ctl.c revision 264264
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 264264 2014-04-08 18:22:03Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * The default is to run with CTL_DONE_THREAD turned on.  Completed
87 * transactions are queued for processing by the CTL work thread.  When
88 * CTL_DONE_THREAD is not defined, completed transactions are processed in
89 * the caller's context.
90 */
91#define CTL_DONE_THREAD
92
93/*
94 * Use the serial number and device ID provided by the backend, rather than
95 * making up our own.
96 */
97#define CTL_USE_BACKEND_SN
98
99/*
100 * Size and alignment macros needed for Copan-specific HA hardware.  These
101 * can go away when the HA code is re-written, and uses busdma for any
102 * hardware.
103 */
104#define	CTL_ALIGN_8B(target, source, type)				\
105	if (((uint32_t)source & 0x7) != 0)				\
106		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
107	else								\
108		target = (type)source;
109
110#define	CTL_SIZE_8B(target, size)					\
111	if ((size & 0x7) != 0)						\
112		target = size + (0x8 - (size & 0x7));			\
113	else								\
114		target = size;
115
116#define CTL_ALIGN_8B_MARGIN	16
117
118/*
119 * Template mode pages.
120 */
121
122/*
123 * Note that these are default values only.  The actual values will be
124 * filled in when the user does a mode sense.
125 */
126static struct copan_power_subpage power_page_default = {
127	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
128	/*subpage*/ PWR_SUBPAGE_CODE,
129	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
130			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
131	/*page_version*/ PWR_VERSION,
132	/* total_luns */ 26,
133	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
134	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
136		      0, 0, 0, 0, 0, 0}
137};
138
139static struct copan_power_subpage power_page_changeable = {
140	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
141	/*subpage*/ PWR_SUBPAGE_CODE,
142	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
143			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
144	/*page_version*/ 0,
145	/* total_luns */ 0,
146	/* max_active_luns*/ 0,
147	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
148		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
149		      0, 0, 0, 0, 0, 0}
150};
151
152static struct copan_aps_subpage aps_page_default = {
153	APS_PAGE_CODE | SMPH_SPF, //page_code
154	APS_SUBPAGE_CODE, //subpage
155	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
156	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
157	APS_VERSION, //page_version
158	0, //lock_active
159	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
160	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
161	0, 0, 0, 0, 0} //reserved
162};
163
164static struct copan_aps_subpage aps_page_changeable = {
165	APS_PAGE_CODE | SMPH_SPF, //page_code
166	APS_SUBPAGE_CODE, //subpage
167	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
168	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
169	0, //page_version
170	0, //lock_active
171	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
173	0, 0, 0, 0, 0} //reserved
174};
175
176static struct copan_debugconf_subpage debugconf_page_default = {
177	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
178	DBGCNF_SUBPAGE_CODE,		/* subpage */
179	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
180	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
181	DBGCNF_VERSION,			/* page_version */
182	{CTL_TIME_IO_DEFAULT_SECS>>8,
183	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
184};
185
186static struct copan_debugconf_subpage debugconf_page_changeable = {
187	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
188	DBGCNF_SUBPAGE_CODE,		/* subpage */
189	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
190	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
191	0,				/* page_version */
192	{0xff,0xff},			/* ctl_time_io_secs */
193};
194
195static struct scsi_format_page format_page_default = {
196	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
197	/*page_length*/sizeof(struct scsi_format_page) - 2,
198	/*tracks_per_zone*/ {0, 0},
199	/*alt_sectors_per_zone*/ {0, 0},
200	/*alt_tracks_per_zone*/ {0, 0},
201	/*alt_tracks_per_lun*/ {0, 0},
202	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
203			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
204	/*bytes_per_sector*/ {0, 0},
205	/*interleave*/ {0, 0},
206	/*track_skew*/ {0, 0},
207	/*cylinder_skew*/ {0, 0},
208	/*flags*/ SFP_HSEC,
209	/*reserved*/ {0, 0, 0}
210};
211
212static struct scsi_format_page format_page_changeable = {
213	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
214	/*page_length*/sizeof(struct scsi_format_page) - 2,
215	/*tracks_per_zone*/ {0, 0},
216	/*alt_sectors_per_zone*/ {0, 0},
217	/*alt_tracks_per_zone*/ {0, 0},
218	/*alt_tracks_per_lun*/ {0, 0},
219	/*sectors_per_track*/ {0, 0},
220	/*bytes_per_sector*/ {0, 0},
221	/*interleave*/ {0, 0},
222	/*track_skew*/ {0, 0},
223	/*cylinder_skew*/ {0, 0},
224	/*flags*/ 0,
225	/*reserved*/ {0, 0, 0}
226};
227
228static struct scsi_rigid_disk_page rigid_disk_page_default = {
229	/*page_code*/SMS_RIGID_DISK_PAGE,
230	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
231	/*cylinders*/ {0, 0, 0},
232	/*heads*/ CTL_DEFAULT_HEADS,
233	/*start_write_precomp*/ {0, 0, 0},
234	/*start_reduced_current*/ {0, 0, 0},
235	/*step_rate*/ {0, 0},
236	/*landing_zone_cylinder*/ {0, 0, 0},
237	/*rpl*/ SRDP_RPL_DISABLED,
238	/*rotational_offset*/ 0,
239	/*reserved1*/ 0,
240	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
241			   CTL_DEFAULT_ROTATION_RATE & 0xff},
242	/*reserved2*/ {0, 0}
243};
244
245static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
246	/*page_code*/SMS_RIGID_DISK_PAGE,
247	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
248	/*cylinders*/ {0, 0, 0},
249	/*heads*/ 0,
250	/*start_write_precomp*/ {0, 0, 0},
251	/*start_reduced_current*/ {0, 0, 0},
252	/*step_rate*/ {0, 0},
253	/*landing_zone_cylinder*/ {0, 0, 0},
254	/*rpl*/ 0,
255	/*rotational_offset*/ 0,
256	/*reserved1*/ 0,
257	/*rotation_rate*/ {0, 0},
258	/*reserved2*/ {0, 0}
259};
260
261static struct scsi_caching_page caching_page_default = {
262	/*page_code*/SMS_CACHING_PAGE,
263	/*page_length*/sizeof(struct scsi_caching_page) - 2,
264	/*flags1*/ SCP_DISC | SCP_WCE,
265	/*ret_priority*/ 0,
266	/*disable_pf_transfer_len*/ {0xff, 0xff},
267	/*min_prefetch*/ {0, 0},
268	/*max_prefetch*/ {0xff, 0xff},
269	/*max_pf_ceiling*/ {0xff, 0xff},
270	/*flags2*/ 0,
271	/*cache_segments*/ 0,
272	/*cache_seg_size*/ {0, 0},
273	/*reserved*/ 0,
274	/*non_cache_seg_size*/ {0, 0, 0}
275};
276
277static struct scsi_caching_page caching_page_changeable = {
278	/*page_code*/SMS_CACHING_PAGE,
279	/*page_length*/sizeof(struct scsi_caching_page) - 2,
280	/*flags1*/ 0,
281	/*ret_priority*/ 0,
282	/*disable_pf_transfer_len*/ {0, 0},
283	/*min_prefetch*/ {0, 0},
284	/*max_prefetch*/ {0, 0},
285	/*max_pf_ceiling*/ {0, 0},
286	/*flags2*/ 0,
287	/*cache_segments*/ 0,
288	/*cache_seg_size*/ {0, 0},
289	/*reserved*/ 0,
290	/*non_cache_seg_size*/ {0, 0, 0}
291};
292
293static struct scsi_control_page control_page_default = {
294	/*page_code*/SMS_CONTROL_MODE_PAGE,
295	/*page_length*/sizeof(struct scsi_control_page) - 2,
296	/*rlec*/0,
297	/*queue_flags*/0,
298	/*eca_and_aen*/0,
299	/*reserved*/0,
300	/*aen_holdoff_period*/{0, 0}
301};
302
303static struct scsi_control_page control_page_changeable = {
304	/*page_code*/SMS_CONTROL_MODE_PAGE,
305	/*page_length*/sizeof(struct scsi_control_page) - 2,
306	/*rlec*/SCP_DSENSE,
307	/*queue_flags*/0,
308	/*eca_and_aen*/0,
309	/*reserved*/0,
310	/*aen_holdoff_period*/{0, 0}
311};
312
313
314/*
315 * XXX KDM move these into the softc.
316 */
317static int rcv_sync_msg;
318static int persis_offset;
319static uint8_t ctl_pause_rtr;
320static int     ctl_is_single = 1;
321static int     index_to_aps_page;
322
323SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
324static int worker_threads = 1;
325TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
326SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
327    &worker_threads, 1, "Number of worker threads");
328static int verbose = 0;
329TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
330SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
331    &verbose, 0, "Show SCSI errors returned to initiator");
332
333/*
334 * Serial number (0x80), device id (0x83), and supported pages (0x00)
335 */
336#define SCSI_EVPD_NUM_SUPPORTED_PAGES	3
337
338static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
339				  int param);
340static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
341static int ctl_init(void);
342void ctl_shutdown(void);
343static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
344static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
345static void ctl_ioctl_online(void *arg);
346static void ctl_ioctl_offline(void *arg);
347static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
348static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
349static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
350static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
351static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
352static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
353static int ctl_ioctl_submit_wait(union ctl_io *io);
354static void ctl_ioctl_datamove(union ctl_io *io);
355static void ctl_ioctl_done(union ctl_io *io);
356static void ctl_ioctl_hard_startstop_callback(void *arg,
357					      struct cfi_metatask *metatask);
358static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
359static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
360			      struct ctl_ooa *ooa_hdr,
361			      struct ctl_ooa_entry *kern_entries);
362static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
363		     struct thread *td);
364uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
365uint32_t ctl_port_idx(int port_num);
366#ifdef unused
367static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
368				   uint32_t targ_target, uint32_t targ_lun,
369				   int can_wait);
370static void ctl_kfree_io(union ctl_io *io);
371#endif /* unused */
372static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
373			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
374static int ctl_free_lun(struct ctl_lun *lun);
375static void ctl_create_lun(struct ctl_be_lun *be_lun);
376/**
377static void ctl_failover_change_pages(struct ctl_softc *softc,
378				      struct ctl_scsiio *ctsio, int master);
379**/
380
381static int ctl_do_mode_select(union ctl_io *io);
382static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
383			   uint64_t res_key, uint64_t sa_res_key,
384			   uint8_t type, uint32_t residx,
385			   struct ctl_scsiio *ctsio,
386			   struct scsi_per_res_out *cdb,
387			   struct scsi_per_res_out_parms* param);
388static void ctl_pro_preempt_other(struct ctl_lun *lun,
389				  union ctl_ha_msg *msg);
390static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
391static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
392static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
394static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
395static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
396static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
397static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
398static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
399					 union ctl_io *ooa_io);
400static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
401				union ctl_io *starting_io);
402static int ctl_check_blocked(struct ctl_lun *lun);
403static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
404				struct ctl_lun *lun,
405				struct ctl_cmd_entry *entry,
406				struct ctl_scsiio *ctsio);
407//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
408static void ctl_failover(void);
409static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
410			       struct ctl_scsiio *ctsio);
411static int ctl_scsiio(struct ctl_scsiio *ctsio);
412
413static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
414static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
415			    ctl_ua_type ua_type);
416static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
417			 ctl_ua_type ua_type);
418static int ctl_abort_task(union ctl_io *io);
419static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
420#ifdef CTL_IO_DELAY
421static void ctl_datamove_timer_wakeup(void *arg);
422static void ctl_done_timer_wakeup(void *arg);
423#endif /* CTL_IO_DELAY */
424
425static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
426static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
427static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
428static void ctl_datamove_remote_write(union ctl_io *io);
429static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
430static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
431static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
432static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
433				    ctl_ha_dt_cb callback);
434static void ctl_datamove_remote_read(union ctl_io *io);
435static void ctl_datamove_remote(union ctl_io *io);
436static int ctl_process_done(union ctl_io *io, int have_lock);
437static void ctl_work_thread(void *arg);
438
439/*
440 * Load the serialization table.  This isn't very pretty, but is probably
441 * the easiest way to do it.
442 */
443#include "ctl_ser_table.c"
444
445/*
446 * We only need to define open, close and ioctl routines for this driver.
447 */
448static struct cdevsw ctl_cdevsw = {
449	.d_version =	D_VERSION,
450	.d_flags =	0,
451	.d_open =	ctl_open,
452	.d_close =	ctl_close,
453	.d_ioctl =	ctl_ioctl,
454	.d_name =	"ctl",
455};
456
457
458MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
459
460static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
461
462static moduledata_t ctl_moduledata = {
463	"ctl",
464	ctl_module_event_handler,
465	NULL
466};
467
468DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
469MODULE_VERSION(ctl, 1);
470
471static void
472ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
473			    union ctl_ha_msg *msg_info)
474{
475	struct ctl_scsiio *ctsio;
476
477	if (msg_info->hdr.original_sc == NULL) {
478		printf("%s: original_sc == NULL!\n", __func__);
479		/* XXX KDM now what? */
480		return;
481	}
482
483	ctsio = &msg_info->hdr.original_sc->scsiio;
484	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
485	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
486	ctsio->io_hdr.status = msg_info->hdr.status;
487	ctsio->scsi_status = msg_info->scsi.scsi_status;
488	ctsio->sense_len = msg_info->scsi.sense_len;
489	ctsio->sense_residual = msg_info->scsi.sense_residual;
490	ctsio->residual = msg_info->scsi.residual;
491	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
492	       sizeof(ctsio->sense_data));
493	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
494	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
495	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
496	ctl_wakeup_thread();
497}
498
499static void
500ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
501				union ctl_ha_msg *msg_info)
502{
503	struct ctl_scsiio *ctsio;
504
505	if (msg_info->hdr.serializing_sc == NULL) {
506		printf("%s: serializing_sc == NULL!\n", __func__);
507		/* XXX KDM now what? */
508		return;
509	}
510
511	ctsio = &msg_info->hdr.serializing_sc->scsiio;
512#if 0
513	/*
514	 * Attempt to catch the situation where an I/O has
515	 * been freed, and we're using it again.
516	 */
517	if (ctsio->io_hdr.io_type == 0xff) {
518		union ctl_io *tmp_io;
519		tmp_io = (union ctl_io *)ctsio;
520		printf("%s: %p use after free!\n", __func__,
521		       ctsio);
522		printf("%s: type %d msg %d cdb %x iptl: "
523		       "%d:%d:%d:%d tag 0x%04x "
524		       "flag %#x status %x\n",
525			__func__,
526			tmp_io->io_hdr.io_type,
527			tmp_io->io_hdr.msg_type,
528			tmp_io->scsiio.cdb[0],
529			tmp_io->io_hdr.nexus.initid.id,
530			tmp_io->io_hdr.nexus.targ_port,
531			tmp_io->io_hdr.nexus.targ_target.id,
532			tmp_io->io_hdr.nexus.targ_lun,
533			(tmp_io->io_hdr.io_type ==
534			CTL_IO_TASK) ?
535			tmp_io->taskio.tag_num :
536			tmp_io->scsiio.tag_num,
537		        tmp_io->io_hdr.flags,
538			tmp_io->io_hdr.status);
539	}
540#endif
541	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
542	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
543	ctl_wakeup_thread();
544}
545
546/*
547 * ISC (Inter Shelf Communication) event handler.  Events from the HA
548 * subsystem come in here.
549 */
550static void
551ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
552{
553	struct ctl_softc *ctl_softc;
554	union ctl_io *io;
555	struct ctl_prio *presio;
556	ctl_ha_status isc_status;
557
558	ctl_softc = control_softc;
559	io = NULL;
560
561
562#if 0
563	printf("CTL: Isc Msg event %d\n", event);
564#endif
565	if (event == CTL_HA_EVT_MSG_RECV) {
566		union ctl_ha_msg msg_info;
567
568		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
569					     sizeof(msg_info), /*wait*/ 0);
570#if 0
571		printf("CTL: msg_type %d\n", msg_info.msg_type);
572#endif
573		if (isc_status != 0) {
574			printf("Error receiving message, status = %d\n",
575			       isc_status);
576			return;
577		}
578		mtx_lock(&ctl_softc->ctl_lock);
579
580		switch (msg_info.hdr.msg_type) {
581		case CTL_MSG_SERIALIZE:
582#if 0
583			printf("Serialize\n");
584#endif
585			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
586			if (io == NULL) {
587				printf("ctl_isc_event_handler: can't allocate "
588				       "ctl_io!\n");
589				/* Bad Juju */
590				/* Need to set busy and send msg back */
591				mtx_unlock(&ctl_softc->ctl_lock);
592				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
593				msg_info.hdr.status = CTL_SCSI_ERROR;
594				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
595				msg_info.scsi.sense_len = 0;
596			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
597				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
598				}
599				goto bailout;
600			}
601			ctl_zero_io(io);
602			// populate ctsio from msg_info
603			io->io_hdr.io_type = CTL_IO_SCSI;
604			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
605			io->io_hdr.original_sc = msg_info.hdr.original_sc;
606#if 0
607			printf("pOrig %x\n", (int)msg_info.original_sc);
608#endif
609			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
610					    CTL_FLAG_IO_ACTIVE;
611			/*
612			 * If we're in serialization-only mode, we don't
613			 * want to go through full done processing.  Thus
614			 * the COPY flag.
615			 *
616			 * XXX KDM add another flag that is more specific.
617			 */
618			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
619				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
620			io->io_hdr.nexus = msg_info.hdr.nexus;
621#if 0
622			printf("targ %d, port %d, iid %d, lun %d\n",
623			       io->io_hdr.nexus.targ_target.id,
624			       io->io_hdr.nexus.targ_port,
625			       io->io_hdr.nexus.initid.id,
626			       io->io_hdr.nexus.targ_lun);
627#endif
628			io->scsiio.tag_num = msg_info.scsi.tag_num;
629			io->scsiio.tag_type = msg_info.scsi.tag_type;
630			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
631			       CTL_MAX_CDBLEN);
632			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
633				struct ctl_cmd_entry *entry;
634				uint8_t opcode;
635
636				opcode = io->scsiio.cdb[0];
637				entry = &ctl_cmd_table[opcode];
638				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
639				io->io_hdr.flags |=
640					entry->flags & CTL_FLAG_DATA_MASK;
641			}
642			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
643					   &io->io_hdr, links);
644			ctl_wakeup_thread();
645			break;
646
647		/* Performed on the Originating SC, XFER mode only */
648		case CTL_MSG_DATAMOVE: {
649			struct ctl_sg_entry *sgl;
650			int i, j;
651
652			io = msg_info.hdr.original_sc;
653			if (io == NULL) {
654				printf("%s: original_sc == NULL!\n", __func__);
655				/* XXX KDM do something here */
656				break;
657			}
658			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
659			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
660			/*
661			 * Keep track of this, we need to send it back over
662			 * when the datamove is complete.
663			 */
664			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
665
666			if (msg_info.dt.sg_sequence == 0) {
667				/*
668				 * XXX KDM we use the preallocated S/G list
669				 * here, but we'll need to change this to
670				 * dynamic allocation if we need larger S/G
671				 * lists.
672				 */
673				if (msg_info.dt.kern_sg_entries >
674				    sizeof(io->io_hdr.remote_sglist) /
675				    sizeof(io->io_hdr.remote_sglist[0])) {
676					printf("%s: number of S/G entries "
677					    "needed %u > allocated num %zd\n",
678					    __func__,
679					    msg_info.dt.kern_sg_entries,
680					    sizeof(io->io_hdr.remote_sglist)/
681					    sizeof(io->io_hdr.remote_sglist[0]));
682
683					/*
684					 * XXX KDM send a message back to
685					 * the other side to shut down the
686					 * DMA.  The error will come back
687					 * through via the normal channel.
688					 */
689					break;
690				}
691				sgl = io->io_hdr.remote_sglist;
692				memset(sgl, 0,
693				       sizeof(io->io_hdr.remote_sglist));
694
695				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
696
697				io->scsiio.kern_sg_entries =
698					msg_info.dt.kern_sg_entries;
699				io->scsiio.rem_sg_entries =
700					msg_info.dt.kern_sg_entries;
701				io->scsiio.kern_data_len =
702					msg_info.dt.kern_data_len;
703				io->scsiio.kern_total_len =
704					msg_info.dt.kern_total_len;
705				io->scsiio.kern_data_resid =
706					msg_info.dt.kern_data_resid;
707				io->scsiio.kern_rel_offset =
708					msg_info.dt.kern_rel_offset;
709				/*
710				 * Clear out per-DMA flags.
711				 */
712				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
713				/*
714				 * Add per-DMA flags that are set for this
715				 * particular DMA request.
716				 */
717				io->io_hdr.flags |= msg_info.dt.flags &
718						    CTL_FLAG_RDMA_MASK;
719			} else
720				sgl = (struct ctl_sg_entry *)
721					io->scsiio.kern_data_ptr;
722
723			for (i = msg_info.dt.sent_sg_entries, j = 0;
724			     i < (msg_info.dt.sent_sg_entries +
725			     msg_info.dt.cur_sg_entries); i++, j++) {
726				sgl[i].addr = msg_info.dt.sg_list[j].addr;
727				sgl[i].len = msg_info.dt.sg_list[j].len;
728
729#if 0
730				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
731				       __func__,
732				       msg_info.dt.sg_list[j].addr,
733				       msg_info.dt.sg_list[j].len,
734				       sgl[i].addr, sgl[i].len, j, i);
735#endif
736			}
737#if 0
738			memcpy(&sgl[msg_info.dt.sent_sg_entries],
739			       msg_info.dt.sg_list,
740			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
741#endif
742
743			/*
744			 * If this is the last piece of the I/O, we've got
745			 * the full S/G list.  Queue processing in the thread.
746			 * Otherwise wait for the next piece.
747			 */
748			if (msg_info.dt.sg_last != 0) {
749				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
750						   &io->io_hdr, links);
751				ctl_wakeup_thread();
752			}
753			break;
754		}
755		/* Performed on the Serializing (primary) SC, XFER mode only */
756		case CTL_MSG_DATAMOVE_DONE: {
757			if (msg_info.hdr.serializing_sc == NULL) {
758				printf("%s: serializing_sc == NULL!\n",
759				       __func__);
760				/* XXX KDM now what? */
761				break;
762			}
763			/*
764			 * We grab the sense information here in case
765			 * there was a failure, so we can return status
766			 * back to the initiator.
767			 */
768			io = msg_info.hdr.serializing_sc;
769			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
770			io->io_hdr.status = msg_info.hdr.status;
771			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
772			io->scsiio.sense_len = msg_info.scsi.sense_len;
773			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
774			io->io_hdr.port_status = msg_info.scsi.fetd_status;
775			io->scsiio.residual = msg_info.scsi.residual;
776			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
777			       sizeof(io->scsiio.sense_data));
778
779			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
780					   &io->io_hdr, links);
781			ctl_wakeup_thread();
782			break;
783		}
784
785		/* Preformed on Originating SC, SER_ONLY mode */
786		case CTL_MSG_R2R:
787			io = msg_info.hdr.original_sc;
788			if (io == NULL) {
789				printf("%s: Major Bummer\n", __func__);
790				mtx_unlock(&ctl_softc->ctl_lock);
791				return;
792			} else {
793#if 0
794				printf("pOrig %x\n",(int) ctsio);
795#endif
796			}
797			io->io_hdr.msg_type = CTL_MSG_R2R;
798			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
799			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
800					   &io->io_hdr, links);
801			ctl_wakeup_thread();
802			break;
803
804		/*
805		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
806		 * mode.
807		 * Performed on the Originating (i.e. secondary) SC in XFER
808		 * mode
809		 */
810		case CTL_MSG_FINISH_IO:
811			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
812				ctl_isc_handler_finish_xfer(ctl_softc,
813							    &msg_info);
814			else
815				ctl_isc_handler_finish_ser_only(ctl_softc,
816								&msg_info);
817			break;
818
819		/* Preformed on Originating SC */
820		case CTL_MSG_BAD_JUJU:
821			io = msg_info.hdr.original_sc;
822			if (io == NULL) {
823				printf("%s: Bad JUJU!, original_sc is NULL!\n",
824				       __func__);
825				break;
826			}
827			ctl_copy_sense_data(&msg_info, io);
828			/*
829			 * IO should have already been cleaned up on other
830			 * SC so clear this flag so we won't send a message
831			 * back to finish the IO there.
832			 */
833			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
834			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
835
836			/* io = msg_info.hdr.serializing_sc; */
837			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
838		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
839					   &io->io_hdr, links);
840			ctl_wakeup_thread();
841			break;
842
843		/* Handle resets sent from the other side */
844		case CTL_MSG_MANAGE_TASKS: {
845			struct ctl_taskio *taskio;
846			taskio = (struct ctl_taskio *)ctl_alloc_io(
847				(void *)ctl_softc->othersc_pool);
848			if (taskio == NULL) {
849				printf("ctl_isc_event_handler: can't allocate "
850				       "ctl_io!\n");
851				/* Bad Juju */
852				/* should I just call the proper reset func
853				   here??? */
854				mtx_unlock(&ctl_softc->ctl_lock);
855				goto bailout;
856			}
857			ctl_zero_io((union ctl_io *)taskio);
858			taskio->io_hdr.io_type = CTL_IO_TASK;
859			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
860			taskio->io_hdr.nexus = msg_info.hdr.nexus;
861			taskio->task_action = msg_info.task.task_action;
862			taskio->tag_num = msg_info.task.tag_num;
863			taskio->tag_type = msg_info.task.tag_type;
864#ifdef CTL_TIME_IO
865			taskio->io_hdr.start_time = time_uptime;
866			getbintime(&taskio->io_hdr.start_bt);
867#if 0
868			cs_prof_gettime(&taskio->io_hdr.start_ticks);
869#endif
870#endif /* CTL_TIME_IO */
871		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
872					   &taskio->io_hdr, links);
873			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
874			ctl_wakeup_thread();
875			break;
876		}
877		/* Persistent Reserve action which needs attention */
878		case CTL_MSG_PERS_ACTION:
879			presio = (struct ctl_prio *)ctl_alloc_io(
880				(void *)ctl_softc->othersc_pool);
881			if (presio == NULL) {
882				printf("ctl_isc_event_handler: can't allocate "
883				       "ctl_io!\n");
884				/* Bad Juju */
885				/* Need to set busy and send msg back */
886				mtx_unlock(&ctl_softc->ctl_lock);
887				goto bailout;
888			}
889			ctl_zero_io((union ctl_io *)presio);
890			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
891			presio->pr_msg = msg_info.pr;
892		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
893					   &presio->io_hdr, links);
894			ctl_wakeup_thread();
895			break;
896		case CTL_MSG_SYNC_FE:
897			rcv_sync_msg = 1;
898			break;
899		case CTL_MSG_APS_LOCK: {
900			// It's quicker to execute this then to
901			// queue it.
902			struct ctl_lun *lun;
903			struct ctl_page_index *page_index;
904			struct copan_aps_subpage *current_sp;
905			uint32_t targ_lun;
906
907			targ_lun = msg_info.hdr.nexus.targ_lun;
908			if (msg_info.hdr.nexus.lun_map_fn != NULL)
909				targ_lun = msg_info.hdr.nexus.lun_map_fn(msg_info.hdr.nexus.lun_map_arg, targ_lun);
910
911			lun = ctl_softc->ctl_luns[targ_lun];
912			page_index = &lun->mode_pages.index[index_to_aps_page];
913			current_sp = (struct copan_aps_subpage *)
914				     (page_index->page_data +
915				     (page_index->page_len * CTL_PAGE_CURRENT));
916
917			current_sp->lock_active = msg_info.aps.lock_flag;
918		        break;
919		}
920		default:
921		        printf("How did I get here?\n");
922		}
923		mtx_unlock(&ctl_softc->ctl_lock);
924	} else if (event == CTL_HA_EVT_MSG_SENT) {
925		if (param != CTL_HA_STATUS_SUCCESS) {
926			printf("Bad status from ctl_ha_msg_send status %d\n",
927			       param);
928		}
929		return;
930	} else if (event == CTL_HA_EVT_DISCONNECT) {
931		printf("CTL: Got a disconnect from Isc\n");
932		return;
933	} else {
934		printf("ctl_isc_event_handler: Unknown event %d\n", event);
935		return;
936	}
937
938bailout:
939	return;
940}
941
942static void
943ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
944{
945	struct scsi_sense_data *sense;
946
947	sense = &dest->scsiio.sense_data;
948	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
949	dest->scsiio.scsi_status = src->scsi.scsi_status;
950	dest->scsiio.sense_len = src->scsi.sense_len;
951	dest->io_hdr.status = src->hdr.status;
952}
953
954static int
955ctl_init(void)
956{
957	struct ctl_softc *softc;
958	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
959	struct ctl_frontend *fe;
960	struct ctl_lun *lun;
961        uint8_t sc_id =0;
962	int i, error, retval;
963	//int isc_retval;
964
965	retval = 0;
966	ctl_pause_rtr = 0;
967        rcv_sync_msg = 0;
968
969	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
970			       M_WAITOK | M_ZERO);
971	softc = control_softc;
972
973	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
974			      "cam/ctl");
975
976	softc->dev->si_drv1 = softc;
977
978	/*
979	 * By default, return a "bad LUN" peripheral qualifier for unknown
980	 * LUNs.  The user can override this default using the tunable or
981	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
982	 */
983	softc->inquiry_pq_no_lun = 1;
984	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
985			  &softc->inquiry_pq_no_lun);
986	sysctl_ctx_init(&softc->sysctl_ctx);
987	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
988		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
989		CTLFLAG_RD, 0, "CAM Target Layer");
990
991	if (softc->sysctl_tree == NULL) {
992		printf("%s: unable to allocate sysctl tree\n", __func__);
993		destroy_dev(softc->dev);
994		free(control_softc, M_DEVBUF);
995		control_softc = NULL;
996		return (ENOMEM);
997	}
998
999	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1000		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1001		       "inquiry_pq_no_lun", CTLFLAG_RW,
1002		       &softc->inquiry_pq_no_lun, 0,
1003		       "Report no lun possible for invalid LUNs");
1004
1005	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1006	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
1007	softc->open_count = 0;
1008
1009	/*
1010	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1011	 * the drive.
1012	 */
1013	softc->flags = CTL_FLAG_REAL_SYNC;
1014
1015	/*
1016	 * In Copan's HA scheme, the "master" and "slave" roles are
1017	 * figured out through the slot the controller is in.  Although it
1018	 * is an active/active system, someone has to be in charge.
1019 	 */
1020#ifdef NEEDTOPORT
1021        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1022#endif
1023
1024        if (sc_id == 0) {
1025		softc->flags |= CTL_FLAG_MASTER_SHELF;
1026		persis_offset = 0;
1027	} else
1028		persis_offset = CTL_MAX_INITIATORS;
1029
1030	/*
1031	 * XXX KDM need to figure out where we want to get our target ID
1032	 * and WWID.  Is it different on each port?
1033	 */
1034	softc->target.id = 0;
1035	softc->target.wwid[0] = 0x12345678;
1036	softc->target.wwid[1] = 0x87654321;
1037	STAILQ_INIT(&softc->lun_list);
1038	STAILQ_INIT(&softc->pending_lun_queue);
1039	STAILQ_INIT(&softc->task_queue);
1040	STAILQ_INIT(&softc->incoming_queue);
1041	STAILQ_INIT(&softc->rtr_queue);
1042	STAILQ_INIT(&softc->done_queue);
1043	STAILQ_INIT(&softc->isc_queue);
1044	STAILQ_INIT(&softc->fe_list);
1045	STAILQ_INIT(&softc->be_list);
1046	STAILQ_INIT(&softc->io_pools);
1047
1048	lun = &softc->lun;
1049
1050	/*
1051	 * We don't bother calling these with ctl_lock held here, because,
1052	 * in theory, no one else can try to do anything while we're in our
1053	 * module init routine.
1054	 */
1055	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1056			    &internal_pool)!= 0){
1057		printf("ctl: can't allocate %d entry internal pool, "
1058		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1059		return (ENOMEM);
1060	}
1061
1062	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1063			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1064		printf("ctl: can't allocate %d entry emergency pool, "
1065		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1066		ctl_pool_free(internal_pool);
1067		return (ENOMEM);
1068	}
1069
1070	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1071	                    &other_pool) != 0)
1072	{
1073		printf("ctl: can't allocate %d entry other SC pool, "
1074		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1075		ctl_pool_free(internal_pool);
1076		ctl_pool_free(emergency_pool);
1077		return (ENOMEM);
1078	}
1079
1080	softc->internal_pool = internal_pool;
1081	softc->emergency_pool = emergency_pool;
1082	softc->othersc_pool = other_pool;
1083
1084	/*
1085	 * We used to allocate a processor LUN here.  The new scheme is to
1086	 * just let the user allocate LUNs as he sees fit.
1087	 */
1088#if 0
1089	mtx_lock(&softc->ctl_lock);
1090	ctl_alloc_lun(softc, lun, /*be_lun*/NULL, /*target*/softc->target);
1091	mtx_unlock(&softc->ctl_lock);
1092#endif
1093
1094	if (worker_threads > MAXCPU || worker_threads == 0) {
1095		printf("invalid kern.cam.ctl.worker_threads value; "
1096		    "setting to 1");
1097		worker_threads = 1;
1098	} else if (worker_threads < 0) {
1099		if (mp_ncpus > 2) {
1100			/*
1101			 * Using more than two worker threads actually hurts
1102			 * performance due to lock contention.
1103			 */
1104			worker_threads = 2;
1105		} else {
1106			worker_threads = 1;
1107		}
1108	}
1109
1110	for (i = 0; i < worker_threads; i++) {
1111		error = kproc_create(ctl_work_thread, softc, &softc->work_thread, 0, 0,
1112				"ctl_thrd%d", i);
1113		if (error != 0) {
1114			printf("error creating CTL work thread!\n");
1115			mtx_lock(&softc->ctl_lock);
1116			ctl_free_lun(lun);
1117			mtx_unlock(&softc->ctl_lock);
1118			ctl_pool_free(internal_pool);
1119			ctl_pool_free(emergency_pool);
1120			ctl_pool_free(other_pool);
1121			return (error);
1122		}
1123	}
1124	if (bootverbose)
1125		printf("ctl: CAM Target Layer loaded\n");
1126
1127	/*
1128	 * Initialize the initiator and portname mappings
1129	 */
1130	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1131
1132	/*
1133	 * Initialize the ioctl front end.
1134	 */
1135	fe = &softc->ioctl_info.fe;
1136	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1137	fe->port_type = CTL_PORT_IOCTL;
1138	fe->num_requested_ctl_io = 100;
1139	fe->port_name = softc->ioctl_info.port_name;
1140	fe->port_online = ctl_ioctl_online;
1141	fe->port_offline = ctl_ioctl_offline;
1142	fe->onoff_arg = &softc->ioctl_info;
1143	fe->targ_enable = ctl_ioctl_targ_enable;
1144	fe->targ_disable = ctl_ioctl_targ_disable;
1145	fe->lun_enable = ctl_ioctl_lun_enable;
1146	fe->lun_disable = ctl_ioctl_lun_disable;
1147	fe->targ_lun_arg = &softc->ioctl_info;
1148	fe->fe_datamove = ctl_ioctl_datamove;
1149	fe->fe_done = ctl_ioctl_done;
1150	fe->max_targets = 15;
1151	fe->max_target_id = 15;
1152
1153	if (ctl_frontend_register(&softc->ioctl_info.fe,
1154	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1155		printf("ctl: ioctl front end registration failed, will "
1156		       "continue anyway\n");
1157	}
1158
1159#ifdef CTL_IO_DELAY
1160	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1161		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1162		       sizeof(struct callout), CTL_TIMER_BYTES);
1163		return (EINVAL);
1164	}
1165#endif /* CTL_IO_DELAY */
1166
1167	return (0);
1168}
1169
1170void
1171ctl_shutdown(void)
1172{
1173	struct ctl_softc *softc;
1174	struct ctl_lun *lun, *next_lun;
1175	struct ctl_io_pool *pool;
1176
1177	softc = (struct ctl_softc *)control_softc;
1178
1179	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1180		printf("ctl: ioctl front end deregistration failed\n");
1181
1182	mtx_lock(&softc->ctl_lock);
1183
1184	/*
1185	 * Free up each LUN.
1186	 */
1187	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1188		next_lun = STAILQ_NEXT(lun, links);
1189		ctl_free_lun(lun);
1190	}
1191
1192	mtx_unlock(&softc->ctl_lock);
1193
1194	/*
1195	 * This will rip the rug out from under any FETDs or anyone else
1196	 * that has a pool allocated.  Since we increment our module
1197	 * refcount any time someone outside the main CTL module allocates
1198	 * a pool, we shouldn't have any problems here.  The user won't be
1199	 * able to unload the CTL module until client modules have
1200	 * successfully unloaded.
1201	 */
1202	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1203		ctl_pool_free(pool);
1204
1205#if 0
1206	ctl_shutdown_thread(softc->work_thread);
1207#endif
1208
1209	mtx_destroy(&softc->pool_lock);
1210	mtx_destroy(&softc->ctl_lock);
1211
1212	destroy_dev(softc->dev);
1213
1214	sysctl_ctx_free(&softc->sysctl_ctx);
1215
1216	free(control_softc, M_DEVBUF);
1217	control_softc = NULL;
1218
1219	if (bootverbose)
1220		printf("ctl: CAM Target Layer unloaded\n");
1221}
1222
1223static int
1224ctl_module_event_handler(module_t mod, int what, void *arg)
1225{
1226
1227	switch (what) {
1228	case MOD_LOAD:
1229		return (ctl_init());
1230	case MOD_UNLOAD:
1231		return (EBUSY);
1232	default:
1233		return (EOPNOTSUPP);
1234	}
1235}
1236
1237/*
1238 * XXX KDM should we do some access checks here?  Bump a reference count to
1239 * prevent a CTL module from being unloaded while someone has it open?
1240 */
1241static int
1242ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1243{
1244	return (0);
1245}
1246
1247static int
1248ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1249{
1250	return (0);
1251}
1252
1253int
1254ctl_port_enable(ctl_port_type port_type)
1255{
1256	struct ctl_softc *softc;
1257	struct ctl_frontend *fe;
1258
1259	if (ctl_is_single == 0) {
1260		union ctl_ha_msg msg_info;
1261		int isc_retval;
1262
1263#if 0
1264		printf("%s: HA mode, synchronizing frontend enable\n",
1265		        __func__);
1266#endif
1267		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1268	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1269		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1270			printf("Sync msg send error retval %d\n", isc_retval);
1271		}
1272		if (!rcv_sync_msg) {
1273			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1274			        sizeof(msg_info), 1);
1275		}
1276#if 0
1277        	printf("CTL:Frontend Enable\n");
1278	} else {
1279		printf("%s: single mode, skipping frontend synchronization\n",
1280		        __func__);
1281#endif
1282	}
1283
1284	softc = control_softc;
1285
1286	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1287		if (port_type & fe->port_type)
1288		{
1289#if 0
1290			printf("port %d\n", fe->targ_port);
1291#endif
1292			ctl_frontend_online(fe);
1293		}
1294	}
1295
1296	return (0);
1297}
1298
1299int
1300ctl_port_disable(ctl_port_type port_type)
1301{
1302	struct ctl_softc *softc;
1303	struct ctl_frontend *fe;
1304
1305	softc = control_softc;
1306
1307	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1308		if (port_type & fe->port_type)
1309			ctl_frontend_offline(fe);
1310	}
1311
1312	return (0);
1313}
1314
1315/*
1316 * Returns 0 for success, 1 for failure.
1317 * Currently the only failure mode is if there aren't enough entries
1318 * allocated.  So, in case of a failure, look at num_entries_dropped,
1319 * reallocate and try again.
1320 */
1321int
1322ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1323	      int *num_entries_filled, int *num_entries_dropped,
1324	      ctl_port_type port_type, int no_virtual)
1325{
1326	struct ctl_softc *softc;
1327	struct ctl_frontend *fe;
1328	int entries_dropped, entries_filled;
1329	int retval;
1330	int i;
1331
1332	softc = control_softc;
1333
1334	retval = 0;
1335	entries_filled = 0;
1336	entries_dropped = 0;
1337
1338	i = 0;
1339	mtx_lock(&softc->ctl_lock);
1340	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1341		struct ctl_port_entry *entry;
1342
1343		if ((fe->port_type & port_type) == 0)
1344			continue;
1345
1346		if ((no_virtual != 0)
1347		 && (fe->virtual_port != 0))
1348			continue;
1349
1350		if (entries_filled >= num_entries_alloced) {
1351			entries_dropped++;
1352			continue;
1353		}
1354		entry = &entries[i];
1355
1356		entry->port_type = fe->port_type;
1357		strlcpy(entry->port_name, fe->port_name,
1358			sizeof(entry->port_name));
1359		entry->physical_port = fe->physical_port;
1360		entry->virtual_port = fe->virtual_port;
1361		entry->wwnn = fe->wwnn;
1362		entry->wwpn = fe->wwpn;
1363
1364		i++;
1365		entries_filled++;
1366	}
1367
1368	mtx_unlock(&softc->ctl_lock);
1369
1370	if (entries_dropped > 0)
1371		retval = 1;
1372
1373	*num_entries_dropped = entries_dropped;
1374	*num_entries_filled = entries_filled;
1375
1376	return (retval);
1377}
1378
1379static void
1380ctl_ioctl_online(void *arg)
1381{
1382	struct ctl_ioctl_info *ioctl_info;
1383
1384	ioctl_info = (struct ctl_ioctl_info *)arg;
1385
1386	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1387}
1388
1389static void
1390ctl_ioctl_offline(void *arg)
1391{
1392	struct ctl_ioctl_info *ioctl_info;
1393
1394	ioctl_info = (struct ctl_ioctl_info *)arg;
1395
1396	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1397}
1398
1399/*
1400 * Remove an initiator by port number and initiator ID.
1401 * Returns 0 for success, 1 for failure.
1402 */
1403int
1404ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1405{
1406	struct ctl_softc *softc;
1407
1408	softc = control_softc;
1409
1410	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1411
1412	if ((targ_port < 0)
1413	 || (targ_port > CTL_MAX_PORTS)) {
1414		printf("%s: invalid port number %d\n", __func__, targ_port);
1415		return (1);
1416	}
1417	if (iid > CTL_MAX_INIT_PER_PORT) {
1418		printf("%s: initiator ID %u > maximun %u!\n",
1419		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1420		return (1);
1421	}
1422
1423	mtx_lock(&softc->ctl_lock);
1424
1425	softc->wwpn_iid[targ_port][iid].in_use = 0;
1426
1427	mtx_unlock(&softc->ctl_lock);
1428
1429	return (0);
1430}
1431
1432/*
1433 * Add an initiator to the initiator map.
1434 * Returns 0 for success, 1 for failure.
1435 */
1436int
1437ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1438{
1439	struct ctl_softc *softc;
1440	int retval;
1441
1442	softc = control_softc;
1443
1444	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1445
1446	retval = 0;
1447
1448	if ((targ_port < 0)
1449	 || (targ_port > CTL_MAX_PORTS)) {
1450		printf("%s: invalid port number %d\n", __func__, targ_port);
1451		return (1);
1452	}
1453	if (iid > CTL_MAX_INIT_PER_PORT) {
1454		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1455		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1456		return (1);
1457	}
1458
1459	mtx_lock(&softc->ctl_lock);
1460
1461	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1462		/*
1463		 * We don't treat this as an error.
1464		 */
1465		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1466			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1467			       __func__, targ_port, iid, (uintmax_t)wwpn);
1468			goto bailout;
1469		}
1470
1471		/*
1472		 * This is an error, but what do we do about it?  The
1473		 * driver is telling us we have a new WWPN for this
1474		 * initiator ID, so we pretty much need to use it.
1475		 */
1476		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1477		       "still at that address\n", __func__, targ_port, iid,
1478		       (uintmax_t)wwpn,
1479		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1480
1481		/*
1482		 * XXX KDM clear have_ca and ua_pending on each LUN for
1483		 * this initiator.
1484		 */
1485	}
1486	softc->wwpn_iid[targ_port][iid].in_use = 1;
1487	softc->wwpn_iid[targ_port][iid].iid = iid;
1488	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1489	softc->wwpn_iid[targ_port][iid].port = targ_port;
1490
1491bailout:
1492
1493	mtx_unlock(&softc->ctl_lock);
1494
1495	return (retval);
1496}
1497
1498/*
1499 * XXX KDM should we pretend to do something in the target/lun
1500 * enable/disable functions?
1501 */
1502static int
1503ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1504{
1505	return (0);
1506}
1507
1508static int
1509ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1510{
1511	return (0);
1512}
1513
1514static int
1515ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1516{
1517	return (0);
1518}
1519
1520static int
1521ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1522{
1523	return (0);
1524}
1525
1526/*
1527 * Data movement routine for the CTL ioctl frontend port.
1528 */
1529static int
1530ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1531{
1532	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1533	struct ctl_sg_entry ext_entry, kern_entry;
1534	int ext_sglen, ext_sg_entries, kern_sg_entries;
1535	int ext_sg_start, ext_offset;
1536	int len_to_copy, len_copied;
1537	int kern_watermark, ext_watermark;
1538	int ext_sglist_malloced;
1539	int i, j;
1540
1541	ext_sglist_malloced = 0;
1542	ext_sg_start = 0;
1543	ext_offset = 0;
1544
1545	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1546
1547	/*
1548	 * If this flag is set, fake the data transfer.
1549	 */
1550	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1551		ctsio->ext_data_filled = ctsio->ext_data_len;
1552		goto bailout;
1553	}
1554
1555	/*
1556	 * To simplify things here, if we have a single buffer, stick it in
1557	 * a S/G entry and just make it a single entry S/G list.
1558	 */
1559	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1560		int len_seen;
1561
1562		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1563
1564		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1565							   M_WAITOK);
1566		ext_sglist_malloced = 1;
1567		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1568				   ext_sglen) != 0) {
1569			ctl_set_internal_failure(ctsio,
1570						 /*sks_valid*/ 0,
1571						 /*retry_count*/ 0);
1572			goto bailout;
1573		}
1574		ext_sg_entries = ctsio->ext_sg_entries;
1575		len_seen = 0;
1576		for (i = 0; i < ext_sg_entries; i++) {
1577			if ((len_seen + ext_sglist[i].len) >=
1578			     ctsio->ext_data_filled) {
1579				ext_sg_start = i;
1580				ext_offset = ctsio->ext_data_filled - len_seen;
1581				break;
1582			}
1583			len_seen += ext_sglist[i].len;
1584		}
1585	} else {
1586		ext_sglist = &ext_entry;
1587		ext_sglist->addr = ctsio->ext_data_ptr;
1588		ext_sglist->len = ctsio->ext_data_len;
1589		ext_sg_entries = 1;
1590		ext_sg_start = 0;
1591		ext_offset = ctsio->ext_data_filled;
1592	}
1593
1594	if (ctsio->kern_sg_entries > 0) {
1595		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1596		kern_sg_entries = ctsio->kern_sg_entries;
1597	} else {
1598		kern_sglist = &kern_entry;
1599		kern_sglist->addr = ctsio->kern_data_ptr;
1600		kern_sglist->len = ctsio->kern_data_len;
1601		kern_sg_entries = 1;
1602	}
1603
1604
1605	kern_watermark = 0;
1606	ext_watermark = ext_offset;
1607	len_copied = 0;
1608	for (i = ext_sg_start, j = 0;
1609	     i < ext_sg_entries && j < kern_sg_entries;) {
1610		uint8_t *ext_ptr, *kern_ptr;
1611
1612		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1613				      kern_sglist[j].len - kern_watermark);
1614
1615		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1616		ext_ptr = ext_ptr + ext_watermark;
1617		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1618			/*
1619			 * XXX KDM fix this!
1620			 */
1621			panic("need to implement bus address support");
1622#if 0
1623			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1624#endif
1625		} else
1626			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1627		kern_ptr = kern_ptr + kern_watermark;
1628
1629		kern_watermark += len_to_copy;
1630		ext_watermark += len_to_copy;
1631
1632		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1633		     CTL_FLAG_DATA_IN) {
1634			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1635					 "bytes to user\n", len_to_copy));
1636			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1637					 "to %p\n", kern_ptr, ext_ptr));
1638			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1639				ctl_set_internal_failure(ctsio,
1640							 /*sks_valid*/ 0,
1641							 /*retry_count*/ 0);
1642				goto bailout;
1643			}
1644		} else {
1645			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1646					 "bytes from user\n", len_to_copy));
1647			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1648					 "to %p\n", ext_ptr, kern_ptr));
1649			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1650				ctl_set_internal_failure(ctsio,
1651							 /*sks_valid*/ 0,
1652							 /*retry_count*/0);
1653				goto bailout;
1654			}
1655		}
1656
1657		len_copied += len_to_copy;
1658
1659		if (ext_sglist[i].len == ext_watermark) {
1660			i++;
1661			ext_watermark = 0;
1662		}
1663
1664		if (kern_sglist[j].len == kern_watermark) {
1665			j++;
1666			kern_watermark = 0;
1667		}
1668	}
1669
1670	ctsio->ext_data_filled += len_copied;
1671
1672	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1673			 "kern_sg_entries: %d\n", ext_sg_entries,
1674			 kern_sg_entries));
1675	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1676			 "kern_data_len = %d\n", ctsio->ext_data_len,
1677			 ctsio->kern_data_len));
1678
1679
1680	/* XXX KDM set residual?? */
1681bailout:
1682
1683	if (ext_sglist_malloced != 0)
1684		free(ext_sglist, M_CTL);
1685
1686	return (CTL_RETVAL_COMPLETE);
1687}
1688
1689/*
1690 * Serialize a command that went down the "wrong" side, and so was sent to
1691 * this controller for execution.  The logic is a little different than the
1692 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1693 * sent back to the other side, but in the success case, we execute the
1694 * command on this side (XFER mode) or tell the other side to execute it
1695 * (SER_ONLY mode).
1696 */
1697static int
1698ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1699{
1700	struct ctl_softc *ctl_softc;
1701	union ctl_ha_msg msg_info;
1702	struct ctl_lun *lun;
1703	int retval = 0;
1704	uint32_t targ_lun;
1705
1706	ctl_softc = control_softc;
1707	if (have_lock == 0)
1708		mtx_lock(&ctl_softc->ctl_lock);
1709
1710	targ_lun = ctsio->io_hdr.nexus.targ_lun;
1711	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
1712		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
1713	lun = ctl_softc->ctl_luns[targ_lun];
1714	if (lun==NULL)
1715	{
1716		/*
1717		 * Why isn't LUN defined? The other side wouldn't
1718		 * send a cmd if the LUN is undefined.
