ctl.c revision 268550
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: stable/10/sys/cam/ctl/ctl.c 268550 2014-07-12 02:24:52Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * The default is to run with CTL_DONE_THREAD turned on.  Completed
87 * transactions are queued for processing by the CTL work thread.  When
88 * CTL_DONE_THREAD is not defined, completed transactions are processed in
89 * the caller's context.
90 */
91#define CTL_DONE_THREAD
92
93/*
94 * Use the serial number and device ID provided by the backend, rather than
95 * making up our own.
96 */
97#define CTL_USE_BACKEND_SN
98
99/*
100 * Size and alignment macros needed for Copan-specific HA hardware.  These
101 * can go away when the HA code is re-written, and uses busdma for any
102 * hardware.
103 */
104#define	CTL_ALIGN_8B(target, source, type)				\
105	if (((uint32_t)source & 0x7) != 0)				\
106		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
107	else								\
108		target = (type)source;
109
110#define	CTL_SIZE_8B(target, size)					\
111	if ((size & 0x7) != 0)						\
112		target = size + (0x8 - (size & 0x7));			\
113	else								\
114		target = size;
115
116#define CTL_ALIGN_8B_MARGIN	16
117
118/*
119 * Template mode pages.
120 */
121
122/*
123 * Note that these are default values only.  The actual values will be
124 * filled in when the user does a mode sense.
125 */
126static struct copan_power_subpage power_page_default = {
127	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
128	/*subpage*/ PWR_SUBPAGE_CODE,
129	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
130			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
131	/*page_version*/ PWR_VERSION,
132	/* total_luns */ 26,
133	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
134	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
136		      0, 0, 0, 0, 0, 0}
137};
138
139static struct copan_power_subpage power_page_changeable = {
140	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
141	/*subpage*/ PWR_SUBPAGE_CODE,
142	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
143			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
144	/*page_version*/ 0,
145	/* total_luns */ 0,
146	/* max_active_luns*/ 0,
147	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
148		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
149		      0, 0, 0, 0, 0, 0}
150};
151
152static struct copan_aps_subpage aps_page_default = {
153	APS_PAGE_CODE | SMPH_SPF, //page_code
154	APS_SUBPAGE_CODE, //subpage
155	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
156	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
157	APS_VERSION, //page_version
158	0, //lock_active
159	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
160	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
161	0, 0, 0, 0, 0} //reserved
162};
163
164static struct copan_aps_subpage aps_page_changeable = {
165	APS_PAGE_CODE | SMPH_SPF, //page_code
166	APS_SUBPAGE_CODE, //subpage
167	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
168	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
169	0, //page_version
170	0, //lock_active
171	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
173	0, 0, 0, 0, 0} //reserved
174};
175
176static struct copan_debugconf_subpage debugconf_page_default = {
177	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
178	DBGCNF_SUBPAGE_CODE,		/* subpage */
179	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
180	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
181	DBGCNF_VERSION,			/* page_version */
182	{CTL_TIME_IO_DEFAULT_SECS>>8,
183	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
184};
185
186static struct copan_debugconf_subpage debugconf_page_changeable = {
187	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
188	DBGCNF_SUBPAGE_CODE,		/* subpage */
189	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
190	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
191	0,				/* page_version */
192	{0xff,0xff},			/* ctl_time_io_secs */
193};
194
195static struct scsi_format_page format_page_default = {
196	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
197	/*page_length*/sizeof(struct scsi_format_page) - 2,
198	/*tracks_per_zone*/ {0, 0},
199	/*alt_sectors_per_zone*/ {0, 0},
200	/*alt_tracks_per_zone*/ {0, 0},
201	/*alt_tracks_per_lun*/ {0, 0},
202	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
203			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
204	/*bytes_per_sector*/ {0, 0},
205	/*interleave*/ {0, 0},
206	/*track_skew*/ {0, 0},
207	/*cylinder_skew*/ {0, 0},
208	/*flags*/ SFP_HSEC,
209	/*reserved*/ {0, 0, 0}
210};
211
212static struct scsi_format_page format_page_changeable = {
213	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
214	/*page_length*/sizeof(struct scsi_format_page) - 2,
215	/*tracks_per_zone*/ {0, 0},
216	/*alt_sectors_per_zone*/ {0, 0},
217	/*alt_tracks_per_zone*/ {0, 0},
218	/*alt_tracks_per_lun*/ {0, 0},
219	/*sectors_per_track*/ {0, 0},
220	/*bytes_per_sector*/ {0, 0},
221	/*interleave*/ {0, 0},
222	/*track_skew*/ {0, 0},
223	/*cylinder_skew*/ {0, 0},
224	/*flags*/ 0,
225	/*reserved*/ {0, 0, 0}
226};
227
228static struct scsi_rigid_disk_page rigid_disk_page_default = {
229	/*page_code*/SMS_RIGID_DISK_PAGE,
230	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
231	/*cylinders*/ {0, 0, 0},
232	/*heads*/ CTL_DEFAULT_HEADS,
233	/*start_write_precomp*/ {0, 0, 0},
234	/*start_reduced_current*/ {0, 0, 0},
235	/*step_rate*/ {0, 0},
236	/*landing_zone_cylinder*/ {0, 0, 0},
237	/*rpl*/ SRDP_RPL_DISABLED,
238	/*rotational_offset*/ 0,
239	/*reserved1*/ 0,
240	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
241			   CTL_DEFAULT_ROTATION_RATE & 0xff},
242	/*reserved2*/ {0, 0}
243};
244
245static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
246	/*page_code*/SMS_RIGID_DISK_PAGE,
247	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
248	/*cylinders*/ {0, 0, 0},
249	/*heads*/ 0,
250	/*start_write_precomp*/ {0, 0, 0},
251	/*start_reduced_current*/ {0, 0, 0},
252	/*step_rate*/ {0, 0},
253	/*landing_zone_cylinder*/ {0, 0, 0},
254	/*rpl*/ 0,
255	/*rotational_offset*/ 0,
256	/*reserved1*/ 0,
257	/*rotation_rate*/ {0, 0},
258	/*reserved2*/ {0, 0}
259};
260
261static struct scsi_caching_page caching_page_default = {
262	/*page_code*/SMS_CACHING_PAGE,
263	/*page_length*/sizeof(struct scsi_caching_page) - 2,
264	/*flags1*/ SCP_DISC | SCP_WCE,
265	/*ret_priority*/ 0,
266	/*disable_pf_transfer_len*/ {0xff, 0xff},
267	/*min_prefetch*/ {0, 0},
268	/*max_prefetch*/ {0xff, 0xff},
269	/*max_pf_ceiling*/ {0xff, 0xff},
270	/*flags2*/ 0,
271	/*cache_segments*/ 0,
272	/*cache_seg_size*/ {0, 0},
273	/*reserved*/ 0,
274	/*non_cache_seg_size*/ {0, 0, 0}
275};
276
277static struct scsi_caching_page caching_page_changeable = {
278	/*page_code*/SMS_CACHING_PAGE,
279	/*page_length*/sizeof(struct scsi_caching_page) - 2,
280	/*flags1*/ 0,
281	/*ret_priority*/ 0,
282	/*disable_pf_transfer_len*/ {0, 0},
283	/*min_prefetch*/ {0, 0},
284	/*max_prefetch*/ {0, 0},
285	/*max_pf_ceiling*/ {0, 0},
286	/*flags2*/ 0,
287	/*cache_segments*/ 0,
288	/*cache_seg_size*/ {0, 0},
289	/*reserved*/ 0,
290	/*non_cache_seg_size*/ {0, 0, 0}
291};
292
293static struct scsi_control_page control_page_default = {
294	/*page_code*/SMS_CONTROL_MODE_PAGE,
295	/*page_length*/sizeof(struct scsi_control_page) - 2,
296	/*rlec*/0,
297	/*queue_flags*/0,
298	/*eca_and_aen*/0,
299	/*reserved*/0,
300	/*aen_holdoff_period*/{0, 0}
301};
302
303static struct scsi_control_page control_page_changeable = {
304	/*page_code*/SMS_CONTROL_MODE_PAGE,
305	/*page_length*/sizeof(struct scsi_control_page) - 2,
306	/*rlec*/SCP_DSENSE,
307	/*queue_flags*/0,
308	/*eca_and_aen*/0,
309	/*reserved*/0,
310	/*aen_holdoff_period*/{0, 0}
311};
312
313
314/*
315 * XXX KDM move these into the softc.
316 */
317static int rcv_sync_msg;
318static int persis_offset;
319static uint8_t ctl_pause_rtr;
320static int     ctl_is_single = 1;
321static int     index_to_aps_page;
322
323SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
324static int worker_threads = 1;
325TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
326SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
327    &worker_threads, 1, "Number of worker threads");
328static int verbose = 0;
329TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
330SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
331    &verbose, 0, "Show SCSI errors returned to initiator");
332
333/*
334 * Serial number (0x80), device id (0x83), supported pages (0x00),
335 * Block limits (0xB0) and Logical Block Provisioning (0xB2)
336 */
337#define SCSI_EVPD_NUM_SUPPORTED_PAGES	5
338
339static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
340				  int param);
341static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
342static int ctl_init(void);
343void ctl_shutdown(void);
344static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
345static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
346static void ctl_ioctl_online(void *arg);
347static void ctl_ioctl_offline(void *arg);
348static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
349static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
350static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
351static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
352static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
353static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
354static int ctl_ioctl_submit_wait(union ctl_io *io);
355static void ctl_ioctl_datamove(union ctl_io *io);
356static void ctl_ioctl_done(union ctl_io *io);
357static void ctl_ioctl_hard_startstop_callback(void *arg,
358					      struct cfi_metatask *metatask);
359static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
360static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
361			      struct ctl_ooa *ooa_hdr,
362			      struct ctl_ooa_entry *kern_entries);
363static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
364		     struct thread *td);
365uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
366uint32_t ctl_port_idx(int port_num);
367#ifdef unused
368static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
369				   uint32_t targ_target, uint32_t targ_lun,
370				   int can_wait);
371static void ctl_kfree_io(union ctl_io *io);
372#endif /* unused */
373static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
374			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
375static int ctl_free_lun(struct ctl_lun *lun);
376static void ctl_create_lun(struct ctl_be_lun *be_lun);
377/**
378static void ctl_failover_change_pages(struct ctl_softc *softc,
379				      struct ctl_scsiio *ctsio, int master);
380**/
381
382static int ctl_do_mode_select(union ctl_io *io);
383static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
384			   uint64_t res_key, uint64_t sa_res_key,
385			   uint8_t type, uint32_t residx,
386			   struct ctl_scsiio *ctsio,
387			   struct scsi_per_res_out *cdb,
388			   struct scsi_per_res_out_parms* param);
389static void ctl_pro_preempt_other(struct ctl_lun *lun,
390				  union ctl_ha_msg *msg);
391static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
392static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
394static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
395static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
396					 int alloc_len);
397static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
398static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
399static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
400static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
401static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
402static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
403					 union ctl_io *ooa_io);
404static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
405				union ctl_io *starting_io);
406static int ctl_check_blocked(struct ctl_lun *lun);
407static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
408				struct ctl_lun *lun,
409				struct ctl_cmd_entry *entry,
410				struct ctl_scsiio *ctsio);
411//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
412static void ctl_failover(void);
413static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
414			       struct ctl_scsiio *ctsio);
415static int ctl_scsiio(struct ctl_scsiio *ctsio);
416
417static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
418static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
419			    ctl_ua_type ua_type);
420static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
421			 ctl_ua_type ua_type);
422static int ctl_abort_task(union ctl_io *io);
423static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
424#ifdef CTL_IO_DELAY
425static void ctl_datamove_timer_wakeup(void *arg);
426static void ctl_done_timer_wakeup(void *arg);
427#endif /* CTL_IO_DELAY */
428
429static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
430static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
431static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
432static void ctl_datamove_remote_write(union ctl_io *io);
433static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
434static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
435static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
436static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
437				    ctl_ha_dt_cb callback);
438static void ctl_datamove_remote_read(union ctl_io *io);
439static void ctl_datamove_remote(union ctl_io *io);
440static int ctl_process_done(union ctl_io *io, int have_lock);
441static void ctl_work_thread(void *arg);
442
443/*
444 * Load the serialization table.  This isn't very pretty, but is probably
445 * the easiest way to do it.
446 */
447#include "ctl_ser_table.c"
448
449/*
450 * We only need to define open, close and ioctl routines for this driver.
451 */
452static struct cdevsw ctl_cdevsw = {
453	.d_version =	D_VERSION,
454	.d_flags =	0,
455	.d_open =	ctl_open,
456	.d_close =	ctl_close,
457	.d_ioctl =	ctl_ioctl,
458	.d_name =	"ctl",
459};
460
461
462MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
463
464static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
465
466static moduledata_t ctl_moduledata = {
467	"ctl",
468	ctl_module_event_handler,
469	NULL
470};
471
472DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
473MODULE_VERSION(ctl, 1);
474
475static void
476ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
477			    union ctl_ha_msg *msg_info)
478{
479	struct ctl_scsiio *ctsio;
480
481	if (msg_info->hdr.original_sc == NULL) {
482		printf("%s: original_sc == NULL!\n", __func__);
483		/* XXX KDM now what? */
484		return;
485	}
486
487	ctsio = &msg_info->hdr.original_sc->scsiio;
488	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
489	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
490	ctsio->io_hdr.status = msg_info->hdr.status;
491	ctsio->scsi_status = msg_info->scsi.scsi_status;
492	ctsio->sense_len = msg_info->scsi.sense_len;
493	ctsio->sense_residual = msg_info->scsi.sense_residual;
494	ctsio->residual = msg_info->scsi.residual;
495	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
496	       sizeof(ctsio->sense_data));
497	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
498	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
499	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
500	ctl_wakeup_thread();
501}
502
503static void
504ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
505				union ctl_ha_msg *msg_info)
506{
507	struct ctl_scsiio *ctsio;
508
509	if (msg_info->hdr.serializing_sc == NULL) {
510		printf("%s: serializing_sc == NULL!\n", __func__);
511		/* XXX KDM now what? */
512		return;
513	}
514
515	ctsio = &msg_info->hdr.serializing_sc->scsiio;
516#if 0
517	/*
518	 * Attempt to catch the situation where an I/O has
519	 * been freed, and we're using it again.
520	 */
521	if (ctsio->io_hdr.io_type == 0xff) {
522		union ctl_io *tmp_io;
523		tmp_io = (union ctl_io *)ctsio;
524		printf("%s: %p use after free!\n", __func__,
525		       ctsio);
526		printf("%s: type %d msg %d cdb %x iptl: "
527		       "%d:%d:%d:%d tag 0x%04x "
528		       "flag %#x status %x\n",
529			__func__,
530			tmp_io->io_hdr.io_type,
531			tmp_io->io_hdr.msg_type,
532			tmp_io->scsiio.cdb[0],
533			tmp_io->io_hdr.nexus.initid.id,
534			tmp_io->io_hdr.nexus.targ_port,
535			tmp_io->io_hdr.nexus.targ_target.id,
536			tmp_io->io_hdr.nexus.targ_lun,
537			(tmp_io->io_hdr.io_type ==
538			CTL_IO_TASK) ?
539			tmp_io->taskio.tag_num :
540			tmp_io->scsiio.tag_num,
541		        tmp_io->io_hdr.flags,
542			tmp_io->io_hdr.status);
543	}
544#endif
545	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
546	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
547	ctl_wakeup_thread();
548}
549
550/*
551 * ISC (Inter Shelf Communication) event handler.  Events from the HA
552 * subsystem come in here.
553 */
554static void
555ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
556{
557	struct ctl_softc *ctl_softc;
558	union ctl_io *io;
559	struct ctl_prio *presio;
560	ctl_ha_status isc_status;
561
562	ctl_softc = control_softc;
563	io = NULL;
564
565
566#if 0
567	printf("CTL: Isc Msg event %d\n", event);
568#endif
569	if (event == CTL_HA_EVT_MSG_RECV) {
570		union ctl_ha_msg msg_info;
571
572		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
573					     sizeof(msg_info), /*wait*/ 0);
574#if 0
575		printf("CTL: msg_type %d\n", msg_info.msg_type);
576#endif
577		if (isc_status != 0) {
578			printf("Error receiving message, status = %d\n",
579			       isc_status);
580			return;
581		}
582		mtx_lock(&ctl_softc->ctl_lock);
583
584		switch (msg_info.hdr.msg_type) {
585		case CTL_MSG_SERIALIZE:
586#if 0
587			printf("Serialize\n");
588#endif
589			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
590			if (io == NULL) {
591				printf("ctl_isc_event_handler: can't allocate "
592				       "ctl_io!\n");
593				/* Bad Juju */
594				/* Need to set busy and send msg back */
595				mtx_unlock(&ctl_softc->ctl_lock);
596				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
597				msg_info.hdr.status = CTL_SCSI_ERROR;
598				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
599				msg_info.scsi.sense_len = 0;
600			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
601				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
602				}
603				goto bailout;
604			}
605			ctl_zero_io(io);
606			// populate ctsio from msg_info
607			io->io_hdr.io_type = CTL_IO_SCSI;
608			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
609			io->io_hdr.original_sc = msg_info.hdr.original_sc;
610#if 0
611			printf("pOrig %x\n", (int)msg_info.original_sc);
612#endif
613			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
614					    CTL_FLAG_IO_ACTIVE;
615			/*
616			 * If we're in serialization-only mode, we don't
617			 * want to go through full done processing.  Thus
618			 * the COPY flag.
619			 *
620			 * XXX KDM add another flag that is more specific.
621			 */
622			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
623				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
624			io->io_hdr.nexus = msg_info.hdr.nexus;
625#if 0
626			printf("targ %d, port %d, iid %d, lun %d\n",
627			       io->io_hdr.nexus.targ_target.id,
628			       io->io_hdr.nexus.targ_port,
629			       io->io_hdr.nexus.initid.id,
630			       io->io_hdr.nexus.targ_lun);
631#endif
632			io->scsiio.tag_num = msg_info.scsi.tag_num;
633			io->scsiio.tag_type = msg_info.scsi.tag_type;
634			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
635			       CTL_MAX_CDBLEN);
636			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
637				struct ctl_cmd_entry *entry;
638				uint8_t opcode;
639
640				opcode = io->scsiio.cdb[0];
641				entry = &ctl_cmd_table[opcode];
642				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
643				io->io_hdr.flags |=
644					entry->flags & CTL_FLAG_DATA_MASK;
645			}
646			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
647					   &io->io_hdr, links);
648			ctl_wakeup_thread();
649			break;
650
651		/* Performed on the Originating SC, XFER mode only */
652		case CTL_MSG_DATAMOVE: {
653			struct ctl_sg_entry *sgl;
654			int i, j;
655
656			io = msg_info.hdr.original_sc;
657			if (io == NULL) {
658				printf("%s: original_sc == NULL!\n", __func__);
659				/* XXX KDM do something here */
660				break;
661			}
662			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
663			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
664			/*
665			 * Keep track of this, we need to send it back over
666			 * when the datamove is complete.
667			 */
668			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
669
670			if (msg_info.dt.sg_sequence == 0) {
671				/*
672				 * XXX KDM we use the preallocated S/G list
673				 * here, but we'll need to change this to
674				 * dynamic allocation if we need larger S/G
675				 * lists.
676				 */
677				if (msg_info.dt.kern_sg_entries >
678				    sizeof(io->io_hdr.remote_sglist) /
679				    sizeof(io->io_hdr.remote_sglist[0])) {
680					printf("%s: number of S/G entries "
681					    "needed %u > allocated num %zd\n",
682					    __func__,
683					    msg_info.dt.kern_sg_entries,
684					    sizeof(io->io_hdr.remote_sglist)/
685					    sizeof(io->io_hdr.remote_sglist[0]));
686
687					/*
688					 * XXX KDM send a message back to
689					 * the other side to shut down the
690					 * DMA.  The error will come back
691					 * through via the normal channel.
692					 */
693					break;
694				}
695				sgl = io->io_hdr.remote_sglist;
696				memset(sgl, 0,
697				       sizeof(io->io_hdr.remote_sglist));
698
699				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
700
701				io->scsiio.kern_sg_entries =
702					msg_info.dt.kern_sg_entries;
703				io->scsiio.rem_sg_entries =
704					msg_info.dt.kern_sg_entries;
705				io->scsiio.kern_data_len =
706					msg_info.dt.kern_data_len;
707				io->scsiio.kern_total_len =
708					msg_info.dt.kern_total_len;
709				io->scsiio.kern_data_resid =
710					msg_info.dt.kern_data_resid;
711				io->scsiio.kern_rel_offset =
712					msg_info.dt.kern_rel_offset;
713				/*
714				 * Clear out per-DMA flags.
715				 */
716				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
717				/*
718				 * Add per-DMA flags that are set for this
719				 * particular DMA request.
720				 */
721				io->io_hdr.flags |= msg_info.dt.flags &
722						    CTL_FLAG_RDMA_MASK;
723			} else
724				sgl = (struct ctl_sg_entry *)
725					io->scsiio.kern_data_ptr;
726
727			for (i = msg_info.dt.sent_sg_entries, j = 0;
728			     i < (msg_info.dt.sent_sg_entries +
729			     msg_info.dt.cur_sg_entries); i++, j++) {
730				sgl[i].addr = msg_info.dt.sg_list[j].addr;
731				sgl[i].len = msg_info.dt.sg_list[j].len;
732
733#if 0
734				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
735				       __func__,
736				       msg_info.dt.sg_list[j].addr,
737				       msg_info.dt.sg_list[j].len,
738				       sgl[i].addr, sgl[i].len, j, i);
739#endif
740			}
741#if 0
742			memcpy(&sgl[msg_info.dt.sent_sg_entries],
743			       msg_info.dt.sg_list,
744			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
745#endif
746
747			/*
748			 * If this is the last piece of the I/O, we've got
749			 * the full S/G list.  Queue processing in the thread.
750			 * Otherwise wait for the next piece.
751			 */
752			if (msg_info.dt.sg_last != 0) {
753				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
754						   &io->io_hdr, links);
755				ctl_wakeup_thread();
756			}
757			break;
758		}
759		/* Performed on the Serializing (primary) SC, XFER mode only */
760		case CTL_MSG_DATAMOVE_DONE: {
761			if (msg_info.hdr.serializing_sc == NULL) {
762				printf("%s: serializing_sc == NULL!\n",
763				       __func__);
764				/* XXX KDM now what? */
765				break;
766			}
767			/*
768			 * We grab the sense information here in case
769			 * there was a failure, so we can return status
770			 * back to the initiator.
771			 */
772			io = msg_info.hdr.serializing_sc;
773			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
774			io->io_hdr.status = msg_info.hdr.status;
775			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
776			io->scsiio.sense_len = msg_info.scsi.sense_len;
777			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
778			io->io_hdr.port_status = msg_info.scsi.fetd_status;
779			io->scsiio.residual = msg_info.scsi.residual;
780			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
781			       sizeof(io->scsiio.sense_data));
782
783			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
784					   &io->io_hdr, links);
785			ctl_wakeup_thread();
786			break;
787		}
788
789		/* Preformed on Originating SC, SER_ONLY mode */
790		case CTL_MSG_R2R:
791			io = msg_info.hdr.original_sc;
792			if (io == NULL) {
793				printf("%s: Major Bummer\n", __func__);
794				mtx_unlock(&ctl_softc->ctl_lock);
795				return;
796			} else {
797#if 0
798				printf("pOrig %x\n",(int) ctsio);
799#endif
800			}
801			io->io_hdr.msg_type = CTL_MSG_R2R;
802			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
803			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
804					   &io->io_hdr, links);
805			ctl_wakeup_thread();
806			break;
807
808		/*
809		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
810		 * mode.
811		 * Performed on the Originating (i.e. secondary) SC in XFER
812		 * mode
813		 */
814		case CTL_MSG_FINISH_IO:
815			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
816				ctl_isc_handler_finish_xfer(ctl_softc,
817							    &msg_info);
818			else
819				ctl_isc_handler_finish_ser_only(ctl_softc,
820								&msg_info);
821			break;
822
823		/* Preformed on Originating SC */
824		case CTL_MSG_BAD_JUJU:
825			io = msg_info.hdr.original_sc;
826			if (io == NULL) {
827				printf("%s: Bad JUJU!, original_sc is NULL!\n",
828				       __func__);
829				break;
830			}
831			ctl_copy_sense_data(&msg_info, io);
832			/*
833			 * IO should have already been cleaned up on other
834			 * SC so clear this flag so we won't send a message
835			 * back to finish the IO there.
836			 */
837			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
838			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
839
840			/* io = msg_info.hdr.serializing_sc; */
841			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
842		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
843					   &io->io_hdr, links);
844			ctl_wakeup_thread();
845			break;
846
847		/* Handle resets sent from the other side */
848		case CTL_MSG_MANAGE_TASKS: {
849			struct ctl_taskio *taskio;
850			taskio = (struct ctl_taskio *)ctl_alloc_io(
851				(void *)ctl_softc->othersc_pool);
852			if (taskio == NULL) {
853				printf("ctl_isc_event_handler: can't allocate "
854				       "ctl_io!\n");
855				/* Bad Juju */
856				/* should I just call the proper reset func
857				   here??? */
858				mtx_unlock(&ctl_softc->ctl_lock);
859				goto bailout;
860			}
861			ctl_zero_io((union ctl_io *)taskio);
862			taskio->io_hdr.io_type = CTL_IO_TASK;
863			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
864			taskio->io_hdr.nexus = msg_info.hdr.nexus;
865			taskio->task_action = msg_info.task.task_action;
866			taskio->tag_num = msg_info.task.tag_num;
867			taskio->tag_type = msg_info.task.tag_type;
868#ifdef CTL_TIME_IO
869			taskio->io_hdr.start_time = time_uptime;
870			getbintime(&taskio->io_hdr.start_bt);
871#if 0
872			cs_prof_gettime(&taskio->io_hdr.start_ticks);
873#endif
874#endif /* CTL_TIME_IO */
875		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
876					   &taskio->io_hdr, links);
877			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
878			ctl_wakeup_thread();
879			break;
880		}
881		/* Persistent Reserve action which needs attention */
882		case CTL_MSG_PERS_ACTION:
883			presio = (struct ctl_prio *)ctl_alloc_io(
884				(void *)ctl_softc->othersc_pool);
885			if (presio == NULL) {
886				printf("ctl_isc_event_handler: can't allocate "
887				       "ctl_io!\n");
888				/* Bad Juju */
889				/* Need to set busy and send msg back */
890				mtx_unlock(&ctl_softc->ctl_lock);
891				goto bailout;
892			}
893			ctl_zero_io((union ctl_io *)presio);
894			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
895			presio->pr_msg = msg_info.pr;
896		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
897					   &presio->io_hdr, links);
898			ctl_wakeup_thread();
899			break;
900		case CTL_MSG_SYNC_FE:
901			rcv_sync_msg = 1;
902			break;
903		case CTL_MSG_APS_LOCK: {
904			// It's quicker to execute this then to
905			// queue it.
906			struct ctl_lun *lun;
907			struct ctl_page_index *page_index;
908			struct copan_aps_subpage *current_sp;
909			uint32_t targ_lun;
910
911			targ_lun = msg_info.hdr.nexus.targ_lun;
912			if (msg_info.hdr.nexus.lun_map_fn != NULL)
913				targ_lun = msg_info.hdr.nexus.lun_map_fn(msg_info.hdr.nexus.lun_map_arg, targ_lun);
914
915			lun = ctl_softc->ctl_luns[targ_lun];
916			page_index = &lun->mode_pages.index[index_to_aps_page];
917			current_sp = (struct copan_aps_subpage *)
918				     (page_index->page_data +
919				     (page_index->page_len * CTL_PAGE_CURRENT));
920
921			current_sp->lock_active = msg_info.aps.lock_flag;
922		        break;
923		}
924		default:
925		        printf("How did I get here?\n");
926		}
927		mtx_unlock(&ctl_softc->ctl_lock);
928	} else if (event == CTL_HA_EVT_MSG_SENT) {
929		if (param != CTL_HA_STATUS_SUCCESS) {
930			printf("Bad status from ctl_ha_msg_send status %d\n",
931			       param);
932		}
933		return;
934	} else if (event == CTL_HA_EVT_DISCONNECT) {
935		printf("CTL: Got a disconnect from Isc\n");
936		return;
937	} else {
938		printf("ctl_isc_event_handler: Unknown event %d\n", event);
939		return;
940	}
941
942bailout:
943	return;
944}
945
946static void
947ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
948{
949	struct scsi_sense_data *sense;
950
951	sense = &dest->scsiio.sense_data;
952	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
953	dest->scsiio.scsi_status = src->scsi.scsi_status;
954	dest->scsiio.sense_len = src->scsi.sense_len;
955	dest->io_hdr.status = src->hdr.status;
956}
957
958static int
959ctl_init(void)
960{
961	struct ctl_softc *softc;
962	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
963	struct ctl_frontend *fe;
964        uint8_t sc_id =0;
965	int i, error, retval;
966	//int isc_retval;
967
968	retval = 0;
969	ctl_pause_rtr = 0;
970        rcv_sync_msg = 0;
971
972	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
973			       M_WAITOK | M_ZERO);
974	softc = control_softc;
975
976	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
977			      "cam/ctl");
978
979	softc->dev->si_drv1 = softc;
980
981	/*
982	 * By default, return a "bad LUN" peripheral qualifier for unknown
983	 * LUNs.  The user can override this default using the tunable or
984	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
985	 */
986	softc->inquiry_pq_no_lun = 1;
987	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
988			  &softc->inquiry_pq_no_lun);
989	sysctl_ctx_init(&softc->sysctl_ctx);
990	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
991		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
992		CTLFLAG_RD, 0, "CAM Target Layer");
993
994	if (softc->sysctl_tree == NULL) {
995		printf("%s: unable to allocate sysctl tree\n", __func__);
996		destroy_dev(softc->dev);
997		free(control_softc, M_DEVBUF);
998		control_softc = NULL;
999		return (ENOMEM);
1000	}
1001
1002	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1003		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1004		       "inquiry_pq_no_lun", CTLFLAG_RW,
1005		       &softc->inquiry_pq_no_lun, 0,
1006		       "Report no lun possible for invalid LUNs");
1007
1008	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1009	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
1010	softc->open_count = 0;
1011
1012	/*
1013	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1014	 * the drive.
1015	 */
1016	softc->flags = CTL_FLAG_REAL_SYNC;
1017
1018	/*
1019	 * In Copan's HA scheme, the "master" and "slave" roles are
1020	 * figured out through the slot the controller is in.  Although it
1021	 * is an active/active system, someone has to be in charge.
1022 	 */
1023#ifdef NEEDTOPORT
1024        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1025#endif
1026
1027        if (sc_id == 0) {
1028		softc->flags |= CTL_FLAG_MASTER_SHELF;
1029		persis_offset = 0;
1030	} else
1031		persis_offset = CTL_MAX_INITIATORS;
1032
1033	/*
1034	 * XXX KDM need to figure out where we want to get our target ID
1035	 * and WWID.  Is it different on each port?
1036	 */
1037	softc->target.id = 0;
1038	softc->target.wwid[0] = 0x12345678;
1039	softc->target.wwid[1] = 0x87654321;
1040	STAILQ_INIT(&softc->lun_list);
1041	STAILQ_INIT(&softc->pending_lun_queue);
1042	STAILQ_INIT(&softc->task_queue);
1043	STAILQ_INIT(&softc->incoming_queue);
1044	STAILQ_INIT(&softc->rtr_queue);
1045	STAILQ_INIT(&softc->done_queue);
1046	STAILQ_INIT(&softc->isc_queue);
1047	STAILQ_INIT(&softc->fe_list);
1048	STAILQ_INIT(&softc->be_list);
1049	STAILQ_INIT(&softc->io_pools);
1050
1051	/*
1052	 * We don't bother calling these with ctl_lock held here, because,
1053	 * in theory, no one else can try to do anything while we're in our
1054	 * module init routine.
1055	 */
1056	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1057			    &internal_pool)!= 0){
1058		printf("ctl: can't allocate %d entry internal pool, "
1059		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1060		return (ENOMEM);
1061	}
1062
1063	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1064			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1065		printf("ctl: can't allocate %d entry emergency pool, "
1066		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1067		ctl_pool_free(internal_pool);
1068		return (ENOMEM);
1069	}
1070
1071	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1072	                    &other_pool) != 0)
1073	{
1074		printf("ctl: can't allocate %d entry other SC pool, "
1075		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1076		ctl_pool_free(internal_pool);
1077		ctl_pool_free(emergency_pool);
1078		return (ENOMEM);
1079	}
1080
1081	softc->internal_pool = internal_pool;
1082	softc->emergency_pool = emergency_pool;
1083	softc->othersc_pool = other_pool;
1084
1085	if (worker_threads > MAXCPU || worker_threads == 0) {
1086		printf("invalid kern.cam.ctl.worker_threads value; "
1087		    "setting to 1");
1088		worker_threads = 1;
1089	} else if (worker_threads < 0) {
1090		if (mp_ncpus > 2) {
1091			/*
1092			 * Using more than two worker threads actually hurts
1093			 * performance due to lock contention.
1094			 */
1095			worker_threads = 2;
1096		} else {
1097			worker_threads = 1;
1098		}
1099	}
1100
1101	for (i = 0; i < worker_threads; i++) {
1102		error = kproc_kthread_add(ctl_work_thread, softc,
1103		    &softc->work_thread, NULL, 0, 0, "ctl", "work%d", i);
1104		if (error != 0) {
1105			printf("error creating CTL work thread!\n");
1106			ctl_pool_free(internal_pool);
1107			ctl_pool_free(emergency_pool);
1108			ctl_pool_free(other_pool);
1109			return (error);
1110		}
1111	}
1112	if (bootverbose)
1113		printf("ctl: CAM Target Layer loaded\n");
1114
1115	/*
1116	 * Initialize the initiator and portname mappings
1117	 */
1118	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1119
1120	/*
1121	 * Initialize the ioctl front end.
1122	 */
1123	fe = &softc->ioctl_info.fe;
1124	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1125	fe->port_type = CTL_PORT_IOCTL;
1126	fe->num_requested_ctl_io = 100;
1127	fe->port_name = softc->ioctl_info.port_name;
1128	fe->port_online = ctl_ioctl_online;
1129	fe->port_offline = ctl_ioctl_offline;
1130	fe->onoff_arg = &softc->ioctl_info;
1131	fe->targ_enable = ctl_ioctl_targ_enable;
1132	fe->targ_disable = ctl_ioctl_targ_disable;
1133	fe->lun_enable = ctl_ioctl_lun_enable;
1134	fe->lun_disable = ctl_ioctl_lun_disable;
1135	fe->targ_lun_arg = &softc->ioctl_info;
1136	fe->fe_datamove = ctl_ioctl_datamove;
1137	fe->fe_done = ctl_ioctl_done;
1138	fe->max_targets = 15;
1139	fe->max_target_id = 15;
1140
1141	if (ctl_frontend_register(&softc->ioctl_info.fe,
1142	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1143		printf("ctl: ioctl front end registration failed, will "
1144		       "continue anyway\n");
1145	}
1146
1147#ifdef CTL_IO_DELAY
1148	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1149		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1150		       sizeof(struct callout), CTL_TIMER_BYTES);
1151		return (EINVAL);
1152	}
1153#endif /* CTL_IO_DELAY */
1154
1155	return (0);
1156}
1157
1158void
1159ctl_shutdown(void)
1160{
1161	struct ctl_softc *softc;
1162	struct ctl_lun *lun, *next_lun;
1163	struct ctl_io_pool *pool;
1164
1165	softc = (struct ctl_softc *)control_softc;
1166
1167	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1168		printf("ctl: ioctl front end deregistration failed\n");
1169
1170	mtx_lock(&softc->ctl_lock);
1171
1172	/*
1173	 * Free up each LUN.
1174	 */
1175	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1176		next_lun = STAILQ_NEXT(lun, links);
1177		ctl_free_lun(lun);
1178	}
1179
1180	mtx_unlock(&softc->ctl_lock);
1181
1182	/*
1183	 * This will rip the rug out from under any FETDs or anyone else
1184	 * that has a pool allocated.  Since we increment our module
1185	 * refcount any time someone outside the main CTL module allocates
1186	 * a pool, we shouldn't have any problems here.  The user won't be
1187	 * able to unload the CTL module until client modules have
1188	 * successfully unloaded.
1189	 */
1190	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1191		ctl_pool_free(pool);
1192
1193#if 0
1194	ctl_shutdown_thread(softc->work_thread);
1195#endif
1196
1197	mtx_destroy(&softc->pool_lock);
1198	mtx_destroy(&softc->ctl_lock);
1199
1200	destroy_dev(softc->dev);
1201
1202	sysctl_ctx_free(&softc->sysctl_ctx);
1203
1204	free(control_softc, M_DEVBUF);
1205	control_softc = NULL;
1206
1207	if (bootverbose)
1208		printf("ctl: CAM Target Layer unloaded\n");
1209}
1210
1211static int
1212ctl_module_event_handler(module_t mod, int what, void *arg)
1213{
1214
1215	switch (what) {
1216	case MOD_LOAD:
1217		return (ctl_init());
1218	case MOD_UNLOAD:
1219		return (EBUSY);
1220	default:
1221		return (EOPNOTSUPP);
1222	}
1223}
1224
1225/*
1226 * XXX KDM should we do some access checks here?  Bump a reference count to
1227 * prevent a CTL module from being unloaded while someone has it open?
1228 */
1229static int
1230ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1231{
1232	return (0);
1233}
1234
1235static int
1236ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1237{
1238	return (0);
1239}
1240
1241int
1242ctl_port_enable(ctl_port_type port_type)
1243{
1244	struct ctl_softc *softc;
1245	struct ctl_frontend *fe;
1246
1247	if (ctl_is_single == 0) {
1248		union ctl_ha_msg msg_info;
1249		int isc_retval;
1250
1251#if 0
1252		printf("%s: HA mode, synchronizing frontend enable\n",
1253		        __func__);
1254#endif
1255		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1256	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1257		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1258			printf("Sync msg send error retval %d\n", isc_retval);
1259		}
1260		if (!rcv_sync_msg) {
1261			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1262			        sizeof(msg_info), 1);
1263		}
1264#if 0
1265        	printf("CTL:Frontend Enable\n");
1266	} else {
1267		printf("%s: single mode, skipping frontend synchronization\n",
1268		        __func__);
1269#endif
1270	}
1271
1272	softc = control_softc;
1273
1274	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1275		if (port_type & fe->port_type)
1276		{
1277#if 0
1278			printf("port %d\n", fe->targ_port);
1279#endif
1280			ctl_frontend_online(fe);
1281		}
1282	}
1283
1284	return (0);
1285}
1286
1287int
1288ctl_port_disable(ctl_port_type port_type)
1289{
1290	struct ctl_softc *softc;
1291	struct ctl_frontend *fe;
1292
1293	softc = control_softc;
1294
1295	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1296		if (port_type & fe->port_type)
1297			ctl_frontend_offline(fe);
1298	}
1299
1300	return (0);
1301}
1302
1303/*
1304 * Returns 0 for success, 1 for failure.
1305 * Currently the only failure mode is if there aren't enough entries
1306 * allocated.  So, in case of a failure, look at num_entries_dropped,
1307 * reallocate and try again.
1308 */
1309int
1310ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1311	      int *num_entries_filled, int *num_entries_dropped,
1312	      ctl_port_type port_type, int no_virtual)
1313{
1314	struct ctl_softc *softc;
1315	struct ctl_frontend *fe;
1316	int entries_dropped, entries_filled;
1317	int retval;
1318	int i;
1319
1320	softc = control_softc;
1321
1322	retval = 0;
1323	entries_filled = 0;
1324	entries_dropped = 0;
1325
1326	i = 0;
1327	mtx_lock(&softc->ctl_lock);
1328	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1329		struct ctl_port_entry *entry;
1330
1331		if ((fe->port_type & port_type) == 0)
1332			continue;
1333
1334		if ((no_virtual != 0)
1335		 && (fe->virtual_port != 0))
1336			continue;
1337
1338		if (entries_filled >= num_entries_alloced) {
1339			entries_dropped++;
1340			continue;
1341		}
1342		entry = &entries[i];
1343
1344		entry->port_type = fe->port_type;
1345		strlcpy(entry->port_name, fe->port_name,
1346			sizeof(entry->port_name));
1347		entry->physical_port = fe->physical_port;
1348		entry->virtual_port = fe->virtual_port;
1349		entry->wwnn = fe->wwnn;
1350		entry->wwpn = fe->wwpn;
1351
1352		i++;
1353		entries_filled++;
1354	}
1355
1356	mtx_unlock(&softc->ctl_lock);
1357
1358	if (entries_dropped > 0)
1359		retval = 1;
1360
1361	*num_entries_dropped = entries_dropped;
1362	*num_entries_filled = entries_filled;
1363
1364	return (retval);
1365}
1366
1367static void
1368ctl_ioctl_online(void *arg)
1369{
1370	struct ctl_ioctl_info *ioctl_info;
1371
1372	ioctl_info = (struct ctl_ioctl_info *)arg;
1373
1374	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1375}
1376
1377static void
1378ctl_ioctl_offline(void *arg)
1379{
1380	struct ctl_ioctl_info *ioctl_info;
1381
1382	ioctl_info = (struct ctl_ioctl_info *)arg;
1383
1384	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1385}
1386
1387/*
1388 * Remove an initiator by port number and initiator ID.
1389 * Returns 0 for success, 1 for failure.
1390 */
1391int
1392ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1393{
1394	struct ctl_softc *softc;
1395
1396	softc = control_softc;
1397
1398	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1399
1400	if ((targ_port < 0)
1401	 || (targ_port > CTL_MAX_PORTS)) {
1402		printf("%s: invalid port number %d\n", __func__, targ_port);
1403		return (1);
1404	}
1405	if (iid > CTL_MAX_INIT_PER_PORT) {
1406		printf("%s: initiator ID %u > maximun %u!\n",
1407		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1408		return (1);
1409	}
1410
1411	mtx_lock(&softc->ctl_lock);
1412
1413	softc->wwpn_iid[targ_port][iid].in_use = 0;
1414
1415	mtx_unlock(&softc->ctl_lock);
1416
1417	return (0);
1418}
1419
1420/*
1421 * Add an initiator to the initiator map.
1422 * Returns 0 for success, 1 for failure.
1423 */
1424int
1425ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1426{
1427	struct ctl_softc *softc;
1428	int retval;
1429
1430	softc = control_softc;
1431
1432	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1433
1434	retval = 0;
1435
1436	if ((targ_port < 0)
1437	 || (targ_port > CTL_MAX_PORTS)) {
1438		printf("%s: invalid port number %d\n", __func__, targ_port);
1439		return (1);
1440	}
1441	if (iid > CTL_MAX_INIT_PER_PORT) {
1442		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1443		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1444		return (1);
1445	}
1446
1447	mtx_lock(&softc->ctl_lock);
1448
1449	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1450		/*
1451		 * We don't treat this as an error.
1452		 */
1453		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1454			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1455			       __func__, targ_port, iid, (uintmax_t)wwpn);
1456			goto bailout;
1457		}
1458
1459		/*
1460		 * This is an error, but what do we do about it?  The
1461		 * driver is telling us we have a new WWPN for this
1462		 * initiator ID, so we pretty much need to use it.
1463		 */
1464		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1465		       "still at that address\n", __func__, targ_port, iid,
1466		       (uintmax_t)wwpn,
1467		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1468
1469		/*
1470		 * XXX KDM clear have_ca and ua_pending on each LUN for
1471		 * this initiator.
1472		 */
1473	}
1474	softc->wwpn_iid[targ_port][iid].in_use = 1;
1475	softc->wwpn_iid[targ_port][iid].iid = iid;
1476	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1477	softc->wwpn_iid[targ_port][iid].port = targ_port;
1478
1479bailout:
1480
1481	mtx_unlock(&softc->ctl_lock);
1482
1483	return (retval);
1484}
1485
1486/*
1487 * XXX KDM should we pretend to do something in the target/lun
1488 * enable/disable functions?
