked.
10486 */
10487static int
10488ctl_check_blocked(struct ctl_lun *lun)
10489{
10490	union ctl_io *cur_blocked, *next_blocked;
10491
10492	mtx_assert(&lun->lun_lock, MA_OWNED);
10493
10494	/*
10495	 * Run forward from the head of the blocked queue, checking each
10496	 * entry against the I/Os prior to it on the OOA queue to see if
10497	 * there is still any blockage.
10498	 *
10499	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10500	 * with our removing a variable on it while it is traversing the
10501	 * list.
10502	 */
10503	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10504	     cur_blocked != NULL; cur_blocked = next_blocked) {
10505		union ctl_io *prev_ooa;
10506		ctl_action action;
10507
10508		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10509							  blocked_links);
10510
10511		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10512						      ctl_ooaq, ooa_links);
10513
10514		/*
10515		 * If cur_blocked happens to be the first item in the OOA
10516		 * queue now, prev_ooa will be NULL, and the action
10517		 * returned will just be CTL_ACTION_PASS.
10518		 */
10519		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10520
10521		switch (action) {
10522		case CTL_ACTION_BLOCK:
10523			/* Nothing to do here, still blocked */
10524			break;
10525		case CTL_ACTION_OVERLAP:
10526		case CTL_ACTION_OVERLAP_TAG:
10527			/*
10528			 * This shouldn't happen!  In theory we've already
10529			 * checked this command for overlap...
10530			 */
10531			break;
10532		case CTL_ACTION_PASS:
10533		case CTL_ACTION_SKIP: {
10534			struct ctl_softc *softc;
10535			struct ctl_cmd_entry *entry;
10536			uint32_t initidx;
10537			uint8_t opcode;
10538			int isc_retval;
10539
10540			/*
10541			 * The skip case shouldn't happen, this transaction
10542			 * should have never made it onto the blocked queue.
10543			 */
10544			/*
10545			 * This I/O is no longer blocked, we can remove it
10546			 * from the blocked queue.  Since this is a TAILQ
10547			 * (doubly linked list), we can do O(1) removals
10548			 * from any place on the list.
10549			 */
10550			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10551				     blocked_links);
10552			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10553
10554			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10555				/*
10556				 * Need to send IO back to original side to
10557				 * run
10558				 */
10559				union ctl_ha_msg msg_info;
10560
10561				msg_info.hdr.original_sc =
10562					cur_blocked->io_hdr.original_sc;
10563				msg_info.hdr.serializing_sc = cur_blocked;
10564				msg_info.hdr.msg_type = CTL_MSG_R2R;
10565				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10566				     &msg_info, sizeof(msg_info), 0)) >
10567				     CTL_HA_STATUS_SUCCESS) {
10568					printf("CTL:Check Blocked error from "
10569					       "ctl_ha_msg_send %d\n",
10570					       isc_retval);
10571				}
10572				break;
10573			}
10574			opcode = cur_blocked->scsiio.cdb[0];
10575			entry = &ctl_cmd_table[opcode];
10576			softc = control_softc;
10577
10578			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10579
10580			/*
10581			 * Check this I/O for LUN state changes that may
10582			 * have happened while this command was blocked.
10583			 * The LUN state may have been changed by a command
10584			 * ahead of us in the queue, so we need to re-check
10585			 * for any states that can be caused by SCSI
10586			 * commands.
10587			 */
10588			if (ctl_scsiio_lun_check(softc, lun, entry,
10589						 &cur_blocked->scsiio) == 0) {
10590				cur_blocked->io_hdr.flags |=
10591				                      CTL_FLAG_IS_WAS_ON_RTR;
10592				ctl_enqueue_rtr(cur_blocked);
10593			} else
10594				ctl_done(cur_blocked);
10595			break;
10596		}
10597		default:
10598			/*
10599			 * This probably shouldn't happen -- we shouldn't
10600			 * get CTL_ACTION_ERROR, or anything else.
10601			 */
10602			break;
10603		}
10604	}
10605
10606	return (CTL_RETVAL_COMPLETE);
10607}
10608
10609/*
10610 * This routine (with one exception) checks LUN flags that can be set by
10611 * commands ahead of us in the OOA queue.  These flags have to be checked
10612 * when a command initially comes in, and when we pull a command off the
10613 * blocked queue and are preparing to execute it.  The reason we have to
10614 * check these flags for commands on the blocked queue is that the LUN
10615 * state may have been changed by a command ahead of us while we're on the
10616 * blocked queue.
10617 *
10618 * Ordering is somewhat important with these checks, so please pay
10619 * careful attention to the placement of any new checks.
10620 */
10621static int
10622ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10623		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10624{
10625	int retval;
10626
10627	retval = 0;
10628
10629	mtx_assert(&lun->lun_lock, MA_OWNED);
10630
10631	/*
10632	 * If this shelf is a secondary shelf controller, we have to reject
10633	 * any media access commands.
10634	 */
10635#if 0
10636	/* No longer needed for HA */
10637	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10638	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10639		ctl_set_lun_standby(ctsio);
10640		retval = 1;
10641		goto bailout;
10642	}
10643#endif
10644
10645	/*
10646	 * Check for a reservation conflict.  If this command isn't allowed
10647	 * even on reserved LUNs, and if this initiator isn't the one who
10648	 * reserved us, reject the command with a reservation conflict.
10649	 */
10650	if ((lun->flags & CTL_LUN_RESERVED)
10651	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10652		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10653		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10654		 || (ctsio->io_hdr.nexus.targ_target.id !=
10655		     lun->rsv_nexus.targ_target.id)) {
10656			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10657			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10658			retval = 1;
10659			goto bailout;
10660		}
10661	}
10662
10663	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10664	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10665		uint32_t residx;
10666
10667		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10668		/*
10669		 * if we aren't registered or it's a res holder type
10670		 * reservation and this isn't the res holder then set a
10671		 * conflict.
10672		 * NOTE: Commands which might be allowed on write exclusive
10673		 * type reservations are checked in the particular command
10674		 * for a conflict. Read and SSU are the only ones.
10675		 */
10676		if (!lun->per_res[residx].registered
10677		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10678			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10679			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10680			retval = 1;
10681			goto bailout;
10682		}
10683
10684	}
10685
10686	if ((lun->flags & CTL_LUN_OFFLINE)
10687	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10688		ctl_set_lun_not_ready(ctsio);
10689		retval = 1;
10690		goto bailout;
10691	}
10692
10693	/*
10694	 * If the LUN is stopped, see if this particular command is allowed
10695	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10696	 */
10697	if ((lun->flags & CTL_LUN_STOPPED)
10698	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10699		/* "Logical unit not ready, initializing cmd. required" */
10700		ctl_set_lun_stopped(ctsio);
10701		retval = 1;
10702		goto bailout;
10703	}
10704
10705	if ((lun->flags & CTL_LUN_INOPERABLE)
10706	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10707		/* "Medium format corrupted" */
10708		ctl_set_medium_format_corrupted(ctsio);
10709		retval = 1;
10710		goto bailout;
10711	}
10712
10713bailout:
10714	return (retval);
10715
10716}
10717
10718static void
10719ctl_failover_io(union ctl_io *io, int have_lock)
10720{
10721	ctl_set_busy(&io->scsiio);
10722	ctl_done(io);
10723}
10724
10725static void
10726ctl_failover(void)
10727{
10728	struct ctl_lun *lun;
10729	struct ctl_softc *ctl_softc;
10730	union ctl_io *next_io, *pending_io;
10731	union ctl_io *io;
10732	int lun_idx;
10733	int i;
10734
10735	ctl_softc = control_softc;
10736
10737	mtx_lock(&ctl_softc->ctl_lock);
10738	/*
10739	 * Remove any cmds from the other SC from the rtr queue.  These
10740	 * will obviously only be for LUNs for which we're the primary.
10741	 * We can't send status or get/send data for these commands.
10742	 * Since they haven't been executed yet, we can just remove them.
10743	 * We'll either abort them or delete them below, depending on
10744	 * which HA mode we're in.
10745	 */
10746#ifdef notyet
10747	mtx_lock(&ctl_softc->queue_lock);
10748	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10749	     io != NULL; io = next_io) {
10750		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10751		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10752			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10753				      ctl_io_hdr, links);
10754	}
10755	mtx_unlock(&ctl_softc->queue_lock);
10756#endif
10757
10758	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10759		lun = ctl_softc->ctl_luns[lun_idx];
10760		if (lun==NULL)
10761			continue;
10762
10763		/*
10764		 * Processor LUNs are primary on both sides.
10765		 * XXX will this always be true?
10766		 */
10767		if (lun->be_lun->lun_type == T_PROCESSOR)
10768			continue;
10769
10770		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10771		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10772			printf("FAILOVER: primary lun %d\n", lun_idx);
10773		        /*
10774			 * Remove all commands from the other SC. First from the
10775			 * blocked queue then from the ooa queue. Once we have
10776			 * removed them. Call ctl_check_blocked to see if there
10777			 * is anything that can run.
10778			 */
10779			for (io = (union ctl_io *)TAILQ_FIRST(
10780			     &lun->blocked_queue); io != NULL; io = next_io) {
10781
10782		        	next_io = (union ctl_io *)TAILQ_NEXT(
10783				    &io->io_hdr, blocked_links);
10784
10785				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10786					TAILQ_REMOVE(&lun->blocked_queue,
10787						     &io->io_hdr,blocked_links);
10788					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10789					TAILQ_REMOVE(&lun->ooa_queue,
10790						     &io->io_hdr, ooa_links);
10791
10792					ctl_free_io(io);
10793				}
10794			}
10795
10796			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10797	     		     io != NULL; io = next_io) {
10798
10799		        	next_io = (union ctl_io *)TAILQ_NEXT(
10800				    &io->io_hdr, ooa_links);
10801
10802				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10803
10804					TAILQ_REMOVE(&lun->ooa_queue,
10805						&io->io_hdr,
10806					     	ooa_links);
10807
10808					ctl_free_io(io);
10809				}
10810			}
10811			ctl_check_blocked(lun);
10812		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10813			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10814
10815			printf("FAILOVER: primary lun %d\n", lun_idx);
10816			/*
10817			 * Abort all commands from the other SC.  We can't
10818			 * send status back for them now.  These should get
10819			 * cleaned up when they are completed or come out
10820			 * for a datamove operation.
10821			 */
10822			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10823	     		     io != NULL; io = next_io) {
10824		        	next_io = (union ctl_io *)TAILQ_NEXT(
10825					&io->io_hdr, ooa_links);
10826
10827				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10828					io->io_hdr.flags |= CTL_FLAG_ABORT;
10829			}
10830		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10831			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10832
10833			printf("FAILOVER: secondary lun %d\n", lun_idx);
10834
10835			lun->flags |= CTL_LUN_PRIMARY_SC;
10836
10837			/*
10838			 * We send all I/O that was sent to this controller
10839			 * and redirected to the other side back with
10840			 * busy status, and have the initiator retry it.
10841			 * Figuring out how much data has been transferred,
10842			 * etc. and picking up where we left off would be
10843			 * very tricky.
10844			 *
10845			 * XXX KDM need to remove I/O from the blocked
10846			 * queue as well!
