ctl.c revision 267537
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 267537 2014-06-16 11:00:14Z 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), supported pages (0x00),
335 * Block limits (0xB0) and Logical Block Provisioning (0xB2)
336 */
337#define SCSI_EVPD_NUM_SUPPORTED_PAGES	5
338
339static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
340				  int param);
341static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
342static int ctl_init(void);
343void ctl_shutdown(void);
344static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
345static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
346static void ctl_ioctl_online(void *arg);
347static void ctl_ioctl_offline(void *arg);
348static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
349static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
350static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
351static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
352static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
353static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
354static int ctl_ioctl_submit_wait(union ctl_io *io);
355static void ctl_ioctl_datamove(union ctl_io *io);
356static void ctl_ioctl_done(union ctl_io *io);
357static void ctl_ioctl_hard_startstop_callback(void *arg,
358					      struct cfi_metatask *metatask);
359static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
360static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
361			      struct ctl_ooa *ooa_hdr,
362			      struct ctl_ooa_entry *kern_entries);
363static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
364		     struct thread *td);
365uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
366uint32_t ctl_port_idx(int port_num);
367#ifdef unused
368static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
369				   uint32_t targ_target, uint32_t targ_lun,
370				   int can_wait);
371static void ctl_kfree_io(union ctl_io *io);
372#endif /* unused */
373static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
374			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
375static int ctl_free_lun(struct ctl_lun *lun);
376static void ctl_create_lun(struct ctl_be_lun *be_lun);
377/**
378static void ctl_failover_change_pages(struct ctl_softc *softc,
379				      struct ctl_scsiio *ctsio, int master);
380**/
381
382static int ctl_do_mode_select(union ctl_io *io);
383static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
384			   uint64_t res_key, uint64_t sa_res_key,
385			   uint8_t type, uint32_t residx,
386			   struct ctl_scsiio *ctsio,
387			   struct scsi_per_res_out *cdb,
388			   struct scsi_per_res_out_parms* param);
389static void ctl_pro_preempt_other(struct ctl_lun *lun,
390				  union ctl_ha_msg *msg);
391static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
392static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
394static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
395static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
396					 int alloc_len);
397static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
398static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
399static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
400static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
401static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
402static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
403					 union ctl_io *ooa_io);
404static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
405				union ctl_io *starting_io);
406static int ctl_check_blocked(struct ctl_lun *lun);
407static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
408				struct ctl_lun *lun,
409				struct ctl_cmd_entry *entry,
410				struct ctl_scsiio *ctsio);
411//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
412static void ctl_failover(void);
413static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
414			       struct ctl_scsiio *ctsio);
415static int ctl_scsiio(struct ctl_scsiio *ctsio);
416
417static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
418static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
419			    ctl_ua_type ua_type);
420static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
421			 ctl_ua_type ua_type);
422static int ctl_abort_task(union ctl_io *io);
423static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
424#ifdef CTL_IO_DELAY
425static void ctl_datamove_timer_wakeup(void *arg);
426static void ctl_done_timer_wakeup(void *arg);
427#endif /* CTL_IO_DELAY */
428
429static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
430static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
431static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
432static void ctl_datamove_remote_write(union ctl_io *io);
433static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
434static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
435static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
436static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
437				    ctl_ha_dt_cb callback);
438static void ctl_datamove_remote_read(union ctl_io *io);
439static void ctl_datamove_remote(union ctl_io *io);
440static int ctl_process_done(union ctl_io *io, int have_lock);
441static void ctl_work_thread(void *arg);
442
443/*
444 * Load the serialization table.  This isn't very pretty, but is probably
445 * the easiest way to do it.
446 */
447#include "ctl_ser_table.c"
448
449/*
450 * We only need to define open, close and ioctl routines for this driver.
451 */
452static struct cdevsw ctl_cdevsw = {
453	.d_version =	D_VERSION,
454	.d_flags =	0,
455	.d_open =	ctl_open,
456	.d_close =	ctl_close,
457	.d_ioctl =	ctl_ioctl,
458	.d_name =	"ctl",
459};
460
461
462MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
463
464static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
465
466static moduledata_t ctl_moduledata = {
467	"ctl",
468	ctl_module_event_handler,
469	NULL
470};
471
472DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
473MODULE_VERSION(ctl, 1);
474
475static void
476ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
477			    union ctl_ha_msg *msg_info)
478{
479	struct ctl_scsiio *ctsio;
480
481	if (msg_info->hdr.original_sc == NULL) {
482		printf("%s: original_sc == NULL!\n", __func__);
483		/* XXX KDM now what? */
484		return;
485	}
486
487	ctsio = &msg_info->hdr.original_sc->scsiio;
488	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
489	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
490	ctsio->io_hdr.status = msg_info->hdr.status;
491	ctsio->scsi_status = msg_info->scsi.scsi_status;
492	ctsio->sense_len = msg_info->scsi.sense_len;
493	ctsio->sense_residual = msg_info->scsi.sense_residual;
494	ctsio->residual = msg_info->scsi.residual;
495	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
496	       sizeof(ctsio->sense_data));
497	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
498	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
499	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
500	ctl_wakeup_thread();
501}
502
503static void
504ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
505				union ctl_ha_msg *msg_info)
506{
507	struct ctl_scsiio *ctsio;
508
509	if (msg_info->hdr.serializing_sc == NULL) {
510		printf("%s: serializing_sc == NULL!\n", __func__);
511		/* XXX KDM now what? */
512		return;
513	}
514
515	ctsio = &msg_info->hdr.serializing_sc->scsiio;
516#if 0
517	/*
518	 * Attempt to catch the situation where an I/O has
519	 * been freed, and we're using it again.
520	 */
521	if (ctsio->io_hdr.io_type == 0xff) {
522		union ctl_io *tmp_io;
523		tmp_io = (union ctl_io *)ctsio;
524		printf("%s: %p use after free!\n", __func__,
525		       ctsio);
526		printf("%s: type %d msg %d cdb %x iptl: "
527		       "%d:%d:%d:%d tag 0x%04x "
528		       "flag %#x status %x\n",
529			__func__,
530			tmp_io->io_hdr.io_type,
531			tmp_io->io_hdr.msg_type,
532			tmp_io->scsiio.cdb[0],
533			tmp_io->io_hdr.nexus.initid.id,
534			tmp_io->io_hdr.nexus.targ_port,
535			tmp_io->io_hdr.nexus.targ_target.id,
536			tmp_io->io_hdr.nexus.targ_lun,
537			(tmp_io->io_hdr.io_type ==
538			CTL_IO_TASK) ?
539			tmp_io->taskio.tag_num :
540			tmp_io->scsiio.tag_num,
541		        tmp_io->io_hdr.flags,
542			tmp_io->io_hdr.status);
543	}
544#endif
545	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
546	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
547	ctl_wakeup_thread();
548}
549
550/*
551 * ISC (Inter Shelf Communication) event handler.  Events from the HA
552 * subsystem come in here.
553 */
554static void
555ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
556{
557	struct ctl_softc *ctl_softc;
558	union ctl_io *io;
559	struct ctl_prio *presio;
560	ctl_ha_status isc_status;
561
562	ctl_softc = control_softc;
563	io = NULL;
564
565
566#if 0
567	printf("CTL: Isc Msg event %d\n", event);
568#endif
569	if (event == CTL_HA_EVT_MSG_RECV) {
570		union ctl_ha_msg msg_info;
571
572		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
573					     sizeof(msg_info), /*wait*/ 0);
574#if 0
575		printf("CTL: msg_type %d\n", msg_info.msg_type);
576#endif
577		if (isc_status != 0) {
578			printf("Error receiving message, status = %d\n",
579			       isc_status);
580			return;
581		}
582		mtx_lock(&ctl_softc->ctl_lock);
583
584		switch (msg_info.hdr.msg_type) {
585		case CTL_MSG_SERIALIZE:
586#if 0
587			printf("Serialize\n");
588#endif
589			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
590			if (io == NULL) {
591				printf("ctl_isc_event_handler: can't allocate "
592				       "ctl_io!\n");
593				/* Bad Juju */
594				/* Need to set busy and send msg back */
595				mtx_unlock(&ctl_softc->ctl_lock);
596				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
597				msg_info.hdr.status = CTL_SCSI_ERROR;
598				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
599				msg_info.scsi.sense_len = 0;
600			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
601				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
602				}
603				goto bailout;
604			}
605			ctl_zero_io(io);
606			// populate ctsio from msg_info
607			io->io_hdr.io_type = CTL_IO_SCSI;
608			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
609			io->io_hdr.original_sc = msg_info.hdr.original_sc;
610#if 0
611			printf("pOrig %x\n", (int)msg_info.original_sc);
612#endif
613			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
614					    CTL_FLAG_IO_ACTIVE;
615			/*
616			 * If we're in serialization-only mode, we don't
617			 * want to go through full done processing.  Thus
618			 * the COPY flag.
619			 *
620			 * XXX KDM add another flag that is more specific.
621			 */
622			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
623				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
624			io->io_hdr.nexus = msg_info.hdr.nexus;
625#if 0
626			printf("targ %d, port %d, iid %d, lun %d\n",
627			       io->io_hdr.nexus.targ_target.id,
628			       io->io_hdr.nexus.targ_port,
629			       io->io_hdr.nexus.initid.id,
630			       io->io_hdr.nexus.targ_lun);
631#endif
632			io->scsiio.tag_num = msg_info.scsi.tag_num;
633			io->scsiio.tag_type = msg_info.scsi.tag_type;
634			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
635			       CTL_MAX_CDBLEN);
636			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
637				struct ctl_cmd_entry *entry;
638				uint8_t opcode;
639
640				opcode = io->scsiio.cdb[0];
641				entry = &ctl_cmd_table[opcode];
642				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
643				io->io_hdr.flags |=
644					entry->flags & CTL_FLAG_DATA_MASK;
645			}
646			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
647					   &io->io_hdr, links);
648			ctl_wakeup_thread();
649			break;
650
651		/* Performed on the Originating SC, XFER mode only */
652		case CTL_MSG_DATAMOVE: {
653			struct ctl_sg_entry *sgl;
654			int i, j;
655
656			io = msg_info.hdr.original_sc;
657			if (io == NULL) {
658				printf("%s: original_sc == NULL!\n", __func__);
659				/* XXX KDM do something here */
660				break;
661			}
662			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
663			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
664			/*
665			 * Keep track of this, we need to send it back over
666			 * when the datamove is complete.
667			 */
668			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
669
670			if (msg_info.dt.sg_sequence == 0) {
671				/*
672				 * XXX KDM we use the preallocated S/G list
673				 * here, but we'll need to change this to
674				 * dynamic allocation if we need larger S/G
675				 * lists.
676				 */
677				if (msg_info.dt.kern_sg_entries >
678				    sizeof(io->io_hdr.remote_sglist) /
679				    sizeof(io->io_hdr.remote_sglist[0])) {
680					printf("%s: number of S/G entries "
681					    "needed %u > allocated num %zd\n",
682					    __func__,
683					    msg_info.dt.kern_sg_entries,
684					    sizeof(io->io_hdr.remote_sglist)/
685					    sizeof(io->io_hdr.remote_sglist[0]));
686
687					/*
688					 * XXX KDM send a message back to
689					 * the other side to shut down the
690					 * DMA.  The error will come back
691					 * through via the normal channel.
692					 */
693					break;
694				}
695				sgl = io->io_hdr.remote_sglist;
696				memset(sgl, 0,
697				       sizeof(io->io_hdr.remote_sglist));
698
699				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
700
701				io->scsiio.kern_sg_entries =
702					msg_info.dt.kern_sg_entries;
703				io->scsiio.rem_sg_entries =
704					msg_info.dt.kern_sg_entries;
705				io->scsiio.kern_data_len =
706					msg_info.dt.kern_data_len;
707				io->scsiio.kern_total_len =
708					msg_info.dt.kern_total_len;
709				io->scsiio.kern_data_resid =
710					msg_info.dt.kern_data_resid;
711				io->scsiio.kern_rel_offset =
712					msg_info.dt.kern_rel_offset;
713				/*
714				 * Clear out per-DMA flags.
715				 */
716				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
717				/*
718				 * Add per-DMA flags that are set for this
719				 * particular DMA request.
720				 */
721				io->io_hdr.flags |= msg_info.dt.flags &
722						    CTL_FLAG_RDMA_MASK;
723			} else
724				sgl = (struct ctl_sg_entry *)
725					io->scsiio.kern_data_ptr;
726
727			for (i = msg_info.dt.sent_sg_entries, j = 0;
728			     i < (msg_info.dt.sent_sg_entries +
729			     msg_info.dt.cur_sg_entries); i++, j++) {
730				sgl[i].addr = msg_info.dt.sg_list[j].addr;
731				sgl[i].len = msg_info.dt.sg_list[j].len;
732
733#if 0
734				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
735				       __func__,
736				       msg_info.dt.sg_list[j].addr,
737				       msg_info.dt.sg_list[j].len,
738				       sgl[i].addr, sgl[i].len, j, i);
739#endif
740			}
741#if 0
742			memcpy(&sgl[msg_info.dt.sent_sg_entries],
743			       msg_info.dt.sg_list,
744			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
745#endif
746
747			/*
748			 * If this is the last piece of the I/O, we've got
749			 * the full S/G list.  Queue processing in the thread.
750			 * Otherwise wait for the next piece.
751			 */
752			if (msg_info.dt.sg_last != 0) {
753				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
754						   &io->io_hdr, links);
755				ctl_wakeup_thread();
756			}
757			break;
758		}
759		/* Performed on the Serializing (primary) SC, XFER mode only */
760		case CTL_MSG_DATAMOVE_DONE: {
761			if (msg_info.hdr.serializing_sc == NULL) {
762				printf("%s: serializing_sc == NULL!\n",
763				       __func__);
764				/* XXX KDM now what? */
765				break;
766			}
767			/*
768			 * We grab the sense information here in case
769			 * there was a failure, so we can return status
770			 * back to the initiator.
771			 */
772			io = msg_info.hdr.serializing_sc;
773			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
774			io->io_hdr.status = msg_info.hdr.status;
775			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
776			io->scsiio.sense_len = msg_info.scsi.sense_len;
777			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
778			io->io_hdr.port_status = msg_info.scsi.fetd_status;
779			io->scsiio.residual = msg_info.scsi.residual;
780			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
781			       sizeof(io->scsiio.sense_data));
782
783			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
784					   &io->io_hdr, links);
785			ctl_wakeup_thread();
786			break;
787		}
788
789		/* Preformed on Originating SC, SER_ONLY mode */
790		case CTL_MSG_R2R:
791			io = msg_info.hdr.original_sc;
792			if (io == NULL) {
793				printf("%s: Major Bummer\n", __func__);
794				mtx_unlock(&ctl_softc->ctl_lock);
795				return;
796			} else {
797#if 0
798				printf("pOrig %x\n",(int) ctsio);
799#endif
800			}
801			io->io_hdr.msg_type = CTL_MSG_R2R;
802			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
803			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
804					   &io->io_hdr, links);
805			ctl_wakeup_thread();
806			break;
807
808		/*
809		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
810		 * mode.
811		 * Performed on the Originating (i.e. secondary) SC in XFER
812		 * mode
813		 */
814		case CTL_MSG_FINISH_IO:
815			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
816				ctl_isc_handler_finish_xfer(ctl_softc,
817							    &msg_info);
818			else
819				ctl_isc_handler_finish_ser_only(ctl_softc,
820								&msg_info);
821			break;
822
823		/* Preformed on Originating SC */
824		case CTL_MSG_BAD_JUJU:
825			io = msg_info.hdr.original_sc;
826			if (io == NULL) {
827				printf("%s: Bad JUJU!, original_sc is NULL!\n",
828				       __func__);
829				break;
830			}
831			ctl_copy_sense_data(&msg_info, io);
832			/*
833			 * IO should have already been cleaned up on other
834			 * SC so clear this flag so we won't send a message
835			 * back to finish the IO there.
836			 */
837			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
838			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
839
840			/* io = msg_info.hdr.serializing_sc; */
841			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
842		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
843					   &io->io_hdr, links);
844			ctl_wakeup_thread();
845			break;
846
847		/* Handle resets sent from the other side */
848		case CTL_MSG_MANAGE_TASKS: {
849			struct ctl_taskio *taskio;
850			taskio = (struct ctl_taskio *)ctl_alloc_io(
851				(void *)ctl_softc->othersc_pool);
852			if (taskio == NULL) {
853				printf("ctl_isc_event_handler: can't allocate "
854				       "ctl_io!\n");
855				/* Bad Juju */
856				/* should I just call the proper reset func
857				   here??? */
858				mtx_unlock(&ctl_softc->ctl_lock);
859				goto bailout;
860			}
861			ctl_zero_io((union ctl_io *)taskio);
862			taskio->io_hdr.io_type = CTL_IO_TASK;
863			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
864			taskio->io_hdr.nexus = msg_info.hdr.nexus;
865			taskio->task_action = msg_info.task.task_action;
866			taskio->tag_num = msg_info.task.tag_num;
867			taskio->tag_type = msg_info.task.tag_type;
868#ifdef CTL_TIME_IO
869			taskio->io_hdr.start_time = time_uptime;
870			getbintime(&taskio->io_hdr.start_bt);
871#if 0
872			cs_prof_gettime(&taskio->io_hdr.start_ticks);
873#endif
874#endif /* CTL_TIME_IO */
875		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
876					   &taskio->io_hdr, links);
877			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
878			ctl_wakeup_thread();
879			break;
880		}
881		/* Persistent Reserve action which needs attention */
882		case CTL_MSG_PERS_ACTION:
883			presio = (struct ctl_prio *)ctl_alloc_io(
884				(void *)ctl_softc->othersc_pool);
885			if (presio == NULL) {
886				printf("ctl_isc_event_handler: can't allocate "
887				       "ctl_io!\n");
888				/* Bad Juju */
889				/* Need to set busy and send msg back */
890				mtx_unlock(&ctl_softc->ctl_lock);
891				goto bailout;
892			}
893			ctl_zero_io((union ctl_io *)presio);
894			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
895			presio->pr_msg = msg_info.pr;
896		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
897					   &presio->io_hdr, links);
898			ctl_wakeup_thread();
899			break;
900		case CTL_MSG_SYNC_FE:
901			rcv_sync_msg = 1;
902			break;
903		case CTL_MSG_APS_LOCK: {
904			// It's quicker to execute this then to
905			// queue it.
906			struct ctl_lun *lun;
907			struct ctl_page_index *page_index;
908			struct copan_aps_subpage *current_sp;
909			uint32_t targ_lun;
910
911			targ_lun = msg_info.hdr.nexus.targ_lun;
912			if (msg_info.hdr.nexus.lun_map_fn != NULL)
913				targ_lun = msg_info.hdr.nexus.lun_map_fn(msg_info.hdr.nexus.lun_map_arg, targ_lun);
914
915			lun = ctl_softc->ctl_luns[targ_lun];
916			page_index = &lun->mode_pages.index[index_to_aps_page];
917			current_sp = (struct copan_aps_subpage *)
918				     (page_index->page_data +
919				     (page_index->page_len * CTL_PAGE_CURRENT));
920
921			current_sp->lock_active = msg_info.aps.lock_flag;
922		        break;
923		}
924		default:
925		        printf("How did I get here?\n");
926		}
927		mtx_unlock(&ctl_softc->ctl_lock);
928	} else if (event == CTL_HA_EVT_MSG_SENT) {
929		if (param != CTL_HA_STATUS_SUCCESS) {
930			printf("Bad status from ctl_ha_msg_send status %d\n",
931			       param);
932		}
933		return;
934	} else if (event == CTL_HA_EVT_DISCONNECT) {
935		printf("CTL: Got a disconnect from Isc\n");
936		return;
937	} else {
938		printf("ctl_isc_event_handler: Unknown event %d\n", event);
939		return;
940	}
941
942bailout:
943	return;
944}
945
946static void
947ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
948{
949	struct scsi_sense_data *sense;
950
951	sense = &dest->scsiio.sense_data;
952	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
953	dest->scsiio.scsi_status = src->scsi.scsi_status;
954	dest->scsiio.sense_len = src->scsi.sense_len;
955	dest->io_hdr.status = src->hdr.status;
956}
957
958static int
959ctl_init(void)
960{
961	struct ctl_softc *softc;
962	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
963	struct ctl_frontend *fe;
964        uint8_t sc_id =0;
965	int i, error, retval;
966	//int isc_retval;
967
968	retval = 0;
969	ctl_pause_rtr = 0;
970        rcv_sync_msg = 0;
971
972	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
973			       M_WAITOK | M_ZERO);
974	softc = control_softc;
975
976	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
977			      "cam/ctl");
978
979	softc->dev->si_drv1 = softc;
980
981	/*
982	 * By default, return a "bad LUN" peripheral qualifier for unknown
983	 * LUNs.  The user can override this default using the tunable or
984	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
985	 */
986	softc->inquiry_pq_no_lun = 1;
987	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
988			  &softc->inquiry_pq_no_lun);
989	sysctl_ctx_init(&softc->sysctl_ctx);
990	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
991		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
992		CTLFLAG_RD, 0, "CAM Target Layer");
993
994	if (softc->sysctl_tree == NULL) {
995		printf("%s: unable to allocate sysctl tree\n", __func__);
996		destroy_dev(softc->dev);
997		free(control_softc, M_DEVBUF);
998		control_softc = NULL;
999		return (ENOMEM);
1000	}
1001
1002	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1003		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1004		       "inquiry_pq_no_lun", CTLFLAG_RW,
1005		       &softc->inquiry_pq_no_lun, 0,
1006		       "Report no lun possible for invalid LUNs");
1007
1008	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1009	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
1010	softc->open_count = 0;
1011
1012	/*
1013	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1014	 * the drive.
1015	 */
1016	softc->flags = CTL_FLAG_REAL_SYNC;
1017
1018	/*
1019	 * In Copan's HA scheme, the "master" and "slave" roles are
1020	 * figured out through the slot the controller is in.  Although it
1021	 * is an active/active system, someone has to be in charge.
1022 	 */
1023#ifdef NEEDTOPORT
1024        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1025#endif
1026
1027        if (sc_id == 0) {
1028		softc->flags |= CTL_FLAG_MASTER_SHELF;
1029		persis_offset = 0;
1030	} else
1031		persis_offset = CTL_MAX_INITIATORS;
1032
1033	/*
1034	 * XXX KDM need to figure out where we want to get our target ID
1035	 * and WWID.  Is it different on each port?
1036	 */
1037	softc->target.id = 0;
1038	softc->target.wwid[0] = 0x12345678;
1039	softc->target.wwid[1] = 0x87654321;
1040	STAILQ_INIT(&softc->lun_list);
1041	STAILQ_INIT(&softc->pending_lun_queue);
1042	STAILQ_INIT(&softc->task_queue);
1043	STAILQ_INIT(&softc->incoming_queue);
1044	STAILQ_INIT(&softc->rtr_queue);
1045	STAILQ_INIT(&softc->done_queue);
1046	STAILQ_INIT(&softc->isc_queue);
1047	STAILQ_INIT(&softc->fe_list);
1048	STAILQ_INIT(&softc->be_list);
1049	STAILQ_INIT(&softc->io_pools);
1050
1051	/*
1052	 * We don't bother calling these with ctl_lock held here, because,
1053	 * in theory, no one else can try to do anything while we're in our
1054	 * module init routine.
1055	 */
1056	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1057			    &internal_pool)!= 0){
1058		printf("ctl: can't allocate %d entry internal pool, "
1059		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1060		return (ENOMEM);
1061	}
1062
1063	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1064			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1065		printf("ctl: can't allocate %d entry emergency pool, "
1066		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1067		ctl_pool_free(internal_pool);
1068		return (ENOMEM);
1069	}
1070
1071	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1072	                    &other_pool) != 0)
1073	{
1074		printf("ctl: can't allocate %d entry other SC pool, "
1075		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1076		ctl_pool_free(internal_pool);
1077		ctl_pool_free(emergency_pool);
1078		return (ENOMEM);
1079	}
1080
1081	softc->internal_pool = internal_pool;
1082	softc->emergency_pool = emergency_pool;
1083	softc->othersc_pool = other_pool;
1084
1085	if (worker_threads > MAXCPU || worker_threads == 0) {
1086		printf("invalid kern.cam.ctl.worker_threads value; "
1087		    "setting to 1");
1088		worker_threads = 1;
1089	} else if (worker_threads < 0) {
1090		if (mp_ncpus > 2) {
1091			/*
1092			 * Using more than two worker threads actually hurts
1093			 * performance due to lock contention.
1094			 */
1095			worker_threads = 2;
1096		} else {
1097			worker_threads = 1;
1098		}
1099	}
1100
1101	for (i = 0; i < worker_threads; i++) {
1102		error = kproc_kthread_add(ctl_work_thread, softc,
1103		    &softc->work_thread, NULL, 0, 0, "ctl", "work%d", i);
1104		if (error != 0) {
1105			printf("error creating CTL work thread!\n");
1106			ctl_pool_free(internal_pool);
1107			ctl_pool_free(emergency_pool);
1108			ctl_pool_free(other_pool);
1109			return (error);
1110		}
1111	}
1112	if (bootverbose)
1113		printf("ctl: CAM Target Layer loaded\n");
1114
1115	/*
1116	 * Initialize the initiator and portname mappings
1117	 */
1118	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1119
1120	/*
1121	 * Initialize the ioctl front end.
1122	 */
1123	fe = &softc->ioctl_info.fe;
1124	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1125	fe->port_type = CTL_PORT_IOCTL;
1126	fe->num_requested_ctl_io = 100;
1127	fe->port_name = softc->ioctl_info.port_name;
1128	fe->port_online = ctl_ioctl_online;
1129	fe->port_offline = ctl_ioctl_offline;
1130	fe->onoff_arg = &softc->ioctl_info;
1131	fe->targ_enable = ctl_ioctl_targ_enable;
1132	fe->targ_disable = ctl_ioctl_targ_disable;
1133	fe->lun_enable = ctl_ioctl_lun_enable;
1134	fe->lun_disable = ctl_ioctl_lun_disable;
1135	fe->targ_lun_arg = &softc->ioctl_info;
1136	fe->fe_datamove = ctl_ioctl_datamove;
1137	fe->fe_done = ctl_ioctl_done;
1138	fe->max_targets = 15;
1139	fe->max_target_id = 15;
1140
1141	if (ctl_frontend_register(&softc->ioctl_info.fe,
1142	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1143		printf("ctl: ioctl front end registration failed, will "
1144		       "continue anyway\n");
1145	}
1146
1147#ifdef CTL_IO_DELAY
1148	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1149		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1150		       sizeof(struct callout), CTL_TIMER_BYTES);
1151		return (EINVAL);
1152	}
1153#endif /* CTL_IO_DELAY */
1154
1155	return (0);
1156}
1157
1158void
1159ctl_shutdown(void)
1160{
1161	struct ctl_softc *softc;
1162	struct ctl_lun *lun, *next_lun;
1163	struct ctl_io_pool *pool;
1164
1165	softc = (struct ctl_softc *)control_softc;
1166
1167	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1168		printf("ctl: ioctl front end deregistration failed\n");
1169
1170	mtx_lock(&softc->ctl_lock);
1171
1172	/*
1173	 * Free up each LUN.
1174	 */
1175	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1176		next_lun = STAILQ_NEXT(lun, links);
1177		ctl_free_lun(lun);
1178	}
1179
1180	mtx_unlock(&softc->ctl_lock);
1181
1182	/*
1183	 * This will rip the rug out from under any FETDs or anyone else
1184	 * that has a pool allocated.  Since we increment our module
1185	 * refcount any time someone outside the main CTL module allocates
1186	 * a pool, we shouldn't have any problems here.  The user won't be
1187	 * able to unload the CTL module until client modules have
1188	 * successfully unloaded.
1189	 */
1190	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1191		ctl_pool_free(pool);
1192
1193#if 0
1194	ctl_shutdown_thread(softc->work_thread);
1195#endif
1196
1197	mtx_destroy(&softc->pool_lock);
1198	mtx_destroy(&softc->ctl_lock);
1199
1200	destroy_dev(softc->dev);
1201
1202	sysctl_ctx_free(&softc->sysctl_ctx);
1203
1204	free(control_softc, M_DEVBUF);
1205	control_softc = NULL;
1206
1207	if (bootverbose)
1208		printf("ctl: CAM Target Layer unloaded\n");
1209}
1210
1211static int
1212ctl_module_event_handler(module_t mod, int what, void *arg)
1213{
1214
1215	switch (what) {
1216	case MOD_LOAD:
1217		return (ctl_init());
1218	case MOD_UNLOAD:
1219		return (EBUSY);
1220	default:
1221		return (EOPNOTSUPP);
1222	}
1223}
1224
1225/*
1226 * XXX KDM should we do some access checks here?  Bump a reference count to
1227 * prevent a CTL module from being unloaded while someone has it open?
1228 */
1229static int
1230ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1231{
1232	return (0);
1233}
1234
1235static int
1236ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1237{
1238	return (0);
1239}
1240
1241int
1242ctl_port_enable(ctl_port_type port_type)
1243{
1244	struct ctl_softc *softc;
1245	struct ctl_frontend *fe;
1246
1247	if (ctl_is_single == 0) {
1248		union ctl_ha_msg msg_info;
1249		int isc_retval;
1250
1251#if 0
1252		printf("%s: HA mode, synchronizing frontend enable\n",
1253		        __func__);
1254#endif
1255		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1256	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1257		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1258			printf("Sync msg send error retval %d\n", isc_retval);
1259		}
1260		if (!rcv_sync_msg) {
1261			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1262			        sizeof(msg_info), 1);
1263		}
1264#if 0
1265        	printf("CTL:Frontend Enable\n");
1266	} else {
1267		printf("%s: single mode, skipping frontend synchronization\n",
1268		        __func__);
1269#endif
1270	}
1271
1272	softc = control_softc;
1273
1274	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1275		if (port_type & fe->port_type)
1276		{
1277#if 0
1278			printf("port %d\n", fe->targ_port);
1279#endif
1280			ctl_frontend_online(fe);
1281		}
1282	}
1283
1284	return (0);
1285}
1286
1287int
1288ctl_port_disable(ctl_port_type port_type)
1289{
1290	struct ctl_softc *softc;
1291	struct ctl_frontend *fe;
1292
1293	softc = control_softc;
1294
1295	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1296		if (port_type & fe->port_type)
1297			ctl_frontend_offline(fe);
1298	}
1299
1300	return (0);
1301}
1302
1303/*
1304 * Returns 0 for success, 1 for failure.
1305 * Currently the only failure mode is if there aren't enough entries
1306 * allocated.  So, in case of a failure, look at num_entries_dropped,
1307 * reallocate and try again.
1308 */
1309int
1310ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1311	      int *num_entries_filled, int *num_entries_dropped,
1312	      ctl_port_type port_type, int no_virtual)
1313{
1314	struct ctl_softc *softc;
1315	struct ctl_frontend *fe;
1316	int entries_dropped, entries_filled;
1317	int retval;
1318	int i;
1319
1320	softc = control_softc;
1321
1322	retval = 0;
1323	entries_filled = 0;
1324	entries_dropped = 0;
1325
1326	i = 0;
1327	mtx_lock(&softc->ctl_lock);
1328	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1329		struct ctl_port_entry *entry;
1330
1331		if ((fe->port_type & port_type) == 0)
1332			continue;
1333
1334		if ((no_virtual != 0)
1335		 && (fe->virtual_port != 0))
1336			continue;
1337
1338		if (entries_filled >= num_entries_alloced) {
1339			entries_dropped++;
1340			continue;
1341		}
1342		entry = &entries[i];
1343
1344		entry->port_type = fe->port_type;
1345		strlcpy(entry->port_name, fe->port_name,
1346			sizeof(entry->port_name));
1347		entry->physical_port = fe->physical_port;
1348		entry->virtual_port = fe->virtual_port;
1349		entry->wwnn = fe->wwnn;
1350		entry->wwpn = fe->wwpn;
1351
1352		i++;
1353		entries_filled++;
1354	}
1355
1356	mtx_unlock(&softc->ctl_lock);
1357
1358	if (entries_dropped > 0)
1359		retval = 1;
1360
1361	*num_entries_dropped = entries_dropped;
1362	*num_entries_filled = entries_filled;
1363
1364	return (retval);
1365}
1366
1367static void
1368ctl_ioctl_online(void *arg)
1369{
1370	struct ctl_ioctl_info *ioctl_info;
1371
1372	ioctl_info = (struct ctl_ioctl_info *)arg;
1373
1374	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1375}
1376
1377static void
1378ctl_ioctl_offline(void *arg)
1379{
1380	struct ctl_ioctl_info *ioctl_info;
1381
1382	ioctl_info = (struct ctl_ioctl_info *)arg;
1383
1384	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1385}
1386
1387/*
1388 * Remove an initiator by port number and initiator ID.
1389 * Returns 0 for success, 1 for failure.
1390 */
1391int
1392ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1393{
1394	struct ctl_softc *softc;
1395
1396	softc = control_softc;
1397
1398	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1399
1400	if ((targ_port < 0)
1401	 || (targ_port > CTL_MAX_PORTS)) {
1402		printf("%s: invalid port number %d\n", __func__, targ_port);
1403		return (1);
1404	}
1405	if (iid > CTL_MAX_INIT_PER_PORT) {
1406		printf("%s: initiator ID %u > maximun %u!\n",
1407		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1408		return (1);
1409	}
1410
1411	mtx_lock(&softc->ctl_lock);
1412
1413	softc->wwpn_iid[targ_port][iid].in_use = 0;
1414
1415	mtx_unlock(&softc->ctl_lock);
1416
1417	return (0);
1418}
1419
1420/*
1421 * Add an initiator to the initiator map.
1422 * Returns 0 for success, 1 for failure.
1423 */
1424int
1425ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1426{
1427	struct ctl_softc *softc;
1428	int retval;
1429
1430	softc = control_softc;
1431
1432	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1433
1434	retval = 0;
1435
1436	if ((targ_port < 0)
1437	 || (targ_port > CTL_MAX_PORTS)) {
1438		printf("%s: invalid port number %d\n", __func__, targ_port);
1439		return (1);
1440	}
1441	if (iid > CTL_MAX_INIT_PER_PORT) {
1442		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1443		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1444		return (1);
1445	}
1446
1447	mtx_lock(&softc->ctl_lock);
1448
1449	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1450		/*
1451		 * We don't treat this as an error.
1452		 */
1453		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1454			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1455			       __func__, targ_port, iid, (uintmax_t)wwpn);
1456			goto bailout;
1457		}
1458
1459		/*
1460		 * This is an error, but what do we do about it?  The
1461		 * driver is telling us we have a new WWPN for this
1462		 * initiator ID, so we pretty much need to use it.
1463		 */
1464		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1465		       "still at that address\n", __func__, targ_port, iid,
1466		       (uintmax_t)wwpn,
1467		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1468
1469		/*
1470		 * XXX KDM clear have_ca and ua_pending on each LUN for
1471		 * this initiator.
1472		 */
1473	}
1474	softc->wwpn_iid[targ_port][iid].in_use = 1;
1475	softc->wwpn_iid[targ_port][iid].iid = iid;
1476	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1477	softc->wwpn_iid[targ_port][iid].port = targ_port;
1478
1479bailout:
1480
1481	mtx_unlock(&softc->ctl_lock);
1482
1483	return (retval);
1484}
1485
1486/*
1487 * XXX KDM should we pretend to do something in the target/lun
1488 * enable/disable functions?
1489 */
1490static int
1491ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1492{
1493	return (0);
1494}
1495
1496static int
1497ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1498{
1499	return (0);
1500}
1501
1502static int
1503ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1504{
1505	return (0);
1506}
1507
1508static int
1509ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1510{
1511	return (0);
1512}
1513
1514/*
1515 * Data movement routine for the CTL ioctl frontend port.
1516 */
1517static int
1518ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1519{
1520	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1521	struct ctl_sg_entry ext_entry, kern_entry;
1522	int ext_sglen, ext_sg_entries, kern_sg_entries;
1523	int ext_sg_start, ext_offset;
1524	int len_to_copy, len_copied;
1525	int kern_watermark, ext_watermark;
1526	int ext_sglist_malloced;
1527	int i, j;
1528
1529	ext_sglist_malloced = 0;
1530	ext_sg_start = 0;
1531	ext_offset = 0;
1532
1533	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1534
1535	/*
1536	 * If this flag is set, fake the data transfer.
1537	 */
1538	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1539		ctsio->ext_data_filled = ctsio->ext_data_len;
1540		goto bailout;
1541	}
1542
1543	/*
1544	 * To simplify things here, if we have a single buffer, stick it in
1545	 * a S/G entry and just make it a single entry S/G list.
1546	 */
1547	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1548		int len_seen;
1549
1550		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1551
1552		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1553							   M_WAITOK);
1554		ext_sglist_malloced = 1;
1555		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1556				   ext_sglen) != 0) {
1557			ctl_set_internal_failure(ctsio,
1558						 /*sks_valid*/ 0,
1559						 /*retry_count*/ 0);
1560			goto bailout;
1561		}
1562		ext_sg_entries = ctsio->ext_sg_entries;
1563		len_seen = 0;
1564		for (i = 0; i < ext_sg_entries; i++) {
1565			if ((len_seen + ext_sglist[i].len) >=
1566			     ctsio->ext_data_filled) {
1567				ext_sg_start = i;
1568				ext_offset = ctsio->ext_data_filled - len_seen;
1569				break;
1570			}
1571			len_seen += ext_sglist[i].len;
1572		}
1573	} else {
1574		ext_sglist = &ext_entry;
1575		ext_sglist->addr = ctsio->ext_data_ptr;
1576		ext_sglist->len = ctsio->ext_data_len;
1577		ext_sg_entries = 1;
1578		ext_sg_start = 0;
1579		ext_offset = ctsio->ext_data_filled;
1580	}
1581
1582	if (ctsio->kern_sg_entries > 0) {
1583		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1584		kern_sg_entries = ctsio->kern_sg_entries;
1585	} else {
1586		kern_sglist = &kern_entry;
1587		kern_sglist->addr = ctsio->kern_data_ptr;
1588		kern_sglist->len = ctsio->kern_data_len;
1589		kern_sg_entries = 1;
1590	}
1591
1592
1593	kern_watermark = 0;
1594	ext_watermark = ext_offset;
1595	len_copied = 0;
1596	for (i = ext_sg_start, j = 0;
1597	     i < ext_sg_entries && j < kern_sg_entries;) {
1598		uint8_t *ext_ptr, *kern_ptr;
1599
1600		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1601				      kern_sglist[j].len - kern_watermark);
1602
1603		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1604		ext_ptr = ext_ptr + ext_watermark;
1605		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1606			/*
1607			 * XXX KDM fix this!
1608			 */
1609			panic("need to implement bus address support");
1610#if 0
1611			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1612#endif
1613		} else
1614			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1615		kern_ptr = kern_ptr + kern_watermark;
1616
1617		kern_watermark += len_to_copy;
1618		ext_watermark += len_to_copy;
1619
1620		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1621		     CTL_FLAG_DATA_IN) {
1622			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1623					 "bytes to user\n", len_to_copy));
1624			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1625					 "to %p\n", kern_ptr, ext_ptr));
1626			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1627				ctl_set_internal_failure(ctsio,
1628							 /*sks_valid*/ 0,
1629							 /*retry_count*/ 0);
1630				goto bailout;
1631			}
1632		} else {
1633			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1634					 "bytes from user\n", len_to_copy));
1635			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1636					 "to %p\n", ext_ptr, kern_ptr));
1637			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1638				ctl_set_internal_failure(ctsio,
1639							 /*sks_valid*/ 0,
1640							 /*retry_count*/0);
1641				goto bailout;
1642			}
1643		}
1644
1645		len_copied += len_to_copy;
1646
1647		if (ext_sglist[i].len == ext_watermark) {
1648			i++;
1649			ext_watermark = 0;
1650		}
1651
1652		if (kern_sglist[j].len == kern_watermark) {
1653			j++;
1654			kern_watermark = 0;
1655		}
1656	}
1657
1658	ctsio->ext_data_filled += len_copied;
1659
1660	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1661			 "kern_sg_entries: %d\n", ext_sg_entries,
1662			 kern_sg_entries));
1663	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1664			 "kern_data_len = %d\n", ctsio->ext_data_len,
1665			 ctsio->kern_data_len));
1666
1667
1668	/* XXX KDM set residual?? */
1669bailout:
1670
1671	if (ext_sglist_malloced != 0)
1672		free(ext_sglist, M_CTL);
1673
1674	return (CTL_RETVAL_COMPLETE);
1675}
1676
1677/*
1678 * Serialize a command that went down the "wrong" side, and so was sent to
1679 * this controller for execution.  The logic is a little different than the
1680 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1681 * sent back to the other side, but in the success case, we execute the
1682 * command on this side (XFER mode) or tell the other side to execute it
1683 * (SER_ONLY mode).
1684 */
1685static int
1686ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1687{
1688	struct ctl_softc *ctl_softc;
1689	union ctl_ha_msg msg_info;
1690	struct ctl_lun *lun;
1691	int retval = 0;
1692	uint32_t targ_lun;
1693
1694	ctl_softc = control_softc;
1695	if (have_lock == 0)
1696		mtx_lock(&ctl_softc->ctl_lock);
1697
1698	targ_lun = ctsio->io_hdr.nexus.targ_lun;
1699	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
1700		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
1701	lun = ctl_softc->ctl_luns[targ_lun];
1702	if (lun==NULL)
1703	{
1704		/*
1705		 * Why isn't LUN defined? The other side wouldn't
1706		 * send a cmd if the LUN is undefined.
1707		 */
1708		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1709
1710		/* "Logical unit not supported" */
1711		ctl_set_sense_data(&msg_info.scsi.sense_data,
1712				   lun,
1713				   /*sense_format*/SSD_TYPE_NONE,
1714				   /*current_error*/ 1,
1715				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1716				   /*asc*/ 0x25,
1717				   /*ascq*/ 0x00,
1718				   SSD_ELEM_NONE);
1719
1720		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1721		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1722		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1723		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1724		msg_info.hdr.serializing_sc = NULL;
1725		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1726	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1727				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1728		}
1729		if (have_lock == 0)
1730			mtx_unlock(&ctl_softc->ctl_lock);
1731		return(1);
1732
1733	}
1734
1735    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1736
1737	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1738		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1739		 ooa_links))) {
1740	case CTL_ACTION_BLOCK:
1741		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1742		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1743				  blocked_links);
1744		break;
1745	case CTL_ACTION_PASS:
1746	case CTL_ACTION_SKIP:
1747		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1748			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1749			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1750					   &ctsio->io_hdr, links);
1751		} else {
1752
1753			/* send msg back to other side */
1754			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1755			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1756			msg_info.hdr.msg_type = CTL_MSG_R2R;
1757#if 0
1758			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1759#endif
1760		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1761			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1762			}
1763		}
1764		break;
1765	case CTL_ACTION_OVERLAP:
1766		/* OVERLAPPED COMMANDS ATTEMPTED */
1767		ctl_set_sense_data(&msg_info.scsi.sense_data,
1768				   lun,
1769				   /*sense_format*/SSD_TYPE_NONE,
1770				   /*current_error*/ 1,
1771				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1772				   /*asc*/ 0x4E,
1773				   /*ascq*/ 0x00,
1774				   SSD_ELEM_NONE);
1775
1776		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1777		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1778		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1779		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1780		msg_info.hdr.serializing_sc = NULL;
1781		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1782#if 0
1783		printf("BAD JUJU:Major Bummer Overlap\n");
1784#endif
1785		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1786		retval = 1;
1787		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1788		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1789		}
1790		break;
1791	case CTL_ACTION_OVERLAP_TAG:
1792		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1793		ctl_set_sense_data(&msg_info.scsi.sense_data,
1794				   lun,
1795				   /*sense_format*/SSD_TYPE_NONE,
1796				   /*current_error*/ 1,
1797				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1798				   /*asc*/ 0x4D,
1799				   /*ascq*/ ctsio->tag_num & 0xff,
1800				   SSD_ELEM_NONE);
1801
1802		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1803		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1804		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1805		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1806		msg_info.hdr.serializing_sc = NULL;
1807		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1808#if 0
1809		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1810#endif
1811		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1812		retval = 1;
1813		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1814		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1815		}
1816		break;
1817	case CTL_ACTION_ERROR:
1818	default:
1819		/* "Internal target failure" */
1820		ctl_set_sense_data(&msg_info.scsi.sense_data,
1821				   lun,
1822				   /*sense_format*/SSD_TYPE_NONE,
1823				   /*current_error*/ 1,
1824				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1825				   /*asc*/ 0x44,
1826				   /*ascq*/ 0x00,
1827				   SSD_ELEM_NONE);
1828
1829		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1830		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1831		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1832		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1833		msg_info.hdr.serializing_sc = NULL;
1834		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1835#if 0
1836		printf("BAD JUJU:Major Bummer HW Error\n");
1837#endif
1838		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1839		retval = 1;
1840		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1841		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1842		}
1843		break;
1844	}
1845	if (have_lock == 0)
1846		mtx_unlock(&ctl_softc->ctl_lock);
1847	return (retval);
1848}
1849
1850static int
1851ctl_ioctl_submit_wait(union ctl_io *io)
1852{
1853	struct ctl_fe_ioctl_params params;
1854	ctl_fe_ioctl_state last_state;
1855	int done, retval;
1856
1857	retval = 0;
1858
1859	bzero(&params, sizeof(params));
1860
1861	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1862	cv_init(&params.sem, "ctlioccv");
1863	params.state = CTL_IOCTL_INPROG;
1864	last_state = params.state;
1865
1866	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1867
1868	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1869
1870	/* This shouldn't happen */
1871	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1872		return (retval);
1873
1874	done = 0;
1875
1876	do {
1877		mtx_lock(&params.ioctl_mtx);
1878		/*
1879		 * Check the state here, and don't sleep if the state has
1880		 * already changed (i.e. wakeup has already occured, but we
1881		 * weren't waiting yet).
1882		 */
1883		if (params.state == last_state) {
1884			/* XXX KDM cv_wait_sig instead? */
1885			cv_wait(&params.sem, &params.ioctl_mtx);
1886		}
1887		last_state = params.state;
1888
1889		switch (params.state) {
1890		case CTL_IOCTL_INPROG:
1891			/* Why did we wake up? */
1892			/* XXX KDM error here? */
1893			mtx_unlock(&params.ioctl_mtx);
1894			break;
1895		case CTL_IOCTL_DATAMOVE:
1896			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1897
1898			/*
1899			 * change last_state back to INPROG to avoid
1900			 * deadlock on subsequent data moves.
1901			 */
1902			params.state = last_state = CTL_IOCTL_INPROG;
1903
1904			mtx_unlock(&params.ioctl_mtx);
1905			ctl_ioctl_do_datamove(&io->scsiio);
1906			/*
1907			 * Note that in some cases, most notably writes,
1908			 * this will queue the I/O and call us back later.
1909			 * In other cases, generally reads, this routine
1910			 * will immediately call back and wake us up,
1911			 * probably using our own context.
1912			 */
1913			io->scsiio.be_move_done(io);
1914			break;
1915		case CTL_IOCTL_DONE:
1916			mtx_unlock(&params.ioctl_mtx);
1917			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1918			done = 1;
1919			break;
1920		default:
1921			mtx_unlock(&params.ioctl_mtx);
1922			/* XXX KDM error here? */
1923			break;
1924		}
1925	} while (done == 0);
1926
1927	mtx_destroy(&params.ioctl_mtx);
1928	cv_destroy(&params.sem);
1929
1930	return (CTL_RETVAL_COMPLETE);
1931}
1932
1933static void
1934ctl_ioctl_datamove(union ctl_io *io)
1935{
1936	struct ctl_fe_ioctl_params *params;
1937
1938	params = (struct ctl_fe_ioctl_params *)
1939		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1940
1941	mtx_lock(&params->ioctl_mtx);
1942	params->state = CTL_IOCTL_DATAMOVE;
1943	cv_broadcast(&params->sem);
1944	mtx_unlock(&params->ioctl_mtx);
1945}
1946
1947static void
1948ctl_ioctl_done(union ctl_io *io)
1949{
1950	struct ctl_fe_ioctl_params *params;
1951
1952	params = (struct ctl_fe_ioctl_params *)
1953		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1954
1955	mtx_lock(&params->ioctl_mtx);
1956	params->state = CTL_IOCTL_DONE;
1957	cv_broadcast(&params->sem);
1958	mtx_unlock(&params->ioctl_mtx);
1959}
1960
1961static void
1962ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1963{
1964	struct ctl_fe_ioctl_startstop_info *sd_info;
1965
1966	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1967
1968	sd_info->hs_info.status = metatask->status;
1969	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1970	sd_info->hs_info.luns_complete =
1971		metatask->taskinfo.startstop.luns_complete;
1972	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1973
1974	cv_broadcast(&sd_info->sem);
1975}
1976
1977static void
1978ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1979{
1980	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1981
1982	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1983
1984	mtx_lock(fe_bbr_info->lock);
1985	fe_bbr_info->bbr_info->status = metatask->status;
1986	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1987	fe_bbr_info->wakeup_done = 1;
1988	mtx_unlock(fe_bbr_info->lock);
1989
1990	cv_broadcast(&fe_bbr_info->sem);
1991}
1992
1993/*
1994 * Returns 0 for success, errno for failure.
1995 */
1996static int
1997ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1998		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1999{
2000	union ctl_io *io;
2001	int retval;
2002
2003	retval = 0;
2004
2005	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
2006
2007	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2008	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2009	     ooa_links)) {
2010		struct ctl_ooa_entry *entry;
2011
2012		/*
2013		 * If we've got more than we can fit, just count the
2014		 * remaining entries.
2015		 */
2016		if (*cur_fill_num >= ooa_hdr->alloc_num)
2017			continue;
2018
2019		entry = &kern_entries[*cur_fill_num];
2020
2021		entry->tag_num = io->scsiio.tag_num;
2022		entry->lun_num = lun->lun;
2023#ifdef CTL_TIME_IO
2024		entry->start_bt = io->io_hdr.start_bt;
2025#endif
2026		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2027		entry->cdb_len = io->scsiio.cdb_len;
2028		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2029			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2030
2031		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2032			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2033
2034		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2035			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2036
2037		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2038			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2039
2040		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2041			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2042	}
2043
2044	return (retval);
2045}
2046
2047static void *
2048ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2049		 size_t error_str_len)
2050{
2051	void *kptr;
2052
2053	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2054
2055	if (copyin(user_addr, kptr, len) != 0) {
2056		snprintf(error_str, error_str_len, "Error copying %d bytes "
2057			 "from user address %p to kernel address %p", len,
2058			 user_addr, kptr);
2059		free(kptr, M_CTL);
2060		return (NULL);
2061	}
2062
2063	return (kptr);
2064}
2065
2066static void
2067ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2068{
2069	int i;
2070
2071	if (be_args == NULL)
2072		return;
2073
2074	for (i = 0; i < num_be_args; i++) {
2075		free(be_args[i].kname, M_CTL);
2076		free(be_args[i].kvalue, M_CTL);
2077	}
2078
2079	free(be_args, M_CTL);
2080}
2081
2082static struct ctl_be_arg *
2083ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2084		char *error_str, size_t error_str_len)
2085{
2086	struct ctl_be_arg *args;
2087	int i;
2088
2089	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2090				error_str, error_str_len);
2091
2092	if (args == NULL)
2093		goto bailout;
2094
2095	for (i = 0; i < num_be_args; i++) {
2096		args[i].kname = NULL;
2097		args[i].kvalue = NULL;
2098	}
2099
2100	for (i = 0; i < num_be_args; i++) {
2101		uint8_t *tmpptr;
2102
2103		args[i].kname = ctl_copyin_alloc(args[i].name,
2104			args[i].namelen, error_str, error_str_len);
2105		if (args[i].kname == NULL)
2106			goto bailout;
2107
2108		if (args[i].kname[args[i].namelen - 1] != '\0') {
2109			snprintf(error_str, error_str_len, "Argument %d "
2110				 "name is not NUL-terminated", i);
2111			goto bailout;
2112		}
2113
2114		args[i].kvalue = NULL;
2115
2116		tmpptr = ctl_copyin_alloc(args[i].value,
2117			args[i].vallen, error_str, error_str_len);
2118		if (tmpptr == NULL)
2119			goto bailout;
2120
2121		args[i].kvalue = tmpptr;
2122
2123		if ((args[i].flags & CTL_BEARG_ASCII)
2124		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2125			snprintf(error_str, error_str_len, "Argument %d "
2126				 "value is not NUL-terminated", i);
2127			goto bailout;
2128		}
2129	}
2130
2131	return (args);
2132bailout:
2133
2134	ctl_free_args(num_be_args, args);
2135
2136	return (NULL);
2137}
2138
2139/*
2140 * Escape characters that are illegal or not recommended in XML.