1719		 */
1720		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1721
1722		/* "Logical unit not supported" */
1723		ctl_set_sense_data(&msg_info.scsi.sense_data,
1724				   lun,
1725				   /*sense_format*/SSD_TYPE_NONE,
1726				   /*current_error*/ 1,
1727				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1728				   /*asc*/ 0x25,
1729				   /*ascq*/ 0x00,
1730				   SSD_ELEM_NONE);
1731
1732		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1733		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1734		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1735		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1736		msg_info.hdr.serializing_sc = NULL;
1737		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1738	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1739				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1740		}
1741		if (have_lock == 0)
1742			mtx_unlock(&ctl_softc->ctl_lock);
1743		return(1);
1744
1745	}
1746
1747    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1748
1749	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1750		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1751		 ooa_links))) {
1752	case CTL_ACTION_BLOCK:
1753		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1754		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1755				  blocked_links);
1756		break;
1757	case CTL_ACTION_PASS:
1758	case CTL_ACTION_SKIP:
1759		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1760			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1761			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1762					   &ctsio->io_hdr, links);
1763		} else {
1764
1765			/* send msg back to other side */
1766			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1767			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1768			msg_info.hdr.msg_type = CTL_MSG_R2R;
1769#if 0
1770			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1771#endif
1772		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1773			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1774			}
1775		}
1776		break;
1777	case CTL_ACTION_OVERLAP:
1778		/* OVERLAPPED COMMANDS ATTEMPTED */
1779		ctl_set_sense_data(&msg_info.scsi.sense_data,
1780				   lun,
1781				   /*sense_format*/SSD_TYPE_NONE,
1782				   /*current_error*/ 1,
1783				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1784				   /*asc*/ 0x4E,
1785				   /*ascq*/ 0x00,
1786				   SSD_ELEM_NONE);
1787
1788		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1789		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1790		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1791		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1792		msg_info.hdr.serializing_sc = NULL;
1793		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1794#if 0
1795		printf("BAD JUJU:Major Bummer Overlap\n");
1796#endif
1797		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1798		retval = 1;
1799		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1800		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1801		}
1802		break;
1803	case CTL_ACTION_OVERLAP_TAG:
1804		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1805		ctl_set_sense_data(&msg_info.scsi.sense_data,
1806				   lun,
1807				   /*sense_format*/SSD_TYPE_NONE,
1808				   /*current_error*/ 1,
1809				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1810				   /*asc*/ 0x4D,
1811				   /*ascq*/ ctsio->tag_num & 0xff,
1812				   SSD_ELEM_NONE);
1813
1814		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1815		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1816		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1817		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1818		msg_info.hdr.serializing_sc = NULL;
1819		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1820#if 0
1821		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1822#endif
1823		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1824		retval = 1;
1825		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1826		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1827		}
1828		break;
1829	case CTL_ACTION_ERROR:
1830	default:
1831		/* "Internal target failure" */
1832		ctl_set_sense_data(&msg_info.scsi.sense_data,
1833				   lun,
1834				   /*sense_format*/SSD_TYPE_NONE,
1835				   /*current_error*/ 1,
1836				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1837				   /*asc*/ 0x44,
1838				   /*ascq*/ 0x00,
1839				   SSD_ELEM_NONE);
1840
1841		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1842		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1843		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1844		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1845		msg_info.hdr.serializing_sc = NULL;
1846		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1847#if 0
1848		printf("BAD JUJU:Major Bummer HW Error\n");
1849#endif
1850		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1851		retval = 1;
1852		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1853		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1854		}
1855		break;
1856	}
1857	if (have_lock == 0)
1858		mtx_unlock(&ctl_softc->ctl_lock);
1859	return (retval);
1860}
1861
1862static int
1863ctl_ioctl_submit_wait(union ctl_io *io)
1864{
1865	struct ctl_fe_ioctl_params params;
1866	ctl_fe_ioctl_state last_state;
1867	int done, retval;
1868
1869	retval = 0;
1870
1871	bzero(&params, sizeof(params));
1872
1873	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1874	cv_init(&params.sem, "ctlioccv");
1875	params.state = CTL_IOCTL_INPROG;
1876	last_state = params.state;
1877
1878	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1879
1880	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1881
1882	/* This shouldn't happen */
1883	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1884		return (retval);
1885
1886	done = 0;
1887
1888	do {
1889		mtx_lock(&params.ioctl_mtx);
1890		/*
1891		 * Check the state here, and don't sleep if the state has
1892		 * already changed (i.e. wakeup has already occured, but we
1893		 * weren't waiting yet).
1894		 */
1895		if (params.state == last_state) {
1896			/* XXX KDM cv_wait_sig instead? */
1897			cv_wait(&params.sem, &params.ioctl_mtx);
1898		}
1899		last_state = params.state;
1900
1901		switch (params.state) {
1902		case CTL_IOCTL_INPROG:
1903			/* Why did we wake up? */
1904			/* XXX KDM error here? */
1905			mtx_unlock(&params.ioctl_mtx);
1906			break;
1907		case CTL_IOCTL_DATAMOVE:
1908			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1909
1910			/*
1911			 * change last_state back to INPROG to avoid
1912			 * deadlock on subsequent data moves.
1913			 */
1914			params.state = last_state = CTL_IOCTL_INPROG;
1915
1916			mtx_unlock(&params.ioctl_mtx);
1917			ctl_ioctl_do_datamove(&io->scsiio);
1918			/*
1919			 * Note that in some cases, most notably writes,
1920			 * this will queue the I/O and call us back later.
1921			 * In other cases, generally reads, this routine
1922			 * will immediately call back and wake us up,
1923			 * probably using our own context.
1924			 */
1925			io->scsiio.be_move_done(io);
1926			break;
1927		case CTL_IOCTL_DONE:
1928			mtx_unlock(&params.ioctl_mtx);
1929			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1930			done = 1;
1931			break;
1932		default:
1933			mtx_unlock(&params.ioctl_mtx);
1934			/* XXX KDM error here? */
1935			break;
1936		}
1937	} while (done == 0);
1938
1939	mtx_destroy(&params.ioctl_mtx);
1940	cv_destroy(&params.sem);
1941
1942	return (CTL_RETVAL_COMPLETE);
1943}
1944
1945static void
1946ctl_ioctl_datamove(union ctl_io *io)
1947{
1948	struct ctl_fe_ioctl_params *params;
1949
1950	params = (struct ctl_fe_ioctl_params *)
1951		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1952
1953	mtx_lock(&params->ioctl_mtx);
1954	params->state = CTL_IOCTL_DATAMOVE;
1955	cv_broadcast(&params->sem);
1956	mtx_unlock(&params->ioctl_mtx);
1957}
1958
1959static void
1960ctl_ioctl_done(union ctl_io *io)
1961{
1962	struct ctl_fe_ioctl_params *params;
1963
1964	params = (struct ctl_fe_ioctl_params *)
1965		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1966
1967	mtx_lock(&params->ioctl_mtx);
1968	params->state = CTL_IOCTL_DONE;
1969	cv_broadcast(&params->sem);
1970	mtx_unlock(&params->ioctl_mtx);
1971}
1972
1973static void
1974ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1975{
1976	struct ctl_fe_ioctl_startstop_info *sd_info;
1977
1978	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1979
1980	sd_info->hs_info.status = metatask->status;
1981	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1982	sd_info->hs_info.luns_complete =
1983		metatask->taskinfo.startstop.luns_complete;
1984	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1985
1986	cv_broadcast(&sd_info->sem);
1987}
1988
1989static void
1990ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1991{
1992	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1993
1994	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1995
1996	mtx_lock(fe_bbr_info->lock);
1997	fe_bbr_info->bbr_info->status = metatask->status;
1998	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1999	fe_bbr_info->wakeup_done = 1;
2000	mtx_unlock(fe_bbr_info->lock);
2001
2002	cv_broadcast(&fe_bbr_info->sem);
2003}
2004
2005/*
2006 * Returns 0 for success, errno for failure.
2007 */
2008static int
2009ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2010		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2011{
2012	union ctl_io *io;
2013	int retval;
2014
2015	retval = 0;
2016
2017	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
2018
2019	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2020	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2021	     ooa_links)) {
2022		struct ctl_ooa_entry *entry;
2023
2024		/*
2025		 * If we've got more than we can fit, just count the
2026		 * remaining entries.
2027		 */
2028		if (*cur_fill_num >= ooa_hdr->alloc_num)
2029			continue;
2030
2031		entry = &kern_entries[*cur_fill_num];
2032
2033		entry->tag_num = io->scsiio.tag_num;
2034		entry->lun_num = lun->lun;
2035#ifdef CTL_TIME_IO
2036		entry->start_bt = io->io_hdr.start_bt;
2037#endif
2038		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2039		entry->cdb_len = io->scsiio.cdb_len;
2040		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2041			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2042
2043		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2044			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2045
2046		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2047			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2048
2049		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2050			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2051
2052		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2053			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2054	}
2055
2056	return (retval);
2057}
2058
2059static void *
2060ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2061		 size_t error_str_len)
2062{
2063	void *kptr;
2064
2065	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2066
2067	if (copyin(user_addr, kptr, len) != 0) {
2068		snprintf(error_str, error_str_len, "Error copying %d bytes "
2069			 "from user address %p to kernel address %p", len,
2070			 user_addr, kptr);
2071		free(kptr, M_CTL);
2072		return (NULL);
2073	}
2074
2075	return (kptr);
2076}
2077
2078static void
2079ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2080{
2081	int i;
2082
2083	if (be_args == NULL)
2084		return;
2085
2086	for (i = 0; i < num_be_args; i++) {
2087		free(be_args[i].kname, M_CTL);
2088		free(be_args[i].kvalue, M_CTL);
2089	}
2090
2091	free(be_args, M_CTL);
2092}
2093
2094static struct ctl_be_arg *
2095ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2096		char *error_str, size_t error_str_len)
2097{
2098	struct ctl_be_arg *args;
2099	int i;
2100
2101	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2102				error_str, error_str_len);
2103
2104	if (args == NULL)
2105		goto bailout;
2106
2107	for (i = 0; i < num_be_args; i++) {
2108		args[i].kname = NULL;
2109		args[i].kvalue = NULL;
2110	}
2111
2112	for (i = 0; i < num_be_args; i++) {
2113		uint8_t *tmpptr;
2114
2115		args[i].kname = ctl_copyin_alloc(args[i].name,
2116			args[i].namelen, error_str, error_str_len);
2117		if (args[i].kname == NULL)
2118			goto bailout;
2119
2120		if (args[i].kname[args[i].namelen - 1] != '\0') {
2121			snprintf(error_str, error_str_len, "Argument %d "
2122				 "name is not NUL-terminated", i);
2123			goto bailout;
2124		}
2125
2126		args[i].kvalue = NULL;
2127
2128		tmpptr = ctl_copyin_alloc(args[i].value,
2129			args[i].vallen, error_str, error_str_len);
2130		if (tmpptr == NULL)
2131			goto bailout;
2132
2133		args[i].kvalue = tmpptr;
2134
2135		if ((args[i].flags & CTL_BEARG_ASCII)
2136		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2137			snprintf(error_str, error_str_len, "Argument %d "
2138				 "value is not NUL-terminated", i);
2139			goto bailout;
2140		}
2141	}
2142
2143	return (args);
2144bailout:
2145
2146	ctl_free_args(num_be_args, args);
2147
2148	return (NULL);
2149}
2150
2151/*
2152 * Escape characters that are illegal or not recommended in XML.
2153 */
2154int
2155ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2156{
2157	int retval;
2158
2159	retval = 0;
2160
2161	for (; *str; str++) {
2162		switch (*str) {
2163		case '&':
2164			retval = sbuf_printf(sb, "&amp;");
2165			break;
2166		case '>':
2167			retval = sbuf_printf(sb, "&gt;");
2168			break;
2169		case '<':
2170			retval = sbuf_printf(sb, "&lt;");
2171			break;
2172		default:
2173			retval = sbuf_putc(sb, *str);
2174			break;
2175		}
2176
2177		if (retval != 0)
2178			break;
2179
2180	}
2181
2182	return (retval);
2183}
2184
2185static int
2186ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2187	  struct thread *td)
2188{
2189	struct ctl_softc *softc;
2190	int retval;
2191
2192	softc = control_softc;
2193
2194	retval = 0;
2195
2196	switch (cmd) {
2197	case CTL_IO: {
2198		union ctl_io *io;
2199		void *pool_tmp;
2200
2201		/*
2202		 * If we haven't been "enabled", don't allow any SCSI I/O
2203		 * to this FETD.
2204		 */
2205		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2206			retval = -EPERM;
2207			break;
2208		}
2209
2210		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2211		if (io == NULL) {
2212			printf("ctl_ioctl: can't allocate ctl_io!\n");
2213			retval = -ENOSPC;
2214			break;
2215		}
2216
2217		/*
2218		 * Need to save the pool reference so it doesn't get
2219		 * spammed by the user's ctl_io.
2220		 */
2221		pool_tmp = io->io_hdr.pool;
2222
2223		memcpy(io, (void *)addr, sizeof(*io));
2224
2225		io->io_hdr.pool = pool_tmp;
2226		/*
2227		 * No status yet, so make sure the status is set properly.
2228		 */
2229		io->io_hdr.status = CTL_STATUS_NONE;
2230
2231		/*
2232		 * The user sets the initiator ID, target and LUN IDs.
2233		 */
2234		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2235		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2236		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2237		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2238			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2239
2240		retval = ctl_ioctl_submit_wait(io);
2241
2242		if (retval != 0) {
2243			ctl_free_io(io);
2244			break;
2245		}
2246
2247		memcpy((void *)addr, io, sizeof(*io));
2248
2249		/* return this to our pool */
2250		ctl_free_io(io);
2251
2252		break;
2253	}
2254	case CTL_ENABLE_PORT:
2255	case CTL_DISABLE_PORT:
2256	case CTL_SET_PORT_WWNS: {
2257		struct ctl_frontend *fe;
2258		struct ctl_port_entry *entry;
2259
2260		entry = (struct ctl_port_entry *)addr;
2261
2262		mtx_lock(&softc->ctl_lock);
2263		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2264			int action, done;
2265
2266			action = 0;
2267			done = 0;
2268
2269			if ((entry->port_type == CTL_PORT_NONE)
2270			 && (entry->targ_port == fe->targ_port)) {
2271				/*
2272				 * If the user only wants to enable or
2273				 * disable or set WWNs on a specific port,
2274				 * do the operation and we're done.
2275				 */
2276				action = 1;
2277				done = 1;
2278			} else if (entry->port_type & fe->port_type) {
2279				/*
2280				 * Compare the user's type mask with the
2281				 * particular frontend type to see if we
2282				 * have a match.
2283				 */
2284				action = 1;
2285				done = 0;
2286
2287				/*
2288				 * Make sure the user isn't trying to set
2289				 * WWNs on multiple ports at the same time.
2290				 */
2291				if (cmd == CTL_SET_PORT_WWNS) {
2292					printf("%s: Can't set WWNs on "
2293					       "multiple ports\n", __func__);
2294					retval = EINVAL;
2295					break;
2296				}
2297			}
2298			if (action != 0) {
2299				/*
2300				 * XXX KDM we have to drop the lock here,
2301				 * because the online/offline operations
2302				 * can potentially block.  We need to
2303				 * reference count the frontends so they
2304				 * can't go away,
2305				 */
2306				mtx_unlock(&softc->ctl_lock);
2307
2308				if (cmd == CTL_ENABLE_PORT) {
2309					struct ctl_lun *lun;
2310
2311					STAILQ_FOREACH(lun, &softc->lun_list,
2312						       links) {
2313						fe->lun_enable(fe->targ_lun_arg,
2314						    lun->target,
2315						    lun->lun);
2316					}
2317
2318					ctl_frontend_online(fe);
2319				} else if (cmd == CTL_DISABLE_PORT) {
2320					struct ctl_lun *lun;
2321
2322					ctl_frontend_offline(fe);
2323
2324					STAILQ_FOREACH(lun, &softc->lun_list,
2325						       links) {
2326						fe->lun_disable(
2327						    fe->targ_lun_arg,
2328						    lun->target,
2329						    lun->lun);
2330					}
2331				}
2332
2333				mtx_lock(&softc->ctl_lock);
2334
2335				if (cmd == CTL_SET_PORT_WWNS)
2336					ctl_frontend_set_wwns(fe,
2337					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2338					    1 : 0, entry->wwnn,
2339					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2340					    1 : 0, entry->wwpn);
2341			}
2342			if (done != 0)
2343				break;
2344		}
2345		mtx_unlock(&softc->ctl_lock);
2346		break;
2347	}
2348	case CTL_GET_PORT_LIST: {
2349		struct ctl_frontend *fe;
2350		struct ctl_port_list *list;
2351		int i;
2352
2353		list = (struct ctl_port_list *)addr;
2354
2355		if (list->alloc_len != (list->alloc_num *
2356		    sizeof(struct ctl_port_entry))) {
2357			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2358			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2359			       "%zu\n", __func__, list->alloc_len,
2360			       list->alloc_num, sizeof(struct ctl_port_entry));
2361			retval = EINVAL;
2362			break;
2363		}
2364		list->fill_len = 0;
2365		list->fill_num = 0;
2366		list->dropped_num = 0;
2367		i = 0;
2368		mtx_lock(&softc->ctl_lock);
2369		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2370			struct ctl_port_entry entry, *list_entry;
2371
2372			if (list->fill_num >= list->alloc_num) {
2373				list->dropped_num++;
2374				continue;
2375			}
2376
2377			entry.port_type = fe->port_type;
2378			strlcpy(entry.port_name, fe->port_name,
2379				sizeof(entry.port_name));
2380			entry.targ_port = fe->targ_port;
2381			entry.physical_port = fe->physical_port;
2382			entry.virtual_port = fe->virtual_port;
2383			entry.wwnn = fe->wwnn;
2384			entry.wwpn = fe->wwpn;
2385			if (fe->status & CTL_PORT_STATUS_ONLINE)
2386				entry.online = 1;
2387			else
2388				entry.online = 0;
2389
2390			list_entry = &list->entries[i];
2391
2392			retval = copyout(&entry, list_entry, sizeof(entry));
2393			if (retval != 0) {
2394				printf("%s: CTL_GET_PORT_LIST: copyout "
2395				       "returned %d\n", __func__, retval);
2396				break;
2397			}
2398			i++;
2399			list->fill_num++;
2400			list->fill_len += sizeof(entry);
2401		}
2402		mtx_unlock(&softc->ctl_lock);
2403
2404		/*
2405		 * If this is non-zero, we had a copyout fault, so there's
2406		 * probably no point in attempting to set the status inside
2407		 * the structure.
2408		 */
2409		if (retval != 0)
2410			break;
2411
2412		if (list->dropped_num > 0)
2413			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2414		else
2415			list->status = CTL_PORT_LIST_OK;
2416		break;
2417	}
2418	case CTL_DUMP_OOA: {
2419		struct ctl_lun *lun;
2420		union ctl_io *io;
2421		char printbuf[128];
2422		struct sbuf sb;
2423
2424		mtx_lock(&softc->ctl_lock);
2425		printf("Dumping OOA queues:\n");
2426		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2427			for (io = (union ctl_io *)TAILQ_FIRST(
2428			     &lun->ooa_queue); io != NULL;
2429			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2430			     ooa_links)) {
2431				sbuf_new(&sb, printbuf, sizeof(printbuf),
2432					 SBUF_FIXEDLEN);
2433				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2434					    (intmax_t)lun->lun,
2435					    io->scsiio.tag_num,
2436					    (io->io_hdr.flags &
2437					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2438					    (io->io_hdr.flags &
2439					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2440					    (io->io_hdr.flags &
2441					    CTL_FLAG_ABORT) ? " ABORT" : "",
2442			                    (io->io_hdr.flags &
2443		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2444				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2445				sbuf_finish(&sb);
2446				printf("%s\n", sbuf_data(&sb));
2447			}
2448		}
2449		printf("OOA queues dump done\n");
2450		mtx_unlock(&softc->ctl_lock);
2451		break;
2452	}
2453	case CTL_GET_OOA: {
2454		struct ctl_lun *lun;
2455		struct ctl_ooa *ooa_hdr;
2456		struct ctl_ooa_entry *entries;
2457		uint32_t cur_fill_num;
2458
2459		ooa_hdr = (struct ctl_ooa *)addr;
2460
2461		if ((ooa_hdr->alloc_len == 0)
2462		 || (ooa_hdr->alloc_num == 0)) {
2463			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2464			       "must be non-zero\n", __func__,
2465			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2466			retval = EINVAL;
2467			break;
2468		}
2469
2470		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2471		    sizeof(struct ctl_ooa_entry))) {
2472			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2473			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2474			       __func__, ooa_hdr->alloc_len,
2475			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2476			retval = EINVAL;
2477			break;
2478		}
2479
2480		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2481		if (entries == NULL) {
2482			printf("%s: could not allocate %d bytes for OOA "
2483			       "dump\n", __func__, ooa_hdr->alloc_len);
2484			retval = ENOMEM;
2485			break;
2486		}
2487
2488		mtx_lock(&softc->ctl_lock);
2489		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2490		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2491		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2492			mtx_unlock(&softc->ctl_lock);
2493			free(entries, M_CTL);
2494			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2495			       __func__, (uintmax_t)ooa_hdr->lun_num);
2496			retval = EINVAL;
2497			break;
2498		}
2499
2500		cur_fill_num = 0;
2501
2502		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2503			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2504				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2505					ooa_hdr, entries);
2506				if (retval != 0)
2507					break;
2508			}
2509			if (retval != 0) {
2510				mtx_unlock(&softc->ctl_lock);
2511				free(entries, M_CTL);
2512				break;
2513			}
2514		} else {
2515			lun = softc->ctl_luns[ooa_hdr->lun_num];
2516
2517			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2518						    entries);
2519		}
2520		mtx_unlock(&softc->ctl_lock);
2521
2522		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2523		ooa_hdr->fill_len = ooa_hdr->fill_num *
2524			sizeof(struct ctl_ooa_entry);
2525		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2526		if (retval != 0) {
2527			printf("%s: error copying out %d bytes for OOA dump\n",
2528			       __func__, ooa_hdr->fill_len);
2529		}
2530
2531		getbintime(&ooa_hdr->cur_bt);
2532
2533		if (cur_fill_num > ooa_hdr->alloc_num) {
2534			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2535			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2536		} else {
2537			ooa_hdr->dropped_num = 0;
2538			ooa_hdr->status = CTL_OOA_OK;
2539		}
2540
2541		free(entries, M_CTL);
2542		break;
2543	}
2544	case CTL_CHECK_OOA: {
2545		union ctl_io *io;
2546		struct ctl_lun *lun;
2547		struct ctl_ooa_info *ooa_info;
2548
2549
2550		ooa_info = (struct ctl_ooa_info *)addr;
2551
2552		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2553			ooa_info->status = CTL_OOA_INVALID_LUN;
2554			break;
2555		}
2556		mtx_lock(&softc->ctl_lock);
2557		lun = softc->ctl_luns[ooa_info->lun_id];
2558		if (lun == NULL) {
2559			mtx_unlock(&softc->ctl_lock);
2560			ooa_info->status = CTL_OOA_INVALID_LUN;
2561			break;
2562		}
2563
2564		ooa_info->num_entries = 0;
2565		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2566		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2567		     &io->io_hdr, ooa_links)) {
2568			ooa_info->num_entries++;
2569		}
2570
2571		mtx_unlock(&softc->ctl_lock);
2572		ooa_info->status = CTL_OOA_SUCCESS;
2573
2574		break;
2575	}
2576	case CTL_HARD_START:
2577	case CTL_HARD_STOP: {
2578		struct ctl_fe_ioctl_startstop_info ss_info;
2579		struct cfi_metatask *metatask;
2580		struct mtx hs_mtx;
2581
2582		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2583
2584		cv_init(&ss_info.sem, "hard start/stop cv" );
2585
2586		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2587		if (metatask == NULL) {
2588			retval = ENOMEM;
2589			mtx_destroy(&hs_mtx);
2590			break;
2591		}
2592
2593		if (cmd == CTL_HARD_START)
2594			metatask->tasktype = CFI_TASK_STARTUP;
2595		else
2596			metatask->tasktype = CFI_TASK_SHUTDOWN;
2597
2598		metatask->callback = ctl_ioctl_hard_startstop_callback;
2599		metatask->callback_arg = &ss_info;
2600
2601		cfi_action(metatask);
2602
2603		/* Wait for the callback */
2604		mtx_lock(&hs_mtx);
2605		cv_wait_sig(&ss_info.sem, &hs_mtx);
2606		mtx_unlock(&hs_mtx);
2607
2608		/*
2609		 * All information has been copied from the metatask by the
2610		 * time cv_broadcast() is called, so we free the metatask here.
2611		 */
2612		cfi_free_metatask(metatask);
2613
2614		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2615
2616		mtx_destroy(&hs_mtx);
2617		break;
2618	}
2619	case CTL_BBRREAD: {
2620		struct ctl_bbrread_info *bbr_info;
2621		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2622		struct mtx bbr_mtx;
2623		struct cfi_metatask *metatask;
2624
2625		bbr_info = (struct ctl_bbrread_info *)addr;
2626
2627		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2628
2629		bzero(&bbr_mtx, sizeof(bbr_mtx));
2630		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2631
2632		fe_bbr_info.bbr_info = bbr_info;
2633		fe_bbr_info.lock = &bbr_mtx;
2634
2635		cv_init(&fe_bbr_info.sem, "BBR read cv");
2636		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2637
2638		if (metatask == NULL) {
2639			mtx_destroy(&bbr_mtx);
2640			cv_destroy(&fe_bbr_info.sem);
2641			retval = ENOMEM;
2642			break;
2643		}
2644		metatask->tasktype = CFI_TASK_BBRREAD;
2645		metatask->callback = ctl_ioctl_bbrread_callback;
2646		metatask->callback_arg = &fe_bbr_info;
2647		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2648		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2649		metatask->taskinfo.bbrread.len = bbr_info->len;
2650
2651		cfi_action(metatask);
2652
2653		mtx_lock(&bbr_mtx);
2654		while (fe_bbr_info.wakeup_done == 0)
2655			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2656		mtx_unlock(&bbr_mtx);
2657
2658		bbr_info->status = metatask->status;
2659		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2660		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2661		memcpy(&bbr_info->sense_data,
2662		       &metatask->taskinfo.bbrread.sense_data,
2663		       ctl_min(sizeof(bbr_info->sense_data),
2664			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2665
2666		cfi_free_metatask(metatask);
2667
2668		mtx_destroy(&bbr_mtx);
2669		cv_destroy(&fe_bbr_info.sem);
2670
2671		break;
2672	}
2673	case CTL_DELAY_IO: {
2674		struct ctl_io_delay_info *delay_info;
2675#ifdef CTL_IO_DELAY
2676		struct ctl_lun *lun;
2677#endif /* CTL_IO_DELAY */
2678
2679		delay_info = (struct ctl_io_delay_info *)addr;
2680
2681#ifdef CTL_IO_DELAY
2682		mtx_lock(&softc->ctl_lock);
2683
2684		if ((delay_info->lun_id > CTL_MAX_LUNS)
2685		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2686			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2687		} else {
2688			lun = softc->ctl_luns[delay_info->lun_id];
2689
2690			delay_info->status = CTL_DELAY_STATUS_OK;
2691
2692			switch (delay_info->delay_type) {
2693			case CTL_DELAY_TYPE_CONT:
2694				break;
2695			case CTL_DELAY_TYPE_ONESHOT:
2696				break;
2697			default:
2698				delay_info->status =
2699					CTL_DELAY_STATUS_INVALID_TYPE;
2700				break;
2701			}
2702
2703			switch (delay_info->delay_loc) {
2704			case CTL_DELAY_LOC_DATAMOVE:
2705				lun->delay_info.datamove_type =
2706					delay_info->delay_type;
2707				lun->delay_info.datamove_delay =
2708					delay_info->delay_secs;
2709				break;
2710			case CTL_DELAY_LOC_DONE:
2711				lun->delay_info.done_type =
2712					delay_info->delay_type;
2713				lun->delay_info.done_delay =
2714					delay_info->delay_secs;
2715				break;
2716			default:
2717				delay_info->status =
2718					CTL_DELAY_STATUS_INVALID_LOC;
2719				break;
2720			}
2721		}
2722
2723		mtx_unlock(&softc->ctl_lock);
2724#else
2725		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2726#endif /* CTL_IO_DELAY */
2727		break;
2728	}
2729	case CTL_REALSYNC_SET: {
2730		int *syncstate;
2731
2732		syncstate = (int *)addr;
2733
2734		mtx_lock(&softc->ctl_lock);
2735		switch (*syncstate) {
2736		case 0:
2737			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2738			break;
2739		case 1:
2740			softc->flags |= CTL_FLAG_REAL_SYNC;
2741			break;
2742		default:
2743			retval = -EINVAL;
2744			break;
2745		}
2746		mtx_unlock(&softc->ctl_lock);
2747		break;
2748	}
2749	case CTL_REALSYNC_GET: {
2750		int *syncstate;
2751
2752		syncstate = (int*)addr;
2753
2754		mtx_lock(&softc->ctl_lock);
2755		if (softc->flags & CTL_FLAG_REAL_SYNC)
2756			*syncstate = 1;
2757		else
2758			*syncstate = 0;
2759		mtx_unlock(&softc->ctl_lock);
2760
2761		break;
2762	}
2763	case CTL_SETSYNC:
2764	case CTL_GETSYNC: {
2765		struct ctl_sync_info *sync_info;
2766		struct ctl_lun *lun;
2767
2768		sync_info = (struct ctl_sync_info *)addr;
2769
2770		mtx_lock(&softc->ctl_lock);
2771		lun = softc->ctl_luns[sync_info->lun_id];
2772		if (lun == NULL) {
2773			mtx_unlock(&softc->ctl_lock);
2774			sync_info->status = CTL_GS_SYNC_NO_LUN;
2775		}
2776		/*
2777		 * Get or set the sync interval.  We're not bounds checking
2778		 * in the set case, hopefully the user won't do something
2779		 * silly.
2780		 */
2781		if (cmd == CTL_GETSYNC)
2782			sync_info->sync_interval = lun->sync_interval;
2783		else
2784			lun->sync_interval = sync_info->sync_interval;
2785
2786		mtx_unlock(&softc->ctl_lock);
2787
2788		sync_info->status = CTL_GS_SYNC_OK;
2789
2790		break;
2791	}
2792	case CTL_GETSTATS: {
2793		struct ctl_stats *stats;
2794		struct ctl_lun *lun;
2795		int i;
2796
2797		stats = (struct ctl_stats *)addr;
2798
2799		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2800		     stats->alloc_len) {
2801			stats->status = CTL_SS_NEED_MORE_SPACE;
2802			stats->num_luns = softc->num_luns;
2803			break;
2804		}
2805		/*
2806		 * XXX KDM no locking here.  If the LUN list changes,
2807		 * things can blow up.
2808		 */
2809		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2810		     i++, lun = STAILQ_NEXT(lun, links)) {
2811			retval = copyout(&lun->stats, &stats->lun_stats[i],
2812					 sizeof(lun->stats));
2813			if (retval != 0)
2814				break;
2815		}
2816		stats->num_luns = softc->num_luns;
2817		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2818				 softc->num_luns;
2819		stats->status = CTL_SS_OK;
2820#ifdef CTL_TIME_IO
2821		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2822#else
2823		stats->flags = CTL_STATS_FLAG_NONE;
2824#endif
2825		getnanouptime(&stats->timestamp);
2826		break;
2827	}
2828	case CTL_ERROR_INJECT: {
2829		struct ctl_error_desc *err_desc, *new_err_desc;
2830		struct ctl_lun *lun;
2831
2832		err_desc = (struct ctl_error_desc *)addr;
2833
2834		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2835				      M_WAITOK | M_ZERO);
2836		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2837
2838		mtx_lock(&softc->ctl_lock);
2839		lun = softc->ctl_luns[err_desc->lun_id];
2840		if (lun == NULL) {
2841			mtx_unlock(&softc->ctl_lock);
2842			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2843			       __func__, (uintmax_t)err_desc->lun_id);
2844			retval = EINVAL;
2845			break;
2846		}
2847
2848		/*
2849		 * We could do some checking here to verify the validity
2850		 * of the request, but given the complexity of error
2851		 * injection requests, the checking logic would be fairly
2852		 * complex.
2853		 *
2854		 * For now, if the request is invalid, it just won't get
2855		 * executed and might get deleted.
2856		 */
2857		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2858
2859		/*
2860		 * XXX KDM check to make sure the serial number is unique,
2861		 * in case we somehow manage to wrap.  That shouldn't
2862		 * happen for a very long time, but it's the right thing to
2863		 * do.
2864		 */
2865		new_err_desc->serial = lun->error_serial;
2866		err_desc->serial = lun->error_serial;
2867		lun->error_serial++;
2868
2869		mtx_unlock(&softc->ctl_lock);
2870		break;
2871	}
2872	case CTL_ERROR_INJECT_DELETE: {
2873		struct ctl_error_desc *delete_desc, *desc, *desc2;
2874		struct ctl_lun *lun;
2875		int delete_done;
2876
2877		delete_desc = (struct ctl_error_desc *)addr;
2878		delete_done = 0;
2879
2880		mtx_lock(&softc->ctl_lock);
2881		lun = softc->ctl_luns[delete_desc->lun_id];
2882		if (lun == NULL) {
2883			mtx_unlock(&softc->ctl_lock);
2884			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2885			       __func__, (uintmax_t)delete_desc->lun_id);
2886			retval = EINVAL;
2887			break;
2888		}
2889		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2890			if (desc->serial != delete_desc->serial)
2891				continue;
2892
2893			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2894				      links);
2895			free(desc, M_CTL);
2896			delete_done = 1;
2897		}
2898		mtx_unlock(&softc->ctl_lock);
2899		if (delete_done == 0) {
2900			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2901			       "error serial %ju on LUN %u\n", __func__,
2902			       delete_desc->serial, delete_desc->lun_id);
2903			retval = EINVAL;
2904			break;
2905		}
2906		break;
2907	}
2908	case CTL_DUMP_STRUCTS: {
2909		int i, j, k;
2910		struct ctl_frontend *fe;
2911
2912		printf("CTL IID to WWPN map start:\n");
2913		for (i = 0; i < CTL_MAX_PORTS; i++) {
2914			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2915				if (softc->wwpn_iid[i][j].in_use == 0)
2916					continue;
2917
2918				printf("port %d iid %u WWPN %#jx\n",
2919				       softc->wwpn_iid[i][j].port,
2920				       softc->wwpn_iid[i][j].iid,
2921				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2922			}
2923		}
2924		printf("CTL IID to WWPN map end\n");
2925		printf("CTL Persistent Reservation information start:\n");
2926		for (i = 0; i < CTL_MAX_LUNS; i++) {
2927			struct ctl_lun *lun;
2928
2929			lun = softc->ctl_luns[i];
2930
2931			if ((lun == NULL)
2932			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2933				continue;
2934
2935			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2936				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2937					if (lun->per_res[j+k].registered == 0)
2938						continue;
2939					printf("LUN %d port %d iid %d key "
2940					       "%#jx\n", i, j, k,
2941					       (uintmax_t)scsi_8btou64(
2942					       lun->per_res[j+k].res_key.key));
2943				}
2944			}
2945		}
2946		printf("CTL Persistent Reservation information end\n");
2947		printf("CTL Frontends:\n");
2948		/*
2949		 * XXX KDM calling this without a lock.  We'd likely want
2950		 * to drop the lock before calling the frontend's dump
2951		 * routine anyway.
2952		 */
2953		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2954			printf("Frontend %s Type %u pport %d vport %d WWNN "
2955			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2956			       fe->physical_port, fe->virtual_port,
2957			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2958
2959			/*
2960			 * Frontends are not required to support the dump
2961			 * routine.
2962			 */
2963			if (fe->fe_dump == NULL)
2964				continue;
2965
2966			fe->fe_dump();
2967		}
2968		printf("CTL Frontend information end\n");
2969		break;
2970	}
2971	case CTL_LUN_REQ: {
2972		struct ctl_lun_req *lun_req;
2973		struct ctl_backend_driver *backend;
2974
2975		lun_req = (struct ctl_lun_req *)addr;
2976
2977		backend = ctl_backend_find(lun_req->backend);
2978		if (backend == NULL) {
2979			lun_req->status = CTL_LUN_ERROR;
2980			snprintf(lun_req->error_str,
2981				 sizeof(lun_req->error_str),
2982				 "Backend \"%s\" not found.",
2983				 lun_req->backend);
2984			break;
2985		}
2986		if (lun_req->num_be_args > 0) {
2987			lun_req->kern_be_args = ctl_copyin_args(
2988				lun_req->num_be_args,
2989				lun_req->be_args,
2990				lun_req->error_str,
2991				sizeof(lun_req->error_str));
2992			if (lun_req->kern_be_args == NULL) {
2993				lun_req->status = CTL_LUN_ERROR;
2994				break;
2995			}
2996		}
2997
2998		retval = backend->ioctl(dev, cmd, addr, flag, td);
2999
3000		if (lun_req->num_be_args > 0) {
3001			ctl_free_args(lun_req->num_be_args,
3002				      lun_req->kern_be_args);
3003		}
3004		break;
3005	}
3006	case CTL_LUN_LIST: {
3007		struct sbuf *sb;
3008		struct ctl_lun *lun;
3009		struct ctl_lun_list *list;
3010		struct ctl_be_lun_option *opt;
3011
3012		list = (struct ctl_lun_list *)addr;
3013
3014		/*
3015		 * Allocate a fixed length sbuf here, based on the length
3016		 * of the user's buffer.  We could allocate an auto-extending
3017		 * buffer, and then tell the user how much larger our
3018		 * amount of data is than his buffer, but that presents
3019		 * some problems:
3020		 *
3021		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3022		 *     we can't hold a lock while calling them with an
3023		 *     auto-extending buffer.
3024 		 *
3025		 * 2.  There is not currently a LUN reference counting
3026		 *     mechanism, outside of outstanding transactions on
3027		 *     the LUN's OOA queue.  So a LUN could go away on us
3028		 *     while we're getting the LUN number, backend-specific
3029		 *     information, etc.  Thus, given the way things
3030		 *     currently work, we need to hold the CTL lock while
3031		 *     grabbing LUN information.
3032		 *
3033		 * So, from the user's standpoint, the best thing to do is
3034		 * allocate what he thinks is a reasonable buffer length,
3035		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3036		 * double the buffer length and try again.  (And repeat
3037		 * that until he succeeds.)
3038		 */
3039		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3040		if (sb == NULL) {
3041			list->status = CTL_LUN_LIST_ERROR;
3042			snprintf(list->error_str, sizeof(list->error_str),
3043				 "Unable to allocate %d bytes for LUN list",
3044				 list->alloc_len);
3045			break;
3046		}
3047
3048		sbuf_printf(sb, "<ctllunlist>\n");
3049
3050		mtx_lock(&softc->ctl_lock);
3051
3052		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3053			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3054					     (uintmax_t)lun->lun);
3055
3056			/*
3057			 * Bail out as soon as we see that we've overfilled
3058			 * the buffer.
3059			 */
3060			if (retval != 0)
3061				break;
3062
3063			retval = sbuf_printf(sb, "<backend_type>%s"
3064					     "</backend_type>\n",
3065					     (lun->backend == NULL) ?  "none" :
3066					     lun->backend->name);
3067
3068			if (retval != 0)
3069				break;
3070
3071			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3072					     lun->be_lun->lun_type);
3073
3074			if (retval != 0)
3075				break;
3076
3077			if (lun->backend == NULL) {
3078				retval = sbuf_printf(sb, "</lun>\n");
3079				if (retval != 0)
3080					break;
3081				continue;
3082			}
3083
3084			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3085					     (lun->be_lun->maxlba > 0) ?
3086					     lun->be_lun->maxlba + 1 : 0);
3087
3088			if (retval != 0)
3089				break;
3090
3091			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3092					     lun->be_lun->blocksize);
3093
3094			if (retval != 0)
3095				break;
3096
3097			retval = sbuf_printf(sb, "<serial_number>");
3098
3099			if (retval != 0)
3100				break;
3101
3102			retval = ctl_sbuf_printf_esc(sb,
3103						     lun->be_lun->serial_num);
3104
3105			if (retval != 0)
3106				break;
3107
3108			retval = sbuf_printf(sb, "</serial_number>\n");
3109
3110			if (retval != 0)
3111				break;
3112
3113			retval = sbuf_printf(sb, "<device_id>");
3114
3115			if (retval != 0)
3116				break;
3117
3118			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3119
3120			if (retval != 0)
3121				break;
3122
3123			retval = sbuf_printf(sb, "</device_id>\n");
3124
3125			if (retval != 0)
3126				break;
3127
3128			if (lun->backend->lun_info != NULL) {
3129				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3130				if (retval != 0)
3131					break;
3132			}
3133			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3134				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3135				if (retval != 0)
3136					break;
3137			}
3138
3139			retval = sbuf_printf(sb, "</lun>\n");
3140
3141			if (retval != 0)
3142				break;
3143		}
3144		mtx_unlock(&softc->ctl_lock);
3145
3146		if ((retval != 0)
3147		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3148			retval = 0;
3149			sbuf_delete(sb);
3150			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3151			snprintf(list->error_str, sizeof(list->error_str),
3152				 "Out of space, %d bytes is too small",
3153				 list->alloc_len);
3154			break;
3155		}
3156
3157		sbuf_finish(sb);
3158
3159		retval = copyout(sbuf_data(sb), list->lun_xml,
3160				 sbuf_len(sb) + 1);
3161
3162		list->fill_len = sbuf_len(sb) + 1;
3163		list->status = CTL_LUN_LIST_OK;
3164		sbuf_delete(sb);
3165		break;
3166	}
3167	case CTL_ISCSI: {
3168		struct ctl_iscsi *ci;
3169		struct ctl_frontend *fe;
3170
3171		ci = (struct ctl_iscsi *)addr;
3172
3173		mtx_lock(&softc->ctl_lock);
3174		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3175			if (strcmp(fe->port_name, "iscsi") == 0)
3176				break;
3177		}
3178		mtx_unlock(&softc->ctl_lock);
3179
3180		if (fe == NULL) {
3181			ci->status = CTL_ISCSI_ERROR;
3182			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3183			break;
3184		}
3185
3186		retval = fe->ioctl(dev, cmd, addr, flag, td);
3187		break;
3188	}
3189	default: {
3190		/* XXX KDM should we fix this? */
3191#if 0
3192		struct ctl_backend_driver *backend;
3193		unsigned int type;
3194		int found;
3195
3196		found = 0;
3197
3198		/*
3199		 * We encode the backend type as the ioctl type for backend
3200		 * ioctls.  So parse it out here, and then search for a
3201		 * backend of this type.
3202		 */
3203		type = _IOC_TYPE(cmd);
3204
3205		STAILQ_FOREACH(backend, &softc->be_list, links) {
3206			if (backend->type == type) {
3207				found = 1;
3208				break;
3209			}
3210		}
3211		if (found == 0) {
3212			printf("ctl: unknown ioctl command %#lx or backend "
3213			       "%d\n", cmd, type);
3214			retval = -EINVAL;
3215			break;
3216		}
3217		retval = backend->ioctl(dev, cmd, addr, flag, td);
3218#endif
3219		retval = ENOTTY;
3220		break;
3221	}
3222	}
3223	return (retval);
3224}
3225
3226uint32_t
3227ctl_get_initindex(struct ctl_nexus *nexus)
3228{
3229	if (nexus->targ_port < CTL_MAX_PORTS)
3230		return (nexus->initid.id +
3231			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3232	else
3233		return (nexus->initid.id +
3234		       ((nexus->targ_port - CTL_MAX_PORTS) *
3235			CTL_MAX_INIT_PER_PORT));
3236}
3237
3238uint32_t
3239ctl_get_resindex(struct ctl_nexus *nexus)
3240{
3241	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3242}
3243
3244uint32_t
3245ctl_port_idx(int port_num)
3246{
3247	if (port_num < CTL_MAX_PORTS)
3248		return(port_num);
3249	else
3250		return(port_num - CTL_MAX_PORTS);
3251}
3252
3253/*
3254 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3255 * that are a power of 2.
3256 */
3257int
3258ctl_ffz(uint32_t *mask, uint32_t size)
3259{
3260	uint32_t num_chunks, num_pieces;
3261	int i, j;
3262
3263	num_chunks = (size >> 5);
3264	if (num_chunks == 0)
3265		num_chunks++;
3266	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3267
3268	for (i = 0; i < num_chunks; i++) {
3269		for (j = 0; j < num_pieces; j++) {
3270			if ((mask[i] & (1 << j)) == 0)
3271				return ((i << 5) + j);
3272		}
3273	}
3274
3275	return (-1);
3276}
3277
3278int
3279ctl_set_mask(uint32_t *mask, uint32_t bit)
3280{
3281	uint32_t chunk, piece;
3282
3283	chunk = bit >> 5;
3284	piece = bit % (sizeof(uint32_t) * 8);
3285
3286	if ((mask[chunk] & (1 << piece)) != 0)
3287		return (-1);
3288	else
3289		mask[chunk] |= (1 << piece);
3290
3291	return (0);
3292}
3293
3294int
3295ctl_clear_mask(uint32_t *mask, uint32_t bit)
3296{
3297	uint32_t chunk, piece;
3298
3299	chunk = bit >> 5;
3300	piece = bit % (sizeof(uint32_t) * 8);
3301
3302	if ((mask[chunk] & (1 << piece)) == 0)
3303		return (-1);
3304	else
3305		mask[chunk] &= ~(1 << piece);
3306
3307	return (0);
3308}
3309
3310int
3311ctl_is_set(uint32_t *mask, uint32_t bit)
3312{
3313	uint32_t chunk, piece;
3314
3315	chunk = bit >> 5;
3316	piece = bit % (sizeof(uint32_t) * 8);
3317
3318	if ((mask[chunk] & (1 << piece)) == 0)
3319		return (0);
3320	else
3321		return (1);
3322}
3323
3324#ifdef unused
3325/*
3326 * The bus, target and lun are optional, they can be filled in later.
3327 * can_wait is used to determine whether we can wait on the malloc or not.
3328 */
3329union ctl_io*
3330ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3331	      uint32_t targ_lun, int can_wait)
3332{
3333	union ctl_io *io;
3334
3335	if (can_wait)
3336		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3337	else
3338		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3339
3340	if (io != NULL) {
3341		io->io_hdr.io_type = io_type;
3342		io->io_hdr.targ_port = targ_port;
3343		/*
3344		 * XXX KDM this needs to change/go away.  We need to move
3345		 * to a preallocated pool of ctl_scsiio structures.
3346		 */
3347		io->io_hdr.nexus.targ_target.id = targ_target;
3348		io->io_hdr.nexus.targ_lun = targ_lun;
3349	}
3350
3351	return (io);
3352}
3353
3354void
3355ctl_kfree_io(union ctl_io *io)
3356{
3357	free(io, M_CTL);
3358}
3359#endif /* unused */
3360
3361/*
3362 * ctl_softc, pool_type, total_ctl_io are passed in.
3363 * npool is passed out.
3364 */
3365int
3366ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3367		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3368{
3369	uint32_t i;
3370	union ctl_io *cur_io, *next_io;
3371	struct ctl_io_pool *pool;
3372	int retval;
3373
3374	retval = 0;
3375
3376	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3377					    M_NOWAIT | M_ZERO);
3378	if (pool == NULL) {
3379		retval = -ENOMEM;
3380		goto bailout;
3381	}
3382
3383	pool->type = pool_type;
3384	pool->ctl_softc = ctl_softc;
3385
3386	mtx_lock(&ctl_softc->pool_lock);
3387	pool->id = ctl_softc->cur_pool_id++;
3388	mtx_unlock(&ctl_softc->pool_lock);
3389
3390	pool->flags = CTL_POOL_FLAG_NONE;
3391	pool->refcount = 1;		/* Reference for validity. */
3392	STAILQ_INIT(&pool->free_queue);
3393
3394	/*
3395	 * XXX KDM other options here:
3396	 * - allocate a page at a time
3397	 * - allocate one big chunk of memory.
3398	 * Page allocation might work well, but would take a little more
3399	 * tracking.
3400	 */
3401	for (i = 0; i < total_ctl_io; i++) {
3402		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3403						M_NOWAIT);
3404		if (cur_io == NULL) {
3405			retval = ENOMEM;
3406			break;
3407		}
3408		cur_io->io_hdr.pool = pool;
3409		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3410		pool->total_ctl_io++;
3411		pool->free_ctl_io++;
3412	}
3413
3414	if (retval != 0) {
3415		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3416		     cur_io != NULL; cur_io = next_io) {
3417			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3418							      links);
3419			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3420				      ctl_io_hdr, links);
3421			free(cur_io, M_CTL);
3422		}
3423
3424		free(pool, M_CTL);
3425		goto bailout;
3426	}
3427	mtx_lock(&ctl_softc->pool_lock);
3428	ctl_softc->num_pools++;
3429	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3430	/*
3431	 * Increment our usage count if this is an external consumer, so we
3432	 * can't get unloaded until the external consumer (most likely a
3433	 * FETD) unloads and frees his pool.
3434	 *
3435	 * XXX KDM will this increment the caller's module use count, or
3436	 * mine?
3437	 */
3438#if 0
3439	if ((pool_type != CTL_POOL_EMERGENCY)
3440	 && (pool_type != CTL_POOL_INTERNAL)
3441	 && (pool_type != CTL_POOL_IOCTL)
3442	 && (pool_type != CTL_POOL_4OTHERSC))
3443		MOD_INC_USE_COUNT;
3444#endif
3445
3446	mtx_unlock(&ctl_softc->pool_lock);
3447
3448	*npool = pool;
3449
3450bailout:
3451
3452	return (retval);
3453}
3454
3455static int
3456ctl_pool_acquire(struct ctl_io_pool *pool)
3457{
3458
3459	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3460
3461	if (pool->flags & CTL_POOL_FLAG_INVALID)
3462		return (-EINVAL);
3463
3464	pool->refcount++;
3465
3466	return (0);
3467}
3468
3469static void
3470ctl_pool_release(struct ctl_io_pool *pool)
3471{
3472	struct ctl_softc *ctl_softc = pool->ctl_softc;
3473	union ctl_io *io;
3474
3475	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3476
3477	if (--pool->refcount != 0)
3478		return;
3479
3480	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3481		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3482			      links);
3483		free(io, M_CTL);
3484	}
3485
3486	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3487	ctl_softc->num_pools--;
3488
3489	/*
3490	 * XXX KDM will this decrement the caller's usage count or mine?