1489 */
1490static int
1491ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1492{
1493	return (0);
1494}
1495
1496static int
1497ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1498{
1499	return (0);
1500}
1501
1502static int
1503ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1504{
1505	return (0);
1506}
1507
1508static int
1509ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1510{
1511	return (0);
1512}
1513
1514/*
1515 * Data movement routine for the CTL ioctl frontend port.
1516 */
1517static int
1518ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1519{
1520	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1521	struct ctl_sg_entry ext_entry, kern_entry;
1522	int ext_sglen, ext_sg_entries, kern_sg_entries;
1523	int ext_sg_start, ext_offset;
1524	int len_to_copy, len_copied;
1525	int kern_watermark, ext_watermark;
1526	int ext_sglist_malloced;
1527	int i, j;
1528
1529	ext_sglist_malloced = 0;
1530	ext_sg_start = 0;
1531	ext_offset = 0;
1532
1533	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1534
1535	/*
1536	 * If this flag is set, fake the data transfer.
1537	 */
1538	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1539		ctsio->ext_data_filled = ctsio->ext_data_len;
1540		goto bailout;
1541	}
1542
1543	/*
1544	 * To simplify things here, if we have a single buffer, stick it in
1545	 * a S/G entry and just make it a single entry S/G list.
1546	 */
1547	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1548		int len_seen;
1549
1550		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1551
1552		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1553							   M_WAITOK);
1554		ext_sglist_malloced = 1;
1555		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1556				   ext_sglen) != 0) {
1557			ctl_set_internal_failure(ctsio,
1558						 /*sks_valid*/ 0,
1559						 /*retry_count*/ 0);
1560			goto bailout;
1561		}
1562		ext_sg_entries = ctsio->ext_sg_entries;
1563		len_seen = 0;
1564		for (i = 0; i < ext_sg_entries; i++) {
1565			if ((len_seen + ext_sglist[i].len) >=
1566			     ctsio->ext_data_filled) {
1567				ext_sg_start = i;
1568				ext_offset = ctsio->ext_data_filled - len_seen;
1569				break;
1570			}
1571			len_seen += ext_sglist[i].len;
1572		}
1573	} else {
1574		ext_sglist = &ext_entry;
1575		ext_sglist->addr = ctsio->ext_data_ptr;
1576		ext_sglist->len = ctsio->ext_data_len;
1577		ext_sg_entries = 1;
1578		ext_sg_start = 0;
1579		ext_offset = ctsio->ext_data_filled;
1580	}
1581
1582	if (ctsio->kern_sg_entries > 0) {
1583		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1584		kern_sg_entries = ctsio->kern_sg_entries;
1585	} else {
1586		kern_sglist = &kern_entry;
1587		kern_sglist->addr = ctsio->kern_data_ptr;
1588		kern_sglist->len = ctsio->kern_data_len;
1589		kern_sg_entries = 1;
1590	}
1591
1592
1593	kern_watermark = 0;
1594	ext_watermark = ext_offset;
1595	len_copied = 0;
1596	for (i = ext_sg_start, j = 0;
1597	     i < ext_sg_entries && j < kern_sg_entries;) {
1598		uint8_t *ext_ptr, *kern_ptr;
1599
1600		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1601				      kern_sglist[j].len - kern_watermark);
1602
1603		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1604		ext_ptr = ext_ptr + ext_watermark;
1605		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1606			/*
1607			 * XXX KDM fix this!
1608			 */
1609			panic("need to implement bus address support");
1610#if 0
1611			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1612#endif
1613		} else
1614			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1615		kern_ptr = kern_ptr + kern_watermark;
1616
1617		kern_watermark += len_to_copy;
1618		ext_watermark += len_to_copy;
1619
1620		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1621		     CTL_FLAG_DATA_IN) {
1622			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1623					 "bytes to user\n", len_to_copy));
1624			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1625					 "to %p\n", kern_ptr, ext_ptr));
1626			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1627				ctl_set_internal_failure(ctsio,
1628							 /*sks_valid*/ 0,
1629							 /*retry_count*/ 0);
1630				goto bailout;
1631			}
1632		} else {
1633			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1634					 "bytes from user\n", len_to_copy));
1635			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1636					 "to %p\n", ext_ptr, kern_ptr));
1637			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1638				ctl_set_internal_failure(ctsio,
1639							 /*sks_valid*/ 0,
1640							 /*retry_count*/0);
1641				goto bailout;
1642			}
1643		}
1644
1645		len_copied += len_to_copy;
1646
1647		if (ext_sglist[i].len == ext_watermark) {
1648			i++;
1649			ext_watermark = 0;
1650		}
1651
1652		if (kern_sglist[j].len == kern_watermark) {
1653			j++;
1654			kern_watermark = 0;
1655		}
1656	}
1657
1658	ctsio->ext_data_filled += len_copied;
1659
1660	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1661			 "kern_sg_entries: %d\n", ext_sg_entries,
1662			 kern_sg_entries));
1663	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1664			 "kern_data_len = %d\n", ctsio->ext_data_len,
1665			 ctsio->kern_data_len));
1666
1667
1668	/* XXX KDM set residual?? */
1669bailout:
1670
1671	if (ext_sglist_malloced != 0)
1672		free(ext_sglist, M_CTL);
1673
1674	return (CTL_RETVAL_COMPLETE);
1675}
1676
1677/*
1678 * Serialize a command that went down the "wrong" side, and so was sent to
1679 * this controller for execution.  The logic is a little different than the
1680 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1681 * sent back to the other side, but in the success case, we execute the
1682 * command on this side (XFER mode) or tell the other side to execute it
1683 * (SER_ONLY mode).
1684 */
1685static int
1686ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1687{
1688	struct ctl_softc *ctl_softc;
1689	union ctl_ha_msg msg_info;
1690	struct ctl_lun *lun;
1691	int retval = 0;
1692	uint32_t targ_lun;
1693
1694	ctl_softc = control_softc;
1695	if (have_lock == 0)
1696		mtx_lock(&ctl_softc->ctl_lock);
1697
1698	targ_lun = ctsio->io_hdr.nexus.targ_lun;
1699	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
1700		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
1701	lun = ctl_softc->ctl_luns[targ_lun];
1702	if (lun==NULL)
1703	{
1704		/*
1705		 * Why isn't LUN defined? The other side wouldn't
1706		 * send a cmd if the LUN is undefined.
1707		 */
1708		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1709
1710		/* "Logical unit not supported" */
1711		ctl_set_sense_data(&msg_info.scsi.sense_data,
1712				   lun,
1713				   /*sense_format*/SSD_TYPE_NONE,
1714				   /*current_error*/ 1,
1715				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1716				   /*asc*/ 0x25,
1717				   /*ascq*/ 0x00,
1718				   SSD_ELEM_NONE);
1719
1720		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1721		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1722		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1723		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1724		msg_info.hdr.serializing_sc = NULL;
1725		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1726	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1727				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1728		}
1729		if (have_lock == 0)
1730			mtx_unlock(&ctl_softc->ctl_lock);
1731		return(1);
1732
1733	}
1734
1735    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1736
1737	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1738		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1739		 ooa_links))) {
1740	case CTL_ACTION_BLOCK:
1741		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1742		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1743				  blocked_links);
1744		break;
1745	case CTL_ACTION_PASS:
1746	case CTL_ACTION_SKIP:
1747		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1748			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1749			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1750					   &ctsio->io_hdr, links);
1751		} else {
1752
1753			/* send msg back to other side */
1754			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1755			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1756			msg_info.hdr.msg_type = CTL_MSG_R2R;
1757#if 0
1758			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1759#endif
1760		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1761			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1762			}
1763		}
1764		break;
1765	case CTL_ACTION_OVERLAP:
1766		/* OVERLAPPED COMMANDS ATTEMPTED */
1767		ctl_set_sense_data(&msg_info.scsi.sense_data,
1768				   lun,
1769				   /*sense_format*/SSD_TYPE_NONE,
1770				   /*current_error*/ 1,
1771				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1772				   /*asc*/ 0x4E,
1773				   /*ascq*/ 0x00,
1774				   SSD_ELEM_NONE);
1775
1776		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1777		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1778		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1779		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1780		msg_info.hdr.serializing_sc = NULL;
1781		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1782#if 0
1783		printf("BAD JUJU:Major Bummer Overlap\n");
1784#endif
1785		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1786		retval = 1;
1787		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1788		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1789		}
1790		break;
1791	case CTL_ACTION_OVERLAP_TAG:
1792		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1793		ctl_set_sense_data(&msg_info.scsi.sense_data,
1794				   lun,
1795				   /*sense_format*/SSD_TYPE_NONE,
1796				   /*current_error*/ 1,
1797				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1798				   /*asc*/ 0x4D,
1799				   /*ascq*/ ctsio->tag_num & 0xff,
1800				   SSD_ELEM_NONE);
1801
1802		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1803		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1804		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1805		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1806		msg_info.hdr.serializing_sc = NULL;
1807		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1808#if 0
1809		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1810#endif
1811		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1812		retval = 1;
1813		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1814		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1815		}
1816		break;
1817	case CTL_ACTION_ERROR:
1818	default:
1819		/* "Internal target failure" */
1820		ctl_set_sense_data(&msg_info.scsi.sense_data,
1821				   lun,
1822				   /*sense_format*/SSD_TYPE_NONE,
1823				   /*current_error*/ 1,
1824				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1825				   /*asc*/ 0x44,
1826				   /*ascq*/ 0x00,
1827				   SSD_ELEM_NONE);
1828
1829		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1830		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1831		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1832		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1833		msg_info.hdr.serializing_sc = NULL;
1834		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1835#if 0
1836		printf("BAD JUJU:Major Bummer HW Error\n");
1837#endif
1838		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1839		retval = 1;
1840		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1841		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1842		}
1843		break;
1844	}
1845	if (have_lock == 0)
1846		mtx_unlock(&ctl_softc->ctl_lock);
1847	return (retval);
1848}
1849
1850static int
1851ctl_ioctl_submit_wait(union ctl_io *io)
1852{
1853	struct ctl_fe_ioctl_params params;
1854	ctl_fe_ioctl_state last_state;
1855	int done, retval;
1856
1857	retval = 0;
1858
1859	bzero(&params, sizeof(params));
1860
1861	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1862	cv_init(&params.sem, "ctlioccv");
1863	params.state = CTL_IOCTL_INPROG;
1864	last_state = params.state;
1865
1866	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1867
1868	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1869
1870	/* This shouldn't happen */
1871	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1872		return (retval);
1873
1874	done = 0;
1875
1876	do {
1877		mtx_lock(&params.ioctl_mtx);
1878		/*
1879		 * Check the state here, and don't sleep if the state has
1880		 * already changed (i.e. wakeup has already occured, but we
1881		 * weren't waiting yet).
1882		 */
1883		if (params.state == last_state) {
1884			/* XXX KDM cv_wait_sig instead? */
1885			cv_wait(&params.sem, &params.ioctl_mtx);
1886		}
1887		last_state = params.state;
1888
1889		switch (params.state) {
1890		case CTL_IOCTL_INPROG:
1891			/* Why did we wake up? */
1892			/* XXX KDM error here? */
1893			mtx_unlock(&params.ioctl_mtx);
1894			break;
1895		case CTL_IOCTL_DATAMOVE:
1896			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1897
1898			/*
1899			 * change last_state back to INPROG to avoid
1900			 * deadlock on subsequent data moves.
1901			 */
1902			params.state = last_state = CTL_IOCTL_INPROG;
1903
1904			mtx_unlock(&params.ioctl_mtx);
1905			ctl_ioctl_do_datamove(&io->scsiio);
1906			/*
1907			 * Note that in some cases, most notably writes,
1908			 * this will queue the I/O and call us back later.
1909			 * In other cases, generally reads, this routine
1910			 * will immediately call back and wake us up,
1911			 * probably using our own context.
1912			 */
1913			io->scsiio.be_move_done(io);
1914			break;
1915		case CTL_IOCTL_DONE:
1916			mtx_unlock(&params.ioctl_mtx);
1917			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1918			done = 1;
1919			break;
1920		default:
1921			mtx_unlock(&params.ioctl_mtx);
1922			/* XXX KDM error here? */
1923			break;
1924		}
1925	} while (done == 0);
1926
1927	mtx_destroy(&params.ioctl_mtx);
1928	cv_destroy(&params.sem);
1929
1930	return (CTL_RETVAL_COMPLETE);
1931}
1932
1933static void
1934ctl_ioctl_datamove(union ctl_io *io)
1935{
1936	struct ctl_fe_ioctl_params *params;
1937
1938	params = (struct ctl_fe_ioctl_params *)
1939		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1940
1941	mtx_lock(&params->ioctl_mtx);
1942	params->state = CTL_IOCTL_DATAMOVE;
1943	cv_broadcast(&params->sem);
1944	mtx_unlock(&params->ioctl_mtx);
1945}
1946
1947static void
1948ctl_ioctl_done(union ctl_io *io)
1949{
1950	struct ctl_fe_ioctl_params *params;
1951
1952	params = (struct ctl_fe_ioctl_params *)
1953		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1954
1955	mtx_lock(&params->ioctl_mtx);
1956	params->state = CTL_IOCTL_DONE;
1957	cv_broadcast(&params->sem);
1958	mtx_unlock(&params->ioctl_mtx);
1959}
1960
1961static void
1962ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1963{
1964	struct ctl_fe_ioctl_startstop_info *sd_info;
1965
1966	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1967
1968	sd_info->hs_info.status = metatask->status;
1969	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1970	sd_info->hs_info.luns_complete =
1971		metatask->taskinfo.startstop.luns_complete;
1972	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1973
1974	cv_broadcast(&sd_info->sem);
1975}
1976
1977static void
1978ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1979{
1980	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1981
1982	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1983
1984	mtx_lock(fe_bbr_info->lock);
1985	fe_bbr_info->bbr_info->status = metatask->status;
1986	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1987	fe_bbr_info->wakeup_done = 1;
1988	mtx_unlock(fe_bbr_info->lock);
1989
1990	cv_broadcast(&fe_bbr_info->sem);
1991}
1992
1993/*
1994 * Returns 0 for success, errno for failure.
1995 */
1996static int
1997ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1998		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1999{
2000	union ctl_io *io;
2001	int retval;
2002
2003	retval = 0;
2004
2005	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
2006
2007	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2008	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2009	     ooa_links)) {
2010		struct ctl_ooa_entry *entry;
2011
2012		/*
2013		 * If we've got more than we can fit, just count the
2014		 * remaining entries.
2015		 */
2016		if (*cur_fill_num >= ooa_hdr->alloc_num)
2017			continue;
2018
2019		entry = &kern_entries[*cur_fill_num];
2020
2021		entry->tag_num = io->scsiio.tag_num;
2022		entry->lun_num = lun->lun;
2023#ifdef CTL_TIME_IO
2024		entry->start_bt = io->io_hdr.start_bt;
2025#endif
2026		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2027		entry->cdb_len = io->scsiio.cdb_len;
2028		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2029			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2030
2031		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2032			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2033
2034		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2035			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2036
2037		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2038			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2039
2040		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2041			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2042	}
2043
2044	return (retval);
2045}
2046
2047static void *
2048ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2049		 size_t error_str_len)
2050{
2051	void *kptr;
2052
2053	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2054
2055	if (copyin(user_addr, kptr, len) != 0) {
2056		snprintf(error_str, error_str_len, "Error copying %d bytes "
2057			 "from user address %p to kernel address %p", len,
2058			 user_addr, kptr);
2059		free(kptr, M_CTL);
2060		return (NULL);
2061	}
2062
2063	return (kptr);
2064}
2065
2066static void
2067ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2068{
2069	int i;
2070
2071	if (be_args == NULL)
2072		return;
2073
2074	for (i = 0; i < num_be_args; i++) {
2075		free(be_args[i].kname, M_CTL);
2076		free(be_args[i].kvalue, M_CTL);
2077	}
2078
2079	free(be_args, M_CTL);
2080}
2081
2082static struct ctl_be_arg *
2083ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2084		char *error_str, size_t error_str_len)
2085{
2086	struct ctl_be_arg *args;
2087	int i;
2088
2089	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2090				error_str, error_str_len);
2091
2092	if (args == NULL)
2093		goto bailout;
2094
2095	for (i = 0; i < num_be_args; i++) {
2096		args[i].kname = NULL;
2097		args[i].kvalue = NULL;
2098	}
2099
2100	for (i = 0; i < num_be_args; i++) {
2101		uint8_t *tmpptr;
2102
2103		args[i].kname = ctl_copyin_alloc(args[i].name,
2104			args[i].namelen, error_str, error_str_len);
2105		if (args[i].kname == NULL)
2106			goto bailout;
2107
2108		if (args[i].kname[args[i].namelen - 1] != '\0') {
2109			snprintf(error_str, error_str_len, "Argument %d "
2110				 "name is not NUL-terminated", i);
2111			goto bailout;
2112		}
2113
2114		args[i].kvalue = NULL;
2115
2116		tmpptr = ctl_copyin_alloc(args[i].value,
2117			args[i].vallen, error_str, error_str_len);
2118		if (tmpptr == NULL)
2119			goto bailout;
2120
2121		args[i].kvalue = tmpptr;
2122
2123		if ((args[i].flags & CTL_BEARG_ASCII)
2124		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2125			snprintf(error_str, error_str_len, "Argument %d "
2126				 "value is not NUL-terminated", i);
2127			goto bailout;
2128		}
2129	}
2130
2131	return (args);
2132bailout:
2133
2134	ctl_free_args(num_be_args, args);
2135
2136	return (NULL);
2137}
2138
2139/*
2140 * Escape characters that are illegal or not recommended in XML.
2141 */
2142int
2143ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2144{
2145	int retval;
2146
2147	retval = 0;
2148
2149	for (; *str; str++) {
2150		switch (*str) {
2151		case '&':
2152			retval = sbuf_printf(sb, "&amp;");
2153			break;
2154		case '>':
2155			retval = sbuf_printf(sb, "&gt;");
2156			break;
2157		case '<':
2158			retval = sbuf_printf(sb, "&lt;");
2159			break;
2160		default:
2161			retval = sbuf_putc(sb, *str);
2162			break;
2163		}
2164
2165		if (retval != 0)
2166			break;
2167
2168	}
2169
2170	return (retval);
2171}
2172
2173static int
2174ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2175	  struct thread *td)
2176{
2177	struct ctl_softc *softc;
2178	int retval;
2179
2180	softc = control_softc;
2181
2182	retval = 0;
2183
2184	switch (cmd) {
2185	case CTL_IO: {
2186		union ctl_io *io;
2187		void *pool_tmp;
2188
2189		/*
2190		 * If we haven't been "enabled", don't allow any SCSI I/O
2191		 * to this FETD.
2192		 */
2193		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2194			retval = EPERM;
2195			break;
2196		}
2197
2198		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2199		if (io == NULL) {
2200			printf("ctl_ioctl: can't allocate ctl_io!\n");
2201			retval = ENOSPC;
2202			break;
2203		}
2204
2205		/*
2206		 * Need to save the pool reference so it doesn't get
2207		 * spammed by the user's ctl_io.
2208		 */
2209		pool_tmp = io->io_hdr.pool;
2210
2211		memcpy(io, (void *)addr, sizeof(*io));
2212
2213		io->io_hdr.pool = pool_tmp;
2214		/*
2215		 * No status yet, so make sure the status is set properly.
2216		 */
2217		io->io_hdr.status = CTL_STATUS_NONE;
2218
2219		/*
2220		 * The user sets the initiator ID, target and LUN IDs.
2221		 */
2222		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2223		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2224		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2225		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2226			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2227
2228		retval = ctl_ioctl_submit_wait(io);
2229
2230		if (retval != 0) {
2231			ctl_free_io(io);
2232			break;
2233		}
2234
2235		memcpy((void *)addr, io, sizeof(*io));
2236
2237		/* return this to our pool */
2238		ctl_free_io(io);
2239
2240		break;
2241	}
2242	case CTL_ENABLE_PORT:
2243	case CTL_DISABLE_PORT:
2244	case CTL_SET_PORT_WWNS: {
2245		struct ctl_frontend *fe;
2246		struct ctl_port_entry *entry;
2247
2248		entry = (struct ctl_port_entry *)addr;
2249
2250		mtx_lock(&softc->ctl_lock);
2251		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2252			int action, done;
2253
2254			action = 0;
2255			done = 0;
2256
2257			if ((entry->port_type == CTL_PORT_NONE)
2258			 && (entry->targ_port == fe->targ_port)) {
2259				/*
2260				 * If the user only wants to enable or
2261				 * disable or set WWNs on a specific port,
2262				 * do the operation and we're done.
2263				 */
2264				action = 1;
2265				done = 1;
2266			} else if (entry->port_type & fe->port_type) {
2267				/*
2268				 * Compare the user's type mask with the
2269				 * particular frontend type to see if we
2270				 * have a match.
2271				 */
2272				action = 1;
2273				done = 0;
2274
2275				/*
2276				 * Make sure the user isn't trying to set
2277				 * WWNs on multiple ports at the same time.
2278				 */
2279				if (cmd == CTL_SET_PORT_WWNS) {
2280					printf("%s: Can't set WWNs on "
2281					       "multiple ports\n", __func__);
2282					retval = EINVAL;
2283					break;
2284				}
2285			}
2286			if (action != 0) {
2287				/*
2288				 * XXX KDM we have to drop the lock here,
2289				 * because the online/offline operations
2290				 * can potentially block.  We need to
2291				 * reference count the frontends so they
2292				 * can't go away,
2293				 */
2294				mtx_unlock(&softc->ctl_lock);
2295
2296				if (cmd == CTL_ENABLE_PORT) {
2297					struct ctl_lun *lun;
2298
2299					STAILQ_FOREACH(lun, &softc->lun_list,
2300						       links) {
2301						fe->lun_enable(fe->targ_lun_arg,
2302						    lun->target,
2303						    lun->lun);
2304					}
2305
2306					ctl_frontend_online(fe);
2307				} else if (cmd == CTL_DISABLE_PORT) {
2308					struct ctl_lun *lun;
2309
2310					ctl_frontend_offline(fe);
2311
2312					STAILQ_FOREACH(lun, &softc->lun_list,
2313						       links) {
2314						fe->lun_disable(
2315						    fe->targ_lun_arg,
2316						    lun->target,
2317						    lun->lun);
2318					}
2319				}
2320
2321				mtx_lock(&softc->ctl_lock);
2322
2323				if (cmd == CTL_SET_PORT_WWNS)
2324					ctl_frontend_set_wwns(fe,
2325					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2326					    1 : 0, entry->wwnn,
2327					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2328					    1 : 0, entry->wwpn);
2329			}
2330			if (done != 0)
2331				break;
2332		}
2333		mtx_unlock(&softc->ctl_lock);
2334		break;
2335	}
2336	case CTL_GET_PORT_LIST: {
2337		struct ctl_frontend *fe;
2338		struct ctl_port_list *list;
2339		int i;
2340
2341		list = (struct ctl_port_list *)addr;
2342
2343		if (list->alloc_len != (list->alloc_num *
2344		    sizeof(struct ctl_port_entry))) {
2345			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2346			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2347			       "%zu\n", __func__, list->alloc_len,
2348			       list->alloc_num, sizeof(struct ctl_port_entry));
2349			retval = EINVAL;
2350			break;
2351		}
2352		list->fill_len = 0;
2353		list->fill_num = 0;
2354		list->dropped_num = 0;
2355		i = 0;
2356		mtx_lock(&softc->ctl_lock);
2357		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2358			struct ctl_port_entry entry, *list_entry;
2359
2360			if (list->fill_num >= list->alloc_num) {
2361				list->dropped_num++;
2362				continue;
2363			}
2364
2365			entry.port_type = fe->port_type;
2366			strlcpy(entry.port_name, fe->port_name,
2367				sizeof(entry.port_name));
2368			entry.targ_port = fe->targ_port;
2369			entry.physical_port = fe->physical_port;
2370			entry.virtual_port = fe->virtual_port;
2371			entry.wwnn = fe->wwnn;
2372			entry.wwpn = fe->wwpn;
2373			if (fe->status & CTL_PORT_STATUS_ONLINE)
2374				entry.online = 1;
2375			else
2376				entry.online = 0;
2377
2378			list_entry = &list->entries[i];
2379
2380			retval = copyout(&entry, list_entry, sizeof(entry));
2381			if (retval != 0) {
2382				printf("%s: CTL_GET_PORT_LIST: copyout "
2383				       "returned %d\n", __func__, retval);
2384				break;
2385			}
2386			i++;
2387			list->fill_num++;
2388			list->fill_len += sizeof(entry);
2389		}
2390		mtx_unlock(&softc->ctl_lock);
2391
2392		/*
2393		 * If this is non-zero, we had a copyout fault, so there's
2394		 * probably no point in attempting to set the status inside
2395		 * the structure.
2396		 */
2397		if (retval != 0)
2398			break;
2399
2400		if (list->dropped_num > 0)
2401			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2402		else
2403			list->status = CTL_PORT_LIST_OK;
2404		break;
2405	}
2406	case CTL_DUMP_OOA: {
2407		struct ctl_lun *lun;
2408		union ctl_io *io;
2409		char printbuf[128];
2410		struct sbuf sb;
2411
2412		mtx_lock(&softc->ctl_lock);
2413		printf("Dumping OOA queues:\n");
2414		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2415			for (io = (union ctl_io *)TAILQ_FIRST(
2416			     &lun->ooa_queue); io != NULL;
2417			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2418			     ooa_links)) {
2419				sbuf_new(&sb, printbuf, sizeof(printbuf),
2420					 SBUF_FIXEDLEN);
2421				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2422					    (intmax_t)lun->lun,
2423					    io->scsiio.tag_num,
2424					    (io->io_hdr.flags &
2425					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2426					    (io->io_hdr.flags &
2427					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2428					    (io->io_hdr.flags &
2429					    CTL_FLAG_ABORT) ? " ABORT" : "",
2430			                    (io->io_hdr.flags &
2431		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2432				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2433				sbuf_finish(&sb);
2434				printf("%s\n", sbuf_data(&sb));
2435			}
2436		}
2437		printf("OOA queues dump done\n");
2438		mtx_unlock(&softc->ctl_lock);
2439		break;
2440	}
2441	case CTL_GET_OOA: {
2442		struct ctl_lun *lun;
2443		struct ctl_ooa *ooa_hdr;
2444		struct ctl_ooa_entry *entries;
2445		uint32_t cur_fill_num;
2446
2447		ooa_hdr = (struct ctl_ooa *)addr;
2448
2449		if ((ooa_hdr->alloc_len == 0)
2450		 || (ooa_hdr->alloc_num == 0)) {
2451			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2452			       "must be non-zero\n", __func__,
2453			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2454			retval = EINVAL;
2455			break;
2456		}
2457
2458		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2459		    sizeof(struct ctl_ooa_entry))) {
2460			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2461			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2462			       __func__, ooa_hdr->alloc_len,
2463			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2464			retval = EINVAL;
2465			break;
2466		}
2467
2468		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2469		if (entries == NULL) {
2470			printf("%s: could not allocate %d bytes for OOA "
2471			       "dump\n", __func__, ooa_hdr->alloc_len);
2472			retval = ENOMEM;
2473			break;
2474		}
2475
2476		mtx_lock(&softc->ctl_lock);
2477		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2478		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2479		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2480			mtx_unlock(&softc->ctl_lock);
2481			free(entries, M_CTL);
2482			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2483			       __func__, (uintmax_t)ooa_hdr->lun_num);
2484			retval = EINVAL;
2485			break;
2486		}
2487
2488		cur_fill_num = 0;
2489
2490		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2491			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2492				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2493					ooa_hdr, entries);
2494				if (retval != 0)
2495					break;
2496			}
2497			if (retval != 0) {
2498				mtx_unlock(&softc->ctl_lock);
2499				free(entries, M_CTL);
2500				break;
2501			}
2502		} else {
2503			lun = softc->ctl_luns[ooa_hdr->lun_num];
2504
2505			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2506						    entries);
2507		}
2508		mtx_unlock(&softc->ctl_lock);
2509
2510		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2511		ooa_hdr->fill_len = ooa_hdr->fill_num *
2512			sizeof(struct ctl_ooa_entry);
2513		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2514		if (retval != 0) {
2515			printf("%s: error copying out %d bytes for OOA dump\n",
2516			       __func__, ooa_hdr->fill_len);
2517		}
2518
2519		getbintime(&ooa_hdr->cur_bt);
2520
2521		if (cur_fill_num > ooa_hdr->alloc_num) {
2522			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2523			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2524		} else {
2525			ooa_hdr->dropped_num = 0;
2526			ooa_hdr->status = CTL_OOA_OK;
2527		}
2528
2529		free(entries, M_CTL);
2530		break;
2531	}
2532	case CTL_CHECK_OOA: {
2533		union ctl_io *io;
2534		struct ctl_lun *lun;
2535		struct ctl_ooa_info *ooa_info;
2536
2537
2538		ooa_info = (struct ctl_ooa_info *)addr;
2539
2540		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2541			ooa_info->status = CTL_OOA_INVALID_LUN;
2542			break;
2543		}
2544		mtx_lock(&softc->ctl_lock);
2545		lun = softc->ctl_luns[ooa_info->lun_id];
2546		if (lun == NULL) {
2547			mtx_unlock(&softc->ctl_lock);
2548			ooa_info->status = CTL_OOA_INVALID_LUN;
2549			break;
2550		}
2551
2552		ooa_info->num_entries = 0;
2553		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2554		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2555		     &io->io_hdr, ooa_links)) {
2556			ooa_info->num_entries++;
2557		}
2558
2559		mtx_unlock(&softc->ctl_lock);
2560		ooa_info->status = CTL_OOA_SUCCESS;
2561
2562		break;
2563	}
2564	case CTL_HARD_START:
2565	case CTL_HARD_STOP: {
2566		struct ctl_fe_ioctl_startstop_info ss_info;
2567		struct cfi_metatask *metatask;
2568		struct mtx hs_mtx;
2569
2570		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2571
2572		cv_init(&ss_info.sem, "hard start/stop cv" );
2573
2574		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2575		if (metatask == NULL) {
2576			retval = ENOMEM;
2577			mtx_destroy(&hs_mtx);
2578			break;
2579		}
2580
2581		if (cmd == CTL_HARD_START)
2582			metatask->tasktype = CFI_TASK_STARTUP;
2583		else
2584			metatask->tasktype = CFI_TASK_SHUTDOWN;
2585
2586		metatask->callback = ctl_ioctl_hard_startstop_callback;
2587		metatask->callback_arg = &ss_info;
2588
2589		cfi_action(metatask);
2590
2591		/* Wait for the callback */
2592		mtx_lock(&hs_mtx);
2593		cv_wait_sig(&ss_info.sem, &hs_mtx);
2594		mtx_unlock(&hs_mtx);
2595
2596		/*
2597		 * All information has been copied from the metatask by the
2598		 * time cv_broadcast() is called, so we free the metatask here.
2599		 */
2600		cfi_free_metatask(metatask);
2601
2602		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2603
2604		mtx_destroy(&hs_mtx);
2605		break;
2606	}
2607	case CTL_BBRREAD: {
2608		struct ctl_bbrread_info *bbr_info;
2609		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2610		struct mtx bbr_mtx;
2611		struct cfi_metatask *metatask;
2612
2613		bbr_info = (struct ctl_bbrread_info *)addr;
2614
2615		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2616
2617		bzero(&bbr_mtx, sizeof(bbr_mtx));
2618		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2619
2620		fe_bbr_info.bbr_info = bbr_info;
2621		fe_bbr_info.lock = &bbr_mtx;
2622
2623		cv_init(&fe_bbr_info.sem, "BBR read cv");
2624		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2625
2626		if (metatask == NULL) {
2627			mtx_destroy(&bbr_mtx);
2628			cv_destroy(&fe_bbr_info.sem);
2629			retval = ENOMEM;
2630			break;
2631		}
2632		metatask->tasktype = CFI_TASK_BBRREAD;
2633		metatask->callback = ctl_ioctl_bbrread_callback;
2634		metatask->callback_arg = &fe_bbr_info;
2635		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2636		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2637		metatask->taskinfo.bbrread.len = bbr_info->len;
2638
2639		cfi_action(metatask);
2640
2641		mtx_lock(&bbr_mtx);
2642		while (fe_bbr_info.wakeup_done == 0)
2643			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2644		mtx_unlock(&bbr_mtx);
2645
2646		bbr_info->status = metatask->status;
2647		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2648		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2649		memcpy(&bbr_info->sense_data,
2650		       &metatask->taskinfo.bbrread.sense_data,
2651		       ctl_min(sizeof(bbr_info->sense_data),
2652			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2653
2654		cfi_free_metatask(metatask);
2655
2656		mtx_destroy(&bbr_mtx);
2657		cv_destroy(&fe_bbr_info.sem);
2658
2659		break;
2660	}
2661	case CTL_DELAY_IO: {
2662		struct ctl_io_delay_info *delay_info;
2663#ifdef CTL_IO_DELAY
2664		struct ctl_lun *lun;
2665#endif /* CTL_IO_DELAY */
2666
2667		delay_info = (struct ctl_io_delay_info *)addr;
2668
2669#ifdef CTL_IO_DELAY
2670		mtx_lock(&softc->ctl_lock);
2671
2672		if ((delay_info->lun_id > CTL_MAX_LUNS)
2673		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2674			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2675		} else {
2676			lun = softc->ctl_luns[delay_info->lun_id];
2677
2678			delay_info->status = CTL_DELAY_STATUS_OK;
2679
2680			switch (delay_info->delay_type) {
2681			case CTL_DELAY_TYPE_CONT:
2682				break;
2683			case CTL_DELAY_TYPE_ONESHOT:
2684				break;
2685			default:
2686				delay_info->status =
2687					CTL_DELAY_STATUS_INVALID_TYPE;
2688				break;
2689			}
2690
2691			switch (delay_info->delay_loc) {
2692			case CTL_DELAY_LOC_DATAMOVE:
2693				lun->delay_info.datamove_type =
2694					delay_info->delay_type;
2695				lun->delay_info.datamove_delay =
2696					delay_info->delay_secs;
2697				break;
2698			case CTL_DELAY_LOC_DONE:
2699				lun->delay_info.done_type =
2700					delay_info->delay_type;
2701				lun->delay_info.done_delay =
2702					delay_info->delay_secs;
2703				break;
2704			default:
2705				delay_info->status =
2706					CTL_DELAY_STATUS_INVALID_LOC;
2707				break;
2708			}
2709		}
2710
2711		mtx_unlock(&softc->ctl_lock);
2712#else
2713		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2714#endif /* CTL_IO_DELAY */
2715		break;
2716	}
2717	case CTL_REALSYNC_SET: {
2718		int *syncstate;
2719
2720		syncstate = (int *)addr;
2721
2722		mtx_lock(&softc->ctl_lock);
2723		switch (*syncstate) {
2724		case 0:
2725			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2726			break;
2727		case 1:
2728			softc->flags |= CTL_FLAG_REAL_SYNC;
2729			break;
2730		default:
2731			retval = EINVAL;
2732			break;
2733		}
2734		mtx_unlock(&softc->ctl_lock);
2735		break;
2736	}
2737	case CTL_REALSYNC_GET: {
2738		int *syncstate;
2739
2740		syncstate = (int*)addr;
2741
2742		mtx_lock(&softc->ctl_lock);
2743		if (softc->flags & CTL_FLAG_REAL_SYNC)
2744			*syncstate = 1;
2745		else
2746			*syncstate = 0;
2747		mtx_unlock(&softc->ctl_lock);
2748
2749		break;
2750	}
2751	case CTL_SETSYNC:
2752	case CTL_GETSYNC: {
2753		struct ctl_sync_info *sync_info;
2754		struct ctl_lun *lun;
2755
2756		sync_info = (struct ctl_sync_info *)addr;
2757
2758		mtx_lock(&softc->ctl_lock);
2759		lun = softc->ctl_luns[sync_info->lun_id];
2760		if (lun == NULL) {
2761			mtx_unlock(&softc->ctl_lock);
2762			sync_info->status = CTL_GS_SYNC_NO_LUN;
2763		}
2764		/*
2765		 * Get or set the sync interval.  We're not bounds checking
2766		 * in the set case, hopefully the user won't do something
2767		 * silly.
2768		 */
2769		if (cmd == CTL_GETSYNC)
2770			sync_info->sync_interval = lun->sync_interval;
2771		else
2772			lun->sync_interval = sync_info->sync_interval;
2773
2774		mtx_unlock(&softc->ctl_lock);
2775
2776		sync_info->status = CTL_GS_SYNC_OK;
2777
2778		break;
2779	}
2780	case CTL_GETSTATS: {
2781		struct ctl_stats *stats;
2782		struct ctl_lun *lun;
2783		int i;
2784
2785		stats = (struct ctl_stats *)addr;
2786
2787		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2788		     stats->alloc_len) {
2789			stats->status = CTL_SS_NEED_MORE_SPACE;
2790			stats->num_luns = softc->num_luns;
2791			break;
2792		}
2793		/*
2794		 * XXX KDM no locking here.  If the LUN list changes,
2795		 * things can blow up.
2796		 */
2797		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2798		     i++, lun = STAILQ_NEXT(lun, links)) {
2799			retval = copyout(&lun->stats, &stats->lun_stats[i],
2800					 sizeof(lun->stats));
2801			if (retval != 0)
2802				break;
2803		}
2804		stats->num_luns = softc->num_luns;
2805		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2806				 softc->num_luns;
2807		stats->status = CTL_SS_OK;
2808#ifdef CTL_TIME_IO
2809		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2810#else
2811		stats->flags = CTL_STATS_FLAG_NONE;
2812#endif
2813		getnanouptime(&stats->timestamp);
2814		break;
2815	}
2816	case CTL_ERROR_INJECT: {
2817		struct ctl_error_desc *err_desc, *new_err_desc;
2818		struct ctl_lun *lun;
2819
2820		err_desc = (struct ctl_error_desc *)addr;
2821
2822		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2823				      M_WAITOK | M_ZERO);
2824		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2825
2826		mtx_lock(&softc->ctl_lock);
2827		lun = softc->ctl_luns[err_desc->lun_id];
2828		if (lun == NULL) {
2829			mtx_unlock(&softc->ctl_lock);
2830			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2831			       __func__, (uintmax_t)err_desc->lun_id);
2832			retval = EINVAL;
2833			break;
2834		}
2835
2836		/*
2837		 * We could do some checking here to verify the validity
2838		 * of the request, but given the complexity of error
2839		 * injection requests, the checking logic would be fairly
2840		 * complex.
2841		 *
2842		 * For now, if the request is invalid, it just won't get
2843		 * executed and might get deleted.
2844		 */
2845		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2846
2847		/*
2848		 * XXX KDM check to make sure the serial number is unique,
2849		 * in case we somehow manage to wrap.  That shouldn't
2850		 * happen for a very long time, but it's the right thing to
2851		 * do.
2852		 */
2853		new_err_desc->serial = lun->error_serial;
2854		err_desc->serial = lun->error_serial;
2855		lun->error_serial++;
2856
2857		mtx_unlock(&softc->ctl_lock);
2858		break;
2859	}
2860	case CTL_ERROR_INJECT_DELETE: {
2861		struct ctl_error_desc *delete_desc, *desc, *desc2;
2862		struct ctl_lun *lun;
2863		int delete_done;
2864
2865		delete_desc = (struct ctl_error_desc *)addr;
2866		delete_done = 0;
2867
2868		mtx_lock(&softc->ctl_lock);
2869		lun = softc->ctl_luns[delete_desc->lun_id];
2870		if (lun == NULL) {
2871			mtx_unlock(&softc->ctl_lock);
2872			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2873			       __func__, (uintmax_t)delete_desc->lun_id);
2874			retval = EINVAL;
2875			break;
2876		}
2877		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2878			if (desc->serial != delete_desc->serial)
2879				continue;
2880
2881			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2882				      links);
2883			free(desc, M_CTL);
2884			delete_done = 1;
2885		}
2886		mtx_unlock(&softc->ctl_lock);
2887		if (delete_done == 0) {
2888			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2889			       "error serial %ju on LUN %u\n", __func__,
2890			       delete_desc->serial, delete_desc->lun_id);
2891			retval = EINVAL;
2892			break;
2893		}
2894		break;
2895	}
2896	case CTL_DUMP_STRUCTS: {
2897		int i, j, k;
2898		struct ctl_frontend *fe;
2899
2900		printf("CTL IID to WWPN map start:\n");
2901		for (i = 0; i < CTL_MAX_PORTS; i++) {
2902			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2903				if (softc->wwpn_iid[i][j].in_use == 0)
2904					continue;
2905
2906				printf("port %d iid %u WWPN %#jx\n",
2907				       softc->wwpn_iid[i][j].port,
2908				       softc->wwpn_iid[i][j].iid,
2909				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2910			}
2911		}
2912		printf("CTL IID to WWPN map end\n");
2913		printf("CTL Persistent Reservation information start:\n");
2914		for (i = 0; i < CTL_MAX_LUNS; i++) {
2915			struct ctl_lun *lun;
2916
2917			lun = softc->ctl_luns[i];
2918
2919			if ((lun == NULL)
2920			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2921				continue;
2922
2923			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2924				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2925					if (lun->per_res[j+k].registered == 0)
2926						continue;
2927					printf("LUN %d port %d iid %d key "
2928					       "%#jx\n", i, j, k,
2929					       (uintmax_t)scsi_8btou64(
2930					       lun->per_res[j+k].res_key.key));
2931				}
2932			}
2933		}
2934		printf("CTL Persistent Reservation information end\n");
2935		printf("CTL Frontends:\n");
2936		/*
2937		 * XXX KDM calling this without a lock.  We'd likely want
2938		 * to drop the lock before calling the frontend's dump
2939		 * routine anyway.
2940		 */
2941		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2942			printf("Frontend %s Type %u pport %d vport %d WWNN "
2943			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2944			       fe->physical_port, fe->virtual_port,
2945			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2946
2947			/*
2948			 * Frontends are not required to support the dump
2949			 * routine.
2950			 */
2951			if (fe->fe_dump == NULL)
2952				continue;
2953
2954			fe->fe_dump();
2955		}
2956		printf("CTL Frontend information end\n");
2957		break;
2958	}
2959	case CTL_LUN_REQ: {
2960		struct ctl_lun_req *lun_req;
2961		struct ctl_backend_driver *backend;
2962
2963		lun_req = (struct ctl_lun_req *)addr;
2964
2965		backend = ctl_backend_find(lun_req->backend);
2966		if (backend == NULL) {
2967			lun_req->status = CTL_LUN_ERROR;
2968			snprintf(lun_req->error_str,
2969				 sizeof(lun_req->error_str),
2970				 "Backend \"%s\" not found.",
2971				 lun_req->backend);
2972			break;
2973		}
2974		if (lun_req->num_be_args > 0) {
2975			lun_req->kern_be_args = ctl_copyin_args(
2976				lun_req->num_be_args,
2977				lun_req->be_args,
2978				lun_req->error_str,
2979				sizeof(lun_req->error_str));
2980			if (lun_req->kern_be_args == NULL) {
2981				lun_req->status = CTL_LUN_ERROR;
2982				break;
2983			}
2984		}
2985
2986		retval = backend->ioctl(dev, cmd, addr, flag, td);
2987
2988		if (lun_req->num_be_args > 0) {
2989			ctl_free_args(lun_req->num_be_args,
2990				      lun_req->kern_be_args);
2991		}
2992		break;
2993	}
2994	case CTL_LUN_LIST: {
2995		struct sbuf *sb;
2996		struct ctl_lun *lun;
2997		struct ctl_lun_list *list;
2998		struct ctl_be_lun_option *opt;
2999
3000		list = (struct ctl_lun_list *)addr;
3001
3002		/*
3003		 * Allocate a fixed length sbuf here, based on the length
3004		 * of the user's buffer.  We could allocate an auto-extending
3005		 * buffer, and then tell the user how much larger our
3006		 * amount of data is than his buffer, but that presents
3007		 * some problems:
3008		 *
3009		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3010		 *     we can't hold a lock while calling them with an
3011		 *     auto-extending buffer.
3012 		 *
3013		 * 2.  There is not currently a LUN reference counting
3014		 *     mechanism, outside of outstanding transactions on
3015		 *     the LUN's OOA queue.  So a LUN could go away on us
3016		 *     while we're getting the LUN number, backend-specific
3017		 *     information, etc.  Thus, given the way things
3018		 *     currently work, we need to hold the CTL lock while
3019		 *     grabbing LUN information.