2141 */
2142int
2143ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2144{
2145	int retval;
2146
2147	retval = 0;
2148
2149	for (; *str; str++) {
2150		switch (*str) {
2151		case '&':
2152			retval = sbuf_printf(sb, "&amp;");
2153			break;
2154		case '>':
2155			retval = sbuf_printf(sb, "&gt;");
2156			break;
2157		case '<':
2158			retval = sbuf_printf(sb, "&lt;");
2159			break;
2160		default:
2161			retval = sbuf_putc(sb, *str);
2162			break;
2163		}
2164
2165		if (retval != 0)
2166			break;
2167
2168	}
2169
2170	return (retval);
2171}
2172
2173static int
2174ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2175	  struct thread *td)
2176{
2177	struct ctl_softc *softc;
2178	int retval;
2179
2180	softc = control_softc;
2181
2182	retval = 0;
2183
2184	switch (cmd) {
2185	case CTL_IO: {
2186		union ctl_io *io;
2187		void *pool_tmp;
2188
2189		/*
2190		 * If we haven't been "enabled", don't allow any SCSI I/O
2191		 * to this FETD.
2192		 */
2193		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2194			retval = -EPERM;
2195			break;
2196		}
2197
2198		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2199		if (io == NULL) {
2200			printf("ctl_ioctl: can't allocate ctl_io!\n");
2201			retval = -ENOSPC;
2202			break;
2203		}
2204
2205		/*
2206		 * Need to save the pool reference so it doesn't get
2207		 * spammed by the user's ctl_io.
2208		 */
2209		pool_tmp = io->io_hdr.pool;
2210
2211		memcpy(io, (void *)addr, sizeof(*io));
2212
2213		io->io_hdr.pool = pool_tmp;
2214		/*
2215		 * No status yet, so make sure the status is set properly.
2216		 */
2217		io->io_hdr.status = CTL_STATUS_NONE;
2218
2219		/*
2220		 * The user sets the initiator ID, target and LUN IDs.
2221		 */
2222		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2223		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2224		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2225		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2226			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2227
2228		retval = ctl_ioctl_submit_wait(io);
2229
2230		if (retval != 0) {
2231			ctl_free_io(io);
2232			break;
2233		}
2234
2235		memcpy((void *)addr, io, sizeof(*io));
2236
2237		/* return this to our pool */
2238		ctl_free_io(io);
2239
2240		break;
2241	}
2242	case CTL_ENABLE_PORT:
2243	case CTL_DISABLE_PORT:
2244	case CTL_SET_PORT_WWNS: {
2245		struct ctl_frontend *fe;
2246		struct ctl_port_entry *entry;
2247
2248		entry = (struct ctl_port_entry *)addr;
2249
2250		mtx_lock(&softc->ctl_lock);
2251		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2252			int action, done;
2253
2254			action = 0;
2255			done = 0;
2256
2257			if ((entry->port_type == CTL_PORT_NONE)
2258			 && (entry->targ_port == fe->targ_port)) {
2259				/*
2260				 * If the user only wants to enable or
2261				 * disable or set WWNs on a specific port,
2262				 * do the operation and we're done.
2263				 */
2264				action = 1;
2265				done = 1;
2266			} else if (entry->port_type & fe->port_type) {
2267				/*
2268				 * Compare the user's type mask with the
2269				 * particular frontend type to see if we
2270				 * have a match.
2271				 */
2272				action = 1;
2273				done = 0;
2274
2275				/*
2276				 * Make sure the user isn't trying to set
2277				 * WWNs on multiple ports at the same time.
2278				 */
2279				if (cmd == CTL_SET_PORT_WWNS) {
2280					printf("%s: Can't set WWNs on "
2281					       "multiple ports\n", __func__);
2282					retval = EINVAL;
2283					break;
2284				}
2285			}
2286			if (action != 0) {
2287				/*
2288				 * XXX KDM we have to drop the lock here,
2289				 * because the online/offline operations
2290				 * can potentially block.  We need to
2291				 * reference count the frontends so they
2292				 * can't go away,
2293				 */
2294				mtx_unlock(&softc->ctl_lock);
2295
2296				if (cmd == CTL_ENABLE_PORT) {
2297					struct ctl_lun *lun;
2298
2299					STAILQ_FOREACH(lun, &softc->lun_list,
2300						       links) {
2301						fe->lun_enable(fe->targ_lun_arg,
2302						    lun->target,
2303						    lun->lun);
2304					}
2305
2306					ctl_frontend_online(fe);
2307				} else if (cmd == CTL_DISABLE_PORT) {
2308					struct ctl_lun *lun;
2309
2310					ctl_frontend_offline(fe);
2311
2312					STAILQ_FOREACH(lun, &softc->lun_list,
2313						       links) {
2314						fe->lun_disable(
2315						    fe->targ_lun_arg,
2316						    lun->target,
2317						    lun->lun);
2318					}
2319				}
2320
2321				mtx_lock(&softc->ctl_lock);
2322
2323				if (cmd == CTL_SET_PORT_WWNS)
2324					ctl_frontend_set_wwns(fe,
2325					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2326					    1 : 0, entry->wwnn,
2327					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2328					    1 : 0, entry->wwpn);
2329			}
2330			if (done != 0)
2331				break;
2332		}
2333		mtx_unlock(&softc->ctl_lock);
2334		break;
2335	}
2336	case CTL_GET_PORT_LIST: {
2337		struct ctl_frontend *fe;
2338		struct ctl_port_list *list;
2339		int i;
2340
2341		list = (struct ctl_port_list *)addr;
2342
2343		if (list->alloc_len != (list->alloc_num *
2344		    sizeof(struct ctl_port_entry))) {
2345			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2346			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2347			       "%zu\n", __func__, list->alloc_len,
2348			       list->alloc_num, sizeof(struct ctl_port_entry));
2349			retval = EINVAL;
2350			break;
2351		}
2352		list->fill_len = 0;
2353		list->fill_num = 0;
2354		list->dropped_num = 0;
2355		i = 0;
2356		mtx_lock(&softc->ctl_lock);
2357		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2358			struct ctl_port_entry entry, *list_entry;
2359
2360			if (list->fill_num >= list->alloc_num) {
2361				list->dropped_num++;
2362				continue;
2363			}
2364
2365			entry.port_type = fe->port_type;
2366			strlcpy(entry.port_name, fe->port_name,
2367				sizeof(entry.port_name));
2368			entry.targ_port = fe->targ_port;
2369			entry.physical_port = fe->physical_port;
2370			entry.virtual_port = fe->virtual_port;
2371			entry.wwnn = fe->wwnn;
2372			entry.wwpn = fe->wwpn;
2373			if (fe->status & CTL_PORT_STATUS_ONLINE)
2374				entry.online = 1;
2375			else
2376				entry.online = 0;
2377
2378			list_entry = &list->entries[i];
2379
2380			retval = copyout(&entry, list_entry, sizeof(entry));
2381			if (retval != 0) {
2382				printf("%s: CTL_GET_PORT_LIST: copyout "
2383				       "returned %d\n", __func__, retval);
2384				break;
2385			}
2386			i++;
2387			list->fill_num++;
2388			list->fill_len += sizeof(entry);
2389		}
2390		mtx_unlock(&softc->ctl_lock);
2391
2392		/*
2393		 * If this is non-zero, we had a copyout fault, so there's
2394		 * probably no point in attempting to set the status inside
2395		 * the structure.
2396		 */
2397		if (retval != 0)
2398			break;
2399
2400		if (list->dropped_num > 0)
2401			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2402		else
2403			list->status = CTL_PORT_LIST_OK;
2404		break;
2405	}
2406	case CTL_DUMP_OOA: {
2407		struct ctl_lun *lun;
2408		union ctl_io *io;
2409		char printbuf[128];
2410		struct sbuf sb;
2411
2412		mtx_lock(&softc->ctl_lock);
2413		printf("Dumping OOA queues:\n");
2414		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2415			for (io = (union ctl_io *)TAILQ_FIRST(
2416			     &lun->ooa_queue); io != NULL;
2417			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2418			     ooa_links)) {
2419				sbuf_new(&sb, printbuf, sizeof(printbuf),
2420					 SBUF_FIXEDLEN);
2421				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2422					    (intmax_t)lun->lun,
2423					    io->scsiio.tag_num,
2424					    (io->io_hdr.flags &
2425					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2426					    (io->io_hdr.flags &
2427					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2428					    (io->io_hdr.flags &
2429					    CTL_FLAG_ABORT) ? " ABORT" : "",
2430			                    (io->io_hdr.flags &
2431		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2432				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2433				sbuf_finish(&sb);
2434				printf("%s\n", sbuf_data(&sb));
2435			}
2436		}
2437		printf("OOA queues dump done\n");
2438		mtx_unlock(&softc->ctl_lock);
2439		break;
2440	}
2441	case CTL_GET_OOA: {
2442		struct ctl_lun *lun;
2443		struct ctl_ooa *ooa_hdr;
2444		struct ctl_ooa_entry *entries;
2445		uint32_t cur_fill_num;
2446
2447		ooa_hdr = (struct ctl_ooa *)addr;
2448
2449		if ((ooa_hdr->alloc_len == 0)
2450		 || (ooa_hdr->alloc_num == 0)) {
2451			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2452			       "must be non-zero\n", __func__,
2453			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2454			retval = EINVAL;
2455			break;
2456		}
2457
2458		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2459		    sizeof(struct ctl_ooa_entry))) {
2460			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2461			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2462			       __func__, ooa_hdr->alloc_len,
2463			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2464			retval = EINVAL;
2465			break;
2466		}
2467
2468		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2469		if (entries == NULL) {
2470			printf("%s: could not allocate %d bytes for OOA "
2471			       "dump\n", __func__, ooa_hdr->alloc_len);
2472			retval = ENOMEM;
2473			break;
2474		}
2475
2476		mtx_lock(&softc->ctl_lock);
2477		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2478		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2479		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2480			mtx_unlock(&softc->ctl_lock);
2481			free(entries, M_CTL);
2482			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2483			       __func__, (uintmax_t)ooa_hdr->lun_num);
2484			retval = EINVAL;
2485			break;
2486		}
2487
2488		cur_fill_num = 0;
2489
2490		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2491			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2492				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2493					ooa_hdr, entries);
2494				if (retval != 0)
2495					break;
2496			}
2497			if (retval != 0) {
2498				mtx_unlock(&softc->ctl_lock);
2499				free(entries, M_CTL);
2500				break;
2501			}
2502		} else {
2503			lun = softc->ctl_luns[ooa_hdr->lun_num];
2504
2505			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2506						    entries);
2507		}
2508		mtx_unlock(&softc->ctl_lock);
2509
2510		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2511		ooa_hdr->fill_len = ooa_hdr->fill_num *
2512			sizeof(struct ctl_ooa_entry);
2513		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2514		if (retval != 0) {
2515			printf("%s: error copying out %d bytes for OOA dump\n",
2516			       __func__, ooa_hdr->fill_len);
2517		}
2518
2519		getbintime(&ooa_hdr->cur_bt);
2520
2521		if (cur_fill_num > ooa_hdr->alloc_num) {
2522			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2523			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2524		} else {
2525			ooa_hdr->dropped_num = 0;
2526			ooa_hdr->status = CTL_OOA_OK;
2527		}
2528
2529		free(entries, M_CTL);
2530		break;
2531	}
2532	case CTL_CHECK_OOA: {
2533		union ctl_io *io;
2534		struct ctl_lun *lun;
2535		struct ctl_ooa_info *ooa_info;
2536
2537
2538		ooa_info = (struct ctl_ooa_info *)addr;
2539
2540		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2541			ooa_info->status = CTL_OOA_INVALID_LUN;
2542			break;
2543		}
2544		mtx_lock(&softc->ctl_lock);
2545		lun = softc->ctl_luns[ooa_info->lun_id];
2546		if (lun == NULL) {
2547			mtx_unlock(&softc->ctl_lock);
2548			ooa_info->status = CTL_OOA_INVALID_LUN;
2549			break;
2550		}
2551
2552		ooa_info->num_entries = 0;
2553		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2554		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2555		     &io->io_hdr, ooa_links)) {
2556			ooa_info->num_entries++;
2557		}
2558
2559		mtx_unlock(&softc->ctl_lock);
2560		ooa_info->status = CTL_OOA_SUCCESS;
2561
2562		break;
2563	}
2564	case CTL_HARD_START:
2565	case CTL_HARD_STOP: {
2566		struct ctl_fe_ioctl_startstop_info ss_info;
2567		struct cfi_metatask *metatask;
2568		struct mtx hs_mtx;
2569
2570		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2571
2572		cv_init(&ss_info.sem, "hard start/stop cv" );
2573
2574		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2575		if (metatask == NULL) {
2576			retval = ENOMEM;
2577			mtx_destroy(&hs_mtx);
2578			break;
2579		}
2580
2581		if (cmd == CTL_HARD_START)
2582			metatask->tasktype = CFI_TASK_STARTUP;
2583		else
2584			metatask->tasktype = CFI_TASK_SHUTDOWN;
2585
2586		metatask->callback = ctl_ioctl_hard_startstop_callback;
2587		metatask->callback_arg = &ss_info;
2588
2589		cfi_action(metatask);
2590
2591		/* Wait for the callback */
2592		mtx_lock(&hs_mtx);
2593		cv_wait_sig(&ss_info.sem, &hs_mtx);
2594		mtx_unlock(&hs_mtx);
2595
2596		/*
2597		 * All information has been copied from the metatask by the
2598		 * time cv_broadcast() is called, so we free the metatask here.
2599		 */
2600		cfi_free_metatask(metatask);
2601
2602		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2603
2604		mtx_destroy(&hs_mtx);
2605		break;
2606	}
2607	case CTL_BBRREAD: {
2608		struct ctl_bbrread_info *bbr_info;
2609		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2610		struct mtx bbr_mtx;
2611		struct cfi_metatask *metatask;
2612
2613		bbr_info = (struct ctl_bbrread_info *)addr;
2614
2615		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2616
2617		bzero(&bbr_mtx, sizeof(bbr_mtx));
2618		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2619
2620		fe_bbr_info.bbr_info = bbr_info;
2621		fe_bbr_info.lock = &bbr_mtx;
2622
2623		cv_init(&fe_bbr_info.sem, "BBR read cv");
2624		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2625
2626		if (metatask == NULL) {
2627			mtx_destroy(&bbr_mtx);
2628			cv_destroy(&fe_bbr_info.sem);
2629			retval = ENOMEM;
2630			break;
2631		}
2632		metatask->tasktype = CFI_TASK_BBRREAD;
2633		metatask->callback = ctl_ioctl_bbrread_callback;
2634		metatask->callback_arg = &fe_bbr_info;
2635		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2636		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2637		metatask->taskinfo.bbrread.len = bbr_info->len;
2638
2639		cfi_action(metatask);
2640
2641		mtx_lock(&bbr_mtx);
2642		while (fe_bbr_info.wakeup_done == 0)
2643			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2644		mtx_unlock(&bbr_mtx);
2645
2646		bbr_info->status = metatask->status;
2647		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2648		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2649		memcpy(&bbr_info->sense_data,
2650		       &metatask->taskinfo.bbrread.sense_data,
2651		       ctl_min(sizeof(bbr_info->sense_data),
2652			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2653
2654		cfi_free_metatask(metatask);
2655
2656		mtx_destroy(&bbr_mtx);
2657		cv_destroy(&fe_bbr_info.sem);
2658
2659		break;
2660	}
2661	case CTL_DELAY_IO: {
2662		struct ctl_io_delay_info *delay_info;
2663#ifdef CTL_IO_DELAY
2664		struct ctl_lun *lun;
2665#endif /* CTL_IO_DELAY */
2666
2667		delay_info = (struct ctl_io_delay_info *)addr;
2668
2669#ifdef CTL_IO_DELAY
2670		mtx_lock(&softc->ctl_lock);
2671
2672		if ((delay_info->lun_id > CTL_MAX_LUNS)
2673		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2674			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2675		} else {
2676			lun = softc->ctl_luns[delay_info->lun_id];
2677
2678			delay_info->status = CTL_DELAY_STATUS_OK;
2679
2680			switch (delay_info->delay_type) {
2681			case CTL_DELAY_TYPE_CONT:
2682				break;
2683			case CTL_DELAY_TYPE_ONESHOT:
2684				break;
2685			default:
2686				delay_info->status =
2687					CTL_DELAY_STATUS_INVALID_TYPE;
2688				break;
2689			}
2690
2691			switch (delay_info->delay_loc) {
2692			case CTL_DELAY_LOC_DATAMOVE:
2693				lun->delay_info.datamove_type =
2694					delay_info->delay_type;
2695				lun->delay_info.datamove_delay =
2696					delay_info->delay_secs;
2697				break;
2698			case CTL_DELAY_LOC_DONE:
2699				lun->delay_info.done_type =
2700					delay_info->delay_type;
2701				lun->delay_info.done_delay =
2702					delay_info->delay_secs;
2703				break;
2704			default:
2705				delay_info->status =
2706					CTL_DELAY_STATUS_INVALID_LOC;
2707				break;
2708			}
2709		}
2710
2711		mtx_unlock(&softc->ctl_lock);
2712#else
2713		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2714#endif /* CTL_IO_DELAY */
2715		break;
2716	}
2717	case CTL_REALSYNC_SET: {
2718		int *syncstate;
2719
2720		syncstate = (int *)addr;
2721
2722		mtx_lock(&softc->ctl_lock);
2723		switch (*syncstate) {
2724		case 0:
2725			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2726			break;
2727		case 1:
2728			softc->flags |= CTL_FLAG_REAL_SYNC;
2729			break;
2730		default:
2731			retval = -EINVAL;
2732			break;
2733		}
2734		mtx_unlock(&softc->ctl_lock);
2735		break;
2736	}
2737	case CTL_REALSYNC_GET: {
2738		int *syncstate;
2739
2740		syncstate = (int*)addr;
2741
2742		mtx_lock(&softc->ctl_lock);
2743		if (softc->flags & CTL_FLAG_REAL_SYNC)
2744			*syncstate = 1;
2745		else
2746			*syncstate = 0;
2747		mtx_unlock(&softc->ctl_lock);
2748
2749		break;
2750	}
2751	case CTL_SETSYNC:
2752	case CTL_GETSYNC: {
2753		struct ctl_sync_info *sync_info;
2754		struct ctl_lun *lun;
2755
2756		sync_info = (struct ctl_sync_info *)addr;
2757
2758		mtx_lock(&softc->ctl_lock);
2759		lun = softc->ctl_luns[sync_info->lun_id];
2760		if (lun == NULL) {
2761			mtx_unlock(&softc->ctl_lock);
2762			sync_info->status = CTL_GS_SYNC_NO_LUN;
2763		}
2764		/*
2765		 * Get or set the sync interval.  We're not bounds checking
2766		 * in the set case, hopefully the user won't do something
2767		 * silly.
2768		 */
2769		if (cmd == CTL_GETSYNC)
2770			sync_info->sync_interval = lun->sync_interval;
2771		else
2772			lun->sync_interval = sync_info->sync_interval;
2773
2774		mtx_unlock(&softc->ctl_lock);
2775
2776		sync_info->status = CTL_GS_SYNC_OK;
2777
2778		break;
2779	}
2780	case CTL_GETSTATS: {
2781		struct ctl_stats *stats;
2782		struct ctl_lun *lun;
2783		int i;
2784
2785		stats = (struct ctl_stats *)addr;
2786
2787		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2788		     stats->alloc_len) {
2789			stats->status = CTL_SS_NEED_MORE_SPACE;
2790			stats->num_luns = softc->num_luns;
2791			break;
2792		}
2793		/*
2794		 * XXX KDM no locking here.  If the LUN list changes,
2795		 * things can blow up.
2796		 */
2797		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2798		     i++, lun = STAILQ_NEXT(lun, links)) {
2799			retval = copyout(&lun->stats, &stats->lun_stats[i],
2800					 sizeof(lun->stats));
2801			if (retval != 0)
2802				break;
2803		}
2804		stats->num_luns = softc->num_luns;
2805		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2806				 softc->num_luns;
2807		stats->status = CTL_SS_OK;
2808#ifdef CTL_TIME_IO
2809		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2810#else
2811		stats->flags = CTL_STATS_FLAG_NONE;
2812#endif
2813		getnanouptime(&stats->timestamp);
2814		break;
2815	}
2816	case CTL_ERROR_INJECT: {
2817		struct ctl_error_desc *err_desc, *new_err_desc;
2818		struct ctl_lun *lun;
2819
2820		err_desc = (struct ctl_error_desc *)addr;
2821
2822		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2823				      M_WAITOK | M_ZERO);
2824		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2825
2826		mtx_lock(&softc->ctl_lock);
2827		lun = softc->ctl_luns[err_desc->lun_id];
2828		if (lun == NULL) {
2829			mtx_unlock(&softc->ctl_lock);
2830			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2831			       __func__, (uintmax_t)err_desc->lun_id);
2832			retval = EINVAL;
2833			break;
2834		}
2835
2836		/*
2837		 * We could do some checking here to verify the validity
2838		 * of the request, but given the complexity of error
2839		 * injection requests, the checking logic would be fairly
2840		 * complex.
2841		 *
2842		 * For now, if the request is invalid, it just won't get
2843		 * executed and might get deleted.
2844		 */
2845		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2846
2847		/*
2848		 * XXX KDM check to make sure the serial number is unique,
2849		 * in case we somehow manage to wrap.  That shouldn't
2850		 * happen for a very long time, but it's the right thing to
2851		 * do.
2852		 */
2853		new_err_desc->serial = lun->error_serial;
2854		err_desc->serial = lun->error_serial;
2855		lun->error_serial++;
2856
2857		mtx_unlock(&softc->ctl_lock);
2858		break;
2859	}
2860	case CTL_ERROR_INJECT_DELETE: {
2861		struct ctl_error_desc *delete_desc, *desc, *desc2;
2862		struct ctl_lun *lun;
2863		int delete_done;
2864
2865		delete_desc = (struct ctl_error_desc *)addr;
2866		delete_done = 0;
2867
2868		mtx_lock(&softc->ctl_lock);
2869		lun = softc->ctl_luns[delete_desc->lun_id];
2870		if (lun == NULL) {
2871			mtx_unlock(&softc->ctl_lock);
2872			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2873			       __func__, (uintmax_t)delete_desc->lun_id);
2874			retval = EINVAL;
2875			break;
2876		}
2877		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2878			if (desc->serial != delete_desc->serial)
2879				continue;
2880
2881			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2882				      links);
2883			free(desc, M_CTL);
2884			delete_done = 1;
2885		}
2886		mtx_unlock(&softc->ctl_lock);
2887		if (delete_done == 0) {
2888			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2889			       "error serial %ju on LUN %u\n", __func__,
2890			       delete_desc->serial, delete_desc->lun_id);
2891			retval = EINVAL;
2892			break;
2893		}
2894		break;
2895	}
2896	case CTL_DUMP_STRUCTS: {
2897		int i, j, k;
2898		struct ctl_frontend *fe;
2899
2900		printf("CTL IID to WWPN map start:\n");
2901		for (i = 0; i < CTL_MAX_PORTS; i++) {
2902			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2903				if (softc->wwpn_iid[i][j].in_use == 0)
2904					continue;
2905
2906				printf("port %d iid %u WWPN %#jx\n",
2907				       softc->wwpn_iid[i][j].port,
2908				       softc->wwpn_iid[i][j].iid,
2909				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2910			}
2911		}
2912		printf("CTL IID to WWPN map end\n");
2913		printf("CTL Persistent Reservation information start:\n");
2914		for (i = 0; i < CTL_MAX_LUNS; i++) {
2915			struct ctl_lun *lun;
2916
2917			lun = softc->ctl_luns[i];
2918
2919			if ((lun == NULL)
2920			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2921				continue;
2922
2923			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2924				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2925					if (lun->per_res[j+k].registered == 0)
2926						continue;
2927					printf("LUN %d port %d iid %d key "
2928					       "%#jx\n", i, j, k,
2929					       (uintmax_t)scsi_8btou64(
2930					       lun->per_res[j+k].res_key.key));
2931				}
2932			}
2933		}
2934		printf("CTL Persistent Reservation information end\n");
2935		printf("CTL Frontends:\n");
2936		/*
2937		 * XXX KDM calling this without a lock.  We'd likely want
2938		 * to drop the lock before calling the frontend's dump
2939		 * routine anyway.
2940		 */
2941		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2942			printf("Frontend %s Type %u pport %d vport %d WWNN "
2943			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2944			       fe->physical_port, fe->virtual_port,
2945			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2946
2947			/*
2948			 * Frontends are not required to support the dump
2949			 * routine.
2950			 */
2951			if (fe->fe_dump == NULL)
2952				continue;
2953
2954			fe->fe_dump();
2955		}
2956		printf("CTL Frontend information end\n");
2957		break;
2958	}
2959	case CTL_LUN_REQ: {
2960		struct ctl_lun_req *lun_req;
2961		struct ctl_backend_driver *backend;
2962
2963		lun_req = (struct ctl_lun_req *)addr;
2964
2965		backend = ctl_backend_find(lun_req->backend);
2966		if (backend == NULL) {
2967			lun_req->status = CTL_LUN_ERROR;
2968			snprintf(lun_req->error_str,
2969				 sizeof(lun_req->error_str),
2970				 "Backend \"%s\" not found.",
2971				 lun_req->backend);
2972			break;
2973		}
2974		if (lun_req->num_be_args > 0) {
2975			lun_req->kern_be_args = ctl_copyin_args(
2976				lun_req->num_be_args,
2977				lun_req->be_args,
2978				lun_req->error_str,
2979				sizeof(lun_req->error_str));
2980			if (lun_req->kern_be_args == NULL) {
2981				lun_req->status = CTL_LUN_ERROR;
2982				break;
2983			}
2984		}
2985
2986		retval = backend->ioctl(dev, cmd, addr, flag, td);
2987
2988		if (lun_req->num_be_args > 0) {
2989			ctl_free_args(lun_req->num_be_args,
2990				      lun_req->kern_be_args);
2991		}
2992		break;
2993	}
2994	case CTL_LUN_LIST: {
2995		struct sbuf *sb;
2996		struct ctl_lun *lun;
2997		struct ctl_lun_list *list;
2998		struct ctl_be_lun_option *opt;
2999
3000		list = (struct ctl_lun_list *)addr;
3001
3002		/*
3003		 * Allocate a fixed length sbuf here, based on the length
3004		 * of the user's buffer.  We could allocate an auto-extending
3005		 * buffer, and then tell the user how much larger our
3006		 * amount of data is than his buffer, but that presents
3007		 * some problems:
3008		 *
3009		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3010		 *     we can't hold a lock while calling them with an
3011		 *     auto-extending buffer.
3012 		 *
3013		 * 2.  There is not currently a LUN reference counting
3014		 *     mechanism, outside of outstanding transactions on
3015		 *     the LUN's OOA queue.  So a LUN could go away on us
3016		 *     while we're getting the LUN number, backend-specific
3017		 *     information, etc.  Thus, given the way things
3018		 *     currently work, we need to hold the CTL lock while
3019		 *     grabbing LUN information.
3020		 *
3021		 * So, from the user's standpoint, the best thing to do is
3022		 * allocate what he thinks is a reasonable buffer length,
3023		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3024		 * double the buffer length and try again.  (And repeat
3025		 * that until he succeeds.)
3026		 */
3027		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3028		if (sb == NULL) {
3029			list->status = CTL_LUN_LIST_ERROR;
3030			snprintf(list->error_str, sizeof(list->error_str),
3031				 "Unable to allocate %d bytes for LUN list",
3032				 list->alloc_len);
3033			break;
3034		}
3035
3036		sbuf_printf(sb, "<ctllunlist>\n");
3037
3038		mtx_lock(&softc->ctl_lock);
3039
3040		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3041			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3042					     (uintmax_t)lun->lun);
3043
3044			/*
3045			 * Bail out as soon as we see that we've overfilled
3046			 * the buffer.
3047			 */
3048			if (retval != 0)
3049				break;
3050
3051			retval = sbuf_printf(sb, "<backend_type>%s"
3052					     "</backend_type>\n",
3053					     (lun->backend == NULL) ?  "none" :
3054					     lun->backend->name);
3055
3056			if (retval != 0)
3057				break;
3058
3059			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3060					     lun->be_lun->lun_type);
3061
3062			if (retval != 0)
3063				break;
3064
3065			if (lun->backend == NULL) {
3066				retval = sbuf_printf(sb, "</lun>\n");
3067				if (retval != 0)
3068					break;
3069				continue;
3070			}
3071
3072			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3073					     (lun->be_lun->maxlba > 0) ?
3074					     lun->be_lun->maxlba + 1 : 0);
3075
3076			if (retval != 0)
3077				break;
3078
3079			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3080					     lun->be_lun->blocksize);
3081
3082			if (retval != 0)
3083				break;
3084
3085			retval = sbuf_printf(sb, "<serial_number>");
3086
3087			if (retval != 0)
3088				break;
3089
3090			retval = ctl_sbuf_printf_esc(sb,
3091						     lun->be_lun->serial_num);
3092
3093			if (retval != 0)
3094				break;
3095
3096			retval = sbuf_printf(sb, "</serial_number>\n");
3097
3098			if (retval != 0)
3099				break;
3100
3101			retval = sbuf_printf(sb, "<device_id>");
3102
3103			if (retval != 0)
3104				break;
3105
3106			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3107
3108			if (retval != 0)
3109				break;
3110
3111			retval = sbuf_printf(sb, "</device_id>\n");
3112
3113			if (retval != 0)
3114				break;
3115
3116			if (lun->backend->lun_info != NULL) {
3117				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3118				if (retval != 0)
3119					break;
3120			}
3121			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3122				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3123				if (retval != 0)
3124					break;
3125			}
3126
3127			retval = sbuf_printf(sb, "</lun>\n");
3128
3129			if (retval != 0)
3130				break;
3131		}
3132		mtx_unlock(&softc->ctl_lock);
3133
3134		if ((retval != 0)
3135		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3136			retval = 0;
3137			sbuf_delete(sb);
3138			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3139			snprintf(list->error_str, sizeof(list->error_str),
3140				 "Out of space, %d bytes is too small",
3141				 list->alloc_len);
3142			break;
3143		}
3144
3145		sbuf_finish(sb);
3146
3147		retval = copyout(sbuf_data(sb), list->lun_xml,
3148				 sbuf_len(sb) + 1);
3149
3150		list->fill_len = sbuf_len(sb) + 1;
3151		list->status = CTL_LUN_LIST_OK;
3152		sbuf_delete(sb);
3153		break;
3154	}
3155	case CTL_ISCSI: {
3156		struct ctl_iscsi *ci;
3157		struct ctl_frontend *fe;
3158
3159		ci = (struct ctl_iscsi *)addr;
3160
3161		mtx_lock(&softc->ctl_lock);
3162		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3163			if (strcmp(fe->port_name, "iscsi") == 0)
3164				break;
3165		}
3166		mtx_unlock(&softc->ctl_lock);
3167
3168		if (fe == NULL) {
3169			ci->status = CTL_ISCSI_ERROR;
3170			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3171			break;
3172		}
3173
3174		retval = fe->ioctl(dev, cmd, addr, flag, td);
3175		break;
3176	}
3177	default: {
3178		/* XXX KDM should we fix this? */
3179#if 0
3180		struct ctl_backend_driver *backend;
3181		unsigned int type;
3182		int found;
3183
3184		found = 0;
3185
3186		/*
3187		 * We encode the backend type as the ioctl type for backend
3188		 * ioctls.  So parse it out here, and then search for a
3189		 * backend of this type.
3190		 */
3191		type = _IOC_TYPE(cmd);
3192
3193		STAILQ_FOREACH(backend, &softc->be_list, links) {
3194			if (backend->type == type) {
3195				found = 1;
3196				break;
3197			}
3198		}
3199		if (found == 0) {
3200			printf("ctl: unknown ioctl command %#lx or backend "
3201			       "%d\n", cmd, type);
3202			retval = -EINVAL;
3203			break;
3204		}
3205		retval = backend->ioctl(dev, cmd, addr, flag, td);
3206#endif
3207		retval = ENOTTY;
3208		break;
3209	}
3210	}
3211	return (retval);
3212}
3213
3214uint32_t
3215ctl_get_initindex(struct ctl_nexus *nexus)
3216{
3217	if (nexus->targ_port < CTL_MAX_PORTS)
3218		return (nexus->initid.id +
3219			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3220	else
3221		return (nexus->initid.id +
3222		       ((nexus->targ_port - CTL_MAX_PORTS) *
3223			CTL_MAX_INIT_PER_PORT));
3224}
3225
3226uint32_t
3227ctl_get_resindex(struct ctl_nexus *nexus)
3228{
3229	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3230}
3231
3232uint32_t
3233ctl_port_idx(int port_num)
3234{
3235	if (port_num < CTL_MAX_PORTS)
3236		return(port_num);
3237	else
3238		return(port_num - CTL_MAX_PORTS);
3239}
3240
3241/*
3242 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3243 * that are a power of 2.
3244 */
3245int
3246ctl_ffz(uint32_t *mask, uint32_t size)
3247{
3248	uint32_t num_chunks, num_pieces;
3249	int i, j;
3250
3251	num_chunks = (size >> 5);
3252	if (num_chunks == 0)
3253		num_chunks++;
3254	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3255
3256	for (i = 0; i < num_chunks; i++) {
3257		for (j = 0; j < num_pieces; j++) {
3258			if ((mask[i] & (1 << j)) == 0)
3259				return ((i << 5) + j);
3260		}
3261	}
3262
3263	return (-1);
3264}
3265
3266int
3267ctl_set_mask(uint32_t *mask, uint32_t bit)
3268{
3269	uint32_t chunk, piece;
3270
3271	chunk = bit >> 5;
3272	piece = bit % (sizeof(uint32_t) * 8);
3273
3274	if ((mask[chunk] & (1 << piece)) != 0)
3275		return (-1);
3276	else
3277		mask[chunk] |= (1 << piece);
3278
3279	return (0);
3280}
3281
3282int
3283ctl_clear_mask(uint32_t *mask, uint32_t bit)
3284{
3285	uint32_t chunk, piece;
3286
3287	chunk = bit >> 5;
3288	piece = bit % (sizeof(uint32_t) * 8);
3289
3290	if ((mask[chunk] & (1 << piece)) == 0)
3291		return (-1);
3292	else
3293		mask[chunk] &= ~(1 << piece);
3294
3295	return (0);
3296}
3297
3298int
3299ctl_is_set(uint32_t *mask, uint32_t bit)
3300{
3301	uint32_t chunk, piece;
3302
3303	chunk = bit >> 5;
3304	piece = bit % (sizeof(uint32_t) * 8);
3305
3306	if ((mask[chunk] & (1 << piece)) == 0)
3307		return (0);
3308	else
3309		return (1);
3310}
3311
3312#ifdef unused
3313/*
3314 * The bus, target and lun are optional, they can be filled in later.
3315 * can_wait is used to determine whether we can wait on the malloc or not.
3316 */
3317union ctl_io*
3318ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3319	      uint32_t targ_lun, int can_wait)
3320{
3321	union ctl_io *io;
3322
3323	if (can_wait)
3324		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3325	else
3326		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3327
3328	if (io != NULL) {
3329		io->io_hdr.io_type = io_type;
3330		io->io_hdr.targ_port = targ_port;
3331		/*
3332		 * XXX KDM this needs to change/go away.  We need to move
3333		 * to a preallocated pool of ctl_scsiio structures.
3334		 */
3335		io->io_hdr.nexus.targ_target.id = targ_target;
3336		io->io_hdr.nexus.targ_lun = targ_lun;
3337	}
3338
3339	return (io);
3340}
3341
3342void
3343ctl_kfree_io(union ctl_io *io)
3344{
3345	free(io, M_CTL);
3346}
3347#endif /* unused */
3348
3349/*
3350 * ctl_softc, pool_type, total_ctl_io are passed in.
3351 * npool is passed out.
3352 */
3353int
3354ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3355		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3356{
3357	uint32_t i;
3358	union ctl_io *cur_io, *next_io;
3359	struct ctl_io_pool *pool;
3360	int retval;
3361
3362	retval = 0;
3363
3364	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3365					    M_NOWAIT | M_ZERO);
3366	if (pool == NULL) {
3367		retval = -ENOMEM;
3368		goto bailout;
3369	}
3370
3371	pool->type = pool_type;
3372	pool->ctl_softc = ctl_softc;
3373
3374	mtx_lock(&ctl_softc->pool_lock);
3375	pool->id = ctl_softc->cur_pool_id++;
3376	mtx_unlock(&ctl_softc->pool_lock);
3377
3378	pool->flags = CTL_POOL_FLAG_NONE;
3379	pool->refcount = 1;		/* Reference for validity. */
3380	STAILQ_INIT(&pool->free_queue);
3381
3382	/*
3383	 * XXX KDM other options here:
3384	 * - allocate a page at a time
3385	 * - allocate one big chunk of memory.
3386	 * Page allocation might work well, but would take a little more
3387	 * tracking.
3388	 */
3389	for (i = 0; i < total_ctl_io; i++) {
3390		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3391						M_NOWAIT);
3392		if (cur_io == NULL) {
3393			retval = ENOMEM;
3394			break;
3395		}
3396		cur_io->io_hdr.pool = pool;
3397		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3398		pool->total_ctl_io++;
3399		pool->free_ctl_io++;
3400	}
3401
3402	if (retval != 0) {
3403		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3404		     cur_io != NULL; cur_io = next_io) {
3405			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3406							      links);
3407			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3408				      ctl_io_hdr, links);
3409			free(cur_io, M_CTL);
3410		}
3411
3412		free(pool, M_CTL);
3413		goto bailout;
3414	}
3415	mtx_lock(&ctl_softc->pool_lock);
3416	ctl_softc->num_pools++;
3417	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3418	/*
3419	 * Increment our usage count if this is an external consumer, so we
3420	 * can't get unloaded until the external consumer (most likely a
3421	 * FETD) unloads and frees his pool.
3422	 *
3423	 * XXX KDM will this increment the caller's module use count, or
3424	 * mine?
3425	 */
3426#if 0
3427	if ((pool_type != CTL_POOL_EMERGENCY)
3428	 && (pool_type != CTL_POOL_INTERNAL)
3429	 && (pool_type != CTL_POOL_IOCTL)
3430	 && (pool_type != CTL_POOL_4OTHERSC))
3431		MOD_INC_USE_COUNT;
3432#endif
3433
3434	mtx_unlock(&ctl_softc->pool_lock);
3435
3436	*npool = pool;
3437
3438bailout:
3439
3440	return (retval);
3441}
3442
3443static int
3444ctl_pool_acquire(struct ctl_io_pool *pool)
3445{
3446
3447	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3448
3449	if (pool->flags & CTL_POOL_FLAG_INVALID)
3450		return (-EINVAL);
3451
3452	pool->refcount++;
3453
3454	return (0);
3455}
3456
3457static void
3458ctl_pool_release(struct ctl_io_pool *pool)
3459{
3460	struct ctl_softc *ctl_softc = pool->ctl_softc;
3461	union ctl_io *io;
3462
3463	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3464
3465	if (--pool->refcount != 0)
3466		return;
3467
3468	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3469		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3470			      links);
3471		free(io, M_CTL);
3472	}
3473
3474	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3475	ctl_softc->num_pools--;
3476
3477	/*
3478	 * XXX KDM will this decrement the caller's usage count or mine?
3479	 */
3480#if 0
3481	if ((pool->type != CTL_POOL_EMERGENCY)
3482	 && (pool->type != CTL_POOL_INTERNAL)
3483	 && (pool->type != CTL_POOL_IOCTL))
3484		MOD_DEC_USE_COUNT;
3485#endif
3486
3487	free(pool, M_CTL);
3488}
3489
3490void
3491ctl_pool_free(struct ctl_io_pool *pool)
3492{
3493	struct ctl_softc *ctl_softc;
3494
3495	if (pool == NULL)
3496		return;
3497
3498	ctl_softc = pool->ctl_softc;
3499	mtx_lock(&ctl_softc->pool_lock);
3500	pool->flags |= CTL_POOL_FLAG_INVALID;
3501	ctl_pool_release(pool);
3502	mtx_unlock(&ctl_softc->pool_lock);
3503}
3504
3505/*
3506 * This routine does not block (except for spinlocks of course).
3507 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3508 * possible.
3509 */
3510union ctl_io *
3511ctl_alloc_io(void *pool_ref)
3512{
3513	union ctl_io *io;
3514	struct ctl_softc *ctl_softc;
3515	struct ctl_io_pool *pool, *npool;
3516	struct ctl_io_pool *emergency_pool;
3517
3518	pool = (struct ctl_io_pool *)pool_ref;
3519
3520	if (pool == NULL) {
3521		printf("%s: pool is NULL\n", __func__);
3522		return (NULL);
3523	}
3524
3525	emergency_pool = NULL;
3526
3527	ctl_softc = pool->ctl_softc;
3528
3529	mtx_lock(&ctl_softc->pool_lock);
3530	/*
3531	 * First, try to get the io structure from the user's pool.
3532	 */
3533	if (ctl_pool_acquire(pool) == 0) {
3534		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3535		if (io != NULL) {
3536			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3537			pool->total_allocated++;
3538			pool->free_ctl_io--;
3539			mtx_unlock(&ctl_softc->pool_lock);
3540			return (io);
3541		} else
3542			ctl_pool_release(pool);
3543	}
3544	/*
3545	 * If he doesn't have any io structures left, search for an
3546	 * emergency pool and grab one from there.
3547	 */
3548	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3549		if (npool->type != CTL_POOL_EMERGENCY)
3550			continue;
3551
3552		if (ctl_pool_acquire(npool) != 0)
3553			continue;
3554
3555		emergency_pool = npool;
3556
3557		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3558		if (io != NULL) {
3559			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3560			npool->total_allocated++;
3561			npool->free_ctl_io--;
3562			mtx_unlock(&ctl_softc->pool_lock);
3563			return (io);
3564		} else
3565			ctl_pool_release(npool);
3566	}
3567
3568	/* Drop the spinlock before we malloc */
3569	mtx_unlock(&ctl_softc->pool_lock);
3570
3571	/*
3572	 * The emergency pool (if it exists) didn't have one, so try an
3573	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3574	 */
3575	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3576	if (io != NULL) {
3577		/*
3578		 * If the emergency pool exists but is empty, add this
3579		 * ctl_io to its list when it gets freed.
3580		 */
3581		if (emergency_pool != NULL) {
3582			mtx_lock(&ctl_softc->pool_lock);
3583			if (ctl_pool_acquire(emergency_pool) == 0) {
3584				io->io_hdr.pool = emergency_pool;
3585				emergency_pool->total_ctl_io++;
3586				/*
3587				 * Need to bump this, otherwise
3588				 * total_allocated and total_freed won't
3589				 * match when we no longer have anything
3590				 * outstanding.
3591				 */
3592				emergency_pool->total_allocated++;
3593			}
3594			mtx_unlock(&ctl_softc->pool_lock);
3595		} else
3596			io->io_hdr.pool = NULL;
3597	}
3598
3599	return (io);
3600}
3601
3602void
3603ctl_free_io(union ctl_io *io)
3604{
3605	if (io == NULL)
3606		return;
3607
3608	/*
3609	 * If this ctl_io has a pool, return it to that pool.
3610	 */
3611	if (io->io_hdr.pool != NULL) {
3612		struct ctl_io_pool *pool;
3613#if 0
3614		struct ctl_softc *ctl_softc;
3615		union ctl_io *tmp_io;
3616		unsigned long xflags;
3617		int i;
3618
3619		ctl_softc = control_softc;
3620#endif
3621
3622		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3623
3624		mtx_lock(&pool->ctl_softc->pool_lock);
3625#if 0
3626		save_flags(xflags);
3627
3628		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3629		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3630		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3631		     links)) {
3632			if (tmp_io == io) {
3633				printf("%s: %p is still on the task queue!\n",
3634				       __func__, tmp_io);
3635				printf("%s: (%d): type %d "
3636				       "msg %d cdb %x iptl: "
3637				       "%d:%d:%d:%d tag 0x%04x "
3638				       "flg %#lx\n",
3639					__func__, i,
3640					tmp_io->io_hdr.io_type,
3641					tmp_io->io_hdr.msg_type,
3642					tmp_io->scsiio.cdb[0],
3643					tmp_io->io_hdr.nexus.initid.id,
3644					tmp_io->io_hdr.nexus.targ_port,
3645					tmp_io->io_hdr.nexus.targ_target.id,
3646					tmp_io->io_hdr.nexus.targ_lun,
3647					(tmp_io->io_hdr.io_type ==
3648					CTL_IO_TASK) ?
3649					tmp_io->taskio.tag_num :
3650					tmp_io->scsiio.tag_num,
3651					xflags);
3652				panic("I/O still on the task queue!");
3653			}
3654		}
3655#endif
3656		io->io_hdr.io_type = 0xff;
3657		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3658		pool->total_freed++;
3659		pool->free_ctl_io++;
3660		ctl_pool_release(pool);
3661		mtx_unlock(&pool->ctl_softc->pool_lock);
3662	} else {
3663		/*
3664		 * Otherwise, just free it.  We probably malloced it and
3665		 * the emergency pool wasn't available.
3666		 */
3667		free(io, M_CTL);
3668	}
3669
3670}
3671
3672void
3673ctl_zero_io(union ctl_io *io)
3674{
3675	void *pool_ref;
3676
3677	if (io == NULL)
3678		return;
3679
3680	/*
3681	 * May need to preserve linked list pointers at some point too.
3682	 */
3683	pool_ref = io->io_hdr.pool;
3684
3685	memset(io, 0, sizeof(*io));
3686
3687	io->io_hdr.pool = pool_ref;
3688}
3689
3690/*
3691 * This routine is currently used for internal copies of ctl_ios that need
3692 * to persist for some reason after we've already returned status to the
3693 * FETD.  (Thus the flag set.)
3694 *
3695 * XXX XXX
3696 * Note that this makes a blind copy of all fields in the ctl_io, except
3697 * for the pool reference.  This includes any memory that has been
3698 * allocated!  That memory will no longer be valid after done has been
3699 * called, so this would be VERY DANGEROUS for command that actually does
3700 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3701 * start and stop commands, which don't transfer any data, so this is not a
3702 * problem.  If it is used for anything else, the caller would also need to
3703 * allocate data buffer space and this routine would need to be modified to
3704 * copy the data buffer(s) as well.
3705 */
3706void
3707ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3708{
3709	void *pool_ref;
3710
3711	if ((src == NULL)
3712	 || (dest == NULL))
3713		return;
3714
3715	/*
3716	 * May need to preserve linked list pointers at some point too.
3717	 */
3718	pool_ref = dest->io_hdr.pool;
3719
3720	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3721
3722	dest->io_hdr.pool = pool_ref;
3723	/*
3724	 * We need to know that this is an internal copy, and doesn't need
3725	 * to get passed back to the FETD that allocated it.
3726	 */
3727	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3728}
3729
3730#ifdef NEEDTOPORT
3731static void
3732ctl_update_power_subpage(struct copan_power_subpage *page)
3733{
3734	int num_luns, num_partitions, config_type;
3735	struct ctl_softc *softc;
3736	cs_BOOL_t aor_present, shelf_50pct_power;
3737	cs_raidset_personality_t rs_type;
3738	int max_active_luns;
3739
3740	softc = control_softc;
3741
3742	/* subtract out the processor LUN */
3743	num_luns = softc->num_luns - 1;
3744	/*
3745	 * Default to 7 LUNs active, which was the only number we allowed
3746	 * in the past.
3747	 */
3748	max_active_luns = 7;
3749
3750	num_partitions = config_GetRsPartitionInfo();
3751	config_type = config_GetConfigType();
3752	shelf_50pct_power = config_GetShelfPowerMode();
3753	aor_present = config_IsAorRsPresent();
3754
3755	rs_type = ddb_GetRsRaidType(1);
3756	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3757	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3758		EPRINT(0, "Unsupported RS type %d!", rs_type);
3759	}
3760
3761
3762	page->total_luns = num_luns;
3763
3764	switch (config_type) {
3765	case 40:
3766		/*
3767		 * In a 40 drive configuration, it doesn't matter what DC
3768		 * cards we have, whether we have AOR enabled or not,
3769		 * partitioning or not, or what type of RAIDset we have.
3770		 * In that scenario, we can power up every LUN we present
3771		 * to the user.
3772		 */
3773		max_active_luns = num_luns;
3774
3775		break;
3776	case 64:
3777		if (shelf_50pct_power == CS_FALSE) {
3778			/* 25% power */
3779			if (aor_present == CS_TRUE) {
3780				if (rs_type ==
3781				     CS_RAIDSET_PERSONALITY_RAID5) {
3782					max_active_luns = 7;
3783				} else if (rs_type ==
3784					 CS_RAIDSET_PERSONALITY_RAID1){
3785					max_active_luns = 14;
3786				} else {
3787					/* XXX KDM now what?? */
3788				}
3789			} else {
3790				if (rs_type ==
3791				     CS_RAIDSET_PERSONALITY_RAID5) {
3792					max_active_luns = 8;
3793				} else if (rs_type ==
3794					 CS_RAIDSET_PERSONALITY_RAID1){
3795					max_active_luns = 16;
3796				} else {
3797					/* XXX KDM now what?? */
3798				}
3799			}
3800		} else {
3801			/* 50% power */
3802			/*
3803			 * With 50% power in a 64 drive configuration, we
3804			 * can power all LUNs we present.
3805			 */
3806			max_active_luns = num_luns;
3807		}
3808		break;
3809	case 112:
3810		if (shelf_50pct_power == CS_FALSE) {
3811			/* 25% power */
3812			if (aor_present == CS_TRUE) {
3813				if (rs_type ==
3814				     CS_RAIDSET_PERSONALITY_RAID5) {
3815					max_active_luns = 7;
3816				} else if (rs_type ==
3817					 CS_RAIDSET_PERSONALITY_RAID1){
3818					max_active_luns = 14;
3819				} else {
3820					/* XXX KDM now what?? */
3821				}
3822			} else {
3823				if (rs_type ==
3824				     CS_RAIDSET_PERSONALITY_RAID5) {
3825					max_active_luns = 8;
3826				} else if (rs_type ==
3827					 CS_RAIDSET_PERSONALITY_RAID1){
3828					max_active_luns = 16;
3829				} else {
3830					/* XXX KDM now what?? */
3831				}
3832			}
3833		} else {
3834			/* 50% power */
3835			if (aor_present == CS_TRUE) {
3836				if (rs_type ==
3837				     CS_RAIDSET_PERSONALITY_RAID5) {
3838					max_active_luns = 14;
3839				} else if (rs_type ==
3840					 CS_RAIDSET_PERSONALITY_RAID1){
3841					/*
3842					 * We're assuming here that disk
3843					 * caching is enabled, and so we're
3844					 * able to power up half of each
3845					 * LUN, and cache all writes.