3491	 */
3492#if 0
3493	if ((pool->type != CTL_POOL_EMERGENCY)
3494	 && (pool->type != CTL_POOL_INTERNAL)
3495	 && (pool->type != CTL_POOL_IOCTL))
3496		MOD_DEC_USE_COUNT;
3497#endif
3498
3499	free(pool, M_CTL);
3500}
3501
3502void
3503ctl_pool_free(struct ctl_io_pool *pool)
3504{
3505	struct ctl_softc *ctl_softc;
3506
3507	if (pool == NULL)
3508		return;
3509
3510	ctl_softc = pool->ctl_softc;
3511	mtx_lock(&ctl_softc->pool_lock);
3512	pool->flags |= CTL_POOL_FLAG_INVALID;
3513	ctl_pool_release(pool);
3514	mtx_unlock(&ctl_softc->pool_lock);
3515}
3516
3517/*
3518 * This routine does not block (except for spinlocks of course).
3519 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3520 * possible.
3521 */
3522union ctl_io *
3523ctl_alloc_io(void *pool_ref)
3524{
3525	union ctl_io *io;
3526	struct ctl_softc *ctl_softc;
3527	struct ctl_io_pool *pool, *npool;
3528	struct ctl_io_pool *emergency_pool;
3529
3530	pool = (struct ctl_io_pool *)pool_ref;
3531
3532	if (pool == NULL) {
3533		printf("%s: pool is NULL\n", __func__);
3534		return (NULL);
3535	}
3536
3537	emergency_pool = NULL;
3538
3539	ctl_softc = pool->ctl_softc;
3540
3541	mtx_lock(&ctl_softc->pool_lock);
3542	/*
3543	 * First, try to get the io structure from the user's pool.
3544	 */
3545	if (ctl_pool_acquire(pool) == 0) {
3546		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3547		if (io != NULL) {
3548			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3549			pool->total_allocated++;
3550			pool->free_ctl_io--;
3551			mtx_unlock(&ctl_softc->pool_lock);
3552			return (io);
3553		} else
3554			ctl_pool_release(pool);
3555	}
3556	/*
3557	 * If he doesn't have any io structures left, search for an
3558	 * emergency pool and grab one from there.
3559	 */
3560	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3561		if (npool->type != CTL_POOL_EMERGENCY)
3562			continue;
3563
3564		if (ctl_pool_acquire(npool) != 0)
3565			continue;
3566
3567		emergency_pool = npool;
3568
3569		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3570		if (io != NULL) {
3571			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3572			npool->total_allocated++;
3573			npool->free_ctl_io--;
3574			mtx_unlock(&ctl_softc->pool_lock);
3575			return (io);
3576		} else
3577			ctl_pool_release(npool);
3578	}
3579
3580	/* Drop the spinlock before we malloc */
3581	mtx_unlock(&ctl_softc->pool_lock);
3582
3583	/*
3584	 * The emergency pool (if it exists) didn't have one, so try an
3585	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3586	 */
3587	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3588	if (io != NULL) {
3589		/*
3590		 * If the emergency pool exists but is empty, add this
3591		 * ctl_io to its list when it gets freed.
3592		 */
3593		if (emergency_pool != NULL) {
3594			mtx_lock(&ctl_softc->pool_lock);
3595			if (ctl_pool_acquire(emergency_pool) == 0) {
3596				io->io_hdr.pool = emergency_pool;
3597				emergency_pool->total_ctl_io++;
3598				/*
3599				 * Need to bump this, otherwise
3600				 * total_allocated and total_freed won't
3601				 * match when we no longer have anything
3602				 * outstanding.
3603				 */
3604				emergency_pool->total_allocated++;
3605			}
3606			mtx_unlock(&ctl_softc->pool_lock);
3607		} else
3608			io->io_hdr.pool = NULL;
3609	}
3610
3611	return (io);
3612}
3613
3614void
3615ctl_free_io(union ctl_io *io)
3616{
3617	if (io == NULL)
3618		return;
3619
3620	/*
3621	 * If this ctl_io has a pool, return it to that pool.
3622	 */
3623	if (io->io_hdr.pool != NULL) {
3624		struct ctl_io_pool *pool;
3625#if 0
3626		struct ctl_softc *ctl_softc;
3627		union ctl_io *tmp_io;
3628		unsigned long xflags;
3629		int i;
3630
3631		ctl_softc = control_softc;
3632#endif
3633
3634		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3635
3636		mtx_lock(&pool->ctl_softc->pool_lock);
3637#if 0
3638		save_flags(xflags);
3639
3640		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3641		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3642		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3643		     links)) {
3644			if (tmp_io == io) {
3645				printf("%s: %p is still on the task queue!\n",
3646				       __func__, tmp_io);
3647				printf("%s: (%d): type %d "
3648				       "msg %d cdb %x iptl: "
3649				       "%d:%d:%d:%d tag 0x%04x "
3650				       "flg %#lx\n",
3651					__func__, i,
3652					tmp_io->io_hdr.io_type,
3653					tmp_io->io_hdr.msg_type,
3654					tmp_io->scsiio.cdb[0],
3655					tmp_io->io_hdr.nexus.initid.id,
3656					tmp_io->io_hdr.nexus.targ_port,
3657					tmp_io->io_hdr.nexus.targ_target.id,
3658					tmp_io->io_hdr.nexus.targ_lun,
3659					(tmp_io->io_hdr.io_type ==
3660					CTL_IO_TASK) ?
3661					tmp_io->taskio.tag_num :
3662					tmp_io->scsiio.tag_num,
3663					xflags);
3664				panic("I/O still on the task queue!");
3665			}
3666		}
3667#endif
3668		io->io_hdr.io_type = 0xff;
3669		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3670		pool->total_freed++;
3671		pool->free_ctl_io++;
3672		ctl_pool_release(pool);
3673		mtx_unlock(&pool->ctl_softc->pool_lock);
3674	} else {
3675		/*
3676		 * Otherwise, just free it.  We probably malloced it and
3677		 * the emergency pool wasn't available.
3678		 */
3679		free(io, M_CTL);
3680	}
3681
3682}
3683
3684void
3685ctl_zero_io(union ctl_io *io)
3686{
3687	void *pool_ref;
3688
3689	if (io == NULL)
3690		return;
3691
3692	/*
3693	 * May need to preserve linked list pointers at some point too.
3694	 */
3695	pool_ref = io->io_hdr.pool;
3696
3697	memset(io, 0, sizeof(*io));
3698
3699	io->io_hdr.pool = pool_ref;
3700}
3701
3702/*
3703 * This routine is currently used for internal copies of ctl_ios that need
3704 * to persist for some reason after we've already returned status to the
3705 * FETD.  (Thus the flag set.)
3706 *
3707 * XXX XXX
3708 * Note that this makes a blind copy of all fields in the ctl_io, except
3709 * for the pool reference.  This includes any memory that has been
3710 * allocated!  That memory will no longer be valid after done has been
3711 * called, so this would be VERY DANGEROUS for command that actually does
3712 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3713 * start and stop commands, which don't transfer any data, so this is not a
3714 * problem.  If it is used for anything else, the caller would also need to
3715 * allocate data buffer space and this routine would need to be modified to
3716 * copy the data buffer(s) as well.
3717 */
3718void
3719ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3720{
3721	void *pool_ref;
3722
3723	if ((src == NULL)
3724	 || (dest == NULL))
3725		return;
3726
3727	/*
3728	 * May need to preserve linked list pointers at some point too.
3729	 */
3730	pool_ref = dest->io_hdr.pool;
3731
3732	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3733
3734	dest->io_hdr.pool = pool_ref;
3735	/*
3736	 * We need to know that this is an internal copy, and doesn't need
3737	 * to get passed back to the FETD that allocated it.
3738	 */
3739	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3740}
3741
3742#ifdef NEEDTOPORT
3743static void
3744ctl_update_power_subpage(struct copan_power_subpage *page)
3745{
3746	int num_luns, num_partitions, config_type;
3747	struct ctl_softc *softc;
3748	cs_BOOL_t aor_present, shelf_50pct_power;
3749	cs_raidset_personality_t rs_type;
3750	int max_active_luns;
3751
3752	softc = control_softc;
3753
3754	/* subtract out the processor LUN */
3755	num_luns = softc->num_luns - 1;
3756	/*
3757	 * Default to 7 LUNs active, which was the only number we allowed
3758	 * in the past.
3759	 */
3760	max_active_luns = 7;
3761
3762	num_partitions = config_GetRsPartitionInfo();
3763	config_type = config_GetConfigType();
3764	shelf_50pct_power = config_GetShelfPowerMode();
3765	aor_present = config_IsAorRsPresent();
3766
3767	rs_type = ddb_GetRsRaidType(1);
3768	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3769	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3770		EPRINT(0, "Unsupported RS type %d!", rs_type);
3771	}
3772
3773
3774	page->total_luns = num_luns;
3775
3776	switch (config_type) {
3777	case 40:
3778		/*
3779		 * In a 40 drive configuration, it doesn't matter what DC
3780		 * cards we have, whether we have AOR enabled or not,
3781		 * partitioning or not, or what type of RAIDset we have.
3782		 * In that scenario, we can power up every LUN we present
3783		 * to the user.
3784		 */
3785		max_active_luns = num_luns;
3786
3787		break;
3788	case 64:
3789		if (shelf_50pct_power == CS_FALSE) {
3790			/* 25% power */
3791			if (aor_present == CS_TRUE) {
3792				if (rs_type ==
3793				     CS_RAIDSET_PERSONALITY_RAID5) {
3794					max_active_luns = 7;
3795				} else if (rs_type ==
3796					 CS_RAIDSET_PERSONALITY_RAID1){
3797					max_active_luns = 14;
3798				} else {
3799					/* XXX KDM now what?? */
3800				}
3801			} else {
3802				if (rs_type ==
3803				     CS_RAIDSET_PERSONALITY_RAID5) {
3804					max_active_luns = 8;
3805				} else if (rs_type ==
3806					 CS_RAIDSET_PERSONALITY_RAID1){
3807					max_active_luns = 16;
3808				} else {
3809					/* XXX KDM now what?? */
3810				}
3811			}
3812		} else {
3813			/* 50% power */
3814			/*
3815			 * With 50% power in a 64 drive configuration, we
3816			 * can power all LUNs we present.
3817			 */
3818			max_active_luns = num_luns;
3819		}
3820		break;
3821	case 112:
3822		if (shelf_50pct_power == CS_FALSE) {
3823			/* 25% power */
3824			if (aor_present == CS_TRUE) {
3825				if (rs_type ==
3826				     CS_RAIDSET_PERSONALITY_RAID5) {
3827					max_active_luns = 7;
3828				} else if (rs_type ==
3829					 CS_RAIDSET_PERSONALITY_RAID1){
3830					max_active_luns = 14;
3831				} else {
3832					/* XXX KDM now what?? */
3833				}
3834			} else {
3835				if (rs_type ==
3836				     CS_RAIDSET_PERSONALITY_RAID5) {
3837					max_active_luns = 8;
3838				} else if (rs_type ==
3839					 CS_RAIDSET_PERSONALITY_RAID1){
3840					max_active_luns = 16;
3841				} else {
3842					/* XXX KDM now what?? */
3843				}
3844			}
3845		} else {
3846			/* 50% power */
3847			if (aor_present == CS_TRUE) {
3848				if (rs_type ==
3849				     CS_RAIDSET_PERSONALITY_RAID5) {
3850					max_active_luns = 14;
3851				} else if (rs_type ==
3852					 CS_RAIDSET_PERSONALITY_RAID1){
3853					/*
3854					 * We're assuming here that disk
3855					 * caching is enabled, and so we're
3856					 * able to power up half of each
3857					 * LUN, and cache all writes.
3858					 */
3859					max_active_luns = num_luns;
3860				} else {
3861					/* XXX KDM now what?? */
3862				}
3863			} else {
3864				if (rs_type ==
3865				     CS_RAIDSET_PERSONALITY_RAID5) {
3866					max_active_luns = 15;
3867				} else if (rs_type ==
3868					 CS_RAIDSET_PERSONALITY_RAID1){
3869					max_active_luns = 30;
3870				} else {
3871					/* XXX KDM now what?? */
3872				}
3873			}
3874		}
3875		break;
3876	default:
3877		/*
3878		 * In this case, we have an unknown configuration, so we
3879		 * just use the default from above.
3880		 */
3881		break;
3882	}
3883
3884	page->max_active_luns = max_active_luns;
3885#if 0
3886	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3887	       page->total_luns, page->max_active_luns);
3888#endif
3889}
3890#endif /* NEEDTOPORT */
3891
3892/*
3893 * This routine could be used in the future to load default and/or saved
3894 * mode page parameters for a particuar lun.
3895 */
3896static int
3897ctl_init_page_index(struct ctl_lun *lun)
3898{
3899	int i;
3900	struct ctl_page_index *page_index;
3901	struct ctl_softc *softc;
3902
3903	memcpy(&lun->mode_pages.index, page_index_template,
3904	       sizeof(page_index_template));
3905
3906	softc = lun->ctl_softc;
3907
3908	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3909
3910		page_index = &lun->mode_pages.index[i];
3911		/*
3912		 * If this is a disk-only mode page, there's no point in
3913		 * setting it up.  For some pages, we have to have some
3914		 * basic information about the disk in order to calculate the
3915		 * mode page data.
3916		 */
3917		if ((lun->be_lun->lun_type != T_DIRECT)
3918		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3919			continue;
3920
3921		switch (page_index->page_code & SMPH_PC_MASK) {
3922		case SMS_FORMAT_DEVICE_PAGE: {
3923			struct scsi_format_page *format_page;
3924
3925			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3926				panic("subpage is incorrect!");
3927
3928			/*
3929			 * Sectors per track are set above.  Bytes per
3930			 * sector need to be set here on a per-LUN basis.
3931			 */
3932			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3933			       &format_page_default,
3934			       sizeof(format_page_default));
3935			memcpy(&lun->mode_pages.format_page[
3936			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3937			       sizeof(format_page_changeable));
3938			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3939			       &format_page_default,
3940			       sizeof(format_page_default));
3941			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3942			       &format_page_default,
3943			       sizeof(format_page_default));
3944
3945			format_page = &lun->mode_pages.format_page[
3946				CTL_PAGE_CURRENT];
3947			scsi_ulto2b(lun->be_lun->blocksize,
3948				    format_page->bytes_per_sector);
3949
3950			format_page = &lun->mode_pages.format_page[
3951				CTL_PAGE_DEFAULT];
3952			scsi_ulto2b(lun->be_lun->blocksize,
3953				    format_page->bytes_per_sector);
3954
3955			format_page = &lun->mode_pages.format_page[
3956				CTL_PAGE_SAVED];
3957			scsi_ulto2b(lun->be_lun->blocksize,
3958				    format_page->bytes_per_sector);
3959
3960			page_index->page_data =
3961				(uint8_t *)lun->mode_pages.format_page;
3962			break;
3963		}
3964		case SMS_RIGID_DISK_PAGE: {
3965			struct scsi_rigid_disk_page *rigid_disk_page;
3966			uint32_t sectors_per_cylinder;
3967			uint64_t cylinders;
3968#ifndef	__XSCALE__
3969			int shift;
3970#endif /* !__XSCALE__ */
3971
3972			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3973				panic("invalid subpage value %d",
3974				      page_index->subpage);
3975
3976			/*
3977			 * Rotation rate and sectors per track are set
3978			 * above.  We calculate the cylinders here based on
3979			 * capacity.  Due to the number of heads and
3980			 * sectors per track we're using, smaller arrays
3981			 * may turn out to have 0 cylinders.  Linux and
3982			 * FreeBSD don't pay attention to these mode pages
3983			 * to figure out capacity, but Solaris does.  It
3984			 * seems to deal with 0 cylinders just fine, and
3985			 * works out a fake geometry based on the capacity.
3986			 */
3987			memcpy(&lun->mode_pages.rigid_disk_page[
3988			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3989			       sizeof(rigid_disk_page_default));
3990			memcpy(&lun->mode_pages.rigid_disk_page[
3991			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3992			       sizeof(rigid_disk_page_changeable));
3993			memcpy(&lun->mode_pages.rigid_disk_page[
3994			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3995			       sizeof(rigid_disk_page_default));
3996			memcpy(&lun->mode_pages.rigid_disk_page[
3997			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3998			       sizeof(rigid_disk_page_default));
3999
4000			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4001				CTL_DEFAULT_HEADS;
4002
4003			/*
4004			 * The divide method here will be more accurate,
4005			 * probably, but results in floating point being
4006			 * used in the kernel on i386 (__udivdi3()).  On the
4007			 * XScale, though, __udivdi3() is implemented in
4008			 * software.
4009			 *
4010			 * The shift method for cylinder calculation is
4011			 * accurate if sectors_per_cylinder is a power of
4012			 * 2.  Otherwise it might be slightly off -- you
4013			 * might have a bit of a truncation problem.
4014			 */
4015#ifdef	__XSCALE__
4016			cylinders = (lun->be_lun->maxlba + 1) /
4017				sectors_per_cylinder;
4018#else
4019			for (shift = 31; shift > 0; shift--) {
4020				if (sectors_per_cylinder & (1 << shift))
4021					break;
4022			}
4023			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4024#endif
4025
4026			/*
4027			 * We've basically got 3 bytes, or 24 bits for the
4028			 * cylinder size in the mode page.  If we're over,
4029			 * just round down to 2^24.
4030			 */
4031			if (cylinders > 0xffffff)
4032				cylinders = 0xffffff;
4033
4034			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4035				CTL_PAGE_CURRENT];
4036			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4037
4038			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4039				CTL_PAGE_DEFAULT];
4040			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4041
4042			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4043				CTL_PAGE_SAVED];
4044			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4045
4046			page_index->page_data =
4047				(uint8_t *)lun->mode_pages.rigid_disk_page;
4048			break;
4049		}
4050		case SMS_CACHING_PAGE: {
4051
4052			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4053				panic("invalid subpage value %d",
4054				      page_index->subpage);
4055			/*
4056			 * Defaults should be okay here, no calculations
4057			 * needed.
4058			 */
4059			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4060			       &caching_page_default,
4061			       sizeof(caching_page_default));
4062			memcpy(&lun->mode_pages.caching_page[
4063			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4064			       sizeof(caching_page_changeable));
4065			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4066			       &caching_page_default,
4067			       sizeof(caching_page_default));
4068			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4069			       &caching_page_default,
4070			       sizeof(caching_page_default));
4071			page_index->page_data =
4072				(uint8_t *)lun->mode_pages.caching_page;
4073			break;
4074		}
4075		case SMS_CONTROL_MODE_PAGE: {
4076
4077			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4078				panic("invalid subpage value %d",
4079				      page_index->subpage);
4080
4081			/*
4082			 * Defaults should be okay here, no calculations
4083			 * needed.
4084			 */
4085			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4086			       &control_page_default,
4087			       sizeof(control_page_default));
4088			memcpy(&lun->mode_pages.control_page[
4089			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4090			       sizeof(control_page_changeable));
4091			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4092			       &control_page_default,
4093			       sizeof(control_page_default));
4094			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4095			       &control_page_default,
4096			       sizeof(control_page_default));
4097			page_index->page_data =
4098				(uint8_t *)lun->mode_pages.control_page;
4099			break;
4100
4101		}
4102		case SMS_VENDOR_SPECIFIC_PAGE:{
4103			switch (page_index->subpage) {
4104			case PWR_SUBPAGE_CODE: {
4105				struct copan_power_subpage *current_page,
4106							   *saved_page;
4107
4108				memcpy(&lun->mode_pages.power_subpage[
4109				       CTL_PAGE_CURRENT],
4110				       &power_page_default,
4111				       sizeof(power_page_default));
4112				memcpy(&lun->mode_pages.power_subpage[
4113				       CTL_PAGE_CHANGEABLE],
4114				       &power_page_changeable,
4115				       sizeof(power_page_changeable));
4116				memcpy(&lun->mode_pages.power_subpage[
4117				       CTL_PAGE_DEFAULT],
4118				       &power_page_default,
4119				       sizeof(power_page_default));
4120				memcpy(&lun->mode_pages.power_subpage[
4121				       CTL_PAGE_SAVED],
4122				       &power_page_default,
4123				       sizeof(power_page_default));
4124				page_index->page_data =
4125				    (uint8_t *)lun->mode_pages.power_subpage;
4126
4127				current_page = (struct copan_power_subpage *)
4128					(page_index->page_data +
4129					 (page_index->page_len *
4130					  CTL_PAGE_CURRENT));
4131			        saved_page = (struct copan_power_subpage *)
4132				        (page_index->page_data +
4133					 (page_index->page_len *
4134					  CTL_PAGE_SAVED));
4135				break;
4136			}
4137			case APS_SUBPAGE_CODE: {
4138				struct copan_aps_subpage *current_page,
4139							 *saved_page;
4140
4141				// This gets set multiple times but
4142				// it should always be the same. It's
4143				// only done during init so who cares.
4144				index_to_aps_page = i;
4145
4146				memcpy(&lun->mode_pages.aps_subpage[
4147				       CTL_PAGE_CURRENT],
4148				       &aps_page_default,
4149				       sizeof(aps_page_default));
4150				memcpy(&lun->mode_pages.aps_subpage[
4151				       CTL_PAGE_CHANGEABLE],
4152				       &aps_page_changeable,
4153				       sizeof(aps_page_changeable));
4154				memcpy(&lun->mode_pages.aps_subpage[
4155				       CTL_PAGE_DEFAULT],
4156				       &aps_page_default,
4157				       sizeof(aps_page_default));
4158				memcpy(&lun->mode_pages.aps_subpage[
4159				       CTL_PAGE_SAVED],
4160				       &aps_page_default,
4161				       sizeof(aps_page_default));
4162				page_index->page_data =
4163					(uint8_t *)lun->mode_pages.aps_subpage;
4164
4165				current_page = (struct copan_aps_subpage *)
4166					(page_index->page_data +
4167					 (page_index->page_len *
4168					  CTL_PAGE_CURRENT));
4169				saved_page = (struct copan_aps_subpage *)
4170					(page_index->page_data +
4171					 (page_index->page_len *
4172					  CTL_PAGE_SAVED));
4173				break;
4174			}
4175			case DBGCNF_SUBPAGE_CODE: {
4176				struct copan_debugconf_subpage *current_page,
4177							       *saved_page;
4178
4179				memcpy(&lun->mode_pages.debugconf_subpage[
4180				       CTL_PAGE_CURRENT],
4181				       &debugconf_page_default,
4182				       sizeof(debugconf_page_default));
4183				memcpy(&lun->mode_pages.debugconf_subpage[
4184				       CTL_PAGE_CHANGEABLE],
4185				       &debugconf_page_changeable,
4186				       sizeof(debugconf_page_changeable));
4187				memcpy(&lun->mode_pages.debugconf_subpage[
4188				       CTL_PAGE_DEFAULT],
4189				       &debugconf_page_default,
4190				       sizeof(debugconf_page_default));
4191				memcpy(&lun->mode_pages.debugconf_subpage[
4192				       CTL_PAGE_SAVED],
4193				       &debugconf_page_default,
4194				       sizeof(debugconf_page_default));
4195				page_index->page_data =
4196					(uint8_t *)lun->mode_pages.debugconf_subpage;
4197
4198				current_page = (struct copan_debugconf_subpage *)
4199					(page_index->page_data +
4200					 (page_index->page_len *
4201					  CTL_PAGE_CURRENT));
4202				saved_page = (struct copan_debugconf_subpage *)
4203					(page_index->page_data +
4204					 (page_index->page_len *
4205					  CTL_PAGE_SAVED));
4206				break;
4207			}
4208			default:
4209				panic("invalid subpage value %d",
4210				      page_index->subpage);
4211				break;
4212			}
4213   			break;
4214		}
4215		default:
4216			panic("invalid page value %d",
4217			      page_index->page_code & SMPH_PC_MASK);
4218			break;
4219    	}
4220	}
4221
4222	return (CTL_RETVAL_COMPLETE);
4223}
4224
4225/*
4226 * LUN allocation.
4227 *
4228 * Requirements:
4229 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4230 *   wants us to allocate the LUN and he can block.
4231 * - ctl_softc is always set
4232 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4233 *
4234 * Returns 0 for success, non-zero (errno) for failure.
4235 */
4236static int
4237ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4238	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4239{
4240	struct ctl_lun *nlun, *lun;
4241	struct ctl_frontend *fe;
4242	int lun_number, i, lun_malloced;
4243
4244	if (be_lun == NULL)
4245		return (EINVAL);
4246
4247	/*
4248	 * We currently only support Direct Access or Processor LUN types.
4249	 */
4250	switch (be_lun->lun_type) {
4251	case T_DIRECT:
4252		break;
4253	case T_PROCESSOR:
4254		break;
4255	case T_SEQUENTIAL:
4256	case T_CHANGER:
4257	default:
4258		be_lun->lun_config_status(be_lun->be_lun,
4259					  CTL_LUN_CONFIG_FAILURE);
4260		break;
4261	}
4262	if (ctl_lun == NULL) {
4263		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4264		lun_malloced = 1;
4265	} else {
4266		lun_malloced = 0;
4267		lun = ctl_lun;
4268	}
4269
4270	memset(lun, 0, sizeof(*lun));
4271	if (lun_malloced)
4272		lun->flags = CTL_LUN_MALLOCED;
4273
4274	mtx_lock(&ctl_softc->ctl_lock);
4275	/*
4276	 * See if the caller requested a particular LUN number.  If so, see
4277	 * if it is available.  Otherwise, allocate the first available LUN.
4278	 */
4279	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4280		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4281		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4282			mtx_unlock(&ctl_softc->ctl_lock);
4283			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4284				printf("ctl: requested LUN ID %d is higher "
4285				       "than CTL_MAX_LUNS - 1 (%d)\n",
4286				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4287			} else {
4288				/*
4289				 * XXX KDM return an error, or just assign
4290				 * another LUN ID in this case??
4291				 */
4292				printf("ctl: requested LUN ID %d is already "
4293				       "in use\n", be_lun->req_lun_id);
4294			}
4295			if (lun->flags & CTL_LUN_MALLOCED)
4296				free(lun, M_CTL);
4297			be_lun->lun_config_status(be_lun->be_lun,
4298						  CTL_LUN_CONFIG_FAILURE);
4299			return (ENOSPC);
4300		}
4301		lun_number = be_lun->req_lun_id;
4302	} else {
4303		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4304		if (lun_number == -1) {
4305			mtx_unlock(&ctl_softc->ctl_lock);
4306			printf("ctl: can't allocate LUN on target %ju, out of "
4307			       "LUNs\n", (uintmax_t)target_id.id);
4308			if (lun->flags & CTL_LUN_MALLOCED)
4309				free(lun, M_CTL);
4310			be_lun->lun_config_status(be_lun->be_lun,
4311						  CTL_LUN_CONFIG_FAILURE);
4312			return (ENOSPC);
4313		}
4314	}
4315	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4316
4317	lun->target = target_id;
4318	lun->lun = lun_number;
4319	lun->be_lun = be_lun;
4320	/*
4321	 * The processor LUN is always enabled.  Disk LUNs come on line
4322	 * disabled, and must be enabled by the backend.
4323	 */
4324	lun->flags |= CTL_LUN_DISABLED;
4325	lun->backend = be_lun->be;
4326	be_lun->ctl_lun = lun;
4327	be_lun->lun_id = lun_number;
4328	atomic_add_int(&be_lun->be->num_luns, 1);
4329	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4330		lun->flags |= CTL_LUN_STOPPED;
4331
4332	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4333		lun->flags |= CTL_LUN_INOPERABLE;
4334
4335	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4336		lun->flags |= CTL_LUN_PRIMARY_SC;
4337
4338	lun->ctl_softc = ctl_softc;
4339	TAILQ_INIT(&lun->ooa_queue);
4340	TAILQ_INIT(&lun->blocked_queue);
4341	STAILQ_INIT(&lun->error_list);
4342
4343	/*
4344	 * Initialize the mode page index.
4345	 */
4346	ctl_init_page_index(lun);
4347
4348	/*
4349	 * Set the poweron UA for all initiators on this LUN only.
4350	 */
4351	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4352		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4353
4354	/*
4355	 * Now, before we insert this lun on the lun list, set the lun
4356	 * inventory changed UA for all other luns.
4357	 */
4358	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4359		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4360			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4361		}
4362	}
4363
4364	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4365
4366	ctl_softc->ctl_luns[lun_number] = lun;
4367
4368	ctl_softc->num_luns++;
4369
4370	/* Setup statistics gathering */
4371	lun->stats.device_type = be_lun->lun_type;
4372	lun->stats.lun_number = lun_number;
4373	if (lun->stats.device_type == T_DIRECT)
4374		lun->stats.blocksize = be_lun->blocksize;
4375	else
4376		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4377	for (i = 0;i < CTL_MAX_PORTS;i++)
4378		lun->stats.ports[i].targ_port = i;
4379
4380	mtx_unlock(&ctl_softc->ctl_lock);
4381
4382	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4383
4384	/*
4385	 * Run through each registered FETD and bring it online if it isn't
4386	 * already.  Enable the target ID if it hasn't been enabled, and
4387	 * enable this particular LUN.
4388	 */
4389	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4390		int retval;
4391
4392		/*
4393		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4394		 * to do things differently if we go to a multiple target
4395		 * ID scheme.
4396		 */
4397		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4398
4399			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4400			if (retval != 0) {
4401				printf("ctl_alloc_lun: FETD %s port %d "
4402				       "returned error %d for targ_enable on "
4403				       "target %ju\n", fe->port_name,
4404				       fe->targ_port, retval,
4405				       (uintmax_t)target_id.id);
4406			} else
4407				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4408		}
4409
4410		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4411		if (retval != 0) {
4412			printf("ctl_alloc_lun: FETD %s port %d returned error "
4413			       "%d for lun_enable on target %ju lun %d\n",
4414			       fe->port_name, fe->targ_port, retval,
4415			       (uintmax_t)target_id.id, lun_number);
4416		} else
4417			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4418	}
4419	return (0);
4420}
4421
4422/*
4423 * Delete a LUN.
4424 * Assumptions:
4425 * - LUN has already been marked invalid and any pending I/O has been taken
4426 *   care of.
4427 */
4428static int
4429ctl_free_lun(struct ctl_lun *lun)
4430{
4431	struct ctl_softc *softc;
4432#if 0
4433	struct ctl_frontend *fe;
4434#endif
4435	struct ctl_lun *nlun;
4436	union ctl_io *io, *next_io;
4437	int i;
4438
4439	softc = lun->ctl_softc;
4440
4441	mtx_assert(&softc->ctl_lock, MA_OWNED);
4442
4443	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4444
4445	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4446
4447	softc->ctl_luns[lun->lun] = NULL;
4448
4449	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4450		printf("ctl_free_lun: aieee!! freeing a LUN with "
4451		       "outstanding I/O!!\n");
4452	}
4453
4454	/*
4455	 * If we have anything pending on the RtR queue, remove it.
4456	 */
4457	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4458	     io = next_io) {
4459		uint32_t targ_lun;
4460
4461		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4462		targ_lun = io->io_hdr.nexus.targ_lun;
4463		if (io->io_hdr.nexus.lun_map_fn != NULL)
4464			targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
4465		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4466		 && (targ_lun == lun->lun))
4467			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4468				      ctl_io_hdr, links);
4469	}
4470
4471	/*
4472	 * Then remove everything from the blocked queue.
4473	 */
4474	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4475	     io = next_io) {
4476		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4477		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4478		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4479	}
4480
4481	/*
4482	 * Now clear out the OOA queue, and free all the I/O.
4483	 * XXX KDM should we notify the FETD here?  We probably need to
4484	 * quiesce the LUN before deleting it.
4485	 */
4486	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4487	     io = next_io) {
4488		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4489		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4490		ctl_free_io(io);
4491	}
4492
4493	softc->num_luns--;
4494
4495	/*
4496	 * XXX KDM this scheme only works for a single target/multiple LUN
4497	 * setup.  It needs to be revamped for a multiple target scheme.
4498	 *
4499	 * XXX KDM this results in fe->lun_disable() getting called twice,
4500	 * once when ctl_disable_lun() is called, and a second time here.
4501	 * We really need to re-think the LUN disable semantics.  There
4502	 * should probably be several steps/levels to LUN removal:
4503	 *  - disable
4504	 *  - invalidate
4505	 *  - free
4506 	 *
4507	 * Right now we only have a disable method when communicating to
4508	 * the front end ports, at least for individual LUNs.
4509	 */
4510#if 0
4511	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4512		int retval;
4513
4514		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4515					 lun->lun);
4516		if (retval != 0) {
4517			printf("ctl_free_lun: FETD %s port %d returned error "
4518			       "%d for lun_disable on target %ju lun %jd\n",
4519			       fe->port_name, fe->targ_port, retval,
4520			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4521		}
4522
4523		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4524			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4525
4526			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4527			if (retval != 0) {
4528				printf("ctl_free_lun: FETD %s port %d "
4529				       "returned error %d for targ_disable on "
4530				       "target %ju\n", fe->port_name,
4531				       fe->targ_port, retval,
4532				       (uintmax_t)lun->target.id);
4533			} else
4534				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4535
4536			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4537				continue;
4538
4539#if 0
4540			fe->port_offline(fe->onoff_arg);
4541			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4542#endif
4543		}
4544	}
4545#endif
4546
4547	/*
4548	 * Tell the backend to free resources, if this LUN has a backend.
4549	 */
4550	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4551	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4552
4553	if (lun->flags & CTL_LUN_MALLOCED)
4554		free(lun, M_CTL);
4555
4556	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4557		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4558			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4559		}
4560	}
4561
4562	return (0);
4563}
4564
4565static void
4566ctl_create_lun(struct ctl_be_lun *be_lun)
4567{
4568	struct ctl_softc *ctl_softc;
4569
4570	ctl_softc = control_softc;
4571
4572	/*
4573	 * ctl_alloc_lun() should handle all potential failure cases.
4574	 */
4575	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4576}
4577
4578int
4579ctl_add_lun(struct ctl_be_lun *be_lun)
4580{
4581	struct ctl_softc *ctl_softc;
4582
4583	ctl_softc = control_softc;
4584
4585	mtx_lock(&ctl_softc->ctl_lock);
4586	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4587	mtx_unlock(&ctl_softc->ctl_lock);
4588
4589	ctl_wakeup_thread();
4590
4591	return (0);
4592}
4593
4594int
4595ctl_enable_lun(struct ctl_be_lun *be_lun)
4596{
4597	struct ctl_softc *ctl_softc;
4598	struct ctl_frontend *fe, *nfe;
4599	struct ctl_lun *lun;
4600	int retval;
4601
4602	ctl_softc = control_softc;
4603
4604	lun = (struct ctl_lun *)be_lun->ctl_lun;
4605
4606	mtx_lock(&ctl_softc->ctl_lock);
4607	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4608		/*
4609		 * eh?  Why did we get called if the LUN is already
4610		 * enabled?
4611		 */
4612		mtx_unlock(&ctl_softc->ctl_lock);
4613		return (0);
4614	}
4615	lun->flags &= ~CTL_LUN_DISABLED;
4616
4617	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4618		nfe = STAILQ_NEXT(fe, links);
4619
4620		/*
4621		 * Drop the lock while we call the FETD's enable routine.
4622		 * This can lead to a callback into CTL (at least in the
4623		 * case of the internal initiator frontend.
4624		 */
4625		mtx_unlock(&ctl_softc->ctl_lock);
4626		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4627		mtx_lock(&ctl_softc->ctl_lock);
4628		if (retval != 0) {
4629			printf("%s: FETD %s port %d returned error "
4630			       "%d for lun_enable on target %ju lun %jd\n",
4631			       __func__, fe->port_name, fe->targ_port, retval,
4632			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4633		}
4634#if 0
4635		 else {
4636            /* NOTE:  TODO:  why does lun enable affect port status? */
4637			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4638		}
4639#endif
4640	}
4641
4642	mtx_unlock(&ctl_softc->ctl_lock);
4643
4644	return (0);
4645}
4646
4647int
4648ctl_disable_lun(struct ctl_be_lun *be_lun)
4649{
4650	struct ctl_softc *ctl_softc;
4651	struct ctl_frontend *fe;
4652	struct ctl_lun *lun;
4653	int retval;
4654
4655	ctl_softc = control_softc;
4656
4657	lun = (struct ctl_lun *)be_lun->ctl_lun;
4658
4659	mtx_lock(&ctl_softc->ctl_lock);
4660
4661	if (lun->flags & CTL_LUN_DISABLED) {
4662		mtx_unlock(&ctl_softc->ctl_lock);
4663		return (0);
4664	}
4665	lun->flags |= CTL_LUN_DISABLED;
4666
4667	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4668		mtx_unlock(&ctl_softc->ctl_lock);
4669		/*
4670		 * Drop the lock before we call the frontend's disable
4671		 * routine, to avoid lock order reversals.
4672		 *
4673		 * XXX KDM what happens if the frontend list changes while
4674		 * we're traversing it?  It's unlikely, but should be handled.
4675		 */
4676		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4677					 lun->lun);
4678		mtx_lock(&ctl_softc->ctl_lock);
4679		if (retval != 0) {
4680			printf("ctl_alloc_lun: FETD %s port %d returned error "
4681			       "%d for lun_disable on target %ju lun %jd\n",
4682			       fe->port_name, fe->targ_port, retval,
4683			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4684		}
4685	}
4686
4687	mtx_unlock(&ctl_softc->ctl_lock);
4688
4689	return (0);
4690}
4691
4692int
4693ctl_start_lun(struct ctl_be_lun *be_lun)
4694{
4695	struct ctl_softc *ctl_softc;
4696	struct ctl_lun *lun;
4697
4698	ctl_softc = control_softc;
4699
4700	lun = (struct ctl_lun *)be_lun->ctl_lun;
4701
4702	mtx_lock(&ctl_softc->ctl_lock);
4703	lun->flags &= ~CTL_LUN_STOPPED;
4704	mtx_unlock(&ctl_softc->ctl_lock);
4705
4706	return (0);
4707}
4708
4709int
4710ctl_stop_lun(struct ctl_be_lun *be_lun)
4711{
4712	struct ctl_softc *ctl_softc;
4713	struct ctl_lun *lun;
4714
4715	ctl_softc = control_softc;
4716
4717	lun = (struct ctl_lun *)be_lun->ctl_lun;
4718
4719	mtx_lock(&ctl_softc->ctl_lock);
4720	lun->flags |= CTL_LUN_STOPPED;
4721	mtx_unlock(&ctl_softc->ctl_lock);
4722
4723	return (0);
4724}
4725
4726int
4727ctl_lun_offline(struct ctl_be_lun *be_lun)
4728{
4729	struct ctl_softc *ctl_softc;
4730	struct ctl_lun *lun;
4731
4732	ctl_softc = control_softc;
4733
4734	lun = (struct ctl_lun *)be_lun->ctl_lun;
4735
4736	mtx_lock(&ctl_softc->ctl_lock);
4737	lun->flags |= CTL_LUN_OFFLINE;
4738	mtx_unlock(&ctl_softc->ctl_lock);
4739
4740	return (0);
4741}
4742
4743int
4744ctl_lun_online(struct ctl_be_lun *be_lun)
4745{
4746	struct ctl_softc *ctl_softc;
4747	struct ctl_lun *lun;
4748
4749	ctl_softc = control_softc;
4750
4751	lun = (struct ctl_lun *)be_lun->ctl_lun;
4752
4753	mtx_lock(&ctl_softc->ctl_lock);
4754	lun->flags &= ~CTL_LUN_OFFLINE;
4755	mtx_unlock(&ctl_softc->ctl_lock);
4756
4757	return (0);
4758}
4759
4760int
4761ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4762{
4763	struct ctl_softc *ctl_softc;
4764	struct ctl_lun *lun;
4765
4766	ctl_softc = control_softc;
4767
4768	lun = (struct ctl_lun *)be_lun->ctl_lun;
4769
4770	mtx_lock(&ctl_softc->ctl_lock);
4771
4772	/*
4773	 * The LUN needs to be disabled before it can be marked invalid.
4774	 */
4775	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4776		mtx_unlock(&ctl_softc->ctl_lock);
4777		return (-1);
4778	}
4779	/*
4780	 * Mark the LUN invalid.
4781	 */
4782	lun->flags |= CTL_LUN_INVALID;
4783
4784	/*
4785	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4786	 * If we have something in the OOA queue, we'll free it when the
4787	 * last I/O completes.
4788	 */
4789	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4790		ctl_free_lun(lun);
4791	mtx_unlock(&ctl_softc->ctl_lock);
4792
4793	return (0);
4794}
4795
4796int
4797ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4798{
4799	struct ctl_softc *ctl_softc;
4800	struct ctl_lun *lun;
4801
4802	ctl_softc = control_softc;
4803	lun = (struct ctl_lun *)be_lun->ctl_lun;
4804
4805	mtx_lock(&ctl_softc->ctl_lock);
4806	lun->flags |= CTL_LUN_INOPERABLE;
4807	mtx_unlock(&ctl_softc->ctl_lock);
4808
4809	return (0);
4810}
4811
4812int
4813ctl_lun_operable(struct ctl_be_lun *be_lun)
4814{
4815	struct ctl_softc *ctl_softc;
4816	struct ctl_lun *lun;
4817
4818	ctl_softc = control_softc;
4819	lun = (struct ctl_lun *)be_lun->ctl_lun;
4820
4821	mtx_lock(&ctl_softc->ctl_lock);
4822	lun->flags &= ~CTL_LUN_INOPERABLE;
4823	mtx_unlock(&ctl_softc->ctl_lock);
4824
4825	return (0);
4826}
4827
4828int
4829ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4830		   int lock)
4831{
4832	struct ctl_softc *softc;
4833	struct ctl_lun *lun;
4834	struct copan_aps_subpage *current_sp;
4835	struct ctl_page_index *page_index;
4836	int i;
4837
4838	softc = control_softc;
4839
4840	mtx_lock(&softc->ctl_lock);
4841
4842	lun = (struct ctl_lun *)be_lun->ctl_lun;
4843
4844	page_index = NULL;
4845	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4846		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4847		     APS_PAGE_CODE)
4848			continue;
4849
4850		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4851			continue;
4852		page_index = &lun->mode_pages.index[i];
4853	}
4854
4855	if (page_index == NULL) {
4856		mtx_unlock(&softc->ctl_lock);
4857		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4858		       (uintmax_t)lun->lun);
4859		return (1);
4860	}
4861#if 0
4862	if ((softc->aps_locked_lun != 0)
4863	 && (softc->aps_locked_lun != lun->lun)) {
4864		printf("%s: attempt to lock LUN %llu when %llu is already "
4865		       "locked\n");
4866		mtx_unlock(&softc->ctl_lock);
4867		return (1);
4868	}
4869#endif
4870
4871	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4872		(page_index->page_len * CTL_PAGE_CURRENT));
4873
4874	if (lock != 0) {
4875		current_sp->lock_active = APS_LOCK_ACTIVE;
4876		softc->aps_locked_lun = lun->lun;
4877	} else {
4878		current_sp->lock_active = 0;
4879		softc->aps_locked_lun = 0;
4880	}
4881
4882
4883	/*
4884	 * If we're in HA mode, try to send the lock message to the other
4885	 * side.
4886	 */
4887	if (ctl_is_single == 0) {
4888		int isc_retval;
4889		union ctl_ha_msg lock_msg;
4890
4891		lock_msg.hdr.nexus = *nexus;
4892		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4893		if (lock != 0)
4894			lock_msg.aps.lock_flag = 1;
4895		else
4896			lock_msg.aps.lock_flag = 0;
4897		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4898					 sizeof(lock_msg), 0);
4899		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4900			printf("%s: APS (lock=%d) error returned from "
4901			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4902			mtx_unlock(&softc->ctl_lock);
4903			return (1);
4904		}
4905	}
4906
4907	mtx_unlock(&softc->ctl_lock);
4908
4909	return (0);
4910}
4911
4912void
4913ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4914{
4915	struct ctl_lun *lun;
4916	struct ctl_softc *softc;
4917	int i;
4918
4919	softc = control_softc;
4920
4921	mtx_lock(&softc->ctl_lock);
4922
4923	lun = (struct ctl_lun *)be_lun->ctl_lun;
4924
4925	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4926		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4927
4928	mtx_unlock(&softc->ctl_lock);
4929}
4930
4931/*
4932 * Backend "memory move is complete" callback for requests that never
4933 * make it down to say RAIDCore's configuration code.
4934 */
4935int
4936ctl_config_move_done(union ctl_io *io)
4937{
4938	int retval;
4939
4940	retval = CTL_RETVAL_COMPLETE;
4941
4942
4943	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4944	/*
4945	 * XXX KDM this shouldn't happen, but what if it does?
4946	 */
4947	if (io->io_hdr.io_type != CTL_IO_SCSI)
4948		panic("I/O type isn't CTL_IO_SCSI!");
4949
4950	if ((io->io_hdr.port_status == 0)
4951	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4952	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4953		io->io_hdr.status = CTL_SUCCESS;
4954	else if ((io->io_hdr.port_status != 0)
4955	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4956	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4957		/*
4958		 * For hardware error sense keys, the sense key
4959		 * specific value is defined to be a retry count,
4960		 * but we use it to pass back an internal FETD
4961		 * error code.  XXX KDM  Hopefully the FETD is only
4962		 * using 16 bits for an error code, since that's
4963		 * all the space we have in the sks field.
4964		 */
4965		ctl_set_internal_failure(&io->scsiio,
4966					 /*sks_valid*/ 1,
4967					 /*retry_count*/
4968					 io->io_hdr.port_status);
4969		free(io->scsiio.kern_data_ptr, M_CTL);
4970		ctl_done(io);
4971		goto bailout;
4972	}
4973
4974	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4975	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4976	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4977		/*
4978		 * XXX KDM just assuming a single pointer here, and not a
4979		 * S/G list.  If we start using S/G lists for config data,
4980		 * we'll need to know how to clean them up here as well.
4981		 */
4982		free(io->scsiio.kern_data_ptr, M_CTL);
4983		/* Hopefully the user has already set the status... */
4984		ctl_done(io);
4985	} else {
4986		/*
4987		 * XXX KDM now we need to continue data movement.  Some
4988		 * options:
4989		 * - call ctl_scsiio() again?  We don't do this for data
4990		 *   writes, because for those at least we know ahead of
4991		 *   time where the write will go and how long it is.  For
4992		 *   config writes, though, that information is largely
4993		 *   contained within the write itself, thus we need to
4994		 *   parse out the data again.
4995		 *
4996		 * - Call some other function once the data is in?
4997		 */
4998
4999		/*
5000		 * XXX KDM call ctl_scsiio() again for now, and check flag
5001		 * bits to see whether we're allocated or not.
5002		 */
5003		retval = ctl_scsiio(&io->scsiio);
5004	}
5005bailout:
5006	return (retval);
5007}
5008
5009/*
5010 * This gets called by a backend driver when it is done with a
5011 * configuration write.
5012 */
5013void
5014ctl_config_write_done(union ctl_io *io)
5015{
5016	/*
5017	 * If the IO_CONT flag is set, we need to call the supplied
5018	 * function to continue processing the I/O, instead of completing
5019	 * the I/O just yet.
5020	 *
5021	 * If there is an error, though, we don't want to keep processing.
5022	 * Instead, just send status back to the initiator.
5023	 */
5024	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5025	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5026	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5027		io->scsiio.io_cont(io);
5028		return;
5029	}
5030	/*
5031	 * Since a configuration write can be done for commands that actually
5032	 * have data allocated, like write buffer, and commands that have
5033	 * no data, like start/stop unit, we need to check here.
5034	 */
5035	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5036		free(io->scsiio.kern_data_ptr, M_CTL);
5037	ctl_done(io);
5038}
5039
5040/*
5041 * SCSI release command.
5042 */
5043int
5044ctl_scsi_release(struct ctl_scsiio *ctsio)
5045{
5046	int length, longid, thirdparty_id, resv_id;
5047	struct ctl_softc *ctl_softc;
5048	struct ctl_lun *lun;
5049
5050	length = 0;
5051	resv_id = 0;
5052
5053	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5054
5055	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5056	ctl_softc = control_softc;
5057
5058	switch (ctsio->cdb[0]) {
5059	case RELEASE: {
5060		struct scsi_release *cdb;
5061
5062		cdb = (struct scsi_release *)ctsio->cdb;
5063		if ((cdb->byte2 & 0x1f) != 0) {
5064			ctl_set_invalid_field(ctsio,
5065					      /*sks_valid*/ 1,
5066					      /*command*/ 1,
5067					      /*field*/ 1,
5068					      /*bit_valid*/ 0,
5069					      /*bit*/ 0);
5070			ctl_done((union ctl_io *)ctsio);
5071			return (CTL_RETVAL_COMPLETE);
5072		}
5073		break;
5074	}
5075	case RELEASE_10: {
5076		struct scsi_release_10 *cdb;
5077
5078		cdb = (struct scsi_release_10 *)ctsio->cdb;
5079
5080		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5081			ctl_set_invalid_field(ctsio,
5082					      /*sks_valid*/ 1,
5083					      /*command*/ 1,
5084					      /*field*/ 1,
5085					      /*bit_valid*/ 1,
5086					      /*bit*/ 0);
5087			ctl_done((union ctl_io *)ctsio);
5088			return (CTL_RETVAL_COMPLETE);
5089
5090		}
5091
5092		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5093			ctl_set_invalid_field(ctsio,
5094					      /*sks_valid*/ 1,
5095					      /*command*/ 1,
5096					      /*field*/ 1,
5097					      /*bit_valid*/ 1,
5098					      /*bit*/ 4);
5099			ctl_done((union ctl_io *)ctsio);
5100			return (CTL_RETVAL_COMPLETE);
5101		}
5102
5103		if (cdb->byte2 & SR10_LONGID)
5104			longid = 1;
5105		else
5106			thirdparty_id = cdb->thirdparty_id;
5107
5108		resv_id = cdb->resv_id;
5109		length = scsi_2btoul(cdb->length);
5110		break;
5111	}
5112	}
5113
5114
5115	/*
5116	 * XXX KDM right now, we only support LUN reservation.  We don't
5117	 * support 3rd party reservations, or extent reservations, which
5118	 * might actually need the parameter list.  If we've gotten this
5119	 * far, we've got a LUN reservation.  Anything else got kicked out
5120	 * above.  So, according to SPC, ignore the length.
5121	 */
5122	length = 0;
5123
5124	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5125	 && (length > 0)) {
5126		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5127		ctsio->kern_data_len = length;
5128		ctsio->kern_total_len = length;
5129		ctsio->kern_data_resid = 0;
5130		ctsio->kern_rel_offset = 0;
5131		ctsio->kern_sg_entries = 0;
5132		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5133		ctsio->be_move_done = ctl_config_move_done;
5134		ctl_datamove((union ctl_io *)ctsio);
5135
5136		return (CTL_RETVAL_COMPLETE);
5137	}
5138
5139	if (length > 0)
5140		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5141
5142	mtx_lock(&ctl_softc->ctl_lock);
5143
5144	/*
5145	 * According to SPC, it is not an error for an intiator to attempt
5146	 * to release a reservation on a LUN that isn't reserved, or that
5147	 * is reserved by another initiator.  The reservation can only be
5148	 * released, though, by the initiator who made it or by one of
5149	 * several reset type events.