3020		 *
3021		 * So, from the user's standpoint, the best thing to do is
3022		 * allocate what he thinks is a reasonable buffer length,
3023		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3024		 * double the buffer length and try again.  (And repeat
3025		 * that until he succeeds.)
3026		 */
3027		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3028		if (sb == NULL) {
3029			list->status = CTL_LUN_LIST_ERROR;
3030			snprintf(list->error_str, sizeof(list->error_str),
3031				 "Unable to allocate %d bytes for LUN list",
3032				 list->alloc_len);
3033			break;
3034		}
3035
3036		sbuf_printf(sb, "<ctllunlist>\n");
3037
3038		mtx_lock(&softc->ctl_lock);
3039
3040		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3041			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3042					     (uintmax_t)lun->lun);
3043
3044			/*
3045			 * Bail out as soon as we see that we've overfilled
3046			 * the buffer.
3047			 */
3048			if (retval != 0)
3049				break;
3050
3051			retval = sbuf_printf(sb, "<backend_type>%s"
3052					     "</backend_type>\n",
3053					     (lun->backend == NULL) ?  "none" :
3054					     lun->backend->name);
3055
3056			if (retval != 0)
3057				break;
3058
3059			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3060					     lun->be_lun->lun_type);
3061
3062			if (retval != 0)
3063				break;
3064
3065			if (lun->backend == NULL) {
3066				retval = sbuf_printf(sb, "</lun>\n");
3067				if (retval != 0)
3068					break;
3069				continue;
3070			}
3071
3072			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3073					     (lun->be_lun->maxlba > 0) ?
3074					     lun->be_lun->maxlba + 1 : 0);
3075
3076			if (retval != 0)
3077				break;
3078
3079			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3080					     lun->be_lun->blocksize);
3081
3082			if (retval != 0)
3083				break;
3084
3085			retval = sbuf_printf(sb, "<serial_number>");
3086
3087			if (retval != 0)
3088				break;
3089
3090			retval = ctl_sbuf_printf_esc(sb,
3091						     lun->be_lun->serial_num);
3092
3093			if (retval != 0)
3094				break;
3095
3096			retval = sbuf_printf(sb, "</serial_number>\n");
3097
3098			if (retval != 0)
3099				break;
3100
3101			retval = sbuf_printf(sb, "<device_id>");
3102
3103			if (retval != 0)
3104				break;
3105
3106			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3107
3108			if (retval != 0)
3109				break;
3110
3111			retval = sbuf_printf(sb, "</device_id>\n");
3112
3113			if (retval != 0)
3114				break;
3115
3116			if (lun->backend->lun_info != NULL) {
3117				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3118				if (retval != 0)
3119					break;
3120			}
3121			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3122				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3123				if (retval != 0)
3124					break;
3125			}
3126
3127			retval = sbuf_printf(sb, "</lun>\n");
3128
3129			if (retval != 0)
3130				break;
3131		}
3132		mtx_unlock(&softc->ctl_lock);
3133
3134		if ((retval != 0)
3135		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3136			retval = 0;
3137			sbuf_delete(sb);
3138			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3139			snprintf(list->error_str, sizeof(list->error_str),
3140				 "Out of space, %d bytes is too small",
3141				 list->alloc_len);
3142			break;
3143		}
3144
3145		sbuf_finish(sb);
3146
3147		retval = copyout(sbuf_data(sb), list->lun_xml,
3148				 sbuf_len(sb) + 1);
3149
3150		list->fill_len = sbuf_len(sb) + 1;
3151		list->status = CTL_LUN_LIST_OK;
3152		sbuf_delete(sb);
3153		break;
3154	}
3155	case CTL_ISCSI: {
3156		struct ctl_iscsi *ci;
3157		struct ctl_frontend *fe;
3158
3159		ci = (struct ctl_iscsi *)addr;
3160
3161		mtx_lock(&softc->ctl_lock);
3162		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3163			if (strcmp(fe->port_name, "iscsi") == 0)
3164				break;
3165		}
3166		mtx_unlock(&softc->ctl_lock);
3167
3168		if (fe == NULL) {
3169			ci->status = CTL_ISCSI_ERROR;
3170			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3171			break;
3172		}
3173
3174		retval = fe->ioctl(dev, cmd, addr, flag, td);
3175		break;
3176	}
3177	default: {
3178		/* XXX KDM should we fix this? */
3179#if 0
3180		struct ctl_backend_driver *backend;
3181		unsigned int type;
3182		int found;
3183
3184		found = 0;
3185
3186		/*
3187		 * We encode the backend type as the ioctl type for backend
3188		 * ioctls.  So parse it out here, and then search for a
3189		 * backend of this type.
3190		 */
3191		type = _IOC_TYPE(cmd);
3192
3193		STAILQ_FOREACH(backend, &softc->be_list, links) {
3194			if (backend->type == type) {
3195				found = 1;
3196				break;
3197			}
3198		}
3199		if (found == 0) {
3200			printf("ctl: unknown ioctl command %#lx or backend "
3201			       "%d\n", cmd, type);
3202			retval = EINVAL;
3203			break;
3204		}
3205		retval = backend->ioctl(dev, cmd, addr, flag, td);
3206#endif
3207		retval = ENOTTY;
3208		break;
3209	}
3210	}
3211	return (retval);
3212}
3213
3214uint32_t
3215ctl_get_initindex(struct ctl_nexus *nexus)
3216{
3217	if (nexus->targ_port < CTL_MAX_PORTS)
3218		return (nexus->initid.id +
3219			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3220	else
3221		return (nexus->initid.id +
3222		       ((nexus->targ_port - CTL_MAX_PORTS) *
3223			CTL_MAX_INIT_PER_PORT));
3224}
3225
3226uint32_t
3227ctl_get_resindex(struct ctl_nexus *nexus)
3228{
3229	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3230}
3231
3232uint32_t
3233ctl_port_idx(int port_num)
3234{
3235	if (port_num < CTL_MAX_PORTS)
3236		return(port_num);
3237	else
3238		return(port_num - CTL_MAX_PORTS);
3239}
3240
3241/*
3242 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3243 * that are a power of 2.
3244 */
3245int
3246ctl_ffz(uint32_t *mask, uint32_t size)
3247{
3248	uint32_t num_chunks, num_pieces;
3249	int i, j;
3250
3251	num_chunks = (size >> 5);
3252	if (num_chunks == 0)
3253		num_chunks++;
3254	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3255
3256	for (i = 0; i < num_chunks; i++) {
3257		for (j = 0; j < num_pieces; j++) {
3258			if ((mask[i] & (1 << j)) == 0)
3259				return ((i << 5) + j);
3260		}
3261	}
3262
3263	return (-1);
3264}
3265
3266int
3267ctl_set_mask(uint32_t *mask, uint32_t bit)
3268{
3269	uint32_t chunk, piece;
3270
3271	chunk = bit >> 5;
3272	piece = bit % (sizeof(uint32_t) * 8);
3273
3274	if ((mask[chunk] & (1 << piece)) != 0)
3275		return (-1);
3276	else
3277		mask[chunk] |= (1 << piece);
3278
3279	return (0);
3280}
3281
3282int
3283ctl_clear_mask(uint32_t *mask, uint32_t bit)
3284{
3285	uint32_t chunk, piece;
3286
3287	chunk = bit >> 5;
3288	piece = bit % (sizeof(uint32_t) * 8);
3289
3290	if ((mask[chunk] & (1 << piece)) == 0)
3291		return (-1);
3292	else
3293		mask[chunk] &= ~(1 << piece);
3294
3295	return (0);
3296}
3297
3298int
3299ctl_is_set(uint32_t *mask, uint32_t bit)
3300{
3301	uint32_t chunk, piece;
3302
3303	chunk = bit >> 5;
3304	piece = bit % (sizeof(uint32_t) * 8);
3305
3306	if ((mask[chunk] & (1 << piece)) == 0)
3307		return (0);
3308	else
3309		return (1);
3310}
3311
3312#ifdef unused
3313/*
3314 * The bus, target and lun are optional, they can be filled in later.
3315 * can_wait is used to determine whether we can wait on the malloc or not.
3316 */
3317union ctl_io*
3318ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3319	      uint32_t targ_lun, int can_wait)
3320{
3321	union ctl_io *io;
3322
3323	if (can_wait)
3324		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3325	else
3326		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3327
3328	if (io != NULL) {
3329		io->io_hdr.io_type = io_type;
3330		io->io_hdr.targ_port = targ_port;
3331		/*
3332		 * XXX KDM this needs to change/go away.  We need to move
3333		 * to a preallocated pool of ctl_scsiio structures.
3334		 */
3335		io->io_hdr.nexus.targ_target.id = targ_target;
3336		io->io_hdr.nexus.targ_lun = targ_lun;
3337	}
3338
3339	return (io);
3340}
3341
3342void
3343ctl_kfree_io(union ctl_io *io)
3344{
3345	free(io, M_CTL);
3346}
3347#endif /* unused */
3348
3349/*
3350 * ctl_softc, pool_type, total_ctl_io are passed in.
3351 * npool is passed out.
3352 */
3353int
3354ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3355		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3356{
3357	uint32_t i;
3358	union ctl_io *cur_io, *next_io;
3359	struct ctl_io_pool *pool;
3360	int retval;
3361
3362	retval = 0;
3363
3364	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3365					    M_NOWAIT | M_ZERO);
3366	if (pool == NULL) {
3367		retval = ENOMEM;
3368		goto bailout;
3369	}
3370
3371	pool->type = pool_type;
3372	pool->ctl_softc = ctl_softc;
3373
3374	mtx_lock(&ctl_softc->pool_lock);
3375	pool->id = ctl_softc->cur_pool_id++;
3376	mtx_unlock(&ctl_softc->pool_lock);
3377
3378	pool->flags = CTL_POOL_FLAG_NONE;
3379	pool->refcount = 1;		/* Reference for validity. */
3380	STAILQ_INIT(&pool->free_queue);
3381
3382	/*
3383	 * XXX KDM other options here:
3384	 * - allocate a page at a time
3385	 * - allocate one big chunk of memory.
3386	 * Page allocation might work well, but would take a little more
3387	 * tracking.
3388	 */
3389	for (i = 0; i < total_ctl_io; i++) {
3390		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3391						M_NOWAIT);
3392		if (cur_io == NULL) {
3393			retval = ENOMEM;
3394			break;
3395		}
3396		cur_io->io_hdr.pool = pool;
3397		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3398		pool->total_ctl_io++;
3399		pool->free_ctl_io++;
3400	}
3401
3402	if (retval != 0) {
3403		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3404		     cur_io != NULL; cur_io = next_io) {
3405			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3406							      links);
3407			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3408				      ctl_io_hdr, links);
3409			free(cur_io, M_CTL);
3410		}
3411
3412		free(pool, M_CTL);
3413		goto bailout;
3414	}
3415	mtx_lock(&ctl_softc->pool_lock);
3416	ctl_softc->num_pools++;
3417	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3418	/*
3419	 * Increment our usage count if this is an external consumer, so we
3420	 * can't get unloaded until the external consumer (most likely a
3421	 * FETD) unloads and frees his pool.
3422	 *
3423	 * XXX KDM will this increment the caller's module use count, or
3424	 * mine?
3425	 */
3426#if 0
3427	if ((pool_type != CTL_POOL_EMERGENCY)
3428	 && (pool_type != CTL_POOL_INTERNAL)
3429	 && (pool_type != CTL_POOL_IOCTL)
3430	 && (pool_type != CTL_POOL_4OTHERSC))
3431		MOD_INC_USE_COUNT;
3432#endif
3433
3434	mtx_unlock(&ctl_softc->pool_lock);
3435
3436	*npool = pool;
3437
3438bailout:
3439
3440	return (retval);
3441}
3442
3443static int
3444ctl_pool_acquire(struct ctl_io_pool *pool)
3445{
3446
3447	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3448
3449	if (pool->flags & CTL_POOL_FLAG_INVALID)
3450		return (EINVAL);
3451
3452	pool->refcount++;
3453
3454	return (0);
3455}
3456
3457static void
3458ctl_pool_release(struct ctl_io_pool *pool)
3459{
3460	struct ctl_softc *ctl_softc = pool->ctl_softc;
3461	union ctl_io *io;
3462
3463	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3464
3465	if (--pool->refcount != 0)
3466		return;
3467
3468	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3469		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3470			      links);
3471		free(io, M_CTL);
3472	}
3473
3474	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3475	ctl_softc->num_pools--;
3476
3477	/*
3478	 * XXX KDM will this decrement the caller's usage count or mine?
3479	 */
3480#if 0
3481	if ((pool->type != CTL_POOL_EMERGENCY)
3482	 && (pool->type != CTL_POOL_INTERNAL)
3483	 && (pool->type != CTL_POOL_IOCTL))
3484		MOD_DEC_USE_COUNT;
3485#endif
3486
3487	free(pool, M_CTL);
3488}
3489
3490void
3491ctl_pool_free(struct ctl_io_pool *pool)
3492{
3493	struct ctl_softc *ctl_softc;
3494
3495	if (pool == NULL)
3496		return;
3497
3498	ctl_softc = pool->ctl_softc;
3499	mtx_lock(&ctl_softc->pool_lock);
3500	pool->flags |= CTL_POOL_FLAG_INVALID;
3501	ctl_pool_release(pool);
3502	mtx_unlock(&ctl_softc->pool_lock);
3503}
3504
3505/*
3506 * This routine does not block (except for spinlocks of course).
3507 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3508 * possible.
3509 */
3510union ctl_io *
3511ctl_alloc_io(void *pool_ref)
3512{
3513	union ctl_io *io;
3514	struct ctl_softc *ctl_softc;
3515	struct ctl_io_pool *pool, *npool;
3516	struct ctl_io_pool *emergency_pool;
3517
3518	pool = (struct ctl_io_pool *)pool_ref;
3519
3520	if (pool == NULL) {
3521		printf("%s: pool is NULL\n", __func__);
3522		return (NULL);
3523	}
3524
3525	emergency_pool = NULL;
3526
3527	ctl_softc = pool->ctl_softc;
3528
3529	mtx_lock(&ctl_softc->pool_lock);
3530	/*
3531	 * First, try to get the io structure from the user's pool.
3532	 */
3533	if (ctl_pool_acquire(pool) == 0) {
3534		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3535		if (io != NULL) {
3536			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3537			pool->total_allocated++;
3538			pool->free_ctl_io--;
3539			mtx_unlock(&ctl_softc->pool_lock);
3540			return (io);
3541		} else
3542			ctl_pool_release(pool);
3543	}
3544	/*
3545	 * If he doesn't have any io structures left, search for an
3546	 * emergency pool and grab one from there.
3547	 */
3548	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3549		if (npool->type != CTL_POOL_EMERGENCY)
3550			continue;
3551
3552		if (ctl_pool_acquire(npool) != 0)
3553			continue;
3554
3555		emergency_pool = npool;
3556
3557		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3558		if (io != NULL) {
3559			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3560			npool->total_allocated++;
3561			npool->free_ctl_io--;
3562			mtx_unlock(&ctl_softc->pool_lock);
3563			return (io);
3564		} else
3565			ctl_pool_release(npool);
3566	}
3567
3568	/* Drop the spinlock before we malloc */
3569	mtx_unlock(&ctl_softc->pool_lock);
3570
3571	/*
3572	 * The emergency pool (if it exists) didn't have one, so try an
3573	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3574	 */
3575	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3576	if (io != NULL) {
3577		/*
3578		 * If the emergency pool exists but is empty, add this
3579		 * ctl_io to its list when it gets freed.
3580		 */
3581		if (emergency_pool != NULL) {
3582			mtx_lock(&ctl_softc->pool_lock);
3583			if (ctl_pool_acquire(emergency_pool) == 0) {
3584				io->io_hdr.pool = emergency_pool;
3585				emergency_pool->total_ctl_io++;
3586				/*
3587				 * Need to bump this, otherwise
3588				 * total_allocated and total_freed won't
3589				 * match when we no longer have anything
3590				 * outstanding.
3591				 */
3592				emergency_pool->total_allocated++;
3593			}
3594			mtx_unlock(&ctl_softc->pool_lock);
3595		} else
3596			io->io_hdr.pool = NULL;
3597	}
3598
3599	return (io);
3600}
3601
3602void
3603ctl_free_io(union ctl_io *io)
3604{
3605	if (io == NULL)
3606		return;
3607
3608	/*
3609	 * If this ctl_io has a pool, return it to that pool.
3610	 */
3611	if (io->io_hdr.pool != NULL) {
3612		struct ctl_io_pool *pool;
3613#if 0
3614		struct ctl_softc *ctl_softc;
3615		union ctl_io *tmp_io;
3616		unsigned long xflags;
3617		int i;
3618
3619		ctl_softc = control_softc;
3620#endif
3621
3622		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3623
3624		mtx_lock(&pool->ctl_softc->pool_lock);
3625#if 0
3626		save_flags(xflags);
3627
3628		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3629		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3630		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3631		     links)) {
3632			if (tmp_io == io) {
3633				printf("%s: %p is still on the task queue!\n",
3634				       __func__, tmp_io);
3635				printf("%s: (%d): type %d "
3636				       "msg %d cdb %x iptl: "
3637				       "%d:%d:%d:%d tag 0x%04x "
3638				       "flg %#lx\n",
3639					__func__, i,
3640					tmp_io->io_hdr.io_type,
3641					tmp_io->io_hdr.msg_type,
3642					tmp_io->scsiio.cdb[0],
3643					tmp_io->io_hdr.nexus.initid.id,
3644					tmp_io->io_hdr.nexus.targ_port,
3645					tmp_io->io_hdr.nexus.targ_target.id,
3646					tmp_io->io_hdr.nexus.targ_lun,
3647					(tmp_io->io_hdr.io_type ==
3648					CTL_IO_TASK) ?
3649					tmp_io->taskio.tag_num :
3650					tmp_io->scsiio.tag_num,
3651					xflags);
3652				panic("I/O still on the task queue!");
3653			}
3654		}
3655#endif
3656		io->io_hdr.io_type = 0xff;
3657		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3658		pool->total_freed++;
3659		pool->free_ctl_io++;
3660		ctl_pool_release(pool);
3661		mtx_unlock(&pool->ctl_softc->pool_lock);
3662	} else {
3663		/*
3664		 * Otherwise, just free it.  We probably malloced it and
3665		 * the emergency pool wasn't available.
3666		 */
3667		free(io, M_CTL);
3668	}
3669
3670}
3671
3672void
3673ctl_zero_io(union ctl_io *io)
3674{
3675	void *pool_ref;
3676
3677	if (io == NULL)
3678		return;
3679
3680	/*
3681	 * May need to preserve linked list pointers at some point too.
3682	 */
3683	pool_ref = io->io_hdr.pool;
3684
3685	memset(io, 0, sizeof(*io));
3686
3687	io->io_hdr.pool = pool_ref;
3688}
3689
3690/*
3691 * This routine is currently used for internal copies of ctl_ios that need
3692 * to persist for some reason after we've already returned status to the
3693 * FETD.  (Thus the flag set.)
3694 *
3695 * XXX XXX
3696 * Note that this makes a blind copy of all fields in the ctl_io, except
3697 * for the pool reference.  This includes any memory that has been
3698 * allocated!  That memory will no longer be valid after done has been
3699 * called, so this would be VERY DANGEROUS for command that actually does
3700 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3701 * start and stop commands, which don't transfer any data, so this is not a
3702 * problem.  If it is used for anything else, the caller would also need to
3703 * allocate data buffer space and this routine would need to be modified to
3704 * copy the data buffer(s) as well.
3705 */
3706void
3707ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3708{
3709	void *pool_ref;
3710
3711	if ((src == NULL)
3712	 || (dest == NULL))
3713		return;
3714
3715	/*
3716	 * May need to preserve linked list pointers at some point too.
3717	 */
3718	pool_ref = dest->io_hdr.pool;
3719
3720	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3721
3722	dest->io_hdr.pool = pool_ref;
3723	/*
3724	 * We need to know that this is an internal copy, and doesn't need
3725	 * to get passed back to the FETD that allocated it.
3726	 */
3727	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3728}
3729
3730#ifdef NEEDTOPORT
3731static void
3732ctl_update_power_subpage(struct copan_power_subpage *page)
3733{
3734	int num_luns, num_partitions, config_type;
3735	struct ctl_softc *softc;
3736	cs_BOOL_t aor_present, shelf_50pct_power;
3737	cs_raidset_personality_t rs_type;
3738	int max_active_luns;
3739
3740	softc = control_softc;
3741
3742	/* subtract out the processor LUN */
3743	num_luns = softc->num_luns - 1;
3744	/*
3745	 * Default to 7 LUNs active, which was the only number we allowed
3746	 * in the past.
3747	 */
3748	max_active_luns = 7;
3749
3750	num_partitions = config_GetRsPartitionInfo();
3751	config_type = config_GetConfigType();
3752	shelf_50pct_power = config_GetShelfPowerMode();
3753	aor_present = config_IsAorRsPresent();
3754
3755	rs_type = ddb_GetRsRaidType(1);
3756	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3757	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3758		EPRINT(0, "Unsupported RS type %d!", rs_type);
3759	}
3760
3761
3762	page->total_luns = num_luns;
3763
3764	switch (config_type) {
3765	case 40:
3766		/*
3767		 * In a 40 drive configuration, it doesn't matter what DC
3768		 * cards we have, whether we have AOR enabled or not,
3769		 * partitioning or not, or what type of RAIDset we have.
3770		 * In that scenario, we can power up every LUN we present
3771		 * to the user.
3772		 */
3773		max_active_luns = num_luns;
3774
3775		break;
3776	case 64:
3777		if (shelf_50pct_power == CS_FALSE) {
3778			/* 25% power */
3779			if (aor_present == CS_TRUE) {
3780				if (rs_type ==
3781				     CS_RAIDSET_PERSONALITY_RAID5) {
3782					max_active_luns = 7;
3783				} else if (rs_type ==
3784					 CS_RAIDSET_PERSONALITY_RAID1){
3785					max_active_luns = 14;
3786				} else {
3787					/* XXX KDM now what?? */
3788				}
3789			} else {
3790				if (rs_type ==
3791				     CS_RAIDSET_PERSONALITY_RAID5) {
3792					max_active_luns = 8;
3793				} else if (rs_type ==
3794					 CS_RAIDSET_PERSONALITY_RAID1){
3795					max_active_luns = 16;
3796				} else {
3797					/* XXX KDM now what?? */
3798				}
3799			}
3800		} else {
3801			/* 50% power */
3802			/*
3803			 * With 50% power in a 64 drive configuration, we
3804			 * can power all LUNs we present.
3805			 */
3806			max_active_luns = num_luns;
3807		}
3808		break;
3809	case 112:
3810		if (shelf_50pct_power == CS_FALSE) {
3811			/* 25% power */
3812			if (aor_present == CS_TRUE) {
3813				if (rs_type ==
3814				     CS_RAIDSET_PERSONALITY_RAID5) {
3815					max_active_luns = 7;
3816				} else if (rs_type ==
3817					 CS_RAIDSET_PERSONALITY_RAID1){
3818					max_active_luns = 14;
3819				} else {
3820					/* XXX KDM now what?? */
3821				}
3822			} else {
3823				if (rs_type ==
3824				     CS_RAIDSET_PERSONALITY_RAID5) {
3825					max_active_luns = 8;
3826				} else if (rs_type ==
3827					 CS_RAIDSET_PERSONALITY_RAID1){
3828					max_active_luns = 16;
3829				} else {
3830					/* XXX KDM now what?? */
3831				}
3832			}
3833		} else {
3834			/* 50% power */
3835			if (aor_present == CS_TRUE) {
3836				if (rs_type ==
3837				     CS_RAIDSET_PERSONALITY_RAID5) {
3838					max_active_luns = 14;
3839				} else if (rs_type ==
3840					 CS_RAIDSET_PERSONALITY_RAID1){
3841					/*
3842					 * We're assuming here that disk
3843					 * caching is enabled, and so we're
3844					 * able to power up half of each
3845					 * LUN, and cache all writes.
3846					 */
3847					max_active_luns = num_luns;
3848				} else {
3849					/* XXX KDM now what?? */
3850				}
3851			} else {
3852				if (rs_type ==
3853				     CS_RAIDSET_PERSONALITY_RAID5) {
3854					max_active_luns = 15;
3855				} else if (rs_type ==
3856					 CS_RAIDSET_PERSONALITY_RAID1){
3857					max_active_luns = 30;
3858				} else {
3859					/* XXX KDM now what?? */
3860				}
3861			}
3862		}
3863		break;
3864	default:
3865		/*
3866		 * In this case, we have an unknown configuration, so we
3867		 * just use the default from above.
3868		 */
3869		break;
3870	}
3871
3872	page->max_active_luns = max_active_luns;
3873#if 0
3874	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3875	       page->total_luns, page->max_active_luns);
3876#endif
3877}
3878#endif /* NEEDTOPORT */
3879
3880/*
3881 * This routine could be used in the future to load default and/or saved
3882 * mode page parameters for a particuar lun.
3883 */
3884static int
3885ctl_init_page_index(struct ctl_lun *lun)
3886{
3887	int i;
3888	struct ctl_page_index *page_index;
3889	struct ctl_softc *softc;
3890
3891	memcpy(&lun->mode_pages.index, page_index_template,
3892	       sizeof(page_index_template));
3893
3894	softc = lun->ctl_softc;
3895
3896	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3897
3898		page_index = &lun->mode_pages.index[i];
3899		/*
3900		 * If this is a disk-only mode page, there's no point in
3901		 * setting it up.  For some pages, we have to have some
3902		 * basic information about the disk in order to calculate the
3903		 * mode page data.
3904		 */
3905		if ((lun->be_lun->lun_type != T_DIRECT)
3906		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3907			continue;
3908
3909		switch (page_index->page_code & SMPH_PC_MASK) {
3910		case SMS_FORMAT_DEVICE_PAGE: {
3911			struct scsi_format_page *format_page;
3912
3913			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3914				panic("subpage is incorrect!");
3915
3916			/*
3917			 * Sectors per track are set above.  Bytes per
3918			 * sector need to be set here on a per-LUN basis.
3919			 */
3920			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3921			       &format_page_default,
3922			       sizeof(format_page_default));
3923			memcpy(&lun->mode_pages.format_page[
3924			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3925			       sizeof(format_page_changeable));
3926			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3927			       &format_page_default,
3928			       sizeof(format_page_default));
3929			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3930			       &format_page_default,
3931			       sizeof(format_page_default));
3932
3933			format_page = &lun->mode_pages.format_page[
3934				CTL_PAGE_CURRENT];
3935			scsi_ulto2b(lun->be_lun->blocksize,
3936				    format_page->bytes_per_sector);
3937
3938			format_page = &lun->mode_pages.format_page[
3939				CTL_PAGE_DEFAULT];
3940			scsi_ulto2b(lun->be_lun->blocksize,
3941				    format_page->bytes_per_sector);
3942
3943			format_page = &lun->mode_pages.format_page[
3944				CTL_PAGE_SAVED];
3945			scsi_ulto2b(lun->be_lun->blocksize,
3946				    format_page->bytes_per_sector);
3947
3948			page_index->page_data =
3949				(uint8_t *)lun->mode_pages.format_page;
3950			break;
3951		}
3952		case SMS_RIGID_DISK_PAGE: {
3953			struct scsi_rigid_disk_page *rigid_disk_page;
3954			uint32_t sectors_per_cylinder;
3955			uint64_t cylinders;
3956#ifndef	__XSCALE__
3957			int shift;
3958#endif /* !__XSCALE__ */
3959
3960			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3961				panic("invalid subpage value %d",
3962				      page_index->subpage);
3963
3964			/*
3965			 * Rotation rate and sectors per track are set
3966			 * above.  We calculate the cylinders here based on
3967			 * capacity.  Due to the number of heads and
3968			 * sectors per track we're using, smaller arrays
3969			 * may turn out to have 0 cylinders.  Linux and
3970			 * FreeBSD don't pay attention to these mode pages
3971			 * to figure out capacity, but Solaris does.  It
3972			 * seems to deal with 0 cylinders just fine, and
3973			 * works out a fake geometry based on the capacity.
3974			 */
3975			memcpy(&lun->mode_pages.rigid_disk_page[
3976			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3977			       sizeof(rigid_disk_page_default));
3978			memcpy(&lun->mode_pages.rigid_disk_page[
3979			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3980			       sizeof(rigid_disk_page_changeable));
3981			memcpy(&lun->mode_pages.rigid_disk_page[
3982			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3983			       sizeof(rigid_disk_page_default));
3984			memcpy(&lun->mode_pages.rigid_disk_page[
3985			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3986			       sizeof(rigid_disk_page_default));
3987
3988			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3989				CTL_DEFAULT_HEADS;
3990
3991			/*
3992			 * The divide method here will be more accurate,
3993			 * probably, but results in floating point being
3994			 * used in the kernel on i386 (__udivdi3()).  On the
3995			 * XScale, though, __udivdi3() is implemented in
3996			 * software.
3997			 *
3998			 * The shift method for cylinder calculation is
3999			 * accurate if sectors_per_cylinder is a power of
4000			 * 2.  Otherwise it might be slightly off -- you
4001			 * might have a bit of a truncation problem.
4002			 */
4003#ifdef	__XSCALE__
4004			cylinders = (lun->be_lun->maxlba + 1) /
4005				sectors_per_cylinder;
4006#else
4007			for (shift = 31; shift > 0; shift--) {
4008				if (sectors_per_cylinder & (1 << shift))
4009					break;
4010			}
4011			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4012#endif
4013
4014			/*
4015			 * We've basically got 3 bytes, or 24 bits for the
4016			 * cylinder size in the mode page.  If we're over,
4017			 * just round down to 2^24.
4018			 */
4019			if (cylinders > 0xffffff)
4020				cylinders = 0xffffff;
4021
4022			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4023				CTL_PAGE_CURRENT];
4024			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4025
4026			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4027				CTL_PAGE_DEFAULT];
4028			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4029
4030			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4031				CTL_PAGE_SAVED];
4032			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4033
4034			page_index->page_data =
4035				(uint8_t *)lun->mode_pages.rigid_disk_page;
4036			break;
4037		}
4038		case SMS_CACHING_PAGE: {
4039
4040			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4041				panic("invalid subpage value %d",
4042				      page_index->subpage);
4043			/*
4044			 * Defaults should be okay here, no calculations
4045			 * needed.
4046			 */
4047			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4048			       &caching_page_default,
4049			       sizeof(caching_page_default));
4050			memcpy(&lun->mode_pages.caching_page[
4051			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4052			       sizeof(caching_page_changeable));
4053			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4054			       &caching_page_default,
4055			       sizeof(caching_page_default));
4056			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4057			       &caching_page_default,
4058			       sizeof(caching_page_default));
4059			page_index->page_data =
4060				(uint8_t *)lun->mode_pages.caching_page;
4061			break;
4062		}
4063		case SMS_CONTROL_MODE_PAGE: {
4064
4065			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4066				panic("invalid subpage value %d",
4067				      page_index->subpage);
4068
4069			/*
4070			 * Defaults should be okay here, no calculations
4071			 * needed.
4072			 */
4073			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4074			       &control_page_default,
4075			       sizeof(control_page_default));
4076			memcpy(&lun->mode_pages.control_page[
4077			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4078			       sizeof(control_page_changeable));
4079			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4080			       &control_page_default,
4081			       sizeof(control_page_default));
4082			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4083			       &control_page_default,
4084			       sizeof(control_page_default));
4085			page_index->page_data =
4086				(uint8_t *)lun->mode_pages.control_page;
4087			break;
4088
4089		}
4090		case SMS_VENDOR_SPECIFIC_PAGE:{
4091			switch (page_index->subpage) {
4092			case PWR_SUBPAGE_CODE: {
4093				struct copan_power_subpage *current_page,
4094							   *saved_page;
4095
4096				memcpy(&lun->mode_pages.power_subpage[
4097				       CTL_PAGE_CURRENT],
4098				       &power_page_default,
4099				       sizeof(power_page_default));
4100				memcpy(&lun->mode_pages.power_subpage[
4101				       CTL_PAGE_CHANGEABLE],
4102				       &power_page_changeable,
4103				       sizeof(power_page_changeable));
4104				memcpy(&lun->mode_pages.power_subpage[
4105				       CTL_PAGE_DEFAULT],
4106				       &power_page_default,
4107				       sizeof(power_page_default));
4108				memcpy(&lun->mode_pages.power_subpage[
4109				       CTL_PAGE_SAVED],
4110				       &power_page_default,
4111				       sizeof(power_page_default));
4112				page_index->page_data =
4113				    (uint8_t *)lun->mode_pages.power_subpage;
4114
4115				current_page = (struct copan_power_subpage *)
4116					(page_index->page_data +
4117					 (page_index->page_len *
4118					  CTL_PAGE_CURRENT));
4119			        saved_page = (struct copan_power_subpage *)
4120				        (page_index->page_data +
4121					 (page_index->page_len *
4122					  CTL_PAGE_SAVED));
4123				break;
4124			}
4125			case APS_SUBPAGE_CODE: {
4126				struct copan_aps_subpage *current_page,
4127							 *saved_page;
4128
4129				// This gets set multiple times but
4130				// it should always be the same. It's
4131				// only done during init so who cares.
4132				index_to_aps_page = i;
4133
4134				memcpy(&lun->mode_pages.aps_subpage[
4135				       CTL_PAGE_CURRENT],
4136				       &aps_page_default,
4137				       sizeof(aps_page_default));
4138				memcpy(&lun->mode_pages.aps_subpage[
4139				       CTL_PAGE_CHANGEABLE],
4140				       &aps_page_changeable,
4141				       sizeof(aps_page_changeable));
4142				memcpy(&lun->mode_pages.aps_subpage[
4143				       CTL_PAGE_DEFAULT],
4144				       &aps_page_default,
4145				       sizeof(aps_page_default));
4146				memcpy(&lun->mode_pages.aps_subpage[
4147				       CTL_PAGE_SAVED],
4148				       &aps_page_default,
4149				       sizeof(aps_page_default));
4150				page_index->page_data =
4151					(uint8_t *)lun->mode_pages.aps_subpage;
4152
4153				current_page = (struct copan_aps_subpage *)
4154					(page_index->page_data +
4155					 (page_index->page_len *
4156					  CTL_PAGE_CURRENT));
4157				saved_page = (struct copan_aps_subpage *)
4158					(page_index->page_data +
4159					 (page_index->page_len *
4160					  CTL_PAGE_SAVED));
4161				break;
4162			}
4163			case DBGCNF_SUBPAGE_CODE: {
4164				struct copan_debugconf_subpage *current_page,
4165							       *saved_page;
4166
4167				memcpy(&lun->mode_pages.debugconf_subpage[
4168				       CTL_PAGE_CURRENT],
4169				       &debugconf_page_default,
4170				       sizeof(debugconf_page_default));
4171				memcpy(&lun->mode_pages.debugconf_subpage[
4172				       CTL_PAGE_CHANGEABLE],
4173				       &debugconf_page_changeable,
4174				       sizeof(debugconf_page_changeable));
4175				memcpy(&lun->mode_pages.debugconf_subpage[
4176				       CTL_PAGE_DEFAULT],
4177				       &debugconf_page_default,
4178				       sizeof(debugconf_page_default));
4179				memcpy(&lun->mode_pages.debugconf_subpage[
4180				       CTL_PAGE_SAVED],
4181				       &debugconf_page_default,
4182				       sizeof(debugconf_page_default));
4183				page_index->page_data =
4184					(uint8_t *)lun->mode_pages.debugconf_subpage;
4185
4186				current_page = (struct copan_debugconf_subpage *)
4187					(page_index->page_data +
4188					 (page_index->page_len *
4189					  CTL_PAGE_CURRENT));
4190				saved_page = (struct copan_debugconf_subpage *)
4191					(page_index->page_data +
4192					 (page_index->page_len *
4193					  CTL_PAGE_SAVED));
4194				break;
4195			}
4196			default:
4197				panic("invalid subpage value %d",
4198				      page_index->subpage);
4199				break;
4200			}
4201   			break;
4202		}
4203		default:
4204			panic("invalid page value %d",
4205			      page_index->page_code & SMPH_PC_MASK);
4206			break;
4207    	}
4208	}
4209
4210	return (CTL_RETVAL_COMPLETE);
4211}
4212
4213/*
4214 * LUN allocation.
4215 *
4216 * Requirements:
4217 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4218 *   wants us to allocate the LUN and he can block.
4219 * - ctl_softc is always set
4220 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4221 *
4222 * Returns 0 for success, non-zero (errno) for failure.
4223 */
4224static int
4225ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4226	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4227{
4228	struct ctl_lun *nlun, *lun;
4229	struct ctl_frontend *fe;
4230	int lun_number, i, lun_malloced;
4231
4232	if (be_lun == NULL)
4233		return (EINVAL);
4234
4235	/*
4236	 * We currently only support Direct Access or Processor LUN types.
4237	 */
4238	switch (be_lun->lun_type) {
4239	case T_DIRECT:
4240		break;
4241	case T_PROCESSOR:
4242		break;
4243	case T_SEQUENTIAL:
4244	case T_CHANGER:
4245	default:
4246		be_lun->lun_config_status(be_lun->be_lun,
4247					  CTL_LUN_CONFIG_FAILURE);
4248		break;
4249	}
4250	if (ctl_lun == NULL) {
4251		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4252		lun_malloced = 1;
4253	} else {
4254		lun_malloced = 0;
4255		lun = ctl_lun;
4256	}
4257
4258	memset(lun, 0, sizeof(*lun));
4259	if (lun_malloced)
4260		lun->flags = CTL_LUN_MALLOCED;
4261
4262	mtx_lock(&ctl_softc->ctl_lock);
4263	/*
4264	 * See if the caller requested a particular LUN number.  If so, see
4265	 * if it is available.  Otherwise, allocate the first available LUN.
4266	 */
4267	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4268		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4269		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4270			mtx_unlock(&ctl_softc->ctl_lock);
4271			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4272				printf("ctl: requested LUN ID %d is higher "
4273				       "than CTL_MAX_LUNS - 1 (%d)\n",
4274				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4275			} else {
4276				/*
4277				 * XXX KDM return an error, or just assign
4278				 * another LUN ID in this case??
4279				 */
4280				printf("ctl: requested LUN ID %d is already "
4281				       "in use\n", be_lun->req_lun_id);
4282			}
4283			if (lun->flags & CTL_LUN_MALLOCED)
4284				free(lun, M_CTL);
4285			be_lun->lun_config_status(be_lun->be_lun,
4286						  CTL_LUN_CONFIG_FAILURE);
4287			return (ENOSPC);
4288		}
4289		lun_number = be_lun->req_lun_id;
4290	} else {
4291		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4292		if (lun_number == -1) {
4293			mtx_unlock(&ctl_softc->ctl_lock);
4294			printf("ctl: can't allocate LUN on target %ju, out of "
4295			       "LUNs\n", (uintmax_t)target_id.id);
4296			if (lun->flags & CTL_LUN_MALLOCED)
4297				free(lun, M_CTL);
4298			be_lun->lun_config_status(be_lun->be_lun,
4299						  CTL_LUN_CONFIG_FAILURE);
4300			return (ENOSPC);
4301		}
4302	}
4303	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4304
4305	lun->target = target_id;
4306	lun->lun = lun_number;
4307	lun->be_lun = be_lun;
4308	/*
4309	 * The processor LUN is always enabled.  Disk LUNs come on line
4310	 * disabled, and must be enabled by the backend.
4311	 */
4312	lun->flags |= CTL_LUN_DISABLED;
4313	lun->backend = be_lun->be;
4314	be_lun->ctl_lun = lun;
4315	be_lun->lun_id = lun_number;
4316	atomic_add_int(&be_lun->be->num_luns, 1);
4317	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4318		lun->flags |= CTL_LUN_STOPPED;
4319
4320	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4321		lun->flags |= CTL_LUN_INOPERABLE;
4322
4323	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4324		lun->flags |= CTL_LUN_PRIMARY_SC;
4325
4326	lun->ctl_softc = ctl_softc;
4327	TAILQ_INIT(&lun->ooa_queue);
4328	TAILQ_INIT(&lun->blocked_queue);
4329	STAILQ_INIT(&lun->error_list);
4330
4331	/*
4332	 * Initialize the mode page index.
4333	 */
4334	ctl_init_page_index(lun);
4335
4336	/*
4337	 * Set the poweron UA for all initiators on this LUN only.
4338	 */
4339	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4340		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4341
4342	/*
4343	 * Now, before we insert this lun on the lun list, set the lun
4344	 * inventory changed UA for all other luns.
4345	 */
4346	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4347		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4348			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4349		}
4350	}
4351
4352	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4353
4354	ctl_softc->ctl_luns[lun_number] = lun;
4355
4356	ctl_softc->num_luns++;
4357
4358	/* Setup statistics gathering */
4359	lun->stats.device_type = be_lun->lun_type;
4360	lun->stats.lun_number = lun_number;
4361	if (lun->stats.device_type == T_DIRECT)
4362		lun->stats.blocksize = be_lun->blocksize;
4363	else
4364		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4365	for (i = 0;i < CTL_MAX_PORTS;i++)
4366		lun->stats.ports[i].targ_port = i;
4367
4368	mtx_unlock(&ctl_softc->ctl_lock);
4369
4370	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4371
4372	/*
4373	 * Run through each registered FETD and bring it online if it isn't
4374	 * already.  Enable the target ID if it hasn't been enabled, and
4375	 * enable this particular LUN.
4376	 */
4377	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4378		int retval;
4379
4380		/*
4381		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4382		 * to do things differently if we go to a multiple target
4383		 * ID scheme.
4384		 */
4385		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4386
4387			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4388			if (retval != 0) {
4389				printf("ctl_alloc_lun: FETD %s port %d "
4390				       "returned error %d for targ_enable on "
4391				       "target %ju\n", fe->port_name,
4392				       fe->targ_port, retval,
4393				       (uintmax_t)target_id.id);
4394			} else
4395				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4396		}
4397
4398		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4399		if (retval != 0) {
4400			printf("ctl_alloc_lun: FETD %s port %d returned error "
4401			       "%d for lun_enable on target %ju lun %d\n",
4402			       fe->port_name, fe->targ_port, retval,
4403			       (uintmax_t)target_id.id, lun_number);
4404		} else
4405			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4406	}
4407	return (0);
4408}
4409
4410/*
4411 * Delete a LUN.
4412 * Assumptions:
4413 * - LUN has already been marked invalid and any pending I/O has been taken
4414 *   care of.
4415 */
4416static int
4417ctl_free_lun(struct ctl_lun *lun)
4418{
4419	struct ctl_softc *softc;
4420#if 0
4421	struct ctl_frontend *fe;
4422#endif
4423	struct ctl_lun *nlun;
4424	union ctl_io *io, *next_io;
4425	int i;
4426
4427	softc = lun->ctl_softc;
4428
4429	mtx_assert(&softc->ctl_lock, MA_OWNED);
4430
4431	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4432
4433	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4434
4435	softc->ctl_luns[lun->lun] = NULL;
4436
4437	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4438		printf("ctl_free_lun: aieee!! freeing a LUN with "
4439		       "outstanding I/O!!\n");
4440	}
4441
4442	/*
4443	 * If we have anything pending on the RtR queue, remove it.
4444	 */
4445	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4446	     io = next_io) {
4447		uint32_t targ_lun;
4448
4449		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4450		targ_lun = io->io_hdr.nexus.targ_lun;
4451		if (io->io_hdr.nexus.lun_map_fn != NULL)
4452			targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
4453		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4454		 && (targ_lun == lun->lun))
4455			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4456				      ctl_io_hdr, links);
4457	}
4458
4459	/*
4460	 * Then remove everything from the blocked queue.
4461	 */
4462	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4463	     io = next_io) {
4464		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4465		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4466		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4467	}
4468
4469	/*
4470	 * Now clear out the OOA queue, and free all the I/O.
4471	 * XXX KDM should we notify the FETD here?  We probably need to
4472	 * quiesce the LUN before deleting it.