3846					 */
3847					max_active_luns = num_luns;
3848				} else {
3849					/* XXX KDM now what?? */
3850				}
3851			} else {
3852				if (rs_type ==
3853				     CS_RAIDSET_PERSONALITY_RAID5) {
3854					max_active_luns = 15;
3855				} else if (rs_type ==
3856					 CS_RAIDSET_PERSONALITY_RAID1){
3857					max_active_luns = 30;
3858				} else {
3859					/* XXX KDM now what?? */
3860				}
3861			}
3862		}
3863		break;
3864	default:
3865		/*
3866		 * In this case, we have an unknown configuration, so we
3867		 * just use the default from above.
3868		 */
3869		break;
3870	}
3871
3872	page->max_active_luns = max_active_luns;
3873#if 0
3874	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3875	       page->total_luns, page->max_active_luns);
3876#endif
3877}
3878#endif /* NEEDTOPORT */
3879
3880/*
3881 * This routine could be used in the future to load default and/or saved
3882 * mode page parameters for a particuar lun.
3883 */
3884static int
3885ctl_init_page_index(struct ctl_lun *lun)
3886{
3887	int i;
3888	struct ctl_page_index *page_index;
3889	struct ctl_softc *softc;
3890
3891	memcpy(&lun->mode_pages.index, page_index_template,
3892	       sizeof(page_index_template));
3893
3894	softc = lun->ctl_softc;
3895
3896	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3897
3898		page_index = &lun->mode_pages.index[i];
3899		/*
3900		 * If this is a disk-only mode page, there's no point in
3901		 * setting it up.  For some pages, we have to have some
3902		 * basic information about the disk in order to calculate the
3903		 * mode page data.
3904		 */
3905		if ((lun->be_lun->lun_type != T_DIRECT)
3906		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3907			continue;
3908
3909		switch (page_index->page_code & SMPH_PC_MASK) {
3910		case SMS_FORMAT_DEVICE_PAGE: {
3911			struct scsi_format_page *format_page;
3912
3913			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3914				panic("subpage is incorrect!");
3915
3916			/*
3917			 * Sectors per track are set above.  Bytes per
3918			 * sector need to be set here on a per-LUN basis.
3919			 */
3920			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3921			       &format_page_default,
3922			       sizeof(format_page_default));
3923			memcpy(&lun->mode_pages.format_page[
3924			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3925			       sizeof(format_page_changeable));
3926			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3927			       &format_page_default,
3928			       sizeof(format_page_default));
3929			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3930			       &format_page_default,
3931			       sizeof(format_page_default));
3932
3933			format_page = &lun->mode_pages.format_page[
3934				CTL_PAGE_CURRENT];
3935			scsi_ulto2b(lun->be_lun->blocksize,
3936				    format_page->bytes_per_sector);
3937
3938			format_page = &lun->mode_pages.format_page[
3939				CTL_PAGE_DEFAULT];
3940			scsi_ulto2b(lun->be_lun->blocksize,
3941				    format_page->bytes_per_sector);
3942
3943			format_page = &lun->mode_pages.format_page[
3944				CTL_PAGE_SAVED];
3945			scsi_ulto2b(lun->be_lun->blocksize,
3946				    format_page->bytes_per_sector);
3947
3948			page_index->page_data =
3949				(uint8_t *)lun->mode_pages.format_page;
3950			break;
3951		}
3952		case SMS_RIGID_DISK_PAGE: {
3953			struct scsi_rigid_disk_page *rigid_disk_page;
3954			uint32_t sectors_per_cylinder;
3955			uint64_t cylinders;
3956#ifndef	__XSCALE__
3957			int shift;
3958#endif /* !__XSCALE__ */
3959
3960			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3961				panic("invalid subpage value %d",
3962				      page_index->subpage);
3963
3964			/*
3965			 * Rotation rate and sectors per track are set
3966			 * above.  We calculate the cylinders here based on
3967			 * capacity.  Due to the number of heads and
3968			 * sectors per track we're using, smaller arrays
3969			 * may turn out to have 0 cylinders.  Linux and
3970			 * FreeBSD don't pay attention to these mode pages
3971			 * to figure out capacity, but Solaris does.  It
3972			 * seems to deal with 0 cylinders just fine, and
3973			 * works out a fake geometry based on the capacity.
3974			 */
3975			memcpy(&lun->mode_pages.rigid_disk_page[
3976			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3977			       sizeof(rigid_disk_page_default));
3978			memcpy(&lun->mode_pages.rigid_disk_page[
3979			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3980			       sizeof(rigid_disk_page_changeable));
3981			memcpy(&lun->mode_pages.rigid_disk_page[
3982			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3983			       sizeof(rigid_disk_page_default));
3984			memcpy(&lun->mode_pages.rigid_disk_page[
3985			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3986			       sizeof(rigid_disk_page_default));
3987
3988			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3989				CTL_DEFAULT_HEADS;
3990
3991			/*
3992			 * The divide method here will be more accurate,
3993			 * probably, but results in floating point being
3994			 * used in the kernel on i386 (__udivdi3()).  On the
3995			 * XScale, though, __udivdi3() is implemented in
3996			 * software.
3997			 *
3998			 * The shift method for cylinder calculation is
3999			 * accurate if sectors_per_cylinder is a power of
4000			 * 2.  Otherwise it might be slightly off -- you
4001			 * might have a bit of a truncation problem.
4002			 */
4003#ifdef	__XSCALE__
4004			cylinders = (lun->be_lun->maxlba + 1) /
4005				sectors_per_cylinder;
4006#else
4007			for (shift = 31; shift > 0; shift--) {
4008				if (sectors_per_cylinder & (1 << shift))
4009					break;
4010			}
4011			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4012#endif
4013
4014			/*
4015			 * We've basically got 3 bytes, or 24 bits for the
4016			 * cylinder size in the mode page.  If we're over,
4017			 * just round down to 2^24.
4018			 */
4019			if (cylinders > 0xffffff)
4020				cylinders = 0xffffff;
4021
4022			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4023				CTL_PAGE_CURRENT];
4024			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4025
4026			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4027				CTL_PAGE_DEFAULT];
4028			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4029
4030			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4031				CTL_PAGE_SAVED];
4032			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4033
4034			page_index->page_data =
4035				(uint8_t *)lun->mode_pages.rigid_disk_page;
4036			break;
4037		}
4038		case SMS_CACHING_PAGE: {
4039
4040			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4041				panic("invalid subpage value %d",
4042				      page_index->subpage);
4043			/*
4044			 * Defaults should be okay here, no calculations
4045			 * needed.
4046			 */
4047			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4048			       &caching_page_default,
4049			       sizeof(caching_page_default));
4050			memcpy(&lun->mode_pages.caching_page[
4051			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4052			       sizeof(caching_page_changeable));
4053			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4054			       &caching_page_default,
4055			       sizeof(caching_page_default));
4056			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4057			       &caching_page_default,
4058			       sizeof(caching_page_default));
4059			page_index->page_data =
4060				(uint8_t *)lun->mode_pages.caching_page;
4061			break;
4062		}
4063		case SMS_CONTROL_MODE_PAGE: {
4064
4065			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4066				panic("invalid subpage value %d",
4067				      page_index->subpage);
4068
4069			/*
4070			 * Defaults should be okay here, no calculations
4071			 * needed.
4072			 */
4073			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4074			       &control_page_default,
4075			       sizeof(control_page_default));
4076			memcpy(&lun->mode_pages.control_page[
4077			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4078			       sizeof(control_page_changeable));
4079			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4080			       &control_page_default,
4081			       sizeof(control_page_default));
4082			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4083			       &control_page_default,
4084			       sizeof(control_page_default));
4085			page_index->page_data =
4086				(uint8_t *)lun->mode_pages.control_page;
4087			break;
4088
4089		}
4090		case SMS_VENDOR_SPECIFIC_PAGE:{
4091			switch (page_index->subpage) {
4092			case PWR_SUBPAGE_CODE: {
4093				struct copan_power_subpage *current_page,
4094							   *saved_page;
4095
4096				memcpy(&lun->mode_pages.power_subpage[
4097				       CTL_PAGE_CURRENT],
4098				       &power_page_default,
4099				       sizeof(power_page_default));
4100				memcpy(&lun->mode_pages.power_subpage[
4101				       CTL_PAGE_CHANGEABLE],
4102				       &power_page_changeable,
4103				       sizeof(power_page_changeable));
4104				memcpy(&lun->mode_pages.power_subpage[
4105				       CTL_PAGE_DEFAULT],
4106				       &power_page_default,
4107				       sizeof(power_page_default));
4108				memcpy(&lun->mode_pages.power_subpage[
4109				       CTL_PAGE_SAVED],
4110				       &power_page_default,
4111				       sizeof(power_page_default));
4112				page_index->page_data =
4113				    (uint8_t *)lun->mode_pages.power_subpage;
4114
4115				current_page = (struct copan_power_subpage *)
4116					(page_index->page_data +
4117					 (page_index->page_len *
4118					  CTL_PAGE_CURRENT));
4119			        saved_page = (struct copan_power_subpage *)
4120				        (page_index->page_data +
4121					 (page_index->page_len *
4122					  CTL_PAGE_SAVED));
4123				break;
4124			}
4125			case APS_SUBPAGE_CODE: {
4126				struct copan_aps_subpage *current_page,
4127							 *saved_page;
4128
4129				// This gets set multiple times but
4130				// it should always be the same. It's
4131				// only done during init so who cares.
4132				index_to_aps_page = i;
4133
4134				memcpy(&lun->mode_pages.aps_subpage[
4135				       CTL_PAGE_CURRENT],
4136				       &aps_page_default,
4137				       sizeof(aps_page_default));
4138				memcpy(&lun->mode_pages.aps_subpage[
4139				       CTL_PAGE_CHANGEABLE],
4140				       &aps_page_changeable,
4141				       sizeof(aps_page_changeable));
4142				memcpy(&lun->mode_pages.aps_subpage[
4143				       CTL_PAGE_DEFAULT],
4144				       &aps_page_default,
4145				       sizeof(aps_page_default));
4146				memcpy(&lun->mode_pages.aps_subpage[
4147				       CTL_PAGE_SAVED],
4148				       &aps_page_default,
4149				       sizeof(aps_page_default));
4150				page_index->page_data =
4151					(uint8_t *)lun->mode_pages.aps_subpage;
4152
4153				current_page = (struct copan_aps_subpage *)
4154					(page_index->page_data +
4155					 (page_index->page_len *
4156					  CTL_PAGE_CURRENT));
4157				saved_page = (struct copan_aps_subpage *)
4158					(page_index->page_data +
4159					 (page_index->page_len *
4160					  CTL_PAGE_SAVED));
4161				break;
4162			}
4163			case DBGCNF_SUBPAGE_CODE: {
4164				struct copan_debugconf_subpage *current_page,
4165							       *saved_page;
4166
4167				memcpy(&lun->mode_pages.debugconf_subpage[
4168				       CTL_PAGE_CURRENT],
4169				       &debugconf_page_default,
4170				       sizeof(debugconf_page_default));
4171				memcpy(&lun->mode_pages.debugconf_subpage[
4172				       CTL_PAGE_CHANGEABLE],
4173				       &debugconf_page_changeable,
4174				       sizeof(debugconf_page_changeable));
4175				memcpy(&lun->mode_pages.debugconf_subpage[
4176				       CTL_PAGE_DEFAULT],
4177				       &debugconf_page_default,
4178				       sizeof(debugconf_page_default));
4179				memcpy(&lun->mode_pages.debugconf_subpage[
4180				       CTL_PAGE_SAVED],
4181				       &debugconf_page_default,
4182				       sizeof(debugconf_page_default));
4183				page_index->page_data =
4184					(uint8_t *)lun->mode_pages.debugconf_subpage;
4185
4186				current_page = (struct copan_debugconf_subpage *)
4187					(page_index->page_data +
4188					 (page_index->page_len *
4189					  CTL_PAGE_CURRENT));
4190				saved_page = (struct copan_debugconf_subpage *)
4191					(page_index->page_data +
4192					 (page_index->page_len *
4193					  CTL_PAGE_SAVED));
4194				break;
4195			}
4196			default:
4197				panic("invalid subpage value %d",
4198				      page_index->subpage);
4199				break;
4200			}
4201   			break;
4202		}
4203		default:
4204			panic("invalid page value %d",
4205			      page_index->page_code & SMPH_PC_MASK);
4206			break;
4207    	}
4208	}
4209
4210	return (CTL_RETVAL_COMPLETE);
4211}
4212
4213/*
4214 * LUN allocation.
4215 *
4216 * Requirements:
4217 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4218 *   wants us to allocate the LUN and he can block.
4219 * - ctl_softc is always set
4220 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4221 *
4222 * Returns 0 for success, non-zero (errno) for failure.
4223 */
4224static int
4225ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4226	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4227{
4228	struct ctl_lun *nlun, *lun;
4229	struct ctl_frontend *fe;
4230	int lun_number, i, lun_malloced;
4231
4232	if (be_lun == NULL)
4233		return (EINVAL);
4234
4235	/*
4236	 * We currently only support Direct Access or Processor LUN types.
4237	 */
4238	switch (be_lun->lun_type) {
4239	case T_DIRECT:
4240		break;
4241	case T_PROCESSOR:
4242		break;
4243	case T_SEQUENTIAL:
4244	case T_CHANGER:
4245	default:
4246		be_lun->lun_config_status(be_lun->be_lun,
4247					  CTL_LUN_CONFIG_FAILURE);
4248		break;
4249	}
4250	if (ctl_lun == NULL) {
4251		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4252		lun_malloced = 1;
4253	} else {
4254		lun_malloced = 0;
4255		lun = ctl_lun;
4256	}
4257
4258	memset(lun, 0, sizeof(*lun));
4259	if (lun_malloced)
4260		lun->flags = CTL_LUN_MALLOCED;
4261
4262	mtx_lock(&ctl_softc->ctl_lock);
4263	/*
4264	 * See if the caller requested a particular LUN number.  If so, see
4265	 * if it is available.  Otherwise, allocate the first available LUN.
4266	 */
4267	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4268		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4269		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4270			mtx_unlock(&ctl_softc->ctl_lock);
4271			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4272				printf("ctl: requested LUN ID %d is higher "
4273				       "than CTL_MAX_LUNS - 1 (%d)\n",
4274				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4275			} else {
4276				/*
4277				 * XXX KDM return an error, or just assign
4278				 * another LUN ID in this case??
4279				 */
4280				printf("ctl: requested LUN ID %d is already "
4281				       "in use\n", be_lun->req_lun_id);
4282			}
4283			if (lun->flags & CTL_LUN_MALLOCED)
4284				free(lun, M_CTL);
4285			be_lun->lun_config_status(be_lun->be_lun,
4286						  CTL_LUN_CONFIG_FAILURE);
4287			return (ENOSPC);
4288		}
4289		lun_number = be_lun->req_lun_id;
4290	} else {
4291		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4292		if (lun_number == -1) {
4293			mtx_unlock(&ctl_softc->ctl_lock);
4294			printf("ctl: can't allocate LUN on target %ju, out of "
4295			       "LUNs\n", (uintmax_t)target_id.id);
4296			if (lun->flags & CTL_LUN_MALLOCED)
4297				free(lun, M_CTL);
4298			be_lun->lun_config_status(be_lun->be_lun,
4299						  CTL_LUN_CONFIG_FAILURE);
4300			return (ENOSPC);
4301		}
4302	}
4303	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4304
4305	lun->target = target_id;
4306	lun->lun = lun_number;
4307	lun->be_lun = be_lun;
4308	/*
4309	 * The processor LUN is always enabled.  Disk LUNs come on line
4310	 * disabled, and must be enabled by the backend.
4311	 */
4312	lun->flags |= CTL_LUN_DISABLED;
4313	lun->backend = be_lun->be;
4314	be_lun->ctl_lun = lun;
4315	be_lun->lun_id = lun_number;
4316	atomic_add_int(&be_lun->be->num_luns, 1);
4317	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4318		lun->flags |= CTL_LUN_STOPPED;
4319
4320	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4321		lun->flags |= CTL_LUN_INOPERABLE;
4322
4323	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4324		lun->flags |= CTL_LUN_PRIMARY_SC;
4325
4326	lun->ctl_softc = ctl_softc;
4327	TAILQ_INIT(&lun->ooa_queue);
4328	TAILQ_INIT(&lun->blocked_queue);
4329	STAILQ_INIT(&lun->error_list);
4330
4331	/*
4332	 * Initialize the mode page index.
4333	 */
4334	ctl_init_page_index(lun);
4335
4336	/*
4337	 * Set the poweron UA for all initiators on this LUN only.
4338	 */
4339	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4340		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4341
4342	/*
4343	 * Now, before we insert this lun on the lun list, set the lun
4344	 * inventory changed UA for all other luns.
4345	 */
4346	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4347		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4348			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4349		}
4350	}
4351
4352	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4353
4354	ctl_softc->ctl_luns[lun_number] = lun;
4355
4356	ctl_softc->num_luns++;
4357
4358	/* Setup statistics gathering */
4359	lun->stats.device_type = be_lun->lun_type;
4360	lun->stats.lun_number = lun_number;
4361	if (lun->stats.device_type == T_DIRECT)
4362		lun->stats.blocksize = be_lun->blocksize;
4363	else
4364		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4365	for (i = 0;i < CTL_MAX_PORTS;i++)
4366		lun->stats.ports[i].targ_port = i;
4367
4368	mtx_unlock(&ctl_softc->ctl_lock);
4369
4370	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4371
4372	/*
4373	 * Run through each registered FETD and bring it online if it isn't
4374	 * already.  Enable the target ID if it hasn't been enabled, and
4375	 * enable this particular LUN.
4376	 */
4377	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4378		int retval;
4379
4380		/*
4381		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4382		 * to do things differently if we go to a multiple target
4383		 * ID scheme.
4384		 */
4385		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4386
4387			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4388			if (retval != 0) {
4389				printf("ctl_alloc_lun: FETD %s port %d "
4390				       "returned error %d for targ_enable on "
4391				       "target %ju\n", fe->port_name,
4392				       fe->targ_port, retval,
4393				       (uintmax_t)target_id.id);
4394			} else
4395				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4396		}
4397
4398		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4399		if (retval != 0) {
4400			printf("ctl_alloc_lun: FETD %s port %d returned error "
4401			       "%d for lun_enable on target %ju lun %d\n",
4402			       fe->port_name, fe->targ_port, retval,
4403			       (uintmax_t)target_id.id, lun_number);
4404		} else
4405			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4406	}
4407	return (0);
4408}
4409
4410/*
4411 * Delete a LUN.
4412 * Assumptions:
4413 * - LUN has already been marked invalid and any pending I/O has been taken
4414 *   care of.
4415 */
4416static int
4417ctl_free_lun(struct ctl_lun *lun)
4418{
4419	struct ctl_softc *softc;
4420#if 0
4421	struct ctl_frontend *fe;
4422#endif
4423	struct ctl_lun *nlun;
4424	union ctl_io *io, *next_io;
4425	int i;
4426
4427	softc = lun->ctl_softc;
4428
4429	mtx_assert(&softc->ctl_lock, MA_OWNED);
4430
4431	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4432
4433	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4434
4435	softc->ctl_luns[lun->lun] = NULL;
4436
4437	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4438		printf("ctl_free_lun: aieee!! freeing a LUN with "
4439		       "outstanding I/O!!\n");
4440	}
4441
4442	/*
4443	 * If we have anything pending on the RtR queue, remove it.
4444	 */
4445	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4446	     io = next_io) {
4447		uint32_t targ_lun;
4448
4449		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4450		targ_lun = io->io_hdr.nexus.targ_lun;
4451		if (io->io_hdr.nexus.lun_map_fn != NULL)
4452			targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
4453		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4454		 && (targ_lun == lun->lun))
4455			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4456				      ctl_io_hdr, links);
4457	}
4458
4459	/*
4460	 * Then remove everything from the blocked queue.
4461	 */
4462	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4463	     io = next_io) {
4464		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4465		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4466		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4467	}
4468
4469	/*
4470	 * Now clear out the OOA queue, and free all the I/O.
4471	 * XXX KDM should we notify the FETD here?  We probably need to
4472	 * quiesce the LUN before deleting it.
4473	 */
4474	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4475	     io = next_io) {
4476		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4477		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4478		ctl_free_io(io);
4479	}
4480
4481	softc->num_luns--;
4482
4483	/*
4484	 * XXX KDM this scheme only works for a single target/multiple LUN
4485	 * setup.  It needs to be revamped for a multiple target scheme.
4486	 *
4487	 * XXX KDM this results in fe->lun_disable() getting called twice,
4488	 * once when ctl_disable_lun() is called, and a second time here.
4489	 * We really need to re-think the LUN disable semantics.  There
4490	 * should probably be several steps/levels to LUN removal:
4491	 *  - disable
4492	 *  - invalidate
4493	 *  - free
4494 	 *
4495	 * Right now we only have a disable method when communicating to
4496	 * the front end ports, at least for individual LUNs.
4497	 */
4498#if 0
4499	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4500		int retval;
4501
4502		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4503					 lun->lun);
4504		if (retval != 0) {
4505			printf("ctl_free_lun: FETD %s port %d returned error "
4506			       "%d for lun_disable on target %ju lun %jd\n",
4507			       fe->port_name, fe->targ_port, retval,
4508			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4509		}
4510
4511		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4512			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4513
4514			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4515			if (retval != 0) {
4516				printf("ctl_free_lun: FETD %s port %d "
4517				       "returned error %d for targ_disable on "
4518				       "target %ju\n", fe->port_name,
4519				       fe->targ_port, retval,
4520				       (uintmax_t)lun->target.id);
4521			} else
4522				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4523
4524			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4525				continue;
4526
4527#if 0
4528			fe->port_offline(fe->onoff_arg);
4529			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4530#endif
4531		}
4532	}
4533#endif
4534
4535	/*
4536	 * Tell the backend to free resources, if this LUN has a backend.
4537	 */
4538	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4539	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4540
4541	if (lun->flags & CTL_LUN_MALLOCED)
4542		free(lun, M_CTL);
4543
4544	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4545		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4546			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4547		}
4548	}
4549
4550	return (0);
4551}
4552
4553static void
4554ctl_create_lun(struct ctl_be_lun *be_lun)
4555{
4556	struct ctl_softc *ctl_softc;
4557
4558	ctl_softc = control_softc;
4559
4560	/*
4561	 * ctl_alloc_lun() should handle all potential failure cases.
4562	 */
4563	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4564}
4565
4566int
4567ctl_add_lun(struct ctl_be_lun *be_lun)
4568{
4569	struct ctl_softc *ctl_softc;
4570
4571	ctl_softc = control_softc;
4572
4573	mtx_lock(&ctl_softc->ctl_lock);
4574	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4575	mtx_unlock(&ctl_softc->ctl_lock);
4576
4577	ctl_wakeup_thread();
4578
4579	return (0);
4580}
4581
4582int
4583ctl_enable_lun(struct ctl_be_lun *be_lun)
4584{
4585	struct ctl_softc *ctl_softc;
4586	struct ctl_frontend *fe, *nfe;
4587	struct ctl_lun *lun;
4588	int retval;
4589
4590	ctl_softc = control_softc;
4591
4592	lun = (struct ctl_lun *)be_lun->ctl_lun;
4593
4594	mtx_lock(&ctl_softc->ctl_lock);
4595	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4596		/*
4597		 * eh?  Why did we get called if the LUN is already
4598		 * enabled?
4599		 */
4600		mtx_unlock(&ctl_softc->ctl_lock);
4601		return (0);
4602	}
4603	lun->flags &= ~CTL_LUN_DISABLED;
4604
4605	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4606		nfe = STAILQ_NEXT(fe, links);
4607
4608		/*
4609		 * Drop the lock while we call the FETD's enable routine.
4610		 * This can lead to a callback into CTL (at least in the
4611		 * case of the internal initiator frontend.
4612		 */
4613		mtx_unlock(&ctl_softc->ctl_lock);
4614		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4615		mtx_lock(&ctl_softc->ctl_lock);
4616		if (retval != 0) {
4617			printf("%s: FETD %s port %d returned error "
4618			       "%d for lun_enable on target %ju lun %jd\n",
4619			       __func__, fe->port_name, fe->targ_port, retval,
4620			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4621		}
4622#if 0
4623		 else {
4624            /* NOTE:  TODO:  why does lun enable affect port status? */
4625			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4626		}
4627#endif
4628	}
4629
4630	mtx_unlock(&ctl_softc->ctl_lock);
4631
4632	return (0);
4633}
4634
4635int
4636ctl_disable_lun(struct ctl_be_lun *be_lun)
4637{
4638	struct ctl_softc *ctl_softc;
4639	struct ctl_frontend *fe;
4640	struct ctl_lun *lun;
4641	int retval;
4642
4643	ctl_softc = control_softc;
4644
4645	lun = (struct ctl_lun *)be_lun->ctl_lun;
4646
4647	mtx_lock(&ctl_softc->ctl_lock);
4648
4649	if (lun->flags & CTL_LUN_DISABLED) {
4650		mtx_unlock(&ctl_softc->ctl_lock);
4651		return (0);
4652	}
4653	lun->flags |= CTL_LUN_DISABLED;
4654
4655	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4656		mtx_unlock(&ctl_softc->ctl_lock);
4657		/*
4658		 * Drop the lock before we call the frontend's disable
4659		 * routine, to avoid lock order reversals.
4660		 *
4661		 * XXX KDM what happens if the frontend list changes while
4662		 * we're traversing it?  It's unlikely, but should be handled.
4663		 */
4664		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4665					 lun->lun);
4666		mtx_lock(&ctl_softc->ctl_lock);
4667		if (retval != 0) {
4668			printf("ctl_alloc_lun: FETD %s port %d returned error "
4669			       "%d for lun_disable on target %ju lun %jd\n",
4670			       fe->port_name, fe->targ_port, retval,
4671			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4672		}
4673	}
4674
4675	mtx_unlock(&ctl_softc->ctl_lock);
4676
4677	return (0);
4678}
4679
4680int
4681ctl_start_lun(struct ctl_be_lun *be_lun)
4682{
4683	struct ctl_softc *ctl_softc;
4684	struct ctl_lun *lun;
4685
4686	ctl_softc = control_softc;
4687
4688	lun = (struct ctl_lun *)be_lun->ctl_lun;
4689
4690	mtx_lock(&ctl_softc->ctl_lock);
4691	lun->flags &= ~CTL_LUN_STOPPED;
4692	mtx_unlock(&ctl_softc->ctl_lock);
4693
4694	return (0);
4695}
4696
4697int
4698ctl_stop_lun(struct ctl_be_lun *be_lun)
4699{
4700	struct ctl_softc *ctl_softc;
4701	struct ctl_lun *lun;
4702
4703	ctl_softc = control_softc;
4704
4705	lun = (struct ctl_lun *)be_lun->ctl_lun;
4706
4707	mtx_lock(&ctl_softc->ctl_lock);
4708	lun->flags |= CTL_LUN_STOPPED;
4709	mtx_unlock(&ctl_softc->ctl_lock);
4710
4711	return (0);
4712}
4713
4714int
4715ctl_lun_offline(struct ctl_be_lun *be_lun)
4716{
4717	struct ctl_softc *ctl_softc;
4718	struct ctl_lun *lun;
4719
4720	ctl_softc = control_softc;
4721
4722	lun = (struct ctl_lun *)be_lun->ctl_lun;
4723
4724	mtx_lock(&ctl_softc->ctl_lock);
4725	lun->flags |= CTL_LUN_OFFLINE;
4726	mtx_unlock(&ctl_softc->ctl_lock);
4727
4728	return (0);
4729}
4730
4731int
4732ctl_lun_online(struct ctl_be_lun *be_lun)
4733{
4734	struct ctl_softc *ctl_softc;
4735	struct ctl_lun *lun;
4736
4737	ctl_softc = control_softc;
4738
4739	lun = (struct ctl_lun *)be_lun->ctl_lun;
4740
4741	mtx_lock(&ctl_softc->ctl_lock);
4742	lun->flags &= ~CTL_LUN_OFFLINE;
4743	mtx_unlock(&ctl_softc->ctl_lock);
4744
4745	return (0);
4746}
4747
4748int
4749ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4750{
4751	struct ctl_softc *ctl_softc;
4752	struct ctl_lun *lun;
4753
4754	ctl_softc = control_softc;
4755
4756	lun = (struct ctl_lun *)be_lun->ctl_lun;
4757
4758	mtx_lock(&ctl_softc->ctl_lock);
4759
4760	/*
4761	 * The LUN needs to be disabled before it can be marked invalid.
4762	 */
4763	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4764		mtx_unlock(&ctl_softc->ctl_lock);
4765		return (-1);
4766	}
4767	/*
4768	 * Mark the LUN invalid.
4769	 */
4770	lun->flags |= CTL_LUN_INVALID;
4771
4772	/*
4773	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4774	 * If we have something in the OOA queue, we'll free it when the
4775	 * last I/O completes.
4776	 */
4777	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4778		ctl_free_lun(lun);
4779	mtx_unlock(&ctl_softc->ctl_lock);
4780
4781	return (0);
4782}
4783
4784int
4785ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4786{
4787	struct ctl_softc *ctl_softc;
4788	struct ctl_lun *lun;
4789
4790	ctl_softc = control_softc;
4791	lun = (struct ctl_lun *)be_lun->ctl_lun;
4792
4793	mtx_lock(&ctl_softc->ctl_lock);
4794	lun->flags |= CTL_LUN_INOPERABLE;
4795	mtx_unlock(&ctl_softc->ctl_lock);
4796
4797	return (0);
4798}
4799
4800int
4801ctl_lun_operable(struct ctl_be_lun *be_lun)
4802{
4803	struct ctl_softc *ctl_softc;
4804	struct ctl_lun *lun;
4805
4806	ctl_softc = control_softc;
4807	lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809	mtx_lock(&ctl_softc->ctl_lock);
4810	lun->flags &= ~CTL_LUN_INOPERABLE;
4811	mtx_unlock(&ctl_softc->ctl_lock);
4812
4813	return (0);
4814}
4815
4816int
4817ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4818		   int lock)
4819{
4820	struct ctl_softc *softc;
4821	struct ctl_lun *lun;
4822	struct copan_aps_subpage *current_sp;
4823	struct ctl_page_index *page_index;
4824	int i;
4825
4826	softc = control_softc;
4827
4828	mtx_lock(&softc->ctl_lock);
4829
4830	lun = (struct ctl_lun *)be_lun->ctl_lun;
4831
4832	page_index = NULL;
4833	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4834		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4835		     APS_PAGE_CODE)
4836			continue;
4837
4838		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4839			continue;
4840		page_index = &lun->mode_pages.index[i];
4841	}
4842
4843	if (page_index == NULL) {
4844		mtx_unlock(&softc->ctl_lock);
4845		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4846		       (uintmax_t)lun->lun);
4847		return (1);
4848	}
4849#if 0
4850	if ((softc->aps_locked_lun != 0)
4851	 && (softc->aps_locked_lun != lun->lun)) {
4852		printf("%s: attempt to lock LUN %llu when %llu is already "
4853		       "locked\n");
4854		mtx_unlock(&softc->ctl_lock);
4855		return (1);
4856	}
4857#endif
4858
4859	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4860		(page_index->page_len * CTL_PAGE_CURRENT));
4861
4862	if (lock != 0) {
4863		current_sp->lock_active = APS_LOCK_ACTIVE;
4864		softc->aps_locked_lun = lun->lun;
4865	} else {
4866		current_sp->lock_active = 0;
4867		softc->aps_locked_lun = 0;
4868	}
4869
4870
4871	/*
4872	 * If we're in HA mode, try to send the lock message to the other
4873	 * side.
4874	 */
4875	if (ctl_is_single == 0) {
4876		int isc_retval;
4877		union ctl_ha_msg lock_msg;
4878
4879		lock_msg.hdr.nexus = *nexus;
4880		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4881		if (lock != 0)
4882			lock_msg.aps.lock_flag = 1;
4883		else
4884			lock_msg.aps.lock_flag = 0;
4885		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4886					 sizeof(lock_msg), 0);
4887		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4888			printf("%s: APS (lock=%d) error returned from "
4889			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4890			mtx_unlock(&softc->ctl_lock);
4891			return (1);
4892		}
4893	}
4894
4895	mtx_unlock(&softc->ctl_lock);
4896
4897	return (0);
4898}
4899
4900void
4901ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4902{
4903	struct ctl_lun *lun;
4904	struct ctl_softc *softc;
4905	int i;
4906
4907	softc = control_softc;
4908
4909	mtx_lock(&softc->ctl_lock);
4910
4911	lun = (struct ctl_lun *)be_lun->ctl_lun;
4912
4913	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4914		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4915
4916	mtx_unlock(&softc->ctl_lock);
4917}
4918
4919/*
4920 * Backend "memory move is complete" callback for requests that never
4921 * make it down to say RAIDCore's configuration code.
4922 */
4923int
4924ctl_config_move_done(union ctl_io *io)
4925{
4926	int retval;
4927
4928	retval = CTL_RETVAL_COMPLETE;
4929
4930
4931	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4932	/*
4933	 * XXX KDM this shouldn't happen, but what if it does?
4934	 */
4935	if (io->io_hdr.io_type != CTL_IO_SCSI)
4936		panic("I/O type isn't CTL_IO_SCSI!");
4937
4938	if ((io->io_hdr.port_status == 0)
4939	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4940	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4941		io->io_hdr.status = CTL_SUCCESS;
4942	else if ((io->io_hdr.port_status != 0)
4943	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4944	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4945		/*
4946		 * For hardware error sense keys, the sense key
4947		 * specific value is defined to be a retry count,
4948		 * but we use it to pass back an internal FETD
4949		 * error code.  XXX KDM  Hopefully the FETD is only
4950		 * using 16 bits for an error code, since that's
4951		 * all the space we have in the sks field.
4952		 */
4953		ctl_set_internal_failure(&io->scsiio,
4954					 /*sks_valid*/ 1,
4955					 /*retry_count*/
4956					 io->io_hdr.port_status);
4957		free(io->scsiio.kern_data_ptr, M_CTL);
4958		ctl_done(io);
4959		goto bailout;
4960	}
4961
4962	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4963	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4964	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4965		/*
4966		 * XXX KDM just assuming a single pointer here, and not a
4967		 * S/G list.  If we start using S/G lists for config data,
4968		 * we'll need to know how to clean them up here as well.
4969		 */
4970		free(io->scsiio.kern_data_ptr, M_CTL);
4971		/* Hopefully the user has already set the status... */
4972		ctl_done(io);
4973	} else {
4974		/*
4975		 * XXX KDM now we need to continue data movement.  Some
4976		 * options:
4977		 * - call ctl_scsiio() again?  We don't do this for data
4978		 *   writes, because for those at least we know ahead of
4979		 *   time where the write will go and how long it is.  For
4980		 *   config writes, though, that information is largely
4981		 *   contained within the write itself, thus we need to
4982		 *   parse out the data again.
4983		 *
4984		 * - Call some other function once the data is in?
4985		 */
4986
4987		/*
4988		 * XXX KDM call ctl_scsiio() again for now, and check flag
4989		 * bits to see whether we're allocated or not.
4990		 */
4991		retval = ctl_scsiio(&io->scsiio);
4992	}
4993bailout:
4994	return (retval);
4995}
4996
4997/*
4998 * This gets called by a backend driver when it is done with a
4999 * data_submit method.
5000 */
5001void
5002ctl_data_submit_done(union ctl_io *io)
5003{
5004	/*
5005	 * If the IO_CONT flag is set, we need to call the supplied
5006	 * function to continue processing the I/O, instead of completing
5007	 * the I/O just yet.
5008	 *
5009	 * If there is an error, though, we don't want to keep processing.
5010	 * Instead, just send status back to the initiator.
5011	 */
5012	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5013	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5014	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5015		io->scsiio.io_cont(io);
5016		return;
5017	}
5018	ctl_done(io);
5019}
5020
5021/*
5022 * This gets called by a backend driver when it is done with a
5023 * configuration write.
5024 */
5025void
5026ctl_config_write_done(union ctl_io *io)
5027{
5028	/*
5029	 * If the IO_CONT flag is set, we need to call the supplied
5030	 * function to continue processing the I/O, instead of completing
5031	 * the I/O just yet.
5032	 *
5033	 * If there is an error, though, we don't want to keep processing.
5034	 * Instead, just send status back to the initiator.
5035	 */
5036	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5037	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5038	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5039		io->scsiio.io_cont(io);
5040		return;
5041	}
5042	/*
5043	 * Since a configuration write can be done for commands that actually
5044	 * have data allocated, like write buffer, and commands that have
5045	 * no data, like start/stop unit, we need to check here.
5046	 */
5047	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5048		free(io->scsiio.kern_data_ptr, M_CTL);
5049	ctl_done(io);
5050}
5051
5052/*
5053 * SCSI release command.
5054 */
5055int
5056ctl_scsi_release(struct ctl_scsiio *ctsio)
5057{
5058	int length, longid, thirdparty_id, resv_id;
5059	struct ctl_softc *ctl_softc;
5060	struct ctl_lun *lun;
5061
5062	length = 0;
5063	resv_id = 0;
5064
5065	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5066
5067	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5068	ctl_softc = control_softc;
5069
5070	switch (ctsio->cdb[0]) {
5071	case RELEASE: {
5072		struct scsi_release *cdb;
5073
5074		cdb = (struct scsi_release *)ctsio->cdb;
5075		if ((cdb->byte2 & 0x1f) != 0) {
5076			ctl_set_invalid_field(ctsio,
5077					      /*sks_valid*/ 1,
5078					      /*command*/ 1,
5079					      /*field*/ 1,
5080					      /*bit_valid*/ 0,
5081					      /*bit*/ 0);
5082			ctl_done((union ctl_io *)ctsio);
5083			return (CTL_RETVAL_COMPLETE);
5084		}
5085		break;
5086	}
5087	case RELEASE_10: {
5088		struct scsi_release_10 *cdb;
5089
5090		cdb = (struct scsi_release_10 *)ctsio->cdb;
5091
5092		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5093			ctl_set_invalid_field(ctsio,
5094					      /*sks_valid*/ 1,
5095					      /*command*/ 1,
5096					      /*field*/ 1,
5097					      /*bit_valid*/ 1,
5098					      /*bit*/ 0);
5099			ctl_done((union ctl_io *)ctsio);
5100			return (CTL_RETVAL_COMPLETE);
5101
5102		}
5103
5104		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5105			ctl_set_invalid_field(ctsio,
5106					      /*sks_valid*/ 1,
5107					      /*command*/ 1,
5108					      /*field*/ 1,
5109					      /*bit_valid*/ 1,
5110					      /*bit*/ 4);
5111			ctl_done((union ctl_io *)ctsio);
5112			return (CTL_RETVAL_COMPLETE);
5113		}
5114
5115		if (cdb->byte2 & SR10_LONGID)
5116			longid = 1;
5117		else
5118			thirdparty_id = cdb->thirdparty_id;
5119
5120		resv_id = cdb->resv_id;
5121		length = scsi_2btoul(cdb->length);
5122		break;
5123	}
5124	}
5125
5126
5127	/*
5128	 * XXX KDM right now, we only support LUN reservation.  We don't
5129	 * support 3rd party reservations, or extent reservations, which
5130	 * might actually need the parameter list.  If we've gotten this
5131	 * far, we've got a LUN reservation.  Anything else got kicked out
5132	 * above.  So, according to SPC, ignore the length.
5133	 */
5134	length = 0;
5135
5136	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5137	 && (length > 0)) {
5138		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5139		ctsio->kern_data_len = length;
5140		ctsio->kern_total_len = length;
5141		ctsio->kern_data_resid = 0;
5142		ctsio->kern_rel_offset = 0;
5143		ctsio->kern_sg_entries = 0;
5144		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5145		ctsio->be_move_done = ctl_config_move_done;
5146		ctl_datamove((union ctl_io *)ctsio);
5147
5148		return (CTL_RETVAL_COMPLETE);
5149	}
5150
5151	if (length > 0)
5152		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5153
5154	mtx_lock(&ctl_softc->ctl_lock);
5155
5156	/*
5157	 * According to SPC, it is not an error for an intiator to attempt
5158	 * to release a reservation on a LUN that isn't reserved, or that
5159	 * is reserved by another initiator.  The reservation can only be
5160	 * released, though, by the initiator who made it or by one of
5161	 * several reset type events.
5162	 */
5163	if (lun->flags & CTL_LUN_RESERVED) {
5164		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5165		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5166		 && (ctsio->io_hdr.nexus.targ_target.id ==
5167		     lun->rsv_nexus.targ_target.id)) {
5168			lun->flags &= ~CTL_LUN_RESERVED;
5169		}
5170	}
5171
5172	ctsio->scsi_status = SCSI_STATUS_OK;
5173	ctsio->io_hdr.status = CTL_SUCCESS;
5174
5175	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5176		free(ctsio->kern_data_ptr, M_CTL);
5177		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5178	}
5179
5180	mtx_unlock(&ctl_softc->ctl_lock);
5181
5182	ctl_done((union ctl_io *)ctsio);
5183	return (CTL_RETVAL_COMPLETE);
5184}
5185
5186int
5187ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5188{
5189	int extent, thirdparty, longid;
5190	int resv_id, length;
5191	uint64_t thirdparty_id;
5192	struct ctl_softc *ctl_softc;
5193	struct ctl_lun *lun;
5194
5195	extent = 0;
5196	thirdparty = 0;
5197	longid = 0;
5198	resv_id = 0;
5199	length = 0;
5200	thirdparty_id = 0;
5201
5202	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5203
5204	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5205	ctl_softc = control_softc;
5206
5207	switch (ctsio->cdb[0]) {
5208	case RESERVE: {
5209		struct scsi_reserve *cdb;
5210
5211		cdb = (struct scsi_reserve *)ctsio->cdb;
5212		if ((cdb->byte2 & 0x1f) != 0) {
5213			ctl_set_invalid_field(ctsio,
5214					      /*sks_valid*/ 1,
5215					      /*command*/ 1,
5216					      /*field*/ 1,
5217					      /*bit_valid*/ 0,
5218					      /*bit*/ 0);
5219			ctl_done((union ctl_io *)ctsio);
5220			return (CTL_RETVAL_COMPLETE);
5221		}
5222		resv_id = cdb->resv_id;
5223		length = scsi_2btoul(cdb->length);
5224		break;
5225	}
5226	case RESERVE_10: {
5227		struct scsi_reserve_10 *cdb;
5228
5229		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5230
5231		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5232			ctl_set_invalid_field(ctsio,
5233					      /*sks_valid*/ 1,
5234					      /*command*/ 1,
5235					      /*field*/ 1,
5236					      /*bit_valid*/ 1,
5237					      /*bit*/ 0);
5238			ctl_done((union ctl_io *)ctsio);
5239			return (CTL_RETVAL_COMPLETE);
5240		}
5241		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5242			ctl_set_invalid_field(ctsio,
5243					      /*sks_valid*/ 1,
5244					      /*command*/ 1,
5245					      /*field*/ 1,
5246					      /*bit_valid*/ 1,
5247					      /*bit*/ 4);
5248			ctl_done((union ctl_io *)ctsio);
5249			return (CTL_RETVAL_COMPLETE);
5250		}
5251		if (cdb->byte2 & SR10_LONGID)
5252			longid = 1;
5253		else
5254			thirdparty_id = cdb->thirdparty_id;
5255
5256		resv_id = cdb->resv_id;
5257		length = scsi_2btoul(cdb->length);
5258		break;
5259	}
5260	}
5261
5262	/*
5263	 * XXX KDM right now, we only support LUN reservation.  We don't
5264	 * support 3rd party reservations, or extent reservations, which
5265	 * might actually need the parameter list.  If we've gotten this
5266	 * far, we've got a LUN reservation.  Anything else got kicked out
5267	 * above.  So, according to SPC, ignore the length.
5268	 */
5269	length = 0;
5270
5271	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5272	 && (length > 0)) {
5273		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5274		ctsio->kern_data_len = length;
5275		ctsio->kern_total_len = length;
5276		ctsio->kern_data_resid = 0;
5277		ctsio->kern_rel_offset = 0;
5278		ctsio->kern_sg_entries = 0;
5279		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5280		ctsio->be_move_done = ctl_config_move_done;
5281		ctl_datamove((union ctl_io *)ctsio);
5282
5283		return (CTL_RETVAL_COMPLETE);
5284	}
5285
5286	if (length > 0)
5287		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5288
5289	mtx_lock(&ctl_softc->ctl_lock);
5290	if (lun->flags & CTL_LUN_RESERVED) {
5291		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5292		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5293		 || (ctsio->io_hdr.nexus.targ_target.id !=
5294		     lun->rsv_nexus.targ_target.id)) {
5295			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5296			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5297			goto bailout;
5298		}
5299	}
5300
5301	lun->flags |= CTL_LUN_RESERVED;
5302	lun->rsv_nexus = ctsio->io_hdr.nexus;
5303
5304	ctsio->scsi_status = SCSI_STATUS_OK;
5305	ctsio->io_hdr.status = CTL_SUCCESS;
5306
5307bailout:
5308	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5309		free(ctsio->kern_data_ptr, M_CTL);
5310		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5311	}
5312
5313	mtx_unlock(&ctl_softc->ctl_lock);
5314
5315	ctl_done((union ctl_io *)ctsio);
5316	return (CTL_RETVAL_COMPLETE);
5317}
5318
5319int
5320ctl_start_stop(struct ctl_scsiio *ctsio)
5321{
5322	struct scsi_start_stop_unit *cdb;
5323	struct ctl_lun *lun;
5324	struct ctl_softc *ctl_softc;
5325	int retval;
5326
5327	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5328
5329	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5330	ctl_softc = control_softc;
5331	retval = 0;
5332
5333	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5334
5335	/*
5336	 * XXX KDM
5337	 * We don't support the immediate bit on a stop unit.  In order to
5338	 * do that, we would need to code up a way to know that a stop is
5339	 * pending, and hold off any new commands until it completes, one
5340	 * way or another.  Then we could accept or reject those commands
5341	 * depending on its status.  We would almost need to do the reverse
5342	 * of what we do below for an immediate start -- return the copy of
5343	 * the ctl_io to the FETD with status to send to the host (and to
5344	 * free the copy!) and then free the original I/O once the stop
5345	 * actually completes.  That way, the OOA queue mechanism can work
5346	 * to block commands that shouldn't proceed.  Another alternative
5347	 * would be to put the copy in the queue in place of the original,
5348	 * and return the original back to the caller.  That could be
5349	 * slightly safer..
5350	 */
5351	if ((cdb->byte2 & SSS_IMMED)
5352	 && ((cdb->how & SSS_START) == 0)) {
5353		ctl_set_invalid_field(ctsio,
5354				      /*sks_valid*/ 1,
5355				      /*command*/ 1,
5356				      /*field*/ 1,
5357				      /*bit_valid*/ 1,
5358				      /*bit*/ 0);
5359		ctl_done((union ctl_io *)ctsio);
5360		return (CTL_RETVAL_COMPLETE);
5361	}
5362
5363	/*
5364	 * We don't support the power conditions field.  We need to check
5365	 * this prior to checking the load/eject and start/stop bits.
5366	 */
5367	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5368		ctl_set_invalid_field(ctsio,
5369				      /*sks_valid*/ 1,
5370				      /*command*/ 1,
5371				      /*field*/ 4,
5372				      /*bit_valid*/ 1,
5373				      /*bit*/ 4);
5374		ctl_done((union ctl_io *)ctsio);
5375		return (CTL_RETVAL_COMPLETE);
5376	}
5377
5378	/*
5379	 * Media isn't removable, so we can't load or eject it.
5380	 */
5381	if ((cdb->how & SSS_LOEJ) != 0) {
5382		ctl_set_invalid_field(ctsio,
5383				      /*sks_valid*/ 1,
5384				      /*command*/ 1,
5385				      /*field*/ 4,
5386				      /*bit_valid*/ 1,
5387				      /*bit*/ 1);
5388		ctl_done((union ctl_io *)ctsio);
5389		return (CTL_RETVAL_COMPLETE);
5390	}
5391
5392	if ((lun->flags & CTL_LUN_PR_RESERVED)
5393	 && ((cdb->how & SSS_START)==0)) {
5394		uint32_t residx;
5395
5396		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5397		if (!lun->per_res[residx].registered
5398		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5399
5400			ctl_set_reservation_conflict(ctsio);
5401			ctl_done((union ctl_io *)ctsio);
5402			return (CTL_RETVAL_COMPLETE);
5403		}
5404	}
5405
5406	/*
5407	 * If there is no backend on this device, we can't start or stop
5408	 * it.  In theory we shouldn't get any start/stop commands in the
5409	 * first place at this level if the LUN doesn't have a backend.
5410	 * That should get stopped by the command decode code.
5411	 */
5412	if (lun->backend == NULL) {
5413		ctl_set_invalid_opcode(ctsio);
5414		ctl_done((union ctl_io *)ctsio);
5415		return (CTL_RETVAL_COMPLETE);
5416	}
5417
5418	/*
5419	 * XXX KDM Copan-specific offline behavior.
5420	 * Figure out a reasonable way to port this?
5421	 */
5422#ifdef NEEDTOPORT
5423	mtx_lock(&ctl_softc->ctl_lock);
5424
5425	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5426	 && (lun->flags & CTL_LUN_OFFLINE)) {
5427		/*
5428		 * If the LUN is offline, and the on/offline bit isn't set,
5429		 * reject the start or stop.  Otherwise, let it through.
5430		 */
5431		mtx_unlock(&ctl_softc->ctl_lock);
5432		ctl_set_lun_not_ready(ctsio);
5433		ctl_done((union ctl_io *)ctsio);
5434	} else {
5435		mtx_unlock(&ctl_softc->ctl_lock);
5436#endif /* NEEDTOPORT */
5437		/*
5438		 * This could be a start or a stop when we're online,
5439		 * or a stop/offline or start/online.  A start or stop when
5440		 * we're offline is covered in the case above.
5441		 */
5442		/*
5443		 * In the non-immediate case, we send the request to
5444		 * the backend and return status to the user when
5445		 * it is done.
5446		 *
5447		 * In the immediate case, we allocate a new ctl_io
5448		 * to hold a copy of the request, and send that to
5449		 * the backend.  We then set good status on the
5450		 * user's request and return it immediately.
5451		 */
5452		if (cdb->byte2 & SSS_IMMED) {
5453			union ctl_io *new_io;
5454
5455			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5456			if (new_io == NULL) {
5457				ctl_set_busy(ctsio);
5458				ctl_done((union ctl_io *)ctsio);
5459			} else {
5460				ctl_copy_io((union ctl_io *)ctsio,
5461					    new_io);
5462				retval = lun->backend->config_write(new_io);
5463				ctl_set_success(ctsio);
5464				ctl_done((union ctl_io *)ctsio);
5465			}
5466		} else {
5467			retval = lun->backend->config_write(
5468				(union ctl_io *)ctsio);
5469		}
5470#ifdef NEEDTOPORT
5471	}
5472#endif
5473	return (retval);
5474}
5475
5476/*
5477 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5478 * we don't really do anything with the LBA and length fields if the user
5479 * passes them in.  Instead we'll just flush out the cache for the entire
5480 * LUN.