5150	 */
5151	if (lun->flags & CTL_LUN_RESERVED) {
5152		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5153		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5154		 && (ctsio->io_hdr.nexus.targ_target.id ==
5155		     lun->rsv_nexus.targ_target.id)) {
5156			lun->flags &= ~CTL_LUN_RESERVED;
5157		}
5158	}
5159
5160	ctsio->scsi_status = SCSI_STATUS_OK;
5161	ctsio->io_hdr.status = CTL_SUCCESS;
5162
5163	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5164		free(ctsio->kern_data_ptr, M_CTL);
5165		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5166	}
5167
5168	mtx_unlock(&ctl_softc->ctl_lock);
5169
5170	ctl_done((union ctl_io *)ctsio);
5171	return (CTL_RETVAL_COMPLETE);
5172}
5173
5174int
5175ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5176{
5177	int extent, thirdparty, longid;
5178	int resv_id, length;
5179	uint64_t thirdparty_id;
5180	struct ctl_softc *ctl_softc;
5181	struct ctl_lun *lun;
5182
5183	extent = 0;
5184	thirdparty = 0;
5185	longid = 0;
5186	resv_id = 0;
5187	length = 0;
5188	thirdparty_id = 0;
5189
5190	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5191
5192	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5193	ctl_softc = control_softc;
5194
5195	switch (ctsio->cdb[0]) {
5196	case RESERVE: {
5197		struct scsi_reserve *cdb;
5198
5199		cdb = (struct scsi_reserve *)ctsio->cdb;
5200		if ((cdb->byte2 & 0x1f) != 0) {
5201			ctl_set_invalid_field(ctsio,
5202					      /*sks_valid*/ 1,
5203					      /*command*/ 1,
5204					      /*field*/ 1,
5205					      /*bit_valid*/ 0,
5206					      /*bit*/ 0);
5207			ctl_done((union ctl_io *)ctsio);
5208			return (CTL_RETVAL_COMPLETE);
5209		}
5210		resv_id = cdb->resv_id;
5211		length = scsi_2btoul(cdb->length);
5212		break;
5213	}
5214	case RESERVE_10: {
5215		struct scsi_reserve_10 *cdb;
5216
5217		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5218
5219		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5220			ctl_set_invalid_field(ctsio,
5221					      /*sks_valid*/ 1,
5222					      /*command*/ 1,
5223					      /*field*/ 1,
5224					      /*bit_valid*/ 1,
5225					      /*bit*/ 0);
5226			ctl_done((union ctl_io *)ctsio);
5227			return (CTL_RETVAL_COMPLETE);
5228		}
5229		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5230			ctl_set_invalid_field(ctsio,
5231					      /*sks_valid*/ 1,
5232					      /*command*/ 1,
5233					      /*field*/ 1,
5234					      /*bit_valid*/ 1,
5235					      /*bit*/ 4);
5236			ctl_done((union ctl_io *)ctsio);
5237			return (CTL_RETVAL_COMPLETE);
5238		}
5239		if (cdb->byte2 & SR10_LONGID)
5240			longid = 1;
5241		else
5242			thirdparty_id = cdb->thirdparty_id;
5243
5244		resv_id = cdb->resv_id;
5245		length = scsi_2btoul(cdb->length);
5246		break;
5247	}
5248	}
5249
5250	/*
5251	 * XXX KDM right now, we only support LUN reservation.  We don't
5252	 * support 3rd party reservations, or extent reservations, which
5253	 * might actually need the parameter list.  If we've gotten this
5254	 * far, we've got a LUN reservation.  Anything else got kicked out
5255	 * above.  So, according to SPC, ignore the length.
5256	 */
5257	length = 0;
5258
5259	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5260	 && (length > 0)) {
5261		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5262		ctsio->kern_data_len = length;
5263		ctsio->kern_total_len = length;
5264		ctsio->kern_data_resid = 0;
5265		ctsio->kern_rel_offset = 0;
5266		ctsio->kern_sg_entries = 0;
5267		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5268		ctsio->be_move_done = ctl_config_move_done;
5269		ctl_datamove((union ctl_io *)ctsio);
5270
5271		return (CTL_RETVAL_COMPLETE);
5272	}
5273
5274	if (length > 0)
5275		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5276
5277	mtx_lock(&ctl_softc->ctl_lock);
5278	if (lun->flags & CTL_LUN_RESERVED) {
5279		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5280		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5281		 || (ctsio->io_hdr.nexus.targ_target.id !=
5282		     lun->rsv_nexus.targ_target.id)) {
5283			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5284			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5285			goto bailout;
5286		}
5287	}
5288
5289	lun->flags |= CTL_LUN_RESERVED;
5290	lun->rsv_nexus = ctsio->io_hdr.nexus;
5291
5292	ctsio->scsi_status = SCSI_STATUS_OK;
5293	ctsio->io_hdr.status = CTL_SUCCESS;
5294
5295bailout:
5296	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5297		free(ctsio->kern_data_ptr, M_CTL);
5298		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5299	}
5300
5301	mtx_unlock(&ctl_softc->ctl_lock);
5302
5303	ctl_done((union ctl_io *)ctsio);
5304	return (CTL_RETVAL_COMPLETE);
5305}
5306
5307int
5308ctl_start_stop(struct ctl_scsiio *ctsio)
5309{
5310	struct scsi_start_stop_unit *cdb;
5311	struct ctl_lun *lun;
5312	struct ctl_softc *ctl_softc;
5313	int retval;
5314
5315	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5316
5317	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5318	ctl_softc = control_softc;
5319	retval = 0;
5320
5321	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5322
5323	/*
5324	 * XXX KDM
5325	 * We don't support the immediate bit on a stop unit.  In order to
5326	 * do that, we would need to code up a way to know that a stop is
5327	 * pending, and hold off any new commands until it completes, one
5328	 * way or another.  Then we could accept or reject those commands
5329	 * depending on its status.  We would almost need to do the reverse
5330	 * of what we do below for an immediate start -- return the copy of
5331	 * the ctl_io to the FETD with status to send to the host (and to
5332	 * free the copy!) and then free the original I/O once the stop
5333	 * actually completes.  That way, the OOA queue mechanism can work
5334	 * to block commands that shouldn't proceed.  Another alternative
5335	 * would be to put the copy in the queue in place of the original,
5336	 * and return the original back to the caller.  That could be
5337	 * slightly safer..
5338	 */
5339	if ((cdb->byte2 & SSS_IMMED)
5340	 && ((cdb->how & SSS_START) == 0)) {
5341		ctl_set_invalid_field(ctsio,
5342				      /*sks_valid*/ 1,
5343				      /*command*/ 1,
5344				      /*field*/ 1,
5345				      /*bit_valid*/ 1,
5346				      /*bit*/ 0);
5347		ctl_done((union ctl_io *)ctsio);
5348		return (CTL_RETVAL_COMPLETE);
5349	}
5350
5351	/*
5352	 * We don't support the power conditions field.  We need to check
5353	 * this prior to checking the load/eject and start/stop bits.
5354	 */
5355	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5356		ctl_set_invalid_field(ctsio,
5357				      /*sks_valid*/ 1,
5358				      /*command*/ 1,
5359				      /*field*/ 4,
5360				      /*bit_valid*/ 1,
5361				      /*bit*/ 4);
5362		ctl_done((union ctl_io *)ctsio);
5363		return (CTL_RETVAL_COMPLETE);
5364	}
5365
5366	/*
5367	 * Media isn't removable, so we can't load or eject it.
5368	 */
5369	if ((cdb->how & SSS_LOEJ) != 0) {
5370		ctl_set_invalid_field(ctsio,
5371				      /*sks_valid*/ 1,
5372				      /*command*/ 1,
5373				      /*field*/ 4,
5374				      /*bit_valid*/ 1,
5375				      /*bit*/ 1);
5376		ctl_done((union ctl_io *)ctsio);
5377		return (CTL_RETVAL_COMPLETE);
5378	}
5379
5380	if ((lun->flags & CTL_LUN_PR_RESERVED)
5381	 && ((cdb->how & SSS_START)==0)) {
5382		uint32_t residx;
5383
5384		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5385		if (!lun->per_res[residx].registered
5386		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5387
5388			ctl_set_reservation_conflict(ctsio);
5389			ctl_done((union ctl_io *)ctsio);
5390			return (CTL_RETVAL_COMPLETE);
5391		}
5392	}
5393
5394	/*
5395	 * If there is no backend on this device, we can't start or stop
5396	 * it.  In theory we shouldn't get any start/stop commands in the
5397	 * first place at this level if the LUN doesn't have a backend.
5398	 * That should get stopped by the command decode code.
5399	 */
5400	if (lun->backend == NULL) {
5401		ctl_set_invalid_opcode(ctsio);
5402		ctl_done((union ctl_io *)ctsio);
5403		return (CTL_RETVAL_COMPLETE);
5404	}
5405
5406	/*
5407	 * XXX KDM Copan-specific offline behavior.
5408	 * Figure out a reasonable way to port this?
5409	 */
5410#ifdef NEEDTOPORT
5411	mtx_lock(&ctl_softc->ctl_lock);
5412
5413	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5414	 && (lun->flags & CTL_LUN_OFFLINE)) {
5415		/*
5416		 * If the LUN is offline, and the on/offline bit isn't set,
5417		 * reject the start or stop.  Otherwise, let it through.
5418		 */
5419		mtx_unlock(&ctl_softc->ctl_lock);
5420		ctl_set_lun_not_ready(ctsio);
5421		ctl_done((union ctl_io *)ctsio);
5422	} else {
5423		mtx_unlock(&ctl_softc->ctl_lock);
5424#endif /* NEEDTOPORT */
5425		/*
5426		 * This could be a start or a stop when we're online,
5427		 * or a stop/offline or start/online.  A start or stop when
5428		 * we're offline is covered in the case above.
5429		 */
5430		/*
5431		 * In the non-immediate case, we send the request to
5432		 * the backend and return status to the user when
5433		 * it is done.
5434		 *
5435		 * In the immediate case, we allocate a new ctl_io
5436		 * to hold a copy of the request, and send that to
5437		 * the backend.  We then set good status on the
5438		 * user's request and return it immediately.
5439		 */
5440		if (cdb->byte2 & SSS_IMMED) {
5441			union ctl_io *new_io;
5442
5443			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5444			if (new_io == NULL) {
5445				ctl_set_busy(ctsio);
5446				ctl_done((union ctl_io *)ctsio);
5447			} else {
5448				ctl_copy_io((union ctl_io *)ctsio,
5449					    new_io);
5450				retval = lun->backend->config_write(new_io);
5451				ctl_set_success(ctsio);
5452				ctl_done((union ctl_io *)ctsio);
5453			}
5454		} else {
5455			retval = lun->backend->config_write(
5456				(union ctl_io *)ctsio);
5457		}
5458#ifdef NEEDTOPORT
5459	}
5460#endif
5461	return (retval);
5462}
5463
5464/*
5465 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5466 * we don't really do anything with the LBA and length fields if the user
5467 * passes them in.  Instead we'll just flush out the cache for the entire
5468 * LUN.
5469 */
5470int
5471ctl_sync_cache(struct ctl_scsiio *ctsio)
5472{
5473	struct ctl_lun *lun;
5474	struct ctl_softc *ctl_softc;
5475	uint64_t starting_lba;
5476	uint32_t block_count;
5477	int reladr, immed;
5478	int retval;
5479
5480	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5481
5482	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5483	ctl_softc = control_softc;
5484	retval = 0;
5485	reladr = 0;
5486	immed = 0;
5487
5488	switch (ctsio->cdb[0]) {
5489	case SYNCHRONIZE_CACHE: {
5490		struct scsi_sync_cache *cdb;
5491		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5492
5493		if (cdb->byte2 & SSC_RELADR)
5494			reladr = 1;
5495
5496		if (cdb->byte2 & SSC_IMMED)
5497			immed = 1;
5498
5499		starting_lba = scsi_4btoul(cdb->begin_lba);
5500		block_count = scsi_2btoul(cdb->lb_count);
5501		break;
5502	}
5503	case SYNCHRONIZE_CACHE_16: {
5504		struct scsi_sync_cache_16 *cdb;
5505		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5506
5507		if (cdb->byte2 & SSC_RELADR)
5508			reladr = 1;
5509
5510		if (cdb->byte2 & SSC_IMMED)
5511			immed = 1;
5512
5513		starting_lba = scsi_8btou64(cdb->begin_lba);
5514		block_count = scsi_4btoul(cdb->lb_count);
5515		break;
5516	}
5517	default:
5518		ctl_set_invalid_opcode(ctsio);
5519		ctl_done((union ctl_io *)ctsio);
5520		goto bailout;
5521		break; /* NOTREACHED */
5522	}
5523
5524	if (immed) {
5525		/*
5526		 * We don't support the immediate bit.  Since it's in the
5527		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5528		 * commands, we can just return the same error in either
5529		 * case.
5530		 */
5531		ctl_set_invalid_field(ctsio,
5532				      /*sks_valid*/ 1,
5533				      /*command*/ 1,
5534				      /*field*/ 1,
5535				      /*bit_valid*/ 1,
5536				      /*bit*/ 1);
5537		ctl_done((union ctl_io *)ctsio);
5538		goto bailout;
5539	}
5540
5541	if (reladr) {
5542		/*
5543		 * We don't support the reladr bit either.  It can only be
5544		 * used with linked commands, and we don't support linked
5545		 * commands.  Since the bit is in the same place for the
5546		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5547		 * just return the same error in either case.
5548		 */
5549		ctl_set_invalid_field(ctsio,
5550				      /*sks_valid*/ 1,
5551				      /*command*/ 1,
5552				      /*field*/ 1,
5553				      /*bit_valid*/ 1,
5554				      /*bit*/ 0);
5555		ctl_done((union ctl_io *)ctsio);
5556		goto bailout;
5557	}
5558
5559	/*
5560	 * We check the LBA and length, but don't do anything with them.
5561	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5562	 * get flushed.  This check will just help satisfy anyone who wants
5563	 * to see an error for an out of range LBA.
5564	 */
5565	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5566		ctl_set_lba_out_of_range(ctsio);
5567		ctl_done((union ctl_io *)ctsio);
5568		goto bailout;
5569	}
5570
5571	/*
5572	 * If this LUN has no backend, we can't flush the cache anyway.
5573	 */
5574	if (lun->backend == NULL) {
5575		ctl_set_invalid_opcode(ctsio);
5576		ctl_done((union ctl_io *)ctsio);
5577		goto bailout;
5578	}
5579
5580	/*
5581	 * Check to see whether we're configured to send the SYNCHRONIZE
5582	 * CACHE command directly to the back end.
5583	 */
5584	mtx_lock(&ctl_softc->ctl_lock);
5585	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5586	 && (++(lun->sync_count) >= lun->sync_interval)) {
5587		lun->sync_count = 0;
5588		mtx_unlock(&ctl_softc->ctl_lock);
5589		retval = lun->backend->config_write((union ctl_io *)ctsio);
5590	} else {
5591		mtx_unlock(&ctl_softc->ctl_lock);
5592		ctl_set_success(ctsio);
5593		ctl_done((union ctl_io *)ctsio);
5594	}
5595
5596bailout:
5597
5598	return (retval);
5599}
5600
5601int
5602ctl_format(struct ctl_scsiio *ctsio)
5603{
5604	struct scsi_format *cdb;
5605	struct ctl_lun *lun;
5606	struct ctl_softc *ctl_softc;
5607	int length, defect_list_len;
5608
5609	CTL_DEBUG_PRINT(("ctl_format\n"));
5610
5611	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5612	ctl_softc = control_softc;
5613
5614	cdb = (struct scsi_format *)ctsio->cdb;
5615
5616	length = 0;
5617	if (cdb->byte2 & SF_FMTDATA) {
5618		if (cdb->byte2 & SF_LONGLIST)
5619			length = sizeof(struct scsi_format_header_long);
5620		else
5621			length = sizeof(struct scsi_format_header_short);
5622	}
5623
5624	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5625	 && (length > 0)) {
5626		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5627		ctsio->kern_data_len = length;
5628		ctsio->kern_total_len = length;
5629		ctsio->kern_data_resid = 0;
5630		ctsio->kern_rel_offset = 0;
5631		ctsio->kern_sg_entries = 0;
5632		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5633		ctsio->be_move_done = ctl_config_move_done;
5634		ctl_datamove((union ctl_io *)ctsio);
5635
5636		return (CTL_RETVAL_COMPLETE);
5637	}
5638
5639	defect_list_len = 0;
5640
5641	if (cdb->byte2 & SF_FMTDATA) {
5642		if (cdb->byte2 & SF_LONGLIST) {
5643			struct scsi_format_header_long *header;
5644
5645			header = (struct scsi_format_header_long *)
5646				ctsio->kern_data_ptr;
5647
5648			defect_list_len = scsi_4btoul(header->defect_list_len);
5649			if (defect_list_len != 0) {
5650				ctl_set_invalid_field(ctsio,
5651						      /*sks_valid*/ 1,
5652						      /*command*/ 0,
5653						      /*field*/ 2,
5654						      /*bit_valid*/ 0,
5655						      /*bit*/ 0);
5656				goto bailout;
5657			}
5658		} else {
5659			struct scsi_format_header_short *header;
5660
5661			header = (struct scsi_format_header_short *)
5662				ctsio->kern_data_ptr;
5663
5664			defect_list_len = scsi_2btoul(header->defect_list_len);
5665			if (defect_list_len != 0) {
5666				ctl_set_invalid_field(ctsio,
5667						      /*sks_valid*/ 1,
5668						      /*command*/ 0,
5669						      /*field*/ 2,
5670						      /*bit_valid*/ 0,
5671						      /*bit*/ 0);
5672				goto bailout;
5673			}
5674		}
5675	}
5676
5677	/*
5678	 * The format command will clear out the "Medium format corrupted"
5679	 * status if set by the configuration code.  That status is really
5680	 * just a way to notify the host that we have lost the media, and
5681	 * get them to issue a command that will basically make them think
5682	 * they're blowing away the media.
5683	 */
5684	mtx_lock(&ctl_softc->ctl_lock);
5685	lun->flags &= ~CTL_LUN_INOPERABLE;
5686	mtx_unlock(&ctl_softc->ctl_lock);
5687
5688	ctsio->scsi_status = SCSI_STATUS_OK;
5689	ctsio->io_hdr.status = CTL_SUCCESS;
5690bailout:
5691
5692	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5693		free(ctsio->kern_data_ptr, M_CTL);
5694		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5695	}
5696
5697	ctl_done((union ctl_io *)ctsio);
5698	return (CTL_RETVAL_COMPLETE);
5699}
5700
5701int
5702ctl_write_buffer(struct ctl_scsiio *ctsio)
5703{
5704	struct scsi_write_buffer *cdb;
5705	struct copan_page_header *header;
5706	struct ctl_lun *lun;
5707	struct ctl_softc *ctl_softc;
5708	int buffer_offset, len;
5709	int retval;
5710
5711	header = NULL;
5712
5713	retval = CTL_RETVAL_COMPLETE;
5714
5715	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5716
5717	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5718	ctl_softc = control_softc;
5719	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5720
5721	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5722		ctl_set_invalid_field(ctsio,
5723				      /*sks_valid*/ 1,
5724				      /*command*/ 1,
5725				      /*field*/ 1,
5726				      /*bit_valid*/ 1,
5727				      /*bit*/ 4);
5728		ctl_done((union ctl_io *)ctsio);
5729		return (CTL_RETVAL_COMPLETE);
5730	}
5731	if (cdb->buffer_id != 0) {
5732		ctl_set_invalid_field(ctsio,
5733				      /*sks_valid*/ 1,
5734				      /*command*/ 1,
5735				      /*field*/ 2,
5736				      /*bit_valid*/ 0,
5737				      /*bit*/ 0);
5738		ctl_done((union ctl_io *)ctsio);
5739		return (CTL_RETVAL_COMPLETE);
5740	}
5741
5742	len = scsi_3btoul(cdb->length);
5743	buffer_offset = scsi_3btoul(cdb->offset);
5744
5745	if (len > sizeof(lun->write_buffer)) {
5746		ctl_set_invalid_field(ctsio,
5747				      /*sks_valid*/ 1,
5748				      /*command*/ 1,
5749				      /*field*/ 6,
5750				      /*bit_valid*/ 0,
5751				      /*bit*/ 0);
5752		ctl_done((union ctl_io *)ctsio);
5753		return (CTL_RETVAL_COMPLETE);
5754	}
5755
5756	if (buffer_offset != 0) {
5757		ctl_set_invalid_field(ctsio,
5758				      /*sks_valid*/ 1,
5759				      /*command*/ 1,
5760				      /*field*/ 3,
5761				      /*bit_valid*/ 0,
5762				      /*bit*/ 0);
5763		ctl_done((union ctl_io *)ctsio);
5764		return (CTL_RETVAL_COMPLETE);
5765	}
5766
5767	/*
5768	 * If we've got a kernel request that hasn't been malloced yet,
5769	 * malloc it and tell the caller the data buffer is here.
5770	 */
5771	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5772		ctsio->kern_data_ptr = lun->write_buffer;
5773		ctsio->kern_data_len = len;
5774		ctsio->kern_total_len = len;
5775		ctsio->kern_data_resid = 0;
5776		ctsio->kern_rel_offset = 0;
5777		ctsio->kern_sg_entries = 0;
5778		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5779		ctsio->be_move_done = ctl_config_move_done;
5780		ctl_datamove((union ctl_io *)ctsio);
5781
5782		return (CTL_RETVAL_COMPLETE);
5783	}
5784
5785	ctl_done((union ctl_io *)ctsio);
5786
5787	return (CTL_RETVAL_COMPLETE);
5788}
5789
5790/*
5791 * Note that this function currently doesn't actually do anything inside
5792 * CTL to enforce things if the DQue bit is turned on.
5793 *
5794 * Also note that this function can't be used in the default case, because
5795 * the DQue bit isn't set in the changeable mask for the control mode page
5796 * anyway.  This is just here as an example for how to implement a page
5797 * handler, and a placeholder in case we want to allow the user to turn
5798 * tagged queueing on and off.
5799 *
5800 * The D_SENSE bit handling is functional, however, and will turn
5801 * descriptor sense on and off for a given LUN.
5802 */
5803int
5804ctl_control_page_handler(struct ctl_scsiio *ctsio,
5805			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5806{
5807	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5808	struct ctl_lun *lun;
5809	struct ctl_softc *softc;
5810	int set_ua;
5811	uint32_t initidx;
5812
5813	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5814	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5815	set_ua = 0;
5816
5817	user_cp = (struct scsi_control_page *)page_ptr;
5818	current_cp = (struct scsi_control_page *)
5819		(page_index->page_data + (page_index->page_len *
5820		CTL_PAGE_CURRENT));
5821	saved_cp = (struct scsi_control_page *)
5822		(page_index->page_data + (page_index->page_len *
5823		CTL_PAGE_SAVED));
5824
5825	softc = control_softc;
5826
5827	mtx_lock(&softc->ctl_lock);
5828	if (((current_cp->rlec & SCP_DSENSE) == 0)
5829	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5830		/*
5831		 * Descriptor sense is currently turned off and the user
5832		 * wants to turn it on.
5833		 */
5834		current_cp->rlec |= SCP_DSENSE;
5835		saved_cp->rlec |= SCP_DSENSE;
5836		lun->flags |= CTL_LUN_SENSE_DESC;
5837		set_ua = 1;
5838	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5839		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5840		/*
5841		 * Descriptor sense is currently turned on, and the user
5842		 * wants to turn it off.
5843		 */
5844		current_cp->rlec &= ~SCP_DSENSE;
5845		saved_cp->rlec &= ~SCP_DSENSE;
5846		lun->flags &= ~CTL_LUN_SENSE_DESC;
5847		set_ua = 1;
5848	}
5849	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5850		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5851#ifdef NEEDTOPORT
5852			csevent_log(CSC_CTL | CSC_SHELF_SW |
5853				    CTL_UNTAG_TO_UNTAG,
5854				    csevent_LogType_Trace,
5855				    csevent_Severity_Information,
5856				    csevent_AlertLevel_Green,
5857				    csevent_FRU_Firmware,
5858				    csevent_FRU_Unknown,
5859				    "Received untagged to untagged transition");
5860#endif /* NEEDTOPORT */
5861		} else {
5862#ifdef NEEDTOPORT
5863			csevent_log(CSC_CTL | CSC_SHELF_SW |
5864				    CTL_UNTAG_TO_TAG,
5865				    csevent_LogType_ConfigChange,
5866				    csevent_Severity_Information,
5867				    csevent_AlertLevel_Green,
5868				    csevent_FRU_Firmware,
5869				    csevent_FRU_Unknown,
5870				    "Received untagged to tagged "
5871				    "queueing transition");
5872#endif /* NEEDTOPORT */
5873
5874			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5875			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5876			set_ua = 1;
5877		}
5878	} else {
5879		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5880#ifdef NEEDTOPORT
5881			csevent_log(CSC_CTL | CSC_SHELF_SW |
5882				    CTL_TAG_TO_UNTAG,
5883				    csevent_LogType_ConfigChange,
5884				    csevent_Severity_Warning,
5885				    csevent_AlertLevel_Yellow,
5886				    csevent_FRU_Firmware,
5887				    csevent_FRU_Unknown,
5888				    "Received tagged queueing to untagged "
5889				    "transition");
5890#endif /* NEEDTOPORT */
5891
5892			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5893			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5894			set_ua = 1;
5895		} else {
5896#ifdef NEEDTOPORT
5897			csevent_log(CSC_CTL | CSC_SHELF_SW |
5898				    CTL_TAG_TO_TAG,
5899				    csevent_LogType_Trace,
5900				    csevent_Severity_Information,
5901				    csevent_AlertLevel_Green,
5902				    csevent_FRU_Firmware,
5903				    csevent_FRU_Unknown,
5904				    "Received tagged queueing to tagged "
5905				    "queueing transition");
5906#endif /* NEEDTOPORT */
5907		}
5908	}
5909	if (set_ua != 0) {
5910		int i;
5911		/*
5912		 * Let other initiators know that the mode
5913		 * parameters for this LUN have changed.
5914		 */
5915		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5916			if (i == initidx)
5917				continue;
5918
5919			lun->pending_sense[i].ua_pending |=
5920				CTL_UA_MODE_CHANGE;
5921		}
5922	}
5923	mtx_unlock(&softc->ctl_lock);
5924
5925	return (0);
5926}
5927
5928int
5929ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5930		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5931{
5932	return (0);
5933}
5934
5935int
5936ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5937			   struct ctl_page_index *page_index, int pc)
5938{
5939	struct copan_power_subpage *page;
5940
5941	page = (struct copan_power_subpage *)page_index->page_data +
5942		(page_index->page_len * pc);
5943
5944	switch (pc) {
5945	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5946		/*
5947		 * We don't update the changable bits for this page.
5948		 */
5949		break;
5950	case SMS_PAGE_CTRL_CURRENT >> 6:
5951	case SMS_PAGE_CTRL_DEFAULT >> 6:
5952	case SMS_PAGE_CTRL_SAVED >> 6:
5953#ifdef NEEDTOPORT
5954		ctl_update_power_subpage(page);
5955#endif
5956		break;
5957	default:
5958#ifdef NEEDTOPORT
5959		EPRINT(0, "Invalid PC %d!!", pc);
5960#endif
5961		break;
5962	}
5963	return (0);
5964}
5965
5966
5967int
5968ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5969		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5970{
5971	struct copan_aps_subpage *user_sp;
5972	struct copan_aps_subpage *current_sp;
5973	union ctl_modepage_info *modepage_info;
5974	struct ctl_softc *softc;
5975	struct ctl_lun *lun;
5976	int retval;
5977
5978	retval = CTL_RETVAL_COMPLETE;
5979	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5980		     (page_index->page_len * CTL_PAGE_CURRENT));
5981	softc = control_softc;
5982	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5983
5984	user_sp = (struct copan_aps_subpage *)page_ptr;
5985
5986	modepage_info = (union ctl_modepage_info *)
5987		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5988
5989	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5990	modepage_info->header.subpage = page_index->subpage;
5991	modepage_info->aps.lock_active = user_sp->lock_active;
5992
5993	mtx_lock(&softc->ctl_lock);
5994
5995	/*
5996	 * If there is a request to lock the LUN and another LUN is locked
5997	 * this is an error. If the requested LUN is already locked ignore
5998	 * the request. If no LUN is locked attempt to lock it.
5999	 * if there is a request to unlock the LUN and the LUN is currently
6000	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6001	 * if another LUN is locked or no LUN is locked.
6002	 */
6003	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6004		if (softc->aps_locked_lun == lun->lun) {
6005			/*
6006			 * This LUN is already locked, so we're done.
6007			 */
6008			retval = CTL_RETVAL_COMPLETE;
6009		} else if (softc->aps_locked_lun == 0) {
6010			/*
6011			 * No one has the lock, pass the request to the
6012			 * backend.
6013			 */
6014			retval = lun->backend->config_write(
6015				(union ctl_io *)ctsio);
6016		} else {
6017			/*
6018			 * Someone else has the lock, throw out the request.
6019			 */
6020			ctl_set_already_locked(ctsio);
6021			free(ctsio->kern_data_ptr, M_CTL);
6022			ctl_done((union ctl_io *)ctsio);
6023
6024			/*
6025			 * Set the return value so that ctl_do_mode_select()
6026			 * won't try to complete the command.  We already
6027			 * completed it here.
6028			 */
6029			retval = CTL_RETVAL_ERROR;
6030		}
6031	} else if (softc->aps_locked_lun == lun->lun) {
6032		/*
6033		 * This LUN is locked, so pass the unlock request to the
6034		 * backend.
6035		 */
6036		retval = lun->backend->config_write((union ctl_io *)ctsio);
6037	}
6038	mtx_unlock(&softc->ctl_lock);
6039
6040	return (retval);
6041}
6042
6043int
6044ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6045				struct ctl_page_index *page_index,
6046				uint8_t *page_ptr)
6047{
6048	uint8_t *c;
6049	int i;
6050
6051	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6052	ctl_time_io_secs =
6053		(c[0] << 8) |
6054		(c[1] << 0) |
6055		0;
6056	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6057	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6058	printf("page data:");
6059	for (i=0; i<8; i++)
6060		printf(" %.2x",page_ptr[i]);
6061	printf("\n");
6062	return (0);
6063}
6064
6065int
6066ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6067			       struct ctl_page_index *page_index,
6068			       int pc)
6069{
6070	struct copan_debugconf_subpage *page;
6071
6072	page = (struct copan_debugconf_subpage *)page_index->page_data +
6073		(page_index->page_len * pc);
6074
6075	switch (pc) {
6076	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6077	case SMS_PAGE_CTRL_DEFAULT >> 6:
6078	case SMS_PAGE_CTRL_SAVED >> 6:
6079		/*
6080		 * We don't update the changable or default bits for this page.
6081		 */
6082		break;
6083	case SMS_PAGE_CTRL_CURRENT >> 6:
6084		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6085		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6086		break;
6087	default:
6088#ifdef NEEDTOPORT
6089		EPRINT(0, "Invalid PC %d!!", pc);
6090#endif /* NEEDTOPORT */
6091		break;
6092	}
6093	return (0);
6094}
6095
6096
6097static int
6098ctl_do_mode_select(union ctl_io *io)
6099{
6100	struct scsi_mode_page_header *page_header;
6101	struct ctl_page_index *page_index;
6102	struct ctl_scsiio *ctsio;
6103	int control_dev, page_len;
6104	int page_len_offset, page_len_size;
6105	union ctl_modepage_info *modepage_info;
6106	struct ctl_lun *lun;
6107	int *len_left, *len_used;
6108	int retval, i;
6109
6110	ctsio = &io->scsiio;
6111	page_index = NULL;
6112	page_len = 0;
6113	retval = CTL_RETVAL_COMPLETE;
6114
6115	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6116
6117	if (lun->be_lun->lun_type != T_DIRECT)
6118		control_dev = 1;
6119	else
6120		control_dev = 0;
6121
6122	modepage_info = (union ctl_modepage_info *)
6123		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6124	len_left = &modepage_info->header.len_left;
6125	len_used = &modepage_info->header.len_used;
6126
6127do_next_page:
6128
6129	page_header = (struct scsi_mode_page_header *)
6130		(ctsio->kern_data_ptr + *len_used);
6131
6132	if (*len_left == 0) {
6133		free(ctsio->kern_data_ptr, M_CTL);
6134		ctl_set_success(ctsio);
6135		ctl_done((union ctl_io *)ctsio);
6136		return (CTL_RETVAL_COMPLETE);
6137	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6138
6139		free(ctsio->kern_data_ptr, M_CTL);
6140		ctl_set_param_len_error(ctsio);
6141		ctl_done((union ctl_io *)ctsio);
6142		return (CTL_RETVAL_COMPLETE);
6143
6144	} else if ((page_header->page_code & SMPH_SPF)
6145		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6146
6147		free(ctsio->kern_data_ptr, M_CTL);
6148		ctl_set_param_len_error(ctsio);
6149		ctl_done((union ctl_io *)ctsio);
6150		return (CTL_RETVAL_COMPLETE);
6151	}
6152
6153
6154	/*
6155	 * XXX KDM should we do something with the block descriptor?
6156	 */
6157	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6158
6159		if ((control_dev != 0)
6160		 && (lun->mode_pages.index[i].page_flags &
6161		     CTL_PAGE_FLAG_DISK_ONLY))
6162			continue;
6163
6164		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6165		    (page_header->page_code & SMPH_PC_MASK))
6166			continue;
6167
6168		/*
6169		 * If neither page has a subpage code, then we've got a
6170		 * match.
6171		 */
6172		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6173		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6174			page_index = &lun->mode_pages.index[i];
6175			page_len = page_header->page_length;
6176			break;
6177		}
6178
6179		/*
6180		 * If both pages have subpages, then the subpage numbers
6181		 * have to match.
6182		 */
6183		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6184		  && (page_header->page_code & SMPH_SPF)) {
6185			struct scsi_mode_page_header_sp *sph;
6186
6187			sph = (struct scsi_mode_page_header_sp *)page_header;
6188
6189			if (lun->mode_pages.index[i].subpage ==
6190			    sph->subpage) {
6191				page_index = &lun->mode_pages.index[i];
6192				page_len = scsi_2btoul(sph->page_length);
6193				break;
6194			}
6195		}
6196	}
6197
6198	/*
6199	 * If we couldn't find the page, or if we don't have a mode select
6200	 * handler for it, send back an error to the user.
6201	 */
6202	if ((page_index == NULL)
6203	 || (page_index->select_handler == NULL)) {
6204		ctl_set_invalid_field(ctsio,
6205				      /*sks_valid*/ 1,
6206				      /*command*/ 0,
6207				      /*field*/ *len_used,
6208				      /*bit_valid*/ 0,
6209				      /*bit*/ 0);
6210		free(ctsio->kern_data_ptr, M_CTL);
6211		ctl_done((union ctl_io *)ctsio);
6212		return (CTL_RETVAL_COMPLETE);
6213	}
6214
6215	if (page_index->page_code & SMPH_SPF) {
6216		page_len_offset = 2;
6217		page_len_size = 2;
6218	} else {
6219		page_len_size = 1;
6220		page_len_offset = 1;
6221	}
6222
6223	/*
6224	 * If the length the initiator gives us isn't the one we specify in
6225	 * the mode page header, or if they didn't specify enough data in
6226	 * the CDB to avoid truncating this page, kick out the request.
6227	 */
6228	if ((page_len != (page_index->page_len - page_len_offset -
6229			  page_len_size))
6230	 || (*len_left < page_index->page_len)) {
6231
6232
6233		ctl_set_invalid_field(ctsio,
6234				      /*sks_valid*/ 1,
6235				      /*command*/ 0,
6236				      /*field*/ *len_used + page_len_offset,
6237				      /*bit_valid*/ 0,
6238				      /*bit*/ 0);
6239		free(ctsio->kern_data_ptr, M_CTL);
6240		ctl_done((union ctl_io *)ctsio);
6241		return (CTL_RETVAL_COMPLETE);
6242	}
6243
6244	/*
6245	 * Run through the mode page, checking to make sure that the bits
6246	 * the user changed are actually legal for him to change.
6247	 */
6248	for (i = 0; i < page_index->page_len; i++) {
6249		uint8_t *user_byte, *change_mask, *current_byte;
6250		int bad_bit;
6251		int j;
6252
6253		user_byte = (uint8_t *)page_header + i;
6254		change_mask = page_index->page_data +
6255			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6256		current_byte = page_index->page_data +
6257			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6258
6259		/*
6260		 * Check to see whether the user set any bits in this byte
6261		 * that he is not allowed to set.
6262		 */
6263		if ((*user_byte & ~(*change_mask)) ==
6264		    (*current_byte & ~(*change_mask)))
6265			continue;
6266
6267		/*
6268		 * Go through bit by bit to determine which one is illegal.
6269		 */
6270		bad_bit = 0;
6271		for (j = 7; j >= 0; j--) {
6272			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6273			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6274				bad_bit = i;
6275				break;
6276			}
6277		}
6278		ctl_set_invalid_field(ctsio,
6279				      /*sks_valid*/ 1,
6280				      /*command*/ 0,
6281				      /*field*/ *len_used + i,
6282				      /*bit_valid*/ 1,
6283				      /*bit*/ bad_bit);
6284		free(ctsio->kern_data_ptr, M_CTL);
6285		ctl_done((union ctl_io *)ctsio);
6286		return (CTL_RETVAL_COMPLETE);
6287	}
6288
6289	/*
6290	 * Decrement these before we call the page handler, since we may
6291	 * end up getting called back one way or another before the handler
6292	 * returns to this context.
6293	 */
6294	*len_left -= page_index->page_len;
6295	*len_used += page_index->page_len;
6296
6297	retval = page_index->select_handler(ctsio, page_index,
6298					    (uint8_t *)page_header);
6299
6300	/*
6301	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6302	 * wait until this queued command completes to finish processing
6303	 * the mode page.  If it returns anything other than
6304	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6305	 * already set the sense information, freed the data pointer, and
6306	 * completed the io for us.
6307	 */
6308	if (retval != CTL_RETVAL_COMPLETE)
6309		goto bailout_no_done;
6310
6311	/*
6312	 * If the initiator sent us more than one page, parse the next one.
6313	 */
6314	if (*len_left > 0)
6315		goto do_next_page;
6316
6317	ctl_set_success(ctsio);
6318	free(ctsio->kern_data_ptr, M_CTL);
6319	ctl_done((union ctl_io *)ctsio);
6320
6321bailout_no_done:
6322
6323	return (CTL_RETVAL_COMPLETE);
6324
6325}
6326
6327int
6328ctl_mode_select(struct ctl_scsiio *ctsio)
6329{
6330	int param_len, pf, sp;
6331	int header_size, bd_len;
6332	int len_left, len_used;
6333	struct ctl_page_index *page_index;
6334	struct ctl_lun *lun;
6335	int control_dev, page_len;
6336	union ctl_modepage_info *modepage_info;
6337	int retval;
6338
6339	pf = 0;
6340	sp = 0;
6341	page_len = 0;
6342	len_used = 0;
6343	len_left = 0;
6344	retval = 0;
6345	bd_len = 0;
6346	page_index = NULL;
6347
6348	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6349
6350	if (lun->be_lun->lun_type != T_DIRECT)
6351		control_dev = 1;
6352	else
6353		control_dev = 0;
6354
6355	switch (ctsio->cdb[0]) {
6356	case MODE_SELECT_6: {
6357		struct scsi_mode_select_6 *cdb;
6358
6359		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6360
6361		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6362		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6363
6364		param_len = cdb->length;
6365		header_size = sizeof(struct scsi_mode_header_6);
6366		break;
6367	}
6368	case MODE_SELECT_10: {
6369		struct scsi_mode_select_10 *cdb;
6370
6371		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6372
6373		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6374		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6375
6376		param_len = scsi_2btoul(cdb->length);
6377		header_size = sizeof(struct scsi_mode_header_10);
6378		break;
6379	}
6380	default:
6381		ctl_set_invalid_opcode(ctsio);
6382		ctl_done((union ctl_io *)ctsio);
6383		return (CTL_RETVAL_COMPLETE);
6384		break; /* NOTREACHED */
6385	}
6386
6387	/*
6388	 * From SPC-3:
6389	 * "A parameter list length of zero indicates that the Data-Out Buffer
6390	 * shall be empty. This condition shall not be considered as an error."
6391	 */
6392	if (param_len == 0) {
6393		ctl_set_success(ctsio);
6394		ctl_done((union ctl_io *)ctsio);
6395		return (CTL_RETVAL_COMPLETE);
6396	}
6397
6398	/*
6399	 * Since we'll hit this the first time through, prior to
6400	 * allocation, we don't need to free a data buffer here.
6401	 */
6402	if (param_len < header_size) {
6403		ctl_set_param_len_error(ctsio);
6404		ctl_done((union ctl_io *)ctsio);
6405		return (CTL_RETVAL_COMPLETE);
6406	}
6407
6408	/*
6409	 * Allocate the data buffer and grab the user's data.  In theory,
6410	 * we shouldn't have to sanity check the parameter list length here
6411	 * because the maximum size is 64K.  We should be able to malloc
6412	 * that much without too many problems.
6413	 */
6414	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6415		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6416		ctsio->kern_data_len = param_len;
6417		ctsio->kern_total_len = param_len;
6418		ctsio->kern_data_resid = 0;
6419		ctsio->kern_rel_offset = 0;
6420		ctsio->kern_sg_entries = 0;
6421		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6422		ctsio->be_move_done = ctl_config_move_done;
6423		ctl_datamove((union ctl_io *)ctsio);
6424
6425		return (CTL_RETVAL_COMPLETE);
6426	}
6427
6428	switch (ctsio->cdb[0]) {
6429	case MODE_SELECT_6: {
6430		struct scsi_mode_header_6 *mh6;
6431
6432		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6433		bd_len = mh6->blk_desc_len;
6434		break;
6435	}
6436	case MODE_SELECT_10: {
6437		struct scsi_mode_header_10 *mh10;
6438
6439		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6440		bd_len = scsi_2btoul(mh10->blk_desc_len);
6441		break;
6442	}
6443	default:
6444		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6445		break;
6446	}
6447
6448	if (param_len < (header_size + bd_len)) {
6449		free(ctsio->kern_data_ptr, M_CTL);
6450		ctl_set_param_len_error(ctsio);
6451		ctl_done((union ctl_io *)ctsio);
6452		return (CTL_RETVAL_COMPLETE);
6453	}
6454
6455	/*
6456	 * Set the IO_CONT flag, so that if this I/O gets passed to
6457	 * ctl_config_write_done(), it'll get passed back to
6458	 * ctl_do_mode_select() for further processing, or completion if
6459	 * we're all done.
6460	 */
6461	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6462	ctsio->io_cont = ctl_do_mode_select;
6463
6464	modepage_info = (union ctl_modepage_info *)
6465		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6466
6467	memset(modepage_info, 0, sizeof(*modepage_info));
6468
6469	len_left = param_len - header_size - bd_len;
6470	len_used = header_size + bd_len;
6471
6472	modepage_info->header.len_left = len_left;
6473	modepage_info->header.len_used = len_used;
6474
6475	return (ctl_do_mode_select((union ctl_io *)ctsio));
6476}
6477
6478int
6479ctl_mode_sense(struct ctl_scsiio *ctsio)
6480{
6481	struct ctl_lun *lun;
6482	int pc, page_code, dbd, llba, subpage;
6483	int alloc_len, page_len, header_len, total_len;
6484	struct scsi_mode_block_descr *block_desc;
6485	struct ctl_page_index *page_index;
6486	int control_dev;
6487
6488	dbd = 0;
6489	llba = 0;
6490	block_desc = NULL;
6491	page_index = NULL;
6492
6493	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6494
6495	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6496
6497	if (lun->be_lun->lun_type != T_DIRECT)
6498		control_dev = 1;
6499	else
6500		control_dev = 0;
6501
6502	switch (ctsio->cdb[0]) {
6503	case MODE_SENSE_6: {
6504		struct scsi_mode_sense_6 *cdb;
6505
6506		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6507
6508		header_len = sizeof(struct scsi_mode_hdr_6);
6509		if (cdb->byte2 & SMS_DBD)
6510			dbd = 1;
6511		else
6512			header_len += sizeof(struct scsi_mode_block_descr);
6513
6514		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6515		page_code = cdb->page & SMS_PAGE_CODE;
6516		subpage = cdb->subpage;
6517		alloc_len = cdb->length;
6518		break;
6519	}
6520	case MODE_SENSE_10: {
6521		struct scsi_mode_sense_10 *cdb;
6522
6523		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6524
6525		header_len = sizeof(struct scsi_mode_hdr_10);
6526
6527		if (cdb->byte2 & SMS_DBD)
6528			dbd = 1;
6529		else
6530			header_len += sizeof(struct scsi_mode_block_descr);
6531		if (cdb->byte2 & SMS10_LLBAA)
6532			llba = 1;
6533		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6534		page_code = cdb->page & SMS_PAGE_CODE;
6535		subpage = cdb->subpage;
6536		alloc_len = scsi_2btoul(cdb->length);
6537		break;
6538	}
6539	default:
6540		ctl_set_invalid_opcode(ctsio);
6541		ctl_done((union ctl_io *)ctsio);
6542		return (CTL_RETVAL_COMPLETE);
6543		break; /* NOTREACHED */
6544	}
6545
6546	/*
6547	 * We have to make a first pass through to calculate the size of
6548	 * the pages that match the user's query.  Then we allocate enough
6549	 * memory to hold it, and actually copy the data into the buffer.
6550	 */
6551	switch (page_code) {
6552	case SMS_ALL_PAGES_PAGE: {
6553		int i;
6554
6555		page_len = 0;
6556
6557		/*
6558		 * At the moment, values other than 0 and 0xff here are
6559		 * reserved according to SPC-3.
6560		 */
6561		if ((subpage != SMS_SUBPAGE_PAGE_0)
6562		 && (subpage != SMS_SUBPAGE_ALL)) {
6563			ctl_set_invalid_field(ctsio,
6564					      /*sks_valid*/ 1,
6565					      /*command*/ 1,
6566					      /*field*/ 3,
6567					      /*bit_valid*/ 0,
6568					      /*bit*/ 0);
6569			ctl_done((union ctl_io *)ctsio);
6570			return (CTL_RETVAL_COMPLETE);
6571		}
6572
6573		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6574			if ((control_dev != 0)
6575			 && (lun->mode_pages.index[i].page_flags &
6576			     CTL_PAGE_FLAG_DISK_ONLY))
6577				continue;
6578
6579			/*
6580			 * We don't use this subpage if the user didn't
6581			 * request all subpages.
6582			 */
6583			if ((lun->mode_pages.index[i].subpage != 0)
6584			 && (subpage == SMS_SUBPAGE_PAGE_0))
6585				continue;
6586
6587#if 0
6588			printf("found page %#x len %d\n",
6589			       lun->mode_pages.index[i].page_code &
6590			       SMPH_PC_MASK,
6591			       lun->mode_pages.index[i].page_len);
6592#endif
6593			page_len += lun->mode_pages.index[i].page_len;
6594		}
6595		break;
6596	}
6597	default: {
6598		int i;
6599
6600		page_len = 0;
6601
6602		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6603			/* Look for the right page code */
6604			if ((lun->mode_pages.index[i].page_code &
6605			     SMPH_PC_MASK) != page_code)
6606				continue;
6607
6608			/* Look for the right subpage or the subpage wildcard*/
6609			if ((lun->mode_pages.index[i].subpage != subpage)
6610			 && (subpage != SMS_SUBPAGE_ALL))
6611				continue;
6612
6613			/* Make sure the page is supported for this dev type */
6614			if ((control_dev != 0)
6615			 && (lun->mode_pages.index[i].page_flags &
6616			     CTL_PAGE_FLAG_DISK_ONLY))
6617				continue;
6618
6619#if 0
6620			printf("found page %#x len %d\n",
6621			       lun->mode_pages.index[i].page_code &
6622			       SMPH_PC_MASK,
6623			       lun->mode_pages.index[i].page_len);
6624#endif
6625
6626			page_len += lun->mode_pages.index[i].page_len;
6627		}
6628
6629		if (page_len == 0) {
6630			ctl_set_invalid_field(ctsio,
6631					      /*sks_valid*/ 1,
6632					      /*command*/ 1,
6633					      /*field*/ 2,
6634					      /*bit_valid*/ 1,
6635					      /*bit*/ 5);
6636			ctl_done((union ctl_io *)ctsio);
6637			return (CTL_RETVAL_COMPLETE);
6638		}
6639		break;
6640	}
6641	}
6642
6643	total_len = header_len + page_len;
6644#if 0
6645	printf("header_len = %d, page_len = %d, total_len = %d\n",
6646	       header_len, page_len, total_len);
6647#endif
6648
6649	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6650	ctsio->kern_sg_entries = 0;
6651	ctsio->kern_data_resid = 0;
6652	ctsio->kern_rel_offset = 0;
6653	if (total_len < alloc_len) {
6654		ctsio->residual = alloc_len - total_len;
6655		ctsio->kern_data_len = total_len;
6656		ctsio->kern_total_len = total_len;
6657	} else {
6658		ctsio->residual = 0;
6659		ctsio->kern_data_len = alloc_len;
6660		ctsio->kern_total_len = alloc_len;
6661	}
6662
6663	switch (ctsio->cdb[0]) {
6664	case MODE_SENSE_6: {
6665		struct scsi_mode_hdr_6 *header;
6666
6667		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6668
6669		header->datalen = ctl_min(total_len - 1, 254);
6670
6671		if (dbd)
6672			header->block_descr_len = 0;
6673		else
6674			header->block_descr_len =
6675				sizeof(struct scsi_mode_block_descr);
6676		block_desc = (struct scsi_mode_block_descr *)&header[1];
6677		break;
6678	}
6679	case MODE_SENSE_10: {
6680		struct scsi_mode_hdr_10 *header;
6681		int datalen;
6682
6683		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6684
6685		datalen = ctl_min(total_len - 2, 65533);
6686		scsi_ulto2b(datalen, header->datalen);
6687		if (dbd)
6688			scsi_ulto2b(0, header->block_descr_len);
6689		else
6690			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6691				    header->block_descr_len);
6692		block_desc = (struct scsi_mode_block_descr *)&header[1];
6693		break;
6694	}
6695	default:
6696		panic("invalid CDB type %#x", ctsio->cdb[0]);
6697		break; /* NOTREACHED */
6698	}
6699
6700	/*
6701	 * If we've got a disk, use its blocksize in the block
6702	 * descriptor.  Otherwise, just set it to 0.
