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("$FreeBSD$");
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/endian.h>
64#include <sys/sysctl.h>
65
66#include <cam/cam.h>
67#include <cam/scsi/scsi_all.h>
68#include <cam/scsi/scsi_da.h>
69#include <cam/ctl/ctl_io.h>
70#include <cam/ctl/ctl.h>
71#include <cam/ctl/ctl_frontend.h>
72#include <cam/ctl/ctl_frontend_internal.h>
73#include <cam/ctl/ctl_util.h>
74#include <cam/ctl/ctl_backend.h>
75#include <cam/ctl/ctl_ioctl.h>
76#include <cam/ctl/ctl_ha.h>
77#include <cam/ctl/ctl_private.h>
78#include <cam/ctl/ctl_debug.h>
79#include <cam/ctl/ctl_scsi_all.h>
80#include <cam/ctl/ctl_error.h>
81
82struct ctl_softc *control_softc = NULL;
83
84/*
85 * The default is to run with CTL_DONE_THREAD turned on.  Completed
86 * transactions are queued for processing by the CTL work thread.  When
87 * CTL_DONE_THREAD is not defined, completed transactions are processed in
88 * the caller's context.
89 */
90#define CTL_DONE_THREAD
91
92/*
93 * Use the serial number and device ID provided by the backend, rather than
94 * making up our own.
95 */
96#define CTL_USE_BACKEND_SN
97
98/*
99 * Size and alignment macros needed for Copan-specific HA hardware.  These
100 * can go away when the HA code is re-written, and uses busdma for any
101 * hardware.
102 */
103#define	CTL_ALIGN_8B(target, source, type)				\
104	if (((uint32_t)source & 0x7) != 0)				\
105		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
106	else								\
107		target = (type)source;
108
109#define	CTL_SIZE_8B(target, size)					\
110	if ((size & 0x7) != 0)						\
111		target = size + (0x8 - (size & 0x7));			\
112	else								\
113		target = size;
114
115#define CTL_ALIGN_8B_MARGIN	16
116
117/*
118 * Template mode pages.
119 */
120
121/*
122 * Note that these are default values only.  The actual values will be
123 * filled in when the user does a mode sense.
124 */
125static struct copan_power_subpage power_page_default = {
126	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
127	/*subpage*/ PWR_SUBPAGE_CODE,
128	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
129			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
130	/*page_version*/ PWR_VERSION,
131	/* total_luns */ 26,
132	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
133	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
134		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0}
136};
137
138static struct copan_power_subpage power_page_changeable = {
139	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
140	/*subpage*/ PWR_SUBPAGE_CODE,
141	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
142			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
143	/*page_version*/ 0,
144	/* total_luns */ 0,
145	/* max_active_luns*/ 0,
146	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
147		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
148		      0, 0, 0, 0, 0, 0}
149};
150
151static struct copan_aps_subpage aps_page_default = {
152	APS_PAGE_CODE | SMPH_SPF, //page_code
153	APS_SUBPAGE_CODE, //subpage
154	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
155	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
156	APS_VERSION, //page_version
157	0, //lock_active
158	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
159	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
160	0, 0, 0, 0, 0} //reserved
161};
162
163static struct copan_aps_subpage aps_page_changeable = {
164	APS_PAGE_CODE | SMPH_SPF, //page_code
165	APS_SUBPAGE_CODE, //subpage
166	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
167	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
168	0, //page_version
169	0, //lock_active
170	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
171	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172	0, 0, 0, 0, 0} //reserved
173};
174
175static struct copan_debugconf_subpage debugconf_page_default = {
176	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
177	DBGCNF_SUBPAGE_CODE,		/* subpage */
178	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
179	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
180	DBGCNF_VERSION,			/* page_version */
181	{CTL_TIME_IO_DEFAULT_SECS>>8,
182	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
183};
184
185static struct copan_debugconf_subpage debugconf_page_changeable = {
186	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
187	DBGCNF_SUBPAGE_CODE,		/* subpage */
188	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
189	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
190	0,				/* page_version */
191	{0xff,0xff},			/* ctl_time_io_secs */
192};
193
194static struct scsi_format_page format_page_default = {
195	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
196	/*page_length*/sizeof(struct scsi_format_page) - 2,
197	/*tracks_per_zone*/ {0, 0},
198	/*alt_sectors_per_zone*/ {0, 0},
199	/*alt_tracks_per_zone*/ {0, 0},
200	/*alt_tracks_per_lun*/ {0, 0},
201	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
202			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
203	/*bytes_per_sector*/ {0, 0},
204	/*interleave*/ {0, 0},
205	/*track_skew*/ {0, 0},
206	/*cylinder_skew*/ {0, 0},
207	/*flags*/ SFP_HSEC,
208	/*reserved*/ {0, 0, 0}
209};
210
211static struct scsi_format_page format_page_changeable = {
212	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
213	/*page_length*/sizeof(struct scsi_format_page) - 2,
214	/*tracks_per_zone*/ {0, 0},
215	/*alt_sectors_per_zone*/ {0, 0},
216	/*alt_tracks_per_zone*/ {0, 0},
217	/*alt_tracks_per_lun*/ {0, 0},
218	/*sectors_per_track*/ {0, 0},
219	/*bytes_per_sector*/ {0, 0},
220	/*interleave*/ {0, 0},
221	/*track_skew*/ {0, 0},
222	/*cylinder_skew*/ {0, 0},
223	/*flags*/ 0,
224	/*reserved*/ {0, 0, 0}
225};
226
227static struct scsi_rigid_disk_page rigid_disk_page_default = {
228	/*page_code*/SMS_RIGID_DISK_PAGE,
229	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
230	/*cylinders*/ {0, 0, 0},
231	/*heads*/ CTL_DEFAULT_HEADS,
232	/*start_write_precomp*/ {0, 0, 0},
233	/*start_reduced_current*/ {0, 0, 0},
234	/*step_rate*/ {0, 0},
235	/*landing_zone_cylinder*/ {0, 0, 0},
236	/*rpl*/ SRDP_RPL_DISABLED,
237	/*rotational_offset*/ 0,
238	/*reserved1*/ 0,
239	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
240			   CTL_DEFAULT_ROTATION_RATE & 0xff},
241	/*reserved2*/ {0, 0}
242};
243
244static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
245	/*page_code*/SMS_RIGID_DISK_PAGE,
246	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
247	/*cylinders*/ {0, 0, 0},
248	/*heads*/ 0,
249	/*start_write_precomp*/ {0, 0, 0},
250	/*start_reduced_current*/ {0, 0, 0},
251	/*step_rate*/ {0, 0},
252	/*landing_zone_cylinder*/ {0, 0, 0},
253	/*rpl*/ 0,
254	/*rotational_offset*/ 0,
255	/*reserved1*/ 0,
256	/*rotation_rate*/ {0, 0},
257	/*reserved2*/ {0, 0}
258};
259
260static struct scsi_caching_page caching_page_default = {
261	/*page_code*/SMS_CACHING_PAGE,
262	/*page_length*/sizeof(struct scsi_caching_page) - 2,
263	/*flags1*/ SCP_DISC | SCP_WCE,
264	/*ret_priority*/ 0,
265	/*disable_pf_transfer_len*/ {0xff, 0xff},
266	/*min_prefetch*/ {0, 0},
267	/*max_prefetch*/ {0xff, 0xff},
268	/*max_pf_ceiling*/ {0xff, 0xff},
269	/*flags2*/ 0,
270	/*cache_segments*/ 0,
271	/*cache_seg_size*/ {0, 0},
272	/*reserved*/ 0,
273	/*non_cache_seg_size*/ {0, 0, 0}
274};
275
276static struct scsi_caching_page caching_page_changeable = {
277	/*page_code*/SMS_CACHING_PAGE,
278	/*page_length*/sizeof(struct scsi_caching_page) - 2,
279	/*flags1*/ 0,
280	/*ret_priority*/ 0,
281	/*disable_pf_transfer_len*/ {0, 0},
282	/*min_prefetch*/ {0, 0},
283	/*max_prefetch*/ {0, 0},
284	/*max_pf_ceiling*/ {0, 0},
285	/*flags2*/ 0,
286	/*cache_segments*/ 0,
287	/*cache_seg_size*/ {0, 0},
288	/*reserved*/ 0,
289	/*non_cache_seg_size*/ {0, 0, 0}
290};
291
292static struct scsi_control_page control_page_default = {
293	/*page_code*/SMS_CONTROL_MODE_PAGE,
294	/*page_length*/sizeof(struct scsi_control_page) - 2,
295	/*rlec*/0,
296	/*queue_flags*/0,
297	/*eca_and_aen*/0,
298	/*reserved*/0,
299	/*aen_holdoff_period*/{0, 0}
300};
301
302static struct scsi_control_page control_page_changeable = {
303	/*page_code*/SMS_CONTROL_MODE_PAGE,
304	/*page_length*/sizeof(struct scsi_control_page) - 2,
305	/*rlec*/SCP_DSENSE,
306	/*queue_flags*/0,
307	/*eca_and_aen*/0,
308	/*reserved*/0,
309	/*aen_holdoff_period*/{0, 0}
310};
311
312
313/*
314 * XXX KDM move these into the softc.
315 */
316static int rcv_sync_msg;
317static int persis_offset;
318static uint8_t ctl_pause_rtr;
319static int     ctl_is_single = 1;
320static int     index_to_aps_page;
321
322SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
323
324/*
325 * Serial number (0x80), device id (0x83), and supported pages (0x00)
326 */
327#define SCSI_EVPD_NUM_SUPPORTED_PAGES	3
328
329static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
330				  int param);
331static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
332static int ctl_init(void);
333void ctl_shutdown(void);
334static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
335static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
336static void ctl_ioctl_online(void *arg);
337static void ctl_ioctl_offline(void *arg);
338static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
339static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
340static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
341static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
342static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
343static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
344static int ctl_ioctl_submit_wait(union ctl_io *io);
345static void ctl_ioctl_datamove(union ctl_io *io);
346static void ctl_ioctl_done(union ctl_io *io);
347static void ctl_ioctl_hard_startstop_callback(void *arg,
348					      struct cfi_metatask *metatask);
349static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
350static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
351			      struct ctl_ooa *ooa_hdr,
352			      struct ctl_ooa_entry *kern_entries);
353static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
354		     struct thread *td);
355uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
356uint32_t ctl_port_idx(int port_num);
357#ifdef unused
358static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
359				   uint32_t targ_target, uint32_t targ_lun,
360				   int can_wait);
361static void ctl_kfree_io(union ctl_io *io);
362#endif /* unused */
363static void ctl_free_io_internal(union ctl_io *io, int have_lock);
364static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
365			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
366static int ctl_free_lun(struct ctl_lun *lun);
367static void ctl_create_lun(struct ctl_be_lun *be_lun);
368/**
369static void ctl_failover_change_pages(struct ctl_softc *softc,
370				      struct ctl_scsiio *ctsio, int master);
371**/
372
373static int ctl_do_mode_select(union ctl_io *io);
374static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
375			   uint64_t res_key, uint64_t sa_res_key,
376			   uint8_t type, uint32_t residx,
377			   struct ctl_scsiio *ctsio,
378			   struct scsi_per_res_out *cdb,
379			   struct scsi_per_res_out_parms* param);
380static void ctl_pro_preempt_other(struct ctl_lun *lun,
381				  union ctl_ha_msg *msg);
382static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
383static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
384static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
385static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
386static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
387static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
388static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
389static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
390static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
391					 union ctl_io *ooa_io);
392static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
393				union ctl_io *starting_io);
394static int ctl_check_blocked(struct ctl_lun *lun);
395static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
396				struct ctl_lun *lun,
397				struct ctl_cmd_entry *entry,
398				struct ctl_scsiio *ctsio);
399//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
400static void ctl_failover(void);
401static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
402			       struct ctl_scsiio *ctsio);
403static int ctl_scsiio(struct ctl_scsiio *ctsio);
404
405static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
406static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
407			    ctl_ua_type ua_type);
408static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
409			 ctl_ua_type ua_type);
410static int ctl_abort_task(union ctl_io *io);
411static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
412#ifdef CTL_IO_DELAY
413static void ctl_datamove_timer_wakeup(void *arg);
414static void ctl_done_timer_wakeup(void *arg);
415#endif /* CTL_IO_DELAY */
416
417static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
418static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
419static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
420static void ctl_datamove_remote_write(union ctl_io *io);
421static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
422static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
423static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
424static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
425				    ctl_ha_dt_cb callback);
426static void ctl_datamove_remote_read(union ctl_io *io);
427static void ctl_datamove_remote(union ctl_io *io);
428static int ctl_process_done(union ctl_io *io, int have_lock);
429static void ctl_work_thread(void *arg);
430
431/*
432 * Load the serialization table.  This isn't very pretty, but is probably
433 * the easiest way to do it.
434 */
435#include "ctl_ser_table.c"
436
437/*
438 * We only need to define open, close and ioctl routines for this driver.
439 */
440static struct cdevsw ctl_cdevsw = {
441	.d_version =	D_VERSION,
442	.d_flags =	0,
443	.d_open =	ctl_open,
444	.d_close =	ctl_close,
445	.d_ioctl =	ctl_ioctl,
446	.d_name =	"ctl",
447};
448
449
450MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
451
452static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
453
454static moduledata_t ctl_moduledata = {
455	"ctl",
456	ctl_module_event_handler,
457	NULL
458};
459
460DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
461MODULE_VERSION(ctl, 1);
462
463static void
464ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
465			    union ctl_ha_msg *msg_info)
466{
467	struct ctl_scsiio *ctsio;
468
469	if (msg_info->hdr.original_sc == NULL) {
470		printf("%s: original_sc == NULL!\n", __func__);
471		/* XXX KDM now what? */
472		return;
473	}
474
475	ctsio = &msg_info->hdr.original_sc->scsiio;
476	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
477	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
478	ctsio->io_hdr.status = msg_info->hdr.status;
479	ctsio->scsi_status = msg_info->scsi.scsi_status;
480	ctsio->sense_len = msg_info->scsi.sense_len;
481	ctsio->sense_residual = msg_info->scsi.sense_residual;
482	ctsio->residual = msg_info->scsi.residual;
483	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
484	       sizeof(ctsio->sense_data));
485	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
486	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
487	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
488	ctl_wakeup_thread();
489}
490
491static void
492ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
493				union ctl_ha_msg *msg_info)
494{
495	struct ctl_scsiio *ctsio;
496
497	if (msg_info->hdr.serializing_sc == NULL) {
498		printf("%s: serializing_sc == NULL!\n", __func__);
499		/* XXX KDM now what? */
500		return;
501	}
502
503	ctsio = &msg_info->hdr.serializing_sc->scsiio;
504#if 0
505	/*
506	 * Attempt to catch the situation where an I/O has
507	 * been freed, and we're using it again.
508	 */
509	if (ctsio->io_hdr.io_type == 0xff) {
510		union ctl_io *tmp_io;
511		tmp_io = (union ctl_io *)ctsio;
512		printf("%s: %p use after free!\n", __func__,
513		       ctsio);
514		printf("%s: type %d msg %d cdb %x iptl: "
515		       "%d:%d:%d:%d tag 0x%04x "
516		       "flag %#x status %x\n",
517			__func__,
518			tmp_io->io_hdr.io_type,
519			tmp_io->io_hdr.msg_type,
520			tmp_io->scsiio.cdb[0],
521			tmp_io->io_hdr.nexus.initid.id,
522			tmp_io->io_hdr.nexus.targ_port,
523			tmp_io->io_hdr.nexus.targ_target.id,
524			tmp_io->io_hdr.nexus.targ_lun,
525			(tmp_io->io_hdr.io_type ==
526			CTL_IO_TASK) ?
527			tmp_io->taskio.tag_num :
528			tmp_io->scsiio.tag_num,
529		        tmp_io->io_hdr.flags,
530			tmp_io->io_hdr.status);
531	}
532#endif
533	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
534	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
535	ctl_wakeup_thread();
536}
537
538/*
539 * ISC (Inter Shelf Communication) event handler.  Events from the HA
540 * subsystem come in here.
541 */
542static void
543ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
544{
545	struct ctl_softc *ctl_softc;
546	union ctl_io *io;
547	struct ctl_prio *presio;
548	ctl_ha_status isc_status;
549
550	ctl_softc = control_softc;
551	io = NULL;
552
553
554#if 0
555	printf("CTL: Isc Msg event %d\n", event);
556#endif
557	if (event == CTL_HA_EVT_MSG_RECV) {
558		union ctl_ha_msg msg_info;
559
560		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
561					     sizeof(msg_info), /*wait*/ 0);
562#if 0
563		printf("CTL: msg_type %d\n", msg_info.msg_type);
564#endif
565		if (isc_status != 0) {
566			printf("Error receiving message, status = %d\n",
567			       isc_status);
568			return;
569		}
570		mtx_lock(&ctl_softc->ctl_lock);
571
572		switch (msg_info.hdr.msg_type) {
573		case CTL_MSG_SERIALIZE:
574#if 0
575			printf("Serialize\n");
576#endif
577			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
578			if (io == NULL) {
579				printf("ctl_isc_event_handler: can't allocate "
580				       "ctl_io!\n");
581				/* Bad Juju */
582				/* Need to set busy and send msg back */
583				mtx_unlock(&ctl_softc->ctl_lock);
584				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
585				msg_info.hdr.status = CTL_SCSI_ERROR;
586				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
587				msg_info.scsi.sense_len = 0;
588			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
589				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
590				}
591				goto bailout;
592			}
593			ctl_zero_io(io);
594			// populate ctsio from msg_info
595			io->io_hdr.io_type = CTL_IO_SCSI;
596			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
597			io->io_hdr.original_sc = msg_info.hdr.original_sc;
598#if 0
599			printf("pOrig %x\n", (int)msg_info.original_sc);
600#endif
601			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
602					    CTL_FLAG_IO_ACTIVE;
603			/*
604			 * If we're in serialization-only mode, we don't
605			 * want to go through full done processing.  Thus
606			 * the COPY flag.
607			 *
608			 * XXX KDM add another flag that is more specific.
609			 */
610			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
611				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
612			io->io_hdr.nexus = msg_info.hdr.nexus;
613#if 0
614			printf("targ %d, port %d, iid %d, lun %d\n",
615			       io->io_hdr.nexus.targ_target.id,
616			       io->io_hdr.nexus.targ_port,
617			       io->io_hdr.nexus.initid.id,
618			       io->io_hdr.nexus.targ_lun);
619#endif
620			io->scsiio.tag_num = msg_info.scsi.tag_num;
621			io->scsiio.tag_type = msg_info.scsi.tag_type;
622			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
623			       CTL_MAX_CDBLEN);
624			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
625				struct ctl_cmd_entry *entry;
626				uint8_t opcode;
627
628				opcode = io->scsiio.cdb[0];
629				entry = &ctl_cmd_table[opcode];
630				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
631				io->io_hdr.flags |=
632					entry->flags & CTL_FLAG_DATA_MASK;
633			}
634			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
635					   &io->io_hdr, links);
636			ctl_wakeup_thread();
637			break;
638
639		/* Performed on the Originating SC, XFER mode only */
640		case CTL_MSG_DATAMOVE: {
641			struct ctl_sg_entry *sgl;
642			int i, j;
643
644			io = msg_info.hdr.original_sc;
645			if (io == NULL) {
646				printf("%s: original_sc == NULL!\n", __func__);
647				/* XXX KDM do something here */
648				break;
649			}
650			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
651			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
652			/*
653			 * Keep track of this, we need to send it back over
654			 * when the datamove is complete.
655			 */
656			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
657
658			if (msg_info.dt.sg_sequence == 0) {
659				/*
660				 * XXX KDM we use the preallocated S/G list
661				 * here, but we'll need to change this to
662				 * dynamic allocation if we need larger S/G
663				 * lists.
664				 */
665				if (msg_info.dt.kern_sg_entries >
666				    sizeof(io->io_hdr.remote_sglist) /
667				    sizeof(io->io_hdr.remote_sglist[0])) {
668					printf("%s: number of S/G entries "
669					    "needed %u > allocated num %zd\n",
670					    __func__,
671					    msg_info.dt.kern_sg_entries,
672					    sizeof(io->io_hdr.remote_sglist)/
673					    sizeof(io->io_hdr.remote_sglist[0]));
674
675					/*
676					 * XXX KDM send a message back to
677					 * the other side to shut down the
678					 * DMA.  The error will come back
679					 * through via the normal channel.
680					 */
681					break;
682				}
683				sgl = io->io_hdr.remote_sglist;
684				memset(sgl, 0,
685				       sizeof(io->io_hdr.remote_sglist));
686
687				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
688
689				io->scsiio.kern_sg_entries =
690					msg_info.dt.kern_sg_entries;
691				io->scsiio.rem_sg_entries =
692					msg_info.dt.kern_sg_entries;
693				io->scsiio.kern_data_len =
694					msg_info.dt.kern_data_len;
695				io->scsiio.kern_total_len =
696					msg_info.dt.kern_total_len;
697				io->scsiio.kern_data_resid =
698					msg_info.dt.kern_data_resid;
699				io->scsiio.kern_rel_offset =
700					msg_info.dt.kern_rel_offset;
701				/*
702				 * Clear out per-DMA flags.
703				 */
704				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
705				/*
706				 * Add per-DMA flags that are set for this
707				 * particular DMA request.
708				 */
709				io->io_hdr.flags |= msg_info.dt.flags &
710						    CTL_FLAG_RDMA_MASK;
711			} else
712				sgl = (struct ctl_sg_entry *)
713					io->scsiio.kern_data_ptr;
714
715			for (i = msg_info.dt.sent_sg_entries, j = 0;
716			     i < (msg_info.dt.sent_sg_entries +
717			     msg_info.dt.cur_sg_entries); i++, j++) {
718				sgl[i].addr = msg_info.dt.sg_list[j].addr;
719				sgl[i].len = msg_info.dt.sg_list[j].len;
720
721#if 0
722				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
723				       __func__,
724				       msg_info.dt.sg_list[j].addr,
725				       msg_info.dt.sg_list[j].len,
726				       sgl[i].addr, sgl[i].len, j, i);
727#endif
728			}
729#if 0
730			memcpy(&sgl[msg_info.dt.sent_sg_entries],
731			       msg_info.dt.sg_list,
732			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
733#endif
734
735			/*
736			 * If this is the last piece of the I/O, we've got
737			 * the full S/G list.  Queue processing in the thread.
738			 * Otherwise wait for the next piece.
739			 */
740			if (msg_info.dt.sg_last != 0) {
741				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
742						   &io->io_hdr, links);
743				ctl_wakeup_thread();
744			}
745			break;
746		}
747		/* Performed on the Serializing (primary) SC, XFER mode only */
748		case CTL_MSG_DATAMOVE_DONE: {
749			if (msg_info.hdr.serializing_sc == NULL) {
750				printf("%s: serializing_sc == NULL!\n",
751				       __func__);
752				/* XXX KDM now what? */
753				break;
754			}
755			/*
756			 * We grab the sense information here in case
757			 * there was a failure, so we can return status
758			 * back to the initiator.
759			 */
760			io = msg_info.hdr.serializing_sc;
761			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
762			io->io_hdr.status = msg_info.hdr.status;
763			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
764			io->scsiio.sense_len = msg_info.scsi.sense_len;
765			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
766			io->io_hdr.port_status = msg_info.scsi.fetd_status;
767			io->scsiio.residual = msg_info.scsi.residual;
768			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
769			       sizeof(io->scsiio.sense_data));
770
771			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
772					   &io->io_hdr, links);
773			ctl_wakeup_thread();
774			break;
775		}
776
777		/* Preformed on Originating SC, SER_ONLY mode */
778		case CTL_MSG_R2R:
779			io = msg_info.hdr.original_sc;
780			if (io == NULL) {
781				printf("%s: Major Bummer\n", __func__);
782				mtx_unlock(&ctl_softc->ctl_lock);
783				return;
784			} else {
785#if 0
786				printf("pOrig %x\n",(int) ctsio);
787#endif
788			}
789			io->io_hdr.msg_type = CTL_MSG_R2R;
790			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
791			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
792					   &io->io_hdr, links);
793			ctl_wakeup_thread();
794			break;
795
796		/*
797		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
798		 * mode.
799		 * Performed on the Originating (i.e. secondary) SC in XFER
800		 * mode
801		 */
802		case CTL_MSG_FINISH_IO:
803			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
804				ctl_isc_handler_finish_xfer(ctl_softc,
805							    &msg_info);
806			else
807				ctl_isc_handler_finish_ser_only(ctl_softc,
808								&msg_info);
809			break;
810
811		/* Preformed on Originating SC */
812		case CTL_MSG_BAD_JUJU:
813			io = msg_info.hdr.original_sc;
814			if (io == NULL) {
815				printf("%s: Bad JUJU!, original_sc is NULL!\n",
816				       __func__);
817				break;
818			}
819			ctl_copy_sense_data(&msg_info, io);
820			/*
821			 * IO should have already been cleaned up on other
822			 * SC so clear this flag so we won't send a message
823			 * back to finish the IO there.
824			 */
825			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
826			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
827
828			/* io = msg_info.hdr.serializing_sc; */
829			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
830		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
831					   &io->io_hdr, links);
832			ctl_wakeup_thread();
833			break;
834
835		/* Handle resets sent from the other side */
836		case CTL_MSG_MANAGE_TASKS: {
837			struct ctl_taskio *taskio;
838			taskio = (struct ctl_taskio *)ctl_alloc_io(
839				(void *)ctl_softc->othersc_pool);
840			if (taskio == NULL) {
841				printf("ctl_isc_event_handler: can't allocate "
842				       "ctl_io!\n");
843				/* Bad Juju */
844				/* should I just call the proper reset func
845				   here??? */
846				mtx_unlock(&ctl_softc->ctl_lock);
847				goto bailout;
848			}
849			ctl_zero_io((union ctl_io *)taskio);
850			taskio->io_hdr.io_type = CTL_IO_TASK;
851			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
852			taskio->io_hdr.nexus = msg_info.hdr.nexus;
853			taskio->task_action = msg_info.task.task_action;
854			taskio->tag_num = msg_info.task.tag_num;
855			taskio->tag_type = msg_info.task.tag_type;
856#ifdef CTL_TIME_IO
857			taskio->io_hdr.start_time = time_uptime;
858			getbintime(&taskio->io_hdr.start_bt);
859#if 0
860			cs_prof_gettime(&taskio->io_hdr.start_ticks);
861#endif
862#endif /* CTL_TIME_IO */
863		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
864					   &taskio->io_hdr, links);
865			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
866			ctl_wakeup_thread();
867			break;
868		}
869		/* Persistent Reserve action which needs attention */
870		case CTL_MSG_PERS_ACTION:
871			presio = (struct ctl_prio *)ctl_alloc_io(
872				(void *)ctl_softc->othersc_pool);
873			if (presio == NULL) {
874				printf("ctl_isc_event_handler: can't allocate "
875				       "ctl_io!\n");
876				/* Bad Juju */
877				/* Need to set busy and send msg back */
878				mtx_unlock(&ctl_softc->ctl_lock);
879				goto bailout;
880			}
881			ctl_zero_io((union ctl_io *)presio);
882			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
883			presio->pr_msg = msg_info.pr;
884		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
885					   &presio->io_hdr, links);
886			ctl_wakeup_thread();
887			break;
888		case CTL_MSG_SYNC_FE:
889			rcv_sync_msg = 1;
890			break;
891		case CTL_MSG_APS_LOCK: {
892			// It's quicker to execute this then to
893			// queue it.
894			struct ctl_lun *lun;
895			struct ctl_page_index *page_index;
896			struct copan_aps_subpage *current_sp;
897			uint32_t targ_lun;
898
899			targ_lun = msg_info.hdr.nexus.targ_lun;
900			if (msg_info.hdr.nexus.lun_map_fn != NULL)
901				targ_lun = msg_info.hdr.nexus.lun_map_fn(msg_info.hdr.nexus.lun_map_arg, targ_lun);
902
903			lun = ctl_softc->ctl_luns[targ_lun];
904			page_index = &lun->mode_pages.index[index_to_aps_page];
905			current_sp = (struct copan_aps_subpage *)
906				     (page_index->page_data +
907				     (page_index->page_len * CTL_PAGE_CURRENT));
908
909			current_sp->lock_active = msg_info.aps.lock_flag;
910		        break;
911		}
912		default:
913		        printf("How did I get here?\n");
914		}
915		mtx_unlock(&ctl_softc->ctl_lock);
916	} else if (event == CTL_HA_EVT_MSG_SENT) {
917		if (param != CTL_HA_STATUS_SUCCESS) {
918			printf("Bad status from ctl_ha_msg_send status %d\n",
919			       param);
920		}
921		return;
922	} else if (event == CTL_HA_EVT_DISCONNECT) {
923		printf("CTL: Got a disconnect from Isc\n");
924		return;
925	} else {
926		printf("ctl_isc_event_handler: Unknown event %d\n", event);
927		return;
928	}
929
930bailout:
931	return;
932}
933
934static void
935ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
936{
937	struct scsi_sense_data *sense;
938
939	sense = &dest->scsiio.sense_data;
940	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
941	dest->scsiio.scsi_status = src->scsi.scsi_status;
942	dest->scsiio.sense_len = src->scsi.sense_len;
943	dest->io_hdr.status = src->hdr.status;
944}
945
946static int
947ctl_init(void)
948{
949	struct ctl_softc *softc;
950	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
951	struct ctl_frontend *fe;
952	struct ctl_lun *lun;
953        uint8_t sc_id =0;
954#if 0
955	int i;
956#endif
957	int error, retval;
958	//int isc_retval;
959
960	retval = 0;
961	ctl_pause_rtr = 0;
962        rcv_sync_msg = 0;
963
964	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
965			       M_WAITOK | M_ZERO);
966	softc = control_softc;
967
968	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
969			      "cam/ctl");
970
971	softc->dev->si_drv1 = softc;
972
973	/*
974	 * By default, return a "bad LUN" peripheral qualifier for unknown
975	 * LUNs.  The user can override this default using the tunable or
976	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
977	 */
978	softc->inquiry_pq_no_lun = 1;
979	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
980			  &softc->inquiry_pq_no_lun);
981	sysctl_ctx_init(&softc->sysctl_ctx);
982	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
983		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
984		CTLFLAG_RD, 0, "CAM Target Layer");
985
986	if (softc->sysctl_tree == NULL) {
987		printf("%s: unable to allocate sysctl tree\n", __func__);
988		destroy_dev(softc->dev);
989		free(control_softc, M_DEVBUF);
990		control_softc = NULL;
991		return (ENOMEM);
992	}
993
994	SYSCTL_ADD_INT(&softc->sysctl_ctx,
995		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
996		       "inquiry_pq_no_lun", CTLFLAG_RW,
997		       &softc->inquiry_pq_no_lun, 0,
998		       "Report no lun possible for invalid LUNs");
999
1000	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1001	softc->open_count = 0;
1002
1003	/*
1004	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1005	 * the drive.
1006	 */
1007	softc->flags = CTL_FLAG_REAL_SYNC;
1008
1009	/*
1010	 * In Copan's HA scheme, the "master" and "slave" roles are
1011	 * figured out through the slot the controller is in.  Although it
1012	 * is an active/active system, someone has to be in charge.
1013 	 */
1014#ifdef NEEDTOPORT
1015        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1016#endif
1017
1018        if (sc_id == 0) {
1019		softc->flags |= CTL_FLAG_MASTER_SHELF;
1020		persis_offset = 0;
1021	} else
1022		persis_offset = CTL_MAX_INITIATORS;
1023
1024	/*
1025	 * XXX KDM need to figure out where we want to get our target ID
1026	 * and WWID.  Is it different on each port?
1027	 */
1028	softc->target.id = 0;
1029	softc->target.wwid[0] = 0x12345678;
1030	softc->target.wwid[1] = 0x87654321;
1031	STAILQ_INIT(&softc->lun_list);
1032	STAILQ_INIT(&softc->pending_lun_queue);
1033	STAILQ_INIT(&softc->task_queue);
1034	STAILQ_INIT(&softc->incoming_queue);
1035	STAILQ_INIT(&softc->rtr_queue);
1036	STAILQ_INIT(&softc->done_queue);
1037	STAILQ_INIT(&softc->isc_queue);
1038	STAILQ_INIT(&softc->fe_list);
1039	STAILQ_INIT(&softc->be_list);
1040	STAILQ_INIT(&softc->io_pools);
1041
1042	lun = &softc->lun;
1043
1044	/*
1045	 * We don't bother calling these with ctl_lock held here, because,
1046	 * in theory, no one else can try to do anything while we're in our
1047	 * module init routine.
1048	 */
1049	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1050			    &internal_pool)!= 0){
1051		printf("ctl: can't allocate %d entry internal pool, "
1052		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1053		return (ENOMEM);
1054	}
1055
1056	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1057			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1058		printf("ctl: can't allocate %d entry emergency pool, "
1059		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1060		ctl_pool_free(softc, internal_pool);
1061		return (ENOMEM);
1062	}
1063
1064	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1065	                    &other_pool) != 0)
1066	{
1067		printf("ctl: can't allocate %d entry other SC pool, "
1068		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1069		ctl_pool_free(softc, internal_pool);
1070		ctl_pool_free(softc, emergency_pool);
1071		return (ENOMEM);
1072	}
1073
1074	softc->internal_pool = internal_pool;
1075	softc->emergency_pool = emergency_pool;
1076	softc->othersc_pool = other_pool;
1077
1078	mtx_lock(&softc->ctl_lock);
1079	ctl_pool_acquire(internal_pool);
1080	ctl_pool_acquire(emergency_pool);
1081	ctl_pool_acquire(other_pool);
1082	mtx_unlock(&softc->ctl_lock);
1083
1084	/*
1085	 * We used to allocate a processor LUN here.  The new scheme is to
1086	 * just let the user allocate LUNs as he sees fit.
1087	 */
1088#if 0
1089	mtx_lock(&softc->ctl_lock);
1090	ctl_alloc_lun(softc, lun, /*be_lun*/NULL, /*target*/softc->target);
1091	mtx_unlock(&softc->ctl_lock);
1092#endif
1093
1094	error = kproc_create(ctl_work_thread, softc, &softc->work_thread, 0, 0,
1095			 "ctl_thrd");
1096	if (error != 0) {
1097		printf("error creating CTL work thread!\n");
1098		mtx_lock(&softc->ctl_lock);
1099		ctl_free_lun(lun);
1100		ctl_pool_free(softc, internal_pool);
1101		ctl_pool_free(softc, emergency_pool);
1102		ctl_pool_free(softc, other_pool);
1103		mtx_unlock(&softc->ctl_lock);
1104		return (error);
1105	}
1106	if (bootverbose)
1107		printf("ctl: CAM Target Layer loaded\n");
1108
1109	/*
1110	 * Initialize the initiator and portname mappings
1111	 */
1112	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1113
1114	/*
1115	 * Initialize the ioctl front end.
1116	 */
1117	fe = &softc->ioctl_info.fe;
1118	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1119	fe->port_type = CTL_PORT_IOCTL;
1120	fe->num_requested_ctl_io = 100;
1121	fe->port_name = softc->ioctl_info.port_name;
1122	fe->port_online = ctl_ioctl_online;
1123	fe->port_offline = ctl_ioctl_offline;
1124	fe->onoff_arg = &softc->ioctl_info;
1125	fe->targ_enable = ctl_ioctl_targ_enable;
1126	fe->targ_disable = ctl_ioctl_targ_disable;
1127	fe->lun_enable = ctl_ioctl_lun_enable;
1128	fe->lun_disable = ctl_ioctl_lun_disable;
1129	fe->targ_lun_arg = &softc->ioctl_info;
1130	fe->fe_datamove = ctl_ioctl_datamove;
1131	fe->fe_done = ctl_ioctl_done;
1132	fe->max_targets = 15;
1133	fe->max_target_id = 15;
1134
1135	if (ctl_frontend_register(&softc->ioctl_info.fe,
1136	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1137		printf("ctl: ioctl front end registration failed, will "
1138		       "continue anyway\n");
1139	}
1140
1141#ifdef CTL_IO_DELAY
1142	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1143		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1144		       sizeof(struct callout), CTL_TIMER_BYTES);
1145		return (EINVAL);
1146	}
1147#endif /* CTL_IO_DELAY */
1148
1149	return (0);
1150}
1151
1152void
1153ctl_shutdown(void)
1154{
1155	struct ctl_softc *softc;
1156	struct ctl_lun *lun, *next_lun;
1157	struct ctl_io_pool *pool, *next_pool;
1158
1159	softc = (struct ctl_softc *)control_softc;
1160
1161	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1162		printf("ctl: ioctl front end deregistration failed\n");
1163
1164	mtx_lock(&softc->ctl_lock);
1165
1166	/*
1167	 * Free up each LUN.
1168	 */
1169	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1170		next_lun = STAILQ_NEXT(lun, links);
1171		ctl_free_lun(lun);
1172	}
1173
1174	/*
1175	 * This will rip the rug out from under any FETDs or anyone else
1176	 * that has a pool allocated.  Since we increment our module
1177	 * refcount any time someone outside the main CTL module allocates
1178	 * a pool, we shouldn't have any problems here.  The user won't be
1179	 * able to unload the CTL module until client modules have
1180	 * successfully unloaded.
1181	 */
1182	for (pool = STAILQ_FIRST(&softc->io_pools); pool != NULL;
1183	     pool = next_pool) {
1184		next_pool = STAILQ_NEXT(pool, links);
1185		ctl_pool_free(softc, pool);
1186	}
1187
1188	mtx_unlock(&softc->ctl_lock);
1189
1190#if 0
1191	ctl_shutdown_thread(softc->work_thread);
1192#endif
1193
1194	mtx_destroy(&softc->ctl_lock);
1195
1196	destroy_dev(softc->dev);
1197
1198	sysctl_ctx_free(&softc->sysctl_ctx);
1199
1200	free(control_softc, M_DEVBUF);
1201	control_softc = NULL;
1202
1203	if (bootverbose)
1204		printf("ctl: CAM Target Layer unloaded\n");
1205}
1206
1207static int
1208ctl_module_event_handler(module_t mod, int what, void *arg)
1209{
1210
1211	switch (what) {
1212	case MOD_LOAD:
1213		return (ctl_init());
1214	case MOD_UNLOAD:
1215		return (EBUSY);
1216	default:
1217		return (EOPNOTSUPP);
1218	}
1219}
1220
1221/*
1222 * XXX KDM should we do some access checks here?  Bump a reference count to
1223 * prevent a CTL module from being unloaded while someone has it open?
1224 */
1225static int
1226ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1227{
1228	return (0);
1229}
1230
1231static int
1232ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1233{
1234	return (0);
1235}
1236
1237int
1238ctl_port_enable(ctl_port_type port_type)
1239{
1240	struct ctl_softc *softc;
1241	struct ctl_frontend *fe;
1242
1243	if (ctl_is_single == 0) {
1244		union ctl_ha_msg msg_info;
1245		int isc_retval;
1246
1247#if 0
1248		printf("%s: HA mode, synchronizing frontend enable\n",
1249		        __func__);
1250#endif
1251		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1252	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1253		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1254			printf("Sync msg send error retval %d\n", isc_retval);
1255		}
1256		if (!rcv_sync_msg) {
1257			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1258			        sizeof(msg_info), 1);
1259		}
1260#if 0
1261        	printf("CTL:Frontend Enable\n");
1262	} else {
1263		printf("%s: single mode, skipping frontend synchronization\n",
1264		        __func__);
1265#endif
1266	}
1267
1268	softc = control_softc;
1269
1270	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1271		if (port_type & fe->port_type)
1272		{
1273#if 0
1274			printf("port %d\n", fe->targ_port);
1275#endif
1276			ctl_frontend_online(fe);
1277		}
1278	}
1279
1280	return (0);
1281}
1282
1283int
1284ctl_port_disable(ctl_port_type port_type)
1285{
1286	struct ctl_softc *softc;
1287	struct ctl_frontend *fe;
1288
1289	softc = control_softc;
1290
1291	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1292		if (port_type & fe->port_type)
1293			ctl_frontend_offline(fe);
1294	}
1295
1296	return (0);
1297}
1298
1299/*
1300 * Returns 0 for success, 1 for failure.
1301 * Currently the only failure mode is if there aren't enough entries
1302 * allocated.  So, in case of a failure, look at num_entries_dropped,
1303 * reallocate and try again.
1304 */
1305int
1306ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1307	      int *num_entries_filled, int *num_entries_dropped,
1308	      ctl_port_type port_type, int no_virtual)
1309{
1310	struct ctl_softc *softc;
1311	struct ctl_frontend *fe;
1312	int entries_dropped, entries_filled;
1313	int retval;
1314	int i;
1315
1316	softc = control_softc;
1317
1318	retval = 0;
1319	entries_filled = 0;
1320	entries_dropped = 0;
1321
1322	i = 0;
1323	mtx_lock(&softc->ctl_lock);
1324	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1325		struct ctl_port_entry *entry;
1326
1327		if ((fe->port_type & port_type) == 0)
1328			continue;
1329
1330		if ((no_virtual != 0)
1331		 && (fe->virtual_port != 0))
1332			continue;
1333
1334		if (entries_filled >= num_entries_alloced) {
1335			entries_dropped++;
1336			continue;
1337		}
1338		entry = &entries[i];
1339
1340		entry->port_type = fe->port_type;
1341		strlcpy(entry->port_name, fe->port_name,
1342			sizeof(entry->port_name));
1343		entry->physical_port = fe->physical_port;
1344		entry->virtual_port = fe->virtual_port;
1345		entry->wwnn = fe->wwnn;
1346		entry->wwpn = fe->wwpn;
1347
1348		i++;
1349		entries_filled++;
1350	}
1351
1352	mtx_unlock(&softc->ctl_lock);
1353
1354	if (entries_dropped > 0)
1355		retval = 1;
1356
1357	*num_entries_dropped = entries_dropped;
1358	*num_entries_filled = entries_filled;
1359
1360	return (retval);
1361}
1362
1363static void
1364ctl_ioctl_online(void *arg)
1365{
1366	struct ctl_ioctl_info *ioctl_info;
1367
1368	ioctl_info = (struct ctl_ioctl_info *)arg;
1369
1370	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1371}
1372
1373static void
1374ctl_ioctl_offline(void *arg)
1375{
1376	struct ctl_ioctl_info *ioctl_info;
1377
1378	ioctl_info = (struct ctl_ioctl_info *)arg;
1379
1380	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1381}
1382
1383/*
1384 * Remove an initiator by port number and initiator ID.
1385 * Returns 0 for success, 1 for failure.
1386 */
1387int
1388ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1389{
1390	struct ctl_softc *softc;
1391
1392	softc = control_softc;
1393
1394	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1395
1396	if ((targ_port < 0)
1397	 || (targ_port > CTL_MAX_PORTS)) {
1398		printf("%s: invalid port number %d\n", __func__, targ_port);
1399		return (1);
1400	}
1401	if (iid > CTL_MAX_INIT_PER_PORT) {
1402		printf("%s: initiator ID %u > maximun %u!\n",
1403		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1404		return (1);
1405	}
1406
1407	mtx_lock(&softc->ctl_lock);
1408
1409	softc->wwpn_iid[targ_port][iid].in_use = 0;
1410
1411	mtx_unlock(&softc->ctl_lock);
1412
1413	return (0);
1414}
1415
1416/*
1417 * Add an initiator to the initiator map.
1418 * Returns 0 for success, 1 for failure.
1419 */
1420int
1421ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1422{
1423	struct ctl_softc *softc;
1424	int retval;
1425
1426	softc = control_softc;
1427
1428	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1429
1430	retval = 0;
1431
1432	if ((targ_port < 0)
1433	 || (targ_port > CTL_MAX_PORTS)) {
1434		printf("%s: invalid port number %d\n", __func__, targ_port);
1435		return (1);
1436	}
1437	if (iid > CTL_MAX_INIT_PER_PORT) {
1438		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1439		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1440		return (1);
1441	}
1442
1443	mtx_lock(&softc->ctl_lock);
1444
1445	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1446		/*
1447		 * We don't treat this as an error.
1448		 */
1449		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1450			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1451			       __func__, targ_port, iid, (uintmax_t)wwpn);
1452			goto bailout;
1453		}
1454
1455		/*
1456		 * This is an error, but what do we do about it?  The
1457		 * driver is telling us we have a new WWPN for this
1458		 * initiator ID, so we pretty much need to use it.
1459		 */
1460		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1461		       "still at that address\n", __func__, targ_port, iid,
1462		       (uintmax_t)wwpn,
1463		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1464
1465		/*
1466		 * XXX KDM clear have_ca and ua_pending on each LUN for
1467		 * this initiator.
1468		 */
1469	}
1470	softc->wwpn_iid[targ_port][iid].in_use = 1;
1471	softc->wwpn_iid[targ_port][iid].iid = iid;
1472	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1473	softc->wwpn_iid[targ_port][iid].port = targ_port;
1474
1475bailout:
1476
1477	mtx_unlock(&softc->ctl_lock);
1478
1479	return (retval);
1480}
1481
1482/*
1483 * XXX KDM should we pretend to do something in the target/lun
1484 * enable/disable functions?
1485 */
1486static int
1487ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1488{
1489	return (0);
1490}
1491
1492static int
1493ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1494{
1495	return (0);
1496}
1497
1498static int
1499ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1500{
1501	return (0);
1502}
1503
1504static int
1505ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1506{
1507	return (0);
1508}
1509
1510/*
1511 * Data movement routine for the CTL ioctl frontend port.
1512 */
1513static int
1514ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1515{
1516	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1517	struct ctl_sg_entry ext_entry, kern_entry;
1518	int ext_sglen, ext_sg_entries, kern_sg_entries;
1519	int ext_sg_start, ext_offset;
1520	int len_to_copy, len_copied;
1521	int kern_watermark, ext_watermark;
1522	int ext_sglist_malloced;
1523	int i, j;
1524
1525	ext_sglist_malloced = 0;
1526	ext_sg_start = 0;
1527	ext_offset = 0;
1528
1529	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1530
1531	/*
1532	 * If this flag is set, fake the data transfer.
1533	 */
1534	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1535		ctsio->ext_data_filled = ctsio->ext_data_len;
1536		goto bailout;
1537	}
1538
1539	/*
1540	 * To simplify things here, if we have a single buffer, stick it in
1541	 * a S/G entry and just make it a single entry S/G list.
1542	 */
1543	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1544		int len_seen;
1545
1546		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1547
1548		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1549							   M_WAITOK);
1550		ext_sglist_malloced = 1;
1551		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1552				   ext_sglen) != 0) {
1553			ctl_set_internal_failure(ctsio,
1554						 /*sks_valid*/ 0,
1555						 /*retry_count*/ 0);
1556			goto bailout;
1557		}
1558		ext_sg_entries = ctsio->ext_sg_entries;
1559		len_seen = 0;
1560		for (i = 0; i < ext_sg_entries; i++) {
1561			if ((len_seen + ext_sglist[i].len) >=
1562			     ctsio->ext_data_filled) {
1563				ext_sg_start = i;
1564				ext_offset = ctsio->ext_data_filled - len_seen;
1565				break;
1566			}
1567			len_seen += ext_sglist[i].len;
1568		}
1569	} else {
1570		ext_sglist = &ext_entry;
1571		ext_sglist->addr = ctsio->ext_data_ptr;
1572		ext_sglist->len = ctsio->ext_data_len;
1573		ext_sg_entries = 1;
1574		ext_sg_start = 0;
1575		ext_offset = ctsio->ext_data_filled;
1576	}
1577
1578	if (ctsio->kern_sg_entries > 0) {
1579		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1580		kern_sg_entries = ctsio->kern_sg_entries;
1581	} else {
1582		kern_sglist = &kern_entry;
1583		kern_sglist->addr = ctsio->kern_data_ptr;
1584		kern_sglist->len = ctsio->kern_data_len;
1585		kern_sg_entries = 1;
1586	}
1587
1588
1589	kern_watermark = 0;
1590	ext_watermark = ext_offset;
1591	len_copied = 0;
1592	for (i = ext_sg_start, j = 0;
1593	     i < ext_sg_entries && j < kern_sg_entries;) {
1594		uint8_t *ext_ptr, *kern_ptr;
1595
1596		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1597				      kern_sglist[j].len - kern_watermark);
1598
1599		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1600		ext_ptr = ext_ptr + ext_watermark;
1601		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1602			/*
1603			 * XXX KDM fix this!
1604			 */
1605			panic("need to implement bus address support");
1606#if 0
1607			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1608#endif
1609		} else
1610			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1611		kern_ptr = kern_ptr + kern_watermark;
1612
1613		kern_watermark += len_to_copy;
1614		ext_watermark += len_to_copy;
1615
1616		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1617		     CTL_FLAG_DATA_IN) {
1618			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1619					 "bytes to user\n", len_to_copy));
1620			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1621					 "to %p\n", kern_ptr, ext_ptr));
1622			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1623				ctl_set_internal_failure(ctsio,
1624							 /*sks_valid*/ 0,
1625							 /*retry_count*/ 0);
1626				goto bailout;
1627			}
1628		} else {
1629			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1630					 "bytes from user\n", len_to_copy));
1631			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1632					 "to %p\n", ext_ptr, kern_ptr));
1633			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1634				ctl_set_internal_failure(ctsio,
1635							 /*sks_valid*/ 0,
1636							 /*retry_count*/0);
1637				goto bailout;
1638			}
1639		}
1640
1641		len_copied += len_to_copy;
1642
1643		if (ext_sglist[i].len == ext_watermark) {
1644			i++;
1645			ext_watermark = 0;
1646		}
1647
1648		if (kern_sglist[j].len == kern_watermark) {
1649			j++;
1650			kern_watermark = 0;
1651		}
1652	}
1653
1654	ctsio->ext_data_filled += len_copied;
1655
1656	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1657			 "kern_sg_entries: %d\n", ext_sg_entries,
1658			 kern_sg_entries));
1659	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1660			 "kern_data_len = %d\n", ctsio->ext_data_len,
1661			 ctsio->kern_data_len));
1662
1663
1664	/* XXX KDM set residual?? */
1665bailout:
1666
1667	if (ext_sglist_malloced != 0)
1668		free(ext_sglist, M_CTL);
1669
1670	return (CTL_RETVAL_COMPLETE);
1671}
1672
1673/*
1674 * Serialize a command that went down the "wrong" side, and so was sent to
1675 * this controller for execution.  The logic is a little different than the
1676 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1677 * sent back to the other side, but in the success case, we execute the
1678 * command on this side (XFER mode) or tell the other side to execute it
1679 * (SER_ONLY mode).
1680 */
1681static int
1682ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1683{
1684	struct ctl_softc *ctl_softc;
1685	union ctl_ha_msg msg_info;
1686	struct ctl_lun *lun;
1687	int retval = 0;
1688	uint32_t targ_lun;
1689
1690	ctl_softc = control_softc;
1691	if (have_lock == 0)
1692		mtx_lock(&ctl_softc->ctl_lock);
1693
1694	targ_lun = ctsio->io_hdr.nexus.targ_lun;
1695	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
1696		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
1697	lun = ctl_softc->ctl_luns[targ_lun];
1698	if (lun==NULL)
1699	{
1700		/*
1701		 * Why isn't LUN defined? The other side wouldn't
1702		 * send a cmd if the LUN is undefined.