5481 */
5482int
5483ctl_sync_cache(struct ctl_scsiio *ctsio)
5484{
5485	struct ctl_lun *lun;
5486	struct ctl_softc *ctl_softc;
5487	uint64_t starting_lba;
5488	uint32_t block_count;
5489	int reladr, immed;
5490	int retval;
5491
5492	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5493
5494	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5495	ctl_softc = control_softc;
5496	retval = 0;
5497	reladr = 0;
5498	immed = 0;
5499
5500	switch (ctsio->cdb[0]) {
5501	case SYNCHRONIZE_CACHE: {
5502		struct scsi_sync_cache *cdb;
5503		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5504
5505		if (cdb->byte2 & SSC_RELADR)
5506			reladr = 1;
5507
5508		if (cdb->byte2 & SSC_IMMED)
5509			immed = 1;
5510
5511		starting_lba = scsi_4btoul(cdb->begin_lba);
5512		block_count = scsi_2btoul(cdb->lb_count);
5513		break;
5514	}
5515	case SYNCHRONIZE_CACHE_16: {
5516		struct scsi_sync_cache_16 *cdb;
5517		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5518
5519		if (cdb->byte2 & SSC_RELADR)
5520			reladr = 1;
5521
5522		if (cdb->byte2 & SSC_IMMED)
5523			immed = 1;
5524
5525		starting_lba = scsi_8btou64(cdb->begin_lba);
5526		block_count = scsi_4btoul(cdb->lb_count);
5527		break;
5528	}
5529	default:
5530		ctl_set_invalid_opcode(ctsio);
5531		ctl_done((union ctl_io *)ctsio);
5532		goto bailout;
5533		break; /* NOTREACHED */
5534	}
5535
5536	if (immed) {
5537		/*
5538		 * We don't support the immediate bit.  Since it's in the
5539		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5540		 * commands, we can just return the same error in either
5541		 * case.
5542		 */
5543		ctl_set_invalid_field(ctsio,
5544				      /*sks_valid*/ 1,
5545				      /*command*/ 1,
5546				      /*field*/ 1,
5547				      /*bit_valid*/ 1,
5548				      /*bit*/ 1);
5549		ctl_done((union ctl_io *)ctsio);
5550		goto bailout;
5551	}
5552
5553	if (reladr) {
5554		/*
5555		 * We don't support the reladr bit either.  It can only be
5556		 * used with linked commands, and we don't support linked
5557		 * commands.  Since the bit is in the same place for the
5558		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5559		 * just return the same error in either case.
5560		 */
5561		ctl_set_invalid_field(ctsio,
5562				      /*sks_valid*/ 1,
5563				      /*command*/ 1,
5564				      /*field*/ 1,
5565				      /*bit_valid*/ 1,
5566				      /*bit*/ 0);
5567		ctl_done((union ctl_io *)ctsio);
5568		goto bailout;
5569	}
5570
5571	/*
5572	 * We check the LBA and length, but don't do anything with them.
5573	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5574	 * get flushed.  This check will just help satisfy anyone who wants
5575	 * to see an error for an out of range LBA.
5576	 */
5577	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5578		ctl_set_lba_out_of_range(ctsio);
5579		ctl_done((union ctl_io *)ctsio);
5580		goto bailout;
5581	}
5582
5583	/*
5584	 * If this LUN has no backend, we can't flush the cache anyway.
5585	 */
5586	if (lun->backend == NULL) {
5587		ctl_set_invalid_opcode(ctsio);
5588		ctl_done((union ctl_io *)ctsio);
5589		goto bailout;
5590	}
5591
5592	/*
5593	 * Check to see whether we're configured to send the SYNCHRONIZE
5594	 * CACHE command directly to the back end.
5595	 */
5596	mtx_lock(&ctl_softc->ctl_lock);
5597	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5598	 && (++(lun->sync_count) >= lun->sync_interval)) {
5599		lun->sync_count = 0;
5600		mtx_unlock(&ctl_softc->ctl_lock);
5601		retval = lun->backend->config_write((union ctl_io *)ctsio);
5602	} else {
5603		mtx_unlock(&ctl_softc->ctl_lock);
5604		ctl_set_success(ctsio);
5605		ctl_done((union ctl_io *)ctsio);
5606	}
5607
5608bailout:
5609
5610	return (retval);
5611}
5612
5613int
5614ctl_format(struct ctl_scsiio *ctsio)
5615{
5616	struct scsi_format *cdb;
5617	struct ctl_lun *lun;
5618	struct ctl_softc *ctl_softc;
5619	int length, defect_list_len;
5620
5621	CTL_DEBUG_PRINT(("ctl_format\n"));
5622
5623	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5624	ctl_softc = control_softc;
5625
5626	cdb = (struct scsi_format *)ctsio->cdb;
5627
5628	length = 0;
5629	if (cdb->byte2 & SF_FMTDATA) {
5630		if (cdb->byte2 & SF_LONGLIST)
5631			length = sizeof(struct scsi_format_header_long);
5632		else
5633			length = sizeof(struct scsi_format_header_short);
5634	}
5635
5636	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5637	 && (length > 0)) {
5638		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5639		ctsio->kern_data_len = length;
5640		ctsio->kern_total_len = length;
5641		ctsio->kern_data_resid = 0;
5642		ctsio->kern_rel_offset = 0;
5643		ctsio->kern_sg_entries = 0;
5644		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5645		ctsio->be_move_done = ctl_config_move_done;
5646		ctl_datamove((union ctl_io *)ctsio);
5647
5648		return (CTL_RETVAL_COMPLETE);
5649	}
5650
5651	defect_list_len = 0;
5652
5653	if (cdb->byte2 & SF_FMTDATA) {
5654		if (cdb->byte2 & SF_LONGLIST) {
5655			struct scsi_format_header_long *header;
5656
5657			header = (struct scsi_format_header_long *)
5658				ctsio->kern_data_ptr;
5659
5660			defect_list_len = scsi_4btoul(header->defect_list_len);
5661			if (defect_list_len != 0) {
5662				ctl_set_invalid_field(ctsio,
5663						      /*sks_valid*/ 1,
5664						      /*command*/ 0,
5665						      /*field*/ 2,
5666						      /*bit_valid*/ 0,
5667						      /*bit*/ 0);
5668				goto bailout;
5669			}
5670		} else {
5671			struct scsi_format_header_short *header;
5672
5673			header = (struct scsi_format_header_short *)
5674				ctsio->kern_data_ptr;
5675
5676			defect_list_len = scsi_2btoul(header->defect_list_len);
5677			if (defect_list_len != 0) {
5678				ctl_set_invalid_field(ctsio,
5679						      /*sks_valid*/ 1,
5680						      /*command*/ 0,
5681						      /*field*/ 2,
5682						      /*bit_valid*/ 0,
5683						      /*bit*/ 0);
5684				goto bailout;
5685			}
5686		}
5687	}
5688
5689	/*
5690	 * The format command will clear out the "Medium format corrupted"
5691	 * status if set by the configuration code.  That status is really
5692	 * just a way to notify the host that we have lost the media, and
5693	 * get them to issue a command that will basically make them think
5694	 * they're blowing away the media.
5695	 */
5696	mtx_lock(&ctl_softc->ctl_lock);
5697	lun->flags &= ~CTL_LUN_INOPERABLE;
5698	mtx_unlock(&ctl_softc->ctl_lock);
5699
5700	ctsio->scsi_status = SCSI_STATUS_OK;
5701	ctsio->io_hdr.status = CTL_SUCCESS;
5702bailout:
5703
5704	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5705		free(ctsio->kern_data_ptr, M_CTL);
5706		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5707	}
5708
5709	ctl_done((union ctl_io *)ctsio);
5710	return (CTL_RETVAL_COMPLETE);
5711}
5712
5713int
5714ctl_write_buffer(struct ctl_scsiio *ctsio)
5715{
5716	struct scsi_write_buffer *cdb;
5717	struct copan_page_header *header;
5718	struct ctl_lun *lun;
5719	struct ctl_softc *ctl_softc;
5720	int buffer_offset, len;
5721	int retval;
5722
5723	header = NULL;
5724
5725	retval = CTL_RETVAL_COMPLETE;
5726
5727	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5728
5729	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5730	ctl_softc = control_softc;
5731	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5732
5733	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5734		ctl_set_invalid_field(ctsio,
5735				      /*sks_valid*/ 1,
5736				      /*command*/ 1,
5737				      /*field*/ 1,
5738				      /*bit_valid*/ 1,
5739				      /*bit*/ 4);
5740		ctl_done((union ctl_io *)ctsio);
5741		return (CTL_RETVAL_COMPLETE);
5742	}
5743	if (cdb->buffer_id != 0) {
5744		ctl_set_invalid_field(ctsio,
5745				      /*sks_valid*/ 1,
5746				      /*command*/ 1,
5747				      /*field*/ 2,
5748				      /*bit_valid*/ 0,
5749				      /*bit*/ 0);
5750		ctl_done((union ctl_io *)ctsio);
5751		return (CTL_RETVAL_COMPLETE);
5752	}
5753
5754	len = scsi_3btoul(cdb->length);
5755	buffer_offset = scsi_3btoul(cdb->offset);
5756
5757	if (len > sizeof(lun->write_buffer)) {
5758		ctl_set_invalid_field(ctsio,
5759				      /*sks_valid*/ 1,
5760				      /*command*/ 1,
5761				      /*field*/ 6,
5762				      /*bit_valid*/ 0,
5763				      /*bit*/ 0);
5764		ctl_done((union ctl_io *)ctsio);
5765		return (CTL_RETVAL_COMPLETE);
5766	}
5767
5768	if (buffer_offset != 0) {
5769		ctl_set_invalid_field(ctsio,
5770				      /*sks_valid*/ 1,
5771				      /*command*/ 1,
5772				      /*field*/ 3,
5773				      /*bit_valid*/ 0,
5774				      /*bit*/ 0);
5775		ctl_done((union ctl_io *)ctsio);
5776		return (CTL_RETVAL_COMPLETE);
5777	}
5778
5779	/*
5780	 * If we've got a kernel request that hasn't been malloced yet,
5781	 * malloc it and tell the caller the data buffer is here.
5782	 */
5783	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5784		ctsio->kern_data_ptr = lun->write_buffer;
5785		ctsio->kern_data_len = len;
5786		ctsio->kern_total_len = len;
5787		ctsio->kern_data_resid = 0;
5788		ctsio->kern_rel_offset = 0;
5789		ctsio->kern_sg_entries = 0;
5790		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5791		ctsio->be_move_done = ctl_config_move_done;
5792		ctl_datamove((union ctl_io *)ctsio);
5793
5794		return (CTL_RETVAL_COMPLETE);
5795	}
5796
5797	ctl_done((union ctl_io *)ctsio);
5798
5799	return (CTL_RETVAL_COMPLETE);
5800}
5801
5802int
5803ctl_write_same(struct ctl_scsiio *ctsio)
5804{
5805	struct ctl_lun *lun;
5806	struct ctl_lba_len_flags *lbalen;
5807	uint64_t lba;
5808	uint32_t num_blocks;
5809	int len, retval;
5810	uint8_t byte2;
5811
5812	retval = CTL_RETVAL_COMPLETE;
5813
5814	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5815
5816	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5817
5818	switch (ctsio->cdb[0]) {
5819	case WRITE_SAME_10: {
5820		struct scsi_write_same_10 *cdb;
5821
5822		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5823
5824		lba = scsi_4btoul(cdb->addr);
5825		num_blocks = scsi_2btoul(cdb->length);
5826		byte2 = cdb->byte2;
5827		break;
5828	}
5829	case WRITE_SAME_16: {
5830		struct scsi_write_same_16 *cdb;
5831
5832		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5833
5834		lba = scsi_8btou64(cdb->addr);
5835		num_blocks = scsi_4btoul(cdb->length);
5836		byte2 = cdb->byte2;
5837		break;
5838	}
5839	default:
5840		/*
5841		 * We got a command we don't support.  This shouldn't
5842		 * happen, commands should be filtered out above us.
5843		 */
5844		ctl_set_invalid_opcode(ctsio);
5845		ctl_done((union ctl_io *)ctsio);
5846
5847		return (CTL_RETVAL_COMPLETE);
5848		break; /* NOTREACHED */
5849	}
5850
5851	/*
5852	 * The first check is to make sure we're in bounds, the second
5853	 * check is to catch wrap-around problems.  If the lba + num blocks
5854	 * is less than the lba, then we've wrapped around and the block
5855	 * range is invalid anyway.
5856	 */
5857	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5858	 || ((lba + num_blocks) < lba)) {
5859		ctl_set_lba_out_of_range(ctsio);
5860		ctl_done((union ctl_io *)ctsio);
5861		return (CTL_RETVAL_COMPLETE);
5862	}
5863
5864	/* Zero number of blocks means "to the last logical block" */
5865	if (num_blocks == 0) {
5866		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5867			ctl_set_invalid_field(ctsio,
5868					      /*sks_valid*/ 0,
5869					      /*command*/ 1,
5870					      /*field*/ 0,
5871					      /*bit_valid*/ 0,
5872					      /*bit*/ 0);
5873			ctl_done((union ctl_io *)ctsio);
5874			return (CTL_RETVAL_COMPLETE);
5875		}
5876		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5877	}
5878
5879	len = lun->be_lun->blocksize;
5880
5881	/*
5882	 * If we've got a kernel request that hasn't been malloced yet,
5883	 * malloc it and tell the caller the data buffer is here.
5884	 */
5885	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5886		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5887		ctsio->kern_data_len = len;
5888		ctsio->kern_total_len = len;
5889		ctsio->kern_data_resid = 0;
5890		ctsio->kern_rel_offset = 0;
5891		ctsio->kern_sg_entries = 0;
5892		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5893		ctsio->be_move_done = ctl_config_move_done;
5894		ctl_datamove((union ctl_io *)ctsio);
5895
5896		return (CTL_RETVAL_COMPLETE);
5897	}
5898
5899	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5900	lbalen->lba = lba;
5901	lbalen->len = num_blocks;
5902	lbalen->flags = byte2;
5903	retval = lun->backend->config_write((union ctl_io *)ctsio);
5904
5905	return (retval);
5906}
5907
5908int
5909ctl_unmap(struct ctl_scsiio *ctsio)
5910{
5911	struct ctl_lun *lun;
5912	struct scsi_unmap *cdb;
5913	struct ctl_ptr_len_flags *ptrlen;
5914	struct scsi_unmap_header *hdr;
5915	struct scsi_unmap_desc *buf, *end;
5916	uint64_t lba;
5917	uint32_t num_blocks;
5918	int len, retval;
5919	uint8_t byte2;
5920
5921	retval = CTL_RETVAL_COMPLETE;
5922
5923	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5924
5925	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5926	cdb = (struct scsi_unmap *)ctsio->cdb;
5927
5928	len = scsi_2btoul(cdb->length);
5929	byte2 = cdb->byte2;
5930
5931	/*
5932	 * If we've got a kernel request that hasn't been malloced yet,
5933	 * malloc it and tell the caller the data buffer is here.
5934	 */
5935	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5936		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5937		ctsio->kern_data_len = len;
5938		ctsio->kern_total_len = len;
5939		ctsio->kern_data_resid = 0;
5940		ctsio->kern_rel_offset = 0;
5941		ctsio->kern_sg_entries = 0;
5942		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5943		ctsio->be_move_done = ctl_config_move_done;
5944		ctl_datamove((union ctl_io *)ctsio);
5945
5946		return (CTL_RETVAL_COMPLETE);
5947	}
5948
5949	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5950	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5951	if (len < sizeof (*hdr) ||
5952	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5953	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5954	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5955		ctl_set_invalid_field(ctsio,
5956				      /*sks_valid*/ 0,
5957				      /*command*/ 0,
5958				      /*field*/ 0,
5959				      /*bit_valid*/ 0,
5960				      /*bit*/ 0);
5961		ctl_done((union ctl_io *)ctsio);
5962		return (CTL_RETVAL_COMPLETE);
5963	}
5964	len = scsi_2btoul(hdr->desc_length);
5965	buf = (struct scsi_unmap_desc *)(hdr + 1);
5966	end = buf + len / sizeof(*buf);
5967
5968	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5969	ptrlen->ptr = (void *)buf;
5970	ptrlen->len = len;
5971	ptrlen->flags = byte2;
5972
5973	for (; buf < end; buf++) {
5974		lba = scsi_8btou64(buf->lba);
5975		num_blocks = scsi_4btoul(buf->length);
5976		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5977		 || ((lba + num_blocks) < lba)) {
5978			ctl_set_lba_out_of_range(ctsio);
5979			ctl_done((union ctl_io *)ctsio);
5980			return (CTL_RETVAL_COMPLETE);
5981		}
5982	}
5983
5984	retval = lun->backend->config_write((union ctl_io *)ctsio);
5985
5986	return (retval);
5987}
5988
5989/*
5990 * Note that this function currently doesn't actually do anything inside
5991 * CTL to enforce things if the DQue bit is turned on.
5992 *
5993 * Also note that this function can't be used in the default case, because
5994 * the DQue bit isn't set in the changeable mask for the control mode page
5995 * anyway.  This is just here as an example for how to implement a page
5996 * handler, and a placeholder in case we want to allow the user to turn
5997 * tagged queueing on and off.
5998 *
5999 * The D_SENSE bit handling is functional, however, and will turn
6000 * descriptor sense on and off for a given LUN.
6001 */
6002int
6003ctl_control_page_handler(struct ctl_scsiio *ctsio,
6004			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6005{
6006	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6007	struct ctl_lun *lun;
6008	struct ctl_softc *softc;
6009	int set_ua;
6010	uint32_t initidx;
6011
6012	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6013	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6014	set_ua = 0;
6015
6016	user_cp = (struct scsi_control_page *)page_ptr;
6017	current_cp = (struct scsi_control_page *)
6018		(page_index->page_data + (page_index->page_len *
6019		CTL_PAGE_CURRENT));
6020	saved_cp = (struct scsi_control_page *)
6021		(page_index->page_data + (page_index->page_len *
6022		CTL_PAGE_SAVED));
6023
6024	softc = control_softc;
6025
6026	mtx_lock(&softc->ctl_lock);
6027	if (((current_cp->rlec & SCP_DSENSE) == 0)
6028	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6029		/*
6030		 * Descriptor sense is currently turned off and the user
6031		 * wants to turn it on.
6032		 */
6033		current_cp->rlec |= SCP_DSENSE;
6034		saved_cp->rlec |= SCP_DSENSE;
6035		lun->flags |= CTL_LUN_SENSE_DESC;
6036		set_ua = 1;
6037	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6038		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6039		/*
6040		 * Descriptor sense is currently turned on, and the user
6041		 * wants to turn it off.
6042		 */
6043		current_cp->rlec &= ~SCP_DSENSE;
6044		saved_cp->rlec &= ~SCP_DSENSE;
6045		lun->flags &= ~CTL_LUN_SENSE_DESC;
6046		set_ua = 1;
6047	}
6048	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6049		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6050#ifdef NEEDTOPORT
6051			csevent_log(CSC_CTL | CSC_SHELF_SW |
6052				    CTL_UNTAG_TO_UNTAG,
6053				    csevent_LogType_Trace,
6054				    csevent_Severity_Information,
6055				    csevent_AlertLevel_Green,
6056				    csevent_FRU_Firmware,
6057				    csevent_FRU_Unknown,
6058				    "Received untagged to untagged transition");
6059#endif /* NEEDTOPORT */
6060		} else {
6061#ifdef NEEDTOPORT
6062			csevent_log(CSC_CTL | CSC_SHELF_SW |
6063				    CTL_UNTAG_TO_TAG,
6064				    csevent_LogType_ConfigChange,
6065				    csevent_Severity_Information,
6066				    csevent_AlertLevel_Green,
6067				    csevent_FRU_Firmware,
6068				    csevent_FRU_Unknown,
6069				    "Received untagged to tagged "
6070				    "queueing transition");
6071#endif /* NEEDTOPORT */
6072
6073			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6074			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6075			set_ua = 1;
6076		}
6077	} else {
6078		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6079#ifdef NEEDTOPORT
6080			csevent_log(CSC_CTL | CSC_SHELF_SW |
6081				    CTL_TAG_TO_UNTAG,
6082				    csevent_LogType_ConfigChange,
6083				    csevent_Severity_Warning,
6084				    csevent_AlertLevel_Yellow,
6085				    csevent_FRU_Firmware,
6086				    csevent_FRU_Unknown,
6087				    "Received tagged queueing to untagged "
6088				    "transition");
6089#endif /* NEEDTOPORT */
6090
6091			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6092			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6093			set_ua = 1;
6094		} else {
6095#ifdef NEEDTOPORT
6096			csevent_log(CSC_CTL | CSC_SHELF_SW |
6097				    CTL_TAG_TO_TAG,
6098				    csevent_LogType_Trace,
6099				    csevent_Severity_Information,
6100				    csevent_AlertLevel_Green,
6101				    csevent_FRU_Firmware,
6102				    csevent_FRU_Unknown,
6103				    "Received tagged queueing to tagged "
6104				    "queueing transition");
6105#endif /* NEEDTOPORT */
6106		}
6107	}
6108	if (set_ua != 0) {
6109		int i;
6110		/*
6111		 * Let other initiators know that the mode
6112		 * parameters for this LUN have changed.
6113		 */
6114		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6115			if (i == initidx)
6116				continue;
6117
6118			lun->pending_sense[i].ua_pending |=
6119				CTL_UA_MODE_CHANGE;
6120		}
6121	}
6122	mtx_unlock(&softc->ctl_lock);
6123
6124	return (0);
6125}
6126
6127int
6128ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6129		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6130{
6131	return (0);
6132}
6133
6134int
6135ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6136			   struct ctl_page_index *page_index, int pc)
6137{
6138	struct copan_power_subpage *page;
6139
6140	page = (struct copan_power_subpage *)page_index->page_data +
6141		(page_index->page_len * pc);
6142
6143	switch (pc) {
6144	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6145		/*
6146		 * We don't update the changable bits for this page.
6147		 */
6148		break;
6149	case SMS_PAGE_CTRL_CURRENT >> 6:
6150	case SMS_PAGE_CTRL_DEFAULT >> 6:
6151	case SMS_PAGE_CTRL_SAVED >> 6:
6152#ifdef NEEDTOPORT
6153		ctl_update_power_subpage(page);
6154#endif
6155		break;
6156	default:
6157#ifdef NEEDTOPORT
6158		EPRINT(0, "Invalid PC %d!!", pc);
6159#endif
6160		break;
6161	}
6162	return (0);
6163}
6164
6165
6166int
6167ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6168		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6169{
6170	struct copan_aps_subpage *user_sp;
6171	struct copan_aps_subpage *current_sp;
6172	union ctl_modepage_info *modepage_info;
6173	struct ctl_softc *softc;
6174	struct ctl_lun *lun;
6175	int retval;
6176
6177	retval = CTL_RETVAL_COMPLETE;
6178	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6179		     (page_index->page_len * CTL_PAGE_CURRENT));
6180	softc = control_softc;
6181	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6182
6183	user_sp = (struct copan_aps_subpage *)page_ptr;
6184
6185	modepage_info = (union ctl_modepage_info *)
6186		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6187
6188	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6189	modepage_info->header.subpage = page_index->subpage;
6190	modepage_info->aps.lock_active = user_sp->lock_active;
6191
6192	mtx_lock(&softc->ctl_lock);
6193
6194	/*
6195	 * If there is a request to lock the LUN and another LUN is locked
6196	 * this is an error. If the requested LUN is already locked ignore
6197	 * the request. If no LUN is locked attempt to lock it.
6198	 * if there is a request to unlock the LUN and the LUN is currently
6199	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6200	 * if another LUN is locked or no LUN is locked.
6201	 */
6202	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6203		if (softc->aps_locked_lun == lun->lun) {
6204			/*
6205			 * This LUN is already locked, so we're done.
6206			 */
6207			retval = CTL_RETVAL_COMPLETE;
6208		} else if (softc->aps_locked_lun == 0) {
6209			/*
6210			 * No one has the lock, pass the request to the
6211			 * backend.
6212			 */
6213			retval = lun->backend->config_write(
6214				(union ctl_io *)ctsio);
6215		} else {
6216			/*
6217			 * Someone else has the lock, throw out the request.
6218			 */
6219			ctl_set_already_locked(ctsio);
6220			free(ctsio->kern_data_ptr, M_CTL);
6221			ctl_done((union ctl_io *)ctsio);
6222
6223			/*
6224			 * Set the return value so that ctl_do_mode_select()
6225			 * won't try to complete the command.  We already
6226			 * completed it here.
6227			 */
6228			retval = CTL_RETVAL_ERROR;
6229		}
6230	} else if (softc->aps_locked_lun == lun->lun) {
6231		/*
6232		 * This LUN is locked, so pass the unlock request to the
6233		 * backend.
6234		 */
6235		retval = lun->backend->config_write((union ctl_io *)ctsio);
6236	}
6237	mtx_unlock(&softc->ctl_lock);
6238
6239	return (retval);
6240}
6241
6242int
6243ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6244				struct ctl_page_index *page_index,
6245				uint8_t *page_ptr)
6246{
6247	uint8_t *c;
6248	int i;
6249
6250	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6251	ctl_time_io_secs =
6252		(c[0] << 8) |
6253		(c[1] << 0) |
6254		0;
6255	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6256	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6257	printf("page data:");
6258	for (i=0; i<8; i++)
6259		printf(" %.2x",page_ptr[i]);
6260	printf("\n");
6261	return (0);
6262}
6263
6264int
6265ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6266			       struct ctl_page_index *page_index,
6267			       int pc)
6268{
6269	struct copan_debugconf_subpage *page;
6270
6271	page = (struct copan_debugconf_subpage *)page_index->page_data +
6272		(page_index->page_len * pc);
6273
6274	switch (pc) {
6275	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6276	case SMS_PAGE_CTRL_DEFAULT >> 6:
6277	case SMS_PAGE_CTRL_SAVED >> 6:
6278		/*
6279		 * We don't update the changable or default bits for this page.
6280		 */
6281		break;
6282	case SMS_PAGE_CTRL_CURRENT >> 6:
6283		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6284		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6285		break;
6286	default:
6287#ifdef NEEDTOPORT
6288		EPRINT(0, "Invalid PC %d!!", pc);
6289#endif /* NEEDTOPORT */
6290		break;
6291	}
6292	return (0);
6293}
6294
6295
6296static int
6297ctl_do_mode_select(union ctl_io *io)
6298{
6299	struct scsi_mode_page_header *page_header;
6300	struct ctl_page_index *page_index;
6301	struct ctl_scsiio *ctsio;
6302	int control_dev, page_len;
6303	int page_len_offset, page_len_size;
6304	union ctl_modepage_info *modepage_info;
6305	struct ctl_lun *lun;
6306	int *len_left, *len_used;
6307	int retval, i;
6308
6309	ctsio = &io->scsiio;
6310	page_index = NULL;
6311	page_len = 0;
6312	retval = CTL_RETVAL_COMPLETE;
6313
6314	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6315
6316	if (lun->be_lun->lun_type != T_DIRECT)
6317		control_dev = 1;
6318	else
6319		control_dev = 0;
6320
6321	modepage_info = (union ctl_modepage_info *)
6322		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6323	len_left = &modepage_info->header.len_left;
6324	len_used = &modepage_info->header.len_used;
6325
6326do_next_page:
6327
6328	page_header = (struct scsi_mode_page_header *)
6329		(ctsio->kern_data_ptr + *len_used);
6330
6331	if (*len_left == 0) {
6332		free(ctsio->kern_data_ptr, M_CTL);
6333		ctl_set_success(ctsio);
6334		ctl_done((union ctl_io *)ctsio);
6335		return (CTL_RETVAL_COMPLETE);
6336	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6337
6338		free(ctsio->kern_data_ptr, M_CTL);
6339		ctl_set_param_len_error(ctsio);
6340		ctl_done((union ctl_io *)ctsio);
6341		return (CTL_RETVAL_COMPLETE);
6342
6343	} else if ((page_header->page_code & SMPH_SPF)
6344		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6345
6346		free(ctsio->kern_data_ptr, M_CTL);
6347		ctl_set_param_len_error(ctsio);
6348		ctl_done((union ctl_io *)ctsio);
6349		return (CTL_RETVAL_COMPLETE);
6350	}
6351
6352
6353	/*
6354	 * XXX KDM should we do something with the block descriptor?
6355	 */
6356	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6357
6358		if ((control_dev != 0)
6359		 && (lun->mode_pages.index[i].page_flags &
6360		     CTL_PAGE_FLAG_DISK_ONLY))
6361			continue;
6362
6363		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6364		    (page_header->page_code & SMPH_PC_MASK))
6365			continue;
6366
6367		/*
6368		 * If neither page has a subpage code, then we've got a
6369		 * match.
6370		 */
6371		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6372		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6373			page_index = &lun->mode_pages.index[i];
6374			page_len = page_header->page_length;
6375			break;
6376		}
6377
6378		/*
6379		 * If both pages have subpages, then the subpage numbers
6380		 * have to match.
6381		 */
6382		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6383		  && (page_header->page_code & SMPH_SPF)) {
6384			struct scsi_mode_page_header_sp *sph;
6385
6386			sph = (struct scsi_mode_page_header_sp *)page_header;
6387
6388			if (lun->mode_pages.index[i].subpage ==
6389			    sph->subpage) {
6390				page_index = &lun->mode_pages.index[i];
6391				page_len = scsi_2btoul(sph->page_length);
6392				break;
6393			}
6394		}
6395	}
6396
6397	/*
6398	 * If we couldn't find the page, or if we don't have a mode select
6399	 * handler for it, send back an error to the user.
6400	 */
6401	if ((page_index == NULL)
6402	 || (page_index->select_handler == NULL)) {
6403		ctl_set_invalid_field(ctsio,
6404				      /*sks_valid*/ 1,
6405				      /*command*/ 0,
6406				      /*field*/ *len_used,
6407				      /*bit_valid*/ 0,
6408				      /*bit*/ 0);
6409		free(ctsio->kern_data_ptr, M_CTL);
6410		ctl_done((union ctl_io *)ctsio);
6411		return (CTL_RETVAL_COMPLETE);
6412	}
6413
6414	if (page_index->page_code & SMPH_SPF) {
6415		page_len_offset = 2;
6416		page_len_size = 2;
6417	} else {
6418		page_len_size = 1;
6419		page_len_offset = 1;
6420	}
6421
6422	/*
6423	 * If the length the initiator gives us isn't the one we specify in
6424	 * the mode page header, or if they didn't specify enough data in
6425	 * the CDB to avoid truncating this page, kick out the request.
6426	 */
6427	if ((page_len != (page_index->page_len - page_len_offset -
6428			  page_len_size))
6429	 || (*len_left < page_index->page_len)) {
6430
6431
6432		ctl_set_invalid_field(ctsio,
6433				      /*sks_valid*/ 1,
6434				      /*command*/ 0,
6435				      /*field*/ *len_used + page_len_offset,
6436				      /*bit_valid*/ 0,
6437				      /*bit*/ 0);
6438		free(ctsio->kern_data_ptr, M_CTL);
6439		ctl_done((union ctl_io *)ctsio);
6440		return (CTL_RETVAL_COMPLETE);
6441	}
6442
6443	/*
6444	 * Run through the mode page, checking to make sure that the bits
6445	 * the user changed are actually legal for him to change.
6446	 */
6447	for (i = 0; i < page_index->page_len; i++) {
6448		uint8_t *user_byte, *change_mask, *current_byte;
6449		int bad_bit;
6450		int j;
6451
6452		user_byte = (uint8_t *)page_header + i;
6453		change_mask = page_index->page_data +
6454			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6455		current_byte = page_index->page_data +
6456			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6457
6458		/*
6459		 * Check to see whether the user set any bits in this byte
6460		 * that he is not allowed to set.
6461		 */
6462		if ((*user_byte & ~(*change_mask)) ==
6463		    (*current_byte & ~(*change_mask)))
6464			continue;
6465
6466		/*
6467		 * Go through bit by bit to determine which one is illegal.
6468		 */
6469		bad_bit = 0;
6470		for (j = 7; j >= 0; j--) {
6471			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6472			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6473				bad_bit = i;
6474				break;
6475			}
6476		}
6477		ctl_set_invalid_field(ctsio,
6478				      /*sks_valid*/ 1,
6479				      /*command*/ 0,
6480				      /*field*/ *len_used + i,
6481				      /*bit_valid*/ 1,
6482				      /*bit*/ bad_bit);
6483		free(ctsio->kern_data_ptr, M_CTL);
6484		ctl_done((union ctl_io *)ctsio);
6485		return (CTL_RETVAL_COMPLETE);
6486	}
6487
6488	/*
6489	 * Decrement these before we call the page handler, since we may
6490	 * end up getting called back one way or another before the handler
6491	 * returns to this context.
6492	 */
6493	*len_left -= page_index->page_len;
6494	*len_used += page_index->page_len;
6495
6496	retval = page_index->select_handler(ctsio, page_index,
6497					    (uint8_t *)page_header);
6498
6499	/*
6500	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6501	 * wait until this queued command completes to finish processing
6502	 * the mode page.  If it returns anything other than
6503	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6504	 * already set the sense information, freed the data pointer, and
6505	 * completed the io for us.
6506	 */
6507	if (retval != CTL_RETVAL_COMPLETE)
6508		goto bailout_no_done;
6509
6510	/*
6511	 * If the initiator sent us more than one page, parse the next one.
6512	 */
6513	if (*len_left > 0)
6514		goto do_next_page;
6515
6516	ctl_set_success(ctsio);
6517	free(ctsio->kern_data_ptr, M_CTL);
6518	ctl_done((union ctl_io *)ctsio);
6519
6520bailout_no_done:
6521
6522	return (CTL_RETVAL_COMPLETE);
6523
6524}
6525
6526int
6527ctl_mode_select(struct ctl_scsiio *ctsio)
6528{
6529	int param_len, pf, sp;
6530	int header_size, bd_len;
6531	int len_left, len_used;
6532	struct ctl_page_index *page_index;
6533	struct ctl_lun *lun;
6534	int control_dev, page_len;
6535	union ctl_modepage_info *modepage_info;
6536	int retval;
6537
6538	pf = 0;
6539	sp = 0;
6540	page_len = 0;
6541	len_used = 0;
6542	len_left = 0;
6543	retval = 0;
6544	bd_len = 0;
6545	page_index = NULL;
6546
6547	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6548
6549	if (lun->be_lun->lun_type != T_DIRECT)
6550		control_dev = 1;
6551	else
6552		control_dev = 0;
6553
6554	switch (ctsio->cdb[0]) {
6555	case MODE_SELECT_6: {
6556		struct scsi_mode_select_6 *cdb;
6557
6558		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6559
6560		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6561		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6562
6563		param_len = cdb->length;
6564		header_size = sizeof(struct scsi_mode_header_6);
6565		break;
6566	}
6567	case MODE_SELECT_10: {
6568		struct scsi_mode_select_10 *cdb;
6569
6570		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6571
6572		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6573		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6574
6575		param_len = scsi_2btoul(cdb->length);
6576		header_size = sizeof(struct scsi_mode_header_10);
6577		break;
6578	}
6579	default:
6580		ctl_set_invalid_opcode(ctsio);
6581		ctl_done((union ctl_io *)ctsio);
6582		return (CTL_RETVAL_COMPLETE);
6583		break; /* NOTREACHED */
6584	}
6585
6586	/*
6587	 * From SPC-3:
6588	 * "A parameter list length of zero indicates that the Data-Out Buffer
6589	 * shall be empty. This condition shall not be considered as an error."
6590	 */
6591	if (param_len == 0) {
6592		ctl_set_success(ctsio);
6593		ctl_done((union ctl_io *)ctsio);
6594		return (CTL_RETVAL_COMPLETE);
6595	}
6596
6597	/*
6598	 * Since we'll hit this the first time through, prior to
6599	 * allocation, we don't need to free a data buffer here.
6600	 */
6601	if (param_len < header_size) {
6602		ctl_set_param_len_error(ctsio);
6603		ctl_done((union ctl_io *)ctsio);
6604		return (CTL_RETVAL_COMPLETE);
6605	}
6606
6607	/*
6608	 * Allocate the data buffer and grab the user's data.  In theory,
6609	 * we shouldn't have to sanity check the parameter list length here
6610	 * because the maximum size is 64K.  We should be able to malloc
6611	 * that much without too many problems.
6612	 */
6613	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6614		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6615		ctsio->kern_data_len = param_len;
6616		ctsio->kern_total_len = param_len;
6617		ctsio->kern_data_resid = 0;
6618		ctsio->kern_rel_offset = 0;
6619		ctsio->kern_sg_entries = 0;
6620		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6621		ctsio->be_move_done = ctl_config_move_done;
6622		ctl_datamove((union ctl_io *)ctsio);
6623
6624		return (CTL_RETVAL_COMPLETE);
6625	}
6626
6627	switch (ctsio->cdb[0]) {
6628	case MODE_SELECT_6: {
6629		struct scsi_mode_header_6 *mh6;
6630
6631		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6632		bd_len = mh6->blk_desc_len;
6633		break;
6634	}
6635	case MODE_SELECT_10: {
6636		struct scsi_mode_header_10 *mh10;
6637
6638		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6639		bd_len = scsi_2btoul(mh10->blk_desc_len);
6640		break;
6641	}
6642	default:
6643		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6644		break;
6645	}
6646
6647	if (param_len < (header_size + bd_len)) {
6648		free(ctsio->kern_data_ptr, M_CTL);
6649		ctl_set_param_len_error(ctsio);
6650		ctl_done((union ctl_io *)ctsio);
6651		return (CTL_RETVAL_COMPLETE);
6652	}
6653
6654	/*
6655	 * Set the IO_CONT flag, so that if this I/O gets passed to
6656	 * ctl_config_write_done(), it'll get passed back to
6657	 * ctl_do_mode_select() for further processing, or completion if
6658	 * we're all done.
6659	 */
6660	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6661	ctsio->io_cont = ctl_do_mode_select;
6662
6663	modepage_info = (union ctl_modepage_info *)
6664		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6665
6666	memset(modepage_info, 0, sizeof(*modepage_info));
6667
6668	len_left = param_len - header_size - bd_len;
6669	len_used = header_size + bd_len;
6670
6671	modepage_info->header.len_left = len_left;
6672	modepage_info->header.len_used = len_used;
6673
6674	return (ctl_do_mode_select((union ctl_io *)ctsio));
6675}
6676
6677int
6678ctl_mode_sense(struct ctl_scsiio *ctsio)
6679{
6680	struct ctl_lun *lun;
6681	int pc, page_code, dbd, llba, subpage;
6682	int alloc_len, page_len, header_len, total_len;
6683	struct scsi_mode_block_descr *block_desc;
6684	struct ctl_page_index *page_index;
6685	int control_dev;
6686
6687	dbd = 0;
6688	llba = 0;
6689	block_desc = NULL;
6690	page_index = NULL;
6691
6692	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6693
6694	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6695
6696	if (lun->be_lun->lun_type != T_DIRECT)
6697		control_dev = 1;
6698	else
6699		control_dev = 0;
6700
6701	switch (ctsio->cdb[0]) {
6702	case MODE_SENSE_6: {
6703		struct scsi_mode_sense_6 *cdb;
6704
6705		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6706
6707		header_len = sizeof(struct scsi_mode_hdr_6);
6708		if (cdb->byte2 & SMS_DBD)
6709			dbd = 1;
6710		else
6711			header_len += sizeof(struct scsi_mode_block_descr);
6712
6713		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6714		page_code = cdb->page & SMS_PAGE_CODE;
6715		subpage = cdb->subpage;
6716		alloc_len = cdb->length;
6717		break;
6718	}
6719	case MODE_SENSE_10: {
6720		struct scsi_mode_sense_10 *cdb;
6721
6722		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6723
6724		header_len = sizeof(struct scsi_mode_hdr_10);
6725
6726		if (cdb->byte2 & SMS_DBD)
6727			dbd = 1;
6728		else
6729			header_len += sizeof(struct scsi_mode_block_descr);
6730		if (cdb->byte2 & SMS10_LLBAA)
6731			llba = 1;
6732		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6733		page_code = cdb->page & SMS_PAGE_CODE;
6734		subpage = cdb->subpage;
6735		alloc_len = scsi_2btoul(cdb->length);
6736		break;
6737	}
6738	default:
6739		ctl_set_invalid_opcode(ctsio);
6740		ctl_done((union ctl_io *)ctsio);
6741		return (CTL_RETVAL_COMPLETE);
6742		break; /* NOTREACHED */
6743	}
6744
6745	/*
6746	 * We have to make a first pass through to calculate the size of
6747	 * the pages that match the user's query.  Then we allocate enough
6748	 * memory to hold it, and actually copy the data into the buffer.
6749	 */
6750	switch (page_code) {
6751	case SMS_ALL_PAGES_PAGE: {
6752		int i;
6753
6754		page_len = 0;
6755
6756		/*
6757		 * At the moment, values other than 0 and 0xff here are
6758		 * reserved according to SPC-3.
6759		 */
6760		if ((subpage != SMS_SUBPAGE_PAGE_0)
6761		 && (subpage != SMS_SUBPAGE_ALL)) {
6762			ctl_set_invalid_field(ctsio,
6763					      /*sks_valid*/ 1,
6764					      /*command*/ 1,
6765					      /*field*/ 3,
6766					      /*bit_valid*/ 0,
6767					      /*bit*/ 0);
6768			ctl_done((union ctl_io *)ctsio);
6769			return (CTL_RETVAL_COMPLETE);
6770		}
6771
6772		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6773			if ((control_dev != 0)
6774			 && (lun->mode_pages.index[i].page_flags &
6775			     CTL_PAGE_FLAG_DISK_ONLY))
6776				continue;
6777
6778			/*
6779			 * We don't use this subpage if the user didn't
6780			 * request all subpages.
6781			 */
6782			if ((lun->mode_pages.index[i].subpage != 0)
6783			 && (subpage == SMS_SUBPAGE_PAGE_0))
6784				continue;
6785
6786#if 0
6787			printf("found page %#x len %d\n",
6788			       lun->mode_pages.index[i].page_code &
6789			       SMPH_PC_MASK,
6790			       lun->mode_pages.index[i].page_len);
6791#endif
6792			page_len += lun->mode_pages.index[i].page_len;
6793		}
6794		break;
6795	}
6796	default: {
6797		int i;
6798
6799		page_len = 0;
6800
6801		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6802			/* Look for the right page code */
6803			if ((lun->mode_pages.index[i].page_code &
6804			     SMPH_PC_MASK) != page_code)
6805				continue;
6806
6807			/* Look for the right subpage or the subpage wildcard*/
6808			if ((lun->mode_pages.index[i].subpage != subpage)
6809			 && (subpage != SMS_SUBPAGE_ALL))
6810				continue;
6811
6812			/* Make sure the page is supported for this dev type */
6813			if ((control_dev != 0)
6814			 && (lun->mode_pages.index[i].page_flags &
6815			     CTL_PAGE_FLAG_DISK_ONLY))
6816				continue;
6817
6818#if 0
6819			printf("found page %#x len %d\n",
6820			       lun->mode_pages.index[i].page_code &
6821			       SMPH_PC_MASK,
6822			       lun->mode_pages.index[i].page_len);
6823#endif
6824
6825			page_len += lun->mode_pages.index[i].page_len;
6826		}
6827
6828		if (page_len == 0) {
6829			ctl_set_invalid_field(ctsio,
6830					      /*sks_valid*/ 1,
6831					      /*command*/ 1,
6832					      /*field*/ 2,
6833					      /*bit_valid*/ 1,
6834					      /*bit*/ 5);
6835			ctl_done((union ctl_io *)ctsio);
6836			return (CTL_RETVAL_COMPLETE);
6837		}
6838		break;
6839	}
6840	}
6841
6842	total_len = header_len + page_len;
6843#if 0
6844	printf("header_len = %d, page_len = %d, total_len = %d\n",
6845	       header_len, page_len, total_len);
6846#endif
6847
6848	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6849	ctsio->kern_sg_entries = 0;
6850	ctsio->kern_data_resid = 0;
6851	ctsio->kern_rel_offset = 0;
6852	if (total_len < alloc_len) {
6853		ctsio->residual = alloc_len - total_len;
6854		ctsio->kern_data_len = total_len;
6855		ctsio->kern_total_len = total_len;
6856	} else {
6857		ctsio->residual = 0;
6858		ctsio->kern_data_len = alloc_len;
6859		ctsio->kern_total_len = alloc_len;
6860	}
6861
6862	switch (ctsio->cdb[0]) {
6863	case MODE_SENSE_6: {
6864		struct scsi_mode_hdr_6 *header;
6865
6866		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6867
6868		header->datalen = ctl_min(total_len - 1, 254);
6869
6870		if (dbd)
6871			header->block_descr_len = 0;
6872		else
6873			header->block_descr_len =
6874				sizeof(struct scsi_mode_block_descr);
6875		block_desc = (struct scsi_mode_block_descr *)&header[1];
6876		break;
6877	}
6878	case MODE_SENSE_10: {
6879		struct scsi_mode_hdr_10 *header;
6880		int datalen;
6881
6882		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6883
6884		datalen = ctl_min(total_len - 2, 65533);
6885		scsi_ulto2b(datalen, header->datalen);
6886		if (dbd)
6887			scsi_ulto2b(0, header->block_descr_len);
6888		else
6889			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6890				    header->block_descr_len);
6891		block_desc = (struct scsi_mode_block_descr *)&header[1];
6892		break;
6893	}
6894	default:
6895		panic("invalid CDB type %#x", ctsio->cdb[0]);
6896		break; /* NOTREACHED */
6897	}
6898
6899	/*
6900	 * If we've got a disk, use its blocksize in the block
6901	 * descriptor.  Otherwise, just set it to 0.
6902	 */
6903	if (dbd == 0) {
6904		if (control_dev != 0)
6905			scsi_ulto3b(lun->be_lun->blocksize,
6906				    block_desc->block_len);
6907		else
6908			scsi_ulto3b(0, block_desc->block_len);
6909	}
6910
6911	switch (page_code) {
6912	case SMS_ALL_PAGES_PAGE: {
6913		int i, data_used;
6914
6915		data_used = header_len;
6916		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6917			struct ctl_page_index *page_index;
6918
6919			page_index = &lun->mode_pages.index[i];
6920
6921			if ((control_dev != 0)
6922			 && (page_index->page_flags &
6923			    CTL_PAGE_FLAG_DISK_ONLY))
6924				continue;
6925
6926			/*
6927			 * We don't use this subpage if the user didn't
6928			 * request all subpages.  We already checked (above)
6929			 * to make sure the user only specified a subpage
6930			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6931			 */
6932			if ((page_index->subpage != 0)
6933			 && (subpage == SMS_SUBPAGE_PAGE_0))
6934				continue;
6935
6936			/*
6937			 * Call the handler, if it exists, to update the
6938			 * page to the latest values.
6939			 */
6940			if (page_index->sense_handler != NULL)
6941				page_index->sense_handler(ctsio, page_index,pc);
6942
6943			memcpy(ctsio->kern_data_ptr + data_used,
6944			       page_index->page_data +
6945			       (page_index->page_len * pc),
6946			       page_index->page_len);
6947			data_used += page_index->page_len;
6948		}
6949		break;
6950	}
6951	default: {
6952		int i, data_used;
6953
6954		data_used = header_len;
6955
6956		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6957			struct ctl_page_index *page_index;
6958
6959			page_index = &lun->mode_pages.index[i];
6960
6961			/* Look for the right page code */
6962			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6963				continue;
6964
6965			/* Look for the right subpage or the subpage wildcard*/
6966			if ((page_index->subpage != subpage)
6967			 && (subpage != SMS_SUBPAGE_ALL))
6968				continue;
6969
6970			/* Make sure the page is supported for this dev type */
6971			if ((control_dev != 0)
6972			 && (page_index->page_flags &
6973			     CTL_PAGE_FLAG_DISK_ONLY))
6974				continue;
6975
6976			/*
6977			 * Call the handler, if it exists, to update the
6978			 * page to the latest values.
6979			 */
6980			if (page_index->sense_handler != NULL)
6981				page_index->sense_handler(ctsio, page_index,pc);
6982
6983			memcpy(ctsio->kern_data_ptr + data_used,
6984			       page_index->page_data +
6985			       (page_index->page_len * pc),
6986			       page_index->page_len);
6987			data_used += page_index->page_len;
6988		}
6989		break;
6990	}
6991	}
6992
6993	ctsio->scsi_status = SCSI_STATUS_OK;
6994
6995	ctsio->be_move_done = ctl_config_move_done;
6996	ctl_datamove((union ctl_io *)ctsio);
6997
6998	return (CTL_RETVAL_COMPLETE);
6999}
7000
7001int
7002ctl_read_capacity(struct ctl_scsiio *ctsio)
7003{
7004	struct scsi_read_capacity *cdb;
7005	struct scsi_read_capacity_data *data;
7006	struct ctl_lun *lun;
7007	uint32_t lba;
7008
7009	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7010
7011	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7012
7013	lba = scsi_4btoul(cdb->addr);
7014	if (((cdb->pmi & SRC_PMI) == 0)
7015	 && (lba != 0)) {
7016		ctl_set_invalid_field(/*ctsio*/ ctsio,
7017				      /*sks_valid*/ 1,
7018				      /*command*/ 1,
7019				      /*field*/ 2,
7020				      /*bit_valid*/ 0,
7021				      /*bit*/ 0);
7022		ctl_done((union ctl_io *)ctsio);
7023		return (CTL_RETVAL_COMPLETE);
7024	}
7025
7026	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7027
7028	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7029	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7030	ctsio->residual = 0;
7031	ctsio->kern_data_len = sizeof(*data);
7032	ctsio->kern_total_len = sizeof(*data);
7033	ctsio->kern_data_resid = 0;
7034	ctsio->kern_rel_offset = 0;
7035	ctsio->kern_sg_entries = 0;
7036
7037	/*
7038	 * If the maximum LBA is greater than 0xfffffffe, the user must
7039	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7040	 * serivce action set.
7041	 */
7042	if (lun->be_lun->maxlba > 0xfffffffe)
7043		scsi_ulto4b(0xffffffff, data->addr);
7044	else
7045		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7046
7047	/*
7048	 * XXX KDM this may not be 512 bytes...