4473	 */
4474	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4475	     io = next_io) {
4476		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4477		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4478		ctl_free_io(io);
4479	}
4480
4481	softc->num_luns--;
4482
4483	/*
4484	 * XXX KDM this scheme only works for a single target/multiple LUN
4485	 * setup.  It needs to be revamped for a multiple target scheme.
4486	 *
4487	 * XXX KDM this results in fe->lun_disable() getting called twice,
4488	 * once when ctl_disable_lun() is called, and a second time here.
4489	 * We really need to re-think the LUN disable semantics.  There
4490	 * should probably be several steps/levels to LUN removal:
4491	 *  - disable
4492	 *  - invalidate
4493	 *  - free
4494 	 *
4495	 * Right now we only have a disable method when communicating to
4496	 * the front end ports, at least for individual LUNs.
4497	 */
4498#if 0
4499	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4500		int retval;
4501
4502		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4503					 lun->lun);
4504		if (retval != 0) {
4505			printf("ctl_free_lun: FETD %s port %d returned error "
4506			       "%d for lun_disable on target %ju lun %jd\n",
4507			       fe->port_name, fe->targ_port, retval,
4508			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4509		}
4510
4511		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4512			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4513
4514			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4515			if (retval != 0) {
4516				printf("ctl_free_lun: FETD %s port %d "
4517				       "returned error %d for targ_disable on "
4518				       "target %ju\n", fe->port_name,
4519				       fe->targ_port, retval,
4520				       (uintmax_t)lun->target.id);
4521			} else
4522				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4523
4524			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4525				continue;
4526
4527#if 0
4528			fe->port_offline(fe->onoff_arg);
4529			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4530#endif
4531		}
4532	}
4533#endif
4534
4535	/*
4536	 * Tell the backend to free resources, if this LUN has a backend.
4537	 */
4538	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4539	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4540
4541	if (lun->flags & CTL_LUN_MALLOCED)
4542		free(lun, M_CTL);
4543
4544	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4545		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4546			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4547		}
4548	}
4549
4550	return (0);
4551}
4552
4553static void
4554ctl_create_lun(struct ctl_be_lun *be_lun)
4555{
4556	struct ctl_softc *ctl_softc;
4557
4558	ctl_softc = control_softc;
4559
4560	/*
4561	 * ctl_alloc_lun() should handle all potential failure cases.
4562	 */
4563	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4564}
4565
4566int
4567ctl_add_lun(struct ctl_be_lun *be_lun)
4568{
4569	struct ctl_softc *ctl_softc;
4570
4571	ctl_softc = control_softc;
4572
4573	mtx_lock(&ctl_softc->ctl_lock);
4574	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4575	mtx_unlock(&ctl_softc->ctl_lock);
4576
4577	ctl_wakeup_thread();
4578
4579	return (0);
4580}
4581
4582int
4583ctl_enable_lun(struct ctl_be_lun *be_lun)
4584{
4585	struct ctl_softc *ctl_softc;
4586	struct ctl_frontend *fe, *nfe;
4587	struct ctl_lun *lun;
4588	int retval;
4589
4590	ctl_softc = control_softc;
4591
4592	lun = (struct ctl_lun *)be_lun->ctl_lun;
4593
4594	mtx_lock(&ctl_softc->ctl_lock);
4595	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4596		/*
4597		 * eh?  Why did we get called if the LUN is already
4598		 * enabled?
4599		 */
4600		mtx_unlock(&ctl_softc->ctl_lock);
4601		return (0);
4602	}
4603	lun->flags &= ~CTL_LUN_DISABLED;
4604
4605	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4606		nfe = STAILQ_NEXT(fe, links);
4607
4608		/*
4609		 * Drop the lock while we call the FETD's enable routine.
4610		 * This can lead to a callback into CTL (at least in the
4611		 * case of the internal initiator frontend.
4612		 */
4613		mtx_unlock(&ctl_softc->ctl_lock);
4614		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4615		mtx_lock(&ctl_softc->ctl_lock);
4616		if (retval != 0) {
4617			printf("%s: FETD %s port %d returned error "
4618			       "%d for lun_enable on target %ju lun %jd\n",
4619			       __func__, fe->port_name, fe->targ_port, retval,
4620			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4621		}
4622#if 0
4623		 else {
4624            /* NOTE:  TODO:  why does lun enable affect port status? */
4625			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4626		}
4627#endif
4628	}
4629
4630	mtx_unlock(&ctl_softc->ctl_lock);
4631
4632	return (0);
4633}
4634
4635int
4636ctl_disable_lun(struct ctl_be_lun *be_lun)
4637{
4638	struct ctl_softc *ctl_softc;
4639	struct ctl_frontend *fe;
4640	struct ctl_lun *lun;
4641	int retval;
4642
4643	ctl_softc = control_softc;
4644
4645	lun = (struct ctl_lun *)be_lun->ctl_lun;
4646
4647	mtx_lock(&ctl_softc->ctl_lock);
4648
4649	if (lun->flags & CTL_LUN_DISABLED) {
4650		mtx_unlock(&ctl_softc->ctl_lock);
4651		return (0);
4652	}
4653	lun->flags |= CTL_LUN_DISABLED;
4654
4655	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4656		mtx_unlock(&ctl_softc->ctl_lock);
4657		/*
4658		 * Drop the lock before we call the frontend's disable
4659		 * routine, to avoid lock order reversals.
4660		 *
4661		 * XXX KDM what happens if the frontend list changes while
4662		 * we're traversing it?  It's unlikely, but should be handled.
4663		 */
4664		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4665					 lun->lun);
4666		mtx_lock(&ctl_softc->ctl_lock);
4667		if (retval != 0) {
4668			printf("ctl_alloc_lun: FETD %s port %d returned error "
4669			       "%d for lun_disable on target %ju lun %jd\n",
4670			       fe->port_name, fe->targ_port, retval,
4671			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4672		}
4673	}
4674
4675	mtx_unlock(&ctl_softc->ctl_lock);
4676
4677	return (0);
4678}
4679
4680int
4681ctl_start_lun(struct ctl_be_lun *be_lun)
4682{
4683	struct ctl_softc *ctl_softc;
4684	struct ctl_lun *lun;
4685
4686	ctl_softc = control_softc;
4687
4688	lun = (struct ctl_lun *)be_lun->ctl_lun;
4689
4690	mtx_lock(&ctl_softc->ctl_lock);
4691	lun->flags &= ~CTL_LUN_STOPPED;
4692	mtx_unlock(&ctl_softc->ctl_lock);
4693
4694	return (0);
4695}
4696
4697int
4698ctl_stop_lun(struct ctl_be_lun *be_lun)
4699{
4700	struct ctl_softc *ctl_softc;
4701	struct ctl_lun *lun;
4702
4703	ctl_softc = control_softc;
4704
4705	lun = (struct ctl_lun *)be_lun->ctl_lun;
4706
4707	mtx_lock(&ctl_softc->ctl_lock);
4708	lun->flags |= CTL_LUN_STOPPED;
4709	mtx_unlock(&ctl_softc->ctl_lock);
4710
4711	return (0);
4712}
4713
4714int
4715ctl_lun_offline(struct ctl_be_lun *be_lun)
4716{
4717	struct ctl_softc *ctl_softc;
4718	struct ctl_lun *lun;
4719
4720	ctl_softc = control_softc;
4721
4722	lun = (struct ctl_lun *)be_lun->ctl_lun;
4723
4724	mtx_lock(&ctl_softc->ctl_lock);
4725	lun->flags |= CTL_LUN_OFFLINE;
4726	mtx_unlock(&ctl_softc->ctl_lock);
4727
4728	return (0);
4729}
4730
4731int
4732ctl_lun_online(struct ctl_be_lun *be_lun)
4733{
4734	struct ctl_softc *ctl_softc;
4735	struct ctl_lun *lun;
4736
4737	ctl_softc = control_softc;
4738
4739	lun = (struct ctl_lun *)be_lun->ctl_lun;
4740
4741	mtx_lock(&ctl_softc->ctl_lock);
4742	lun->flags &= ~CTL_LUN_OFFLINE;
4743	mtx_unlock(&ctl_softc->ctl_lock);
4744
4745	return (0);
4746}
4747
4748int
4749ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4750{
4751	struct ctl_softc *ctl_softc;
4752	struct ctl_lun *lun;
4753
4754	ctl_softc = control_softc;
4755
4756	lun = (struct ctl_lun *)be_lun->ctl_lun;
4757
4758	mtx_lock(&ctl_softc->ctl_lock);
4759
4760	/*
4761	 * The LUN needs to be disabled before it can be marked invalid.
4762	 */
4763	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4764		mtx_unlock(&ctl_softc->ctl_lock);
4765		return (-1);
4766	}
4767	/*
4768	 * Mark the LUN invalid.
4769	 */
4770	lun->flags |= CTL_LUN_INVALID;
4771
4772	/*
4773	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4774	 * If we have something in the OOA queue, we'll free it when the
4775	 * last I/O completes.
4776	 */
4777	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4778		ctl_free_lun(lun);
4779	mtx_unlock(&ctl_softc->ctl_lock);
4780
4781	return (0);
4782}
4783
4784int
4785ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4786{
4787	struct ctl_softc *ctl_softc;
4788	struct ctl_lun *lun;
4789
4790	ctl_softc = control_softc;
4791	lun = (struct ctl_lun *)be_lun->ctl_lun;
4792
4793	mtx_lock(&ctl_softc->ctl_lock);
4794	lun->flags |= CTL_LUN_INOPERABLE;
4795	mtx_unlock(&ctl_softc->ctl_lock);
4796
4797	return (0);
4798}
4799
4800int
4801ctl_lun_operable(struct ctl_be_lun *be_lun)
4802{
4803	struct ctl_softc *ctl_softc;
4804	struct ctl_lun *lun;
4805
4806	ctl_softc = control_softc;
4807	lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809	mtx_lock(&ctl_softc->ctl_lock);
4810	lun->flags &= ~CTL_LUN_INOPERABLE;
4811	mtx_unlock(&ctl_softc->ctl_lock);
4812
4813	return (0);
4814}
4815
4816int
4817ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4818		   int lock)
4819{
4820	struct ctl_softc *softc;
4821	struct ctl_lun *lun;
4822	struct copan_aps_subpage *current_sp;
4823	struct ctl_page_index *page_index;
4824	int i;
4825
4826	softc = control_softc;
4827
4828	mtx_lock(&softc->ctl_lock);
4829
4830	lun = (struct ctl_lun *)be_lun->ctl_lun;
4831
4832	page_index = NULL;
4833	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4834		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4835		     APS_PAGE_CODE)
4836			continue;
4837
4838		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4839			continue;
4840		page_index = &lun->mode_pages.index[i];
4841	}
4842
4843	if (page_index == NULL) {
4844		mtx_unlock(&softc->ctl_lock);
4845		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4846		       (uintmax_t)lun->lun);
4847		return (1);
4848	}
4849#if 0
4850	if ((softc->aps_locked_lun != 0)
4851	 && (softc->aps_locked_lun != lun->lun)) {
4852		printf("%s: attempt to lock LUN %llu when %llu is already "
4853		       "locked\n");
4854		mtx_unlock(&softc->ctl_lock);
4855		return (1);
4856	}
4857#endif
4858
4859	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4860		(page_index->page_len * CTL_PAGE_CURRENT));
4861
4862	if (lock != 0) {
4863		current_sp->lock_active = APS_LOCK_ACTIVE;
4864		softc->aps_locked_lun = lun->lun;
4865	} else {
4866		current_sp->lock_active = 0;
4867		softc->aps_locked_lun = 0;
4868	}
4869
4870
4871	/*
4872	 * If we're in HA mode, try to send the lock message to the other
4873	 * side.
4874	 */
4875	if (ctl_is_single == 0) {
4876		int isc_retval;
4877		union ctl_ha_msg lock_msg;
4878
4879		lock_msg.hdr.nexus = *nexus;
4880		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4881		if (lock != 0)
4882			lock_msg.aps.lock_flag = 1;
4883		else
4884			lock_msg.aps.lock_flag = 0;
4885		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4886					 sizeof(lock_msg), 0);
4887		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4888			printf("%s: APS (lock=%d) error returned from "
4889			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4890			mtx_unlock(&softc->ctl_lock);
4891			return (1);
4892		}
4893	}
4894
4895	mtx_unlock(&softc->ctl_lock);
4896
4897	return (0);
4898}
4899
4900void
4901ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4902{
4903	struct ctl_lun *lun;
4904	struct ctl_softc *softc;
4905	int i;
4906
4907	softc = control_softc;
4908
4909	mtx_lock(&softc->ctl_lock);
4910
4911	lun = (struct ctl_lun *)be_lun->ctl_lun;
4912
4913	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4914		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4915
4916	mtx_unlock(&softc->ctl_lock);
4917}
4918
4919/*
4920 * Backend "memory move is complete" callback for requests that never
4921 * make it down to say RAIDCore's configuration code.
4922 */
4923int
4924ctl_config_move_done(union ctl_io *io)
4925{
4926	int retval;
4927
4928	retval = CTL_RETVAL_COMPLETE;
4929
4930
4931	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4932	/*
4933	 * XXX KDM this shouldn't happen, but what if it does?
4934	 */
4935	if (io->io_hdr.io_type != CTL_IO_SCSI)
4936		panic("I/O type isn't CTL_IO_SCSI!");
4937
4938	if ((io->io_hdr.port_status == 0)
4939	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4940	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4941		io->io_hdr.status = CTL_SUCCESS;
4942	else if ((io->io_hdr.port_status != 0)
4943	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4944	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4945		/*
4946		 * For hardware error sense keys, the sense key
4947		 * specific value is defined to be a retry count,
4948		 * but we use it to pass back an internal FETD
4949		 * error code.  XXX KDM  Hopefully the FETD is only
4950		 * using 16 bits for an error code, since that's
4951		 * all the space we have in the sks field.
4952		 */
4953		ctl_set_internal_failure(&io->scsiio,
4954					 /*sks_valid*/ 1,
4955					 /*retry_count*/
4956					 io->io_hdr.port_status);
4957		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4958			free(io->scsiio.kern_data_ptr, M_CTL);
4959		ctl_done(io);
4960		goto bailout;
4961	}
4962
4963	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4964	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4965	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4966		/*
4967		 * XXX KDM just assuming a single pointer here, and not a
4968		 * S/G list.  If we start using S/G lists for config data,
4969		 * we'll need to know how to clean them up here as well.
4970		 */
4971		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4972			free(io->scsiio.kern_data_ptr, M_CTL);
4973		/* Hopefully the user has already set the status... */
4974		ctl_done(io);
4975	} else {
4976		/*
4977		 * XXX KDM now we need to continue data movement.  Some
4978		 * options:
4979		 * - call ctl_scsiio() again?  We don't do this for data
4980		 *   writes, because for those at least we know ahead of
4981		 *   time where the write will go and how long it is.  For
4982		 *   config writes, though, that information is largely
4983		 *   contained within the write itself, thus we need to
4984		 *   parse out the data again.
4985		 *
4986		 * - Call some other function once the data is in?
4987		 */
4988
4989		/*
4990		 * XXX KDM call ctl_scsiio() again for now, and check flag
4991		 * bits to see whether we're allocated or not.
4992		 */
4993		retval = ctl_scsiio(&io->scsiio);
4994	}
4995bailout:
4996	return (retval);
4997}
4998
4999/*
5000 * This gets called by a backend driver when it is done with a
5001 * data_submit method.
5002 */
5003void
5004ctl_data_submit_done(union ctl_io *io)
5005{
5006	/*
5007	 * If the IO_CONT flag is set, we need to call the supplied
5008	 * function to continue processing the I/O, instead of completing
5009	 * the I/O just yet.
5010	 *
5011	 * If there is an error, though, we don't want to keep processing.
5012	 * Instead, just send status back to the initiator.
5013	 */
5014	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5015	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5016	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5017	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5018		io->scsiio.io_cont(io);
5019		return;
5020	}
5021	ctl_done(io);
5022}
5023
5024/*
5025 * This gets called by a backend driver when it is done with a
5026 * configuration write.
5027 */
5028void
5029ctl_config_write_done(union ctl_io *io)
5030{
5031	/*
5032	 * If the IO_CONT flag is set, we need to call the supplied
5033	 * function to continue processing the I/O, instead of completing
5034	 * the I/O just yet.
5035	 *
5036	 * If there is an error, though, we don't want to keep processing.
5037	 * Instead, just send status back to the initiator.
5038	 */
5039	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5040	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5041	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5042		io->scsiio.io_cont(io);
5043		return;
5044	}
5045	/*
5046	 * Since a configuration write can be done for commands that actually
5047	 * have data allocated, like write buffer, and commands that have
5048	 * no data, like start/stop unit, we need to check here.
5049	 */
5050	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5051		free(io->scsiio.kern_data_ptr, M_CTL);
5052	ctl_done(io);
5053}
5054
5055/*
5056 * SCSI release command.
5057 */
5058int
5059ctl_scsi_release(struct ctl_scsiio *ctsio)
5060{
5061	int length, longid, thirdparty_id, resv_id;
5062	struct ctl_softc *ctl_softc;
5063	struct ctl_lun *lun;
5064
5065	length = 0;
5066	resv_id = 0;
5067
5068	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5069
5070	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5071	ctl_softc = control_softc;
5072
5073	switch (ctsio->cdb[0]) {
5074	case RELEASE: {
5075		struct scsi_release *cdb;
5076
5077		cdb = (struct scsi_release *)ctsio->cdb;
5078		if ((cdb->byte2 & 0x1f) != 0) {
5079			ctl_set_invalid_field(ctsio,
5080					      /*sks_valid*/ 1,
5081					      /*command*/ 1,
5082					      /*field*/ 1,
5083					      /*bit_valid*/ 0,
5084					      /*bit*/ 0);
5085			ctl_done((union ctl_io *)ctsio);
5086			return (CTL_RETVAL_COMPLETE);
5087		}
5088		break;
5089	}
5090	case RELEASE_10: {
5091		struct scsi_release_10 *cdb;
5092
5093		cdb = (struct scsi_release_10 *)ctsio->cdb;
5094
5095		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5096			ctl_set_invalid_field(ctsio,
5097					      /*sks_valid*/ 1,
5098					      /*command*/ 1,
5099					      /*field*/ 1,
5100					      /*bit_valid*/ 1,
5101					      /*bit*/ 0);
5102			ctl_done((union ctl_io *)ctsio);
5103			return (CTL_RETVAL_COMPLETE);
5104
5105		}
5106
5107		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5108			ctl_set_invalid_field(ctsio,
5109					      /*sks_valid*/ 1,
5110					      /*command*/ 1,
5111					      /*field*/ 1,
5112					      /*bit_valid*/ 1,
5113					      /*bit*/ 4);
5114			ctl_done((union ctl_io *)ctsio);
5115			return (CTL_RETVAL_COMPLETE);
5116		}
5117
5118		if (cdb->byte2 & SR10_LONGID)
5119			longid = 1;
5120		else
5121			thirdparty_id = cdb->thirdparty_id;
5122
5123		resv_id = cdb->resv_id;
5124		length = scsi_2btoul(cdb->length);
5125		break;
5126	}
5127	}
5128
5129
5130	/*
5131	 * XXX KDM right now, we only support LUN reservation.  We don't
5132	 * support 3rd party reservations, or extent reservations, which
5133	 * might actually need the parameter list.  If we've gotten this
5134	 * far, we've got a LUN reservation.  Anything else got kicked out
5135	 * above.  So, according to SPC, ignore the length.
5136	 */
5137	length = 0;
5138
5139	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5140	 && (length > 0)) {
5141		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5142		ctsio->kern_data_len = length;
5143		ctsio->kern_total_len = length;
5144		ctsio->kern_data_resid = 0;
5145		ctsio->kern_rel_offset = 0;
5146		ctsio->kern_sg_entries = 0;
5147		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5148		ctsio->be_move_done = ctl_config_move_done;
5149		ctl_datamove((union ctl_io *)ctsio);
5150
5151		return (CTL_RETVAL_COMPLETE);
5152	}
5153
5154	if (length > 0)
5155		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5156
5157	mtx_lock(&ctl_softc->ctl_lock);
5158
5159	/*
5160	 * According to SPC, it is not an error for an intiator to attempt
5161	 * to release a reservation on a LUN that isn't reserved, or that
5162	 * is reserved by another initiator.  The reservation can only be
5163	 * released, though, by the initiator who made it or by one of
5164	 * several reset type events.
5165	 */
5166	if (lun->flags & CTL_LUN_RESERVED) {
5167		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5168		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5169		 && (ctsio->io_hdr.nexus.targ_target.id ==
5170		     lun->rsv_nexus.targ_target.id)) {
5171			lun->flags &= ~CTL_LUN_RESERVED;
5172		}
5173	}
5174
5175	ctsio->scsi_status = SCSI_STATUS_OK;
5176	ctsio->io_hdr.status = CTL_SUCCESS;
5177
5178	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5179		free(ctsio->kern_data_ptr, M_CTL);
5180		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5181	}
5182
5183	mtx_unlock(&ctl_softc->ctl_lock);
5184
5185	ctl_done((union ctl_io *)ctsio);
5186	return (CTL_RETVAL_COMPLETE);
5187}
5188
5189int
5190ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5191{
5192	int extent, thirdparty, longid;
5193	int resv_id, length;
5194	uint64_t thirdparty_id;
5195	struct ctl_softc *ctl_softc;
5196	struct ctl_lun *lun;
5197
5198	extent = 0;
5199	thirdparty = 0;
5200	longid = 0;
5201	resv_id = 0;
5202	length = 0;
5203	thirdparty_id = 0;
5204
5205	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5206
5207	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5208	ctl_softc = control_softc;
5209
5210	switch (ctsio->cdb[0]) {
5211	case RESERVE: {
5212		struct scsi_reserve *cdb;
5213
5214		cdb = (struct scsi_reserve *)ctsio->cdb;
5215		if ((cdb->byte2 & 0x1f) != 0) {
5216			ctl_set_invalid_field(ctsio,
5217					      /*sks_valid*/ 1,
5218					      /*command*/ 1,
5219					      /*field*/ 1,
5220					      /*bit_valid*/ 0,
5221					      /*bit*/ 0);
5222			ctl_done((union ctl_io *)ctsio);
5223			return (CTL_RETVAL_COMPLETE);
5224		}
5225		resv_id = cdb->resv_id;
5226		length = scsi_2btoul(cdb->length);
5227		break;
5228	}
5229	case RESERVE_10: {
5230		struct scsi_reserve_10 *cdb;
5231
5232		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5233
5234		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5235			ctl_set_invalid_field(ctsio,
5236					      /*sks_valid*/ 1,
5237					      /*command*/ 1,
5238					      /*field*/ 1,
5239					      /*bit_valid*/ 1,
5240					      /*bit*/ 0);
5241			ctl_done((union ctl_io *)ctsio);
5242			return (CTL_RETVAL_COMPLETE);
5243		}
5244		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5245			ctl_set_invalid_field(ctsio,
5246					      /*sks_valid*/ 1,
5247					      /*command*/ 1,
5248					      /*field*/ 1,
5249					      /*bit_valid*/ 1,
5250					      /*bit*/ 4);
5251			ctl_done((union ctl_io *)ctsio);
5252			return (CTL_RETVAL_COMPLETE);
5253		}
5254		if (cdb->byte2 & SR10_LONGID)
5255			longid = 1;
5256		else
5257			thirdparty_id = cdb->thirdparty_id;
5258
5259		resv_id = cdb->resv_id;
5260		length = scsi_2btoul(cdb->length);
5261		break;
5262	}
5263	}
5264
5265	/*
5266	 * XXX KDM right now, we only support LUN reservation.  We don't
5267	 * support 3rd party reservations, or extent reservations, which
5268	 * might actually need the parameter list.  If we've gotten this
5269	 * far, we've got a LUN reservation.  Anything else got kicked out
5270	 * above.  So, according to SPC, ignore the length.
5271	 */
5272	length = 0;
5273
5274	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5275	 && (length > 0)) {
5276		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5277		ctsio->kern_data_len = length;
5278		ctsio->kern_total_len = length;
5279		ctsio->kern_data_resid = 0;
5280		ctsio->kern_rel_offset = 0;
5281		ctsio->kern_sg_entries = 0;
5282		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5283		ctsio->be_move_done = ctl_config_move_done;
5284		ctl_datamove((union ctl_io *)ctsio);
5285
5286		return (CTL_RETVAL_COMPLETE);
5287	}
5288
5289	if (length > 0)
5290		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5291
5292	mtx_lock(&ctl_softc->ctl_lock);
5293	if (lun->flags & CTL_LUN_RESERVED) {
5294		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5295		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5296		 || (ctsio->io_hdr.nexus.targ_target.id !=
5297		     lun->rsv_nexus.targ_target.id)) {
5298			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5299			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5300			goto bailout;
5301		}
5302	}
5303
5304	lun->flags |= CTL_LUN_RESERVED;
5305	lun->rsv_nexus = ctsio->io_hdr.nexus;
5306
5307	ctsio->scsi_status = SCSI_STATUS_OK;
5308	ctsio->io_hdr.status = CTL_SUCCESS;
5309
5310bailout:
5311	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5312		free(ctsio->kern_data_ptr, M_CTL);
5313		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5314	}
5315
5316	mtx_unlock(&ctl_softc->ctl_lock);
5317
5318	ctl_done((union ctl_io *)ctsio);
5319	return (CTL_RETVAL_COMPLETE);
5320}
5321
5322int
5323ctl_start_stop(struct ctl_scsiio *ctsio)
5324{
5325	struct scsi_start_stop_unit *cdb;
5326	struct ctl_lun *lun;
5327	struct ctl_softc *ctl_softc;
5328	int retval;
5329
5330	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5331
5332	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5333	ctl_softc = control_softc;
5334	retval = 0;
5335
5336	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5337
5338	/*
5339	 * XXX KDM
5340	 * We don't support the immediate bit on a stop unit.  In order to
5341	 * do that, we would need to code up a way to know that a stop is
5342	 * pending, and hold off any new commands until it completes, one
5343	 * way or another.  Then we could accept or reject those commands
5344	 * depending on its status.  We would almost need to do the reverse
5345	 * of what we do below for an immediate start -- return the copy of
5346	 * the ctl_io to the FETD with status to send to the host (and to
5347	 * free the copy!) and then free the original I/O once the stop
5348	 * actually completes.  That way, the OOA queue mechanism can work
5349	 * to block commands that shouldn't proceed.  Another alternative
5350	 * would be to put the copy in the queue in place of the original,
5351	 * and return the original back to the caller.  That could be
5352	 * slightly safer..
5353	 */
5354	if ((cdb->byte2 & SSS_IMMED)
5355	 && ((cdb->how & SSS_START) == 0)) {
5356		ctl_set_invalid_field(ctsio,
5357				      /*sks_valid*/ 1,
5358				      /*command*/ 1,
5359				      /*field*/ 1,
5360				      /*bit_valid*/ 1,
5361				      /*bit*/ 0);
5362		ctl_done((union ctl_io *)ctsio);
5363		return (CTL_RETVAL_COMPLETE);
5364	}
5365
5366	/*
5367	 * We don't support the power conditions field.  We need to check
5368	 * this prior to checking the load/eject and start/stop bits.
5369	 */
5370	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5371		ctl_set_invalid_field(ctsio,
5372				      /*sks_valid*/ 1,
5373				      /*command*/ 1,
5374				      /*field*/ 4,
5375				      /*bit_valid*/ 1,
5376				      /*bit*/ 4);
5377		ctl_done((union ctl_io *)ctsio);
5378		return (CTL_RETVAL_COMPLETE);
5379	}
5380
5381	/*
5382	 * Media isn't removable, so we can't load or eject it.
5383	 */
5384	if ((cdb->how & SSS_LOEJ) != 0) {
5385		ctl_set_invalid_field(ctsio,
5386				      /*sks_valid*/ 1,
5387				      /*command*/ 1,
5388				      /*field*/ 4,
5389				      /*bit_valid*/ 1,
5390				      /*bit*/ 1);
5391		ctl_done((union ctl_io *)ctsio);
5392		return (CTL_RETVAL_COMPLETE);
5393	}
5394
5395	if ((lun->flags & CTL_LUN_PR_RESERVED)
5396	 && ((cdb->how & SSS_START)==0)) {
5397		uint32_t residx;
5398
5399		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5400		if (!lun->per_res[residx].registered
5401		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5402
5403			ctl_set_reservation_conflict(ctsio);
5404			ctl_done((union ctl_io *)ctsio);
5405			return (CTL_RETVAL_COMPLETE);
5406		}
5407	}
5408
5409	/*
5410	 * If there is no backend on this device, we can't start or stop
5411	 * it.  In theory we shouldn't get any start/stop commands in the
5412	 * first place at this level if the LUN doesn't have a backend.
5413	 * That should get stopped by the command decode code.
5414	 */
5415	if (lun->backend == NULL) {
5416		ctl_set_invalid_opcode(ctsio);
5417		ctl_done((union ctl_io *)ctsio);
5418		return (CTL_RETVAL_COMPLETE);
5419	}
5420
5421	/*
5422	 * XXX KDM Copan-specific offline behavior.
5423	 * Figure out a reasonable way to port this?
5424	 */
5425#ifdef NEEDTOPORT
5426	mtx_lock(&ctl_softc->ctl_lock);
5427
5428	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5429	 && (lun->flags & CTL_LUN_OFFLINE)) {
5430		/*
5431		 * If the LUN is offline, and the on/offline bit isn't set,
5432		 * reject the start or stop.  Otherwise, let it through.
5433		 */
5434		mtx_unlock(&ctl_softc->ctl_lock);
5435		ctl_set_lun_not_ready(ctsio);
5436		ctl_done((union ctl_io *)ctsio);
5437	} else {
5438		mtx_unlock(&ctl_softc->ctl_lock);
5439#endif /* NEEDTOPORT */
5440		/*
5441		 * This could be a start or a stop when we're online,
5442		 * or a stop/offline or start/online.  A start or stop when
5443		 * we're offline is covered in the case above.
5444		 */
5445		/*
5446		 * In the non-immediate case, we send the request to
5447		 * the backend and return status to the user when
5448		 * it is done.
5449		 *
5450		 * In the immediate case, we allocate a new ctl_io
5451		 * to hold a copy of the request, and send that to
5452		 * the backend.  We then set good status on the
5453		 * user's request and return it immediately.
5454		 */
5455		if (cdb->byte2 & SSS_IMMED) {
5456			union ctl_io *new_io;
5457
5458			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5459			if (new_io == NULL) {
5460				ctl_set_busy(ctsio);
5461				ctl_done((union ctl_io *)ctsio);
5462			} else {
5463				ctl_copy_io((union ctl_io *)ctsio,
5464					    new_io);
5465				retval = lun->backend->config_write(new_io);
5466				ctl_set_success(ctsio);
5467				ctl_done((union ctl_io *)ctsio);
5468			}
5469		} else {
5470			retval = lun->backend->config_write(
5471				(union ctl_io *)ctsio);
5472		}
5473#ifdef NEEDTOPORT
5474	}
5475#endif
5476	return (retval);
5477}
5478
5479/*
5480 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5481 * we don't really do anything with the LBA and length fields if the user
5482 * passes them in.  Instead we'll just flush out the cache for the entire
5483 * LUN.
5484 */
5485int
5486ctl_sync_cache(struct ctl_scsiio *ctsio)
5487{
5488	struct ctl_lun *lun;
5489	struct ctl_softc *ctl_softc;
5490	uint64_t starting_lba;
5491	uint32_t block_count;
5492	int reladr, immed;
5493	int retval;
5494
5495	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5496
5497	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5498	ctl_softc = control_softc;
5499	retval = 0;
5500	reladr = 0;
5501	immed = 0;
5502
5503	switch (ctsio->cdb[0]) {
5504	case SYNCHRONIZE_CACHE: {
5505		struct scsi_sync_cache *cdb;
5506		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5507
5508		if (cdb->byte2 & SSC_RELADR)
5509			reladr = 1;
5510
5511		if (cdb->byte2 & SSC_IMMED)
5512			immed = 1;
5513
5514		starting_lba = scsi_4btoul(cdb->begin_lba);
5515		block_count = scsi_2btoul(cdb->lb_count);
5516		break;
5517	}
5518	case SYNCHRONIZE_CACHE_16: {
5519		struct scsi_sync_cache_16 *cdb;
5520		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5521
5522		if (cdb->byte2 & SSC_RELADR)
5523			reladr = 1;
5524
5525		if (cdb->byte2 & SSC_IMMED)
5526			immed = 1;
5527
5528		starting_lba = scsi_8btou64(cdb->begin_lba);
5529		block_count = scsi_4btoul(cdb->lb_count);
5530		break;
5531	}
5532	default:
5533		ctl_set_invalid_opcode(ctsio);
5534		ctl_done((union ctl_io *)ctsio);
5535		goto bailout;
5536		break; /* NOTREACHED */
5537	}
5538
5539	if (immed) {
5540		/*
5541		 * We don't support the immediate bit.  Since it's in the
5542		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5543		 * commands, we can just return the same error in either
5544		 * case.
5545		 */
5546		ctl_set_invalid_field(ctsio,
5547				      /*sks_valid*/ 1,
5548				      /*command*/ 1,
5549				      /*field*/ 1,
5550				      /*bit_valid*/ 1,
5551				      /*bit*/ 1);
5552		ctl_done((union ctl_io *)ctsio);
5553		goto bailout;
5554	}
5555
5556	if (reladr) {
5557		/*
5558		 * We don't support the reladr bit either.  It can only be
5559		 * used with linked commands, and we don't support linked
5560		 * commands.  Since the bit is in the same place for the
5561		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5562		 * just return the same error in either case.
5563		 */
5564		ctl_set_invalid_field(ctsio,
5565				      /*sks_valid*/ 1,
5566				      /*command*/ 1,
5567				      /*field*/ 1,
5568				      /*bit_valid*/ 1,
5569				      /*bit*/ 0);
5570		ctl_done((union ctl_io *)ctsio);
5571		goto bailout;
5572	}
5573
5574	/*
5575	 * We check the LBA and length, but don't do anything with them.
5576	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5577	 * get flushed.  This check will just help satisfy anyone who wants
5578	 * to see an error for an out of range LBA.
5579	 */
5580	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5581		ctl_set_lba_out_of_range(ctsio);
5582		ctl_done((union ctl_io *)ctsio);
5583		goto bailout;
5584	}
5585
5586	/*
5587	 * If this LUN has no backend, we can't flush the cache anyway.
5588	 */
5589	if (lun->backend == NULL) {
5590		ctl_set_invalid_opcode(ctsio);
5591		ctl_done((union ctl_io *)ctsio);
5592		goto bailout;
5593	}
5594
5595	/*
5596	 * Check to see whether we're configured to send the SYNCHRONIZE
5597	 * CACHE command directly to the back end.
5598	 */
5599	mtx_lock(&ctl_softc->ctl_lock);
5600	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5601	 && (++(lun->sync_count) >= lun->sync_interval)) {
5602		lun->sync_count = 0;
5603		mtx_unlock(&ctl_softc->ctl_lock);
5604		retval = lun->backend->config_write((union ctl_io *)ctsio);
5605	} else {
5606		mtx_unlock(&ctl_softc->ctl_lock);
5607		ctl_set_success(ctsio);
5608		ctl_done((union ctl_io *)ctsio);
5609	}
5610
5611bailout:
5612
5613	return (retval);
5614}
5615
5616int
5617ctl_format(struct ctl_scsiio *ctsio)
5618{
5619	struct scsi_format *cdb;
5620	struct ctl_lun *lun;
5621	struct ctl_softc *ctl_softc;
5622	int length, defect_list_len;
5623
5624	CTL_DEBUG_PRINT(("ctl_format\n"));
5625
5626	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5627	ctl_softc = control_softc;
5628
5629	cdb = (struct scsi_format *)ctsio->cdb;
5630
5631	length = 0;
5632	if (cdb->byte2 & SF_FMTDATA) {
5633		if (cdb->byte2 & SF_LONGLIST)
5634			length = sizeof(struct scsi_format_header_long);
5635		else
5636			length = sizeof(struct scsi_format_header_short);
5637	}
5638
5639	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5640	 && (length > 0)) {
5641		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5642		ctsio->kern_data_len = length;
5643		ctsio->kern_total_len = length;
5644		ctsio->kern_data_resid = 0;
5645		ctsio->kern_rel_offset = 0;
5646		ctsio->kern_sg_entries = 0;
5647		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5648		ctsio->be_move_done = ctl_config_move_done;
5649		ctl_datamove((union ctl_io *)ctsio);
5650
5651		return (CTL_RETVAL_COMPLETE);
5652	}
5653
5654	defect_list_len = 0;
5655
5656	if (cdb->byte2 & SF_FMTDATA) {
5657		if (cdb->byte2 & SF_LONGLIST) {
5658			struct scsi_format_header_long *header;
5659
5660			header = (struct scsi_format_header_long *)
5661				ctsio->kern_data_ptr;
5662
5663			defect_list_len = scsi_4btoul(header->defect_list_len);
5664			if (defect_list_len != 0) {
5665				ctl_set_invalid_field(ctsio,
5666						      /*sks_valid*/ 1,
5667						      /*command*/ 0,
5668						      /*field*/ 2,
5669						      /*bit_valid*/ 0,
5670						      /*bit*/ 0);
5671				goto bailout;
5672			}
5673		} else {
5674			struct scsi_format_header_short *header;
5675
5676			header = (struct scsi_format_header_short *)
5677				ctsio->kern_data_ptr;
5678
5679			defect_list_len = scsi_2btoul(header->defect_list_len);
5680			if (defect_list_len != 0) {
5681				ctl_set_invalid_field(ctsio,
5682						      /*sks_valid*/ 1,
5683						      /*command*/ 0,
5684						      /*field*/ 2,
5685						      /*bit_valid*/ 0,
5686						      /*bit*/ 0);
5687				goto bailout;
5688			}
5689		}
5690	}
5691
5692	/*
5693	 * The format command will clear out the "Medium format corrupted"
5694	 * status if set by the configuration code.  That status is really
5695	 * just a way to notify the host that we have lost the media, and
5696	 * get them to issue a command that will basically make them think
5697	 * they're blowing away the media.
5698	 */
5699	mtx_lock(&ctl_softc->ctl_lock);
5700	lun->flags &= ~CTL_LUN_INOPERABLE;
5701	mtx_unlock(&ctl_softc->ctl_lock);
5702
5703	ctsio->scsi_status = SCSI_STATUS_OK;
5704	ctsio->io_hdr.status = CTL_SUCCESS;
5705bailout:
5706
5707	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5708		free(ctsio->kern_data_ptr, M_CTL);
5709		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5710	}
5711
5712	ctl_done((union ctl_io *)ctsio);
5713	return (CTL_RETVAL_COMPLETE);
5714}
5715
5716int
5717ctl_read_buffer(struct ctl_scsiio *ctsio)
5718{
5719	struct scsi_read_buffer *cdb;
5720	struct ctl_lun *lun;
5721	int buffer_offset, len;
5722	static uint8_t descr[4];
5723	static uint8_t echo_descr[4] = { 0 };
5724
5725	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5726
5727	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5728	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5729
5730	if (lun->flags & CTL_LUN_PR_RESERVED) {
5731		uint32_t residx;
5732
5733		/*
5734		 * XXX KDM need a lock here.
5735		 */
5736		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5737		if ((lun->res_type == SPR_TYPE_EX_AC
5738		  && residx != lun->pr_res_idx)
5739		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
5740		   || lun->res_type == SPR_TYPE_EX_AC_AR)
5741		  && !lun->per_res[residx].registered)) {
5742			ctl_set_reservation_conflict(ctsio);
5743			ctl_done((union ctl_io *)ctsio);
5744			return (CTL_RETVAL_COMPLETE);
5745	        }
5746	}
5747
5748	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5749	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5750	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5751		ctl_set_invalid_field(ctsio,
5752				      /*sks_valid*/ 1,
5753				      /*command*/ 1,
5754				      /*field*/ 1,
5755				      /*bit_valid*/ 1,
5756				      /*bit*/ 4);
5757		ctl_done((union ctl_io *)ctsio);
5758		return (CTL_RETVAL_COMPLETE);
5759	}
5760	if (cdb->buffer_id != 0) {
5761		ctl_set_invalid_field(ctsio,
5762				      /*sks_valid*/ 1,
5763				      /*command*/ 1,
5764				      /*field*/ 2,
5765				      /*bit_valid*/ 0,
5766				      /*bit*/ 0);
5767		ctl_done((union ctl_io *)ctsio);
5768		return (CTL_RETVAL_COMPLETE);
5769	}
5770
5771	len = scsi_3btoul(cdb->length);
5772	buffer_offset = scsi_3btoul(cdb->offset);
5773
5774	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5775		ctl_set_invalid_field(ctsio,
5776				      /*sks_valid*/ 1,
5777				      /*command*/ 1,
5778				      /*field*/ 6,
5779				      /*bit_valid*/ 0,
5780				      /*bit*/ 0);
5781		ctl_done((union ctl_io *)ctsio);
5782		return (CTL_RETVAL_COMPLETE);
5783	}
5784
5785	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5786		descr[0] = 0;
5787		scsi_ulto3b(sizeof(lun->write_buffer), &descr[1]);
5788		ctsio->kern_data_ptr = descr;
5789		len = min(len, sizeof(descr));
5790	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5791		ctsio->kern_data_ptr = echo_descr;
5792		len = min(len, sizeof(echo_descr));
5793	} else
5794		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5795	ctsio->kern_data_len = len;
5796	ctsio->kern_total_len = len;
5797	ctsio->kern_data_resid = 0;
5798	ctsio->kern_rel_offset = 0;
5799	ctsio->kern_sg_entries = 0;
5800	ctsio->be_move_done = ctl_config_move_done;
5801	ctl_datamove((union ctl_io *)ctsio);
5802
5803	return (CTL_RETVAL_COMPLETE);
5804}
5805
5806int
5807ctl_write_buffer(struct ctl_scsiio *ctsio)
5808{
5809	struct scsi_write_buffer *cdb;
5810	struct ctl_lun *lun;
5811	int buffer_offset, len;
5812
5813	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5814
5815	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5816	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5817
5818	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5819		ctl_set_invalid_field(ctsio,
5820				      /*sks_valid*/ 1,
5821				      /*command*/ 1,
5822				      /*field*/ 1,
5823				      /*bit_valid*/ 1,
5824				      /*bit*/ 4);
5825		ctl_done((union ctl_io *)ctsio);
5826		return (CTL_RETVAL_COMPLETE);
5827	}
5828	if (cdb->buffer_id != 0) {
5829		ctl_set_invalid_field(ctsio,
5830				      /*sks_valid*/ 1,
5831				      /*command*/ 1,
5832				      /*field*/ 2,
5833				      /*bit_valid*/ 0,
5834				      /*bit*/ 0);
5835		ctl_done((union ctl_io *)ctsio);
5836		return (CTL_RETVAL_COMPLETE);
5837	}
5838
5839	len = scsi_3btoul(cdb->length);
5840	buffer_offset = scsi_3btoul(cdb->offset);
5841
5842	if (buffer_offset + len > sizeof(lun->write_buffer)) {
5843		ctl_set_invalid_field(ctsio,
5844				      /*sks_valid*/ 1,
5845				      /*command*/ 1,
5846				      /*field*/ 6,
5847				      /*bit_valid*/ 0,
5848				      /*bit*/ 0);
5849		ctl_done((union ctl_io *)ctsio);
5850		return (CTL_RETVAL_COMPLETE);
5851	}
5852
5853	/*
5854	 * If we've got a kernel request that hasn't been malloced yet,
5855	 * malloc it and tell the caller the data buffer is here.