10847			 */
10848			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10849			     &lun->ooa_queue); pending_io != NULL;
10850			     pending_io = next_io) {
10851
10852				next_io =  (union ctl_io *)TAILQ_NEXT(
10853					&pending_io->io_hdr, ooa_links);
10854
10855				pending_io->io_hdr.flags &=
10856					~CTL_FLAG_SENT_2OTHER_SC;
10857
10858				if (pending_io->io_hdr.flags &
10859				    CTL_FLAG_IO_ACTIVE) {
10860					pending_io->io_hdr.flags |=
10861						CTL_FLAG_FAILOVER;
10862				} else {
10863					ctl_set_busy(&pending_io->scsiio);
10864					ctl_done(pending_io);
10865				}
10866			}
10867
10868			/*
10869			 * Build Unit Attention
10870			 */
10871			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10872				lun->pending_sense[i].ua_pending |=
10873				                     CTL_UA_ASYM_ACC_CHANGE;
10874			}
10875		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10876			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10877			printf("FAILOVER: secondary lun %d\n", lun_idx);
10878			/*
10879			 * if the first io on the OOA is not on the RtR queue
10880			 * add it.
10881			 */
10882			lun->flags |= CTL_LUN_PRIMARY_SC;
10883
10884			pending_io = (union ctl_io *)TAILQ_FIRST(
10885			    &lun->ooa_queue);
10886			if (pending_io==NULL) {
10887				printf("Nothing on OOA queue\n");
10888				continue;
10889			}
10890
10891			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10892			if ((pending_io->io_hdr.flags &
10893			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10894				pending_io->io_hdr.flags |=
10895				    CTL_FLAG_IS_WAS_ON_RTR;
10896				ctl_enqueue_rtr(pending_io);
10897			}
10898#if 0
10899			else
10900			{
10901				printf("Tag 0x%04x is running\n",
10902				      pending_io->scsiio.tag_num);
10903			}
10904#endif
10905
10906			next_io = (union ctl_io *)TAILQ_NEXT(
10907			    &pending_io->io_hdr, ooa_links);
10908			for (pending_io=next_io; pending_io != NULL;
10909			     pending_io = next_io) {
10910				pending_io->io_hdr.flags &=
10911				    ~CTL_FLAG_SENT_2OTHER_SC;
10912				next_io = (union ctl_io *)TAILQ_NEXT(
10913					&pending_io->io_hdr, ooa_links);
10914				if (pending_io->io_hdr.flags &
10915				    CTL_FLAG_IS_WAS_ON_RTR) {
10916#if 0
10917				        printf("Tag 0x%04x is running\n",
10918				      		pending_io->scsiio.tag_num);
10919#endif
10920					continue;
10921				}
10922
10923				switch (ctl_check_ooa(lun, pending_io,
10924			            (union ctl_io *)TAILQ_PREV(
10925				    &pending_io->io_hdr, ctl_ooaq,
10926				    ooa_links))) {
10927
10928				case CTL_ACTION_BLOCK:
10929					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10930							  &pending_io->io_hdr,
10931							  blocked_links);
10932					pending_io->io_hdr.flags |=
10933					    CTL_FLAG_BLOCKED;
10934					break;
10935				case CTL_ACTION_PASS:
10936				case CTL_ACTION_SKIP:
10937					pending_io->io_hdr.flags |=
10938					    CTL_FLAG_IS_WAS_ON_RTR;
10939					ctl_enqueue_rtr(pending_io);
10940					break;
10941				case CTL_ACTION_OVERLAP:
10942					ctl_set_overlapped_cmd(
10943					    (struct ctl_scsiio *)pending_io);
10944					ctl_done(pending_io);
10945					break;
10946				case CTL_ACTION_OVERLAP_TAG:
10947					ctl_set_overlapped_tag(
10948					    (struct ctl_scsiio *)pending_io,
10949					    pending_io->scsiio.tag_num & 0xff);
10950					ctl_done(pending_io);
10951					break;
10952				case CTL_ACTION_ERROR:
10953				default:
10954					ctl_set_internal_failure(
10955						(struct ctl_scsiio *)pending_io,
10956						0,  // sks_valid
10957						0); //retry count
10958					ctl_done(pending_io);
10959					break;
10960				}
10961			}
10962
10963			/*
10964			 * Build Unit Attention
10965			 */
10966			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10967				lun->pending_sense[i].ua_pending |=
10968				                     CTL_UA_ASYM_ACC_CHANGE;
10969			}
10970		} else {
10971			panic("Unhandled HA mode failover, LUN flags = %#x, "
10972			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10973		}
10974	}
10975	ctl_pause_rtr = 0;
10976	mtx_unlock(&ctl_softc->ctl_lock);
10977}
10978
10979static int
10980ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10981{
10982	struct ctl_lun *lun;
10983	struct ctl_cmd_entry *entry;
10984	uint8_t opcode;
10985	uint32_t initidx, targ_lun;
10986	int retval;
10987
10988	retval = 0;
10989
10990	lun = NULL;
10991
10992	opcode = ctsio->cdb[0];
10993
10994	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
10995	if ((targ_lun < CTL_MAX_LUNS)
10996	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
10997		lun = ctl_softc->ctl_luns[targ_lun];
10998		/*
10999		 * If the LUN is invalid, pretend that it doesn't exist.
11000		 * It will go away as soon as all pending I/O has been
11001		 * completed.
11002		 */
11003		if (lun->flags & CTL_LUN_DISABLED) {
11004			lun = NULL;
11005		} else {
11006			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11007			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11008				lun->be_lun;
11009			if (lun->be_lun->lun_type == T_PROCESSOR) {
11010				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11011			}
11012		}
11013	} else {
11014		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11015		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11016	}
11017
11018	entry = &ctl_cmd_table[opcode];
11019
11020	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11021	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11022
11023	/*
11024	 * Check to see whether we can send this command to LUNs that don't
11025	 * exist.  This should pretty much only be the case for inquiry
11026	 * and request sense.  Further checks, below, really require having
11027	 * a LUN, so we can't really check the command anymore.  Just put
11028	 * it on the rtr queue.
11029	 */
11030	if (lun == NULL) {
11031		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11032			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11033			ctl_enqueue_rtr((union ctl_io *)ctsio);
11034			return (retval);
11035		}
11036
11037		ctl_set_unsupported_lun(ctsio);
11038		ctl_done((union ctl_io *)ctsio);
11039		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11040		return (retval);
11041	} else {
11042		mtx_lock(&lun->lun_lock);
11043
11044		/*
11045		 * Every I/O goes into the OOA queue for a particular LUN, and
11046		 * stays there until completion.
11047		 */
11048		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11049
11050		/*
11051		 * Make sure we support this particular command on this LUN.
11052		 * e.g., we don't support writes to the control LUN.
11053		 */
11054		switch (lun->be_lun->lun_type) {
11055		case T_PROCESSOR:
11056		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11057			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11058			      == 0)) {
11059				mtx_unlock(&lun->lun_lock);
11060				ctl_set_invalid_opcode(ctsio);
11061				ctl_done((union ctl_io *)ctsio);
11062				return (retval);
11063			}
11064			break;
11065		case T_DIRECT:
11066			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11067			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11068			      == 0)){
11069				mtx_unlock(&lun->lun_lock);
11070				ctl_set_invalid_opcode(ctsio);
11071				ctl_done((union ctl_io *)ctsio);
11072				return (retval);
11073			}
11074			break;
11075		default:
11076			mtx_unlock(&lun->lun_lock);
11077			panic("Unsupported CTL LUN type %d\n",
11078			      lun->be_lun->lun_type);
11079		}
11080	}
11081
11082	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11083
11084	/*
11085	 * If we've got a request sense, it'll clear the contingent
11086	 * allegiance condition.  Otherwise, if we have a CA condition for
11087	 * this initiator, clear it, because it sent down a command other
11088	 * than request sense.
11089	 */
11090	if ((opcode != REQUEST_SENSE)
11091	 && (ctl_is_set(lun->have_ca, initidx)))
11092		ctl_clear_mask(lun->have_ca, initidx);
11093
11094	/*
11095	 * If the command has this flag set, it handles its own unit
11096	 * attention reporting, we shouldn't do anything.  Otherwise we
11097	 * check for any pending unit attentions, and send them back to the
11098	 * initiator.  We only do this when a command initially comes in,
11099	 * not when we pull it off the blocked queue.
11100	 *
11101	 * According to SAM-3, section 5.3.2, the order that things get
11102	 * presented back to the host is basically unit attentions caused
11103	 * by some sort of reset event, busy status, reservation conflicts
11104	 * or task set full, and finally any other status.
11105	 *
11106	 * One issue here is that some of the unit attentions we report
11107	 * don't fall into the "reset" category (e.g. "reported luns data
11108	 * has changed").  So reporting it here, before the reservation
11109	 * check, may be technically wrong.  I guess the only thing to do
11110	 * would be to check for and report the reset events here, and then
11111	 * check for the other unit attention types after we check for a
11112	 * reservation conflict.
11113	 *
11114	 * XXX KDM need to fix this
11115	 */
11116	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11117		ctl_ua_type ua_type;
11118
11119		ua_type = lun->pending_sense[initidx].ua_pending;
11120		if (ua_type != CTL_UA_NONE) {
11121			scsi_sense_data_type sense_format;
11122
11123			if (lun != NULL)
11124				sense_format = (lun->flags &
11125				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11126				    SSD_TYPE_FIXED;
11127			else
11128				sense_format = SSD_TYPE_FIXED;
11129
11130			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11131					       sense_format);
11132			if (ua_type != CTL_UA_NONE) {
11133				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11134				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11135						       CTL_AUTOSENSE;
11136				ctsio->sense_len = SSD_FULL_SIZE;
11137				lun->pending_sense[initidx].ua_pending &=
11138					~ua_type;
11139				mtx_unlock(&lun->lun_lock);
11140				ctl_done((union ctl_io *)ctsio);
11141				return (retval);
11142			}
11143		}
11144	}
11145
11146
11147	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11148		mtx_unlock(&lun->lun_lock);
11149		ctl_done((union ctl_io *)ctsio);
11150		return (retval);
11151	}
11152
11153	/*
11154	 * XXX CHD this is where we want to send IO to other side if
11155	 * this LUN is secondary on this SC. We will need to make a copy
11156	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11157	 * the copy we send as FROM_OTHER.
11158	 * We also need to stuff the address of the original IO so we can
11159	 * find it easily. Something similar will need be done on the other
11160	 * side so when we are done we can find the copy.
11161	 */
11162	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11163		union ctl_ha_msg msg_info;
11164		int isc_retval;
11165
11166		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11167
11168		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11169		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11170#if 0
11171		printf("1. ctsio %p\n", ctsio);
11172#endif
11173		msg_info.hdr.serializing_sc = NULL;
11174		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11175		msg_info.scsi.tag_num = ctsio->tag_num;
11176		msg_info.scsi.tag_type = ctsio->tag_type;
11177		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11178
11179		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11180
11181		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11182		    (void *)&msg_info, sizeof(msg_info), 0)) >
11183		    CTL_HA_STATUS_SUCCESS) {
11184			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11185			       isc_retval);
11186			printf("CTL:opcode is %x\n",opcode);
11187		} else {
11188#if 0
11189			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11190#endif
11191		}
11192
11193		/*
11194		 * XXX KDM this I/O is off the incoming queue, but hasn't
11195		 * been inserted on any other queue.  We may need to come
11196		 * up with a holding queue while we wait for serialization
11197		 * so that we have an idea of what we're waiting for from
11198		 * the other side.