6703	 */
6704	if (dbd == 0) {
6705		if (control_dev != 0)
6706			scsi_ulto3b(lun->be_lun->blocksize,
6707				    block_desc->block_len);
6708		else
6709			scsi_ulto3b(0, block_desc->block_len);
6710	}
6711
6712	switch (page_code) {
6713	case SMS_ALL_PAGES_PAGE: {
6714		int i, data_used;
6715
6716		data_used = header_len;
6717		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6718			struct ctl_page_index *page_index;
6719
6720			page_index = &lun->mode_pages.index[i];
6721
6722			if ((control_dev != 0)
6723			 && (page_index->page_flags &
6724			    CTL_PAGE_FLAG_DISK_ONLY))
6725				continue;
6726
6727			/*
6728			 * We don't use this subpage if the user didn't
6729			 * request all subpages.  We already checked (above)
6730			 * to make sure the user only specified a subpage
6731			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6732			 */
6733			if ((page_index->subpage != 0)
6734			 && (subpage == SMS_SUBPAGE_PAGE_0))
6735				continue;
6736
6737			/*
6738			 * Call the handler, if it exists, to update the
6739			 * page to the latest values.
6740			 */
6741			if (page_index->sense_handler != NULL)
6742				page_index->sense_handler(ctsio, page_index,pc);
6743
6744			memcpy(ctsio->kern_data_ptr + data_used,
6745			       page_index->page_data +
6746			       (page_index->page_len * pc),
6747			       page_index->page_len);
6748			data_used += page_index->page_len;
6749		}
6750		break;
6751	}
6752	default: {
6753		int i, data_used;
6754
6755		data_used = header_len;
6756
6757		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6758			struct ctl_page_index *page_index;
6759
6760			page_index = &lun->mode_pages.index[i];
6761
6762			/* Look for the right page code */
6763			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6764				continue;
6765
6766			/* Look for the right subpage or the subpage wildcard*/
6767			if ((page_index->subpage != subpage)
6768			 && (subpage != SMS_SUBPAGE_ALL))
6769				continue;
6770
6771			/* Make sure the page is supported for this dev type */
6772			if ((control_dev != 0)
6773			 && (page_index->page_flags &
6774			     CTL_PAGE_FLAG_DISK_ONLY))
6775				continue;
6776
6777			/*
6778			 * Call the handler, if it exists, to update the
6779			 * page to the latest values.
6780			 */
6781			if (page_index->sense_handler != NULL)
6782				page_index->sense_handler(ctsio, page_index,pc);
6783
6784			memcpy(ctsio->kern_data_ptr + data_used,
6785			       page_index->page_data +
6786			       (page_index->page_len * pc),
6787			       page_index->page_len);
6788			data_used += page_index->page_len;
6789		}
6790		break;
6791	}
6792	}
6793
6794	ctsio->scsi_status = SCSI_STATUS_OK;
6795
6796	ctsio->be_move_done = ctl_config_move_done;
6797	ctl_datamove((union ctl_io *)ctsio);
6798
6799	return (CTL_RETVAL_COMPLETE);
6800}
6801
6802int
6803ctl_read_capacity(struct ctl_scsiio *ctsio)
6804{
6805	struct scsi_read_capacity *cdb;
6806	struct scsi_read_capacity_data *data;
6807	struct ctl_lun *lun;
6808	uint32_t lba;
6809
6810	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6811
6812	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6813
6814	lba = scsi_4btoul(cdb->addr);
6815	if (((cdb->pmi & SRC_PMI) == 0)
6816	 && (lba != 0)) {
6817		ctl_set_invalid_field(/*ctsio*/ ctsio,
6818				      /*sks_valid*/ 1,
6819				      /*command*/ 1,
6820				      /*field*/ 2,
6821				      /*bit_valid*/ 0,
6822				      /*bit*/ 0);
6823		ctl_done((union ctl_io *)ctsio);
6824		return (CTL_RETVAL_COMPLETE);
6825	}
6826
6827	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6828
6829	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6830	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6831	ctsio->residual = 0;
6832	ctsio->kern_data_len = sizeof(*data);
6833	ctsio->kern_total_len = sizeof(*data);
6834	ctsio->kern_data_resid = 0;
6835	ctsio->kern_rel_offset = 0;
6836	ctsio->kern_sg_entries = 0;
6837
6838	/*
6839	 * If the maximum LBA is greater than 0xfffffffe, the user must
6840	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6841	 * serivce action set.
6842	 */
6843	if (lun->be_lun->maxlba > 0xfffffffe)
6844		scsi_ulto4b(0xffffffff, data->addr);
6845	else
6846		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6847
6848	/*
6849	 * XXX KDM this may not be 512 bytes...
6850	 */
6851	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6852
6853	ctsio->scsi_status = SCSI_STATUS_OK;
6854
6855	ctsio->be_move_done = ctl_config_move_done;
6856	ctl_datamove((union ctl_io *)ctsio);
6857
6858	return (CTL_RETVAL_COMPLETE);
6859}
6860
6861static int
6862ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6863{
6864	struct scsi_read_capacity_16 *cdb;
6865	struct scsi_read_capacity_data_long *data;
6866	struct ctl_lun *lun;
6867	uint64_t lba;
6868	uint32_t alloc_len;
6869
6870	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6871
6872	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6873
6874	alloc_len = scsi_4btoul(cdb->alloc_len);
6875	lba = scsi_8btou64(cdb->addr);
6876
6877	if ((cdb->reladr & SRC16_PMI)
6878	 && (lba != 0)) {
6879		ctl_set_invalid_field(/*ctsio*/ ctsio,
6880				      /*sks_valid*/ 1,
6881				      /*command*/ 1,
6882				      /*field*/ 2,
6883				      /*bit_valid*/ 0,
6884				      /*bit*/ 0);
6885		ctl_done((union ctl_io *)ctsio);
6886		return (CTL_RETVAL_COMPLETE);
6887	}
6888
6889	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6890
6891	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6892	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6893
6894	if (sizeof(*data) < alloc_len) {
6895		ctsio->residual = alloc_len - sizeof(*data);
6896		ctsio->kern_data_len = sizeof(*data);
6897		ctsio->kern_total_len = sizeof(*data);
6898	} else {
6899		ctsio->residual = 0;
6900		ctsio->kern_data_len = alloc_len;
6901		ctsio->kern_total_len = alloc_len;
6902	}
6903	ctsio->kern_data_resid = 0;
6904	ctsio->kern_rel_offset = 0;
6905	ctsio->kern_sg_entries = 0;
6906
6907	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6908	/* XXX KDM this may not be 512 bytes... */
6909	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6910	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6911	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6912
6913	ctsio->scsi_status = SCSI_STATUS_OK;
6914
6915	ctsio->be_move_done = ctl_config_move_done;
6916	ctl_datamove((union ctl_io *)ctsio);
6917
6918	return (CTL_RETVAL_COMPLETE);
6919}
6920
6921int
6922ctl_service_action_in(struct ctl_scsiio *ctsio)
6923{
6924	struct scsi_service_action_in *cdb;
6925	int retval;
6926
6927	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
6928
6929	cdb = (struct scsi_service_action_in *)ctsio->cdb;
6930
6931	retval = CTL_RETVAL_COMPLETE;
6932
6933	switch (cdb->service_action) {
6934	case SRC16_SERVICE_ACTION:
6935		retval = ctl_read_capacity_16(ctsio);
6936		break;
6937	default:
6938		ctl_set_invalid_field(/*ctsio*/ ctsio,
6939				      /*sks_valid*/ 1,
6940				      /*command*/ 1,
6941				      /*field*/ 1,
6942				      /*bit_valid*/ 1,
6943				      /*bit*/ 4);
6944		ctl_done((union ctl_io *)ctsio);
6945		break;
6946	}
6947
6948	return (retval);
6949}
6950
6951int
6952ctl_maintenance_in(struct ctl_scsiio *ctsio)
6953{
6954	struct scsi_maintenance_in *cdb;
6955	int retval;
6956	int alloc_len, total_len = 0;
6957	int num_target_port_groups, single;
6958	struct ctl_lun *lun;
6959	struct ctl_softc *softc;
6960	struct scsi_target_group_data *rtg_ptr;
6961	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6962	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6963	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6964
6965	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
6966
6967	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6968	softc = control_softc;
6969	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6970
6971	retval = CTL_RETVAL_COMPLETE;
6972
6973	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
6974		ctl_set_invalid_field(/*ctsio*/ ctsio,
6975				      /*sks_valid*/ 1,
6976				      /*command*/ 1,
6977				      /*field*/ 1,
6978				      /*bit_valid*/ 1,
6979				      /*bit*/ 4);
6980		ctl_done((union ctl_io *)ctsio);
6981		return(retval);
6982	}
6983
6984	mtx_lock(&softc->ctl_lock);
6985	single = ctl_is_single;
6986	mtx_unlock(&softc->ctl_lock);
6987
6988	if (single)
6989        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6990	else
6991        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6992
6993	total_len = sizeof(struct scsi_target_group_data) +
6994		sizeof(struct scsi_target_port_group_descriptor) *
6995		num_target_port_groups +
6996		sizeof(struct scsi_target_port_descriptor) *
6997		NUM_PORTS_PER_GRP * num_target_port_groups;
6998
6999	alloc_len = scsi_4btoul(cdb->length);
7000
7001	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7002
7003	ctsio->kern_sg_entries = 0;
7004
7005	if (total_len < alloc_len) {
7006		ctsio->residual = alloc_len - total_len;
7007		ctsio->kern_data_len = total_len;
7008		ctsio->kern_total_len = total_len;
7009	} else {
7010		ctsio->residual = 0;
7011		ctsio->kern_data_len = alloc_len;
7012		ctsio->kern_total_len = alloc_len;
7013	}
7014	ctsio->kern_data_resid = 0;
7015	ctsio->kern_rel_offset = 0;
7016
7017	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7018
7019	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7020	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7021	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7022	        &tp_desc_ptr1_1->desc_list[0];
7023
7024	if (single == 0) {
7025		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7026	                &tp_desc_ptr1_2->desc_list[0];
7027		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7028		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7029	        	&tp_desc_ptr2_1->desc_list[0];
7030        } else {
7031		tpg_desc_ptr2 = NULL;
7032		tp_desc_ptr2_1 = NULL;
7033		tp_desc_ptr2_2 = NULL;
7034	}
7035
7036	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7037	if (single == 0) {
7038        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7039			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7040				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7041				tpg_desc_ptr2->pref_state =
7042					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7043			} else {
7044				tpg_desc_ptr1->pref_state =
7045					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7046				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7047			}
7048		} else {
7049			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7050				tpg_desc_ptr1->pref_state =
7051					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7052				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7053			} else {
7054				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7055				tpg_desc_ptr2->pref_state =
7056					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7057			}
7058		}
7059	} else {
7060		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7061	}
7062	tpg_desc_ptr1->support = 0;
7063	tpg_desc_ptr1->target_port_group[1] = 1;
7064	tpg_desc_ptr1->status = TPG_IMPLICIT;
7065	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7066
7067	if (single == 0) {
7068		tpg_desc_ptr2->support = 0;
7069		tpg_desc_ptr2->target_port_group[1] = 2;
7070		tpg_desc_ptr2->status = TPG_IMPLICIT;
7071		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7072
7073		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7074		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7075
7076		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7077		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7078	} else {
7079        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7080			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7081			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7082		} else {
7083			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7084			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7085		}
7086	}
7087
7088	ctsio->be_move_done = ctl_config_move_done;
7089
7090	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7091			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7092			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7093			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7094			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7095
7096	ctl_datamove((union ctl_io *)ctsio);
7097	return(retval);
7098}
7099
7100int
7101ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7102{
7103	struct scsi_per_res_in *cdb;
7104	int alloc_len, total_len = 0;
7105	/* struct scsi_per_res_in_rsrv in_data; */
7106	struct ctl_lun *lun;
7107	struct ctl_softc *softc;
7108
7109	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7110
7111	softc = control_softc;
7112
7113	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7114
7115	alloc_len = scsi_2btoul(cdb->length);
7116
7117	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7118
7119retry:
7120	mtx_lock(&softc->ctl_lock);
7121	switch (cdb->action) {
7122	case SPRI_RK: /* read keys */
7123		total_len = sizeof(struct scsi_per_res_in_keys) +
7124			lun->pr_key_count *
7125			sizeof(struct scsi_per_res_key);
7126		break;
7127	case SPRI_RR: /* read reservation */
7128		if (lun->flags & CTL_LUN_PR_RESERVED)
7129			total_len = sizeof(struct scsi_per_res_in_rsrv);
7130		else
7131			total_len = sizeof(struct scsi_per_res_in_header);
7132		break;
7133	case SPRI_RC: /* report capabilities */
7134		total_len = sizeof(struct scsi_per_res_cap);
7135		break;
7136	case SPRI_RS: /* read full status */
7137	default:
7138		mtx_unlock(&softc->ctl_lock);
7139		ctl_set_invalid_field(ctsio,
7140				      /*sks_valid*/ 1,
7141				      /*command*/ 1,
7142				      /*field*/ 1,
7143				      /*bit_valid*/ 1,
7144				      /*bit*/ 0);
7145		ctl_done((union ctl_io *)ctsio);
7146		return (CTL_RETVAL_COMPLETE);
7147		break; /* NOTREACHED */
7148	}
7149	mtx_unlock(&softc->ctl_lock);
7150
7151	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7152
7153	if (total_len < alloc_len) {
7154		ctsio->residual = alloc_len - total_len;
7155		ctsio->kern_data_len = total_len;
7156		ctsio->kern_total_len = total_len;
7157	} else {
7158		ctsio->residual = 0;
7159		ctsio->kern_data_len = alloc_len;
7160		ctsio->kern_total_len = alloc_len;
7161	}
7162
7163	ctsio->kern_data_resid = 0;
7164	ctsio->kern_rel_offset = 0;
7165	ctsio->kern_sg_entries = 0;
7166
7167	mtx_lock(&softc->ctl_lock);
7168	switch (cdb->action) {
7169	case SPRI_RK: { // read keys
7170        struct scsi_per_res_in_keys *res_keys;
7171		int i, key_count;
7172
7173		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7174
7175		/*
7176		 * We had to drop the lock to allocate our buffer, which
7177		 * leaves time for someone to come in with another
7178		 * persistent reservation.  (That is unlikely, though,
7179		 * since this should be the only persistent reservation
7180		 * command active right now.)
7181		 */
7182		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7183		    (lun->pr_key_count *
7184		     sizeof(struct scsi_per_res_key)))){
7185			mtx_unlock(&softc->ctl_lock);
7186			free(ctsio->kern_data_ptr, M_CTL);
7187			printf("%s: reservation length changed, retrying\n",
7188			       __func__);
7189			goto retry;
7190		}
7191
7192		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7193
7194		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7195			     lun->pr_key_count, res_keys->header.length);
7196
7197		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7198			if (!lun->per_res[i].registered)
7199				continue;
7200
7201			/*
7202			 * We used lun->pr_key_count to calculate the
7203			 * size to allocate.  If it turns out the number of
7204			 * initiators with the registered flag set is
7205			 * larger than that (i.e. they haven't been kept in
7206			 * sync), we've got a problem.
7207			 */
7208			if (key_count >= lun->pr_key_count) {
7209#ifdef NEEDTOPORT
7210				csevent_log(CSC_CTL | CSC_SHELF_SW |
7211					    CTL_PR_ERROR,
7212					    csevent_LogType_Fault,
7213					    csevent_AlertLevel_Yellow,
7214					    csevent_FRU_ShelfController,
7215					    csevent_FRU_Firmware,
7216				        csevent_FRU_Unknown,
7217					    "registered keys %d >= key "
7218					    "count %d", key_count,
7219					    lun->pr_key_count);
7220#endif
7221				key_count++;
7222				continue;
7223			}
7224			memcpy(res_keys->keys[key_count].key,
7225			       lun->per_res[i].res_key.key,
7226			       ctl_min(sizeof(res_keys->keys[key_count].key),
7227			       sizeof(lun->per_res[i].res_key)));
7228			key_count++;
7229		}
7230		break;
7231	}
7232	case SPRI_RR: { // read reservation
7233		struct scsi_per_res_in_rsrv *res;
7234		int tmp_len, header_only;
7235
7236		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7237
7238		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7239
7240		if (lun->flags & CTL_LUN_PR_RESERVED)
7241		{
7242			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7243			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7244				    res->header.length);
7245			header_only = 0;
7246		} else {
7247			tmp_len = sizeof(struct scsi_per_res_in_header);
7248			scsi_ulto4b(0, res->header.length);
7249			header_only = 1;
7250		}
7251
7252		/*
7253		 * We had to drop the lock to allocate our buffer, which
7254		 * leaves time for someone to come in with another
7255		 * persistent reservation.  (That is unlikely, though,
7256		 * since this should be the only persistent reservation
7257		 * command active right now.)
7258		 */
7259		if (tmp_len != total_len) {
7260			mtx_unlock(&softc->ctl_lock);
7261			free(ctsio->kern_data_ptr, M_CTL);
7262			printf("%s: reservation status changed, retrying\n",
7263			       __func__);
7264			goto retry;
7265		}
7266
7267		/*
7268		 * No reservation held, so we're done.
7269		 */
7270		if (header_only != 0)
7271			break;
7272
7273		/*
7274		 * If the registration is an All Registrants type, the key
7275		 * is 0, since it doesn't really matter.
7276		 */
7277		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7278			memcpy(res->data.reservation,
7279			       &lun->per_res[lun->pr_res_idx].res_key,
7280			       sizeof(struct scsi_per_res_key));
7281		}
7282		res->data.scopetype = lun->res_type;
7283		break;
7284	}
7285	case SPRI_RC:     //report capabilities
7286	{
7287		struct scsi_per_res_cap *res_cap;
7288		uint16_t type_mask;
7289
7290		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7291		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7292		res_cap->flags2 |= SPRI_TMV;
7293		type_mask = SPRI_TM_WR_EX_AR |
7294			    SPRI_TM_EX_AC_RO |
7295			    SPRI_TM_WR_EX_RO |
7296			    SPRI_TM_EX_AC |
7297			    SPRI_TM_WR_EX |
7298			    SPRI_TM_EX_AC_AR;
7299		scsi_ulto2b(type_mask, res_cap->type_mask);
7300		break;
7301	}
7302	case SPRI_RS: //read full status
7303	default:
7304		/*
7305		 * This is a bug, because we just checked for this above,
7306		 * and should have returned an error.
7307		 */
7308		panic("Invalid PR type %x", cdb->action);
7309		break; /* NOTREACHED */
7310	}
7311	mtx_unlock(&softc->ctl_lock);
7312
7313	ctsio->be_move_done = ctl_config_move_done;
7314
7315	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7316			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7317			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7318			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7319			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7320
7321	ctl_datamove((union ctl_io *)ctsio);
7322
7323	return (CTL_RETVAL_COMPLETE);
7324}
7325
7326/*
7327 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7328 * it should return.
7329 */
7330static int
7331ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7332		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7333		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7334		struct scsi_per_res_out_parms* param)
7335{
7336	union ctl_ha_msg persis_io;
7337	int retval, i;
7338	int isc_retval;
7339
7340	retval = 0;
7341
7342	if (sa_res_key == 0) {
7343		mtx_lock(&softc->ctl_lock);
7344		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7345			/* validate scope and type */
7346			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7347			     SPR_LU_SCOPE) {
7348				mtx_unlock(&softc->ctl_lock);
7349				ctl_set_invalid_field(/*ctsio*/ ctsio,
7350						      /*sks_valid*/ 1,
7351						      /*command*/ 1,
7352						      /*field*/ 2,
7353						      /*bit_valid*/ 1,
7354						      /*bit*/ 4);
7355				ctl_done((union ctl_io *)ctsio);
7356				return (1);
7357			}
7358
7359		        if (type>8 || type==2 || type==4 || type==0) {
7360				mtx_unlock(&softc->ctl_lock);
7361				ctl_set_invalid_field(/*ctsio*/ ctsio,
7362       	           				      /*sks_valid*/ 1,
7363						      /*command*/ 1,
7364						      /*field*/ 2,
7365						      /*bit_valid*/ 1,
7366						      /*bit*/ 0);
7367				ctl_done((union ctl_io *)ctsio);
7368				return (1);
7369		        }
7370
7371			/* temporarily unregister this nexus */
7372			lun->per_res[residx].registered = 0;
7373
7374			/*
7375			 * Unregister everybody else and build UA for
7376			 * them
7377			 */
7378			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7379				if (lun->per_res[i].registered == 0)
7380					continue;
7381
7382				if (!persis_offset
7383				 && i <CTL_MAX_INITIATORS)
7384					lun->pending_sense[i].ua_pending |=
7385						CTL_UA_REG_PREEMPT;
7386				else if (persis_offset
7387				      && i >= persis_offset)
7388					lun->pending_sense[i-persis_offset
7389						].ua_pending |=
7390						CTL_UA_REG_PREEMPT;
7391				lun->per_res[i].registered = 0;
7392				memset(&lun->per_res[i].res_key, 0,
7393				       sizeof(struct scsi_per_res_key));
7394			}
7395			lun->per_res[residx].registered = 1;
7396			lun->pr_key_count = 1;
7397			lun->res_type = type;
7398			if (lun->res_type != SPR_TYPE_WR_EX_AR
7399			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7400				lun->pr_res_idx = residx;
7401
7402			mtx_unlock(&softc->ctl_lock);
7403			/* send msg to other side */
7404			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7405			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7406			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7407			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7408			persis_io.pr.pr_info.res_type = type;
7409			memcpy(persis_io.pr.pr_info.sa_res_key,
7410			       param->serv_act_res_key,
7411			       sizeof(param->serv_act_res_key));
7412			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7413			     &persis_io, sizeof(persis_io), 0)) >
7414			     CTL_HA_STATUS_SUCCESS) {
7415				printf("CTL:Persis Out error returned "
7416				       "from ctl_ha_msg_send %d\n",
7417				       isc_retval);
7418			}
7419		} else {
7420			/* not all registrants */
7421			mtx_unlock(&softc->ctl_lock);
7422			free(ctsio->kern_data_ptr, M_CTL);
7423			ctl_set_invalid_field(ctsio,
7424					      /*sks_valid*/ 1,
7425					      /*command*/ 0,
7426					      /*field*/ 8,
7427					      /*bit_valid*/ 0,
7428					      /*bit*/ 0);
7429			ctl_done((union ctl_io *)ctsio);
7430			return (1);
7431		}
7432	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7433		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7434		int found = 0;
7435
7436		mtx_lock(&softc->ctl_lock);
7437		if (res_key == sa_res_key) {
7438			/* special case */
7439			/*
7440			 * The spec implies this is not good but doesn't
7441			 * say what to do. There are two choices either
7442			 * generate a res conflict or check condition
7443			 * with illegal field in parameter data. Since
7444			 * that is what is done when the sa_res_key is
7445			 * zero I'll take that approach since this has
7446			 * to do with the sa_res_key.
7447			 */
7448			mtx_unlock(&softc->ctl_lock);
7449			free(ctsio->kern_data_ptr, M_CTL);
7450			ctl_set_invalid_field(ctsio,
7451					      /*sks_valid*/ 1,
7452					      /*command*/ 0,
7453					      /*field*/ 8,
7454					      /*bit_valid*/ 0,
7455					      /*bit*/ 0);
7456			ctl_done((union ctl_io *)ctsio);
7457			return (1);
7458		}
7459
7460		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7461			if (lun->per_res[i].registered
7462			 && memcmp(param->serv_act_res_key,
7463			    lun->per_res[i].res_key.key,
7464			    sizeof(struct scsi_per_res_key)) != 0)
7465				continue;
7466
7467			found = 1;
7468			lun->per_res[i].registered = 0;
7469			memset(&lun->per_res[i].res_key, 0,
7470			       sizeof(struct scsi_per_res_key));
7471			lun->pr_key_count--;
7472
7473			if (!persis_offset
7474			 && i < CTL_MAX_INITIATORS)
7475				lun->pending_sense[i].ua_pending |=
7476					CTL_UA_REG_PREEMPT;
7477			else if (persis_offset
7478			      && i >= persis_offset)
7479				lun->pending_sense[i-persis_offset].ua_pending|=
7480					CTL_UA_REG_PREEMPT;
7481		}
7482		mtx_unlock(&softc->ctl_lock);
7483		if (!found) {
7484			free(ctsio->kern_data_ptr, M_CTL);
7485			ctl_set_reservation_conflict(ctsio);
7486			ctl_done((union ctl_io *)ctsio);
7487			return (CTL_RETVAL_COMPLETE);
7488		}
7489		/* send msg to other side */
7490		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7491		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7492		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7493		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7494		persis_io.pr.pr_info.res_type = type;
7495		memcpy(persis_io.pr.pr_info.sa_res_key,
7496		       param->serv_act_res_key,
7497		       sizeof(param->serv_act_res_key));
7498		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7499		     &persis_io, sizeof(persis_io), 0)) >
7500		     CTL_HA_STATUS_SUCCESS) {
7501			printf("CTL:Persis Out error returned from "
7502			       "ctl_ha_msg_send %d\n", isc_retval);
7503		}
7504	} else {
7505		/* Reserved but not all registrants */
7506		/* sa_res_key is res holder */
7507		if (memcmp(param->serv_act_res_key,
7508                   lun->per_res[lun->pr_res_idx].res_key.key,
7509                   sizeof(struct scsi_per_res_key)) == 0) {
7510			/* validate scope and type */
7511			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7512			     SPR_LU_SCOPE) {
7513				ctl_set_invalid_field(/*ctsio*/ ctsio,
7514						      /*sks_valid*/ 1,
7515						      /*command*/ 1,
7516						      /*field*/ 2,
7517						      /*bit_valid*/ 1,
7518						      /*bit*/ 4);
7519				ctl_done((union ctl_io *)ctsio);
7520				return (1);
7521			}
7522
7523			if (type>8 || type==2 || type==4 || type==0) {
7524				ctl_set_invalid_field(/*ctsio*/ ctsio,
7525						      /*sks_valid*/ 1,
7526						      /*command*/ 1,
7527						      /*field*/ 2,
7528						      /*bit_valid*/ 1,
7529						      /*bit*/ 0);
7530				ctl_done((union ctl_io *)ctsio);
7531				return (1);
7532			}
7533
7534			/*
7535			 * Do the following:
7536			 * if sa_res_key != res_key remove all
7537			 * registrants w/sa_res_key and generate UA
7538			 * for these registrants(Registrations
7539			 * Preempted) if it wasn't an exclusive
7540			 * reservation generate UA(Reservations
7541			 * Preempted) for all other registered nexuses
7542			 * if the type has changed. Establish the new
7543			 * reservation and holder. If res_key and
7544			 * sa_res_key are the same do the above
7545			 * except don't unregister the res holder.
7546			 */
7547
7548			/*
7549			 * Temporarily unregister so it won't get
7550			 * removed or UA generated
7551			 */
7552			lun->per_res[residx].registered = 0;
7553			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7554				if (lun->per_res[i].registered == 0)
7555					continue;
7556
7557				if (memcmp(param->serv_act_res_key,
7558				    lun->per_res[i].res_key.key,
7559				    sizeof(struct scsi_per_res_key)) == 0) {
7560					lun->per_res[i].registered = 0;
7561					memset(&lun->per_res[i].res_key,
7562					       0,
7563					       sizeof(struct scsi_per_res_key));
7564					lun->pr_key_count--;
7565
7566					if (!persis_offset
7567					 && i < CTL_MAX_INITIATORS)
7568						lun->pending_sense[i
7569							].ua_pending |=
7570							CTL_UA_REG_PREEMPT;
7571					else if (persis_offset
7572					      && i >= persis_offset)
7573						lun->pending_sense[
7574						  i-persis_offset].ua_pending |=
7575						  CTL_UA_REG_PREEMPT;
7576				} else if (type != lun->res_type
7577					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7578					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7579						if (!persis_offset
7580						 && i < CTL_MAX_INITIATORS)
7581							lun->pending_sense[i
7582							].ua_pending |=
7583							CTL_UA_RES_RELEASE;
7584						else if (persis_offset
7585						      && i >= persis_offset)
7586							lun->pending_sense[
7587							i-persis_offset
7588							].ua_pending |=
7589							CTL_UA_RES_RELEASE;
7590				}
7591			}
7592			lun->per_res[residx].registered = 1;
7593			lun->res_type = type;
7594			if (lun->res_type != SPR_TYPE_WR_EX_AR
7595			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7596				lun->pr_res_idx = residx;
7597			else
7598				lun->pr_res_idx =
7599					CTL_PR_ALL_REGISTRANTS;
7600
7601			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7602			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7603			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7604			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7605			persis_io.pr.pr_info.res_type = type;
7606			memcpy(persis_io.pr.pr_info.sa_res_key,
7607			       param->serv_act_res_key,
7608			       sizeof(param->serv_act_res_key));
7609			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7610			     &persis_io, sizeof(persis_io), 0)) >
7611			     CTL_HA_STATUS_SUCCESS) {
7612				printf("CTL:Persis Out error returned "
7613				       "from ctl_ha_msg_send %d\n",
7614				       isc_retval);
7615			}
7616		} else {
7617			/*
7618			 * sa_res_key is not the res holder just
7619			 * remove registrants
7620			 */
7621			int found=0;
7622			mtx_lock(&softc->ctl_lock);
7623
7624			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7625				if (memcmp(param->serv_act_res_key,
7626				    lun->per_res[i].res_key.key,
7627				    sizeof(struct scsi_per_res_key)) != 0)
7628					continue;
7629
7630				found = 1;
7631				lun->per_res[i].registered = 0;
7632				memset(&lun->per_res[i].res_key, 0,
7633				       sizeof(struct scsi_per_res_key));
7634				lun->pr_key_count--;
7635
7636				if (!persis_offset
7637				 && i < CTL_MAX_INITIATORS)
7638					lun->pending_sense[i].ua_pending |=
7639						CTL_UA_REG_PREEMPT;
7640				else if (persis_offset
7641				      && i >= persis_offset)
7642					lun->pending_sense[
7643						i-persis_offset].ua_pending |=
7644						CTL_UA_REG_PREEMPT;
7645			}
7646
7647			if (!found) {
7648				mtx_unlock(&softc->ctl_lock);
7649				free(ctsio->kern_data_ptr, M_CTL);
7650				ctl_set_reservation_conflict(ctsio);
7651				ctl_done((union ctl_io *)ctsio);
7652		        	return (1);
7653			}
7654			mtx_unlock(&softc->ctl_lock);
7655			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7656			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7657			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7658			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7659			persis_io.pr.pr_info.res_type = type;
7660			memcpy(persis_io.pr.pr_info.sa_res_key,
7661			       param->serv_act_res_key,
7662			       sizeof(param->serv_act_res_key));
7663			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7664			     &persis_io, sizeof(persis_io), 0)) >
7665			     CTL_HA_STATUS_SUCCESS) {
7666				printf("CTL:Persis Out error returned "
7667				       "from ctl_ha_msg_send %d\n",
7668				isc_retval);
7669			}
7670		}
7671	}
7672
7673	lun->PRGeneration++;
7674
7675	return (retval);
7676}
7677
7678static void
7679ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7680{
7681	int i;
7682
7683	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7684	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7685	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7686		   msg->pr.pr_info.sa_res_key,
7687		   sizeof(struct scsi_per_res_key)) != 0) {
7688		uint64_t sa_res_key;
7689		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7690
7691		if (sa_res_key == 0) {
7692			/* temporarily unregister this nexus */
7693			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7694
7695			/*
7696			 * Unregister everybody else and build UA for
7697			 * them
7698			 */
7699			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7700				if (lun->per_res[i].registered == 0)
7701					continue;
7702
7703				if (!persis_offset
7704				 && i < CTL_MAX_INITIATORS)
7705					lun->pending_sense[i].ua_pending |=
7706						CTL_UA_REG_PREEMPT;
7707				else if (persis_offset && i >= persis_offset)
7708					lun->pending_sense[i -
7709						persis_offset].ua_pending |=
7710						CTL_UA_REG_PREEMPT;
7711				lun->per_res[i].registered = 0;
7712				memset(&lun->per_res[i].res_key, 0,
7713				       sizeof(struct scsi_per_res_key));
7714			}
7715
7716			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7717			lun->pr_key_count = 1;
7718			lun->res_type = msg->pr.pr_info.res_type;
7719			if (lun->res_type != SPR_TYPE_WR_EX_AR
7720			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7721				lun->pr_res_idx = msg->pr.pr_info.residx;
7722		} else {
7723		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7724				if (memcmp(msg->pr.pr_info.sa_res_key,
7725		                   lun->per_res[i].res_key.key,
7726		                   sizeof(struct scsi_per_res_key)) != 0)
7727					continue;
7728
7729				lun->per_res[i].registered = 0;
7730				memset(&lun->per_res[i].res_key, 0,
7731				       sizeof(struct scsi_per_res_key));
7732				lun->pr_key_count--;
7733
7734				if (!persis_offset
7735				 && i < persis_offset)
7736					lun->pending_sense[i].ua_pending |=
7737						CTL_UA_REG_PREEMPT;
7738				else if (persis_offset
7739				      && i >= persis_offset)
7740					lun->pending_sense[i -
7741						persis_offset].ua_pending |=
7742						CTL_UA_REG_PREEMPT;
7743			}
7744		}
7745	} else {
7746		/*
7747		 * Temporarily unregister so it won't get removed
7748		 * or UA generated
7749		 */
7750		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7751		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7752			if (lun->per_res[i].registered == 0)
7753				continue;
7754
7755			if (memcmp(msg->pr.pr_info.sa_res_key,
7756	                   lun->per_res[i].res_key.key,
7757	                   sizeof(struct scsi_per_res_key)) == 0) {
7758				lun->per_res[i].registered = 0;
7759				memset(&lun->per_res[i].res_key, 0,
7760				       sizeof(struct scsi_per_res_key));
7761				lun->pr_key_count--;
7762				if (!persis_offset
7763				 && i < CTL_MAX_INITIATORS)
7764					lun->pending_sense[i].ua_pending |=
7765						CTL_UA_REG_PREEMPT;
7766				else if (persis_offset
7767				      && i >= persis_offset)
7768					lun->pending_sense[i -
7769						persis_offset].ua_pending |=
7770						CTL_UA_REG_PREEMPT;
7771			} else if (msg->pr.pr_info.res_type != lun->res_type
7772				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7773				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7774					if (!persis_offset
7775					 && i < persis_offset)
7776						lun->pending_sense[i
7777							].ua_pending |=
7778							CTL_UA_RES_RELEASE;
7779					else if (persis_offset
7780					      && i >= persis_offset)
7781					lun->pending_sense[i -
7782						persis_offset].ua_pending |=
7783						CTL_UA_RES_RELEASE;
7784			}
7785		}
7786		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7787		lun->res_type = msg->pr.pr_info.res_type;
7788		if (lun->res_type != SPR_TYPE_WR_EX_AR
7789		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7790			lun->pr_res_idx = msg->pr.pr_info.residx;
7791		else
7792			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7793	}
7794	lun->PRGeneration++;
7795
7796}
7797
7798
7799int
7800ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7801{
7802	int retval;
7803	int isc_retval;
7804	u_int32_t param_len;
7805	struct scsi_per_res_out *cdb;
7806	struct ctl_lun *lun;
7807	struct scsi_per_res_out_parms* param;
7808	struct ctl_softc *softc;
7809	uint32_t residx;
7810	uint64_t res_key, sa_res_key;
7811	uint8_t type;
7812	union ctl_ha_msg persis_io;
7813	int    i;
7814
7815	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7816
7817	retval = CTL_RETVAL_COMPLETE;
7818
7819	softc = control_softc;
7820
7821	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7822	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7823
7824	/*
7825	 * We only support whole-LUN scope.  The scope & type are ignored for
7826	 * register, register and ignore existing key and clear.
7827	 * We sometimes ignore scope and type on preempts too!!
7828	 * Verify reservation type here as well.
7829	 */
7830	type = cdb->scope_type & SPR_TYPE_MASK;
7831	if ((cdb->action == SPRO_RESERVE)
7832	 || (cdb->action == SPRO_RELEASE)) {
7833		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7834			ctl_set_invalid_field(/*ctsio*/ ctsio,
7835					      /*sks_valid*/ 1,
7836					      /*command*/ 1,
7837					      /*field*/ 2,
7838					      /*bit_valid*/ 1,
7839					      /*bit*/ 4);
7840			ctl_done((union ctl_io *)ctsio);
7841			return (CTL_RETVAL_COMPLETE);
7842		}
7843
7844		if (type>8 || type==2 || type==4 || type==0) {
7845			ctl_set_invalid_field(/*ctsio*/ ctsio,
7846					      /*sks_valid*/ 1,
7847					      /*command*/ 1,
7848					      /*field*/ 2,
7849					      /*bit_valid*/ 1,
7850					      /*bit*/ 0);
7851			ctl_done((union ctl_io *)ctsio);
7852			return (CTL_RETVAL_COMPLETE);
7853		}
7854	}
7855
7856	switch (cdb->action & SPRO_ACTION_MASK) {
7857	case SPRO_REGISTER:
7858	case SPRO_RESERVE:
7859	case SPRO_RELEASE:
7860	case SPRO_CLEAR:
7861	case SPRO_PREEMPT:
7862	case SPRO_REG_IGNO:
7863		break;
7864	case SPRO_REG_MOVE:
7865	case SPRO_PRE_ABO:
7866	default:
7867		ctl_set_invalid_field(/*ctsio*/ ctsio,
7868				      /*sks_valid*/ 1,
7869				      /*command*/ 1,
7870				      /*field*/ 1,
7871				      /*bit_valid*/ 1,
7872				      /*bit*/ 0);
7873		ctl_done((union ctl_io *)ctsio);
7874		return (CTL_RETVAL_COMPLETE);
7875		break; /* NOTREACHED */
7876	}
7877
7878	param_len = scsi_4btoul(cdb->length);
7879
7880	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7881		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7882		ctsio->kern_data_len = param_len;
7883		ctsio->kern_total_len = param_len;
7884		ctsio->kern_data_resid = 0;
7885		ctsio->kern_rel_offset = 0;
7886		ctsio->kern_sg_entries = 0;
7887		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7888		ctsio->be_move_done = ctl_config_move_done;
7889		ctl_datamove((union ctl_io *)ctsio);
7890
7891		return (CTL_RETVAL_COMPLETE);
7892	}
7893
7894	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7895
7896	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7897	res_key = scsi_8btou64(param->res_key.key);
7898	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7899
7900	/*
7901	 * Validate the reservation key here except for SPRO_REG_IGNO
7902	 * This must be done for all other service actions
7903	 */
7904	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7905		mtx_lock(&softc->ctl_lock);
7906		if (lun->per_res[residx].registered) {
7907		    if (memcmp(param->res_key.key,
7908			       lun->per_res[residx].res_key.key,
7909			       ctl_min(sizeof(param->res_key),
7910			       sizeof(lun->per_res[residx].res_key))) != 0) {
7911				/*
7912				 * The current key passed in doesn't match
7913				 * the one the initiator previously
7914				 * registered.
7915				 */
7916				mtx_unlock(&softc->ctl_lock);
7917				free(ctsio->kern_data_ptr, M_CTL);
7918				ctl_set_reservation_conflict(ctsio);
7919				ctl_done((union ctl_io *)ctsio);
7920				return (CTL_RETVAL_COMPLETE);
7921			}
7922		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7923			/*
7924			 * We are not registered
7925			 */
7926			mtx_unlock(&softc->ctl_lock);
7927			free(ctsio->kern_data_ptr, M_CTL);
7928			ctl_set_reservation_conflict(ctsio);
7929			ctl_done((union ctl_io *)ctsio);
7930			return (CTL_RETVAL_COMPLETE);
7931		} else if (res_key != 0) {
7932			/*
7933			 * We are not registered and trying to register but
7934			 * the register key isn't zero.
7935			 */
7936			mtx_unlock(&softc->ctl_lock);
7937			free(ctsio->kern_data_ptr, M_CTL);
7938			ctl_set_reservation_conflict(ctsio);
7939			ctl_done((union ctl_io *)ctsio);
7940			return (CTL_RETVAL_COMPLETE);
7941		}
7942		mtx_unlock(&softc->ctl_lock);
7943	}
7944
7945	switch (cdb->action & SPRO_ACTION_MASK) {
7946	case SPRO_REGISTER:
7947	case SPRO_REG_IGNO: {
7948
7949#if 0
7950		printf("Registration received\n");
7951#endif
7952
7953		/*
7954		 * We don't support any of these options, as we report in
7955		 * the read capabilities request (see
7956		 * ctl_persistent_reserve_in(), above).
7957		 */
7958		if ((param->flags & SPR_SPEC_I_PT)
7959		 || (param->flags & SPR_ALL_TG_PT)
7960		 || (param->flags & SPR_APTPL)) {
7961			int bit_ptr;
7962
7963			if (param->flags & SPR_APTPL)
7964				bit_ptr = 0;
7965			else if (param->flags & SPR_ALL_TG_PT)
7966				bit_ptr = 2;
7967			else /* SPR_SPEC_I_PT */
7968				bit_ptr = 3;
7969
7970			free(ctsio->kern_data_ptr, M_CTL);
7971			ctl_set_invalid_field(ctsio,
7972					      /*sks_valid*/ 1,
7973					      /*command*/ 0,
7974					      /*field*/ 20,
7975					      /*bit_valid*/ 1,
7976					      /*bit*/ bit_ptr);
7977			ctl_done((union ctl_io *)ctsio);
7978			return (CTL_RETVAL_COMPLETE);
7979		}
7980
7981		mtx_lock(&softc->ctl_lock);
7982
7983		/*
7984		 * The initiator wants to clear the
7985		 * key/unregister.
7986		 */
7987		if (sa_res_key == 0) {
7988			if ((res_key == 0
7989			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
7990			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
7991			  && !lun->per_res[residx].registered)) {
7992				mtx_unlock(&softc->ctl_lock);
7993				goto done;
7994			}
7995
7996			lun->per_res[residx].registered = 0;
7997			memset(&lun->per_res[residx].res_key,
7998			       0, sizeof(lun->per_res[residx].res_key));
7999			lun->pr_key_count--;
8000
8001			if (residx == lun->pr_res_idx) {
8002				lun->flags &= ~CTL_LUN_PR_RESERVED;
8003				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8004
8005				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8006				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8007				 && lun->pr_key_count) {
8008					/*
8009					 * If the reservation is a registrants
8010					 * only type we need to generate a UA
8011					 * for other registered inits.  The
8012					 * sense code should be RESERVATIONS
8013					 * RELEASED
8014					 */
8015
8016					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8017						if (lun->per_res[
8018						    i+persis_offset].registered
8019						    == 0)
8020							continue;
8021						lun->pending_sense[i
8022							].ua_pending |=
8023							CTL_UA_RES_RELEASE;
8024					}
8025				}
8026				lun->res_type = 0;
8027			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8028				if (lun->pr_key_count==0) {
8029					lun->flags &= ~CTL_LUN_PR_RESERVED;
8030					lun->res_type = 0;
8031					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8032				}
8033			}
8034			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8035			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8036			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8037			persis_io.pr.pr_info.residx = residx;
8038			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8039			     &persis_io, sizeof(persis_io), 0 )) >
8040			     CTL_HA_STATUS_SUCCESS) {
8041				printf("CTL:Persis Out error returned from "
8042				       "ctl_ha_msg_send %d\n", isc_retval);
8043			}
8044			mtx_unlock(&softc->ctl_lock);
8045		} else /* sa_res_key != 0 */ {
8046
8047			/*
8048			 * If we aren't registered currently then increment
8049			 * the key count and set the registered flag.
8050			 */
8051			if (!lun->per_res[residx].registered) {
8052				lun->pr_key_count++;
8053				lun->per_res[residx].registered = 1;
8054			}
8055
8056			memcpy(&lun->per_res[residx].res_key,
8057			       param->serv_act_res_key,
8058			       ctl_min(sizeof(param->serv_act_res_key),
8059			       sizeof(lun->per_res[residx].res_key)));
8060
8061			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8062			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8063			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8064			persis_io.pr.pr_info.residx = residx;
8065			memcpy(persis_io.pr.pr_info.sa_res_key,
8066			       param->serv_act_res_key,
8067			       sizeof(param->serv_act_res_key));
8068			mtx_unlock(&softc->ctl_lock);
8069			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8070			     &persis_io, sizeof(persis_io), 0)) >
8071			     CTL_HA_STATUS_SUCCESS) {
8072				printf("CTL:Persis Out error returned from "
8073				       "ctl_ha_msg_send %d\n", isc_retval);
8074			}
8075		}
8076		lun->PRGeneration++;
8077
8078		break;
8079	}
8080	case SPRO_RESERVE:
8081#if 0
8082                printf("Reserve executed type %d\n", type);
8083#endif
8084		mtx_lock(&softc->ctl_lock);
8085		if (lun->flags & CTL_LUN_PR_RESERVED) {
8086			/*
8087			 * if this isn't the reservation holder and it's
8088			 * not a "all registrants" type or if the type is
8089			 * different then we have a conflict
8090			 */
8091			if ((lun->pr_res_idx != residx
8092			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8093			 || lun->res_type != type) {
8094				mtx_unlock(&softc->ctl_lock);
8095				free(ctsio->kern_data_ptr, M_CTL);
8096				ctl_set_reservation_conflict(ctsio);
8097				ctl_done((union ctl_io *)ctsio);
8098				return (CTL_RETVAL_COMPLETE);
8099			}
8100			mtx_unlock(&softc->ctl_lock);
8101		} else /* create a reservation */ {
8102			/*
8103			 * If it's not an "all registrants" type record
8104			 * reservation holder
8105			 */
8106			if (type != SPR_TYPE_WR_EX_AR
8107			 && type != SPR_TYPE_EX_AC_AR)
8108				lun->pr_res_idx = residx; /* Res holder */
8109			else
8110				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8111
8112			lun->flags |= CTL_LUN_PR_RESERVED;
8113			lun->res_type = type;
8114
8115			mtx_unlock(&softc->ctl_lock);
8116
8117			/* send msg to other side */
8118			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8119			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8120			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8121			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8122			persis_io.pr.pr_info.res_type = type;
8123			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8124			     &persis_io, sizeof(persis_io), 0)) >
8125			     CTL_HA_STATUS_SUCCESS) {
8126				printf("CTL:Persis Out error returned from "
8127				       "ctl_ha_msg_send %d\n", isc_retval);
8128			}
8129		}
8130		break;
8131
8132	case SPRO_RELEASE:
8133		mtx_lock(&softc->ctl_lock);
8134		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8135			/* No reservation exists return good status */
8136			mtx_unlock(&softc->ctl_lock);
8137			goto done;
8138		}
8139		/*
8140		 * Is this nexus a reservation holder?
8141		 */
8142		if (lun->pr_res_idx != residx
8143		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8144			/*
8145			 * not a res holder return good status but
8146			 * do nothing
8147			 */
8148			mtx_unlock(&softc->ctl_lock);
8149			goto done;
8150		}
8151
8152		if (lun->res_type != type) {
8153			mtx_unlock(&softc->ctl_lock);
8154			free(ctsio->kern_data_ptr, M_CTL);
8155			ctl_set_illegal_pr_release(ctsio);
8156			ctl_done((union ctl_io *)ctsio);
8157			return (CTL_RETVAL_COMPLETE);
8158		}
8159
8160		/* okay to release */
8161		lun->flags &= ~CTL_LUN_PR_RESERVED;
8162		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8163		lun->res_type = 0;
8164
8165		/*
8166		 * if this isn't an exclusive access
8167		 * res generate UA for all other
8168		 * registrants.
8169		 */
8170		if (type != SPR_TYPE_EX_AC
8171		 && type != SPR_TYPE_WR_EX) {
8172			/*
8173			 * temporarily unregister so we don't generate UA
8174			 */
8175			lun->per_res[residx].registered = 0;
8176
8177			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8178				if (lun->per_res[i+persis_offset].registered
8179				    == 0)
8180					continue;
8181				lun->pending_sense[i].ua_pending |=
8182					CTL_UA_RES_RELEASE;
8183			}
8184
8185			lun->per_res[residx].registered = 1;
8186		}
8187		mtx_unlock(&softc->ctl_lock);
8188		/* Send msg to other side */
8189		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8190		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8191		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8192		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8193		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8194			printf("CTL:Persis Out error returned from "
8195			       "ctl_ha_msg_send %d\n", isc_retval);
8196		}
8197		break;
8198
8199	case SPRO_CLEAR:
8200		/* send msg to other side */
8201
8202		mtx_lock(&softc->ctl_lock);
8203		lun->flags &= ~CTL_LUN_PR_RESERVED;
8204		lun->res_type = 0;
8205		lun->pr_key_count = 0;
8206		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8207
8208
8209		memset(&lun->per_res[residx].res_key,
8210		       0, sizeof(lun->per_res[residx].res_key));
8211		lun->per_res[residx].registered = 0;
8212
8213		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8214			if (lun->per_res[i].registered) {
8215				if (!persis_offset && i < CTL_MAX_INITIATORS)
8216					lun->pending_sense[i].ua_pending |=
8217						CTL_UA_RES_PREEMPT;
8218				else if (persis_offset && i >= persis_offset)
8219					lun->pending_sense[i-persis_offset
8220					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8221
8222				memset(&lun->per_res[i].res_key,
8223				       0, sizeof(struct scsi_per_res_key));
8224				lun->per_res[i].registered = 0;
8225			}
8226		lun->PRGeneration++;
8227		mtx_unlock(&softc->ctl_lock);
8228		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8229		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8230		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8231		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8232		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8233			printf("CTL:Persis Out error returned from "
8234			       "ctl_ha_msg_send %d\n", isc_retval);
8235		}
8236		break;
8237
8238	case SPRO_PREEMPT: {
8239		int nretval;
8240
8241		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8242					  residx, ctsio, cdb, param);
8243		if (nretval != 0)
8244			return (CTL_RETVAL_COMPLETE);
8245		break;
8246	}
8247	case SPRO_REG_MOVE:
8248	case SPRO_PRE_ABO:
8249	default:
8250		free(ctsio->kern_data_ptr, M_CTL);
8251		ctl_set_invalid_field(/*ctsio*/ ctsio,
8252				      /*sks_valid*/ 1,
8253				      /*command*/ 1,
8254				      /*field*/ 1,
8255				      /*bit_valid*/ 1,
8256				      /*bit*/ 0);
8257		ctl_done((union ctl_io *)ctsio);
8258		return (CTL_RETVAL_COMPLETE);
8259		break; /* NOTREACHED */
8260	}
8261
8262done:
8263	free(ctsio->kern_data_ptr, M_CTL);
8264	ctl_set_success(ctsio);
8265	ctl_done((union ctl_io *)ctsio);
8266
8267	return (retval);
8268}
8269
8270/*
8271 * This routine is for handling a message from the other SC pertaining to
8272 * persistent reserve out. All the error checking will have been done
8273 * so only perorming the action need be done here to keep the two
8274 * in sync.