7049	 */
7050	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7051
7052	ctsio->scsi_status = SCSI_STATUS_OK;
7053
7054	ctsio->be_move_done = ctl_config_move_done;
7055	ctl_datamove((union ctl_io *)ctsio);
7056
7057	return (CTL_RETVAL_COMPLETE);
7058}
7059
7060static int
7061ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7062{
7063	struct scsi_read_capacity_16 *cdb;
7064	struct scsi_read_capacity_data_long *data;
7065	struct ctl_lun *lun;
7066	uint64_t lba;
7067	uint32_t alloc_len;
7068
7069	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7070
7071	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7072
7073	alloc_len = scsi_4btoul(cdb->alloc_len);
7074	lba = scsi_8btou64(cdb->addr);
7075
7076	if ((cdb->reladr & SRC16_PMI)
7077	 && (lba != 0)) {
7078		ctl_set_invalid_field(/*ctsio*/ ctsio,
7079				      /*sks_valid*/ 1,
7080				      /*command*/ 1,
7081				      /*field*/ 2,
7082				      /*bit_valid*/ 0,
7083				      /*bit*/ 0);
7084		ctl_done((union ctl_io *)ctsio);
7085		return (CTL_RETVAL_COMPLETE);
7086	}
7087
7088	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7089
7090	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7091	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7092
7093	if (sizeof(*data) < alloc_len) {
7094		ctsio->residual = alloc_len - sizeof(*data);
7095		ctsio->kern_data_len = sizeof(*data);
7096		ctsio->kern_total_len = sizeof(*data);
7097	} else {
7098		ctsio->residual = 0;
7099		ctsio->kern_data_len = alloc_len;
7100		ctsio->kern_total_len = alloc_len;
7101	}
7102	ctsio->kern_data_resid = 0;
7103	ctsio->kern_rel_offset = 0;
7104	ctsio->kern_sg_entries = 0;
7105
7106	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7107	/* XXX KDM this may not be 512 bytes... */
7108	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7109	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7110	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7111	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7112		data->lalba_lbp[0] |= SRC16_LBPME;
7113
7114	ctsio->scsi_status = SCSI_STATUS_OK;
7115
7116	ctsio->be_move_done = ctl_config_move_done;
7117	ctl_datamove((union ctl_io *)ctsio);
7118
7119	return (CTL_RETVAL_COMPLETE);
7120}
7121
7122int
7123ctl_service_action_in(struct ctl_scsiio *ctsio)
7124{
7125	struct scsi_service_action_in *cdb;
7126	int retval;
7127
7128	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
7129
7130	cdb = (struct scsi_service_action_in *)ctsio->cdb;
7131
7132	retval = CTL_RETVAL_COMPLETE;
7133
7134	switch (cdb->service_action) {
7135	case SRC16_SERVICE_ACTION:
7136		retval = ctl_read_capacity_16(ctsio);
7137		break;
7138	default:
7139		ctl_set_invalid_field(/*ctsio*/ ctsio,
7140				      /*sks_valid*/ 1,
7141				      /*command*/ 1,
7142				      /*field*/ 1,
7143				      /*bit_valid*/ 1,
7144				      /*bit*/ 4);
7145		ctl_done((union ctl_io *)ctsio);
7146		break;
7147	}
7148
7149	return (retval);
7150}
7151
7152int
7153ctl_maintenance_in(struct ctl_scsiio *ctsio)
7154{
7155	struct scsi_maintenance_in *cdb;
7156	int retval;
7157	int alloc_len, total_len = 0;
7158	int num_target_port_groups, single;
7159	struct ctl_lun *lun;
7160	struct ctl_softc *softc;
7161	struct scsi_target_group_data *rtg_ptr;
7162	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
7163	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
7164	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
7165
7166	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
7167
7168	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7169	softc = control_softc;
7170	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7171
7172	retval = CTL_RETVAL_COMPLETE;
7173
7174	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
7175		ctl_set_invalid_field(/*ctsio*/ ctsio,
7176				      /*sks_valid*/ 1,
7177				      /*command*/ 1,
7178				      /*field*/ 1,
7179				      /*bit_valid*/ 1,
7180				      /*bit*/ 4);
7181		ctl_done((union ctl_io *)ctsio);
7182		return(retval);
7183	}
7184
7185	mtx_lock(&softc->ctl_lock);
7186	single = ctl_is_single;
7187	mtx_unlock(&softc->ctl_lock);
7188
7189	if (single)
7190        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
7191	else
7192        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7193
7194	total_len = sizeof(struct scsi_target_group_data) +
7195		sizeof(struct scsi_target_port_group_descriptor) *
7196		num_target_port_groups +
7197		sizeof(struct scsi_target_port_descriptor) *
7198		NUM_PORTS_PER_GRP * num_target_port_groups;
7199
7200	alloc_len = scsi_4btoul(cdb->length);
7201
7202	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7203
7204	ctsio->kern_sg_entries = 0;
7205
7206	if (total_len < alloc_len) {
7207		ctsio->residual = alloc_len - total_len;
7208		ctsio->kern_data_len = total_len;
7209		ctsio->kern_total_len = total_len;
7210	} else {
7211		ctsio->residual = 0;
7212		ctsio->kern_data_len = alloc_len;
7213		ctsio->kern_total_len = alloc_len;
7214	}
7215	ctsio->kern_data_resid = 0;
7216	ctsio->kern_rel_offset = 0;
7217
7218	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7219
7220	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7221	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7222	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7223	        &tp_desc_ptr1_1->desc_list[0];
7224
7225	if (single == 0) {
7226		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7227	                &tp_desc_ptr1_2->desc_list[0];
7228		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7229		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7230	        	&tp_desc_ptr2_1->desc_list[0];
7231        } else {
7232		tpg_desc_ptr2 = NULL;
7233		tp_desc_ptr2_1 = NULL;
7234		tp_desc_ptr2_2 = NULL;
7235	}
7236
7237	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7238	if (single == 0) {
7239        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7240			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7241				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7242				tpg_desc_ptr2->pref_state =
7243					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7244			} else {
7245				tpg_desc_ptr1->pref_state =
7246					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7247				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7248			}
7249		} else {
7250			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7251				tpg_desc_ptr1->pref_state =
7252					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7253				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7254			} else {
7255				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7256				tpg_desc_ptr2->pref_state =
7257					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7258			}
7259		}
7260	} else {
7261		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7262	}
7263	tpg_desc_ptr1->support = 0;
7264	tpg_desc_ptr1->target_port_group[1] = 1;
7265	tpg_desc_ptr1->status = TPG_IMPLICIT;
7266	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7267
7268	if (single == 0) {
7269		tpg_desc_ptr2->support = 0;
7270		tpg_desc_ptr2->target_port_group[1] = 2;
7271		tpg_desc_ptr2->status = TPG_IMPLICIT;
7272		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7273
7274		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7275		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7276
7277		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7278		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7279	} else {
7280        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7281			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7282			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7283		} else {
7284			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7285			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7286		}
7287	}
7288
7289	ctsio->be_move_done = ctl_config_move_done;
7290
7291	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7292			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7293			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7294			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7295			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7296
7297	ctl_datamove((union ctl_io *)ctsio);
7298	return(retval);
7299}
7300
7301int
7302ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7303{
7304	struct scsi_per_res_in *cdb;
7305	int alloc_len, total_len = 0;
7306	/* struct scsi_per_res_in_rsrv in_data; */
7307	struct ctl_lun *lun;
7308	struct ctl_softc *softc;
7309
7310	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7311
7312	softc = control_softc;
7313
7314	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7315
7316	alloc_len = scsi_2btoul(cdb->length);
7317
7318	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7319
7320retry:
7321	mtx_lock(&softc->ctl_lock);
7322	switch (cdb->action) {
7323	case SPRI_RK: /* read keys */
7324		total_len = sizeof(struct scsi_per_res_in_keys) +
7325			lun->pr_key_count *
7326			sizeof(struct scsi_per_res_key);
7327		break;
7328	case SPRI_RR: /* read reservation */
7329		if (lun->flags & CTL_LUN_PR_RESERVED)
7330			total_len = sizeof(struct scsi_per_res_in_rsrv);
7331		else
7332			total_len = sizeof(struct scsi_per_res_in_header);
7333		break;
7334	case SPRI_RC: /* report capabilities */
7335		total_len = sizeof(struct scsi_per_res_cap);
7336		break;
7337	case SPRI_RS: /* read full status */
7338	default:
7339		mtx_unlock(&softc->ctl_lock);
7340		ctl_set_invalid_field(ctsio,
7341				      /*sks_valid*/ 1,
7342				      /*command*/ 1,
7343				      /*field*/ 1,
7344				      /*bit_valid*/ 1,
7345				      /*bit*/ 0);
7346		ctl_done((union ctl_io *)ctsio);
7347		return (CTL_RETVAL_COMPLETE);
7348		break; /* NOTREACHED */
7349	}
7350	mtx_unlock(&softc->ctl_lock);
7351
7352	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7353
7354	if (total_len < alloc_len) {
7355		ctsio->residual = alloc_len - total_len;
7356		ctsio->kern_data_len = total_len;
7357		ctsio->kern_total_len = total_len;
7358	} else {
7359		ctsio->residual = 0;
7360		ctsio->kern_data_len = alloc_len;
7361		ctsio->kern_total_len = alloc_len;
7362	}
7363
7364	ctsio->kern_data_resid = 0;
7365	ctsio->kern_rel_offset = 0;
7366	ctsio->kern_sg_entries = 0;
7367
7368	mtx_lock(&softc->ctl_lock);
7369	switch (cdb->action) {
7370	case SPRI_RK: { // read keys
7371        struct scsi_per_res_in_keys *res_keys;
7372		int i, key_count;
7373
7374		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7375
7376		/*
7377		 * We had to drop the lock to allocate our buffer, which
7378		 * leaves time for someone to come in with another
7379		 * persistent reservation.  (That is unlikely, though,
7380		 * since this should be the only persistent reservation
7381		 * command active right now.)
7382		 */
7383		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7384		    (lun->pr_key_count *
7385		     sizeof(struct scsi_per_res_key)))){
7386			mtx_unlock(&softc->ctl_lock);
7387			free(ctsio->kern_data_ptr, M_CTL);
7388			printf("%s: reservation length changed, retrying\n",
7389			       __func__);
7390			goto retry;
7391		}
7392
7393		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7394
7395		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7396			     lun->pr_key_count, res_keys->header.length);
7397
7398		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7399			if (!lun->per_res[i].registered)
7400				continue;
7401
7402			/*
7403			 * We used lun->pr_key_count to calculate the
7404			 * size to allocate.  If it turns out the number of
7405			 * initiators with the registered flag set is
7406			 * larger than that (i.e. they haven't been kept in
7407			 * sync), we've got a problem.
7408			 */
7409			if (key_count >= lun->pr_key_count) {
7410#ifdef NEEDTOPORT
7411				csevent_log(CSC_CTL | CSC_SHELF_SW |
7412					    CTL_PR_ERROR,
7413					    csevent_LogType_Fault,
7414					    csevent_AlertLevel_Yellow,
7415					    csevent_FRU_ShelfController,
7416					    csevent_FRU_Firmware,
7417				        csevent_FRU_Unknown,
7418					    "registered keys %d >= key "
7419					    "count %d", key_count,
7420					    lun->pr_key_count);
7421#endif
7422				key_count++;
7423				continue;
7424			}
7425			memcpy(res_keys->keys[key_count].key,
7426			       lun->per_res[i].res_key.key,
7427			       ctl_min(sizeof(res_keys->keys[key_count].key),
7428			       sizeof(lun->per_res[i].res_key)));
7429			key_count++;
7430		}
7431		break;
7432	}
7433	case SPRI_RR: { // read reservation
7434		struct scsi_per_res_in_rsrv *res;
7435		int tmp_len, header_only;
7436
7437		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7438
7439		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7440
7441		if (lun->flags & CTL_LUN_PR_RESERVED)
7442		{
7443			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7444			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7445				    res->header.length);
7446			header_only = 0;
7447		} else {
7448			tmp_len = sizeof(struct scsi_per_res_in_header);
7449			scsi_ulto4b(0, res->header.length);
7450			header_only = 1;
7451		}
7452
7453		/*
7454		 * We had to drop the lock to allocate our buffer, which
7455		 * leaves time for someone to come in with another
7456		 * persistent reservation.  (That is unlikely, though,
7457		 * since this should be the only persistent reservation
7458		 * command active right now.)
7459		 */
7460		if (tmp_len != total_len) {
7461			mtx_unlock(&softc->ctl_lock);
7462			free(ctsio->kern_data_ptr, M_CTL);
7463			printf("%s: reservation status changed, retrying\n",
7464			       __func__);
7465			goto retry;
7466		}
7467
7468		/*
7469		 * No reservation held, so we're done.
7470		 */
7471		if (header_only != 0)
7472			break;
7473
7474		/*
7475		 * If the registration is an All Registrants type, the key
7476		 * is 0, since it doesn't really matter.
7477		 */
7478		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7479			memcpy(res->data.reservation,
7480			       &lun->per_res[lun->pr_res_idx].res_key,
7481			       sizeof(struct scsi_per_res_key));
7482		}
7483		res->data.scopetype = lun->res_type;
7484		break;
7485	}
7486	case SPRI_RC:     //report capabilities
7487	{
7488		struct scsi_per_res_cap *res_cap;
7489		uint16_t type_mask;
7490
7491		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7492		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7493		res_cap->flags2 |= SPRI_TMV;
7494		type_mask = SPRI_TM_WR_EX_AR |
7495			    SPRI_TM_EX_AC_RO |
7496			    SPRI_TM_WR_EX_RO |
7497			    SPRI_TM_EX_AC |
7498			    SPRI_TM_WR_EX |
7499			    SPRI_TM_EX_AC_AR;
7500		scsi_ulto2b(type_mask, res_cap->type_mask);
7501		break;
7502	}
7503	case SPRI_RS: //read full status
7504	default:
7505		/*
7506		 * This is a bug, because we just checked for this above,
7507		 * and should have returned an error.
7508		 */
7509		panic("Invalid PR type %x", cdb->action);
7510		break; /* NOTREACHED */
7511	}
7512	mtx_unlock(&softc->ctl_lock);
7513
7514	ctsio->be_move_done = ctl_config_move_done;
7515
7516	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7517			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7518			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7519			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7520			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7521
7522	ctl_datamove((union ctl_io *)ctsio);
7523
7524	return (CTL_RETVAL_COMPLETE);
7525}
7526
7527/*
7528 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7529 * it should return.
7530 */
7531static int
7532ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7533		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7534		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7535		struct scsi_per_res_out_parms* param)
7536{
7537	union ctl_ha_msg persis_io;
7538	int retval, i;
7539	int isc_retval;
7540
7541	retval = 0;
7542
7543	if (sa_res_key == 0) {
7544		mtx_lock(&softc->ctl_lock);
7545		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7546			/* validate scope and type */
7547			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7548			     SPR_LU_SCOPE) {
7549				mtx_unlock(&softc->ctl_lock);
7550				ctl_set_invalid_field(/*ctsio*/ ctsio,
7551						      /*sks_valid*/ 1,
7552						      /*command*/ 1,
7553						      /*field*/ 2,
7554						      /*bit_valid*/ 1,
7555						      /*bit*/ 4);
7556				ctl_done((union ctl_io *)ctsio);
7557				return (1);
7558			}
7559
7560		        if (type>8 || type==2 || type==4 || type==0) {
7561				mtx_unlock(&softc->ctl_lock);
7562				ctl_set_invalid_field(/*ctsio*/ ctsio,
7563       	           				      /*sks_valid*/ 1,
7564						      /*command*/ 1,
7565						      /*field*/ 2,
7566						      /*bit_valid*/ 1,
7567						      /*bit*/ 0);
7568				ctl_done((union ctl_io *)ctsio);
7569				return (1);
7570		        }
7571
7572			/* temporarily unregister this nexus */
7573			lun->per_res[residx].registered = 0;
7574
7575			/*
7576			 * Unregister everybody else and build UA for
7577			 * them
7578			 */
7579			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7580				if (lun->per_res[i].registered == 0)
7581					continue;
7582
7583				if (!persis_offset
7584				 && i <CTL_MAX_INITIATORS)
7585					lun->pending_sense[i].ua_pending |=
7586						CTL_UA_REG_PREEMPT;
7587				else if (persis_offset
7588				      && i >= persis_offset)
7589					lun->pending_sense[i-persis_offset
7590						].ua_pending |=
7591						CTL_UA_REG_PREEMPT;
7592				lun->per_res[i].registered = 0;
7593				memset(&lun->per_res[i].res_key, 0,
7594				       sizeof(struct scsi_per_res_key));
7595			}
7596			lun->per_res[residx].registered = 1;
7597			lun->pr_key_count = 1;
7598			lun->res_type = type;
7599			if (lun->res_type != SPR_TYPE_WR_EX_AR
7600			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7601				lun->pr_res_idx = residx;
7602
7603			mtx_unlock(&softc->ctl_lock);
7604			/* send msg to other side */
7605			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7606			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7607			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7608			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7609			persis_io.pr.pr_info.res_type = type;
7610			memcpy(persis_io.pr.pr_info.sa_res_key,
7611			       param->serv_act_res_key,
7612			       sizeof(param->serv_act_res_key));
7613			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7614			     &persis_io, sizeof(persis_io), 0)) >
7615			     CTL_HA_STATUS_SUCCESS) {
7616				printf("CTL:Persis Out error returned "
7617				       "from ctl_ha_msg_send %d\n",
7618				       isc_retval);
7619			}
7620		} else {
7621			/* not all registrants */
7622			mtx_unlock(&softc->ctl_lock);
7623			free(ctsio->kern_data_ptr, M_CTL);
7624			ctl_set_invalid_field(ctsio,
7625					      /*sks_valid*/ 1,
7626					      /*command*/ 0,
7627					      /*field*/ 8,
7628					      /*bit_valid*/ 0,
7629					      /*bit*/ 0);
7630			ctl_done((union ctl_io *)ctsio);
7631			return (1);
7632		}
7633	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7634		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7635		int found = 0;
7636
7637		mtx_lock(&softc->ctl_lock);
7638		if (res_key == sa_res_key) {
7639			/* special case */
7640			/*
7641			 * The spec implies this is not good but doesn't
7642			 * say what to do. There are two choices either
7643			 * generate a res conflict or check condition
7644			 * with illegal field in parameter data. Since
7645			 * that is what is done when the sa_res_key is
7646			 * zero I'll take that approach since this has
7647			 * to do with the sa_res_key.
7648			 */
7649			mtx_unlock(&softc->ctl_lock);
7650			free(ctsio->kern_data_ptr, M_CTL);
7651			ctl_set_invalid_field(ctsio,
7652					      /*sks_valid*/ 1,
7653					      /*command*/ 0,
7654					      /*field*/ 8,
7655					      /*bit_valid*/ 0,
7656					      /*bit*/ 0);
7657			ctl_done((union ctl_io *)ctsio);
7658			return (1);
7659		}
7660
7661		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7662			if (lun->per_res[i].registered
7663			 && memcmp(param->serv_act_res_key,
7664			    lun->per_res[i].res_key.key,
7665			    sizeof(struct scsi_per_res_key)) != 0)
7666				continue;
7667
7668			found = 1;
7669			lun->per_res[i].registered = 0;
7670			memset(&lun->per_res[i].res_key, 0,
7671			       sizeof(struct scsi_per_res_key));
7672			lun->pr_key_count--;
7673
7674			if (!persis_offset
7675			 && i < CTL_MAX_INITIATORS)
7676				lun->pending_sense[i].ua_pending |=
7677					CTL_UA_REG_PREEMPT;
7678			else if (persis_offset
7679			      && i >= persis_offset)
7680				lun->pending_sense[i-persis_offset].ua_pending|=
7681					CTL_UA_REG_PREEMPT;
7682		}
7683		mtx_unlock(&softc->ctl_lock);
7684		if (!found) {
7685			free(ctsio->kern_data_ptr, M_CTL);
7686			ctl_set_reservation_conflict(ctsio);
7687			ctl_done((union ctl_io *)ctsio);
7688			return (CTL_RETVAL_COMPLETE);
7689		}
7690		/* send msg to other side */
7691		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7692		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7693		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7694		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7695		persis_io.pr.pr_info.res_type = type;
7696		memcpy(persis_io.pr.pr_info.sa_res_key,
7697		       param->serv_act_res_key,
7698		       sizeof(param->serv_act_res_key));
7699		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7700		     &persis_io, sizeof(persis_io), 0)) >
7701		     CTL_HA_STATUS_SUCCESS) {
7702			printf("CTL:Persis Out error returned from "
7703			       "ctl_ha_msg_send %d\n", isc_retval);
7704		}
7705	} else {
7706		/* Reserved but not all registrants */
7707		/* sa_res_key is res holder */
7708		if (memcmp(param->serv_act_res_key,
7709                   lun->per_res[lun->pr_res_idx].res_key.key,
7710                   sizeof(struct scsi_per_res_key)) == 0) {
7711			/* validate scope and type */
7712			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7713			     SPR_LU_SCOPE) {
7714				ctl_set_invalid_field(/*ctsio*/ ctsio,
7715						      /*sks_valid*/ 1,
7716						      /*command*/ 1,
7717						      /*field*/ 2,
7718						      /*bit_valid*/ 1,
7719						      /*bit*/ 4);
7720				ctl_done((union ctl_io *)ctsio);
7721				return (1);
7722			}
7723
7724			if (type>8 || type==2 || type==4 || type==0) {
7725				ctl_set_invalid_field(/*ctsio*/ ctsio,
7726						      /*sks_valid*/ 1,
7727						      /*command*/ 1,
7728						      /*field*/ 2,
7729						      /*bit_valid*/ 1,
7730						      /*bit*/ 0);
7731				ctl_done((union ctl_io *)ctsio);
7732				return (1);
7733			}
7734
7735			/*
7736			 * Do the following:
7737			 * if sa_res_key != res_key remove all
7738			 * registrants w/sa_res_key and generate UA
7739			 * for these registrants(Registrations
7740			 * Preempted) if it wasn't an exclusive
7741			 * reservation generate UA(Reservations
7742			 * Preempted) for all other registered nexuses
7743			 * if the type has changed. Establish the new
7744			 * reservation and holder. If res_key and
7745			 * sa_res_key are the same do the above
7746			 * except don't unregister the res holder.
7747			 */
7748
7749			/*
7750			 * Temporarily unregister so it won't get
7751			 * removed or UA generated
7752			 */
7753			lun->per_res[residx].registered = 0;
7754			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7755				if (lun->per_res[i].registered == 0)
7756					continue;
7757
7758				if (memcmp(param->serv_act_res_key,
7759				    lun->per_res[i].res_key.key,
7760				    sizeof(struct scsi_per_res_key)) == 0) {
7761					lun->per_res[i].registered = 0;
7762					memset(&lun->per_res[i].res_key,
7763					       0,
7764					       sizeof(struct scsi_per_res_key));
7765					lun->pr_key_count--;
7766
7767					if (!persis_offset
7768					 && i < CTL_MAX_INITIATORS)
7769						lun->pending_sense[i
7770							].ua_pending |=
7771							CTL_UA_REG_PREEMPT;
7772					else if (persis_offset
7773					      && i >= persis_offset)
7774						lun->pending_sense[
7775						  i-persis_offset].ua_pending |=
7776						  CTL_UA_REG_PREEMPT;
7777				} else if (type != lun->res_type
7778					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7779					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7780						if (!persis_offset
7781						 && i < CTL_MAX_INITIATORS)
7782							lun->pending_sense[i
7783							].ua_pending |=
7784							CTL_UA_RES_RELEASE;
7785						else if (persis_offset
7786						      && i >= persis_offset)
7787							lun->pending_sense[
7788							i-persis_offset
7789							].ua_pending |=
7790							CTL_UA_RES_RELEASE;
7791				}
7792			}
7793			lun->per_res[residx].registered = 1;
7794			lun->res_type = type;
7795			if (lun->res_type != SPR_TYPE_WR_EX_AR
7796			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7797				lun->pr_res_idx = residx;
7798			else
7799				lun->pr_res_idx =
7800					CTL_PR_ALL_REGISTRANTS;
7801
7802			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7803			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7804			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7805			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7806			persis_io.pr.pr_info.res_type = type;
7807			memcpy(persis_io.pr.pr_info.sa_res_key,
7808			       param->serv_act_res_key,
7809			       sizeof(param->serv_act_res_key));
7810			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7811			     &persis_io, sizeof(persis_io), 0)) >
7812			     CTL_HA_STATUS_SUCCESS) {
7813				printf("CTL:Persis Out error returned "
7814				       "from ctl_ha_msg_send %d\n",
7815				       isc_retval);
7816			}
7817		} else {
7818			/*
7819			 * sa_res_key is not the res holder just
7820			 * remove registrants
7821			 */
7822			int found=0;
7823			mtx_lock(&softc->ctl_lock);
7824
7825			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7826				if (memcmp(param->serv_act_res_key,
7827				    lun->per_res[i].res_key.key,
7828				    sizeof(struct scsi_per_res_key)) != 0)
7829					continue;
7830
7831				found = 1;
7832				lun->per_res[i].registered = 0;
7833				memset(&lun->per_res[i].res_key, 0,
7834				       sizeof(struct scsi_per_res_key));
7835				lun->pr_key_count--;
7836
7837				if (!persis_offset
7838				 && i < CTL_MAX_INITIATORS)
7839					lun->pending_sense[i].ua_pending |=
7840						CTL_UA_REG_PREEMPT;
7841				else if (persis_offset
7842				      && i >= persis_offset)
7843					lun->pending_sense[
7844						i-persis_offset].ua_pending |=
7845						CTL_UA_REG_PREEMPT;
7846			}
7847
7848			if (!found) {
7849				mtx_unlock(&softc->ctl_lock);
7850				free(ctsio->kern_data_ptr, M_CTL);
7851				ctl_set_reservation_conflict(ctsio);
7852				ctl_done((union ctl_io *)ctsio);
7853		        	return (1);
7854			}
7855			mtx_unlock(&softc->ctl_lock);
7856			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7857			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7858			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7859			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7860			persis_io.pr.pr_info.res_type = type;
7861			memcpy(persis_io.pr.pr_info.sa_res_key,
7862			       param->serv_act_res_key,
7863			       sizeof(param->serv_act_res_key));
7864			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7865			     &persis_io, sizeof(persis_io), 0)) >
7866			     CTL_HA_STATUS_SUCCESS) {
7867				printf("CTL:Persis Out error returned "
7868				       "from ctl_ha_msg_send %d\n",
7869				isc_retval);
7870			}
7871		}
7872	}
7873
7874	lun->PRGeneration++;
7875
7876	return (retval);
7877}
7878
7879static void
7880ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7881{
7882	int i;
7883
7884	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7885	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7886	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7887		   msg->pr.pr_info.sa_res_key,
7888		   sizeof(struct scsi_per_res_key)) != 0) {
7889		uint64_t sa_res_key;
7890		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7891
7892		if (sa_res_key == 0) {
7893			/* temporarily unregister this nexus */
7894			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7895
7896			/*
7897			 * Unregister everybody else and build UA for
7898			 * them
7899			 */
7900			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7901				if (lun->per_res[i].registered == 0)
7902					continue;
7903
7904				if (!persis_offset
7905				 && i < CTL_MAX_INITIATORS)
7906					lun->pending_sense[i].ua_pending |=
7907						CTL_UA_REG_PREEMPT;
7908				else if (persis_offset && i >= persis_offset)
7909					lun->pending_sense[i -
7910						persis_offset].ua_pending |=
7911						CTL_UA_REG_PREEMPT;
7912				lun->per_res[i].registered = 0;
7913				memset(&lun->per_res[i].res_key, 0,
7914				       sizeof(struct scsi_per_res_key));
7915			}
7916
7917			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7918			lun->pr_key_count = 1;
7919			lun->res_type = msg->pr.pr_info.res_type;
7920			if (lun->res_type != SPR_TYPE_WR_EX_AR
7921			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7922				lun->pr_res_idx = msg->pr.pr_info.residx;
7923		} else {
7924		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7925				if (memcmp(msg->pr.pr_info.sa_res_key,
7926		                   lun->per_res[i].res_key.key,
7927		                   sizeof(struct scsi_per_res_key)) != 0)
7928					continue;
7929
7930				lun->per_res[i].registered = 0;
7931				memset(&lun->per_res[i].res_key, 0,
7932				       sizeof(struct scsi_per_res_key));
7933				lun->pr_key_count--;
7934
7935				if (!persis_offset
7936				 && i < persis_offset)
7937					lun->pending_sense[i].ua_pending |=
7938						CTL_UA_REG_PREEMPT;
7939				else if (persis_offset
7940				      && i >= persis_offset)
7941					lun->pending_sense[i -
7942						persis_offset].ua_pending |=
7943						CTL_UA_REG_PREEMPT;
7944			}
7945		}
7946	} else {
7947		/*
7948		 * Temporarily unregister so it won't get removed
7949		 * or UA generated
7950		 */
7951		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7952		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7953			if (lun->per_res[i].registered == 0)
7954				continue;
7955
7956			if (memcmp(msg->pr.pr_info.sa_res_key,
7957	                   lun->per_res[i].res_key.key,
7958	                   sizeof(struct scsi_per_res_key)) == 0) {
7959				lun->per_res[i].registered = 0;
7960				memset(&lun->per_res[i].res_key, 0,
7961				       sizeof(struct scsi_per_res_key));
7962				lun->pr_key_count--;
7963				if (!persis_offset
7964				 && i < CTL_MAX_INITIATORS)
7965					lun->pending_sense[i].ua_pending |=
7966						CTL_UA_REG_PREEMPT;
7967				else if (persis_offset
7968				      && i >= persis_offset)
7969					lun->pending_sense[i -
7970						persis_offset].ua_pending |=
7971						CTL_UA_REG_PREEMPT;
7972			} else if (msg->pr.pr_info.res_type != lun->res_type
7973				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7974				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7975					if (!persis_offset
7976					 && i < persis_offset)
7977						lun->pending_sense[i
7978							].ua_pending |=
7979							CTL_UA_RES_RELEASE;
7980					else if (persis_offset
7981					      && i >= persis_offset)
7982					lun->pending_sense[i -
7983						persis_offset].ua_pending |=
7984						CTL_UA_RES_RELEASE;
7985			}
7986		}
7987		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7988		lun->res_type = msg->pr.pr_info.res_type;
7989		if (lun->res_type != SPR_TYPE_WR_EX_AR
7990		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7991			lun->pr_res_idx = msg->pr.pr_info.residx;
7992		else
7993			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7994	}
7995	lun->PRGeneration++;
7996
7997}
7998
7999
8000int
8001ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8002{
8003	int retval;
8004	int isc_retval;
8005	u_int32_t param_len;
8006	struct scsi_per_res_out *cdb;
8007	struct ctl_lun *lun;
8008	struct scsi_per_res_out_parms* param;
8009	struct ctl_softc *softc;
8010	uint32_t residx;
8011	uint64_t res_key, sa_res_key;
8012	uint8_t type;
8013	union ctl_ha_msg persis_io;
8014	int    i;
8015
8016	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8017
8018	retval = CTL_RETVAL_COMPLETE;
8019
8020	softc = control_softc;
8021
8022	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8023	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8024
8025	/*
8026	 * We only support whole-LUN scope.  The scope & type are ignored for
8027	 * register, register and ignore existing key and clear.
8028	 * We sometimes ignore scope and type on preempts too!!
8029	 * Verify reservation type here as well.
8030	 */
8031	type = cdb->scope_type & SPR_TYPE_MASK;
8032	if ((cdb->action == SPRO_RESERVE)
8033	 || (cdb->action == SPRO_RELEASE)) {
8034		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8035			ctl_set_invalid_field(/*ctsio*/ ctsio,
8036					      /*sks_valid*/ 1,
8037					      /*command*/ 1,
8038					      /*field*/ 2,
8039					      /*bit_valid*/ 1,
8040					      /*bit*/ 4);
8041			ctl_done((union ctl_io *)ctsio);
8042			return (CTL_RETVAL_COMPLETE);
8043		}
8044
8045		if (type>8 || type==2 || type==4 || type==0) {
8046			ctl_set_invalid_field(/*ctsio*/ ctsio,
8047					      /*sks_valid*/ 1,
8048					      /*command*/ 1,
8049					      /*field*/ 2,
8050					      /*bit_valid*/ 1,
8051					      /*bit*/ 0);
8052			ctl_done((union ctl_io *)ctsio);
8053			return (CTL_RETVAL_COMPLETE);
8054		}
8055	}
8056
8057	switch (cdb->action & SPRO_ACTION_MASK) {
8058	case SPRO_REGISTER:
8059	case SPRO_RESERVE:
8060	case SPRO_RELEASE:
8061	case SPRO_CLEAR:
8062	case SPRO_PREEMPT:
8063	case SPRO_REG_IGNO:
8064		break;
8065	case SPRO_REG_MOVE:
8066	case SPRO_PRE_ABO:
8067	default:
8068		ctl_set_invalid_field(/*ctsio*/ ctsio,
8069				      /*sks_valid*/ 1,
8070				      /*command*/ 1,
8071				      /*field*/ 1,
8072				      /*bit_valid*/ 1,
8073				      /*bit*/ 0);
8074		ctl_done((union ctl_io *)ctsio);
8075		return (CTL_RETVAL_COMPLETE);
8076		break; /* NOTREACHED */
8077	}
8078
8079	param_len = scsi_4btoul(cdb->length);
8080
8081	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8082		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8083		ctsio->kern_data_len = param_len;
8084		ctsio->kern_total_len = param_len;
8085		ctsio->kern_data_resid = 0;
8086		ctsio->kern_rel_offset = 0;
8087		ctsio->kern_sg_entries = 0;
8088		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8089		ctsio->be_move_done = ctl_config_move_done;
8090		ctl_datamove((union ctl_io *)ctsio);
8091
8092		return (CTL_RETVAL_COMPLETE);
8093	}
8094
8095	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8096
8097	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8098	res_key = scsi_8btou64(param->res_key.key);
8099	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8100
8101	/*
8102	 * Validate the reservation key here except for SPRO_REG_IGNO
8103	 * This must be done for all other service actions
8104	 */
8105	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8106		mtx_lock(&softc->ctl_lock);
8107		if (lun->per_res[residx].registered) {
8108		    if (memcmp(param->res_key.key,
8109			       lun->per_res[residx].res_key.key,
8110			       ctl_min(sizeof(param->res_key),
8111			       sizeof(lun->per_res[residx].res_key))) != 0) {
8112				/*
8113				 * The current key passed in doesn't match
8114				 * the one the initiator previously
8115				 * registered.
8116				 */
8117				mtx_unlock(&softc->ctl_lock);
8118				free(ctsio->kern_data_ptr, M_CTL);
8119				ctl_set_reservation_conflict(ctsio);
8120				ctl_done((union ctl_io *)ctsio);
8121				return (CTL_RETVAL_COMPLETE);
8122			}
8123		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8124			/*
8125			 * We are not registered
8126			 */
8127			mtx_unlock(&softc->ctl_lock);
8128			free(ctsio->kern_data_ptr, M_CTL);
8129			ctl_set_reservation_conflict(ctsio);
8130			ctl_done((union ctl_io *)ctsio);
8131			return (CTL_RETVAL_COMPLETE);
8132		} else if (res_key != 0) {
8133			/*
8134			 * We are not registered and trying to register but
8135			 * the register key isn't zero.
8136			 */
8137			mtx_unlock(&softc->ctl_lock);
8138			free(ctsio->kern_data_ptr, M_CTL);
8139			ctl_set_reservation_conflict(ctsio);
8140			ctl_done((union ctl_io *)ctsio);
8141			return (CTL_RETVAL_COMPLETE);
8142		}
8143		mtx_unlock(&softc->ctl_lock);
8144	}
8145
8146	switch (cdb->action & SPRO_ACTION_MASK) {
8147	case SPRO_REGISTER:
8148	case SPRO_REG_IGNO: {
8149
8150#if 0
8151		printf("Registration received\n");
8152#endif
8153
8154		/*
8155		 * We don't support any of these options, as we report in
8156		 * the read capabilities request (see
8157		 * ctl_persistent_reserve_in(), above).
8158		 */
8159		if ((param->flags & SPR_SPEC_I_PT)
8160		 || (param->flags & SPR_ALL_TG_PT)
8161		 || (param->flags & SPR_APTPL)) {
8162			int bit_ptr;
8163
8164			if (param->flags & SPR_APTPL)
8165				bit_ptr = 0;
8166			else if (param->flags & SPR_ALL_TG_PT)
8167				bit_ptr = 2;
8168			else /* SPR_SPEC_I_PT */
8169				bit_ptr = 3;
8170
8171			free(ctsio->kern_data_ptr, M_CTL);
8172			ctl_set_invalid_field(ctsio,
8173					      /*sks_valid*/ 1,
8174					      /*command*/ 0,
8175					      /*field*/ 20,
8176					      /*bit_valid*/ 1,
8177					      /*bit*/ bit_ptr);
8178			ctl_done((union ctl_io *)ctsio);
8179			return (CTL_RETVAL_COMPLETE);
8180		}
8181
8182		mtx_lock(&softc->ctl_lock);
8183
8184		/*
8185		 * The initiator wants to clear the
8186		 * key/unregister.
8187		 */
8188		if (sa_res_key == 0) {
8189			if ((res_key == 0
8190			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8191			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8192			  && !lun->per_res[residx].registered)) {
8193				mtx_unlock(&softc->ctl_lock);
8194				goto done;
8195			}
8196
8197			lun->per_res[residx].registered = 0;
8198			memset(&lun->per_res[residx].res_key,
8199			       0, sizeof(lun->per_res[residx].res_key));
8200			lun->pr_key_count--;
8201
8202			if (residx == lun->pr_res_idx) {
8203				lun->flags &= ~CTL_LUN_PR_RESERVED;
8204				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8205
8206				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8207				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8208				 && lun->pr_key_count) {
8209					/*
8210					 * If the reservation is a registrants
8211					 * only type we need to generate a UA
8212					 * for other registered inits.  The
8213					 * sense code should be RESERVATIONS
8214					 * RELEASED
8215					 */
8216
8217					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8218						if (lun->per_res[
8219						    i+persis_offset].registered
8220						    == 0)
8221							continue;
8222						lun->pending_sense[i
8223							].ua_pending |=
8224							CTL_UA_RES_RELEASE;
8225					}
8226				}
8227				lun->res_type = 0;
8228			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8229				if (lun->pr_key_count==0) {
8230					lun->flags &= ~CTL_LUN_PR_RESERVED;
8231					lun->res_type = 0;
8232					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8233				}
8234			}
8235			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8236			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8237			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8238			persis_io.pr.pr_info.residx = residx;
8239			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8240			     &persis_io, sizeof(persis_io), 0 )) >
8241			     CTL_HA_STATUS_SUCCESS) {
8242				printf("CTL:Persis Out error returned from "
8243				       "ctl_ha_msg_send %d\n", isc_retval);
8244			}
8245			mtx_unlock(&softc->ctl_lock);
8246		} else /* sa_res_key != 0 */ {
8247
8248			/*
8249			 * If we aren't registered currently then increment
8250			 * the key count and set the registered flag.
8251			 */
8252			if (!lun->per_res[residx].registered) {
8253				lun->pr_key_count++;
8254				lun->per_res[residx].registered = 1;
8255			}
8256
8257			memcpy(&lun->per_res[residx].res_key,
8258			       param->serv_act_res_key,
8259			       ctl_min(sizeof(param->serv_act_res_key),
8260			       sizeof(lun->per_res[residx].res_key)));
8261
8262			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8263			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8264			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8265			persis_io.pr.pr_info.residx = residx;
8266			memcpy(persis_io.pr.pr_info.sa_res_key,
8267			       param->serv_act_res_key,
8268			       sizeof(param->serv_act_res_key));
8269			mtx_unlock(&softc->ctl_lock);
8270			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8271			     &persis_io, sizeof(persis_io), 0)) >
8272			     CTL_HA_STATUS_SUCCESS) {
8273				printf("CTL:Persis Out error returned from "
8274				       "ctl_ha_msg_send %d\n", isc_retval);
8275			}
8276		}
8277		lun->PRGeneration++;
8278
8279		break;
8280	}
8281	case SPRO_RESERVE:
8282#if 0
8283                printf("Reserve executed type %d\n", type);
8284#endif
8285		mtx_lock(&softc->ctl_lock);
8286		if (lun->flags & CTL_LUN_PR_RESERVED) {
8287			/*
8288			 * if this isn't the reservation holder and it's
8289			 * not a "all registrants" type or if the type is
8290			 * different then we have a conflict
8291			 */
8292			if ((lun->pr_res_idx != residx
8293			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8294			 || lun->res_type != type) {
8295				mtx_unlock(&softc->ctl_lock);
8296				free(ctsio->kern_data_ptr, M_CTL);
8297				ctl_set_reservation_conflict(ctsio);
8298				ctl_done((union ctl_io *)ctsio);
8299				return (CTL_RETVAL_COMPLETE);
8300			}
8301			mtx_unlock(&softc->ctl_lock);
8302		} else /* create a reservation */ {
8303			/*
8304			 * If it's not an "all registrants" type record
8305			 * reservation holder
8306			 */
8307			if (type != SPR_TYPE_WR_EX_AR
8308			 && type != SPR_TYPE_EX_AC_AR)
8309				lun->pr_res_idx = residx; /* Res holder */
8310			else
8311				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8312
8313			lun->flags |= CTL_LUN_PR_RESERVED;
8314			lun->res_type = type;
8315
8316			mtx_unlock(&softc->ctl_lock);
8317
8318			/* send msg to other side */
8319			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8320			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8321			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8322			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8323			persis_io.pr.pr_info.res_type = type;
8324			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8325			     &persis_io, sizeof(persis_io), 0)) >
8326			     CTL_HA_STATUS_SUCCESS) {
8327				printf("CTL:Persis Out error returned from "
8328				       "ctl_ha_msg_send %d\n", isc_retval);
8329			}
8330		}
8331		break;
8332
8333	case SPRO_RELEASE:
8334		mtx_lock(&softc->ctl_lock);
8335		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8336			/* No reservation exists return good status */
8337			mtx_unlock(&softc->ctl_lock);
8338			goto done;
8339		}
8340		/*
8341		 * Is this nexus a reservation holder?
8342		 */
8343		if (lun->pr_res_idx != residx
8344		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8345			/*
8346			 * not a res holder return good status but
8347			 * do nothing
8348			 */
8349			mtx_unlock(&softc->ctl_lock);
8350			goto done;
8351		}
8352
8353		if (lun->res_type != type) {
8354			mtx_unlock(&softc->ctl_lock);
8355			free(ctsio->kern_data_ptr, M_CTL);
8356			ctl_set_illegal_pr_release(ctsio);
8357			ctl_done((union ctl_io *)ctsio);
8358			return (CTL_RETVAL_COMPLETE);
8359		}
8360
8361		/* okay to release */
8362		lun->flags &= ~CTL_LUN_PR_RESERVED;
8363		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8364		lun->res_type = 0;
8365
8366		/*
8367		 * if this isn't an exclusive access
8368		 * res generate UA for all other
8369		 * registrants.
8370		 */
8371		if (type != SPR_TYPE_EX_AC
8372		 && type != SPR_TYPE_WR_EX) {
8373			/*
8374			 * temporarily unregister so we don't generate UA
8375			 */
8376			lun->per_res[residx].registered = 0;
8377
8378			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8379				if (lun->per_res[i+persis_offset].registered
8380				    == 0)
8381					continue;
8382				lun->pending_sense[i].ua_pending |=
8383					CTL_UA_RES_RELEASE;
8384			}
8385
8386			lun->per_res[residx].registered = 1;
8387		}
8388		mtx_unlock(&softc->ctl_lock);
8389		/* Send msg to other side */
8390		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8391		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8392		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8393		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8394		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8395			printf("CTL:Persis Out error returned from "
8396			       "ctl_ha_msg_send %d\n", isc_retval);
8397		}
8398		break;
8399
8400	case SPRO_CLEAR:
8401		/* send msg to other side */
8402
8403		mtx_lock(&softc->ctl_lock);
8404		lun->flags &= ~CTL_LUN_PR_RESERVED;
8405		lun->res_type = 0;
8406		lun->pr_key_count = 0;
8407		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8408
8409
8410		memset(&lun->per_res[residx].res_key,
8411		       0, sizeof(lun->per_res[residx].res_key));
8412		lun->per_res[residx].registered = 0;
8413
8414		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8415			if (lun->per_res[i].registered) {
8416				if (!persis_offset && i < CTL_MAX_INITIATORS)
8417					lun->pending_sense[i].ua_pending |=
8418						CTL_UA_RES_PREEMPT;
8419				else if (persis_offset && i >= persis_offset)
8420					lun->pending_sense[i-persis_offset
8421					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8422
8423				memset(&lun->per_res[i].res_key,
8424				       0, sizeof(struct scsi_per_res_key));
8425				lun->per_res[i].registered = 0;
8426			}
8427		lun->PRGeneration++;
8428		mtx_unlock(&softc->ctl_lock);
8429		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8430		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8431		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8432		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8433		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8434			printf("CTL:Persis Out error returned from "
8435			       "ctl_ha_msg_send %d\n", isc_retval);
8436		}
8437		break;
8438
8439	case SPRO_PREEMPT: {
8440		int nretval;
8441
8442		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8443					  residx, ctsio, cdb, param);
8444		if (nretval != 0)
8445			return (CTL_RETVAL_COMPLETE);
8446		break;
8447	}
8448	case SPRO_REG_MOVE:
8449	case SPRO_PRE_ABO:
8450	default:
8451		free(ctsio->kern_data_ptr, M_CTL);
8452		ctl_set_invalid_field(/*ctsio*/ ctsio,
8453				      /*sks_valid*/ 1,
8454				      /*command*/ 1,
8455				      /*field*/ 1,
8456				      /*bit_valid*/ 1,
8457				      /*bit*/ 0);
8458		ctl_done((union ctl_io *)ctsio);
8459		return (CTL_RETVAL_COMPLETE);
8460		break; /* NOTREACHED */
8461	}
8462
8463done:
8464	free(ctsio->kern_data_ptr, M_CTL);
8465	ctl_set_success(ctsio);
8466	ctl_done((union ctl_io *)ctsio);
8467
8468	return (retval);
8469}
8470
8471/*
8472 * This routine is for handling a message from the other SC pertaining to
8473 * persistent reserve out. All the error checking will have been done
8474 * so only perorming the action need be done here to keep the two
8475 * in sync.
8476 */
8477static void
8478ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8479{
8480	struct ctl_lun *lun;
8481	struct ctl_softc *softc;
8482	int i;
8483	uint32_t targ_lun;
8484
8485	softc = control_softc;
8486
8487	mtx_lock(&softc->ctl_lock);
8488
8489	targ_lun = msg->hdr.nexus.targ_lun;
8490	if (msg->hdr.nexus.lun_map_fn != NULL)
8491		targ_lun = msg->hdr.nexus.lun_map_fn(msg->hdr.nexus.lun_map_arg, targ_lun);
8492	lun = softc->ctl_luns[targ_lun];
8493	switch(msg->pr.pr_info.action) {
8494	case CTL_PR_REG_KEY:
8495		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8496			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8497			lun->pr_key_count++;
8498		}
8499		lun->PRGeneration++;
8500		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8501		       msg->pr.pr_info.sa_res_key,
8502		       sizeof(struct scsi_per_res_key));
8503		break;
8504
8505	case CTL_PR_UNREG_KEY:
8506		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8507		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8508		       0, sizeof(struct scsi_per_res_key));
8509		lun->pr_key_count--;
8510
8511		/* XXX Need to see if the reservation has been released */
8512		/* if so do we need to generate UA? */
8513		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8514			lun->flags &= ~CTL_LUN_PR_RESERVED;
8515			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8516
8517			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8518			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8519			 && lun->pr_key_count) {
8520				/*
8521				 * If the reservation is a registrants
8522				 * only type we need to generate a UA
8523				 * for other registered inits.  The
8524				 * sense code should be RESERVATIONS
8525				 * RELEASED
8526				 */
8527
8528				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8529					if (lun->per_res[i+
8530					    persis_offset].registered == 0)
8531						continue;
8532
8533					lun->pending_sense[i
8534						].ua_pending |=
8535						CTL_UA_RES_RELEASE;
8536				}
8537			}
8538			lun->res_type = 0;
8539		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8540			if (lun->pr_key_count==0) {
8541				lun->flags &= ~CTL_LUN_PR_RESERVED;
8542				lun->res_type = 0;
8543				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8544			}
8545		}
8546		lun->PRGeneration++;
8547		break;
8548
8549	case CTL_PR_RESERVE:
8550		lun->flags |= CTL_LUN_PR_RESERVED;
8551		lun->res_type = msg->pr.pr_info.res_type;
8552		lun->pr_res_idx = msg->pr.pr_info.residx;
8553
8554		break;
8555
8556	case CTL_PR_RELEASE:
8557		/*
8558		 * if this isn't an exclusive access res generate UA for all
8559		 * other registrants.
8560		 */
8561		if (lun->res_type != SPR_TYPE_EX_AC
8562		 && lun->res_type != SPR_TYPE_WR_EX) {
8563			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8564				if (lun->per_res[i+persis_offset].registered)
8565					lun->pending_sense[i].ua_pending |=
8566						CTL_UA_RES_RELEASE;
8567		}
8568
8569		lun->flags &= ~CTL_LUN_PR_RESERVED;
8570		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8571		lun->res_type = 0;
8572		break;
8573
8574	case CTL_PR_PREEMPT:
8575		ctl_pro_preempt_other(lun, msg);
8576		break;
8577	case CTL_PR_CLEAR:
8578		lun->flags &= ~CTL_LUN_PR_RESERVED;
8579		lun->res_type = 0;
8580		lun->pr_key_count = 0;
8581		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8582
8583		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8584			if (lun->per_res[i].registered == 0)
8585				continue;
8586			if (!persis_offset
8587			 && i < CTL_MAX_INITIATORS)
8588				lun->pending_sense[i].ua_pending |=
8589					CTL_UA_RES_PREEMPT;
8590			else if (persis_offset
8591			      && i >= persis_offset)
8592   				lun->pending_sense[i-persis_offset].ua_pending|=
8593					CTL_UA_RES_PREEMPT;
8594			memset(&lun->per_res[i].res_key, 0,
8595			       sizeof(struct scsi_per_res_key));
8596			lun->per_res[i].registered = 0;
8597		}
8598		lun->PRGeneration++;
8599		break;
8600	}
8601
8602	mtx_unlock(&softc->ctl_lock);
8603}
8604
8605int
8606ctl_read_write(struct ctl_scsiio *ctsio)
8607{
8608	struct ctl_lun *lun;
8609	struct ctl_lba_len_flags *lbalen;
8610	uint64_t lba;
8611	uint32_t num_blocks;
8612	int reladdr, fua, dpo, ebp;
8613	int retval;
8614	int isread;
8615
8616	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8617
8618	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8619
8620	reladdr = 0;
8621	fua = 0;
8622	dpo = 0;
8623	ebp = 0;
8624
8625	retval = CTL_RETVAL_COMPLETE;
8626
8627	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8628	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8629	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8630		uint32_t residx;
8631
8632		/*
8633		 * XXX KDM need a lock here.
8634		 */
8635		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8636		if ((lun->res_type == SPR_TYPE_EX_AC
8637		  && residx != lun->pr_res_idx)
8638		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8639		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8640		  && !lun->per_res[residx].registered)) {
8641			ctl_set_reservation_conflict(ctsio);
8642			ctl_done((union ctl_io *)ctsio);
8643			return (CTL_RETVAL_COMPLETE);
8644	        }
8645	}
8646
8647	switch (ctsio->cdb[0]) {
8648	case READ_6:
8649	case WRITE_6: {
8650		struct scsi_rw_6 *cdb;
8651
8652		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8653
8654		lba = scsi_3btoul(cdb->addr);
8655		/* only 5 bits are valid in the most significant address byte */
8656		lba &= 0x1fffff;
8657		num_blocks = cdb->length;
8658		/*
8659		 * This is correct according to SBC-2.
8660		 */
8661		if (num_blocks == 0)
8662			num_blocks = 256;
8663		break;
8664	}
8665	case READ_10:
8666	case WRITE_10: {
8667		struct scsi_rw_10 *cdb;
8668
8669		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8670
8671		if (cdb->byte2 & SRW10_RELADDR)
8672			reladdr = 1;
8673		if (cdb->byte2 & SRW10_FUA)
8674			fua = 1;
8675		if (cdb->byte2 & SRW10_DPO)
8676			dpo = 1;
8677
8678		if ((cdb->opcode == WRITE_10)
8679		 && (cdb->byte2 & SRW10_EBP))
8680			ebp = 1;
8681
8682		lba = scsi_4btoul(cdb->addr);
8683		num_blocks = scsi_2btoul(cdb->length);
8684		break;
8685	}
8686	case WRITE_VERIFY_10: {
8687		struct scsi_write_verify_10 *cdb;
8688
8689		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8690
8691		/*
8692		 * XXX KDM we should do actual write verify support at some
8693		 * point.  This is obviously fake, we're just translating
8694		 * things to a write.  So we don't even bother checking the
8695		 * BYTCHK field, since we don't do any verification.  If
8696		 * the user asks for it, we'll just pretend we did it.