11199		 */
11200		mtx_unlock(&lun->lun_lock);
11201		return (retval);
11202	}
11203
11204	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11205			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11206			      ctl_ooaq, ooa_links))) {
11207	case CTL_ACTION_BLOCK:
11208		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11209		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11210				  blocked_links);
11211		mtx_unlock(&lun->lun_lock);
11212		return (retval);
11213	case CTL_ACTION_PASS:
11214	case CTL_ACTION_SKIP:
11215		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11216		mtx_unlock(&lun->lun_lock);
11217		ctl_enqueue_rtr((union ctl_io *)ctsio);
11218		break;
11219	case CTL_ACTION_OVERLAP:
11220		mtx_unlock(&lun->lun_lock);
11221		ctl_set_overlapped_cmd(ctsio);
11222		ctl_done((union ctl_io *)ctsio);
11223		break;
11224	case CTL_ACTION_OVERLAP_TAG:
11225		mtx_unlock(&lun->lun_lock);
11226		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11227		ctl_done((union ctl_io *)ctsio);
11228		break;
11229	case CTL_ACTION_ERROR:
11230	default:
11231		mtx_unlock(&lun->lun_lock);
11232		ctl_set_internal_failure(ctsio,
11233					 /*sks_valid*/ 0,
11234					 /*retry_count*/ 0);
11235		ctl_done((union ctl_io *)ctsio);
11236		break;
11237	}
11238	return (retval);
11239}
11240
11241static int
11242ctl_scsiio(struct ctl_scsiio *ctsio)
11243{
11244	int retval;
11245	struct ctl_cmd_entry *entry;
11246
11247	retval = CTL_RETVAL_COMPLETE;
11248
11249	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11250
11251	entry = &ctl_cmd_table[ctsio->cdb[0]];
11252
11253	/*
11254	 * If this I/O has been aborted, just send it straight to
11255	 * ctl_done() without executing it.
11256	 */
11257	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11258		ctl_done((union ctl_io *)ctsio);
11259		goto bailout;
11260	}
11261
11262	/*
11263	 * All the checks should have been handled by ctl_scsiio_precheck().
11264	 * We should be clear now to just execute the I/O.
11265	 */
11266	retval = entry->execute(ctsio);
11267
11268bailout:
11269	return (retval);
11270}
11271
11272/*
11273 * Since we only implement one target right now, a bus reset simply resets
11274 * our single target.
11275 */
11276static int
11277ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11278{
11279	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11280}
11281
11282static int
11283ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11284		 ctl_ua_type ua_type)
11285{
11286	struct ctl_lun *lun;
11287	int retval;
11288
11289	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11290		union ctl_ha_msg msg_info;
11291
11292		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11293		msg_info.hdr.nexus = io->io_hdr.nexus;
11294		if (ua_type==CTL_UA_TARG_RESET)
11295			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11296		else
11297			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11298		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11299		msg_info.hdr.original_sc = NULL;
11300		msg_info.hdr.serializing_sc = NULL;
11301		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11302		    (void *)&msg_info, sizeof(msg_info), 0)) {
11303		}
11304	}
11305	retval = 0;
11306
11307	mtx_lock(&ctl_softc->ctl_lock);
11308	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11309		retval += ctl_lun_reset(lun, io, ua_type);
11310	mtx_unlock(&ctl_softc->ctl_lock);
11311
11312	return (retval);
11313}
11314
11315/*
11316 * The LUN should always be set.  The I/O is optional, and is used to
11317 * distinguish between I/Os sent by this initiator, and by other
11318 * initiators.  We set unit attention for initiators other than this one.
11319 * SAM-3 is vague on this point.  It does say that a unit attention should
11320 * be established for other initiators when a LUN is reset (see section
11321 * 5.7.3), but it doesn't specifically say that the unit attention should
11322 * be established for this particular initiator when a LUN is reset.  Here
11323 * is the relevant text, from SAM-3 rev 8:
11324 *
11325 * 5.7.2 When a SCSI initiator port aborts its own tasks
11326 *
11327 * When a SCSI initiator port causes its own task(s) to be aborted, no
11328 * notification that the task(s) have been aborted shall be returned to
11329 * the SCSI initiator port other than the completion response for the
11330 * command or task management function action that caused the task(s) to
11331 * be aborted and notification(s) associated with related effects of the
11332 * action (e.g., a reset unit attention condition).
11333 *
11334 * XXX KDM for now, we're setting unit attention for all initiators.
11335 */
11336static int
11337ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11338{
11339	union ctl_io *xio;
11340#if 0
11341	uint32_t initindex;
11342#endif
11343	int i;
11344
11345	mtx_lock(&lun->lun_lock);
11346	/*
11347	 * Run through the OOA queue and abort each I/O.
11348	 */
11349#if 0
11350	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11351#endif
11352	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11353	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11354		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11355	}
11356
11357	/*
11358	 * This version sets unit attention for every
11359	 */
11360#if 0
11361	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11362	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11363		if (initindex == i)
11364			continue;
11365		lun->pending_sense[i].ua_pending |= ua_type;
11366	}
11367#endif
11368
11369	/*
11370	 * A reset (any kind, really) clears reservations established with
11371	 * RESERVE/RELEASE.  It does not clear reservations established
11372	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11373	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11374	 * reservations made with the RESERVE/RELEASE commands, because
11375	 * those commands are obsolete in SPC-3.
11376	 */
11377	lun->flags &= ~CTL_LUN_RESERVED;
11378
11379	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11380		ctl_clear_mask(lun->have_ca, i);
11381		lun->pending_sense[i].ua_pending |= ua_type;
11382	}
11383	mtx_lock(&lun->lun_lock);
11384
11385	return (0);
11386}
11387
11388static int
11389ctl_abort_task(union ctl_io *io)
11390{
11391	union ctl_io *xio;
11392	struct ctl_lun *lun;
11393	struct ctl_softc *ctl_softc;
11394#if 0
11395	struct sbuf sb;
11396	char printbuf[128];
11397#endif
11398	int found;
11399	uint32_t targ_lun;
11400
11401	ctl_softc = control_softc;
11402	found = 0;
11403
11404	/*
11405	 * Look up the LUN.
11406	 */
11407	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11408	mtx_lock(&ctl_softc->ctl_lock);
11409	if ((targ_lun < CTL_MAX_LUNS)
11410	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11411		lun = ctl_softc->ctl_luns[targ_lun];
11412	else {
11413		mtx_unlock(&ctl_softc->ctl_lock);
11414		goto bailout;
11415	}
11416
11417#if 0
11418	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11419	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11420#endif
11421
11422	mtx_lock(&lun->lun_lock);
11423	mtx_unlock(&ctl_softc->ctl_lock);
11424	/*
11425	 * Run through the OOA queue and attempt to find the given I/O.
11426	 * The target port, initiator ID, tag type and tag number have to
11427	 * match the values that we got from the initiator.  If we have an
11428	 * untagged command to abort, simply abort the first untagged command
11429	 * we come to.  We only allow one untagged command at a time of course.
11430	 */
11431#if 0
11432	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11433#endif
11434	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11435	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11436#if 0
11437		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11438
11439		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11440			    lun->lun, xio->scsiio.tag_num,
11441			    xio->scsiio.tag_type,
11442			    (xio->io_hdr.blocked_links.tqe_prev
11443			    == NULL) ? "" : " BLOCKED",
11444			    (xio->io_hdr.flags &
11445			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11446			    (xio->io_hdr.flags &
11447			    CTL_FLAG_ABORT) ? " ABORT" : "",
11448			    (xio->io_hdr.flags &
11449			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11450		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11451		sbuf_finish(&sb);
11452		printf("%s\n", sbuf_data(&sb));
11453#endif
11454
11455		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11456		 && (xio->io_hdr.nexus.initid.id ==
11457		     io->io_hdr.nexus.initid.id)) {
11458			/*
11459			 * If the abort says that the task is untagged, the
11460			 * task in the queue must be untagged.  Otherwise,
11461			 * we just check to see whether the tag numbers
11462			 * match.  This is because the QLogic firmware
11463			 * doesn't pass back the tag type in an abort
11464			 * request.
11465			 */
11466#if 0
11467			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11468			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11469			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11470#endif
11471			/*
11472			 * XXX KDM we've got problems with FC, because it
11473			 * doesn't send down a tag type with aborts.  So we
11474			 * can only really go by the tag number...
11475			 * This may cause problems with parallel SCSI.
11476			 * Need to figure that out!!
11477			 */
11478			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11479				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11480				found = 1;
11481				if ((io->io_hdr.flags &
11482				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11483				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11484					union ctl_ha_msg msg_info;
11485
11486					io->io_hdr.flags |=
11487					                CTL_FLAG_SENT_2OTHER_SC;
11488					msg_info.hdr.nexus = io->io_hdr.nexus;
11489					msg_info.task.task_action =
11490						CTL_TASK_ABORT_TASK;
11491					msg_info.task.tag_num =
11492						io->taskio.tag_num;
11493					msg_info.task.tag_type =
11494						io->taskio.tag_type;
11495					msg_info.hdr.msg_type =
11496						CTL_MSG_MANAGE_TASKS;
11497					msg_info.hdr.original_sc = NULL;
11498					msg_info.hdr.serializing_sc = NULL;
11499#if 0
11500					printf("Sent Abort to other side\n");
11501#endif
11502					if (CTL_HA_STATUS_SUCCESS !=
11503					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11504		    				(void *)&msg_info,
11505						sizeof(msg_info), 0)) {
11506					}
11507				}
11508#if 0
11509				printf("ctl_abort_task: found I/O to abort\n");
11510#endif
11511				break;
11512			}
11513		}
11514	}
11515	mtx_unlock(&lun->lun_lock);
11516
11517bailout:
11518
11519	if (found == 0) {
11520		/*
11521		 * This isn't really an error.  It's entirely possible for
11522		 * the abort and command completion to cross on the wire.
11523		 * This is more of an informative/diagnostic error.
11524		 */
11525#if 0
11526		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11527		       "%d:%d:%d:%d tag %d type %d\n",
11528		       io->io_hdr.nexus.initid.id,
11529		       io->io_hdr.nexus.targ_port,
11530		       io->io_hdr.nexus.targ_target.id,
11531		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11532		       io->taskio.tag_type);
11533#endif
11534		return (1);
11535	} else
11536		return (0);
11537}
11538
11539/*
11540 * This routine cannot block!  It must be callable from an interrupt
11541 * handler as well as from the work thread.