8275 */
8276static void
8277ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8278{
8279	struct ctl_lun *lun;
8280	struct ctl_softc *softc;
8281	int i;
8282	uint32_t targ_lun;
8283
8284	softc = control_softc;
8285
8286	mtx_lock(&softc->ctl_lock);
8287
8288	targ_lun = msg->hdr.nexus.targ_lun;
8289	if (msg->hdr.nexus.lun_map_fn != NULL)
8290		targ_lun = msg->hdr.nexus.lun_map_fn(msg->hdr.nexus.lun_map_arg, targ_lun);
8291	lun = softc->ctl_luns[targ_lun];
8292	switch(msg->pr.pr_info.action) {
8293	case CTL_PR_REG_KEY:
8294		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8295			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8296			lun->pr_key_count++;
8297		}
8298		lun->PRGeneration++;
8299		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8300		       msg->pr.pr_info.sa_res_key,
8301		       sizeof(struct scsi_per_res_key));
8302		break;
8303
8304	case CTL_PR_UNREG_KEY:
8305		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8306		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8307		       0, sizeof(struct scsi_per_res_key));
8308		lun->pr_key_count--;
8309
8310		/* XXX Need to see if the reservation has been released */
8311		/* if so do we need to generate UA? */
8312		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8313			lun->flags &= ~CTL_LUN_PR_RESERVED;
8314			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8315
8316			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8317			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8318			 && lun->pr_key_count) {
8319				/*
8320				 * If the reservation is a registrants
8321				 * only type we need to generate a UA
8322				 * for other registered inits.  The
8323				 * sense code should be RESERVATIONS
8324				 * RELEASED
8325				 */
8326
8327				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8328					if (lun->per_res[i+
8329					    persis_offset].registered == 0)
8330						continue;
8331
8332					lun->pending_sense[i
8333						].ua_pending |=
8334						CTL_UA_RES_RELEASE;
8335				}
8336			}
8337			lun->res_type = 0;
8338		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8339			if (lun->pr_key_count==0) {
8340				lun->flags &= ~CTL_LUN_PR_RESERVED;
8341				lun->res_type = 0;
8342				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8343			}
8344		}
8345		lun->PRGeneration++;
8346		break;
8347
8348	case CTL_PR_RESERVE:
8349		lun->flags |= CTL_LUN_PR_RESERVED;
8350		lun->res_type = msg->pr.pr_info.res_type;
8351		lun->pr_res_idx = msg->pr.pr_info.residx;
8352
8353		break;
8354
8355	case CTL_PR_RELEASE:
8356		/*
8357		 * if this isn't an exclusive access res generate UA for all
8358		 * other registrants.
8359		 */
8360		if (lun->res_type != SPR_TYPE_EX_AC
8361		 && lun->res_type != SPR_TYPE_WR_EX) {
8362			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8363				if (lun->per_res[i+persis_offset].registered)
8364					lun->pending_sense[i].ua_pending |=
8365						CTL_UA_RES_RELEASE;
8366		}
8367
8368		lun->flags &= ~CTL_LUN_PR_RESERVED;
8369		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8370		lun->res_type = 0;
8371		break;
8372
8373	case CTL_PR_PREEMPT:
8374		ctl_pro_preempt_other(lun, msg);
8375		break;
8376	case CTL_PR_CLEAR:
8377		lun->flags &= ~CTL_LUN_PR_RESERVED;
8378		lun->res_type = 0;
8379		lun->pr_key_count = 0;
8380		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8381
8382		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8383			if (lun->per_res[i].registered == 0)
8384				continue;
8385			if (!persis_offset
8386			 && i < CTL_MAX_INITIATORS)
8387				lun->pending_sense[i].ua_pending |=
8388					CTL_UA_RES_PREEMPT;
8389			else if (persis_offset
8390			      && i >= persis_offset)
8391   				lun->pending_sense[i-persis_offset].ua_pending|=
8392					CTL_UA_RES_PREEMPT;
8393			memset(&lun->per_res[i].res_key, 0,
8394			       sizeof(struct scsi_per_res_key));
8395			lun->per_res[i].registered = 0;
8396		}
8397		lun->PRGeneration++;
8398		break;
8399	}
8400
8401	mtx_unlock(&softc->ctl_lock);
8402}
8403
8404int
8405ctl_read_write(struct ctl_scsiio *ctsio)
8406{
8407	struct ctl_lun *lun;
8408	struct ctl_lba_len lbalen;
8409	uint64_t lba;
8410	uint32_t num_blocks;
8411	int reladdr, fua, dpo, ebp;
8412	int retval;
8413	int isread;
8414
8415	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8416
8417	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8418
8419	reladdr = 0;
8420	fua = 0;
8421	dpo = 0;
8422	ebp = 0;
8423
8424	retval = CTL_RETVAL_COMPLETE;
8425
8426	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8427	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8428	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8429		uint32_t residx;
8430
8431		/*
8432		 * XXX KDM need a lock here.
8433		 */
8434		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8435		if ((lun->res_type == SPR_TYPE_EX_AC
8436		  && residx != lun->pr_res_idx)
8437		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8438		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8439		  && !lun->per_res[residx].registered)) {
8440			ctl_set_reservation_conflict(ctsio);
8441			ctl_done((union ctl_io *)ctsio);
8442			return (CTL_RETVAL_COMPLETE);
8443	        }
8444	}
8445
8446	switch (ctsio->cdb[0]) {
8447	case READ_6:
8448	case WRITE_6: {
8449		struct scsi_rw_6 *cdb;
8450
8451		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8452
8453		lba = scsi_3btoul(cdb->addr);
8454		/* only 5 bits are valid in the most significant address byte */
8455		lba &= 0x1fffff;
8456		num_blocks = cdb->length;
8457		/*
8458		 * This is correct according to SBC-2.
8459		 */
8460		if (num_blocks == 0)
8461			num_blocks = 256;
8462		break;
8463	}
8464	case READ_10:
8465	case WRITE_10: {
8466		struct scsi_rw_10 *cdb;
8467
8468		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8469
8470		if (cdb->byte2 & SRW10_RELADDR)
8471			reladdr = 1;
8472		if (cdb->byte2 & SRW10_FUA)
8473			fua = 1;
8474		if (cdb->byte2 & SRW10_DPO)
8475			dpo = 1;
8476
8477		if ((cdb->opcode == WRITE_10)
8478		 && (cdb->byte2 & SRW10_EBP))
8479			ebp = 1;
8480
8481		lba = scsi_4btoul(cdb->addr);
8482		num_blocks = scsi_2btoul(cdb->length);
8483		break;
8484	}
8485	case WRITE_VERIFY_10: {
8486		struct scsi_write_verify_10 *cdb;
8487
8488		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8489
8490		/*
8491		 * XXX KDM we should do actual write verify support at some
8492		 * point.  This is obviously fake, we're just translating
8493		 * things to a write.  So we don't even bother checking the
8494		 * BYTCHK field, since we don't do any verification.  If
8495		 * the user asks for it, we'll just pretend we did it.
8496		 */
8497		if (cdb->byte2 & SWV_DPO)
8498			dpo = 1;
8499
8500		lba = scsi_4btoul(cdb->addr);
8501		num_blocks = scsi_2btoul(cdb->length);
8502		break;
8503	}
8504	case READ_12:
8505	case WRITE_12: {
8506		struct scsi_rw_12 *cdb;
8507
8508		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8509
8510		if (cdb->byte2 & SRW12_RELADDR)
8511			reladdr = 1;
8512		if (cdb->byte2 & SRW12_FUA)
8513			fua = 1;
8514		if (cdb->byte2 & SRW12_DPO)
8515			dpo = 1;
8516		lba = scsi_4btoul(cdb->addr);
8517		num_blocks = scsi_4btoul(cdb->length);
8518		break;
8519	}
8520	case WRITE_VERIFY_12: {
8521		struct scsi_write_verify_12 *cdb;
8522
8523		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8524
8525		if (cdb->byte2 & SWV_DPO)
8526			dpo = 1;
8527
8528		lba = scsi_4btoul(cdb->addr);
8529		num_blocks = scsi_4btoul(cdb->length);
8530
8531		break;
8532	}
8533	case READ_16:
8534	case WRITE_16: {
8535		struct scsi_rw_16 *cdb;
8536
8537		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8538
8539		if (cdb->byte2 & SRW12_RELADDR)
8540			reladdr = 1;
8541		if (cdb->byte2 & SRW12_FUA)
8542			fua = 1;
8543		if (cdb->byte2 & SRW12_DPO)
8544			dpo = 1;
8545
8546		lba = scsi_8btou64(cdb->addr);
8547		num_blocks = scsi_4btoul(cdb->length);
8548		break;
8549	}
8550	case WRITE_VERIFY_16: {
8551		struct scsi_write_verify_16 *cdb;
8552
8553		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8554
8555		if (cdb->byte2 & SWV_DPO)
8556			dpo = 1;
8557
8558		lba = scsi_8btou64(cdb->addr);
8559		num_blocks = scsi_4btoul(cdb->length);
8560		break;
8561	}
8562	default:
8563		/*
8564		 * We got a command we don't support.  This shouldn't
8565		 * happen, commands should be filtered out above us.
8566		 */
8567		ctl_set_invalid_opcode(ctsio);
8568		ctl_done((union ctl_io *)ctsio);
8569
8570		return (CTL_RETVAL_COMPLETE);
8571		break; /* NOTREACHED */
8572	}
8573
8574	/*
8575	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8576	 * interesting for us, but if RAIDCore is in write-back mode,
8577	 * getting it to do write-through for a particular transaction may
8578	 * not be possible.
8579	 */
8580	/*
8581	 * We don't support relative addressing.  That also requires
8582	 * supporting linked commands, which we don't do.
8583	 */
8584	if (reladdr != 0) {
8585		ctl_set_invalid_field(ctsio,
8586				      /*sks_valid*/ 1,
8587				      /*command*/ 1,
8588				      /*field*/ 1,
8589				      /*bit_valid*/ 1,
8590				      /*bit*/ 0);
8591		ctl_done((union ctl_io *)ctsio);
8592		return (CTL_RETVAL_COMPLETE);
8593	}
8594
8595	/*
8596	 * The first check is to make sure we're in bounds, the second
8597	 * check is to catch wrap-around problems.  If the lba + num blocks
8598	 * is less than the lba, then we've wrapped around and the block
8599	 * range is invalid anyway.
8600	 */
8601	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8602	 || ((lba + num_blocks) < lba)) {
8603		ctl_set_lba_out_of_range(ctsio);
8604		ctl_done((union ctl_io *)ctsio);
8605		return (CTL_RETVAL_COMPLETE);
8606	}
8607
8608	/*
8609	 * According to SBC-3, a transfer length of 0 is not an error.
8610	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8611	 * translates to 256 blocks for those commands.
8612	 */
8613	if (num_blocks == 0) {
8614		ctl_set_success(ctsio);
8615		ctl_done((union ctl_io *)ctsio);
8616		return (CTL_RETVAL_COMPLETE);
8617	}
8618
8619	lbalen.lba = lba;
8620	lbalen.len = num_blocks;
8621	memcpy(ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
8622	       sizeof(lbalen));
8623
8624	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8625
8626	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8627
8628	return (retval);
8629}
8630
8631int
8632ctl_report_luns(struct ctl_scsiio *ctsio)
8633{
8634	struct scsi_report_luns *cdb;
8635	struct scsi_report_luns_data *lun_data;
8636	struct ctl_lun *lun, *request_lun;
8637	int num_luns, retval;
8638	uint32_t alloc_len, lun_datalen;
8639	int num_filled, well_known;
8640	uint32_t initidx, targ_lun_id, lun_id;
8641
8642	retval = CTL_RETVAL_COMPLETE;
8643	well_known = 0;
8644
8645	cdb = (struct scsi_report_luns *)ctsio->cdb;
8646
8647	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8648
8649	mtx_lock(&control_softc->ctl_lock);
8650	num_luns = control_softc->num_luns;
8651	mtx_unlock(&control_softc->ctl_lock);
8652
8653	switch (cdb->select_report) {
8654	case RPL_REPORT_DEFAULT:
8655	case RPL_REPORT_ALL:
8656		break;
8657	case RPL_REPORT_WELLKNOWN:
8658		well_known = 1;
8659		num_luns = 0;
8660		break;
8661	default:
8662		ctl_set_invalid_field(ctsio,
8663				      /*sks_valid*/ 1,
8664				      /*command*/ 1,
8665				      /*field*/ 2,
8666				      /*bit_valid*/ 0,
8667				      /*bit*/ 0);
8668		ctl_done((union ctl_io *)ctsio);
8669		return (retval);
8670		break; /* NOTREACHED */
8671	}
8672
8673	alloc_len = scsi_4btoul(cdb->length);
8674	/*
8675	 * The initiator has to allocate at least 16 bytes for this request,
8676	 * so he can at least get the header and the first LUN.  Otherwise
8677	 * we reject the request (per SPC-3 rev 14, section 6.21).
8678	 */
8679	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8680	    sizeof(struct scsi_report_luns_lundata))) {
8681		ctl_set_invalid_field(ctsio,
8682				      /*sks_valid*/ 1,
8683				      /*command*/ 1,
8684				      /*field*/ 6,
8685				      /*bit_valid*/ 0,
8686				      /*bit*/ 0);
8687		ctl_done((union ctl_io *)ctsio);
8688		return (retval);
8689	}
8690
8691	request_lun = (struct ctl_lun *)
8692		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8693
8694	lun_datalen = sizeof(*lun_data) +
8695		(num_luns * sizeof(struct scsi_report_luns_lundata));
8696
8697	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8698	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8699	ctsio->kern_sg_entries = 0;
8700
8701	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8702
8703	mtx_lock(&control_softc->ctl_lock);
8704	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
8705		lun_id = targ_lun_id;
8706		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
8707			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
8708		if (lun_id >= CTL_MAX_LUNS)
8709			continue;
8710		lun = control_softc->ctl_luns[lun_id];
8711		if (lun == NULL)
8712			continue;
8713
8714		if (targ_lun_id <= 0xff) {
8715			/*
8716			 * Peripheral addressing method, bus number 0.
8717			 */
8718			lun_data->luns[num_filled].lundata[0] =
8719				RPL_LUNDATA_ATYP_PERIPH;
8720			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
8721			num_filled++;
8722		} else if (targ_lun_id <= 0x3fff) {
8723			/*
8724			 * Flat addressing method.
8725			 */
8726			lun_data->luns[num_filled].lundata[0] =
8727				RPL_LUNDATA_ATYP_FLAT |
8728				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
8729#ifdef OLDCTLHEADERS
8730				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
8731				(targ_lun_id & SRLD_BUS_LUN_MASK);
8732#endif
8733			lun_data->luns[num_filled].lundata[1] =
8734#ifdef OLDCTLHEADERS
8735				targ_lun_id >> SRLD_BUS_LUN_BITS;
8736#endif
8737				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
8738			num_filled++;
8739		} else {
8740			printf("ctl_report_luns: bogus LUN number %jd, "
8741			       "skipping\n", (intmax_t)targ_lun_id);
8742		}
8743		/*
8744		 * According to SPC-3, rev 14 section 6.21:
8745		 *
8746		 * "The execution of a REPORT LUNS command to any valid and
8747		 * installed logical unit shall clear the REPORTED LUNS DATA
8748		 * HAS CHANGED unit attention condition for all logical
8749		 * units of that target with respect to the requesting
8750		 * initiator. A valid and installed logical unit is one
8751		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8752		 * INQUIRY data (see 6.4.2)."
8753		 *
8754		 * If request_lun is NULL, the LUN this report luns command
8755		 * was issued to is either disabled or doesn't exist. In that
8756		 * case, we shouldn't clear any pending lun change unit
8757		 * attention.
8758		 */
8759		if (request_lun != NULL)
8760			lun->pending_sense[initidx].ua_pending &=
8761				~CTL_UA_LUN_CHANGE;
8762	}
8763	mtx_unlock(&control_softc->ctl_lock);
8764
8765	/*
8766	 * It's quite possible that we've returned fewer LUNs than we allocated
8767	 * space for.  Trim it.
8768	 */
8769	lun_datalen = sizeof(*lun_data) +
8770		(num_filled * sizeof(struct scsi_report_luns_lundata));
8771
8772	if (lun_datalen < alloc_len) {
8773		ctsio->residual = alloc_len - lun_datalen;
8774		ctsio->kern_data_len = lun_datalen;
8775		ctsio->kern_total_len = lun_datalen;
8776	} else {
8777		ctsio->residual = 0;
8778		ctsio->kern_data_len = alloc_len;
8779		ctsio->kern_total_len = alloc_len;
8780	}
8781	ctsio->kern_data_resid = 0;
8782	ctsio->kern_rel_offset = 0;
8783	ctsio->kern_sg_entries = 0;
8784
8785	/*
8786	 * We set this to the actual data length, regardless of how much
8787	 * space we actually have to return results.  If the user looks at
8788	 * this value, he'll know whether or not he allocated enough space
8789	 * and reissue the command if necessary.  We don't support well
8790	 * known logical units, so if the user asks for that, return none.
8791	 */
8792	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8793
8794	/*
8795	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8796	 * this request.
8797	 */
8798	ctsio->scsi_status = SCSI_STATUS_OK;
8799
8800	ctsio->be_move_done = ctl_config_move_done;
8801	ctl_datamove((union ctl_io *)ctsio);
8802
8803	return (retval);
8804}
8805
8806int
8807ctl_request_sense(struct ctl_scsiio *ctsio)
8808{
8809	struct scsi_request_sense *cdb;
8810	struct scsi_sense_data *sense_ptr;
8811	struct ctl_lun *lun;
8812	uint32_t initidx;
8813	int have_error;
8814	scsi_sense_data_type sense_format;
8815
8816	cdb = (struct scsi_request_sense *)ctsio->cdb;
8817
8818	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8819
8820	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
8821
8822	/*
8823	 * Determine which sense format the user wants.
8824	 */
8825	if (cdb->byte2 & SRS_DESC)
8826		sense_format = SSD_TYPE_DESC;
8827	else
8828		sense_format = SSD_TYPE_FIXED;
8829
8830	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
8831	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
8832	ctsio->kern_sg_entries = 0;
8833
8834	/*
8835	 * struct scsi_sense_data, which is currently set to 256 bytes, is
8836	 * larger than the largest allowed value for the length field in the
8837	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
8838	 */
8839	ctsio->residual = 0;
8840	ctsio->kern_data_len = cdb->length;
8841	ctsio->kern_total_len = cdb->length;
8842
8843	ctsio->kern_data_resid = 0;
8844	ctsio->kern_rel_offset = 0;
8845	ctsio->kern_sg_entries = 0;
8846
8847	/*
8848	 * If we don't have a LUN, we don't have any pending sense.
8849	 */
8850	if (lun == NULL)
8851		goto no_sense;
8852
8853	have_error = 0;
8854	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8855	/*
8856	 * Check for pending sense, and then for pending unit attentions.
8857	 * Pending sense gets returned first, then pending unit attentions.
8858	 */
8859	mtx_lock(&lun->ctl_softc->ctl_lock);
8860	if (ctl_is_set(lun->have_ca, initidx)) {
8861		scsi_sense_data_type stored_format;
8862
8863		/*
8864		 * Check to see which sense format was used for the stored
8865		 * sense data.
8866		 */
8867		stored_format = scsi_sense_type(
8868		    &lun->pending_sense[initidx].sense);
8869
8870		/*
8871		 * If the user requested a different sense format than the
8872		 * one we stored, then we need to convert it to the other
8873		 * format.  If we're going from descriptor to fixed format
8874		 * sense data, we may lose things in translation, depending
8875		 * on what options were used.
8876		 *
8877		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
8878		 * for some reason we'll just copy it out as-is.
8879		 */
8880		if ((stored_format == SSD_TYPE_FIXED)
8881		 && (sense_format == SSD_TYPE_DESC))
8882			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
8883			    &lun->pending_sense[initidx].sense,
8884			    (struct scsi_sense_data_desc *)sense_ptr);
8885		else if ((stored_format == SSD_TYPE_DESC)
8886		      && (sense_format == SSD_TYPE_FIXED))
8887			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
8888			    &lun->pending_sense[initidx].sense,
8889			    (struct scsi_sense_data_fixed *)sense_ptr);
8890		else
8891			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
8892			       ctl_min(sizeof(*sense_ptr),
8893			       sizeof(lun->pending_sense[initidx].sense)));
8894
8895		ctl_clear_mask(lun->have_ca, initidx);
8896		have_error = 1;
8897	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
8898		ctl_ua_type ua_type;
8899
8900		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
8901				       sense_ptr, sense_format);
8902		if (ua_type != CTL_UA_NONE) {
8903			have_error = 1;
8904			/* We're reporting this UA, so clear it */
8905			lun->pending_sense[initidx].ua_pending &= ~ua_type;
8906		}
8907	}
8908	mtx_unlock(&lun->ctl_softc->ctl_lock);
8909
8910	/*
8911	 * We already have a pending error, return it.
8912	 */
8913	if (have_error != 0) {
8914		/*
8915		 * We report the SCSI status as OK, since the status of the
8916		 * request sense command itself is OK.
8917		 */
8918		ctsio->scsi_status = SCSI_STATUS_OK;
8919
8920		/*
8921		 * We report 0 for the sense length, because we aren't doing
8922		 * autosense in this case.  We're reporting sense as
8923		 * parameter data.
8924		 */
8925		ctsio->sense_len = 0;
8926
8927		ctsio->be_move_done = ctl_config_move_done;
8928		ctl_datamove((union ctl_io *)ctsio);
8929
8930		return (CTL_RETVAL_COMPLETE);
8931	}
8932
8933no_sense:
8934
8935	/*
8936	 * No sense information to report, so we report that everything is
8937	 * okay.
8938	 */
8939	ctl_set_sense_data(sense_ptr,
8940			   lun,
8941			   sense_format,
8942			   /*current_error*/ 1,
8943			   /*sense_key*/ SSD_KEY_NO_SENSE,
8944			   /*asc*/ 0x00,
8945			   /*ascq*/ 0x00,
8946			   SSD_ELEM_NONE);
8947
8948	ctsio->scsi_status = SCSI_STATUS_OK;
8949
8950	/*
8951	 * We report 0 for the sense length, because we aren't doing
8952	 * autosense in this case.  We're reporting sense as parameter data.
8953	 */
8954	ctsio->sense_len = 0;
8955	ctsio->be_move_done = ctl_config_move_done;
8956	ctl_datamove((union ctl_io *)ctsio);
8957
8958	return (CTL_RETVAL_COMPLETE);
8959}
8960
8961int
8962ctl_tur(struct ctl_scsiio *ctsio)
8963{
8964	struct ctl_lun *lun;
8965
8966	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8967
8968	CTL_DEBUG_PRINT(("ctl_tur\n"));
8969
8970	if (lun == NULL)
8971		return (-EINVAL);
8972
8973	ctsio->scsi_status = SCSI_STATUS_OK;
8974	ctsio->io_hdr.status = CTL_SUCCESS;
8975
8976	ctl_done((union ctl_io *)ctsio);
8977
8978	return (CTL_RETVAL_COMPLETE);
8979}
8980
8981#ifdef notyet
8982static int
8983ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
8984{
8985
8986}
8987#endif
8988
8989static int
8990ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
8991{
8992	struct scsi_vpd_supported_pages *pages;
8993	int sup_page_size;
8994	struct ctl_lun *lun;
8995
8996	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8997
8998	sup_page_size = sizeof(struct scsi_vpd_supported_pages) +
8999		SCSI_EVPD_NUM_SUPPORTED_PAGES;
9000	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9001	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9002	ctsio->kern_sg_entries = 0;
9003
9004	if (sup_page_size < alloc_len) {
9005		ctsio->residual = alloc_len - sup_page_size;
9006		ctsio->kern_data_len = sup_page_size;
9007		ctsio->kern_total_len = sup_page_size;
9008	} else {
9009		ctsio->residual = 0;
9010		ctsio->kern_data_len = alloc_len;
9011		ctsio->kern_total_len = alloc_len;
9012	}
9013	ctsio->kern_data_resid = 0;
9014	ctsio->kern_rel_offset = 0;
9015	ctsio->kern_sg_entries = 0;
9016
9017	/*
9018	 * The control device is always connected.  The disk device, on the
9019	 * other hand, may not be online all the time.  Need to change this
9020	 * to figure out whether the disk device is actually online or not.
9021	 */
9022	if (lun != NULL)
9023		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9024				lun->be_lun->lun_type;
9025	else
9026		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9027
9028	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9029	/* Supported VPD pages */
9030	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9031	/* Serial Number */
9032	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9033	/* Device Identification */
9034	pages->page_list[2] = SVPD_DEVICE_ID;
9035
9036	ctsio->scsi_status = SCSI_STATUS_OK;
9037
9038	ctsio->be_move_done = ctl_config_move_done;
9039	ctl_datamove((union ctl_io *)ctsio);
9040
9041	return (CTL_RETVAL_COMPLETE);
9042}
9043
9044static int
9045ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9046{
9047	struct scsi_vpd_unit_serial_number *sn_ptr;
9048	struct ctl_lun *lun;
9049#ifndef CTL_USE_BACKEND_SN
9050	char tmpstr[32];
9051#endif
9052
9053	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9054
9055	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9056	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9057	ctsio->kern_sg_entries = 0;
9058
9059	if (sizeof(*sn_ptr) < alloc_len) {
9060		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9061		ctsio->kern_data_len = sizeof(*sn_ptr);
9062		ctsio->kern_total_len = sizeof(*sn_ptr);
9063	} else {
9064		ctsio->residual = 0;
9065		ctsio->kern_data_len = alloc_len;
9066		ctsio->kern_total_len = alloc_len;
9067	}
9068	ctsio->kern_data_resid = 0;
9069	ctsio->kern_rel_offset = 0;
9070	ctsio->kern_sg_entries = 0;
9071
9072	/*
9073	 * The control device is always connected.  The disk device, on the
9074	 * other hand, may not be online all the time.  Need to change this
9075	 * to figure out whether the disk device is actually online or not.
9076	 */
9077	if (lun != NULL)
9078		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9079				  lun->be_lun->lun_type;
9080	else
9081		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9082
9083	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9084	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9085#ifdef CTL_USE_BACKEND_SN
9086	/*
9087	 * If we don't have a LUN, we just leave the serial number as
9088	 * all spaces.
9089	 */
9090	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9091	if (lun != NULL) {
9092		strncpy((char *)sn_ptr->serial_num,
9093			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9094	}
9095#else
9096	/*
9097	 * Note that we're using a non-unique serial number here,
9098	 */
9099	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9100	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9101	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9102		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9103#endif
9104	ctsio->scsi_status = SCSI_STATUS_OK;
9105
9106	ctsio->be_move_done = ctl_config_move_done;
9107	ctl_datamove((union ctl_io *)ctsio);
9108
9109	return (CTL_RETVAL_COMPLETE);
9110}
9111
9112
9113static int
9114ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9115{
9116	struct scsi_vpd_device_id *devid_ptr;
9117	struct scsi_vpd_id_descriptor *desc, *desc1;
9118	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9119	struct scsi_vpd_id_t10 *t10id;
9120	struct ctl_softc *ctl_softc;
9121	struct ctl_lun *lun;
9122	struct ctl_frontend *fe;
9123#ifndef CTL_USE_BACKEND_SN
9124	char tmpstr[32];
9125#endif /* CTL_USE_BACKEND_SN */
9126	int devid_len;
9127
9128	ctl_softc = control_softc;
9129
9130	mtx_lock(&ctl_softc->ctl_lock);
9131	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9132	mtx_unlock(&ctl_softc->ctl_lock);
9133
9134	if (fe->devid != NULL)
9135		return ((fe->devid)(ctsio, alloc_len));
9136
9137	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9138
9139	devid_len = sizeof(struct scsi_vpd_device_id) +
9140		sizeof(struct scsi_vpd_id_descriptor) +
9141		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9142		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9143		sizeof(struct scsi_vpd_id_descriptor) +
9144		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9145		sizeof(struct scsi_vpd_id_descriptor) +
9146		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9147
9148	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9149	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9150	ctsio->kern_sg_entries = 0;
9151
9152	if (devid_len < alloc_len) {
9153		ctsio->residual = alloc_len - devid_len;
9154		ctsio->kern_data_len = devid_len;
9155		ctsio->kern_total_len = devid_len;
9156	} else {
9157		ctsio->residual = 0;
9158		ctsio->kern_data_len = alloc_len;
9159		ctsio->kern_total_len = alloc_len;
9160	}
9161	ctsio->kern_data_resid = 0;
9162	ctsio->kern_rel_offset = 0;
9163	ctsio->kern_sg_entries = 0;
9164
9165	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9166	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9167	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9168		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9169	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9170	          CTL_WWPN_LEN);
9171	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9172	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9173
9174	/*
9175	 * The control device is always connected.  The disk device, on the
9176	 * other hand, may not be online all the time.
9177	 */
9178	if (lun != NULL)
9179		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9180				     lun->be_lun->lun_type;
9181	else
9182		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9183
9184	devid_ptr->page_code = SVPD_DEVICE_ID;
9185
9186	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9187
9188	mtx_lock(&ctl_softc->ctl_lock);
9189
9190	/*
9191	 * For Fibre channel,
9192	 */
9193	if (fe->port_type == CTL_PORT_FC)
9194	{
9195		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9196				      SVPD_ID_CODESET_ASCII;
9197        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9198		              SVPD_ID_CODESET_BINARY;
9199	}
9200	else
9201	{
9202		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9203				      SVPD_ID_CODESET_ASCII;
9204        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9205		              SVPD_ID_CODESET_BINARY;
9206	}
9207	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9208	mtx_unlock(&ctl_softc->ctl_lock);
9209
9210	/*
9211	 * We're using a LUN association here.  i.e., this device ID is a
9212	 * per-LUN identifier.
9213	 */
9214	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9215	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9216	strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9217
9218	/*
9219	 * desc1 is for the WWPN which is a port asscociation.
9220	 */
9221	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9222	desc1->length = CTL_WWPN_LEN;
9223	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9224	/* For testing just create the WWPN */
9225#if 0
9226	ddb_GetWWNN((char *)desc1->identifier);
9227
9228	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9229	/* This is so Copancontrol will return something sane */
9230	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9231	    ctsio->io_hdr.nexus.targ_port!=8)
9232		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9233	else
9234		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9235#endif
9236
9237	be64enc(desc1->identifier, fe->wwpn);
9238
9239	/*
9240	 * desc2 is for the Relative Target Port(type 4h) identifier
9241	 */
9242	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9243	                 | SVPD_ID_TYPE_RELTARG;
9244	desc2->length = 4;
9245//#if 0
9246	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9247	/* This is so Copancontrol will return something sane */
9248	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9249	    ctsio->io_hdr.nexus.targ_port!=8)
9250		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9251	else
9252	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9253//#endif
9254
9255	/*
9256	 * desc3 is for the Target Port Group(type 5h) identifier
9257	 */
9258	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9259	                 | SVPD_ID_TYPE_TPORTGRP;
9260	desc3->length = 4;
9261	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9262		desc3->identifier[3] = 1;
9263	else
9264		desc3->identifier[3] = 2;
9265
9266#ifdef CTL_USE_BACKEND_SN
9267	/*
9268	 * If we've actually got a backend, copy the device id from the
9269	 * per-LUN data.  Otherwise, set it to all spaces.
9270	 */
9271	if (lun != NULL) {
9272		/*
9273		 * Copy the backend's LUN ID.
9274		 */
9275		strncpy((char *)t10id->vendor_spec_id,
9276			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9277	} else {
9278		/*
9279		 * No backend, set this to spaces.
9280		 */
9281		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9282	}
9283#else
9284	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9285		 (lun != NULL) ?  (int)lun->lun : 0);
9286	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9287		sizeof(tmpstr)));
9288#endif
9289
9290	ctsio->scsi_status = SCSI_STATUS_OK;
9291
9292	ctsio->be_move_done = ctl_config_move_done;
9293	ctl_datamove((union ctl_io *)ctsio);
9294
9295	return (CTL_RETVAL_COMPLETE);
9296}
9297
9298static int
9299ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9300{
9301	struct scsi_inquiry *cdb;
9302	int alloc_len, retval;
9303
9304	cdb = (struct scsi_inquiry *)ctsio->cdb;
9305
9306	retval = CTL_RETVAL_COMPLETE;
9307
9308	alloc_len = scsi_2btoul(cdb->length);
9309
9310	switch (cdb->page_code) {
9311	case SVPD_SUPPORTED_PAGES:
9312		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9313		break;
9314	case SVPD_UNIT_SERIAL_NUMBER:
9315		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9316		break;
9317	case SVPD_DEVICE_ID:
9318		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9319		break;
9320	default:
9321		ctl_set_invalid_field(ctsio,
9322				      /*sks_valid*/ 1,
9323				      /*command*/ 1,
9324				      /*field*/ 2,
9325				      /*bit_valid*/ 0,
9326				      /*bit*/ 0);
9327		ctl_done((union ctl_io *)ctsio);
9328		retval = CTL_RETVAL_COMPLETE;
9329		break;
9330	}
9331
9332	return (retval);
9333}
9334
9335static int
9336ctl_inquiry_std(struct ctl_scsiio *ctsio)
9337{
9338	struct scsi_inquiry_data *inq_ptr;
9339	struct scsi_inquiry *cdb;
9340	struct ctl_softc *ctl_softc;
9341	struct ctl_lun *lun;
9342	uint32_t alloc_len;
9343	int is_fc;
9344
9345	ctl_softc = control_softc;
9346
9347	/*
9348	 * Figure out whether we're talking to a Fibre Channel port or not.
9349	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9350	 * SCSI front ends.
9351	 */
9352	mtx_lock(&ctl_softc->ctl_lock);
9353	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9354	    CTL_PORT_FC)
9355		is_fc = 0;
9356	else
9357		is_fc = 1;
9358	mtx_unlock(&ctl_softc->ctl_lock);
9359
9360	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9361	cdb = (struct scsi_inquiry *)ctsio->cdb;
9362	alloc_len = scsi_2btoul(cdb->length);
9363
9364	/*
9365	 * We malloc the full inquiry data size here and fill it
9366	 * in.  If the user only asks for less, we'll give him
9367	 * that much.
9368	 */
9369	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9370	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9371	ctsio->kern_sg_entries = 0;
9372	ctsio->kern_data_resid = 0;
9373	ctsio->kern_rel_offset = 0;
9374
9375	if (sizeof(*inq_ptr) < alloc_len) {
9376		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9377		ctsio->kern_data_len = sizeof(*inq_ptr);
9378		ctsio->kern_total_len = sizeof(*inq_ptr);
9379	} else {
9380		ctsio->residual = 0;
9381		ctsio->kern_data_len = alloc_len;
9382		ctsio->kern_total_len = alloc_len;
9383	}
9384
9385	/*
9386	 * If we have a LUN configured, report it as connected.  Otherwise,
9387	 * report that it is offline or no device is supported, depending
9388	 * on the value of inquiry_pq_no_lun.
9389	 *
9390	 * According to the spec (SPC-4 r34), the peripheral qualifier
9391	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9392	 *
9393	 * "A peripheral device having the specified peripheral device type
9394	 * is not connected to this logical unit. However, the device
9395	 * server is capable of supporting the specified peripheral device
9396	 * type on this logical unit."
9397	 *
9398	 * According to the same spec, the peripheral qualifier
9399	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9400	 *
9401	 * "The device server is not capable of supporting a peripheral
9402	 * device on this logical unit. For this peripheral qualifier the
9403	 * peripheral device type shall be set to 1Fh. All other peripheral
9404	 * device type values are reserved for this peripheral qualifier."
9405	 *
9406	 * Given the text, it would seem that we probably want to report that
9407	 * the LUN is offline here.  There is no LUN connected, but we can
9408	 * support a LUN at the given LUN number.
9409	 *
9410	 * In the real world, though, it sounds like things are a little
9411	 * different:
9412	 *
9413	 * - Linux, when presented with a LUN with the offline peripheral
9414	 *   qualifier, will create an sg driver instance for it.  So when
9415	 *   you attach it to CTL, you wind up with a ton of sg driver
9416	 *   instances.  (One for every LUN that Linux bothered to probe.)
9417	 *   Linux does this despite the fact that it issues a REPORT LUNs
9418	 *   to LUN 0 to get the inventory of supported LUNs.
9419	 *
9420	 * - There is other anecdotal evidence (from Emulex folks) about
9421	 *   arrays that use the offline peripheral qualifier for LUNs that
9422	 *   are on the "passive" path in an active/passive array.
9423	 *
9424	 * So the solution is provide a hopefully reasonable default
9425	 * (return bad/no LUN) and allow the user to change the behavior
9426	 * with a tunable/sysctl variable.
9427	 */
9428	if (lun != NULL)
9429		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9430				  lun->be_lun->lun_type;
9431	else if (ctl_softc->inquiry_pq_no_lun == 0)
9432		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9433	else
9434		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9435
9436	/* RMB in byte 2 is 0 */
9437	inq_ptr->version = SCSI_REV_SPC3;
9438
9439	/*
9440	 * According to SAM-3, even if a device only supports a single
9441	 * level of LUN addressing, it should still set the HISUP bit:
9442	 *
9443	 * 4.9.1 Logical unit numbers overview
9444	 *
9445	 * All logical unit number formats described in this standard are
9446	 * hierarchical in structure even when only a single level in that
9447	 * hierarchy is used. The HISUP bit shall be set to one in the
9448	 * standard INQUIRY data (see SPC-2) when any logical unit number
9449	 * format described in this standard is used.  Non-hierarchical
9450	 * formats are outside the scope of this standard.
9451	 *
9452	 * Therefore we set the HiSup bit here.
9453	 *
9454	 * The reponse format is 2, per SPC-3.
9455	 */
9456	inq_ptr->response_format = SID_HiSup | 2;
9457
9458	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9459	CTL_DEBUG_PRINT(("additional_length = %d\n",
9460			 inq_ptr->additional_length));
9461
9462	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9463	/* 16 bit addressing */
9464	if (is_fc == 0)
9465		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9466	/* XXX set the SID_MultiP bit here if we're actually going to
9467	   respond on multiple ports */
9468	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9469
9470	/* 16 bit data bus, synchronous transfers */
9471	/* XXX these flags don't apply for FC */
9472	if (is_fc == 0)
9473		inq_ptr->flags = SID_WBus16 | SID_Sync;
9474	/*
9475	 * XXX KDM do we want to support tagged queueing on the control
9476	 * device at all?
9477	 */
9478	if ((lun == NULL)
9479	 || (lun->be_lun->lun_type != T_PROCESSOR))
9480		inq_ptr->flags |= SID_CmdQue;
9481	/*
9482	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9483	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9484	 * name and 4 bytes for the revision.
9485	 */
9486	strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
9487	if (lun == NULL) {
9488		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9489	} else {
9490		switch (lun->be_lun->lun_type) {
9491		case T_DIRECT:
9492			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9493			break;
9494		case T_PROCESSOR:
9495			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9496			break;
9497		default:
9498			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9499			break;
9500		}
9501	}
9502
9503	/*
9504	 * XXX make this a macro somewhere so it automatically gets
9505	 * incremented when we make changes.
9506	 */
9507	strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9508
9509	/*
9510	 * For parallel SCSI, we support double transition and single
9511	 * transition clocking.  We also support QAS (Quick Arbitration
9512	 * and Selection) and Information Unit transfers on both the
9513	 * control and array devices.
9514	 */
9515	if (is_fc == 0)
9516		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9517				    SID_SPI_IUS;
9518
9519	/* SAM-3 */
9520	scsi_ulto2b(0x0060, inq_ptr->version1);
9521	/* SPC-3 (no version claimed) XXX should we claim a version? */
9522	scsi_ulto2b(0x0300, inq_ptr->version2);
9523	if (is_fc) {
9524		/* FCP-2 ANSI INCITS.350:2003 */
9525		scsi_ulto2b(0x0917, inq_ptr->version3);
9526	} else {
9527		/* SPI-4 ANSI INCITS.362:200x */
9528		scsi_ulto2b(0x0B56, inq_ptr->version3);
9529	}
9530
9531	if (lun == NULL) {
9532		/* SBC-2 (no version claimed) XXX should we claim a version? */
9533		scsi_ulto2b(0x0320, inq_ptr->version4);
9534	} else {
9535		switch (lun->be_lun->lun_type) {
9536		case T_DIRECT:
9537			/*
9538			 * SBC-2 (no version claimed) XXX should we claim a
9539			 * version?
9540			 */
9541			scsi_ulto2b(0x0320, inq_ptr->version4);
9542			break;
9543		case T_PROCESSOR:
9544		default:
9545			break;
9546		}
9547	}
9548
9549	ctsio->scsi_status = SCSI_STATUS_OK;
9550	if (ctsio->kern_data_len > 0) {
9551		ctsio->be_move_done = ctl_config_move_done;
9552		ctl_datamove((union ctl_io *)ctsio);
9553	} else {
9554		ctsio->io_hdr.status = CTL_SUCCESS;
9555		ctl_done((union ctl_io *)ctsio);
9556	}
9557
9558	return (CTL_RETVAL_COMPLETE);
9559}
9560
9561int
9562ctl_inquiry(struct ctl_scsiio *ctsio)
9563{
9564	struct scsi_inquiry *cdb;
9565	int retval;
9566
9567	cdb = (struct scsi_inquiry *)ctsio->cdb;
9568
9569	retval = 0;
9570
9571	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
9572
9573	/*
9574	 * Right now, we don't support the CmdDt inquiry information.
9575	 * This would be nice to support in the future.  When we do
9576	 * support it, we should change this test so that it checks to make
9577	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
9578	 */
9579#ifdef notyet
9580	if (((cdb->byte2 & SI_EVPD)
9581	 && (cdb->byte2 & SI_CMDDT)))
9582#endif
9583	if (cdb->byte2 & SI_CMDDT) {
9584		/*
9585		 * Point to the SI_CMDDT bit.  We might change this
9586		 * when we support SI_CMDDT, but since both bits would be
9587		 * "wrong", this should probably just stay as-is then.
9588		 */
9589		ctl_set_invalid_field(ctsio,
9590				      /*sks_valid*/ 1,
9591				      /*command*/ 1,
9592				      /*field*/ 1,
9593				      /*bit_valid*/ 1,
9594				      /*bit*/ 1);
9595		ctl_done((union ctl_io *)ctsio);
9596		return (CTL_RETVAL_COMPLETE);
9597	}
9598	if (cdb->byte2 & SI_EVPD)
9599		retval = ctl_inquiry_evpd(ctsio);
9600#ifdef notyet
9601	else if (cdb->byte2 & SI_CMDDT)
9602		retval = ctl_inquiry_cmddt(ctsio);
9603#endif
9604	else
9605		retval = ctl_inquiry_std(ctsio);
9606
9607	return (retval);
9608}
9609
9610/*
9611 * For known CDB types, parse the LBA and length.
9612 */
9613static int
9614ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
9615{
9616	if (io->io_hdr.io_type != CTL_IO_SCSI)
9617		return (1);
9618
9619	switch (io->scsiio.cdb[0]) {
9620	case READ_6:
9621	case WRITE_6: {
9622		struct scsi_rw_6 *cdb;
9623
9624		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
9625
9626		*lba = scsi_3btoul(cdb->addr);
9627		/* only 5 bits are valid in the most significant address byte */
9628		*lba &= 0x1fffff;
9629		*len = cdb->length;
9630		break;
9631	}
9632	case READ_10:
9633	case WRITE_10: {
9634		struct scsi_rw_10 *cdb;
9635
9636		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
9637
9638		*lba = scsi_4btoul(cdb->addr);
9639		*len = scsi_2btoul(cdb->length);
9640		break;
9641	}
9642	case WRITE_VERIFY_10: {
9643		struct scsi_write_verify_10 *cdb;
9644
9645		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
9646
9647		*lba = scsi_4btoul(cdb->addr);
9648		*len = scsi_2btoul(cdb->length);
9649		break;
9650	}
9651	case READ_12:
9652	case WRITE_12: {
9653		struct scsi_rw_12 *cdb;
9654
9655		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
9656
9657		*lba = scsi_4btoul(cdb->addr);
9658		*len = scsi_4btoul(cdb->length);
9659		break;
9660	}
9661	case WRITE_VERIFY_12: {
9662		struct scsi_write_verify_12 *cdb;
9663
9664		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
9665
9666		*lba = scsi_4btoul(cdb->addr);
9667		*len = scsi_4btoul(cdb->length);
9668		break;
9669	}
9670	case READ_16:
9671	case WRITE_16: {
9672		struct scsi_rw_16 *cdb;
9673
9674		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
9675
9676		*lba = scsi_8btou64(cdb->addr);
9677		*len = scsi_4btoul(cdb->length);
9678		break;
9679	}
9680	case WRITE_VERIFY_16: {
9681		struct scsi_write_verify_16 *cdb;
9682
9683		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
9684
9685
9686		*lba = scsi_8btou64(cdb->addr);
9687		*len = scsi_4btoul(cdb->length);
9688		break;
9689	}
9690	default:
9691		return (1);
9692		break; /* NOTREACHED */
9693	}
9694
9695	return (0);
9696}
9697
9698static ctl_action
9699ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
9700{
9701	uint64_t endlba1, endlba2;
9702
9703	endlba1 = lba1 + len1 - 1;
9704	endlba2 = lba2 + len2 - 1;
9705
9706	if ((endlba1 < lba2)
9707	 || (endlba2 < lba1))
9708		return (CTL_ACTION_PASS);
9709	else
9710		return (CTL_ACTION_BLOCK);
9711}
9712
9713static ctl_action
9714ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
9715{
9716	uint64_t lba1, lba2;
9717	uint32_t len1, len2;
9718	int retval;
9719
9720	retval = ctl_get_lba_len(io1, &lba1, &len1);
9721	if (retval != 0)
9722		return (CTL_ACTION_ERROR);
9723
9724	retval = ctl_get_lba_len(io2, &lba2, &len2);
9725	if (retval != 0)
9726		return (CTL_ACTION_ERROR);
9727
9728	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
9729}
9730
9731static ctl_action
9732ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
9733{
9734	struct ctl_cmd_entry *pending_entry, *ooa_entry;
9735	ctl_serialize_action *serialize_row;
9736
9737	/*
9738	 * The initiator attempted multiple untagged commands at the same
9739	 * time.  Can't do that.
9740	 */
9741	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9742	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9743	 && ((pending_io->io_hdr.nexus.targ_port ==
9744	      ooa_io->io_hdr.nexus.targ_port)
9745	  && (pending_io->io_hdr.nexus.initid.id ==
9746	      ooa_io->io_hdr.nexus.initid.id))
9747	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9748		return (CTL_ACTION_OVERLAP);
9749
9750	/*
9751	 * The initiator attempted to send multiple tagged commands with
9752	 * the same ID.  (It's fine if different initiators have the same
9753	 * tag ID.)
9754	 *
9755	 * Even if all of those conditions are true, we don't kill the I/O
9756	 * if the command ahead of us has been aborted.  We won't end up
9757	 * sending it to the FETD, and it's perfectly legal to resend a
9758	 * command with the same tag number as long as the previous
9759	 * instance of this tag number has been aborted somehow.
9760	 */
9761	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9762	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9763	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
9764	 && ((pending_io->io_hdr.nexus.targ_port ==
9765	      ooa_io->io_hdr.nexus.targ_port)
9766	  && (pending_io->io_hdr.nexus.initid.id ==
9767	      ooa_io->io_hdr.nexus.initid.id))
9768	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9769		return (CTL_ACTION_OVERLAP_TAG);
9770
9771	/*
9772	 * If we get a head of queue tag, SAM-3 says that we should
9773	 * immediately execute it.
9774	 *
9775	 * What happens if this command would normally block for some other
9776	 * reason?  e.g. a request sense with a head of queue tag
9777	 * immediately after a write.  Normally that would block, but this
9778	 * will result in its getting executed immediately...
9779	 *
9780	 * We currently return "pass" instead of "skip", so we'll end up
9781	 * going through the rest of the queue to check for overlapped tags.
9782	 *
9783	 * XXX KDM check for other types of blockage first??
9784	 */
9785	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9786		return (CTL_ACTION_PASS);
9787
9788	/*
9789	 * Ordered tags have to block until all items ahead of them
9790	 * have completed.  If we get called with an ordered tag, we always
9791	 * block, if something else is ahead of us in the queue.
9792	 */
9793	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
9794		return (CTL_ACTION_BLOCK);
9795
9796	/*
9797	 * Simple tags get blocked until all head of queue and ordered tags
9798	 * ahead of them have completed.  I'm lumping untagged commands in
9799	 * with simple tags here.  XXX KDM is that the right thing to do?
9800	 */
9801	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9802	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
9803	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9804	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
9805		return (CTL_ACTION_BLOCK);
9806
9807	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
9808	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
9809
9810	serialize_row = ctl_serialize_table[ooa_entry->seridx];
9811
9812	switch (serialize_row[pending_entry->seridx]) {
9813	case CTL_SER_BLOCK:
9814		return (CTL_ACTION_BLOCK);
9815		break; /* NOTREACHED */
9816	case CTL_SER_EXTENT:
9817		return (ctl_extent_check(pending_io, ooa_io));
9818		break; /* NOTREACHED */
9819	case CTL_SER_PASS:
9820		return (CTL_ACTION_PASS);
9821		break; /* NOTREACHED */
9822	case CTL_SER_SKIP:
9823		return (CTL_ACTION_SKIP);
9824		break;
9825	default:
9826		panic("invalid serialization value %d",
9827		      serialize_row[pending_entry->seridx]);
9828		break; /* NOTREACHED */
9829	}
9830
9831	return (CTL_ACTION_ERROR);
9832}
9833
9834/*
9835 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
9836 * Assumptions:
9837 * - pending_io is generally either incoming, or on the blocked queue
9838 * - starting I/O is the I/O we want to start the check with.
9839 */
9840static ctl_action
9841ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
9842	      union ctl_io *starting_io)
9843{
9844	union ctl_io *ooa_io;
9845	ctl_action action;
9846
9847	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
9848
9849	/*
9850	 * Run back along the OOA queue, starting with the current
9851	 * blocked I/O and going through every I/O before it on the
9852	 * queue.  If starting_io is NULL, we'll just end up returning
9853	 * CTL_ACTION_PASS.
9854	 */
9855	for (ooa_io = starting_io; ooa_io != NULL;
9856	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
9857	     ooa_links)){
9858
9859		/*
9860		 * This routine just checks to see whether
9861		 * cur_blocked is blocked by ooa_io, which is ahead
9862		 * of it in the queue.  It doesn't queue/dequeue
9863		 * cur_blocked.