8697		 */
8698		if (cdb->byte2 & SWV_DPO)
8699			dpo = 1;
8700
8701		lba = scsi_4btoul(cdb->addr);
8702		num_blocks = scsi_2btoul(cdb->length);
8703		break;
8704	}
8705	case READ_12:
8706	case WRITE_12: {
8707		struct scsi_rw_12 *cdb;
8708
8709		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8710
8711		if (cdb->byte2 & SRW12_RELADDR)
8712			reladdr = 1;
8713		if (cdb->byte2 & SRW12_FUA)
8714			fua = 1;
8715		if (cdb->byte2 & SRW12_DPO)
8716			dpo = 1;
8717		lba = scsi_4btoul(cdb->addr);
8718		num_blocks = scsi_4btoul(cdb->length);
8719		break;
8720	}
8721	case WRITE_VERIFY_12: {
8722		struct scsi_write_verify_12 *cdb;
8723
8724		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8725
8726		if (cdb->byte2 & SWV_DPO)
8727			dpo = 1;
8728
8729		lba = scsi_4btoul(cdb->addr);
8730		num_blocks = scsi_4btoul(cdb->length);
8731
8732		break;
8733	}
8734	case READ_16:
8735	case WRITE_16: {
8736		struct scsi_rw_16 *cdb;
8737
8738		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8739
8740		if (cdb->byte2 & SRW12_RELADDR)
8741			reladdr = 1;
8742		if (cdb->byte2 & SRW12_FUA)
8743			fua = 1;
8744		if (cdb->byte2 & SRW12_DPO)
8745			dpo = 1;
8746
8747		lba = scsi_8btou64(cdb->addr);
8748		num_blocks = scsi_4btoul(cdb->length);
8749		break;
8750	}
8751	case WRITE_VERIFY_16: {
8752		struct scsi_write_verify_16 *cdb;
8753
8754		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8755
8756		if (cdb->byte2 & SWV_DPO)
8757			dpo = 1;
8758
8759		lba = scsi_8btou64(cdb->addr);
8760		num_blocks = scsi_4btoul(cdb->length);
8761		break;
8762	}
8763	default:
8764		/*
8765		 * We got a command we don't support.  This shouldn't
8766		 * happen, commands should be filtered out above us.
8767		 */
8768		ctl_set_invalid_opcode(ctsio);
8769		ctl_done((union ctl_io *)ctsio);
8770
8771		return (CTL_RETVAL_COMPLETE);
8772		break; /* NOTREACHED */
8773	}
8774
8775	/*
8776	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8777	 * interesting for us, but if RAIDCore is in write-back mode,
8778	 * getting it to do write-through for a particular transaction may
8779	 * not be possible.
8780	 */
8781	/*
8782	 * We don't support relative addressing.  That also requires
8783	 * supporting linked commands, which we don't do.
8784	 */
8785	if (reladdr != 0) {
8786		ctl_set_invalid_field(ctsio,
8787				      /*sks_valid*/ 1,
8788				      /*command*/ 1,
8789				      /*field*/ 1,
8790				      /*bit_valid*/ 1,
8791				      /*bit*/ 0);
8792		ctl_done((union ctl_io *)ctsio);
8793		return (CTL_RETVAL_COMPLETE);
8794	}
8795
8796	/*
8797	 * The first check is to make sure we're in bounds, the second
8798	 * check is to catch wrap-around problems.  If the lba + num blocks
8799	 * is less than the lba, then we've wrapped around and the block
8800	 * range is invalid anyway.
8801	 */
8802	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8803	 || ((lba + num_blocks) < lba)) {
8804		ctl_set_lba_out_of_range(ctsio);
8805		ctl_done((union ctl_io *)ctsio);
8806		return (CTL_RETVAL_COMPLETE);
8807	}
8808
8809	/*
8810	 * According to SBC-3, a transfer length of 0 is not an error.
8811	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8812	 * translates to 256 blocks for those commands.
8813	 */
8814	if (num_blocks == 0) {
8815		ctl_set_success(ctsio);
8816		ctl_done((union ctl_io *)ctsio);
8817		return (CTL_RETVAL_COMPLETE);
8818	}
8819
8820	lbalen = (struct ctl_lba_len_flags *)
8821	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8822	lbalen->lba = lba;
8823	lbalen->len = num_blocks;
8824	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8825
8826	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8827	ctsio->kern_rel_offset = 0;
8828
8829	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8830
8831	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8832
8833	return (retval);
8834}
8835
8836static int
8837ctl_cnw_cont(union ctl_io *io)
8838{
8839	struct ctl_scsiio *ctsio;
8840	struct ctl_lun *lun;
8841	struct ctl_lba_len_flags *lbalen;
8842	int retval;
8843
8844	ctsio = &io->scsiio;
8845	ctsio->io_hdr.status = CTL_STATUS_NONE;
8846	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8847	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8848	lbalen = (struct ctl_lba_len_flags *)
8849	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8850	lbalen->flags = CTL_LLF_WRITE;
8851
8852	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8853	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8854	return (retval);
8855}
8856
8857int
8858ctl_cnw(struct ctl_scsiio *ctsio)
8859{
8860	struct ctl_lun *lun;
8861	struct ctl_lba_len_flags *lbalen;
8862	uint64_t lba;
8863	uint32_t num_blocks;
8864	int fua, dpo;
8865	int retval;
8866
8867	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8868
8869	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8870
8871	fua = 0;
8872	dpo = 0;
8873
8874	retval = CTL_RETVAL_COMPLETE;
8875
8876	switch (ctsio->cdb[0]) {
8877	case COMPARE_AND_WRITE: {
8878		struct scsi_compare_and_write *cdb;
8879
8880		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8881
8882		if (cdb->byte2 & SRW10_FUA)
8883			fua = 1;
8884		if (cdb->byte2 & SRW10_DPO)
8885			dpo = 1;
8886		lba = scsi_8btou64(cdb->addr);
8887		num_blocks = cdb->length;
8888		break;
8889	}
8890	default:
8891		/*
8892		 * We got a command we don't support.  This shouldn't
8893		 * happen, commands should be filtered out above us.
8894		 */
8895		ctl_set_invalid_opcode(ctsio);
8896		ctl_done((union ctl_io *)ctsio);
8897
8898		return (CTL_RETVAL_COMPLETE);
8899		break; /* NOTREACHED */
8900	}
8901
8902	/*
8903	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8904	 * interesting for us, but if RAIDCore is in write-back mode,
8905	 * getting it to do write-through for a particular transaction may
8906	 * not be possible.
8907	 */
8908
8909	/*
8910	 * The first check is to make sure we're in bounds, the second
8911	 * check is to catch wrap-around problems.  If the lba + num blocks
8912	 * is less than the lba, then we've wrapped around and the block
8913	 * range is invalid anyway.
8914	 */
8915	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8916	 || ((lba + num_blocks) < lba)) {
8917		ctl_set_lba_out_of_range(ctsio);
8918		ctl_done((union ctl_io *)ctsio);
8919		return (CTL_RETVAL_COMPLETE);
8920	}
8921
8922	/*
8923	 * According to SBC-3, a transfer length of 0 is not an error.
8924	 */
8925	if (num_blocks == 0) {
8926		ctl_set_success(ctsio);
8927		ctl_done((union ctl_io *)ctsio);
8928		return (CTL_RETVAL_COMPLETE);
8929	}
8930
8931	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8932	ctsio->kern_rel_offset = 0;
8933
8934	/*
8935	 * Set the IO_CONT flag, so that if this I/O gets passed to
8936	 * ctl_data_submit_done(), it'll get passed back to
8937	 * ctl_ctl_cnw_cont() for further processing.
8938	 */
8939	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8940	ctsio->io_cont = ctl_cnw_cont;
8941
8942	lbalen = (struct ctl_lba_len_flags *)
8943	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8944	lbalen->lba = lba;
8945	lbalen->len = num_blocks;
8946	lbalen->flags = CTL_LLF_COMPARE;
8947
8948	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8949	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8950	return (retval);
8951}
8952
8953int
8954ctl_verify(struct ctl_scsiio *ctsio)
8955{
8956	struct ctl_lun *lun;
8957	struct ctl_lba_len_flags *lbalen;
8958	uint64_t lba;
8959	uint32_t num_blocks;
8960	int bytchk, dpo;
8961	int retval;
8962
8963	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8964
8965	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8966
8967	bytchk = 0;
8968	dpo = 0;
8969	retval = CTL_RETVAL_COMPLETE;
8970
8971	switch (ctsio->cdb[0]) {
8972	case VERIFY_10: {
8973		struct scsi_verify_10 *cdb;
8974
8975		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8976		if (cdb->byte2 & SVFY_BYTCHK)
8977			bytchk = 1;
8978		if (cdb->byte2 & SVFY_DPO)
8979			dpo = 1;
8980		lba = scsi_4btoul(cdb->addr);
8981		num_blocks = scsi_2btoul(cdb->length);
8982		break;
8983	}
8984	case VERIFY_12: {
8985		struct scsi_verify_12 *cdb;
8986
8987		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8988		if (cdb->byte2 & SVFY_BYTCHK)
8989			bytchk = 1;
8990		if (cdb->byte2 & SVFY_DPO)
8991			dpo = 1;
8992		lba = scsi_4btoul(cdb->addr);
8993		num_blocks = scsi_4btoul(cdb->length);
8994		break;
8995	}
8996	case VERIFY_16: {
8997		struct scsi_rw_16 *cdb;
8998
8999		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9000		if (cdb->byte2 & SVFY_BYTCHK)
9001			bytchk = 1;
9002		if (cdb->byte2 & SVFY_DPO)
9003			dpo = 1;
9004		lba = scsi_8btou64(cdb->addr);
9005		num_blocks = scsi_4btoul(cdb->length);
9006		break;
9007	}
9008	default:
9009		/*
9010		 * We got a command we don't support.  This shouldn't
9011		 * happen, commands should be filtered out above us.
9012		 */
9013		ctl_set_invalid_opcode(ctsio);
9014		ctl_done((union ctl_io *)ctsio);
9015		return (CTL_RETVAL_COMPLETE);
9016	}
9017
9018	/*
9019	 * The first check is to make sure we're in bounds, the second
9020	 * check is to catch wrap-around problems.  If the lba + num blocks
9021	 * is less than the lba, then we've wrapped around and the block
9022	 * range is invalid anyway.
9023	 */
9024	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9025	 || ((lba + num_blocks) < lba)) {
9026		ctl_set_lba_out_of_range(ctsio);
9027		ctl_done((union ctl_io *)ctsio);
9028		return (CTL_RETVAL_COMPLETE);
9029	}
9030
9031	/*
9032	 * According to SBC-3, a transfer length of 0 is not an error.
9033	 */
9034	if (num_blocks == 0) {
9035		ctl_set_success(ctsio);
9036		ctl_done((union ctl_io *)ctsio);
9037		return (CTL_RETVAL_COMPLETE);
9038	}
9039
9040	lbalen = (struct ctl_lba_len_flags *)
9041	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9042	lbalen->lba = lba;
9043	lbalen->len = num_blocks;
9044	if (bytchk) {
9045		lbalen->flags = CTL_LLF_COMPARE;
9046		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9047	} else {
9048		lbalen->flags = CTL_LLF_VERIFY;
9049		ctsio->kern_total_len = 0;
9050	}
9051	ctsio->kern_rel_offset = 0;
9052
9053	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9054	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9055	return (retval);
9056}
9057
9058int
9059ctl_report_luns(struct ctl_scsiio *ctsio)
9060{
9061	struct scsi_report_luns *cdb;
9062	struct scsi_report_luns_data *lun_data;
9063	struct ctl_lun *lun, *request_lun;
9064	int num_luns, retval;
9065	uint32_t alloc_len, lun_datalen;
9066	int num_filled, well_known;
9067	uint32_t initidx, targ_lun_id, lun_id;
9068
9069	retval = CTL_RETVAL_COMPLETE;
9070	well_known = 0;
9071
9072	cdb = (struct scsi_report_luns *)ctsio->cdb;
9073
9074	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9075
9076	mtx_lock(&control_softc->ctl_lock);
9077	num_luns = control_softc->num_luns;
9078	mtx_unlock(&control_softc->ctl_lock);
9079
9080	switch (cdb->select_report) {
9081	case RPL_REPORT_DEFAULT:
9082	case RPL_REPORT_ALL:
9083		break;
9084	case RPL_REPORT_WELLKNOWN:
9085		well_known = 1;
9086		num_luns = 0;
9087		break;
9088	default:
9089		ctl_set_invalid_field(ctsio,
9090				      /*sks_valid*/ 1,
9091				      /*command*/ 1,
9092				      /*field*/ 2,
9093				      /*bit_valid*/ 0,
9094				      /*bit*/ 0);
9095		ctl_done((union ctl_io *)ctsio);
9096		return (retval);
9097		break; /* NOTREACHED */
9098	}
9099
9100	alloc_len = scsi_4btoul(cdb->length);
9101	/*
9102	 * The initiator has to allocate at least 16 bytes for this request,
9103	 * so he can at least get the header and the first LUN.  Otherwise
9104	 * we reject the request (per SPC-3 rev 14, section 6.21).
9105	 */
9106	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9107	    sizeof(struct scsi_report_luns_lundata))) {
9108		ctl_set_invalid_field(ctsio,
9109				      /*sks_valid*/ 1,
9110				      /*command*/ 1,
9111				      /*field*/ 6,
9112				      /*bit_valid*/ 0,
9113				      /*bit*/ 0);
9114		ctl_done((union ctl_io *)ctsio);
9115		return (retval);
9116	}
9117
9118	request_lun = (struct ctl_lun *)
9119		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9120
9121	lun_datalen = sizeof(*lun_data) +
9122		(num_luns * sizeof(struct scsi_report_luns_lundata));
9123
9124	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9125	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9126	ctsio->kern_sg_entries = 0;
9127
9128	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9129
9130	mtx_lock(&control_softc->ctl_lock);
9131	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9132		lun_id = targ_lun_id;
9133		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9134			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9135		if (lun_id >= CTL_MAX_LUNS)
9136			continue;
9137		lun = control_softc->ctl_luns[lun_id];
9138		if (lun == NULL)
9139			continue;
9140
9141		if (targ_lun_id <= 0xff) {
9142			/*
9143			 * Peripheral addressing method, bus number 0.
9144			 */
9145			lun_data->luns[num_filled].lundata[0] =
9146				RPL_LUNDATA_ATYP_PERIPH;
9147			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9148			num_filled++;
9149		} else if (targ_lun_id <= 0x3fff) {
9150			/*
9151			 * Flat addressing method.
9152			 */
9153			lun_data->luns[num_filled].lundata[0] =
9154				RPL_LUNDATA_ATYP_FLAT |
9155				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9156#ifdef OLDCTLHEADERS
9157				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9158				(targ_lun_id & SRLD_BUS_LUN_MASK);
9159#endif
9160			lun_data->luns[num_filled].lundata[1] =
9161#ifdef OLDCTLHEADERS
9162				targ_lun_id >> SRLD_BUS_LUN_BITS;
9163#endif
9164				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9165			num_filled++;
9166		} else {
9167			printf("ctl_report_luns: bogus LUN number %jd, "
9168			       "skipping\n", (intmax_t)targ_lun_id);
9169		}
9170		/*
9171		 * According to SPC-3, rev 14 section 6.21:
9172		 *
9173		 * "The execution of a REPORT LUNS command to any valid and
9174		 * installed logical unit shall clear the REPORTED LUNS DATA
9175		 * HAS CHANGED unit attention condition for all logical
9176		 * units of that target with respect to the requesting
9177		 * initiator. A valid and installed logical unit is one
9178		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9179		 * INQUIRY data (see 6.4.2)."
9180		 *
9181		 * If request_lun is NULL, the LUN this report luns command
9182		 * was issued to is either disabled or doesn't exist. In that
9183		 * case, we shouldn't clear any pending lun change unit
9184		 * attention.
9185		 */
9186		if (request_lun != NULL)
9187			lun->pending_sense[initidx].ua_pending &=
9188				~CTL_UA_LUN_CHANGE;
9189	}
9190	mtx_unlock(&control_softc->ctl_lock);
9191
9192	/*
9193	 * It's quite possible that we've returned fewer LUNs than we allocated
9194	 * space for.  Trim it.
9195	 */
9196	lun_datalen = sizeof(*lun_data) +
9197		(num_filled * sizeof(struct scsi_report_luns_lundata));
9198
9199	if (lun_datalen < alloc_len) {
9200		ctsio->residual = alloc_len - lun_datalen;
9201		ctsio->kern_data_len = lun_datalen;
9202		ctsio->kern_total_len = lun_datalen;
9203	} else {
9204		ctsio->residual = 0;
9205		ctsio->kern_data_len = alloc_len;
9206		ctsio->kern_total_len = alloc_len;
9207	}
9208	ctsio->kern_data_resid = 0;
9209	ctsio->kern_rel_offset = 0;
9210	ctsio->kern_sg_entries = 0;
9211
9212	/*
9213	 * We set this to the actual data length, regardless of how much
9214	 * space we actually have to return results.  If the user looks at
9215	 * this value, he'll know whether or not he allocated enough space
9216	 * and reissue the command if necessary.  We don't support well
9217	 * known logical units, so if the user asks for that, return none.
9218	 */
9219	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9220
9221	/*
9222	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9223	 * this request.
9224	 */
9225	ctsio->scsi_status = SCSI_STATUS_OK;
9226
9227	ctsio->be_move_done = ctl_config_move_done;
9228	ctl_datamove((union ctl_io *)ctsio);
9229
9230	return (retval);
9231}
9232
9233int
9234ctl_request_sense(struct ctl_scsiio *ctsio)
9235{
9236	struct scsi_request_sense *cdb;
9237	struct scsi_sense_data *sense_ptr;
9238	struct ctl_lun *lun;
9239	uint32_t initidx;
9240	int have_error;
9241	scsi_sense_data_type sense_format;
9242
9243	cdb = (struct scsi_request_sense *)ctsio->cdb;
9244
9245	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9246
9247	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9248
9249	/*
9250	 * Determine which sense format the user wants.
9251	 */
9252	if (cdb->byte2 & SRS_DESC)
9253		sense_format = SSD_TYPE_DESC;
9254	else
9255		sense_format = SSD_TYPE_FIXED;
9256
9257	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9258	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9259	ctsio->kern_sg_entries = 0;
9260
9261	/*
9262	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9263	 * larger than the largest allowed value for the length field in the
9264	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9265	 */
9266	ctsio->residual = 0;
9267	ctsio->kern_data_len = cdb->length;
9268	ctsio->kern_total_len = cdb->length;
9269
9270	ctsio->kern_data_resid = 0;
9271	ctsio->kern_rel_offset = 0;
9272	ctsio->kern_sg_entries = 0;
9273
9274	/*
9275	 * If we don't have a LUN, we don't have any pending sense.
9276	 */
9277	if (lun == NULL)
9278		goto no_sense;
9279
9280	have_error = 0;
9281	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9282	/*
9283	 * Check for pending sense, and then for pending unit attentions.
9284	 * Pending sense gets returned first, then pending unit attentions.
9285	 */
9286	mtx_lock(&lun->ctl_softc->ctl_lock);
9287	if (ctl_is_set(lun->have_ca, initidx)) {
9288		scsi_sense_data_type stored_format;
9289
9290		/*
9291		 * Check to see which sense format was used for the stored
9292		 * sense data.
9293		 */
9294		stored_format = scsi_sense_type(
9295		    &lun->pending_sense[initidx].sense);
9296
9297		/*
9298		 * If the user requested a different sense format than the
9299		 * one we stored, then we need to convert it to the other
9300		 * format.  If we're going from descriptor to fixed format
9301		 * sense data, we may lose things in translation, depending
9302		 * on what options were used.
9303		 *
9304		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9305		 * for some reason we'll just copy it out as-is.
9306		 */
9307		if ((stored_format == SSD_TYPE_FIXED)
9308		 && (sense_format == SSD_TYPE_DESC))
9309			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9310			    &lun->pending_sense[initidx].sense,
9311			    (struct scsi_sense_data_desc *)sense_ptr);
9312		else if ((stored_format == SSD_TYPE_DESC)
9313		      && (sense_format == SSD_TYPE_FIXED))
9314			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9315			    &lun->pending_sense[initidx].sense,
9316			    (struct scsi_sense_data_fixed *)sense_ptr);
9317		else
9318			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9319			       ctl_min(sizeof(*sense_ptr),
9320			       sizeof(lun->pending_sense[initidx].sense)));
9321
9322		ctl_clear_mask(lun->have_ca, initidx);
9323		have_error = 1;
9324	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9325		ctl_ua_type ua_type;
9326
9327		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9328				       sense_ptr, sense_format);
9329		if (ua_type != CTL_UA_NONE) {
9330			have_error = 1;
9331			/* We're reporting this UA, so clear it */
9332			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9333		}
9334	}
9335	mtx_unlock(&lun->ctl_softc->ctl_lock);
9336
9337	/*
9338	 * We already have a pending error, return it.
9339	 */
9340	if (have_error != 0) {
9341		/*
9342		 * We report the SCSI status as OK, since the status of the
9343		 * request sense command itself is OK.
9344		 */
9345		ctsio->scsi_status = SCSI_STATUS_OK;
9346
9347		/*
9348		 * We report 0 for the sense length, because we aren't doing
9349		 * autosense in this case.  We're reporting sense as
9350		 * parameter data.
9351		 */
9352		ctsio->sense_len = 0;
9353
9354		ctsio->be_move_done = ctl_config_move_done;
9355		ctl_datamove((union ctl_io *)ctsio);
9356
9357		return (CTL_RETVAL_COMPLETE);
9358	}
9359
9360no_sense:
9361
9362	/*
9363	 * No sense information to report, so we report that everything is
9364	 * okay.
9365	 */
9366	ctl_set_sense_data(sense_ptr,
9367			   lun,
9368			   sense_format,
9369			   /*current_error*/ 1,
9370			   /*sense_key*/ SSD_KEY_NO_SENSE,
9371			   /*asc*/ 0x00,
9372			   /*ascq*/ 0x00,
9373			   SSD_ELEM_NONE);
9374
9375	ctsio->scsi_status = SCSI_STATUS_OK;
9376
9377	/*
9378	 * We report 0 for the sense length, because we aren't doing
9379	 * autosense in this case.  We're reporting sense as parameter data.
9380	 */
9381	ctsio->sense_len = 0;
9382	ctsio->be_move_done = ctl_config_move_done;
9383	ctl_datamove((union ctl_io *)ctsio);
9384
9385	return (CTL_RETVAL_COMPLETE);
9386}
9387
9388int
9389ctl_tur(struct ctl_scsiio *ctsio)
9390{
9391	struct ctl_lun *lun;
9392
9393	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9394
9395	CTL_DEBUG_PRINT(("ctl_tur\n"));
9396
9397	if (lun == NULL)
9398		return (-EINVAL);
9399
9400	ctsio->scsi_status = SCSI_STATUS_OK;
9401	ctsio->io_hdr.status = CTL_SUCCESS;
9402
9403	ctl_done((union ctl_io *)ctsio);
9404
9405	return (CTL_RETVAL_COMPLETE);
9406}
9407
9408#ifdef notyet
9409static int
9410ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9411{
9412
9413}
9414#endif
9415
9416static int
9417ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9418{
9419	struct scsi_vpd_supported_pages *pages;
9420	int sup_page_size;
9421	struct ctl_lun *lun;
9422
9423	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9424
9425	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9426	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9427	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9428	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9429	ctsio->kern_sg_entries = 0;
9430
9431	if (sup_page_size < alloc_len) {
9432		ctsio->residual = alloc_len - sup_page_size;
9433		ctsio->kern_data_len = sup_page_size;
9434		ctsio->kern_total_len = sup_page_size;
9435	} else {
9436		ctsio->residual = 0;
9437		ctsio->kern_data_len = alloc_len;
9438		ctsio->kern_total_len = alloc_len;
9439	}
9440	ctsio->kern_data_resid = 0;
9441	ctsio->kern_rel_offset = 0;
9442	ctsio->kern_sg_entries = 0;
9443
9444	/*
9445	 * The control device is always connected.  The disk device, on the
9446	 * other hand, may not be online all the time.  Need to change this
9447	 * to figure out whether the disk device is actually online or not.
9448	 */
9449	if (lun != NULL)
9450		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9451				lun->be_lun->lun_type;
9452	else
9453		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9454
9455	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9456	/* Supported VPD pages */
9457	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9458	/* Serial Number */
9459	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9460	/* Device Identification */
9461	pages->page_list[2] = SVPD_DEVICE_ID;
9462	/* Block limits */
9463	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9464	/* Logical Block Provisioning */
9465	pages->page_list[4] = SVPD_LBP;
9466
9467	ctsio->scsi_status = SCSI_STATUS_OK;
9468
9469	ctsio->be_move_done = ctl_config_move_done;
9470	ctl_datamove((union ctl_io *)ctsio);
9471
9472	return (CTL_RETVAL_COMPLETE);
9473}
9474
9475static int
9476ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9477{
9478	struct scsi_vpd_unit_serial_number *sn_ptr;
9479	struct ctl_lun *lun;
9480#ifndef CTL_USE_BACKEND_SN
9481	char tmpstr[32];
9482#endif
9483
9484	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9485
9486	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9487	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9488	ctsio->kern_sg_entries = 0;
9489
9490	if (sizeof(*sn_ptr) < alloc_len) {
9491		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9492		ctsio->kern_data_len = sizeof(*sn_ptr);
9493		ctsio->kern_total_len = sizeof(*sn_ptr);
9494	} else {
9495		ctsio->residual = 0;
9496		ctsio->kern_data_len = alloc_len;
9497		ctsio->kern_total_len = alloc_len;
9498	}
9499	ctsio->kern_data_resid = 0;
9500	ctsio->kern_rel_offset = 0;
9501	ctsio->kern_sg_entries = 0;
9502
9503	/*
9504	 * The control device is always connected.  The disk device, on the
9505	 * other hand, may not be online all the time.  Need to change this
9506	 * to figure out whether the disk device is actually online or not.
9507	 */
9508	if (lun != NULL)
9509		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9510				  lun->be_lun->lun_type;
9511	else
9512		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9513
9514	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9515	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9516#ifdef CTL_USE_BACKEND_SN
9517	/*
9518	 * If we don't have a LUN, we just leave the serial number as
9519	 * all spaces.
9520	 */
9521	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9522	if (lun != NULL) {
9523		strncpy((char *)sn_ptr->serial_num,
9524			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9525	}
9526#else
9527	/*
9528	 * Note that we're using a non-unique serial number here,
9529	 */
9530	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9531	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9532	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9533		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9534#endif
9535	ctsio->scsi_status = SCSI_STATUS_OK;
9536
9537	ctsio->be_move_done = ctl_config_move_done;
9538	ctl_datamove((union ctl_io *)ctsio);
9539
9540	return (CTL_RETVAL_COMPLETE);
9541}
9542
9543
9544static int
9545ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9546{
9547	struct scsi_vpd_device_id *devid_ptr;
9548	struct scsi_vpd_id_descriptor *desc, *desc1;
9549	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9550	struct scsi_vpd_id_t10 *t10id;
9551	struct ctl_softc *ctl_softc;
9552	struct ctl_lun *lun;
9553	struct ctl_frontend *fe;
9554	char *val;
9555#ifndef CTL_USE_BACKEND_SN
9556	char tmpstr[32];
9557#endif /* CTL_USE_BACKEND_SN */
9558	int devid_len;
9559
9560	ctl_softc = control_softc;
9561
9562	mtx_lock(&ctl_softc->ctl_lock);
9563	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9564	mtx_unlock(&ctl_softc->ctl_lock);
9565
9566	if (fe->devid != NULL)
9567		return ((fe->devid)(ctsio, alloc_len));
9568
9569	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9570
9571	devid_len = sizeof(struct scsi_vpd_device_id) +
9572		sizeof(struct scsi_vpd_id_descriptor) +
9573		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9574		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9575		sizeof(struct scsi_vpd_id_descriptor) +
9576		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9577		sizeof(struct scsi_vpd_id_descriptor) +
9578		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9579
9580	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9581	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9582	ctsio->kern_sg_entries = 0;
9583
9584	if (devid_len < alloc_len) {
9585		ctsio->residual = alloc_len - devid_len;
9586		ctsio->kern_data_len = devid_len;
9587		ctsio->kern_total_len = devid_len;
9588	} else {
9589		ctsio->residual = 0;
9590		ctsio->kern_data_len = alloc_len;
9591		ctsio->kern_total_len = alloc_len;
9592	}
9593	ctsio->kern_data_resid = 0;
9594	ctsio->kern_rel_offset = 0;
9595	ctsio->kern_sg_entries = 0;
9596
9597	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9598	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9599	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9600		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9601	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9602	          CTL_WWPN_LEN);
9603	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9604	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9605
9606	/*
9607	 * The control device is always connected.  The disk device, on the
9608	 * other hand, may not be online all the time.
9609	 */
9610	if (lun != NULL)
9611		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9612				     lun->be_lun->lun_type;
9613	else
9614		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9615
9616	devid_ptr->page_code = SVPD_DEVICE_ID;
9617
9618	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9619
9620	mtx_lock(&ctl_softc->ctl_lock);
9621
9622	/*
9623	 * For Fibre channel,
9624	 */
9625	if (fe->port_type == CTL_PORT_FC)
9626	{
9627		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9628				      SVPD_ID_CODESET_ASCII;
9629        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9630		              SVPD_ID_CODESET_BINARY;
9631	}
9632	else
9633	{
9634		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9635				      SVPD_ID_CODESET_ASCII;
9636        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9637		              SVPD_ID_CODESET_BINARY;
9638	}
9639	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9640	mtx_unlock(&ctl_softc->ctl_lock);
9641
9642	/*
9643	 * We're using a LUN association here.  i.e., this device ID is a
9644	 * per-LUN identifier.
9645	 */
9646	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9647	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9648	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9649		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9650	} else {
9651		memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9652		strncpy(t10id->vendor, val,
9653		    min(sizeof(t10id->vendor), strlen(val)));
9654	}
9655
9656	/*
9657	 * desc1 is for the WWPN which is a port asscociation.
9658	 */
9659	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9660	desc1->length = CTL_WWPN_LEN;
9661	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9662	/* For testing just create the WWPN */
9663#if 0
9664	ddb_GetWWNN((char *)desc1->identifier);
9665
9666	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9667	/* This is so Copancontrol will return something sane */
9668	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9669	    ctsio->io_hdr.nexus.targ_port!=8)
9670		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9671	else
9672		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9673#endif
9674
9675	be64enc(desc1->identifier, fe->wwpn);
9676
9677	/*
9678	 * desc2 is for the Relative Target Port(type 4h) identifier
9679	 */
9680	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9681	                 | SVPD_ID_TYPE_RELTARG;
9682	desc2->length = 4;
9683//#if 0
9684	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9685	/* This is so Copancontrol will return something sane */
9686	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9687	    ctsio->io_hdr.nexus.targ_port!=8)
9688		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9689	else
9690	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9691//#endif
9692
9693	/*
9694	 * desc3 is for the Target Port Group(type 5h) identifier
9695	 */
9696	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9697	                 | SVPD_ID_TYPE_TPORTGRP;
9698	desc3->length = 4;
9699	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9700		desc3->identifier[3] = 1;
9701	else
9702		desc3->identifier[3] = 2;
9703
9704#ifdef CTL_USE_BACKEND_SN
9705	/*
9706	 * If we've actually got a backend, copy the device id from the
9707	 * per-LUN data.  Otherwise, set it to all spaces.
9708	 */
9709	if (lun != NULL) {
9710		/*
9711		 * Copy the backend's LUN ID.
9712		 */
9713		strncpy((char *)t10id->vendor_spec_id,
9714			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9715	} else {
9716		/*
9717		 * No backend, set this to spaces.
9718		 */
9719		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9720	}
9721#else
9722	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9723		 (lun != NULL) ?  (int)lun->lun : 0);
9724	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9725		sizeof(tmpstr)));
9726#endif
9727
9728	ctsio->scsi_status = SCSI_STATUS_OK;
9729
9730	ctsio->be_move_done = ctl_config_move_done;
9731	ctl_datamove((union ctl_io *)ctsio);
9732
9733	return (CTL_RETVAL_COMPLETE);
9734}
9735
9736static int
9737ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9738{
9739	struct scsi_vpd_block_limits *bl_ptr;
9740	struct ctl_lun *lun;
9741	int bs;
9742
9743	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9744	bs = lun->be_lun->blocksize;
9745
9746	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9747	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9748	ctsio->kern_sg_entries = 0;
9749
9750	if (sizeof(*bl_ptr) < alloc_len) {
9751		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9752		ctsio->kern_data_len = sizeof(*bl_ptr);
9753		ctsio->kern_total_len = sizeof(*bl_ptr);
9754	} else {
9755		ctsio->residual = 0;
9756		ctsio->kern_data_len = alloc_len;
9757		ctsio->kern_total_len = alloc_len;
9758	}
9759	ctsio->kern_data_resid = 0;
9760	ctsio->kern_rel_offset = 0;
9761	ctsio->kern_sg_entries = 0;
9762
9763	/*
9764	 * The control device is always connected.  The disk device, on the
9765	 * other hand, may not be online all the time.  Need to change this
9766	 * to figure out whether the disk device is actually online or not.
9767	 */
9768	if (lun != NULL)
9769		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9770				  lun->be_lun->lun_type;
9771	else
9772		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9773
9774	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9775	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9776	bl_ptr->max_cmp_write_len = 0xff;
9777	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9778	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9779	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9780		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9781		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9782	}
9783	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9784
9785	ctsio->scsi_status = SCSI_STATUS_OK;
9786	ctsio->be_move_done = ctl_config_move_done;
9787	ctl_datamove((union ctl_io *)ctsio);
9788
9789	return (CTL_RETVAL_COMPLETE);
9790}
9791
9792static int
9793ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9794{
9795	struct scsi_vpd_logical_block_prov *lbp_ptr;
9796	struct ctl_lun *lun;
9797	int bs;
9798
9799	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9800	bs = lun->be_lun->blocksize;
9801
9802	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9803	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9804	ctsio->kern_sg_entries = 0;
9805
9806	if (sizeof(*lbp_ptr) < alloc_len) {
9807		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9808		ctsio->kern_data_len = sizeof(*lbp_ptr);
9809		ctsio->kern_total_len = sizeof(*lbp_ptr);
9810	} else {
9811		ctsio->residual = 0;
9812		ctsio->kern_data_len = alloc_len;
9813		ctsio->kern_total_len = alloc_len;
9814	}
9815	ctsio->kern_data_resid = 0;
9816	ctsio->kern_rel_offset = 0;
9817	ctsio->kern_sg_entries = 0;
9818
9819	/*
9820	 * The control device is always connected.  The disk device, on the
9821	 * other hand, may not be online all the time.  Need to change this
9822	 * to figure out whether the disk device is actually online or not.
9823	 */
9824	if (lun != NULL)
9825		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9826				  lun->be_lun->lun_type;
9827	else
9828		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9829
9830	lbp_ptr->page_code = SVPD_LBP;
9831	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9832		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9833
9834	ctsio->scsi_status = SCSI_STATUS_OK;
9835	ctsio->be_move_done = ctl_config_move_done;
9836	ctl_datamove((union ctl_io *)ctsio);
9837
9838	return (CTL_RETVAL_COMPLETE);
9839}
9840
9841static int
9842ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9843{
9844	struct scsi_inquiry *cdb;
9845	struct ctl_lun *lun;
9846	int alloc_len, retval;
9847
9848	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9849	cdb = (struct scsi_inquiry *)ctsio->cdb;
9850
9851	retval = CTL_RETVAL_COMPLETE;
9852
9853	alloc_len = scsi_2btoul(cdb->length);
9854
9855	switch (cdb->page_code) {
9856	case SVPD_SUPPORTED_PAGES:
9857		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9858		break;
9859	case SVPD_UNIT_SERIAL_NUMBER:
9860		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9861		break;
9862	case SVPD_DEVICE_ID:
9863		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9864		break;
9865	case SVPD_BLOCK_LIMITS:
9866		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9867		break;
9868	case SVPD_LBP:
9869		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9870		break;
9871	default:
9872		ctl_set_invalid_field(ctsio,
9873				      /*sks_valid*/ 1,
9874				      /*command*/ 1,
9875				      /*field*/ 2,
9876				      /*bit_valid*/ 0,
9877				      /*bit*/ 0);
9878		ctl_done((union ctl_io *)ctsio);
9879		retval = CTL_RETVAL_COMPLETE;
9880		break;
9881	}
9882
9883	return (retval);
9884}
9885
9886static int
9887ctl_inquiry_std(struct ctl_scsiio *ctsio)
9888{
9889	struct scsi_inquiry_data *inq_ptr;
9890	struct scsi_inquiry *cdb;
9891	struct ctl_softc *ctl_softc;
9892	struct ctl_lun *lun;
9893	char *val;
9894	uint32_t alloc_len;
9895	int is_fc;
9896
9897	ctl_softc = control_softc;
9898
9899	/*
9900	 * Figure out whether we're talking to a Fibre Channel port or not.
9901	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9902	 * SCSI front ends.
9903	 */
9904	mtx_lock(&ctl_softc->ctl_lock);
9905	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9906	    CTL_PORT_FC)
9907		is_fc = 0;
9908	else
9909		is_fc = 1;
9910	mtx_unlock(&ctl_softc->ctl_lock);
9911
9912	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9913	cdb = (struct scsi_inquiry *)ctsio->cdb;
9914	alloc_len = scsi_2btoul(cdb->length);
9915
9916	/*
9917	 * We malloc the full inquiry data size here and fill it
9918	 * in.  If the user only asks for less, we'll give him
9919	 * that much.
9920	 */
9921	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9922	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9923	ctsio->kern_sg_entries = 0;
9924	ctsio->kern_data_resid = 0;
9925	ctsio->kern_rel_offset = 0;
9926
9927	if (sizeof(*inq_ptr) < alloc_len) {
9928		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9929		ctsio->kern_data_len = sizeof(*inq_ptr);
9930		ctsio->kern_total_len = sizeof(*inq_ptr);
9931	} else {
9932		ctsio->residual = 0;
9933		ctsio->kern_data_len = alloc_len;
9934		ctsio->kern_total_len = alloc_len;
9935	}
9936
9937	/*
9938	 * If we have a LUN configured, report it as connected.  Otherwise,
9939	 * report that it is offline or no device is supported, depending
9940	 * on the value of inquiry_pq_no_lun.
9941	 *
9942	 * According to the spec (SPC-4 r34), the peripheral qualifier
9943	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9944	 *
9945	 * "A peripheral device having the specified peripheral device type
9946	 * is not connected to this logical unit. However, the device
9947	 * server is capable of supporting the specified peripheral device
9948	 * type on this logical unit."
9949	 *
9950	 * According to the same spec, the peripheral qualifier
9951	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9952	 *
9953	 * "The device server is not capable of supporting a peripheral
9954	 * device on this logical unit. For this peripheral qualifier the
9955	 * peripheral device type shall be set to 1Fh. All other peripheral
9956	 * device type values are reserved for this peripheral qualifier."
9957	 *
9958	 * Given the text, it would seem that we probably want to report that
9959	 * the LUN is offline here.  There is no LUN connected, but we can
9960	 * support a LUN at the given LUN number.
9961	 *
9962	 * In the real world, though, it sounds like things are a little
9963	 * different:
9964	 *
9965	 * - Linux, when presented with a LUN with the offline peripheral
9966	 *   qualifier, will create an sg driver instance for it.  So when
9967	 *   you attach it to CTL, you wind up with a ton of sg driver
9968	 *   instances.  (One for every LUN that Linux bothered to probe.)
9969	 *   Linux does this despite the fact that it issues a REPORT LUNs
9970	 *   to LUN 0 to get the inventory of supported LUNs.
9971	 *
9972	 * - There is other anecdotal evidence (from Emulex folks) about
9973	 *   arrays that use the offline peripheral qualifier for LUNs that
9974	 *   are on the "passive" path in an active/passive array.
9975	 *
9976	 * So the solution is provide a hopefully reasonable default
9977	 * (return bad/no LUN) and allow the user to change the behavior
9978	 * with a tunable/sysctl variable.
9979	 */
9980	if (lun != NULL)
9981		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9982				  lun->be_lun->lun_type;
9983	else if (ctl_softc->inquiry_pq_no_lun == 0)
9984		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9985	else
9986		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9987
9988	/* RMB in byte 2 is 0 */
9989	inq_ptr->version = SCSI_REV_SPC3;
9990
9991	/*
9992	 * According to SAM-3, even if a device only supports a single
9993	 * level of LUN addressing, it should still set the HISUP bit:
9994	 *
9995	 * 4.9.1 Logical unit numbers overview
9996	 *
9997	 * All logical unit number formats described in this standard are
9998	 * hierarchical in structure even when only a single level in that
9999	 * hierarchy is used. The HISUP bit shall be set to one in the
10000	 * standard INQUIRY data (see SPC-2) when any logical unit number
10001	 * format described in this standard is used.  Non-hierarchical
10002	 * formats are outside the scope of this standard.
10003	 *
10004	 * Therefore we set the HiSup bit here.
10005	 *
10006	 * The reponse format is 2, per SPC-3.
10007	 */
10008	inq_ptr->response_format = SID_HiSup | 2;
10009
10010	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10011	CTL_DEBUG_PRINT(("additional_length = %d\n",
10012			 inq_ptr->additional_length));
10013
10014	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10015	/* 16 bit addressing */
10016	if (is_fc == 0)
10017		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10018	/* XXX set the SID_MultiP bit here if we're actually going to
10019	   respond on multiple ports */
10020	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10021
10022	/* 16 bit data bus, synchronous transfers */
10023	/* XXX these flags don't apply for FC */
10024	if (is_fc == 0)
10025		inq_ptr->flags = SID_WBus16 | SID_Sync;
10026	/*
10027	 * XXX KDM do we want to support tagged queueing on the control
10028	 * device at all?
10029	 */
10030	if ((lun == NULL)
10031	 || (lun->be_lun->lun_type != T_PROCESSOR))
10032		inq_ptr->flags |= SID_CmdQue;
10033	/*
10034	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10035	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10036	 * name and 4 bytes for the revision.
10037	 */
10038	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
10039		strcpy(inq_ptr->vendor, CTL_VENDOR);
10040	} else {
10041		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10042		strncpy(inq_ptr->vendor, val,
10043		    min(sizeof(inq_ptr->vendor), strlen(val)));
10044	}
10045	if (lun == NULL) {
10046		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10047	} else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
10048		switch (lun->be_lun->lun_type) {
10049		case T_DIRECT:
10050			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10051			break;
10052		case T_PROCESSOR:
10053			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10054			break;
10055		default:
10056			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10057			break;
10058		}
10059	} else {
10060		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10061		strncpy(inq_ptr->product, val,
10062		    min(sizeof(inq_ptr->product), strlen(val)));
10063	}
10064
10065	/*
10066	 * XXX make this a macro somewhere so it automatically gets
10067	 * incremented when we make changes.
10068	 */
10069	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
10070		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10071	} else {
10072		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10073		strncpy(inq_ptr->revision, val,
10074		    min(sizeof(inq_ptr->revision), strlen(val)));
10075	}
10076
10077	/*
10078	 * For parallel SCSI, we support double transition and single
10079	 * transition clocking.  We also support QAS (Quick Arbitration
10080	 * and Selection) and Information Unit transfers on both the
10081	 * control and array devices.
10082	 */
10083	if (is_fc == 0)
10084		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10085				    SID_SPI_IUS;
10086
10087	/* SAM-3 */
10088	scsi_ulto2b(0x0060, inq_ptr->version1);
10089	/* SPC-3 (no version claimed) XXX should we claim a version? */
10090	scsi_ulto2b(0x0300, inq_ptr->version2);
10091	if (is_fc) {
10092		/* FCP-2 ANSI INCITS.350:2003 */
10093		scsi_ulto2b(0x0917, inq_ptr->version3);
10094	} else {
10095		/* SPI-4 ANSI INCITS.362:200x */
10096		scsi_ulto2b(0x0B56, inq_ptr->version3);
10097	}
10098
10099	if (lun == NULL) {
10100		/* SBC-2 (no version claimed) XXX should we claim a version? */
10101		scsi_ulto2b(0x0320, inq_ptr->version4);
10102	} else {
10103		switch (lun->be_lun->lun_type) {
10104		case T_DIRECT:
10105			/*
10106			 * SBC-2 (no version claimed) XXX should we claim a
10107			 * version?
10108			 */
10109			scsi_ulto2b(0x0320, inq_ptr->version4);
10110			break;
10111		case T_PROCESSOR:
10112		default:
10113			break;
10114		}
10115	}
10116
10117	ctsio->scsi_status = SCSI_STATUS_OK;
10118	if (ctsio->kern_data_len > 0) {
10119		ctsio->be_move_done = ctl_config_move_done;
10120		ctl_datamove((union ctl_io *)ctsio);
10121	} else {
10122		ctsio->io_hdr.status = CTL_SUCCESS;
10123		ctl_done((union ctl_io *)ctsio);
10124	}
10125
10126	return (CTL_RETVAL_COMPLETE);
10127}
10128
10129int
10130ctl_inquiry(struct ctl_scsiio *ctsio)
10131{
10132	struct scsi_inquiry *cdb;
10133	int retval;
10134
10135	cdb = (struct scsi_inquiry *)ctsio->cdb;
10136
10137	retval = 0;
10138
10139	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10140
10141	/*
10142	 * Right now, we don't support the CmdDt inquiry information.
10143	 * This would be nice to support in the future.  When we do
10144	 * support it, we should change this test so that it checks to make
10145	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10146	 */
10147#ifdef notyet
10148	if (((cdb->byte2 & SI_EVPD)
10149	 && (cdb->byte2 & SI_CMDDT)))
10150#endif
10151	if (cdb->byte2 & SI_CMDDT) {
10152		/*
10153		 * Point to the SI_CMDDT bit.  We might change this
10154		 * when we support SI_CMDDT, but since both bits would be
10155		 * "wrong", this should probably just stay as-is then.
10156		 */
10157		ctl_set_invalid_field(ctsio,
10158				      /*sks_valid*/ 1,
10159				      /*command*/ 1,
10160				      /*field*/ 1,
10161				      /*bit_valid*/ 1,
10162				      /*bit*/ 1);
10163		ctl_done((union ctl_io *)ctsio);
10164		return (CTL_RETVAL_COMPLETE);
10165	}
10166	if (cdb->byte2 & SI_EVPD)
10167		retval = ctl_inquiry_evpd(ctsio);
10168#ifdef notyet
10169	else if (cdb->byte2 & SI_CMDDT)
10170		retval = ctl_inquiry_cmddt(ctsio);
10171#endif
10172	else
10173		retval = ctl_inquiry_std(ctsio);
10174
10175	return (retval);
10176}
10177
10178/*
10179 * For known CDB types, parse the LBA and length.
10180 */
10181static int
10182ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10183{
10184	if (io->io_hdr.io_type != CTL_IO_SCSI)
10185		return (1);
10186
10187	switch (io->scsiio.cdb[0]) {
10188	case COMPARE_AND_WRITE: {
10189		struct scsi_compare_and_write *cdb;
10190
10191		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10192
10193		*lba = scsi_8btou64(cdb->addr);
10194		*len = cdb->length;
10195		break;
10196	}
10197	case READ_6:
10198	case WRITE_6: {
10199		struct scsi_rw_6 *cdb;
10200
10201		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10202
10203		*lba = scsi_3btoul(cdb->addr);
10204		/* only 5 bits are valid in the most significant address byte */
10205		*lba &= 0x1fffff;
10206		*len = cdb->length;
10207		break;
10208	}
10209	case READ_10:
10210	case WRITE_10: {
10211		struct scsi_rw_10 *cdb;
10212
10213		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10214
10215		*lba = scsi_4btoul(cdb->addr);
10216		*len = scsi_2btoul(cdb->length);
10217		break;
10218	}
10219	case WRITE_VERIFY_10: {
10220		struct scsi_write_verify_10 *cdb;
10221
10222		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10223
10224		*lba = scsi_4btoul(cdb->addr);
10225		*len = scsi_2btoul(cdb->length);
10226		break;
10227	}
10228	case READ_12:
10229	case WRITE_12: {
10230		struct scsi_rw_12 *cdb;
10231
10232		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10233
10234		*lba = scsi_4btoul(cdb->addr);
10235		*len = scsi_4btoul(cdb->length);
10236		break;
10237	}
10238	case WRITE_VERIFY_12: {
10239		struct scsi_write_verify_12 *cdb;
10240
10241		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10242
10243		*lba = scsi_4btoul(cdb->addr);
10244		*len = scsi_4btoul(cdb->length);
10245		break;
10246	}
10247	case READ_16:
10248	case WRITE_16: {
10249		struct scsi_rw_16 *cdb;
10250
10251		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10252
10253		*lba = scsi_8btou64(cdb->addr);
10254		*len = scsi_4btoul(cdb->length);
10255		break;
10256	}
10257	case WRITE_VERIFY_16: {
10258		struct scsi_write_verify_16 *cdb;
10259
10260		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10261
10262
10263		*lba = scsi_8btou64(cdb->addr);
10264		*len = scsi_4btoul(cdb->length);
10265		break;
10266	}
10267	case WRITE_SAME_10: {
10268		struct scsi_write_same_10 *cdb;
10269
10270		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10271
10272		*lba = scsi_4btoul(cdb->addr);
10273		*len = scsi_2btoul(cdb->length);
10274		break;
10275	}
10276	case WRITE_SAME_16: {
10277		struct scsi_write_same_16 *cdb;
10278
10279		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10280
10281		*lba = scsi_8btou64(cdb->addr);
10282		*len = scsi_4btoul(cdb->length);
10283		break;
10284	}
10285	case VERIFY_10: {
10286		struct scsi_verify_10 *cdb;
10287
10288		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10289
10290		*lba = scsi_4btoul(cdb->addr);
10291		*len = scsi_2btoul(cdb->length);
10292		break;
10293	}
10294	case VERIFY_12: {
10295		struct scsi_verify_12 *cdb;
10296
10297		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10298
10299		*lba = scsi_4btoul(cdb->addr);
10300		*len = scsi_4btoul(cdb->length);
10301		break;
10302	}
10303	case VERIFY_16: {
10304		struct scsi_verify_16 *cdb;
10305
10306		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10307
10308		*lba = scsi_8btou64(cdb->addr);
10309		*len = scsi_4btoul(cdb->length);
10310		break;
10311	}
10312	default:
10313		return (1);
10314		break; /* NOTREACHED */
10315	}
10316
10317	return (0);
10318}
10319
10320static ctl_action
10321ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10322{
10323	uint64_t endlba1, endlba2;
10324
10325	endlba1 = lba1 + len1 - 1;
10326	endlba2 = lba2 + len2 - 1;
10327
10328	if ((endlba1 < lba2)
10329	 || (endlba2 < lba1))
10330		return (CTL_ACTION_PASS);
10331	else
10332		return (CTL_ACTION_BLOCK);
10333}
10334
10335static ctl_action
10336ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10337{
10338	uint64_t lba1, lba2;
10339	uint32_t len1, len2;
10340	int retval;
10341
10342	retval = ctl_get_lba_len(io1, &lba1, &len1);
10343	if (retval != 0)
10344		return (CTL_ACTION_ERROR);
10345
10346	retval = ctl_get_lba_len(io2, &lba2, &len2);
10347	if (retval != 0)
10348		return (CTL_ACTION_ERROR);
10349
10350	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10351}
10352
10353static ctl_action
10354ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10355{
10356	struct ctl_cmd_entry *pending_entry, *ooa_entry;
10357	ctl_serialize_action *serialize_row;
10358
10359	/*
10360	 * The initiator attempted multiple untagged commands at the same
10361	 * time.  Can't do that.
10362	 */
10363	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10364	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10365	 && ((pending_io->io_hdr.nexus.targ_port ==
10366	      ooa_io->io_hdr.nexus.targ_port)
10367	  && (pending_io->io_hdr.nexus.initid.id ==
10368	      ooa_io->io_hdr.nexus.initid.id))
10369	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10370		return (CTL_ACTION_OVERLAP);
10371
10372	/*
10373	 * The initiator attempted to send multiple tagged commands with
10374	 * the same ID.  (It's fine if different initiators have the same
10375	 * tag ID.)
10376	 *
10377	 * Even if all of those conditions are true, we don't kill the I/O
10378	 * if the command ahead of us has been aborted.  We won't end up
10379	 * sending it to the FETD, and it's perfectly legal to resend a
10380	 * command with the same tag number as long as the previous
10381	 * instance of this tag number has been aborted somehow.