11542 */
11543static void
11544ctl_run_task(union ctl_io *io)
11545{
11546	struct ctl_softc *ctl_softc;
11547	int retval;
11548	const char *task_desc;
11549
11550	CTL_DEBUG_PRINT(("ctl_run_task\n"));
11551
11552	ctl_softc = control_softc;
11553	retval = 0;
11554
11555	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11556	    ("ctl_run_task: Unextected io_type %d\n",
11557	     io->io_hdr.io_type));
11558
11559	task_desc = ctl_scsi_task_string(&io->taskio);
11560	if (task_desc != NULL) {
11561#ifdef NEEDTOPORT
11562		csevent_log(CSC_CTL | CSC_SHELF_SW |
11563			    CTL_TASK_REPORT,
11564			    csevent_LogType_Trace,
11565			    csevent_Severity_Information,
11566			    csevent_AlertLevel_Green,
11567			    csevent_FRU_Firmware,
11568			    csevent_FRU_Unknown,
11569			    "CTL: received task: %s",task_desc);
11570#endif
11571	} else {
11572#ifdef NEEDTOPORT
11573		csevent_log(CSC_CTL | CSC_SHELF_SW |
11574			    CTL_TASK_REPORT,
11575			    csevent_LogType_Trace,
11576			    csevent_Severity_Information,
11577			    csevent_AlertLevel_Green,
11578			    csevent_FRU_Firmware,
11579			    csevent_FRU_Unknown,
11580			    "CTL: received unknown task "
11581			    "type: %d (%#x)",
11582			    io->taskio.task_action,
11583			    io->taskio.task_action);
11584#endif
11585	}
11586	switch (io->taskio.task_action) {
11587	case CTL_TASK_ABORT_TASK:
11588		retval = ctl_abort_task(io);
11589		break;
11590	case CTL_TASK_ABORT_TASK_SET:
11591		break;
11592	case CTL_TASK_CLEAR_ACA:
11593		break;
11594	case CTL_TASK_CLEAR_TASK_SET:
11595		break;
11596	case CTL_TASK_LUN_RESET: {
11597		struct ctl_lun *lun;
11598		uint32_t targ_lun;
11599		int retval;
11600
11601		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11602		mtx_lock(&ctl_softc->ctl_lock);
11603		if ((targ_lun < CTL_MAX_LUNS)
11604		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11605			lun = ctl_softc->ctl_luns[targ_lun];
11606		else {
11607			mtx_unlock(&ctl_softc->ctl_lock);
11608			retval = 1;
11609			break;
11610		}
11611
11612		if (!(io->io_hdr.flags &
11613		    CTL_FLAG_FROM_OTHER_SC)) {
11614			union ctl_ha_msg msg_info;
11615
11616			io->io_hdr.flags |=
11617				CTL_FLAG_SENT_2OTHER_SC;
11618			msg_info.hdr.msg_type =
11619				CTL_MSG_MANAGE_TASKS;
11620			msg_info.hdr.nexus = io->io_hdr.nexus;
11621			msg_info.task.task_action =
11622				CTL_TASK_LUN_RESET;
11623			msg_info.hdr.original_sc = NULL;
11624			msg_info.hdr.serializing_sc = NULL;
11625			if (CTL_HA_STATUS_SUCCESS !=
11626			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11627			    (void *)&msg_info,
11628			    sizeof(msg_info), 0)) {
11629			}
11630		}
11631
11632		retval = ctl_lun_reset(lun, io,
11633				       CTL_UA_LUN_RESET);
11634		mtx_unlock(&ctl_softc->ctl_lock);
11635		break;
11636	}
11637	case CTL_TASK_TARGET_RESET:
11638		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
11639		break;
11640	case CTL_TASK_BUS_RESET:
11641		retval = ctl_bus_reset(ctl_softc, io);
11642		break;
11643	case CTL_TASK_PORT_LOGIN:
11644		break;
11645	case CTL_TASK_PORT_LOGOUT:
11646		break;
11647	default:
11648		printf("ctl_run_task: got unknown task management event %d\n",
11649		       io->taskio.task_action);
11650		break;
11651	}
11652	if (retval == 0)
11653		io->io_hdr.status = CTL_SUCCESS;
11654	else
11655		io->io_hdr.status = CTL_ERROR;
11656
11657	/*
11658	 * This will queue this I/O to the done queue, but the
11659	 * work thread won't be able to process it until we
11660	 * return and the lock is released.
11661	 */
11662	ctl_done(io);
11663}
11664
11665/*
11666 * For HA operation.  Handle commands that come in from the other
11667 * controller.
11668 */
11669static void
11670ctl_handle_isc(union ctl_io *io)
11671{
11672	int free_io;
11673	struct ctl_lun *lun;
11674	struct ctl_softc *ctl_softc;
11675	uint32_t targ_lun;
11676
11677	ctl_softc = control_softc;
11678
11679	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11680	lun = ctl_softc->ctl_luns[targ_lun];
11681
11682	switch (io->io_hdr.msg_type) {
11683	case CTL_MSG_SERIALIZE:
11684		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11685		break;
11686	case CTL_MSG_R2R: {
11687		uint8_t opcode;
11688		struct ctl_cmd_entry *entry;
11689
11690		/*
11691		 * This is only used in SER_ONLY mode.
11692		 */
11693		free_io = 0;
11694		opcode = io->scsiio.cdb[0];
11695		entry = &ctl_cmd_table[opcode];
11696		mtx_lock(&lun->lun_lock);
11697		if (ctl_scsiio_lun_check(ctl_softc, lun,
11698		    entry, (struct ctl_scsiio *)io) != 0) {
11699			mtx_unlock(&lun->lun_lock);
11700			ctl_done(io);
11701			break;
11702		}
11703		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11704		mtx_unlock(&lun->lun_lock);
11705		ctl_enqueue_rtr(io);
11706		break;
11707	}
11708	case CTL_MSG_FINISH_IO:
11709		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11710			free_io = 0;
11711			ctl_done(io);
11712		} else {
11713			free_io = 1;
11714			mtx_lock(&lun->lun_lock);
11715			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11716				     ooa_links);
11717			ctl_check_blocked(lun);
11718			mtx_unlock(&lun->lun_lock);
11719		}
11720		break;
11721	case CTL_MSG_PERS_ACTION:
11722		ctl_hndl_per_res_out_on_other_sc(
11723			(union ctl_ha_msg *)&io->presio.pr_msg);
11724		free_io = 1;
11725		break;
11726	case CTL_MSG_BAD_JUJU:
11727		free_io = 0;
11728		ctl_done(io);
11729		break;
11730	case CTL_MSG_DATAMOVE:
11731		/* Only used in XFER mode */
11732		free_io = 0;
11733		ctl_datamove_remote(io);
11734		break;
11735	case CTL_MSG_DATAMOVE_DONE:
11736		/* Only used in XFER mode */
11737		free_io = 0;
11738		io->scsiio.be_move_done(io);
11739		break;
11740	default:
11741		free_io = 1;
11742		printf("%s: Invalid message type %d\n",
11743		       __func__, io->io_hdr.msg_type);
11744		break;
11745	}
11746	if (free_io)
11747		ctl_free_io(io);
11748
11749}
11750
11751
11752/*
11753 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11754 * there is no match.
11755 */
11756static ctl_lun_error_pattern
11757ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11758{
11759	struct ctl_cmd_entry *entry;
11760	ctl_lun_error_pattern filtered_pattern, pattern;
11761	uint8_t opcode;
11762
11763	pattern = desc->error_pattern;
11764
11765	/*
11766	 * XXX KDM we need more data passed into this function to match a
11767	 * custom pattern, and we actually need to implement custom pattern
11768	 * matching.
11769	 */
11770	if (pattern & CTL_LUN_PAT_CMD)
11771		return (CTL_LUN_PAT_CMD);
11772
11773	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11774		return (CTL_LUN_PAT_ANY);
11775
11776	opcode = ctsio->cdb[0];
11777	entry = &ctl_cmd_table[opcode];
11778
11779	filtered_pattern = entry->pattern & pattern;
11780
11781	/*
11782	 * If the user requested specific flags in the pattern (e.g.
11783	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11784	 * flags.
11785	 *
11786	 * If the user did not specify any flags, it doesn't matter whether
11787	 * or not the command supports the flags.
11788	 */
11789	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11790	     (pattern & ~CTL_LUN_PAT_MASK))
11791		return (CTL_LUN_PAT_NONE);
11792
11793	/*
11794	 * If the user asked for a range check, see if the requested LBA
11795	 * range overlaps with this command's LBA range.
11796	 */
11797	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11798		uint64_t lba1;
11799		uint32_t len1;
11800		ctl_action action;
11801		int retval;
11802
11803		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11804		if (retval != 0)
11805			return (CTL_LUN_PAT_NONE);
11806
11807		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11808					      desc->lba_range.len);
11809		/*
11810		 * A "pass" means that the LBA ranges don't overlap, so
11811		 * this doesn't match the user's range criteria.
11812		 */
11813		if (action == CTL_ACTION_PASS)
11814			return (CTL_LUN_PAT_NONE);
11815	}
11816
11817	return (filtered_pattern);
11818}
11819
11820static void
11821ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11822{
11823	struct ctl_error_desc *desc, *desc2;
11824
11825	mtx_assert(&lun->lun_lock, MA_OWNED);
11826
11827	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11828		ctl_lun_error_pattern pattern;
11829		/*
11830		 * Check to see whether this particular command matches
11831		 * the pattern in the descriptor.
11832		 */
11833		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11834		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11835			continue;
11836
11837		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11838		case CTL_LUN_INJ_ABORTED:
11839			ctl_set_aborted(&io->scsiio);
11840			break;
11841		case CTL_LUN_INJ_MEDIUM_ERR:
11842			ctl_set_medium_error(&io->scsiio);
11843			break;
11844		case CTL_LUN_INJ_UA:
11845			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11846			 * OCCURRED */
11847			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11848			break;
11849		case CTL_LUN_INJ_CUSTOM:
11850			/*
11851			 * We're assuming the user knows what he is doing.
11852			 * Just copy the sense information without doing
11853			 * checks.
11854			 */
11855			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11856			      ctl_min(sizeof(desc->custom_sense),
11857				      sizeof(io->scsiio.sense_data)));
11858			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11859			io->scsiio.sense_len = SSD_FULL_SIZE;
11860			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11861			break;
11862		case CTL_LUN_INJ_NONE:
11863		default:
11864			/*
11865			 * If this is an error injection type we don't know
11866			 * about, clear the continuous flag (if it is set)
11867			 * so it will get deleted below.
11868			 */
11869			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11870			break;
11871		}
11872		/*
11873		 * By default, each error injection action is a one-shot
11874		 */
11875		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11876			continue;
11877
11878		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11879
11880		free(desc, M_CTL);
11881	}
11882}
11883
11884#ifdef CTL_IO_DELAY
11885static void
11886ctl_datamove_timer_wakeup(void *arg)
11887{
11888	union ctl_io *io;
11889
11890	io = (union ctl_io *)arg;
11891
11892	ctl_datamove(io);
11893}
11894#endif /* CTL_IO_DELAY */
11895
11896void
11897ctl_datamove(union ctl_io *io)
11898{
11899	void (*fe_datamove)(union ctl_io *io);
11900
11901	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11902
11903	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11904
11905#ifdef CTL_TIME_IO
11906	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11907		char str[256];
11908		char path_str[64];
11909		struct sbuf sb;
11910
11911		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11912		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11913
11914		sbuf_cat(&sb, path_str);
11915		switch (io->io_hdr.io_type) {
11916		case CTL_IO_SCSI:
11917			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11918			sbuf_printf(&sb, "\n");
11919			sbuf_cat(&sb, path_str);
11920			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11921				    io->scsiio.tag_num, io->scsiio.tag_type);
11922			break;
11923		case CTL_IO_TASK:
11924			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11925				    "Tag Type: %d\n", io->taskio.task_action,
11926				    io->taskio.tag_num, io->taskio.tag_type);
11927			break;
11928		default:
11929			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11930			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11931			break;
11932		}
11933		sbuf_cat(&sb, path_str);
11934		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11935			    (intmax_t)time_uptime - io->io_hdr.start_time);
11936		sbuf_finish(&sb);
11937		printf("%s", sbuf_data(&sb));
11938	}
11939#endif /* CTL_TIME_IO */
11940
11941#ifdef CTL_IO_DELAY
11942	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11943		struct ctl_lun *lun;
11944
11945		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11946
11947		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11948	} else {
11949		struct ctl_lun *lun;
11950
11951		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11952		if ((lun != NULL)
11953		 && (lun->delay_info.datamove_delay > 0)) {
11954			struct callout *callout;
11955
11956			callout = (struct callout *)&io->io_hdr.timer_bytes;
11957			callout_init(callout, /*mpsafe*/ 1);
11958			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11959			callout_reset(callout,
11960				      lun->delay_info.datamove_delay * hz,
11961				      ctl_datamove_timer_wakeup, io);
11962			if (lun->delay_info.datamove_type ==
11963			    CTL_DELAY_TYPE_ONESHOT)
11964				lun->delay_info.datamove_delay = 0;
11965			return;
11966		}
11967	}
11968#endif
11969
11970	/*
11971	 * This command has been aborted.  Set the port status, so we fail
11972	 * the data move.
11973	 */
11974	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11975		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11976		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11977		       io->io_hdr.nexus.targ_port,
11978		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11979		       io->io_hdr.nexus.targ_lun);
11980		io->io_hdr.status = CTL_CMD_ABORTED;
11981		io->io_hdr.port_status = 31337;
11982		/*
11983		 * Note that the backend, in this case, will get the
11984		 * callback in its context.  In other cases it may get
11985		 * called in the frontend's interrupt thread context.