1703		 */
1704		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1705
1706		/* "Logical unit not supported" */
1707		ctl_set_sense_data(&msg_info.scsi.sense_data,
1708				   lun,
1709				   /*sense_format*/SSD_TYPE_NONE,
1710				   /*current_error*/ 1,
1711				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1712				   /*asc*/ 0x25,
1713				   /*ascq*/ 0x00,
1714				   SSD_ELEM_NONE);
1715
1716		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1717		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1718		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1719		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1720		msg_info.hdr.serializing_sc = NULL;
1721		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1722	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1723				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1724		}
1725		if (have_lock == 0)
1726			mtx_unlock(&ctl_softc->ctl_lock);
1727		return(1);
1728
1729	}
1730
1731    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1732
1733	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1734		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1735		 ooa_links))) {
1736	case CTL_ACTION_BLOCK:
1737		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1738		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1739				  blocked_links);
1740		break;
1741	case CTL_ACTION_PASS:
1742	case CTL_ACTION_SKIP:
1743		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1744			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1745			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1746					   &ctsio->io_hdr, links);
1747		} else {
1748
1749			/* send msg back to other side */
1750			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1751			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1752			msg_info.hdr.msg_type = CTL_MSG_R2R;
1753#if 0
1754			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1755#endif
1756		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1757			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1758			}
1759		}
1760		break;
1761	case CTL_ACTION_OVERLAP:
1762		/* OVERLAPPED COMMANDS ATTEMPTED */
1763		ctl_set_sense_data(&msg_info.scsi.sense_data,
1764				   lun,
1765				   /*sense_format*/SSD_TYPE_NONE,
1766				   /*current_error*/ 1,
1767				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1768				   /*asc*/ 0x4E,
1769				   /*ascq*/ 0x00,
1770				   SSD_ELEM_NONE);
1771
1772		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1773		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1774		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1775		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1776		msg_info.hdr.serializing_sc = NULL;
1777		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1778#if 0
1779		printf("BAD JUJU:Major Bummer Overlap\n");
1780#endif
1781		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1782		retval = 1;
1783		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1784		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1785		}
1786		break;
1787	case CTL_ACTION_OVERLAP_TAG:
1788		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1789		ctl_set_sense_data(&msg_info.scsi.sense_data,
1790				   lun,
1791				   /*sense_format*/SSD_TYPE_NONE,
1792				   /*current_error*/ 1,
1793				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1794				   /*asc*/ 0x4D,
1795				   /*ascq*/ ctsio->tag_num & 0xff,
1796				   SSD_ELEM_NONE);
1797
1798		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1799		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1800		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1801		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1802		msg_info.hdr.serializing_sc = NULL;
1803		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1804#if 0
1805		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1806#endif
1807		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1808		retval = 1;
1809		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1810		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1811		}
1812		break;
1813	case CTL_ACTION_ERROR:
1814	default:
1815		/* "Internal target failure" */
1816		ctl_set_sense_data(&msg_info.scsi.sense_data,
1817				   lun,
1818				   /*sense_format*/SSD_TYPE_NONE,
1819				   /*current_error*/ 1,
1820				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1821				   /*asc*/ 0x44,
1822				   /*ascq*/ 0x00,
1823				   SSD_ELEM_NONE);
1824
1825		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1826		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1827		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1828		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1829		msg_info.hdr.serializing_sc = NULL;
1830		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1831#if 0
1832		printf("BAD JUJU:Major Bummer HW Error\n");
1833#endif
1834		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1835		retval = 1;
1836		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1837		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1838		}
1839		break;
1840	}
1841	if (have_lock == 0)
1842		mtx_unlock(&ctl_softc->ctl_lock);
1843	return (retval);
1844}
1845
1846static int
1847ctl_ioctl_submit_wait(union ctl_io *io)
1848{
1849	struct ctl_fe_ioctl_params params;
1850	ctl_fe_ioctl_state last_state;
1851	int done, retval;
1852
1853	retval = 0;
1854
1855	bzero(&params, sizeof(params));
1856
1857	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1858	cv_init(&params.sem, "ctlioccv");
1859	params.state = CTL_IOCTL_INPROG;
1860	last_state = params.state;
1861
1862	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1863
1864	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1865
1866	/* This shouldn't happen */
1867	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1868		return (retval);
1869
1870	done = 0;
1871
1872	do {
1873		mtx_lock(&params.ioctl_mtx);
1874		/*
1875		 * Check the state here, and don't sleep if the state has
1876		 * already changed (i.e. wakeup has already occured, but we
1877		 * weren't waiting yet).
1878		 */
1879		if (params.state == last_state) {
1880			/* XXX KDM cv_wait_sig instead? */
1881			cv_wait(&params.sem, &params.ioctl_mtx);
1882		}
1883		last_state = params.state;
1884
1885		switch (params.state) {
1886		case CTL_IOCTL_INPROG:
1887			/* Why did we wake up? */
1888			/* XXX KDM error here? */
1889			mtx_unlock(&params.ioctl_mtx);
1890			break;
1891		case CTL_IOCTL_DATAMOVE:
1892			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1893
1894			/*
1895			 * change last_state back to INPROG to avoid
1896			 * deadlock on subsequent data moves.
1897			 */
1898			params.state = last_state = CTL_IOCTL_INPROG;
1899
1900			mtx_unlock(&params.ioctl_mtx);
1901			ctl_ioctl_do_datamove(&io->scsiio);
1902			/*
1903			 * Note that in some cases, most notably writes,
1904			 * this will queue the I/O and call us back later.
1905			 * In other cases, generally reads, this routine
1906			 * will immediately call back and wake us up,
1907			 * probably using our own context.
1908			 */
1909			io->scsiio.be_move_done(io);
1910			break;
1911		case CTL_IOCTL_DONE:
1912			mtx_unlock(&params.ioctl_mtx);
1913			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1914			done = 1;
1915			break;
1916		default:
1917			mtx_unlock(&params.ioctl_mtx);
1918			/* XXX KDM error here? */
1919			break;
1920		}
1921	} while (done == 0);
1922
1923	mtx_destroy(&params.ioctl_mtx);
1924	cv_destroy(&params.sem);
1925
1926	return (CTL_RETVAL_COMPLETE);
1927}
1928
1929static void
1930ctl_ioctl_datamove(union ctl_io *io)
1931{
1932	struct ctl_fe_ioctl_params *params;
1933
1934	params = (struct ctl_fe_ioctl_params *)
1935		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1936
1937	mtx_lock(&params->ioctl_mtx);
1938	params->state = CTL_IOCTL_DATAMOVE;
1939	cv_broadcast(&params->sem);
1940	mtx_unlock(&params->ioctl_mtx);
1941}
1942
1943static void
1944ctl_ioctl_done(union ctl_io *io)
1945{
1946	struct ctl_fe_ioctl_params *params;
1947
1948	params = (struct ctl_fe_ioctl_params *)
1949		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1950
1951	mtx_lock(&params->ioctl_mtx);
1952	params->state = CTL_IOCTL_DONE;
1953	cv_broadcast(&params->sem);
1954	mtx_unlock(&params->ioctl_mtx);
1955}
1956
1957static void
1958ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1959{
1960	struct ctl_fe_ioctl_startstop_info *sd_info;
1961
1962	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1963
1964	sd_info->hs_info.status = metatask->status;
1965	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1966	sd_info->hs_info.luns_complete =
1967		metatask->taskinfo.startstop.luns_complete;
1968	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1969
1970	cv_broadcast(&sd_info->sem);
1971}
1972
1973static void
1974ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1975{
1976	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1977
1978	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1979
1980	mtx_lock(fe_bbr_info->lock);
1981	fe_bbr_info->bbr_info->status = metatask->status;
1982	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1983	fe_bbr_info->wakeup_done = 1;
1984	mtx_unlock(fe_bbr_info->lock);
1985
1986	cv_broadcast(&fe_bbr_info->sem);
1987}
1988
1989/*
1990 * Returns 0 for success, errno for failure.
1991 */
1992static int
1993ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1994		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1995{
1996	union ctl_io *io;
1997	int retval;
1998
1999	retval = 0;
2000
2001	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
2002
2003	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2004	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2005	     ooa_links)) {
2006		struct ctl_ooa_entry *entry;
2007
2008		/*
2009		 * If we've got more than we can fit, just count the
2010		 * remaining entries.
2011		 */
2012		if (*cur_fill_num >= ooa_hdr->alloc_num)
2013			continue;
2014
2015		entry = &kern_entries[*cur_fill_num];
2016
2017		entry->tag_num = io->scsiio.tag_num;
2018		entry->lun_num = lun->lun;
2019#ifdef CTL_TIME_IO
2020		entry->start_bt = io->io_hdr.start_bt;
2021#endif
2022		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2023		entry->cdb_len = io->scsiio.cdb_len;
2024		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2025			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2026
2027		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2028			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2029
2030		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2031			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2032
2033		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2034			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2035
2036		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2037			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2038	}
2039
2040	return (retval);
2041}
2042
2043static void *
2044ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2045		 size_t error_str_len)
2046{
2047	void *kptr;
2048
2049	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2050
2051	if (copyin(user_addr, kptr, len) != 0) {
2052		snprintf(error_str, error_str_len, "Error copying %d bytes "
2053			 "from user address %p to kernel address %p", len,
2054			 user_addr, kptr);
2055		free(kptr, M_CTL);
2056		return (NULL);
2057	}
2058
2059	return (kptr);
2060}
2061
2062static void
2063ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2064{
2065	int i;
2066
2067	if (be_args == NULL)
2068		return;
2069
2070	for (i = 0; i < num_be_args; i++) {
2071		free(be_args[i].kname, M_CTL);
2072		free(be_args[i].kvalue, M_CTL);
2073	}
2074
2075	free(be_args, M_CTL);
2076}
2077
2078static struct ctl_be_arg *
2079ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2080		char *error_str, size_t error_str_len)
2081{
2082	struct ctl_be_arg *args;
2083	int i;
2084
2085	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2086				error_str, error_str_len);
2087
2088	if (args == NULL)
2089		goto bailout;
2090
2091	for (i = 0; i < num_be_args; i++) {
2092		args[i].kname = NULL;
2093		args[i].kvalue = NULL;
2094	}
2095
2096	for (i = 0; i < num_be_args; i++) {
2097		uint8_t *tmpptr;
2098
2099		args[i].kname = ctl_copyin_alloc(args[i].name,
2100			args[i].namelen, error_str, error_str_len);
2101		if (args[i].kname == NULL)
2102			goto bailout;
2103
2104		if (args[i].kname[args[i].namelen - 1] != '\0') {
2105			snprintf(error_str, error_str_len, "Argument %d "
2106				 "name is not NUL-terminated", i);
2107			goto bailout;
2108		}
2109
2110		args[i].kvalue = NULL;
2111
2112		tmpptr = ctl_copyin_alloc(args[i].value,
2113			args[i].vallen, error_str, error_str_len);
2114		if (tmpptr == NULL)
2115			goto bailout;
2116
2117		args[i].kvalue = tmpptr;
2118
2119		if ((args[i].flags & CTL_BEARG_ASCII)
2120		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2121			snprintf(error_str, error_str_len, "Argument %d "
2122				 "value is not NUL-terminated", i);
2123			goto bailout;
2124		}
2125	}
2126
2127	return (args);
2128bailout:
2129
2130	ctl_free_args(num_be_args, args);
2131
2132	return (NULL);
2133}
2134
2135/*
2136 * Escape characters that are illegal or not recommended in XML.
2137 */
2138int
2139ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2140{
2141	int retval;
2142
2143	retval = 0;
2144
2145	for (; *str; str++) {
2146		switch (*str) {
2147		case '&':
2148			retval = sbuf_printf(sb, "&amp;");
2149			break;
2150		case '>':
2151			retval = sbuf_printf(sb, "&gt;");
2152			break;
2153		case '<':
2154			retval = sbuf_printf(sb, "&lt;");
2155			break;
2156		default:
2157			retval = sbuf_putc(sb, *str);
2158			break;
2159		}
2160
2161		if (retval != 0)
2162			break;
2163
2164	}
2165
2166	return (retval);
2167}
2168
2169static int
2170ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2171	  struct thread *td)
2172{
2173	struct ctl_softc *softc;
2174	int retval;
2175
2176	softc = control_softc;
2177
2178	retval = 0;
2179
2180	switch (cmd) {
2181	case CTL_IO: {
2182		union ctl_io *io;
2183		void *pool_tmp;
2184
2185		/*
2186		 * If we haven't been "enabled", don't allow any SCSI I/O
2187		 * to this FETD.
2188		 */
2189		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2190			retval = -EPERM;
2191			break;
2192		}
2193
2194		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2195		if (io == NULL) {
2196			printf("ctl_ioctl: can't allocate ctl_io!\n");
2197			retval = -ENOSPC;
2198			break;
2199		}
2200
2201		/*
2202		 * Need to save the pool reference so it doesn't get
2203		 * spammed by the user's ctl_io.
2204		 */
2205		pool_tmp = io->io_hdr.pool;
2206
2207		memcpy(io, (void *)addr, sizeof(*io));
2208
2209		io->io_hdr.pool = pool_tmp;
2210		/*
2211		 * No status yet, so make sure the status is set properly.
2212		 */
2213		io->io_hdr.status = CTL_STATUS_NONE;
2214
2215		/*
2216		 * The user sets the initiator ID, target and LUN IDs.
2217		 */
2218		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2219		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2220		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2221		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2222			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2223
2224		retval = ctl_ioctl_submit_wait(io);
2225
2226		if (retval != 0) {
2227			ctl_free_io(io);
2228			break;
2229		}
2230
2231		memcpy((void *)addr, io, sizeof(*io));
2232
2233		/* return this to our pool */
2234		ctl_free_io(io);
2235
2236		break;
2237	}
2238	case CTL_ENABLE_PORT:
2239	case CTL_DISABLE_PORT:
2240	case CTL_SET_PORT_WWNS: {
2241		struct ctl_frontend *fe;
2242		struct ctl_port_entry *entry;
2243
2244		entry = (struct ctl_port_entry *)addr;
2245
2246		mtx_lock(&softc->ctl_lock);
2247		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2248			int action, done;
2249
2250			action = 0;
2251			done = 0;
2252
2253			if ((entry->port_type == CTL_PORT_NONE)
2254			 && (entry->targ_port == fe->targ_port)) {
2255				/*
2256				 * If the user only wants to enable or
2257				 * disable or set WWNs on a specific port,
2258				 * do the operation and we're done.
2259				 */
2260				action = 1;
2261				done = 1;
2262			} else if (entry->port_type & fe->port_type) {
2263				/*
2264				 * Compare the user's type mask with the
2265				 * particular frontend type to see if we
2266				 * have a match.
2267				 */
2268				action = 1;
2269				done = 0;
2270
2271				/*
2272				 * Make sure the user isn't trying to set
2273				 * WWNs on multiple ports at the same time.
2274				 */
2275				if (cmd == CTL_SET_PORT_WWNS) {
2276					printf("%s: Can't set WWNs on "
2277					       "multiple ports\n", __func__);
2278					retval = EINVAL;
2279					break;
2280				}
2281			}
2282			if (action != 0) {
2283				/*
2284				 * XXX KDM we have to drop the lock here,
2285				 * because the online/offline operations
2286				 * can potentially block.  We need to
2287				 * reference count the frontends so they
2288				 * can't go away,
2289				 */
2290				mtx_unlock(&softc->ctl_lock);
2291
2292				if (cmd == CTL_ENABLE_PORT) {
2293					struct ctl_lun *lun;
2294
2295					STAILQ_FOREACH(lun, &softc->lun_list,
2296						       links) {
2297						fe->lun_enable(fe->targ_lun_arg,
2298						    lun->target,
2299						    lun->lun);
2300					}
2301
2302					ctl_frontend_online(fe);
2303				} else if (cmd == CTL_DISABLE_PORT) {
2304					struct ctl_lun *lun;
2305
2306					ctl_frontend_offline(fe);
2307
2308					STAILQ_FOREACH(lun, &softc->lun_list,
2309						       links) {
2310						fe->lun_disable(
2311						    fe->targ_lun_arg,
2312						    lun->target,
2313						    lun->lun);
2314					}
2315				}
2316
2317				mtx_lock(&softc->ctl_lock);
2318
2319				if (cmd == CTL_SET_PORT_WWNS)
2320					ctl_frontend_set_wwns(fe,
2321					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2322					    1 : 0, entry->wwnn,
2323					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2324					    1 : 0, entry->wwpn);
2325			}
2326			if (done != 0)
2327				break;
2328		}
2329		mtx_unlock(&softc->ctl_lock);
2330		break;
2331	}
2332	case CTL_GET_PORT_LIST: {
2333		struct ctl_frontend *fe;
2334		struct ctl_port_list *list;
2335		int i;
2336
2337		list = (struct ctl_port_list *)addr;
2338
2339		if (list->alloc_len != (list->alloc_num *
2340		    sizeof(struct ctl_port_entry))) {
2341			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2342			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2343			       "%zu\n", __func__, list->alloc_len,
2344			       list->alloc_num, sizeof(struct ctl_port_entry));
2345			retval = EINVAL;
2346			break;
2347		}
2348		list->fill_len = 0;
2349		list->fill_num = 0;
2350		list->dropped_num = 0;
2351		i = 0;
2352		mtx_lock(&softc->ctl_lock);
2353		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2354			struct ctl_port_entry entry, *list_entry;
2355
2356			if (list->fill_num >= list->alloc_num) {
2357				list->dropped_num++;
2358				continue;
2359			}
2360
2361			entry.port_type = fe->port_type;
2362			strlcpy(entry.port_name, fe->port_name,
2363				sizeof(entry.port_name));
2364			entry.targ_port = fe->targ_port;
2365			entry.physical_port = fe->physical_port;
2366			entry.virtual_port = fe->virtual_port;
2367			entry.wwnn = fe->wwnn;
2368			entry.wwpn = fe->wwpn;
2369			if (fe->status & CTL_PORT_STATUS_ONLINE)
2370				entry.online = 1;
2371			else
2372				entry.online = 0;
2373
2374			list_entry = &list->entries[i];
2375
2376			retval = copyout(&entry, list_entry, sizeof(entry));
2377			if (retval != 0) {
2378				printf("%s: CTL_GET_PORT_LIST: copyout "
2379				       "returned %d\n", __func__, retval);
2380				break;
2381			}
2382			i++;
2383			list->fill_num++;
2384			list->fill_len += sizeof(entry);
2385		}
2386		mtx_unlock(&softc->ctl_lock);
2387
2388		/*
2389		 * If this is non-zero, we had a copyout fault, so there's
2390		 * probably no point in attempting to set the status inside
2391		 * the structure.
2392		 */
2393		if (retval != 0)
2394			break;
2395
2396		if (list->dropped_num > 0)
2397			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2398		else
2399			list->status = CTL_PORT_LIST_OK;
2400		break;
2401	}
2402	case CTL_DUMP_OOA: {
2403		struct ctl_lun *lun;
2404		union ctl_io *io;
2405		char printbuf[128];
2406		struct sbuf sb;
2407
2408		mtx_lock(&softc->ctl_lock);
2409		printf("Dumping OOA queues:\n");
2410		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2411			for (io = (union ctl_io *)TAILQ_FIRST(
2412			     &lun->ooa_queue); io != NULL;
2413			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2414			     ooa_links)) {
2415				sbuf_new(&sb, printbuf, sizeof(printbuf),
2416					 SBUF_FIXEDLEN);
2417				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2418					    (intmax_t)lun->lun,
2419					    io->scsiio.tag_num,
2420					    (io->io_hdr.flags &
2421					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2422					    (io->io_hdr.flags &
2423					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2424					    (io->io_hdr.flags &
2425					    CTL_FLAG_ABORT) ? " ABORT" : "",
2426			                    (io->io_hdr.flags &
2427		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2428				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2429				sbuf_finish(&sb);
2430				printf("%s\n", sbuf_data(&sb));
2431			}
2432		}
2433		printf("OOA queues dump done\n");
2434		mtx_unlock(&softc->ctl_lock);
2435		break;
2436	}
2437	case CTL_GET_OOA: {
2438		struct ctl_lun *lun;
2439		struct ctl_ooa *ooa_hdr;
2440		struct ctl_ooa_entry *entries;
2441		uint32_t cur_fill_num;
2442
2443		ooa_hdr = (struct ctl_ooa *)addr;
2444
2445		if ((ooa_hdr->alloc_len == 0)
2446		 || (ooa_hdr->alloc_num == 0)) {
2447			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2448			       "must be non-zero\n", __func__,
2449			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2450			retval = EINVAL;
2451			break;
2452		}
2453
2454		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2455		    sizeof(struct ctl_ooa_entry))) {
2456			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2457			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2458			       __func__, ooa_hdr->alloc_len,
2459			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2460			retval = EINVAL;
2461			break;
2462		}
2463
2464		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2465		if (entries == NULL) {
2466			printf("%s: could not allocate %d bytes for OOA "
2467			       "dump\n", __func__, ooa_hdr->alloc_len);
2468			retval = ENOMEM;
2469			break;
2470		}
2471
2472		mtx_lock(&softc->ctl_lock);
2473		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2474		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2475		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2476			mtx_unlock(&softc->ctl_lock);
2477			free(entries, M_CTL);
2478			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2479			       __func__, (uintmax_t)ooa_hdr->lun_num);
2480			retval = EINVAL;
2481			break;
2482		}
2483
2484		cur_fill_num = 0;
2485
2486		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2487			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2488				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2489					ooa_hdr, entries);
2490				if (retval != 0)
2491					break;
2492			}
2493			if (retval != 0) {
2494				mtx_unlock(&softc->ctl_lock);
2495				free(entries, M_CTL);
2496				break;
2497			}
2498		} else {
2499			lun = softc->ctl_luns[ooa_hdr->lun_num];
2500
2501			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2502						    entries);
2503		}
2504		mtx_unlock(&softc->ctl_lock);
2505
2506		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2507		ooa_hdr->fill_len = ooa_hdr->fill_num *
2508			sizeof(struct ctl_ooa_entry);
2509		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2510		if (retval != 0) {
2511			printf("%s: error copying out %d bytes for OOA dump\n",
2512			       __func__, ooa_hdr->fill_len);
2513		}
2514
2515		getbintime(&ooa_hdr->cur_bt);
2516
2517		if (cur_fill_num > ooa_hdr->alloc_num) {
2518			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2519			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2520		} else {
2521			ooa_hdr->dropped_num = 0;
2522			ooa_hdr->status = CTL_OOA_OK;
2523		}
2524
2525		free(entries, M_CTL);
2526		break;
2527	}
2528	case CTL_CHECK_OOA: {
2529		union ctl_io *io;
2530		struct ctl_lun *lun;
2531		struct ctl_ooa_info *ooa_info;
2532
2533
2534		ooa_info = (struct ctl_ooa_info *)addr;
2535
2536		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2537			ooa_info->status = CTL_OOA_INVALID_LUN;
2538			break;
2539		}
2540		mtx_lock(&softc->ctl_lock);
2541		lun = softc->ctl_luns[ooa_info->lun_id];
2542		if (lun == NULL) {
2543			mtx_unlock(&softc->ctl_lock);
2544			ooa_info->status = CTL_OOA_INVALID_LUN;
2545			break;
2546		}
2547
2548		ooa_info->num_entries = 0;
2549		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2550		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2551		     &io->io_hdr, ooa_links)) {
2552			ooa_info->num_entries++;
2553		}
2554
2555		mtx_unlock(&softc->ctl_lock);
2556		ooa_info->status = CTL_OOA_SUCCESS;
2557
2558		break;
2559	}
2560	case CTL_HARD_START:
2561	case CTL_HARD_STOP: {
2562		struct ctl_fe_ioctl_startstop_info ss_info;
2563		struct cfi_metatask *metatask;
2564		struct mtx hs_mtx;
2565
2566		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2567
2568		cv_init(&ss_info.sem, "hard start/stop cv" );
2569
2570		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2571		if (metatask == NULL) {
2572			retval = ENOMEM;
2573			mtx_destroy(&hs_mtx);
2574			break;
2575		}
2576
2577		if (cmd == CTL_HARD_START)
2578			metatask->tasktype = CFI_TASK_STARTUP;
2579		else
2580			metatask->tasktype = CFI_TASK_SHUTDOWN;
2581
2582		metatask->callback = ctl_ioctl_hard_startstop_callback;
2583		metatask->callback_arg = &ss_info;
2584
2585		cfi_action(metatask);
2586
2587		/* Wait for the callback */
2588		mtx_lock(&hs_mtx);
2589		cv_wait_sig(&ss_info.sem, &hs_mtx);
2590		mtx_unlock(&hs_mtx);
2591
2592		/*
2593		 * All information has been copied from the metatask by the
2594		 * time cv_broadcast() is called, so we free the metatask here.
2595		 */
2596		cfi_free_metatask(metatask);
2597
2598		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2599
2600		mtx_destroy(&hs_mtx);
2601		break;
2602	}
2603	case CTL_BBRREAD: {
2604		struct ctl_bbrread_info *bbr_info;
2605		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2606		struct mtx bbr_mtx;
2607		struct cfi_metatask *metatask;
2608
2609		bbr_info = (struct ctl_bbrread_info *)addr;
2610
2611		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2612
2613		bzero(&bbr_mtx, sizeof(bbr_mtx));
2614		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2615
2616		fe_bbr_info.bbr_info = bbr_info;
2617		fe_bbr_info.lock = &bbr_mtx;
2618
2619		cv_init(&fe_bbr_info.sem, "BBR read cv");
2620		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2621
2622		if (metatask == NULL) {
2623			mtx_destroy(&bbr_mtx);
2624			cv_destroy(&fe_bbr_info.sem);
2625			retval = ENOMEM;
2626			break;
2627		}
2628		metatask->tasktype = CFI_TASK_BBRREAD;
2629		metatask->callback = ctl_ioctl_bbrread_callback;
2630		metatask->callback_arg = &fe_bbr_info;
2631		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2632		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2633		metatask->taskinfo.bbrread.len = bbr_info->len;
2634
2635		cfi_action(metatask);
2636
2637		mtx_lock(&bbr_mtx);
2638		while (fe_bbr_info.wakeup_done == 0)
2639			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2640		mtx_unlock(&bbr_mtx);
2641
2642		bbr_info->status = metatask->status;
2643		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2644		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2645		memcpy(&bbr_info->sense_data,
2646		       &metatask->taskinfo.bbrread.sense_data,
2647		       ctl_min(sizeof(bbr_info->sense_data),
2648			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2649
2650		cfi_free_metatask(metatask);
2651
2652		mtx_destroy(&bbr_mtx);
2653		cv_destroy(&fe_bbr_info.sem);
2654
2655		break;
2656	}
2657	case CTL_DELAY_IO: {
2658		struct ctl_io_delay_info *delay_info;
2659#ifdef CTL_IO_DELAY
2660		struct ctl_lun *lun;
2661#endif /* CTL_IO_DELAY */
2662
2663		delay_info = (struct ctl_io_delay_info *)addr;
2664
2665#ifdef CTL_IO_DELAY
2666		mtx_lock(&softc->ctl_lock);
2667
2668		if ((delay_info->lun_id > CTL_MAX_LUNS)
2669		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2670			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2671		} else {
2672			lun = softc->ctl_luns[delay_info->lun_id];
2673
2674			delay_info->status = CTL_DELAY_STATUS_OK;
2675
2676			switch (delay_info->delay_type) {
2677			case CTL_DELAY_TYPE_CONT:
2678				break;
2679			case CTL_DELAY_TYPE_ONESHOT:
2680				break;
2681			default:
2682				delay_info->status =
2683					CTL_DELAY_STATUS_INVALID_TYPE;
2684				break;
2685			}
2686
2687			switch (delay_info->delay_loc) {
2688			case CTL_DELAY_LOC_DATAMOVE:
2689				lun->delay_info.datamove_type =
2690					delay_info->delay_type;
2691				lun->delay_info.datamove_delay =
2692					delay_info->delay_secs;
2693				break;
2694			case CTL_DELAY_LOC_DONE:
2695				lun->delay_info.done_type =
2696					delay_info->delay_type;
2697				lun->delay_info.done_delay =
2698					delay_info->delay_secs;
2699				break;
2700			default:
2701				delay_info->status =
2702					CTL_DELAY_STATUS_INVALID_LOC;
2703				break;
2704			}
2705		}
2706
2707		mtx_unlock(&softc->ctl_lock);
2708#else
2709		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2710#endif /* CTL_IO_DELAY */
2711		break;
2712	}
2713	case CTL_REALSYNC_SET: {
2714		int *syncstate;
2715
2716		syncstate = (int *)addr;
2717
2718		mtx_lock(&softc->ctl_lock);
2719		switch (*syncstate) {
2720		case 0:
2721			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2722			break;
2723		case 1:
2724			softc->flags |= CTL_FLAG_REAL_SYNC;
2725			break;
2726		default:
2727			retval = -EINVAL;
2728			break;
2729		}
2730		mtx_unlock(&softc->ctl_lock);
2731		break;
2732	}
2733	case CTL_REALSYNC_GET: {
2734		int *syncstate;
2735
2736		syncstate = (int*)addr;
2737
2738		mtx_lock(&softc->ctl_lock);
2739		if (softc->flags & CTL_FLAG_REAL_SYNC)
2740			*syncstate = 1;
2741		else
2742			*syncstate = 0;
2743		mtx_unlock(&softc->ctl_lock);
2744
2745		break;
2746	}
2747	case CTL_SETSYNC:
2748	case CTL_GETSYNC: {
2749		struct ctl_sync_info *sync_info;
2750		struct ctl_lun *lun;
2751
2752		sync_info = (struct ctl_sync_info *)addr;
2753
2754		mtx_lock(&softc->ctl_lock);
2755		lun = softc->ctl_luns[sync_info->lun_id];
2756		if (lun == NULL) {
2757			mtx_unlock(&softc->ctl_lock);
2758			sync_info->status = CTL_GS_SYNC_NO_LUN;
2759		}
2760		/*
2761		 * Get or set the sync interval.  We're not bounds checking
2762		 * in the set case, hopefully the user won't do something
2763		 * silly.
2764		 */
2765		if (cmd == CTL_GETSYNC)
2766			sync_info->sync_interval = lun->sync_interval;
2767		else
2768			lun->sync_interval = sync_info->sync_interval;
2769
2770		mtx_unlock(&softc->ctl_lock);
2771
2772		sync_info->status = CTL_GS_SYNC_OK;
2773
2774		break;
2775	}
2776	case CTL_GETSTATS: {
2777		struct ctl_stats *stats;
2778		struct ctl_lun *lun;
2779		int i;
2780
2781		stats = (struct ctl_stats *)addr;
2782
2783		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2784		     stats->alloc_len) {
2785			stats->status = CTL_SS_NEED_MORE_SPACE;
2786			stats->num_luns = softc->num_luns;
2787			break;
2788		}
2789		/*
2790		 * XXX KDM no locking here.  If the LUN list changes,
2791		 * things can blow up.
2792		 */
2793		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2794		     i++, lun = STAILQ_NEXT(lun, links)) {
2795			retval = copyout(&lun->stats, &stats->lun_stats[i],
2796					 sizeof(lun->stats));
2797			if (retval != 0)
2798				break;
2799		}
2800		stats->num_luns = softc->num_luns;
2801		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2802				 softc->num_luns;
2803		stats->status = CTL_SS_OK;
2804#ifdef CTL_TIME_IO
2805		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2806#else
2807		stats->flags = CTL_STATS_FLAG_NONE;
2808#endif
2809		getnanouptime(&stats->timestamp);
2810		break;
2811	}
2812	case CTL_ERROR_INJECT: {
2813		struct ctl_error_desc *err_desc, *new_err_desc;
2814		struct ctl_lun *lun;
2815
2816		err_desc = (struct ctl_error_desc *)addr;
2817
2818		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2819				      M_WAITOK | M_ZERO);
2820		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2821
2822		mtx_lock(&softc->ctl_lock);
2823		lun = softc->ctl_luns[err_desc->lun_id];
2824		if (lun == NULL) {
2825			mtx_unlock(&softc->ctl_lock);
2826			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2827			       __func__, (uintmax_t)err_desc->lun_id);
2828			retval = EINVAL;
2829			break;
2830		}
2831
2832		/*
2833		 * We could do some checking here to verify the validity
2834		 * of the request, but given the complexity of error
2835		 * injection requests, the checking logic would be fairly
2836		 * complex.
2837		 *
2838		 * For now, if the request is invalid, it just won't get
2839		 * executed and might get deleted.
2840		 */
2841		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2842
2843		/*
2844		 * XXX KDM check to make sure the serial number is unique,
2845		 * in case we somehow manage to wrap.  That shouldn't
2846		 * happen for a very long time, but it's the right thing to
2847		 * do.
2848		 */
2849		new_err_desc->serial = lun->error_serial;
2850		err_desc->serial = lun->error_serial;
2851		lun->error_serial++;
2852
2853		mtx_unlock(&softc->ctl_lock);
2854		break;
2855	}
2856	case CTL_ERROR_INJECT_DELETE: {
2857		struct ctl_error_desc *delete_desc, *desc, *desc2;
2858		struct ctl_lun *lun;
2859		int delete_done;
2860
2861		delete_desc = (struct ctl_error_desc *)addr;
2862		delete_done = 0;
2863
2864		mtx_lock(&softc->ctl_lock);
2865		lun = softc->ctl_luns[delete_desc->lun_id];
2866		if (lun == NULL) {
2867			mtx_unlock(&softc->ctl_lock);
2868			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2869			       __func__, (uintmax_t)delete_desc->lun_id);
2870			retval = EINVAL;
2871			break;
2872		}
2873		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2874			if (desc->serial != delete_desc->serial)
2875				continue;
2876
2877			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2878				      links);
2879			free(desc, M_CTL);
2880			delete_done = 1;
2881		}
2882		mtx_unlock(&softc->ctl_lock);
2883		if (delete_done == 0) {
2884			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2885			       "error serial %ju on LUN %u\n", __func__,
2886			       delete_desc->serial, delete_desc->lun_id);
2887			retval = EINVAL;
2888			break;
2889		}
2890		break;
2891	}
2892	case CTL_DUMP_STRUCTS: {
2893		int i, j, k;
2894		struct ctl_frontend *fe;
2895
2896		printf("CTL IID to WWPN map start:\n");
2897		for (i = 0; i < CTL_MAX_PORTS; i++) {
2898			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2899				if (softc->wwpn_iid[i][j].in_use == 0)
2900					continue;
2901
2902				printf("port %d iid %u WWPN %#jx\n",
2903				       softc->wwpn_iid[i][j].port,
2904				       softc->wwpn_iid[i][j].iid,
2905				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2906			}
2907		}
2908		printf("CTL IID to WWPN map end\n");
2909		printf("CTL Persistent Reservation information start:\n");
2910		for (i = 0; i < CTL_MAX_LUNS; i++) {
2911			struct ctl_lun *lun;
2912
2913			lun = softc->ctl_luns[i];
2914
2915			if ((lun == NULL)
2916			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2917				continue;
2918
2919			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2920				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2921					if (lun->per_res[j+k].registered == 0)
2922						continue;
2923					printf("LUN %d port %d iid %d key "
2924					       "%#jx\n", i, j, k,
2925					       (uintmax_t)scsi_8btou64(
2926					       lun->per_res[j+k].res_key.key));
2927				}
2928			}
2929		}
2930		printf("CTL Persistent Reservation information end\n");
2931		printf("CTL Frontends:\n");
2932		/*
2933		 * XXX KDM calling this without a lock.  We'd likely want
2934		 * to drop the lock before calling the frontend's dump
2935		 * routine anyway.
2936		 */
2937		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2938			printf("Frontend %s Type %u pport %d vport %d WWNN "
2939			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2940			       fe->physical_port, fe->virtual_port,
2941			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2942
2943			/*
2944			 * Frontends are not required to support the dump
2945			 * routine.
2946			 */
2947			if (fe->fe_dump == NULL)
2948				continue;
2949
2950			fe->fe_dump();
2951		}
2952		printf("CTL Frontend information end\n");
2953		break;
2954	}
2955	case CTL_LUN_REQ: {
2956		struct ctl_lun_req *lun_req;
2957		struct ctl_backend_driver *backend;
2958
2959		lun_req = (struct ctl_lun_req *)addr;
2960
2961		backend = ctl_backend_find(lun_req->backend);
2962		if (backend == NULL) {
2963			lun_req->status = CTL_LUN_ERROR;
2964			snprintf(lun_req->error_str,
2965				 sizeof(lun_req->error_str),
2966				 "Backend \"%s\" not found.",
2967				 lun_req->backend);
2968			break;
2969		}
2970		if (lun_req->num_be_args > 0) {
2971			lun_req->kern_be_args = ctl_copyin_args(
2972				lun_req->num_be_args,
2973				lun_req->be_args,
2974				lun_req->error_str,
2975				sizeof(lun_req->error_str));
2976			if (lun_req->kern_be_args == NULL) {
2977				lun_req->status = CTL_LUN_ERROR;
2978				break;
2979			}
2980		}
2981
2982		retval = backend->ioctl(dev, cmd, addr, flag, td);
2983
2984		if (lun_req->num_be_args > 0) {
2985			ctl_free_args(lun_req->num_be_args,
2986				      lun_req->kern_be_args);
2987		}
2988		break;
2989	}
2990	case CTL_LUN_LIST: {
2991		struct sbuf *sb;
2992		struct ctl_lun *lun;
2993		struct ctl_lun_list *list;
2994		struct ctl_be_lun_option *opt;
2995
2996		list = (struct ctl_lun_list *)addr;
2997
2998		/*
2999		 * Allocate a fixed length sbuf here, based on the length
3000		 * of the user's buffer.  We could allocate an auto-extending
3001		 * buffer, and then tell the user how much larger our
3002		 * amount of data is than his buffer, but that presents
3003		 * some problems:
3004		 *
3005		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3006		 *     we can't hold a lock while calling them with an
3007		 *     auto-extending buffer.
3008 		 *
3009		 * 2.  There is not currently a LUN reference counting
3010		 *     mechanism, outside of outstanding transactions on
3011		 *     the LUN's OOA queue.  So a LUN could go away on us
3012		 *     while we're getting the LUN number, backend-specific
3013		 *     information, etc.  Thus, given the way things
3014		 *     currently work, we need to hold the CTL lock while
3015		 *     grabbing LUN information.
3016		 *
3017		 * So, from the user's standpoint, the best thing to do is
3018		 * allocate what he thinks is a reasonable buffer length,
3019		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3020		 * double the buffer length and try again.  (And repeat
3021		 * that until he succeeds.)
3022		 */
3023		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3024		if (sb == NULL) {
3025			list->status = CTL_LUN_LIST_ERROR;
3026			snprintf(list->error_str, sizeof(list->error_str),
3027				 "Unable to allocate %d bytes for LUN list",
3028				 list->alloc_len);
3029			break;
3030		}
3031
3032		sbuf_printf(sb, "<ctllunlist>\n");
3033
3034		mtx_lock(&softc->ctl_lock);
3035
3036		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3037			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3038					     (uintmax_t)lun->lun);
3039
3040			/*
3041			 * Bail out as soon as we see that we've overfilled
3042			 * the buffer.
3043			 */
3044			if (retval != 0)
3045				break;
3046
3047			retval = sbuf_printf(sb, "<backend_type>%s"
3048					     "</backend_type>\n",
3049					     (lun->backend == NULL) ?  "none" :
3050					     lun->backend->name);
3051
3052			if (retval != 0)
3053				break;
3054
3055			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3056					     lun->be_lun->lun_type);
3057
3058			if (retval != 0)
3059				break;
3060
3061			if (lun->backend == NULL) {
3062				retval = sbuf_printf(sb, "</lun>\n");
3063				if (retval != 0)
3064					break;
3065				continue;
3066			}
3067
3068			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3069					     (lun->be_lun->maxlba > 0) ?
3070					     lun->be_lun->maxlba + 1 : 0);
3071
3072			if (retval != 0)
3073				break;
3074
3075			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3076					     lun->be_lun->blocksize);
3077
3078			if (retval != 0)
3079				break;
3080
3081			retval = sbuf_printf(sb, "<serial_number>");
3082
3083			if (retval != 0)
3084				break;
3085
3086			retval = ctl_sbuf_printf_esc(sb,
3087						     lun->be_lun->serial_num);
3088
3089			if (retval != 0)
3090				break;
3091
3092			retval = sbuf_printf(sb, "</serial_number>\n");
3093
3094			if (retval != 0)
3095				break;
3096
3097			retval = sbuf_printf(sb, "<device_id>");
3098
3099			if (retval != 0)
3100				break;
3101
3102			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3103
3104			if (retval != 0)
3105				break;
3106
3107			retval = sbuf_printf(sb, "</device_id>\n");
3108
3109			if (retval != 0)
3110				break;
3111
3112			if (lun->backend->lun_info != NULL) {
3113				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3114				if (retval != 0)
3115					break;
3116			}
3117			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3118				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3119				if (retval != 0)
3120					break;
3121			}
3122
3123			retval = sbuf_printf(sb, "</lun>\n");
3124
3125			if (retval != 0)
3126				break;
3127		}
3128		mtx_unlock(&softc->ctl_lock);
3129
3130		if ((retval != 0)
3131		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3132			retval = 0;
3133			sbuf_delete(sb);
3134			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3135			snprintf(list->error_str, sizeof(list->error_str),
3136				 "Out of space, %d bytes is too small",
3137				 list->alloc_len);
3138			break;
3139		}
3140
3141		sbuf_finish(sb);
3142
3143		retval = copyout(sbuf_data(sb), list->lun_xml,
3144				 sbuf_len(sb) + 1);
3145
3146		list->fill_len = sbuf_len(sb) + 1;
3147		list->status = CTL_LUN_LIST_OK;
3148		sbuf_delete(sb);
3149		break;
3150	}
3151	case CTL_ISCSI: {
3152		struct ctl_iscsi *ci;
3153		struct ctl_frontend *fe;
3154
3155		ci = (struct ctl_iscsi *)addr;
3156
3157		mtx_lock(&softc->ctl_lock);
3158		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3159			if (strcmp(fe->port_name, "iscsi") == 0)
3160				break;
3161		}
3162		mtx_unlock(&softc->ctl_lock);
3163
3164		if (fe == NULL) {
3165			ci->status = CTL_ISCSI_ERROR;
3166			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3167			break;
3168		}
3169
3170		retval = fe->ioctl(dev, cmd, addr, flag, td);
3171		break;
3172	}
3173	default: {
3174		/* XXX KDM should we fix this? */
3175#if 0
3176		struct ctl_backend_driver *backend;
3177		unsigned int type;
3178		int found;
3179
3180		found = 0;
3181
3182		/*
3183		 * We encode the backend type as the ioctl type for backend
3184		 * ioctls.  So parse it out here, and then search for a
3185		 * backend of this type.
3186		 */
3187		type = _IOC_TYPE(cmd);
3188
3189		STAILQ_FOREACH(backend, &softc->be_list, links) {
3190			if (backend->type == type) {
3191				found = 1;
3192				break;
3193			}
3194		}
3195		if (found == 0) {
3196			printf("ctl: unknown ioctl command %#lx or backend "
3197			       "%d\n", cmd, type);
3198			retval = -EINVAL;
3199			break;
3200		}
3201		retval = backend->ioctl(dev, cmd, addr, flag, td);
3202#endif
3203		retval = ENOTTY;
3204		break;
3205	}
3206	}
3207	return (retval);
3208}
3209
3210uint32_t
3211ctl_get_initindex(struct ctl_nexus *nexus)
3212{
3213	if (nexus->targ_port < CTL_MAX_PORTS)
3214		return (nexus->initid.id +
3215			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3216	else
3217		return (nexus->initid.id +
3218		       ((nexus->targ_port - CTL_MAX_PORTS) *
3219			CTL_MAX_INIT_PER_PORT));
3220}
3221
3222uint32_t
3223ctl_get_resindex(struct ctl_nexus *nexus)
3224{
3225	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3226}
3227
3228uint32_t
3229ctl_port_idx(int port_num)
3230{
3231	if (port_num < CTL_MAX_PORTS)
3232		return(port_num);
3233	else
3234		return(port_num - CTL_MAX_PORTS);
3235}
3236
3237/*
3238 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3239 * that are a power of 2.
3240 */
3241int
3242ctl_ffz(uint32_t *mask, uint32_t size)
3243{
3244	uint32_t num_chunks, num_pieces;
3245	int i, j;
3246
3247	num_chunks = (size >> 5);
3248	if (num_chunks == 0)
3249		num_chunks++;
3250	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3251
3252	for (i = 0; i < num_chunks; i++) {
3253		for (j = 0; j < num_pieces; j++) {
3254			if ((mask[i] & (1 << j)) == 0)
3255				return ((i << 5) + j);
3256		}
3257	}
3258
3259	return (-1);
3260}
3261
3262int
3263ctl_set_mask(uint32_t *mask, uint32_t bit)
3264{
3265	uint32_t chunk, piece;
3266
3267	chunk = bit >> 5;
3268	piece = bit % (sizeof(uint32_t) * 8);
3269
3270	if ((mask[chunk] & (1 << piece)) != 0)
3271		return (-1);
3272	else
3273		mask[chunk] |= (1 << piece);
3274
3275	return (0);
3276}
3277
3278int
3279ctl_clear_mask(uint32_t *mask, uint32_t bit)
3280{
3281	uint32_t chunk, piece;
3282
3283	chunk = bit >> 5;
3284	piece = bit % (sizeof(uint32_t) * 8);
3285
3286	if ((mask[chunk] & (1 << piece)) == 0)
3287		return (-1);
3288	else
3289		mask[chunk] &= ~(1 << piece);
3290
3291	return (0);
3292}
3293
3294int
3295ctl_is_set(uint32_t *mask, uint32_t bit)
3296{
3297	uint32_t chunk, piece;
3298
3299	chunk = bit >> 5;
3300	piece = bit % (sizeof(uint32_t) * 8);
3301
3302	if ((mask[chunk] & (1 << piece)) == 0)
3303		return (0);
3304	else
3305		return (1);
3306}
3307
3308#ifdef unused
3309/*
3310 * The bus, target and lun are optional, they can be filled in later.
3311 * can_wait is used to determine whether we can wait on the malloc or not.
3312 */
3313union ctl_io*
3314ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3315	      uint32_t targ_lun, int can_wait)
3316{
3317	union ctl_io *io;
3318
3319	if (can_wait)
3320		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3321	else
3322		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3323
3324	if (io != NULL) {
3325		io->io_hdr.io_type = io_type;
3326		io->io_hdr.targ_port = targ_port;
3327		/*
3328		 * XXX KDM this needs to change/go away.  We need to move
3329		 * to a preallocated pool of ctl_scsiio structures.
3330		 */
3331		io->io_hdr.nexus.targ_target.id = targ_target;
3332		io->io_hdr.nexus.targ_lun = targ_lun;
3333	}
3334
3335	return (io);
3336}
3337
3338void
3339ctl_kfree_io(union ctl_io *io)
3340{
3341	free(io, M_CTL);
3342}
3343#endif /* unused */
3344
3345/*
3346 * ctl_softc, pool_type, total_ctl_io are passed in.
3347 * npool is passed out.
3348 */
3349int
3350ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3351		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3352{
3353	uint32_t i;
3354	union ctl_io *cur_io, *next_io;
3355	struct ctl_io_pool *pool;
3356	int retval;
3357
3358	retval = 0;
3359
3360	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3361					    M_NOWAIT | M_ZERO);
3362	if (pool == NULL) {
3363		retval = -ENOMEM;
3364		goto bailout;
3365	}
3366
3367	pool->type = pool_type;
3368	pool->ctl_softc = ctl_softc;
3369
3370	mtx_lock(&ctl_softc->ctl_lock);
3371	pool->id = ctl_softc->cur_pool_id++;
3372	mtx_unlock(&ctl_softc->ctl_lock);
3373
3374	pool->flags = CTL_POOL_FLAG_NONE;
3375	STAILQ_INIT(&pool->free_queue);
3376
3377	/*
3378	 * XXX KDM other options here:
3379	 * - allocate a page at a time
3380	 * - allocate one big chunk of memory.
3381	 * Page allocation might work well, but would take a little more
3382	 * tracking.
3383	 */
3384	for (i = 0; i < total_ctl_io; i++) {
3385		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3386						M_NOWAIT);
3387		if (cur_io == NULL) {
3388			retval = ENOMEM;
3389			break;
3390		}
3391		cur_io->io_hdr.pool = pool;
3392		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3393		pool->total_ctl_io++;
3394		pool->free_ctl_io++;
3395	}
3396
3397	if (retval != 0) {
3398		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3399		     cur_io != NULL; cur_io = next_io) {
3400			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3401							      links);
3402			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3403				      ctl_io_hdr, links);
3404			free(cur_io, M_CTL);
3405		}
3406
3407		free(pool, M_CTL);
3408		goto bailout;
3409	}
3410	mtx_lock(&ctl_softc->ctl_lock);
3411	ctl_softc->num_pools++;
3412	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3413	/*
3414	 * Increment our usage count if this is an external consumer, so we
3415	 * can't get unloaded until the external consumer (most likely a
3416	 * FETD) unloads and frees his pool.
3417	 *
3418	 * XXX KDM will this increment the caller's module use count, or
3419	 * mine?
3420	 */
3421#if 0
3422	if ((pool_type != CTL_POOL_EMERGENCY)
3423	 && (pool_type != CTL_POOL_INTERNAL)
3424	 && (pool_type != CTL_POOL_IOCTL)
3425	 && (pool_type != CTL_POOL_4OTHERSC))
3426		MOD_INC_USE_COUNT;
3427#endif
3428
3429	mtx_unlock(&ctl_softc->ctl_lock);
3430
3431	*npool = pool;
3432
3433bailout:
3434
3435	return (retval);
3436}
3437
3438int
3439ctl_pool_acquire(struct ctl_io_pool *pool)
3440{
3441
3442	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
3443
3444	if (pool == NULL)
3445		return (-EINVAL);
3446
3447	if (pool->flags & CTL_POOL_FLAG_INVALID)
3448		return (-EINVAL);
3449
3450	pool->refcount++;
3451
3452	return (0);
3453}
3454
3455int
3456ctl_pool_invalidate(struct ctl_io_pool *pool)
3457{
3458
3459	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
3460
3461	if (pool == NULL)
3462		return (-EINVAL);
3463
3464	pool->flags |= CTL_POOL_FLAG_INVALID;
3465
3466	return (0);
3467}
3468
3469int
3470ctl_pool_release(struct ctl_io_pool *pool)
3471{
3472
3473	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
3474
3475	if (pool == NULL)
3476		return (-EINVAL);
3477
3478	if ((--pool->refcount == 0)
3479	 && (pool->flags & CTL_POOL_FLAG_INVALID)) {
3480		ctl_pool_free(pool->ctl_softc, pool);
3481	}
3482
3483	return (0);
3484}
3485
3486void
3487ctl_pool_free(struct ctl_softc *ctl_softc, struct ctl_io_pool *pool)
3488{
3489	union ctl_io *cur_io, *next_io;
3490
3491	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
3492
3493	for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3494	     cur_io != NULL; cur_io = next_io) {
3495		next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3496						      links);
3497		STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr, ctl_io_hdr,
3498			      links);
3499		free(cur_io, M_CTL);
3500	}
3501
3502	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3503	ctl_softc->num_pools--;
3504
3505	/*
3506	 * XXX KDM will this decrement the caller's usage count or mine?