5856	 */
5857	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5858		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5859		ctsio->kern_data_len = len;
5860		ctsio->kern_total_len = len;
5861		ctsio->kern_data_resid = 0;
5862		ctsio->kern_rel_offset = 0;
5863		ctsio->kern_sg_entries = 0;
5864		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5865		ctsio->be_move_done = ctl_config_move_done;
5866		ctl_datamove((union ctl_io *)ctsio);
5867
5868		return (CTL_RETVAL_COMPLETE);
5869	}
5870
5871	ctl_done((union ctl_io *)ctsio);
5872
5873	return (CTL_RETVAL_COMPLETE);
5874}
5875
5876int
5877ctl_write_same(struct ctl_scsiio *ctsio)
5878{
5879	struct ctl_lun *lun;
5880	struct ctl_lba_len_flags *lbalen;
5881	uint64_t lba;
5882	uint32_t num_blocks;
5883	int len, retval;
5884	uint8_t byte2;
5885
5886	retval = CTL_RETVAL_COMPLETE;
5887
5888	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5889
5890	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5891
5892	switch (ctsio->cdb[0]) {
5893	case WRITE_SAME_10: {
5894		struct scsi_write_same_10 *cdb;
5895
5896		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5897
5898		lba = scsi_4btoul(cdb->addr);
5899		num_blocks = scsi_2btoul(cdb->length);
5900		byte2 = cdb->byte2;
5901		break;
5902	}
5903	case WRITE_SAME_16: {
5904		struct scsi_write_same_16 *cdb;
5905
5906		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5907
5908		lba = scsi_8btou64(cdb->addr);
5909		num_blocks = scsi_4btoul(cdb->length);
5910		byte2 = cdb->byte2;
5911		break;
5912	}
5913	default:
5914		/*
5915		 * We got a command we don't support.  This shouldn't
5916		 * happen, commands should be filtered out above us.
5917		 */
5918		ctl_set_invalid_opcode(ctsio);
5919		ctl_done((union ctl_io *)ctsio);
5920
5921		return (CTL_RETVAL_COMPLETE);
5922		break; /* NOTREACHED */
5923	}
5924
5925	/*
5926	 * The first check is to make sure we're in bounds, the second
5927	 * check is to catch wrap-around problems.  If the lba + num blocks
5928	 * is less than the lba, then we've wrapped around and the block
5929	 * range is invalid anyway.
5930	 */
5931	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5932	 || ((lba + num_blocks) < lba)) {
5933		ctl_set_lba_out_of_range(ctsio);
5934		ctl_done((union ctl_io *)ctsio);
5935		return (CTL_RETVAL_COMPLETE);
5936	}
5937
5938	/* Zero number of blocks means "to the last logical block" */
5939	if (num_blocks == 0) {
5940		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5941			ctl_set_invalid_field(ctsio,
5942					      /*sks_valid*/ 0,
5943					      /*command*/ 1,
5944					      /*field*/ 0,
5945					      /*bit_valid*/ 0,
5946					      /*bit*/ 0);
5947			ctl_done((union ctl_io *)ctsio);
5948			return (CTL_RETVAL_COMPLETE);
5949		}
5950		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5951	}
5952
5953	len = lun->be_lun->blocksize;
5954
5955	/*
5956	 * If we've got a kernel request that hasn't been malloced yet,
5957	 * malloc it and tell the caller the data buffer is here.
5958	 */
5959	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5960		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5961		ctsio->kern_data_len = len;
5962		ctsio->kern_total_len = len;
5963		ctsio->kern_data_resid = 0;
5964		ctsio->kern_rel_offset = 0;
5965		ctsio->kern_sg_entries = 0;
5966		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5967		ctsio->be_move_done = ctl_config_move_done;
5968		ctl_datamove((union ctl_io *)ctsio);
5969
5970		return (CTL_RETVAL_COMPLETE);
5971	}
5972
5973	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5974	lbalen->lba = lba;
5975	lbalen->len = num_blocks;
5976	lbalen->flags = byte2;
5977	retval = lun->backend->config_write((union ctl_io *)ctsio);
5978
5979	return (retval);
5980}
5981
5982int
5983ctl_unmap(struct ctl_scsiio *ctsio)
5984{
5985	struct ctl_lun *lun;
5986	struct scsi_unmap *cdb;
5987	struct ctl_ptr_len_flags *ptrlen;
5988	struct scsi_unmap_header *hdr;
5989	struct scsi_unmap_desc *buf, *end;
5990	uint64_t lba;
5991	uint32_t num_blocks;
5992	int len, retval;
5993	uint8_t byte2;
5994
5995	retval = CTL_RETVAL_COMPLETE;
5996
5997	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5998
5999	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6000	cdb = (struct scsi_unmap *)ctsio->cdb;
6001
6002	len = scsi_2btoul(cdb->length);
6003	byte2 = cdb->byte2;
6004
6005	/*
6006	 * If we've got a kernel request that hasn't been malloced yet,
6007	 * malloc it and tell the caller the data buffer is here.
6008	 */
6009	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6010		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
6011		ctsio->kern_data_len = len;
6012		ctsio->kern_total_len = len;
6013		ctsio->kern_data_resid = 0;
6014		ctsio->kern_rel_offset = 0;
6015		ctsio->kern_sg_entries = 0;
6016		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6017		ctsio->be_move_done = ctl_config_move_done;
6018		ctl_datamove((union ctl_io *)ctsio);
6019
6020		return (CTL_RETVAL_COMPLETE);
6021	}
6022
6023	len = ctsio->kern_total_len - ctsio->kern_data_resid;
6024	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
6025	if (len < sizeof (*hdr) ||
6026	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6027	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6028	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6029		ctl_set_invalid_field(ctsio,
6030				      /*sks_valid*/ 0,
6031				      /*command*/ 0,
6032				      /*field*/ 0,
6033				      /*bit_valid*/ 0,
6034				      /*bit*/ 0);
6035		ctl_done((union ctl_io *)ctsio);
6036		return (CTL_RETVAL_COMPLETE);
6037	}
6038	len = scsi_2btoul(hdr->desc_length);
6039	buf = (struct scsi_unmap_desc *)(hdr + 1);
6040	end = buf + len / sizeof(*buf);
6041
6042	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6043	ptrlen->ptr = (void *)buf;
6044	ptrlen->len = len;
6045	ptrlen->flags = byte2;
6046
6047	for (; buf < end; buf++) {
6048		lba = scsi_8btou64(buf->lba);
6049		num_blocks = scsi_4btoul(buf->length);
6050		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6051		 || ((lba + num_blocks) < lba)) {
6052			ctl_set_lba_out_of_range(ctsio);
6053			ctl_done((union ctl_io *)ctsio);
6054			return (CTL_RETVAL_COMPLETE);
6055		}
6056	}
6057
6058	retval = lun->backend->config_write((union ctl_io *)ctsio);
6059
6060	return (retval);
6061}
6062
6063/*
6064 * Note that this function currently doesn't actually do anything inside
6065 * CTL to enforce things if the DQue bit is turned on.
6066 *
6067 * Also note that this function can't be used in the default case, because
6068 * the DQue bit isn't set in the changeable mask for the control mode page
6069 * anyway.  This is just here as an example for how to implement a page
6070 * handler, and a placeholder in case we want to allow the user to turn
6071 * tagged queueing on and off.
6072 *
6073 * The D_SENSE bit handling is functional, however, and will turn
6074 * descriptor sense on and off for a given LUN.
6075 */
6076int
6077ctl_control_page_handler(struct ctl_scsiio *ctsio,
6078			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6079{
6080	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6081	struct ctl_lun *lun;
6082	struct ctl_softc *softc;
6083	int set_ua;
6084	uint32_t initidx;
6085
6086	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6087	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6088	set_ua = 0;
6089
6090	user_cp = (struct scsi_control_page *)page_ptr;
6091	current_cp = (struct scsi_control_page *)
6092		(page_index->page_data + (page_index->page_len *
6093		CTL_PAGE_CURRENT));
6094	saved_cp = (struct scsi_control_page *)
6095		(page_index->page_data + (page_index->page_len *
6096		CTL_PAGE_SAVED));
6097
6098	softc = control_softc;
6099
6100	mtx_lock(&softc->ctl_lock);
6101	if (((current_cp->rlec & SCP_DSENSE) == 0)
6102	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6103		/*
6104		 * Descriptor sense is currently turned off and the user
6105		 * wants to turn it on.
6106		 */
6107		current_cp->rlec |= SCP_DSENSE;
6108		saved_cp->rlec |= SCP_DSENSE;
6109		lun->flags |= CTL_LUN_SENSE_DESC;
6110		set_ua = 1;
6111	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6112		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6113		/*
6114		 * Descriptor sense is currently turned on, and the user
6115		 * wants to turn it off.
6116		 */
6117		current_cp->rlec &= ~SCP_DSENSE;
6118		saved_cp->rlec &= ~SCP_DSENSE;
6119		lun->flags &= ~CTL_LUN_SENSE_DESC;
6120		set_ua = 1;
6121	}
6122	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6123		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6124#ifdef NEEDTOPORT
6125			csevent_log(CSC_CTL | CSC_SHELF_SW |
6126				    CTL_UNTAG_TO_UNTAG,
6127				    csevent_LogType_Trace,
6128				    csevent_Severity_Information,
6129				    csevent_AlertLevel_Green,
6130				    csevent_FRU_Firmware,
6131				    csevent_FRU_Unknown,
6132				    "Received untagged to untagged transition");
6133#endif /* NEEDTOPORT */
6134		} else {
6135#ifdef NEEDTOPORT
6136			csevent_log(CSC_CTL | CSC_SHELF_SW |
6137				    CTL_UNTAG_TO_TAG,
6138				    csevent_LogType_ConfigChange,
6139				    csevent_Severity_Information,
6140				    csevent_AlertLevel_Green,
6141				    csevent_FRU_Firmware,
6142				    csevent_FRU_Unknown,
6143				    "Received untagged to tagged "
6144				    "queueing transition");
6145#endif /* NEEDTOPORT */
6146
6147			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6148			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6149			set_ua = 1;
6150		}
6151	} else {
6152		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6153#ifdef NEEDTOPORT
6154			csevent_log(CSC_CTL | CSC_SHELF_SW |
6155				    CTL_TAG_TO_UNTAG,
6156				    csevent_LogType_ConfigChange,
6157				    csevent_Severity_Warning,
6158				    csevent_AlertLevel_Yellow,
6159				    csevent_FRU_Firmware,
6160				    csevent_FRU_Unknown,
6161				    "Received tagged queueing to untagged "
6162				    "transition");
6163#endif /* NEEDTOPORT */
6164
6165			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6166			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6167			set_ua = 1;
6168		} else {
6169#ifdef NEEDTOPORT
6170			csevent_log(CSC_CTL | CSC_SHELF_SW |
6171				    CTL_TAG_TO_TAG,
6172				    csevent_LogType_Trace,
6173				    csevent_Severity_Information,
6174				    csevent_AlertLevel_Green,
6175				    csevent_FRU_Firmware,
6176				    csevent_FRU_Unknown,
6177				    "Received tagged queueing to tagged "
6178				    "queueing transition");
6179#endif /* NEEDTOPORT */
6180		}
6181	}
6182	if (set_ua != 0) {
6183		int i;
6184		/*
6185		 * Let other initiators know that the mode
6186		 * parameters for this LUN have changed.
6187		 */
6188		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6189			if (i == initidx)
6190				continue;
6191
6192			lun->pending_sense[i].ua_pending |=
6193				CTL_UA_MODE_CHANGE;
6194		}
6195	}
6196	mtx_unlock(&softc->ctl_lock);
6197
6198	return (0);
6199}
6200
6201int
6202ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6203		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6204{
6205	return (0);
6206}
6207
6208int
6209ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6210			   struct ctl_page_index *page_index, int pc)
6211{
6212	struct copan_power_subpage *page;
6213
6214	page = (struct copan_power_subpage *)page_index->page_data +
6215		(page_index->page_len * pc);
6216
6217	switch (pc) {
6218	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6219		/*
6220		 * We don't update the changable bits for this page.
6221		 */
6222		break;
6223	case SMS_PAGE_CTRL_CURRENT >> 6:
6224	case SMS_PAGE_CTRL_DEFAULT >> 6:
6225	case SMS_PAGE_CTRL_SAVED >> 6:
6226#ifdef NEEDTOPORT
6227		ctl_update_power_subpage(page);
6228#endif
6229		break;
6230	default:
6231#ifdef NEEDTOPORT
6232		EPRINT(0, "Invalid PC %d!!", pc);
6233#endif
6234		break;
6235	}
6236	return (0);
6237}
6238
6239
6240int
6241ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6242		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6243{
6244	struct copan_aps_subpage *user_sp;
6245	struct copan_aps_subpage *current_sp;
6246	union ctl_modepage_info *modepage_info;
6247	struct ctl_softc *softc;
6248	struct ctl_lun *lun;
6249	int retval;
6250
6251	retval = CTL_RETVAL_COMPLETE;
6252	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6253		     (page_index->page_len * CTL_PAGE_CURRENT));
6254	softc = control_softc;
6255	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6256
6257	user_sp = (struct copan_aps_subpage *)page_ptr;
6258
6259	modepage_info = (union ctl_modepage_info *)
6260		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6261
6262	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6263	modepage_info->header.subpage = page_index->subpage;
6264	modepage_info->aps.lock_active = user_sp->lock_active;
6265
6266	mtx_lock(&softc->ctl_lock);
6267
6268	/*
6269	 * If there is a request to lock the LUN and another LUN is locked
6270	 * this is an error. If the requested LUN is already locked ignore
6271	 * the request. If no LUN is locked attempt to lock it.
6272	 * if there is a request to unlock the LUN and the LUN is currently
6273	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6274	 * if another LUN is locked or no LUN is locked.
6275	 */
6276	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6277		if (softc->aps_locked_lun == lun->lun) {
6278			/*
6279			 * This LUN is already locked, so we're done.
6280			 */
6281			retval = CTL_RETVAL_COMPLETE;
6282		} else if (softc->aps_locked_lun == 0) {
6283			/*
6284			 * No one has the lock, pass the request to the
6285			 * backend.
6286			 */
6287			retval = lun->backend->config_write(
6288				(union ctl_io *)ctsio);
6289		} else {
6290			/*
6291			 * Someone else has the lock, throw out the request.
6292			 */
6293			ctl_set_already_locked(ctsio);
6294			free(ctsio->kern_data_ptr, M_CTL);
6295			ctl_done((union ctl_io *)ctsio);
6296
6297			/*
6298			 * Set the return value so that ctl_do_mode_select()
6299			 * won't try to complete the command.  We already
6300			 * completed it here.
6301			 */
6302			retval = CTL_RETVAL_ERROR;
6303		}
6304	} else if (softc->aps_locked_lun == lun->lun) {
6305		/*
6306		 * This LUN is locked, so pass the unlock request to the
6307		 * backend.
6308		 */
6309		retval = lun->backend->config_write((union ctl_io *)ctsio);
6310	}
6311	mtx_unlock(&softc->ctl_lock);
6312
6313	return (retval);
6314}
6315
6316int
6317ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6318				struct ctl_page_index *page_index,
6319				uint8_t *page_ptr)
6320{
6321	uint8_t *c;
6322	int i;
6323
6324	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6325	ctl_time_io_secs =
6326		(c[0] << 8) |
6327		(c[1] << 0) |
6328		0;
6329	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6330	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6331	printf("page data:");
6332	for (i=0; i<8; i++)
6333		printf(" %.2x",page_ptr[i]);
6334	printf("\n");
6335	return (0);
6336}
6337
6338int
6339ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6340			       struct ctl_page_index *page_index,
6341			       int pc)
6342{
6343	struct copan_debugconf_subpage *page;
6344
6345	page = (struct copan_debugconf_subpage *)page_index->page_data +
6346		(page_index->page_len * pc);
6347
6348	switch (pc) {
6349	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6350	case SMS_PAGE_CTRL_DEFAULT >> 6:
6351	case SMS_PAGE_CTRL_SAVED >> 6:
6352		/*
6353		 * We don't update the changable or default bits for this page.
6354		 */
6355		break;
6356	case SMS_PAGE_CTRL_CURRENT >> 6:
6357		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6358		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6359		break;
6360	default:
6361#ifdef NEEDTOPORT
6362		EPRINT(0, "Invalid PC %d!!", pc);
6363#endif /* NEEDTOPORT */
6364		break;
6365	}
6366	return (0);
6367}
6368
6369
6370static int
6371ctl_do_mode_select(union ctl_io *io)
6372{
6373	struct scsi_mode_page_header *page_header;
6374	struct ctl_page_index *page_index;
6375	struct ctl_scsiio *ctsio;
6376	int control_dev, page_len;
6377	int page_len_offset, page_len_size;
6378	union ctl_modepage_info *modepage_info;
6379	struct ctl_lun *lun;
6380	int *len_left, *len_used;
6381	int retval, i;
6382
6383	ctsio = &io->scsiio;
6384	page_index = NULL;
6385	page_len = 0;
6386	retval = CTL_RETVAL_COMPLETE;
6387
6388	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6389
6390	if (lun->be_lun->lun_type != T_DIRECT)
6391		control_dev = 1;
6392	else
6393		control_dev = 0;
6394
6395	modepage_info = (union ctl_modepage_info *)
6396		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6397	len_left = &modepage_info->header.len_left;
6398	len_used = &modepage_info->header.len_used;
6399
6400do_next_page:
6401
6402	page_header = (struct scsi_mode_page_header *)
6403		(ctsio->kern_data_ptr + *len_used);
6404
6405	if (*len_left == 0) {
6406		free(ctsio->kern_data_ptr, M_CTL);
6407		ctl_set_success(ctsio);
6408		ctl_done((union ctl_io *)ctsio);
6409		return (CTL_RETVAL_COMPLETE);
6410	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6411
6412		free(ctsio->kern_data_ptr, M_CTL);
6413		ctl_set_param_len_error(ctsio);
6414		ctl_done((union ctl_io *)ctsio);
6415		return (CTL_RETVAL_COMPLETE);
6416
6417	} else if ((page_header->page_code & SMPH_SPF)
6418		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6419
6420		free(ctsio->kern_data_ptr, M_CTL);
6421		ctl_set_param_len_error(ctsio);
6422		ctl_done((union ctl_io *)ctsio);
6423		return (CTL_RETVAL_COMPLETE);
6424	}
6425
6426
6427	/*
6428	 * XXX KDM should we do something with the block descriptor?
6429	 */
6430	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6431
6432		if ((control_dev != 0)
6433		 && (lun->mode_pages.index[i].page_flags &
6434		     CTL_PAGE_FLAG_DISK_ONLY))
6435			continue;
6436
6437		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6438		    (page_header->page_code & SMPH_PC_MASK))
6439			continue;
6440
6441		/*
6442		 * If neither page has a subpage code, then we've got a
6443		 * match.
6444		 */
6445		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6446		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6447			page_index = &lun->mode_pages.index[i];
6448			page_len = page_header->page_length;
6449			break;
6450		}
6451
6452		/*
6453		 * If both pages have subpages, then the subpage numbers
6454		 * have to match.
6455		 */
6456		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6457		  && (page_header->page_code & SMPH_SPF)) {
6458			struct scsi_mode_page_header_sp *sph;
6459
6460			sph = (struct scsi_mode_page_header_sp *)page_header;
6461
6462			if (lun->mode_pages.index[i].subpage ==
6463			    sph->subpage) {
6464				page_index = &lun->mode_pages.index[i];
6465				page_len = scsi_2btoul(sph->page_length);
6466				break;
6467			}
6468		}
6469	}
6470
6471	/*
6472	 * If we couldn't find the page, or if we don't have a mode select
6473	 * handler for it, send back an error to the user.
6474	 */
6475	if ((page_index == NULL)
6476	 || (page_index->select_handler == NULL)) {
6477		ctl_set_invalid_field(ctsio,
6478				      /*sks_valid*/ 1,
6479				      /*command*/ 0,
6480				      /*field*/ *len_used,
6481				      /*bit_valid*/ 0,
6482				      /*bit*/ 0);
6483		free(ctsio->kern_data_ptr, M_CTL);
6484		ctl_done((union ctl_io *)ctsio);
6485		return (CTL_RETVAL_COMPLETE);
6486	}
6487
6488	if (page_index->page_code & SMPH_SPF) {
6489		page_len_offset = 2;
6490		page_len_size = 2;
6491	} else {
6492		page_len_size = 1;
6493		page_len_offset = 1;
6494	}
6495
6496	/*
6497	 * If the length the initiator gives us isn't the one we specify in
6498	 * the mode page header, or if they didn't specify enough data in
6499	 * the CDB to avoid truncating this page, kick out the request.
6500	 */
6501	if ((page_len != (page_index->page_len - page_len_offset -
6502			  page_len_size))
6503	 || (*len_left < page_index->page_len)) {
6504
6505
6506		ctl_set_invalid_field(ctsio,
6507				      /*sks_valid*/ 1,
6508				      /*command*/ 0,
6509				      /*field*/ *len_used + page_len_offset,
6510				      /*bit_valid*/ 0,
6511				      /*bit*/ 0);
6512		free(ctsio->kern_data_ptr, M_CTL);
6513		ctl_done((union ctl_io *)ctsio);
6514		return (CTL_RETVAL_COMPLETE);
6515	}
6516
6517	/*
6518	 * Run through the mode page, checking to make sure that the bits
6519	 * the user changed are actually legal for him to change.
6520	 */
6521	for (i = 0; i < page_index->page_len; i++) {
6522		uint8_t *user_byte, *change_mask, *current_byte;
6523		int bad_bit;
6524		int j;
6525
6526		user_byte = (uint8_t *)page_header + i;
6527		change_mask = page_index->page_data +
6528			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6529		current_byte = page_index->page_data +
6530			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6531
6532		/*
6533		 * Check to see whether the user set any bits in this byte
6534		 * that he is not allowed to set.
6535		 */
6536		if ((*user_byte & ~(*change_mask)) ==
6537		    (*current_byte & ~(*change_mask)))
6538			continue;
6539
6540		/*
6541		 * Go through bit by bit to determine which one is illegal.
6542		 */
6543		bad_bit = 0;
6544		for (j = 7; j >= 0; j--) {
6545			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6546			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6547				bad_bit = i;
6548				break;
6549			}
6550		}
6551		ctl_set_invalid_field(ctsio,
6552				      /*sks_valid*/ 1,
6553				      /*command*/ 0,
6554				      /*field*/ *len_used + i,
6555				      /*bit_valid*/ 1,
6556				      /*bit*/ bad_bit);
6557		free(ctsio->kern_data_ptr, M_CTL);
6558		ctl_done((union ctl_io *)ctsio);
6559		return (CTL_RETVAL_COMPLETE);
6560	}
6561
6562	/*
6563	 * Decrement these before we call the page handler, since we may
6564	 * end up getting called back one way or another before the handler
6565	 * returns to this context.
6566	 */
6567	*len_left -= page_index->page_len;
6568	*len_used += page_index->page_len;
6569
6570	retval = page_index->select_handler(ctsio, page_index,
6571					    (uint8_t *)page_header);
6572
6573	/*
6574	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6575	 * wait until this queued command completes to finish processing
6576	 * the mode page.  If it returns anything other than
6577	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6578	 * already set the sense information, freed the data pointer, and
6579	 * completed the io for us.
6580	 */
6581	if (retval != CTL_RETVAL_COMPLETE)
6582		goto bailout_no_done;
6583
6584	/*
6585	 * If the initiator sent us more than one page, parse the next one.
6586	 */
6587	if (*len_left > 0)
6588		goto do_next_page;
6589
6590	ctl_set_success(ctsio);
6591	free(ctsio->kern_data_ptr, M_CTL);
6592	ctl_done((union ctl_io *)ctsio);
6593
6594bailout_no_done:
6595
6596	return (CTL_RETVAL_COMPLETE);
6597
6598}
6599
6600int
6601ctl_mode_select(struct ctl_scsiio *ctsio)
6602{
6603	int param_len, pf, sp;
6604	int header_size, bd_len;
6605	int len_left, len_used;
6606	struct ctl_page_index *page_index;
6607	struct ctl_lun *lun;
6608	int control_dev, page_len;
6609	union ctl_modepage_info *modepage_info;
6610	int retval;
6611
6612	pf = 0;
6613	sp = 0;
6614	page_len = 0;
6615	len_used = 0;
6616	len_left = 0;
6617	retval = 0;
6618	bd_len = 0;
6619	page_index = NULL;
6620
6621	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6622
6623	if (lun->be_lun->lun_type != T_DIRECT)
6624		control_dev = 1;
6625	else
6626		control_dev = 0;
6627
6628	switch (ctsio->cdb[0]) {
6629	case MODE_SELECT_6: {
6630		struct scsi_mode_select_6 *cdb;
6631
6632		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6633
6634		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6635		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6636
6637		param_len = cdb->length;
6638		header_size = sizeof(struct scsi_mode_header_6);
6639		break;
6640	}
6641	case MODE_SELECT_10: {
6642		struct scsi_mode_select_10 *cdb;
6643
6644		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6645
6646		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6647		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6648
6649		param_len = scsi_2btoul(cdb->length);
6650		header_size = sizeof(struct scsi_mode_header_10);
6651		break;
6652	}
6653	default:
6654		ctl_set_invalid_opcode(ctsio);
6655		ctl_done((union ctl_io *)ctsio);
6656		return (CTL_RETVAL_COMPLETE);
6657		break; /* NOTREACHED */
6658	}
6659
6660	/*
6661	 * From SPC-3:
6662	 * "A parameter list length of zero indicates that the Data-Out Buffer
6663	 * shall be empty. This condition shall not be considered as an error."
6664	 */
6665	if (param_len == 0) {
6666		ctl_set_success(ctsio);
6667		ctl_done((union ctl_io *)ctsio);
6668		return (CTL_RETVAL_COMPLETE);
6669	}
6670
6671	/*
6672	 * Since we'll hit this the first time through, prior to
6673	 * allocation, we don't need to free a data buffer here.
6674	 */
6675	if (param_len < header_size) {
6676		ctl_set_param_len_error(ctsio);
6677		ctl_done((union ctl_io *)ctsio);
6678		return (CTL_RETVAL_COMPLETE);
6679	}
6680
6681	/*
6682	 * Allocate the data buffer and grab the user's data.  In theory,
6683	 * we shouldn't have to sanity check the parameter list length here
6684	 * because the maximum size is 64K.  We should be able to malloc
6685	 * that much without too many problems.
6686	 */
6687	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6688		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6689		ctsio->kern_data_len = param_len;
6690		ctsio->kern_total_len = param_len;
6691		ctsio->kern_data_resid = 0;
6692		ctsio->kern_rel_offset = 0;
6693		ctsio->kern_sg_entries = 0;
6694		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6695		ctsio->be_move_done = ctl_config_move_done;
6696		ctl_datamove((union ctl_io *)ctsio);
6697
6698		return (CTL_RETVAL_COMPLETE);
6699	}
6700
6701	switch (ctsio->cdb[0]) {
6702	case MODE_SELECT_6: {
6703		struct scsi_mode_header_6 *mh6;
6704
6705		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6706		bd_len = mh6->blk_desc_len;
6707		break;
6708	}
6709	case MODE_SELECT_10: {
6710		struct scsi_mode_header_10 *mh10;
6711
6712		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6713		bd_len = scsi_2btoul(mh10->blk_desc_len);
6714		break;
6715	}
6716	default:
6717		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6718		break;
6719	}
6720
6721	if (param_len < (header_size + bd_len)) {
6722		free(ctsio->kern_data_ptr, M_CTL);
6723		ctl_set_param_len_error(ctsio);
6724		ctl_done((union ctl_io *)ctsio);
6725		return (CTL_RETVAL_COMPLETE);
6726	}
6727
6728	/*
6729	 * Set the IO_CONT flag, so that if this I/O gets passed to
6730	 * ctl_config_write_done(), it'll get passed back to
6731	 * ctl_do_mode_select() for further processing, or completion if
6732	 * we're all done.
6733	 */
6734	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6735	ctsio->io_cont = ctl_do_mode_select;
6736
6737	modepage_info = (union ctl_modepage_info *)
6738		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6739
6740	memset(modepage_info, 0, sizeof(*modepage_info));
6741
6742	len_left = param_len - header_size - bd_len;
6743	len_used = header_size + bd_len;
6744
6745	modepage_info->header.len_left = len_left;
6746	modepage_info->header.len_used = len_used;
6747
6748	return (ctl_do_mode_select((union ctl_io *)ctsio));
6749}
6750
6751int
6752ctl_mode_sense(struct ctl_scsiio *ctsio)
6753{
6754	struct ctl_lun *lun;
6755	int pc, page_code, dbd, llba, subpage;
6756	int alloc_len, page_len, header_len, total_len;
6757	struct scsi_mode_block_descr *block_desc;
6758	struct ctl_page_index *page_index;
6759	int control_dev;
6760
6761	dbd = 0;
6762	llba = 0;
6763	block_desc = NULL;
6764	page_index = NULL;
6765
6766	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6767
6768	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6769
6770	if (lun->be_lun->lun_type != T_DIRECT)
6771		control_dev = 1;
6772	else
6773		control_dev = 0;
6774
6775	switch (ctsio->cdb[0]) {
6776	case MODE_SENSE_6: {
6777		struct scsi_mode_sense_6 *cdb;
6778
6779		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6780
6781		header_len = sizeof(struct scsi_mode_hdr_6);
6782		if (cdb->byte2 & SMS_DBD)
6783			dbd = 1;
6784		else
6785			header_len += sizeof(struct scsi_mode_block_descr);
6786
6787		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6788		page_code = cdb->page & SMS_PAGE_CODE;
6789		subpage = cdb->subpage;
6790		alloc_len = cdb->length;
6791		break;
6792	}
6793	case MODE_SENSE_10: {
6794		struct scsi_mode_sense_10 *cdb;
6795
6796		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6797
6798		header_len = sizeof(struct scsi_mode_hdr_10);
6799
6800		if (cdb->byte2 & SMS_DBD)
6801			dbd = 1;
6802		else
6803			header_len += sizeof(struct scsi_mode_block_descr);
6804		if (cdb->byte2 & SMS10_LLBAA)
6805			llba = 1;
6806		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6807		page_code = cdb->page & SMS_PAGE_CODE;
6808		subpage = cdb->subpage;
6809		alloc_len = scsi_2btoul(cdb->length);
6810		break;
6811	}
6812	default:
6813		ctl_set_invalid_opcode(ctsio);
6814		ctl_done((union ctl_io *)ctsio);
6815		return (CTL_RETVAL_COMPLETE);
6816		break; /* NOTREACHED */
6817	}
6818
6819	/*
6820	 * We have to make a first pass through to calculate the size of
6821	 * the pages that match the user's query.  Then we allocate enough
6822	 * memory to hold it, and actually copy the data into the buffer.
6823	 */
6824	switch (page_code) {
6825	case SMS_ALL_PAGES_PAGE: {
6826		int i;
6827
6828		page_len = 0;
6829
6830		/*
6831		 * At the moment, values other than 0 and 0xff here are
6832		 * reserved according to SPC-3.
6833		 */
6834		if ((subpage != SMS_SUBPAGE_PAGE_0)
6835		 && (subpage != SMS_SUBPAGE_ALL)) {
6836			ctl_set_invalid_field(ctsio,
6837					      /*sks_valid*/ 1,
6838					      /*command*/ 1,
6839					      /*field*/ 3,
6840					      /*bit_valid*/ 0,
6841					      /*bit*/ 0);
6842			ctl_done((union ctl_io *)ctsio);
6843			return (CTL_RETVAL_COMPLETE);
6844		}
6845
6846		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6847			if ((control_dev != 0)
6848			 && (lun->mode_pages.index[i].page_flags &
6849			     CTL_PAGE_FLAG_DISK_ONLY))
6850				continue;
6851
6852			/*
6853			 * We don't use this subpage if the user didn't
6854			 * request all subpages.
6855			 */
6856			if ((lun->mode_pages.index[i].subpage != 0)
6857			 && (subpage == SMS_SUBPAGE_PAGE_0))
6858				continue;
6859
6860#if 0
6861			printf("found page %#x len %d\n",
6862			       lun->mode_pages.index[i].page_code &
6863			       SMPH_PC_MASK,
6864			       lun->mode_pages.index[i].page_len);
6865#endif
6866			page_len += lun->mode_pages.index[i].page_len;
6867		}
6868		break;
6869	}
6870	default: {
6871		int i;
6872
6873		page_len = 0;
6874
6875		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6876			/* Look for the right page code */
6877			if ((lun->mode_pages.index[i].page_code &
6878			     SMPH_PC_MASK) != page_code)
6879				continue;
6880
6881			/* Look for the right subpage or the subpage wildcard*/
6882			if ((lun->mode_pages.index[i].subpage != subpage)
6883			 && (subpage != SMS_SUBPAGE_ALL))
6884				continue;
6885
6886			/* Make sure the page is supported for this dev type */
6887			if ((control_dev != 0)
6888			 && (lun->mode_pages.index[i].page_flags &
6889			     CTL_PAGE_FLAG_DISK_ONLY))
6890				continue;
6891
6892#if 0
6893			printf("found page %#x len %d\n",
6894			       lun->mode_pages.index[i].page_code &
6895			       SMPH_PC_MASK,
6896			       lun->mode_pages.index[i].page_len);
6897#endif
6898
6899			page_len += lun->mode_pages.index[i].page_len;
6900		}
6901
6902		if (page_len == 0) {
6903			ctl_set_invalid_field(ctsio,
6904					      /*sks_valid*/ 1,
6905					      /*command*/ 1,
6906					      /*field*/ 2,
6907					      /*bit_valid*/ 1,
6908					      /*bit*/ 5);
6909			ctl_done((union ctl_io *)ctsio);
6910			return (CTL_RETVAL_COMPLETE);
6911		}
6912		break;
6913	}
6914	}
6915
6916	total_len = header_len + page_len;
6917#if 0
6918	printf("header_len = %d, page_len = %d, total_len = %d\n",
6919	       header_len, page_len, total_len);
6920#endif
6921
6922	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6923	ctsio->kern_sg_entries = 0;
6924	ctsio->kern_data_resid = 0;
6925	ctsio->kern_rel_offset = 0;
6926	if (total_len < alloc_len) {
6927		ctsio->residual = alloc_len - total_len;
6928		ctsio->kern_data_len = total_len;
6929		ctsio->kern_total_len = total_len;
6930	} else {
6931		ctsio->residual = 0;
6932		ctsio->kern_data_len = alloc_len;
6933		ctsio->kern_total_len = alloc_len;
6934	}
6935
6936	switch (ctsio->cdb[0]) {
6937	case MODE_SENSE_6: {
6938		struct scsi_mode_hdr_6 *header;
6939
6940		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6941
6942		header->datalen = ctl_min(total_len - 1, 254);
6943
6944		if (dbd)
6945			header->block_descr_len = 0;
6946		else
6947			header->block_descr_len =
6948				sizeof(struct scsi_mode_block_descr);
6949		block_desc = (struct scsi_mode_block_descr *)&header[1];
6950		break;
6951	}
6952	case MODE_SENSE_10: {
6953		struct scsi_mode_hdr_10 *header;
6954		int datalen;
6955
6956		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6957
6958		datalen = ctl_min(total_len - 2, 65533);
6959		scsi_ulto2b(datalen, header->datalen);
6960		if (dbd)
6961			scsi_ulto2b(0, header->block_descr_len);
6962		else
6963			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6964				    header->block_descr_len);
6965		block_desc = (struct scsi_mode_block_descr *)&header[1];
6966		break;
6967	}
6968	default:
6969		panic("invalid CDB type %#x", ctsio->cdb[0]);
6970		break; /* NOTREACHED */
6971	}
6972
6973	/*
6974	 * If we've got a disk, use its blocksize in the block
6975	 * descriptor.  Otherwise, just set it to 0.
6976	 */
6977	if (dbd == 0) {
6978		if (control_dev != 0)
6979			scsi_ulto3b(lun->be_lun->blocksize,
6980				    block_desc->block_len);
6981		else
6982			scsi_ulto3b(0, block_desc->block_len);
6983	}
6984
6985	switch (page_code) {
6986	case SMS_ALL_PAGES_PAGE: {
6987		int i, data_used;
6988
6989		data_used = header_len;
6990		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6991			struct ctl_page_index *page_index;
6992
6993			page_index = &lun->mode_pages.index[i];
6994
6995			if ((control_dev != 0)
6996			 && (page_index->page_flags &
6997			    CTL_PAGE_FLAG_DISK_ONLY))
6998				continue;
6999
7000			/*
7001			 * We don't use this subpage if the user didn't
7002			 * request all subpages.  We already checked (above)
7003			 * to make sure the user only specified a subpage
7004			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
7005			 */
7006			if ((page_index->subpage != 0)
7007			 && (subpage == SMS_SUBPAGE_PAGE_0))
7008				continue;
7009
7010			/*
7011			 * Call the handler, if it exists, to update the
7012			 * page to the latest values.
7013			 */
7014			if (page_index->sense_handler != NULL)
7015				page_index->sense_handler(ctsio, page_index,pc);
7016
7017			memcpy(ctsio->kern_data_ptr + data_used,
7018			       page_index->page_data +
7019			       (page_index->page_len * pc),
7020			       page_index->page_len);
7021			data_used += page_index->page_len;
7022		}
7023		break;
7024	}
7025	default: {
7026		int i, data_used;
7027
7028		data_used = header_len;
7029
7030		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
7031			struct ctl_page_index *page_index;
7032
7033			page_index = &lun->mode_pages.index[i];
7034
7035			/* Look for the right page code */
7036			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
7037				continue;
7038
7039			/* Look for the right subpage or the subpage wildcard*/
7040			if ((page_index->subpage != subpage)
7041			 && (subpage != SMS_SUBPAGE_ALL))
7042				continue;
7043
7044			/* Make sure the page is supported for this dev type */
7045			if ((control_dev != 0)
7046			 && (page_index->page_flags &
7047			     CTL_PAGE_FLAG_DISK_ONLY))
7048				continue;
7049
7050			/*
7051			 * Call the handler, if it exists, to update the
7052			 * page to the latest values.
7053			 */
7054			if (page_index->sense_handler != NULL)
7055				page_index->sense_handler(ctsio, page_index,pc);
7056
7057			memcpy(ctsio->kern_data_ptr + data_used,
7058			       page_index->page_data +
7059			       (page_index->page_len * pc),
7060			       page_index->page_len);
7061			data_used += page_index->page_len;
7062		}
7063		break;
7064	}
7065	}
7066
7067	ctsio->scsi_status = SCSI_STATUS_OK;
7068
7069	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7070	ctsio->be_move_done = ctl_config_move_done;
7071	ctl_datamove((union ctl_io *)ctsio);
7072
7073	return (CTL_RETVAL_COMPLETE);
7074}
7075
7076int
7077ctl_read_capacity(struct ctl_scsiio *ctsio)
7078{
7079	struct scsi_read_capacity *cdb;
7080	struct scsi_read_capacity_data *data;
7081	struct ctl_lun *lun;
7082	uint32_t lba;
7083
7084	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7085
7086	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7087
7088	lba = scsi_4btoul(cdb->addr);
7089	if (((cdb->pmi & SRC_PMI) == 0)
7090	 && (lba != 0)) {
7091		ctl_set_invalid_field(/*ctsio*/ ctsio,
7092				      /*sks_valid*/ 1,
7093				      /*command*/ 1,
7094				      /*field*/ 2,
7095				      /*bit_valid*/ 0,
7096				      /*bit*/ 0);
7097		ctl_done((union ctl_io *)ctsio);
7098		return (CTL_RETVAL_COMPLETE);
7099	}
7100
7101	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7102
7103	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7104	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7105	ctsio->residual = 0;
7106	ctsio->kern_data_len = sizeof(*data);
7107	ctsio->kern_total_len = sizeof(*data);
7108	ctsio->kern_data_resid = 0;
7109	ctsio->kern_rel_offset = 0;
7110	ctsio->kern_sg_entries = 0;
7111
7112	/*
7113	 * If the maximum LBA is greater than 0xfffffffe, the user must
7114	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7115	 * serivce action set.
7116	 */
7117	if (lun->be_lun->maxlba > 0xfffffffe)
7118		scsi_ulto4b(0xffffffff, data->addr);
7119	else
7120		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7121
7122	/*
7123	 * XXX KDM this may not be 512 bytes...