11986		 */
11987		io->scsiio.be_move_done(io);
11988		return;
11989	}
11990
11991	/*
11992	 * If we're in XFER mode and this I/O is from the other shelf
11993	 * controller, we need to send the DMA to the other side to
11994	 * actually transfer the data to/from the host.  In serialize only
11995	 * mode the transfer happens below CTL and ctl_datamove() is only
11996	 * called on the machine that originally received the I/O.
11997	 */
11998	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11999	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12000		union ctl_ha_msg msg;
12001		uint32_t sg_entries_sent;
12002		int do_sg_copy;
12003		int i;
12004
12005		memset(&msg, 0, sizeof(msg));
12006		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12007		msg.hdr.original_sc = io->io_hdr.original_sc;
12008		msg.hdr.serializing_sc = io;
12009		msg.hdr.nexus = io->io_hdr.nexus;
12010		msg.dt.flags = io->io_hdr.flags;
12011		/*
12012		 * We convert everything into a S/G list here.  We can't
12013		 * pass by reference, only by value between controllers.
12014		 * So we can't pass a pointer to the S/G list, only as many
12015		 * S/G entries as we can fit in here.  If it's possible for
12016		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12017		 * then we need to break this up into multiple transfers.
12018		 */
12019		if (io->scsiio.kern_sg_entries == 0) {
12020			msg.dt.kern_sg_entries = 1;
12021			/*
12022			 * If this is in cached memory, flush the cache
12023			 * before we send the DMA request to the other
12024			 * controller.  We want to do this in either the
12025			 * read or the write case.  The read case is
12026			 * straightforward.  In the write case, we want to
12027			 * make sure nothing is in the local cache that
12028			 * could overwrite the DMAed data.
12029			 */
12030			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12031				/*
12032				 * XXX KDM use bus_dmamap_sync() here.
12033				 */
12034			}
12035
12036			/*
12037			 * Convert to a physical address if this is a
12038			 * virtual address.
12039			 */
12040			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12041				msg.dt.sg_list[0].addr =
12042					io->scsiio.kern_data_ptr;
12043			} else {
12044				/*
12045				 * XXX KDM use busdma here!
12046				 */
12047#if 0
12048				msg.dt.sg_list[0].addr = (void *)
12049					vtophys(io->scsiio.kern_data_ptr);
12050#endif
12051			}
12052
12053			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12054			do_sg_copy = 0;
12055		} else {
12056			struct ctl_sg_entry *sgl;
12057
12058			do_sg_copy = 1;
12059			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12060			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12061			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12062				/*
12063				 * XXX KDM use bus_dmamap_sync() here.
12064				 */
12065			}
12066		}
12067
12068		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12069		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12070		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12071		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12072		msg.dt.sg_sequence = 0;
12073
12074		/*
12075		 * Loop until we've sent all of the S/G entries.  On the
12076		 * other end, we'll recompose these S/G entries into one
12077		 * contiguous list before passing it to the
12078		 */
12079		for (sg_entries_sent = 0; sg_entries_sent <
12080		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12081			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12082				sizeof(msg.dt.sg_list[0])),
12083				msg.dt.kern_sg_entries - sg_entries_sent);
12084
12085			if (do_sg_copy != 0) {
12086				struct ctl_sg_entry *sgl;
12087				int j;
12088
12089				sgl = (struct ctl_sg_entry *)
12090					io->scsiio.kern_data_ptr;
12091				/*
12092				 * If this is in cached memory, flush the cache
12093				 * before we send the DMA request to the other
12094				 * controller.  We want to do this in either
12095				 * the * read or the write case.  The read
12096				 * case is straightforward.  In the write
12097				 * case, we want to make sure nothing is
12098				 * in the local cache that could overwrite
12099				 * the DMAed data.
12100				 */
12101
12102				for (i = sg_entries_sent, j = 0;
12103				     i < msg.dt.cur_sg_entries; i++, j++) {
12104					if ((io->io_hdr.flags &
12105					     CTL_FLAG_NO_DATASYNC) == 0) {
12106						/*
12107						 * XXX KDM use bus_dmamap_sync()
12108						 */
12109					}
12110					if ((io->io_hdr.flags &
12111					     CTL_FLAG_BUS_ADDR) == 0) {
12112						/*
12113						 * XXX KDM use busdma.
12114						 */
12115#if 0
12116						msg.dt.sg_list[j].addr =(void *)
12117						       vtophys(sgl[i].addr);
12118#endif
12119					} else {
12120						msg.dt.sg_list[j].addr =
12121							sgl[i].addr;
12122					}
12123					msg.dt.sg_list[j].len = sgl[i].len;
12124				}
12125			}
12126
12127			sg_entries_sent += msg.dt.cur_sg_entries;
12128			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12129				msg.dt.sg_last = 1;
12130			else
12131				msg.dt.sg_last = 0;
12132
12133			/*
12134			 * XXX KDM drop and reacquire the lock here?
12135			 */
12136			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12137			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12138				/*
12139				 * XXX do something here.
12140				 */
12141			}
12142
12143			msg.dt.sent_sg_entries = sg_entries_sent;
12144		}
12145		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12146		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12147			ctl_failover_io(io, /*have_lock*/ 0);
12148
12149	} else {
12150
12151		/*
12152		 * Lookup the fe_datamove() function for this particular
12153		 * front end.
12154		 */
12155		fe_datamove =
12156		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12157
12158		fe_datamove(io);
12159	}
12160}
12161
12162static void
12163ctl_send_datamove_done(union ctl_io *io, int have_lock)
12164{
12165	union ctl_ha_msg msg;
12166	int isc_status;
12167
12168	memset(&msg, 0, sizeof(msg));
12169
12170	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12171	msg.hdr.original_sc = io;
12172	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12173	msg.hdr.nexus = io->io_hdr.nexus;
12174	msg.hdr.status = io->io_hdr.status;
12175	msg.scsi.tag_num = io->scsiio.tag_num;
12176	msg.scsi.tag_type = io->scsiio.tag_type;
12177	msg.scsi.scsi_status = io->scsiio.scsi_status;
12178	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12179	       sizeof(io->scsiio.sense_data));
12180	msg.scsi.sense_len = io->scsiio.sense_len;
12181	msg.scsi.sense_residual = io->scsiio.sense_residual;
12182	msg.scsi.fetd_status = io->io_hdr.port_status;
12183	msg.scsi.residual = io->scsiio.residual;
12184	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12185
12186	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12187		ctl_failover_io(io, /*have_lock*/ have_lock);
12188		return;
12189	}
12190
12191	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12192	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12193		/* XXX do something if this fails */
12194	}
12195
12196}
12197
12198/*
12199 * The DMA to the remote side is done, now we need to tell the other side
12200 * we're done so it can continue with its data movement.
12201 */
12202static void
12203ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12204{
12205	union ctl_io *io;
12206
12207	io = rq->context;
12208
12209	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12210		printf("%s: ISC DMA write failed with error %d", __func__,
12211		       rq->ret);
12212		ctl_set_internal_failure(&io->scsiio,
12213					 /*sks_valid*/ 1,
12214					 /*retry_count*/ rq->ret);
12215	}
12216
12217	ctl_dt_req_free(rq);
12218
12219	/*
12220	 * In this case, we had to malloc the memory locally.  Free it.
12221	 */
12222	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12223		int i;
12224		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12225			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12226	}
12227	/*
12228	 * The data is in local and remote memory, so now we need to send
12229	 * status (good or back) back to the other side.
12230	 */
12231	ctl_send_datamove_done(io, /*have_lock*/ 0);
12232}
12233
12234/*
12235 * We've moved the data from the host/controller into local memory.  Now we
12236 * need to push it over to the remote controller's memory.
12237 */
12238static int
12239ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12240{
12241	int retval;
12242
12243	retval = 0;
12244
12245	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12246					  ctl_datamove_remote_write_cb);
12247
12248	return (retval);
12249}
12250
12251static void
12252ctl_datamove_remote_write(union ctl_io *io)
12253{
12254	int retval;
12255	void (*fe_datamove)(union ctl_io *io);
12256
12257	/*
12258	 * - Get the data from the host/HBA into local memory.
12259	 * - DMA memory from the local controller to the remote controller.
12260	 * - Send status back to the remote controller.
12261	 */
12262
12263	retval = ctl_datamove_remote_sgl_setup(io);
12264	if (retval != 0)
12265		return;
12266
12267	/* Switch the pointer over so the FETD knows what to do */
12268	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12269
12270	/*
12271	 * Use a custom move done callback, since we need to send completion
12272	 * back to the other controller, not to the backend on this side.
12273	 */
12274	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12275
12276	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12277
12278	fe_datamove(io);
12279
12280	return;
12281
12282}
12283
12284static int
12285ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12286{
12287#if 0
12288	char str[256];
12289	char path_str[64];
12290	struct sbuf sb;
12291#endif
12292
12293	/*
12294	 * In this case, we had to malloc the memory locally.  Free it.
12295	 */
12296	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12297		int i;
12298		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12299			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12300	}
12301
12302#if 0
12303	scsi_path_string(io, path_str, sizeof(path_str));
12304	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12305	sbuf_cat(&sb, path_str);
12306	scsi_command_string(&io->scsiio, NULL, &sb);
12307	sbuf_printf(&sb, "\n");
12308	sbuf_cat(&sb, path_str);
12309	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12310		    io->scsiio.tag_num, io->scsiio.tag_type);
12311	sbuf_cat(&sb, path_str);
12312	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12313		    io->io_hdr.flags, io->io_hdr.status);
12314	sbuf_finish(&sb);
12315	printk("%s", sbuf_data(&sb));
12316#endif
12317
12318
12319	/*
12320	 * The read is done, now we need to send status (good or bad) back
12321	 * to the other side.
12322	 */
12323	ctl_send_datamove_done(io, /*have_lock*/ 0);
12324
12325	return (0);
12326}
12327
12328static void
12329ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12330{
12331	union ctl_io *io;
12332	void (*fe_datamove)(union ctl_io *io);
12333
12334	io = rq->context;
12335
12336	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12337		printf("%s: ISC DMA read failed with error %d", __func__,
12338		       rq->ret);
12339		ctl_set_internal_failure(&io->scsiio,
12340					 /*sks_valid*/ 1,
12341					 /*retry_count*/ rq->ret);
12342	}
12343
12344	ctl_dt_req_free(rq);
12345
12346	/* Switch the pointer over so the FETD knows what to do */
12347	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12348
12349	/*
12350	 * Use a custom move done callback, since we need to send completion
12351	 * back to the other controller, not to the backend on this side.
12352	 */
12353	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12354
12355	/* XXX KDM add checks like the ones in ctl_datamove? */
12356
12357	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12358
12359	fe_datamove(io);
12360}
12361
12362static int
12363ctl_datamove_remote_sgl_setup(union ctl_io *io)
12364{
12365	struct ctl_sg_entry *local_sglist, *remote_sglist;
12366	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12367	struct ctl_softc *softc;
12368	int retval;
12369	int i;
12370
12371	retval = 0;
12372	softc = control_softc;
12373
12374	local_sglist = io->io_hdr.local_sglist;
12375	local_dma_sglist = io->io_hdr.local_dma_sglist;
12376	remote_sglist = io->io_hdr.remote_sglist;
12377	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12378
12379	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12380		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12381			local_sglist[i].len = remote_sglist[i].len;
12382
12383			/*
12384			 * XXX Detect the situation where the RS-level I/O
12385			 * redirector on the other side has already read the
12386			 * data off of the AOR RS on this side, and
12387			 * transferred it to remote (mirror) memory on the
12388			 * other side.  Since we already have the data in
12389			 * memory here, we just need to use it.