9864		 */
9865		action = ctl_check_for_blockage(pending_io, ooa_io);
9866		switch (action) {
9867		case CTL_ACTION_BLOCK:
9868		case CTL_ACTION_OVERLAP:
9869		case CTL_ACTION_OVERLAP_TAG:
9870		case CTL_ACTION_SKIP:
9871		case CTL_ACTION_ERROR:
9872			return (action);
9873			break; /* NOTREACHED */
9874		case CTL_ACTION_PASS:
9875			break;
9876		default:
9877			panic("invalid action %d", action);
9878			break;  /* NOTREACHED */
9879		}
9880	}
9881
9882	return (CTL_ACTION_PASS);
9883}
9884
9885/*
9886 * Assumptions:
9887 * - An I/O has just completed, and has been removed from the per-LUN OOA
9888 *   queue, so some items on the blocked queue may now be unblocked.
9889 */
9890static int
9891ctl_check_blocked(struct ctl_lun *lun)
9892{
9893	union ctl_io *cur_blocked, *next_blocked;
9894
9895	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
9896
9897	/*
9898	 * Run forward from the head of the blocked queue, checking each
9899	 * entry against the I/Os prior to it on the OOA queue to see if
9900	 * there is still any blockage.
9901	 *
9902	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
9903	 * with our removing a variable on it while it is traversing the
9904	 * list.
9905	 */
9906	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
9907	     cur_blocked != NULL; cur_blocked = next_blocked) {
9908		union ctl_io *prev_ooa;
9909		ctl_action action;
9910
9911		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
9912							  blocked_links);
9913
9914		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
9915						      ctl_ooaq, ooa_links);
9916
9917		/*
9918		 * If cur_blocked happens to be the first item in the OOA
9919		 * queue now, prev_ooa will be NULL, and the action
9920		 * returned will just be CTL_ACTION_PASS.
9921		 */
9922		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
9923
9924		switch (action) {
9925		case CTL_ACTION_BLOCK:
9926			/* Nothing to do here, still blocked */
9927			break;
9928		case CTL_ACTION_OVERLAP:
9929		case CTL_ACTION_OVERLAP_TAG:
9930			/*
9931			 * This shouldn't happen!  In theory we've already
9932			 * checked this command for overlap...
9933			 */
9934			break;
9935		case CTL_ACTION_PASS:
9936		case CTL_ACTION_SKIP: {
9937			struct ctl_softc *softc;
9938			struct ctl_cmd_entry *entry;
9939			uint32_t initidx;
9940			uint8_t opcode;
9941			int isc_retval;
9942
9943			/*
9944			 * The skip case shouldn't happen, this transaction
9945			 * should have never made it onto the blocked queue.
9946			 */
9947			/*
9948			 * This I/O is no longer blocked, we can remove it
9949			 * from the blocked queue.  Since this is a TAILQ
9950			 * (doubly linked list), we can do O(1) removals
9951			 * from any place on the list.
9952			 */
9953			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
9954				     blocked_links);
9955			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
9956
9957			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
9958				/*
9959				 * Need to send IO back to original side to
9960				 * run
9961				 */
9962				union ctl_ha_msg msg_info;
9963
9964				msg_info.hdr.original_sc =
9965					cur_blocked->io_hdr.original_sc;
9966				msg_info.hdr.serializing_sc = cur_blocked;
9967				msg_info.hdr.msg_type = CTL_MSG_R2R;
9968				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
9969				     &msg_info, sizeof(msg_info), 0)) >
9970				     CTL_HA_STATUS_SUCCESS) {
9971					printf("CTL:Check Blocked error from "
9972					       "ctl_ha_msg_send %d\n",
9973					       isc_retval);
9974				}
9975				break;
9976			}
9977			opcode = cur_blocked->scsiio.cdb[0];
9978			entry = &ctl_cmd_table[opcode];
9979			softc = control_softc;
9980
9981			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
9982
9983			/*
9984			 * Check this I/O for LUN state changes that may
9985			 * have happened while this command was blocked.
9986			 * The LUN state may have been changed by a command
9987			 * ahead of us in the queue, so we need to re-check
9988			 * for any states that can be caused by SCSI
9989			 * commands.
9990			 */
9991			if (ctl_scsiio_lun_check(softc, lun, entry,
9992						 &cur_blocked->scsiio) == 0) {
9993				cur_blocked->io_hdr.flags |=
9994				                      CTL_FLAG_IS_WAS_ON_RTR;
9995				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
9996						   &cur_blocked->io_hdr, links);
9997				/*
9998				 * In the non CTL_DONE_THREAD case, we need
9999				 * to wake up the work thread here.  When
10000				 * we're processing completed requests from
10001				 * the work thread context, we'll pop back
10002				 * around and end up pulling things off the
10003				 * RtR queue.  When we aren't processing
10004				 * things from the work thread context,
10005				 * though, we won't ever check the RtR queue.
10006				 * So we need to wake up the thread to clear
10007				 * things off the queue.  Otherwise this
10008				 * transaction will just sit on the RtR queue
10009				 * until a new I/O comes in.  (Which may or
10010				 * may not happen...)
10011				 */
10012#ifndef CTL_DONE_THREAD
10013				ctl_wakeup_thread();
10014#endif
10015			} else
10016				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
10017			break;
10018		}
10019		default:
10020			/*
10021			 * This probably shouldn't happen -- we shouldn't
10022			 * get CTL_ACTION_ERROR, or anything else.
10023			 */
10024			break;
10025		}
10026	}
10027
10028	return (CTL_RETVAL_COMPLETE);
10029}
10030
10031/*
10032 * This routine (with one exception) checks LUN flags that can be set by
10033 * commands ahead of us in the OOA queue.  These flags have to be checked
10034 * when a command initially comes in, and when we pull a command off the
10035 * blocked queue and are preparing to execute it.  The reason we have to
10036 * check these flags for commands on the blocked queue is that the LUN
10037 * state may have been changed by a command ahead of us while we're on the
10038 * blocked queue.
10039 *
10040 * Ordering is somewhat important with these checks, so please pay
10041 * careful attention to the placement of any new checks.
10042 */
10043static int
10044ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10045		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10046{
10047	int retval;
10048
10049	retval = 0;
10050
10051	/*
10052	 * If this shelf is a secondary shelf controller, we have to reject
10053	 * any media access commands.
10054	 */
10055#if 0
10056	/* No longer needed for HA */
10057	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10058	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10059		ctl_set_lun_standby(ctsio);
10060		retval = 1;
10061		goto bailout;
10062	}
10063#endif
10064
10065	/*
10066	 * Check for a reservation conflict.  If this command isn't allowed
10067	 * even on reserved LUNs, and if this initiator isn't the one who
10068	 * reserved us, reject the command with a reservation conflict.
10069	 */
10070	if ((lun->flags & CTL_LUN_RESERVED)
10071	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10072		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10073		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10074		 || (ctsio->io_hdr.nexus.targ_target.id !=
10075		     lun->rsv_nexus.targ_target.id)) {
10076			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10077			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10078			retval = 1;
10079			goto bailout;
10080		}
10081	}
10082
10083	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10084	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10085		uint32_t residx;
10086
10087		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10088		/*
10089		 * if we aren't registered or it's a res holder type
10090		 * reservation and this isn't the res holder then set a
10091		 * conflict.
10092		 * NOTE: Commands which might be allowed on write exclusive
10093		 * type reservations are checked in the particular command
10094		 * for a conflict. Read and SSU are the only ones.
10095		 */
10096		if (!lun->per_res[residx].registered
10097		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10098			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10099			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10100			retval = 1;
10101			goto bailout;
10102		}
10103
10104	}
10105
10106	if ((lun->flags & CTL_LUN_OFFLINE)
10107	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10108		ctl_set_lun_not_ready(ctsio);
10109		retval = 1;
10110		goto bailout;
10111	}
10112
10113	/*
10114	 * If the LUN is stopped, see if this particular command is allowed
10115	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10116	 */
10117	if ((lun->flags & CTL_LUN_STOPPED)
10118	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10119		/* "Logical unit not ready, initializing cmd. required" */
10120		ctl_set_lun_stopped(ctsio);
10121		retval = 1;
10122		goto bailout;
10123	}
10124
10125	if ((lun->flags & CTL_LUN_INOPERABLE)
10126	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10127		/* "Medium format corrupted" */
10128		ctl_set_medium_format_corrupted(ctsio);
10129		retval = 1;
10130		goto bailout;
10131	}
10132
10133bailout:
10134	return (retval);
10135
10136}
10137
10138static void
10139ctl_failover_io(union ctl_io *io, int have_lock)
10140{
10141	ctl_set_busy(&io->scsiio);
10142	ctl_done_lock(io, have_lock);
10143}
10144
10145static void
10146ctl_failover(void)
10147{
10148	struct ctl_lun *lun;
10149	struct ctl_softc *ctl_softc;
10150	union ctl_io *next_io, *pending_io;
10151	union ctl_io *io;
10152	int lun_idx;
10153	int i;
10154
10155	ctl_softc = control_softc;
10156
10157	mtx_lock(&ctl_softc->ctl_lock);
10158	/*
10159	 * Remove any cmds from the other SC from the rtr queue.  These
10160	 * will obviously only be for LUNs for which we're the primary.
10161	 * We can't send status or get/send data for these commands.
10162	 * Since they haven't been executed yet, we can just remove them.
10163	 * We'll either abort them or delete them below, depending on
10164	 * which HA mode we're in.
10165	 */
10166	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10167	     io != NULL; io = next_io) {
10168		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10169		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10170			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10171				      ctl_io_hdr, links);
10172	}
10173
10174	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10175		lun = ctl_softc->ctl_luns[lun_idx];
10176		if (lun==NULL)
10177			continue;
10178
10179		/*
10180		 * Processor LUNs are primary on both sides.
10181		 * XXX will this always be true?
10182		 */
10183		if (lun->be_lun->lun_type == T_PROCESSOR)
10184			continue;
10185
10186		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10187		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10188			printf("FAILOVER: primary lun %d\n", lun_idx);
10189		        /*
10190			 * Remove all commands from the other SC. First from the
10191			 * blocked queue then from the ooa queue. Once we have
10192			 * removed them. Call ctl_check_blocked to see if there
10193			 * is anything that can run.
10194			 */
10195			for (io = (union ctl_io *)TAILQ_FIRST(
10196			     &lun->blocked_queue); io != NULL; io = next_io) {
10197
10198		        	next_io = (union ctl_io *)TAILQ_NEXT(
10199				    &io->io_hdr, blocked_links);
10200
10201				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10202					TAILQ_REMOVE(&lun->blocked_queue,
10203						     &io->io_hdr,blocked_links);
10204					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10205					TAILQ_REMOVE(&lun->ooa_queue,
10206						     &io->io_hdr, ooa_links);
10207
10208					ctl_free_io(io);
10209				}
10210			}
10211
10212			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10213	     		     io != NULL; io = next_io) {
10214
10215		        	next_io = (union ctl_io *)TAILQ_NEXT(
10216				    &io->io_hdr, ooa_links);
10217
10218				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10219
10220					TAILQ_REMOVE(&lun->ooa_queue,
10221						&io->io_hdr,
10222					     	ooa_links);
10223
10224					ctl_free_io(io);
10225				}
10226			}
10227			ctl_check_blocked(lun);
10228		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10229			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10230
10231			printf("FAILOVER: primary lun %d\n", lun_idx);
10232			/*
10233			 * Abort all commands from the other SC.  We can't
10234			 * send status back for them now.  These should get
10235			 * cleaned up when they are completed or come out
10236			 * for a datamove operation.
10237			 */
10238			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10239	     		     io != NULL; io = next_io) {
10240		        	next_io = (union ctl_io *)TAILQ_NEXT(
10241					&io->io_hdr, ooa_links);
10242
10243				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10244					io->io_hdr.flags |= CTL_FLAG_ABORT;
10245			}
10246		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10247			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10248
10249			printf("FAILOVER: secondary lun %d\n", lun_idx);
10250
10251			lun->flags |= CTL_LUN_PRIMARY_SC;
10252
10253			/*
10254			 * We send all I/O that was sent to this controller
10255			 * and redirected to the other side back with
10256			 * busy status, and have the initiator retry it.
10257			 * Figuring out how much data has been transferred,
10258			 * etc. and picking up where we left off would be
10259			 * very tricky.
10260			 *
10261			 * XXX KDM need to remove I/O from the blocked
10262			 * queue as well!
10263			 */
10264			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10265			     &lun->ooa_queue); pending_io != NULL;
10266			     pending_io = next_io) {
10267
10268				next_io =  (union ctl_io *)TAILQ_NEXT(
10269					&pending_io->io_hdr, ooa_links);
10270
10271				pending_io->io_hdr.flags &=
10272					~CTL_FLAG_SENT_2OTHER_SC;
10273
10274				if (pending_io->io_hdr.flags &
10275				    CTL_FLAG_IO_ACTIVE) {
10276					pending_io->io_hdr.flags |=
10277						CTL_FLAG_FAILOVER;
10278				} else {
10279					ctl_set_busy(&pending_io->scsiio);
10280					ctl_done_lock(pending_io,
10281						      /*have_lock*/1);
10282				}
10283			}
10284
10285			/*
10286			 * Build Unit Attention
10287			 */
10288			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10289				lun->pending_sense[i].ua_pending |=
10290				                     CTL_UA_ASYM_ACC_CHANGE;
10291			}
10292		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10293			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10294			printf("FAILOVER: secondary lun %d\n", lun_idx);
10295			/*
10296			 * if the first io on the OOA is not on the RtR queue
10297			 * add it.
10298			 */
10299			lun->flags |= CTL_LUN_PRIMARY_SC;
10300
10301			pending_io = (union ctl_io *)TAILQ_FIRST(
10302			    &lun->ooa_queue);
10303			if (pending_io==NULL) {
10304				printf("Nothing on OOA queue\n");
10305				continue;
10306			}
10307
10308			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10309			if ((pending_io->io_hdr.flags &
10310			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10311				pending_io->io_hdr.flags |=
10312				    CTL_FLAG_IS_WAS_ON_RTR;
10313				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10314						   &pending_io->io_hdr, links);
10315			}
10316#if 0
10317			else
10318			{
10319				printf("Tag 0x%04x is running\n",
10320				      pending_io->scsiio.tag_num);
10321			}
10322#endif
10323
10324			next_io = (union ctl_io *)TAILQ_NEXT(
10325			    &pending_io->io_hdr, ooa_links);
10326			for (pending_io=next_io; pending_io != NULL;
10327			     pending_io = next_io) {
10328				pending_io->io_hdr.flags &=
10329				    ~CTL_FLAG_SENT_2OTHER_SC;
10330				next_io = (union ctl_io *)TAILQ_NEXT(
10331					&pending_io->io_hdr, ooa_links);
10332				if (pending_io->io_hdr.flags &
10333				    CTL_FLAG_IS_WAS_ON_RTR) {
10334#if 0
10335				        printf("Tag 0x%04x is running\n",
10336				      		pending_io->scsiio.tag_num);
10337#endif
10338					continue;
10339				}
10340
10341				switch (ctl_check_ooa(lun, pending_io,
10342			            (union ctl_io *)TAILQ_PREV(
10343				    &pending_io->io_hdr, ctl_ooaq,
10344				    ooa_links))) {
10345
10346				case CTL_ACTION_BLOCK:
10347					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10348							  &pending_io->io_hdr,
10349							  blocked_links);
10350					pending_io->io_hdr.flags |=
10351					    CTL_FLAG_BLOCKED;
10352					break;
10353				case CTL_ACTION_PASS:
10354				case CTL_ACTION_SKIP:
10355					pending_io->io_hdr.flags |=
10356					    CTL_FLAG_IS_WAS_ON_RTR;
10357					STAILQ_INSERT_TAIL(
10358					    &ctl_softc->rtr_queue,
10359					    &pending_io->io_hdr, links);
10360					break;
10361				case CTL_ACTION_OVERLAP:
10362					ctl_set_overlapped_cmd(
10363					    (struct ctl_scsiio *)pending_io);
10364					ctl_done_lock(pending_io,
10365						      /*have_lock*/ 1);
10366					break;
10367				case CTL_ACTION_OVERLAP_TAG:
10368					ctl_set_overlapped_tag(
10369					    (struct ctl_scsiio *)pending_io,
10370					    pending_io->scsiio.tag_num & 0xff);
10371					ctl_done_lock(pending_io,
10372						      /*have_lock*/ 1);
10373					break;
10374				case CTL_ACTION_ERROR:
10375				default:
10376					ctl_set_internal_failure(
10377						(struct ctl_scsiio *)pending_io,
10378						0,  // sks_valid
10379						0); //retry count
10380					ctl_done_lock(pending_io,
10381						      /*have_lock*/ 1);
10382					break;
10383				}
10384			}
10385
10386			/*
10387			 * Build Unit Attention
10388			 */
10389			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10390				lun->pending_sense[i].ua_pending |=
10391				                     CTL_UA_ASYM_ACC_CHANGE;
10392			}
10393		} else {
10394			panic("Unhandled HA mode failover, LUN flags = %#x, "
10395			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10396		}
10397	}
10398	ctl_pause_rtr = 0;
10399	mtx_unlock(&ctl_softc->ctl_lock);
10400}
10401
10402static int
10403ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10404{
10405	struct ctl_lun *lun;
10406	struct ctl_cmd_entry *entry;
10407	uint8_t opcode;
10408	uint32_t initidx, targ_lun;
10409	int retval;
10410
10411	retval = 0;
10412
10413	lun = NULL;
10414
10415	opcode = ctsio->cdb[0];
10416
10417	mtx_lock(&ctl_softc->ctl_lock);
10418
10419	targ_lun = ctsio->io_hdr.nexus.targ_lun;
10420	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
10421		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
10422	if ((targ_lun < CTL_MAX_LUNS)
10423	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
10424		lun = ctl_softc->ctl_luns[targ_lun];
10425		/*
10426		 * If the LUN is invalid, pretend that it doesn't exist.
10427		 * It will go away as soon as all pending I/O has been
10428		 * completed.
10429		 */
10430		if (lun->flags & CTL_LUN_DISABLED) {
10431			lun = NULL;
10432		} else {
10433			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10434			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10435				lun->be_lun;
10436			if (lun->be_lun->lun_type == T_PROCESSOR) {
10437				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10438			}
10439		}
10440	} else {
10441		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10442		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10443	}
10444
10445	entry = &ctl_cmd_table[opcode];
10446
10447	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10448	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10449
10450	/*
10451	 * Check to see whether we can send this command to LUNs that don't
10452	 * exist.  This should pretty much only be the case for inquiry
10453	 * and request sense.  Further checks, below, really require having
10454	 * a LUN, so we can't really check the command anymore.  Just put
10455	 * it on the rtr queue.
10456	 */
10457	if (lun == NULL) {
10458		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10459			goto queue_rtr;
10460
10461		ctl_set_unsupported_lun(ctsio);
10462		mtx_unlock(&ctl_softc->ctl_lock);
10463		ctl_done((union ctl_io *)ctsio);
10464		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
10465		goto bailout;
10466	} else {
10467		/*
10468		 * Every I/O goes into the OOA queue for a particular LUN, and
10469		 * stays there until completion.
10470		 */
10471		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
10472
10473		/*
10474		 * Make sure we support this particular command on this LUN.
10475		 * e.g., we don't support writes to the control LUN.
10476		 */
10477		switch (lun->be_lun->lun_type) {
10478		case T_PROCESSOR:
10479		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
10480			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10481			      == 0)) {
10482				ctl_set_invalid_opcode(ctsio);
10483				mtx_unlock(&ctl_softc->ctl_lock);
10484				ctl_done((union ctl_io *)ctsio);
10485				goto bailout;
10486			}
10487			break;
10488		case T_DIRECT:
10489			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
10490			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10491			      == 0)){
10492				ctl_set_invalid_opcode(ctsio);
10493				mtx_unlock(&ctl_softc->ctl_lock);
10494				ctl_done((union ctl_io *)ctsio);
10495				goto bailout;
10496			}
10497			break;
10498		default:
10499			printf("Unsupported CTL LUN type %d\n",
10500			       lun->be_lun->lun_type);
10501			panic("Unsupported CTL LUN type %d\n",
10502			      lun->be_lun->lun_type);
10503			break; /* NOTREACHED */
10504		}
10505	}
10506
10507	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10508
10509	/*
10510	 * If we've got a request sense, it'll clear the contingent
10511	 * allegiance condition.  Otherwise, if we have a CA condition for
10512	 * this initiator, clear it, because it sent down a command other
10513	 * than request sense.
10514	 */
10515	if ((opcode != REQUEST_SENSE)
10516	 && (ctl_is_set(lun->have_ca, initidx)))
10517		ctl_clear_mask(lun->have_ca, initidx);
10518
10519	/*
10520	 * If the command has this flag set, it handles its own unit
10521	 * attention reporting, we shouldn't do anything.  Otherwise we
10522	 * check for any pending unit attentions, and send them back to the
10523	 * initiator.  We only do this when a command initially comes in,
10524	 * not when we pull it off the blocked queue.
10525	 *
10526	 * According to SAM-3, section 5.3.2, the order that things get
10527	 * presented back to the host is basically unit attentions caused
10528	 * by some sort of reset event, busy status, reservation conflicts
10529	 * or task set full, and finally any other status.
10530	 *
10531	 * One issue here is that some of the unit attentions we report
10532	 * don't fall into the "reset" category (e.g. "reported luns data
10533	 * has changed").  So reporting it here, before the reservation
10534	 * check, may be technically wrong.  I guess the only thing to do
10535	 * would be to check for and report the reset events here, and then
10536	 * check for the other unit attention types after we check for a
10537	 * reservation conflict.
10538	 *
10539	 * XXX KDM need to fix this
10540	 */
10541	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
10542		ctl_ua_type ua_type;
10543
10544		ua_type = lun->pending_sense[initidx].ua_pending;
10545		if (ua_type != CTL_UA_NONE) {
10546			scsi_sense_data_type sense_format;
10547
10548			if (lun != NULL)
10549				sense_format = (lun->flags &
10550				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
10551				    SSD_TYPE_FIXED;
10552			else
10553				sense_format = SSD_TYPE_FIXED;
10554
10555			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
10556					       sense_format);
10557			if (ua_type != CTL_UA_NONE) {
10558				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
10559				ctsio->io_hdr.status = CTL_SCSI_ERROR |
10560						       CTL_AUTOSENSE;
10561				ctsio->sense_len = SSD_FULL_SIZE;
10562				lun->pending_sense[initidx].ua_pending &=
10563					~ua_type;
10564				mtx_unlock(&ctl_softc->ctl_lock);
10565				ctl_done((union ctl_io *)ctsio);
10566				goto bailout;
10567			}
10568		}
10569	}
10570
10571
10572	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
10573		mtx_unlock(&ctl_softc->ctl_lock);
10574		ctl_done((union ctl_io *)ctsio);
10575		goto bailout;
10576	}
10577
10578	/*
10579	 * XXX CHD this is where we want to send IO to other side if
10580	 * this LUN is secondary on this SC. We will need to make a copy
10581	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
10582	 * the copy we send as FROM_OTHER.
10583	 * We also need to stuff the address of the original IO so we can
10584	 * find it easily. Something similar will need be done on the other
10585	 * side so when we are done we can find the copy.
10586	 */
10587	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10588		union ctl_ha_msg msg_info;
10589		int isc_retval;
10590
10591		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10592
10593		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
10594		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
10595#if 0
10596		printf("1. ctsio %p\n", ctsio);
10597#endif
10598		msg_info.hdr.serializing_sc = NULL;
10599		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
10600		msg_info.scsi.tag_num = ctsio->tag_num;
10601		msg_info.scsi.tag_type = ctsio->tag_type;
10602		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
10603
10604		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10605
10606		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10607		    (void *)&msg_info, sizeof(msg_info), 0)) >
10608		    CTL_HA_STATUS_SUCCESS) {
10609			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
10610			       isc_retval);
10611			printf("CTL:opcode is %x\n",opcode);
10612		} else {
10613#if 0
10614			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
10615#endif
10616		}
10617
10618		/*
10619		 * XXX KDM this I/O is off the incoming queue, but hasn't
10620		 * been inserted on any other queue.  We may need to come
10621		 * up with a holding queue while we wait for serialization
10622		 * so that we have an idea of what we're waiting for from
10623		 * the other side.
10624		 */
10625		goto bailout_unlock;
10626	}
10627
10628	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
10629			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
10630			      ctl_ooaq, ooa_links))) {
10631	case CTL_ACTION_BLOCK:
10632		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
10633		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
10634				  blocked_links);
10635		goto bailout_unlock;
10636		break; /* NOTREACHED */
10637	case CTL_ACTION_PASS:
10638	case CTL_ACTION_SKIP:
10639		goto queue_rtr;
10640		break; /* NOTREACHED */
10641	case CTL_ACTION_OVERLAP:
10642		ctl_set_overlapped_cmd(ctsio);
10643		mtx_unlock(&ctl_softc->ctl_lock);
10644		ctl_done((union ctl_io *)ctsio);
10645		goto bailout;
10646		break; /* NOTREACHED */
10647	case CTL_ACTION_OVERLAP_TAG:
10648		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
10649		mtx_unlock(&ctl_softc->ctl_lock);
10650		ctl_done((union ctl_io *)ctsio);
10651		goto bailout;
10652		break; /* NOTREACHED */
10653	case CTL_ACTION_ERROR:
10654	default:
10655		ctl_set_internal_failure(ctsio,
10656					 /*sks_valid*/ 0,
10657					 /*retry_count*/ 0);
10658		mtx_unlock(&ctl_softc->ctl_lock);
10659		ctl_done((union ctl_io *)ctsio);
10660		goto bailout;
10661		break; /* NOTREACHED */
10662	}
10663
10664	goto bailout_unlock;
10665
10666queue_rtr:
10667	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10668	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
10669
10670bailout_unlock:
10671	mtx_unlock(&ctl_softc->ctl_lock);
10672
10673bailout:
10674	return (retval);
10675}
10676
10677static int
10678ctl_scsiio(struct ctl_scsiio *ctsio)
10679{
10680	int retval;
10681	struct ctl_cmd_entry *entry;
10682
10683	retval = CTL_RETVAL_COMPLETE;
10684
10685	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
10686
10687	entry = &ctl_cmd_table[ctsio->cdb[0]];
10688
10689	/*
10690	 * If this I/O has been aborted, just send it straight to
10691	 * ctl_done() without executing it.
10692	 */
10693	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
10694		ctl_done((union ctl_io *)ctsio);
10695		goto bailout;
10696	}
10697
10698	/*
10699	 * All the checks should have been handled by ctl_scsiio_precheck().
10700	 * We should be clear now to just execute the I/O.
10701	 */
10702	retval = entry->execute(ctsio);
10703
10704bailout:
10705	return (retval);
10706}
10707
10708/*
10709 * Since we only implement one target right now, a bus reset simply resets
10710 * our single target.
10711 */
10712static int
10713ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
10714{
10715	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
10716}
10717
10718static int
10719ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
10720		 ctl_ua_type ua_type)
10721{
10722	struct ctl_lun *lun;
10723	int retval;
10724
10725	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
10726		union ctl_ha_msg msg_info;
10727
10728		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10729		msg_info.hdr.nexus = io->io_hdr.nexus;
10730		if (ua_type==CTL_UA_TARG_RESET)
10731			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
10732		else
10733			msg_info.task.task_action = CTL_TASK_BUS_RESET;
10734		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
10735		msg_info.hdr.original_sc = NULL;
10736		msg_info.hdr.serializing_sc = NULL;
10737		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10738		    (void *)&msg_info, sizeof(msg_info), 0)) {
10739		}
10740	}
10741	retval = 0;
10742
10743	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
10744		retval += ctl_lun_reset(lun, io, ua_type);
10745
10746	return (retval);
10747}
10748
10749/*
10750 * The LUN should always be set.  The I/O is optional, and is used to
10751 * distinguish between I/Os sent by this initiator, and by other
10752 * initiators.  We set unit attention for initiators other than this one.
10753 * SAM-3 is vague on this point.  It does say that a unit attention should
10754 * be established for other initiators when a LUN is reset (see section
10755 * 5.7.3), but it doesn't specifically say that the unit attention should
10756 * be established for this particular initiator when a LUN is reset.  Here
10757 * is the relevant text, from SAM-3 rev 8:
10758 *
10759 * 5.7.2 When a SCSI initiator port aborts its own tasks
10760 *
10761 * When a SCSI initiator port causes its own task(s) to be aborted, no
10762 * notification that the task(s) have been aborted shall be returned to
10763 * the SCSI initiator port other than the completion response for the
10764 * command or task management function action that caused the task(s) to
10765 * be aborted and notification(s) associated with related effects of the
10766 * action (e.g., a reset unit attention condition).
10767 *
10768 * XXX KDM for now, we're setting unit attention for all initiators.
10769 */
10770static int
10771ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
10772{
10773	union ctl_io *xio;
10774#if 0
10775	uint32_t initindex;
10776#endif
10777	int i;
10778
10779	/*
10780	 * Run through the OOA queue and abort each I/O.
10781	 */
10782#if 0
10783	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10784#endif
10785	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10786	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10787		xio->io_hdr.flags |= CTL_FLAG_ABORT;
10788	}
10789
10790	/*
10791	 * This version sets unit attention for every
10792	 */
10793#if 0
10794	initindex = ctl_get_initindex(&io->io_hdr.nexus);
10795	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10796		if (initindex == i)
10797			continue;
10798		lun->pending_sense[i].ua_pending |= ua_type;
10799	}
10800#endif
10801
10802	/*
10803	 * A reset (any kind, really) clears reservations established with
10804	 * RESERVE/RELEASE.  It does not clear reservations established
10805	 * with PERSISTENT RESERVE OUT, but we don't support that at the
10806	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
10807	 * reservations made with the RESERVE/RELEASE commands, because
10808	 * those commands are obsolete in SPC-3.
10809	 */
10810	lun->flags &= ~CTL_LUN_RESERVED;
10811
10812	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10813		ctl_clear_mask(lun->have_ca, i);
10814		lun->pending_sense[i].ua_pending |= ua_type;
10815	}
10816
10817	return (0);
10818}
10819
10820static int
10821ctl_abort_task(union ctl_io *io)
10822{
10823	union ctl_io *xio;
10824	struct ctl_lun *lun;
10825	struct ctl_softc *ctl_softc;
10826#if 0
10827	struct sbuf sb;
10828	char printbuf[128];
10829#endif
10830	int found;
10831	uint32_t targ_lun;
10832
10833	ctl_softc = control_softc;
10834	found = 0;
10835
10836	/*
10837	 * Look up the LUN.
10838	 */
10839	targ_lun = io->io_hdr.nexus.targ_lun;
10840	if (io->io_hdr.nexus.lun_map_fn != NULL)
10841		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
10842	if ((targ_lun < CTL_MAX_LUNS)
10843	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
10844		lun = ctl_softc->ctl_luns[targ_lun];
10845	else
10846		goto bailout;
10847
10848#if 0
10849	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
10850	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
10851#endif
10852
10853	/*
10854	 * Run through the OOA queue and attempt to find the given I/O.
10855	 * The target port, initiator ID, tag type and tag number have to
10856	 * match the values that we got from the initiator.  If we have an
10857	 * untagged command to abort, simply abort the first untagged command
10858	 * we come to.  We only allow one untagged command at a time of course.
10859	 */
10860#if 0
10861	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10862#endif
10863	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10864	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10865#if 0
10866		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
10867
10868		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
10869			    lun->lun, xio->scsiio.tag_num,
10870			    xio->scsiio.tag_type,
10871			    (xio->io_hdr.blocked_links.tqe_prev
10872			    == NULL) ? "" : " BLOCKED",
10873			    (xio->io_hdr.flags &
10874			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
10875			    (xio->io_hdr.flags &
10876			    CTL_FLAG_ABORT) ? " ABORT" : "",
10877			    (xio->io_hdr.flags &
10878			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
10879		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
10880		sbuf_finish(&sb);
10881		printf("%s\n", sbuf_data(&sb));
10882#endif
10883
10884		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
10885		 && (xio->io_hdr.nexus.initid.id ==
10886		     io->io_hdr.nexus.initid.id)) {
10887			/*
10888			 * If the abort says that the task is untagged, the
10889			 * task in the queue must be untagged.  Otherwise,
10890			 * we just check to see whether the tag numbers
10891			 * match.  This is because the QLogic firmware
10892			 * doesn't pass back the tag type in an abort
10893			 * request.
10894			 */
10895#if 0
10896			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
10897			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
10898			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
10899#endif
10900			/*
10901			 * XXX KDM we've got problems with FC, because it
10902			 * doesn't send down a tag type with aborts.  So we
10903			 * can only really go by the tag number...
10904			 * This may cause problems with parallel SCSI.
10905			 * Need to figure that out!!
10906			 */
10907			if (xio->scsiio.tag_num == io->taskio.tag_num) {
10908				xio->io_hdr.flags |= CTL_FLAG_ABORT;
10909				found = 1;
10910				if ((io->io_hdr.flags &
10911				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
10912				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
10913					union ctl_ha_msg msg_info;
10914
10915					io->io_hdr.flags |=
10916					                CTL_FLAG_SENT_2OTHER_SC;
10917					msg_info.hdr.nexus = io->io_hdr.nexus;
10918					msg_info.task.task_action =
10919						CTL_TASK_ABORT_TASK;
10920					msg_info.task.tag_num =
10921						io->taskio.tag_num;
10922					msg_info.task.tag_type =
10923						io->taskio.tag_type;
10924					msg_info.hdr.msg_type =
10925						CTL_MSG_MANAGE_TASKS;
10926					msg_info.hdr.original_sc = NULL;
10927					msg_info.hdr.serializing_sc = NULL;
10928#if 0
10929					printf("Sent Abort to other side\n");
10930#endif
10931					if (CTL_HA_STATUS_SUCCESS !=
10932					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10933		    				(void *)&msg_info,
10934						sizeof(msg_info), 0)) {
10935					}
10936				}
10937#if 0
10938				printf("ctl_abort_task: found I/O to abort\n");
10939#endif
10940				break;
10941			}
10942		}
10943	}
10944
10945bailout:
10946
10947	if (found == 0) {
10948		/*
10949		 * This isn't really an error.  It's entirely possible for
10950		 * the abort and command completion to cross on the wire.
10951		 * This is more of an informative/diagnostic error.
10952		 */
10953#if 0
10954		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
10955		       "%d:%d:%d:%d tag %d type %d\n",
10956		       io->io_hdr.nexus.initid.id,
10957		       io->io_hdr.nexus.targ_port,
10958		       io->io_hdr.nexus.targ_target.id,
10959		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
10960		       io->taskio.tag_type);
10961#endif
10962		return (1);
10963	} else
10964		return (0);
10965}
10966
10967/*
10968 * This routine cannot block!  It must be callable from an interrupt
10969 * handler as well as from the work thread.
10970 */
10971static void
10972ctl_run_task_queue(struct ctl_softc *ctl_softc)
10973{
10974	union ctl_io *io, *next_io;
10975
10976	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
10977
10978	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
10979
10980	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
10981	     io != NULL; io = next_io) {
10982		int retval;
10983		const char *task_desc;
10984
10985		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10986
10987		retval = 0;
10988
10989		switch (io->io_hdr.io_type) {
10990		case CTL_IO_TASK: {
10991			task_desc = ctl_scsi_task_string(&io->taskio);
10992			if (task_desc != NULL) {
10993#ifdef NEEDTOPORT
10994				csevent_log(CSC_CTL | CSC_SHELF_SW |
10995					    CTL_TASK_REPORT,
10996					    csevent_LogType_Trace,
10997					    csevent_Severity_Information,
10998					    csevent_AlertLevel_Green,
10999					    csevent_FRU_Firmware,
11000					    csevent_FRU_Unknown,
11001					    "CTL: received task: %s",task_desc);
11002#endif
11003			} else {
11004#ifdef NEEDTOPORT
11005				csevent_log(CSC_CTL | CSC_SHELF_SW |
11006					    CTL_TASK_REPORT,
11007					    csevent_LogType_Trace,
11008					    csevent_Severity_Information,
11009					    csevent_AlertLevel_Green,
11010					    csevent_FRU_Firmware,
11011					    csevent_FRU_Unknown,
11012					    "CTL: received unknown task "
11013					    "type: %d (%#x)",
11014					    io->taskio.task_action,
11015					    io->taskio.task_action);
11016#endif
11017			}
11018			switch (io->taskio.task_action) {
11019			case CTL_TASK_ABORT_TASK:
11020				retval = ctl_abort_task(io);
11021				break;
11022			case CTL_TASK_ABORT_TASK_SET:
11023				break;
11024			case CTL_TASK_CLEAR_ACA:
11025				break;
11026			case CTL_TASK_CLEAR_TASK_SET:
11027				break;
11028			case CTL_TASK_LUN_RESET: {
11029				struct ctl_lun *lun;
11030				uint32_t targ_lun;
11031				int retval;
11032
11033				targ_lun = io->io_hdr.nexus.targ_lun;
11034				if (io->io_hdr.nexus.lun_map_fn != NULL)
11035					targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11036
11037				if ((targ_lun < CTL_MAX_LUNS)
11038				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11039					lun = ctl_softc->ctl_luns[targ_lun];
11040				else {
11041					retval = 1;
11042					break;
11043				}
11044
11045				if (!(io->io_hdr.flags &
11046				    CTL_FLAG_FROM_OTHER_SC)) {
11047					union ctl_ha_msg msg_info;
11048
11049					io->io_hdr.flags |=
11050						CTL_FLAG_SENT_2OTHER_SC;
11051					msg_info.hdr.msg_type =
11052						CTL_MSG_MANAGE_TASKS;
11053					msg_info.hdr.nexus = io->io_hdr.nexus;
11054					msg_info.task.task_action =
11055						CTL_TASK_LUN_RESET;
11056					msg_info.hdr.original_sc = NULL;
11057					msg_info.hdr.serializing_sc = NULL;
11058					if (CTL_HA_STATUS_SUCCESS !=
11059					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11060					    (void *)&msg_info,
11061					    sizeof(msg_info), 0)) {
11062					}
11063				}
11064
11065				retval = ctl_lun_reset(lun, io,
11066						       CTL_UA_LUN_RESET);
11067				break;
11068			}
11069			case CTL_TASK_TARGET_RESET:
11070				retval = ctl_target_reset(ctl_softc, io,
11071							  CTL_UA_TARG_RESET);
11072				break;
11073			case CTL_TASK_BUS_RESET:
11074				retval = ctl_bus_reset(ctl_softc, io);
11075				break;
11076			case CTL_TASK_PORT_LOGIN:
11077				break;
11078			case CTL_TASK_PORT_LOGOUT:
11079				break;
11080			default:
11081				printf("ctl_run_task_queue: got unknown task "
11082				       "management event %d\n",
11083				       io->taskio.task_action);
11084				break;
11085			}
11086			if (retval == 0)
11087				io->io_hdr.status = CTL_SUCCESS;
11088			else
11089				io->io_hdr.status = CTL_ERROR;
11090
11091			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11092				      ctl_io_hdr, links);
11093			/*
11094			 * This will queue this I/O to the done queue, but the
11095			 * work thread won't be able to process it until we
11096			 * return and the lock is released.
11097			 */
11098			ctl_done_lock(io, /*have_lock*/ 1);
11099			break;
11100		}
11101		default: {
11102
11103			printf("%s: invalid I/O type %d msg %d cdb %x"
11104			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11105			       __func__, io->io_hdr.io_type,
11106			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11107			       (uintmax_t)io->io_hdr.nexus.initid.id,
11108			       io->io_hdr.nexus.targ_port,
11109			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11110			       io->io_hdr.nexus.targ_lun /* XXX */,
11111			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11112			       io->taskio.tag_num : io->scsiio.tag_num);
11113			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11114				      ctl_io_hdr, links);
11115			ctl_free_io(io);
11116			break;
11117		}
11118		}
11119	}
11120
11121	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11122}
11123
11124/*
11125 * For HA operation.  Handle commands that come in from the other
11126 * controller.
11127 */
11128static void
11129ctl_handle_isc(union ctl_io *io)
11130{
11131	int free_io;
11132	struct ctl_lun *lun;
11133	struct ctl_softc *ctl_softc;
11134	uint32_t targ_lun;
11135
11136	ctl_softc = control_softc;
11137
11138	targ_lun = io->io_hdr.nexus.targ_lun;
11139	if (io->io_hdr.nexus.lun_map_fn != NULL)
11140		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11141	lun = ctl_softc->ctl_luns[targ_lun];
11142
11143	switch (io->io_hdr.msg_type) {
11144	case CTL_MSG_SERIALIZE:
11145		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11146						     /*have_lock*/ 0);
11147		break;
11148	case CTL_MSG_R2R: {
11149		uint8_t opcode;
11150		struct ctl_cmd_entry *entry;
11151
11152		/*
11153		 * This is only used in SER_ONLY mode.
11154		 */
11155		free_io = 0;
11156		opcode = io->scsiio.cdb[0];
11157		entry = &ctl_cmd_table[opcode];
11158		mtx_lock(&ctl_softc->ctl_lock);
11159		if (ctl_scsiio_lun_check(ctl_softc, lun,
11160		    entry, (struct ctl_scsiio *)io) != 0) {
11161			ctl_done_lock(io, /*have_lock*/ 1);
11162			mtx_unlock(&ctl_softc->ctl_lock);
11163			break;
11164		}
11165		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11166		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11167				   &io->io_hdr, links);
11168		mtx_unlock(&ctl_softc->ctl_lock);
11169		break;
11170	}
11171	case CTL_MSG_FINISH_IO:
11172		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11173			free_io = 0;
11174			ctl_done_lock(io, /*have_lock*/ 0);
11175		} else {
11176			free_io = 1;
11177			mtx_lock(&ctl_softc->ctl_lock);
11178			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11179				     ooa_links);
11180			STAILQ_REMOVE(&ctl_softc->task_queue,
11181				      &io->io_hdr, ctl_io_hdr, links);
11182			ctl_check_blocked(lun);
11183			mtx_unlock(&ctl_softc->ctl_lock);
11184		}
11185		break;
11186	case CTL_MSG_PERS_ACTION:
11187		ctl_hndl_per_res_out_on_other_sc(
11188			(union ctl_ha_msg *)&io->presio.pr_msg);
11189		free_io = 1;
11190		break;
11191	case CTL_MSG_BAD_JUJU:
11192		free_io = 0;
11193		ctl_done_lock(io, /*have_lock*/ 0);
11194		break;
11195	case CTL_MSG_DATAMOVE:
11196		/* Only used in XFER mode */
11197		free_io = 0;
11198		ctl_datamove_remote(io);
11199		break;
11200	case CTL_MSG_DATAMOVE_DONE:
11201		/* Only used in XFER mode */
11202		free_io = 0;
11203		io->scsiio.be_move_done(io);
11204		break;
11205	default:
11206		free_io = 1;
11207		printf("%s: Invalid message type %d\n",
11208		       __func__, io->io_hdr.msg_type);
11209		break;
11210	}
11211	if (free_io)
11212		ctl_free_io(io);
11213
11214}
11215
11216
11217/*
11218 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11219 * there is no match.
11220 */
11221static ctl_lun_error_pattern
11222ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11223{
11224	struct ctl_cmd_entry *entry;
11225	ctl_lun_error_pattern filtered_pattern, pattern;
11226	uint8_t opcode;
11227
11228	pattern = desc->error_pattern;
11229
11230	/*
11231	 * XXX KDM we need more data passed into this function to match a
11232	 * custom pattern, and we actually need to implement custom pattern
11233	 * matching.
11234	 */
11235	if (pattern & CTL_LUN_PAT_CMD)
11236		return (CTL_LUN_PAT_CMD);
11237
11238	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11239		return (CTL_LUN_PAT_ANY);
11240
11241	opcode = ctsio->cdb[0];
11242	entry = &ctl_cmd_table[opcode];
11243
11244	filtered_pattern = entry->pattern & pattern;
11245
11246	/*
11247	 * If the user requested specific flags in the pattern (e.g.
11248	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11249	 * flags.
11250	 *
11251	 * If the user did not specify any flags, it doesn't matter whether
11252	 * or not the command supports the flags.
11253	 */
11254	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11255	     (pattern & ~CTL_LUN_PAT_MASK))
11256		return (CTL_LUN_PAT_NONE);
11257
11258	/*
11259	 * If the user asked for a range check, see if the requested LBA
11260	 * range overlaps with this command's LBA range.
11261	 */
11262	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11263		uint64_t lba1;
11264		uint32_t len1;
11265		ctl_action action;
11266		int retval;
11267
11268		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11269		if (retval != 0)
11270			return (CTL_LUN_PAT_NONE);
11271
11272		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11273					      desc->lba_range.len);
11274		/*
11275		 * A "pass" means that the LBA ranges don't overlap, so
11276		 * this doesn't match the user's range criteria.
11277		 */
11278		if (action == CTL_ACTION_PASS)
11279			return (CTL_LUN_PAT_NONE);
11280	}
11281
11282	return (filtered_pattern);
11283}
11284
11285static void
11286ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11287{
11288	struct ctl_error_desc *desc, *desc2;
11289
11290	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
11291
11292	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11293		ctl_lun_error_pattern pattern;
11294		/*
11295		 * Check to see whether this particular command matches
11296		 * the pattern in the descriptor.
11297		 */
11298		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11299		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11300			continue;
11301
11302		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11303		case CTL_LUN_INJ_ABORTED:
11304			ctl_set_aborted(&io->scsiio);
11305			break;
11306		case CTL_LUN_INJ_MEDIUM_ERR:
11307			ctl_set_medium_error(&io->scsiio);
11308			break;
11309		case CTL_LUN_INJ_UA:
11310			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11311			 * OCCURRED */
11312			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11313			break;
11314		case CTL_LUN_INJ_CUSTOM:
11315			/*
11316			 * We're assuming the user knows what he is doing.
11317			 * Just copy the sense information without doing
11318			 * checks.
11319			 */
11320			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11321			      ctl_min(sizeof(desc->custom_sense),
11322				      sizeof(io->scsiio.sense_data)));
11323			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11324			io->scsiio.sense_len = SSD_FULL_SIZE;
11325			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11326			break;
11327		case CTL_LUN_INJ_NONE:
11328		default:
11329			/*
11330			 * If this is an error injection type we don't know
11331			 * about, clear the continuous flag (if it is set)
11332			 * so it will get deleted below.
11333			 */
11334			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11335			break;
11336		}
11337		/*
11338		 * By default, each error injection action is a one-shot
11339		 */
11340		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11341			continue;
11342
11343		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11344
11345		free(desc, M_CTL);
11346	}
11347}
11348
11349#ifdef CTL_IO_DELAY
11350static void
11351ctl_datamove_timer_wakeup(void *arg)
11352{
11353	union ctl_io *io;
11354
11355	io = (union ctl_io *)arg;
11356
11357	ctl_datamove(io);
11358}
11359#endif /* CTL_IO_DELAY */
11360
11361void
11362ctl_datamove(union ctl_io *io)
11363{
11364	void (*fe_datamove)(union ctl_io *io);
11365
11366	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11367
11368	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11369
11370#ifdef CTL_TIME_IO
11371	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11372		char str[256];
11373		char path_str[64];
11374		struct sbuf sb;
11375
11376		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11377		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11378
11379		sbuf_cat(&sb, path_str);
11380		switch (io->io_hdr.io_type) {
11381		case CTL_IO_SCSI:
11382			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11383			sbuf_printf(&sb, "\n");
11384			sbuf_cat(&sb, path_str);
11385			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11386				    io->scsiio.tag_num, io->scsiio.tag_type);
11387			break;
11388		case CTL_IO_TASK:
11389			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11390				    "Tag Type: %d\n", io->taskio.task_action,
11391				    io->taskio.tag_num, io->taskio.tag_type);
11392			break;
11393		default:
11394			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11395			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11396			break;
11397		}
11398		sbuf_cat(&sb, path_str);
11399		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11400			    (intmax_t)time_uptime - io->io_hdr.start_time);
11401		sbuf_finish(&sb);
11402		printf("%s", sbuf_data(&sb));
11403	}
11404#endif /* CTL_TIME_IO */
11405
11406	mtx_lock(&control_softc->ctl_lock);
11407#ifdef CTL_IO_DELAY
11408	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11409		struct ctl_lun *lun;
11410
11411		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11412
11413		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11414	} else {
11415		struct ctl_lun *lun;
11416
11417		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11418		if ((lun != NULL)
11419		 && (lun->delay_info.datamove_delay > 0)) {
11420			struct callout *callout;
11421
11422			callout = (struct callout *)&io->io_hdr.timer_bytes;
11423			callout_init(callout, /*mpsafe*/ 1);
11424			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11425			callout_reset(callout,
11426				      lun->delay_info.datamove_delay * hz,
11427				      ctl_datamove_timer_wakeup, io);
11428			if (lun->delay_info.datamove_type ==
11429			    CTL_DELAY_TYPE_ONESHOT)
11430				lun->delay_info.datamove_delay = 0;
11431			mtx_unlock(&control_softc->ctl_lock);
11432			return;
11433		}
11434	}
11435#endif
11436	/*
11437	 * If we have any pending task management commands, process them
11438	 * first.  This is necessary to eliminate a race condition with the
11439	 * FETD:
11440	 *
11441	 * - FETD submits a task management command, like an abort.
11442	 * - Back end calls fe_datamove() to move the data for the aborted
11443	 *   command.  The FETD can't really accept it, but if it did, it
11444	 *   would end up transmitting data for a command that the initiator
11445	 *   told us to abort.
11446	 *
11447	 * We close the race by processing all pending task management
11448	 * commands here (we can't block!), and then check this I/O to see
11449	 * if it has been aborted.  If so, return it to the back end with
11450	 * bad status, so the back end can say return an error to the back end
11451	 * and then when the back end returns an error, we can return the
11452	 * aborted command to the FETD, so it can clean up its resources.
11453	 */
11454	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
11455		ctl_run_task_queue(control_softc);
11456
11457	/*
11458	 * This command has been aborted.  Set the port status, so we fail
11459	 * the data move.
11460	 */
11461	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11462		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11463		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11464		       io->io_hdr.nexus.targ_port,
11465		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11466		       io->io_hdr.nexus.targ_lun);
11467		io->io_hdr.status = CTL_CMD_ABORTED;
11468		io->io_hdr.port_status = 31337;
11469		mtx_unlock(&control_softc->ctl_lock);
11470		/*
11471		 * Note that the backend, in this case, will get the
11472		 * callback in its context.  In other cases it may get
11473		 * called in the frontend's interrupt thread context.