3507	 */
3508#if 0
3509	if ((pool->type != CTL_POOL_EMERGENCY)
3510	 && (pool->type != CTL_POOL_INTERNAL)
3511	 && (pool->type != CTL_POOL_IOCTL))
3512		MOD_DEC_USE_COUNT;
3513#endif
3514
3515	free(pool, M_CTL);
3516}
3517
3518/*
3519 * This routine does not block (except for spinlocks of course).
3520 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3521 * possible.
3522 */
3523union ctl_io *
3524ctl_alloc_io(void *pool_ref)
3525{
3526	union ctl_io *io;
3527	struct ctl_softc *ctl_softc;
3528	struct ctl_io_pool *pool, *npool;
3529	struct ctl_io_pool *emergency_pool;
3530
3531	pool = (struct ctl_io_pool *)pool_ref;
3532
3533	if (pool == NULL) {
3534		printf("%s: pool is NULL\n", __func__);
3535		return (NULL);
3536	}
3537
3538	emergency_pool = NULL;
3539
3540	ctl_softc = pool->ctl_softc;
3541
3542	mtx_lock(&ctl_softc->ctl_lock);
3543	/*
3544	 * First, try to get the io structure from the user's pool.
3545	 */
3546	if (ctl_pool_acquire(pool) == 0) {
3547		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3548		if (io != NULL) {
3549			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3550			pool->total_allocated++;
3551			pool->free_ctl_io--;
3552			mtx_unlock(&ctl_softc->ctl_lock);
3553			return (io);
3554		} else
3555			ctl_pool_release(pool);
3556	}
3557	/*
3558	 * If he doesn't have any io structures left, search for an
3559	 * emergency pool and grab one from there.
3560	 */
3561	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3562		if (npool->type != CTL_POOL_EMERGENCY)
3563			continue;
3564
3565		if (ctl_pool_acquire(npool) != 0)
3566			continue;
3567
3568		emergency_pool = npool;
3569
3570		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3571		if (io != NULL) {
3572			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3573			npool->total_allocated++;
3574			npool->free_ctl_io--;
3575			mtx_unlock(&ctl_softc->ctl_lock);
3576			return (io);
3577		} else
3578			ctl_pool_release(npool);
3579	}
3580
3581	/* Drop the spinlock before we malloc */
3582	mtx_unlock(&ctl_softc->ctl_lock);
3583
3584	/*
3585	 * The emergency pool (if it exists) didn't have one, so try an
3586	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3587	 */
3588	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3589	if (io != NULL) {
3590		/*
3591		 * If the emergency pool exists but is empty, add this
3592		 * ctl_io to its list when it gets freed.
3593		 */
3594		if (emergency_pool != NULL) {
3595			mtx_lock(&ctl_softc->ctl_lock);
3596			if (ctl_pool_acquire(emergency_pool) == 0) {
3597				io->io_hdr.pool = emergency_pool;
3598				emergency_pool->total_ctl_io++;
3599				/*
3600				 * Need to bump this, otherwise
3601				 * total_allocated and total_freed won't
3602				 * match when we no longer have anything
3603				 * outstanding.
3604				 */
3605				emergency_pool->total_allocated++;
3606			}
3607			mtx_unlock(&ctl_softc->ctl_lock);
3608		} else
3609			io->io_hdr.pool = NULL;
3610	}
3611
3612	return (io);
3613}
3614
3615static void
3616ctl_free_io_internal(union ctl_io *io, int have_lock)
3617{
3618	if (io == NULL)
3619		return;
3620
3621	/*
3622	 * If this ctl_io has a pool, return it to that pool.
3623	 */
3624	if (io->io_hdr.pool != NULL) {
3625		struct ctl_io_pool *pool;
3626#if 0
3627		struct ctl_softc *ctl_softc;
3628		union ctl_io *tmp_io;
3629		unsigned long xflags;
3630		int i;
3631
3632		ctl_softc = control_softc;
3633#endif
3634
3635		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3636
3637		if (have_lock == 0)
3638			mtx_lock(&pool->ctl_softc->ctl_lock);
3639#if 0
3640		save_flags(xflags);
3641
3642		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3643		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3644		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3645		     links)) {
3646			if (tmp_io == io) {
3647				printf("%s: %p is still on the task queue!\n",
3648				       __func__, tmp_io);
3649				printf("%s: (%d): type %d "
3650				       "msg %d cdb %x iptl: "
3651				       "%d:%d:%d:%d tag 0x%04x "
3652				       "flg %#lx\n",
3653					__func__, i,
3654					tmp_io->io_hdr.io_type,
3655					tmp_io->io_hdr.msg_type,
3656					tmp_io->scsiio.cdb[0],
3657					tmp_io->io_hdr.nexus.initid.id,
3658					tmp_io->io_hdr.nexus.targ_port,
3659					tmp_io->io_hdr.nexus.targ_target.id,
3660					tmp_io->io_hdr.nexus.targ_lun,
3661					(tmp_io->io_hdr.io_type ==
3662					CTL_IO_TASK) ?
3663					tmp_io->taskio.tag_num :
3664					tmp_io->scsiio.tag_num,
3665					xflags);
3666				panic("I/O still on the task queue!");
3667			}
3668		}
3669#endif
3670		io->io_hdr.io_type = 0xff;
3671		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3672		pool->total_freed++;
3673		pool->free_ctl_io++;
3674		ctl_pool_release(pool);
3675		if (have_lock == 0)
3676			mtx_unlock(&pool->ctl_softc->ctl_lock);
3677	} else {
3678		/*
3679		 * Otherwise, just free it.  We probably malloced it and
3680		 * the emergency pool wasn't available.
3681		 */
3682		free(io, M_CTL);
3683	}
3684
3685}
3686
3687void
3688ctl_free_io(union ctl_io *io)
3689{
3690	ctl_free_io_internal(io, /*have_lock*/ 0);
3691}
3692
3693void
3694ctl_zero_io(union ctl_io *io)
3695{
3696	void *pool_ref;
3697
3698	if (io == NULL)
3699		return;
3700
3701	/*
3702	 * May need to preserve linked list pointers at some point too.
3703	 */
3704	pool_ref = io->io_hdr.pool;
3705
3706	memset(io, 0, sizeof(*io));
3707
3708	io->io_hdr.pool = pool_ref;
3709}
3710
3711/*
3712 * This routine is currently used for internal copies of ctl_ios that need
3713 * to persist for some reason after we've already returned status to the
3714 * FETD.  (Thus the flag set.)
3715 *
3716 * XXX XXX
3717 * Note that this makes a blind copy of all fields in the ctl_io, except
3718 * for the pool reference.  This includes any memory that has been
3719 * allocated!  That memory will no longer be valid after done has been
3720 * called, so this would be VERY DANGEROUS for command that actually does
3721 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3722 * start and stop commands, which don't transfer any data, so this is not a
3723 * problem.  If it is used for anything else, the caller would also need to
3724 * allocate data buffer space and this routine would need to be modified to
3725 * copy the data buffer(s) as well.
3726 */
3727void
3728ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3729{
3730	void *pool_ref;
3731
3732	if ((src == NULL)
3733	 || (dest == NULL))
3734		return;
3735
3736	/*
3737	 * May need to preserve linked list pointers at some point too.
3738	 */
3739	pool_ref = dest->io_hdr.pool;
3740
3741	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3742
3743	dest->io_hdr.pool = pool_ref;
3744	/*
3745	 * We need to know that this is an internal copy, and doesn't need
3746	 * to get passed back to the FETD that allocated it.
3747	 */
3748	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3749}
3750
3751#ifdef NEEDTOPORT
3752static void
3753ctl_update_power_subpage(struct copan_power_subpage *page)
3754{
3755	int num_luns, num_partitions, config_type;
3756	struct ctl_softc *softc;
3757	cs_BOOL_t aor_present, shelf_50pct_power;
3758	cs_raidset_personality_t rs_type;
3759	int max_active_luns;
3760
3761	softc = control_softc;
3762
3763	/* subtract out the processor LUN */
3764	num_luns = softc->num_luns - 1;
3765	/*
3766	 * Default to 7 LUNs active, which was the only number we allowed
3767	 * in the past.
3768	 */
3769	max_active_luns = 7;
3770
3771	num_partitions = config_GetRsPartitionInfo();
3772	config_type = config_GetConfigType();
3773	shelf_50pct_power = config_GetShelfPowerMode();
3774	aor_present = config_IsAorRsPresent();
3775
3776	rs_type = ddb_GetRsRaidType(1);
3777	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3778	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3779		EPRINT(0, "Unsupported RS type %d!", rs_type);
3780	}
3781
3782
3783	page->total_luns = num_luns;
3784
3785	switch (config_type) {
3786	case 40:
3787		/*
3788		 * In a 40 drive configuration, it doesn't matter what DC
3789		 * cards we have, whether we have AOR enabled or not,
3790		 * partitioning or not, or what type of RAIDset we have.
3791		 * In that scenario, we can power up every LUN we present
3792		 * to the user.
3793		 */
3794		max_active_luns = num_luns;
3795
3796		break;
3797	case 64:
3798		if (shelf_50pct_power == CS_FALSE) {
3799			/* 25% power */
3800			if (aor_present == CS_TRUE) {
3801				if (rs_type ==
3802				     CS_RAIDSET_PERSONALITY_RAID5) {
3803					max_active_luns = 7;
3804				} else if (rs_type ==
3805					 CS_RAIDSET_PERSONALITY_RAID1){
3806					max_active_luns = 14;
3807				} else {
3808					/* XXX KDM now what?? */
3809				}
3810			} else {
3811				if (rs_type ==
3812				     CS_RAIDSET_PERSONALITY_RAID5) {
3813					max_active_luns = 8;
3814				} else if (rs_type ==
3815					 CS_RAIDSET_PERSONALITY_RAID1){
3816					max_active_luns = 16;
3817				} else {
3818					/* XXX KDM now what?? */
3819				}
3820			}
3821		} else {
3822			/* 50% power */
3823			/*
3824			 * With 50% power in a 64 drive configuration, we
3825			 * can power all LUNs we present.
3826			 */
3827			max_active_luns = num_luns;
3828		}
3829		break;
3830	case 112:
3831		if (shelf_50pct_power == CS_FALSE) {
3832			/* 25% power */
3833			if (aor_present == CS_TRUE) {
3834				if (rs_type ==
3835				     CS_RAIDSET_PERSONALITY_RAID5) {
3836					max_active_luns = 7;
3837				} else if (rs_type ==
3838					 CS_RAIDSET_PERSONALITY_RAID1){
3839					max_active_luns = 14;
3840				} else {
3841					/* XXX KDM now what?? */
3842				}
3843			} else {
3844				if (rs_type ==
3845				     CS_RAIDSET_PERSONALITY_RAID5) {
3846					max_active_luns = 8;
3847				} else if (rs_type ==
3848					 CS_RAIDSET_PERSONALITY_RAID1){
3849					max_active_luns = 16;
3850				} else {
3851					/* XXX KDM now what?? */
3852				}
3853			}
3854		} else {
3855			/* 50% power */
3856			if (aor_present == CS_TRUE) {
3857				if (rs_type ==
3858				     CS_RAIDSET_PERSONALITY_RAID5) {
3859					max_active_luns = 14;
3860				} else if (rs_type ==
3861					 CS_RAIDSET_PERSONALITY_RAID1){
3862					/*
3863					 * We're assuming here that disk
3864					 * caching is enabled, and so we're
3865					 * able to power up half of each
3866					 * LUN, and cache all writes.
3867					 */
3868					max_active_luns = num_luns;
3869				} else {
3870					/* XXX KDM now what?? */
3871				}
3872			} else {
3873				if (rs_type ==
3874				     CS_RAIDSET_PERSONALITY_RAID5) {
3875					max_active_luns = 15;
3876				} else if (rs_type ==
3877					 CS_RAIDSET_PERSONALITY_RAID1){
3878					max_active_luns = 30;
3879				} else {
3880					/* XXX KDM now what?? */
3881				}
3882			}
3883		}
3884		break;
3885	default:
3886		/*
3887		 * In this case, we have an unknown configuration, so we
3888		 * just use the default from above.
3889		 */
3890		break;
3891	}
3892
3893	page->max_active_luns = max_active_luns;
3894#if 0
3895	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3896	       page->total_luns, page->max_active_luns);
3897#endif
3898}
3899#endif /* NEEDTOPORT */
3900
3901/*
3902 * This routine could be used in the future to load default and/or saved
3903 * mode page parameters for a particuar lun.
3904 */
3905static int
3906ctl_init_page_index(struct ctl_lun *lun)
3907{
3908	int i;
3909	struct ctl_page_index *page_index;
3910	struct ctl_softc *softc;
3911
3912	memcpy(&lun->mode_pages.index, page_index_template,
3913	       sizeof(page_index_template));
3914
3915	softc = lun->ctl_softc;
3916
3917	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3918
3919		page_index = &lun->mode_pages.index[i];
3920		/*
3921		 * If this is a disk-only mode page, there's no point in
3922		 * setting it up.  For some pages, we have to have some
3923		 * basic information about the disk in order to calculate the
3924		 * mode page data.
3925		 */
3926		if ((lun->be_lun->lun_type != T_DIRECT)
3927		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3928			continue;
3929
3930		switch (page_index->page_code & SMPH_PC_MASK) {
3931		case SMS_FORMAT_DEVICE_PAGE: {
3932			struct scsi_format_page *format_page;
3933
3934			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3935				panic("subpage is incorrect!");
3936
3937			/*
3938			 * Sectors per track are set above.  Bytes per
3939			 * sector need to be set here on a per-LUN basis.
3940			 */
3941			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3942			       &format_page_default,
3943			       sizeof(format_page_default));
3944			memcpy(&lun->mode_pages.format_page[
3945			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3946			       sizeof(format_page_changeable));
3947			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3948			       &format_page_default,
3949			       sizeof(format_page_default));
3950			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3951			       &format_page_default,
3952			       sizeof(format_page_default));
3953
3954			format_page = &lun->mode_pages.format_page[
3955				CTL_PAGE_CURRENT];
3956			scsi_ulto2b(lun->be_lun->blocksize,
3957				    format_page->bytes_per_sector);
3958
3959			format_page = &lun->mode_pages.format_page[
3960				CTL_PAGE_DEFAULT];
3961			scsi_ulto2b(lun->be_lun->blocksize,
3962				    format_page->bytes_per_sector);
3963
3964			format_page = &lun->mode_pages.format_page[
3965				CTL_PAGE_SAVED];
3966			scsi_ulto2b(lun->be_lun->blocksize,
3967				    format_page->bytes_per_sector);
3968
3969			page_index->page_data =
3970				(uint8_t *)lun->mode_pages.format_page;
3971			break;
3972		}
3973		case SMS_RIGID_DISK_PAGE: {
3974			struct scsi_rigid_disk_page *rigid_disk_page;
3975			uint32_t sectors_per_cylinder;
3976			uint64_t cylinders;
3977#ifndef	__XSCALE__
3978			int shift;
3979#endif /* !__XSCALE__ */
3980
3981			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3982				panic("invalid subpage value %d",
3983				      page_index->subpage);
3984
3985			/*
3986			 * Rotation rate and sectors per track are set
3987			 * above.  We calculate the cylinders here based on
3988			 * capacity.  Due to the number of heads and
3989			 * sectors per track we're using, smaller arrays
3990			 * may turn out to have 0 cylinders.  Linux and
3991			 * FreeBSD don't pay attention to these mode pages
3992			 * to figure out capacity, but Solaris does.  It
3993			 * seems to deal with 0 cylinders just fine, and
3994			 * works out a fake geometry based on the capacity.
3995			 */
3996			memcpy(&lun->mode_pages.rigid_disk_page[
3997			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3998			       sizeof(rigid_disk_page_default));
3999			memcpy(&lun->mode_pages.rigid_disk_page[
4000			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4001			       sizeof(rigid_disk_page_changeable));
4002			memcpy(&lun->mode_pages.rigid_disk_page[
4003			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4004			       sizeof(rigid_disk_page_default));
4005			memcpy(&lun->mode_pages.rigid_disk_page[
4006			       CTL_PAGE_SAVED], &rigid_disk_page_default,
4007			       sizeof(rigid_disk_page_default));
4008
4009			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4010				CTL_DEFAULT_HEADS;
4011
4012			/*
4013			 * The divide method here will be more accurate,
4014			 * probably, but results in floating point being
4015			 * used in the kernel on i386 (__udivdi3()).  On the
4016			 * XScale, though, __udivdi3() is implemented in
4017			 * software.
4018			 *
4019			 * The shift method for cylinder calculation is
4020			 * accurate if sectors_per_cylinder is a power of
4021			 * 2.  Otherwise it might be slightly off -- you
4022			 * might have a bit of a truncation problem.
4023			 */
4024#ifdef	__XSCALE__
4025			cylinders = (lun->be_lun->maxlba + 1) /
4026				sectors_per_cylinder;
4027#else
4028			for (shift = 31; shift > 0; shift--) {
4029				if (sectors_per_cylinder & (1 << shift))
4030					break;
4031			}
4032			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4033#endif
4034
4035			/*
4036			 * We've basically got 3 bytes, or 24 bits for the
4037			 * cylinder size in the mode page.  If we're over,
4038			 * just round down to 2^24.
4039			 */
4040			if (cylinders > 0xffffff)
4041				cylinders = 0xffffff;
4042
4043			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4044				CTL_PAGE_CURRENT];
4045			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4046
4047			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4048				CTL_PAGE_DEFAULT];
4049			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4050
4051			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4052				CTL_PAGE_SAVED];
4053			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4054
4055			page_index->page_data =
4056				(uint8_t *)lun->mode_pages.rigid_disk_page;
4057			break;
4058		}
4059		case SMS_CACHING_PAGE: {
4060
4061			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4062				panic("invalid subpage value %d",
4063				      page_index->subpage);
4064			/*
4065			 * Defaults should be okay here, no calculations
4066			 * needed.
4067			 */
4068			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4069			       &caching_page_default,
4070			       sizeof(caching_page_default));
4071			memcpy(&lun->mode_pages.caching_page[
4072			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4073			       sizeof(caching_page_changeable));
4074			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4075			       &caching_page_default,
4076			       sizeof(caching_page_default));
4077			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4078			       &caching_page_default,
4079			       sizeof(caching_page_default));
4080			page_index->page_data =
4081				(uint8_t *)lun->mode_pages.caching_page;
4082			break;
4083		}
4084		case SMS_CONTROL_MODE_PAGE: {
4085
4086			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4087				panic("invalid subpage value %d",
4088				      page_index->subpage);
4089
4090			/*
4091			 * Defaults should be okay here, no calculations
4092			 * needed.
4093			 */
4094			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4095			       &control_page_default,
4096			       sizeof(control_page_default));
4097			memcpy(&lun->mode_pages.control_page[
4098			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4099			       sizeof(control_page_changeable));
4100			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4101			       &control_page_default,
4102			       sizeof(control_page_default));
4103			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4104			       &control_page_default,
4105			       sizeof(control_page_default));
4106			page_index->page_data =
4107				(uint8_t *)lun->mode_pages.control_page;
4108			break;
4109
4110		}
4111		case SMS_VENDOR_SPECIFIC_PAGE:{
4112			switch (page_index->subpage) {
4113			case PWR_SUBPAGE_CODE: {
4114				struct copan_power_subpage *current_page,
4115							   *saved_page;
4116
4117				memcpy(&lun->mode_pages.power_subpage[
4118				       CTL_PAGE_CURRENT],
4119				       &power_page_default,
4120				       sizeof(power_page_default));
4121				memcpy(&lun->mode_pages.power_subpage[
4122				       CTL_PAGE_CHANGEABLE],
4123				       &power_page_changeable,
4124				       sizeof(power_page_changeable));
4125				memcpy(&lun->mode_pages.power_subpage[
4126				       CTL_PAGE_DEFAULT],
4127				       &power_page_default,
4128				       sizeof(power_page_default));
4129				memcpy(&lun->mode_pages.power_subpage[
4130				       CTL_PAGE_SAVED],
4131				       &power_page_default,
4132				       sizeof(power_page_default));
4133				page_index->page_data =
4134				    (uint8_t *)lun->mode_pages.power_subpage;
4135
4136				current_page = (struct copan_power_subpage *)
4137					(page_index->page_data +
4138					 (page_index->page_len *
4139					  CTL_PAGE_CURRENT));
4140			        saved_page = (struct copan_power_subpage *)
4141				        (page_index->page_data +
4142					 (page_index->page_len *
4143					  CTL_PAGE_SAVED));
4144				break;
4145			}
4146			case APS_SUBPAGE_CODE: {
4147				struct copan_aps_subpage *current_page,
4148							 *saved_page;
4149
4150				// This gets set multiple times but
4151				// it should always be the same. It's
4152				// only done during init so who cares.
4153				index_to_aps_page = i;
4154
4155				memcpy(&lun->mode_pages.aps_subpage[
4156				       CTL_PAGE_CURRENT],
4157				       &aps_page_default,
4158				       sizeof(aps_page_default));
4159				memcpy(&lun->mode_pages.aps_subpage[
4160				       CTL_PAGE_CHANGEABLE],
4161				       &aps_page_changeable,
4162				       sizeof(aps_page_changeable));
4163				memcpy(&lun->mode_pages.aps_subpage[
4164				       CTL_PAGE_DEFAULT],
4165				       &aps_page_default,
4166				       sizeof(aps_page_default));
4167				memcpy(&lun->mode_pages.aps_subpage[
4168				       CTL_PAGE_SAVED],
4169				       &aps_page_default,
4170				       sizeof(aps_page_default));
4171				page_index->page_data =
4172					(uint8_t *)lun->mode_pages.aps_subpage;
4173
4174				current_page = (struct copan_aps_subpage *)
4175					(page_index->page_data +
4176					 (page_index->page_len *
4177					  CTL_PAGE_CURRENT));
4178				saved_page = (struct copan_aps_subpage *)
4179					(page_index->page_data +
4180					 (page_index->page_len *
4181					  CTL_PAGE_SAVED));
4182				break;
4183			}
4184			case DBGCNF_SUBPAGE_CODE: {
4185				struct copan_debugconf_subpage *current_page,
4186							       *saved_page;
4187
4188				memcpy(&lun->mode_pages.debugconf_subpage[
4189				       CTL_PAGE_CURRENT],
4190				       &debugconf_page_default,
4191				       sizeof(debugconf_page_default));
4192				memcpy(&lun->mode_pages.debugconf_subpage[
4193				       CTL_PAGE_CHANGEABLE],
4194				       &debugconf_page_changeable,
4195				       sizeof(debugconf_page_changeable));
4196				memcpy(&lun->mode_pages.debugconf_subpage[
4197				       CTL_PAGE_DEFAULT],
4198				       &debugconf_page_default,
4199				       sizeof(debugconf_page_default));
4200				memcpy(&lun->mode_pages.debugconf_subpage[
4201				       CTL_PAGE_SAVED],
4202				       &debugconf_page_default,
4203				       sizeof(debugconf_page_default));
4204				page_index->page_data =
4205					(uint8_t *)lun->mode_pages.debugconf_subpage;
4206
4207				current_page = (struct copan_debugconf_subpage *)
4208					(page_index->page_data +
4209					 (page_index->page_len *
4210					  CTL_PAGE_CURRENT));
4211				saved_page = (struct copan_debugconf_subpage *)
4212					(page_index->page_data +
4213					 (page_index->page_len *
4214					  CTL_PAGE_SAVED));
4215				break;
4216			}
4217			default:
4218				panic("invalid subpage value %d",
4219				      page_index->subpage);
4220				break;
4221			}
4222   			break;
4223		}
4224		default:
4225			panic("invalid page value %d",
4226			      page_index->page_code & SMPH_PC_MASK);
4227			break;
4228    	}
4229	}
4230
4231	return (CTL_RETVAL_COMPLETE);
4232}
4233
4234/*
4235 * LUN allocation.
4236 *
4237 * Requirements:
4238 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4239 *   wants us to allocate the LUN and he can block.
4240 * - ctl_softc is always set
4241 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4242 *
4243 * Returns 0 for success, non-zero (errno) for failure.
4244 */
4245static int
4246ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4247	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4248{
4249	struct ctl_lun *nlun, *lun;
4250	struct ctl_frontend *fe;
4251	int lun_number, i, lun_malloced;
4252
4253	if (be_lun == NULL)
4254		return (EINVAL);
4255
4256	/*
4257	 * We currently only support Direct Access or Processor LUN types.
4258	 */
4259	switch (be_lun->lun_type) {
4260	case T_DIRECT:
4261		break;
4262	case T_PROCESSOR:
4263		break;
4264	case T_SEQUENTIAL:
4265	case T_CHANGER:
4266	default:
4267		be_lun->lun_config_status(be_lun->be_lun,
4268					  CTL_LUN_CONFIG_FAILURE);
4269		break;
4270	}
4271	if (ctl_lun == NULL) {
4272		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4273		lun_malloced = 1;
4274	} else {
4275		lun_malloced = 0;
4276		lun = ctl_lun;
4277	}
4278
4279	memset(lun, 0, sizeof(*lun));
4280	if (lun_malloced)
4281		lun->flags = CTL_LUN_MALLOCED;
4282
4283	mtx_lock(&ctl_softc->ctl_lock);
4284	/*
4285	 * See if the caller requested a particular LUN number.  If so, see
4286	 * if it is available.  Otherwise, allocate the first available LUN.
4287	 */
4288	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4289		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4290		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4291			mtx_unlock(&ctl_softc->ctl_lock);
4292			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4293				printf("ctl: requested LUN ID %d is higher "
4294				       "than CTL_MAX_LUNS - 1 (%d)\n",
4295				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4296			} else {
4297				/*
4298				 * XXX KDM return an error, or just assign
4299				 * another LUN ID in this case??
4300				 */
4301				printf("ctl: requested LUN ID %d is already "
4302				       "in use\n", be_lun->req_lun_id);
4303			}
4304			if (lun->flags & CTL_LUN_MALLOCED)
4305				free(lun, M_CTL);
4306			be_lun->lun_config_status(be_lun->be_lun,
4307						  CTL_LUN_CONFIG_FAILURE);
4308			return (ENOSPC);
4309		}
4310		lun_number = be_lun->req_lun_id;
4311	} else {
4312		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4313		if (lun_number == -1) {
4314			mtx_unlock(&ctl_softc->ctl_lock);
4315			printf("ctl: can't allocate LUN on target %ju, out of "
4316			       "LUNs\n", (uintmax_t)target_id.id);
4317			if (lun->flags & CTL_LUN_MALLOCED)
4318				free(lun, M_CTL);
4319			be_lun->lun_config_status(be_lun->be_lun,
4320						  CTL_LUN_CONFIG_FAILURE);
4321			return (ENOSPC);
4322		}
4323	}
4324	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4325
4326	lun->target = target_id;
4327	lun->lun = lun_number;
4328	lun->be_lun = be_lun;
4329	/*
4330	 * The processor LUN is always enabled.  Disk LUNs come on line
4331	 * disabled, and must be enabled by the backend.
4332	 */
4333	lun->flags |= CTL_LUN_DISABLED;
4334	lun->backend = be_lun->be;
4335	be_lun->ctl_lun = lun;
4336	be_lun->lun_id = lun_number;
4337	atomic_add_int(&be_lun->be->num_luns, 1);
4338	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4339		lun->flags |= CTL_LUN_STOPPED;
4340
4341	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4342		lun->flags |= CTL_LUN_INOPERABLE;
4343
4344	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4345		lun->flags |= CTL_LUN_PRIMARY_SC;
4346
4347	lun->ctl_softc = ctl_softc;
4348	TAILQ_INIT(&lun->ooa_queue);
4349	TAILQ_INIT(&lun->blocked_queue);
4350	STAILQ_INIT(&lun->error_list);
4351
4352	/*
4353	 * Initialize the mode page index.
4354	 */
4355	ctl_init_page_index(lun);
4356
4357	/*
4358	 * Set the poweron UA for all initiators on this LUN only.
4359	 */
4360	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4361		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4362
4363	/*
4364	 * Now, before we insert this lun on the lun list, set the lun
4365	 * inventory changed UA for all other luns.
4366	 */
4367	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4368		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4369			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4370		}
4371	}
4372
4373	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4374
4375	ctl_softc->ctl_luns[lun_number] = lun;
4376
4377	ctl_softc->num_luns++;
4378
4379	/* Setup statistics gathering */
4380	lun->stats.device_type = be_lun->lun_type;
4381	lun->stats.lun_number = lun_number;
4382	if (lun->stats.device_type == T_DIRECT)
4383		lun->stats.blocksize = be_lun->blocksize;
4384	else
4385		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4386	for (i = 0;i < CTL_MAX_PORTS;i++)
4387		lun->stats.ports[i].targ_port = i;
4388
4389	mtx_unlock(&ctl_softc->ctl_lock);
4390
4391	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4392
4393	/*
4394	 * Run through each registered FETD and bring it online if it isn't
4395	 * already.  Enable the target ID if it hasn't been enabled, and
4396	 * enable this particular LUN.
4397	 */
4398	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4399		int retval;
4400
4401		/*
4402		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4403		 * to do things differently if we go to a multiple target
4404		 * ID scheme.
4405		 */
4406		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4407
4408			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4409			if (retval != 0) {
4410				printf("ctl_alloc_lun: FETD %s port %d "
4411				       "returned error %d for targ_enable on "
4412				       "target %ju\n", fe->port_name,
4413				       fe->targ_port, retval,
4414				       (uintmax_t)target_id.id);
4415			} else
4416				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4417		}
4418
4419		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4420		if (retval != 0) {
4421			printf("ctl_alloc_lun: FETD %s port %d returned error "
4422			       "%d for lun_enable on target %ju lun %d\n",
4423			       fe->port_name, fe->targ_port, retval,
4424			       (uintmax_t)target_id.id, lun_number);
4425		} else
4426			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4427	}
4428	return (0);
4429}
4430
4431/*
4432 * Delete a LUN.
4433 * Assumptions:
4434 * - LUN has already been marked invalid and any pending I/O has been taken
4435 *   care of.
4436 */
4437static int
4438ctl_free_lun(struct ctl_lun *lun)
4439{
4440	struct ctl_softc *softc;
4441#if 0
4442	struct ctl_frontend *fe;
4443#endif
4444	struct ctl_lun *nlun;
4445	union ctl_io *io, *next_io;
4446	int i;
4447
4448	softc = lun->ctl_softc;
4449
4450	mtx_assert(&softc->ctl_lock, MA_OWNED);
4451
4452	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4453
4454	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4455
4456	softc->ctl_luns[lun->lun] = NULL;
4457
4458	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4459		printf("ctl_free_lun: aieee!! freeing a LUN with "
4460		       "outstanding I/O!!\n");
4461	}
4462
4463	/*
4464	 * If we have anything pending on the RtR queue, remove it.
4465	 */
4466	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4467	     io = next_io) {
4468		uint32_t targ_lun;
4469
4470		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4471		targ_lun = io->io_hdr.nexus.targ_lun;
4472		if (io->io_hdr.nexus.lun_map_fn != NULL)
4473			targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
4474		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4475		 && (targ_lun == lun->lun))
4476			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4477				      ctl_io_hdr, links);
4478	}
4479
4480	/*
4481	 * Then remove everything from the blocked queue.
4482	 */
4483	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4484	     io = next_io) {
4485		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4486		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4487		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4488	}
4489
4490	/*
4491	 * Now clear out the OOA queue, and free all the I/O.
4492	 * XXX KDM should we notify the FETD here?  We probably need to
4493	 * quiesce the LUN before deleting it.
4494	 */
4495	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4496	     io = next_io) {
4497		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4498		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4499		ctl_free_io_internal(io, /*have_lock*/ 1);
4500	}
4501
4502	softc->num_luns--;
4503
4504	/*
4505	 * XXX KDM this scheme only works for a single target/multiple LUN
4506	 * setup.  It needs to be revamped for a multiple target scheme.
4507	 *
4508	 * XXX KDM this results in fe->lun_disable() getting called twice,
4509	 * once when ctl_disable_lun() is called, and a second time here.
4510	 * We really need to re-think the LUN disable semantics.  There
4511	 * should probably be several steps/levels to LUN removal:
4512	 *  - disable
4513	 *  - invalidate
4514	 *  - free
4515 	 *
4516	 * Right now we only have a disable method when communicating to
4517	 * the front end ports, at least for individual LUNs.
4518	 */
4519#if 0
4520	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4521		int retval;
4522
4523		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4524					 lun->lun);
4525		if (retval != 0) {
4526			printf("ctl_free_lun: FETD %s port %d returned error "
4527			       "%d for lun_disable on target %ju lun %jd\n",
4528			       fe->port_name, fe->targ_port, retval,
4529			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4530		}
4531
4532		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4533			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4534
4535			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4536			if (retval != 0) {
4537				printf("ctl_free_lun: FETD %s port %d "
4538				       "returned error %d for targ_disable on "
4539				       "target %ju\n", fe->port_name,
4540				       fe->targ_port, retval,
4541				       (uintmax_t)lun->target.id);
4542			} else
4543				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4544
4545			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4546				continue;
4547
4548#if 0
4549			fe->port_offline(fe->onoff_arg);
4550			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4551#endif
4552		}
4553	}
4554#endif
4555
4556	/*
4557	 * Tell the backend to free resources, if this LUN has a backend.
4558	 */
4559	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4560	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4561
4562	if (lun->flags & CTL_LUN_MALLOCED)
4563		free(lun, M_CTL);
4564
4565	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4566		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4567			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4568		}
4569	}
4570
4571	return (0);
4572}
4573
4574static void
4575ctl_create_lun(struct ctl_be_lun *be_lun)
4576{
4577	struct ctl_softc *ctl_softc;
4578
4579	ctl_softc = control_softc;
4580
4581	/*
4582	 * ctl_alloc_lun() should handle all potential failure cases.
4583	 */
4584	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4585}
4586
4587int
4588ctl_add_lun(struct ctl_be_lun *be_lun)
4589{
4590	struct ctl_softc *ctl_softc;
4591
4592	ctl_softc = control_softc;
4593
4594	mtx_lock(&ctl_softc->ctl_lock);
4595	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4596	mtx_unlock(&ctl_softc->ctl_lock);
4597
4598	ctl_wakeup_thread();
4599
4600	return (0);
4601}
4602
4603int
4604ctl_enable_lun(struct ctl_be_lun *be_lun)
4605{
4606	struct ctl_softc *ctl_softc;
4607	struct ctl_frontend *fe, *nfe;
4608	struct ctl_lun *lun;
4609	int retval;
4610
4611	ctl_softc = control_softc;
4612
4613	lun = (struct ctl_lun *)be_lun->ctl_lun;
4614
4615	mtx_lock(&ctl_softc->ctl_lock);
4616	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4617		/*
4618		 * eh?  Why did we get called if the LUN is already
4619		 * enabled?
4620		 */
4621		mtx_unlock(&ctl_softc->ctl_lock);
4622		return (0);
4623	}
4624	lun->flags &= ~CTL_LUN_DISABLED;
4625
4626	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4627		nfe = STAILQ_NEXT(fe, links);
4628
4629		/*
4630		 * Drop the lock while we call the FETD's enable routine.
4631		 * This can lead to a callback into CTL (at least in the
4632		 * case of the internal initiator frontend.
4633		 */
4634		mtx_unlock(&ctl_softc->ctl_lock);
4635		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4636		mtx_lock(&ctl_softc->ctl_lock);
4637		if (retval != 0) {
4638			printf("%s: FETD %s port %d returned error "
4639			       "%d for lun_enable on target %ju lun %jd\n",
4640			       __func__, fe->port_name, fe->targ_port, retval,
4641			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4642		}
4643#if 0
4644		 else {
4645            /* NOTE:  TODO:  why does lun enable affect port status? */
4646			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4647		}
4648#endif
4649	}
4650
4651	mtx_unlock(&ctl_softc->ctl_lock);
4652
4653	return (0);
4654}
4655
4656int
4657ctl_disable_lun(struct ctl_be_lun *be_lun)
4658{
4659	struct ctl_softc *ctl_softc;
4660	struct ctl_frontend *fe;
4661	struct ctl_lun *lun;
4662	int retval;
4663
4664	ctl_softc = control_softc;
4665
4666	lun = (struct ctl_lun *)be_lun->ctl_lun;
4667
4668	mtx_lock(&ctl_softc->ctl_lock);
4669
4670	if (lun->flags & CTL_LUN_DISABLED) {
4671		mtx_unlock(&ctl_softc->ctl_lock);
4672		return (0);
4673	}
4674	lun->flags |= CTL_LUN_DISABLED;
4675
4676	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4677		mtx_unlock(&ctl_softc->ctl_lock);
4678		/*
4679		 * Drop the lock before we call the frontend's disable
4680		 * routine, to avoid lock order reversals.
4681		 *
4682		 * XXX KDM what happens if the frontend list changes while
4683		 * we're traversing it?  It's unlikely, but should be handled.
4684		 */
4685		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4686					 lun->lun);
4687		mtx_lock(&ctl_softc->ctl_lock);
4688		if (retval != 0) {
4689			printf("ctl_alloc_lun: FETD %s port %d returned error "
4690			       "%d for lun_disable on target %ju lun %jd\n",
4691			       fe->port_name, fe->targ_port, retval,
4692			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4693		}
4694	}
4695
4696	mtx_unlock(&ctl_softc->ctl_lock);
4697
4698	return (0);
4699}
4700
4701int
4702ctl_start_lun(struct ctl_be_lun *be_lun)
4703{
4704	struct ctl_softc *ctl_softc;
4705	struct ctl_lun *lun;
4706
4707	ctl_softc = control_softc;
4708
4709	lun = (struct ctl_lun *)be_lun->ctl_lun;
4710
4711	mtx_lock(&ctl_softc->ctl_lock);
4712	lun->flags &= ~CTL_LUN_STOPPED;
4713	mtx_unlock(&ctl_softc->ctl_lock);
4714
4715	return (0);
4716}
4717
4718int
4719ctl_stop_lun(struct ctl_be_lun *be_lun)
4720{
4721	struct ctl_softc *ctl_softc;
4722	struct ctl_lun *lun;
4723
4724	ctl_softc = control_softc;
4725
4726	lun = (struct ctl_lun *)be_lun->ctl_lun;
4727
4728	mtx_lock(&ctl_softc->ctl_lock);
4729	lun->flags |= CTL_LUN_STOPPED;
4730	mtx_unlock(&ctl_softc->ctl_lock);
4731
4732	return (0);
4733}
4734
4735int
4736ctl_lun_offline(struct ctl_be_lun *be_lun)
4737{
4738	struct ctl_softc *ctl_softc;
4739	struct ctl_lun *lun;
4740
4741	ctl_softc = control_softc;
4742
4743	lun = (struct ctl_lun *)be_lun->ctl_lun;
4744
4745	mtx_lock(&ctl_softc->ctl_lock);
4746	lun->flags |= CTL_LUN_OFFLINE;
4747	mtx_unlock(&ctl_softc->ctl_lock);
4748
4749	return (0);
4750}
4751
4752int
4753ctl_lun_online(struct ctl_be_lun *be_lun)
4754{
4755	struct ctl_softc *ctl_softc;
4756	struct ctl_lun *lun;
4757
4758	ctl_softc = control_softc;
4759
4760	lun = (struct ctl_lun *)be_lun->ctl_lun;
4761
4762	mtx_lock(&ctl_softc->ctl_lock);
4763	lun->flags &= ~CTL_LUN_OFFLINE;
4764	mtx_unlock(&ctl_softc->ctl_lock);
4765
4766	return (0);
4767}
4768
4769int
4770ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4771{
4772	struct ctl_softc *ctl_softc;
4773	struct ctl_lun *lun;
4774
4775	ctl_softc = control_softc;
4776
4777	lun = (struct ctl_lun *)be_lun->ctl_lun;
4778
4779	mtx_lock(&ctl_softc->ctl_lock);
4780
4781	/*
4782	 * The LUN needs to be disabled before it can be marked invalid.
4783	 */
4784	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4785		mtx_unlock(&ctl_softc->ctl_lock);
4786		return (-1);
4787	}
4788	/*
4789	 * Mark the LUN invalid.
4790	 */
4791	lun->flags |= CTL_LUN_INVALID;
4792
4793	/*
4794	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4795	 * If we have something in the OOA queue, we'll free it when the
4796	 * last I/O completes.
4797	 */
4798	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4799		ctl_free_lun(lun);
4800	mtx_unlock(&ctl_softc->ctl_lock);
4801
4802	return (0);
4803}
4804
4805int
4806ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4807{
4808	struct ctl_softc *ctl_softc;
4809	struct ctl_lun *lun;
4810
4811	ctl_softc = control_softc;
4812	lun = (struct ctl_lun *)be_lun->ctl_lun;
4813
4814	mtx_lock(&ctl_softc->ctl_lock);
4815	lun->flags |= CTL_LUN_INOPERABLE;
4816	mtx_unlock(&ctl_softc->ctl_lock);
4817
4818	return (0);
4819}
4820
4821int
4822ctl_lun_operable(struct ctl_be_lun *be_lun)
4823{
4824	struct ctl_softc *ctl_softc;
4825	struct ctl_lun *lun;
4826
4827	ctl_softc = control_softc;
4828	lun = (struct ctl_lun *)be_lun->ctl_lun;
4829
4830	mtx_lock(&ctl_softc->ctl_lock);
4831	lun->flags &= ~CTL_LUN_INOPERABLE;
4832	mtx_unlock(&ctl_softc->ctl_lock);
4833
4834	return (0);
4835}
4836
4837int
4838ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4839		   int lock)
4840{
4841	struct ctl_softc *softc;
4842	struct ctl_lun *lun;
4843	struct copan_aps_subpage *current_sp;
4844	struct ctl_page_index *page_index;
4845	int i;
4846
4847	softc = control_softc;
4848
4849	mtx_lock(&softc->ctl_lock);
4850
4851	lun = (struct ctl_lun *)be_lun->ctl_lun;
4852
4853	page_index = NULL;
4854	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4855		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4856		     APS_PAGE_CODE)
4857			continue;
4858
4859		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4860			continue;
4861		page_index = &lun->mode_pages.index[i];
4862	}
4863
4864	if (page_index == NULL) {
4865		mtx_unlock(&softc->ctl_lock);
4866		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4867		       (uintmax_t)lun->lun);
4868		return (1);
4869	}
4870#if 0
4871	if ((softc->aps_locked_lun != 0)
4872	 && (softc->aps_locked_lun != lun->lun)) {
4873		printf("%s: attempt to lock LUN %llu when %llu is already "
4874		       "locked\n");
4875		mtx_unlock(&softc->ctl_lock);
4876		return (1);
4877	}
4878#endif
4879
4880	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4881		(page_index->page_len * CTL_PAGE_CURRENT));
4882
4883	if (lock != 0) {
4884		current_sp->lock_active = APS_LOCK_ACTIVE;
4885		softc->aps_locked_lun = lun->lun;
4886	} else {
4887		current_sp->lock_active = 0;
4888		softc->aps_locked_lun = 0;
4889	}
4890
4891
4892	/*
4893	 * If we're in HA mode, try to send the lock message to the other
4894	 * side.
4895	 */
4896	if (ctl_is_single == 0) {
4897		int isc_retval;
4898		union ctl_ha_msg lock_msg;
4899
4900		lock_msg.hdr.nexus = *nexus;
4901		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4902		if (lock != 0)
4903			lock_msg.aps.lock_flag = 1;
4904		else
4905			lock_msg.aps.lock_flag = 0;
4906		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4907					 sizeof(lock_msg), 0);
4908		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4909			printf("%s: APS (lock=%d) error returned from "
4910			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4911			mtx_unlock(&softc->ctl_lock);
4912			return (1);
4913		}
4914	}
4915
4916	mtx_unlock(&softc->ctl_lock);
4917
4918	return (0);
4919}
4920
4921void
4922ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4923{
4924	struct ctl_lun *lun;
4925	struct ctl_softc *softc;
4926	int i;
4927
4928	softc = control_softc;
4929
4930	mtx_lock(&softc->ctl_lock);
4931
4932	lun = (struct ctl_lun *)be_lun->ctl_lun;
4933
4934	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4935		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4936
4937	mtx_unlock(&softc->ctl_lock);
4938}
4939
4940/*
4941 * Backend "memory move is complete" callback for requests that never
4942 * make it down to say RAIDCore's configuration code.
4943 */
4944int
4945ctl_config_move_done(union ctl_io *io)
4946{
4947	int retval;
4948
4949	retval = CTL_RETVAL_COMPLETE;
4950
4951
4952	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4953	/*
4954	 * XXX KDM this shouldn't happen, but what if it does?
4955	 */
4956	if (io->io_hdr.io_type != CTL_IO_SCSI)
4957		panic("I/O type isn't CTL_IO_SCSI!");
4958
4959	if ((io->io_hdr.port_status == 0)
4960	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4961	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4962		io->io_hdr.status = CTL_SUCCESS;
4963	else if ((io->io_hdr.port_status != 0)
4964	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4965	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4966		/*
4967		 * For hardware error sense keys, the sense key
4968		 * specific value is defined to be a retry count,
4969		 * but we use it to pass back an internal FETD
4970		 * error code.  XXX KDM  Hopefully the FETD is only
4971		 * using 16 bits for an error code, since that's
4972		 * all the space we have in the sks field.
4973		 */
4974		ctl_set_internal_failure(&io->scsiio,
4975					 /*sks_valid*/ 1,
4976					 /*retry_count*/
4977					 io->io_hdr.port_status);
4978		free(io->scsiio.kern_data_ptr, M_CTL);
4979		ctl_done(io);
4980		goto bailout;
4981	}
4982
4983	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4984	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4985	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4986		/*
4987		 * XXX KDM just assuming a single pointer here, and not a
4988		 * S/G list.  If we start using S/G lists for config data,
4989		 * we'll need to know how to clean them up here as well.
4990		 */
4991		free(io->scsiio.kern_data_ptr, M_CTL);
4992		/* Hopefully the user has already set the status... */
4993		ctl_done(io);
4994	} else {
4995		/*
4996		 * XXX KDM now we need to continue data movement.  Some
4997		 * options:
4998		 * - call ctl_scsiio() again?  We don't do this for data
4999		 *   writes, because for those at least we know ahead of
5000		 *   time where the write will go and how long it is.  For
5001		 *   config writes, though, that information is largely
5002		 *   contained within the write itself, thus we need to
5003		 *   parse out the data again.
5004		 *
5005		 * - Call some other function once the data is in?
5006		 */
5007
5008		/*
5009		 * XXX KDM call ctl_scsiio() again for now, and check flag
5010		 * bits to see whether we're allocated or not.
5011		 */
5012		retval = ctl_scsiio(&io->scsiio);
5013	}
5014bailout:
5015	return (retval);
5016}
5017
5018/*
5019 * This gets called by a backend driver when it is done with a
5020 * configuration write.
5021 */
5022void
5023ctl_config_write_done(union ctl_io *io)
5024{
5025	/*
5026	 * If the IO_CONT flag is set, we need to call the supplied
5027	 * function to continue processing the I/O, instead of completing
5028	 * the I/O just yet.
5029	 *
5030	 * If there is an error, though, we don't want to keep processing.
5031	 * Instead, just send status back to the initiator.
5032	 */
5033	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5034	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5035	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5036		io->scsiio.io_cont(io);
5037		return;
5038	}
5039	/*
5040	 * Since a configuration write can be done for commands that actually
5041	 * have data allocated, like write buffer, and commands that have
5042	 * no data, like start/stop unit, we need to check here.
5043	 */
5044	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5045		free(io->scsiio.kern_data_ptr, M_CTL);
5046	ctl_done(io);
5047}
5048
5049/*
5050 * SCSI release command.
5051 */
5052int
5053ctl_scsi_release(struct ctl_scsiio *ctsio)
5054{
5055	int length, longid, thirdparty_id, resv_id;
5056	struct ctl_softc *ctl_softc;
5057	struct ctl_lun *lun;
5058
5059	length = 0;
5060	resv_id = 0;
5061
5062	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5063
5064	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5065	ctl_softc = control_softc;
5066
5067	switch (ctsio->cdb[0]) {
5068	case RELEASE: {
5069		struct scsi_release *cdb;
5070
5071		cdb = (struct scsi_release *)ctsio->cdb;
5072		if ((cdb->byte2 & 0x1f) != 0) {
5073			ctl_set_invalid_field(ctsio,
5074					      /*sks_valid*/ 1,
5075					      /*command*/ 1,
5076					      /*field*/ 1,
5077					      /*bit_valid*/ 0,
5078					      /*bit*/ 0);
5079			ctl_done((union ctl_io *)ctsio);
5080			return (CTL_RETVAL_COMPLETE);
5081		}
5082		break;
5083	}
5084	case RELEASE_10: {
5085		struct scsi_release_10 *cdb;
5086
5087		cdb = (struct scsi_release_10 *)ctsio->cdb;
5088
5089		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5090			ctl_set_invalid_field(ctsio,
5091					      /*sks_valid*/ 1,
5092					      /*command*/ 1,
5093					      /*field*/ 1,
5094					      /*bit_valid*/ 1,
5095					      /*bit*/ 0);
5096			ctl_done((union ctl_io *)ctsio);
5097			return (CTL_RETVAL_COMPLETE);
5098
5099		}
5100
5101		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5102			ctl_set_invalid_field(ctsio,
5103					      /*sks_valid*/ 1,
5104					      /*command*/ 1,
5105					      /*field*/ 1,
5106					      /*bit_valid*/ 1,
5107					      /*bit*/ 4);
5108			ctl_done((union ctl_io *)ctsio);
5109			return (CTL_RETVAL_COMPLETE);
5110		}
5111
5112		if (cdb->byte2 & SR10_LONGID)
5113			longid = 1;
5114		else
5115			thirdparty_id = cdb->thirdparty_id;
5116
5117		resv_id = cdb->resv_id;
5118		length = scsi_2btoul(cdb->length);
5119		break;
5120	}
5121	}
5122
5123
5124	/*
5125	 * XXX KDM right now, we only support LUN reservation.  We don't
5126	 * support 3rd party reservations, or extent reservations, which
5127	 * might actually need the parameter list.  If we've gotten this
5128	 * far, we've got a LUN reservation.  Anything else got kicked out
5129	 * above.  So, according to SPC, ignore the length.