7124	 */
7125	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7126
7127	ctsio->scsi_status = SCSI_STATUS_OK;
7128
7129	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7130	ctsio->be_move_done = ctl_config_move_done;
7131	ctl_datamove((union ctl_io *)ctsio);
7132
7133	return (CTL_RETVAL_COMPLETE);
7134}
7135
7136static int
7137ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7138{
7139	struct scsi_read_capacity_16 *cdb;
7140	struct scsi_read_capacity_data_long *data;
7141	struct ctl_lun *lun;
7142	uint64_t lba;
7143	uint32_t alloc_len;
7144
7145	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7146
7147	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7148
7149	alloc_len = scsi_4btoul(cdb->alloc_len);
7150	lba = scsi_8btou64(cdb->addr);
7151
7152	if ((cdb->reladr & SRC16_PMI)
7153	 && (lba != 0)) {
7154		ctl_set_invalid_field(/*ctsio*/ ctsio,
7155				      /*sks_valid*/ 1,
7156				      /*command*/ 1,
7157				      /*field*/ 2,
7158				      /*bit_valid*/ 0,
7159				      /*bit*/ 0);
7160		ctl_done((union ctl_io *)ctsio);
7161		return (CTL_RETVAL_COMPLETE);
7162	}
7163
7164	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7165
7166	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7167	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7168
7169	if (sizeof(*data) < alloc_len) {
7170		ctsio->residual = alloc_len - sizeof(*data);
7171		ctsio->kern_data_len = sizeof(*data);
7172		ctsio->kern_total_len = sizeof(*data);
7173	} else {
7174		ctsio->residual = 0;
7175		ctsio->kern_data_len = alloc_len;
7176		ctsio->kern_total_len = alloc_len;
7177	}
7178	ctsio->kern_data_resid = 0;
7179	ctsio->kern_rel_offset = 0;
7180	ctsio->kern_sg_entries = 0;
7181
7182	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7183	/* XXX KDM this may not be 512 bytes... */
7184	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7185	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7186	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7187	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7188		data->lalba_lbp[0] |= SRC16_LBPME;
7189
7190	ctsio->scsi_status = SCSI_STATUS_OK;
7191
7192	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7193	ctsio->be_move_done = ctl_config_move_done;
7194	ctl_datamove((union ctl_io *)ctsio);
7195
7196	return (CTL_RETVAL_COMPLETE);
7197}
7198
7199int
7200ctl_service_action_in(struct ctl_scsiio *ctsio)
7201{
7202	struct scsi_service_action_in *cdb;
7203	int retval;
7204
7205	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
7206
7207	cdb = (struct scsi_service_action_in *)ctsio->cdb;
7208
7209	retval = CTL_RETVAL_COMPLETE;
7210
7211	switch (cdb->service_action) {
7212	case SRC16_SERVICE_ACTION:
7213		retval = ctl_read_capacity_16(ctsio);
7214		break;
7215	default:
7216		ctl_set_invalid_field(/*ctsio*/ ctsio,
7217				      /*sks_valid*/ 1,
7218				      /*command*/ 1,
7219				      /*field*/ 1,
7220				      /*bit_valid*/ 1,
7221				      /*bit*/ 4);
7222		ctl_done((union ctl_io *)ctsio);
7223		break;
7224	}
7225
7226	return (retval);
7227}
7228
7229int
7230ctl_maintenance_in(struct ctl_scsiio *ctsio)
7231{
7232	struct scsi_maintenance_in *cdb;
7233	int retval;
7234	int alloc_len, total_len = 0;
7235	int num_target_port_groups, single;
7236	struct ctl_lun *lun;
7237	struct ctl_softc *softc;
7238	struct scsi_target_group_data *rtg_ptr;
7239	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
7240	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
7241	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
7242
7243	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
7244
7245	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7246	softc = control_softc;
7247	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7248
7249	retval = CTL_RETVAL_COMPLETE;
7250
7251	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
7252		ctl_set_invalid_field(/*ctsio*/ ctsio,
7253				      /*sks_valid*/ 1,
7254				      /*command*/ 1,
7255				      /*field*/ 1,
7256				      /*bit_valid*/ 1,
7257				      /*bit*/ 4);
7258		ctl_done((union ctl_io *)ctsio);
7259		return(retval);
7260	}
7261
7262	mtx_lock(&softc->ctl_lock);
7263	single = ctl_is_single;
7264	mtx_unlock(&softc->ctl_lock);
7265
7266	if (single)
7267        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
7268	else
7269        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7270
7271	total_len = sizeof(struct scsi_target_group_data) +
7272		sizeof(struct scsi_target_port_group_descriptor) *
7273		num_target_port_groups +
7274		sizeof(struct scsi_target_port_descriptor) *
7275		NUM_PORTS_PER_GRP * num_target_port_groups;
7276
7277	alloc_len = scsi_4btoul(cdb->length);
7278
7279	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7280
7281	ctsio->kern_sg_entries = 0;
7282
7283	if (total_len < alloc_len) {
7284		ctsio->residual = alloc_len - total_len;
7285		ctsio->kern_data_len = total_len;
7286		ctsio->kern_total_len = total_len;
7287	} else {
7288		ctsio->residual = 0;
7289		ctsio->kern_data_len = alloc_len;
7290		ctsio->kern_total_len = alloc_len;
7291	}
7292	ctsio->kern_data_resid = 0;
7293	ctsio->kern_rel_offset = 0;
7294
7295	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7296
7297	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7298	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7299	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7300	        &tp_desc_ptr1_1->desc_list[0];
7301
7302	if (single == 0) {
7303		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7304	                &tp_desc_ptr1_2->desc_list[0];
7305		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7306		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7307	        	&tp_desc_ptr2_1->desc_list[0];
7308        } else {
7309		tpg_desc_ptr2 = NULL;
7310		tp_desc_ptr2_1 = NULL;
7311		tp_desc_ptr2_2 = NULL;
7312	}
7313
7314	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7315	if (single == 0) {
7316        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7317			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7318				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7319				tpg_desc_ptr2->pref_state =
7320					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7321			} else {
7322				tpg_desc_ptr1->pref_state =
7323					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7324				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7325			}
7326		} else {
7327			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7328				tpg_desc_ptr1->pref_state =
7329					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7330				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7331			} else {
7332				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7333				tpg_desc_ptr2->pref_state =
7334					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7335			}
7336		}
7337	} else {
7338		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7339	}
7340	tpg_desc_ptr1->support = 0;
7341	tpg_desc_ptr1->target_port_group[1] = 1;
7342	tpg_desc_ptr1->status = TPG_IMPLICIT;
7343	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7344
7345	if (single == 0) {
7346		tpg_desc_ptr2->support = 0;
7347		tpg_desc_ptr2->target_port_group[1] = 2;
7348		tpg_desc_ptr2->status = TPG_IMPLICIT;
7349		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7350
7351		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7352		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7353
7354		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7355		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7356	} else {
7357        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7358			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7359			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7360		} else {
7361			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7362			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7363		}
7364	}
7365
7366	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7367	ctsio->be_move_done = ctl_config_move_done;
7368
7369	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7370			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7371			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7372			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7373			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7374
7375	ctl_datamove((union ctl_io *)ctsio);
7376	return(retval);
7377}
7378
7379int
7380ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7381{
7382	struct scsi_per_res_in *cdb;
7383	int alloc_len, total_len = 0;
7384	/* struct scsi_per_res_in_rsrv in_data; */
7385	struct ctl_lun *lun;
7386	struct ctl_softc *softc;
7387
7388	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7389
7390	softc = control_softc;
7391
7392	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7393
7394	alloc_len = scsi_2btoul(cdb->length);
7395
7396	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7397
7398retry:
7399	mtx_lock(&softc->ctl_lock);
7400	switch (cdb->action) {
7401	case SPRI_RK: /* read keys */
7402		total_len = sizeof(struct scsi_per_res_in_keys) +
7403			lun->pr_key_count *
7404			sizeof(struct scsi_per_res_key);
7405		break;
7406	case SPRI_RR: /* read reservation */
7407		if (lun->flags & CTL_LUN_PR_RESERVED)
7408			total_len = sizeof(struct scsi_per_res_in_rsrv);
7409		else
7410			total_len = sizeof(struct scsi_per_res_in_header);
7411		break;
7412	case SPRI_RC: /* report capabilities */
7413		total_len = sizeof(struct scsi_per_res_cap);
7414		break;
7415	case SPRI_RS: /* read full status */
7416	default:
7417		mtx_unlock(&softc->ctl_lock);
7418		ctl_set_invalid_field(ctsio,
7419				      /*sks_valid*/ 1,
7420				      /*command*/ 1,
7421				      /*field*/ 1,
7422				      /*bit_valid*/ 1,
7423				      /*bit*/ 0);
7424		ctl_done((union ctl_io *)ctsio);
7425		return (CTL_RETVAL_COMPLETE);
7426		break; /* NOTREACHED */
7427	}
7428	mtx_unlock(&softc->ctl_lock);
7429
7430	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7431
7432	if (total_len < alloc_len) {
7433		ctsio->residual = alloc_len - total_len;
7434		ctsio->kern_data_len = total_len;
7435		ctsio->kern_total_len = total_len;
7436	} else {
7437		ctsio->residual = 0;
7438		ctsio->kern_data_len = alloc_len;
7439		ctsio->kern_total_len = alloc_len;
7440	}
7441
7442	ctsio->kern_data_resid = 0;
7443	ctsio->kern_rel_offset = 0;
7444	ctsio->kern_sg_entries = 0;
7445
7446	mtx_lock(&softc->ctl_lock);
7447	switch (cdb->action) {
7448	case SPRI_RK: { // read keys
7449        struct scsi_per_res_in_keys *res_keys;
7450		int i, key_count;
7451
7452		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7453
7454		/*
7455		 * We had to drop the lock to allocate our buffer, which
7456		 * leaves time for someone to come in with another
7457		 * persistent reservation.  (That is unlikely, though,
7458		 * since this should be the only persistent reservation
7459		 * command active right now.)
7460		 */
7461		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7462		    (lun->pr_key_count *
7463		     sizeof(struct scsi_per_res_key)))){
7464			mtx_unlock(&softc->ctl_lock);
7465			free(ctsio->kern_data_ptr, M_CTL);
7466			printf("%s: reservation length changed, retrying\n",
7467			       __func__);
7468			goto retry;
7469		}
7470
7471		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7472
7473		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7474			     lun->pr_key_count, res_keys->header.length);
7475
7476		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7477			if (!lun->per_res[i].registered)
7478				continue;
7479
7480			/*
7481			 * We used lun->pr_key_count to calculate the
7482			 * size to allocate.  If it turns out the number of
7483			 * initiators with the registered flag set is
7484			 * larger than that (i.e. they haven't been kept in
7485			 * sync), we've got a problem.
7486			 */
7487			if (key_count >= lun->pr_key_count) {
7488#ifdef NEEDTOPORT
7489				csevent_log(CSC_CTL | CSC_SHELF_SW |
7490					    CTL_PR_ERROR,
7491					    csevent_LogType_Fault,
7492					    csevent_AlertLevel_Yellow,
7493					    csevent_FRU_ShelfController,
7494					    csevent_FRU_Firmware,
7495				        csevent_FRU_Unknown,
7496					    "registered keys %d >= key "
7497					    "count %d", key_count,
7498					    lun->pr_key_count);
7499#endif
7500				key_count++;
7501				continue;
7502			}
7503			memcpy(res_keys->keys[key_count].key,
7504			       lun->per_res[i].res_key.key,
7505			       ctl_min(sizeof(res_keys->keys[key_count].key),
7506			       sizeof(lun->per_res[i].res_key)));
7507			key_count++;
7508		}
7509		break;
7510	}
7511	case SPRI_RR: { // read reservation
7512		struct scsi_per_res_in_rsrv *res;
7513		int tmp_len, header_only;
7514
7515		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7516
7517		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7518
7519		if (lun->flags & CTL_LUN_PR_RESERVED)
7520		{
7521			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7522			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7523				    res->header.length);
7524			header_only = 0;
7525		} else {
7526			tmp_len = sizeof(struct scsi_per_res_in_header);
7527			scsi_ulto4b(0, res->header.length);
7528			header_only = 1;
7529		}
7530
7531		/*
7532		 * We had to drop the lock to allocate our buffer, which
7533		 * leaves time for someone to come in with another
7534		 * persistent reservation.  (That is unlikely, though,
7535		 * since this should be the only persistent reservation
7536		 * command active right now.)
7537		 */
7538		if (tmp_len != total_len) {
7539			mtx_unlock(&softc->ctl_lock);
7540			free(ctsio->kern_data_ptr, M_CTL);
7541			printf("%s: reservation status changed, retrying\n",
7542			       __func__);
7543			goto retry;
7544		}
7545
7546		/*
7547		 * No reservation held, so we're done.
7548		 */
7549		if (header_only != 0)
7550			break;
7551
7552		/*
7553		 * If the registration is an All Registrants type, the key
7554		 * is 0, since it doesn't really matter.
7555		 */
7556		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7557			memcpy(res->data.reservation,
7558			       &lun->per_res[lun->pr_res_idx].res_key,
7559			       sizeof(struct scsi_per_res_key));
7560		}
7561		res->data.scopetype = lun->res_type;
7562		break;
7563	}
7564	case SPRI_RC:     //report capabilities
7565	{
7566		struct scsi_per_res_cap *res_cap;
7567		uint16_t type_mask;
7568
7569		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7570		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7571		res_cap->flags2 |= SPRI_TMV;
7572		type_mask = SPRI_TM_WR_EX_AR |
7573			    SPRI_TM_EX_AC_RO |
7574			    SPRI_TM_WR_EX_RO |
7575			    SPRI_TM_EX_AC |
7576			    SPRI_TM_WR_EX |
7577			    SPRI_TM_EX_AC_AR;
7578		scsi_ulto2b(type_mask, res_cap->type_mask);
7579		break;
7580	}
7581	case SPRI_RS: //read full status
7582	default:
7583		/*
7584		 * This is a bug, because we just checked for this above,
7585		 * and should have returned an error.
7586		 */
7587		panic("Invalid PR type %x", cdb->action);
7588		break; /* NOTREACHED */
7589	}
7590	mtx_unlock(&softc->ctl_lock);
7591
7592	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7593	ctsio->be_move_done = ctl_config_move_done;
7594
7595	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7596			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7597			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7598			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7599			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7600
7601	ctl_datamove((union ctl_io *)ctsio);
7602
7603	return (CTL_RETVAL_COMPLETE);
7604}
7605
7606/*
7607 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7608 * it should return.
7609 */
7610static int
7611ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7612		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7613		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7614		struct scsi_per_res_out_parms* param)
7615{
7616	union ctl_ha_msg persis_io;
7617	int retval, i;
7618	int isc_retval;
7619
7620	retval = 0;
7621
7622	if (sa_res_key == 0) {
7623		mtx_lock(&softc->ctl_lock);
7624		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7625			/* validate scope and type */
7626			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7627			     SPR_LU_SCOPE) {
7628				mtx_unlock(&softc->ctl_lock);
7629				ctl_set_invalid_field(/*ctsio*/ ctsio,
7630						      /*sks_valid*/ 1,
7631						      /*command*/ 1,
7632						      /*field*/ 2,
7633						      /*bit_valid*/ 1,
7634						      /*bit*/ 4);
7635				ctl_done((union ctl_io *)ctsio);
7636				return (1);
7637			}
7638
7639		        if (type>8 || type==2 || type==4 || type==0) {
7640				mtx_unlock(&softc->ctl_lock);
7641				ctl_set_invalid_field(/*ctsio*/ ctsio,
7642       	           				      /*sks_valid*/ 1,
7643						      /*command*/ 1,
7644						      /*field*/ 2,
7645						      /*bit_valid*/ 1,
7646						      /*bit*/ 0);
7647				ctl_done((union ctl_io *)ctsio);
7648				return (1);
7649		        }
7650
7651			/* temporarily unregister this nexus */
7652			lun->per_res[residx].registered = 0;
7653
7654			/*
7655			 * Unregister everybody else and build UA for
7656			 * them
7657			 */
7658			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7659				if (lun->per_res[i].registered == 0)
7660					continue;
7661
7662				if (!persis_offset
7663				 && i <CTL_MAX_INITIATORS)
7664					lun->pending_sense[i].ua_pending |=
7665						CTL_UA_REG_PREEMPT;
7666				else if (persis_offset
7667				      && i >= persis_offset)
7668					lun->pending_sense[i-persis_offset
7669						].ua_pending |=
7670						CTL_UA_REG_PREEMPT;
7671				lun->per_res[i].registered = 0;
7672				memset(&lun->per_res[i].res_key, 0,
7673				       sizeof(struct scsi_per_res_key));
7674			}
7675			lun->per_res[residx].registered = 1;
7676			lun->pr_key_count = 1;
7677			lun->res_type = type;
7678			if (lun->res_type != SPR_TYPE_WR_EX_AR
7679			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7680				lun->pr_res_idx = residx;
7681
7682			mtx_unlock(&softc->ctl_lock);
7683			/* send msg to other side */
7684			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7685			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7686			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7687			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7688			persis_io.pr.pr_info.res_type = type;
7689			memcpy(persis_io.pr.pr_info.sa_res_key,
7690			       param->serv_act_res_key,
7691			       sizeof(param->serv_act_res_key));
7692			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7693			     &persis_io, sizeof(persis_io), 0)) >
7694			     CTL_HA_STATUS_SUCCESS) {
7695				printf("CTL:Persis Out error returned "
7696				       "from ctl_ha_msg_send %d\n",
7697				       isc_retval);
7698			}
7699		} else {
7700			/* not all registrants */
7701			mtx_unlock(&softc->ctl_lock);
7702			free(ctsio->kern_data_ptr, M_CTL);
7703			ctl_set_invalid_field(ctsio,
7704					      /*sks_valid*/ 1,
7705					      /*command*/ 0,
7706					      /*field*/ 8,
7707					      /*bit_valid*/ 0,
7708					      /*bit*/ 0);
7709			ctl_done((union ctl_io *)ctsio);
7710			return (1);
7711		}
7712	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7713		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7714		int found = 0;
7715
7716		mtx_lock(&softc->ctl_lock);
7717		if (res_key == sa_res_key) {
7718			/* special case */
7719			/*
7720			 * The spec implies this is not good but doesn't
7721			 * say what to do. There are two choices either
7722			 * generate a res conflict or check condition
7723			 * with illegal field in parameter data. Since
7724			 * that is what is done when the sa_res_key is
7725			 * zero I'll take that approach since this has
7726			 * to do with the sa_res_key.
7727			 */
7728			mtx_unlock(&softc->ctl_lock);
7729			free(ctsio->kern_data_ptr, M_CTL);
7730			ctl_set_invalid_field(ctsio,
7731					      /*sks_valid*/ 1,
7732					      /*command*/ 0,
7733					      /*field*/ 8,
7734					      /*bit_valid*/ 0,
7735					      /*bit*/ 0);
7736			ctl_done((union ctl_io *)ctsio);
7737			return (1);
7738		}
7739
7740		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7741			if (lun->per_res[i].registered
7742			 && memcmp(param->serv_act_res_key,
7743			    lun->per_res[i].res_key.key,
7744			    sizeof(struct scsi_per_res_key)) != 0)
7745				continue;
7746
7747			found = 1;
7748			lun->per_res[i].registered = 0;
7749			memset(&lun->per_res[i].res_key, 0,
7750			       sizeof(struct scsi_per_res_key));
7751			lun->pr_key_count--;
7752
7753			if (!persis_offset
7754			 && i < CTL_MAX_INITIATORS)
7755				lun->pending_sense[i].ua_pending |=
7756					CTL_UA_REG_PREEMPT;
7757			else if (persis_offset
7758			      && i >= persis_offset)
7759				lun->pending_sense[i-persis_offset].ua_pending|=
7760					CTL_UA_REG_PREEMPT;
7761		}
7762		mtx_unlock(&softc->ctl_lock);
7763		if (!found) {
7764			free(ctsio->kern_data_ptr, M_CTL);
7765			ctl_set_reservation_conflict(ctsio);
7766			ctl_done((union ctl_io *)ctsio);
7767			return (CTL_RETVAL_COMPLETE);
7768		}
7769		/* send msg to other side */
7770		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7771		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7772		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7773		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7774		persis_io.pr.pr_info.res_type = type;
7775		memcpy(persis_io.pr.pr_info.sa_res_key,
7776		       param->serv_act_res_key,
7777		       sizeof(param->serv_act_res_key));
7778		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7779		     &persis_io, sizeof(persis_io), 0)) >
7780		     CTL_HA_STATUS_SUCCESS) {
7781			printf("CTL:Persis Out error returned from "
7782			       "ctl_ha_msg_send %d\n", isc_retval);
7783		}
7784	} else {
7785		/* Reserved but not all registrants */
7786		/* sa_res_key is res holder */
7787		if (memcmp(param->serv_act_res_key,
7788                   lun->per_res[lun->pr_res_idx].res_key.key,
7789                   sizeof(struct scsi_per_res_key)) == 0) {
7790			/* validate scope and type */
7791			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7792			     SPR_LU_SCOPE) {
7793				ctl_set_invalid_field(/*ctsio*/ ctsio,
7794						      /*sks_valid*/ 1,
7795						      /*command*/ 1,
7796						      /*field*/ 2,
7797						      /*bit_valid*/ 1,
7798						      /*bit*/ 4);
7799				ctl_done((union ctl_io *)ctsio);
7800				return (1);
7801			}
7802
7803			if (type>8 || type==2 || type==4 || type==0) {
7804				ctl_set_invalid_field(/*ctsio*/ ctsio,
7805						      /*sks_valid*/ 1,
7806						      /*command*/ 1,
7807						      /*field*/ 2,
7808						      /*bit_valid*/ 1,
7809						      /*bit*/ 0);
7810				ctl_done((union ctl_io *)ctsio);
7811				return (1);
7812			}
7813
7814			/*
7815			 * Do the following:
7816			 * if sa_res_key != res_key remove all
7817			 * registrants w/sa_res_key and generate UA
7818			 * for these registrants(Registrations
7819			 * Preempted) if it wasn't an exclusive
7820			 * reservation generate UA(Reservations
7821			 * Preempted) for all other registered nexuses
7822			 * if the type has changed. Establish the new
7823			 * reservation and holder. If res_key and
7824			 * sa_res_key are the same do the above
7825			 * except don't unregister the res holder.
7826			 */
7827
7828			/*
7829			 * Temporarily unregister so it won't get
7830			 * removed or UA generated
7831			 */
7832			lun->per_res[residx].registered = 0;
7833			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7834				if (lun->per_res[i].registered == 0)
7835					continue;
7836
7837				if (memcmp(param->serv_act_res_key,
7838				    lun->per_res[i].res_key.key,
7839				    sizeof(struct scsi_per_res_key)) == 0) {
7840					lun->per_res[i].registered = 0;
7841					memset(&lun->per_res[i].res_key,
7842					       0,
7843					       sizeof(struct scsi_per_res_key));
7844					lun->pr_key_count--;
7845
7846					if (!persis_offset
7847					 && i < CTL_MAX_INITIATORS)
7848						lun->pending_sense[i
7849							].ua_pending |=
7850							CTL_UA_REG_PREEMPT;
7851					else if (persis_offset
7852					      && i >= persis_offset)
7853						lun->pending_sense[
7854						  i-persis_offset].ua_pending |=
7855						  CTL_UA_REG_PREEMPT;
7856				} else if (type != lun->res_type
7857					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7858					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7859						if (!persis_offset
7860						 && i < CTL_MAX_INITIATORS)
7861							lun->pending_sense[i
7862							].ua_pending |=
7863							CTL_UA_RES_RELEASE;
7864						else if (persis_offset
7865						      && i >= persis_offset)
7866							lun->pending_sense[
7867							i-persis_offset
7868							].ua_pending |=
7869							CTL_UA_RES_RELEASE;
7870				}
7871			}
7872			lun->per_res[residx].registered = 1;
7873			lun->res_type = type;
7874			if (lun->res_type != SPR_TYPE_WR_EX_AR
7875			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7876				lun->pr_res_idx = residx;
7877			else
7878				lun->pr_res_idx =
7879					CTL_PR_ALL_REGISTRANTS;
7880
7881			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7882			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7883			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7884			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7885			persis_io.pr.pr_info.res_type = type;
7886			memcpy(persis_io.pr.pr_info.sa_res_key,
7887			       param->serv_act_res_key,
7888			       sizeof(param->serv_act_res_key));
7889			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7890			     &persis_io, sizeof(persis_io), 0)) >
7891			     CTL_HA_STATUS_SUCCESS) {
7892				printf("CTL:Persis Out error returned "
7893				       "from ctl_ha_msg_send %d\n",
7894				       isc_retval);
7895			}
7896		} else {
7897			/*
7898			 * sa_res_key is not the res holder just
7899			 * remove registrants
7900			 */
7901			int found=0;
7902			mtx_lock(&softc->ctl_lock);
7903
7904			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7905				if (memcmp(param->serv_act_res_key,
7906				    lun->per_res[i].res_key.key,
7907				    sizeof(struct scsi_per_res_key)) != 0)
7908					continue;
7909
7910				found = 1;
7911				lun->per_res[i].registered = 0;
7912				memset(&lun->per_res[i].res_key, 0,
7913				       sizeof(struct scsi_per_res_key));
7914				lun->pr_key_count--;
7915
7916				if (!persis_offset
7917				 && i < CTL_MAX_INITIATORS)
7918					lun->pending_sense[i].ua_pending |=
7919						CTL_UA_REG_PREEMPT;
7920				else if (persis_offset
7921				      && i >= persis_offset)
7922					lun->pending_sense[
7923						i-persis_offset].ua_pending |=
7924						CTL_UA_REG_PREEMPT;
7925			}
7926
7927			if (!found) {
7928				mtx_unlock(&softc->ctl_lock);
7929				free(ctsio->kern_data_ptr, M_CTL);
7930				ctl_set_reservation_conflict(ctsio);
7931				ctl_done((union ctl_io *)ctsio);
7932		        	return (1);
7933			}
7934			mtx_unlock(&softc->ctl_lock);
7935			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7936			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7937			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7938			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7939			persis_io.pr.pr_info.res_type = type;
7940			memcpy(persis_io.pr.pr_info.sa_res_key,
7941			       param->serv_act_res_key,
7942			       sizeof(param->serv_act_res_key));
7943			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7944			     &persis_io, sizeof(persis_io), 0)) >
7945			     CTL_HA_STATUS_SUCCESS) {
7946				printf("CTL:Persis Out error returned "
7947				       "from ctl_ha_msg_send %d\n",
7948				isc_retval);
7949			}
7950		}
7951	}
7952
7953	lun->PRGeneration++;
7954
7955	return (retval);
7956}
7957
7958static void
7959ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7960{
7961	int i;
7962
7963	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7964	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7965	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7966		   msg->pr.pr_info.sa_res_key,
7967		   sizeof(struct scsi_per_res_key)) != 0) {
7968		uint64_t sa_res_key;
7969		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7970
7971		if (sa_res_key == 0) {
7972			/* temporarily unregister this nexus */
7973			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7974
7975			/*
7976			 * Unregister everybody else and build UA for
7977			 * them
7978			 */
7979			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7980				if (lun->per_res[i].registered == 0)
7981					continue;
7982
7983				if (!persis_offset
7984				 && i < CTL_MAX_INITIATORS)
7985					lun->pending_sense[i].ua_pending |=
7986						CTL_UA_REG_PREEMPT;
7987				else if (persis_offset && i >= persis_offset)
7988					lun->pending_sense[i -
7989						persis_offset].ua_pending |=
7990						CTL_UA_REG_PREEMPT;
7991				lun->per_res[i].registered = 0;
7992				memset(&lun->per_res[i].res_key, 0,
7993				       sizeof(struct scsi_per_res_key));
7994			}
7995
7996			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7997			lun->pr_key_count = 1;
7998			lun->res_type = msg->pr.pr_info.res_type;
7999			if (lun->res_type != SPR_TYPE_WR_EX_AR
8000			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8001				lun->pr_res_idx = msg->pr.pr_info.residx;
8002		} else {
8003		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8004				if (memcmp(msg->pr.pr_info.sa_res_key,
8005		                   lun->per_res[i].res_key.key,
8006		                   sizeof(struct scsi_per_res_key)) != 0)
8007					continue;
8008
8009				lun->per_res[i].registered = 0;
8010				memset(&lun->per_res[i].res_key, 0,
8011				       sizeof(struct scsi_per_res_key));
8012				lun->pr_key_count--;
8013
8014				if (!persis_offset
8015				 && i < persis_offset)
8016					lun->pending_sense[i].ua_pending |=
8017						CTL_UA_REG_PREEMPT;
8018				else if (persis_offset
8019				      && i >= persis_offset)
8020					lun->pending_sense[i -
8021						persis_offset].ua_pending |=
8022						CTL_UA_REG_PREEMPT;
8023			}
8024		}
8025	} else {
8026		/*
8027		 * Temporarily unregister so it won't get removed
8028		 * or UA generated
8029		 */
8030		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8031		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8032			if (lun->per_res[i].registered == 0)
8033				continue;
8034
8035			if (memcmp(msg->pr.pr_info.sa_res_key,
8036	                   lun->per_res[i].res_key.key,
8037	                   sizeof(struct scsi_per_res_key)) == 0) {
8038				lun->per_res[i].registered = 0;
8039				memset(&lun->per_res[i].res_key, 0,
8040				       sizeof(struct scsi_per_res_key));
8041				lun->pr_key_count--;
8042				if (!persis_offset
8043				 && i < CTL_MAX_INITIATORS)
8044					lun->pending_sense[i].ua_pending |=
8045						CTL_UA_REG_PREEMPT;
8046				else if (persis_offset
8047				      && i >= persis_offset)
8048					lun->pending_sense[i -
8049						persis_offset].ua_pending |=
8050						CTL_UA_REG_PREEMPT;
8051			} else if (msg->pr.pr_info.res_type != lun->res_type
8052				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8053				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8054					if (!persis_offset
8055					 && i < persis_offset)
8056						lun->pending_sense[i
8057							].ua_pending |=
8058							CTL_UA_RES_RELEASE;
8059					else if (persis_offset
8060					      && i >= persis_offset)
8061					lun->pending_sense[i -
8062						persis_offset].ua_pending |=
8063						CTL_UA_RES_RELEASE;
8064			}
8065		}
8066		lun->per_res[msg->pr.pr_info.residx].registered = 1;
8067		lun->res_type = msg->pr.pr_info.res_type;
8068		if (lun->res_type != SPR_TYPE_WR_EX_AR
8069		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8070			lun->pr_res_idx = msg->pr.pr_info.residx;
8071		else
8072			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8073	}
8074	lun->PRGeneration++;
8075
8076}
8077
8078
8079int
8080ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8081{
8082	int retval;
8083	int isc_retval;
8084	u_int32_t param_len;
8085	struct scsi_per_res_out *cdb;
8086	struct ctl_lun *lun;
8087	struct scsi_per_res_out_parms* param;
8088	struct ctl_softc *softc;
8089	uint32_t residx;
8090	uint64_t res_key, sa_res_key;
8091	uint8_t type;
8092	union ctl_ha_msg persis_io;
8093	int    i;
8094
8095	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8096
8097	retval = CTL_RETVAL_COMPLETE;
8098
8099	softc = control_softc;
8100
8101	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8102	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8103
8104	/*
8105	 * We only support whole-LUN scope.  The scope & type are ignored for
8106	 * register, register and ignore existing key and clear.
8107	 * We sometimes ignore scope and type on preempts too!!
8108	 * Verify reservation type here as well.
8109	 */
8110	type = cdb->scope_type & SPR_TYPE_MASK;
8111	if ((cdb->action == SPRO_RESERVE)
8112	 || (cdb->action == SPRO_RELEASE)) {
8113		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8114			ctl_set_invalid_field(/*ctsio*/ ctsio,
8115					      /*sks_valid*/ 1,
8116					      /*command*/ 1,
8117					      /*field*/ 2,
8118					      /*bit_valid*/ 1,
8119					      /*bit*/ 4);
8120			ctl_done((union ctl_io *)ctsio);
8121			return (CTL_RETVAL_COMPLETE);
8122		}
8123
8124		if (type>8 || type==2 || type==4 || type==0) {
8125			ctl_set_invalid_field(/*ctsio*/ ctsio,
8126					      /*sks_valid*/ 1,
8127					      /*command*/ 1,
8128					      /*field*/ 2,
8129					      /*bit_valid*/ 1,
8130					      /*bit*/ 0);
8131			ctl_done((union ctl_io *)ctsio);
8132			return (CTL_RETVAL_COMPLETE);
8133		}
8134	}
8135
8136	switch (cdb->action & SPRO_ACTION_MASK) {
8137	case SPRO_REGISTER:
8138	case SPRO_RESERVE:
8139	case SPRO_RELEASE:
8140	case SPRO_CLEAR:
8141	case SPRO_PREEMPT:
8142	case SPRO_REG_IGNO:
8143		break;
8144	case SPRO_REG_MOVE:
8145	case SPRO_PRE_ABO:
8146	default:
8147		ctl_set_invalid_field(/*ctsio*/ ctsio,
8148				      /*sks_valid*/ 1,
8149				      /*command*/ 1,
8150				      /*field*/ 1,
8151				      /*bit_valid*/ 1,
8152				      /*bit*/ 0);
8153		ctl_done((union ctl_io *)ctsio);
8154		return (CTL_RETVAL_COMPLETE);
8155		break; /* NOTREACHED */
8156	}
8157
8158	param_len = scsi_4btoul(cdb->length);
8159
8160	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8161		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8162		ctsio->kern_data_len = param_len;
8163		ctsio->kern_total_len = param_len;
8164		ctsio->kern_data_resid = 0;
8165		ctsio->kern_rel_offset = 0;
8166		ctsio->kern_sg_entries = 0;
8167		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8168		ctsio->be_move_done = ctl_config_move_done;
8169		ctl_datamove((union ctl_io *)ctsio);
8170
8171		return (CTL_RETVAL_COMPLETE);
8172	}
8173
8174	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8175
8176	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8177	res_key = scsi_8btou64(param->res_key.key);
8178	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8179
8180	/*
8181	 * Validate the reservation key here except for SPRO_REG_IGNO
8182	 * This must be done for all other service actions
8183	 */
8184	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8185		mtx_lock(&softc->ctl_lock);
8186		if (lun->per_res[residx].registered) {
8187		    if (memcmp(param->res_key.key,
8188			       lun->per_res[residx].res_key.key,
8189			       ctl_min(sizeof(param->res_key),
8190			       sizeof(lun->per_res[residx].res_key))) != 0) {
8191				/*
8192				 * The current key passed in doesn't match
8193				 * the one the initiator previously
8194				 * registered.
8195				 */
8196				mtx_unlock(&softc->ctl_lock);
8197				free(ctsio->kern_data_ptr, M_CTL);
8198				ctl_set_reservation_conflict(ctsio);
8199				ctl_done((union ctl_io *)ctsio);
8200				return (CTL_RETVAL_COMPLETE);
8201			}
8202		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8203			/*
8204			 * We are not registered
8205			 */
8206			mtx_unlock(&softc->ctl_lock);
8207			free(ctsio->kern_data_ptr, M_CTL);
8208			ctl_set_reservation_conflict(ctsio);
8209			ctl_done((union ctl_io *)ctsio);
8210			return (CTL_RETVAL_COMPLETE);
8211		} else if (res_key != 0) {
8212			/*
8213			 * We are not registered and trying to register but
8214			 * the register key isn't zero.
8215			 */
8216			mtx_unlock(&softc->ctl_lock);
8217			free(ctsio->kern_data_ptr, M_CTL);
8218			ctl_set_reservation_conflict(ctsio);
8219			ctl_done((union ctl_io *)ctsio);
8220			return (CTL_RETVAL_COMPLETE);
8221		}
8222		mtx_unlock(&softc->ctl_lock);
8223	}
8224
8225	switch (cdb->action & SPRO_ACTION_MASK) {
8226	case SPRO_REGISTER:
8227	case SPRO_REG_IGNO: {
8228
8229#if 0
8230		printf("Registration received\n");
8231#endif
8232
8233		/*
8234		 * We don't support any of these options, as we report in
8235		 * the read capabilities request (see
8236		 * ctl_persistent_reserve_in(), above).
8237		 */
8238		if ((param->flags & SPR_SPEC_I_PT)
8239		 || (param->flags & SPR_ALL_TG_PT)
8240		 || (param->flags & SPR_APTPL)) {
8241			int bit_ptr;
8242
8243			if (param->flags & SPR_APTPL)
8244				bit_ptr = 0;
8245			else if (param->flags & SPR_ALL_TG_PT)
8246				bit_ptr = 2;
8247			else /* SPR_SPEC_I_PT */
8248				bit_ptr = 3;
8249
8250			free(ctsio->kern_data_ptr, M_CTL);
8251			ctl_set_invalid_field(ctsio,
8252					      /*sks_valid*/ 1,
8253					      /*command*/ 0,
8254					      /*field*/ 20,
8255					      /*bit_valid*/ 1,
8256					      /*bit*/ bit_ptr);
8257			ctl_done((union ctl_io *)ctsio);
8258			return (CTL_RETVAL_COMPLETE);
8259		}
8260
8261		mtx_lock(&softc->ctl_lock);
8262
8263		/*
8264		 * The initiator wants to clear the
8265		 * key/unregister.
8266		 */
8267		if (sa_res_key == 0) {
8268			if ((res_key == 0
8269			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8270			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8271			  && !lun->per_res[residx].registered)) {
8272				mtx_unlock(&softc->ctl_lock);
8273				goto done;
8274			}
8275
8276			lun->per_res[residx].registered = 0;
8277			memset(&lun->per_res[residx].res_key,
8278			       0, sizeof(lun->per_res[residx].res_key));
8279			lun->pr_key_count--;
8280
8281			if (residx == lun->pr_res_idx) {
8282				lun->flags &= ~CTL_LUN_PR_RESERVED;
8283				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8284
8285				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8286				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8287				 && lun->pr_key_count) {
8288					/*
8289					 * If the reservation is a registrants
8290					 * only type we need to generate a UA
8291					 * for other registered inits.  The
8292					 * sense code should be RESERVATIONS
8293					 * RELEASED
8294					 */
8295
8296					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8297						if (lun->per_res[
8298						    i+persis_offset].registered
8299						    == 0)
8300							continue;
8301						lun->pending_sense[i
8302							].ua_pending |=
8303							CTL_UA_RES_RELEASE;
8304					}
8305				}
8306				lun->res_type = 0;
8307			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8308				if (lun->pr_key_count==0) {
8309					lun->flags &= ~CTL_LUN_PR_RESERVED;
8310					lun->res_type = 0;
8311					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8312				}
8313			}
8314			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8315			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8316			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8317			persis_io.pr.pr_info.residx = residx;
8318			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8319			     &persis_io, sizeof(persis_io), 0 )) >
8320			     CTL_HA_STATUS_SUCCESS) {
8321				printf("CTL:Persis Out error returned from "
8322				       "ctl_ha_msg_send %d\n", isc_retval);
8323			}
8324			mtx_unlock(&softc->ctl_lock);
8325		} else /* sa_res_key != 0 */ {
8326
8327			/*
8328			 * If we aren't registered currently then increment
8329			 * the key count and set the registered flag.
8330			 */
8331			if (!lun->per_res[residx].registered) {
8332				lun->pr_key_count++;
8333				lun->per_res[residx].registered = 1;
8334			}
8335
8336			memcpy(&lun->per_res[residx].res_key,
8337			       param->serv_act_res_key,
8338			       ctl_min(sizeof(param->serv_act_res_key),
8339			       sizeof(lun->per_res[residx].res_key)));
8340
8341			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8342			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8343			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8344			persis_io.pr.pr_info.residx = residx;
8345			memcpy(persis_io.pr.pr_info.sa_res_key,
8346			       param->serv_act_res_key,
8347			       sizeof(param->serv_act_res_key));
8348			mtx_unlock(&softc->ctl_lock);
8349			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8350			     &persis_io, sizeof(persis_io), 0)) >
8351			     CTL_HA_STATUS_SUCCESS) {
8352				printf("CTL:Persis Out error returned from "
8353				       "ctl_ha_msg_send %d\n", isc_retval);
8354			}
8355		}
8356		lun->PRGeneration++;
8357
8358		break;
8359	}
8360	case SPRO_RESERVE:
8361#if 0
8362                printf("Reserve executed type %d\n", type);
8363#endif
8364		mtx_lock(&softc->ctl_lock);
8365		if (lun->flags & CTL_LUN_PR_RESERVED) {
8366			/*
8367			 * if this isn't the reservation holder and it's
8368			 * not a "all registrants" type or if the type is
8369			 * different then we have a conflict
8370			 */
8371			if ((lun->pr_res_idx != residx
8372			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8373			 || lun->res_type != type) {
8374				mtx_unlock(&softc->ctl_lock);
8375				free(ctsio->kern_data_ptr, M_CTL);
8376				ctl_set_reservation_conflict(ctsio);
8377				ctl_done((union ctl_io *)ctsio);
8378				return (CTL_RETVAL_COMPLETE);
8379			}
8380			mtx_unlock(&softc->ctl_lock);
8381		} else /* create a reservation */ {
8382			/*
8383			 * If it's not an "all registrants" type record
8384			 * reservation holder
8385			 */
8386			if (type != SPR_TYPE_WR_EX_AR
8387			 && type != SPR_TYPE_EX_AC_AR)
8388				lun->pr_res_idx = residx; /* Res holder */
8389			else
8390				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8391
8392			lun->flags |= CTL_LUN_PR_RESERVED;
8393			lun->res_type = type;
8394
8395			mtx_unlock(&softc->ctl_lock);
8396
8397			/* send msg to other side */
8398			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8399			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8400			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8401			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8402			persis_io.pr.pr_info.res_type = type;
8403			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8404			     &persis_io, sizeof(persis_io), 0)) >
8405			     CTL_HA_STATUS_SUCCESS) {
8406				printf("CTL:Persis Out error returned from "
8407				       "ctl_ha_msg_send %d\n", isc_retval);
8408			}
8409		}
8410		break;
8411
8412	case SPRO_RELEASE:
8413		mtx_lock(&softc->ctl_lock);
8414		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8415			/* No reservation exists return good status */
8416			mtx_unlock(&softc->ctl_lock);
8417			goto done;
8418		}
8419		/*
8420		 * Is this nexus a reservation holder?
8421		 */
8422		if (lun->pr_res_idx != residx
8423		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8424			/*
8425			 * not a res holder return good status but
8426			 * do nothing
8427			 */
8428			mtx_unlock(&softc->ctl_lock);
8429			goto done;
8430		}
8431
8432		if (lun->res_type != type) {
8433			mtx_unlock(&softc->ctl_lock);
8434			free(ctsio->kern_data_ptr, M_CTL);
8435			ctl_set_illegal_pr_release(ctsio);
8436			ctl_done((union ctl_io *)ctsio);
8437			return (CTL_RETVAL_COMPLETE);
8438		}
8439
8440		/* okay to release */
8441		lun->flags &= ~CTL_LUN_PR_RESERVED;
8442		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8443		lun->res_type = 0;
8444
8445		/*
8446		 * if this isn't an exclusive access
8447		 * res generate UA for all other
8448		 * registrants.
8449		 */
8450		if (type != SPR_TYPE_EX_AC
8451		 && type != SPR_TYPE_WR_EX) {
8452			/*
8453			 * temporarily unregister so we don't generate UA
8454			 */
8455			lun->per_res[residx].registered = 0;
8456
8457			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8458				if (lun->per_res[i+persis_offset].registered
8459				    == 0)
8460					continue;
8461				lun->pending_sense[i].ua_pending |=
8462					CTL_UA_RES_RELEASE;
8463			}
8464
8465			lun->per_res[residx].registered = 1;
8466		}
8467		mtx_unlock(&softc->ctl_lock);
8468		/* Send msg to other side */
8469		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8470		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8471		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8472		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8473		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8474			printf("CTL:Persis Out error returned from "
8475			       "ctl_ha_msg_send %d\n", isc_retval);
8476		}
8477		break;
8478
8479	case SPRO_CLEAR:
8480		/* send msg to other side */
8481
8482		mtx_lock(&softc->ctl_lock);
8483		lun->flags &= ~CTL_LUN_PR_RESERVED;
8484		lun->res_type = 0;
8485		lun->pr_key_count = 0;
8486		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8487
8488
8489		memset(&lun->per_res[residx].res_key,
8490		       0, sizeof(lun->per_res[residx].res_key));
8491		lun->per_res[residx].registered = 0;
8492
8493		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8494			if (lun->per_res[i].registered) {
8495				if (!persis_offset && i < CTL_MAX_INITIATORS)
8496					lun->pending_sense[i].ua_pending |=
8497						CTL_UA_RES_PREEMPT;
8498				else if (persis_offset && i >= persis_offset)
8499					lun->pending_sense[i-persis_offset
8500					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8501
8502				memset(&lun->per_res[i].res_key,
8503				       0, sizeof(struct scsi_per_res_key));
8504				lun->per_res[i].registered = 0;
8505			}
8506		lun->PRGeneration++;
8507		mtx_unlock(&softc->ctl_lock);
8508		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8509		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8510		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8511		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8512		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8513			printf("CTL:Persis Out error returned from "
8514			       "ctl_ha_msg_send %d\n", isc_retval);
8515		}
8516		break;
8517
8518	case SPRO_PREEMPT: {
8519		int nretval;
8520
8521		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8522					  residx, ctsio, cdb, param);
8523		if (nretval != 0)
8524			return (CTL_RETVAL_COMPLETE);
8525		break;
8526	}
8527	case SPRO_REG_MOVE:
8528	case SPRO_PRE_ABO:
8529	default:
8530		free(ctsio->kern_data_ptr, M_CTL);
8531		ctl_set_invalid_field(/*ctsio*/ ctsio,
8532				      /*sks_valid*/ 1,
8533				      /*command*/ 1,
8534				      /*field*/ 1,
8535				      /*bit_valid*/ 1,
8536				      /*bit*/ 0);
8537		ctl_done((union ctl_io *)ctsio);
8538		return (CTL_RETVAL_COMPLETE);
8539		break; /* NOTREACHED */
8540	}
8541
8542done:
8543	free(ctsio->kern_data_ptr, M_CTL);
8544	ctl_set_success(ctsio);
8545	ctl_done((union ctl_io *)ctsio);
8546
8547	return (retval);
8548}
8549
8550/*
8551 * This routine is for handling a message from the other SC pertaining to
8552 * persistent reserve out. All the error checking will have been done
8553 * so only perorming the action need be done here to keep the two
8554 * in sync.
8555 */
8556static void
8557ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8558{
8559	struct ctl_lun *lun;
8560	struct ctl_softc *softc;
8561	int i;
8562	uint32_t targ_lun;
8563
8564	softc = control_softc;
8565
8566	mtx_lock(&softc->ctl_lock);
8567
8568	targ_lun = msg->hdr.nexus.targ_lun;
8569	if (msg->hdr.nexus.lun_map_fn != NULL)
8570		targ_lun = msg->hdr.nexus.lun_map_fn(msg->hdr.nexus.lun_map_arg, targ_lun);
8571	lun = softc->ctl_luns[targ_lun];
8572	switch(msg->pr.pr_info.action) {
8573	case CTL_PR_REG_KEY:
8574		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8575			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8576			lun->pr_key_count++;
8577		}
8578		lun->PRGeneration++;
8579		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8580		       msg->pr.pr_info.sa_res_key,
8581		       sizeof(struct scsi_per_res_key));
8582		break;
8583
8584	case CTL_PR_UNREG_KEY:
8585		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8586		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8587		       0, sizeof(struct scsi_per_res_key));
8588		lun->pr_key_count--;
8589
8590		/* XXX Need to see if the reservation has been released */
8591		/* if so do we need to generate UA? */
8592		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8593			lun->flags &= ~CTL_LUN_PR_RESERVED;
8594			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8595
8596			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8597			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8598			 && lun->pr_key_count) {
8599				/*
8600				 * If the reservation is a registrants
8601				 * only type we need to generate a UA
8602				 * for other registered inits.  The
8603				 * sense code should be RESERVATIONS
8604				 * RELEASED
8605				 */
8606
8607				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8608					if (lun->per_res[i+
8609					    persis_offset].registered == 0)
8610						continue;
8611
8612					lun->pending_sense[i
8613						].ua_pending |=
8614						CTL_UA_RES_RELEASE;
8615				}
8616			}
8617			lun->res_type = 0;
8618		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8619			if (lun->pr_key_count==0) {
8620				lun->flags &= ~CTL_LUN_PR_RESERVED;
8621				lun->res_type = 0;
8622				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8623			}
8624		}
8625		lun->PRGeneration++;
8626		break;
8627
8628	case CTL_PR_RESERVE:
8629		lun->flags |= CTL_LUN_PR_RESERVED;
8630		lun->res_type = msg->pr.pr_info.res_type;
8631		lun->pr_res_idx = msg->pr.pr_info.residx;
8632
8633		break;
8634
8635	case CTL_PR_RELEASE:
8636		/*
8637		 * if this isn't an exclusive access res generate UA for all
8638		 * other registrants.