12390			 *
12391			 * XXX KDM this can probably be removed once we
12392			 * get the cache device code in and take the
12393			 * current AOR implementation out.
12394			 */
12395#ifdef NEEDTOPORT
12396			if ((remote_sglist[i].addr >=
12397			     (void *)vtophys(softc->mirr->addr))
12398			 && (remote_sglist[i].addr <
12399			     ((void *)vtophys(softc->mirr->addr) +
12400			     CacheMirrorOffset))) {
12401				local_sglist[i].addr = remote_sglist[i].addr -
12402					CacheMirrorOffset;
12403				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12404				     CTL_FLAG_DATA_IN)
12405					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12406			} else {
12407				local_sglist[i].addr = remote_sglist[i].addr +
12408					CacheMirrorOffset;
12409			}
12410#endif
12411#if 0
12412			printf("%s: local %p, remote %p, len %d\n",
12413			       __func__, local_sglist[i].addr,
12414			       remote_sglist[i].addr, local_sglist[i].len);
12415#endif
12416		}
12417	} else {
12418		uint32_t len_to_go;
12419
12420		/*
12421		 * In this case, we don't have automatically allocated
12422		 * memory for this I/O on this controller.  This typically
12423		 * happens with internal CTL I/O -- e.g. inquiry, mode
12424		 * sense, etc.  Anything coming from RAIDCore will have
12425		 * a mirror area available.
12426		 */
12427		len_to_go = io->scsiio.kern_data_len;
12428
12429		/*
12430		 * Clear the no datasync flag, we have to use malloced
12431		 * buffers.
12432		 */
12433		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12434
12435		/*
12436		 * The difficult thing here is that the size of the various
12437		 * S/G segments may be different than the size from the
12438		 * remote controller.  That'll make it harder when DMAing
12439		 * the data back to the other side.
12440		 */
12441		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12442		     sizeof(io->io_hdr.remote_sglist[0])) &&
12443		     (len_to_go > 0); i++) {
12444			local_sglist[i].len = ctl_min(len_to_go, 131072);
12445			CTL_SIZE_8B(local_dma_sglist[i].len,
12446				    local_sglist[i].len);
12447			local_sglist[i].addr =
12448				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12449
12450			local_dma_sglist[i].addr = local_sglist[i].addr;
12451
12452			if (local_sglist[i].addr == NULL) {
12453				int j;
12454
12455				printf("malloc failed for %zd bytes!",
12456				       local_dma_sglist[i].len);
12457				for (j = 0; j < i; j++) {
12458					free(local_sglist[j].addr, M_CTL);
12459				}
12460				ctl_set_internal_failure(&io->scsiio,
12461							 /*sks_valid*/ 1,
12462							 /*retry_count*/ 4857);
12463				retval = 1;
12464				goto bailout_error;
12465
12466			}
12467			/* XXX KDM do we need a sync here? */
12468
12469			len_to_go -= local_sglist[i].len;
12470		}
12471		/*
12472		 * Reset the number of S/G entries accordingly.  The
12473		 * original number of S/G entries is available in
12474		 * rem_sg_entries.
12475		 */
12476		io->scsiio.kern_sg_entries = i;
12477
12478#if 0
12479		printf("%s: kern_sg_entries = %d\n", __func__,
12480		       io->scsiio.kern_sg_entries);
12481		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12482			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12483			       local_sglist[i].addr, local_sglist[i].len,
12484			       local_dma_sglist[i].len);
12485#endif
12486	}
12487
12488
12489	return (retval);
12490
12491bailout_error:
12492
12493	ctl_send_datamove_done(io, /*have_lock*/ 0);
12494
12495	return (retval);
12496}
12497
12498static int
12499ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12500			 ctl_ha_dt_cb callback)
12501{
12502	struct ctl_ha_dt_req *rq;
12503	struct ctl_sg_entry *remote_sglist, *local_sglist;
12504	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12505	uint32_t local_used, remote_used, total_used;
12506	int retval;
12507	int i, j;
12508
12509	retval = 0;
12510
12511	rq = ctl_dt_req_alloc();
12512
12513	/*
12514	 * If we failed to allocate the request, and if the DMA didn't fail
12515	 * anyway, set busy status.  This is just a resource allocation
12516	 * failure.
12517	 */
12518	if ((rq == NULL)
12519	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12520		ctl_set_busy(&io->scsiio);
12521
12522	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12523
12524		if (rq != NULL)
12525			ctl_dt_req_free(rq);
12526
12527		/*
12528		 * The data move failed.  We need to return status back
12529		 * to the other controller.  No point in trying to DMA
12530		 * data to the remote controller.
12531		 */
12532
12533		ctl_send_datamove_done(io, /*have_lock*/ 0);
12534
12535		retval = 1;
12536
12537		goto bailout;
12538	}
12539
12540	local_sglist = io->io_hdr.local_sglist;
12541	local_dma_sglist = io->io_hdr.local_dma_sglist;
12542	remote_sglist = io->io_hdr.remote_sglist;
12543	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12544	local_used = 0;
12545	remote_used = 0;
12546	total_used = 0;
12547
12548	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12549		rq->ret = CTL_HA_STATUS_SUCCESS;
12550		rq->context = io;
12551		callback(rq);
12552		goto bailout;
12553	}
12554
12555	/*
12556	 * Pull/push the data over the wire from/to the other controller.
12557	 * This takes into account the possibility that the local and
12558	 * remote sglists may not be identical in terms of the size of
12559	 * the elements and the number of elements.
12560	 *
12561	 * One fundamental assumption here is that the length allocated for
12562	 * both the local and remote sglists is identical.  Otherwise, we've
12563	 * essentially got a coding error of some sort.
12564	 */
12565	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12566		int isc_ret;
12567		uint32_t cur_len, dma_length;
12568		uint8_t *tmp_ptr;
12569
12570		rq->id = CTL_HA_DATA_CTL;
12571		rq->command = command;
12572		rq->context = io;
12573
12574		/*
12575		 * Both pointers should be aligned.  But it is possible
12576		 * that the allocation length is not.  They should both
12577		 * also have enough slack left over at the end, though,
12578		 * to round up to the next 8 byte boundary.
12579		 */
12580		cur_len = ctl_min(local_sglist[i].len - local_used,
12581				  remote_sglist[j].len - remote_used);
12582
12583		/*
12584		 * In this case, we have a size issue and need to decrease
12585		 * the size, except in the case where we actually have less
12586		 * than 8 bytes left.  In that case, we need to increase
12587		 * the DMA length to get the last bit.
12588		 */
12589		if ((cur_len & 0x7) != 0) {
12590			if (cur_len > 0x7) {
12591				cur_len = cur_len - (cur_len & 0x7);
12592				dma_length = cur_len;
12593			} else {
12594				CTL_SIZE_8B(dma_length, cur_len);
12595			}
12596
12597		} else
12598			dma_length = cur_len;
12599
12600		/*
12601		 * If we had to allocate memory for this I/O, instead of using
12602		 * the non-cached mirror memory, we'll need to flush the cache
12603		 * before trying to DMA to the other controller.
12604		 *
12605		 * We could end up doing this multiple times for the same
12606		 * segment if we have a larger local segment than remote
12607		 * segment.  That shouldn't be an issue.
12608		 */
12609		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12610			/*
12611			 * XXX KDM use bus_dmamap_sync() here.
12612			 */
12613		}
12614
12615		rq->size = dma_length;
12616
12617		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12618		tmp_ptr += local_used;
12619
12620		/* Use physical addresses when talking to ISC hardware */
12621		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12622			/* XXX KDM use busdma */
12623#if 0
12624			rq->local = vtophys(tmp_ptr);
12625#endif
12626		} else
12627			rq->local = tmp_ptr;
12628
12629		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12630		tmp_ptr += remote_used;
12631		rq->remote = tmp_ptr;
12632
12633		rq->callback = NULL;
12634
12635		local_used += cur_len;
12636		if (local_used >= local_sglist[i].len) {
12637			i++;
12638			local_used = 0;
12639		}
12640
12641		remote_used += cur_len;
12642		if (remote_used >= remote_sglist[j].len) {
12643			j++;
12644			remote_used = 0;
12645		}
12646		total_used += cur_len;
12647
12648		if (total_used >= io->scsiio.kern_data_len)
12649			rq->callback = callback;
12650
12651		if ((rq->size & 0x7) != 0) {
12652			printf("%s: warning: size %d is not on 8b boundary\n",
12653			       __func__, rq->size);
12654		}
12655		if (((uintptr_t)rq->local & 0x7) != 0) {
12656			printf("%s: warning: local %p not on 8b boundary\n",
12657			       __func__, rq->local);
12658		}
12659		if (((uintptr_t)rq->remote & 0x7) != 0) {
12660			printf("%s: warning: remote %p not on 8b boundary\n",
12661			       __func__, rq->local);
12662		}
12663#if 0
12664		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12665		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12666		       rq->local, rq->remote, rq->size);
12667#endif
12668
12669		isc_ret = ctl_dt_single(rq);
12670		if (isc_ret == CTL_HA_STATUS_WAIT)
12671			continue;
12672
12673		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12674			rq->ret = CTL_HA_STATUS_SUCCESS;
12675		} else {
12676			rq->ret = isc_ret;
12677		}
12678		callback(rq);
12679		goto bailout;
12680	}
12681
12682bailout:
12683	return (retval);
12684
12685}
12686
12687static void
12688ctl_datamove_remote_read(union ctl_io *io)
12689{
12690	int retval;
12691	int i;
12692
12693	/*
12694	 * This will send an error to the other controller in the case of a
12695	 * failure.
12696	 */
12697	retval = ctl_datamove_remote_sgl_setup(io);
12698	if (retval != 0)
12699		return;
12700
12701	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12702					  ctl_datamove_remote_read_cb);
12703	if ((retval != 0)
12704	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12705		/*
12706		 * Make sure we free memory if there was an error..  The
12707		 * ctl_datamove_remote_xfer() function will send the
12708		 * datamove done message, or call the callback with an
12709		 * error if there is a problem.
12710		 */
12711		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12712			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12713	}
12714
12715	return;
12716}
12717
12718/*
12719 * Process a datamove request from the other controller.  This is used for
12720 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12721 * first.  Once that is complete, the data gets DMAed into the remote
12722 * controller's memory.  For reads, we DMA from the remote controller's
12723 * memory into our memory first, and then move it out to the FETD.
12724 */
12725static void
12726ctl_datamove_remote(union ctl_io *io)
12727{
12728	struct ctl_softc *softc;
12729
12730	softc = control_softc;
12731
12732	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12733
12734	/*
12735	 * Note that we look for an aborted I/O here, but don't do some of
12736	 * the other checks that ctl_datamove() normally does.  We don't
12737	 * need to run the task queue, because this I/O is on the ISC
12738	 * queue, which is executed by the work thread after the task queue.
12739	 * We don't need to run the datamove delay code, since that should
12740	 * have been done if need be on the other controller.