5130	 */
5131	length = 0;
5132
5133	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5134	 && (length > 0)) {
5135		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5136		ctsio->kern_data_len = length;
5137		ctsio->kern_total_len = length;
5138		ctsio->kern_data_resid = 0;
5139		ctsio->kern_rel_offset = 0;
5140		ctsio->kern_sg_entries = 0;
5141		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5142		ctsio->be_move_done = ctl_config_move_done;
5143		ctl_datamove((union ctl_io *)ctsio);
5144
5145		return (CTL_RETVAL_COMPLETE);
5146	}
5147
5148	if (length > 0)
5149		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5150
5151	mtx_lock(&ctl_softc->ctl_lock);
5152
5153	/*
5154	 * According to SPC, it is not an error for an intiator to attempt
5155	 * to release a reservation on a LUN that isn't reserved, or that
5156	 * is reserved by another initiator.  The reservation can only be
5157	 * released, though, by the initiator who made it or by one of
5158	 * several reset type events.
5159	 */
5160	if (lun->flags & CTL_LUN_RESERVED) {
5161		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5162		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5163		 && (ctsio->io_hdr.nexus.targ_target.id ==
5164		     lun->rsv_nexus.targ_target.id)) {
5165			lun->flags &= ~CTL_LUN_RESERVED;
5166		}
5167	}
5168
5169	ctsio->scsi_status = SCSI_STATUS_OK;
5170	ctsio->io_hdr.status = CTL_SUCCESS;
5171
5172	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5173		free(ctsio->kern_data_ptr, M_CTL);
5174		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5175	}
5176
5177	mtx_unlock(&ctl_softc->ctl_lock);
5178
5179	ctl_done((union ctl_io *)ctsio);
5180	return (CTL_RETVAL_COMPLETE);
5181}
5182
5183int
5184ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5185{
5186	int extent, thirdparty, longid;
5187	int resv_id, length;
5188	uint64_t thirdparty_id;
5189	struct ctl_softc *ctl_softc;
5190	struct ctl_lun *lun;
5191
5192	extent = 0;
5193	thirdparty = 0;
5194	longid = 0;
5195	resv_id = 0;
5196	length = 0;
5197	thirdparty_id = 0;
5198
5199	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5200
5201	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5202	ctl_softc = control_softc;
5203
5204	switch (ctsio->cdb[0]) {
5205	case RESERVE: {
5206		struct scsi_reserve *cdb;
5207
5208		cdb = (struct scsi_reserve *)ctsio->cdb;
5209		if ((cdb->byte2 & 0x1f) != 0) {
5210			ctl_set_invalid_field(ctsio,
5211					      /*sks_valid*/ 1,
5212					      /*command*/ 1,
5213					      /*field*/ 1,
5214					      /*bit_valid*/ 0,
5215					      /*bit*/ 0);
5216			ctl_done((union ctl_io *)ctsio);
5217			return (CTL_RETVAL_COMPLETE);
5218		}
5219		resv_id = cdb->resv_id;
5220		length = scsi_2btoul(cdb->length);
5221		break;
5222	}
5223	case RESERVE_10: {
5224		struct scsi_reserve_10 *cdb;
5225
5226		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5227
5228		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5229			ctl_set_invalid_field(ctsio,
5230					      /*sks_valid*/ 1,
5231					      /*command*/ 1,
5232					      /*field*/ 1,
5233					      /*bit_valid*/ 1,
5234					      /*bit*/ 0);
5235			ctl_done((union ctl_io *)ctsio);
5236			return (CTL_RETVAL_COMPLETE);
5237		}
5238		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5239			ctl_set_invalid_field(ctsio,
5240					      /*sks_valid*/ 1,
5241					      /*command*/ 1,
5242					      /*field*/ 1,
5243					      /*bit_valid*/ 1,
5244					      /*bit*/ 4);
5245			ctl_done((union ctl_io *)ctsio);
5246			return (CTL_RETVAL_COMPLETE);
5247		}
5248		if (cdb->byte2 & SR10_LONGID)
5249			longid = 1;
5250		else
5251			thirdparty_id = cdb->thirdparty_id;
5252
5253		resv_id = cdb->resv_id;
5254		length = scsi_2btoul(cdb->length);
5255		break;
5256	}
5257	}
5258
5259	/*
5260	 * XXX KDM right now, we only support LUN reservation.  We don't
5261	 * support 3rd party reservations, or extent reservations, which
5262	 * might actually need the parameter list.  If we've gotten this
5263	 * far, we've got a LUN reservation.  Anything else got kicked out
5264	 * above.  So, according to SPC, ignore the length.
5265	 */
5266	length = 0;
5267
5268	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5269	 && (length > 0)) {
5270		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5271		ctsio->kern_data_len = length;
5272		ctsio->kern_total_len = length;
5273		ctsio->kern_data_resid = 0;
5274		ctsio->kern_rel_offset = 0;
5275		ctsio->kern_sg_entries = 0;
5276		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5277		ctsio->be_move_done = ctl_config_move_done;
5278		ctl_datamove((union ctl_io *)ctsio);
5279
5280		return (CTL_RETVAL_COMPLETE);
5281	}
5282
5283	if (length > 0)
5284		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5285
5286	mtx_lock(&ctl_softc->ctl_lock);
5287	if (lun->flags & CTL_LUN_RESERVED) {
5288		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5289		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5290		 || (ctsio->io_hdr.nexus.targ_target.id !=
5291		     lun->rsv_nexus.targ_target.id)) {
5292			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5293			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5294			goto bailout;
5295		}
5296	}
5297
5298	lun->flags |= CTL_LUN_RESERVED;
5299	lun->rsv_nexus = ctsio->io_hdr.nexus;
5300
5301	ctsio->scsi_status = SCSI_STATUS_OK;
5302	ctsio->io_hdr.status = CTL_SUCCESS;
5303
5304bailout:
5305	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5306		free(ctsio->kern_data_ptr, M_CTL);
5307		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5308	}
5309
5310	mtx_unlock(&ctl_softc->ctl_lock);
5311
5312	ctl_done((union ctl_io *)ctsio);
5313	return (CTL_RETVAL_COMPLETE);
5314}
5315
5316int
5317ctl_start_stop(struct ctl_scsiio *ctsio)
5318{
5319	struct scsi_start_stop_unit *cdb;
5320	struct ctl_lun *lun;
5321	struct ctl_softc *ctl_softc;
5322	int retval;
5323
5324	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5325
5326	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5327	ctl_softc = control_softc;
5328	retval = 0;
5329
5330	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5331
5332	/*
5333	 * XXX KDM
5334	 * We don't support the immediate bit on a stop unit.  In order to
5335	 * do that, we would need to code up a way to know that a stop is
5336	 * pending, and hold off any new commands until it completes, one
5337	 * way or another.  Then we could accept or reject those commands
5338	 * depending on its status.  We would almost need to do the reverse
5339	 * of what we do below for an immediate start -- return the copy of
5340	 * the ctl_io to the FETD with status to send to the host (and to
5341	 * free the copy!) and then free the original I/O once the stop
5342	 * actually completes.  That way, the OOA queue mechanism can work
5343	 * to block commands that shouldn't proceed.  Another alternative
5344	 * would be to put the copy in the queue in place of the original,
5345	 * and return the original back to the caller.  That could be
5346	 * slightly safer..
5347	 */
5348	if ((cdb->byte2 & SSS_IMMED)
5349	 && ((cdb->how & SSS_START) == 0)) {
5350		ctl_set_invalid_field(ctsio,
5351				      /*sks_valid*/ 1,
5352				      /*command*/ 1,
5353				      /*field*/ 1,
5354				      /*bit_valid*/ 1,
5355				      /*bit*/ 0);
5356		ctl_done((union ctl_io *)ctsio);
5357		return (CTL_RETVAL_COMPLETE);
5358	}
5359
5360	/*
5361	 * We don't support the power conditions field.  We need to check
5362	 * this prior to checking the load/eject and start/stop bits.
5363	 */
5364	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5365		ctl_set_invalid_field(ctsio,
5366				      /*sks_valid*/ 1,
5367				      /*command*/ 1,
5368				      /*field*/ 4,
5369				      /*bit_valid*/ 1,
5370				      /*bit*/ 4);
5371		ctl_done((union ctl_io *)ctsio);
5372		return (CTL_RETVAL_COMPLETE);
5373	}
5374
5375	/*
5376	 * Media isn't removable, so we can't load or eject it.
5377	 */
5378	if ((cdb->how & SSS_LOEJ) != 0) {
5379		ctl_set_invalid_field(ctsio,
5380				      /*sks_valid*/ 1,
5381				      /*command*/ 1,
5382				      /*field*/ 4,
5383				      /*bit_valid*/ 1,
5384				      /*bit*/ 1);
5385		ctl_done((union ctl_io *)ctsio);
5386		return (CTL_RETVAL_COMPLETE);
5387	}
5388
5389	if ((lun->flags & CTL_LUN_PR_RESERVED)
5390	 && ((cdb->how & SSS_START)==0)) {
5391		uint32_t residx;
5392
5393		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5394		if (!lun->per_res[residx].registered
5395		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5396
5397			ctl_set_reservation_conflict(ctsio);
5398			ctl_done((union ctl_io *)ctsio);
5399			return (CTL_RETVAL_COMPLETE);
5400		}
5401	}
5402
5403	/*
5404	 * If there is no backend on this device, we can't start or stop
5405	 * it.  In theory we shouldn't get any start/stop commands in the
5406	 * first place at this level if the LUN doesn't have a backend.
5407	 * That should get stopped by the command decode code.
5408	 */
5409	if (lun->backend == NULL) {
5410		ctl_set_invalid_opcode(ctsio);
5411		ctl_done((union ctl_io *)ctsio);
5412		return (CTL_RETVAL_COMPLETE);
5413	}
5414
5415	/*
5416	 * XXX KDM Copan-specific offline behavior.
5417	 * Figure out a reasonable way to port this?
5418	 */
5419#ifdef NEEDTOPORT
5420	mtx_lock(&ctl_softc->ctl_lock);
5421
5422	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5423	 && (lun->flags & CTL_LUN_OFFLINE)) {
5424		/*
5425		 * If the LUN is offline, and the on/offline bit isn't set,
5426		 * reject the start or stop.  Otherwise, let it through.
5427		 */
5428		mtx_unlock(&ctl_softc->ctl_lock);
5429		ctl_set_lun_not_ready(ctsio);
5430		ctl_done((union ctl_io *)ctsio);
5431	} else {
5432		mtx_unlock(&ctl_softc->ctl_lock);
5433#endif /* NEEDTOPORT */
5434		/*
5435		 * This could be a start or a stop when we're online,
5436		 * or a stop/offline or start/online.  A start or stop when
5437		 * we're offline is covered in the case above.
5438		 */
5439		/*
5440		 * In the non-immediate case, we send the request to
5441		 * the backend and return status to the user when
5442		 * it is done.
5443		 *
5444		 * In the immediate case, we allocate a new ctl_io
5445		 * to hold a copy of the request, and send that to
5446		 * the backend.  We then set good status on the
5447		 * user's request and return it immediately.
5448		 */
5449		if (cdb->byte2 & SSS_IMMED) {
5450			union ctl_io *new_io;
5451
5452			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5453			if (new_io == NULL) {
5454				ctl_set_busy(ctsio);
5455				ctl_done((union ctl_io *)ctsio);
5456			} else {
5457				ctl_copy_io((union ctl_io *)ctsio,
5458					    new_io);
5459				retval = lun->backend->config_write(new_io);
5460				ctl_set_success(ctsio);
5461				ctl_done((union ctl_io *)ctsio);
5462			}
5463		} else {
5464			retval = lun->backend->config_write(
5465				(union ctl_io *)ctsio);
5466		}
5467#ifdef NEEDTOPORT
5468	}
5469#endif
5470	return (retval);
5471}
5472
5473/*
5474 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5475 * we don't really do anything with the LBA and length fields if the user
5476 * passes them in.  Instead we'll just flush out the cache for the entire
5477 * LUN.
5478 */
5479int
5480ctl_sync_cache(struct ctl_scsiio *ctsio)
5481{
5482	struct ctl_lun *lun;
5483	struct ctl_softc *ctl_softc;
5484	uint64_t starting_lba;
5485	uint32_t block_count;
5486	int reladr, immed;
5487	int retval;
5488
5489	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5490
5491	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5492	ctl_softc = control_softc;
5493	retval = 0;
5494	reladr = 0;
5495	immed = 0;
5496
5497	switch (ctsio->cdb[0]) {
5498	case SYNCHRONIZE_CACHE: {
5499		struct scsi_sync_cache *cdb;
5500		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5501
5502		if (cdb->byte2 & SSC_RELADR)
5503			reladr = 1;
5504
5505		if (cdb->byte2 & SSC_IMMED)
5506			immed = 1;
5507
5508		starting_lba = scsi_4btoul(cdb->begin_lba);
5509		block_count = scsi_2btoul(cdb->lb_count);
5510		break;
5511	}
5512	case SYNCHRONIZE_CACHE_16: {
5513		struct scsi_sync_cache_16 *cdb;
5514		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5515
5516		if (cdb->byte2 & SSC_RELADR)
5517			reladr = 1;
5518
5519		if (cdb->byte2 & SSC_IMMED)
5520			immed = 1;
5521
5522		starting_lba = scsi_8btou64(cdb->begin_lba);
5523		block_count = scsi_4btoul(cdb->lb_count);
5524		break;
5525	}
5526	default:
5527		ctl_set_invalid_opcode(ctsio);
5528		ctl_done((union ctl_io *)ctsio);
5529		goto bailout;
5530		break; /* NOTREACHED */
5531	}
5532
5533	if (immed) {
5534		/*
5535		 * We don't support the immediate bit.  Since it's in the
5536		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5537		 * commands, we can just return the same error in either
5538		 * case.
5539		 */
5540		ctl_set_invalid_field(ctsio,
5541				      /*sks_valid*/ 1,
5542				      /*command*/ 1,
5543				      /*field*/ 1,
5544				      /*bit_valid*/ 1,
5545				      /*bit*/ 1);
5546		ctl_done((union ctl_io *)ctsio);
5547		goto bailout;
5548	}
5549
5550	if (reladr) {
5551		/*
5552		 * We don't support the reladr bit either.  It can only be
5553		 * used with linked commands, and we don't support linked
5554		 * commands.  Since the bit is in the same place for the
5555		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5556		 * just return the same error in either case.
5557		 */
5558		ctl_set_invalid_field(ctsio,
5559				      /*sks_valid*/ 1,
5560				      /*command*/ 1,
5561				      /*field*/ 1,
5562				      /*bit_valid*/ 1,
5563				      /*bit*/ 0);
5564		ctl_done((union ctl_io *)ctsio);
5565		goto bailout;
5566	}
5567
5568	/*
5569	 * We check the LBA and length, but don't do anything with them.
5570	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5571	 * get flushed.  This check will just help satisfy anyone who wants
5572	 * to see an error for an out of range LBA.
5573	 */
5574	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5575		ctl_set_lba_out_of_range(ctsio);
5576		ctl_done((union ctl_io *)ctsio);
5577		goto bailout;
5578	}
5579
5580	/*
5581	 * If this LUN has no backend, we can't flush the cache anyway.
5582	 */
5583	if (lun->backend == NULL) {
5584		ctl_set_invalid_opcode(ctsio);
5585		ctl_done((union ctl_io *)ctsio);
5586		goto bailout;
5587	}
5588
5589	/*
5590	 * Check to see whether we're configured to send the SYNCHRONIZE
5591	 * CACHE command directly to the back end.
5592	 */
5593	mtx_lock(&ctl_softc->ctl_lock);
5594	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5595	 && (++(lun->sync_count) >= lun->sync_interval)) {
5596		lun->sync_count = 0;
5597		mtx_unlock(&ctl_softc->ctl_lock);
5598		retval = lun->backend->config_write((union ctl_io *)ctsio);
5599	} else {
5600		mtx_unlock(&ctl_softc->ctl_lock);
5601		ctl_set_success(ctsio);
5602		ctl_done((union ctl_io *)ctsio);
5603	}
5604
5605bailout:
5606
5607	return (retval);
5608}
5609
5610int
5611ctl_format(struct ctl_scsiio *ctsio)
5612{
5613	struct scsi_format *cdb;
5614	struct ctl_lun *lun;
5615	struct ctl_softc *ctl_softc;
5616	int length, defect_list_len;
5617
5618	CTL_DEBUG_PRINT(("ctl_format\n"));
5619
5620	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5621	ctl_softc = control_softc;
5622
5623	cdb = (struct scsi_format *)ctsio->cdb;
5624
5625	length = 0;
5626	if (cdb->byte2 & SF_FMTDATA) {
5627		if (cdb->byte2 & SF_LONGLIST)
5628			length = sizeof(struct scsi_format_header_long);
5629		else
5630			length = sizeof(struct scsi_format_header_short);
5631	}
5632
5633	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5634	 && (length > 0)) {
5635		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5636		ctsio->kern_data_len = length;
5637		ctsio->kern_total_len = length;
5638		ctsio->kern_data_resid = 0;
5639		ctsio->kern_rel_offset = 0;
5640		ctsio->kern_sg_entries = 0;
5641		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5642		ctsio->be_move_done = ctl_config_move_done;
5643		ctl_datamove((union ctl_io *)ctsio);
5644
5645		return (CTL_RETVAL_COMPLETE);
5646	}
5647
5648	defect_list_len = 0;
5649
5650	if (cdb->byte2 & SF_FMTDATA) {
5651		if (cdb->byte2 & SF_LONGLIST) {
5652			struct scsi_format_header_long *header;
5653
5654			header = (struct scsi_format_header_long *)
5655				ctsio->kern_data_ptr;
5656
5657			defect_list_len = scsi_4btoul(header->defect_list_len);
5658			if (defect_list_len != 0) {
5659				ctl_set_invalid_field(ctsio,
5660						      /*sks_valid*/ 1,
5661						      /*command*/ 0,
5662						      /*field*/ 2,
5663						      /*bit_valid*/ 0,
5664						      /*bit*/ 0);
5665				goto bailout;
5666			}
5667		} else {
5668			struct scsi_format_header_short *header;
5669
5670			header = (struct scsi_format_header_short *)
5671				ctsio->kern_data_ptr;
5672
5673			defect_list_len = scsi_2btoul(header->defect_list_len);
5674			if (defect_list_len != 0) {
5675				ctl_set_invalid_field(ctsio,
5676						      /*sks_valid*/ 1,
5677						      /*command*/ 0,
5678						      /*field*/ 2,
5679						      /*bit_valid*/ 0,
5680						      /*bit*/ 0);
5681				goto bailout;
5682			}
5683		}
5684	}
5685
5686	/*
5687	 * The format command will clear out the "Medium format corrupted"
5688	 * status if set by the configuration code.  That status is really
5689	 * just a way to notify the host that we have lost the media, and
5690	 * get them to issue a command that will basically make them think
5691	 * they're blowing away the media.
5692	 */
5693	mtx_lock(&ctl_softc->ctl_lock);
5694	lun->flags &= ~CTL_LUN_INOPERABLE;
5695	mtx_unlock(&ctl_softc->ctl_lock);
5696
5697	ctsio->scsi_status = SCSI_STATUS_OK;
5698	ctsio->io_hdr.status = CTL_SUCCESS;
5699bailout:
5700
5701	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5702		free(ctsio->kern_data_ptr, M_CTL);
5703		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5704	}
5705
5706	ctl_done((union ctl_io *)ctsio);
5707	return (CTL_RETVAL_COMPLETE);
5708}
5709
5710int
5711ctl_write_buffer(struct ctl_scsiio *ctsio)
5712{
5713	struct scsi_write_buffer *cdb;
5714	struct copan_page_header *header;
5715	struct ctl_lun *lun;
5716	struct ctl_softc *ctl_softc;
5717	int buffer_offset, len;
5718	int retval;
5719
5720	header = NULL;
5721
5722	retval = CTL_RETVAL_COMPLETE;
5723
5724	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5725
5726	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5727	ctl_softc = control_softc;
5728	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5729
5730	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5731		ctl_set_invalid_field(ctsio,
5732				      /*sks_valid*/ 1,
5733				      /*command*/ 1,
5734				      /*field*/ 1,
5735				      /*bit_valid*/ 1,
5736				      /*bit*/ 4);
5737		ctl_done((union ctl_io *)ctsio);
5738		return (CTL_RETVAL_COMPLETE);
5739	}
5740	if (cdb->buffer_id != 0) {
5741		ctl_set_invalid_field(ctsio,
5742				      /*sks_valid*/ 1,
5743				      /*command*/ 1,
5744				      /*field*/ 2,
5745				      /*bit_valid*/ 0,
5746				      /*bit*/ 0);
5747		ctl_done((union ctl_io *)ctsio);
5748		return (CTL_RETVAL_COMPLETE);
5749	}
5750
5751	len = scsi_3btoul(cdb->length);
5752	buffer_offset = scsi_3btoul(cdb->offset);
5753
5754	if (len > sizeof(lun->write_buffer)) {
5755		ctl_set_invalid_field(ctsio,
5756				      /*sks_valid*/ 1,
5757				      /*command*/ 1,
5758				      /*field*/ 6,
5759				      /*bit_valid*/ 0,
5760				      /*bit*/ 0);
5761		ctl_done((union ctl_io *)ctsio);
5762		return (CTL_RETVAL_COMPLETE);
5763	}
5764
5765	if (buffer_offset != 0) {
5766		ctl_set_invalid_field(ctsio,
5767				      /*sks_valid*/ 1,
5768				      /*command*/ 1,
5769				      /*field*/ 3,
5770				      /*bit_valid*/ 0,
5771				      /*bit*/ 0);
5772		ctl_done((union ctl_io *)ctsio);
5773		return (CTL_RETVAL_COMPLETE);
5774	}
5775
5776	/*
5777	 * If we've got a kernel request that hasn't been malloced yet,
5778	 * malloc it and tell the caller the data buffer is here.
5779	 */
5780	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5781		ctsio->kern_data_ptr = lun->write_buffer;
5782		ctsio->kern_data_len = len;
5783		ctsio->kern_total_len = len;
5784		ctsio->kern_data_resid = 0;
5785		ctsio->kern_rel_offset = 0;
5786		ctsio->kern_sg_entries = 0;
5787		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5788		ctsio->be_move_done = ctl_config_move_done;
5789		ctl_datamove((union ctl_io *)ctsio);
5790
5791		return (CTL_RETVAL_COMPLETE);
5792	}
5793
5794	ctl_done((union ctl_io *)ctsio);
5795
5796	return (CTL_RETVAL_COMPLETE);
5797}
5798
5799/*
5800 * Note that this function currently doesn't actually do anything inside
5801 * CTL to enforce things if the DQue bit is turned on.
5802 *
5803 * Also note that this function can't be used in the default case, because
5804 * the DQue bit isn't set in the changeable mask for the control mode page
5805 * anyway.  This is just here as an example for how to implement a page
5806 * handler, and a placeholder in case we want to allow the user to turn
5807 * tagged queueing on and off.
5808 *
5809 * The D_SENSE bit handling is functional, however, and will turn
5810 * descriptor sense on and off for a given LUN.
5811 */
5812int
5813ctl_control_page_handler(struct ctl_scsiio *ctsio,
5814			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5815{
5816	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5817	struct ctl_lun *lun;
5818	struct ctl_softc *softc;
5819	int set_ua;
5820	uint32_t initidx;
5821
5822	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5823	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5824	set_ua = 0;
5825
5826	user_cp = (struct scsi_control_page *)page_ptr;
5827	current_cp = (struct scsi_control_page *)
5828		(page_index->page_data + (page_index->page_len *
5829		CTL_PAGE_CURRENT));
5830	saved_cp = (struct scsi_control_page *)
5831		(page_index->page_data + (page_index->page_len *
5832		CTL_PAGE_SAVED));
5833
5834	softc = control_softc;
5835
5836	mtx_lock(&softc->ctl_lock);
5837	if (((current_cp->rlec & SCP_DSENSE) == 0)
5838	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5839		/*
5840		 * Descriptor sense is currently turned off and the user
5841		 * wants to turn it on.
5842		 */
5843		current_cp->rlec |= SCP_DSENSE;
5844		saved_cp->rlec |= SCP_DSENSE;
5845		lun->flags |= CTL_LUN_SENSE_DESC;
5846		set_ua = 1;
5847	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
5848		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
5849		/*
5850		 * Descriptor sense is currently turned on, and the user
5851		 * wants to turn it off.
5852		 */
5853		current_cp->rlec &= ~SCP_DSENSE;
5854		saved_cp->rlec &= ~SCP_DSENSE;
5855		lun->flags &= ~CTL_LUN_SENSE_DESC;
5856		set_ua = 1;
5857	}
5858	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
5859		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5860#ifdef NEEDTOPORT
5861			csevent_log(CSC_CTL | CSC_SHELF_SW |
5862				    CTL_UNTAG_TO_UNTAG,
5863				    csevent_LogType_Trace,
5864				    csevent_Severity_Information,
5865				    csevent_AlertLevel_Green,
5866				    csevent_FRU_Firmware,
5867				    csevent_FRU_Unknown,
5868				    "Received untagged to untagged transition");
5869#endif /* NEEDTOPORT */
5870		} else {
5871#ifdef NEEDTOPORT
5872			csevent_log(CSC_CTL | CSC_SHELF_SW |
5873				    CTL_UNTAG_TO_TAG,
5874				    csevent_LogType_ConfigChange,
5875				    csevent_Severity_Information,
5876				    csevent_AlertLevel_Green,
5877				    csevent_FRU_Firmware,
5878				    csevent_FRU_Unknown,
5879				    "Received untagged to tagged "
5880				    "queueing transition");
5881#endif /* NEEDTOPORT */
5882
5883			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5884			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
5885			set_ua = 1;
5886		}
5887	} else {
5888		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
5889#ifdef NEEDTOPORT
5890			csevent_log(CSC_CTL | CSC_SHELF_SW |
5891				    CTL_TAG_TO_UNTAG,
5892				    csevent_LogType_ConfigChange,
5893				    csevent_Severity_Warning,
5894				    csevent_AlertLevel_Yellow,
5895				    csevent_FRU_Firmware,
5896				    csevent_FRU_Unknown,
5897				    "Received tagged queueing to untagged "
5898				    "transition");
5899#endif /* NEEDTOPORT */
5900
5901			current_cp->queue_flags |= SCP_QUEUE_DQUE;
5902			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
5903			set_ua = 1;
5904		} else {
5905#ifdef NEEDTOPORT
5906			csevent_log(CSC_CTL | CSC_SHELF_SW |
5907				    CTL_TAG_TO_TAG,
5908				    csevent_LogType_Trace,
5909				    csevent_Severity_Information,
5910				    csevent_AlertLevel_Green,
5911				    csevent_FRU_Firmware,
5912				    csevent_FRU_Unknown,
5913				    "Received tagged queueing to tagged "
5914				    "queueing transition");
5915#endif /* NEEDTOPORT */
5916		}
5917	}
5918	if (set_ua != 0) {
5919		int i;
5920		/*
5921		 * Let other initiators know that the mode
5922		 * parameters for this LUN have changed.
5923		 */
5924		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
5925			if (i == initidx)
5926				continue;
5927
5928			lun->pending_sense[i].ua_pending |=
5929				CTL_UA_MODE_CHANGE;
5930		}
5931	}
5932	mtx_unlock(&softc->ctl_lock);
5933
5934	return (0);
5935}
5936
5937int
5938ctl_power_sp_handler(struct ctl_scsiio *ctsio,
5939		     struct ctl_page_index *page_index, uint8_t *page_ptr)
5940{
5941	return (0);
5942}
5943
5944int
5945ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
5946			   struct ctl_page_index *page_index, int pc)
5947{
5948	struct copan_power_subpage *page;
5949
5950	page = (struct copan_power_subpage *)page_index->page_data +
5951		(page_index->page_len * pc);
5952
5953	switch (pc) {
5954	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5955		/*
5956		 * We don't update the changable bits for this page.
5957		 */
5958		break;
5959	case SMS_PAGE_CTRL_CURRENT >> 6:
5960	case SMS_PAGE_CTRL_DEFAULT >> 6:
5961	case SMS_PAGE_CTRL_SAVED >> 6:
5962#ifdef NEEDTOPORT
5963		ctl_update_power_subpage(page);
5964#endif
5965		break;
5966	default:
5967#ifdef NEEDTOPORT
5968		EPRINT(0, "Invalid PC %d!!", pc);
5969#endif
5970		break;
5971	}
5972	return (0);
5973}
5974
5975
5976int
5977ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
5978		   struct ctl_page_index *page_index, uint8_t *page_ptr)
5979{
5980	struct copan_aps_subpage *user_sp;
5981	struct copan_aps_subpage *current_sp;
5982	union ctl_modepage_info *modepage_info;
5983	struct ctl_softc *softc;
5984	struct ctl_lun *lun;
5985	int retval;
5986
5987	retval = CTL_RETVAL_COMPLETE;
5988	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
5989		     (page_index->page_len * CTL_PAGE_CURRENT));
5990	softc = control_softc;
5991	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5992
5993	user_sp = (struct copan_aps_subpage *)page_ptr;
5994
5995	modepage_info = (union ctl_modepage_info *)
5996		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5997
5998	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
5999	modepage_info->header.subpage = page_index->subpage;
6000	modepage_info->aps.lock_active = user_sp->lock_active;
6001
6002	mtx_lock(&softc->ctl_lock);
6003
6004	/*
6005	 * If there is a request to lock the LUN and another LUN is locked
6006	 * this is an error. If the requested LUN is already locked ignore
6007	 * the request. If no LUN is locked attempt to lock it.
6008	 * if there is a request to unlock the LUN and the LUN is currently
6009	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6010	 * if another LUN is locked or no LUN is locked.
6011	 */
6012	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6013		if (softc->aps_locked_lun == lun->lun) {
6014			/*
6015			 * This LUN is already locked, so we're done.
6016			 */
6017			retval = CTL_RETVAL_COMPLETE;
6018		} else if (softc->aps_locked_lun == 0) {
6019			/*
6020			 * No one has the lock, pass the request to the
6021			 * backend.
6022			 */
6023			retval = lun->backend->config_write(
6024				(union ctl_io *)ctsio);
6025		} else {
6026			/*
6027			 * Someone else has the lock, throw out the request.
6028			 */
6029			ctl_set_already_locked(ctsio);
6030			free(ctsio->kern_data_ptr, M_CTL);
6031			ctl_done((union ctl_io *)ctsio);
6032
6033			/*
6034			 * Set the return value so that ctl_do_mode_select()
6035			 * won't try to complete the command.  We already
6036			 * completed it here.
6037			 */
6038			retval = CTL_RETVAL_ERROR;
6039		}
6040	} else if (softc->aps_locked_lun == lun->lun) {
6041		/*
6042		 * This LUN is locked, so pass the unlock request to the
6043		 * backend.
6044		 */
6045		retval = lun->backend->config_write((union ctl_io *)ctsio);
6046	}
6047	mtx_unlock(&softc->ctl_lock);
6048
6049	return (retval);
6050}
6051
6052int
6053ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6054				struct ctl_page_index *page_index,
6055				uint8_t *page_ptr)
6056{
6057	uint8_t *c;
6058	int i;
6059
6060	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6061	ctl_time_io_secs =
6062		(c[0] << 8) |
6063		(c[1] << 0) |
6064		0;
6065	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6066	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6067	printf("page data:");
6068	for (i=0; i<8; i++)
6069		printf(" %.2x",page_ptr[i]);
6070	printf("\n");
6071	return (0);
6072}
6073
6074int
6075ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6076			       struct ctl_page_index *page_index,
6077			       int pc)
6078{
6079	struct copan_debugconf_subpage *page;
6080
6081	page = (struct copan_debugconf_subpage *)page_index->page_data +
6082		(page_index->page_len * pc);
6083
6084	switch (pc) {
6085	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6086	case SMS_PAGE_CTRL_DEFAULT >> 6:
6087	case SMS_PAGE_CTRL_SAVED >> 6:
6088		/*
6089		 * We don't update the changable or default bits for this page.
6090		 */
6091		break;
6092	case SMS_PAGE_CTRL_CURRENT >> 6:
6093		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6094		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6095		break;
6096	default:
6097#ifdef NEEDTOPORT
6098		EPRINT(0, "Invalid PC %d!!", pc);
6099#endif /* NEEDTOPORT */
6100		break;
6101	}
6102	return (0);
6103}
6104
6105
6106static int
6107ctl_do_mode_select(union ctl_io *io)
6108{
6109	struct scsi_mode_page_header *page_header;
6110	struct ctl_page_index *page_index;
6111	struct ctl_scsiio *ctsio;
6112	int control_dev, page_len;
6113	int page_len_offset, page_len_size;
6114	union ctl_modepage_info *modepage_info;
6115	struct ctl_lun *lun;
6116	int *len_left, *len_used;
6117	int retval, i;
6118
6119	ctsio = &io->scsiio;
6120	page_index = NULL;
6121	page_len = 0;
6122	retval = CTL_RETVAL_COMPLETE;
6123
6124	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6125
6126	if (lun->be_lun->lun_type != T_DIRECT)
6127		control_dev = 1;
6128	else
6129		control_dev = 0;
6130
6131	modepage_info = (union ctl_modepage_info *)
6132		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6133	len_left = &modepage_info->header.len_left;
6134	len_used = &modepage_info->header.len_used;
6135
6136do_next_page:
6137
6138	page_header = (struct scsi_mode_page_header *)
6139		(ctsio->kern_data_ptr + *len_used);
6140
6141	if (*len_left == 0) {
6142		free(ctsio->kern_data_ptr, M_CTL);
6143		ctl_set_success(ctsio);
6144		ctl_done((union ctl_io *)ctsio);
6145		return (CTL_RETVAL_COMPLETE);
6146	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6147
6148		free(ctsio->kern_data_ptr, M_CTL);
6149		ctl_set_param_len_error(ctsio);
6150		ctl_done((union ctl_io *)ctsio);
6151		return (CTL_RETVAL_COMPLETE);
6152
6153	} else if ((page_header->page_code & SMPH_SPF)
6154		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6155
6156		free(ctsio->kern_data_ptr, M_CTL);
6157		ctl_set_param_len_error(ctsio);
6158		ctl_done((union ctl_io *)ctsio);
6159		return (CTL_RETVAL_COMPLETE);
6160	}
6161
6162
6163	/*
6164	 * XXX KDM should we do something with the block descriptor?
6165	 */
6166	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6167
6168		if ((control_dev != 0)
6169		 && (lun->mode_pages.index[i].page_flags &
6170		     CTL_PAGE_FLAG_DISK_ONLY))
6171			continue;
6172
6173		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6174		    (page_header->page_code & SMPH_PC_MASK))
6175			continue;
6176
6177		/*
6178		 * If neither page has a subpage code, then we've got a
6179		 * match.
6180		 */
6181		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6182		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6183			page_index = &lun->mode_pages.index[i];
6184			page_len = page_header->page_length;
6185			break;
6186		}
6187
6188		/*
6189		 * If both pages have subpages, then the subpage numbers
6190		 * have to match.
6191		 */
6192		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6193		  && (page_header->page_code & SMPH_SPF)) {
6194			struct scsi_mode_page_header_sp *sph;
6195
6196			sph = (struct scsi_mode_page_header_sp *)page_header;
6197
6198			if (lun->mode_pages.index[i].subpage ==
6199			    sph->subpage) {
6200				page_index = &lun->mode_pages.index[i];
6201				page_len = scsi_2btoul(sph->page_length);
6202				break;
6203			}
6204		}
6205	}
6206
6207	/*
6208	 * If we couldn't find the page, or if we don't have a mode select
6209	 * handler for it, send back an error to the user.
6210	 */
6211	if ((page_index == NULL)
6212	 || (page_index->select_handler == NULL)) {
6213		ctl_set_invalid_field(ctsio,
6214				      /*sks_valid*/ 1,
6215				      /*command*/ 0,
6216				      /*field*/ *len_used,
6217				      /*bit_valid*/ 0,
6218				      /*bit*/ 0);
6219		free(ctsio->kern_data_ptr, M_CTL);
6220		ctl_done((union ctl_io *)ctsio);
6221		return (CTL_RETVAL_COMPLETE);
6222	}
6223
6224	if (page_index->page_code & SMPH_SPF) {
6225		page_len_offset = 2;
6226		page_len_size = 2;
6227	} else {
6228		page_len_size = 1;
6229		page_len_offset = 1;
6230	}
6231
6232	/*
6233	 * If the length the initiator gives us isn't the one we specify in
6234	 * the mode page header, or if they didn't specify enough data in
6235	 * the CDB to avoid truncating this page, kick out the request.
6236	 */
6237	if ((page_len != (page_index->page_len - page_len_offset -
6238			  page_len_size))
6239	 || (*len_left < page_index->page_len)) {
6240
6241
6242		ctl_set_invalid_field(ctsio,
6243				      /*sks_valid*/ 1,
6244				      /*command*/ 0,
6245				      /*field*/ *len_used + page_len_offset,
6246				      /*bit_valid*/ 0,
6247				      /*bit*/ 0);
6248		free(ctsio->kern_data_ptr, M_CTL);
6249		ctl_done((union ctl_io *)ctsio);
6250		return (CTL_RETVAL_COMPLETE);
6251	}
6252
6253	/*
6254	 * Run through the mode page, checking to make sure that the bits
6255	 * the user changed are actually legal for him to change.
6256	 */
6257	for (i = 0; i < page_index->page_len; i++) {
6258		uint8_t *user_byte, *change_mask, *current_byte;
6259		int bad_bit;
6260		int j;
6261
6262		user_byte = (uint8_t *)page_header + i;
6263		change_mask = page_index->page_data +
6264			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6265		current_byte = page_index->page_data +
6266			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6267
6268		/*
6269		 * Check to see whether the user set any bits in this byte
6270		 * that he is not allowed to set.
6271		 */
6272		if ((*user_byte & ~(*change_mask)) ==
6273		    (*current_byte & ~(*change_mask)))
6274			continue;
6275
6276		/*
6277		 * Go through bit by bit to determine which one is illegal.
6278		 */
6279		bad_bit = 0;
6280		for (j = 7; j >= 0; j--) {
6281			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6282			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6283				bad_bit = i;
6284				break;
6285			}
6286		}
6287		ctl_set_invalid_field(ctsio,
6288				      /*sks_valid*/ 1,
6289				      /*command*/ 0,
6290				      /*field*/ *len_used + i,
6291				      /*bit_valid*/ 1,
6292				      /*bit*/ bad_bit);
6293		free(ctsio->kern_data_ptr, M_CTL);
6294		ctl_done((union ctl_io *)ctsio);
6295		return (CTL_RETVAL_COMPLETE);
6296	}
6297
6298	/*
6299	 * Decrement these before we call the page handler, since we may
6300	 * end up getting called back one way or another before the handler
6301	 * returns to this context.
6302	 */
6303	*len_left -= page_index->page_len;
6304	*len_used += page_index->page_len;
6305
6306	retval = page_index->select_handler(ctsio, page_index,
6307					    (uint8_t *)page_header);
6308
6309	/*
6310	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6311	 * wait until this queued command completes to finish processing
6312	 * the mode page.  If it returns anything other than
6313	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6314	 * already set the sense information, freed the data pointer, and
6315	 * completed the io for us.
6316	 */
6317	if (retval != CTL_RETVAL_COMPLETE)
6318		goto bailout_no_done;
6319
6320	/*
6321	 * If the initiator sent us more than one page, parse the next one.
6322	 */
6323	if (*len_left > 0)
6324		goto do_next_page;
6325
6326	ctl_set_success(ctsio);
6327	free(ctsio->kern_data_ptr, M_CTL);
6328	ctl_done((union ctl_io *)ctsio);
6329
6330bailout_no_done:
6331
6332	return (CTL_RETVAL_COMPLETE);
6333
6334}
6335
6336int
6337ctl_mode_select(struct ctl_scsiio *ctsio)
6338{
6339	int param_len, pf, sp;
6340	int header_size, bd_len;
6341	int len_left, len_used;
6342	struct ctl_page_index *page_index;
6343	struct ctl_lun *lun;
6344	int control_dev, page_len;
6345	union ctl_modepage_info *modepage_info;
6346	int retval;
6347
6348	pf = 0;
6349	sp = 0;
6350	page_len = 0;
6351	len_used = 0;
6352	len_left = 0;
6353	retval = 0;
6354	bd_len = 0;
6355	page_index = NULL;
6356
6357	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6358
6359	if (lun->be_lun->lun_type != T_DIRECT)
6360		control_dev = 1;
6361	else
6362		control_dev = 0;
6363
6364	switch (ctsio->cdb[0]) {
6365	case MODE_SELECT_6: {
6366		struct scsi_mode_select_6 *cdb;
6367
6368		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6369
6370		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6371		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6372
6373		param_len = cdb->length;
6374		header_size = sizeof(struct scsi_mode_header_6);
6375		break;
6376	}
6377	case MODE_SELECT_10: {
6378		struct scsi_mode_select_10 *cdb;
6379
6380		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6381
6382		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6383		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6384
6385		param_len = scsi_2btoul(cdb->length);
6386		header_size = sizeof(struct scsi_mode_header_10);
6387		break;
6388	}
6389	default:
6390		ctl_set_invalid_opcode(ctsio);
6391		ctl_done((union ctl_io *)ctsio);
6392		return (CTL_RETVAL_COMPLETE);
6393		break; /* NOTREACHED */
6394	}
6395
6396	/*
6397	 * From SPC-3:
6398	 * "A parameter list length of zero indicates that the Data-Out Buffer
6399	 * shall be empty. This condition shall not be considered as an error."
6400	 */
6401	if (param_len == 0) {
6402		ctl_set_success(ctsio);
6403		ctl_done((union ctl_io *)ctsio);
6404		return (CTL_RETVAL_COMPLETE);
6405	}
6406
6407	/*
6408	 * Since we'll hit this the first time through, prior to
6409	 * allocation, we don't need to free a data buffer here.
6410	 */
6411	if (param_len < header_size) {
6412		ctl_set_param_len_error(ctsio);
6413		ctl_done((union ctl_io *)ctsio);
6414		return (CTL_RETVAL_COMPLETE);
6415	}
6416
6417	/*
6418	 * Allocate the data buffer and grab the user's data.  In theory,
6419	 * we shouldn't have to sanity check the parameter list length here
6420	 * because the maximum size is 64K.  We should be able to malloc
6421	 * that much without too many problems.
6422	 */
6423	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6424		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6425		ctsio->kern_data_len = param_len;
6426		ctsio->kern_total_len = param_len;
6427		ctsio->kern_data_resid = 0;
6428		ctsio->kern_rel_offset = 0;
6429		ctsio->kern_sg_entries = 0;
6430		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6431		ctsio->be_move_done = ctl_config_move_done;
6432		ctl_datamove((union ctl_io *)ctsio);
6433
6434		return (CTL_RETVAL_COMPLETE);
6435	}
6436
6437	switch (ctsio->cdb[0]) {
6438	case MODE_SELECT_6: {
6439		struct scsi_mode_header_6 *mh6;
6440
6441		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6442		bd_len = mh6->blk_desc_len;
6443		break;
6444	}
6445	case MODE_SELECT_10: {
6446		struct scsi_mode_header_10 *mh10;
6447
6448		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6449		bd_len = scsi_2btoul(mh10->blk_desc_len);
6450		break;
6451	}
6452	default:
6453		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6454		break;
6455	}
6456
6457	if (param_len < (header_size + bd_len)) {
6458		free(ctsio->kern_data_ptr, M_CTL);
6459		ctl_set_param_len_error(ctsio);
6460		ctl_done((union ctl_io *)ctsio);
6461		return (CTL_RETVAL_COMPLETE);
6462	}
6463
6464	/*
6465	 * Set the IO_CONT flag, so that if this I/O gets passed to
6466	 * ctl_config_write_done(), it'll get passed back to
6467	 * ctl_do_mode_select() for further processing, or completion if
6468	 * we're all done.
6469	 */
6470	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6471	ctsio->io_cont = ctl_do_mode_select;
6472
6473	modepage_info = (union ctl_modepage_info *)
6474		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6475
6476	memset(modepage_info, 0, sizeof(*modepage_info));
6477
6478	len_left = param_len - header_size - bd_len;
6479	len_used = header_size + bd_len;
6480
6481	modepage_info->header.len_left = len_left;
6482	modepage_info->header.len_used = len_used;
6483
6484	return (ctl_do_mode_select((union ctl_io *)ctsio));
6485}
6486
6487int
6488ctl_mode_sense(struct ctl_scsiio *ctsio)
6489{
6490	struct ctl_lun *lun;
6491	int pc, page_code, dbd, llba, subpage;
6492	int alloc_len, page_len, header_len, total_len;
6493	struct scsi_mode_block_descr *block_desc;
6494	struct ctl_page_index *page_index;
6495	int control_dev;
6496
6497	dbd = 0;
6498	llba = 0;
6499	block_desc = NULL;
6500	page_index = NULL;
6501
6502	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6503
6504	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6505
6506	if (lun->be_lun->lun_type != T_DIRECT)
6507		control_dev = 1;
6508	else
6509		control_dev = 0;
6510
6511	switch (ctsio->cdb[0]) {
6512	case MODE_SENSE_6: {
6513		struct scsi_mode_sense_6 *cdb;
6514
6515		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6516
6517		header_len = sizeof(struct scsi_mode_hdr_6);
6518		if (cdb->byte2 & SMS_DBD)
6519			dbd = 1;
6520		else
6521			header_len += sizeof(struct scsi_mode_block_descr);
6522
6523		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6524		page_code = cdb->page & SMS_PAGE_CODE;
6525		subpage = cdb->subpage;
6526		alloc_len = cdb->length;
6527		break;
6528	}
6529	case MODE_SENSE_10: {
6530		struct scsi_mode_sense_10 *cdb;
6531
6532		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6533
6534		header_len = sizeof(struct scsi_mode_hdr_10);
6535
6536		if (cdb->byte2 & SMS_DBD)
6537			dbd = 1;
6538		else
6539			header_len += sizeof(struct scsi_mode_block_descr);
6540		if (cdb->byte2 & SMS10_LLBAA)
6541			llba = 1;
6542		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6543		page_code = cdb->page & SMS_PAGE_CODE;
6544		subpage = cdb->subpage;
6545		alloc_len = scsi_2btoul(cdb->length);
6546		break;
6547	}
6548	default:
6549		ctl_set_invalid_opcode(ctsio);
6550		ctl_done((union ctl_io *)ctsio);
6551		return (CTL_RETVAL_COMPLETE);
6552		break; /* NOTREACHED */
6553	}
6554
6555	/*
6556	 * We have to make a first pass through to calculate the size of
6557	 * the pages that match the user's query.  Then we allocate enough
6558	 * memory to hold it, and actually copy the data into the buffer.
6559	 */
6560	switch (page_code) {
6561	case SMS_ALL_PAGES_PAGE: {
6562		int i;
6563
6564		page_len = 0;
6565
6566		/*
6567		 * At the moment, values other than 0 and 0xff here are
6568		 * reserved according to SPC-3.
6569		 */
6570		if ((subpage != SMS_SUBPAGE_PAGE_0)
6571		 && (subpage != SMS_SUBPAGE_ALL)) {
6572			ctl_set_invalid_field(ctsio,
6573					      /*sks_valid*/ 1,
6574					      /*command*/ 1,
6575					      /*field*/ 3,
6576					      /*bit_valid*/ 0,
6577					      /*bit*/ 0);
6578			ctl_done((union ctl_io *)ctsio);
6579			return (CTL_RETVAL_COMPLETE);
6580		}
6581
6582		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6583			if ((control_dev != 0)
6584			 && (lun->mode_pages.index[i].page_flags &
6585			     CTL_PAGE_FLAG_DISK_ONLY))
6586				continue;
6587
6588			/*
6589			 * We don't use this subpage if the user didn't
6590			 * request all subpages.
6591			 */
6592			if ((lun->mode_pages.index[i].subpage != 0)
6593			 && (subpage == SMS_SUBPAGE_PAGE_0))
6594				continue;
6595
6596#if 0
6597			printf("found page %#x len %d\n",
6598			       lun->mode_pages.index[i].page_code &
6599			       SMPH_PC_MASK,
6600			       lun->mode_pages.index[i].page_len);
6601#endif
6602			page_len += lun->mode_pages.index[i].page_len;
6603		}
6604		break;
6605	}
6606	default: {
6607		int i;
6608
6609		page_len = 0;
6610
6611		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6612			/* Look for the right page code */
6613			if ((lun->mode_pages.index[i].page_code &
6614			     SMPH_PC_MASK) != page_code)
6615				continue;
6616
6617			/* Look for the right subpage or the subpage wildcard*/
6618			if ((lun->mode_pages.index[i].subpage != subpage)
6619			 && (subpage != SMS_SUBPAGE_ALL))
6620				continue;
6621
6622			/* Make sure the page is supported for this dev type */
6623			if ((control_dev != 0)
6624			 && (lun->mode_pages.index[i].page_flags &
6625			     CTL_PAGE_FLAG_DISK_ONLY))
6626				continue;
6627
6628#if 0
6629			printf("found page %#x len %d\n",
6630			       lun->mode_pages.index[i].page_code &
6631			       SMPH_PC_MASK,
6632			       lun->mode_pages.index[i].page_len);
6633#endif
6634
6635			page_len += lun->mode_pages.index[i].page_len;
6636		}
6637
6638		if (page_len == 0) {
6639			ctl_set_invalid_field(ctsio,
6640					      /*sks_valid*/ 1,
6641					      /*command*/ 1,
6642					      /*field*/ 2,
6643					      /*bit_valid*/ 1,
6644					      /*bit*/ 5);
6645			ctl_done((union ctl_io *)ctsio);
6646			return (CTL_RETVAL_COMPLETE);
6647		}
6648		break;
6649	}
6650	}
6651
6652	total_len = header_len + page_len;
6653#if 0
6654	printf("header_len = %d, page_len = %d, total_len = %d\n",
6655	       header_len, page_len, total_len);
6656#endif
6657
6658	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6659	ctsio->kern_sg_entries = 0;
6660	ctsio->kern_data_resid = 0;
6661	ctsio->kern_rel_offset = 0;
6662	if (total_len < alloc_len) {
6663		ctsio->residual = alloc_len - total_len;
6664		ctsio->kern_data_len = total_len;
6665		ctsio->kern_total_len = total_len;
6666	} else {
6667		ctsio->residual = 0;
6668		ctsio->kern_data_len = alloc_len;
6669		ctsio->kern_total_len = alloc_len;
6670	}
6671
6672	switch (ctsio->cdb[0]) {
6673	case MODE_SENSE_6: {
6674		struct scsi_mode_hdr_6 *header;
6675
6676		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6677
6678		header->datalen = ctl_min(total_len - 1, 254);
6679
6680		if (dbd)
6681			header->block_descr_len = 0;
6682		else
6683			header->block_descr_len =
6684				sizeof(struct scsi_mode_block_descr);
6685		block_desc = (struct scsi_mode_block_descr *)&header[1];
6686		break;
6687	}
6688	case MODE_SENSE_10: {
6689		struct scsi_mode_hdr_10 *header;
6690		int datalen;
6691
6692		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6693
6694		datalen = ctl_min(total_len - 2, 65533);
6695		scsi_ulto2b(datalen, header->datalen);
6696		if (dbd)
6697			scsi_ulto2b(0, header->block_descr_len);
6698		else
6699			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6700				    header->block_descr_len);
6701		block_desc = (struct scsi_mode_block_descr *)&header[1];
6702		break;
6703	}
6704	default:
6705		panic("invalid CDB type %#x", ctsio->cdb[0]);
6706		break; /* NOTREACHED */
6707	}
6708
6709	/*
6710	 * If we've got a disk, use its blocksize in the block
6711	 * descriptor.  Otherwise, just set it to 0.