8639		 */
8640		if (lun->res_type != SPR_TYPE_EX_AC
8641		 && lun->res_type != SPR_TYPE_WR_EX) {
8642			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8643				if (lun->per_res[i+persis_offset].registered)
8644					lun->pending_sense[i].ua_pending |=
8645						CTL_UA_RES_RELEASE;
8646		}
8647
8648		lun->flags &= ~CTL_LUN_PR_RESERVED;
8649		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8650		lun->res_type = 0;
8651		break;
8652
8653	case CTL_PR_PREEMPT:
8654		ctl_pro_preempt_other(lun, msg);
8655		break;
8656	case CTL_PR_CLEAR:
8657		lun->flags &= ~CTL_LUN_PR_RESERVED;
8658		lun->res_type = 0;
8659		lun->pr_key_count = 0;
8660		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8661
8662		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8663			if (lun->per_res[i].registered == 0)
8664				continue;
8665			if (!persis_offset
8666			 && i < CTL_MAX_INITIATORS)
8667				lun->pending_sense[i].ua_pending |=
8668					CTL_UA_RES_PREEMPT;
8669			else if (persis_offset
8670			      && i >= persis_offset)
8671   				lun->pending_sense[i-persis_offset].ua_pending|=
8672					CTL_UA_RES_PREEMPT;
8673			memset(&lun->per_res[i].res_key, 0,
8674			       sizeof(struct scsi_per_res_key));
8675			lun->per_res[i].registered = 0;
8676		}
8677		lun->PRGeneration++;
8678		break;
8679	}
8680
8681	mtx_unlock(&softc->ctl_lock);
8682}
8683
8684int
8685ctl_read_write(struct ctl_scsiio *ctsio)
8686{
8687	struct ctl_lun *lun;
8688	struct ctl_lba_len_flags *lbalen;
8689	uint64_t lba;
8690	uint32_t num_blocks;
8691	int reladdr, fua, dpo, ebp;
8692	int retval;
8693	int isread;
8694
8695	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8696
8697	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8698
8699	reladdr = 0;
8700	fua = 0;
8701	dpo = 0;
8702	ebp = 0;
8703
8704	retval = CTL_RETVAL_COMPLETE;
8705
8706	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8707	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8708	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8709		uint32_t residx;
8710
8711		/*
8712		 * XXX KDM need a lock here.
8713		 */
8714		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8715		if ((lun->res_type == SPR_TYPE_EX_AC
8716		  && residx != lun->pr_res_idx)
8717		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8718		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8719		  && !lun->per_res[residx].registered)) {
8720			ctl_set_reservation_conflict(ctsio);
8721			ctl_done((union ctl_io *)ctsio);
8722			return (CTL_RETVAL_COMPLETE);
8723	        }
8724	}
8725
8726	switch (ctsio->cdb[0]) {
8727	case READ_6:
8728	case WRITE_6: {
8729		struct scsi_rw_6 *cdb;
8730
8731		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8732
8733		lba = scsi_3btoul(cdb->addr);
8734		/* only 5 bits are valid in the most significant address byte */
8735		lba &= 0x1fffff;
8736		num_blocks = cdb->length;
8737		/*
8738		 * This is correct according to SBC-2.
8739		 */
8740		if (num_blocks == 0)
8741			num_blocks = 256;
8742		break;
8743	}
8744	case READ_10:
8745	case WRITE_10: {
8746		struct scsi_rw_10 *cdb;
8747
8748		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8749
8750		if (cdb->byte2 & SRW10_RELADDR)
8751			reladdr = 1;
8752		if (cdb->byte2 & SRW10_FUA)
8753			fua = 1;
8754		if (cdb->byte2 & SRW10_DPO)
8755			dpo = 1;
8756
8757		if ((cdb->opcode == WRITE_10)
8758		 && (cdb->byte2 & SRW10_EBP))
8759			ebp = 1;
8760
8761		lba = scsi_4btoul(cdb->addr);
8762		num_blocks = scsi_2btoul(cdb->length);
8763		break;
8764	}
8765	case WRITE_VERIFY_10: {
8766		struct scsi_write_verify_10 *cdb;
8767
8768		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8769
8770		/*
8771		 * XXX KDM we should do actual write verify support at some
8772		 * point.  This is obviously fake, we're just translating
8773		 * things to a write.  So we don't even bother checking the
8774		 * BYTCHK field, since we don't do any verification.  If
8775		 * the user asks for it, we'll just pretend we did it.
8776		 */
8777		if (cdb->byte2 & SWV_DPO)
8778			dpo = 1;
8779
8780		lba = scsi_4btoul(cdb->addr);
8781		num_blocks = scsi_2btoul(cdb->length);
8782		break;
8783	}
8784	case READ_12:
8785	case WRITE_12: {
8786		struct scsi_rw_12 *cdb;
8787
8788		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8789
8790		if (cdb->byte2 & SRW12_RELADDR)
8791			reladdr = 1;
8792		if (cdb->byte2 & SRW12_FUA)
8793			fua = 1;
8794		if (cdb->byte2 & SRW12_DPO)
8795			dpo = 1;
8796		lba = scsi_4btoul(cdb->addr);
8797		num_blocks = scsi_4btoul(cdb->length);
8798		break;
8799	}
8800	case WRITE_VERIFY_12: {
8801		struct scsi_write_verify_12 *cdb;
8802
8803		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8804
8805		if (cdb->byte2 & SWV_DPO)
8806			dpo = 1;
8807
8808		lba = scsi_4btoul(cdb->addr);
8809		num_blocks = scsi_4btoul(cdb->length);
8810
8811		break;
8812	}
8813	case READ_16:
8814	case WRITE_16: {
8815		struct scsi_rw_16 *cdb;
8816
8817		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8818
8819		if (cdb->byte2 & SRW12_RELADDR)
8820			reladdr = 1;
8821		if (cdb->byte2 & SRW12_FUA)
8822			fua = 1;
8823		if (cdb->byte2 & SRW12_DPO)
8824			dpo = 1;
8825
8826		lba = scsi_8btou64(cdb->addr);
8827		num_blocks = scsi_4btoul(cdb->length);
8828		break;
8829	}
8830	case WRITE_VERIFY_16: {
8831		struct scsi_write_verify_16 *cdb;
8832
8833		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8834
8835		if (cdb->byte2 & SWV_DPO)
8836			dpo = 1;
8837
8838		lba = scsi_8btou64(cdb->addr);
8839		num_blocks = scsi_4btoul(cdb->length);
8840		break;
8841	}
8842	default:
8843		/*
8844		 * We got a command we don't support.  This shouldn't
8845		 * happen, commands should be filtered out above us.
8846		 */
8847		ctl_set_invalid_opcode(ctsio);
8848		ctl_done((union ctl_io *)ctsio);
8849
8850		return (CTL_RETVAL_COMPLETE);
8851		break; /* NOTREACHED */
8852	}
8853
8854	/*
8855	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8856	 * interesting for us, but if RAIDCore is in write-back mode,
8857	 * getting it to do write-through for a particular transaction may
8858	 * not be possible.
8859	 */
8860	/*
8861	 * We don't support relative addressing.  That also requires
8862	 * supporting linked commands, which we don't do.
8863	 */
8864	if (reladdr != 0) {
8865		ctl_set_invalid_field(ctsio,
8866				      /*sks_valid*/ 1,
8867				      /*command*/ 1,
8868				      /*field*/ 1,
8869				      /*bit_valid*/ 1,
8870				      /*bit*/ 0);
8871		ctl_done((union ctl_io *)ctsio);
8872		return (CTL_RETVAL_COMPLETE);
8873	}
8874
8875	/*
8876	 * The first check is to make sure we're in bounds, the second
8877	 * check is to catch wrap-around problems.  If the lba + num blocks
8878	 * is less than the lba, then we've wrapped around and the block
8879	 * range is invalid anyway.
8880	 */
8881	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8882	 || ((lba + num_blocks) < lba)) {
8883		ctl_set_lba_out_of_range(ctsio);
8884		ctl_done((union ctl_io *)ctsio);
8885		return (CTL_RETVAL_COMPLETE);
8886	}
8887
8888	/*
8889	 * According to SBC-3, a transfer length of 0 is not an error.
8890	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8891	 * translates to 256 blocks for those commands.
8892	 */
8893	if (num_blocks == 0) {
8894		ctl_set_success(ctsio);
8895		ctl_done((union ctl_io *)ctsio);
8896		return (CTL_RETVAL_COMPLETE);
8897	}
8898
8899	lbalen = (struct ctl_lba_len_flags *)
8900	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8901	lbalen->lba = lba;
8902	lbalen->len = num_blocks;
8903	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8904
8905	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8906	ctsio->kern_rel_offset = 0;
8907
8908	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8909
8910	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8911
8912	return (retval);
8913}
8914
8915static int
8916ctl_cnw_cont(union ctl_io *io)
8917{
8918	struct ctl_scsiio *ctsio;
8919	struct ctl_lun *lun;
8920	struct ctl_lba_len_flags *lbalen;
8921	int retval;
8922
8923	ctsio = &io->scsiio;
8924	ctsio->io_hdr.status = CTL_STATUS_NONE;
8925	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8926	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8927	lbalen = (struct ctl_lba_len_flags *)
8928	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8929	lbalen->flags = CTL_LLF_WRITE;
8930
8931	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8932	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8933	return (retval);
8934}
8935
8936int
8937ctl_cnw(struct ctl_scsiio *ctsio)
8938{
8939	struct ctl_lun *lun;
8940	struct ctl_lba_len_flags *lbalen;
8941	uint64_t lba;
8942	uint32_t num_blocks;
8943	int fua, dpo;
8944	int retval;
8945
8946	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8947
8948	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8949
8950	fua = 0;
8951	dpo = 0;
8952
8953	retval = CTL_RETVAL_COMPLETE;
8954
8955	switch (ctsio->cdb[0]) {
8956	case COMPARE_AND_WRITE: {
8957		struct scsi_compare_and_write *cdb;
8958
8959		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8960
8961		if (cdb->byte2 & SRW10_FUA)
8962			fua = 1;
8963		if (cdb->byte2 & SRW10_DPO)
8964			dpo = 1;
8965		lba = scsi_8btou64(cdb->addr);
8966		num_blocks = cdb->length;
8967		break;
8968	}
8969	default:
8970		/*
8971		 * We got a command we don't support.  This shouldn't
8972		 * happen, commands should be filtered out above us.
8973		 */
8974		ctl_set_invalid_opcode(ctsio);
8975		ctl_done((union ctl_io *)ctsio);
8976
8977		return (CTL_RETVAL_COMPLETE);
8978		break; /* NOTREACHED */
8979	}
8980
8981	/*
8982	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8983	 * interesting for us, but if RAIDCore is in write-back mode,
8984	 * getting it to do write-through for a particular transaction may
8985	 * not be possible.
8986	 */
8987
8988	/*
8989	 * The first check is to make sure we're in bounds, the second
8990	 * check is to catch wrap-around problems.  If the lba + num blocks
8991	 * is less than the lba, then we've wrapped around and the block
8992	 * range is invalid anyway.
8993	 */
8994	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8995	 || ((lba + num_blocks) < lba)) {
8996		ctl_set_lba_out_of_range(ctsio);
8997		ctl_done((union ctl_io *)ctsio);
8998		return (CTL_RETVAL_COMPLETE);
8999	}
9000
9001	/*
9002	 * According to SBC-3, a transfer length of 0 is not an error.
9003	 */
9004	if (num_blocks == 0) {
9005		ctl_set_success(ctsio);
9006		ctl_done((union ctl_io *)ctsio);
9007		return (CTL_RETVAL_COMPLETE);
9008	}
9009
9010	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9011	ctsio->kern_rel_offset = 0;
9012
9013	/*
9014	 * Set the IO_CONT flag, so that if this I/O gets passed to
9015	 * ctl_data_submit_done(), it'll get passed back to
9016	 * ctl_ctl_cnw_cont() for further processing.
9017	 */
9018	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9019	ctsio->io_cont = ctl_cnw_cont;
9020
9021	lbalen = (struct ctl_lba_len_flags *)
9022	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9023	lbalen->lba = lba;
9024	lbalen->len = num_blocks;
9025	lbalen->flags = CTL_LLF_COMPARE;
9026
9027	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9028	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9029	return (retval);
9030}
9031
9032int
9033ctl_verify(struct ctl_scsiio *ctsio)
9034{
9035	struct ctl_lun *lun;
9036	struct ctl_lba_len_flags *lbalen;
9037	uint64_t lba;
9038	uint32_t num_blocks;
9039	int bytchk, dpo;
9040	int retval;
9041
9042	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9043
9044	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9045
9046	bytchk = 0;
9047	dpo = 0;
9048	retval = CTL_RETVAL_COMPLETE;
9049
9050	switch (ctsio->cdb[0]) {
9051	case VERIFY_10: {
9052		struct scsi_verify_10 *cdb;
9053
9054		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9055		if (cdb->byte2 & SVFY_BYTCHK)
9056			bytchk = 1;
9057		if (cdb->byte2 & SVFY_DPO)
9058			dpo = 1;
9059		lba = scsi_4btoul(cdb->addr);
9060		num_blocks = scsi_2btoul(cdb->length);
9061		break;
9062	}
9063	case VERIFY_12: {
9064		struct scsi_verify_12 *cdb;
9065
9066		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9067		if (cdb->byte2 & SVFY_BYTCHK)
9068			bytchk = 1;
9069		if (cdb->byte2 & SVFY_DPO)
9070			dpo = 1;
9071		lba = scsi_4btoul(cdb->addr);
9072		num_blocks = scsi_4btoul(cdb->length);
9073		break;
9074	}
9075	case VERIFY_16: {
9076		struct scsi_rw_16 *cdb;
9077
9078		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9079		if (cdb->byte2 & SVFY_BYTCHK)
9080			bytchk = 1;
9081		if (cdb->byte2 & SVFY_DPO)
9082			dpo = 1;
9083		lba = scsi_8btou64(cdb->addr);
9084		num_blocks = scsi_4btoul(cdb->length);
9085		break;
9086	}
9087	default:
9088		/*
9089		 * We got a command we don't support.  This shouldn't
9090		 * happen, commands should be filtered out above us.
9091		 */
9092		ctl_set_invalid_opcode(ctsio);
9093		ctl_done((union ctl_io *)ctsio);
9094		return (CTL_RETVAL_COMPLETE);
9095	}
9096
9097	/*
9098	 * The first check is to make sure we're in bounds, the second
9099	 * check is to catch wrap-around problems.  If the lba + num blocks
9100	 * is less than the lba, then we've wrapped around and the block
9101	 * range is invalid anyway.
9102	 */
9103	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9104	 || ((lba + num_blocks) < lba)) {
9105		ctl_set_lba_out_of_range(ctsio);
9106		ctl_done((union ctl_io *)ctsio);
9107		return (CTL_RETVAL_COMPLETE);
9108	}
9109
9110	/*
9111	 * According to SBC-3, a transfer length of 0 is not an error.
9112	 */
9113	if (num_blocks == 0) {
9114		ctl_set_success(ctsio);
9115		ctl_done((union ctl_io *)ctsio);
9116		return (CTL_RETVAL_COMPLETE);
9117	}
9118
9119	lbalen = (struct ctl_lba_len_flags *)
9120	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9121	lbalen->lba = lba;
9122	lbalen->len = num_blocks;
9123	if (bytchk) {
9124		lbalen->flags = CTL_LLF_COMPARE;
9125		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9126	} else {
9127		lbalen->flags = CTL_LLF_VERIFY;
9128		ctsio->kern_total_len = 0;
9129	}
9130	ctsio->kern_rel_offset = 0;
9131
9132	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9133	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9134	return (retval);
9135}
9136
9137int
9138ctl_report_luns(struct ctl_scsiio *ctsio)
9139{
9140	struct scsi_report_luns *cdb;
9141	struct scsi_report_luns_data *lun_data;
9142	struct ctl_lun *lun, *request_lun;
9143	int num_luns, retval;
9144	uint32_t alloc_len, lun_datalen;
9145	int num_filled, well_known;
9146	uint32_t initidx, targ_lun_id, lun_id;
9147
9148	retval = CTL_RETVAL_COMPLETE;
9149	well_known = 0;
9150
9151	cdb = (struct scsi_report_luns *)ctsio->cdb;
9152
9153	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9154
9155	mtx_lock(&control_softc->ctl_lock);
9156	num_luns = control_softc->num_luns;
9157	mtx_unlock(&control_softc->ctl_lock);
9158
9159	switch (cdb->select_report) {
9160	case RPL_REPORT_DEFAULT:
9161	case RPL_REPORT_ALL:
9162		break;
9163	case RPL_REPORT_WELLKNOWN:
9164		well_known = 1;
9165		num_luns = 0;
9166		break;
9167	default:
9168		ctl_set_invalid_field(ctsio,
9169				      /*sks_valid*/ 1,
9170				      /*command*/ 1,
9171				      /*field*/ 2,
9172				      /*bit_valid*/ 0,
9173				      /*bit*/ 0);
9174		ctl_done((union ctl_io *)ctsio);
9175		return (retval);
9176		break; /* NOTREACHED */
9177	}
9178
9179	alloc_len = scsi_4btoul(cdb->length);
9180	/*
9181	 * The initiator has to allocate at least 16 bytes for this request,
9182	 * so he can at least get the header and the first LUN.  Otherwise
9183	 * we reject the request (per SPC-3 rev 14, section 6.21).
9184	 */
9185	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9186	    sizeof(struct scsi_report_luns_lundata))) {
9187		ctl_set_invalid_field(ctsio,
9188				      /*sks_valid*/ 1,
9189				      /*command*/ 1,
9190				      /*field*/ 6,
9191				      /*bit_valid*/ 0,
9192				      /*bit*/ 0);
9193		ctl_done((union ctl_io *)ctsio);
9194		return (retval);
9195	}
9196
9197	request_lun = (struct ctl_lun *)
9198		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9199
9200	lun_datalen = sizeof(*lun_data) +
9201		(num_luns * sizeof(struct scsi_report_luns_lundata));
9202
9203	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9204	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9205	ctsio->kern_sg_entries = 0;
9206
9207	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9208
9209	mtx_lock(&control_softc->ctl_lock);
9210	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9211		lun_id = targ_lun_id;
9212		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9213			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9214		if (lun_id >= CTL_MAX_LUNS)
9215			continue;
9216		lun = control_softc->ctl_luns[lun_id];
9217		if (lun == NULL)
9218			continue;
9219
9220		if (targ_lun_id <= 0xff) {
9221			/*
9222			 * Peripheral addressing method, bus number 0.
9223			 */
9224			lun_data->luns[num_filled].lundata[0] =
9225				RPL_LUNDATA_ATYP_PERIPH;
9226			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9227			num_filled++;
9228		} else if (targ_lun_id <= 0x3fff) {
9229			/*
9230			 * Flat addressing method.
9231			 */
9232			lun_data->luns[num_filled].lundata[0] =
9233				RPL_LUNDATA_ATYP_FLAT |
9234				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9235#ifdef OLDCTLHEADERS
9236				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9237				(targ_lun_id & SRLD_BUS_LUN_MASK);
9238#endif
9239			lun_data->luns[num_filled].lundata[1] =
9240#ifdef OLDCTLHEADERS
9241				targ_lun_id >> SRLD_BUS_LUN_BITS;
9242#endif
9243				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9244			num_filled++;
9245		} else {
9246			printf("ctl_report_luns: bogus LUN number %jd, "
9247			       "skipping\n", (intmax_t)targ_lun_id);
9248		}
9249		/*
9250		 * According to SPC-3, rev 14 section 6.21:
9251		 *
9252		 * "The execution of a REPORT LUNS command to any valid and
9253		 * installed logical unit shall clear the REPORTED LUNS DATA
9254		 * HAS CHANGED unit attention condition for all logical
9255		 * units of that target with respect to the requesting
9256		 * initiator. A valid and installed logical unit is one
9257		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9258		 * INQUIRY data (see 6.4.2)."
9259		 *
9260		 * If request_lun is NULL, the LUN this report luns command
9261		 * was issued to is either disabled or doesn't exist. In that
9262		 * case, we shouldn't clear any pending lun change unit
9263		 * attention.
9264		 */
9265		if (request_lun != NULL)
9266			lun->pending_sense[initidx].ua_pending &=
9267				~CTL_UA_LUN_CHANGE;
9268	}
9269	mtx_unlock(&control_softc->ctl_lock);
9270
9271	/*
9272	 * It's quite possible that we've returned fewer LUNs than we allocated
9273	 * space for.  Trim it.
9274	 */
9275	lun_datalen = sizeof(*lun_data) +
9276		(num_filled * sizeof(struct scsi_report_luns_lundata));
9277
9278	if (lun_datalen < alloc_len) {
9279		ctsio->residual = alloc_len - lun_datalen;
9280		ctsio->kern_data_len = lun_datalen;
9281		ctsio->kern_total_len = lun_datalen;
9282	} else {
9283		ctsio->residual = 0;
9284		ctsio->kern_data_len = alloc_len;
9285		ctsio->kern_total_len = alloc_len;
9286	}
9287	ctsio->kern_data_resid = 0;
9288	ctsio->kern_rel_offset = 0;
9289	ctsio->kern_sg_entries = 0;
9290
9291	/*
9292	 * We set this to the actual data length, regardless of how much
9293	 * space we actually have to return results.  If the user looks at
9294	 * this value, he'll know whether or not he allocated enough space
9295	 * and reissue the command if necessary.  We don't support well
9296	 * known logical units, so if the user asks for that, return none.
9297	 */
9298	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9299
9300	/*
9301	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9302	 * this request.
9303	 */
9304	ctsio->scsi_status = SCSI_STATUS_OK;
9305
9306	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9307	ctsio->be_move_done = ctl_config_move_done;
9308	ctl_datamove((union ctl_io *)ctsio);
9309
9310	return (retval);
9311}
9312
9313int
9314ctl_request_sense(struct ctl_scsiio *ctsio)
9315{
9316	struct scsi_request_sense *cdb;
9317	struct scsi_sense_data *sense_ptr;
9318	struct ctl_lun *lun;
9319	uint32_t initidx;
9320	int have_error;
9321	scsi_sense_data_type sense_format;
9322
9323	cdb = (struct scsi_request_sense *)ctsio->cdb;
9324
9325	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9326
9327	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9328
9329	/*
9330	 * Determine which sense format the user wants.
9331	 */
9332	if (cdb->byte2 & SRS_DESC)
9333		sense_format = SSD_TYPE_DESC;
9334	else
9335		sense_format = SSD_TYPE_FIXED;
9336
9337	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9338	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9339	ctsio->kern_sg_entries = 0;
9340
9341	/*
9342	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9343	 * larger than the largest allowed value for the length field in the
9344	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9345	 */
9346	ctsio->residual = 0;
9347	ctsio->kern_data_len = cdb->length;
9348	ctsio->kern_total_len = cdb->length;
9349
9350	ctsio->kern_data_resid = 0;
9351	ctsio->kern_rel_offset = 0;
9352	ctsio->kern_sg_entries = 0;
9353
9354	/*
9355	 * If we don't have a LUN, we don't have any pending sense.
9356	 */
9357	if (lun == NULL)
9358		goto no_sense;
9359
9360	have_error = 0;
9361	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9362	/*
9363	 * Check for pending sense, and then for pending unit attentions.
9364	 * Pending sense gets returned first, then pending unit attentions.
9365	 */
9366	mtx_lock(&lun->ctl_softc->ctl_lock);
9367	if (ctl_is_set(lun->have_ca, initidx)) {
9368		scsi_sense_data_type stored_format;
9369
9370		/*
9371		 * Check to see which sense format was used for the stored
9372		 * sense data.
9373		 */
9374		stored_format = scsi_sense_type(
9375		    &lun->pending_sense[initidx].sense);
9376
9377		/*
9378		 * If the user requested a different sense format than the
9379		 * one we stored, then we need to convert it to the other
9380		 * format.  If we're going from descriptor to fixed format
9381		 * sense data, we may lose things in translation, depending
9382		 * on what options were used.
9383		 *
9384		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9385		 * for some reason we'll just copy it out as-is.
9386		 */
9387		if ((stored_format == SSD_TYPE_FIXED)
9388		 && (sense_format == SSD_TYPE_DESC))
9389			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9390			    &lun->pending_sense[initidx].sense,
9391			    (struct scsi_sense_data_desc *)sense_ptr);
9392		else if ((stored_format == SSD_TYPE_DESC)
9393		      && (sense_format == SSD_TYPE_FIXED))
9394			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9395			    &lun->pending_sense[initidx].sense,
9396			    (struct scsi_sense_data_fixed *)sense_ptr);
9397		else
9398			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9399			       ctl_min(sizeof(*sense_ptr),
9400			       sizeof(lun->pending_sense[initidx].sense)));
9401
9402		ctl_clear_mask(lun->have_ca, initidx);
9403		have_error = 1;
9404	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9405		ctl_ua_type ua_type;
9406
9407		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9408				       sense_ptr, sense_format);
9409		if (ua_type != CTL_UA_NONE) {
9410			have_error = 1;
9411			/* We're reporting this UA, so clear it */
9412			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9413		}
9414	}
9415	mtx_unlock(&lun->ctl_softc->ctl_lock);
9416
9417	/*
9418	 * We already have a pending error, return it.
9419	 */
9420	if (have_error != 0) {
9421		/*
9422		 * We report the SCSI status as OK, since the status of the
9423		 * request sense command itself is OK.
9424		 */
9425		ctsio->scsi_status = SCSI_STATUS_OK;
9426
9427		/*
9428		 * We report 0 for the sense length, because we aren't doing
9429		 * autosense in this case.  We're reporting sense as
9430		 * parameter data.
9431		 */
9432		ctsio->sense_len = 0;
9433		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9434		ctsio->be_move_done = ctl_config_move_done;
9435		ctl_datamove((union ctl_io *)ctsio);
9436
9437		return (CTL_RETVAL_COMPLETE);
9438	}
9439
9440no_sense:
9441
9442	/*
9443	 * No sense information to report, so we report that everything is
9444	 * okay.
9445	 */
9446	ctl_set_sense_data(sense_ptr,
9447			   lun,
9448			   sense_format,
9449			   /*current_error*/ 1,
9450			   /*sense_key*/ SSD_KEY_NO_SENSE,
9451			   /*asc*/ 0x00,
9452			   /*ascq*/ 0x00,
9453			   SSD_ELEM_NONE);
9454
9455	ctsio->scsi_status = SCSI_STATUS_OK;
9456
9457	/*
9458	 * We report 0 for the sense length, because we aren't doing
9459	 * autosense in this case.  We're reporting sense as parameter data.
9460	 */
9461	ctsio->sense_len = 0;
9462	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9463	ctsio->be_move_done = ctl_config_move_done;
9464	ctl_datamove((union ctl_io *)ctsio);
9465
9466	return (CTL_RETVAL_COMPLETE);
9467}
9468
9469int
9470ctl_tur(struct ctl_scsiio *ctsio)
9471{
9472	struct ctl_lun *lun;
9473
9474	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9475
9476	CTL_DEBUG_PRINT(("ctl_tur\n"));
9477
9478	if (lun == NULL)
9479		return (EINVAL);
9480
9481	ctsio->scsi_status = SCSI_STATUS_OK;
9482	ctsio->io_hdr.status = CTL_SUCCESS;
9483
9484	ctl_done((union ctl_io *)ctsio);
9485
9486	return (CTL_RETVAL_COMPLETE);
9487}
9488
9489#ifdef notyet
9490static int
9491ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9492{
9493
9494}
9495#endif
9496
9497static int
9498ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9499{
9500	struct scsi_vpd_supported_pages *pages;
9501	int sup_page_size;
9502	struct ctl_lun *lun;
9503
9504	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9505
9506	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9507	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9508	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9509	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9510	ctsio->kern_sg_entries = 0;
9511
9512	if (sup_page_size < alloc_len) {
9513		ctsio->residual = alloc_len - sup_page_size;
9514		ctsio->kern_data_len = sup_page_size;
9515		ctsio->kern_total_len = sup_page_size;
9516	} else {
9517		ctsio->residual = 0;
9518		ctsio->kern_data_len = alloc_len;
9519		ctsio->kern_total_len = alloc_len;
9520	}
9521	ctsio->kern_data_resid = 0;
9522	ctsio->kern_rel_offset = 0;
9523	ctsio->kern_sg_entries = 0;
9524
9525	/*
9526	 * The control device is always connected.  The disk device, on the
9527	 * other hand, may not be online all the time.  Need to change this
9528	 * to figure out whether the disk device is actually online or not.
9529	 */
9530	if (lun != NULL)
9531		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9532				lun->be_lun->lun_type;
9533	else
9534		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9535
9536	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9537	/* Supported VPD pages */
9538	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9539	/* Serial Number */
9540	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9541	/* Device Identification */
9542	pages->page_list[2] = SVPD_DEVICE_ID;
9543	/* Block limits */
9544	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9545	/* Logical Block Provisioning */
9546	pages->page_list[4] = SVPD_LBP;
9547
9548	ctsio->scsi_status = SCSI_STATUS_OK;
9549
9550	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9551	ctsio->be_move_done = ctl_config_move_done;
9552	ctl_datamove((union ctl_io *)ctsio);
9553
9554	return (CTL_RETVAL_COMPLETE);
9555}
9556
9557static int
9558ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9559{
9560	struct scsi_vpd_unit_serial_number *sn_ptr;
9561	struct ctl_lun *lun;
9562#ifndef CTL_USE_BACKEND_SN
9563	char tmpstr[32];
9564#endif
9565
9566	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9567
9568	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9569	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9570	ctsio->kern_sg_entries = 0;
9571
9572	if (sizeof(*sn_ptr) < alloc_len) {
9573		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9574		ctsio->kern_data_len = sizeof(*sn_ptr);
9575		ctsio->kern_total_len = sizeof(*sn_ptr);
9576	} else {
9577		ctsio->residual = 0;
9578		ctsio->kern_data_len = alloc_len;
9579		ctsio->kern_total_len = alloc_len;
9580	}
9581	ctsio->kern_data_resid = 0;
9582	ctsio->kern_rel_offset = 0;
9583	ctsio->kern_sg_entries = 0;
9584
9585	/*
9586	 * The control device is always connected.  The disk device, on the
9587	 * other hand, may not be online all the time.  Need to change this
9588	 * to figure out whether the disk device is actually online or not.
9589	 */
9590	if (lun != NULL)
9591		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9592				  lun->be_lun->lun_type;
9593	else
9594		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9595
9596	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9597	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9598#ifdef CTL_USE_BACKEND_SN
9599	/*
9600	 * If we don't have a LUN, we just leave the serial number as
9601	 * all spaces.
9602	 */
9603	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9604	if (lun != NULL) {
9605		strncpy((char *)sn_ptr->serial_num,
9606			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9607	}
9608#else
9609	/*
9610	 * Note that we're using a non-unique serial number here,
9611	 */
9612	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9613	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9614	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9615		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9616#endif
9617	ctsio->scsi_status = SCSI_STATUS_OK;
9618
9619	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9620	ctsio->be_move_done = ctl_config_move_done;
9621	ctl_datamove((union ctl_io *)ctsio);
9622
9623	return (CTL_RETVAL_COMPLETE);
9624}
9625
9626
9627static int
9628ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9629{
9630	struct scsi_vpd_device_id *devid_ptr;
9631	struct scsi_vpd_id_descriptor *desc, *desc1;
9632	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9633	struct scsi_vpd_id_t10 *t10id;
9634	struct ctl_softc *ctl_softc;
9635	struct ctl_lun *lun;
9636	struct ctl_frontend *fe;
9637	char *val;
9638#ifndef CTL_USE_BACKEND_SN
9639	char tmpstr[32];
9640#endif /* CTL_USE_BACKEND_SN */
9641	int devid_len;
9642
9643	ctl_softc = control_softc;
9644
9645	mtx_lock(&ctl_softc->ctl_lock);
9646	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9647	mtx_unlock(&ctl_softc->ctl_lock);
9648
9649	if (fe->devid != NULL)
9650		return ((fe->devid)(ctsio, alloc_len));
9651
9652	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9653
9654	devid_len = sizeof(struct scsi_vpd_device_id) +
9655		sizeof(struct scsi_vpd_id_descriptor) +
9656		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9657		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9658		sizeof(struct scsi_vpd_id_descriptor) +
9659		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9660		sizeof(struct scsi_vpd_id_descriptor) +
9661		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9662
9663	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9664	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9665	ctsio->kern_sg_entries = 0;
9666
9667	if (devid_len < alloc_len) {
9668		ctsio->residual = alloc_len - devid_len;
9669		ctsio->kern_data_len = devid_len;
9670		ctsio->kern_total_len = devid_len;
9671	} else {
9672		ctsio->residual = 0;
9673		ctsio->kern_data_len = alloc_len;
9674		ctsio->kern_total_len = alloc_len;
9675	}
9676	ctsio->kern_data_resid = 0;
9677	ctsio->kern_rel_offset = 0;
9678	ctsio->kern_sg_entries = 0;
9679
9680	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9681	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9682	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9683		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9684	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9685	          CTL_WWPN_LEN);
9686	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9687	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9688
9689	/*
9690	 * The control device is always connected.  The disk device, on the
9691	 * other hand, may not be online all the time.
9692	 */
9693	if (lun != NULL)
9694		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9695				     lun->be_lun->lun_type;
9696	else
9697		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9698
9699	devid_ptr->page_code = SVPD_DEVICE_ID;
9700
9701	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9702
9703	mtx_lock(&ctl_softc->ctl_lock);
9704
9705	/*
9706	 * For Fibre channel,
9707	 */
9708	if (fe->port_type == CTL_PORT_FC)
9709	{
9710		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9711				      SVPD_ID_CODESET_ASCII;
9712        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9713		              SVPD_ID_CODESET_BINARY;
9714	}
9715	else
9716	{
9717		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9718				      SVPD_ID_CODESET_ASCII;
9719        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9720		              SVPD_ID_CODESET_BINARY;
9721	}
9722	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9723	mtx_unlock(&ctl_softc->ctl_lock);
9724
9725	/*
9726	 * We're using a LUN association here.  i.e., this device ID is a
9727	 * per-LUN identifier.
9728	 */
9729	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9730	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9731	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9732		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9733	} else {
9734		memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9735		strncpy(t10id->vendor, val,
9736		    min(sizeof(t10id->vendor), strlen(val)));
9737	}
9738
9739	/*
9740	 * desc1 is for the WWPN which is a port asscociation.
9741	 */
9742	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9743	desc1->length = CTL_WWPN_LEN;
9744	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9745	/* For testing just create the WWPN */
9746#if 0
9747	ddb_GetWWNN((char *)desc1->identifier);
9748
9749	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9750	/* This is so Copancontrol will return something sane */
9751	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9752	    ctsio->io_hdr.nexus.targ_port!=8)
9753		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9754	else
9755		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9756#endif
9757
9758	be64enc(desc1->identifier, fe->wwpn);
9759
9760	/*
9761	 * desc2 is for the Relative Target Port(type 4h) identifier
9762	 */
9763	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9764	                 | SVPD_ID_TYPE_RELTARG;
9765	desc2->length = 4;
9766//#if 0
9767	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9768	/* This is so Copancontrol will return something sane */
9769	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9770	    ctsio->io_hdr.nexus.targ_port!=8)
9771		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9772	else
9773	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9774//#endif
9775
9776	/*
9777	 * desc3 is for the Target Port Group(type 5h) identifier
9778	 */
9779	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9780	                 | SVPD_ID_TYPE_TPORTGRP;
9781	desc3->length = 4;
9782	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9783		desc3->identifier[3] = 1;
9784	else
9785		desc3->identifier[3] = 2;
9786
9787#ifdef CTL_USE_BACKEND_SN
9788	/*
9789	 * If we've actually got a backend, copy the device id from the
9790	 * per-LUN data.  Otherwise, set it to all spaces.
9791	 */
9792	if (lun != NULL) {
9793		/*
9794		 * Copy the backend's LUN ID.
9795		 */
9796		strncpy((char *)t10id->vendor_spec_id,
9797			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9798	} else {
9799		/*
9800		 * No backend, set this to spaces.
9801		 */
9802		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9803	}
9804#else
9805	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9806		 (lun != NULL) ?  (int)lun->lun : 0);
9807	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9808		sizeof(tmpstr)));
9809#endif
9810
9811	ctsio->scsi_status = SCSI_STATUS_OK;
9812
9813	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9814	ctsio->be_move_done = ctl_config_move_done;
9815	ctl_datamove((union ctl_io *)ctsio);
9816
9817	return (CTL_RETVAL_COMPLETE);
9818}
9819
9820static int
9821ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9822{
9823	struct scsi_vpd_block_limits *bl_ptr;
9824	struct ctl_lun *lun;
9825	int bs;
9826
9827	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9828	bs = lun->be_lun->blocksize;
9829
9830	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9831	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9832	ctsio->kern_sg_entries = 0;
9833
9834	if (sizeof(*bl_ptr) < alloc_len) {
9835		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9836		ctsio->kern_data_len = sizeof(*bl_ptr);
9837		ctsio->kern_total_len = sizeof(*bl_ptr);
9838	} else {
9839		ctsio->residual = 0;
9840		ctsio->kern_data_len = alloc_len;
9841		ctsio->kern_total_len = alloc_len;
9842	}
9843	ctsio->kern_data_resid = 0;
9844	ctsio->kern_rel_offset = 0;
9845	ctsio->kern_sg_entries = 0;
9846
9847	/*
9848	 * The control device is always connected.  The disk device, on the
9849	 * other hand, may not be online all the time.  Need to change this
9850	 * to figure out whether the disk device is actually online or not.
9851	 */
9852	if (lun != NULL)
9853		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9854				  lun->be_lun->lun_type;
9855	else
9856		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9857
9858	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9859	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9860	bl_ptr->max_cmp_write_len = 0xff;
9861	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9862	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9863	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9864		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9865		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9866	}
9867	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9868
9869	ctsio->scsi_status = SCSI_STATUS_OK;
9870	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9871	ctsio->be_move_done = ctl_config_move_done;
9872	ctl_datamove((union ctl_io *)ctsio);
9873
9874	return (CTL_RETVAL_COMPLETE);
9875}
9876
9877static int
9878ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9879{
9880	struct scsi_vpd_logical_block_prov *lbp_ptr;
9881	struct ctl_lun *lun;
9882	int bs;
9883
9884	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9885	bs = lun->be_lun->blocksize;
9886
9887	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9888	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9889	ctsio->kern_sg_entries = 0;
9890
9891	if (sizeof(*lbp_ptr) < alloc_len) {
9892		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9893		ctsio->kern_data_len = sizeof(*lbp_ptr);
9894		ctsio->kern_total_len = sizeof(*lbp_ptr);
9895	} else {
9896		ctsio->residual = 0;
9897		ctsio->kern_data_len = alloc_len;
9898		ctsio->kern_total_len = alloc_len;
9899	}
9900	ctsio->kern_data_resid = 0;
9901	ctsio->kern_rel_offset = 0;
9902	ctsio->kern_sg_entries = 0;
9903
9904	/*
9905	 * The control device is always connected.  The disk device, on the
9906	 * other hand, may not be online all the time.  Need to change this
9907	 * to figure out whether the disk device is actually online or not.
9908	 */
9909	if (lun != NULL)
9910		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9911				  lun->be_lun->lun_type;
9912	else
9913		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9914
9915	lbp_ptr->page_code = SVPD_LBP;
9916	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9917		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9918
9919	ctsio->scsi_status = SCSI_STATUS_OK;
9920	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9921	ctsio->be_move_done = ctl_config_move_done;
9922	ctl_datamove((union ctl_io *)ctsio);
9923
9924	return (CTL_RETVAL_COMPLETE);
9925}
9926
9927static int
9928ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9929{
9930	struct scsi_inquiry *cdb;
9931	struct ctl_lun *lun;
9932	int alloc_len, retval;
9933
9934	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9935	cdb = (struct scsi_inquiry *)ctsio->cdb;
9936
9937	retval = CTL_RETVAL_COMPLETE;
9938
9939	alloc_len = scsi_2btoul(cdb->length);
9940
9941	switch (cdb->page_code) {
9942	case SVPD_SUPPORTED_PAGES:
9943		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9944		break;
9945	case SVPD_UNIT_SERIAL_NUMBER:
9946		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9947		break;
9948	case SVPD_DEVICE_ID:
9949		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9950		break;
9951	case SVPD_BLOCK_LIMITS:
9952		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9953		break;
9954	case SVPD_LBP:
9955		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9956		break;
9957	default:
9958		ctl_set_invalid_field(ctsio,
9959				      /*sks_valid*/ 1,
9960				      /*command*/ 1,
9961				      /*field*/ 2,
9962				      /*bit_valid*/ 0,
9963				      /*bit*/ 0);
9964		ctl_done((union ctl_io *)ctsio);
9965		retval = CTL_RETVAL_COMPLETE;
9966		break;
9967	}
9968
9969	return (retval);
9970}
9971
9972static int
9973ctl_inquiry_std(struct ctl_scsiio *ctsio)
9974{
9975	struct scsi_inquiry_data *inq_ptr;
9976	struct scsi_inquiry *cdb;
9977	struct ctl_softc *ctl_softc;
9978	struct ctl_lun *lun;
9979	char *val;
9980	uint32_t alloc_len;
9981	int is_fc;
9982
9983	ctl_softc = control_softc;
9984
9985	/*
9986	 * Figure out whether we're talking to a Fibre Channel port or not.
9987	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9988	 * SCSI front ends.
9989	 */
9990	mtx_lock(&ctl_softc->ctl_lock);
9991	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9992	    CTL_PORT_FC)
9993		is_fc = 0;
9994	else
9995		is_fc = 1;
9996	mtx_unlock(&ctl_softc->ctl_lock);
9997
9998	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9999	cdb = (struct scsi_inquiry *)ctsio->cdb;
10000	alloc_len = scsi_2btoul(cdb->length);
10001
10002	/*
10003	 * We malloc the full inquiry data size here and fill it
10004	 * in.  If the user only asks for less, we'll give him
10005	 * that much.
10006	 */
10007	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
10008	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10009	ctsio->kern_sg_entries = 0;
10010	ctsio->kern_data_resid = 0;
10011	ctsio->kern_rel_offset = 0;
10012
10013	if (sizeof(*inq_ptr) < alloc_len) {
10014		ctsio->residual = alloc_len - sizeof(*inq_ptr);
10015		ctsio->kern_data_len = sizeof(*inq_ptr);
10016		ctsio->kern_total_len = sizeof(*inq_ptr);
10017	} else {
10018		ctsio->residual = 0;
10019		ctsio->kern_data_len = alloc_len;
10020		ctsio->kern_total_len = alloc_len;
10021	}
10022
10023	/*
10024	 * If we have a LUN configured, report it as connected.  Otherwise,
10025	 * report that it is offline or no device is supported, depending
10026	 * on the value of inquiry_pq_no_lun.
10027	 *
10028	 * According to the spec (SPC-4 r34), the peripheral qualifier
10029	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10030	 *
10031	 * "A peripheral device having the specified peripheral device type
10032	 * is not connected to this logical unit. However, the device
10033	 * server is capable of supporting the specified peripheral device
10034	 * type on this logical unit."
10035	 *
10036	 * According to the same spec, the peripheral qualifier
10037	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10038	 *
10039	 * "The device server is not capable of supporting a peripheral
10040	 * device on this logical unit. For this peripheral qualifier the
10041	 * peripheral device type shall be set to 1Fh. All other peripheral
10042	 * device type values are reserved for this peripheral qualifier."