12741	 */
12742	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12743
12744		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12745		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12746		       io->io_hdr.nexus.targ_port,
12747		       io->io_hdr.nexus.targ_target.id,
12748		       io->io_hdr.nexus.targ_lun);
12749		io->io_hdr.status = CTL_CMD_ABORTED;
12750		io->io_hdr.port_status = 31338;
12751
12752		ctl_send_datamove_done(io, /*have_lock*/ 0);
12753
12754		return;
12755	}
12756
12757	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12758		ctl_datamove_remote_write(io);
12759	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12760		ctl_datamove_remote_read(io);
12761	} else {
12762		union ctl_ha_msg msg;
12763		struct scsi_sense_data *sense;
12764		uint8_t sks[3];
12765		int retry_count;
12766
12767		memset(&msg, 0, sizeof(msg));
12768
12769		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12770		msg.hdr.status = CTL_SCSI_ERROR;
12771		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12772
12773		retry_count = 4243;
12774
12775		sense = &msg.scsi.sense_data;
12776		sks[0] = SSD_SCS_VALID;
12777		sks[1] = (retry_count >> 8) & 0xff;
12778		sks[2] = retry_count & 0xff;
12779
12780		/* "Internal target failure" */
12781		scsi_set_sense_data(sense,
12782				    /*sense_format*/ SSD_TYPE_NONE,
12783				    /*current_error*/ 1,
12784				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12785				    /*asc*/ 0x44,
12786				    /*ascq*/ 0x00,
12787				    /*type*/ SSD_ELEM_SKS,
12788				    /*size*/ sizeof(sks),
12789				    /*data*/ sks,
12790				    SSD_ELEM_NONE);
12791
12792		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12793		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12794			ctl_failover_io(io, /*have_lock*/ 1);
12795			return;
12796		}
12797
12798		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12799		    CTL_HA_STATUS_SUCCESS) {
12800			/* XXX KDM what to do if this fails? */
12801		}
12802		return;
12803	}
12804
12805}
12806
12807static int
12808ctl_process_done(union ctl_io *io)
12809{
12810	struct ctl_lun *lun;
12811	struct ctl_softc *ctl_softc;
12812	void (*fe_done)(union ctl_io *io);
12813	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12814
12815	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12816
12817	fe_done =
12818	    control_softc->ctl_ports[targ_port]->fe_done;
12819
12820#ifdef CTL_TIME_IO
12821	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12822		char str[256];
12823		char path_str[64];
12824		struct sbuf sb;
12825
12826		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12827		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12828
12829		sbuf_cat(&sb, path_str);
12830		switch (io->io_hdr.io_type) {
12831		case CTL_IO_SCSI:
12832			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12833			sbuf_printf(&sb, "\n");
12834			sbuf_cat(&sb, path_str);
12835			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12836				    io->scsiio.tag_num, io->scsiio.tag_type);
12837			break;
12838		case CTL_IO_TASK:
12839			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12840				    "Tag Type: %d\n", io->taskio.task_action,
12841				    io->taskio.tag_num, io->taskio.tag_type);
12842			break;
12843		default:
12844			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12845			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12846			break;
12847		}
12848		sbuf_cat(&sb, path_str);
12849		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12850			    (intmax_t)time_uptime - io->io_hdr.start_time);
12851		sbuf_finish(&sb);
12852		printf("%s", sbuf_data(&sb));
12853	}
12854#endif /* CTL_TIME_IO */
12855
12856	switch (io->io_hdr.io_type) {
12857	case CTL_IO_SCSI:
12858		break;
12859	case CTL_IO_TASK:
12860		if (bootverbose || verbose > 0)
12861			ctl_io_error_print(io, NULL);
12862		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12863			ctl_free_io(io);
12864		else
12865			fe_done(io);
12866		return (CTL_RETVAL_COMPLETE);
12867		break;
12868	default:
12869		printf("ctl_process_done: invalid io type %d\n",
12870		       io->io_hdr.io_type);
12871		panic("ctl_process_done: invalid io type %d\n",
12872		      io->io_hdr.io_type);
12873		break; /* NOTREACHED */
12874	}
12875
12876	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12877	if (lun == NULL) {
12878		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12879				 io->io_hdr.nexus.targ_mapped_lun));
12880		fe_done(io);
12881		goto bailout;
12882	}
12883	ctl_softc = lun->ctl_softc;
12884
12885	mtx_lock(&lun->lun_lock);
12886
12887	/*
12888	 * Check to see if we have any errors to inject here.  We only
12889	 * inject errors for commands that don't already have errors set.
12890	 */
12891	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12892	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12893		ctl_inject_error(lun, io);
12894
12895	/*
12896	 * XXX KDM how do we treat commands that aren't completed
12897	 * successfully?
12898	 *
12899	 * XXX KDM should we also track I/O latency?
12900	 */
12901	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12902	    io->io_hdr.io_type == CTL_IO_SCSI) {
12903#ifdef CTL_TIME_IO
12904		struct bintime cur_bt;
12905#endif
12906		int type;
12907
12908		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12909		    CTL_FLAG_DATA_IN)
12910			type = CTL_STATS_READ;
12911		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12912		    CTL_FLAG_DATA_OUT)
12913			type = CTL_STATS_WRITE;
12914		else
12915			type = CTL_STATS_NO_IO;
12916
12917		lun->stats.ports[targ_port].bytes[type] +=
12918		    io->scsiio.kern_total_len;
12919		lun->stats.ports[targ_port].operations[type]++;
12920#ifdef CTL_TIME_IO
12921		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
12922		   &io->io_hdr.dma_bt);
12923		lun->stats.ports[targ_port].num_dmas[type] +=
12924		    io->io_hdr.num_dmas;
12925		getbintime(&cur_bt);
12926		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12927		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
12928#endif
12929	}
12930
12931	/*
12932	 * Remove this from the OOA queue.
12933	 */
12934	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12935
12936	/*
12937	 * Run through the blocked queue on this LUN and see if anything
12938	 * has become unblocked, now that this transaction is done.
12939	 */
12940	ctl_check_blocked(lun);
12941
12942	/*
12943	 * If the LUN has been invalidated, free it if there is nothing
12944	 * left on its OOA queue.
12945	 */
12946	if ((lun->flags & CTL_LUN_INVALID)
12947	 && TAILQ_EMPTY(&lun->ooa_queue)) {
12948		mtx_unlock(&lun->lun_lock);
12949		mtx_lock(&ctl_softc->ctl_lock);
12950		ctl_free_lun(lun);
12951		mtx_unlock(&ctl_softc->ctl_lock);
12952	} else
12953		mtx_unlock(&lun->lun_lock);
12954
12955	/*
12956	 * If this command has been aborted, make sure we set the status
12957	 * properly.  The FETD is responsible for freeing the I/O and doing
12958	 * whatever it needs to do to clean up its state.
12959	 */
12960	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12961		io->io_hdr.status = CTL_CMD_ABORTED;
12962
12963	/*
12964	 * We print out status for every task management command.  For SCSI
12965	 * commands, we filter out any unit attention errors; they happen
12966	 * on every boot, and would clutter up the log.  Note:  task
12967	 * management commands aren't printed here, they are printed above,
12968	 * since they should never even make it down here.
12969	 */
12970	switch (io->io_hdr.io_type) {
12971	case CTL_IO_SCSI: {
12972		int error_code, sense_key, asc, ascq;
12973
12974		sense_key = 0;
12975
12976		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12977		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12978			/*
12979			 * Since this is just for printing, no need to
12980			 * show errors here.
12981			 */
12982			scsi_extract_sense_len(&io->scsiio.sense_data,
12983					       io->scsiio.sense_len,
12984					       &error_code,
12985					       &sense_key,
12986					       &asc,
12987					       &ascq,
12988					       /*show_errors*/ 0);
12989		}
12990
12991		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12992		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12993		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12994		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12995
12996			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12997				ctl_softc->skipped_prints++;
12998			} else {
12999				uint32_t skipped_prints;
13000
13001				skipped_prints = ctl_softc->skipped_prints;
13002
13003				ctl_softc->skipped_prints = 0;
13004				ctl_softc->last_print_jiffies = time_uptime;
13005
13006				if (skipped_prints > 0) {
13007#ifdef NEEDTOPORT
13008					csevent_log(CSC_CTL | CSC_SHELF_SW |
13009					    CTL_ERROR_REPORT,
13010					    csevent_LogType_Trace,
13011					    csevent_Severity_Information,
13012					    csevent_AlertLevel_Green,
13013					    csevent_FRU_Firmware,
13014					    csevent_FRU_Unknown,
13015					    "High CTL error volume, %d prints "
13016					    "skipped", skipped_prints);
13017#endif
13018				}
13019				if (bootverbose || verbose > 0)
13020					ctl_io_error_print(io, NULL);
13021			}
13022		}
13023		break;
13024	}
13025	case CTL_IO_TASK:
13026		if (bootverbose || verbose > 0)
13027			ctl_io_error_print(io, NULL);
13028		break;
13029	default:
13030		break;
13031	}
13032
13033	/*
13034	 * Tell the FETD or the other shelf controller we're done with this
13035	 * command.  Note that only SCSI commands get to this point.  Task
13036	 * management commands are completed above.
13037	 *
13038	 * We only send status to the other controller if we're in XFER
13039	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13040	 * received the I/O (from CTL's perspective), and so the status is
13041	 * generated there.
13042	 *
13043	 * XXX KDM if we hold the lock here, we could cause a deadlock
13044	 * if the frontend comes back in in this context to queue
13045	 * something.
13046	 */
13047	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13048	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13049		union ctl_ha_msg msg;
13050
13051		memset(&msg, 0, sizeof(msg));
13052		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13053		msg.hdr.original_sc = io->io_hdr.original_sc;
13054		msg.hdr.nexus = io->io_hdr.nexus;
13055		msg.hdr.status = io->io_hdr.status;
13056		msg.scsi.scsi_status = io->scsiio.scsi_status;
13057		msg.scsi.tag_num = io->scsiio.tag_num;
13058		msg.scsi.tag_type = io->scsiio.tag_type;
13059		msg.scsi.sense_len = io->scsiio.sense_len;
13060		msg.scsi.sense_residual = io->scsiio.sense_residual;
13061		msg.scsi.residual = io->scsiio.residual;
13062		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13063		       sizeof(io->scsiio.sense_data));
13064		/*
13065		 * We copy this whether or not this is an I/O-related
13066		 * command.  Otherwise, we'd have to go and check to see
13067		 * whether it's a read/write command, and it really isn't
13068		 * worth it.
13069		 */
13070		memcpy(&msg.scsi.lbalen,
13071		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13072		       sizeof(msg.scsi.lbalen));
13073
13074		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13075				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13076			/* XXX do something here */
13077		}
13078
13079		ctl_free_io(io);
13080	} else
13081		fe_done(io);
13082
13083bailout:
13084
13085	return (CTL_RETVAL_COMPLETE);
13086}
13087
13088/*
13089 * Front end should call this if it doesn't do autosense.  When the request
13090 * sense comes back in from the initiator, we'll dequeue this and send it.
13091 */
13092int
13093ctl_queue_sense(union ctl_io *io)
13094{
13095	struct ctl_lun *lun;
13096	struct ctl_softc *ctl_softc;
13097	uint32_t initidx, targ_lun;
13098
13099	ctl_softc = control_softc;
13100
13101	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13102
13103	/*
13104	 * LUN lookup will likely move to the ctl_work_thread() once we
13105	 * have our new queueing infrastructure (that doesn't put things on
13106	 * a per-LUN queue initially).  That is so that we can handle
13107	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13108	 * can't deal with that right now.
13109	 */
13110	mtx_lock(&ctl_softc->ctl_lock);
13111
13112	/*
13113	 * If we don't have a LUN for this, just toss the sense
13114	 * information.
13115	 */
13116	targ_lun = io->io_hdr.nexus.targ_lun;
13117	if (io->io_hdr.nexus.lun_map_fn != NULL)
13118		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13119	if ((targ_lun < CTL_MAX_LUNS)
13120	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13121		lun = ctl_softc->ctl_luns[targ_lun];
13122	else
13123		goto bailout;
13124
13125	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13126
13127	mtx_lock(&lun->lun_lock);
13128	/*
13129	 * Already have CA set for this LUN...toss the sense information.