6712	 */
6713	if (dbd == 0) {
6714		if (control_dev != 0)
6715			scsi_ulto3b(lun->be_lun->blocksize,
6716				    block_desc->block_len);
6717		else
6718			scsi_ulto3b(0, block_desc->block_len);
6719	}
6720
6721	switch (page_code) {
6722	case SMS_ALL_PAGES_PAGE: {
6723		int i, data_used;
6724
6725		data_used = header_len;
6726		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6727			struct ctl_page_index *page_index;
6728
6729			page_index = &lun->mode_pages.index[i];
6730
6731			if ((control_dev != 0)
6732			 && (page_index->page_flags &
6733			    CTL_PAGE_FLAG_DISK_ONLY))
6734				continue;
6735
6736			/*
6737			 * We don't use this subpage if the user didn't
6738			 * request all subpages.  We already checked (above)
6739			 * to make sure the user only specified a subpage
6740			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6741			 */
6742			if ((page_index->subpage != 0)
6743			 && (subpage == SMS_SUBPAGE_PAGE_0))
6744				continue;
6745
6746			/*
6747			 * Call the handler, if it exists, to update the
6748			 * page to the latest values.
6749			 */
6750			if (page_index->sense_handler != NULL)
6751				page_index->sense_handler(ctsio, page_index,pc);
6752
6753			memcpy(ctsio->kern_data_ptr + data_used,
6754			       page_index->page_data +
6755			       (page_index->page_len * pc),
6756			       page_index->page_len);
6757			data_used += page_index->page_len;
6758		}
6759		break;
6760	}
6761	default: {
6762		int i, data_used;
6763
6764		data_used = header_len;
6765
6766		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6767			struct ctl_page_index *page_index;
6768
6769			page_index = &lun->mode_pages.index[i];
6770
6771			/* Look for the right page code */
6772			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6773				continue;
6774
6775			/* Look for the right subpage or the subpage wildcard*/
6776			if ((page_index->subpage != subpage)
6777			 && (subpage != SMS_SUBPAGE_ALL))
6778				continue;
6779
6780			/* Make sure the page is supported for this dev type */
6781			if ((control_dev != 0)
6782			 && (page_index->page_flags &
6783			     CTL_PAGE_FLAG_DISK_ONLY))
6784				continue;
6785
6786			/*
6787			 * Call the handler, if it exists, to update the
6788			 * page to the latest values.
6789			 */
6790			if (page_index->sense_handler != NULL)
6791				page_index->sense_handler(ctsio, page_index,pc);
6792
6793			memcpy(ctsio->kern_data_ptr + data_used,
6794			       page_index->page_data +
6795			       (page_index->page_len * pc),
6796			       page_index->page_len);
6797			data_used += page_index->page_len;
6798		}
6799		break;
6800	}
6801	}
6802
6803	ctsio->scsi_status = SCSI_STATUS_OK;
6804
6805	ctsio->be_move_done = ctl_config_move_done;
6806	ctl_datamove((union ctl_io *)ctsio);
6807
6808	return (CTL_RETVAL_COMPLETE);
6809}
6810
6811int
6812ctl_read_capacity(struct ctl_scsiio *ctsio)
6813{
6814	struct scsi_read_capacity *cdb;
6815	struct scsi_read_capacity_data *data;
6816	struct ctl_lun *lun;
6817	uint32_t lba;
6818
6819	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6820
6821	cdb = (struct scsi_read_capacity *)ctsio->cdb;
6822
6823	lba = scsi_4btoul(cdb->addr);
6824	if (((cdb->pmi & SRC_PMI) == 0)
6825	 && (lba != 0)) {
6826		ctl_set_invalid_field(/*ctsio*/ ctsio,
6827				      /*sks_valid*/ 1,
6828				      /*command*/ 1,
6829				      /*field*/ 2,
6830				      /*bit_valid*/ 0,
6831				      /*bit*/ 0);
6832		ctl_done((union ctl_io *)ctsio);
6833		return (CTL_RETVAL_COMPLETE);
6834	}
6835
6836	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6837
6838	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6839	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6840	ctsio->residual = 0;
6841	ctsio->kern_data_len = sizeof(*data);
6842	ctsio->kern_total_len = sizeof(*data);
6843	ctsio->kern_data_resid = 0;
6844	ctsio->kern_rel_offset = 0;
6845	ctsio->kern_sg_entries = 0;
6846
6847	/*
6848	 * If the maximum LBA is greater than 0xfffffffe, the user must
6849	 * issue a SERVICE ACTION IN (16) command, with the read capacity
6850	 * serivce action set.
6851	 */
6852	if (lun->be_lun->maxlba > 0xfffffffe)
6853		scsi_ulto4b(0xffffffff, data->addr);
6854	else
6855		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6856
6857	/*
6858	 * XXX KDM this may not be 512 bytes...
6859	 */
6860	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6861
6862	ctsio->scsi_status = SCSI_STATUS_OK;
6863
6864	ctsio->be_move_done = ctl_config_move_done;
6865	ctl_datamove((union ctl_io *)ctsio);
6866
6867	return (CTL_RETVAL_COMPLETE);
6868}
6869
6870static int
6871ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6872{
6873	struct scsi_read_capacity_16 *cdb;
6874	struct scsi_read_capacity_data_long *data;
6875	struct ctl_lun *lun;
6876	uint64_t lba;
6877	uint32_t alloc_len;
6878
6879	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6880
6881	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6882
6883	alloc_len = scsi_4btoul(cdb->alloc_len);
6884	lba = scsi_8btou64(cdb->addr);
6885
6886	if ((cdb->reladr & SRC16_PMI)
6887	 && (lba != 0)) {
6888		ctl_set_invalid_field(/*ctsio*/ ctsio,
6889				      /*sks_valid*/ 1,
6890				      /*command*/ 1,
6891				      /*field*/ 2,
6892				      /*bit_valid*/ 0,
6893				      /*bit*/ 0);
6894		ctl_done((union ctl_io *)ctsio);
6895		return (CTL_RETVAL_COMPLETE);
6896	}
6897
6898	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6899
6900	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6901	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6902
6903	if (sizeof(*data) < alloc_len) {
6904		ctsio->residual = alloc_len - sizeof(*data);
6905		ctsio->kern_data_len = sizeof(*data);
6906		ctsio->kern_total_len = sizeof(*data);
6907	} else {
6908		ctsio->residual = 0;
6909		ctsio->kern_data_len = alloc_len;
6910		ctsio->kern_total_len = alloc_len;
6911	}
6912	ctsio->kern_data_resid = 0;
6913	ctsio->kern_rel_offset = 0;
6914	ctsio->kern_sg_entries = 0;
6915
6916	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6917	/* XXX KDM this may not be 512 bytes... */
6918	scsi_ulto4b(lun->be_lun->blocksize, data->length);
6919
6920	ctsio->scsi_status = SCSI_STATUS_OK;
6921
6922	ctsio->be_move_done = ctl_config_move_done;
6923	ctl_datamove((union ctl_io *)ctsio);
6924
6925	return (CTL_RETVAL_COMPLETE);
6926}
6927
6928int
6929ctl_service_action_in(struct ctl_scsiio *ctsio)
6930{
6931	struct scsi_service_action_in *cdb;
6932	int retval;
6933
6934	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
6935
6936	cdb = (struct scsi_service_action_in *)ctsio->cdb;
6937
6938	retval = CTL_RETVAL_COMPLETE;
6939
6940	switch (cdb->service_action) {
6941	case SRC16_SERVICE_ACTION:
6942		retval = ctl_read_capacity_16(ctsio);
6943		break;
6944	default:
6945		ctl_set_invalid_field(/*ctsio*/ ctsio,
6946				      /*sks_valid*/ 1,
6947				      /*command*/ 1,
6948				      /*field*/ 1,
6949				      /*bit_valid*/ 1,
6950				      /*bit*/ 4);
6951		ctl_done((union ctl_io *)ctsio);
6952		break;
6953	}
6954
6955	return (retval);
6956}
6957
6958int
6959ctl_maintenance_in(struct ctl_scsiio *ctsio)
6960{
6961	struct scsi_maintenance_in *cdb;
6962	int retval;
6963	int alloc_len, total_len = 0;
6964	int num_target_port_groups, single;
6965	struct ctl_lun *lun;
6966	struct ctl_softc *softc;
6967	struct scsi_target_group_data *rtg_ptr;
6968	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
6969	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
6970	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
6971
6972	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
6973
6974	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
6975	softc = control_softc;
6976	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6977
6978	retval = CTL_RETVAL_COMPLETE;
6979
6980	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
6981		ctl_set_invalid_field(/*ctsio*/ ctsio,
6982				      /*sks_valid*/ 1,
6983				      /*command*/ 1,
6984				      /*field*/ 1,
6985				      /*bit_valid*/ 1,
6986				      /*bit*/ 4);
6987		ctl_done((union ctl_io *)ctsio);
6988		return(retval);
6989	}
6990
6991	mtx_lock(&softc->ctl_lock);
6992	single = ctl_is_single;
6993	mtx_unlock(&softc->ctl_lock);
6994
6995	if (single)
6996        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
6997	else
6998        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
6999
7000	total_len = sizeof(struct scsi_target_group_data) +
7001		sizeof(struct scsi_target_port_group_descriptor) *
7002		num_target_port_groups +
7003		sizeof(struct scsi_target_port_descriptor) *
7004		NUM_PORTS_PER_GRP * num_target_port_groups;
7005
7006	alloc_len = scsi_4btoul(cdb->length);
7007
7008	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7009
7010	ctsio->kern_sg_entries = 0;
7011
7012	if (total_len < alloc_len) {
7013		ctsio->residual = alloc_len - total_len;
7014		ctsio->kern_data_len = total_len;
7015		ctsio->kern_total_len = total_len;
7016	} else {
7017		ctsio->residual = 0;
7018		ctsio->kern_data_len = alloc_len;
7019		ctsio->kern_total_len = alloc_len;
7020	}
7021	ctsio->kern_data_resid = 0;
7022	ctsio->kern_rel_offset = 0;
7023
7024	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7025
7026	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7027	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7028	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7029	        &tp_desc_ptr1_1->desc_list[0];
7030
7031	if (single == 0) {
7032		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7033	                &tp_desc_ptr1_2->desc_list[0];
7034		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7035		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7036	        	&tp_desc_ptr2_1->desc_list[0];
7037        } else {
7038		tpg_desc_ptr2 = NULL;
7039		tp_desc_ptr2_1 = NULL;
7040		tp_desc_ptr2_2 = NULL;
7041	}
7042
7043	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7044	if (single == 0) {
7045        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7046			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7047				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7048				tpg_desc_ptr2->pref_state =
7049					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7050			} else {
7051				tpg_desc_ptr1->pref_state =
7052					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7053				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7054			}
7055		} else {
7056			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7057				tpg_desc_ptr1->pref_state =
7058					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7059				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7060			} else {
7061				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7062				tpg_desc_ptr2->pref_state =
7063					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7064			}
7065		}
7066	} else {
7067		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7068	}
7069	tpg_desc_ptr1->support = 0;
7070	tpg_desc_ptr1->target_port_group[1] = 1;
7071	tpg_desc_ptr1->status = TPG_IMPLICIT;
7072	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7073
7074	if (single == 0) {
7075		tpg_desc_ptr2->support = 0;
7076		tpg_desc_ptr2->target_port_group[1] = 2;
7077		tpg_desc_ptr2->status = TPG_IMPLICIT;
7078		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7079
7080		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7081		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7082
7083		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7084		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7085	} else {
7086        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7087			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7088			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7089		} else {
7090			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7091			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7092		}
7093	}
7094
7095	ctsio->be_move_done = ctl_config_move_done;
7096
7097	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7098			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7099			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7100			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7101			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7102
7103	ctl_datamove((union ctl_io *)ctsio);
7104	return(retval);
7105}
7106
7107int
7108ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7109{
7110	struct scsi_per_res_in *cdb;
7111	int alloc_len, total_len = 0;
7112	/* struct scsi_per_res_in_rsrv in_data; */
7113	struct ctl_lun *lun;
7114	struct ctl_softc *softc;
7115
7116	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7117
7118	softc = control_softc;
7119
7120	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7121
7122	alloc_len = scsi_2btoul(cdb->length);
7123
7124	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7125
7126retry:
7127	mtx_lock(&softc->ctl_lock);
7128	switch (cdb->action) {
7129	case SPRI_RK: /* read keys */
7130		total_len = sizeof(struct scsi_per_res_in_keys) +
7131			lun->pr_key_count *
7132			sizeof(struct scsi_per_res_key);
7133		break;
7134	case SPRI_RR: /* read reservation */
7135		if (lun->flags & CTL_LUN_PR_RESERVED)
7136			total_len = sizeof(struct scsi_per_res_in_rsrv);
7137		else
7138			total_len = sizeof(struct scsi_per_res_in_header);
7139		break;
7140	case SPRI_RC: /* report capabilities */
7141		total_len = sizeof(struct scsi_per_res_cap);
7142		break;
7143	case SPRI_RS: /* read full status */
7144	default:
7145		mtx_unlock(&softc->ctl_lock);
7146		ctl_set_invalid_field(ctsio,
7147				      /*sks_valid*/ 1,
7148				      /*command*/ 1,
7149				      /*field*/ 1,
7150				      /*bit_valid*/ 1,
7151				      /*bit*/ 0);
7152		ctl_done((union ctl_io *)ctsio);
7153		return (CTL_RETVAL_COMPLETE);
7154		break; /* NOTREACHED */
7155	}
7156	mtx_unlock(&softc->ctl_lock);
7157
7158	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7159
7160	if (total_len < alloc_len) {
7161		ctsio->residual = alloc_len - total_len;
7162		ctsio->kern_data_len = total_len;
7163		ctsio->kern_total_len = total_len;
7164	} else {
7165		ctsio->residual = 0;
7166		ctsio->kern_data_len = alloc_len;
7167		ctsio->kern_total_len = alloc_len;
7168	}
7169
7170	ctsio->kern_data_resid = 0;
7171	ctsio->kern_rel_offset = 0;
7172	ctsio->kern_sg_entries = 0;
7173
7174	mtx_lock(&softc->ctl_lock);
7175	switch (cdb->action) {
7176	case SPRI_RK: { // read keys
7177        struct scsi_per_res_in_keys *res_keys;
7178		int i, key_count;
7179
7180		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7181
7182		/*
7183		 * We had to drop the lock to allocate our buffer, which
7184		 * leaves time for someone to come in with another
7185		 * persistent reservation.  (That is unlikely, though,
7186		 * since this should be the only persistent reservation
7187		 * command active right now.)
7188		 */
7189		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7190		    (lun->pr_key_count *
7191		     sizeof(struct scsi_per_res_key)))){
7192			mtx_unlock(&softc->ctl_lock);
7193			free(ctsio->kern_data_ptr, M_CTL);
7194			printf("%s: reservation length changed, retrying\n",
7195			       __func__);
7196			goto retry;
7197		}
7198
7199		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7200
7201		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7202			     lun->pr_key_count, res_keys->header.length);
7203
7204		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7205			if (!lun->per_res[i].registered)
7206				continue;
7207
7208			/*
7209			 * We used lun->pr_key_count to calculate the
7210			 * size to allocate.  If it turns out the number of
7211			 * initiators with the registered flag set is
7212			 * larger than that (i.e. they haven't been kept in
7213			 * sync), we've got a problem.
7214			 */
7215			if (key_count >= lun->pr_key_count) {
7216#ifdef NEEDTOPORT
7217				csevent_log(CSC_CTL | CSC_SHELF_SW |
7218					    CTL_PR_ERROR,
7219					    csevent_LogType_Fault,
7220					    csevent_AlertLevel_Yellow,
7221					    csevent_FRU_ShelfController,
7222					    csevent_FRU_Firmware,
7223				        csevent_FRU_Unknown,
7224					    "registered keys %d >= key "
7225					    "count %d", key_count,
7226					    lun->pr_key_count);
7227#endif
7228				key_count++;
7229				continue;
7230			}
7231			memcpy(res_keys->keys[key_count].key,
7232			       lun->per_res[i].res_key.key,
7233			       ctl_min(sizeof(res_keys->keys[key_count].key),
7234			       sizeof(lun->per_res[i].res_key)));
7235			key_count++;
7236		}
7237		break;
7238	}
7239	case SPRI_RR: { // read reservation
7240		struct scsi_per_res_in_rsrv *res;
7241		int tmp_len, header_only;
7242
7243		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7244
7245		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7246
7247		if (lun->flags & CTL_LUN_PR_RESERVED)
7248		{
7249			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7250			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7251				    res->header.length);
7252			header_only = 0;
7253		} else {
7254			tmp_len = sizeof(struct scsi_per_res_in_header);
7255			scsi_ulto4b(0, res->header.length);
7256			header_only = 1;
7257		}
7258
7259		/*
7260		 * We had to drop the lock to allocate our buffer, which
7261		 * leaves time for someone to come in with another
7262		 * persistent reservation.  (That is unlikely, though,
7263		 * since this should be the only persistent reservation
7264		 * command active right now.)
7265		 */
7266		if (tmp_len != total_len) {
7267			mtx_unlock(&softc->ctl_lock);
7268			free(ctsio->kern_data_ptr, M_CTL);
7269			printf("%s: reservation status changed, retrying\n",
7270			       __func__);
7271			goto retry;
7272		}
7273
7274		/*
7275		 * No reservation held, so we're done.
7276		 */
7277		if (header_only != 0)
7278			break;
7279
7280		/*
7281		 * If the registration is an All Registrants type, the key
7282		 * is 0, since it doesn't really matter.
7283		 */
7284		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7285			memcpy(res->data.reservation,
7286			       &lun->per_res[lun->pr_res_idx].res_key,
7287			       sizeof(struct scsi_per_res_key));
7288		}
7289		res->data.scopetype = lun->res_type;
7290		break;
7291	}
7292	case SPRI_RC:     //report capabilities
7293	{
7294		struct scsi_per_res_cap *res_cap;
7295		uint16_t type_mask;
7296
7297		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7298		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7299		res_cap->flags2 |= SPRI_TMV;
7300		type_mask = SPRI_TM_WR_EX_AR |
7301			    SPRI_TM_EX_AC_RO |
7302			    SPRI_TM_WR_EX_RO |
7303			    SPRI_TM_EX_AC |
7304			    SPRI_TM_WR_EX |
7305			    SPRI_TM_EX_AC_AR;
7306		scsi_ulto2b(type_mask, res_cap->type_mask);
7307		break;
7308	}
7309	case SPRI_RS: //read full status
7310	default:
7311		/*
7312		 * This is a bug, because we just checked for this above,
7313		 * and should have returned an error.
7314		 */
7315		panic("Invalid PR type %x", cdb->action);
7316		break; /* NOTREACHED */
7317	}
7318	mtx_unlock(&softc->ctl_lock);
7319
7320	ctsio->be_move_done = ctl_config_move_done;
7321
7322	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7323			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7324			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7325			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7326			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7327
7328	ctl_datamove((union ctl_io *)ctsio);
7329
7330	return (CTL_RETVAL_COMPLETE);
7331}
7332
7333/*
7334 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7335 * it should return.
7336 */
7337static int
7338ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7339		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7340		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7341		struct scsi_per_res_out_parms* param)
7342{
7343	union ctl_ha_msg persis_io;
7344	int retval, i;
7345	int isc_retval;
7346
7347	retval = 0;
7348
7349	if (sa_res_key == 0) {
7350		mtx_lock(&softc->ctl_lock);
7351		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7352			/* validate scope and type */
7353			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7354			     SPR_LU_SCOPE) {
7355				mtx_unlock(&softc->ctl_lock);
7356				ctl_set_invalid_field(/*ctsio*/ ctsio,
7357						      /*sks_valid*/ 1,
7358						      /*command*/ 1,
7359						      /*field*/ 2,
7360						      /*bit_valid*/ 1,
7361						      /*bit*/ 4);
7362				ctl_done((union ctl_io *)ctsio);
7363				return (1);
7364			}
7365
7366		        if (type>8 || type==2 || type==4 || type==0) {
7367				mtx_unlock(&softc->ctl_lock);
7368				ctl_set_invalid_field(/*ctsio*/ ctsio,
7369       	           				      /*sks_valid*/ 1,
7370						      /*command*/ 1,
7371						      /*field*/ 2,
7372						      /*bit_valid*/ 1,
7373						      /*bit*/ 0);
7374				ctl_done((union ctl_io *)ctsio);
7375				return (1);
7376		        }
7377
7378			/* temporarily unregister this nexus */
7379			lun->per_res[residx].registered = 0;
7380
7381			/*
7382			 * Unregister everybody else and build UA for
7383			 * them
7384			 */
7385			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7386				if (lun->per_res[i].registered == 0)
7387					continue;
7388
7389				if (!persis_offset
7390				 && i <CTL_MAX_INITIATORS)
7391					lun->pending_sense[i].ua_pending |=
7392						CTL_UA_REG_PREEMPT;
7393				else if (persis_offset
7394				      && i >= persis_offset)
7395					lun->pending_sense[i-persis_offset
7396						].ua_pending |=
7397						CTL_UA_REG_PREEMPT;
7398				lun->per_res[i].registered = 0;
7399				memset(&lun->per_res[i].res_key, 0,
7400				       sizeof(struct scsi_per_res_key));
7401			}
7402			lun->per_res[residx].registered = 1;
7403			lun->pr_key_count = 1;
7404			lun->res_type = type;
7405			if (lun->res_type != SPR_TYPE_WR_EX_AR
7406			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7407				lun->pr_res_idx = residx;
7408
7409			mtx_unlock(&softc->ctl_lock);
7410			/* send msg to other side */
7411			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7412			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7413			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7414			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7415			persis_io.pr.pr_info.res_type = type;
7416			memcpy(persis_io.pr.pr_info.sa_res_key,
7417			       param->serv_act_res_key,
7418			       sizeof(param->serv_act_res_key));
7419			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7420			     &persis_io, sizeof(persis_io), 0)) >
7421			     CTL_HA_STATUS_SUCCESS) {
7422				printf("CTL:Persis Out error returned "
7423				       "from ctl_ha_msg_send %d\n",
7424				       isc_retval);
7425			}
7426		} else {
7427			/* not all registrants */
7428			mtx_unlock(&softc->ctl_lock);
7429			free(ctsio->kern_data_ptr, M_CTL);
7430			ctl_set_invalid_field(ctsio,
7431					      /*sks_valid*/ 1,
7432					      /*command*/ 0,
7433					      /*field*/ 8,
7434					      /*bit_valid*/ 0,
7435					      /*bit*/ 0);
7436			ctl_done((union ctl_io *)ctsio);
7437			return (1);
7438		}
7439	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7440		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7441		int found = 0;
7442
7443		mtx_lock(&softc->ctl_lock);
7444		if (res_key == sa_res_key) {
7445			/* special case */
7446			/*
7447			 * The spec implies this is not good but doesn't
7448			 * say what to do. There are two choices either
7449			 * generate a res conflict or check condition
7450			 * with illegal field in parameter data. Since
7451			 * that is what is done when the sa_res_key is
7452			 * zero I'll take that approach since this has
7453			 * to do with the sa_res_key.
7454			 */
7455			mtx_unlock(&softc->ctl_lock);
7456			free(ctsio->kern_data_ptr, M_CTL);
7457			ctl_set_invalid_field(ctsio,
7458					      /*sks_valid*/ 1,
7459					      /*command*/ 0,
7460					      /*field*/ 8,
7461					      /*bit_valid*/ 0,
7462					      /*bit*/ 0);
7463			ctl_done((union ctl_io *)ctsio);
7464			return (1);
7465		}
7466
7467		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7468			if (lun->per_res[i].registered
7469			 && memcmp(param->serv_act_res_key,
7470			    lun->per_res[i].res_key.key,
7471			    sizeof(struct scsi_per_res_key)) != 0)
7472				continue;
7473
7474			found = 1;
7475			lun->per_res[i].registered = 0;
7476			memset(&lun->per_res[i].res_key, 0,
7477			       sizeof(struct scsi_per_res_key));
7478			lun->pr_key_count--;
7479
7480			if (!persis_offset
7481			 && i < CTL_MAX_INITIATORS)
7482				lun->pending_sense[i].ua_pending |=
7483					CTL_UA_REG_PREEMPT;
7484			else if (persis_offset
7485			      && i >= persis_offset)
7486				lun->pending_sense[i-persis_offset].ua_pending|=
7487					CTL_UA_REG_PREEMPT;
7488		}
7489		mtx_unlock(&softc->ctl_lock);
7490		if (!found) {
7491			free(ctsio->kern_data_ptr, M_CTL);
7492			ctl_set_reservation_conflict(ctsio);
7493			ctl_done((union ctl_io *)ctsio);
7494			return (CTL_RETVAL_COMPLETE);
7495		}
7496		/* send msg to other side */
7497		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7498		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7499		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7500		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7501		persis_io.pr.pr_info.res_type = type;
7502		memcpy(persis_io.pr.pr_info.sa_res_key,
7503		       param->serv_act_res_key,
7504		       sizeof(param->serv_act_res_key));
7505		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7506		     &persis_io, sizeof(persis_io), 0)) >
7507		     CTL_HA_STATUS_SUCCESS) {
7508			printf("CTL:Persis Out error returned from "
7509			       "ctl_ha_msg_send %d\n", isc_retval);
7510		}
7511	} else {
7512		/* Reserved but not all registrants */
7513		/* sa_res_key is res holder */
7514		if (memcmp(param->serv_act_res_key,
7515                   lun->per_res[lun->pr_res_idx].res_key.key,
7516                   sizeof(struct scsi_per_res_key)) == 0) {
7517			/* validate scope and type */
7518			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7519			     SPR_LU_SCOPE) {
7520				ctl_set_invalid_field(/*ctsio*/ ctsio,
7521						      /*sks_valid*/ 1,
7522						      /*command*/ 1,
7523						      /*field*/ 2,
7524						      /*bit_valid*/ 1,
7525						      /*bit*/ 4);
7526				ctl_done((union ctl_io *)ctsio);
7527				return (1);
7528			}
7529
7530			if (type>8 || type==2 || type==4 || type==0) {
7531				ctl_set_invalid_field(/*ctsio*/ ctsio,
7532						      /*sks_valid*/ 1,
7533						      /*command*/ 1,
7534						      /*field*/ 2,
7535						      /*bit_valid*/ 1,
7536						      /*bit*/ 0);
7537				ctl_done((union ctl_io *)ctsio);
7538				return (1);
7539			}
7540
7541			/*
7542			 * Do the following:
7543			 * if sa_res_key != res_key remove all
7544			 * registrants w/sa_res_key and generate UA
7545			 * for these registrants(Registrations
7546			 * Preempted) if it wasn't an exclusive
7547			 * reservation generate UA(Reservations
7548			 * Preempted) for all other registered nexuses
7549			 * if the type has changed. Establish the new
7550			 * reservation and holder. If res_key and
7551			 * sa_res_key are the same do the above
7552			 * except don't unregister the res holder.
7553			 */
7554
7555			/*
7556			 * Temporarily unregister so it won't get
7557			 * removed or UA generated
7558			 */
7559			lun->per_res[residx].registered = 0;
7560			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7561				if (lun->per_res[i].registered == 0)
7562					continue;
7563
7564				if (memcmp(param->serv_act_res_key,
7565				    lun->per_res[i].res_key.key,
7566				    sizeof(struct scsi_per_res_key)) == 0) {
7567					lun->per_res[i].registered = 0;
7568					memset(&lun->per_res[i].res_key,
7569					       0,
7570					       sizeof(struct scsi_per_res_key));
7571					lun->pr_key_count--;
7572
7573					if (!persis_offset
7574					 && i < CTL_MAX_INITIATORS)
7575						lun->pending_sense[i
7576							].ua_pending |=
7577							CTL_UA_REG_PREEMPT;
7578					else if (persis_offset
7579					      && i >= persis_offset)
7580						lun->pending_sense[
7581						  i-persis_offset].ua_pending |=
7582						  CTL_UA_REG_PREEMPT;
7583				} else if (type != lun->res_type
7584					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7585					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7586						if (!persis_offset
7587						 && i < CTL_MAX_INITIATORS)
7588							lun->pending_sense[i
7589							].ua_pending |=
7590							CTL_UA_RES_RELEASE;
7591						else if (persis_offset
7592						      && i >= persis_offset)
7593							lun->pending_sense[
7594							i-persis_offset
7595							].ua_pending |=
7596							CTL_UA_RES_RELEASE;
7597				}
7598			}
7599			lun->per_res[residx].registered = 1;
7600			lun->res_type = type;
7601			if (lun->res_type != SPR_TYPE_WR_EX_AR
7602			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7603				lun->pr_res_idx = residx;
7604			else
7605				lun->pr_res_idx =
7606					CTL_PR_ALL_REGISTRANTS;
7607
7608			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7609			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7610			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7611			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7612			persis_io.pr.pr_info.res_type = type;
7613			memcpy(persis_io.pr.pr_info.sa_res_key,
7614			       param->serv_act_res_key,
7615			       sizeof(param->serv_act_res_key));
7616			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7617			     &persis_io, sizeof(persis_io), 0)) >
7618			     CTL_HA_STATUS_SUCCESS) {
7619				printf("CTL:Persis Out error returned "
7620				       "from ctl_ha_msg_send %d\n",
7621				       isc_retval);
7622			}
7623		} else {
7624			/*
7625			 * sa_res_key is not the res holder just
7626			 * remove registrants
7627			 */
7628			int found=0;
7629			mtx_lock(&softc->ctl_lock);
7630
7631			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7632				if (memcmp(param->serv_act_res_key,
7633				    lun->per_res[i].res_key.key,
7634				    sizeof(struct scsi_per_res_key)) != 0)
7635					continue;
7636
7637				found = 1;
7638				lun->per_res[i].registered = 0;
7639				memset(&lun->per_res[i].res_key, 0,
7640				       sizeof(struct scsi_per_res_key));
7641				lun->pr_key_count--;
7642
7643				if (!persis_offset
7644				 && i < CTL_MAX_INITIATORS)
7645					lun->pending_sense[i].ua_pending |=
7646						CTL_UA_REG_PREEMPT;
7647				else if (persis_offset
7648				      && i >= persis_offset)
7649					lun->pending_sense[
7650						i-persis_offset].ua_pending |=
7651						CTL_UA_REG_PREEMPT;
7652			}
7653
7654			if (!found) {
7655				mtx_unlock(&softc->ctl_lock);
7656				free(ctsio->kern_data_ptr, M_CTL);
7657				ctl_set_reservation_conflict(ctsio);
7658				ctl_done((union ctl_io *)ctsio);
7659		        	return (1);
7660			}
7661			mtx_unlock(&softc->ctl_lock);
7662			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7663			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7664			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7665			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7666			persis_io.pr.pr_info.res_type = type;
7667			memcpy(persis_io.pr.pr_info.sa_res_key,
7668			       param->serv_act_res_key,
7669			       sizeof(param->serv_act_res_key));
7670			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7671			     &persis_io, sizeof(persis_io), 0)) >
7672			     CTL_HA_STATUS_SUCCESS) {
7673				printf("CTL:Persis Out error returned "
7674				       "from ctl_ha_msg_send %d\n",
7675				isc_retval);
7676			}
7677		}
7678	}
7679
7680	lun->PRGeneration++;
7681
7682	return (retval);
7683}
7684
7685static void
7686ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7687{
7688	int i;
7689
7690	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7691	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7692	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7693		   msg->pr.pr_info.sa_res_key,
7694		   sizeof(struct scsi_per_res_key)) != 0) {
7695		uint64_t sa_res_key;
7696		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7697
7698		if (sa_res_key == 0) {
7699			/* temporarily unregister this nexus */
7700			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7701
7702			/*
7703			 * Unregister everybody else and build UA for
7704			 * them
7705			 */
7706			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7707				if (lun->per_res[i].registered == 0)
7708					continue;
7709
7710				if (!persis_offset
7711				 && i < CTL_MAX_INITIATORS)
7712					lun->pending_sense[i].ua_pending |=
7713						CTL_UA_REG_PREEMPT;
7714				else if (persis_offset && i >= persis_offset)
7715					lun->pending_sense[i -
7716						persis_offset].ua_pending |=
7717						CTL_UA_REG_PREEMPT;
7718				lun->per_res[i].registered = 0;
7719				memset(&lun->per_res[i].res_key, 0,
7720				       sizeof(struct scsi_per_res_key));
7721			}
7722
7723			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7724			lun->pr_key_count = 1;
7725			lun->res_type = msg->pr.pr_info.res_type;
7726			if (lun->res_type != SPR_TYPE_WR_EX_AR
7727			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7728				lun->pr_res_idx = msg->pr.pr_info.residx;
7729		} else {
7730		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7731				if (memcmp(msg->pr.pr_info.sa_res_key,
7732		                   lun->per_res[i].res_key.key,
7733		                   sizeof(struct scsi_per_res_key)) != 0)
7734					continue;
7735
7736				lun->per_res[i].registered = 0;
7737				memset(&lun->per_res[i].res_key, 0,
7738				       sizeof(struct scsi_per_res_key));
7739				lun->pr_key_count--;
7740
7741				if (!persis_offset
7742				 && i < persis_offset)
7743					lun->pending_sense[i].ua_pending |=
7744						CTL_UA_REG_PREEMPT;
7745				else if (persis_offset
7746				      && i >= persis_offset)
7747					lun->pending_sense[i -
7748						persis_offset].ua_pending |=
7749						CTL_UA_REG_PREEMPT;
7750			}
7751		}
7752	} else {
7753		/*
7754		 * Temporarily unregister so it won't get removed
7755		 * or UA generated
7756		 */
7757		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7758		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7759			if (lun->per_res[i].registered == 0)
7760				continue;
7761
7762			if (memcmp(msg->pr.pr_info.sa_res_key,
7763	                   lun->per_res[i].res_key.key,
7764	                   sizeof(struct scsi_per_res_key)) == 0) {
7765				lun->per_res[i].registered = 0;
7766				memset(&lun->per_res[i].res_key, 0,
7767				       sizeof(struct scsi_per_res_key));
7768				lun->pr_key_count--;
7769				if (!persis_offset
7770				 && i < CTL_MAX_INITIATORS)
7771					lun->pending_sense[i].ua_pending |=
7772						CTL_UA_REG_PREEMPT;
7773				else if (persis_offset
7774				      && i >= persis_offset)
7775					lun->pending_sense[i -
7776						persis_offset].ua_pending |=
7777						CTL_UA_REG_PREEMPT;
7778			} else if (msg->pr.pr_info.res_type != lun->res_type
7779				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7780				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7781					if (!persis_offset
7782					 && i < persis_offset)
7783						lun->pending_sense[i
7784							].ua_pending |=
7785							CTL_UA_RES_RELEASE;
7786					else if (persis_offset
7787					      && i >= persis_offset)
7788					lun->pending_sense[i -
7789						persis_offset].ua_pending |=
7790						CTL_UA_RES_RELEASE;
7791			}
7792		}
7793		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7794		lun->res_type = msg->pr.pr_info.res_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 = msg->pr.pr_info.residx;
7798		else
7799			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7800	}
7801	lun->PRGeneration++;
7802
7803}
7804
7805
7806int
7807ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7808{
7809	int retval;
7810	int isc_retval;
7811	u_int32_t param_len;
7812	struct scsi_per_res_out *cdb;
7813	struct ctl_lun *lun;
7814	struct scsi_per_res_out_parms* param;
7815	struct ctl_softc *softc;
7816	uint32_t residx;
7817	uint64_t res_key, sa_res_key;
7818	uint8_t type;
7819	union ctl_ha_msg persis_io;
7820	int    i;
7821
7822	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
7823
7824	retval = CTL_RETVAL_COMPLETE;
7825
7826	softc = control_softc;
7827
7828	cdb = (struct scsi_per_res_out *)ctsio->cdb;
7829	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7830
7831	/*
7832	 * We only support whole-LUN scope.  The scope & type are ignored for
7833	 * register, register and ignore existing key and clear.
7834	 * We sometimes ignore scope and type on preempts too!!
7835	 * Verify reservation type here as well.
7836	 */
7837	type = cdb->scope_type & SPR_TYPE_MASK;
7838	if ((cdb->action == SPRO_RESERVE)
7839	 || (cdb->action == SPRO_RELEASE)) {
7840		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
7841			ctl_set_invalid_field(/*ctsio*/ ctsio,
7842					      /*sks_valid*/ 1,
7843					      /*command*/ 1,
7844					      /*field*/ 2,
7845					      /*bit_valid*/ 1,
7846					      /*bit*/ 4);
7847			ctl_done((union ctl_io *)ctsio);
7848			return (CTL_RETVAL_COMPLETE);
7849		}
7850
7851		if (type>8 || type==2 || type==4 || type==0) {
7852			ctl_set_invalid_field(/*ctsio*/ ctsio,
7853					      /*sks_valid*/ 1,
7854					      /*command*/ 1,
7855					      /*field*/ 2,
7856					      /*bit_valid*/ 1,
7857					      /*bit*/ 0);
7858			ctl_done((union ctl_io *)ctsio);
7859			return (CTL_RETVAL_COMPLETE);
7860		}
7861	}
7862
7863	switch (cdb->action & SPRO_ACTION_MASK) {
7864	case SPRO_REGISTER:
7865	case SPRO_RESERVE:
7866	case SPRO_RELEASE:
7867	case SPRO_CLEAR:
7868	case SPRO_PREEMPT:
7869	case SPRO_REG_IGNO:
7870		break;
7871	case SPRO_REG_MOVE:
7872	case SPRO_PRE_ABO:
7873	default:
7874		ctl_set_invalid_field(/*ctsio*/ ctsio,
7875				      /*sks_valid*/ 1,
7876				      /*command*/ 1,
7877				      /*field*/ 1,
7878				      /*bit_valid*/ 1,
7879				      /*bit*/ 0);
7880		ctl_done((union ctl_io *)ctsio);
7881		return (CTL_RETVAL_COMPLETE);
7882		break; /* NOTREACHED */
7883	}
7884
7885	param_len = scsi_4btoul(cdb->length);
7886
7887	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
7888		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
7889		ctsio->kern_data_len = param_len;
7890		ctsio->kern_total_len = param_len;
7891		ctsio->kern_data_resid = 0;
7892		ctsio->kern_rel_offset = 0;
7893		ctsio->kern_sg_entries = 0;
7894		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7895		ctsio->be_move_done = ctl_config_move_done;
7896		ctl_datamove((union ctl_io *)ctsio);
7897
7898		return (CTL_RETVAL_COMPLETE);
7899	}
7900
7901	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
7902
7903	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
7904	res_key = scsi_8btou64(param->res_key.key);
7905	sa_res_key = scsi_8btou64(param->serv_act_res_key);
7906
7907	/*
7908	 * Validate the reservation key here except for SPRO_REG_IGNO
7909	 * This must be done for all other service actions
7910	 */
7911	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
7912		mtx_lock(&softc->ctl_lock);
7913		if (lun->per_res[residx].registered) {
7914		    if (memcmp(param->res_key.key,
7915			       lun->per_res[residx].res_key.key,
7916			       ctl_min(sizeof(param->res_key),
7917			       sizeof(lun->per_res[residx].res_key))) != 0) {
7918				/*
7919				 * The current key passed in doesn't match
7920				 * the one the initiator previously
7921				 * registered.
7922				 */
7923				mtx_unlock(&softc->ctl_lock);
7924				free(ctsio->kern_data_ptr, M_CTL);
7925				ctl_set_reservation_conflict(ctsio);
7926				ctl_done((union ctl_io *)ctsio);
7927				return (CTL_RETVAL_COMPLETE);
7928			}
7929		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
7930			/*
7931			 * We are not registered
7932			 */
7933			mtx_unlock(&softc->ctl_lock);
7934			free(ctsio->kern_data_ptr, M_CTL);
7935			ctl_set_reservation_conflict(ctsio);
7936			ctl_done((union ctl_io *)ctsio);
7937			return (CTL_RETVAL_COMPLETE);
7938		} else if (res_key != 0) {
7939			/*
7940			 * We are not registered and trying to register but
7941			 * the register key isn't zero.
7942			 */
7943			mtx_unlock(&softc->ctl_lock);
7944			free(ctsio->kern_data_ptr, M_CTL);
7945			ctl_set_reservation_conflict(ctsio);
7946			ctl_done((union ctl_io *)ctsio);
7947			return (CTL_RETVAL_COMPLETE);
7948		}
7949		mtx_unlock(&softc->ctl_lock);
7950	}
7951
7952	switch (cdb->action & SPRO_ACTION_MASK) {
7953	case SPRO_REGISTER:
7954	case SPRO_REG_IGNO: {
7955
7956#if 0
7957		printf("Registration received\n");
7958#endif
7959
7960		/*
7961		 * We don't support any of these options, as we report in
7962		 * the read capabilities request (see
7963		 * ctl_persistent_reserve_in(), above).
7964		 */
7965		if ((param->flags & SPR_SPEC_I_PT)
7966		 || (param->flags & SPR_ALL_TG_PT)
7967		 || (param->flags & SPR_APTPL)) {
7968			int bit_ptr;
7969
7970			if (param->flags & SPR_APTPL)
7971				bit_ptr = 0;
7972			else if (param->flags & SPR_ALL_TG_PT)
7973				bit_ptr = 2;
7974			else /* SPR_SPEC_I_PT */
7975				bit_ptr = 3;
7976
7977			free(ctsio->kern_data_ptr, M_CTL);
7978			ctl_set_invalid_field(ctsio,
7979					      /*sks_valid*/ 1,
7980					      /*command*/ 0,
7981					      /*field*/ 20,
7982					      /*bit_valid*/ 1,
7983					      /*bit*/ bit_ptr);
7984			ctl_done((union ctl_io *)ctsio);
7985			return (CTL_RETVAL_COMPLETE);
7986		}
7987
7988		mtx_lock(&softc->ctl_lock);
7989
7990		/*
7991		 * The initiator wants to clear the
7992		 * key/unregister.
7993		 */
7994		if (sa_res_key == 0) {
7995			if ((res_key == 0
7996			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
7997			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
7998			  && !lun->per_res[residx].registered)) {
7999				mtx_unlock(&softc->ctl_lock);
8000				goto done;
8001			}
8002
8003			lun->per_res[residx].registered = 0;
8004			memset(&lun->per_res[residx].res_key,
8005			       0, sizeof(lun->per_res[residx].res_key));
8006			lun->pr_key_count--;
8007
8008			if (residx == lun->pr_res_idx) {
8009				lun->flags &= ~CTL_LUN_PR_RESERVED;
8010				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8011
8012				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8013				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8014				 && lun->pr_key_count) {
8015					/*
8016					 * If the reservation is a registrants
8017					 * only type we need to generate a UA
8018					 * for other registered inits.  The
8019					 * sense code should be RESERVATIONS
8020					 * RELEASED
8021					 */
8022
8023					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8024						if (lun->per_res[
8025						    i+persis_offset].registered
8026						    == 0)
8027							continue;
8028						lun->pending_sense[i
8029							].ua_pending |=
8030							CTL_UA_RES_RELEASE;
8031					}
8032				}
8033				lun->res_type = 0;
8034			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8035				if (lun->pr_key_count==0) {
8036					lun->flags &= ~CTL_LUN_PR_RESERVED;
8037					lun->res_type = 0;
8038					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8039				}
8040			}
8041			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8042			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8043			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8044			persis_io.pr.pr_info.residx = residx;
8045			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8046			     &persis_io, sizeof(persis_io), 0 )) >
8047			     CTL_HA_STATUS_SUCCESS) {
8048				printf("CTL:Persis Out error returned from "
8049				       "ctl_ha_msg_send %d\n", isc_retval);
8050			}
8051			mtx_unlock(&softc->ctl_lock);
8052		} else /* sa_res_key != 0 */ {
8053
8054			/*
8055			 * If we aren't registered currently then increment
8056			 * the key count and set the registered flag.
8057			 */
8058			if (!lun->per_res[residx].registered) {
8059				lun->pr_key_count++;
8060				lun->per_res[residx].registered = 1;
8061			}
8062
8063			memcpy(&lun->per_res[residx].res_key,
8064			       param->serv_act_res_key,
8065			       ctl_min(sizeof(param->serv_act_res_key),
8066			       sizeof(lun->per_res[residx].res_key)));
8067
8068			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8069			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8070			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8071			persis_io.pr.pr_info.residx = residx;
8072			memcpy(persis_io.pr.pr_info.sa_res_key,
8073			       param->serv_act_res_key,
8074			       sizeof(param->serv_act_res_key));
8075			mtx_unlock(&softc->ctl_lock);
8076			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8077			     &persis_io, sizeof(persis_io), 0)) >
8078			     CTL_HA_STATUS_SUCCESS) {
8079				printf("CTL:Persis Out error returned from "
8080				       "ctl_ha_msg_send %d\n", isc_retval);
8081			}
8082		}
8083		lun->PRGeneration++;
8084
8085		break;
8086	}
8087	case SPRO_RESERVE:
8088#if 0
8089                printf("Reserve executed type %d\n", type);
8090#endif
8091		mtx_lock(&softc->ctl_lock);
8092		if (lun->flags & CTL_LUN_PR_RESERVED) {
8093			/*
8094			 * if this isn't the reservation holder and it's
8095			 * not a "all registrants" type or if the type is
8096			 * different then we have a conflict
8097			 */
8098			if ((lun->pr_res_idx != residx
8099			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8100			 || lun->res_type != type) {
8101				mtx_unlock(&softc->ctl_lock);
8102				free(ctsio->kern_data_ptr, M_CTL);
8103				ctl_set_reservation_conflict(ctsio);
8104				ctl_done((union ctl_io *)ctsio);
8105				return (CTL_RETVAL_COMPLETE);
8106			}
8107		} else /* create a reservation */ {
8108			/*
8109			 * If it's not an "all registrants" type record
8110			 * reservation holder
8111			 */
8112			if (type != SPR_TYPE_WR_EX_AR
8113			 && type != SPR_TYPE_EX_AC_AR)
8114				lun->pr_res_idx = residx; /* Res holder */
8115			else
8116				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8117
8118			lun->flags |= CTL_LUN_PR_RESERVED;
8119			lun->res_type = type;
8120
8121			mtx_unlock(&softc->ctl_lock);
8122
8123			/* send msg to other side */
8124			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8125			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8126			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8127			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8128			persis_io.pr.pr_info.res_type = type;
8129			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8130			     &persis_io, sizeof(persis_io), 0)) >
8131			     CTL_HA_STATUS_SUCCESS) {
8132				printf("CTL:Persis Out error returned from "
8133				       "ctl_ha_msg_send %d\n", isc_retval);
8134			}
8135		}
8136		break;
8137
8138	case SPRO_RELEASE:
8139		mtx_lock(&softc->ctl_lock);
8140		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8141			/* No reservation exists return good status */
8142			mtx_unlock(&softc->ctl_lock);
8143			goto done;
8144		}
8145		/*
8146		 * Is this nexus a reservation holder?
8147		 */
8148		if (lun->pr_res_idx != residx
8149		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8150			/*
8151			 * not a res holder return good status but
8152			 * do nothing
8153			 */
8154			mtx_unlock(&softc->ctl_lock);
8155			goto done;
8156		}
8157
8158		if (lun->res_type != type) {
8159			mtx_unlock(&softc->ctl_lock);
8160			free(ctsio->kern_data_ptr, M_CTL);
8161			ctl_set_illegal_pr_release(ctsio);
8162			ctl_done((union ctl_io *)ctsio);
8163			return (CTL_RETVAL_COMPLETE);
8164		}
8165
8166		/* okay to release */
8167		lun->flags &= ~CTL_LUN_PR_RESERVED;
8168		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8169		lun->res_type = 0;
8170
8171		/*
8172		 * if this isn't an exclusive access
8173		 * res generate UA for all other
8174		 * registrants.
8175		 */
8176		if (type != SPR_TYPE_EX_AC
8177		 && type != SPR_TYPE_WR_EX) {
8178			/*
8179			 * temporarily unregister so we don't generate UA
8180			 */
8181			lun->per_res[residx].registered = 0;
8182
8183			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8184				if (lun->per_res[i+persis_offset].registered
8185				    == 0)
8186					continue;
8187				lun->pending_sense[i].ua_pending |=
8188					CTL_UA_RES_RELEASE;
8189			}
8190
8191			lun->per_res[residx].registered = 1;
8192		}
8193		mtx_unlock(&softc->ctl_lock);
8194		/* Send msg to other side */
8195		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8196		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8197		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8198		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8199		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8200			printf("CTL:Persis Out error returned from "
8201			       "ctl_ha_msg_send %d\n", isc_retval);
8202		}
8203		break;
8204
8205	case SPRO_CLEAR:
8206		/* send msg to other side */
8207
8208		mtx_lock(&softc->ctl_lock);
8209		lun->flags &= ~CTL_LUN_PR_RESERVED;
8210		lun->res_type = 0;
8211		lun->pr_key_count = 0;
8212		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8213
8214
8215		memset(&lun->per_res[residx].res_key,
8216		       0, sizeof(lun->per_res[residx].res_key));
8217		lun->per_res[residx].registered = 0;
8218
8219		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8220			if (lun->per_res[i].registered) {
8221				if (!persis_offset && i < CTL_MAX_INITIATORS)
8222					lun->pending_sense[i].ua_pending |=
8223						CTL_UA_RES_PREEMPT;
8224				else if (persis_offset && i >= persis_offset)
8225					lun->pending_sense[i-persis_offset
8226					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8227
8228				memset(&lun->per_res[i].res_key,
8229				       0, sizeof(struct scsi_per_res_key));
8230				lun->per_res[i].registered = 0;
8231			}
8232		lun->PRGeneration++;
8233		mtx_unlock(&softc->ctl_lock);
8234		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8235		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8236		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8237		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8238		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8239			printf("CTL:Persis Out error returned from "
8240			       "ctl_ha_msg_send %d\n", isc_retval);
8241		}
8242		break;
8243
8244	case SPRO_PREEMPT: {
8245		int nretval;
8246
8247		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8248					  residx, ctsio, cdb, param);
8249		if (nretval != 0)
8250			return (CTL_RETVAL_COMPLETE);
8251		break;
8252	}
8253	case SPRO_REG_MOVE:
8254	case SPRO_PRE_ABO:
8255	default:
8256		free(ctsio->kern_data_ptr, M_CTL);
8257		ctl_set_invalid_field(/*ctsio*/ ctsio,
8258				      /*sks_valid*/ 1,
8259				      /*command*/ 1,
8260				      /*field*/ 1,
8261				      /*bit_valid*/ 1,
8262				      /*bit*/ 0);
8263		ctl_done((union ctl_io *)ctsio);
8264		return (CTL_RETVAL_COMPLETE);
8265		break; /* NOTREACHED */
8266	}
8267
8268done:
8269	free(ctsio->kern_data_ptr, M_CTL);
8270	ctl_set_success(ctsio);
8271	ctl_done((union ctl_io *)ctsio);
8272
8273	return (retval);
8274}
8275
8276/*
8277 * This routine is for handling a message from the other SC pertaining to
8278 * persistent reserve out. All the error checking will have been done
8279 * so only perorming the action need be done here to keep the two
8280 * in sync.