11474		 */
11475		io->scsiio.be_move_done(io);
11476		return;
11477	}
11478
11479	/*
11480	 * If we're in XFER mode and this I/O is from the other shelf
11481	 * controller, we need to send the DMA to the other side to
11482	 * actually transfer the data to/from the host.  In serialize only
11483	 * mode the transfer happens below CTL and ctl_datamove() is only
11484	 * called on the machine that originally received the I/O.
11485	 */
11486	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11487	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11488		union ctl_ha_msg msg;
11489		uint32_t sg_entries_sent;
11490		int do_sg_copy;
11491		int i;
11492
11493		memset(&msg, 0, sizeof(msg));
11494		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11495		msg.hdr.original_sc = io->io_hdr.original_sc;
11496		msg.hdr.serializing_sc = io;
11497		msg.hdr.nexus = io->io_hdr.nexus;
11498		msg.dt.flags = io->io_hdr.flags;
11499		/*
11500		 * We convert everything into a S/G list here.  We can't
11501		 * pass by reference, only by value between controllers.
11502		 * So we can't pass a pointer to the S/G list, only as many
11503		 * S/G entries as we can fit in here.  If it's possible for
11504		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11505		 * then we need to break this up into multiple transfers.
11506		 */
11507		if (io->scsiio.kern_sg_entries == 0) {
11508			msg.dt.kern_sg_entries = 1;
11509			/*
11510			 * If this is in cached memory, flush the cache
11511			 * before we send the DMA request to the other
11512			 * controller.  We want to do this in either the
11513			 * read or the write case.  The read case is
11514			 * straightforward.  In the write case, we want to
11515			 * make sure nothing is in the local cache that
11516			 * could overwrite the DMAed data.
11517			 */
11518			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11519				/*
11520				 * XXX KDM use bus_dmamap_sync() here.
11521				 */
11522			}
11523
11524			/*
11525			 * Convert to a physical address if this is a
11526			 * virtual address.
11527			 */
11528			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
11529				msg.dt.sg_list[0].addr =
11530					io->scsiio.kern_data_ptr;
11531			} else {
11532				/*
11533				 * XXX KDM use busdma here!
11534				 */
11535#if 0
11536				msg.dt.sg_list[0].addr = (void *)
11537					vtophys(io->scsiio.kern_data_ptr);
11538#endif
11539			}
11540
11541			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
11542			do_sg_copy = 0;
11543		} else {
11544			struct ctl_sg_entry *sgl;
11545
11546			do_sg_copy = 1;
11547			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
11548			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
11549			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11550				/*
11551				 * XXX KDM use bus_dmamap_sync() here.
11552				 */
11553			}
11554		}
11555
11556		msg.dt.kern_data_len = io->scsiio.kern_data_len;
11557		msg.dt.kern_total_len = io->scsiio.kern_total_len;
11558		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
11559		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
11560		msg.dt.sg_sequence = 0;
11561
11562		/*
11563		 * Loop until we've sent all of the S/G entries.  On the
11564		 * other end, we'll recompose these S/G entries into one
11565		 * contiguous list before passing it to the
11566		 */
11567		for (sg_entries_sent = 0; sg_entries_sent <
11568		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
11569			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
11570				sizeof(msg.dt.sg_list[0])),
11571				msg.dt.kern_sg_entries - sg_entries_sent);
11572
11573			if (do_sg_copy != 0) {
11574				struct ctl_sg_entry *sgl;
11575				int j;
11576
11577				sgl = (struct ctl_sg_entry *)
11578					io->scsiio.kern_data_ptr;
11579				/*
11580				 * If this is in cached memory, flush the cache
11581				 * before we send the DMA request to the other
11582				 * controller.  We want to do this in either
11583				 * the * read or the write case.  The read
11584				 * case is straightforward.  In the write
11585				 * case, we want to make sure nothing is
11586				 * in the local cache that could overwrite
11587				 * the DMAed data.
11588				 */
11589
11590				for (i = sg_entries_sent, j = 0;
11591				     i < msg.dt.cur_sg_entries; i++, j++) {
11592					if ((io->io_hdr.flags &
11593					     CTL_FLAG_NO_DATASYNC) == 0) {
11594						/*
11595						 * XXX KDM use bus_dmamap_sync()
11596						 */
11597					}
11598					if ((io->io_hdr.flags &
11599					     CTL_FLAG_BUS_ADDR) == 0) {
11600						/*
11601						 * XXX KDM use busdma.
11602						 */
11603#if 0
11604						msg.dt.sg_list[j].addr =(void *)
11605						       vtophys(sgl[i].addr);
11606#endif
11607					} else {
11608						msg.dt.sg_list[j].addr =
11609							sgl[i].addr;
11610					}
11611					msg.dt.sg_list[j].len = sgl[i].len;
11612				}
11613			}
11614
11615			sg_entries_sent += msg.dt.cur_sg_entries;
11616			if (sg_entries_sent >= msg.dt.kern_sg_entries)
11617				msg.dt.sg_last = 1;
11618			else
11619				msg.dt.sg_last = 0;
11620
11621			/*
11622			 * XXX KDM drop and reacquire the lock here?
11623			 */
11624			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
11625			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
11626				/*
11627				 * XXX do something here.
11628				 */
11629			}
11630
11631			msg.dt.sent_sg_entries = sg_entries_sent;
11632		}
11633		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11634		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
11635			ctl_failover_io(io, /*have_lock*/ 1);
11636
11637	} else {
11638
11639		/*
11640		 * Lookup the fe_datamove() function for this particular
11641		 * front end.
11642		 */
11643		fe_datamove =
11644		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11645		mtx_unlock(&control_softc->ctl_lock);
11646
11647		fe_datamove(io);
11648	}
11649}
11650
11651static void
11652ctl_send_datamove_done(union ctl_io *io, int have_lock)
11653{
11654	union ctl_ha_msg msg;
11655	int isc_status;
11656
11657	memset(&msg, 0, sizeof(msg));
11658
11659	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
11660	msg.hdr.original_sc = io;
11661	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
11662	msg.hdr.nexus = io->io_hdr.nexus;
11663	msg.hdr.status = io->io_hdr.status;
11664	msg.scsi.tag_num = io->scsiio.tag_num;
11665	msg.scsi.tag_type = io->scsiio.tag_type;
11666	msg.scsi.scsi_status = io->scsiio.scsi_status;
11667	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
11668	       sizeof(io->scsiio.sense_data));
11669	msg.scsi.sense_len = io->scsiio.sense_len;
11670	msg.scsi.sense_residual = io->scsiio.sense_residual;
11671	msg.scsi.fetd_status = io->io_hdr.port_status;
11672	msg.scsi.residual = io->scsiio.residual;
11673	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11674
11675	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
11676		ctl_failover_io(io, /*have_lock*/ have_lock);
11677		return;
11678	}
11679
11680	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
11681	if (isc_status > CTL_HA_STATUS_SUCCESS) {
11682		/* XXX do something if this fails */
11683	}
11684
11685}
11686
11687/*
11688 * The DMA to the remote side is done, now we need to tell the other side
11689 * we're done so it can continue with its data movement.
11690 */
11691static void
11692ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
11693{
11694	union ctl_io *io;
11695
11696	io = rq->context;
11697
11698	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11699		printf("%s: ISC DMA write failed with error %d", __func__,
11700		       rq->ret);
11701		ctl_set_internal_failure(&io->scsiio,
11702					 /*sks_valid*/ 1,
11703					 /*retry_count*/ rq->ret);
11704	}
11705
11706	ctl_dt_req_free(rq);
11707
11708	/*
11709	 * In this case, we had to malloc the memory locally.  Free it.
11710	 */
11711	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11712		int i;
11713		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11714			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11715	}
11716	/*
11717	 * The data is in local and remote memory, so now we need to send
11718	 * status (good or back) back to the other side.
11719	 */
11720	ctl_send_datamove_done(io, /*have_lock*/ 0);
11721}
11722
11723/*
11724 * We've moved the data from the host/controller into local memory.  Now we
11725 * need to push it over to the remote controller's memory.
11726 */
11727static int
11728ctl_datamove_remote_dm_write_cb(union ctl_io *io)
11729{
11730	int retval;
11731
11732	retval = 0;
11733
11734	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
11735					  ctl_datamove_remote_write_cb);
11736
11737	return (retval);
11738}
11739
11740static void
11741ctl_datamove_remote_write(union ctl_io *io)
11742{
11743	int retval;
11744	void (*fe_datamove)(union ctl_io *io);
11745
11746	/*
11747	 * - Get the data from the host/HBA into local memory.
11748	 * - DMA memory from the local controller to the remote controller.
11749	 * - Send status back to the remote controller.
11750	 */
11751
11752	retval = ctl_datamove_remote_sgl_setup(io);
11753	if (retval != 0)
11754		return;
11755
11756	/* Switch the pointer over so the FETD knows what to do */
11757	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11758
11759	/*
11760	 * Use a custom move done callback, since we need to send completion
11761	 * back to the other controller, not to the backend on this side.
11762	 */
11763	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
11764
11765	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11766
11767	fe_datamove(io);
11768
11769	return;
11770
11771}
11772
11773static int
11774ctl_datamove_remote_dm_read_cb(union ctl_io *io)
11775{
11776#if 0
11777	char str[256];
11778	char path_str[64];
11779	struct sbuf sb;
11780#endif
11781
11782	/*
11783	 * In this case, we had to malloc the memory locally.  Free it.
11784	 */
11785	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11786		int i;
11787		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11788			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11789	}
11790
11791#if 0
11792	scsi_path_string(io, path_str, sizeof(path_str));
11793	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11794	sbuf_cat(&sb, path_str);
11795	scsi_command_string(&io->scsiio, NULL, &sb);
11796	sbuf_printf(&sb, "\n");
11797	sbuf_cat(&sb, path_str);
11798	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11799		    io->scsiio.tag_num, io->scsiio.tag_type);
11800	sbuf_cat(&sb, path_str);
11801	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
11802		    io->io_hdr.flags, io->io_hdr.status);
11803	sbuf_finish(&sb);
11804	printk("%s", sbuf_data(&sb));
11805#endif
11806
11807
11808	/*
11809	 * The read is done, now we need to send status (good or bad) back
11810	 * to the other side.
11811	 */
11812	ctl_send_datamove_done(io, /*have_lock*/ 0);
11813
11814	return (0);
11815}
11816
11817static void
11818ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
11819{
11820	union ctl_io *io;
11821	void (*fe_datamove)(union ctl_io *io);
11822
11823	io = rq->context;
11824
11825	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11826		printf("%s: ISC DMA read failed with error %d", __func__,
11827		       rq->ret);
11828		ctl_set_internal_failure(&io->scsiio,
11829					 /*sks_valid*/ 1,
11830					 /*retry_count*/ rq->ret);
11831	}
11832
11833	ctl_dt_req_free(rq);
11834
11835	/* Switch the pointer over so the FETD knows what to do */
11836	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11837
11838	/*
11839	 * Use a custom move done callback, since we need to send completion
11840	 * back to the other controller, not to the backend on this side.
11841	 */
11842	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
11843
11844	/* XXX KDM add checks like the ones in ctl_datamove? */
11845
11846	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11847
11848	fe_datamove(io);
11849}
11850
11851static int
11852ctl_datamove_remote_sgl_setup(union ctl_io *io)
11853{
11854	struct ctl_sg_entry *local_sglist, *remote_sglist;
11855	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
11856	struct ctl_softc *softc;
11857	int retval;
11858	int i;
11859
11860	retval = 0;
11861	softc = control_softc;
11862
11863	local_sglist = io->io_hdr.local_sglist;
11864	local_dma_sglist = io->io_hdr.local_dma_sglist;
11865	remote_sglist = io->io_hdr.remote_sglist;
11866	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11867
11868	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
11869		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
11870			local_sglist[i].len = remote_sglist[i].len;
11871
11872			/*
11873			 * XXX Detect the situation where the RS-level I/O
11874			 * redirector on the other side has already read the
11875			 * data off of the AOR RS on this side, and
11876			 * transferred it to remote (mirror) memory on the
11877			 * other side.  Since we already have the data in
11878			 * memory here, we just need to use it.
11879			 *
11880			 * XXX KDM this can probably be removed once we
11881			 * get the cache device code in and take the
11882			 * current AOR implementation out.
11883			 */
11884#ifdef NEEDTOPORT
11885			if ((remote_sglist[i].addr >=
11886			     (void *)vtophys(softc->mirr->addr))
11887			 && (remote_sglist[i].addr <
11888			     ((void *)vtophys(softc->mirr->addr) +
11889			     CacheMirrorOffset))) {
11890				local_sglist[i].addr = remote_sglist[i].addr -
11891					CacheMirrorOffset;
11892				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
11893				     CTL_FLAG_DATA_IN)
11894					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
11895			} else {
11896				local_sglist[i].addr = remote_sglist[i].addr +
11897					CacheMirrorOffset;
11898			}
11899#endif
11900#if 0
11901			printf("%s: local %p, remote %p, len %d\n",
11902			       __func__, local_sglist[i].addr,
11903			       remote_sglist[i].addr, local_sglist[i].len);
11904#endif
11905		}
11906	} else {
11907		uint32_t len_to_go;
11908
11909		/*
11910		 * In this case, we don't have automatically allocated
11911		 * memory for this I/O on this controller.  This typically
11912		 * happens with internal CTL I/O -- e.g. inquiry, mode
11913		 * sense, etc.  Anything coming from RAIDCore will have
11914		 * a mirror area available.
11915		 */
11916		len_to_go = io->scsiio.kern_data_len;
11917
11918		/*
11919		 * Clear the no datasync flag, we have to use malloced
11920		 * buffers.
11921		 */
11922		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
11923
11924		/*
11925		 * The difficult thing here is that the size of the various
11926		 * S/G segments may be different than the size from the
11927		 * remote controller.  That'll make it harder when DMAing
11928		 * the data back to the other side.
11929		 */
11930		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
11931		     sizeof(io->io_hdr.remote_sglist[0])) &&
11932		     (len_to_go > 0); i++) {
11933			local_sglist[i].len = ctl_min(len_to_go, 131072);
11934			CTL_SIZE_8B(local_dma_sglist[i].len,
11935				    local_sglist[i].len);
11936			local_sglist[i].addr =
11937				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
11938
11939			local_dma_sglist[i].addr = local_sglist[i].addr;
11940
11941			if (local_sglist[i].addr == NULL) {
11942				int j;
11943
11944				printf("malloc failed for %zd bytes!",
11945				       local_dma_sglist[i].len);
11946				for (j = 0; j < i; j++) {
11947					free(local_sglist[j].addr, M_CTL);
11948				}
11949				ctl_set_internal_failure(&io->scsiio,
11950							 /*sks_valid*/ 1,
11951							 /*retry_count*/ 4857);
11952				retval = 1;
11953				goto bailout_error;
11954
11955			}
11956			/* XXX KDM do we need a sync here? */
11957
11958			len_to_go -= local_sglist[i].len;
11959		}
11960		/*
11961		 * Reset the number of S/G entries accordingly.  The
11962		 * original number of S/G entries is available in
11963		 * rem_sg_entries.
11964		 */
11965		io->scsiio.kern_sg_entries = i;
11966
11967#if 0
11968		printf("%s: kern_sg_entries = %d\n", __func__,
11969		       io->scsiio.kern_sg_entries);
11970		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11971			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
11972			       local_sglist[i].addr, local_sglist[i].len,
11973			       local_dma_sglist[i].len);
11974#endif
11975	}
11976
11977
11978	return (retval);
11979
11980bailout_error:
11981
11982	ctl_send_datamove_done(io, /*have_lock*/ 0);
11983
11984	return (retval);
11985}
11986
11987static int
11988ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
11989			 ctl_ha_dt_cb callback)
11990{
11991	struct ctl_ha_dt_req *rq;
11992	struct ctl_sg_entry *remote_sglist, *local_sglist;
11993	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
11994	uint32_t local_used, remote_used, total_used;
11995	int retval;
11996	int i, j;
11997
11998	retval = 0;
11999
12000	rq = ctl_dt_req_alloc();
12001
12002	/*
12003	 * If we failed to allocate the request, and if the DMA didn't fail
12004	 * anyway, set busy status.  This is just a resource allocation
12005	 * failure.
12006	 */
12007	if ((rq == NULL)
12008	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12009		ctl_set_busy(&io->scsiio);
12010
12011	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12012
12013		if (rq != NULL)
12014			ctl_dt_req_free(rq);
12015
12016		/*
12017		 * The data move failed.  We need to return status back
12018		 * to the other controller.  No point in trying to DMA
12019		 * data to the remote controller.
12020		 */
12021
12022		ctl_send_datamove_done(io, /*have_lock*/ 0);
12023
12024		retval = 1;
12025
12026		goto bailout;
12027	}
12028
12029	local_sglist = io->io_hdr.local_sglist;
12030	local_dma_sglist = io->io_hdr.local_dma_sglist;
12031	remote_sglist = io->io_hdr.remote_sglist;
12032	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12033	local_used = 0;
12034	remote_used = 0;
12035	total_used = 0;
12036
12037	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12038		rq->ret = CTL_HA_STATUS_SUCCESS;
12039		rq->context = io;
12040		callback(rq);
12041		goto bailout;
12042	}
12043
12044	/*
12045	 * Pull/push the data over the wire from/to the other controller.
12046	 * This takes into account the possibility that the local and
12047	 * remote sglists may not be identical in terms of the size of
12048	 * the elements and the number of elements.
12049	 *
12050	 * One fundamental assumption here is that the length allocated for
12051	 * both the local and remote sglists is identical.  Otherwise, we've
12052	 * essentially got a coding error of some sort.
12053	 */
12054	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12055		int isc_ret;
12056		uint32_t cur_len, dma_length;
12057		uint8_t *tmp_ptr;
12058
12059		rq->id = CTL_HA_DATA_CTL;
12060		rq->command = command;
12061		rq->context = io;
12062
12063		/*
12064		 * Both pointers should be aligned.  But it is possible
12065		 * that the allocation length is not.  They should both
12066		 * also have enough slack left over at the end, though,
12067		 * to round up to the next 8 byte boundary.
12068		 */
12069		cur_len = ctl_min(local_sglist[i].len - local_used,
12070				  remote_sglist[j].len - remote_used);
12071
12072		/*
12073		 * In this case, we have a size issue and need to decrease
12074		 * the size, except in the case where we actually have less
12075		 * than 8 bytes left.  In that case, we need to increase
12076		 * the DMA length to get the last bit.
12077		 */
12078		if ((cur_len & 0x7) != 0) {
12079			if (cur_len > 0x7) {
12080				cur_len = cur_len - (cur_len & 0x7);
12081				dma_length = cur_len;
12082			} else {
12083				CTL_SIZE_8B(dma_length, cur_len);
12084			}
12085
12086		} else
12087			dma_length = cur_len;
12088
12089		/*
12090		 * If we had to allocate memory for this I/O, instead of using
12091		 * the non-cached mirror memory, we'll need to flush the cache
12092		 * before trying to DMA to the other controller.
12093		 *
12094		 * We could end up doing this multiple times for the same
12095		 * segment if we have a larger local segment than remote
12096		 * segment.  That shouldn't be an issue.
12097		 */
12098		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12099			/*
12100			 * XXX KDM use bus_dmamap_sync() here.
12101			 */
12102		}
12103
12104		rq->size = dma_length;
12105
12106		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12107		tmp_ptr += local_used;
12108
12109		/* Use physical addresses when talking to ISC hardware */
12110		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12111			/* XXX KDM use busdma */
12112#if 0
12113			rq->local = vtophys(tmp_ptr);
12114#endif
12115		} else
12116			rq->local = tmp_ptr;
12117
12118		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12119		tmp_ptr += remote_used;
12120		rq->remote = tmp_ptr;
12121
12122		rq->callback = NULL;
12123
12124		local_used += cur_len;
12125		if (local_used >= local_sglist[i].len) {
12126			i++;
12127			local_used = 0;
12128		}
12129
12130		remote_used += cur_len;
12131		if (remote_used >= remote_sglist[j].len) {
12132			j++;
12133			remote_used = 0;
12134		}
12135		total_used += cur_len;
12136
12137		if (total_used >= io->scsiio.kern_data_len)
12138			rq->callback = callback;
12139
12140		if ((rq->size & 0x7) != 0) {
12141			printf("%s: warning: size %d is not on 8b boundary\n",
12142			       __func__, rq->size);
12143		}
12144		if (((uintptr_t)rq->local & 0x7) != 0) {
12145			printf("%s: warning: local %p not on 8b boundary\n",
12146			       __func__, rq->local);
12147		}
12148		if (((uintptr_t)rq->remote & 0x7) != 0) {
12149			printf("%s: warning: remote %p not on 8b boundary\n",
12150			       __func__, rq->local);
12151		}
12152#if 0
12153		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12154		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12155		       rq->local, rq->remote, rq->size);
12156#endif
12157
12158		isc_ret = ctl_dt_single(rq);
12159		if (isc_ret == CTL_HA_STATUS_WAIT)
12160			continue;
12161
12162		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12163			rq->ret = CTL_HA_STATUS_SUCCESS;
12164		} else {
12165			rq->ret = isc_ret;
12166		}
12167		callback(rq);
12168		goto bailout;
12169	}
12170
12171bailout:
12172	return (retval);
12173
12174}
12175
12176static void
12177ctl_datamove_remote_read(union ctl_io *io)
12178{
12179	int retval;
12180	int i;
12181
12182	/*
12183	 * This will send an error to the other controller in the case of a
12184	 * failure.
12185	 */
12186	retval = ctl_datamove_remote_sgl_setup(io);
12187	if (retval != 0)
12188		return;
12189
12190	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12191					  ctl_datamove_remote_read_cb);
12192	if ((retval != 0)
12193	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12194		/*
12195		 * Make sure we free memory if there was an error..  The
12196		 * ctl_datamove_remote_xfer() function will send the
12197		 * datamove done message, or call the callback with an
12198		 * error if there is a problem.
12199		 */
12200		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12201			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12202	}
12203
12204	return;
12205}
12206
12207/*
12208 * Process a datamove request from the other controller.  This is used for
12209 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12210 * first.  Once that is complete, the data gets DMAed into the remote
12211 * controller's memory.  For reads, we DMA from the remote controller's
12212 * memory into our memory first, and then move it out to the FETD.
12213 */
12214static void
12215ctl_datamove_remote(union ctl_io *io)
12216{
12217	struct ctl_softc *softc;
12218
12219	softc = control_softc;
12220
12221	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12222
12223	/*
12224	 * Note that we look for an aborted I/O here, but don't do some of
12225	 * the other checks that ctl_datamove() normally does.  We don't
12226	 * need to run the task queue, because this I/O is on the ISC
12227	 * queue, which is executed by the work thread after the task queue.
12228	 * We don't need to run the datamove delay code, since that should
12229	 * have been done if need be on the other controller.
12230	 */
12231	mtx_lock(&softc->ctl_lock);
12232
12233	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12234
12235		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12236		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12237		       io->io_hdr.nexus.targ_port,
12238		       io->io_hdr.nexus.targ_target.id,
12239		       io->io_hdr.nexus.targ_lun);
12240		io->io_hdr.status = CTL_CMD_ABORTED;
12241		io->io_hdr.port_status = 31338;
12242
12243		mtx_unlock(&softc->ctl_lock);
12244
12245		ctl_send_datamove_done(io, /*have_lock*/ 0);
12246
12247		return;
12248	}
12249
12250	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12251		mtx_unlock(&softc->ctl_lock);
12252		ctl_datamove_remote_write(io);
12253	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12254		mtx_unlock(&softc->ctl_lock);
12255		ctl_datamove_remote_read(io);
12256	} else {
12257		union ctl_ha_msg msg;
12258		struct scsi_sense_data *sense;
12259		uint8_t sks[3];
12260		int retry_count;
12261
12262		memset(&msg, 0, sizeof(msg));
12263
12264		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12265		msg.hdr.status = CTL_SCSI_ERROR;
12266		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12267
12268		retry_count = 4243;
12269
12270		sense = &msg.scsi.sense_data;
12271		sks[0] = SSD_SCS_VALID;
12272		sks[1] = (retry_count >> 8) & 0xff;
12273		sks[2] = retry_count & 0xff;
12274
12275		/* "Internal target failure" */
12276		scsi_set_sense_data(sense,
12277				    /*sense_format*/ SSD_TYPE_NONE,
12278				    /*current_error*/ 1,
12279				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12280				    /*asc*/ 0x44,
12281				    /*ascq*/ 0x00,
12282				    /*type*/ SSD_ELEM_SKS,
12283				    /*size*/ sizeof(sks),
12284				    /*data*/ sks,
12285				    SSD_ELEM_NONE);
12286
12287		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12288		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12289			ctl_failover_io(io, /*have_lock*/ 1);
12290			mtx_unlock(&softc->ctl_lock);
12291			return;
12292		}
12293
12294		mtx_unlock(&softc->ctl_lock);
12295
12296		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12297		    CTL_HA_STATUS_SUCCESS) {
12298			/* XXX KDM what to do if this fails? */
12299		}
12300		return;
12301	}
12302
12303}
12304
12305static int
12306ctl_process_done(union ctl_io *io, int have_lock)
12307{
12308	struct ctl_lun *lun;
12309	struct ctl_softc *ctl_softc;
12310	void (*fe_done)(union ctl_io *io);
12311	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12312
12313	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12314
12315	fe_done =
12316	    control_softc->ctl_ports[targ_port]->fe_done;
12317
12318#ifdef CTL_TIME_IO
12319	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12320		char str[256];
12321		char path_str[64];
12322		struct sbuf sb;
12323
12324		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12325		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12326
12327		sbuf_cat(&sb, path_str);
12328		switch (io->io_hdr.io_type) {
12329		case CTL_IO_SCSI:
12330			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12331			sbuf_printf(&sb, "\n");
12332			sbuf_cat(&sb, path_str);
12333			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12334				    io->scsiio.tag_num, io->scsiio.tag_type);
12335			break;
12336		case CTL_IO_TASK:
12337			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12338				    "Tag Type: %d\n", io->taskio.task_action,
12339				    io->taskio.tag_num, io->taskio.tag_type);
12340			break;
12341		default:
12342			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12343			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12344			break;
12345		}
12346		sbuf_cat(&sb, path_str);
12347		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12348			    (intmax_t)time_uptime - io->io_hdr.start_time);
12349		sbuf_finish(&sb);
12350		printf("%s", sbuf_data(&sb));
12351	}
12352#endif /* CTL_TIME_IO */
12353
12354	switch (io->io_hdr.io_type) {
12355	case CTL_IO_SCSI:
12356		break;
12357	case CTL_IO_TASK:
12358		if (bootverbose || verbose > 0)
12359			ctl_io_error_print(io, NULL);
12360		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12361			ctl_free_io(io);
12362		else
12363			fe_done(io);
12364		return (CTL_RETVAL_COMPLETE);
12365		break;
12366	default:
12367		printf("ctl_process_done: invalid io type %d\n",
12368		       io->io_hdr.io_type);
12369		panic("ctl_process_done: invalid io type %d\n",
12370		      io->io_hdr.io_type);
12371		break; /* NOTREACHED */
12372	}
12373
12374	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12375	if (lun == NULL) {
12376		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12377				 io->io_hdr.nexus.targ_lun));
12378		fe_done(io);
12379		goto bailout;
12380	}
12381	ctl_softc = lun->ctl_softc;
12382
12383	/*
12384	 * Remove this from the OOA queue.
12385	 */
12386	if (have_lock == 0)
12387		mtx_lock(&ctl_softc->ctl_lock);
12388
12389	/*
12390	 * Check to see if we have any errors to inject here.  We only
12391	 * inject errors for commands that don't already have errors set.
12392	 */
12393	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12394	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12395		ctl_inject_error(lun, io);
12396
12397	/*
12398	 * XXX KDM how do we treat commands that aren't completed
12399	 * successfully?
12400	 *
12401	 * XXX KDM should we also track I/O latency?
12402	 */
12403	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
12404		uint32_t blocksize;
12405#ifdef CTL_TIME_IO
12406		struct bintime cur_bt;
12407#endif
12408
12409		if ((lun->be_lun != NULL)
12410		 && (lun->be_lun->blocksize != 0))
12411			blocksize = lun->be_lun->blocksize;
12412		else
12413			blocksize = 512;
12414
12415		switch (io->io_hdr.io_type) {
12416		case CTL_IO_SCSI: {
12417			int isread;
12418			struct ctl_lba_len lbalen;
12419
12420			isread = 0;
12421			switch (io->scsiio.cdb[0]) {
12422			case READ_6:
12423			case READ_10:
12424			case READ_12:
12425			case READ_16:
12426				isread = 1;
12427				/* FALLTHROUGH */
12428			case WRITE_6:
12429			case WRITE_10:
12430			case WRITE_12:
12431			case WRITE_16:
12432			case WRITE_VERIFY_10:
12433			case WRITE_VERIFY_12:
12434			case WRITE_VERIFY_16:
12435				memcpy(&lbalen, io->io_hdr.ctl_private[
12436				       CTL_PRIV_LBA_LEN].bytes, sizeof(lbalen));
12437
12438				if (isread) {
12439					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
12440						lbalen.len * blocksize;
12441					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
12442
12443#ifdef CTL_TIME_IO
12444					bintime_add(
12445					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
12446					   &io->io_hdr.dma_bt);
12447					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
12448						io->io_hdr.num_dmas;
12449					getbintime(&cur_bt);
12450					bintime_sub(&cur_bt,
12451						    &io->io_hdr.start_bt);
12452
12453					bintime_add(
12454					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
12455					    &cur_bt);
12456
12457#if 0
12458					cs_prof_gettime(&cur_ticks);
12459					lun->stats.time[CTL_STATS_READ] +=
12460						cur_ticks -
12461						io->io_hdr.start_ticks;
12462#endif
12463#if 0
12464					lun->stats.time[CTL_STATS_READ] +=
12465						jiffies - io->io_hdr.start_time;
12466#endif
12467#endif /* CTL_TIME_IO */
12468				} else {
12469					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
12470						lbalen.len * blocksize;
12471					lun->stats.ports[targ_port].operations[
12472						CTL_STATS_WRITE]++;
12473
12474#ifdef CTL_TIME_IO
12475					bintime_add(
12476					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
12477					  &io->io_hdr.dma_bt);
12478					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
12479						io->io_hdr.num_dmas;
12480					getbintime(&cur_bt);
12481					bintime_sub(&cur_bt,
12482						    &io->io_hdr.start_bt);
12483
12484					bintime_add(
12485					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
12486					    &cur_bt);
12487#if 0
12488					cs_prof_gettime(&cur_ticks);
12489					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12490						cur_ticks -
12491						io->io_hdr.start_ticks;
12492					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12493						jiffies - io->io_hdr.start_time;
12494#endif
12495#endif /* CTL_TIME_IO */
12496				}
12497				break;
12498			default:
12499				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
12500
12501#ifdef CTL_TIME_IO
12502				bintime_add(
12503				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
12504				  &io->io_hdr.dma_bt);
12505				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
12506					io->io_hdr.num_dmas;
12507				getbintime(&cur_bt);
12508				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12509
12510				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
12511					    &cur_bt);
12512
12513#if 0
12514				cs_prof_gettime(&cur_ticks);
12515				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12516					cur_ticks -
12517					io->io_hdr.start_ticks;
12518				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12519					jiffies - io->io_hdr.start_time;
12520#endif
12521#endif /* CTL_TIME_IO */
12522				break;
12523			}
12524			break;
12525		}
12526		default:
12527			break;
12528		}
12529	}
12530
12531	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12532
12533	/*
12534	 * Run through the blocked queue on this LUN and see if anything
12535	 * has become unblocked, now that this transaction is done.
12536	 */
12537	ctl_check_blocked(lun);
12538
12539	/*
12540	 * If the LUN has been invalidated, free it if there is nothing
12541	 * left on its OOA queue.
12542	 */
12543	if ((lun->flags & CTL_LUN_INVALID)
12544	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
12545		ctl_free_lun(lun);
12546
12547	/*
12548	 * If this command has been aborted, make sure we set the status
12549	 * properly.  The FETD is responsible for freeing the I/O and doing
12550	 * whatever it needs to do to clean up its state.
12551	 */
12552	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12553		io->io_hdr.status = CTL_CMD_ABORTED;
12554
12555	/*
12556	 * We print out status for every task management command.  For SCSI
12557	 * commands, we filter out any unit attention errors; they happen
12558	 * on every boot, and would clutter up the log.  Note:  task
12559	 * management commands aren't printed here, they are printed above,
12560	 * since they should never even make it down here.
12561	 */
12562	switch (io->io_hdr.io_type) {
12563	case CTL_IO_SCSI: {
12564		int error_code, sense_key, asc, ascq;
12565
12566		sense_key = 0;
12567
12568		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12569		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12570			/*
12571			 * Since this is just for printing, no need to
12572			 * show errors here.
12573			 */
12574			scsi_extract_sense_len(&io->scsiio.sense_data,
12575					       io->scsiio.sense_len,
12576					       &error_code,
12577					       &sense_key,
12578					       &asc,
12579					       &ascq,
12580					       /*show_errors*/ 0);
12581		}
12582
12583		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12584		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12585		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12586		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12587
12588			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12589				ctl_softc->skipped_prints++;
12590				if (have_lock == 0)
12591					mtx_unlock(&ctl_softc->ctl_lock);
12592			} else {
12593				uint32_t skipped_prints;
12594
12595				skipped_prints = ctl_softc->skipped_prints;
12596
12597				ctl_softc->skipped_prints = 0;
12598				ctl_softc->last_print_jiffies = time_uptime;
12599
12600				if (have_lock == 0)
12601					mtx_unlock(&ctl_softc->ctl_lock);
12602				if (skipped_prints > 0) {
12603#ifdef NEEDTOPORT
12604					csevent_log(CSC_CTL | CSC_SHELF_SW |
12605					    CTL_ERROR_REPORT,
12606					    csevent_LogType_Trace,
12607					    csevent_Severity_Information,
12608					    csevent_AlertLevel_Green,
12609					    csevent_FRU_Firmware,
12610					    csevent_FRU_Unknown,
12611					    "High CTL error volume, %d prints "
12612					    "skipped", skipped_prints);
12613#endif
12614				}
12615				if (bootverbose || verbose > 0)
12616					ctl_io_error_print(io, NULL);
12617			}
12618		} else {
12619			if (have_lock == 0)
12620				mtx_unlock(&ctl_softc->ctl_lock);
12621		}
12622		break;
12623	}
12624	case CTL_IO_TASK:
12625		if (have_lock == 0)
12626			mtx_unlock(&ctl_softc->ctl_lock);
12627		if (bootverbose || verbose > 0)
12628			ctl_io_error_print(io, NULL);
12629		break;
12630	default:
12631		if (have_lock == 0)
12632			mtx_unlock(&ctl_softc->ctl_lock);
12633		break;
12634	}
12635
12636	/*
12637	 * Tell the FETD or the other shelf controller we're done with this
12638	 * command.  Note that only SCSI commands get to this point.  Task
12639	 * management commands are completed above.
12640	 *
12641	 * We only send status to the other controller if we're in XFER
12642	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
12643	 * received the I/O (from CTL's perspective), and so the status is
12644	 * generated there.
12645	 *
12646	 * XXX KDM if we hold the lock here, we could cause a deadlock
12647	 * if the frontend comes back in in this context to queue
12648	 * something.
12649	 */
12650	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
12651	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12652		union ctl_ha_msg msg;
12653
12654		memset(&msg, 0, sizeof(msg));
12655		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12656		msg.hdr.original_sc = io->io_hdr.original_sc;
12657		msg.hdr.nexus = io->io_hdr.nexus;
12658		msg.hdr.status = io->io_hdr.status;
12659		msg.scsi.scsi_status = io->scsiio.scsi_status;
12660		msg.scsi.tag_num = io->scsiio.tag_num;
12661		msg.scsi.tag_type = io->scsiio.tag_type;
12662		msg.scsi.sense_len = io->scsiio.sense_len;
12663		msg.scsi.sense_residual = io->scsiio.sense_residual;
12664		msg.scsi.residual = io->scsiio.residual;
12665		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12666		       sizeof(io->scsiio.sense_data));
12667		/*
12668		 * We copy this whether or not this is an I/O-related
12669		 * command.  Otherwise, we'd have to go and check to see
12670		 * whether it's a read/write command, and it really isn't
12671		 * worth it.
12672		 */
12673		memcpy(&msg.scsi.lbalen,
12674		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12675		       sizeof(msg.scsi.lbalen));
12676
12677		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12678				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12679			/* XXX do something here */
12680		}
12681
12682		ctl_free_io(io);
12683	} else
12684		fe_done(io);
12685
12686bailout:
12687
12688	return (CTL_RETVAL_COMPLETE);
12689}
12690
12691/*
12692 * Front end should call this if it doesn't do autosense.  When the request
12693 * sense comes back in from the initiator, we'll dequeue this and send it.
12694 */
12695int
12696ctl_queue_sense(union ctl_io *io)
12697{
12698	struct ctl_lun *lun;
12699	struct ctl_softc *ctl_softc;
12700	uint32_t initidx, targ_lun;
12701
12702	ctl_softc = control_softc;
12703
12704	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12705
12706	/*
12707	 * LUN lookup will likely move to the ctl_work_thread() once we
12708	 * have our new queueing infrastructure (that doesn't put things on
12709	 * a per-LUN queue initially).  That is so that we can handle
12710	 * things like an INQUIRY to a LUN that we don't have enabled.  We
12711	 * can't deal with that right now.
12712	 */
12713	mtx_lock(&ctl_softc->ctl_lock);
12714
12715	/*
12716	 * If we don't have a LUN for this, just toss the sense
12717	 * information.
12718	 */
12719	targ_lun = io->io_hdr.nexus.targ_lun;
12720	if (io->io_hdr.nexus.lun_map_fn != NULL)
12721		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
12722	if ((targ_lun < CTL_MAX_LUNS)
12723	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12724		lun = ctl_softc->ctl_luns[targ_lun];
12725	else
12726		goto bailout;
12727
12728	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12729
12730	/*
12731	 * Already have CA set for this LUN...toss the sense information.
12732	 */
12733	if (ctl_is_set(lun->have_ca, initidx))
12734		goto bailout;
12735
12736	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
12737	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
12738	       sizeof(io->scsiio.sense_data)));
12739	ctl_set_mask(lun->have_ca, initidx);
12740
12741bailout:
12742	mtx_unlock(&ctl_softc->ctl_lock);
12743
12744	ctl_free_io(io);
12745
12746	return (CTL_RETVAL_COMPLETE);
12747}
12748
12749/*
12750 * Primary command inlet from frontend ports.  All SCSI and task I/O
12751 * requests must go through this function.
12752 */
12753int
12754ctl_queue(union ctl_io *io)
12755{
12756	struct ctl_softc *ctl_softc;
12757
12758	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12759
12760	ctl_softc = control_softc;
12761
12762#ifdef CTL_TIME_IO
12763	io->io_hdr.start_time = time_uptime;
12764	getbintime(&io->io_hdr.start_bt);
12765#endif /* CTL_TIME_IO */
12766
12767	mtx_lock(&ctl_softc->ctl_lock);
12768
12769	switch (io->io_hdr.io_type) {
12770	case CTL_IO_SCSI:
12771		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
12772				   links);
12773		break;
12774	case CTL_IO_TASK:
12775		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
12776		/*
12777		 * Set the task pending flag.  This is necessary to close a
12778		 * race condition with the FETD:
12779		 *
12780		 * - FETD submits a task management command, like an abort.
12781		 * - Back end calls fe_datamove() to move the data for the
12782		 *   aborted command.  The FETD can't really accept it, but
12783		 *   if it did, it would end up transmitting data for a
12784		 *   command that the initiator told us to abort.
12785		 *
12786		 * We close the race condition by setting the flag here,
12787		 * and checking it in ctl_datamove(), before calling the
12788		 * FETD's fe_datamove routine.  If we've got a task
12789		 * pending, we run the task queue and then check to see
12790		 * whether our particular I/O has been aborted.
12791		 */
12792		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
12793		break;
12794	default:
12795		mtx_unlock(&ctl_softc->ctl_lock);
12796		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12797		return (-EINVAL);
12798		break; /* NOTREACHED */
12799	}
12800	mtx_unlock(&ctl_softc->ctl_lock);
12801
12802	ctl_wakeup_thread();
12803
12804	return (CTL_RETVAL_COMPLETE);
12805}
12806
12807#ifdef CTL_IO_DELAY
12808static void
12809ctl_done_timer_wakeup(void *arg)
12810{
12811	union ctl_io *io;
12812
12813	io = (union ctl_io *)arg;
12814	ctl_done_lock(io, /*have_lock*/ 0);
12815}
12816#endif /* CTL_IO_DELAY */
12817
12818void
12819ctl_done_lock(union ctl_io *io, int have_lock)
12820{
12821	struct ctl_softc *ctl_softc;
12822#ifndef CTL_DONE_THREAD
12823	union ctl_io *xio;
12824#endif /* !CTL_DONE_THREAD */
12825
12826	ctl_softc = control_softc;
12827
12828	if (have_lock == 0)
12829		mtx_lock(&ctl_softc->ctl_lock);
12830
12831	/*
12832	 * Enable this to catch duplicate completion issues.
12833	 */
12834#if 0
12835	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12836		printf("%s: type %d msg %d cdb %x iptl: "
12837		       "%d:%d:%d:%d tag 0x%04x "
12838		       "flag %#x status %x\n",
12839			__func__,
12840			io->io_hdr.io_type,
12841			io->io_hdr.msg_type,
12842			io->scsiio.cdb[0],
12843			io->io_hdr.nexus.initid.id,
12844			io->io_hdr.nexus.targ_port,
12845			io->io_hdr.nexus.targ_target.id,
12846			io->io_hdr.nexus.targ_lun,
12847			(io->io_hdr.io_type ==
12848			CTL_IO_TASK) ?
12849			io->taskio.tag_num :
12850			io->scsiio.tag_num,
12851		        io->io_hdr.flags,
12852			io->io_hdr.status);
12853	} else
12854		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12855#endif
12856
12857	/*
12858	 * This is an internal copy of an I/O, and should not go through
12859	 * the normal done processing logic.
12860	 */
12861	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
12862		if (have_lock == 0)
12863			mtx_unlock(&ctl_softc->ctl_lock);
12864		return;
12865	}
12866
12867	/*
12868	 * We need to send a msg to the serializing shelf to finish the IO
12869	 * as well.  We don't send a finish message to the other shelf if
12870	 * this is a task management command.  Task management commands
12871	 * aren't serialized in the OOA queue, but rather just executed on
12872	 * both shelf controllers for commands that originated on that
12873	 * controller.
12874	 */
12875	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
12876	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
12877		union ctl_ha_msg msg_io;
12878
12879		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
12880		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
12881		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
12882		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
12883		}
12884		/* continue on to finish IO */
12885	}
12886#ifdef CTL_IO_DELAY
12887	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12888		struct ctl_lun *lun;
12889
12890		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12891
12892		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12893	} else {
12894		struct ctl_lun *lun;
12895
12896		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12897
12898		if ((lun != NULL)
12899		 && (lun->delay_info.done_delay > 0)) {
12900			struct callout *callout;
12901
12902			callout = (struct callout *)&io->io_hdr.timer_bytes;
12903			callout_init(callout, /*mpsafe*/ 1);
12904			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12905			callout_reset(callout,
12906				      lun->delay_info.done_delay * hz,
12907				      ctl_done_timer_wakeup, io);
12908			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
12909				lun->delay_info.done_delay = 0;
12910			if (have_lock == 0)
12911				mtx_unlock(&ctl_softc->ctl_lock);
12912			return;
12913		}
12914	}
12915#endif /* CTL_IO_DELAY */
12916
12917	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
12918
12919#ifdef CTL_DONE_THREAD
12920	if (have_lock == 0)
12921		mtx_unlock(&ctl_softc->ctl_lock);
12922
12923	ctl_wakeup_thread();
12924#else /* CTL_DONE_THREAD */
12925	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
12926	     xio != NULL;
12927	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
12928
12929		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
12930
12931		ctl_process_done(xio, /*have_lock*/ 1);
12932	}
12933	if (have_lock == 0)
12934		mtx_unlock(&ctl_softc->ctl_lock);
12935#endif /* CTL_DONE_THREAD */
12936}
12937
12938void
12939ctl_done(union ctl_io *io)
12940{
12941	ctl_done_lock(io, /*have_lock*/ 0);
12942}
12943
12944int
12945ctl_isc(struct ctl_scsiio *ctsio)
12946{
12947	struct ctl_lun *lun;
12948	int retval;
12949
12950	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12951
12952	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
12953
12954	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
12955
12956	retval = lun->backend->data_submit((union ctl_io *)ctsio);
12957
12958	return (retval);
12959}
12960
12961
12962static void
12963ctl_work_thread(void *arg)
12964{
12965	struct ctl_softc *softc;
12966	union ctl_io *io;
12967	struct ctl_be_lun *be_lun;
12968	int retval;
12969
12970	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
12971
12972	softc = (struct ctl_softc *)arg;
12973	if (softc == NULL)
12974		return;
12975
12976	mtx_lock(&softc->ctl_lock);
12977	for (;;) {
12978		retval = 0;
12979
12980		/*
12981		 * We handle the queues in this order:
12982		 * - task management
12983		 * - ISC
12984		 * - done queue (to free up resources, unblock other commands)
12985		 * - RtR queue
12986		 * - incoming queue
12987		 *
12988		 * If those queues are empty, we break out of the loop and
12989		 * go to sleep.
12990		 */
12991		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
12992		if (io != NULL) {
12993			ctl_run_task_queue(softc);
12994			continue;
12995		}
12996		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
12997		if (io != NULL) {
12998			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
12999			ctl_handle_isc(io);
13000			continue;
13001		}
13002		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
13003		if (io != NULL) {
13004			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
13005			/* clear any blocked commands, call fe_done */
13006			mtx_unlock(&softc->ctl_lock);
13007			/*
13008			 * XXX KDM
13009			 * Call this without a lock for now.  This will
13010			 * depend on whether there is any way the FETD can
13011			 * sleep or deadlock if called with the CTL lock
13012			 * held.
13013			 */
13014			retval = ctl_process_done(io, /*have_lock*/ 0);
13015			mtx_lock(&softc->ctl_lock);
13016			continue;
13017		}
13018		if (!ctl_pause_rtr) {
13019			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
13020			if (io != NULL) {
13021				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
13022				mtx_unlock(&softc->ctl_lock);
13023				retval = ctl_scsiio(&io->scsiio);
13024				if (retval != CTL_RETVAL_COMPLETE)
13025					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13026				mtx_lock(&softc->ctl_lock);
13027				continue;
13028			}
13029		}
13030		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
13031		if (io != NULL) {
13032			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
13033			mtx_unlock(&softc->ctl_lock);
13034			ctl_scsiio_precheck(softc, &io->scsiio);
13035			mtx_lock(&softc->ctl_lock);
13036			continue;
13037		}
13038		/*
13039		 * We might want to move this to a separate thread, so that
13040		 * configuration requests (in this case LUN creations)
13041		 * won't impact the I/O path.
13042		 */
13043		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13044		if (be_lun != NULL) {
13045			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13046			mtx_unlock(&softc->ctl_lock);
13047			ctl_create_lun(be_lun);
13048			mtx_lock(&softc->ctl_lock);
13049			continue;
13050		}
13051
13052		/* XXX KDM use the PDROP flag?? */
13053		/* Sleep until we have something to do. */
13054		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "ctl_work", 0);
13055
13056		/* Back to the top of the loop to see what woke us up. */
13057		continue;
13058	}
13059}
13060
13061void
13062ctl_wakeup_thread()
13063{
13064	struct ctl_softc *softc;
13065
13066	softc = control_softc;
13067
13068	wakeup_one(softc);
13069}
13070
13071/* Initialization and failover */
13072
13073void
13074ctl_init_isc_msg(void)
13075{
13076	printf("CTL: Still calling this thing\n");
13077}
13078
13079/*
13080 * Init component
13081 * 	Initializes component into configuration defined by bootMode
13082 *	(see hasc-sv.c)
13083 *  	returns hasc_Status:
13084 * 		OK
13085 *		ERROR - fatal error
13086 */
13087static ctl_ha_comp_status
13088ctl_isc_init(struct ctl_ha_component *c)
13089{
13090	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13091
13092	c->status = ret;
13093	return ret;
13094}
13095
13096/* Start component
13097 * 	Starts component in state requested. If component starts successfully,
13098 *	it must set its own state to the requestrd state
13099 *	When requested state is HASC_STATE_HA, the component may refine it
13100 * 	by adding _SLAVE or _MASTER flags.
13101 *	Currently allowed state transitions are:
13102 *	UNKNOWN->HA		- initial startup
13103 *	UNKNOWN->SINGLE - initial startup when no parter detected
13104 *	HA->SINGLE		- failover
13105 * returns ctl_ha_comp_status:
13106 * 		OK	- component successfully started in requested state
13107 *		FAILED  - could not start the requested state, failover may
13108 * 			  be possible
13109 *		ERROR	- fatal error detected, no future startup possible
13110 */
13111static ctl_ha_comp_status
13112ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13113{
13114	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13115
13116	printf("%s: go\n", __func__);
13117
13118	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13119	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13120		ctl_is_single = 0;
13121		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13122		    != CTL_HA_STATUS_SUCCESS) {
13123			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13124			ret = CTL_HA_COMP_STATUS_ERROR;
13125		}
13126	} else if (CTL_HA_STATE_IS_HA(c->state)
13127		&& CTL_HA_STATE_IS_SINGLE(state)){
13128		// HA->SINGLE transition
13129	        ctl_failover();
13130		ctl_is_single = 1;
13131	} else {
13132		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13133		       c->state, state);
13134		ret = CTL_HA_COMP_STATUS_ERROR;
13135	}
13136	if (CTL_HA_STATE_IS_SINGLE(state))
13137		ctl_is_single = 1;
13138
13139	c->state = state;
13140	c->status = ret;
13141	return ret;
13142}
13143
13144/*
13145 * Quiesce component
13146 * The component must clear any error conditions (set status to OK) and
13147 * prepare itself to another Start call
13148 * returns ctl_ha_comp_status:
13149 * 	OK
13150 *	ERROR
13151 */
13152static ctl_ha_comp_status
13153ctl_isc_quiesce(struct ctl_ha_component *c)
13154{
13155	int ret = CTL_HA_COMP_STATUS_OK;
13156
13157	ctl_pause_rtr = 1;
13158	c->status = ret;
13159	return ret;
13160}
13161
13162struct ctl_ha_component ctl_ha_component_ctlisc =
13163{
13164	.name = "CTL ISC",
13165	.state = CTL_HA_STATE_UNKNOWN,
13166	.init = ctl_isc_init,
13167	.start = ctl_isc_start,
13168	.quiesce = ctl_isc_quiesce
13169};
13170
13171/*
13172 *  vim: ts=8
13173 */
13174