10382	 */
10383	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10384	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10385	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10386	 && ((pending_io->io_hdr.nexus.targ_port ==
10387	      ooa_io->io_hdr.nexus.targ_port)
10388	  && (pending_io->io_hdr.nexus.initid.id ==
10389	      ooa_io->io_hdr.nexus.initid.id))
10390	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10391		return (CTL_ACTION_OVERLAP_TAG);
10392
10393	/*
10394	 * If we get a head of queue tag, SAM-3 says that we should
10395	 * immediately execute it.
10396	 *
10397	 * What happens if this command would normally block for some other
10398	 * reason?  e.g. a request sense with a head of queue tag
10399	 * immediately after a write.  Normally that would block, but this
10400	 * will result in its getting executed immediately...
10401	 *
10402	 * We currently return "pass" instead of "skip", so we'll end up
10403	 * going through the rest of the queue to check for overlapped tags.
10404	 *
10405	 * XXX KDM check for other types of blockage first??
10406	 */
10407	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10408		return (CTL_ACTION_PASS);
10409
10410	/*
10411	 * Ordered tags have to block until all items ahead of them
10412	 * have completed.  If we get called with an ordered tag, we always
10413	 * block, if something else is ahead of us in the queue.
10414	 */
10415	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10416		return (CTL_ACTION_BLOCK);
10417
10418	/*
10419	 * Simple tags get blocked until all head of queue and ordered tags
10420	 * ahead of them have completed.  I'm lumping untagged commands in
10421	 * with simple tags here.  XXX KDM is that the right thing to do?
10422	 */
10423	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10424	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10425	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10426	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10427		return (CTL_ACTION_BLOCK);
10428
10429	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
10430	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
10431
10432	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10433
10434	switch (serialize_row[pending_entry->seridx]) {
10435	case CTL_SER_BLOCK:
10436		return (CTL_ACTION_BLOCK);
10437		break; /* NOTREACHED */
10438	case CTL_SER_EXTENT:
10439		return (ctl_extent_check(pending_io, ooa_io));
10440		break; /* NOTREACHED */
10441	case CTL_SER_PASS:
10442		return (CTL_ACTION_PASS);
10443		break; /* NOTREACHED */
10444	case CTL_SER_SKIP:
10445		return (CTL_ACTION_SKIP);
10446		break;
10447	default:
10448		panic("invalid serialization value %d",
10449		      serialize_row[pending_entry->seridx]);
10450		break; /* NOTREACHED */
10451	}
10452
10453	return (CTL_ACTION_ERROR);
10454}
10455
10456/*
10457 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10458 * Assumptions:
10459 * - pending_io is generally either incoming, or on the blocked queue
10460 * - starting I/O is the I/O we want to start the check with.
10461 */
10462static ctl_action
10463ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10464	      union ctl_io *starting_io)
10465{
10466	union ctl_io *ooa_io;
10467	ctl_action action;
10468
10469	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10470
10471	/*
10472	 * Run back along the OOA queue, starting with the current
10473	 * blocked I/O and going through every I/O before it on the
10474	 * queue.  If starting_io is NULL, we'll just end up returning
10475	 * CTL_ACTION_PASS.
10476	 */
10477	for (ooa_io = starting_io; ooa_io != NULL;
10478	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10479	     ooa_links)){
10480
10481		/*
10482		 * This routine just checks to see whether
10483		 * cur_blocked is blocked by ooa_io, which is ahead
10484		 * of it in the queue.  It doesn't queue/dequeue
10485		 * cur_blocked.
10486		 */
10487		action = ctl_check_for_blockage(pending_io, ooa_io);
10488		switch (action) {
10489		case CTL_ACTION_BLOCK:
10490		case CTL_ACTION_OVERLAP:
10491		case CTL_ACTION_OVERLAP_TAG:
10492		case CTL_ACTION_SKIP:
10493		case CTL_ACTION_ERROR:
10494			return (action);
10495			break; /* NOTREACHED */
10496		case CTL_ACTION_PASS:
10497			break;
10498		default:
10499			panic("invalid action %d", action);
10500			break;  /* NOTREACHED */
10501		}
10502	}
10503
10504	return (CTL_ACTION_PASS);
10505}
10506
10507/*
10508 * Assumptions:
10509 * - An I/O has just completed, and has been removed from the per-LUN OOA
10510 *   queue, so some items on the blocked queue may now be unblocked.
10511 */
10512static int
10513ctl_check_blocked(struct ctl_lun *lun)
10514{
10515	union ctl_io *cur_blocked, *next_blocked;
10516
10517	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10518
10519	/*
10520	 * Run forward from the head of the blocked queue, checking each
10521	 * entry against the I/Os prior to it on the OOA queue to see if
10522	 * there is still any blockage.
10523	 *
10524	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10525	 * with our removing a variable on it while it is traversing the
10526	 * list.
10527	 */
10528	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10529	     cur_blocked != NULL; cur_blocked = next_blocked) {
10530		union ctl_io *prev_ooa;
10531		ctl_action action;
10532
10533		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10534							  blocked_links);
10535
10536		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10537						      ctl_ooaq, ooa_links);
10538
10539		/*
10540		 * If cur_blocked happens to be the first item in the OOA
10541		 * queue now, prev_ooa will be NULL, and the action
10542		 * returned will just be CTL_ACTION_PASS.
10543		 */
10544		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10545
10546		switch (action) {
10547		case CTL_ACTION_BLOCK:
10548			/* Nothing to do here, still blocked */
10549			break;
10550		case CTL_ACTION_OVERLAP:
10551		case CTL_ACTION_OVERLAP_TAG:
10552			/*
10553			 * This shouldn't happen!  In theory we've already
10554			 * checked this command for overlap...
10555			 */
10556			break;
10557		case CTL_ACTION_PASS:
10558		case CTL_ACTION_SKIP: {
10559			struct ctl_softc *softc;
10560			struct ctl_cmd_entry *entry;
10561			uint32_t initidx;
10562			uint8_t opcode;
10563			int isc_retval;
10564
10565			/*
10566			 * The skip case shouldn't happen, this transaction
10567			 * should have never made it onto the blocked queue.
10568			 */
10569			/*
10570			 * This I/O is no longer blocked, we can remove it
10571			 * from the blocked queue.  Since this is a TAILQ
10572			 * (doubly linked list), we can do O(1) removals
10573			 * from any place on the list.
10574			 */
10575			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10576				     blocked_links);
10577			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10578
10579			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10580				/*
10581				 * Need to send IO back to original side to
10582				 * run
10583				 */
10584				union ctl_ha_msg msg_info;
10585
10586				msg_info.hdr.original_sc =
10587					cur_blocked->io_hdr.original_sc;
10588				msg_info.hdr.serializing_sc = cur_blocked;
10589				msg_info.hdr.msg_type = CTL_MSG_R2R;
10590				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10591				     &msg_info, sizeof(msg_info), 0)) >
10592				     CTL_HA_STATUS_SUCCESS) {
10593					printf("CTL:Check Blocked error from "
10594					       "ctl_ha_msg_send %d\n",
10595					       isc_retval);
10596				}
10597				break;
10598			}
10599			opcode = cur_blocked->scsiio.cdb[0];
10600			entry = &ctl_cmd_table[opcode];
10601			softc = control_softc;
10602
10603			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10604
10605			/*
10606			 * Check this I/O for LUN state changes that may
10607			 * have happened while this command was blocked.
10608			 * The LUN state may have been changed by a command
10609			 * ahead of us in the queue, so we need to re-check
10610			 * for any states that can be caused by SCSI
10611			 * commands.
10612			 */
10613			if (ctl_scsiio_lun_check(softc, lun, entry,
10614						 &cur_blocked->scsiio) == 0) {
10615				cur_blocked->io_hdr.flags |=
10616				                      CTL_FLAG_IS_WAS_ON_RTR;
10617				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
10618						   &cur_blocked->io_hdr, links);
10619				/*
10620				 * In the non CTL_DONE_THREAD case, we need
10621				 * to wake up the work thread here.  When
10622				 * we're processing completed requests from
10623				 * the work thread context, we'll pop back
10624				 * around and end up pulling things off the
10625				 * RtR queue.  When we aren't processing
10626				 * things from the work thread context,
10627				 * though, we won't ever check the RtR queue.
10628				 * So we need to wake up the thread to clear
10629				 * things off the queue.  Otherwise this
10630				 * transaction will just sit on the RtR queue
10631				 * until a new I/O comes in.  (Which may or
10632				 * may not happen...)
10633				 */
10634#ifndef CTL_DONE_THREAD
10635				ctl_wakeup_thread();
10636#endif
10637			} else
10638				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
10639			break;
10640		}
10641		default:
10642			/*
10643			 * This probably shouldn't happen -- we shouldn't
10644			 * get CTL_ACTION_ERROR, or anything else.
10645			 */
10646			break;
10647		}
10648	}
10649
10650	return (CTL_RETVAL_COMPLETE);
10651}
10652
10653/*
10654 * This routine (with one exception) checks LUN flags that can be set by
10655 * commands ahead of us in the OOA queue.  These flags have to be checked
10656 * when a command initially comes in, and when we pull a command off the
10657 * blocked queue and are preparing to execute it.  The reason we have to
10658 * check these flags for commands on the blocked queue is that the LUN
10659 * state may have been changed by a command ahead of us while we're on the
10660 * blocked queue.
10661 *
10662 * Ordering is somewhat important with these checks, so please pay
10663 * careful attention to the placement of any new checks.
10664 */
10665static int
10666ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10667		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10668{
10669	int retval;
10670
10671	retval = 0;
10672
10673	/*
10674	 * If this shelf is a secondary shelf controller, we have to reject
10675	 * any media access commands.
10676	 */
10677#if 0
10678	/* No longer needed for HA */
10679	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10680	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10681		ctl_set_lun_standby(ctsio);
10682		retval = 1;
10683		goto bailout;
10684	}
10685#endif
10686
10687	/*
10688	 * Check for a reservation conflict.  If this command isn't allowed
10689	 * even on reserved LUNs, and if this initiator isn't the one who
10690	 * reserved us, reject the command with a reservation conflict.
10691	 */
10692	if ((lun->flags & CTL_LUN_RESERVED)
10693	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10694		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10695		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10696		 || (ctsio->io_hdr.nexus.targ_target.id !=
10697		     lun->rsv_nexus.targ_target.id)) {
10698			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10699			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10700			retval = 1;
10701			goto bailout;
10702		}
10703	}
10704
10705	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10706	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10707		uint32_t residx;
10708
10709		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10710		/*
10711		 * if we aren't registered or it's a res holder type
10712		 * reservation and this isn't the res holder then set a
10713		 * conflict.
10714		 * NOTE: Commands which might be allowed on write exclusive
10715		 * type reservations are checked in the particular command
10716		 * for a conflict. Read and SSU are the only ones.
10717		 */
10718		if (!lun->per_res[residx].registered
10719		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10720			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10721			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10722			retval = 1;
10723			goto bailout;
10724		}
10725
10726	}
10727
10728	if ((lun->flags & CTL_LUN_OFFLINE)
10729	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10730		ctl_set_lun_not_ready(ctsio);
10731		retval = 1;
10732		goto bailout;
10733	}
10734
10735	/*
10736	 * If the LUN is stopped, see if this particular command is allowed
10737	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10738	 */
10739	if ((lun->flags & CTL_LUN_STOPPED)
10740	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10741		/* "Logical unit not ready, initializing cmd. required" */
10742		ctl_set_lun_stopped(ctsio);
10743		retval = 1;
10744		goto bailout;
10745	}
10746
10747	if ((lun->flags & CTL_LUN_INOPERABLE)
10748	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10749		/* "Medium format corrupted" */
10750		ctl_set_medium_format_corrupted(ctsio);
10751		retval = 1;
10752		goto bailout;
10753	}
10754
10755bailout:
10756	return (retval);
10757
10758}
10759
10760static void
10761ctl_failover_io(union ctl_io *io, int have_lock)
10762{
10763	ctl_set_busy(&io->scsiio);
10764	ctl_done_lock(io, have_lock);
10765}
10766
10767static void
10768ctl_failover(void)
10769{
10770	struct ctl_lun *lun;
10771	struct ctl_softc *ctl_softc;
10772	union ctl_io *next_io, *pending_io;
10773	union ctl_io *io;
10774	int lun_idx;
10775	int i;
10776
10777	ctl_softc = control_softc;
10778
10779	mtx_lock(&ctl_softc->ctl_lock);
10780	/*
10781	 * Remove any cmds from the other SC from the rtr queue.  These
10782	 * will obviously only be for LUNs for which we're the primary.
10783	 * We can't send status or get/send data for these commands.
10784	 * Since they haven't been executed yet, we can just remove them.
10785	 * We'll either abort them or delete them below, depending on
10786	 * which HA mode we're in.
10787	 */
10788	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10789	     io != NULL; io = next_io) {
10790		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10791		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10792			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10793				      ctl_io_hdr, links);
10794	}
10795
10796	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10797		lun = ctl_softc->ctl_luns[lun_idx];
10798		if (lun==NULL)
10799			continue;
10800
10801		/*
10802		 * Processor LUNs are primary on both sides.
10803		 * XXX will this always be true?
10804		 */
10805		if (lun->be_lun->lun_type == T_PROCESSOR)
10806			continue;
10807
10808		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10809		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10810			printf("FAILOVER: primary lun %d\n", lun_idx);
10811		        /*
10812			 * Remove all commands from the other SC. First from the
10813			 * blocked queue then from the ooa queue. Once we have
10814			 * removed them. Call ctl_check_blocked to see if there
10815			 * is anything that can run.
10816			 */
10817			for (io = (union ctl_io *)TAILQ_FIRST(
10818			     &lun->blocked_queue); io != NULL; io = next_io) {
10819
10820		        	next_io = (union ctl_io *)TAILQ_NEXT(
10821				    &io->io_hdr, blocked_links);
10822
10823				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10824					TAILQ_REMOVE(&lun->blocked_queue,
10825						     &io->io_hdr,blocked_links);
10826					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10827					TAILQ_REMOVE(&lun->ooa_queue,
10828						     &io->io_hdr, ooa_links);
10829
10830					ctl_free_io(io);
10831				}
10832			}
10833
10834			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10835	     		     io != NULL; io = next_io) {
10836
10837		        	next_io = (union ctl_io *)TAILQ_NEXT(
10838				    &io->io_hdr, ooa_links);
10839
10840				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10841
10842					TAILQ_REMOVE(&lun->ooa_queue,
10843						&io->io_hdr,
10844					     	ooa_links);
10845
10846					ctl_free_io(io);
10847				}
10848			}
10849			ctl_check_blocked(lun);
10850		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10851			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10852
10853			printf("FAILOVER: primary lun %d\n", lun_idx);
10854			/*
10855			 * Abort all commands from the other SC.  We can't
10856			 * send status back for them now.  These should get
10857			 * cleaned up when they are completed or come out
10858			 * for a datamove operation.
10859			 */
10860			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10861	     		     io != NULL; io = next_io) {
10862		        	next_io = (union ctl_io *)TAILQ_NEXT(
10863					&io->io_hdr, ooa_links);
10864
10865				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10866					io->io_hdr.flags |= CTL_FLAG_ABORT;
10867			}
10868		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10869			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10870
10871			printf("FAILOVER: secondary lun %d\n", lun_idx);
10872
10873			lun->flags |= CTL_LUN_PRIMARY_SC;
10874
10875			/*
10876			 * We send all I/O that was sent to this controller
10877			 * and redirected to the other side back with
10878			 * busy status, and have the initiator retry it.
10879			 * Figuring out how much data has been transferred,
10880			 * etc. and picking up where we left off would be
10881			 * very tricky.
10882			 *
10883			 * XXX KDM need to remove I/O from the blocked
10884			 * queue as well!
10885			 */
10886			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10887			     &lun->ooa_queue); pending_io != NULL;
10888			     pending_io = next_io) {
10889
10890				next_io =  (union ctl_io *)TAILQ_NEXT(
10891					&pending_io->io_hdr, ooa_links);
10892
10893				pending_io->io_hdr.flags &=
10894					~CTL_FLAG_SENT_2OTHER_SC;
10895
10896				if (pending_io->io_hdr.flags &
10897				    CTL_FLAG_IO_ACTIVE) {
10898					pending_io->io_hdr.flags |=
10899						CTL_FLAG_FAILOVER;
10900				} else {
10901					ctl_set_busy(&pending_io->scsiio);
10902					ctl_done_lock(pending_io,
10903						      /*have_lock*/1);
10904				}
10905			}
10906
10907			/*
10908			 * Build Unit Attention
10909			 */
10910			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10911				lun->pending_sense[i].ua_pending |=
10912				                     CTL_UA_ASYM_ACC_CHANGE;
10913			}
10914		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10915			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10916			printf("FAILOVER: secondary lun %d\n", lun_idx);
10917			/*
10918			 * if the first io on the OOA is not on the RtR queue
10919			 * add it.
10920			 */
10921			lun->flags |= CTL_LUN_PRIMARY_SC;
10922
10923			pending_io = (union ctl_io *)TAILQ_FIRST(
10924			    &lun->ooa_queue);
10925			if (pending_io==NULL) {
10926				printf("Nothing on OOA queue\n");
10927				continue;
10928			}
10929
10930			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10931			if ((pending_io->io_hdr.flags &
10932			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10933				pending_io->io_hdr.flags |=
10934				    CTL_FLAG_IS_WAS_ON_RTR;
10935				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10936						   &pending_io->io_hdr, links);
10937			}
10938#if 0
10939			else
10940			{
10941				printf("Tag 0x%04x is running\n",
10942				      pending_io->scsiio.tag_num);
10943			}
10944#endif
10945
10946			next_io = (union ctl_io *)TAILQ_NEXT(
10947			    &pending_io->io_hdr, ooa_links);
10948			for (pending_io=next_io; pending_io != NULL;
10949			     pending_io = next_io) {
10950				pending_io->io_hdr.flags &=
10951				    ~CTL_FLAG_SENT_2OTHER_SC;
10952				next_io = (union ctl_io *)TAILQ_NEXT(
10953					&pending_io->io_hdr, ooa_links);
10954				if (pending_io->io_hdr.flags &
10955				    CTL_FLAG_IS_WAS_ON_RTR) {
10956#if 0
10957				        printf("Tag 0x%04x is running\n",
10958				      		pending_io->scsiio.tag_num);
10959#endif
10960					continue;
10961				}
10962
10963				switch (ctl_check_ooa(lun, pending_io,
10964			            (union ctl_io *)TAILQ_PREV(
10965				    &pending_io->io_hdr, ctl_ooaq,
10966				    ooa_links))) {
10967
10968				case CTL_ACTION_BLOCK:
10969					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10970							  &pending_io->io_hdr,
10971							  blocked_links);
10972					pending_io->io_hdr.flags |=
10973					    CTL_FLAG_BLOCKED;
10974					break;
10975				case CTL_ACTION_PASS:
10976				case CTL_ACTION_SKIP:
10977					pending_io->io_hdr.flags |=
10978					    CTL_FLAG_IS_WAS_ON_RTR;
10979					STAILQ_INSERT_TAIL(
10980					    &ctl_softc->rtr_queue,
10981					    &pending_io->io_hdr, links);
10982					break;
10983				case CTL_ACTION_OVERLAP:
10984					ctl_set_overlapped_cmd(
10985					    (struct ctl_scsiio *)pending_io);
10986					ctl_done_lock(pending_io,
10987						      /*have_lock*/ 1);
10988					break;
10989				case CTL_ACTION_OVERLAP_TAG:
10990					ctl_set_overlapped_tag(
10991					    (struct ctl_scsiio *)pending_io,
10992					    pending_io->scsiio.tag_num & 0xff);
10993					ctl_done_lock(pending_io,
10994						      /*have_lock*/ 1);
10995					break;
10996				case CTL_ACTION_ERROR:
10997				default:
10998					ctl_set_internal_failure(
10999						(struct ctl_scsiio *)pending_io,
11000						0,  // sks_valid
11001						0); //retry count
11002					ctl_done_lock(pending_io,
11003						      /*have_lock*/ 1);
11004					break;
11005				}
11006			}
11007
11008			/*
11009			 * Build Unit Attention
11010			 */
11011			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11012				lun->pending_sense[i].ua_pending |=
11013				                     CTL_UA_ASYM_ACC_CHANGE;
11014			}
11015		} else {
11016			panic("Unhandled HA mode failover, LUN flags = %#x, "
11017			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11018		}
11019	}
11020	ctl_pause_rtr = 0;
11021	mtx_unlock(&ctl_softc->ctl_lock);
11022}
11023
11024static int
11025ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11026{
11027	struct ctl_lun *lun;
11028	struct ctl_cmd_entry *entry;
11029	uint8_t opcode;
11030	uint32_t initidx, targ_lun;
11031	int retval;
11032
11033	retval = 0;
11034
11035	lun = NULL;
11036
11037	opcode = ctsio->cdb[0];
11038
11039	mtx_lock(&ctl_softc->ctl_lock);
11040
11041	targ_lun = ctsio->io_hdr.nexus.targ_lun;
11042	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
11043		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
11044	if ((targ_lun < CTL_MAX_LUNS)
11045	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11046		lun = ctl_softc->ctl_luns[targ_lun];
11047		/*
11048		 * If the LUN is invalid, pretend that it doesn't exist.
11049		 * It will go away as soon as all pending I/O has been
11050		 * completed.
11051		 */
11052		if (lun->flags & CTL_LUN_DISABLED) {
11053			lun = NULL;
11054		} else {
11055			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11056			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11057				lun->be_lun;
11058			if (lun->be_lun->lun_type == T_PROCESSOR) {
11059				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11060			}
11061		}
11062	} else {
11063		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11064		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11065	}
11066
11067	entry = &ctl_cmd_table[opcode];
11068
11069	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11070	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11071
11072	/*
11073	 * Check to see whether we can send this command to LUNs that don't
11074	 * exist.  This should pretty much only be the case for inquiry
11075	 * and request sense.  Further checks, below, really require having
11076	 * a LUN, so we can't really check the command anymore.  Just put
11077	 * it on the rtr queue.
11078	 */
11079	if (lun == NULL) {
11080		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11081			goto queue_rtr;
11082
11083		ctl_set_unsupported_lun(ctsio);
11084		mtx_unlock(&ctl_softc->ctl_lock);
11085		ctl_done((union ctl_io *)ctsio);
11086		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11087		goto bailout;
11088	} else {
11089		/*
11090		 * Every I/O goes into the OOA queue for a particular LUN, and
11091		 * stays there until completion.
11092		 */
11093		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11094
11095		/*
11096		 * Make sure we support this particular command on this LUN.
11097		 * e.g., we don't support writes to the control LUN.
11098		 */
11099		switch (lun->be_lun->lun_type) {
11100		case T_PROCESSOR:
11101		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11102			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11103			      == 0)) {
11104				ctl_set_invalid_opcode(ctsio);
11105				mtx_unlock(&ctl_softc->ctl_lock);
11106				ctl_done((union ctl_io *)ctsio);
11107				goto bailout;
11108			}
11109			break;
11110		case T_DIRECT:
11111			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11112			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11113			      == 0)){
11114				ctl_set_invalid_opcode(ctsio);
11115				mtx_unlock(&ctl_softc->ctl_lock);
11116				ctl_done((union ctl_io *)ctsio);
11117				goto bailout;
11118			}
11119			break;
11120		default:
11121			printf("Unsupported CTL LUN type %d\n",
11122			       lun->be_lun->lun_type);
11123			panic("Unsupported CTL LUN type %d\n",
11124			      lun->be_lun->lun_type);
11125			break; /* NOTREACHED */
11126		}
11127	}
11128
11129	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11130
11131	/*
11132	 * If we've got a request sense, it'll clear the contingent
11133	 * allegiance condition.  Otherwise, if we have a CA condition for
11134	 * this initiator, clear it, because it sent down a command other
11135	 * than request sense.
11136	 */
11137	if ((opcode != REQUEST_SENSE)
11138	 && (ctl_is_set(lun->have_ca, initidx)))
11139		ctl_clear_mask(lun->have_ca, initidx);
11140
11141	/*
11142	 * If the command has this flag set, it handles its own unit
11143	 * attention reporting, we shouldn't do anything.  Otherwise we
11144	 * check for any pending unit attentions, and send them back to the
11145	 * initiator.  We only do this when a command initially comes in,
11146	 * not when we pull it off the blocked queue.
11147	 *
11148	 * According to SAM-3, section 5.3.2, the order that things get
11149	 * presented back to the host is basically unit attentions caused
11150	 * by some sort of reset event, busy status, reservation conflicts
11151	 * or task set full, and finally any other status.
11152	 *
11153	 * One issue here is that some of the unit attentions we report
11154	 * don't fall into the "reset" category (e.g. "reported luns data
11155	 * has changed").  So reporting it here, before the reservation
11156	 * check, may be technically wrong.  I guess the only thing to do
11157	 * would be to check for and report the reset events here, and then
11158	 * check for the other unit attention types after we check for a
11159	 * reservation conflict.
11160	 *
11161	 * XXX KDM need to fix this
11162	 */
11163	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11164		ctl_ua_type ua_type;
11165
11166		ua_type = lun->pending_sense[initidx].ua_pending;
11167		if (ua_type != CTL_UA_NONE) {
11168			scsi_sense_data_type sense_format;
11169
11170			if (lun != NULL)
11171				sense_format = (lun->flags &
11172				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11173				    SSD_TYPE_FIXED;
11174			else
11175				sense_format = SSD_TYPE_FIXED;
11176
11177			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11178					       sense_format);
11179			if (ua_type != CTL_UA_NONE) {
11180				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11181				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11182						       CTL_AUTOSENSE;
11183				ctsio->sense_len = SSD_FULL_SIZE;
11184				lun->pending_sense[initidx].ua_pending &=
11185					~ua_type;
11186				mtx_unlock(&ctl_softc->ctl_lock);
11187				ctl_done((union ctl_io *)ctsio);
11188				goto bailout;
11189			}
11190		}
11191	}
11192
11193
11194	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11195		mtx_unlock(&ctl_softc->ctl_lock);
11196		ctl_done((union ctl_io *)ctsio);
11197		goto bailout;
11198	}
11199
11200	/*
11201	 * XXX CHD this is where we want to send IO to other side if
11202	 * this LUN is secondary on this SC. We will need to make a copy
11203	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11204	 * the copy we send as FROM_OTHER.
11205	 * We also need to stuff the address of the original IO so we can
11206	 * find it easily. Something similar will need be done on the other
11207	 * side so when we are done we can find the copy.
11208	 */
11209	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11210		union ctl_ha_msg msg_info;
11211		int isc_retval;
11212
11213		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11214
11215		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11216		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11217#if 0
11218		printf("1. ctsio %p\n", ctsio);
11219#endif
11220		msg_info.hdr.serializing_sc = NULL;
11221		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11222		msg_info.scsi.tag_num = ctsio->tag_num;
11223		msg_info.scsi.tag_type = ctsio->tag_type;
11224		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11225
11226		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11227
11228		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11229		    (void *)&msg_info, sizeof(msg_info), 0)) >
11230		    CTL_HA_STATUS_SUCCESS) {
11231			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11232			       isc_retval);
11233			printf("CTL:opcode is %x\n",opcode);
11234		} else {
11235#if 0
11236			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11237#endif
11238		}
11239
11240		/*
11241		 * XXX KDM this I/O is off the incoming queue, but hasn't
11242		 * been inserted on any other queue.  We may need to come
11243		 * up with a holding queue while we wait for serialization
11244		 * so that we have an idea of what we're waiting for from
11245		 * the other side.
11246		 */
11247		goto bailout_unlock;
11248	}
11249
11250	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11251			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11252			      ctl_ooaq, ooa_links))) {
11253	case CTL_ACTION_BLOCK:
11254		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11255		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11256				  blocked_links);
11257		goto bailout_unlock;
11258		break; /* NOTREACHED */
11259	case CTL_ACTION_PASS:
11260	case CTL_ACTION_SKIP:
11261		goto queue_rtr;
11262		break; /* NOTREACHED */
11263	case CTL_ACTION_OVERLAP:
11264		ctl_set_overlapped_cmd(ctsio);
11265		mtx_unlock(&ctl_softc->ctl_lock);
11266		ctl_done((union ctl_io *)ctsio);
11267		goto bailout;
11268		break; /* NOTREACHED */
11269	case CTL_ACTION_OVERLAP_TAG:
11270		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11271		mtx_unlock(&ctl_softc->ctl_lock);
11272		ctl_done((union ctl_io *)ctsio);
11273		goto bailout;
11274		break; /* NOTREACHED */
11275	case CTL_ACTION_ERROR:
11276	default:
11277		ctl_set_internal_failure(ctsio,
11278					 /*sks_valid*/ 0,
11279					 /*retry_count*/ 0);
11280		mtx_unlock(&ctl_softc->ctl_lock);
11281		ctl_done((union ctl_io *)ctsio);
11282		goto bailout;
11283		break; /* NOTREACHED */
11284	}
11285
11286	goto bailout_unlock;
11287
11288queue_rtr:
11289	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11290	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
11291
11292bailout_unlock:
11293	mtx_unlock(&ctl_softc->ctl_lock);
11294
11295bailout:
11296	return (retval);
11297}
11298
11299static int
11300ctl_scsiio(struct ctl_scsiio *ctsio)
11301{
11302	int retval;
11303	struct ctl_cmd_entry *entry;
11304
11305	retval = CTL_RETVAL_COMPLETE;
11306
11307	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11308
11309	entry = &ctl_cmd_table[ctsio->cdb[0]];
11310
11311	/*
11312	 * If this I/O has been aborted, just send it straight to
11313	 * ctl_done() without executing it.
11314	 */
11315	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11316		ctl_done((union ctl_io *)ctsio);
11317		goto bailout;
11318	}
11319
11320	/*
11321	 * All the checks should have been handled by ctl_scsiio_precheck().
11322	 * We should be clear now to just execute the I/O.
11323	 */
11324	retval = entry->execute(ctsio);
11325
11326bailout:
11327	return (retval);
11328}
11329
11330/*
11331 * Since we only implement one target right now, a bus reset simply resets
11332 * our single target.
11333 */
11334static int
11335ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11336{
11337	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11338}
11339
11340static int
11341ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11342		 ctl_ua_type ua_type)
11343{
11344	struct ctl_lun *lun;
11345	int retval;
11346
11347	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11348		union ctl_ha_msg msg_info;
11349
11350		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11351		msg_info.hdr.nexus = io->io_hdr.nexus;
11352		if (ua_type==CTL_UA_TARG_RESET)
11353			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11354		else
11355			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11356		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11357		msg_info.hdr.original_sc = NULL;
11358		msg_info.hdr.serializing_sc = NULL;
11359		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11360		    (void *)&msg_info, sizeof(msg_info), 0)) {
11361		}
11362	}
11363	retval = 0;
11364
11365	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11366		retval += ctl_lun_reset(lun, io, ua_type);
11367
11368	return (retval);
11369}
11370
11371/*
11372 * The LUN should always be set.  The I/O is optional, and is used to
11373 * distinguish between I/Os sent by this initiator, and by other
11374 * initiators.  We set unit attention for initiators other than this one.
11375 * SAM-3 is vague on this point.  It does say that a unit attention should
11376 * be established for other initiators when a LUN is reset (see section
11377 * 5.7.3), but it doesn't specifically say that the unit attention should
11378 * be established for this particular initiator when a LUN is reset.  Here
11379 * is the relevant text, from SAM-3 rev 8:
11380 *
11381 * 5.7.2 When a SCSI initiator port aborts its own tasks
11382 *
11383 * When a SCSI initiator port causes its own task(s) to be aborted, no
11384 * notification that the task(s) have been aborted shall be returned to
11385 * the SCSI initiator port other than the completion response for the
11386 * command or task management function action that caused the task(s) to
11387 * be aborted and notification(s) associated with related effects of the
11388 * action (e.g., a reset unit attention condition).
11389 *
11390 * XXX KDM for now, we're setting unit attention for all initiators.
11391 */
11392static int
11393ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11394{
11395	union ctl_io *xio;
11396#if 0
11397	uint32_t initindex;
11398#endif
11399	int i;
11400
11401	/*
11402	 * Run through the OOA queue and abort each I/O.
11403	 */
11404#if 0
11405	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11406#endif
11407	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11408	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11409		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11410	}
11411
11412	/*
11413	 * This version sets unit attention for every
11414	 */
11415#if 0
11416	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11417	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11418		if (initindex == i)
11419			continue;
11420		lun->pending_sense[i].ua_pending |= ua_type;
11421	}
11422#endif
11423
11424	/*
11425	 * A reset (any kind, really) clears reservations established with
11426	 * RESERVE/RELEASE.  It does not clear reservations established
11427	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11428	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11429	 * reservations made with the RESERVE/RELEASE commands, because
11430	 * those commands are obsolete in SPC-3.
11431	 */
11432	lun->flags &= ~CTL_LUN_RESERVED;
11433
11434	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11435		ctl_clear_mask(lun->have_ca, i);
11436		lun->pending_sense[i].ua_pending |= ua_type;
11437	}
11438
11439	return (0);
11440}
11441
11442static int
11443ctl_abort_task(union ctl_io *io)
11444{
11445	union ctl_io *xio;
11446	struct ctl_lun *lun;
11447	struct ctl_softc *ctl_softc;
11448#if 0
11449	struct sbuf sb;
11450	char printbuf[128];
11451#endif
11452	int found;
11453	uint32_t targ_lun;
11454
11455	ctl_softc = control_softc;
11456	found = 0;
11457
11458	/*
11459	 * Look up the LUN.
11460	 */
11461	targ_lun = io->io_hdr.nexus.targ_lun;
11462	if (io->io_hdr.nexus.lun_map_fn != NULL)
11463		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11464	if ((targ_lun < CTL_MAX_LUNS)
11465	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11466		lun = ctl_softc->ctl_luns[targ_lun];
11467	else
11468		goto bailout;
11469
11470#if 0
11471	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11472	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11473#endif
11474
11475	/*
11476	 * Run through the OOA queue and attempt to find the given I/O.
11477	 * The target port, initiator ID, tag type and tag number have to
11478	 * match the values that we got from the initiator.  If we have an
11479	 * untagged command to abort, simply abort the first untagged command
11480	 * we come to.  We only allow one untagged command at a time of course.
11481	 */
11482#if 0
11483	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11484#endif
11485	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11486	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11487#if 0
11488		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11489
11490		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11491			    lun->lun, xio->scsiio.tag_num,
11492			    xio->scsiio.tag_type,
11493			    (xio->io_hdr.blocked_links.tqe_prev
11494			    == NULL) ? "" : " BLOCKED",
11495			    (xio->io_hdr.flags &
11496			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11497			    (xio->io_hdr.flags &
11498			    CTL_FLAG_ABORT) ? " ABORT" : "",
11499			    (xio->io_hdr.flags &
11500			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11501		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11502		sbuf_finish(&sb);
11503		printf("%s\n", sbuf_data(&sb));
11504#endif
11505
11506		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11507		 && (xio->io_hdr.nexus.initid.id ==
11508		     io->io_hdr.nexus.initid.id)) {
11509			/*
11510			 * If the abort says that the task is untagged, the
11511			 * task in the queue must be untagged.  Otherwise,
11512			 * we just check to see whether the tag numbers
11513			 * match.  This is because the QLogic firmware
11514			 * doesn't pass back the tag type in an abort
11515			 * request.
11516			 */
11517#if 0
11518			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11519			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11520			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11521#endif
11522			/*
11523			 * XXX KDM we've got problems with FC, because it
11524			 * doesn't send down a tag type with aborts.  So we
11525			 * can only really go by the tag number...
11526			 * This may cause problems with parallel SCSI.
11527			 * Need to figure that out!!
11528			 */
11529			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11530				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11531				found = 1;
11532				if ((io->io_hdr.flags &
11533				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11534				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11535					union ctl_ha_msg msg_info;
11536
11537					io->io_hdr.flags |=
11538					                CTL_FLAG_SENT_2OTHER_SC;
11539					msg_info.hdr.nexus = io->io_hdr.nexus;
11540					msg_info.task.task_action =
11541						CTL_TASK_ABORT_TASK;
11542					msg_info.task.tag_num =
11543						io->taskio.tag_num;
11544					msg_info.task.tag_type =
11545						io->taskio.tag_type;
11546					msg_info.hdr.msg_type =
11547						CTL_MSG_MANAGE_TASKS;
11548					msg_info.hdr.original_sc = NULL;
11549					msg_info.hdr.serializing_sc = NULL;
11550#if 0
11551					printf("Sent Abort to other side\n");
11552#endif
11553					if (CTL_HA_STATUS_SUCCESS !=
11554					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11555		    				(void *)&msg_info,
11556						sizeof(msg_info), 0)) {
11557					}
11558				}
11559#if 0
11560				printf("ctl_abort_task: found I/O to abort\n");
11561#endif
11562				break;
11563			}
11564		}
11565	}
11566
11567bailout:
11568
11569	if (found == 0) {
11570		/*
11571		 * This isn't really an error.  It's entirely possible for
11572		 * the abort and command completion to cross on the wire.
11573		 * This is more of an informative/diagnostic error.
11574		 */
11575#if 0
11576		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11577		       "%d:%d:%d:%d tag %d type %d\n",
11578		       io->io_hdr.nexus.initid.id,
11579		       io->io_hdr.nexus.targ_port,
11580		       io->io_hdr.nexus.targ_target.id,
11581		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11582		       io->taskio.tag_type);
11583#endif
11584		return (1);
11585	} else
11586		return (0);
11587}
11588
11589/*
11590 * This routine cannot block!  It must be callable from an interrupt
11591 * handler as well as from the work thread.
11592 */
11593static void
11594ctl_run_task_queue(struct ctl_softc *ctl_softc)
11595{
11596	union ctl_io *io, *next_io;
11597
11598	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
11599
11600	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
11601
11602	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
11603	     io != NULL; io = next_io) {
11604		int retval;
11605		const char *task_desc;
11606
11607		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11608
11609		retval = 0;
11610
11611		switch (io->io_hdr.io_type) {
11612		case CTL_IO_TASK: {
11613			task_desc = ctl_scsi_task_string(&io->taskio);
11614			if (task_desc != NULL) {
11615#ifdef NEEDTOPORT
11616				csevent_log(CSC_CTL | CSC_SHELF_SW |
11617					    CTL_TASK_REPORT,
11618					    csevent_LogType_Trace,
11619					    csevent_Severity_Information,
11620					    csevent_AlertLevel_Green,
11621					    csevent_FRU_Firmware,
11622					    csevent_FRU_Unknown,
11623					    "CTL: received task: %s",task_desc);
11624#endif
11625			} else {
11626#ifdef NEEDTOPORT
11627				csevent_log(CSC_CTL | CSC_SHELF_SW |
11628					    CTL_TASK_REPORT,
11629					    csevent_LogType_Trace,
11630					    csevent_Severity_Information,
11631					    csevent_AlertLevel_Green,
11632					    csevent_FRU_Firmware,
11633					    csevent_FRU_Unknown,
11634					    "CTL: received unknown task "
11635					    "type: %d (%#x)",
11636					    io->taskio.task_action,
11637					    io->taskio.task_action);
11638#endif
11639			}
11640			switch (io->taskio.task_action) {
11641			case CTL_TASK_ABORT_TASK:
11642				retval = ctl_abort_task(io);
11643				break;
11644			case CTL_TASK_ABORT_TASK_SET:
11645				break;
11646			case CTL_TASK_CLEAR_ACA:
11647				break;
11648			case CTL_TASK_CLEAR_TASK_SET:
11649				break;
11650			case CTL_TASK_LUN_RESET: {
11651				struct ctl_lun *lun;
11652				uint32_t targ_lun;
11653				int retval;
11654
11655				targ_lun = io->io_hdr.nexus.targ_lun;
11656				if (io->io_hdr.nexus.lun_map_fn != NULL)
11657					targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11658
11659				if ((targ_lun < CTL_MAX_LUNS)
11660				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11661					lun = ctl_softc->ctl_luns[targ_lun];
11662				else {
11663					retval = 1;
11664					break;
11665				}
11666
11667				if (!(io->io_hdr.flags &
11668				    CTL_FLAG_FROM_OTHER_SC)) {
11669					union ctl_ha_msg msg_info;
11670
11671					io->io_hdr.flags |=
11672						CTL_FLAG_SENT_2OTHER_SC;
11673					msg_info.hdr.msg_type =
11674						CTL_MSG_MANAGE_TASKS;
11675					msg_info.hdr.nexus = io->io_hdr.nexus;
11676					msg_info.task.task_action =
11677						CTL_TASK_LUN_RESET;
11678					msg_info.hdr.original_sc = NULL;
11679					msg_info.hdr.serializing_sc = NULL;
11680					if (CTL_HA_STATUS_SUCCESS !=
11681					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11682					    (void *)&msg_info,
11683					    sizeof(msg_info), 0)) {
11684					}
11685				}
11686
11687				retval = ctl_lun_reset(lun, io,
11688						       CTL_UA_LUN_RESET);
11689				break;
11690			}
11691			case CTL_TASK_TARGET_RESET:
11692				retval = ctl_target_reset(ctl_softc, io,
11693							  CTL_UA_TARG_RESET);
11694				break;
11695			case CTL_TASK_BUS_RESET:
11696				retval = ctl_bus_reset(ctl_softc, io);
11697				break;
11698			case CTL_TASK_PORT_LOGIN:
11699				break;
11700			case CTL_TASK_PORT_LOGOUT:
11701				break;
11702			default:
11703				printf("ctl_run_task_queue: got unknown task "
11704				       "management event %d\n",
11705				       io->taskio.task_action);
11706				break;
11707			}
11708			if (retval == 0)
11709				io->io_hdr.status = CTL_SUCCESS;
11710			else
11711				io->io_hdr.status = CTL_ERROR;
11712
11713			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11714				      ctl_io_hdr, links);
11715			/*
11716			 * This will queue this I/O to the done queue, but the
11717			 * work thread won't be able to process it until we
11718			 * return and the lock is released.
11719			 */
11720			ctl_done_lock(io, /*have_lock*/ 1);
11721			break;
11722		}
11723		default: {
11724
11725			printf("%s: invalid I/O type %d msg %d cdb %x"
11726			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11727			       __func__, io->io_hdr.io_type,
11728			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11729			       (uintmax_t)io->io_hdr.nexus.initid.id,
11730			       io->io_hdr.nexus.targ_port,
11731			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11732			       io->io_hdr.nexus.targ_lun /* XXX */,
11733			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11734			       io->taskio.tag_num : io->scsiio.tag_num);
11735			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11736				      ctl_io_hdr, links);
11737			ctl_free_io(io);
11738			break;
11739		}
11740		}
11741	}
11742
11743	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11744}
11745
11746/*
11747 * For HA operation.  Handle commands that come in from the other
11748 * controller.
11749 */
11750static void
11751ctl_handle_isc(union ctl_io *io)
11752{
11753	int free_io;
11754	struct ctl_lun *lun;
11755	struct ctl_softc *ctl_softc;
11756	uint32_t targ_lun;
11757
11758	ctl_softc = control_softc;
11759
11760	targ_lun = io->io_hdr.nexus.targ_lun;
11761	if (io->io_hdr.nexus.lun_map_fn != NULL)
11762		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11763	lun = ctl_softc->ctl_luns[targ_lun];
11764
11765	switch (io->io_hdr.msg_type) {
11766	case CTL_MSG_SERIALIZE:
11767		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11768						     /*have_lock*/ 0);
11769		break;
11770	case CTL_MSG_R2R: {
11771		uint8_t opcode;
11772		struct ctl_cmd_entry *entry;
11773
11774		/*
11775		 * This is only used in SER_ONLY mode.
11776		 */
11777		free_io = 0;
11778		opcode = io->scsiio.cdb[0];
11779		entry = &ctl_cmd_table[opcode];
11780		mtx_lock(&ctl_softc->ctl_lock);
11781		if (ctl_scsiio_lun_check(ctl_softc, lun,
11782		    entry, (struct ctl_scsiio *)io) != 0) {
11783			ctl_done_lock(io, /*have_lock*/ 1);
11784			mtx_unlock(&ctl_softc->ctl_lock);
11785			break;
11786		}
11787		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11788		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11789				   &io->io_hdr, links);
11790		mtx_unlock(&ctl_softc->ctl_lock);
11791		break;
11792	}
11793	case CTL_MSG_FINISH_IO:
11794		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11795			free_io = 0;
11796			ctl_done_lock(io, /*have_lock*/ 0);
11797		} else {
11798			free_io = 1;
11799			mtx_lock(&ctl_softc->ctl_lock);
11800			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11801				     ooa_links);
11802			STAILQ_REMOVE(&ctl_softc->task_queue,
11803				      &io->io_hdr, ctl_io_hdr, links);
11804			ctl_check_blocked(lun);
11805			mtx_unlock(&ctl_softc->ctl_lock);
11806		}
11807		break;
11808	case CTL_MSG_PERS_ACTION:
11809		ctl_hndl_per_res_out_on_other_sc(
11810			(union ctl_ha_msg *)&io->presio.pr_msg);
11811		free_io = 1;
11812		break;
11813	case CTL_MSG_BAD_JUJU:
11814		free_io = 0;
11815		ctl_done_lock(io, /*have_lock*/ 0);
11816		break;
11817	case CTL_MSG_DATAMOVE:
11818		/* Only used in XFER mode */
11819		free_io = 0;
11820		ctl_datamove_remote(io);
11821		break;
11822	case CTL_MSG_DATAMOVE_DONE:
11823		/* Only used in XFER mode */
11824		free_io = 0;
11825		io->scsiio.be_move_done(io);
11826		break;
11827	default:
11828		free_io = 1;
11829		printf("%s: Invalid message type %d\n",
11830		       __func__, io->io_hdr.msg_type);
11831		break;
11832	}
11833	if (free_io)
11834		ctl_free_io(io);
11835
11836}
11837
11838
11839/*
11840 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11841 * there is no match.
11842 */
11843static ctl_lun_error_pattern
11844ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11845{
11846	struct ctl_cmd_entry *entry;
11847	ctl_lun_error_pattern filtered_pattern, pattern;
11848	uint8_t opcode;
11849
11850	pattern = desc->error_pattern;
11851
11852	/*
11853	 * XXX KDM we need more data passed into this function to match a
11854	 * custom pattern, and we actually need to implement custom pattern
11855	 * matching.
11856	 */
11857	if (pattern & CTL_LUN_PAT_CMD)
11858		return (CTL_LUN_PAT_CMD);
11859
11860	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11861		return (CTL_LUN_PAT_ANY);
11862
11863	opcode = ctsio->cdb[0];
11864	entry = &ctl_cmd_table[opcode];
11865
11866	filtered_pattern = entry->pattern & pattern;
11867
11868	/*
11869	 * If the user requested specific flags in the pattern (e.g.
11870	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11871	 * flags.
11872	 *
11873	 * If the user did not specify any flags, it doesn't matter whether
11874	 * or not the command supports the flags.
11875	 */
11876	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11877	     (pattern & ~CTL_LUN_PAT_MASK))
11878		return (CTL_LUN_PAT_NONE);
11879
11880	/*
11881	 * If the user asked for a range check, see if the requested LBA
11882	 * range overlaps with this command's LBA range.
11883	 */
11884	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11885		uint64_t lba1;
11886		uint32_t len1;
11887		ctl_action action;
11888		int retval;
11889
11890		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11891		if (retval != 0)
11892			return (CTL_LUN_PAT_NONE);
11893
11894		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11895					      desc->lba_range.len);
11896		/*
11897		 * A "pass" means that the LBA ranges don't overlap, so
11898		 * this doesn't match the user's range criteria.
11899		 */
11900		if (action == CTL_ACTION_PASS)
11901			return (CTL_LUN_PAT_NONE);
11902	}
11903
11904	return (filtered_pattern);
11905}
11906
11907static void
11908ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11909{
11910	struct ctl_error_desc *desc, *desc2;
11911
11912	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
11913
11914	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11915		ctl_lun_error_pattern pattern;
11916		/*
11917		 * Check to see whether this particular command matches
11918		 * the pattern in the descriptor.
11919		 */
11920		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11921		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11922			continue;
11923
11924		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11925		case CTL_LUN_INJ_ABORTED:
11926			ctl_set_aborted(&io->scsiio);
11927			break;
11928		case CTL_LUN_INJ_MEDIUM_ERR:
11929			ctl_set_medium_error(&io->scsiio);
11930			break;
11931		case CTL_LUN_INJ_UA:
11932			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11933			 * OCCURRED */
11934			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11935			break;
11936		case CTL_LUN_INJ_CUSTOM:
11937			/*
11938			 * We're assuming the user knows what he is doing.
11939			 * Just copy the sense information without doing
11940			 * checks.
11941			 */
11942			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11943			      ctl_min(sizeof(desc->custom_sense),
11944				      sizeof(io->scsiio.sense_data)));
11945			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11946			io->scsiio.sense_len = SSD_FULL_SIZE;
11947			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11948			break;
11949		case CTL_LUN_INJ_NONE:
11950		default:
11951			/*
11952			 * If this is an error injection type we don't know
11953			 * about, clear the continuous flag (if it is set)
11954			 * so it will get deleted below.
11955			 */
11956			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11957			break;
11958		}
11959		/*
11960		 * By default, each error injection action is a one-shot
11961		 */
11962		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11963			continue;
11964
11965		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11966
11967		free(desc, M_CTL);
11968	}
11969}
11970
11971#ifdef CTL_IO_DELAY
11972static void
11973ctl_datamove_timer_wakeup(void *arg)
11974{
11975	union ctl_io *io;
11976
11977	io = (union ctl_io *)arg;
11978
11979	ctl_datamove(io);
11980}
11981#endif /* CTL_IO_DELAY */
11982
11983void
11984ctl_datamove(union ctl_io *io)
11985{
11986	void (*fe_datamove)(union ctl_io *io);
11987
11988	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11989
11990	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11991
11992#ifdef CTL_TIME_IO
11993	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11994		char str[256];
11995		char path_str[64];
11996		struct sbuf sb;
11997
11998		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11999		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12000
12001		sbuf_cat(&sb, path_str);
12002		switch (io->io_hdr.io_type) {
12003		case CTL_IO_SCSI:
12004			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12005			sbuf_printf(&sb, "\n");
12006			sbuf_cat(&sb, path_str);
12007			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12008				    io->scsiio.tag_num, io->scsiio.tag_type);
12009			break;
12010		case CTL_IO_TASK:
12011			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12012				    "Tag Type: %d\n", io->taskio.task_action,
12013				    io->taskio.tag_num, io->taskio.tag_type);
12014			break;
12015		default:
12016			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12017			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12018			break;
12019		}
12020		sbuf_cat(&sb, path_str);
12021		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12022			    (intmax_t)time_uptime - io->io_hdr.start_time);
12023		sbuf_finish(&sb);
12024		printf("%s", sbuf_data(&sb));
12025	}
12026#endif /* CTL_TIME_IO */
12027
12028	mtx_lock(&control_softc->ctl_lock);
12029#ifdef CTL_IO_DELAY
12030	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12031		struct ctl_lun *lun;
12032
12033		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12034
12035		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12036	} else {
12037		struct ctl_lun *lun;
12038
12039		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12040		if ((lun != NULL)
12041		 && (lun->delay_info.datamove_delay > 0)) {
12042			struct callout *callout;
12043
12044			callout = (struct callout *)&io->io_hdr.timer_bytes;
12045			callout_init(callout, /*mpsafe*/ 1);
12046			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12047			callout_reset(callout,
12048				      lun->delay_info.datamove_delay * hz,
12049				      ctl_datamove_timer_wakeup, io);
12050			if (lun->delay_info.datamove_type ==
12051			    CTL_DELAY_TYPE_ONESHOT)
12052				lun->delay_info.datamove_delay = 0;
12053			mtx_unlock(&control_softc->ctl_lock);
12054			return;
12055		}
12056	}
12057#endif
12058	/*
12059	 * If we have any pending task management commands, process them
12060	 * first.  This is necessary to eliminate a race condition with the
12061	 * FETD:
12062	 *
12063	 * - FETD submits a task management command, like an abort.