8281 */
8282static void
8283ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8284{
8285	struct ctl_lun *lun;
8286	struct ctl_softc *softc;
8287	int i;
8288	uint32_t targ_lun;
8289
8290	softc = control_softc;
8291
8292	mtx_lock(&softc->ctl_lock);
8293
8294	targ_lun = msg->hdr.nexus.targ_lun;
8295	if (msg->hdr.nexus.lun_map_fn != NULL)
8296		targ_lun = msg->hdr.nexus.lun_map_fn(msg->hdr.nexus.lun_map_arg, targ_lun);
8297	lun = softc->ctl_luns[targ_lun];
8298	switch(msg->pr.pr_info.action) {
8299	case CTL_PR_REG_KEY:
8300		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8301			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8302			lun->pr_key_count++;
8303		}
8304		lun->PRGeneration++;
8305		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8306		       msg->pr.pr_info.sa_res_key,
8307		       sizeof(struct scsi_per_res_key));
8308		break;
8309
8310	case CTL_PR_UNREG_KEY:
8311		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8312		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8313		       0, sizeof(struct scsi_per_res_key));
8314		lun->pr_key_count--;
8315
8316		/* XXX Need to see if the reservation has been released */
8317		/* if so do we need to generate UA? */
8318		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8319			lun->flags &= ~CTL_LUN_PR_RESERVED;
8320			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8321
8322			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8323			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8324			 && lun->pr_key_count) {
8325				/*
8326				 * If the reservation is a registrants
8327				 * only type we need to generate a UA
8328				 * for other registered inits.  The
8329				 * sense code should be RESERVATIONS
8330				 * RELEASED
8331				 */
8332
8333				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8334					if (lun->per_res[i+
8335					    persis_offset].registered == 0)
8336						continue;
8337
8338					lun->pending_sense[i
8339						].ua_pending |=
8340						CTL_UA_RES_RELEASE;
8341				}
8342			}
8343			lun->res_type = 0;
8344		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8345			if (lun->pr_key_count==0) {
8346				lun->flags &= ~CTL_LUN_PR_RESERVED;
8347				lun->res_type = 0;
8348				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8349			}
8350		}
8351		lun->PRGeneration++;
8352		break;
8353
8354	case CTL_PR_RESERVE:
8355		lun->flags |= CTL_LUN_PR_RESERVED;
8356		lun->res_type = msg->pr.pr_info.res_type;
8357		lun->pr_res_idx = msg->pr.pr_info.residx;
8358
8359		break;
8360
8361	case CTL_PR_RELEASE:
8362		/*
8363		 * if this isn't an exclusive access res generate UA for all
8364		 * other registrants.
8365		 */
8366		if (lun->res_type != SPR_TYPE_EX_AC
8367		 && lun->res_type != SPR_TYPE_WR_EX) {
8368			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8369				if (lun->per_res[i+persis_offset].registered)
8370					lun->pending_sense[i].ua_pending |=
8371						CTL_UA_RES_RELEASE;
8372		}
8373
8374		lun->flags &= ~CTL_LUN_PR_RESERVED;
8375		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8376		lun->res_type = 0;
8377		break;
8378
8379	case CTL_PR_PREEMPT:
8380		ctl_pro_preempt_other(lun, msg);
8381		break;
8382	case CTL_PR_CLEAR:
8383		lun->flags &= ~CTL_LUN_PR_RESERVED;
8384		lun->res_type = 0;
8385		lun->pr_key_count = 0;
8386		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8387
8388		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8389			if (lun->per_res[i].registered == 0)
8390				continue;
8391			if (!persis_offset
8392			 && i < CTL_MAX_INITIATORS)
8393				lun->pending_sense[i].ua_pending |=
8394					CTL_UA_RES_PREEMPT;
8395			else if (persis_offset
8396			      && i >= persis_offset)
8397   				lun->pending_sense[i-persis_offset].ua_pending|=
8398					CTL_UA_RES_PREEMPT;
8399			memset(&lun->per_res[i].res_key, 0,
8400			       sizeof(struct scsi_per_res_key));
8401			lun->per_res[i].registered = 0;
8402		}
8403		lun->PRGeneration++;
8404		break;
8405	}
8406
8407	mtx_unlock(&softc->ctl_lock);
8408}
8409
8410int
8411ctl_read_write(struct ctl_scsiio *ctsio)
8412{
8413	struct ctl_lun *lun;
8414	struct ctl_lba_len lbalen;
8415	uint64_t lba;
8416	uint32_t num_blocks;
8417	int reladdr, fua, dpo, ebp;
8418	int retval;
8419	int isread;
8420
8421	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8422
8423	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8424
8425	reladdr = 0;
8426	fua = 0;
8427	dpo = 0;
8428	ebp = 0;
8429
8430	retval = CTL_RETVAL_COMPLETE;
8431
8432	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8433	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8434	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8435		uint32_t residx;
8436
8437		/*
8438		 * XXX KDM need a lock here.
8439		 */
8440		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8441		if ((lun->res_type == SPR_TYPE_EX_AC
8442		  && residx != lun->pr_res_idx)
8443		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8444		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8445		  && !lun->per_res[residx].registered)) {
8446			ctl_set_reservation_conflict(ctsio);
8447			ctl_done((union ctl_io *)ctsio);
8448			return (CTL_RETVAL_COMPLETE);
8449	        }
8450	}
8451
8452	switch (ctsio->cdb[0]) {
8453	case READ_6:
8454	case WRITE_6: {
8455		struct scsi_rw_6 *cdb;
8456
8457		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8458
8459		lba = scsi_3btoul(cdb->addr);
8460		/* only 5 bits are valid in the most significant address byte */
8461		lba &= 0x1fffff;
8462		num_blocks = cdb->length;
8463		/*
8464		 * This is correct according to SBC-2.
8465		 */
8466		if (num_blocks == 0)
8467			num_blocks = 256;
8468		break;
8469	}
8470	case READ_10:
8471	case WRITE_10: {
8472		struct scsi_rw_10 *cdb;
8473
8474		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8475
8476		if (cdb->byte2 & SRW10_RELADDR)
8477			reladdr = 1;
8478		if (cdb->byte2 & SRW10_FUA)
8479			fua = 1;
8480		if (cdb->byte2 & SRW10_DPO)
8481			dpo = 1;
8482
8483		if ((cdb->opcode == WRITE_10)
8484		 && (cdb->byte2 & SRW10_EBP))
8485			ebp = 1;
8486
8487		lba = scsi_4btoul(cdb->addr);
8488		num_blocks = scsi_2btoul(cdb->length);
8489		break;
8490	}
8491	case WRITE_VERIFY_10: {
8492		struct scsi_write_verify_10 *cdb;
8493
8494		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8495
8496		/*
8497		 * XXX KDM we should do actual write verify support at some
8498		 * point.  This is obviously fake, we're just translating
8499		 * things to a write.  So we don't even bother checking the
8500		 * BYTCHK field, since we don't do any verification.  If
8501		 * the user asks for it, we'll just pretend we did it.
8502		 */
8503		if (cdb->byte2 & SWV_DPO)
8504			dpo = 1;
8505
8506		lba = scsi_4btoul(cdb->addr);
8507		num_blocks = scsi_2btoul(cdb->length);
8508		break;
8509	}
8510	case READ_12:
8511	case WRITE_12: {
8512		struct scsi_rw_12 *cdb;
8513
8514		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8515
8516		if (cdb->byte2 & SRW12_RELADDR)
8517			reladdr = 1;
8518		if (cdb->byte2 & SRW12_FUA)
8519			fua = 1;
8520		if (cdb->byte2 & SRW12_DPO)
8521			dpo = 1;
8522		lba = scsi_4btoul(cdb->addr);
8523		num_blocks = scsi_4btoul(cdb->length);
8524		break;
8525	}
8526	case WRITE_VERIFY_12: {
8527		struct scsi_write_verify_12 *cdb;
8528
8529		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8530
8531		if (cdb->byte2 & SWV_DPO)
8532			dpo = 1;
8533
8534		lba = scsi_4btoul(cdb->addr);
8535		num_blocks = scsi_4btoul(cdb->length);
8536
8537		break;
8538	}
8539	case READ_16:
8540	case WRITE_16: {
8541		struct scsi_rw_16 *cdb;
8542
8543		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8544
8545		if (cdb->byte2 & SRW12_RELADDR)
8546			reladdr = 1;
8547		if (cdb->byte2 & SRW12_FUA)
8548			fua = 1;
8549		if (cdb->byte2 & SRW12_DPO)
8550			dpo = 1;
8551
8552		lba = scsi_8btou64(cdb->addr);
8553		num_blocks = scsi_4btoul(cdb->length);
8554		break;
8555	}
8556	case WRITE_VERIFY_16: {
8557		struct scsi_write_verify_16 *cdb;
8558
8559		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8560
8561		if (cdb->byte2 & SWV_DPO)
8562			dpo = 1;
8563
8564		lba = scsi_8btou64(cdb->addr);
8565		num_blocks = scsi_4btoul(cdb->length);
8566		break;
8567	}
8568	default:
8569		/*
8570		 * We got a command we don't support.  This shouldn't
8571		 * happen, commands should be filtered out above us.
8572		 */
8573		ctl_set_invalid_opcode(ctsio);
8574		ctl_done((union ctl_io *)ctsio);
8575
8576		return (CTL_RETVAL_COMPLETE);
8577		break; /* NOTREACHED */
8578	}
8579
8580	/*
8581	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8582	 * interesting for us, but if RAIDCore is in write-back mode,
8583	 * getting it to do write-through for a particular transaction may
8584	 * not be possible.
8585	 */
8586	/*
8587	 * We don't support relative addressing.  That also requires
8588	 * supporting linked commands, which we don't do.
8589	 */
8590	if (reladdr != 0) {
8591		ctl_set_invalid_field(ctsio,
8592				      /*sks_valid*/ 1,
8593				      /*command*/ 1,
8594				      /*field*/ 1,
8595				      /*bit_valid*/ 1,
8596				      /*bit*/ 0);
8597		ctl_done((union ctl_io *)ctsio);
8598		return (CTL_RETVAL_COMPLETE);
8599	}
8600
8601	/*
8602	 * The first check is to make sure we're in bounds, the second
8603	 * check is to catch wrap-around problems.  If the lba + num blocks
8604	 * is less than the lba, then we've wrapped around and the block
8605	 * range is invalid anyway.
8606	 */
8607	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8608	 || ((lba + num_blocks) < lba)) {
8609		ctl_set_lba_out_of_range(ctsio);
8610		ctl_done((union ctl_io *)ctsio);
8611		return (CTL_RETVAL_COMPLETE);
8612	}
8613
8614	/*
8615	 * According to SBC-3, a transfer length of 0 is not an error.
8616	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8617	 * translates to 256 blocks for those commands.
8618	 */
8619	if (num_blocks == 0) {
8620		ctl_set_success(ctsio);
8621		ctl_done((union ctl_io *)ctsio);
8622		return (CTL_RETVAL_COMPLETE);
8623	}
8624
8625	lbalen.lba = lba;
8626	lbalen.len = num_blocks;
8627	memcpy(ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &lbalen,
8628	       sizeof(lbalen));
8629
8630	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8631
8632	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8633
8634	return (retval);
8635}
8636
8637int
8638ctl_report_luns(struct ctl_scsiio *ctsio)
8639{
8640	struct scsi_report_luns *cdb;
8641	struct scsi_report_luns_data *lun_data;
8642	struct ctl_lun *lun, *request_lun;
8643	int num_luns, retval;
8644	uint32_t alloc_len, lun_datalen;
8645	int num_filled, well_known;
8646	uint32_t initidx, targ_lun_id, lun_id;
8647
8648	retval = CTL_RETVAL_COMPLETE;
8649	well_known = 0;
8650
8651	cdb = (struct scsi_report_luns *)ctsio->cdb;
8652
8653	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8654
8655	mtx_lock(&control_softc->ctl_lock);
8656	num_luns = control_softc->num_luns;
8657	mtx_unlock(&control_softc->ctl_lock);
8658
8659	switch (cdb->select_report) {
8660	case RPL_REPORT_DEFAULT:
8661	case RPL_REPORT_ALL:
8662		break;
8663	case RPL_REPORT_WELLKNOWN:
8664		well_known = 1;
8665		num_luns = 0;
8666		break;
8667	default:
8668		ctl_set_invalid_field(ctsio,
8669				      /*sks_valid*/ 1,
8670				      /*command*/ 1,
8671				      /*field*/ 2,
8672				      /*bit_valid*/ 0,
8673				      /*bit*/ 0);
8674		ctl_done((union ctl_io *)ctsio);
8675		return (retval);
8676		break; /* NOTREACHED */
8677	}
8678
8679	alloc_len = scsi_4btoul(cdb->length);
8680	/*
8681	 * The initiator has to allocate at least 16 bytes for this request,
8682	 * so he can at least get the header and the first LUN.  Otherwise
8683	 * we reject the request (per SPC-3 rev 14, section 6.21).
8684	 */
8685	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
8686	    sizeof(struct scsi_report_luns_lundata))) {
8687		ctl_set_invalid_field(ctsio,
8688				      /*sks_valid*/ 1,
8689				      /*command*/ 1,
8690				      /*field*/ 6,
8691				      /*bit_valid*/ 0,
8692				      /*bit*/ 0);
8693		ctl_done((union ctl_io *)ctsio);
8694		return (retval);
8695	}
8696
8697	request_lun = (struct ctl_lun *)
8698		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8699
8700	lun_datalen = sizeof(*lun_data) +
8701		(num_luns * sizeof(struct scsi_report_luns_lundata));
8702
8703	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
8704	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
8705	ctsio->kern_sg_entries = 0;
8706
8707	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8708
8709	mtx_lock(&control_softc->ctl_lock);
8710	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
8711		lun_id = targ_lun_id;
8712		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
8713			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
8714		if (lun_id >= CTL_MAX_LUNS)
8715			continue;
8716		lun = control_softc->ctl_luns[lun_id];
8717		if (lun == NULL)
8718			continue;
8719
8720		if (targ_lun_id <= 0xff) {
8721			/*
8722			 * Peripheral addressing method, bus number 0.
8723			 */
8724			lun_data->luns[num_filled].lundata[0] =
8725				RPL_LUNDATA_ATYP_PERIPH;
8726			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
8727			num_filled++;
8728		} else if (targ_lun_id <= 0x3fff) {
8729			/*
8730			 * Flat addressing method.
8731			 */
8732			lun_data->luns[num_filled].lundata[0] =
8733				RPL_LUNDATA_ATYP_FLAT |
8734				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
8735#ifdef OLDCTLHEADERS
8736				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
8737				(targ_lun_id & SRLD_BUS_LUN_MASK);
8738#endif
8739			lun_data->luns[num_filled].lundata[1] =
8740#ifdef OLDCTLHEADERS
8741				targ_lun_id >> SRLD_BUS_LUN_BITS;
8742#endif
8743				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
8744			num_filled++;
8745		} else {
8746			printf("ctl_report_luns: bogus LUN number %jd, "
8747			       "skipping\n", (intmax_t)targ_lun_id);
8748		}
8749		/*
8750		 * According to SPC-3, rev 14 section 6.21:
8751		 *
8752		 * "The execution of a REPORT LUNS command to any valid and
8753		 * installed logical unit shall clear the REPORTED LUNS DATA
8754		 * HAS CHANGED unit attention condition for all logical
8755		 * units of that target with respect to the requesting
8756		 * initiator. A valid and installed logical unit is one
8757		 * having a PERIPHERAL QUALIFIER of 000b in the standard
8758		 * INQUIRY data (see 6.4.2)."
8759		 *
8760		 * If request_lun is NULL, the LUN this report luns command
8761		 * was issued to is either disabled or doesn't exist. In that
8762		 * case, we shouldn't clear any pending lun change unit
8763		 * attention.
8764		 */
8765		if (request_lun != NULL)
8766			lun->pending_sense[initidx].ua_pending &=
8767				~CTL_UA_LUN_CHANGE;
8768	}
8769	mtx_unlock(&control_softc->ctl_lock);
8770
8771	/*
8772	 * It's quite possible that we've returned fewer LUNs than we allocated
8773	 * space for.  Trim it.
8774	 */
8775	lun_datalen = sizeof(*lun_data) +
8776		(num_filled * sizeof(struct scsi_report_luns_lundata));
8777
8778	if (lun_datalen < alloc_len) {
8779		ctsio->residual = alloc_len - lun_datalen;
8780		ctsio->kern_data_len = lun_datalen;
8781		ctsio->kern_total_len = lun_datalen;
8782	} else {
8783		ctsio->residual = 0;
8784		ctsio->kern_data_len = alloc_len;
8785		ctsio->kern_total_len = alloc_len;
8786	}
8787	ctsio->kern_data_resid = 0;
8788	ctsio->kern_rel_offset = 0;
8789	ctsio->kern_sg_entries = 0;
8790
8791	/*
8792	 * We set this to the actual data length, regardless of how much
8793	 * space we actually have to return results.  If the user looks at
8794	 * this value, he'll know whether or not he allocated enough space
8795	 * and reissue the command if necessary.  We don't support well
8796	 * known logical units, so if the user asks for that, return none.
8797	 */
8798	scsi_ulto4b(lun_datalen - 8, lun_data->length);
8799
8800	/*
8801	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
8802	 * this request.
8803	 */
8804	ctsio->scsi_status = SCSI_STATUS_OK;
8805
8806	ctsio->be_move_done = ctl_config_move_done;
8807	ctl_datamove((union ctl_io *)ctsio);
8808
8809	return (retval);
8810}
8811
8812int
8813ctl_request_sense(struct ctl_scsiio *ctsio)
8814{
8815	struct scsi_request_sense *cdb;
8816	struct scsi_sense_data *sense_ptr;
8817	struct ctl_lun *lun;
8818	uint32_t initidx;
8819	int have_error;
8820	scsi_sense_data_type sense_format;
8821
8822	cdb = (struct scsi_request_sense *)ctsio->cdb;
8823
8824	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8825
8826	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
8827
8828	/*
8829	 * Determine which sense format the user wants.
8830	 */
8831	if (cdb->byte2 & SRS_DESC)
8832		sense_format = SSD_TYPE_DESC;
8833	else
8834		sense_format = SSD_TYPE_FIXED;
8835
8836	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
8837	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
8838	ctsio->kern_sg_entries = 0;
8839
8840	/*
8841	 * struct scsi_sense_data, which is currently set to 256 bytes, is
8842	 * larger than the largest allowed value for the length field in the
8843	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
8844	 */
8845	ctsio->residual = 0;
8846	ctsio->kern_data_len = cdb->length;
8847	ctsio->kern_total_len = cdb->length;
8848
8849	ctsio->kern_data_resid = 0;
8850	ctsio->kern_rel_offset = 0;
8851	ctsio->kern_sg_entries = 0;
8852
8853	/*
8854	 * If we don't have a LUN, we don't have any pending sense.
8855	 */
8856	if (lun == NULL)
8857		goto no_sense;
8858
8859	have_error = 0;
8860	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8861	/*
8862	 * Check for pending sense, and then for pending unit attentions.
8863	 * Pending sense gets returned first, then pending unit attentions.
8864	 */
8865	mtx_lock(&lun->ctl_softc->ctl_lock);
8866	if (ctl_is_set(lun->have_ca, initidx)) {
8867		scsi_sense_data_type stored_format;
8868
8869		/*
8870		 * Check to see which sense format was used for the stored
8871		 * sense data.
8872		 */
8873		stored_format = scsi_sense_type(
8874		    &lun->pending_sense[initidx].sense);
8875
8876		/*
8877		 * If the user requested a different sense format than the
8878		 * one we stored, then we need to convert it to the other
8879		 * format.  If we're going from descriptor to fixed format
8880		 * sense data, we may lose things in translation, depending
8881		 * on what options were used.
8882		 *
8883		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
8884		 * for some reason we'll just copy it out as-is.
8885		 */
8886		if ((stored_format == SSD_TYPE_FIXED)
8887		 && (sense_format == SSD_TYPE_DESC))
8888			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
8889			    &lun->pending_sense[initidx].sense,
8890			    (struct scsi_sense_data_desc *)sense_ptr);
8891		else if ((stored_format == SSD_TYPE_DESC)
8892		      && (sense_format == SSD_TYPE_FIXED))
8893			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
8894			    &lun->pending_sense[initidx].sense,
8895			    (struct scsi_sense_data_fixed *)sense_ptr);
8896		else
8897			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
8898			       ctl_min(sizeof(*sense_ptr),
8899			       sizeof(lun->pending_sense[initidx].sense)));
8900
8901		ctl_clear_mask(lun->have_ca, initidx);
8902		have_error = 1;
8903	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
8904		ctl_ua_type ua_type;
8905
8906		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
8907				       sense_ptr, sense_format);
8908		if (ua_type != CTL_UA_NONE) {
8909			have_error = 1;
8910			/* We're reporting this UA, so clear it */
8911			lun->pending_sense[initidx].ua_pending &= ~ua_type;
8912		}
8913	}
8914	mtx_unlock(&lun->ctl_softc->ctl_lock);
8915
8916	/*
8917	 * We already have a pending error, return it.
8918	 */
8919	if (have_error != 0) {
8920		/*
8921		 * We report the SCSI status as OK, since the status of the
8922		 * request sense command itself is OK.
8923		 */
8924		ctsio->scsi_status = SCSI_STATUS_OK;
8925
8926		/*
8927		 * We report 0 for the sense length, because we aren't doing
8928		 * autosense in this case.  We're reporting sense as
8929		 * parameter data.
8930		 */
8931		ctsio->sense_len = 0;
8932
8933		ctsio->be_move_done = ctl_config_move_done;
8934		ctl_datamove((union ctl_io *)ctsio);
8935
8936		return (CTL_RETVAL_COMPLETE);
8937	}
8938
8939no_sense:
8940
8941	/*
8942	 * No sense information to report, so we report that everything is
8943	 * okay.
8944	 */
8945	ctl_set_sense_data(sense_ptr,
8946			   lun,
8947			   sense_format,
8948			   /*current_error*/ 1,
8949			   /*sense_key*/ SSD_KEY_NO_SENSE,
8950			   /*asc*/ 0x00,
8951			   /*ascq*/ 0x00,
8952			   SSD_ELEM_NONE);
8953
8954	ctsio->scsi_status = SCSI_STATUS_OK;
8955
8956	/*
8957	 * We report 0 for the sense length, because we aren't doing
8958	 * autosense in this case.  We're reporting sense as parameter data.
8959	 */
8960	ctsio->sense_len = 0;
8961	ctsio->be_move_done = ctl_config_move_done;
8962	ctl_datamove((union ctl_io *)ctsio);
8963
8964	return (CTL_RETVAL_COMPLETE);
8965}
8966
8967int
8968ctl_tur(struct ctl_scsiio *ctsio)
8969{
8970	struct ctl_lun *lun;
8971
8972	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8973
8974	CTL_DEBUG_PRINT(("ctl_tur\n"));
8975
8976	if (lun == NULL)
8977		return (-EINVAL);
8978
8979	ctsio->scsi_status = SCSI_STATUS_OK;
8980	ctsio->io_hdr.status = CTL_SUCCESS;
8981
8982	ctl_done((union ctl_io *)ctsio);
8983
8984	return (CTL_RETVAL_COMPLETE);
8985}
8986
8987#ifdef notyet
8988static int
8989ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
8990{
8991
8992}
8993#endif
8994
8995static int
8996ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
8997{
8998	struct scsi_vpd_supported_pages *pages;
8999	int sup_page_size;
9000	struct ctl_lun *lun;
9001
9002	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9003
9004	sup_page_size = sizeof(struct scsi_vpd_supported_pages) +
9005		SCSI_EVPD_NUM_SUPPORTED_PAGES;
9006	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9007	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9008	ctsio->kern_sg_entries = 0;
9009
9010	if (sup_page_size < alloc_len) {
9011		ctsio->residual = alloc_len - sup_page_size;
9012		ctsio->kern_data_len = sup_page_size;
9013		ctsio->kern_total_len = sup_page_size;
9014	} else {
9015		ctsio->residual = 0;
9016		ctsio->kern_data_len = alloc_len;
9017		ctsio->kern_total_len = alloc_len;
9018	}
9019	ctsio->kern_data_resid = 0;
9020	ctsio->kern_rel_offset = 0;
9021	ctsio->kern_sg_entries = 0;
9022
9023	/*
9024	 * The control device is always connected.  The disk device, on the
9025	 * other hand, may not be online all the time.  Need to change this
9026	 * to figure out whether the disk device is actually online or not.
9027	 */
9028	if (lun != NULL)
9029		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9030				lun->be_lun->lun_type;
9031	else
9032		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9033
9034	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9035	/* Supported VPD pages */
9036	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9037	/* Serial Number */
9038	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9039	/* Device Identification */
9040	pages->page_list[2] = SVPD_DEVICE_ID;
9041
9042	ctsio->scsi_status = SCSI_STATUS_OK;
9043
9044	ctsio->be_move_done = ctl_config_move_done;
9045	ctl_datamove((union ctl_io *)ctsio);
9046
9047	return (CTL_RETVAL_COMPLETE);
9048}
9049
9050static int
9051ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9052{
9053	struct scsi_vpd_unit_serial_number *sn_ptr;
9054	struct ctl_lun *lun;
9055#ifndef CTL_USE_BACKEND_SN
9056	char tmpstr[32];
9057#endif
9058
9059	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9060
9061	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9062	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9063	ctsio->kern_sg_entries = 0;
9064
9065	if (sizeof(*sn_ptr) < alloc_len) {
9066		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9067		ctsio->kern_data_len = sizeof(*sn_ptr);
9068		ctsio->kern_total_len = sizeof(*sn_ptr);
9069	} else {
9070		ctsio->residual = 0;
9071		ctsio->kern_data_len = alloc_len;
9072		ctsio->kern_total_len = alloc_len;
9073	}
9074	ctsio->kern_data_resid = 0;
9075	ctsio->kern_rel_offset = 0;
9076	ctsio->kern_sg_entries = 0;
9077
9078	/*
9079	 * The control device is always connected.  The disk device, on the
9080	 * other hand, may not be online all the time.  Need to change this
9081	 * to figure out whether the disk device is actually online or not.
9082	 */
9083	if (lun != NULL)
9084		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9085				  lun->be_lun->lun_type;
9086	else
9087		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9088
9089	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9090	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9091#ifdef CTL_USE_BACKEND_SN
9092	/*
9093	 * If we don't have a LUN, we just leave the serial number as
9094	 * all spaces.
9095	 */
9096	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9097	if (lun != NULL) {
9098		strncpy((char *)sn_ptr->serial_num,
9099			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9100	}
9101#else
9102	/*
9103	 * Note that we're using a non-unique serial number here,
9104	 */
9105	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9106	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9107	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9108		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9109#endif
9110	ctsio->scsi_status = SCSI_STATUS_OK;
9111
9112	ctsio->be_move_done = ctl_config_move_done;
9113	ctl_datamove((union ctl_io *)ctsio);
9114
9115	return (CTL_RETVAL_COMPLETE);
9116}
9117
9118
9119static int
9120ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9121{
9122	struct scsi_vpd_device_id *devid_ptr;
9123	struct scsi_vpd_id_descriptor *desc, *desc1;
9124	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9125	struct scsi_vpd_id_t10 *t10id;
9126	struct ctl_softc *ctl_softc;
9127	struct ctl_lun *lun;
9128	struct ctl_frontend *fe;
9129#ifndef CTL_USE_BACKEND_SN
9130	char tmpstr[32];
9131#endif /* CTL_USE_BACKEND_SN */
9132	int devid_len;
9133
9134	ctl_softc = control_softc;
9135
9136	mtx_lock(&ctl_softc->ctl_lock);
9137	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9138	mtx_unlock(&ctl_softc->ctl_lock);
9139
9140	if (fe->devid != NULL)
9141		return ((fe->devid)(ctsio, alloc_len));
9142
9143	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9144
9145	devid_len = sizeof(struct scsi_vpd_device_id) +
9146		sizeof(struct scsi_vpd_id_descriptor) +
9147		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9148		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9149		sizeof(struct scsi_vpd_id_descriptor) +
9150		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9151		sizeof(struct scsi_vpd_id_descriptor) +
9152		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9153
9154	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9155	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9156	ctsio->kern_sg_entries = 0;
9157
9158	if (devid_len < alloc_len) {
9159		ctsio->residual = alloc_len - devid_len;
9160		ctsio->kern_data_len = devid_len;
9161		ctsio->kern_total_len = devid_len;
9162	} else {
9163		ctsio->residual = 0;
9164		ctsio->kern_data_len = alloc_len;
9165		ctsio->kern_total_len = alloc_len;
9166	}
9167	ctsio->kern_data_resid = 0;
9168	ctsio->kern_rel_offset = 0;
9169	ctsio->kern_sg_entries = 0;
9170
9171	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9172	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9173	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9174		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9175	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9176	          CTL_WWPN_LEN);
9177	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9178	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9179
9180	/*
9181	 * The control device is always connected.  The disk device, on the
9182	 * other hand, may not be online all the time.
9183	 */
9184	if (lun != NULL)
9185		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9186				     lun->be_lun->lun_type;
9187	else
9188		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9189
9190	devid_ptr->page_code = SVPD_DEVICE_ID;
9191
9192	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9193
9194	mtx_lock(&ctl_softc->ctl_lock);
9195
9196	/*
9197	 * For Fibre channel,
9198	 */
9199	if (fe->port_type == CTL_PORT_FC)
9200	{
9201		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9202				      SVPD_ID_CODESET_ASCII;
9203        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9204		              SVPD_ID_CODESET_BINARY;
9205	}
9206	else
9207	{
9208		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9209				      SVPD_ID_CODESET_ASCII;
9210        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9211		              SVPD_ID_CODESET_BINARY;
9212	}
9213	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9214	mtx_unlock(&ctl_softc->ctl_lock);
9215
9216	/*
9217	 * We're using a LUN association here.  i.e., this device ID is a
9218	 * per-LUN identifier.
9219	 */
9220	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9221	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9222	strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9223
9224	/*
9225	 * desc1 is for the WWPN which is a port asscociation.
9226	 */
9227	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9228	desc1->length = CTL_WWPN_LEN;
9229	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9230	/* For testing just create the WWPN */
9231#if 0
9232	ddb_GetWWNN((char *)desc1->identifier);
9233
9234	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9235	/* This is so Copancontrol will return something sane */
9236	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9237	    ctsio->io_hdr.nexus.targ_port!=8)
9238		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9239	else
9240		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9241#endif
9242
9243	be64enc(desc1->identifier, fe->wwpn);
9244
9245	/*
9246	 * desc2 is for the Relative Target Port(type 4h) identifier
9247	 */
9248	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9249	                 | SVPD_ID_TYPE_RELTARG;
9250	desc2->length = 4;
9251//#if 0
9252	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9253	/* This is so Copancontrol will return something sane */
9254	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9255	    ctsio->io_hdr.nexus.targ_port!=8)
9256		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9257	else
9258	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9259//#endif
9260
9261	/*
9262	 * desc3 is for the Target Port Group(type 5h) identifier
9263	 */
9264	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9265	                 | SVPD_ID_TYPE_TPORTGRP;
9266	desc3->length = 4;
9267	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9268		desc3->identifier[3] = 1;
9269	else
9270		desc3->identifier[3] = 2;
9271
9272#ifdef CTL_USE_BACKEND_SN
9273	/*
9274	 * If we've actually got a backend, copy the device id from the
9275	 * per-LUN data.  Otherwise, set it to all spaces.
9276	 */
9277	if (lun != NULL) {
9278		/*
9279		 * Copy the backend's LUN ID.
9280		 */
9281		strncpy((char *)t10id->vendor_spec_id,
9282			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9283	} else {
9284		/*
9285		 * No backend, set this to spaces.
9286		 */
9287		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9288	}
9289#else
9290	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9291		 (lun != NULL) ?  (int)lun->lun : 0);
9292	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9293		sizeof(tmpstr)));
9294#endif
9295
9296	ctsio->scsi_status = SCSI_STATUS_OK;
9297
9298	ctsio->be_move_done = ctl_config_move_done;
9299	ctl_datamove((union ctl_io *)ctsio);
9300
9301	return (CTL_RETVAL_COMPLETE);
9302}
9303
9304static int
9305ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9306{
9307	struct scsi_inquiry *cdb;
9308	int alloc_len, retval;
9309
9310	cdb = (struct scsi_inquiry *)ctsio->cdb;
9311
9312	retval = CTL_RETVAL_COMPLETE;
9313
9314	alloc_len = scsi_2btoul(cdb->length);
9315
9316	switch (cdb->page_code) {
9317	case SVPD_SUPPORTED_PAGES:
9318		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9319		break;
9320	case SVPD_UNIT_SERIAL_NUMBER:
9321		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9322		break;
9323	case SVPD_DEVICE_ID:
9324		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9325		break;
9326	default:
9327		ctl_set_invalid_field(ctsio,
9328				      /*sks_valid*/ 1,
9329				      /*command*/ 1,
9330				      /*field*/ 2,
9331				      /*bit_valid*/ 0,
9332				      /*bit*/ 0);
9333		ctl_done((union ctl_io *)ctsio);
9334		retval = CTL_RETVAL_COMPLETE;
9335		break;
9336	}
9337
9338	return (retval);
9339}
9340
9341static int
9342ctl_inquiry_std(struct ctl_scsiio *ctsio)
9343{
9344	struct scsi_inquiry_data *inq_ptr;
9345	struct scsi_inquiry *cdb;
9346	struct ctl_softc *ctl_softc;
9347	struct ctl_lun *lun;
9348	uint32_t alloc_len;
9349	int is_fc;
9350
9351	ctl_softc = control_softc;
9352
9353	/*
9354	 * Figure out whether we're talking to a Fibre Channel port or not.
9355	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9356	 * SCSI front ends.
9357	 */
9358	mtx_lock(&ctl_softc->ctl_lock);
9359	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9360	    CTL_PORT_FC)
9361		is_fc = 0;
9362	else
9363		is_fc = 1;
9364	mtx_unlock(&ctl_softc->ctl_lock);
9365
9366	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9367	cdb = (struct scsi_inquiry *)ctsio->cdb;
9368	alloc_len = scsi_2btoul(cdb->length);
9369
9370	/*
9371	 * We malloc the full inquiry data size here and fill it
9372	 * in.  If the user only asks for less, we'll give him
9373	 * that much.
9374	 */
9375	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9376	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9377	ctsio->kern_sg_entries = 0;
9378	ctsio->kern_data_resid = 0;
9379	ctsio->kern_rel_offset = 0;
9380
9381	if (sizeof(*inq_ptr) < alloc_len) {
9382		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9383		ctsio->kern_data_len = sizeof(*inq_ptr);
9384		ctsio->kern_total_len = sizeof(*inq_ptr);
9385	} else {
9386		ctsio->residual = 0;
9387		ctsio->kern_data_len = alloc_len;
9388		ctsio->kern_total_len = alloc_len;
9389	}
9390
9391	/*
9392	 * If we have a LUN configured, report it as connected.  Otherwise,
9393	 * report that it is offline or no device is supported, depending
9394	 * on the value of inquiry_pq_no_lun.
9395	 *
9396	 * According to the spec (SPC-4 r34), the peripheral qualifier
9397	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9398	 *
9399	 * "A peripheral device having the specified peripheral device type
9400	 * is not connected to this logical unit. However, the device
9401	 * server is capable of supporting the specified peripheral device
9402	 * type on this logical unit."
9403	 *
9404	 * According to the same spec, the peripheral qualifier
9405	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9406	 *
9407	 * "The device server is not capable of supporting a peripheral
9408	 * device on this logical unit. For this peripheral qualifier the
9409	 * peripheral device type shall be set to 1Fh. All other peripheral
9410	 * device type values are reserved for this peripheral qualifier."
9411	 *
9412	 * Given the text, it would seem that we probably want to report that
9413	 * the LUN is offline here.  There is no LUN connected, but we can
9414	 * support a LUN at the given LUN number.
9415	 *
9416	 * In the real world, though, it sounds like things are a little
9417	 * different:
9418	 *
9419	 * - Linux, when presented with a LUN with the offline peripheral
9420	 *   qualifier, will create an sg driver instance for it.  So when
9421	 *   you attach it to CTL, you wind up with a ton of sg driver
9422	 *   instances.  (One for every LUN that Linux bothered to probe.)
9423	 *   Linux does this despite the fact that it issues a REPORT LUNs
9424	 *   to LUN 0 to get the inventory of supported LUNs.
9425	 *
9426	 * - There is other anecdotal evidence (from Emulex folks) about
9427	 *   arrays that use the offline peripheral qualifier for LUNs that
9428	 *   are on the "passive" path in an active/passive array.
9429	 *
9430	 * So the solution is provide a hopefully reasonable default
9431	 * (return bad/no LUN) and allow the user to change the behavior
9432	 * with a tunable/sysctl variable.
9433	 */
9434	if (lun != NULL)
9435		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9436				  lun->be_lun->lun_type;
9437	else if (ctl_softc->inquiry_pq_no_lun == 0)
9438		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9439	else
9440		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9441
9442	/* RMB in byte 2 is 0 */
9443	inq_ptr->version = SCSI_REV_SPC3;
9444
9445	/*
9446	 * According to SAM-3, even if a device only supports a single
9447	 * level of LUN addressing, it should still set the HISUP bit:
9448	 *
9449	 * 4.9.1 Logical unit numbers overview
9450	 *
9451	 * All logical unit number formats described in this standard are
9452	 * hierarchical in structure even when only a single level in that
9453	 * hierarchy is used. The HISUP bit shall be set to one in the
9454	 * standard INQUIRY data (see SPC-2) when any logical unit number
9455	 * format described in this standard is used.  Non-hierarchical
9456	 * formats are outside the scope of this standard.
9457	 *
9458	 * Therefore we set the HiSup bit here.
9459	 *
9460	 * The reponse format is 2, per SPC-3.
9461	 */
9462	inq_ptr->response_format = SID_HiSup | 2;
9463
9464	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9465	CTL_DEBUG_PRINT(("additional_length = %d\n",
9466			 inq_ptr->additional_length));
9467
9468	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9469	/* 16 bit addressing */
9470	if (is_fc == 0)
9471		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9472	/* XXX set the SID_MultiP bit here if we're actually going to
9473	   respond on multiple ports */
9474	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9475
9476	/* 16 bit data bus, synchronous transfers */
9477	/* XXX these flags don't apply for FC */
9478	if (is_fc == 0)
9479		inq_ptr->flags = SID_WBus16 | SID_Sync;
9480	/*
9481	 * XXX KDM do we want to support tagged queueing on the control
9482	 * device at all?
9483	 */
9484	if ((lun == NULL)
9485	 || (lun->be_lun->lun_type != T_PROCESSOR))
9486		inq_ptr->flags |= SID_CmdQue;
9487	/*
9488	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
9489	 * We have 8 bytes for the vendor name, and 16 bytes for the device
9490	 * name and 4 bytes for the revision.
9491	 */
9492	strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
9493	if (lun == NULL) {
9494		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9495	} else {
9496		switch (lun->be_lun->lun_type) {
9497		case T_DIRECT:
9498			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
9499			break;
9500		case T_PROCESSOR:
9501			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
9502			break;
9503		default:
9504			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
9505			break;
9506		}
9507	}
9508
9509	/*
9510	 * XXX make this a macro somewhere so it automatically gets
9511	 * incremented when we make changes.
9512	 */
9513	strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
9514
9515	/*
9516	 * For parallel SCSI, we support double transition and single
9517	 * transition clocking.  We also support QAS (Quick Arbitration
9518	 * and Selection) and Information Unit transfers on both the
9519	 * control and array devices.
9520	 */
9521	if (is_fc == 0)
9522		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
9523				    SID_SPI_IUS;
9524
9525	/* SAM-3 */
9526	scsi_ulto2b(0x0060, inq_ptr->version1);
9527	/* SPC-3 (no version claimed) XXX should we claim a version? */
9528	scsi_ulto2b(0x0300, inq_ptr->version2);
9529	if (is_fc) {
9530		/* FCP-2 ANSI INCITS.350:2003 */
9531		scsi_ulto2b(0x0917, inq_ptr->version3);
9532	} else {
9533		/* SPI-4 ANSI INCITS.362:200x */
9534		scsi_ulto2b(0x0B56, inq_ptr->version3);
9535	}
9536
9537	if (lun == NULL) {
9538		/* SBC-2 (no version claimed) XXX should we claim a version? */
9539		scsi_ulto2b(0x0320, inq_ptr->version4);
9540	} else {
9541		switch (lun->be_lun->lun_type) {
9542		case T_DIRECT:
9543			/*
9544			 * SBC-2 (no version claimed) XXX should we claim a
9545			 * version?
9546			 */
9547			scsi_ulto2b(0x0320, inq_ptr->version4);
9548			break;
9549		case T_PROCESSOR:
9550		default:
9551			break;
9552		}
9553	}
9554
9555	ctsio->scsi_status = SCSI_STATUS_OK;
9556	if (ctsio->kern_data_len > 0) {
9557		ctsio->be_move_done = ctl_config_move_done;
9558		ctl_datamove((union ctl_io *)ctsio);
9559	} else {
9560		ctsio->io_hdr.status = CTL_SUCCESS;
9561		ctl_done((union ctl_io *)ctsio);
9562	}
9563
9564	return (CTL_RETVAL_COMPLETE);
9565}
9566
9567int
9568ctl_inquiry(struct ctl_scsiio *ctsio)
9569{
9570	struct scsi_inquiry *cdb;
9571	int retval;
9572
9573	cdb = (struct scsi_inquiry *)ctsio->cdb;
9574
9575	retval = 0;
9576
9577	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
9578
9579	/*
9580	 * Right now, we don't support the CmdDt inquiry information.
9581	 * This would be nice to support in the future.  When we do
9582	 * support it, we should change this test so that it checks to make
9583	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
9584	 */
9585#ifdef notyet
9586	if (((cdb->byte2 & SI_EVPD)
9587	 && (cdb->byte2 & SI_CMDDT)))
9588#endif
9589	if (cdb->byte2 & SI_CMDDT) {
9590		/*
9591		 * Point to the SI_CMDDT bit.  We might change this
9592		 * when we support SI_CMDDT, but since both bits would be
9593		 * "wrong", this should probably just stay as-is then.
9594		 */
9595		ctl_set_invalid_field(ctsio,
9596				      /*sks_valid*/ 1,
9597				      /*command*/ 1,
9598				      /*field*/ 1,
9599				      /*bit_valid*/ 1,
9600				      /*bit*/ 1);
9601		ctl_done((union ctl_io *)ctsio);
9602		return (CTL_RETVAL_COMPLETE);
9603	}
9604	if (cdb->byte2 & SI_EVPD)
9605		retval = ctl_inquiry_evpd(ctsio);
9606#ifdef notyet
9607	else if (cdb->byte2 & SI_CMDDT)
9608		retval = ctl_inquiry_cmddt(ctsio);
9609#endif
9610	else
9611		retval = ctl_inquiry_std(ctsio);
9612
9613	return (retval);
9614}
9615
9616/*
9617 * For known CDB types, parse the LBA and length.
9618 */
9619static int
9620ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
9621{
9622	if (io->io_hdr.io_type != CTL_IO_SCSI)
9623		return (1);
9624
9625	switch (io->scsiio.cdb[0]) {
9626	case READ_6:
9627	case WRITE_6: {
9628		struct scsi_rw_6 *cdb;
9629
9630		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
9631
9632		*lba = scsi_3btoul(cdb->addr);
9633		/* only 5 bits are valid in the most significant address byte */
9634		*lba &= 0x1fffff;
9635		*len = cdb->length;
9636		break;
9637	}
9638	case READ_10:
9639	case WRITE_10: {
9640		struct scsi_rw_10 *cdb;
9641
9642		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
9643
9644		*lba = scsi_4btoul(cdb->addr);
9645		*len = scsi_2btoul(cdb->length);
9646		break;
9647	}
9648	case WRITE_VERIFY_10: {
9649		struct scsi_write_verify_10 *cdb;
9650
9651		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
9652
9653		*lba = scsi_4btoul(cdb->addr);
9654		*len = scsi_2btoul(cdb->length);
9655		break;
9656	}
9657	case READ_12:
9658	case WRITE_12: {
9659		struct scsi_rw_12 *cdb;
9660
9661		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
9662
9663		*lba = scsi_4btoul(cdb->addr);
9664		*len = scsi_4btoul(cdb->length);
9665		break;
9666	}
9667	case WRITE_VERIFY_12: {
9668		struct scsi_write_verify_12 *cdb;
9669
9670		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
9671
9672		*lba = scsi_4btoul(cdb->addr);
9673		*len = scsi_4btoul(cdb->length);
9674		break;
9675	}
9676	case READ_16:
9677	case WRITE_16: {
9678		struct scsi_rw_16 *cdb;
9679
9680		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
9681
9682		*lba = scsi_8btou64(cdb->addr);
9683		*len = scsi_4btoul(cdb->length);
9684		break;
9685	}
9686	case WRITE_VERIFY_16: {
9687		struct scsi_write_verify_16 *cdb;
9688
9689		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
9690
9691
9692		*lba = scsi_8btou64(cdb->addr);
9693		*len = scsi_4btoul(cdb->length);
9694		break;
9695	}
9696	default:
9697		return (1);
9698		break; /* NOTREACHED */
9699	}
9700
9701	return (0);
9702}
9703
9704static ctl_action
9705ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
9706{
9707	uint64_t endlba1, endlba2;
9708
9709	endlba1 = lba1 + len1 - 1;
9710	endlba2 = lba2 + len2 - 1;
9711
9712	if ((endlba1 < lba2)
9713	 || (endlba2 < lba1))
9714		return (CTL_ACTION_PASS);
9715	else
9716		return (CTL_ACTION_BLOCK);
9717}
9718
9719static ctl_action
9720ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
9721{
9722	uint64_t lba1, lba2;
9723	uint32_t len1, len2;
9724	int retval;
9725
9726	retval = ctl_get_lba_len(io1, &lba1, &len1);
9727	if (retval != 0)
9728		return (CTL_ACTION_ERROR);
9729
9730	retval = ctl_get_lba_len(io2, &lba2, &len2);
9731	if (retval != 0)
9732		return (CTL_ACTION_ERROR);
9733
9734	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
9735}
9736
9737static ctl_action
9738ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
9739{
9740	struct ctl_cmd_entry *pending_entry, *ooa_entry;
9741	ctl_serialize_action *serialize_row;
9742
9743	/*
9744	 * The initiator attempted multiple untagged commands at the same
9745	 * time.  Can't do that.
9746	 */
9747	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9748	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9749	 && ((pending_io->io_hdr.nexus.targ_port ==
9750	      ooa_io->io_hdr.nexus.targ_port)
9751	  && (pending_io->io_hdr.nexus.initid.id ==
9752	      ooa_io->io_hdr.nexus.initid.id))
9753	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9754		return (CTL_ACTION_OVERLAP);
9755
9756	/*
9757	 * The initiator attempted to send multiple tagged commands with
9758	 * the same ID.  (It's fine if different initiators have the same
9759	 * tag ID.)
9760	 *
9761	 * Even if all of those conditions are true, we don't kill the I/O
9762	 * if the command ahead of us has been aborted.  We won't end up
9763	 * sending it to the FETD, and it's perfectly legal to resend a
9764	 * command with the same tag number as long as the previous
9765	 * instance of this tag number has been aborted somehow.
9766	 */
9767	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9768	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
9769	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
9770	 && ((pending_io->io_hdr.nexus.targ_port ==
9771	      ooa_io->io_hdr.nexus.targ_port)
9772	  && (pending_io->io_hdr.nexus.initid.id ==
9773	      ooa_io->io_hdr.nexus.initid.id))
9774	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
9775		return (CTL_ACTION_OVERLAP_TAG);
9776
9777	/*
9778	 * If we get a head of queue tag, SAM-3 says that we should
9779	 * immediately execute it.
9780	 *
9781	 * What happens if this command would normally block for some other
9782	 * reason?  e.g. a request sense with a head of queue tag
9783	 * immediately after a write.  Normally that would block, but this
9784	 * will result in its getting executed immediately...
9785	 *
9786	 * We currently return "pass" instead of "skip", so we'll end up
9787	 * going through the rest of the queue to check for overlapped tags.
9788	 *
9789	 * XXX KDM check for other types of blockage first??
9790	 */
9791	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9792		return (CTL_ACTION_PASS);
9793
9794	/*
9795	 * Ordered tags have to block until all items ahead of them
9796	 * have completed.  If we get called with an ordered tag, we always
9797	 * block, if something else is ahead of us in the queue.
9798	 */
9799	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
9800		return (CTL_ACTION_BLOCK);
9801
9802	/*
9803	 * Simple tags get blocked until all head of queue and ordered tags
9804	 * ahead of them have completed.  I'm lumping untagged commands in
9805	 * with simple tags here.  XXX KDM is that the right thing to do?
9806	 */
9807	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
9808	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
9809	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
9810	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
9811		return (CTL_ACTION_BLOCK);
9812
9813	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
9814	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
9815
9816	serialize_row = ctl_serialize_table[ooa_entry->seridx];
9817
9818	switch (serialize_row[pending_entry->seridx]) {
9819	case CTL_SER_BLOCK:
9820		return (CTL_ACTION_BLOCK);
9821		break; /* NOTREACHED */
9822	case CTL_SER_EXTENT:
9823		return (ctl_extent_check(pending_io, ooa_io));
9824		break; /* NOTREACHED */
9825	case CTL_SER_PASS:
9826		return (CTL_ACTION_PASS);
9827		break; /* NOTREACHED */
9828	case CTL_SER_SKIP:
9829		return (CTL_ACTION_SKIP);
9830		break;
9831	default:
9832		panic("invalid serialization value %d",
9833		      serialize_row[pending_entry->seridx]);
9834		break; /* NOTREACHED */
9835	}
9836
9837	return (CTL_ACTION_ERROR);
9838}
9839
9840/*
9841 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
9842 * Assumptions:
9843 * - pending_io is generally either incoming, or on the blocked queue
9844 * - starting I/O is the I/O we want to start the check with.
9845 */
9846static ctl_action
9847ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
9848	      union ctl_io *starting_io)
9849{
9850	union ctl_io *ooa_io;
9851	ctl_action action;
9852
9853	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
9854
9855	/*
9856	 * Run back along the OOA queue, starting with the current
9857	 * blocked I/O and going through every I/O before it on the
9858	 * queue.  If starting_io is NULL, we'll just end up returning
9859	 * CTL_ACTION_PASS.