13130	 */
13131	if (ctl_is_set(lun->have_ca, initidx)) {
13132		mtx_unlock(&lun->lun_lock);
13133		goto bailout;
13134	}
13135
13136	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13137	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13138	       sizeof(io->scsiio.sense_data)));
13139	ctl_set_mask(lun->have_ca, initidx);
13140	mtx_unlock(&lun->lun_lock);
13141
13142bailout:
13143	mtx_unlock(&ctl_softc->ctl_lock);
13144
13145	ctl_free_io(io);
13146
13147	return (CTL_RETVAL_COMPLETE);
13148}
13149
13150/*
13151 * Primary command inlet from frontend ports.  All SCSI and task I/O
13152 * requests must go through this function.
13153 */
13154int
13155ctl_queue(union ctl_io *io)
13156{
13157	struct ctl_softc *ctl_softc;
13158
13159	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13160
13161	ctl_softc = control_softc;
13162
13163#ifdef CTL_TIME_IO
13164	io->io_hdr.start_time = time_uptime;
13165	getbintime(&io->io_hdr.start_bt);
13166#endif /* CTL_TIME_IO */
13167
13168	/* Map FE-specific LUN ID into global one. */
13169	if (io->io_hdr.nexus.lun_map_fn != NULL)
13170		io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.lun_map_fn(
13171		    io->io_hdr.nexus.lun_map_arg, io->io_hdr.nexus.targ_lun);
13172	else
13173		io->io_hdr.nexus.targ_mapped_lun = io->io_hdr.nexus.targ_lun;
13174
13175	switch (io->io_hdr.io_type) {
13176	case CTL_IO_SCSI:
13177		ctl_enqueue_incoming(io);
13178		break;
13179	case CTL_IO_TASK:
13180		ctl_run_task(io);
13181		break;
13182	default:
13183		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13184		return (EINVAL);
13185	}
13186
13187	return (CTL_RETVAL_COMPLETE);
13188}
13189
13190#ifdef CTL_IO_DELAY
13191static void
13192ctl_done_timer_wakeup(void *arg)
13193{
13194	union ctl_io *io;
13195
13196	io = (union ctl_io *)arg;
13197	ctl_done(io);
13198}
13199#endif /* CTL_IO_DELAY */
13200
13201void
13202ctl_done(union ctl_io *io)
13203{
13204	struct ctl_softc *ctl_softc;
13205
13206	ctl_softc = control_softc;
13207
13208	/*
13209	 * Enable this to catch duplicate completion issues.
13210	 */
13211#if 0
13212	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13213		printf("%s: type %d msg %d cdb %x iptl: "
13214		       "%d:%d:%d:%d tag 0x%04x "
13215		       "flag %#x status %x\n",
13216			__func__,
13217			io->io_hdr.io_type,
13218			io->io_hdr.msg_type,
13219			io->scsiio.cdb[0],
13220			io->io_hdr.nexus.initid.id,
13221			io->io_hdr.nexus.targ_port,
13222			io->io_hdr.nexus.targ_target.id,
13223			io->io_hdr.nexus.targ_lun,
13224			(io->io_hdr.io_type ==
13225			CTL_IO_TASK) ?
13226			io->taskio.tag_num :
13227			io->scsiio.tag_num,
13228		        io->io_hdr.flags,
13229			io->io_hdr.status);
13230	} else
13231		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13232#endif
13233
13234	/*
13235	 * This is an internal copy of an I/O, and should not go through
13236	 * the normal done processing logic.
13237	 */
13238	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13239		return;
13240
13241	/*
13242	 * We need to send a msg to the serializing shelf to finish the IO
13243	 * as well.  We don't send a finish message to the other shelf if
13244	 * this is a task management command.  Task management commands
13245	 * aren't serialized in the OOA queue, but rather just executed on
13246	 * both shelf controllers for commands that originated on that
13247	 * controller.
13248	 */
13249	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13250	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13251		union ctl_ha_msg msg_io;
13252
13253		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13254		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13255		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13256		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13257		}
13258		/* continue on to finish IO */
13259	}
13260#ifdef CTL_IO_DELAY
13261	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13262		struct ctl_lun *lun;
13263
13264		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13265
13266		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13267	} else {
13268		struct ctl_lun *lun;
13269
13270		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13271
13272		if ((lun != NULL)
13273		 && (lun->delay_info.done_delay > 0)) {
13274			struct callout *callout;
13275
13276			callout = (struct callout *)&io->io_hdr.timer_bytes;
13277			callout_init(callout, /*mpsafe*/ 1);
13278			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13279			callout_reset(callout,
13280				      lun->delay_info.done_delay * hz,
13281				      ctl_done_timer_wakeup, io);
13282			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13283				lun->delay_info.done_delay = 0;
13284			return;
13285		}
13286	}
13287#endif /* CTL_IO_DELAY */
13288
13289	ctl_enqueue_done(io);
13290}
13291
13292int
13293ctl_isc(struct ctl_scsiio *ctsio)
13294{
13295	struct ctl_lun *lun;
13296	int retval;
13297
13298	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13299
13300	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13301
13302	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13303
13304	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13305
13306	return (retval);
13307}
13308
13309
13310static void
13311ctl_work_thread(void *arg)
13312{
13313	struct ctl_thread *thr = (struct ctl_thread *)arg;
13314	struct ctl_softc *softc = thr->ctl_softc;
13315	union ctl_io *io;
13316	int retval;
13317
13318	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13319
13320	for (;;) {
13321		retval = 0;
13322
13323		/*
13324		 * We handle the queues in this order:
13325		 * - ISC
13326		 * - done queue (to free up resources, unblock other commands)
13327		 * - RtR queue
13328		 * - incoming queue
13329		 *
13330		 * If those queues are empty, we break out of the loop and
13331		 * go to sleep.
13332		 */
13333		mtx_lock(&thr->queue_lock);
13334		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13335		if (io != NULL) {
13336			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13337			mtx_unlock(&thr->queue_lock);
13338			ctl_handle_isc(io);
13339			continue;
13340		}
13341		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13342		if (io != NULL) {
13343			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13344			/* clear any blocked commands, call fe_done */
13345			mtx_unlock(&thr->queue_lock);
13346			retval = ctl_process_done(io);
13347			continue;
13348		}
13349		if (!ctl_pause_rtr) {
13350			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13351			if (io != NULL) {
13352				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13353				mtx_unlock(&thr->queue_lock);
13354				retval = ctl_scsiio(&io->scsiio);
13355				if (retval != CTL_RETVAL_COMPLETE)
13356					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13357				continue;
13358			}
13359		}
13360		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13361		if (io != NULL) {
13362			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13363			mtx_unlock(&thr->queue_lock);
13364			ctl_scsiio_precheck(softc, &io->scsiio);
13365			continue;
13366		}
13367
13368		/* Sleep until we have something to do. */
13369		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13370	}
13371}
13372
13373static void
13374ctl_lun_thread(void *arg)
13375{
13376	struct ctl_softc *softc = (struct ctl_softc *)arg;
13377	struct ctl_be_lun *be_lun;
13378	int retval;
13379
13380	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13381
13382	for (;;) {
13383		retval = 0;
13384		mtx_lock(&softc->ctl_lock);
13385		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13386		if (be_lun != NULL) {
13387			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13388			mtx_unlock(&softc->ctl_lock);
13389			ctl_create_lun(be_lun);
13390			continue;
13391		}
13392
13393		/* Sleep until we have something to do. */
13394		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13395		    PDROP | PRIBIO, "-", 0);
13396	}
13397}
13398
13399static void
13400ctl_enqueue_incoming(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->incoming_queue, &io->io_hdr, links);
13408	mtx_unlock(&thr->queue_lock);
13409	wakeup(thr);
13410}
13411
13412static void
13413ctl_enqueue_rtr(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->rtr_queue, &io->io_hdr, links);
13421	mtx_unlock(&thr->queue_lock);
13422	wakeup(thr);
13423}
13424
13425static void
13426ctl_enqueue_done(union ctl_io *io)
13427{
13428	struct ctl_softc *softc = control_softc;
13429	struct ctl_thread *thr;
13430
13431	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13432	mtx_lock(&thr->queue_lock);
13433	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13434	mtx_unlock(&thr->queue_lock);
13435	wakeup(thr);
13436}
13437
13438static void
13439ctl_enqueue_isc(union ctl_io *io)
13440{
13441	struct ctl_softc *softc = control_softc;
13442	struct ctl_thread *thr;
13443
13444	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13445	mtx_lock(&thr->queue_lock);
13446	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13447	mtx_unlock(&thr->queue_lock);
13448	wakeup(thr);
13449}
13450
13451/* Initialization and failover */
13452
13453void
13454ctl_init_isc_msg(void)
13455{
13456	printf("CTL: Still calling this thing\n");
13457}
13458
13459/*
13460 * Init component
13461 * 	Initializes component into configuration defined by bootMode
13462 *	(see hasc-sv.c)
13463 *  	returns hasc_Status:
13464 * 		OK
13465 *		ERROR - fatal error
13466 */
13467static ctl_ha_comp_status
13468ctl_isc_init(struct ctl_ha_component *c)
13469{
13470	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13471
13472	c->status = ret;
13473	return ret;
13474}
13475
13476/* Start component
13477 * 	Starts component in state requested. If component starts successfully,
13478 *	it must set its own state to the requestrd state
13479 *	When requested state is HASC_STATE_HA, the component may refine it
13480 * 	by adding _SLAVE or _MASTER flags.
13481 *	Currently allowed state transitions are:
13482 *	UNKNOWN->HA		- initial startup
13483 *	UNKNOWN->SINGLE - initial startup when no parter detected
13484 *	HA->SINGLE		- failover
13485 * returns ctl_ha_comp_status:
13486 * 		OK	- component successfully started in requested state
13487 *		FAILED  - could not start the requested state, failover may
13488 * 			  be possible
13489 *		ERROR	- fatal error detected, no future startup possible
13490 */
13491static ctl_ha_comp_status
13492ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13493{
13494	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13495
13496	printf("%s: go\n", __func__);
13497
13498	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13499	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13500		ctl_is_single = 0;
13501		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13502		    != CTL_HA_STATUS_SUCCESS) {
13503			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13504			ret = CTL_HA_COMP_STATUS_ERROR;
13505		}
13506	} else if (CTL_HA_STATE_IS_HA(c->state)
13507		&& CTL_HA_STATE_IS_SINGLE(state)){
13508		// HA->SINGLE transition
13509	        ctl_failover();
13510		ctl_is_single = 1;
13511	} else {
13512		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13513		       c->state, state);
13514		ret = CTL_HA_COMP_STATUS_ERROR;
13515	}
13516	if (CTL_HA_STATE_IS_SINGLE(state))
13517		ctl_is_single = 1;
13518
13519	c->state = state;
13520	c->status = ret;
13521	return ret;
13522}
13523
13524/*
13525 * Quiesce component
13526 * The component must clear any error conditions (set status to OK) and
13527 * prepare itself to another Start call
13528 * returns ctl_ha_comp_status:
13529 * 	OK
13530 *	ERROR
13531 */
13532static ctl_ha_comp_status
13533ctl_isc_quiesce(struct ctl_ha_component *c)
13534{
13535	int ret = CTL_HA_COMP_STATUS_OK;
13536
13537	ctl_pause_rtr = 1;
13538	c->status = ret;
13539	return ret;
13540}
13541
13542struct ctl_ha_component ctl_ha_component_ctlisc =
13543{
13544	.name = "CTL ISC",
13545	.state = CTL_HA_STATE_UNKNOWN,
13546	.init = ctl_isc_init,
13547	.start = ctl_isc_start,
13548	.quiesce = ctl_isc_quiesce
13549};
13550
13551/*
13552 *  vim: ts=8
13553 */
13554