12064	 * - Back end calls fe_datamove() to move the data for the aborted
12065	 *   command.  The FETD can't really accept it, but if it did, it
12066	 *   would end up transmitting data for a command that the initiator
12067	 *   told us to abort.
12068	 *
12069	 * We close the race by processing all pending task management
12070	 * commands here (we can't block!), and then check this I/O to see
12071	 * if it has been aborted.  If so, return it to the back end with
12072	 * bad status, so the back end can say return an error to the back end
12073	 * and then when the back end returns an error, we can return the
12074	 * aborted command to the FETD, so it can clean up its resources.
12075	 */
12076	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
12077		ctl_run_task_queue(control_softc);
12078
12079	/*
12080	 * This command has been aborted.  Set the port status, so we fail
12081	 * the data move.
12082	 */
12083	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12084		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12085		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12086		       io->io_hdr.nexus.targ_port,
12087		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12088		       io->io_hdr.nexus.targ_lun);
12089		io->io_hdr.status = CTL_CMD_ABORTED;
12090		io->io_hdr.port_status = 31337;
12091		mtx_unlock(&control_softc->ctl_lock);
12092		/*
12093		 * Note that the backend, in this case, will get the
12094		 * callback in its context.  In other cases it may get
12095		 * called in the frontend's interrupt thread context.
12096		 */
12097		io->scsiio.be_move_done(io);
12098		return;
12099	}
12100
12101	/*
12102	 * If we're in XFER mode and this I/O is from the other shelf
12103	 * controller, we need to send the DMA to the other side to
12104	 * actually transfer the data to/from the host.  In serialize only
12105	 * mode the transfer happens below CTL and ctl_datamove() is only
12106	 * called on the machine that originally received the I/O.
12107	 */
12108	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12109	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12110		union ctl_ha_msg msg;
12111		uint32_t sg_entries_sent;
12112		int do_sg_copy;
12113		int i;
12114
12115		memset(&msg, 0, sizeof(msg));
12116		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12117		msg.hdr.original_sc = io->io_hdr.original_sc;
12118		msg.hdr.serializing_sc = io;
12119		msg.hdr.nexus = io->io_hdr.nexus;
12120		msg.dt.flags = io->io_hdr.flags;
12121		/*
12122		 * We convert everything into a S/G list here.  We can't
12123		 * pass by reference, only by value between controllers.
12124		 * So we can't pass a pointer to the S/G list, only as many
12125		 * S/G entries as we can fit in here.  If it's possible for
12126		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12127		 * then we need to break this up into multiple transfers.
12128		 */
12129		if (io->scsiio.kern_sg_entries == 0) {
12130			msg.dt.kern_sg_entries = 1;
12131			/*
12132			 * If this is in cached memory, flush the cache
12133			 * before we send the DMA request to the other
12134			 * controller.  We want to do this in either the
12135			 * read or the write case.  The read case is
12136			 * straightforward.  In the write case, we want to
12137			 * make sure nothing is in the local cache that
12138			 * could overwrite the DMAed data.
12139			 */
12140			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12141				/*
12142				 * XXX KDM use bus_dmamap_sync() here.
12143				 */
12144			}
12145
12146			/*
12147			 * Convert to a physical address if this is a
12148			 * virtual address.
12149			 */
12150			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12151				msg.dt.sg_list[0].addr =
12152					io->scsiio.kern_data_ptr;
12153			} else {
12154				/*
12155				 * XXX KDM use busdma here!
12156				 */
12157#if 0
12158				msg.dt.sg_list[0].addr = (void *)
12159					vtophys(io->scsiio.kern_data_ptr);
12160#endif
12161			}
12162
12163			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12164			do_sg_copy = 0;
12165		} else {
12166			struct ctl_sg_entry *sgl;
12167
12168			do_sg_copy = 1;
12169			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12170			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12171			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12172				/*
12173				 * XXX KDM use bus_dmamap_sync() here.
12174				 */
12175			}
12176		}
12177
12178		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12179		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12180		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12181		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12182		msg.dt.sg_sequence = 0;
12183
12184		/*
12185		 * Loop until we've sent all of the S/G entries.  On the
12186		 * other end, we'll recompose these S/G entries into one
12187		 * contiguous list before passing it to the
12188		 */
12189		for (sg_entries_sent = 0; sg_entries_sent <
12190		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12191			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12192				sizeof(msg.dt.sg_list[0])),
12193				msg.dt.kern_sg_entries - sg_entries_sent);
12194
12195			if (do_sg_copy != 0) {
12196				struct ctl_sg_entry *sgl;
12197				int j;
12198
12199				sgl = (struct ctl_sg_entry *)
12200					io->scsiio.kern_data_ptr;
12201				/*
12202				 * If this is in cached memory, flush the cache
12203				 * before we send the DMA request to the other
12204				 * controller.  We want to do this in either
12205				 * the * read or the write case.  The read
12206				 * case is straightforward.  In the write
12207				 * case, we want to make sure nothing is
12208				 * in the local cache that could overwrite
12209				 * the DMAed data.
12210				 */
12211
12212				for (i = sg_entries_sent, j = 0;
12213				     i < msg.dt.cur_sg_entries; i++, j++) {
12214					if ((io->io_hdr.flags &
12215					     CTL_FLAG_NO_DATASYNC) == 0) {
12216						/*
12217						 * XXX KDM use bus_dmamap_sync()
12218						 */
12219					}
12220					if ((io->io_hdr.flags &
12221					     CTL_FLAG_BUS_ADDR) == 0) {
12222						/*
12223						 * XXX KDM use busdma.
12224						 */
12225#if 0
12226						msg.dt.sg_list[j].addr =(void *)
12227						       vtophys(sgl[i].addr);
12228#endif
12229					} else {
12230						msg.dt.sg_list[j].addr =
12231							sgl[i].addr;
12232					}
12233					msg.dt.sg_list[j].len = sgl[i].len;
12234				}
12235			}
12236
12237			sg_entries_sent += msg.dt.cur_sg_entries;
12238			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12239				msg.dt.sg_last = 1;
12240			else
12241				msg.dt.sg_last = 0;
12242
12243			/*
12244			 * XXX KDM drop and reacquire the lock here?
12245			 */
12246			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12247			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12248				/*
12249				 * XXX do something here.
12250				 */
12251			}
12252
12253			msg.dt.sent_sg_entries = sg_entries_sent;
12254		}
12255		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12256		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12257			ctl_failover_io(io, /*have_lock*/ 1);
12258
12259	} else {
12260
12261		/*
12262		 * Lookup the fe_datamove() function for this particular
12263		 * front end.
12264		 */
12265		fe_datamove =
12266		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12267		mtx_unlock(&control_softc->ctl_lock);
12268
12269		fe_datamove(io);
12270	}
12271}
12272
12273static void
12274ctl_send_datamove_done(union ctl_io *io, int have_lock)
12275{
12276	union ctl_ha_msg msg;
12277	int isc_status;
12278
12279	memset(&msg, 0, sizeof(msg));
12280
12281	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12282	msg.hdr.original_sc = io;
12283	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12284	msg.hdr.nexus = io->io_hdr.nexus;
12285	msg.hdr.status = io->io_hdr.status;
12286	msg.scsi.tag_num = io->scsiio.tag_num;
12287	msg.scsi.tag_type = io->scsiio.tag_type;
12288	msg.scsi.scsi_status = io->scsiio.scsi_status;
12289	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12290	       sizeof(io->scsiio.sense_data));
12291	msg.scsi.sense_len = io->scsiio.sense_len;
12292	msg.scsi.sense_residual = io->scsiio.sense_residual;
12293	msg.scsi.fetd_status = io->io_hdr.port_status;
12294	msg.scsi.residual = io->scsiio.residual;
12295	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12296
12297	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12298		ctl_failover_io(io, /*have_lock*/ have_lock);
12299		return;
12300	}
12301
12302	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12303	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12304		/* XXX do something if this fails */
12305	}
12306
12307}
12308
12309/*
12310 * The DMA to the remote side is done, now we need to tell the other side
12311 * we're done so it can continue with its data movement.
12312 */
12313static void
12314ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12315{
12316	union ctl_io *io;
12317
12318	io = rq->context;
12319
12320	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12321		printf("%s: ISC DMA write failed with error %d", __func__,
12322		       rq->ret);
12323		ctl_set_internal_failure(&io->scsiio,
12324					 /*sks_valid*/ 1,
12325					 /*retry_count*/ rq->ret);
12326	}
12327
12328	ctl_dt_req_free(rq);
12329
12330	/*
12331	 * In this case, we had to malloc the memory locally.  Free it.
12332	 */
12333	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12334		int i;
12335		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12336			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12337	}
12338	/*
12339	 * The data is in local and remote memory, so now we need to send
12340	 * status (good or back) back to the other side.
12341	 */
12342	ctl_send_datamove_done(io, /*have_lock*/ 0);
12343}
12344
12345/*
12346 * We've moved the data from the host/controller into local memory.  Now we
12347 * need to push it over to the remote controller's memory.
12348 */
12349static int
12350ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12351{
12352	int retval;
12353
12354	retval = 0;
12355
12356	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12357					  ctl_datamove_remote_write_cb);
12358
12359	return (retval);
12360}
12361
12362static void
12363ctl_datamove_remote_write(union ctl_io *io)
12364{
12365	int retval;
12366	void (*fe_datamove)(union ctl_io *io);
12367
12368	/*
12369	 * - Get the data from the host/HBA into local memory.
12370	 * - DMA memory from the local controller to the remote controller.
12371	 * - Send status back to the remote controller.
12372	 */
12373
12374	retval = ctl_datamove_remote_sgl_setup(io);
12375	if (retval != 0)
12376		return;
12377
12378	/* Switch the pointer over so the FETD knows what to do */
12379	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12380
12381	/*
12382	 * Use a custom move done callback, since we need to send completion
12383	 * back to the other controller, not to the backend on this side.
12384	 */
12385	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12386
12387	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12388
12389	fe_datamove(io);
12390
12391	return;
12392
12393}
12394
12395static int
12396ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12397{
12398#if 0
12399	char str[256];
12400	char path_str[64];
12401	struct sbuf sb;
12402#endif
12403
12404	/*
12405	 * In this case, we had to malloc the memory locally.  Free it.
12406	 */
12407	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12408		int i;
12409		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12410			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12411	}
12412
12413#if 0
12414	scsi_path_string(io, path_str, sizeof(path_str));
12415	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12416	sbuf_cat(&sb, path_str);
12417	scsi_command_string(&io->scsiio, NULL, &sb);
12418	sbuf_printf(&sb, "\n");
12419	sbuf_cat(&sb, path_str);
12420	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12421		    io->scsiio.tag_num, io->scsiio.tag_type);
12422	sbuf_cat(&sb, path_str);
12423	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12424		    io->io_hdr.flags, io->io_hdr.status);
12425	sbuf_finish(&sb);
12426	printk("%s", sbuf_data(&sb));
12427#endif
12428
12429
12430	/*
12431	 * The read is done, now we need to send status (good or bad) back
12432	 * to the other side.
12433	 */
12434	ctl_send_datamove_done(io, /*have_lock*/ 0);
12435
12436	return (0);
12437}
12438
12439static void
12440ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12441{
12442	union ctl_io *io;
12443	void (*fe_datamove)(union ctl_io *io);
12444
12445	io = rq->context;
12446
12447	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12448		printf("%s: ISC DMA read failed with error %d", __func__,
12449		       rq->ret);
12450		ctl_set_internal_failure(&io->scsiio,
12451					 /*sks_valid*/ 1,
12452					 /*retry_count*/ rq->ret);
12453	}
12454
12455	ctl_dt_req_free(rq);
12456
12457	/* Switch the pointer over so the FETD knows what to do */
12458	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12459
12460	/*
12461	 * Use a custom move done callback, since we need to send completion
12462	 * back to the other controller, not to the backend on this side.
12463	 */
12464	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12465
12466	/* XXX KDM add checks like the ones in ctl_datamove? */
12467
12468	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12469
12470	fe_datamove(io);
12471}
12472
12473static int
12474ctl_datamove_remote_sgl_setup(union ctl_io *io)
12475{
12476	struct ctl_sg_entry *local_sglist, *remote_sglist;
12477	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12478	struct ctl_softc *softc;
12479	int retval;
12480	int i;
12481
12482	retval = 0;
12483	softc = control_softc;
12484
12485	local_sglist = io->io_hdr.local_sglist;
12486	local_dma_sglist = io->io_hdr.local_dma_sglist;
12487	remote_sglist = io->io_hdr.remote_sglist;
12488	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12489
12490	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12491		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12492			local_sglist[i].len = remote_sglist[i].len;
12493
12494			/*
12495			 * XXX Detect the situation where the RS-level I/O
12496			 * redirector on the other side has already read the
12497			 * data off of the AOR RS on this side, and
12498			 * transferred it to remote (mirror) memory on the
12499			 * other side.  Since we already have the data in
12500			 * memory here, we just need to use it.
12501			 *
12502			 * XXX KDM this can probably be removed once we
12503			 * get the cache device code in and take the
12504			 * current AOR implementation out.
12505			 */
12506#ifdef NEEDTOPORT
12507			if ((remote_sglist[i].addr >=
12508			     (void *)vtophys(softc->mirr->addr))
12509			 && (remote_sglist[i].addr <
12510			     ((void *)vtophys(softc->mirr->addr) +
12511			     CacheMirrorOffset))) {
12512				local_sglist[i].addr = remote_sglist[i].addr -
12513					CacheMirrorOffset;
12514				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12515				     CTL_FLAG_DATA_IN)
12516					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12517			} else {
12518				local_sglist[i].addr = remote_sglist[i].addr +
12519					CacheMirrorOffset;
12520			}
12521#endif
12522#if 0
12523			printf("%s: local %p, remote %p, len %d\n",
12524			       __func__, local_sglist[i].addr,
12525			       remote_sglist[i].addr, local_sglist[i].len);
12526#endif
12527		}
12528	} else {
12529		uint32_t len_to_go;
12530
12531		/*
12532		 * In this case, we don't have automatically allocated
12533		 * memory for this I/O on this controller.  This typically
12534		 * happens with internal CTL I/O -- e.g. inquiry, mode
12535		 * sense, etc.  Anything coming from RAIDCore will have
12536		 * a mirror area available.
12537		 */
12538		len_to_go = io->scsiio.kern_data_len;
12539
12540		/*
12541		 * Clear the no datasync flag, we have to use malloced
12542		 * buffers.
12543		 */
12544		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12545
12546		/*
12547		 * The difficult thing here is that the size of the various
12548		 * S/G segments may be different than the size from the
12549		 * remote controller.  That'll make it harder when DMAing
12550		 * the data back to the other side.
12551		 */
12552		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12553		     sizeof(io->io_hdr.remote_sglist[0])) &&
12554		     (len_to_go > 0); i++) {
12555			local_sglist[i].len = ctl_min(len_to_go, 131072);
12556			CTL_SIZE_8B(local_dma_sglist[i].len,
12557				    local_sglist[i].len);
12558			local_sglist[i].addr =
12559				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12560
12561			local_dma_sglist[i].addr = local_sglist[i].addr;
12562
12563			if (local_sglist[i].addr == NULL) {
12564				int j;
12565
12566				printf("malloc failed for %zd bytes!",
12567				       local_dma_sglist[i].len);
12568				for (j = 0; j < i; j++) {
12569					free(local_sglist[j].addr, M_CTL);
12570				}
12571				ctl_set_internal_failure(&io->scsiio,
12572							 /*sks_valid*/ 1,
12573							 /*retry_count*/ 4857);
12574				retval = 1;
12575				goto bailout_error;
12576
12577			}
12578			/* XXX KDM do we need a sync here? */
12579
12580			len_to_go -= local_sglist[i].len;
12581		}
12582		/*
12583		 * Reset the number of S/G entries accordingly.  The
12584		 * original number of S/G entries is available in
12585		 * rem_sg_entries.
12586		 */
12587		io->scsiio.kern_sg_entries = i;
12588
12589#if 0
12590		printf("%s: kern_sg_entries = %d\n", __func__,
12591		       io->scsiio.kern_sg_entries);
12592		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12593			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12594			       local_sglist[i].addr, local_sglist[i].len,
12595			       local_dma_sglist[i].len);
12596#endif
12597	}
12598
12599
12600	return (retval);
12601
12602bailout_error:
12603
12604	ctl_send_datamove_done(io, /*have_lock*/ 0);
12605
12606	return (retval);
12607}
12608
12609static int
12610ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12611			 ctl_ha_dt_cb callback)
12612{
12613	struct ctl_ha_dt_req *rq;
12614	struct ctl_sg_entry *remote_sglist, *local_sglist;
12615	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12616	uint32_t local_used, remote_used, total_used;
12617	int retval;
12618	int i, j;
12619
12620	retval = 0;
12621
12622	rq = ctl_dt_req_alloc();
12623
12624	/*
12625	 * If we failed to allocate the request, and if the DMA didn't fail
12626	 * anyway, set busy status.  This is just a resource allocation
12627	 * failure.
12628	 */
12629	if ((rq == NULL)
12630	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12631		ctl_set_busy(&io->scsiio);
12632
12633	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12634
12635		if (rq != NULL)
12636			ctl_dt_req_free(rq);
12637
12638		/*
12639		 * The data move failed.  We need to return status back
12640		 * to the other controller.  No point in trying to DMA
12641		 * data to the remote controller.
12642		 */
12643
12644		ctl_send_datamove_done(io, /*have_lock*/ 0);
12645
12646		retval = 1;
12647
12648		goto bailout;
12649	}
12650
12651	local_sglist = io->io_hdr.local_sglist;
12652	local_dma_sglist = io->io_hdr.local_dma_sglist;
12653	remote_sglist = io->io_hdr.remote_sglist;
12654	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12655	local_used = 0;
12656	remote_used = 0;
12657	total_used = 0;
12658
12659	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12660		rq->ret = CTL_HA_STATUS_SUCCESS;
12661		rq->context = io;
12662		callback(rq);
12663		goto bailout;
12664	}
12665
12666	/*
12667	 * Pull/push the data over the wire from/to the other controller.
12668	 * This takes into account the possibility that the local and
12669	 * remote sglists may not be identical in terms of the size of
12670	 * the elements and the number of elements.
12671	 *
12672	 * One fundamental assumption here is that the length allocated for
12673	 * both the local and remote sglists is identical.  Otherwise, we've
12674	 * essentially got a coding error of some sort.
12675	 */
12676	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12677		int isc_ret;
12678		uint32_t cur_len, dma_length;
12679		uint8_t *tmp_ptr;
12680
12681		rq->id = CTL_HA_DATA_CTL;
12682		rq->command = command;
12683		rq->context = io;
12684
12685		/*
12686		 * Both pointers should be aligned.  But it is possible
12687		 * that the allocation length is not.  They should both
12688		 * also have enough slack left over at the end, though,
12689		 * to round up to the next 8 byte boundary.
12690		 */
12691		cur_len = ctl_min(local_sglist[i].len - local_used,
12692				  remote_sglist[j].len - remote_used);
12693
12694		/*
12695		 * In this case, we have a size issue and need to decrease
12696		 * the size, except in the case where we actually have less
12697		 * than 8 bytes left.  In that case, we need to increase
12698		 * the DMA length to get the last bit.
12699		 */
12700		if ((cur_len & 0x7) != 0) {
12701			if (cur_len > 0x7) {
12702				cur_len = cur_len - (cur_len & 0x7);
12703				dma_length = cur_len;
12704			} else {
12705				CTL_SIZE_8B(dma_length, cur_len);
12706			}
12707
12708		} else
12709			dma_length = cur_len;
12710
12711		/*
12712		 * If we had to allocate memory for this I/O, instead of using
12713		 * the non-cached mirror memory, we'll need to flush the cache
12714		 * before trying to DMA to the other controller.
12715		 *
12716		 * We could end up doing this multiple times for the same
12717		 * segment if we have a larger local segment than remote
12718		 * segment.  That shouldn't be an issue.
12719		 */
12720		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12721			/*
12722			 * XXX KDM use bus_dmamap_sync() here.
12723			 */
12724		}
12725
12726		rq->size = dma_length;
12727
12728		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12729		tmp_ptr += local_used;
12730
12731		/* Use physical addresses when talking to ISC hardware */
12732		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12733			/* XXX KDM use busdma */
12734#if 0
12735			rq->local = vtophys(tmp_ptr);
12736#endif
12737		} else
12738			rq->local = tmp_ptr;
12739
12740		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12741		tmp_ptr += remote_used;
12742		rq->remote = tmp_ptr;
12743
12744		rq->callback = NULL;
12745
12746		local_used += cur_len;
12747		if (local_used >= local_sglist[i].len) {
12748			i++;
12749			local_used = 0;
12750		}
12751
12752		remote_used += cur_len;
12753		if (remote_used >= remote_sglist[j].len) {
12754			j++;
12755			remote_used = 0;
12756		}
12757		total_used += cur_len;
12758
12759		if (total_used >= io->scsiio.kern_data_len)
12760			rq->callback = callback;
12761
12762		if ((rq->size & 0x7) != 0) {
12763			printf("%s: warning: size %d is not on 8b boundary\n",
12764			       __func__, rq->size);
12765		}
12766		if (((uintptr_t)rq->local & 0x7) != 0) {
12767			printf("%s: warning: local %p not on 8b boundary\n",
12768			       __func__, rq->local);
12769		}
12770		if (((uintptr_t)rq->remote & 0x7) != 0) {
12771			printf("%s: warning: remote %p not on 8b boundary\n",
12772			       __func__, rq->local);
12773		}
12774#if 0
12775		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12776		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12777		       rq->local, rq->remote, rq->size);
12778#endif
12779
12780		isc_ret = ctl_dt_single(rq);
12781		if (isc_ret == CTL_HA_STATUS_WAIT)
12782			continue;
12783
12784		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12785			rq->ret = CTL_HA_STATUS_SUCCESS;
12786		} else {
12787			rq->ret = isc_ret;
12788		}
12789		callback(rq);
12790		goto bailout;
12791	}
12792
12793bailout:
12794	return (retval);
12795
12796}
12797
12798static void
12799ctl_datamove_remote_read(union ctl_io *io)
12800{
12801	int retval;
12802	int i;
12803
12804	/*
12805	 * This will send an error to the other controller in the case of a
12806	 * failure.
12807	 */
12808	retval = ctl_datamove_remote_sgl_setup(io);
12809	if (retval != 0)
12810		return;
12811
12812	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12813					  ctl_datamove_remote_read_cb);
12814	if ((retval != 0)
12815	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12816		/*
12817		 * Make sure we free memory if there was an error..  The
12818		 * ctl_datamove_remote_xfer() function will send the
12819		 * datamove done message, or call the callback with an
12820		 * error if there is a problem.
12821		 */
12822		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12823			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12824	}
12825
12826	return;
12827}
12828
12829/*
12830 * Process a datamove request from the other controller.  This is used for
12831 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12832 * first.  Once that is complete, the data gets DMAed into the remote
12833 * controller's memory.  For reads, we DMA from the remote controller's
12834 * memory into our memory first, and then move it out to the FETD.
12835 */
12836static void
12837ctl_datamove_remote(union ctl_io *io)
12838{
12839	struct ctl_softc *softc;
12840
12841	softc = control_softc;
12842
12843	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12844
12845	/*
12846	 * Note that we look for an aborted I/O here, but don't do some of
12847	 * the other checks that ctl_datamove() normally does.  We don't
12848	 * need to run the task queue, because this I/O is on the ISC
12849	 * queue, which is executed by the work thread after the task queue.
12850	 * We don't need to run the datamove delay code, since that should
12851	 * have been done if need be on the other controller.
12852	 */
12853	mtx_lock(&softc->ctl_lock);
12854
12855	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12856
12857		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12858		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12859		       io->io_hdr.nexus.targ_port,
12860		       io->io_hdr.nexus.targ_target.id,
12861		       io->io_hdr.nexus.targ_lun);
12862		io->io_hdr.status = CTL_CMD_ABORTED;
12863		io->io_hdr.port_status = 31338;
12864
12865		mtx_unlock(&softc->ctl_lock);
12866
12867		ctl_send_datamove_done(io, /*have_lock*/ 0);
12868
12869		return;
12870	}
12871
12872	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12873		mtx_unlock(&softc->ctl_lock);
12874		ctl_datamove_remote_write(io);
12875	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12876		mtx_unlock(&softc->ctl_lock);
12877		ctl_datamove_remote_read(io);
12878	} else {
12879		union ctl_ha_msg msg;
12880		struct scsi_sense_data *sense;
12881		uint8_t sks[3];
12882		int retry_count;
12883
12884		memset(&msg, 0, sizeof(msg));
12885
12886		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12887		msg.hdr.status = CTL_SCSI_ERROR;
12888		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12889
12890		retry_count = 4243;
12891
12892		sense = &msg.scsi.sense_data;
12893		sks[0] = SSD_SCS_VALID;
12894		sks[1] = (retry_count >> 8) & 0xff;
12895		sks[2] = retry_count & 0xff;
12896
12897		/* "Internal target failure" */
12898		scsi_set_sense_data(sense,
12899				    /*sense_format*/ SSD_TYPE_NONE,
12900				    /*current_error*/ 1,
12901				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12902				    /*asc*/ 0x44,
12903				    /*ascq*/ 0x00,
12904				    /*type*/ SSD_ELEM_SKS,
12905				    /*size*/ sizeof(sks),
12906				    /*data*/ sks,
12907				    SSD_ELEM_NONE);
12908
12909		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12910		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12911			ctl_failover_io(io, /*have_lock*/ 1);
12912			mtx_unlock(&softc->ctl_lock);
12913			return;
12914		}
12915
12916		mtx_unlock(&softc->ctl_lock);
12917
12918		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12919		    CTL_HA_STATUS_SUCCESS) {
12920			/* XXX KDM what to do if this fails? */
12921		}
12922		return;
12923	}
12924
12925}
12926
12927static int
12928ctl_process_done(union ctl_io *io, int have_lock)
12929{
12930	struct ctl_lun *lun;
12931	struct ctl_softc *ctl_softc;
12932	void (*fe_done)(union ctl_io *io);
12933	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12934
12935	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12936
12937	fe_done =
12938	    control_softc->ctl_ports[targ_port]->fe_done;
12939
12940#ifdef CTL_TIME_IO
12941	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12942		char str[256];
12943		char path_str[64];
12944		struct sbuf sb;
12945
12946		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12947		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12948
12949		sbuf_cat(&sb, path_str);
12950		switch (io->io_hdr.io_type) {
12951		case CTL_IO_SCSI:
12952			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12953			sbuf_printf(&sb, "\n");
12954			sbuf_cat(&sb, path_str);
12955			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12956				    io->scsiio.tag_num, io->scsiio.tag_type);
12957			break;
12958		case CTL_IO_TASK:
12959			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12960				    "Tag Type: %d\n", io->taskio.task_action,
12961				    io->taskio.tag_num, io->taskio.tag_type);
12962			break;
12963		default:
12964			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12965			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12966			break;
12967		}
12968		sbuf_cat(&sb, path_str);
12969		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12970			    (intmax_t)time_uptime - io->io_hdr.start_time);
12971		sbuf_finish(&sb);
12972		printf("%s", sbuf_data(&sb));
12973	}
12974#endif /* CTL_TIME_IO */
12975
12976	switch (io->io_hdr.io_type) {
12977	case CTL_IO_SCSI:
12978		break;
12979	case CTL_IO_TASK:
12980		if (bootverbose || verbose > 0)
12981			ctl_io_error_print(io, NULL);
12982		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12983			ctl_free_io(io);
12984		else
12985			fe_done(io);
12986		return (CTL_RETVAL_COMPLETE);
12987		break;
12988	default:
12989		printf("ctl_process_done: invalid io type %d\n",
12990		       io->io_hdr.io_type);
12991		panic("ctl_process_done: invalid io type %d\n",
12992		      io->io_hdr.io_type);
12993		break; /* NOTREACHED */
12994	}
12995
12996	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12997	if (lun == NULL) {
12998		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12999				 io->io_hdr.nexus.targ_lun));
13000		fe_done(io);
13001		goto bailout;
13002	}
13003	ctl_softc = lun->ctl_softc;
13004
13005	/*
13006	 * Remove this from the OOA queue.
13007	 */
13008	if (have_lock == 0)
13009		mtx_lock(&ctl_softc->ctl_lock);
13010
13011	/*
13012	 * Check to see if we have any errors to inject here.  We only
13013	 * inject errors for commands that don't already have errors set.
13014	 */
13015	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13016	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13017		ctl_inject_error(lun, io);
13018
13019	/*
13020	 * XXX KDM how do we treat commands that aren't completed
13021	 * successfully?
13022	 *
13023	 * XXX KDM should we also track I/O latency?
13024	 */
13025	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
13026		uint32_t blocksize;
13027#ifdef CTL_TIME_IO
13028		struct bintime cur_bt;
13029#endif
13030
13031		if ((lun->be_lun != NULL)
13032		 && (lun->be_lun->blocksize != 0))
13033			blocksize = lun->be_lun->blocksize;
13034		else
13035			blocksize = 512;
13036
13037		switch (io->io_hdr.io_type) {
13038		case CTL_IO_SCSI: {
13039			int isread;
13040			struct ctl_lba_len_flags *lbalen;
13041
13042			isread = 0;
13043			switch (io->scsiio.cdb[0]) {
13044			case READ_6:
13045			case READ_10:
13046			case READ_12:
13047			case READ_16:
13048				isread = 1;
13049				/* FALLTHROUGH */
13050			case WRITE_6:
13051			case WRITE_10:
13052			case WRITE_12:
13053			case WRITE_16:
13054			case WRITE_VERIFY_10:
13055			case WRITE_VERIFY_12:
13056			case WRITE_VERIFY_16:
13057				lbalen = (struct ctl_lba_len_flags *)
13058				    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
13059
13060				if (isread) {
13061					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
13062					    lbalen->len * blocksize;
13063					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
13064
13065#ifdef CTL_TIME_IO
13066					bintime_add(
13067					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
13068					   &io->io_hdr.dma_bt);
13069					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
13070						io->io_hdr.num_dmas;
13071					getbintime(&cur_bt);
13072					bintime_sub(&cur_bt,
13073						    &io->io_hdr.start_bt);
13074
13075					bintime_add(
13076					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
13077					    &cur_bt);
13078
13079#if 0
13080					cs_prof_gettime(&cur_ticks);
13081					lun->stats.time[CTL_STATS_READ] +=
13082						cur_ticks -
13083						io->io_hdr.start_ticks;
13084#endif
13085#if 0
13086					lun->stats.time[CTL_STATS_READ] +=
13087						jiffies - io->io_hdr.start_time;
13088#endif
13089#endif /* CTL_TIME_IO */
13090				} else {
13091					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
13092					    lbalen->len * blocksize;
13093					lun->stats.ports[targ_port].operations[
13094						CTL_STATS_WRITE]++;
13095
13096#ifdef CTL_TIME_IO
13097					bintime_add(
13098					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
13099					  &io->io_hdr.dma_bt);
13100					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
13101						io->io_hdr.num_dmas;
13102					getbintime(&cur_bt);
13103					bintime_sub(&cur_bt,
13104						    &io->io_hdr.start_bt);
13105
13106					bintime_add(
13107					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
13108					    &cur_bt);
13109#if 0
13110					cs_prof_gettime(&cur_ticks);
13111					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13112						cur_ticks -
13113						io->io_hdr.start_ticks;
13114					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13115						jiffies - io->io_hdr.start_time;
13116#endif
13117#endif /* CTL_TIME_IO */
13118				}
13119				break;
13120			default:
13121				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
13122
13123#ifdef CTL_TIME_IO
13124				bintime_add(
13125				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
13126				  &io->io_hdr.dma_bt);
13127				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
13128					io->io_hdr.num_dmas;
13129				getbintime(&cur_bt);
13130				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13131
13132				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
13133					    &cur_bt);
13134
13135#if 0
13136				cs_prof_gettime(&cur_ticks);
13137				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13138					cur_ticks -
13139					io->io_hdr.start_ticks;
13140				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13141					jiffies - io->io_hdr.start_time;
13142#endif
13143#endif /* CTL_TIME_IO */
13144				break;
13145			}
13146			break;
13147		}
13148		default:
13149			break;
13150		}
13151	}
13152
13153	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13154
13155	/*
13156	 * Run through the blocked queue on this LUN and see if anything
13157	 * has become unblocked, now that this transaction is done.
13158	 */
13159	ctl_check_blocked(lun);
13160
13161	/*
13162	 * If the LUN has been invalidated, free it if there is nothing
13163	 * left on its OOA queue.
13164	 */
13165	if ((lun->flags & CTL_LUN_INVALID)
13166	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
13167		ctl_free_lun(lun);
13168
13169	/*
13170	 * If this command has been aborted, make sure we set the status
13171	 * properly.  The FETD is responsible for freeing the I/O and doing
13172	 * whatever it needs to do to clean up its state.
13173	 */
13174	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13175		io->io_hdr.status = CTL_CMD_ABORTED;
13176
13177	/*
13178	 * We print out status for every task management command.  For SCSI
13179	 * commands, we filter out any unit attention errors; they happen
13180	 * on every boot, and would clutter up the log.  Note:  task
13181	 * management commands aren't printed here, they are printed above,
13182	 * since they should never even make it down here.
13183	 */
13184	switch (io->io_hdr.io_type) {
13185	case CTL_IO_SCSI: {
13186		int error_code, sense_key, asc, ascq;
13187
13188		sense_key = 0;
13189
13190		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13191		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13192			/*
13193			 * Since this is just for printing, no need to
13194			 * show errors here.
13195			 */
13196			scsi_extract_sense_len(&io->scsiio.sense_data,
13197					       io->scsiio.sense_len,
13198					       &error_code,
13199					       &sense_key,
13200					       &asc,
13201					       &ascq,
13202					       /*show_errors*/ 0);
13203		}
13204
13205		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13206		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13207		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13208		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13209
13210			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13211				ctl_softc->skipped_prints++;
13212				if (have_lock == 0)
13213					mtx_unlock(&ctl_softc->ctl_lock);
13214			} else {
13215				uint32_t skipped_prints;
13216
13217				skipped_prints = ctl_softc->skipped_prints;
13218
13219				ctl_softc->skipped_prints = 0;
13220				ctl_softc->last_print_jiffies = time_uptime;
13221
13222				if (have_lock == 0)
13223					mtx_unlock(&ctl_softc->ctl_lock);
13224				if (skipped_prints > 0) {
13225#ifdef NEEDTOPORT
13226					csevent_log(CSC_CTL | CSC_SHELF_SW |
13227					    CTL_ERROR_REPORT,
13228					    csevent_LogType_Trace,
13229					    csevent_Severity_Information,
13230					    csevent_AlertLevel_Green,
13231					    csevent_FRU_Firmware,
13232					    csevent_FRU_Unknown,
13233					    "High CTL error volume, %d prints "
13234					    "skipped", skipped_prints);
13235#endif
13236				}
13237				if (bootverbose || verbose > 0)
13238					ctl_io_error_print(io, NULL);
13239			}
13240		} else {
13241			if (have_lock == 0)
13242				mtx_unlock(&ctl_softc->ctl_lock);
13243		}
13244		break;
13245	}
13246	case CTL_IO_TASK:
13247		if (have_lock == 0)
13248			mtx_unlock(&ctl_softc->ctl_lock);
13249		if (bootverbose || verbose > 0)
13250			ctl_io_error_print(io, NULL);
13251		break;
13252	default:
13253		if (have_lock == 0)
13254			mtx_unlock(&ctl_softc->ctl_lock);
13255		break;
13256	}
13257
13258	/*
13259	 * Tell the FETD or the other shelf controller we're done with this
13260	 * command.  Note that only SCSI commands get to this point.  Task
13261	 * management commands are completed above.
13262	 *
13263	 * We only send status to the other controller if we're in XFER
13264	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13265	 * received the I/O (from CTL's perspective), and so the status is
13266	 * generated there.
13267	 *
13268	 * XXX KDM if we hold the lock here, we could cause a deadlock
13269	 * if the frontend comes back in in this context to queue
13270	 * something.
13271	 */
13272	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13273	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13274		union ctl_ha_msg msg;
13275
13276		memset(&msg, 0, sizeof(msg));
13277		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13278		msg.hdr.original_sc = io->io_hdr.original_sc;
13279		msg.hdr.nexus = io->io_hdr.nexus;
13280		msg.hdr.status = io->io_hdr.status;
13281		msg.scsi.scsi_status = io->scsiio.scsi_status;
13282		msg.scsi.tag_num = io->scsiio.tag_num;
13283		msg.scsi.tag_type = io->scsiio.tag_type;
13284		msg.scsi.sense_len = io->scsiio.sense_len;
13285		msg.scsi.sense_residual = io->scsiio.sense_residual;
13286		msg.scsi.residual = io->scsiio.residual;
13287		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13288		       sizeof(io->scsiio.sense_data));
13289		/*
13290		 * We copy this whether or not this is an I/O-related
13291		 * command.  Otherwise, we'd have to go and check to see
13292		 * whether it's a read/write command, and it really isn't
13293		 * worth it.
13294		 */
13295		memcpy(&msg.scsi.lbalen,
13296		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13297		       sizeof(msg.scsi.lbalen));
13298
13299		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13300				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13301			/* XXX do something here */
13302		}
13303
13304		ctl_free_io(io);
13305	} else
13306		fe_done(io);
13307
13308bailout:
13309
13310	return (CTL_RETVAL_COMPLETE);
13311}
13312
13313/*
13314 * Front end should call this if it doesn't do autosense.  When the request
13315 * sense comes back in from the initiator, we'll dequeue this and send it.
13316 */
13317int
13318ctl_queue_sense(union ctl_io *io)
13319{
13320	struct ctl_lun *lun;
13321	struct ctl_softc *ctl_softc;
13322	uint32_t initidx, targ_lun;
13323
13324	ctl_softc = control_softc;
13325
13326	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13327
13328	/*
13329	 * LUN lookup will likely move to the ctl_work_thread() once we
13330	 * have our new queueing infrastructure (that doesn't put things on
13331	 * a per-LUN queue initially).  That is so that we can handle
13332	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13333	 * can't deal with that right now.
13334	 */
13335	mtx_lock(&ctl_softc->ctl_lock);
13336
13337	/*
13338	 * If we don't have a LUN for this, just toss the sense
13339	 * information.
13340	 */
13341	targ_lun = io->io_hdr.nexus.targ_lun;
13342	if (io->io_hdr.nexus.lun_map_fn != NULL)
13343		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13344	if ((targ_lun < CTL_MAX_LUNS)
13345	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13346		lun = ctl_softc->ctl_luns[targ_lun];
13347	else
13348		goto bailout;
13349
13350	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13351
13352	/*
13353	 * Already have CA set for this LUN...toss the sense information.
13354	 */
13355	if (ctl_is_set(lun->have_ca, initidx))
13356		goto bailout;
13357
13358	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13359	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13360	       sizeof(io->scsiio.sense_data)));
13361	ctl_set_mask(lun->have_ca, initidx);
13362
13363bailout:
13364	mtx_unlock(&ctl_softc->ctl_lock);
13365
13366	ctl_free_io(io);
13367
13368	return (CTL_RETVAL_COMPLETE);
13369}
13370
13371/*
13372 * Primary command inlet from frontend ports.  All SCSI and task I/O
13373 * requests must go through this function.
13374 */
13375int
13376ctl_queue(union ctl_io *io)
13377{
13378	struct ctl_softc *ctl_softc;
13379
13380	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13381
13382	ctl_softc = control_softc;
13383
13384#ifdef CTL_TIME_IO
13385	io->io_hdr.start_time = time_uptime;
13386	getbintime(&io->io_hdr.start_bt);
13387#endif /* CTL_TIME_IO */
13388
13389	mtx_lock(&ctl_softc->ctl_lock);
13390
13391	switch (io->io_hdr.io_type) {
13392	case CTL_IO_SCSI:
13393		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
13394				   links);
13395		break;
13396	case CTL_IO_TASK:
13397		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
13398		/*
13399		 * Set the task pending flag.  This is necessary to close a
13400		 * race condition with the FETD:
13401		 *
13402		 * - FETD submits a task management command, like an abort.
13403		 * - Back end calls fe_datamove() to move the data for the
13404		 *   aborted command.  The FETD can't really accept it, but
13405		 *   if it did, it would end up transmitting data for a
13406		 *   command that the initiator told us to abort.
13407		 *
13408		 * We close the race condition by setting the flag here,
13409		 * and checking it in ctl_datamove(), before calling the
13410		 * FETD's fe_datamove routine.  If we've got a task
13411		 * pending, we run the task queue and then check to see
13412		 * whether our particular I/O has been aborted.
13413		 */
13414		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
13415		break;
13416	default:
13417		mtx_unlock(&ctl_softc->ctl_lock);
13418		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13419		return (-EINVAL);
13420		break; /* NOTREACHED */
13421	}
13422	mtx_unlock(&ctl_softc->ctl_lock);
13423
13424	ctl_wakeup_thread();
13425
13426	return (CTL_RETVAL_COMPLETE);
13427}
13428
13429#ifdef CTL_IO_DELAY
13430static void
13431ctl_done_timer_wakeup(void *arg)
13432{
13433	union ctl_io *io;
13434
13435	io = (union ctl_io *)arg;
13436	ctl_done_lock(io, /*have_lock*/ 0);
13437}
13438#endif /* CTL_IO_DELAY */
13439
13440void
13441ctl_done_lock(union ctl_io *io, int have_lock)
13442{
13443	struct ctl_softc *ctl_softc;
13444#ifndef CTL_DONE_THREAD
13445	union ctl_io *xio;
13446#endif /* !CTL_DONE_THREAD */
13447
13448	ctl_softc = control_softc;
13449
13450	if (have_lock == 0)
13451		mtx_lock(&ctl_softc->ctl_lock);
13452
13453	/*
13454	 * Enable this to catch duplicate completion issues.
13455	 */
13456#if 0
13457	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13458		printf("%s: type %d msg %d cdb %x iptl: "
13459		       "%d:%d:%d:%d tag 0x%04x "
13460		       "flag %#x status %x\n",
13461			__func__,
13462			io->io_hdr.io_type,
13463			io->io_hdr.msg_type,
13464			io->scsiio.cdb[0],
13465			io->io_hdr.nexus.initid.id,
13466			io->io_hdr.nexus.targ_port,
13467			io->io_hdr.nexus.targ_target.id,
13468			io->io_hdr.nexus.targ_lun,
13469			(io->io_hdr.io_type ==
13470			CTL_IO_TASK) ?
13471			io->taskio.tag_num :
13472			io->scsiio.tag_num,
13473		        io->io_hdr.flags,
13474			io->io_hdr.status);
13475	} else
13476		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13477#endif
13478
13479	/*
13480	 * This is an internal copy of an I/O, and should not go through
13481	 * the normal done processing logic.
13482	 */
13483	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
13484		if (have_lock == 0)
13485			mtx_unlock(&ctl_softc->ctl_lock);
13486		return;
13487	}
13488
13489	/*
13490	 * We need to send a msg to the serializing shelf to finish the IO
13491	 * as well.  We don't send a finish message to the other shelf if
13492	 * this is a task management command.  Task management commands
13493	 * aren't serialized in the OOA queue, but rather just executed on
13494	 * both shelf controllers for commands that originated on that
13495	 * controller.
13496	 */
13497	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13498	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13499		union ctl_ha_msg msg_io;
13500
13501		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13502		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13503		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13504		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13505		}
13506		/* continue on to finish IO */
13507	}
13508#ifdef CTL_IO_DELAY
13509	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13510		struct ctl_lun *lun;
13511
13512		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13513
13514		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13515	} else {
13516		struct ctl_lun *lun;
13517
13518		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13519
13520		if ((lun != NULL)
13521		 && (lun->delay_info.done_delay > 0)) {
13522			struct callout *callout;
13523
13524			callout = (struct callout *)&io->io_hdr.timer_bytes;
13525			callout_init(callout, /*mpsafe*/ 1);
13526			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13527			callout_reset(callout,
13528				      lun->delay_info.done_delay * hz,
13529				      ctl_done_timer_wakeup, io);
13530			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13531				lun->delay_info.done_delay = 0;
13532			if (have_lock == 0)
13533				mtx_unlock(&ctl_softc->ctl_lock);
13534			return;
13535		}
13536	}
13537#endif /* CTL_IO_DELAY */
13538
13539	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
13540
13541#ifdef CTL_DONE_THREAD
13542	if (have_lock == 0)
13543		mtx_unlock(&ctl_softc->ctl_lock);
13544
13545	ctl_wakeup_thread();
13546#else /* CTL_DONE_THREAD */
13547	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
13548	     xio != NULL;
13549	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
13550
13551		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
13552
13553		ctl_process_done(xio, /*have_lock*/ 1);
13554	}
13555	if (have_lock == 0)
13556		mtx_unlock(&ctl_softc->ctl_lock);
13557#endif /* CTL_DONE_THREAD */
13558}
13559
13560void
13561ctl_done(union ctl_io *io)
13562{
13563	ctl_done_lock(io, /*have_lock*/ 0);
13564}
13565
13566int
13567ctl_isc(struct ctl_scsiio *ctsio)
13568{
13569	struct ctl_lun *lun;
13570	int retval;
13571
13572	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13573
13574	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13575
13576	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13577
13578	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13579
13580	return (retval);
13581}
13582
13583
13584static void
13585ctl_work_thread(void *arg)
13586{
13587	struct ctl_softc *softc;
13588	union ctl_io *io;
13589	struct ctl_be_lun *be_lun;
13590	int retval;
13591
13592	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13593
13594	softc = (struct ctl_softc *)arg;
13595	if (softc == NULL)
13596		return;
13597
13598	mtx_lock(&softc->ctl_lock);
13599	for (;;) {
13600		retval = 0;
13601
13602		/*
13603		 * We handle the queues in this order:
13604		 * - task management
13605		 * - ISC
13606		 * - done queue (to free up resources, unblock other commands)
13607		 * - RtR queue
13608		 * - incoming queue
13609		 *
13610		 * If those queues are empty, we break out of the loop and
13611		 * go to sleep.
13612		 */
13613		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
13614		if (io != NULL) {
13615			ctl_run_task_queue(softc);
13616			continue;
13617		}
13618		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
13619		if (io != NULL) {
13620			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
13621			ctl_handle_isc(io);
13622			continue;
13623		}
13624		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
13625		if (io != NULL) {
13626			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
13627			/* clear any blocked commands, call fe_done */
13628			mtx_unlock(&softc->ctl_lock);
13629			/*
13630			 * XXX KDM
13631			 * Call this without a lock for now.  This will
13632			 * depend on whether there is any way the FETD can
13633			 * sleep or deadlock if called with the CTL lock
13634			 * held.
13635			 */
13636			retval = ctl_process_done(io, /*have_lock*/ 0);
13637			mtx_lock(&softc->ctl_lock);
13638			continue;
13639		}
13640		if (!ctl_pause_rtr) {
13641			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
13642			if (io != NULL) {
13643				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
13644				mtx_unlock(&softc->ctl_lock);
13645				retval = ctl_scsiio(&io->scsiio);
13646				if (retval != CTL_RETVAL_COMPLETE)
13647					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13648				mtx_lock(&softc->ctl_lock);
13649				continue;
13650			}
13651		}
13652		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
13653		if (io != NULL) {
13654			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
13655			mtx_unlock(&softc->ctl_lock);
13656			ctl_scsiio_precheck(softc, &io->scsiio);
13657			mtx_lock(&softc->ctl_lock);
13658			continue;
13659		}
13660		/*
13661		 * We might want to move this to a separate thread, so that
13662		 * configuration requests (in this case LUN creations)
13663		 * won't impact the I/O path.
13664		 */
13665		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13666		if (be_lun != NULL) {
13667			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13668			mtx_unlock(&softc->ctl_lock);
13669			ctl_create_lun(be_lun);
13670			mtx_lock(&softc->ctl_lock);
13671			continue;
13672		}
13673
13674		/* XXX KDM use the PDROP flag?? */
13675		/* Sleep until we have something to do. */
13676		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "-", 0);
13677
13678		/* Back to the top of the loop to see what woke us up. */
13679		continue;
13680	}
13681}
13682
13683void
13684ctl_wakeup_thread()
13685{
13686	struct ctl_softc *softc;
13687
13688	softc = control_softc;
13689
13690	wakeup_one(softc);
13691}
13692
13693/* Initialization and failover */
13694
13695void
13696ctl_init_isc_msg(void)
13697{
13698	printf("CTL: Still calling this thing\n");
13699}
13700
13701/*
13702 * Init component
13703 * 	Initializes component into configuration defined by bootMode
13704 *	(see hasc-sv.c)
13705 *  	returns hasc_Status:
13706 * 		OK
13707 *		ERROR - fatal error
13708 */
13709static ctl_ha_comp_status
13710ctl_isc_init(struct ctl_ha_component *c)
13711{
13712	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13713
13714	c->status = ret;
13715	return ret;
13716}
13717
13718/* Start component
13719 * 	Starts component in state requested. If component starts successfully,
13720 *	it must set its own state to the requestrd state
13721 *	When requested state is HASC_STATE_HA, the component may refine it
13722 * 	by adding _SLAVE or _MASTER flags.
13723 *	Currently allowed state transitions are:
13724 *	UNKNOWN->HA		- initial startup
13725 *	UNKNOWN->SINGLE - initial startup when no parter detected
13726 *	HA->SINGLE		- failover
13727 * returns ctl_ha_comp_status:
13728 * 		OK	- component successfully started in requested state
13729 *		FAILED  - could not start the requested state, failover may
13730 * 			  be possible
13731 *		ERROR	- fatal error detected, no future startup possible
13732 */
13733static ctl_ha_comp_status
13734ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13735{
13736	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13737
13738	printf("%s: go\n", __func__);
13739
13740	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13741	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13742		ctl_is_single = 0;
13743		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13744		    != CTL_HA_STATUS_SUCCESS) {
13745			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13746			ret = CTL_HA_COMP_STATUS_ERROR;
13747		}
13748	} else if (CTL_HA_STATE_IS_HA(c->state)
13749		&& CTL_HA_STATE_IS_SINGLE(state)){
13750		// HA->SINGLE transition
13751	        ctl_failover();
13752		ctl_is_single = 1;
13753	} else {
13754		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13755		       c->state, state);
13756		ret = CTL_HA_COMP_STATUS_ERROR;
13757	}
13758	if (CTL_HA_STATE_IS_SINGLE(state))
13759		ctl_is_single = 1;
13760
13761	c->state = state;
13762	c->status = ret;
13763	return ret;
13764}
13765
13766/*
13767 * Quiesce component
13768 * The component must clear any error conditions (set status to OK) and
13769 * prepare itself to another Start call
13770 * returns ctl_ha_comp_status:
13771 * 	OK
13772 *	ERROR
13773 */
13774static ctl_ha_comp_status
13775ctl_isc_quiesce(struct ctl_ha_component *c)
13776{
13777	int ret = CTL_HA_COMP_STATUS_OK;
13778
13779	ctl_pause_rtr = 1;
13780	c->status = ret;
13781	return ret;
13782}
13783
13784struct ctl_ha_component ctl_ha_component_ctlisc =
13785{
13786	.name = "CTL ISC",
13787	.state = CTL_HA_STATE_UNKNOWN,
13788	.init = ctl_isc_init,
13789	.start = ctl_isc_start,
13790	.quiesce = ctl_isc_quiesce
13791};
13792
13793/*
13794 *  vim: ts=8
13795 */
13796