9860	 */
9861	for (ooa_io = starting_io; ooa_io != NULL;
9862	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
9863	     ooa_links)){
9864
9865		/*
9866		 * This routine just checks to see whether
9867		 * cur_blocked is blocked by ooa_io, which is ahead
9868		 * of it in the queue.  It doesn't queue/dequeue
9869		 * cur_blocked.
9870		 */
9871		action = ctl_check_for_blockage(pending_io, ooa_io);
9872		switch (action) {
9873		case CTL_ACTION_BLOCK:
9874		case CTL_ACTION_OVERLAP:
9875		case CTL_ACTION_OVERLAP_TAG:
9876		case CTL_ACTION_SKIP:
9877		case CTL_ACTION_ERROR:
9878			return (action);
9879			break; /* NOTREACHED */
9880		case CTL_ACTION_PASS:
9881			break;
9882		default:
9883			panic("invalid action %d", action);
9884			break;  /* NOTREACHED */
9885		}
9886	}
9887
9888	return (CTL_ACTION_PASS);
9889}
9890
9891/*
9892 * Assumptions:
9893 * - An I/O has just completed, and has been removed from the per-LUN OOA
9894 *   queue, so some items on the blocked queue may now be unblocked.
9895 */
9896static int
9897ctl_check_blocked(struct ctl_lun *lun)
9898{
9899	union ctl_io *cur_blocked, *next_blocked;
9900
9901	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
9902
9903	/*
9904	 * Run forward from the head of the blocked queue, checking each
9905	 * entry against the I/Os prior to it on the OOA queue to see if
9906	 * there is still any blockage.
9907	 *
9908	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
9909	 * with our removing a variable on it while it is traversing the
9910	 * list.
9911	 */
9912	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
9913	     cur_blocked != NULL; cur_blocked = next_blocked) {
9914		union ctl_io *prev_ooa;
9915		ctl_action action;
9916
9917		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
9918							  blocked_links);
9919
9920		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
9921						      ctl_ooaq, ooa_links);
9922
9923		/*
9924		 * If cur_blocked happens to be the first item in the OOA
9925		 * queue now, prev_ooa will be NULL, and the action
9926		 * returned will just be CTL_ACTION_PASS.
9927		 */
9928		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
9929
9930		switch (action) {
9931		case CTL_ACTION_BLOCK:
9932			/* Nothing to do here, still blocked */
9933			break;
9934		case CTL_ACTION_OVERLAP:
9935		case CTL_ACTION_OVERLAP_TAG:
9936			/*
9937			 * This shouldn't happen!  In theory we've already
9938			 * checked this command for overlap...
9939			 */
9940			break;
9941		case CTL_ACTION_PASS:
9942		case CTL_ACTION_SKIP: {
9943			struct ctl_softc *softc;
9944			struct ctl_cmd_entry *entry;
9945			uint32_t initidx;
9946			uint8_t opcode;
9947			int isc_retval;
9948
9949			/*
9950			 * The skip case shouldn't happen, this transaction
9951			 * should have never made it onto the blocked queue.
9952			 */
9953			/*
9954			 * This I/O is no longer blocked, we can remove it
9955			 * from the blocked queue.  Since this is a TAILQ
9956			 * (doubly linked list), we can do O(1) removals
9957			 * from any place on the list.
9958			 */
9959			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
9960				     blocked_links);
9961			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
9962
9963			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
9964				/*
9965				 * Need to send IO back to original side to
9966				 * run
9967				 */
9968				union ctl_ha_msg msg_info;
9969
9970				msg_info.hdr.original_sc =
9971					cur_blocked->io_hdr.original_sc;
9972				msg_info.hdr.serializing_sc = cur_blocked;
9973				msg_info.hdr.msg_type = CTL_MSG_R2R;
9974				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
9975				     &msg_info, sizeof(msg_info), 0)) >
9976				     CTL_HA_STATUS_SUCCESS) {
9977					printf("CTL:Check Blocked error from "
9978					       "ctl_ha_msg_send %d\n",
9979					       isc_retval);
9980				}
9981				break;
9982			}
9983			opcode = cur_blocked->scsiio.cdb[0];
9984			entry = &ctl_cmd_table[opcode];
9985			softc = control_softc;
9986
9987			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
9988
9989			/*
9990			 * Check this I/O for LUN state changes that may
9991			 * have happened while this command was blocked.
9992			 * The LUN state may have been changed by a command
9993			 * ahead of us in the queue, so we need to re-check
9994			 * for any states that can be caused by SCSI
9995			 * commands.
9996			 */
9997			if (ctl_scsiio_lun_check(softc, lun, entry,
9998						 &cur_blocked->scsiio) == 0) {
9999				cur_blocked->io_hdr.flags |=
10000				                      CTL_FLAG_IS_WAS_ON_RTR;
10001				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
10002						   &cur_blocked->io_hdr, links);
10003				/*
10004				 * In the non CTL_DONE_THREAD case, we need
10005				 * to wake up the work thread here.  When
10006				 * we're processing completed requests from
10007				 * the work thread context, we'll pop back
10008				 * around and end up pulling things off the
10009				 * RtR queue.  When we aren't processing
10010				 * things from the work thread context,
10011				 * though, we won't ever check the RtR queue.
10012				 * So we need to wake up the thread to clear
10013				 * things off the queue.  Otherwise this
10014				 * transaction will just sit on the RtR queue
10015				 * until a new I/O comes in.  (Which may or
10016				 * may not happen...)
10017				 */
10018#ifndef CTL_DONE_THREAD
10019				ctl_wakeup_thread();
10020#endif
10021			} else
10022				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
10023			break;
10024		}
10025		default:
10026			/*
10027			 * This probably shouldn't happen -- we shouldn't
10028			 * get CTL_ACTION_ERROR, or anything else.
10029			 */
10030			break;
10031		}
10032	}
10033
10034	return (CTL_RETVAL_COMPLETE);
10035}
10036
10037/*
10038 * This routine (with one exception) checks LUN flags that can be set by
10039 * commands ahead of us in the OOA queue.  These flags have to be checked
10040 * when a command initially comes in, and when we pull a command off the
10041 * blocked queue and are preparing to execute it.  The reason we have to
10042 * check these flags for commands on the blocked queue is that the LUN
10043 * state may have been changed by a command ahead of us while we're on the
10044 * blocked queue.
10045 *
10046 * Ordering is somewhat important with these checks, so please pay
10047 * careful attention to the placement of any new checks.
10048 */
10049static int
10050ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10051		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10052{
10053	int retval;
10054
10055	retval = 0;
10056
10057	/*
10058	 * If this shelf is a secondary shelf controller, we have to reject
10059	 * any media access commands.
10060	 */
10061#if 0
10062	/* No longer needed for HA */
10063	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10064	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10065		ctl_set_lun_standby(ctsio);
10066		retval = 1;
10067		goto bailout;
10068	}
10069#endif
10070
10071	/*
10072	 * Check for a reservation conflict.  If this command isn't allowed
10073	 * even on reserved LUNs, and if this initiator isn't the one who
10074	 * reserved us, reject the command with a reservation conflict.
10075	 */
10076	if ((lun->flags & CTL_LUN_RESERVED)
10077	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10078		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10079		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10080		 || (ctsio->io_hdr.nexus.targ_target.id !=
10081		     lun->rsv_nexus.targ_target.id)) {
10082			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10083			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10084			retval = 1;
10085			goto bailout;
10086		}
10087	}
10088
10089	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10090	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10091		uint32_t residx;
10092
10093		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10094		/*
10095		 * if we aren't registered or it's a res holder type
10096		 * reservation and this isn't the res holder then set a
10097		 * conflict.
10098		 * NOTE: Commands which might be allowed on write exclusive
10099		 * type reservations are checked in the particular command
10100		 * for a conflict. Read and SSU are the only ones.
10101		 */
10102		if (!lun->per_res[residx].registered
10103		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10104			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10105			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10106			retval = 1;
10107			goto bailout;
10108		}
10109
10110	}
10111
10112	if ((lun->flags & CTL_LUN_OFFLINE)
10113	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10114		ctl_set_lun_not_ready(ctsio);
10115		retval = 1;
10116		goto bailout;
10117	}
10118
10119	/*
10120	 * If the LUN is stopped, see if this particular command is allowed
10121	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10122	 */
10123	if ((lun->flags & CTL_LUN_STOPPED)
10124	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10125		/* "Logical unit not ready, initializing cmd. required" */
10126		ctl_set_lun_stopped(ctsio);
10127		retval = 1;
10128		goto bailout;
10129	}
10130
10131	if ((lun->flags & CTL_LUN_INOPERABLE)
10132	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10133		/* "Medium format corrupted" */
10134		ctl_set_medium_format_corrupted(ctsio);
10135		retval = 1;
10136		goto bailout;
10137	}
10138
10139bailout:
10140	return (retval);
10141
10142}
10143
10144static void
10145ctl_failover_io(union ctl_io *io, int have_lock)
10146{
10147	ctl_set_busy(&io->scsiio);
10148	ctl_done_lock(io, have_lock);
10149}
10150
10151static void
10152ctl_failover(void)
10153{
10154	struct ctl_lun *lun;
10155	struct ctl_softc *ctl_softc;
10156	union ctl_io *next_io, *pending_io;
10157	union ctl_io *io;
10158	int lun_idx;
10159	int i;
10160
10161	ctl_softc = control_softc;
10162
10163	mtx_lock(&ctl_softc->ctl_lock);
10164	/*
10165	 * Remove any cmds from the other SC from the rtr queue.  These
10166	 * will obviously only be for LUNs for which we're the primary.
10167	 * We can't send status or get/send data for these commands.
10168	 * Since they haven't been executed yet, we can just remove them.
10169	 * We'll either abort them or delete them below, depending on
10170	 * which HA mode we're in.
10171	 */
10172	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10173	     io != NULL; io = next_io) {
10174		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10175		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10176			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10177				      ctl_io_hdr, links);
10178	}
10179
10180	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10181		lun = ctl_softc->ctl_luns[lun_idx];
10182		if (lun==NULL)
10183			continue;
10184
10185		/*
10186		 * Processor LUNs are primary on both sides.
10187		 * XXX will this always be true?
10188		 */
10189		if (lun->be_lun->lun_type == T_PROCESSOR)
10190			continue;
10191
10192		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10193		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10194			printf("FAILOVER: primary lun %d\n", lun_idx);
10195		        /*
10196			 * Remove all commands from the other SC. First from the
10197			 * blocked queue then from the ooa queue. Once we have
10198			 * removed them. Call ctl_check_blocked to see if there
10199			 * is anything that can run.
10200			 */
10201			for (io = (union ctl_io *)TAILQ_FIRST(
10202			     &lun->blocked_queue); io != NULL; io = next_io) {
10203
10204		        	next_io = (union ctl_io *)TAILQ_NEXT(
10205				    &io->io_hdr, blocked_links);
10206
10207				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10208					TAILQ_REMOVE(&lun->blocked_queue,
10209						     &io->io_hdr,blocked_links);
10210					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10211					TAILQ_REMOVE(&lun->ooa_queue,
10212						     &io->io_hdr, ooa_links);
10213
10214					ctl_free_io_internal(io, 1);
10215				}
10216			}
10217
10218			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10219	     		     io != NULL; io = next_io) {
10220
10221		        	next_io = (union ctl_io *)TAILQ_NEXT(
10222				    &io->io_hdr, ooa_links);
10223
10224				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10225
10226					TAILQ_REMOVE(&lun->ooa_queue,
10227						&io->io_hdr,
10228					     	ooa_links);
10229
10230					ctl_free_io_internal(io, 1);
10231				}
10232			}
10233			ctl_check_blocked(lun);
10234		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10235			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10236
10237			printf("FAILOVER: primary lun %d\n", lun_idx);
10238			/*
10239			 * Abort all commands from the other SC.  We can't
10240			 * send status back for them now.  These should get
10241			 * cleaned up when they are completed or come out
10242			 * for a datamove operation.
10243			 */
10244			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10245	     		     io != NULL; io = next_io) {
10246		        	next_io = (union ctl_io *)TAILQ_NEXT(
10247					&io->io_hdr, ooa_links);
10248
10249				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10250					io->io_hdr.flags |= CTL_FLAG_ABORT;
10251			}
10252		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10253			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10254
10255			printf("FAILOVER: secondary lun %d\n", lun_idx);
10256
10257			lun->flags |= CTL_LUN_PRIMARY_SC;
10258
10259			/*
10260			 * We send all I/O that was sent to this controller
10261			 * and redirected to the other side back with
10262			 * busy status, and have the initiator retry it.
10263			 * Figuring out how much data has been transferred,
10264			 * etc. and picking up where we left off would be
10265			 * very tricky.
10266			 *
10267			 * XXX KDM need to remove I/O from the blocked
10268			 * queue as well!
10269			 */
10270			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10271			     &lun->ooa_queue); pending_io != NULL;
10272			     pending_io = next_io) {
10273
10274				next_io =  (union ctl_io *)TAILQ_NEXT(
10275					&pending_io->io_hdr, ooa_links);
10276
10277				pending_io->io_hdr.flags &=
10278					~CTL_FLAG_SENT_2OTHER_SC;
10279
10280				if (pending_io->io_hdr.flags &
10281				    CTL_FLAG_IO_ACTIVE) {
10282					pending_io->io_hdr.flags |=
10283						CTL_FLAG_FAILOVER;
10284				} else {
10285					ctl_set_busy(&pending_io->scsiio);
10286					ctl_done_lock(pending_io,
10287						      /*have_lock*/1);
10288				}
10289			}
10290
10291			/*
10292			 * Build Unit Attention
10293			 */
10294			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10295				lun->pending_sense[i].ua_pending |=
10296				                     CTL_UA_ASYM_ACC_CHANGE;
10297			}
10298		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10299			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10300			printf("FAILOVER: secondary lun %d\n", lun_idx);
10301			/*
10302			 * if the first io on the OOA is not on the RtR queue
10303			 * add it.
10304			 */
10305			lun->flags |= CTL_LUN_PRIMARY_SC;
10306
10307			pending_io = (union ctl_io *)TAILQ_FIRST(
10308			    &lun->ooa_queue);
10309			if (pending_io==NULL) {
10310				printf("Nothing on OOA queue\n");
10311				continue;
10312			}
10313
10314			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10315			if ((pending_io->io_hdr.flags &
10316			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10317				pending_io->io_hdr.flags |=
10318				    CTL_FLAG_IS_WAS_ON_RTR;
10319				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10320						   &pending_io->io_hdr, links);
10321			}
10322#if 0
10323			else
10324			{
10325				printf("Tag 0x%04x is running\n",
10326				      pending_io->scsiio.tag_num);
10327			}
10328#endif
10329
10330			next_io = (union ctl_io *)TAILQ_NEXT(
10331			    &pending_io->io_hdr, ooa_links);
10332			for (pending_io=next_io; pending_io != NULL;
10333			     pending_io = next_io) {
10334				pending_io->io_hdr.flags &=
10335				    ~CTL_FLAG_SENT_2OTHER_SC;
10336				next_io = (union ctl_io *)TAILQ_NEXT(
10337					&pending_io->io_hdr, ooa_links);
10338				if (pending_io->io_hdr.flags &
10339				    CTL_FLAG_IS_WAS_ON_RTR) {
10340#if 0
10341				        printf("Tag 0x%04x is running\n",
10342				      		pending_io->scsiio.tag_num);
10343#endif
10344					continue;
10345				}
10346
10347				switch (ctl_check_ooa(lun, pending_io,
10348			            (union ctl_io *)TAILQ_PREV(
10349				    &pending_io->io_hdr, ctl_ooaq,
10350				    ooa_links))) {
10351
10352				case CTL_ACTION_BLOCK:
10353					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10354							  &pending_io->io_hdr,
10355							  blocked_links);
10356					pending_io->io_hdr.flags |=
10357					    CTL_FLAG_BLOCKED;
10358					break;
10359				case CTL_ACTION_PASS:
10360				case CTL_ACTION_SKIP:
10361					pending_io->io_hdr.flags |=
10362					    CTL_FLAG_IS_WAS_ON_RTR;
10363					STAILQ_INSERT_TAIL(
10364					    &ctl_softc->rtr_queue,
10365					    &pending_io->io_hdr, links);
10366					break;
10367				case CTL_ACTION_OVERLAP:
10368					ctl_set_overlapped_cmd(
10369					    (struct ctl_scsiio *)pending_io);
10370					ctl_done_lock(pending_io,
10371						      /*have_lock*/ 1);
10372					break;
10373				case CTL_ACTION_OVERLAP_TAG:
10374					ctl_set_overlapped_tag(
10375					    (struct ctl_scsiio *)pending_io,
10376					    pending_io->scsiio.tag_num & 0xff);
10377					ctl_done_lock(pending_io,
10378						      /*have_lock*/ 1);
10379					break;
10380				case CTL_ACTION_ERROR:
10381				default:
10382					ctl_set_internal_failure(
10383						(struct ctl_scsiio *)pending_io,
10384						0,  // sks_valid
10385						0); //retry count
10386					ctl_done_lock(pending_io,
10387						      /*have_lock*/ 1);
10388					break;
10389				}
10390			}
10391
10392			/*
10393			 * Build Unit Attention
10394			 */
10395			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10396				lun->pending_sense[i].ua_pending |=
10397				                     CTL_UA_ASYM_ACC_CHANGE;
10398			}
10399		} else {
10400			panic("Unhandled HA mode failover, LUN flags = %#x, "
10401			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10402		}
10403	}
10404	ctl_pause_rtr = 0;
10405	mtx_unlock(&ctl_softc->ctl_lock);
10406}
10407
10408static int
10409ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
10410{
10411	struct ctl_lun *lun;
10412	struct ctl_cmd_entry *entry;
10413	uint8_t opcode;
10414	uint32_t initidx, targ_lun;
10415	int retval;
10416
10417	retval = 0;
10418
10419	lun = NULL;
10420
10421	opcode = ctsio->cdb[0];
10422
10423	mtx_lock(&ctl_softc->ctl_lock);
10424
10425	targ_lun = ctsio->io_hdr.nexus.targ_lun;
10426	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
10427		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
10428	if ((targ_lun < CTL_MAX_LUNS)
10429	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
10430		lun = ctl_softc->ctl_luns[targ_lun];
10431		/*
10432		 * If the LUN is invalid, pretend that it doesn't exist.
10433		 * It will go away as soon as all pending I/O has been
10434		 * completed.
10435		 */
10436		if (lun->flags & CTL_LUN_DISABLED) {
10437			lun = NULL;
10438		} else {
10439			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
10440			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
10441				lun->be_lun;
10442			if (lun->be_lun->lun_type == T_PROCESSOR) {
10443				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
10444			}
10445		}
10446	} else {
10447		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
10448		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
10449	}
10450
10451	entry = &ctl_cmd_table[opcode];
10452
10453	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
10454	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
10455
10456	/*
10457	 * Check to see whether we can send this command to LUNs that don't
10458	 * exist.  This should pretty much only be the case for inquiry
10459	 * and request sense.  Further checks, below, really require having
10460	 * a LUN, so we can't really check the command anymore.  Just put
10461	 * it on the rtr queue.
10462	 */
10463	if (lun == NULL) {
10464		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10465			goto queue_rtr;
10466
10467		ctl_set_unsupported_lun(ctsio);
10468		mtx_unlock(&ctl_softc->ctl_lock);
10469		ctl_done((union ctl_io *)ctsio);
10470		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
10471		goto bailout;
10472	} else {
10473		/*
10474		 * Every I/O goes into the OOA queue for a particular LUN, and
10475		 * stays there until completion.
10476		 */
10477		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
10478
10479		/*
10480		 * Make sure we support this particular command on this LUN.
10481		 * e.g., we don't support writes to the control LUN.
10482		 */
10483		switch (lun->be_lun->lun_type) {
10484		case T_PROCESSOR:
10485		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
10486			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10487			      == 0)) {
10488				ctl_set_invalid_opcode(ctsio);
10489				mtx_unlock(&ctl_softc->ctl_lock);
10490				ctl_done((union ctl_io *)ctsio);
10491				goto bailout;
10492			}
10493			break;
10494		case T_DIRECT:
10495			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
10496			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
10497			      == 0)){
10498				ctl_set_invalid_opcode(ctsio);
10499				mtx_unlock(&ctl_softc->ctl_lock);
10500				ctl_done((union ctl_io *)ctsio);
10501				goto bailout;
10502			}
10503			break;
10504		default:
10505			printf("Unsupported CTL LUN type %d\n",
10506			       lun->be_lun->lun_type);
10507			panic("Unsupported CTL LUN type %d\n",
10508			      lun->be_lun->lun_type);
10509			break; /* NOTREACHED */
10510		}
10511	}
10512
10513	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10514
10515	/*
10516	 * If we've got a request sense, it'll clear the contingent
10517	 * allegiance condition.  Otherwise, if we have a CA condition for
10518	 * this initiator, clear it, because it sent down a command other
10519	 * than request sense.
10520	 */
10521	if ((opcode != REQUEST_SENSE)
10522	 && (ctl_is_set(lun->have_ca, initidx)))
10523		ctl_clear_mask(lun->have_ca, initidx);
10524
10525	/*
10526	 * If the command has this flag set, it handles its own unit
10527	 * attention reporting, we shouldn't do anything.  Otherwise we
10528	 * check for any pending unit attentions, and send them back to the
10529	 * initiator.  We only do this when a command initially comes in,
10530	 * not when we pull it off the blocked queue.
10531	 *
10532	 * According to SAM-3, section 5.3.2, the order that things get
10533	 * presented back to the host is basically unit attentions caused
10534	 * by some sort of reset event, busy status, reservation conflicts
10535	 * or task set full, and finally any other status.
10536	 *
10537	 * One issue here is that some of the unit attentions we report
10538	 * don't fall into the "reset" category (e.g. "reported luns data
10539	 * has changed").  So reporting it here, before the reservation
10540	 * check, may be technically wrong.  I guess the only thing to do
10541	 * would be to check for and report the reset events here, and then
10542	 * check for the other unit attention types after we check for a
10543	 * reservation conflict.
10544	 *
10545	 * XXX KDM need to fix this
10546	 */
10547	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
10548		ctl_ua_type ua_type;
10549
10550		ua_type = lun->pending_sense[initidx].ua_pending;
10551		if (ua_type != CTL_UA_NONE) {
10552			scsi_sense_data_type sense_format;
10553
10554			if (lun != NULL)
10555				sense_format = (lun->flags &
10556				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
10557				    SSD_TYPE_FIXED;
10558			else
10559				sense_format = SSD_TYPE_FIXED;
10560
10561			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
10562					       sense_format);
10563			if (ua_type != CTL_UA_NONE) {
10564				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
10565				ctsio->io_hdr.status = CTL_SCSI_ERROR |
10566						       CTL_AUTOSENSE;
10567				ctsio->sense_len = SSD_FULL_SIZE;
10568				lun->pending_sense[initidx].ua_pending &=
10569					~ua_type;
10570				mtx_unlock(&ctl_softc->ctl_lock);
10571				ctl_done((union ctl_io *)ctsio);
10572				goto bailout;
10573			}
10574		}
10575	}
10576
10577
10578	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
10579		mtx_unlock(&ctl_softc->ctl_lock);
10580		ctl_done((union ctl_io *)ctsio);
10581		goto bailout;
10582	}
10583
10584	/*
10585	 * XXX CHD this is where we want to send IO to other side if
10586	 * this LUN is secondary on this SC. We will need to make a copy
10587	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
10588	 * the copy we send as FROM_OTHER.
10589	 * We also need to stuff the address of the original IO so we can
10590	 * find it easily. Something similar will need be done on the other
10591	 * side so when we are done we can find the copy.
10592	 */
10593	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10594		union ctl_ha_msg msg_info;
10595		int isc_retval;
10596
10597		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10598
10599		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
10600		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
10601#if 0
10602		printf("1. ctsio %p\n", ctsio);
10603#endif
10604		msg_info.hdr.serializing_sc = NULL;
10605		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
10606		msg_info.scsi.tag_num = ctsio->tag_num;
10607		msg_info.scsi.tag_type = ctsio->tag_type;
10608		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
10609
10610		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10611
10612		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10613		    (void *)&msg_info, sizeof(msg_info), 0)) >
10614		    CTL_HA_STATUS_SUCCESS) {
10615			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
10616			       isc_retval);
10617			printf("CTL:opcode is %x\n",opcode);
10618		} else {
10619#if 0
10620			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
10621#endif
10622		}
10623
10624		/*
10625		 * XXX KDM this I/O is off the incoming queue, but hasn't
10626		 * been inserted on any other queue.  We may need to come
10627		 * up with a holding queue while we wait for serialization
10628		 * so that we have an idea of what we're waiting for from
10629		 * the other side.
10630		 */
10631		goto bailout_unlock;
10632	}
10633
10634	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
10635			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
10636			      ctl_ooaq, ooa_links))) {
10637	case CTL_ACTION_BLOCK:
10638		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
10639		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
10640				  blocked_links);
10641		goto bailout_unlock;
10642		break; /* NOTREACHED */
10643	case CTL_ACTION_PASS:
10644	case CTL_ACTION_SKIP:
10645		goto queue_rtr;
10646		break; /* NOTREACHED */
10647	case CTL_ACTION_OVERLAP:
10648		ctl_set_overlapped_cmd(ctsio);
10649		mtx_unlock(&ctl_softc->ctl_lock);
10650		ctl_done((union ctl_io *)ctsio);
10651		goto bailout;
10652		break; /* NOTREACHED */
10653	case CTL_ACTION_OVERLAP_TAG:
10654		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
10655		mtx_unlock(&ctl_softc->ctl_lock);
10656		ctl_done((union ctl_io *)ctsio);
10657		goto bailout;
10658		break; /* NOTREACHED */
10659	case CTL_ACTION_ERROR:
10660	default:
10661		ctl_set_internal_failure(ctsio,
10662					 /*sks_valid*/ 0,
10663					 /*retry_count*/ 0);
10664		mtx_unlock(&ctl_softc->ctl_lock);
10665		ctl_done((union ctl_io *)ctsio);
10666		goto bailout;
10667		break; /* NOTREACHED */
10668	}
10669
10670	goto bailout_unlock;
10671
10672queue_rtr:
10673	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
10674	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
10675
10676bailout_unlock:
10677	mtx_unlock(&ctl_softc->ctl_lock);
10678
10679bailout:
10680	return (retval);
10681}
10682
10683static int
10684ctl_scsiio(struct ctl_scsiio *ctsio)
10685{
10686	int retval;
10687	struct ctl_cmd_entry *entry;
10688
10689	retval = CTL_RETVAL_COMPLETE;
10690
10691	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
10692
10693	entry = &ctl_cmd_table[ctsio->cdb[0]];
10694
10695	/*
10696	 * If this I/O has been aborted, just send it straight to
10697	 * ctl_done() without executing it.
10698	 */
10699	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
10700		ctl_done((union ctl_io *)ctsio);
10701		goto bailout;
10702	}
10703
10704	/*
10705	 * All the checks should have been handled by ctl_scsiio_precheck().
10706	 * We should be clear now to just execute the I/O.
10707	 */
10708	retval = entry->execute(ctsio);
10709
10710bailout:
10711	return (retval);
10712}
10713
10714/*
10715 * Since we only implement one target right now, a bus reset simply resets
10716 * our single target.
10717 */
10718static int
10719ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
10720{
10721	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
10722}
10723
10724static int
10725ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
10726		 ctl_ua_type ua_type)
10727{
10728	struct ctl_lun *lun;
10729	int retval;
10730
10731	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
10732		union ctl_ha_msg msg_info;
10733
10734		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
10735		msg_info.hdr.nexus = io->io_hdr.nexus;
10736		if (ua_type==CTL_UA_TARG_RESET)
10737			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
10738		else
10739			msg_info.task.task_action = CTL_TASK_BUS_RESET;
10740		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
10741		msg_info.hdr.original_sc = NULL;
10742		msg_info.hdr.serializing_sc = NULL;
10743		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10744		    (void *)&msg_info, sizeof(msg_info), 0)) {
10745		}
10746	}
10747	retval = 0;
10748
10749	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
10750		retval += ctl_lun_reset(lun, io, ua_type);
10751
10752	return (retval);
10753}
10754
10755/*
10756 * The LUN should always be set.  The I/O is optional, and is used to
10757 * distinguish between I/Os sent by this initiator, and by other
10758 * initiators.  We set unit attention for initiators other than this one.
10759 * SAM-3 is vague on this point.  It does say that a unit attention should
10760 * be established for other initiators when a LUN is reset (see section
10761 * 5.7.3), but it doesn't specifically say that the unit attention should
10762 * be established for this particular initiator when a LUN is reset.  Here
10763 * is the relevant text, from SAM-3 rev 8:
10764 *
10765 * 5.7.2 When a SCSI initiator port aborts its own tasks
10766 *
10767 * When a SCSI initiator port causes its own task(s) to be aborted, no
10768 * notification that the task(s) have been aborted shall be returned to
10769 * the SCSI initiator port other than the completion response for the
10770 * command or task management function action that caused the task(s) to
10771 * be aborted and notification(s) associated with related effects of the
10772 * action (e.g., a reset unit attention condition).
10773 *
10774 * XXX KDM for now, we're setting unit attention for all initiators.
10775 */
10776static int
10777ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
10778{
10779	union ctl_io *xio;
10780#if 0
10781	uint32_t initindex;
10782#endif
10783	int i;
10784
10785	/*
10786	 * Run through the OOA queue and abort each I/O.
10787	 */
10788#if 0
10789	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10790#endif
10791	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10792	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10793		xio->io_hdr.flags |= CTL_FLAG_ABORT;
10794	}
10795
10796	/*
10797	 * This version sets unit attention for every
10798	 */
10799#if 0
10800	initindex = ctl_get_initindex(&io->io_hdr.nexus);
10801	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10802		if (initindex == i)
10803			continue;
10804		lun->pending_sense[i].ua_pending |= ua_type;
10805	}
10806#endif
10807
10808	/*
10809	 * A reset (any kind, really) clears reservations established with
10810	 * RESERVE/RELEASE.  It does not clear reservations established
10811	 * with PERSISTENT RESERVE OUT, but we don't support that at the
10812	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
10813	 * reservations made with the RESERVE/RELEASE commands, because
10814	 * those commands are obsolete in SPC-3.
10815	 */
10816	lun->flags &= ~CTL_LUN_RESERVED;
10817
10818	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10819		ctl_clear_mask(lun->have_ca, i);
10820		lun->pending_sense[i].ua_pending |= ua_type;
10821	}
10822
10823	return (0);
10824}
10825
10826static int
10827ctl_abort_task(union ctl_io *io)
10828{
10829	union ctl_io *xio;
10830	struct ctl_lun *lun;
10831	struct ctl_softc *ctl_softc;
10832#if 0
10833	struct sbuf sb;
10834	char printbuf[128];
10835#endif
10836	int found;
10837	uint32_t targ_lun;
10838
10839	ctl_softc = control_softc;
10840	found = 0;
10841
10842	/*
10843	 * Look up the LUN.
10844	 */
10845	targ_lun = io->io_hdr.nexus.targ_lun;
10846	if (io->io_hdr.nexus.lun_map_fn != NULL)
10847		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
10848	if ((targ_lun < CTL_MAX_LUNS)
10849	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
10850		lun = ctl_softc->ctl_luns[targ_lun];
10851	else
10852		goto bailout;
10853
10854#if 0
10855	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
10856	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
10857#endif
10858
10859	/*
10860	 * Run through the OOA queue and attempt to find the given I/O.
10861	 * The target port, initiator ID, tag type and tag number have to
10862	 * match the values that we got from the initiator.  If we have an
10863	 * untagged command to abort, simply abort the first untagged command
10864	 * we come to.  We only allow one untagged command at a time of course.
10865	 */
10866#if 0
10867	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
10868#endif
10869	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
10870	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
10871#if 0
10872		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
10873
10874		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
10875			    lun->lun, xio->scsiio.tag_num,
10876			    xio->scsiio.tag_type,
10877			    (xio->io_hdr.blocked_links.tqe_prev
10878			    == NULL) ? "" : " BLOCKED",
10879			    (xio->io_hdr.flags &
10880			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
10881			    (xio->io_hdr.flags &
10882			    CTL_FLAG_ABORT) ? " ABORT" : "",
10883			    (xio->io_hdr.flags &
10884			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
10885		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
10886		sbuf_finish(&sb);
10887		printf("%s\n", sbuf_data(&sb));
10888#endif
10889
10890		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
10891		 && (xio->io_hdr.nexus.initid.id ==
10892		     io->io_hdr.nexus.initid.id)) {
10893			/*
10894			 * If the abort says that the task is untagged, the
10895			 * task in the queue must be untagged.  Otherwise,
10896			 * we just check to see whether the tag numbers
10897			 * match.  This is because the QLogic firmware
10898			 * doesn't pass back the tag type in an abort
10899			 * request.
10900			 */
10901#if 0
10902			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
10903			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
10904			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
10905#endif
10906			/*
10907			 * XXX KDM we've got problems with FC, because it
10908			 * doesn't send down a tag type with aborts.  So we
10909			 * can only really go by the tag number...
10910			 * This may cause problems with parallel SCSI.
10911			 * Need to figure that out!!
10912			 */
10913			if (xio->scsiio.tag_num == io->taskio.tag_num) {
10914				xio->io_hdr.flags |= CTL_FLAG_ABORT;
10915				found = 1;
10916				if ((io->io_hdr.flags &
10917				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
10918				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
10919					union ctl_ha_msg msg_info;
10920
10921					io->io_hdr.flags |=
10922					                CTL_FLAG_SENT_2OTHER_SC;
10923					msg_info.hdr.nexus = io->io_hdr.nexus;
10924					msg_info.task.task_action =
10925						CTL_TASK_ABORT_TASK;
10926					msg_info.task.tag_num =
10927						io->taskio.tag_num;
10928					msg_info.task.tag_type =
10929						io->taskio.tag_type;
10930					msg_info.hdr.msg_type =
10931						CTL_MSG_MANAGE_TASKS;
10932					msg_info.hdr.original_sc = NULL;
10933					msg_info.hdr.serializing_sc = NULL;
10934#if 0
10935					printf("Sent Abort to other side\n");
10936#endif
10937					if (CTL_HA_STATUS_SUCCESS !=
10938					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10939		    				(void *)&msg_info,
10940						sizeof(msg_info), 0)) {
10941					}
10942				}
10943#if 0
10944				printf("ctl_abort_task: found I/O to abort\n");
10945#endif
10946				break;
10947			}
10948		}
10949	}
10950
10951bailout:
10952
10953	if (found == 0) {
10954		/*
10955		 * This isn't really an error.  It's entirely possible for
10956		 * the abort and command completion to cross on the wire.
10957		 * This is more of an informative/diagnostic error.
10958		 */
10959#if 0
10960		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
10961		       "%d:%d:%d:%d tag %d type %d\n",
10962		       io->io_hdr.nexus.initid.id,
10963		       io->io_hdr.nexus.targ_port,
10964		       io->io_hdr.nexus.targ_target.id,
10965		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
10966		       io->taskio.tag_type);
10967#endif
10968		return (1);
10969	} else
10970		return (0);
10971}
10972
10973/*
10974 * This routine cannot block!  It must be callable from an interrupt
10975 * handler as well as from the work thread.
10976 */
10977static void
10978ctl_run_task_queue(struct ctl_softc *ctl_softc)
10979{
10980	union ctl_io *io, *next_io;
10981
10982	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
10983
10984	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
10985
10986	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
10987	     io != NULL; io = next_io) {
10988		int retval;
10989		const char *task_desc;
10990
10991		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10992
10993		retval = 0;
10994
10995		switch (io->io_hdr.io_type) {
10996		case CTL_IO_TASK: {
10997			task_desc = ctl_scsi_task_string(&io->taskio);
10998			if (task_desc != NULL) {
10999#ifdef NEEDTOPORT
11000				csevent_log(CSC_CTL | CSC_SHELF_SW |
11001					    CTL_TASK_REPORT,
11002					    csevent_LogType_Trace,
11003					    csevent_Severity_Information,
11004					    csevent_AlertLevel_Green,
11005					    csevent_FRU_Firmware,
11006					    csevent_FRU_Unknown,
11007					    "CTL: received task: %s",task_desc);
11008#endif
11009			} else {
11010#ifdef NEEDTOPORT
11011				csevent_log(CSC_CTL | CSC_SHELF_SW |
11012					    CTL_TASK_REPORT,
11013					    csevent_LogType_Trace,
11014					    csevent_Severity_Information,
11015					    csevent_AlertLevel_Green,
11016					    csevent_FRU_Firmware,
11017					    csevent_FRU_Unknown,
11018					    "CTL: received unknown task "
11019					    "type: %d (%#x)",
11020					    io->taskio.task_action,
11021					    io->taskio.task_action);
11022#endif
11023			}
11024			switch (io->taskio.task_action) {
11025			case CTL_TASK_ABORT_TASK:
11026				retval = ctl_abort_task(io);
11027				break;
11028			case CTL_TASK_ABORT_TASK_SET:
11029				break;
11030			case CTL_TASK_CLEAR_ACA:
11031				break;
11032			case CTL_TASK_CLEAR_TASK_SET:
11033				break;
11034			case CTL_TASK_LUN_RESET: {
11035				struct ctl_lun *lun;
11036				uint32_t targ_lun;
11037				int retval;
11038
11039				targ_lun = io->io_hdr.nexus.targ_lun;
11040				if (io->io_hdr.nexus.lun_map_fn != NULL)
11041					targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11042
11043				if ((targ_lun < CTL_MAX_LUNS)
11044				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11045					lun = ctl_softc->ctl_luns[targ_lun];
11046				else {
11047					retval = 1;
11048					break;
11049				}
11050
11051				if (!(io->io_hdr.flags &
11052				    CTL_FLAG_FROM_OTHER_SC)) {
11053					union ctl_ha_msg msg_info;
11054
11055					io->io_hdr.flags |=
11056						CTL_FLAG_SENT_2OTHER_SC;
11057					msg_info.hdr.msg_type =
11058						CTL_MSG_MANAGE_TASKS;
11059					msg_info.hdr.nexus = io->io_hdr.nexus;
11060					msg_info.task.task_action =
11061						CTL_TASK_LUN_RESET;
11062					msg_info.hdr.original_sc = NULL;
11063					msg_info.hdr.serializing_sc = NULL;
11064					if (CTL_HA_STATUS_SUCCESS !=
11065					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11066					    (void *)&msg_info,
11067					    sizeof(msg_info), 0)) {
11068					}
11069				}
11070
11071				retval = ctl_lun_reset(lun, io,
11072						       CTL_UA_LUN_RESET);
11073				break;
11074			}
11075			case CTL_TASK_TARGET_RESET:
11076				retval = ctl_target_reset(ctl_softc, io,
11077							  CTL_UA_TARG_RESET);
11078				break;
11079			case CTL_TASK_BUS_RESET:
11080				retval = ctl_bus_reset(ctl_softc, io);
11081				break;
11082			case CTL_TASK_PORT_LOGIN:
11083				break;
11084			case CTL_TASK_PORT_LOGOUT:
11085				break;
11086			default:
11087				printf("ctl_run_task_queue: got unknown task "
11088				       "management event %d\n",
11089				       io->taskio.task_action);
11090				break;
11091			}
11092			if (retval == 0)
11093				io->io_hdr.status = CTL_SUCCESS;
11094			else
11095				io->io_hdr.status = CTL_ERROR;
11096
11097			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11098				      ctl_io_hdr, links);
11099			/*
11100			 * This will queue this I/O to the done queue, but the
11101			 * work thread won't be able to process it until we
11102			 * return and the lock is released.
11103			 */
11104			ctl_done_lock(io, /*have_lock*/ 1);
11105			break;
11106		}
11107		default: {
11108
11109			printf("%s: invalid I/O type %d msg %d cdb %x"
11110			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11111			       __func__, io->io_hdr.io_type,
11112			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11113			       (uintmax_t)io->io_hdr.nexus.initid.id,
11114			       io->io_hdr.nexus.targ_port,
11115			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11116			       io->io_hdr.nexus.targ_lun /* XXX */,
11117			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11118			       io->taskio.tag_num : io->scsiio.tag_num);
11119			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11120				      ctl_io_hdr, links);
11121			ctl_free_io_internal(io, 1);
11122			break;
11123		}
11124		}
11125	}
11126
11127	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11128}
11129
11130/*
11131 * For HA operation.  Handle commands that come in from the other
11132 * controller.
11133 */
11134static void
11135ctl_handle_isc(union ctl_io *io)
11136{
11137	int free_io;
11138	struct ctl_lun *lun;
11139	struct ctl_softc *ctl_softc;
11140	uint32_t targ_lun;
11141
11142	ctl_softc = control_softc;
11143
11144	targ_lun = io->io_hdr.nexus.targ_lun;
11145	if (io->io_hdr.nexus.lun_map_fn != NULL)
11146		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11147	lun = ctl_softc->ctl_luns[targ_lun];
11148
11149	switch (io->io_hdr.msg_type) {
11150	case CTL_MSG_SERIALIZE:
11151		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11152						     /*have_lock*/ 0);
11153		break;
11154	case CTL_MSG_R2R: {
11155		uint8_t opcode;
11156		struct ctl_cmd_entry *entry;
11157
11158		/*
11159		 * This is only used in SER_ONLY mode.
11160		 */
11161		free_io = 0;
11162		opcode = io->scsiio.cdb[0];
11163		entry = &ctl_cmd_table[opcode];
11164		mtx_lock(&ctl_softc->ctl_lock);
11165		if (ctl_scsiio_lun_check(ctl_softc, lun,
11166		    entry, (struct ctl_scsiio *)io) != 0) {
11167			ctl_done_lock(io, /*have_lock*/ 1);
11168			mtx_unlock(&ctl_softc->ctl_lock);
11169			break;
11170		}
11171		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11172		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11173				   &io->io_hdr, links);
11174		mtx_unlock(&ctl_softc->ctl_lock);
11175		break;
11176	}
11177	case CTL_MSG_FINISH_IO:
11178		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11179			free_io = 0;
11180			ctl_done_lock(io, /*have_lock*/ 0);
11181		} else {
11182			free_io = 1;
11183			mtx_lock(&ctl_softc->ctl_lock);
11184			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11185				     ooa_links);
11186			STAILQ_REMOVE(&ctl_softc->task_queue,
11187				      &io->io_hdr, ctl_io_hdr, links);
11188			ctl_check_blocked(lun);
11189			mtx_unlock(&ctl_softc->ctl_lock);
11190		}
11191		break;
11192	case CTL_MSG_PERS_ACTION:
11193		ctl_hndl_per_res_out_on_other_sc(
11194			(union ctl_ha_msg *)&io->presio.pr_msg);
11195		free_io = 1;
11196		break;
11197	case CTL_MSG_BAD_JUJU:
11198		free_io = 0;
11199		ctl_done_lock(io, /*have_lock*/ 0);
11200		break;
11201	case CTL_MSG_DATAMOVE:
11202		/* Only used in XFER mode */
11203		free_io = 0;
11204		ctl_datamove_remote(io);
11205		break;
11206	case CTL_MSG_DATAMOVE_DONE:
11207		/* Only used in XFER mode */
11208		free_io = 0;
11209		io->scsiio.be_move_done(io);
11210		break;
11211	default:
11212		free_io = 1;
11213		printf("%s: Invalid message type %d\n",
11214		       __func__, io->io_hdr.msg_type);
11215		break;
11216	}
11217	if (free_io)
11218		ctl_free_io_internal(io, 0);
11219
11220}
11221
11222
11223/*
11224 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11225 * there is no match.
11226 */
11227static ctl_lun_error_pattern
11228ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11229{
11230	struct ctl_cmd_entry *entry;
11231	ctl_lun_error_pattern filtered_pattern, pattern;
11232	uint8_t opcode;
11233
11234	pattern = desc->error_pattern;
11235
11236	/*
11237	 * XXX KDM we need more data passed into this function to match a
11238	 * custom pattern, and we actually need to implement custom pattern
11239	 * matching.
11240	 */
11241	if (pattern & CTL_LUN_PAT_CMD)
11242		return (CTL_LUN_PAT_CMD);
11243
11244	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11245		return (CTL_LUN_PAT_ANY);
11246
11247	opcode = ctsio->cdb[0];
11248	entry = &ctl_cmd_table[opcode];
11249
11250	filtered_pattern = entry->pattern & pattern;
11251
11252	/*
11253	 * If the user requested specific flags in the pattern (e.g.
11254	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11255	 * flags.
11256	 *
11257	 * If the user did not specify any flags, it doesn't matter whether
11258	 * or not the command supports the flags.
11259	 */
11260	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11261	     (pattern & ~CTL_LUN_PAT_MASK))
11262		return (CTL_LUN_PAT_NONE);
11263
11264	/*
11265	 * If the user asked for a range check, see if the requested LBA
11266	 * range overlaps with this command's LBA range.
11267	 */
11268	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11269		uint64_t lba1;
11270		uint32_t len1;
11271		ctl_action action;
11272		int retval;
11273
11274		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11275		if (retval != 0)
11276			return (CTL_LUN_PAT_NONE);
11277
11278		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11279					      desc->lba_range.len);
11280		/*
11281		 * A "pass" means that the LBA ranges don't overlap, so
11282		 * this doesn't match the user's range criteria.
11283		 */
11284		if (action == CTL_ACTION_PASS)
11285			return (CTL_LUN_PAT_NONE);
11286	}
11287
11288	return (filtered_pattern);
11289}
11290
11291static void
11292ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11293{
11294	struct ctl_error_desc *desc, *desc2;
11295
11296	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
11297
11298	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11299		ctl_lun_error_pattern pattern;
11300		/*
11301		 * Check to see whether this particular command matches
11302		 * the pattern in the descriptor.
11303		 */
11304		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11305		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11306			continue;
11307
11308		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11309		case CTL_LUN_INJ_ABORTED:
11310			ctl_set_aborted(&io->scsiio);
11311			break;
11312		case CTL_LUN_INJ_MEDIUM_ERR:
11313			ctl_set_medium_error(&io->scsiio);
11314			break;
11315		case CTL_LUN_INJ_UA:
11316			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11317			 * OCCURRED */
11318			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11319			break;
11320		case CTL_LUN_INJ_CUSTOM:
11321			/*
11322			 * We're assuming the user knows what he is doing.
11323			 * Just copy the sense information without doing
11324			 * checks.
11325			 */
11326			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11327			      ctl_min(sizeof(desc->custom_sense),
11328				      sizeof(io->scsiio.sense_data)));
11329			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11330			io->scsiio.sense_len = SSD_FULL_SIZE;
11331			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11332			break;
11333		case CTL_LUN_INJ_NONE:
11334		default:
11335			/*
11336			 * If this is an error injection type we don't know
11337			 * about, clear the continuous flag (if it is set)
11338			 * so it will get deleted below.
11339			 */
11340			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11341			break;
11342		}
11343		/*
11344		 * By default, each error injection action is a one-shot
11345		 */
11346		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11347			continue;
11348
11349		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11350
11351		free(desc, M_CTL);
11352	}
11353}
11354
11355#ifdef CTL_IO_DELAY
11356static void
11357ctl_datamove_timer_wakeup(void *arg)
11358{
11359	union ctl_io *io;
11360
11361	io = (union ctl_io *)arg;
11362
11363	ctl_datamove(io);
11364}
11365#endif /* CTL_IO_DELAY */
11366
11367void
11368ctl_datamove(union ctl_io *io)
11369{
11370	void (*fe_datamove)(union ctl_io *io);
11371
11372	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11373
11374	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11375
11376#ifdef CTL_TIME_IO
11377	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11378		char str[256];
11379		char path_str[64];
11380		struct sbuf sb;
11381
11382		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11383		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11384
11385		sbuf_cat(&sb, path_str);
11386		switch (io->io_hdr.io_type) {
11387		case CTL_IO_SCSI:
11388			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11389			sbuf_printf(&sb, "\n");
11390			sbuf_cat(&sb, path_str);
11391			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11392				    io->scsiio.tag_num, io->scsiio.tag_type);
11393			break;
11394		case CTL_IO_TASK:
11395			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11396				    "Tag Type: %d\n", io->taskio.task_action,
11397				    io->taskio.tag_num, io->taskio.tag_type);
11398			break;
11399		default:
11400			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11401			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11402			break;
11403		}
11404		sbuf_cat(&sb, path_str);
11405		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
11406			    (intmax_t)time_uptime - io->io_hdr.start_time);
11407		sbuf_finish(&sb);
11408		printf("%s", sbuf_data(&sb));
11409	}
11410#endif /* CTL_TIME_IO */
11411
11412	mtx_lock(&control_softc->ctl_lock);
11413#ifdef CTL_IO_DELAY
11414	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
11415		struct ctl_lun *lun;
11416
11417		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11418
11419		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
11420	} else {
11421		struct ctl_lun *lun;
11422
11423		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
11424		if ((lun != NULL)
11425		 && (lun->delay_info.datamove_delay > 0)) {
11426			struct callout *callout;
11427
11428			callout = (struct callout *)&io->io_hdr.timer_bytes;
11429			callout_init(callout, /*mpsafe*/ 1);
11430			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
11431			callout_reset(callout,
11432				      lun->delay_info.datamove_delay * hz,
11433				      ctl_datamove_timer_wakeup, io);
11434			if (lun->delay_info.datamove_type ==
11435			    CTL_DELAY_TYPE_ONESHOT)
11436				lun->delay_info.datamove_delay = 0;
11437			mtx_unlock(&control_softc->ctl_lock);
11438			return;
11439		}
11440	}
11441#endif
11442	/*
11443	 * If we have any pending task management commands, process them
11444	 * first.  This is necessary to eliminate a race condition with the
11445	 * FETD:
11446	 *
11447	 * - FETD submits a task management command, like an abort.
11448	 * - Back end calls fe_datamove() to move the data for the aborted
11449	 *   command.  The FETD can't really accept it, but if it did, it
11450	 *   would end up transmitting data for a command that the initiator
11451	 *   told us to abort.
11452	 *
11453	 * We close the race by processing all pending task management
11454	 * commands here (we can't block!), and then check this I/O to see
11455	 * if it has been aborted.  If so, return it to the back end with
11456	 * bad status, so the back end can say return an error to the back end
11457	 * and then when the back end returns an error, we can return the
11458	 * aborted command to the FETD, so it can clean up its resources.
11459	 */
11460	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
11461		ctl_run_task_queue(control_softc);
11462
11463	/*
11464	 * This command has been aborted.  Set the port status, so we fail
11465	 * the data move.
11466	 */
11467	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
11468		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
11469		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
11470		       io->io_hdr.nexus.targ_port,
11471		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11472		       io->io_hdr.nexus.targ_lun);
11473		io->io_hdr.status = CTL_CMD_ABORTED;
11474		io->io_hdr.port_status = 31337;
11475		mtx_unlock(&control_softc->ctl_lock);
11476		/*
11477		 * Note that the backend, in this case, will get the
11478		 * callback in its context.  In other cases it may get
11479		 * called in the frontend's interrupt thread context.
11480		 */
11481		io->scsiio.be_move_done(io);
11482		return;
11483	}
11484
11485	/*
11486	 * If we're in XFER mode and this I/O is from the other shelf
11487	 * controller, we need to send the DMA to the other side to
11488	 * actually transfer the data to/from the host.  In serialize only
11489	 * mode the transfer happens below CTL and ctl_datamove() is only
11490	 * called on the machine that originally received the I/O.
11491	 */
11492	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
11493	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11494		union ctl_ha_msg msg;
11495		uint32_t sg_entries_sent;
11496		int do_sg_copy;
11497		int i;
11498
11499		memset(&msg, 0, sizeof(msg));
11500		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
11501		msg.hdr.original_sc = io->io_hdr.original_sc;
11502		msg.hdr.serializing_sc = io;
11503		msg.hdr.nexus = io->io_hdr.nexus;
11504		msg.dt.flags = io->io_hdr.flags;
11505		/*
11506		 * We convert everything into a S/G list here.  We can't
11507		 * pass by reference, only by value between controllers.
11508		 * So we can't pass a pointer to the S/G list, only as many
11509		 * S/G entries as we can fit in here.  If it's possible for
11510		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
11511		 * then we need to break this up into multiple transfers.
11512		 */
11513		if (io->scsiio.kern_sg_entries == 0) {
11514			msg.dt.kern_sg_entries = 1;
11515			/*
11516			 * If this is in cached memory, flush the cache
11517			 * before we send the DMA request to the other
11518			 * controller.  We want to do this in either the
11519			 * read or the write case.  The read case is
11520			 * straightforward.  In the write case, we want to
11521			 * make sure nothing is in the local cache that
11522			 * could overwrite the DMAed data.
11523			 */
11524			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11525				/*
11526				 * XXX KDM use bus_dmamap_sync() here.
11527				 */
11528			}
11529
11530			/*
11531			 * Convert to a physical address if this is a
11532			 * virtual address.
11533			 */
11534			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
11535				msg.dt.sg_list[0].addr =
11536					io->scsiio.kern_data_ptr;
11537			} else {
11538				/*
11539				 * XXX KDM use busdma here!
11540				 */
11541#if 0
11542				msg.dt.sg_list[0].addr = (void *)
11543					vtophys(io->scsiio.kern_data_ptr);
11544#endif
11545			}
11546
11547			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
11548			do_sg_copy = 0;
11549		} else {
11550			struct ctl_sg_entry *sgl;
11551
11552			do_sg_copy = 1;
11553			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
11554			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
11555			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
11556				/*
11557				 * XXX KDM use bus_dmamap_sync() here.
11558				 */
11559			}
11560		}
11561
11562		msg.dt.kern_data_len = io->scsiio.kern_data_len;
11563		msg.dt.kern_total_len = io->scsiio.kern_total_len;
11564		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
11565		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
11566		msg.dt.sg_sequence = 0;
11567
11568		/*
11569		 * Loop until we've sent all of the S/G entries.  On the
11570		 * other end, we'll recompose these S/G entries into one
11571		 * contiguous list before passing it to the
11572		 */
11573		for (sg_entries_sent = 0; sg_entries_sent <
11574		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
11575			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
11576				sizeof(msg.dt.sg_list[0])),
11577				msg.dt.kern_sg_entries - sg_entries_sent);
11578
11579			if (do_sg_copy != 0) {
11580				struct ctl_sg_entry *sgl;
11581				int j;
11582
11583				sgl = (struct ctl_sg_entry *)
11584					io->scsiio.kern_data_ptr;
11585				/*
11586				 * If this is in cached memory, flush the cache
11587				 * before we send the DMA request to the other
11588				 * controller.  We want to do this in either
11589				 * the * read or the write case.  The read
11590				 * case is straightforward.  In the write
11591				 * case, we want to make sure nothing is
11592				 * in the local cache that could overwrite
11593				 * the DMAed data.
11594				 */
11595
11596				for (i = sg_entries_sent, j = 0;
11597				     i < msg.dt.cur_sg_entries; i++, j++) {
11598					if ((io->io_hdr.flags &
11599					     CTL_FLAG_NO_DATASYNC) == 0) {
11600						/*
11601						 * XXX KDM use bus_dmamap_sync()
11602						 */
11603					}
11604					if ((io->io_hdr.flags &
11605					     CTL_FLAG_BUS_ADDR) == 0) {
11606						/*
11607						 * XXX KDM use busdma.
11608						 */
11609#if 0
11610						msg.dt.sg_list[j].addr =(void *)
11611						       vtophys(sgl[i].addr);
11612#endif
11613					} else {
11614						msg.dt.sg_list[j].addr =
11615							sgl[i].addr;
11616					}
11617					msg.dt.sg_list[j].len = sgl[i].len;
11618				}
11619			}
11620
11621			sg_entries_sent += msg.dt.cur_sg_entries;
11622			if (sg_entries_sent >= msg.dt.kern_sg_entries)
11623				msg.dt.sg_last = 1;
11624			else
11625				msg.dt.sg_last = 0;
11626
11627			/*
11628			 * XXX KDM drop and reacquire the lock here?
11629			 */
11630			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
11631			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
11632				/*
11633				 * XXX do something here.
11634				 */
11635			}
11636
11637			msg.dt.sent_sg_entries = sg_entries_sent;
11638		}
11639		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11640		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
11641			ctl_failover_io(io, /*have_lock*/ 1);
11642
11643	} else {
11644
11645		/*
11646		 * Lookup the fe_datamove() function for this particular
11647		 * front end.
11648		 */
11649		fe_datamove =
11650		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11651		mtx_unlock(&control_softc->ctl_lock);
11652
11653		fe_datamove(io);
11654	}
11655}
11656
11657static void
11658ctl_send_datamove_done(union ctl_io *io, int have_lock)
11659{
11660	union ctl_ha_msg msg;
11661	int isc_status;
11662
11663	memset(&msg, 0, sizeof(msg));
11664
11665	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
11666	msg.hdr.original_sc = io;
11667	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
11668	msg.hdr.nexus = io->io_hdr.nexus;
11669	msg.hdr.status = io->io_hdr.status;
11670	msg.scsi.tag_num = io->scsiio.tag_num;
11671	msg.scsi.tag_type = io->scsiio.tag_type;
11672	msg.scsi.scsi_status = io->scsiio.scsi_status;
11673	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
11674	       sizeof(io->scsiio.sense_data));
11675	msg.scsi.sense_len = io->scsiio.sense_len;
11676	msg.scsi.sense_residual = io->scsiio.sense_residual;
11677	msg.scsi.fetd_status = io->io_hdr.port_status;
11678	msg.scsi.residual = io->scsiio.residual;
11679	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11680
11681	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
11682		ctl_failover_io(io, /*have_lock*/ have_lock);
11683		return;
11684	}
11685
11686	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
11687	if (isc_status > CTL_HA_STATUS_SUCCESS) {
11688		/* XXX do something if this fails */
11689	}
11690
11691}
11692
11693/*
11694 * The DMA to the remote side is done, now we need to tell the other side
11695 * we're done so it can continue with its data movement.
11696 */
11697static void
11698ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
11699{
11700	union ctl_io *io;
11701
11702	io = rq->context;
11703
11704	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11705		printf("%s: ISC DMA write failed with error %d", __func__,
11706		       rq->ret);
11707		ctl_set_internal_failure(&io->scsiio,
11708					 /*sks_valid*/ 1,
11709					 /*retry_count*/ rq->ret);
11710	}
11711
11712	ctl_dt_req_free(rq);
11713
11714	/*
11715	 * In this case, we had to malloc the memory locally.  Free it.
11716	 */
11717	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11718		int i;
11719		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11720			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11721	}
11722	/*
11723	 * The data is in local and remote memory, so now we need to send
11724	 * status (good or back) back to the other side.
11725	 */
11726	ctl_send_datamove_done(io, /*have_lock*/ 0);
11727}
11728
11729/*
11730 * We've moved the data from the host/controller into local memory.  Now we
11731 * need to push it over to the remote controller's memory.
11732 */
11733static int
11734ctl_datamove_remote_dm_write_cb(union ctl_io *io)
11735{
11736	int retval;
11737
11738	retval = 0;
11739
11740	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
11741					  ctl_datamove_remote_write_cb);
11742
11743	return (retval);
11744}
11745
11746static void
11747ctl_datamove_remote_write(union ctl_io *io)
11748{
11749	int retval;
11750	void (*fe_datamove)(union ctl_io *io);
11751
11752	/*
11753	 * - Get the data from the host/HBA into local memory.
11754	 * - DMA memory from the local controller to the remote controller.
11755	 * - Send status back to the remote controller.
11756	 */
11757
11758	retval = ctl_datamove_remote_sgl_setup(io);
11759	if (retval != 0)
11760		return;
11761
11762	/* Switch the pointer over so the FETD knows what to do */
11763	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11764
11765	/*
11766	 * Use a custom move done callback, since we need to send completion
11767	 * back to the other controller, not to the backend on this side.
11768	 */
11769	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
11770
11771	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11772
11773	fe_datamove(io);
11774
11775	return;
11776
11777}
11778
11779static int
11780ctl_datamove_remote_dm_read_cb(union ctl_io *io)
11781{
11782#if 0
11783	char str[256];
11784	char path_str[64];
11785	struct sbuf sb;
11786#endif
11787
11788	/*
11789	 * In this case, we had to malloc the memory locally.  Free it.
11790	 */
11791	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
11792		int i;
11793		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11794			free(io->io_hdr.local_sglist[i].addr, M_CTL);
11795	}
11796
11797#if 0
11798	scsi_path_string(io, path_str, sizeof(path_str));
11799	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11800	sbuf_cat(&sb, path_str);
11801	scsi_command_string(&io->scsiio, NULL, &sb);
11802	sbuf_printf(&sb, "\n");
11803	sbuf_cat(&sb, path_str);
11804	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11805		    io->scsiio.tag_num, io->scsiio.tag_type);
11806	sbuf_cat(&sb, path_str);
11807	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
11808		    io->io_hdr.flags, io->io_hdr.status);
11809	sbuf_finish(&sb);
11810	printk("%s", sbuf_data(&sb));
11811#endif
11812
11813
11814	/*
11815	 * The read is done, now we need to send status (good or bad) back
11816	 * to the other side.
11817	 */
11818	ctl_send_datamove_done(io, /*have_lock*/ 0);
11819
11820	return (0);
11821}
11822
11823static void
11824ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
11825{
11826	union ctl_io *io;
11827	void (*fe_datamove)(union ctl_io *io);
11828
11829	io = rq->context;
11830
11831	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
11832		printf("%s: ISC DMA read failed with error %d", __func__,
11833		       rq->ret);
11834		ctl_set_internal_failure(&io->scsiio,
11835					 /*sks_valid*/ 1,
11836					 /*retry_count*/ rq->ret);
11837	}
11838
11839	ctl_dt_req_free(rq);
11840
11841	/* Switch the pointer over so the FETD knows what to do */
11842	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
11843
11844	/*
11845	 * Use a custom move done callback, since we need to send completion
11846	 * back to the other controller, not to the backend on this side.
11847	 */
11848	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
11849
11850	/* XXX KDM add checks like the ones in ctl_datamove? */
11851
11852	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
11853
11854	fe_datamove(io);
11855}
11856
11857static int
11858ctl_datamove_remote_sgl_setup(union ctl_io *io)
11859{
11860	struct ctl_sg_entry *local_sglist, *remote_sglist;
11861	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
11862	struct ctl_softc *softc;
11863	int retval;
11864	int i;
11865
11866	retval = 0;
11867	softc = control_softc;
11868
11869	local_sglist = io->io_hdr.local_sglist;
11870	local_dma_sglist = io->io_hdr.local_dma_sglist;
11871	remote_sglist = io->io_hdr.remote_sglist;
11872	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
11873
11874	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
11875		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
11876			local_sglist[i].len = remote_sglist[i].len;
11877
11878			/*
11879			 * XXX Detect the situation where the RS-level I/O
11880			 * redirector on the other side has already read the
11881			 * data off of the AOR RS on this side, and
11882			 * transferred it to remote (mirror) memory on the
11883			 * other side.  Since we already have the data in
11884			 * memory here, we just need to use it.
11885			 *
11886			 * XXX KDM this can probably be removed once we
11887			 * get the cache device code in and take the
11888			 * current AOR implementation out.
11889			 */
11890#ifdef NEEDTOPORT
11891			if ((remote_sglist[i].addr >=
11892			     (void *)vtophys(softc->mirr->addr))
11893			 && (remote_sglist[i].addr <
11894			     ((void *)vtophys(softc->mirr->addr) +
11895			     CacheMirrorOffset))) {
11896				local_sglist[i].addr = remote_sglist[i].addr -
11897					CacheMirrorOffset;
11898				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
11899				     CTL_FLAG_DATA_IN)
11900					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
11901			} else {
11902				local_sglist[i].addr = remote_sglist[i].addr +
11903					CacheMirrorOffset;
11904			}
11905#endif
11906#if 0
11907			printf("%s: local %p, remote %p, len %d\n",
11908			       __func__, local_sglist[i].addr,
11909			       remote_sglist[i].addr, local_sglist[i].len);
11910#endif
11911		}
11912	} else {
11913		uint32_t len_to_go;
11914
11915		/*
11916		 * In this case, we don't have automatically allocated
11917		 * memory for this I/O on this controller.  This typically
11918		 * happens with internal CTL I/O -- e.g. inquiry, mode
11919		 * sense, etc.  Anything coming from RAIDCore will have
11920		 * a mirror area available.
11921		 */
11922		len_to_go = io->scsiio.kern_data_len;
11923
11924		/*
11925		 * Clear the no datasync flag, we have to use malloced
11926		 * buffers.
11927		 */
11928		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
11929
11930		/*
11931		 * The difficult thing here is that the size of the various
11932		 * S/G segments may be different than the size from the
11933		 * remote controller.  That'll make it harder when DMAing
11934		 * the data back to the other side.
11935		 */
11936		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
11937		     sizeof(io->io_hdr.remote_sglist[0])) &&
11938		     (len_to_go > 0); i++) {
11939			local_sglist[i].len = ctl_min(len_to_go, 131072);
11940			CTL_SIZE_8B(local_dma_sglist[i].len,
11941				    local_sglist[i].len);
11942			local_sglist[i].addr =
11943				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
11944
11945			local_dma_sglist[i].addr = local_sglist[i].addr;
11946
11947			if (local_sglist[i].addr == NULL) {
11948				int j;
11949
11950				printf("malloc failed for %zd bytes!",
11951				       local_dma_sglist[i].len);
11952				for (j = 0; j < i; j++) {
11953					free(local_sglist[j].addr, M_CTL);
11954				}
11955				ctl_set_internal_failure(&io->scsiio,
11956							 /*sks_valid*/ 1,
11957							 /*retry_count*/ 4857);
11958				retval = 1;
11959				goto bailout_error;
11960
11961			}
11962			/* XXX KDM do we need a sync here? */
11963
11964			len_to_go -= local_sglist[i].len;
11965		}
11966		/*
11967		 * Reset the number of S/G entries accordingly.  The
11968		 * original number of S/G entries is available in
11969		 * rem_sg_entries.
11970		 */
11971		io->scsiio.kern_sg_entries = i;
11972
11973#if 0
11974		printf("%s: kern_sg_entries = %d\n", __func__,
11975		       io->scsiio.kern_sg_entries);
11976		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
11977			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
11978			       local_sglist[i].addr, local_sglist[i].len,
11979			       local_dma_sglist[i].len);
11980#endif
11981	}
11982
11983
11984	return (retval);
11985
11986bailout_error:
11987
11988	ctl_send_datamove_done(io, /*have_lock*/ 0);
11989
11990	return (retval);
11991}
11992
11993static int
11994ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
11995			 ctl_ha_dt_cb callback)
11996{
11997	struct ctl_ha_dt_req *rq;
11998	struct ctl_sg_entry *remote_sglist, *local_sglist;
11999	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12000	uint32_t local_used, remote_used, total_used;
12001	int retval;
12002	int i, j;
12003
12004	retval = 0;
12005
12006	rq = ctl_dt_req_alloc();
12007
12008	/*
12009	 * If we failed to allocate the request, and if the DMA didn't fail
12010	 * anyway, set busy status.  This is just a resource allocation
12011	 * failure.
12012	 */
12013	if ((rq == NULL)
12014	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12015		ctl_set_busy(&io->scsiio);
12016
12017	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12018
12019		if (rq != NULL)
12020			ctl_dt_req_free(rq);
12021
12022		/*
12023		 * The data move failed.  We need to return status back
12024		 * to the other controller.  No point in trying to DMA
12025		 * data to the remote controller.
12026		 */
12027
12028		ctl_send_datamove_done(io, /*have_lock*/ 0);
12029
12030		retval = 1;
12031
12032		goto bailout;
12033	}
12034
12035	local_sglist = io->io_hdr.local_sglist;
12036	local_dma_sglist = io->io_hdr.local_dma_sglist;
12037	remote_sglist = io->io_hdr.remote_sglist;
12038	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12039	local_used = 0;
12040	remote_used = 0;
12041	total_used = 0;
12042
12043	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12044		rq->ret = CTL_HA_STATUS_SUCCESS;
12045		rq->context = io;
12046		callback(rq);
12047		goto bailout;
12048	}
12049
12050	/*
12051	 * Pull/push the data over the wire from/to the other controller.
12052	 * This takes into account the possibility that the local and
12053	 * remote sglists may not be identical in terms of the size of
12054	 * the elements and the number of elements.
12055	 *
12056	 * One fundamental assumption here is that the length allocated for
12057	 * both the local and remote sglists is identical.  Otherwise, we've
12058	 * essentially got a coding error of some sort.
12059	 */
12060	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12061		int isc_ret;
12062		uint32_t cur_len, dma_length;
12063		uint8_t *tmp_ptr;
12064
12065		rq->id = CTL_HA_DATA_CTL;
12066		rq->command = command;
12067		rq->context = io;
12068
12069		/*
12070		 * Both pointers should be aligned.  But it is possible
12071		 * that the allocation length is not.  They should both
12072		 * also have enough slack left over at the end, though,
12073		 * to round up to the next 8 byte boundary.
12074		 */
12075		cur_len = ctl_min(local_sglist[i].len - local_used,
12076				  remote_sglist[j].len - remote_used);
12077
12078		/*
12079		 * In this case, we have a size issue and need to decrease
12080		 * the size, except in the case where we actually have less
12081		 * than 8 bytes left.  In that case, we need to increase
12082		 * the DMA length to get the last bit.
12083		 */
12084		if ((cur_len & 0x7) != 0) {
12085			if (cur_len > 0x7) {
12086				cur_len = cur_len - (cur_len & 0x7);
12087				dma_length = cur_len;
12088			} else {
12089				CTL_SIZE_8B(dma_length, cur_len);
12090			}
12091
12092		} else
12093			dma_length = cur_len;
12094
12095		/*
12096		 * If we had to allocate memory for this I/O, instead of using
12097		 * the non-cached mirror memory, we'll need to flush the cache
12098		 * before trying to DMA to the other controller.
12099		 *
12100		 * We could end up doing this multiple times for the same
12101		 * segment if we have a larger local segment than remote
12102		 * segment.  That shouldn't be an issue.
12103		 */
12104		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12105			/*
12106			 * XXX KDM use bus_dmamap_sync() here.
12107			 */
12108		}
12109
12110		rq->size = dma_length;
12111
12112		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12113		tmp_ptr += local_used;
12114
12115		/* Use physical addresses when talking to ISC hardware */
12116		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12117			/* XXX KDM use busdma */
12118#if 0
12119			rq->local = vtophys(tmp_ptr);
12120#endif
12121		} else
12122			rq->local = tmp_ptr;
12123
12124		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12125		tmp_ptr += remote_used;
12126		rq->remote = tmp_ptr;
12127
12128		rq->callback = NULL;
12129
12130		local_used += cur_len;
12131		if (local_used >= local_sglist[i].len) {
12132			i++;
12133			local_used = 0;
12134		}
12135
12136		remote_used += cur_len;
12137		if (remote_used >= remote_sglist[j].len) {
12138			j++;
12139			remote_used = 0;
12140		}
12141		total_used += cur_len;
12142
12143		if (total_used >= io->scsiio.kern_data_len)
12144			rq->callback = callback;
12145
12146		if ((rq->size & 0x7) != 0) {
12147			printf("%s: warning: size %d is not on 8b boundary\n",
12148			       __func__, rq->size);
12149		}
12150		if (((uintptr_t)rq->local & 0x7) != 0) {
12151			printf("%s: warning: local %p not on 8b boundary\n",
12152			       __func__, rq->local);
12153		}
12154		if (((uintptr_t)rq->remote & 0x7) != 0) {
12155			printf("%s: warning: remote %p not on 8b boundary\n",
12156			       __func__, rq->local);
12157		}
12158#if 0
12159		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12160		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12161		       rq->local, rq->remote, rq->size);
12162#endif
12163
12164		isc_ret = ctl_dt_single(rq);
12165		if (isc_ret == CTL_HA_STATUS_WAIT)
12166			continue;
12167
12168		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12169			rq->ret = CTL_HA_STATUS_SUCCESS;
12170		} else {
12171			rq->ret = isc_ret;
12172		}
12173		callback(rq);
12174		goto bailout;
12175	}
12176
12177bailout:
12178	return (retval);
12179
12180}
12181
12182static void
12183ctl_datamove_remote_read(union ctl_io *io)
12184{
12185	int retval;
12186	int i;
12187
12188	/*
12189	 * This will send an error to the other controller in the case of a
12190	 * failure.
12191	 */
12192	retval = ctl_datamove_remote_sgl_setup(io);
12193	if (retval != 0)
12194		return;
12195
12196	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12197					  ctl_datamove_remote_read_cb);
12198	if ((retval != 0)
12199	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12200		/*
12201		 * Make sure we free memory if there was an error..  The
12202		 * ctl_datamove_remote_xfer() function will send the
12203		 * datamove done message, or call the callback with an
12204		 * error if there is a problem.
12205		 */
12206		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12207			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12208	}
12209
12210	return;
12211}
12212
12213/*
12214 * Process a datamove request from the other controller.  This is used for
12215 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12216 * first.  Once that is complete, the data gets DMAed into the remote
12217 * controller's memory.  For reads, we DMA from the remote controller's
12218 * memory into our memory first, and then move it out to the FETD.
12219 */
12220static void
12221ctl_datamove_remote(union ctl_io *io)
12222{
12223	struct ctl_softc *softc;
12224
12225	softc = control_softc;
12226
12227	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12228
12229	/*
12230	 * Note that we look for an aborted I/O here, but don't do some of
12231	 * the other checks that ctl_datamove() normally does.  We don't
12232	 * need to run the task queue, because this I/O is on the ISC
12233	 * queue, which is executed by the work thread after the task queue.
12234	 * We don't need to run the datamove delay code, since that should
12235	 * have been done if need be on the other controller.
12236	 */
12237	mtx_lock(&softc->ctl_lock);
12238
12239	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12240
12241		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12242		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12243		       io->io_hdr.nexus.targ_port,
12244		       io->io_hdr.nexus.targ_target.id,
12245		       io->io_hdr.nexus.targ_lun);
12246		io->io_hdr.status = CTL_CMD_ABORTED;
12247		io->io_hdr.port_status = 31338;
12248
12249		mtx_unlock(&softc->ctl_lock);
12250
12251		ctl_send_datamove_done(io, /*have_lock*/ 0);
12252
12253		return;
12254	}
12255
12256	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12257		mtx_unlock(&softc->ctl_lock);
12258		ctl_datamove_remote_write(io);
12259	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12260		mtx_unlock(&softc->ctl_lock);
12261		ctl_datamove_remote_read(io);
12262	} else {
12263		union ctl_ha_msg msg;
12264		struct scsi_sense_data *sense;
12265		uint8_t sks[3];
12266		int retry_count;
12267
12268		memset(&msg, 0, sizeof(msg));
12269
12270		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12271		msg.hdr.status = CTL_SCSI_ERROR;
12272		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12273
12274		retry_count = 4243;
12275
12276		sense = &msg.scsi.sense_data;
12277		sks[0] = SSD_SCS_VALID;
12278		sks[1] = (retry_count >> 8) & 0xff;
12279		sks[2] = retry_count & 0xff;
12280
12281		/* "Internal target failure" */
12282		scsi_set_sense_data(sense,
12283				    /*sense_format*/ SSD_TYPE_NONE,
12284				    /*current_error*/ 1,
12285				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12286				    /*asc*/ 0x44,
12287				    /*ascq*/ 0x00,
12288				    /*type*/ SSD_ELEM_SKS,
12289				    /*size*/ sizeof(sks),
12290				    /*data*/ sks,
12291				    SSD_ELEM_NONE);
12292
12293		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12294		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12295			ctl_failover_io(io, /*have_lock*/ 1);
12296			mtx_unlock(&softc->ctl_lock);
12297			return;
12298		}
12299
12300		mtx_unlock(&softc->ctl_lock);
12301
12302		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12303		    CTL_HA_STATUS_SUCCESS) {
12304			/* XXX KDM what to do if this fails? */
12305		}
12306		return;
12307	}
12308
12309}
12310
12311static int
12312ctl_process_done(union ctl_io *io, int have_lock)
12313{
12314	struct ctl_lun *lun;
12315	struct ctl_softc *ctl_softc;
12316	void (*fe_done)(union ctl_io *io);
12317	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12318
12319	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12320
12321	fe_done =
12322	    control_softc->ctl_ports[targ_port]->fe_done;
12323
12324#ifdef CTL_TIME_IO
12325	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12326		char str[256];
12327		char path_str[64];
12328		struct sbuf sb;
12329
12330		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12331		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12332
12333		sbuf_cat(&sb, path_str);
12334		switch (io->io_hdr.io_type) {
12335		case CTL_IO_SCSI:
12336			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12337			sbuf_printf(&sb, "\n");
12338			sbuf_cat(&sb, path_str);
12339			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12340				    io->scsiio.tag_num, io->scsiio.tag_type);
12341			break;
12342		case CTL_IO_TASK:
12343			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12344				    "Tag Type: %d\n", io->taskio.task_action,
12345				    io->taskio.tag_num, io->taskio.tag_type);
12346			break;
12347		default:
12348			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12349			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12350			break;
12351		}
12352		sbuf_cat(&sb, path_str);
12353		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12354			    (intmax_t)time_uptime - io->io_hdr.start_time);
12355		sbuf_finish(&sb);
12356		printf("%s", sbuf_data(&sb));
12357	}
12358#endif /* CTL_TIME_IO */
12359
12360	switch (io->io_hdr.io_type) {
12361	case CTL_IO_SCSI:
12362		break;
12363	case CTL_IO_TASK:
12364		ctl_io_error_print(io, NULL);
12365		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12366			ctl_free_io_internal(io, /*have_lock*/ 0);
12367		else
12368			fe_done(io);
12369		return (CTL_RETVAL_COMPLETE);
12370		break;
12371	default:
12372		printf("ctl_process_done: invalid io type %d\n",
12373		       io->io_hdr.io_type);
12374		panic("ctl_process_done: invalid io type %d\n",
12375		      io->io_hdr.io_type);
12376		break; /* NOTREACHED */
12377	}
12378
12379	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12380	if (lun == NULL) {
12381		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12382				 io->io_hdr.nexus.targ_lun));
12383		fe_done(io);
12384		goto bailout;
12385	}
12386	ctl_softc = lun->ctl_softc;
12387
12388	/*
12389	 * Remove this from the OOA queue.
12390	 */
12391	if (have_lock == 0)
12392		mtx_lock(&ctl_softc->ctl_lock);
12393
12394	/*
12395	 * Check to see if we have any errors to inject here.  We only
12396	 * inject errors for commands that don't already have errors set.
12397	 */
12398	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12399	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12400		ctl_inject_error(lun, io);
12401
12402	/*
12403	 * XXX KDM how do we treat commands that aren't completed
12404	 * successfully?
12405	 *
12406	 * XXX KDM should we also track I/O latency?
12407	 */
12408	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
12409		uint32_t blocksize;
12410#ifdef CTL_TIME_IO
12411		struct bintime cur_bt;
12412#endif
12413
12414		if ((lun->be_lun != NULL)
12415		 && (lun->be_lun->blocksize != 0))
12416			blocksize = lun->be_lun->blocksize;
12417		else
12418			blocksize = 512;
12419
12420		switch (io->io_hdr.io_type) {
12421		case CTL_IO_SCSI: {
12422			int isread;
12423			struct ctl_lba_len lbalen;
12424
12425			isread = 0;
12426			switch (io->scsiio.cdb[0]) {
12427			case READ_6:
12428			case READ_10:
12429			case READ_12:
12430			case READ_16:
12431				isread = 1;
12432				/* FALLTHROUGH */
12433			case WRITE_6:
12434			case WRITE_10:
12435			case WRITE_12:
12436			case WRITE_16:
12437			case WRITE_VERIFY_10:
12438			case WRITE_VERIFY_12:
12439			case WRITE_VERIFY_16:
12440				memcpy(&lbalen, io->io_hdr.ctl_private[
12441				       CTL_PRIV_LBA_LEN].bytes, sizeof(lbalen));
12442
12443				if (isread) {
12444					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
12445						lbalen.len * blocksize;
12446					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
12447
12448#ifdef CTL_TIME_IO
12449					bintime_add(
12450					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
12451					   &io->io_hdr.dma_bt);
12452					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
12453						io->io_hdr.num_dmas;
12454					getbintime(&cur_bt);
12455					bintime_sub(&cur_bt,
12456						    &io->io_hdr.start_bt);
12457
12458					bintime_add(
12459					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
12460					    &cur_bt);
12461
12462#if 0
12463					cs_prof_gettime(&cur_ticks);
12464					lun->stats.time[CTL_STATS_READ] +=
12465						cur_ticks -
12466						io->io_hdr.start_ticks;
12467#endif
12468#if 0
12469					lun->stats.time[CTL_STATS_READ] +=
12470						jiffies - io->io_hdr.start_time;
12471#endif
12472#endif /* CTL_TIME_IO */
12473				} else {
12474					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
12475						lbalen.len * blocksize;
12476					lun->stats.ports[targ_port].operations[
12477						CTL_STATS_WRITE]++;
12478
12479#ifdef CTL_TIME_IO
12480					bintime_add(
12481					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
12482					  &io->io_hdr.dma_bt);
12483					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
12484						io->io_hdr.num_dmas;
12485					getbintime(&cur_bt);
12486					bintime_sub(&cur_bt,
12487						    &io->io_hdr.start_bt);
12488
12489					bintime_add(
12490					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
12491					    &cur_bt);
12492#if 0
12493					cs_prof_gettime(&cur_ticks);
12494					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12495						cur_ticks -
12496						io->io_hdr.start_ticks;
12497					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
12498						jiffies - io->io_hdr.start_time;
12499#endif
12500#endif /* CTL_TIME_IO */
12501				}
12502				break;
12503			default:
12504				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
12505
12506#ifdef CTL_TIME_IO
12507				bintime_add(
12508				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
12509				  &io->io_hdr.dma_bt);
12510				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
12511					io->io_hdr.num_dmas;
12512				getbintime(&cur_bt);
12513				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12514
12515				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
12516					    &cur_bt);
12517
12518#if 0
12519				cs_prof_gettime(&cur_ticks);
12520				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12521					cur_ticks -
12522					io->io_hdr.start_ticks;
12523				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
12524					jiffies - io->io_hdr.start_time;
12525#endif
12526#endif /* CTL_TIME_IO */
12527				break;
12528			}
12529			break;
12530		}
12531		default:
12532			break;
12533		}
12534	}
12535
12536	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12537
12538	/*
12539	 * Run through the blocked queue on this LUN and see if anything
12540	 * has become unblocked, now that this transaction is done.
12541	 */
12542	ctl_check_blocked(lun);
12543
12544	/*
12545	 * If the LUN has been invalidated, free it if there is nothing
12546	 * left on its OOA queue.
12547	 */
12548	if ((lun->flags & CTL_LUN_INVALID)
12549	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
12550		ctl_free_lun(lun);
12551
12552	/*
12553	 * If this command has been aborted, make sure we set the status
12554	 * properly.  The FETD is responsible for freeing the I/O and doing
12555	 * whatever it needs to do to clean up its state.
12556	 */
12557	if (io->io_hdr.flags & CTL_FLAG_ABORT)
12558		io->io_hdr.status = CTL_CMD_ABORTED;
12559
12560	/*
12561	 * We print out status for every task management command.  For SCSI
12562	 * commands, we filter out any unit attention errors; they happen
12563	 * on every boot, and would clutter up the log.  Note:  task
12564	 * management commands aren't printed here, they are printed above,
12565	 * since they should never even make it down here.
12566	 */
12567	switch (io->io_hdr.io_type) {
12568	case CTL_IO_SCSI: {
12569		int error_code, sense_key, asc, ascq;
12570
12571		sense_key = 0;
12572
12573		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
12574		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
12575			/*
12576			 * Since this is just for printing, no need to
12577			 * show errors here.
12578			 */
12579			scsi_extract_sense_len(&io->scsiio.sense_data,
12580					       io->scsiio.sense_len,
12581					       &error_code,
12582					       &sense_key,
12583					       &asc,
12584					       &ascq,
12585					       /*show_errors*/ 0);
12586		}
12587
12588		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
12589		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
12590		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
12591		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
12592
12593			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
12594				ctl_softc->skipped_prints++;
12595				if (have_lock == 0)
12596					mtx_unlock(&ctl_softc->ctl_lock);
12597			} else {
12598				uint32_t skipped_prints;
12599
12600				skipped_prints = ctl_softc->skipped_prints;
12601
12602				ctl_softc->skipped_prints = 0;
12603				ctl_softc->last_print_jiffies = time_uptime;
12604
12605				if (have_lock == 0)
12606					mtx_unlock(&ctl_softc->ctl_lock);
12607				if (skipped_prints > 0) {
12608#ifdef NEEDTOPORT
12609					csevent_log(CSC_CTL | CSC_SHELF_SW |
12610					    CTL_ERROR_REPORT,
12611					    csevent_LogType_Trace,
12612					    csevent_Severity_Information,
12613					    csevent_AlertLevel_Green,
12614					    csevent_FRU_Firmware,
12615					    csevent_FRU_Unknown,
12616					    "High CTL error volume, %d prints "
12617					    "skipped", skipped_prints);
12618#endif
12619				}
12620				ctl_io_error_print(io, NULL);
12621			}
12622		} else {
12623			if (have_lock == 0)
12624				mtx_unlock(&ctl_softc->ctl_lock);
12625		}
12626		break;
12627	}
12628	case CTL_IO_TASK:
12629		if (have_lock == 0)
12630			mtx_unlock(&ctl_softc->ctl_lock);
12631		ctl_io_error_print(io, NULL);
12632		break;
12633	default:
12634		if (have_lock == 0)
12635			mtx_unlock(&ctl_softc->ctl_lock);
12636		break;
12637	}
12638
12639	/*
12640	 * Tell the FETD or the other shelf controller we're done with this
12641	 * command.  Note that only SCSI commands get to this point.  Task
12642	 * management commands are completed above.
12643	 *
12644	 * We only send status to the other controller if we're in XFER
12645	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
12646	 * received the I/O (from CTL's perspective), and so the status is
12647	 * generated there.
12648	 *
12649	 * XXX KDM if we hold the lock here, we could cause a deadlock
12650	 * if the frontend comes back in in this context to queue
12651	 * something.
12652	 */
12653	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
12654	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12655		union ctl_ha_msg msg;
12656
12657		memset(&msg, 0, sizeof(msg));
12658		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12659		msg.hdr.original_sc = io->io_hdr.original_sc;
12660		msg.hdr.nexus = io->io_hdr.nexus;
12661		msg.hdr.status = io->io_hdr.status;
12662		msg.scsi.scsi_status = io->scsiio.scsi_status;
12663		msg.scsi.tag_num = io->scsiio.tag_num;
12664		msg.scsi.tag_type = io->scsiio.tag_type;
12665		msg.scsi.sense_len = io->scsiio.sense_len;
12666		msg.scsi.sense_residual = io->scsiio.sense_residual;
12667		msg.scsi.residual = io->scsiio.residual;
12668		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12669		       sizeof(io->scsiio.sense_data));
12670		/*
12671		 * We copy this whether or not this is an I/O-related
12672		 * command.  Otherwise, we'd have to go and check to see
12673		 * whether it's a read/write command, and it really isn't
12674		 * worth it.
12675		 */
12676		memcpy(&msg.scsi.lbalen,
12677		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12678		       sizeof(msg.scsi.lbalen));
12679
12680		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12681				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12682			/* XXX do something here */
12683		}
12684
12685		ctl_free_io_internal(io, /*have_lock*/ 0);
12686	} else
12687		fe_done(io);
12688
12689bailout:
12690
12691	return (CTL_RETVAL_COMPLETE);
12692}
12693
12694/*
12695 * Front end should call this if it doesn't do autosense.  When the request
12696 * sense comes back in from the initiator, we'll dequeue this and send it.
12697 */
12698int
12699ctl_queue_sense(union ctl_io *io)
12700{
12701	struct ctl_lun *lun;
12702	struct ctl_softc *ctl_softc;
12703	uint32_t initidx, targ_lun;
12704
12705	ctl_softc = control_softc;
12706
12707	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12708
12709	/*
12710	 * LUN lookup will likely move to the ctl_work_thread() once we
12711	 * have our new queueing infrastructure (that doesn't put things on
12712	 * a per-LUN queue initially).  That is so that we can handle
12713	 * things like an INQUIRY to a LUN that we don't have enabled.  We
12714	 * can't deal with that right now.
12715	 */
12716	mtx_lock(&ctl_softc->ctl_lock);
12717
12718	/*
12719	 * If we don't have a LUN for this, just toss the sense
12720	 * information.
12721	 */
12722	targ_lun = io->io_hdr.nexus.targ_lun;
12723	if (io->io_hdr.nexus.lun_map_fn != NULL)
12724		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
12725	if ((targ_lun < CTL_MAX_LUNS)
12726	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12727		lun = ctl_softc->ctl_luns[targ_lun];
12728	else
12729		goto bailout;
12730
12731	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12732
12733	/*
12734	 * Already have CA set for this LUN...toss the sense information.
12735	 */
12736	if (ctl_is_set(lun->have_ca, initidx))
12737		goto bailout;
12738
12739	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
12740	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
12741	       sizeof(io->scsiio.sense_data)));
12742	ctl_set_mask(lun->have_ca, initidx);
12743
12744bailout:
12745	mtx_unlock(&ctl_softc->ctl_lock);
12746
12747	ctl_free_io(io);
12748
12749	return (CTL_RETVAL_COMPLETE);
12750}
12751
12752/*
12753 * Primary command inlet from frontend ports.  All SCSI and task I/O
12754 * requests must go through this function.
12755 */
12756int
12757ctl_queue(union ctl_io *io)
12758{
12759	struct ctl_softc *ctl_softc;
12760
12761	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12762
12763	ctl_softc = control_softc;
12764
12765#ifdef CTL_TIME_IO
12766	io->io_hdr.start_time = time_uptime;
12767	getbintime(&io->io_hdr.start_bt);
12768#endif /* CTL_TIME_IO */
12769
12770	mtx_lock(&ctl_softc->ctl_lock);
12771
12772	switch (io->io_hdr.io_type) {
12773	case CTL_IO_SCSI:
12774		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
12775				   links);
12776		break;
12777	case CTL_IO_TASK:
12778		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
12779		/*
12780		 * Set the task pending flag.  This is necessary to close a
12781		 * race condition with the FETD:
12782		 *
12783		 * - FETD submits a task management command, like an abort.
12784		 * - Back end calls fe_datamove() to move the data for the
12785		 *   aborted command.  The FETD can't really accept it, but
12786		 *   if it did, it would end up transmitting data for a
12787		 *   command that the initiator told us to abort.
12788		 *
12789		 * We close the race condition by setting the flag here,
12790		 * and checking it in ctl_datamove(), before calling the
12791		 * FETD's fe_datamove routine.  If we've got a task
12792		 * pending, we run the task queue and then check to see
12793		 * whether our particular I/O has been aborted.
12794		 */
12795		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
12796		break;
12797	default:
12798		mtx_unlock(&ctl_softc->ctl_lock);
12799		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12800		return (-EINVAL);
12801		break; /* NOTREACHED */
12802	}
12803	mtx_unlock(&ctl_softc->ctl_lock);
12804
12805	ctl_wakeup_thread();
12806
12807	return (CTL_RETVAL_COMPLETE);
12808}
12809
12810#ifdef CTL_IO_DELAY
12811static void
12812ctl_done_timer_wakeup(void *arg)
12813{
12814	union ctl_io *io;
12815
12816	io = (union ctl_io *)arg;
12817	ctl_done_lock(io, /*have_lock*/ 0);
12818}
12819#endif /* CTL_IO_DELAY */
12820
12821void
12822ctl_done_lock(union ctl_io *io, int have_lock)
12823{
12824	struct ctl_softc *ctl_softc;
12825#ifndef CTL_DONE_THREAD
12826	union ctl_io *xio;
12827#endif /* !CTL_DONE_THREAD */
12828
12829	ctl_softc = control_softc;
12830
12831	if (have_lock == 0)
12832		mtx_lock(&ctl_softc->ctl_lock);
12833
12834	/*
12835	 * Enable this to catch duplicate completion issues.
12836	 */
12837#if 0
12838	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12839		printf("%s: type %d msg %d cdb %x iptl: "
12840		       "%d:%d:%d:%d tag 0x%04x "
12841		       "flag %#x status %x\n",
12842			__func__,
12843			io->io_hdr.io_type,
12844			io->io_hdr.msg_type,
12845			io->scsiio.cdb[0],
12846			io->io_hdr.nexus.initid.id,
12847			io->io_hdr.nexus.targ_port,
12848			io->io_hdr.nexus.targ_target.id,
12849			io->io_hdr.nexus.targ_lun,
12850			(io->io_hdr.io_type ==
12851			CTL_IO_TASK) ?
12852			io->taskio.tag_num :
12853			io->scsiio.tag_num,
12854		        io->io_hdr.flags,
12855			io->io_hdr.status);
12856	} else
12857		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12858#endif
12859
12860	/*
12861	 * This is an internal copy of an I/O, and should not go through
12862	 * the normal done processing logic.
12863	 */
12864	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
12865		if (have_lock == 0)
12866			mtx_unlock(&ctl_softc->ctl_lock);
12867		return;
12868	}
12869
12870	/*
12871	 * We need to send a msg to the serializing shelf to finish the IO
12872	 * as well.  We don't send a finish message to the other shelf if
12873	 * this is a task management command.  Task management commands
12874	 * aren't serialized in the OOA queue, but rather just executed on
12875	 * both shelf controllers for commands that originated on that
12876	 * controller.
12877	 */
12878	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
12879	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
12880		union ctl_ha_msg msg_io;
12881
12882		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
12883		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
12884		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
12885		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
12886		}
12887		/* continue on to finish IO */
12888	}
12889#ifdef CTL_IO_DELAY
12890	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12891		struct ctl_lun *lun;
12892
12893		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12894
12895		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12896	} else {
12897		struct ctl_lun *lun;
12898
12899		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12900
12901		if ((lun != NULL)
12902		 && (lun->delay_info.done_delay > 0)) {
12903			struct callout *callout;
12904
12905			callout = (struct callout *)&io->io_hdr.timer_bytes;
12906			callout_init(callout, /*mpsafe*/ 1);
12907			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12908			callout_reset(callout,
12909				      lun->delay_info.done_delay * hz,
12910				      ctl_done_timer_wakeup, io);
12911			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
12912				lun->delay_info.done_delay = 0;
12913			if (have_lock == 0)
12914				mtx_unlock(&ctl_softc->ctl_lock);
12915			return;
12916		}
12917	}
12918#endif /* CTL_IO_DELAY */
12919
12920	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
12921
12922#ifdef CTL_DONE_THREAD
12923	if (have_lock == 0)
12924		mtx_unlock(&ctl_softc->ctl_lock);
12925
12926	ctl_wakeup_thread();
12927#else /* CTL_DONE_THREAD */
12928	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
12929	     xio != NULL;
12930	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
12931
12932		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
12933
12934		ctl_process_done(xio, /*have_lock*/ 1);
12935	}
12936	if (have_lock == 0)
12937		mtx_unlock(&ctl_softc->ctl_lock);
12938#endif /* CTL_DONE_THREAD */
12939}
12940
12941void
12942ctl_done(union ctl_io *io)
12943{
12944	ctl_done_lock(io, /*have_lock*/ 0);
12945}
12946
12947int
12948ctl_isc(struct ctl_scsiio *ctsio)
12949{
12950	struct ctl_lun *lun;
12951	int retval;
12952
12953	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12954
12955	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
12956
12957	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
12958
12959	retval = lun->backend->data_submit((union ctl_io *)ctsio);
12960
12961	return (retval);
12962}
12963
12964
12965static void
12966ctl_work_thread(void *arg)
12967{
12968	struct ctl_softc *softc;
12969	union ctl_io *io;
12970	struct ctl_be_lun *be_lun;
12971	int retval;
12972
12973	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
12974
12975	softc = (struct ctl_softc *)arg;
12976	if (softc == NULL)
12977		return;
12978
12979	mtx_lock(&softc->ctl_lock);
12980	for (;;) {
12981		retval = 0;
12982
12983		/*
12984		 * We handle the queues in this order:
12985		 * - task management
12986		 * - ISC
12987		 * - done queue (to free up resources, unblock other commands)
12988		 * - RtR queue
12989		 * - incoming queue
12990		 *
12991		 * If those queues are empty, we break out of the loop and
12992		 * go to sleep.
12993		 */
12994		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
12995		if (io != NULL) {
12996			ctl_run_task_queue(softc);
12997			continue;
12998		}
12999		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
13000		if (io != NULL) {
13001			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
13002			ctl_handle_isc(io);
13003			continue;
13004		}
13005		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
13006		if (io != NULL) {
13007			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
13008			/* clear any blocked commands, call fe_done */
13009			mtx_unlock(&softc->ctl_lock);
13010			/*
13011			 * XXX KDM
13012			 * Call this without a lock for now.  This will
13013			 * depend on whether there is any way the FETD can
13014			 * sleep or deadlock if called with the CTL lock
13015			 * held.
13016			 */
13017			retval = ctl_process_done(io, /*have_lock*/ 0);
13018			mtx_lock(&softc->ctl_lock);
13019			continue;
13020		}
13021		if (!ctl_pause_rtr) {
13022			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
13023			if (io != NULL) {
13024				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
13025				mtx_unlock(&softc->ctl_lock);
13026				goto execute;
13027			}
13028		}
13029		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
13030		if (io != NULL) {
13031			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
13032			mtx_unlock(&softc->ctl_lock);
13033			ctl_scsiio_precheck(softc, &io->scsiio);
13034			mtx_lock(&softc->ctl_lock);
13035			continue;
13036		}
13037		/*
13038		 * We might want to move this to a separate thread, so that
13039		 * configuration requests (in this case LUN creations)
13040		 * won't impact the I/O path.
13041		 */
13042		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13043		if (be_lun != NULL) {
13044			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13045			mtx_unlock(&softc->ctl_lock);
13046			ctl_create_lun(be_lun);
13047			mtx_lock(&softc->ctl_lock);
13048			continue;
13049		}
13050
13051		/* XXX KDM use the PDROP flag?? */
13052		/* Sleep until we have something to do. */
13053		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "ctl_work", 0);
13054
13055		/* Back to the top of the loop to see what woke us up. */
13056		continue;
13057
13058execute:
13059		retval = ctl_scsiio(&io->scsiio);
13060		switch (retval) {
13061		case CTL_RETVAL_COMPLETE:
13062			break;
13063		default:
13064			/*
13065			 * Probably need to make sure this doesn't happen.
13066			 */
13067			break;
13068		}
13069		mtx_lock(&softc->ctl_lock);
13070	}
13071}
13072
13073void
13074ctl_wakeup_thread()
13075{
13076	struct ctl_softc *softc;
13077
13078	softc = control_softc;
13079
13080	wakeup(softc);
13081}
13082
13083/* Initialization and failover */
13084
13085void
13086ctl_init_isc_msg(void)
13087{
13088	printf("CTL: Still calling this thing\n");
13089}
13090
13091/*
13092 * Init component
13093 * 	Initializes component into configuration defined by bootMode
13094 *	(see hasc-sv.c)
13095 *  	returns hasc_Status:
13096 * 		OK
13097 *		ERROR - fatal error
13098 */
13099static ctl_ha_comp_status
13100ctl_isc_init(struct ctl_ha_component *c)
13101{
13102	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13103
13104	c->status = ret;
13105	return ret;
13106}
13107
13108/* Start component
13109 * 	Starts component in state requested. If component starts successfully,
13110 *	it must set its own state to the requestrd state
13111 *	When requested state is HASC_STATE_HA, the component may refine it
13112 * 	by adding _SLAVE or _MASTER flags.
13113 *	Currently allowed state transitions are:
13114 *	UNKNOWN->HA		- initial startup
13115 *	UNKNOWN->SINGLE - initial startup when no parter detected
13116 *	HA->SINGLE		- failover
13117 * returns ctl_ha_comp_status:
13118 * 		OK	- component successfully started in requested state
13119 *		FAILED  - could not start the requested state, failover may
13120 * 			  be possible
13121 *		ERROR	- fatal error detected, no future startup possible
13122 */
13123static ctl_ha_comp_status
13124ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13125{
13126	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13127
13128	printf("%s: go\n", __func__);
13129
13130	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13131	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13132		ctl_is_single = 0;
13133		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13134		    != CTL_HA_STATUS_SUCCESS) {
13135			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13136			ret = CTL_HA_COMP_STATUS_ERROR;
13137		}
13138	} else if (CTL_HA_STATE_IS_HA(c->state)
13139		&& CTL_HA_STATE_IS_SINGLE(state)){
13140		// HA->SINGLE transition
13141	        ctl_failover();
13142		ctl_is_single = 1;
13143	} else {
13144		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13145		       c->state, state);
13146		ret = CTL_HA_COMP_STATUS_ERROR;
13147	}
13148	if (CTL_HA_STATE_IS_SINGLE(state))
13149		ctl_is_single = 1;
13150
13151	c->state = state;
13152	c->status = ret;
13153	return ret;
13154}
13155
13156/*
13157 * Quiesce component
13158 * The component must clear any error conditions (set status to OK) and
13159 * prepare itself to another Start call
13160 * returns ctl_ha_comp_status:
13161 * 	OK
13162 *	ERROR
13163 */
13164static ctl_ha_comp_status
13165ctl_isc_quiesce(struct ctl_ha_component *c)
13166{
13167	int ret = CTL_HA_COMP_STATUS_OK;
13168
13169	ctl_pause_rtr = 1;
13170	c->status = ret;
13171	return ret;
13172}
13173
13174struct ctl_ha_component ctl_ha_component_ctlisc =
13175{
13176	.name = "CTL ISC",
13177	.state = CTL_HA_STATE_UNKNOWN,
13178	.init = ctl_isc_init,
13179	.start = ctl_isc_start,
13180	.quiesce = ctl_isc_quiesce
13181};
13182
13183/*
13184 *  vim: ts=8
13185 */
13186