10043	 *
10044	 * Given the text, it would seem that we probably want to report that
10045	 * the LUN is offline here.  There is no LUN connected, but we can
10046	 * support a LUN at the given LUN number.
10047	 *
10048	 * In the real world, though, it sounds like things are a little
10049	 * different:
10050	 *
10051	 * - Linux, when presented with a LUN with the offline peripheral
10052	 *   qualifier, will create an sg driver instance for it.  So when
10053	 *   you attach it to CTL, you wind up with a ton of sg driver
10054	 *   instances.  (One for every LUN that Linux bothered to probe.)
10055	 *   Linux does this despite the fact that it issues a REPORT LUNs
10056	 *   to LUN 0 to get the inventory of supported LUNs.
10057	 *
10058	 * - There is other anecdotal evidence (from Emulex folks) about
10059	 *   arrays that use the offline peripheral qualifier for LUNs that
10060	 *   are on the "passive" path in an active/passive array.
10061	 *
10062	 * So the solution is provide a hopefully reasonable default
10063	 * (return bad/no LUN) and allow the user to change the behavior
10064	 * with a tunable/sysctl variable.
10065	 */
10066	if (lun != NULL)
10067		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10068				  lun->be_lun->lun_type;
10069	else if (ctl_softc->inquiry_pq_no_lun == 0)
10070		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10071	else
10072		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10073
10074	/* RMB in byte 2 is 0 */
10075	inq_ptr->version = SCSI_REV_SPC3;
10076
10077	/*
10078	 * According to SAM-3, even if a device only supports a single
10079	 * level of LUN addressing, it should still set the HISUP bit:
10080	 *
10081	 * 4.9.1 Logical unit numbers overview
10082	 *
10083	 * All logical unit number formats described in this standard are
10084	 * hierarchical in structure even when only a single level in that
10085	 * hierarchy is used. The HISUP bit shall be set to one in the
10086	 * standard INQUIRY data (see SPC-2) when any logical unit number
10087	 * format described in this standard is used.  Non-hierarchical
10088	 * formats are outside the scope of this standard.
10089	 *
10090	 * Therefore we set the HiSup bit here.
10091	 *
10092	 * The reponse format is 2, per SPC-3.
10093	 */
10094	inq_ptr->response_format = SID_HiSup | 2;
10095
10096	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10097	CTL_DEBUG_PRINT(("additional_length = %d\n",
10098			 inq_ptr->additional_length));
10099
10100	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10101	/* 16 bit addressing */
10102	if (is_fc == 0)
10103		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10104	/* XXX set the SID_MultiP bit here if we're actually going to
10105	   respond on multiple ports */
10106	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10107
10108	/* 16 bit data bus, synchronous transfers */
10109	/* XXX these flags don't apply for FC */
10110	if (is_fc == 0)
10111		inq_ptr->flags = SID_WBus16 | SID_Sync;
10112	/*
10113	 * XXX KDM do we want to support tagged queueing on the control
10114	 * device at all?
10115	 */
10116	if ((lun == NULL)
10117	 || (lun->be_lun->lun_type != T_PROCESSOR))
10118		inq_ptr->flags |= SID_CmdQue;
10119	/*
10120	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10121	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10122	 * name and 4 bytes for the revision.
10123	 */
10124	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
10125		strcpy(inq_ptr->vendor, CTL_VENDOR);
10126	} else {
10127		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10128		strncpy(inq_ptr->vendor, val,
10129		    min(sizeof(inq_ptr->vendor), strlen(val)));
10130	}
10131	if (lun == NULL) {
10132		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10133	} else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
10134		switch (lun->be_lun->lun_type) {
10135		case T_DIRECT:
10136			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10137			break;
10138		case T_PROCESSOR:
10139			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10140			break;
10141		default:
10142			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10143			break;
10144		}
10145	} else {
10146		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10147		strncpy(inq_ptr->product, val,
10148		    min(sizeof(inq_ptr->product), strlen(val)));
10149	}
10150
10151	/*
10152	 * XXX make this a macro somewhere so it automatically gets
10153	 * incremented when we make changes.
10154	 */
10155	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
10156		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10157	} else {
10158		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10159		strncpy(inq_ptr->revision, val,
10160		    min(sizeof(inq_ptr->revision), strlen(val)));
10161	}
10162
10163	/*
10164	 * For parallel SCSI, we support double transition and single
10165	 * transition clocking.  We also support QAS (Quick Arbitration
10166	 * and Selection) and Information Unit transfers on both the
10167	 * control and array devices.
10168	 */
10169	if (is_fc == 0)
10170		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10171				    SID_SPI_IUS;
10172
10173	/* SAM-3 */
10174	scsi_ulto2b(0x0060, inq_ptr->version1);
10175	/* SPC-3 (no version claimed) XXX should we claim a version? */
10176	scsi_ulto2b(0x0300, inq_ptr->version2);
10177	if (is_fc) {
10178		/* FCP-2 ANSI INCITS.350:2003 */
10179		scsi_ulto2b(0x0917, inq_ptr->version3);
10180	} else {
10181		/* SPI-4 ANSI INCITS.362:200x */
10182		scsi_ulto2b(0x0B56, inq_ptr->version3);
10183	}
10184
10185	if (lun == NULL) {
10186		/* SBC-2 (no version claimed) XXX should we claim a version? */
10187		scsi_ulto2b(0x0320, inq_ptr->version4);
10188	} else {
10189		switch (lun->be_lun->lun_type) {
10190		case T_DIRECT:
10191			/*
10192			 * SBC-2 (no version claimed) XXX should we claim a
10193			 * version?
10194			 */
10195			scsi_ulto2b(0x0320, inq_ptr->version4);
10196			break;
10197		case T_PROCESSOR:
10198		default:
10199			break;
10200		}
10201	}
10202
10203	ctsio->scsi_status = SCSI_STATUS_OK;
10204	if (ctsio->kern_data_len > 0) {
10205		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10206		ctsio->be_move_done = ctl_config_move_done;
10207		ctl_datamove((union ctl_io *)ctsio);
10208	} else {
10209		ctsio->io_hdr.status = CTL_SUCCESS;
10210		ctl_done((union ctl_io *)ctsio);
10211	}
10212
10213	return (CTL_RETVAL_COMPLETE);
10214}
10215
10216int
10217ctl_inquiry(struct ctl_scsiio *ctsio)
10218{
10219	struct scsi_inquiry *cdb;
10220	int retval;
10221
10222	cdb = (struct scsi_inquiry *)ctsio->cdb;
10223
10224	retval = 0;
10225
10226	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10227
10228	/*
10229	 * Right now, we don't support the CmdDt inquiry information.
10230	 * This would be nice to support in the future.  When we do
10231	 * support it, we should change this test so that it checks to make
10232	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10233	 */
10234#ifdef notyet
10235	if (((cdb->byte2 & SI_EVPD)
10236	 && (cdb->byte2 & SI_CMDDT)))
10237#endif
10238	if (cdb->byte2 & SI_CMDDT) {
10239		/*
10240		 * Point to the SI_CMDDT bit.  We might change this
10241		 * when we support SI_CMDDT, but since both bits would be
10242		 * "wrong", this should probably just stay as-is then.
10243		 */
10244		ctl_set_invalid_field(ctsio,
10245				      /*sks_valid*/ 1,
10246				      /*command*/ 1,
10247				      /*field*/ 1,
10248				      /*bit_valid*/ 1,
10249				      /*bit*/ 1);
10250		ctl_done((union ctl_io *)ctsio);
10251		return (CTL_RETVAL_COMPLETE);
10252	}
10253	if (cdb->byte2 & SI_EVPD)
10254		retval = ctl_inquiry_evpd(ctsio);
10255#ifdef notyet
10256	else if (cdb->byte2 & SI_CMDDT)
10257		retval = ctl_inquiry_cmddt(ctsio);
10258#endif
10259	else
10260		retval = ctl_inquiry_std(ctsio);
10261
10262	return (retval);
10263}
10264
10265/*
10266 * For known CDB types, parse the LBA and length.
10267 */
10268static int
10269ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10270{
10271	if (io->io_hdr.io_type != CTL_IO_SCSI)
10272		return (1);
10273
10274	switch (io->scsiio.cdb[0]) {
10275	case COMPARE_AND_WRITE: {
10276		struct scsi_compare_and_write *cdb;
10277
10278		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10279
10280		*lba = scsi_8btou64(cdb->addr);
10281		*len = cdb->length;
10282		break;
10283	}
10284	case READ_6:
10285	case WRITE_6: {
10286		struct scsi_rw_6 *cdb;
10287
10288		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10289
10290		*lba = scsi_3btoul(cdb->addr);
10291		/* only 5 bits are valid in the most significant address byte */
10292		*lba &= 0x1fffff;
10293		*len = cdb->length;
10294		break;
10295	}
10296	case READ_10:
10297	case WRITE_10: {
10298		struct scsi_rw_10 *cdb;
10299
10300		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10301
10302		*lba = scsi_4btoul(cdb->addr);
10303		*len = scsi_2btoul(cdb->length);
10304		break;
10305	}
10306	case WRITE_VERIFY_10: {
10307		struct scsi_write_verify_10 *cdb;
10308
10309		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10310
10311		*lba = scsi_4btoul(cdb->addr);
10312		*len = scsi_2btoul(cdb->length);
10313		break;
10314	}
10315	case READ_12:
10316	case WRITE_12: {
10317		struct scsi_rw_12 *cdb;
10318
10319		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10320
10321		*lba = scsi_4btoul(cdb->addr);
10322		*len = scsi_4btoul(cdb->length);
10323		break;
10324	}
10325	case WRITE_VERIFY_12: {
10326		struct scsi_write_verify_12 *cdb;
10327
10328		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10329
10330		*lba = scsi_4btoul(cdb->addr);
10331		*len = scsi_4btoul(cdb->length);
10332		break;
10333	}
10334	case READ_16:
10335	case WRITE_16: {
10336		struct scsi_rw_16 *cdb;
10337
10338		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10339
10340		*lba = scsi_8btou64(cdb->addr);
10341		*len = scsi_4btoul(cdb->length);
10342		break;
10343	}
10344	case WRITE_VERIFY_16: {
10345		struct scsi_write_verify_16 *cdb;
10346
10347		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10348
10349
10350		*lba = scsi_8btou64(cdb->addr);
10351		*len = scsi_4btoul(cdb->length);
10352		break;
10353	}
10354	case WRITE_SAME_10: {
10355		struct scsi_write_same_10 *cdb;
10356
10357		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10358
10359		*lba = scsi_4btoul(cdb->addr);
10360		*len = scsi_2btoul(cdb->length);
10361		break;
10362	}
10363	case WRITE_SAME_16: {
10364		struct scsi_write_same_16 *cdb;
10365
10366		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10367
10368		*lba = scsi_8btou64(cdb->addr);
10369		*len = scsi_4btoul(cdb->length);
10370		break;
10371	}
10372	case VERIFY_10: {
10373		struct scsi_verify_10 *cdb;
10374
10375		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10376
10377		*lba = scsi_4btoul(cdb->addr);
10378		*len = scsi_2btoul(cdb->length);
10379		break;
10380	}
10381	case VERIFY_12: {
10382		struct scsi_verify_12 *cdb;
10383
10384		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10385
10386		*lba = scsi_4btoul(cdb->addr);
10387		*len = scsi_4btoul(cdb->length);
10388		break;
10389	}
10390	case VERIFY_16: {
10391		struct scsi_verify_16 *cdb;
10392
10393		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10394
10395		*lba = scsi_8btou64(cdb->addr);
10396		*len = scsi_4btoul(cdb->length);
10397		break;
10398	}
10399	default:
10400		return (1);
10401		break; /* NOTREACHED */
10402	}
10403
10404	return (0);
10405}
10406
10407static ctl_action
10408ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10409{
10410	uint64_t endlba1, endlba2;
10411
10412	endlba1 = lba1 + len1 - 1;
10413	endlba2 = lba2 + len2 - 1;
10414
10415	if ((endlba1 < lba2)
10416	 || (endlba2 < lba1))
10417		return (CTL_ACTION_PASS);
10418	else
10419		return (CTL_ACTION_BLOCK);
10420}
10421
10422static ctl_action
10423ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10424{
10425	uint64_t lba1, lba2;
10426	uint32_t len1, len2;
10427	int retval;
10428
10429	retval = ctl_get_lba_len(io1, &lba1, &len1);
10430	if (retval != 0)
10431		return (CTL_ACTION_ERROR);
10432
10433	retval = ctl_get_lba_len(io2, &lba2, &len2);
10434	if (retval != 0)
10435		return (CTL_ACTION_ERROR);
10436
10437	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10438}
10439
10440static ctl_action
10441ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10442{
10443	struct ctl_cmd_entry *pending_entry, *ooa_entry;
10444	ctl_serialize_action *serialize_row;
10445
10446	/*
10447	 * The initiator attempted multiple untagged commands at the same
10448	 * time.  Can't do that.
10449	 */
10450	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10451	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10452	 && ((pending_io->io_hdr.nexus.targ_port ==
10453	      ooa_io->io_hdr.nexus.targ_port)
10454	  && (pending_io->io_hdr.nexus.initid.id ==
10455	      ooa_io->io_hdr.nexus.initid.id))
10456	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10457		return (CTL_ACTION_OVERLAP);
10458
10459	/*
10460	 * The initiator attempted to send multiple tagged commands with
10461	 * the same ID.  (It's fine if different initiators have the same
10462	 * tag ID.)
10463	 *
10464	 * Even if all of those conditions are true, we don't kill the I/O
10465	 * if the command ahead of us has been aborted.  We won't end up
10466	 * sending it to the FETD, and it's perfectly legal to resend a
10467	 * command with the same tag number as long as the previous
10468	 * instance of this tag number has been aborted somehow.
10469	 */
10470	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10471	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10472	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10473	 && ((pending_io->io_hdr.nexus.targ_port ==
10474	      ooa_io->io_hdr.nexus.targ_port)
10475	  && (pending_io->io_hdr.nexus.initid.id ==
10476	      ooa_io->io_hdr.nexus.initid.id))
10477	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10478		return (CTL_ACTION_OVERLAP_TAG);
10479
10480	/*
10481	 * If we get a head of queue tag, SAM-3 says that we should
10482	 * immediately execute it.
10483	 *
10484	 * What happens if this command would normally block for some other
10485	 * reason?  e.g. a request sense with a head of queue tag
10486	 * immediately after a write.  Normally that would block, but this
10487	 * will result in its getting executed immediately...
10488	 *
10489	 * We currently return "pass" instead of "skip", so we'll end up
10490	 * going through the rest of the queue to check for overlapped tags.
10491	 *
10492	 * XXX KDM check for other types of blockage first??
10493	 */
10494	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10495		return (CTL_ACTION_PASS);
10496
10497	/*
10498	 * Ordered tags have to block until all items ahead of them
10499	 * have completed.  If we get called with an ordered tag, we always
10500	 * block, if something else is ahead of us in the queue.
10501	 */
10502	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10503		return (CTL_ACTION_BLOCK);
10504
10505	/*
10506	 * Simple tags get blocked until all head of queue and ordered tags
10507	 * ahead of them have completed.  I'm lumping untagged commands in
10508	 * with simple tags here.  XXX KDM is that the right thing to do?
10509	 */
10510	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10511	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10512	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10513	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10514		return (CTL_ACTION_BLOCK);
10515
10516	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
10517	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
10518
10519	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10520
10521	switch (serialize_row[pending_entry->seridx]) {
10522	case CTL_SER_BLOCK:
10523		return (CTL_ACTION_BLOCK);
10524		break; /* NOTREACHED */
10525	case CTL_SER_EXTENT:
10526		return (ctl_extent_check(pending_io, ooa_io));
10527		break; /* NOTREACHED */
10528	case CTL_SER_PASS:
10529		return (CTL_ACTION_PASS);
10530		break; /* NOTREACHED */
10531	case CTL_SER_SKIP:
10532		return (CTL_ACTION_SKIP);
10533		break;
10534	default:
10535		panic("invalid serialization value %d",
10536		      serialize_row[pending_entry->seridx]);
10537		break; /* NOTREACHED */
10538	}
10539
10540	return (CTL_ACTION_ERROR);
10541}
10542
10543/*
10544 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10545 * Assumptions:
10546 * - pending_io is generally either incoming, or on the blocked queue
10547 * - starting I/O is the I/O we want to start the check with.
10548 */
10549static ctl_action
10550ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10551	      union ctl_io *starting_io)
10552{
10553	union ctl_io *ooa_io;
10554	ctl_action action;
10555
10556	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10557
10558	/*
10559	 * Run back along the OOA queue, starting with the current
10560	 * blocked I/O and going through every I/O before it on the
10561	 * queue.  If starting_io is NULL, we'll just end up returning
10562	 * CTL_ACTION_PASS.
10563	 */
10564	for (ooa_io = starting_io; ooa_io != NULL;
10565	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10566	     ooa_links)){
10567
10568		/*
10569		 * This routine just checks to see whether
10570		 * cur_blocked is blocked by ooa_io, which is ahead
10571		 * of it in the queue.  It doesn't queue/dequeue
10572		 * cur_blocked.
10573		 */
10574		action = ctl_check_for_blockage(pending_io, ooa_io);
10575		switch (action) {
10576		case CTL_ACTION_BLOCK:
10577		case CTL_ACTION_OVERLAP:
10578		case CTL_ACTION_OVERLAP_TAG:
10579		case CTL_ACTION_SKIP:
10580		case CTL_ACTION_ERROR:
10581			return (action);
10582			break; /* NOTREACHED */
10583		case CTL_ACTION_PASS:
10584			break;
10585		default:
10586			panic("invalid action %d", action);
10587			break;  /* NOTREACHED */
10588		}
10589	}
10590
10591	return (CTL_ACTION_PASS);
10592}
10593
10594/*
10595 * Assumptions:
10596 * - An I/O has just completed, and has been removed from the per-LUN OOA
10597 *   queue, so some items on the blocked queue may now be unblocked.
10598 */
10599static int
10600ctl_check_blocked(struct ctl_lun *lun)
10601{
10602	union ctl_io *cur_blocked, *next_blocked;
10603
10604	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10605
10606	/*
10607	 * Run forward from the head of the blocked queue, checking each
10608	 * entry against the I/Os prior to it on the OOA queue to see if
10609	 * there is still any blockage.
10610	 *
10611	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10612	 * with our removing a variable on it while it is traversing the
10613	 * list.
10614	 */
10615	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10616	     cur_blocked != NULL; cur_blocked = next_blocked) {
10617		union ctl_io *prev_ooa;
10618		ctl_action action;
10619
10620		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10621							  blocked_links);
10622
10623		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10624						      ctl_ooaq, ooa_links);
10625
10626		/*
10627		 * If cur_blocked happens to be the first item in the OOA
10628		 * queue now, prev_ooa will be NULL, and the action
10629		 * returned will just be CTL_ACTION_PASS.
10630		 */
10631		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10632
10633		switch (action) {
10634		case CTL_ACTION_BLOCK:
10635			/* Nothing to do here, still blocked */
10636			break;
10637		case CTL_ACTION_OVERLAP:
10638		case CTL_ACTION_OVERLAP_TAG:
10639			/*
10640			 * This shouldn't happen!  In theory we've already
10641			 * checked this command for overlap...
10642			 */
10643			break;
10644		case CTL_ACTION_PASS:
10645		case CTL_ACTION_SKIP: {
10646			struct ctl_softc *softc;
10647			struct ctl_cmd_entry *entry;
10648			uint32_t initidx;
10649			uint8_t opcode;
10650			int isc_retval;
10651
10652			/*
10653			 * The skip case shouldn't happen, this transaction
10654			 * should have never made it onto the blocked queue.
10655			 */
10656			/*
10657			 * This I/O is no longer blocked, we can remove it
10658			 * from the blocked queue.  Since this is a TAILQ
10659			 * (doubly linked list), we can do O(1) removals
10660			 * from any place on the list.
10661			 */
10662			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10663				     blocked_links);
10664			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10665
10666			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10667				/*
10668				 * Need to send IO back to original side to
10669				 * run
10670				 */
10671				union ctl_ha_msg msg_info;
10672
10673				msg_info.hdr.original_sc =
10674					cur_blocked->io_hdr.original_sc;
10675				msg_info.hdr.serializing_sc = cur_blocked;
10676				msg_info.hdr.msg_type = CTL_MSG_R2R;
10677				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10678				     &msg_info, sizeof(msg_info), 0)) >
10679				     CTL_HA_STATUS_SUCCESS) {
10680					printf("CTL:Check Blocked error from "
10681					       "ctl_ha_msg_send %d\n",
10682					       isc_retval);
10683				}
10684				break;
10685			}
10686			opcode = cur_blocked->scsiio.cdb[0];
10687			entry = &ctl_cmd_table[opcode];
10688			softc = control_softc;
10689
10690			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10691
10692			/*
10693			 * Check this I/O for LUN state changes that may
10694			 * have happened while this command was blocked.
10695			 * The LUN state may have been changed by a command
10696			 * ahead of us in the queue, so we need to re-check
10697			 * for any states that can be caused by SCSI
10698			 * commands.
10699			 */
10700			if (ctl_scsiio_lun_check(softc, lun, entry,
10701						 &cur_blocked->scsiio) == 0) {
10702				cur_blocked->io_hdr.flags |=
10703				                      CTL_FLAG_IS_WAS_ON_RTR;
10704				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
10705						   &cur_blocked->io_hdr, links);
10706				/*
10707				 * In the non CTL_DONE_THREAD case, we need
10708				 * to wake up the work thread here.  When
10709				 * we're processing completed requests from
10710				 * the work thread context, we'll pop back
10711				 * around and end up pulling things off the
10712				 * RtR queue.  When we aren't processing
10713				 * things from the work thread context,
10714				 * though, we won't ever check the RtR queue.
10715				 * So we need to wake up the thread to clear
10716				 * things off the queue.  Otherwise this
10717				 * transaction will just sit on the RtR queue
10718				 * until a new I/O comes in.  (Which may or
10719				 * may not happen...)
10720				 */
10721#ifndef CTL_DONE_THREAD
10722				ctl_wakeup_thread();
10723#endif
10724			} else
10725				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
10726			break;
10727		}
10728		default:
10729			/*
10730			 * This probably shouldn't happen -- we shouldn't
10731			 * get CTL_ACTION_ERROR, or anything else.
10732			 */
10733			break;
10734		}
10735	}
10736
10737	return (CTL_RETVAL_COMPLETE);
10738}
10739
10740/*
10741 * This routine (with one exception) checks LUN flags that can be set by
10742 * commands ahead of us in the OOA queue.  These flags have to be checked
10743 * when a command initially comes in, and when we pull a command off the
10744 * blocked queue and are preparing to execute it.  The reason we have to
10745 * check these flags for commands on the blocked queue is that the LUN
10746 * state may have been changed by a command ahead of us while we're on the
10747 * blocked queue.
10748 *
10749 * Ordering is somewhat important with these checks, so please pay
10750 * careful attention to the placement of any new checks.
10751 */
10752static int
10753ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10754		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10755{
10756	int retval;
10757
10758	retval = 0;
10759
10760	/*
10761	 * If this shelf is a secondary shelf controller, we have to reject
10762	 * any media access commands.
10763	 */
10764#if 0
10765	/* No longer needed for HA */
10766	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10767	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10768		ctl_set_lun_standby(ctsio);
10769		retval = 1;
10770		goto bailout;
10771	}
10772#endif
10773
10774	/*
10775	 * Check for a reservation conflict.  If this command isn't allowed
10776	 * even on reserved LUNs, and if this initiator isn't the one who
10777	 * reserved us, reject the command with a reservation conflict.
10778	 */
10779	if ((lun->flags & CTL_LUN_RESERVED)
10780	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10781		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10782		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10783		 || (ctsio->io_hdr.nexus.targ_target.id !=
10784		     lun->rsv_nexus.targ_target.id)) {
10785			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10786			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10787			retval = 1;
10788			goto bailout;
10789		}
10790	}
10791
10792	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10793	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10794		uint32_t residx;
10795
10796		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10797		/*
10798		 * if we aren't registered or it's a res holder type
10799		 * reservation and this isn't the res holder then set a
10800		 * conflict.
10801		 * NOTE: Commands which might be allowed on write exclusive
10802		 * type reservations are checked in the particular command
10803		 * for a conflict. Read and SSU are the only ones.
10804		 */
10805		if (!lun->per_res[residx].registered
10806		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10807			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10808			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10809			retval = 1;
10810			goto bailout;
10811		}
10812
10813	}
10814
10815	if ((lun->flags & CTL_LUN_OFFLINE)
10816	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10817		ctl_set_lun_not_ready(ctsio);
10818		retval = 1;
10819		goto bailout;
10820	}
10821
10822	/*
10823	 * If the LUN is stopped, see if this particular command is allowed
10824	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10825	 */
10826	if ((lun->flags & CTL_LUN_STOPPED)
10827	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10828		/* "Logical unit not ready, initializing cmd. required" */
10829		ctl_set_lun_stopped(ctsio);
10830		retval = 1;
10831		goto bailout;
10832	}
10833
10834	if ((lun->flags & CTL_LUN_INOPERABLE)
10835	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10836		/* "Medium format corrupted" */
10837		ctl_set_medium_format_corrupted(ctsio);
10838		retval = 1;
10839		goto bailout;
10840	}
10841
10842bailout:
10843	return (retval);
10844
10845}
10846
10847static void
10848ctl_failover_io(union ctl_io *io, int have_lock)
10849{
10850	ctl_set_busy(&io->scsiio);
10851	ctl_done_lock(io, have_lock);
10852}
10853
10854static void
10855ctl_failover(void)
10856{
10857	struct ctl_lun *lun;
10858	struct ctl_softc *ctl_softc;
10859	union ctl_io *next_io, *pending_io;
10860	union ctl_io *io;
10861	int lun_idx;
10862	int i;
10863
10864	ctl_softc = control_softc;
10865
10866	mtx_lock(&ctl_softc->ctl_lock);
10867	/*
10868	 * Remove any cmds from the other SC from the rtr queue.  These
10869	 * will obviously only be for LUNs for which we're the primary.
10870	 * We can't send status or get/send data for these commands.
10871	 * Since they haven't been executed yet, we can just remove them.
10872	 * We'll either abort them or delete them below, depending on
10873	 * which HA mode we're in.
10874	 */
10875	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10876	     io != NULL; io = next_io) {
10877		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10878		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10879			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10880				      ctl_io_hdr, links);
10881	}
10882
10883	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10884		lun = ctl_softc->ctl_luns[lun_idx];
10885		if (lun==NULL)
10886			continue;
10887
10888		/*
10889		 * Processor LUNs are primary on both sides.
10890		 * XXX will this always be true?
10891		 */
10892		if (lun->be_lun->lun_type == T_PROCESSOR)
10893			continue;
10894
10895		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10896		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10897			printf("FAILOVER: primary lun %d\n", lun_idx);
10898		        /*
10899			 * Remove all commands from the other SC. First from the
10900			 * blocked queue then from the ooa queue. Once we have
10901			 * removed them. Call ctl_check_blocked to see if there
10902			 * is anything that can run.
10903			 */
10904			for (io = (union ctl_io *)TAILQ_FIRST(
10905			     &lun->blocked_queue); io != NULL; io = next_io) {
10906
10907		        	next_io = (union ctl_io *)TAILQ_NEXT(
10908				    &io->io_hdr, blocked_links);
10909
10910				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10911					TAILQ_REMOVE(&lun->blocked_queue,
10912						     &io->io_hdr,blocked_links);
10913					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10914					TAILQ_REMOVE(&lun->ooa_queue,
10915						     &io->io_hdr, ooa_links);
10916
10917					ctl_free_io(io);
10918				}
10919			}
10920
10921			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10922	     		     io != NULL; io = next_io) {
10923
10924		        	next_io = (union ctl_io *)TAILQ_NEXT(
10925				    &io->io_hdr, ooa_links);
10926
10927				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10928
10929					TAILQ_REMOVE(&lun->ooa_queue,
10930						&io->io_hdr,
10931					     	ooa_links);
10932
10933					ctl_free_io(io);
10934				}
10935			}
10936			ctl_check_blocked(lun);
10937		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10938			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10939
10940			printf("FAILOVER: primary lun %d\n", lun_idx);
10941			/*
10942			 * Abort all commands from the other SC.  We can't
10943			 * send status back for them now.  These should get
10944			 * cleaned up when they are completed or come out
10945			 * for a datamove operation.
10946			 */
10947			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10948	     		     io != NULL; io = next_io) {
10949		        	next_io = (union ctl_io *)TAILQ_NEXT(
10950					&io->io_hdr, ooa_links);
10951
10952				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10953					io->io_hdr.flags |= CTL_FLAG_ABORT;
10954			}
10955		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10956			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10957
10958			printf("FAILOVER: secondary lun %d\n", lun_idx);
10959
10960			lun->flags |= CTL_LUN_PRIMARY_SC;
10961
10962			/*
10963			 * We send all I/O that was sent to this controller
10964			 * and redirected to the other side back with
10965			 * busy status, and have the initiator retry it.
10966			 * Figuring out how much data has been transferred,
10967			 * etc. and picking up where we left off would be
10968			 * very tricky.
10969			 *
10970			 * XXX KDM need to remove I/O from the blocked
10971			 * queue as well!
10972			 */
10973			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10974			     &lun->ooa_queue); pending_io != NULL;
10975			     pending_io = next_io) {
10976
10977				next_io =  (union ctl_io *)TAILQ_NEXT(
10978					&pending_io->io_hdr, ooa_links);
10979
10980				pending_io->io_hdr.flags &=
10981					~CTL_FLAG_SENT_2OTHER_SC;
10982
10983				if (pending_io->io_hdr.flags &
10984				    CTL_FLAG_IO_ACTIVE) {
10985					pending_io->io_hdr.flags |=
10986						CTL_FLAG_FAILOVER;
10987				} else {
10988					ctl_set_busy(&pending_io->scsiio);
10989					ctl_done_lock(pending_io,
10990						      /*have_lock*/1);
10991				}
10992			}
10993
10994			/*
10995			 * Build Unit Attention
10996			 */
10997			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10998				lun->pending_sense[i].ua_pending |=
10999				                     CTL_UA_ASYM_ACC_CHANGE;
11000			}
11001		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11002			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11003			printf("FAILOVER: secondary lun %d\n", lun_idx);
11004			/*
11005			 * if the first io on the OOA is not on the RtR queue
11006			 * add it.
11007			 */
11008			lun->flags |= CTL_LUN_PRIMARY_SC;
11009
11010			pending_io = (union ctl_io *)TAILQ_FIRST(
11011			    &lun->ooa_queue);
11012			if (pending_io==NULL) {
11013				printf("Nothing on OOA queue\n");
11014				continue;
11015			}
11016
11017			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11018			if ((pending_io->io_hdr.flags &
11019			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11020				pending_io->io_hdr.flags |=
11021				    CTL_FLAG_IS_WAS_ON_RTR;
11022				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11023						   &pending_io->io_hdr, links);
11024			}
11025#if 0
11026			else
11027			{
11028				printf("Tag 0x%04x is running\n",
11029				      pending_io->scsiio.tag_num);
11030			}
11031#endif
11032
11033			next_io = (union ctl_io *)TAILQ_NEXT(
11034			    &pending_io->io_hdr, ooa_links);
11035			for (pending_io=next_io; pending_io != NULL;
11036			     pending_io = next_io) {
11037				pending_io->io_hdr.flags &=
11038				    ~CTL_FLAG_SENT_2OTHER_SC;
11039				next_io = (union ctl_io *)TAILQ_NEXT(
11040					&pending_io->io_hdr, ooa_links);
11041				if (pending_io->io_hdr.flags &
11042				    CTL_FLAG_IS_WAS_ON_RTR) {
11043#if 0
11044				        printf("Tag 0x%04x is running\n",
11045				      		pending_io->scsiio.tag_num);
11046#endif
11047					continue;
11048				}
11049
11050				switch (ctl_check_ooa(lun, pending_io,
11051			            (union ctl_io *)TAILQ_PREV(
11052				    &pending_io->io_hdr, ctl_ooaq,
11053				    ooa_links))) {
11054
11055				case CTL_ACTION_BLOCK:
11056					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11057							  &pending_io->io_hdr,
11058							  blocked_links);
11059					pending_io->io_hdr.flags |=
11060					    CTL_FLAG_BLOCKED;
11061					break;
11062				case CTL_ACTION_PASS:
11063				case CTL_ACTION_SKIP:
11064					pending_io->io_hdr.flags |=
11065					    CTL_FLAG_IS_WAS_ON_RTR;
11066					STAILQ_INSERT_TAIL(
11067					    &ctl_softc->rtr_queue,
11068					    &pending_io->io_hdr, links);
11069					break;
11070				case CTL_ACTION_OVERLAP:
11071					ctl_set_overlapped_cmd(
11072					    (struct ctl_scsiio *)pending_io);
11073					ctl_done_lock(pending_io,
11074						      /*have_lock*/ 1);
11075					break;
11076				case CTL_ACTION_OVERLAP_TAG:
11077					ctl_set_overlapped_tag(
11078					    (struct ctl_scsiio *)pending_io,
11079					    pending_io->scsiio.tag_num & 0xff);
11080					ctl_done_lock(pending_io,
11081						      /*have_lock*/ 1);
11082					break;
11083				case CTL_ACTION_ERROR:
11084				default:
11085					ctl_set_internal_failure(
11086						(struct ctl_scsiio *)pending_io,
11087						0,  // sks_valid
11088						0); //retry count
11089					ctl_done_lock(pending_io,
11090						      /*have_lock*/ 1);
11091					break;
11092				}
11093			}
11094
11095			/*
11096			 * Build Unit Attention
11097			 */
11098			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11099				lun->pending_sense[i].ua_pending |=
11100				                     CTL_UA_ASYM_ACC_CHANGE;
11101			}
11102		} else {
11103			panic("Unhandled HA mode failover, LUN flags = %#x, "
11104			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11105		}
11106	}
11107	ctl_pause_rtr = 0;
11108	mtx_unlock(&ctl_softc->ctl_lock);
11109}
11110
11111static int
11112ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11113{
11114	struct ctl_lun *lun;
11115	struct ctl_cmd_entry *entry;
11116	uint8_t opcode;
11117	uint32_t initidx, targ_lun;
11118	int retval;
11119
11120	retval = 0;
11121
11122	lun = NULL;
11123
11124	opcode = ctsio->cdb[0];
11125
11126	mtx_lock(&ctl_softc->ctl_lock);
11127
11128	targ_lun = ctsio->io_hdr.nexus.targ_lun;
11129	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
11130		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
11131	if ((targ_lun < CTL_MAX_LUNS)
11132	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11133		lun = ctl_softc->ctl_luns[targ_lun];
11134		/*
11135		 * If the LUN is invalid, pretend that it doesn't exist.
11136		 * It will go away as soon as all pending I/O has been
11137		 * completed.
11138		 */
11139		if (lun->flags & CTL_LUN_DISABLED) {
11140			lun = NULL;
11141		} else {
11142			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11143			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11144				lun->be_lun;
11145			if (lun->be_lun->lun_type == T_PROCESSOR) {
11146				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11147			}
11148		}
11149	} else {
11150		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11151		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11152	}
11153
11154	entry = &ctl_cmd_table[opcode];
11155
11156	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11157	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11158
11159	/*
11160	 * Check to see whether we can send this command to LUNs that don't
11161	 * exist.  This should pretty much only be the case for inquiry
11162	 * and request sense.  Further checks, below, really require having
11163	 * a LUN, so we can't really check the command anymore.  Just put
11164	 * it on the rtr queue.
11165	 */
11166	if (lun == NULL) {
11167		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11168			goto queue_rtr;
11169
11170		ctl_set_unsupported_lun(ctsio);
11171		mtx_unlock(&ctl_softc->ctl_lock);
11172		ctl_done((union ctl_io *)ctsio);
11173		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11174		goto bailout;
11175	} else {
11176		/*
11177		 * Every I/O goes into the OOA queue for a particular LUN, and
11178		 * stays there until completion.
11179		 */
11180		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11181
11182		/*
11183		 * Make sure we support this particular command on this LUN.
11184		 * e.g., we don't support writes to the control LUN.
11185		 */
11186		switch (lun->be_lun->lun_type) {
11187		case T_PROCESSOR:
11188		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11189			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11190			      == 0)) {
11191				ctl_set_invalid_opcode(ctsio);
11192				mtx_unlock(&ctl_softc->ctl_lock);
11193				ctl_done((union ctl_io *)ctsio);
11194				goto bailout;
11195			}
11196			break;
11197		case T_DIRECT:
11198			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11199			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11200			      == 0)){
11201				ctl_set_invalid_opcode(ctsio);
11202				mtx_unlock(&ctl_softc->ctl_lock);
11203				ctl_done((union ctl_io *)ctsio);
11204				goto bailout;
11205			}
11206			break;
11207		default:
11208			printf("Unsupported CTL LUN type %d\n",
11209			       lun->be_lun->lun_type);
11210			panic("Unsupported CTL LUN type %d\n",
11211			      lun->be_lun->lun_type);
11212			break; /* NOTREACHED */
11213		}
11214	}
11215
11216	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11217
11218	/*
11219	 * If we've got a request sense, it'll clear the contingent
11220	 * allegiance condition.  Otherwise, if we have a CA condition for
11221	 * this initiator, clear it, because it sent down a command other
11222	 * than request sense.
11223	 */
11224	if ((opcode != REQUEST_SENSE)
11225	 && (ctl_is_set(lun->have_ca, initidx)))
11226		ctl_clear_mask(lun->have_ca, initidx);
11227
11228	/*
11229	 * If the command has this flag set, it handles its own unit
11230	 * attention reporting, we shouldn't do anything.  Otherwise we
11231	 * check for any pending unit attentions, and send them back to the
11232	 * initiator.  We only do this when a command initially comes in,
11233	 * not when we pull it off the blocked queue.
11234	 *
11235	 * According to SAM-3, section 5.3.2, the order that things get
11236	 * presented back to the host is basically unit attentions caused
11237	 * by some sort of reset event, busy status, reservation conflicts
11238	 * or task set full, and finally any other status.
11239	 *
11240	 * One issue here is that some of the unit attentions we report
11241	 * don't fall into the "reset" category (e.g. "reported luns data
11242	 * has changed").  So reporting it here, before the reservation
11243	 * check, may be technically wrong.  I guess the only thing to do
11244	 * would be to check for and report the reset events here, and then
11245	 * check for the other unit attention types after we check for a
11246	 * reservation conflict.
11247	 *
11248	 * XXX KDM need to fix this
11249	 */
11250	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11251		ctl_ua_type ua_type;
11252
11253		ua_type = lun->pending_sense[initidx].ua_pending;
11254		if (ua_type != CTL_UA_NONE) {
11255			scsi_sense_data_type sense_format;
11256
11257			if (lun != NULL)
11258				sense_format = (lun->flags &
11259				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11260				    SSD_TYPE_FIXED;
11261			else
11262				sense_format = SSD_TYPE_FIXED;
11263
11264			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11265					       sense_format);
11266			if (ua_type != CTL_UA_NONE) {
11267				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11268				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11269						       CTL_AUTOSENSE;
11270				ctsio->sense_len = SSD_FULL_SIZE;
11271				lun->pending_sense[initidx].ua_pending &=
11272					~ua_type;
11273				mtx_unlock(&ctl_softc->ctl_lock);
11274				ctl_done((union ctl_io *)ctsio);
11275				goto bailout;
11276			}
11277		}
11278	}
11279
11280
11281	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11282		mtx_unlock(&ctl_softc->ctl_lock);
11283		ctl_done((union ctl_io *)ctsio);
11284		goto bailout;
11285	}
11286
11287	/*
11288	 * XXX CHD this is where we want to send IO to other side if
11289	 * this LUN is secondary on this SC. We will need to make a copy
11290	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11291	 * the copy we send as FROM_OTHER.
11292	 * We also need to stuff the address of the original IO so we can
11293	 * find it easily. Something similar will need be done on the other
11294	 * side so when we are done we can find the copy.
11295	 */
11296	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11297		union ctl_ha_msg msg_info;
11298		int isc_retval;
11299
11300		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11301
11302		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11303		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11304#if 0
11305		printf("1. ctsio %p\n", ctsio);
11306#endif
11307		msg_info.hdr.serializing_sc = NULL;
11308		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11309		msg_info.scsi.tag_num = ctsio->tag_num;
11310		msg_info.scsi.tag_type = ctsio->tag_type;
11311		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11312
11313		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11314
11315		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11316		    (void *)&msg_info, sizeof(msg_info), 0)) >
11317		    CTL_HA_STATUS_SUCCESS) {
11318			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11319			       isc_retval);
11320			printf("CTL:opcode is %x\n",opcode);
11321		} else {
11322#if 0
11323			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11324#endif
11325		}
11326
11327		/*
11328		 * XXX KDM this I/O is off the incoming queue, but hasn't
11329		 * been inserted on any other queue.  We may need to come
11330		 * up with a holding queue while we wait for serialization
11331		 * so that we have an idea of what we're waiting for from
11332		 * the other side.
11333		 */
11334		goto bailout_unlock;
11335	}
11336
11337	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11338			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11339			      ctl_ooaq, ooa_links))) {
11340	case CTL_ACTION_BLOCK:
11341		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11342		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11343				  blocked_links);
11344		goto bailout_unlock;
11345		break; /* NOTREACHED */
11346	case CTL_ACTION_PASS:
11347	case CTL_ACTION_SKIP:
11348		goto queue_rtr;
11349		break; /* NOTREACHED */
11350	case CTL_ACTION_OVERLAP:
11351		ctl_set_overlapped_cmd(ctsio);
11352		mtx_unlock(&ctl_softc->ctl_lock);
11353		ctl_done((union ctl_io *)ctsio);
11354		goto bailout;
11355		break; /* NOTREACHED */
11356	case CTL_ACTION_OVERLAP_TAG:
11357		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11358		mtx_unlock(&ctl_softc->ctl_lock);
11359		ctl_done((union ctl_io *)ctsio);
11360		goto bailout;
11361		break; /* NOTREACHED */
11362	case CTL_ACTION_ERROR:
11363	default:
11364		ctl_set_internal_failure(ctsio,
11365					 /*sks_valid*/ 0,
11366					 /*retry_count*/ 0);
11367		mtx_unlock(&ctl_softc->ctl_lock);
11368		ctl_done((union ctl_io *)ctsio);
11369		goto bailout;
11370		break; /* NOTREACHED */
11371	}
11372
11373	goto bailout_unlock;
11374
11375queue_rtr:
11376	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11377	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
11378
11379bailout_unlock:
11380	mtx_unlock(&ctl_softc->ctl_lock);
11381
11382bailout:
11383	return (retval);
11384}
11385
11386static int
11387ctl_scsiio(struct ctl_scsiio *ctsio)
11388{
11389	int retval;
11390	struct ctl_cmd_entry *entry;
11391
11392	retval = CTL_RETVAL_COMPLETE;
11393
11394	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11395
11396	entry = &ctl_cmd_table[ctsio->cdb[0]];
11397
11398	/*
11399	 * If this I/O has been aborted, just send it straight to
11400	 * ctl_done() without executing it.
11401	 */
11402	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11403		ctl_done((union ctl_io *)ctsio);
11404		goto bailout;
11405	}
11406
11407	/*
11408	 * All the checks should have been handled by ctl_scsiio_precheck().
11409	 * We should be clear now to just execute the I/O.
11410	 */
11411	retval = entry->execute(ctsio);
11412
11413bailout:
11414	return (retval);
11415}
11416
11417/*
11418 * Since we only implement one target right now, a bus reset simply resets
11419 * our single target.
11420 */
11421static int
11422ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11423{
11424	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11425}
11426
11427static int
11428ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11429		 ctl_ua_type ua_type)
11430{
11431	struct ctl_lun *lun;
11432	int retval;
11433
11434	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11435		union ctl_ha_msg msg_info;
11436
11437		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11438		msg_info.hdr.nexus = io->io_hdr.nexus;
11439		if (ua_type==CTL_UA_TARG_RESET)
11440			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11441		else
11442			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11443		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11444		msg_info.hdr.original_sc = NULL;
11445		msg_info.hdr.serializing_sc = NULL;
11446		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11447		    (void *)&msg_info, sizeof(msg_info), 0)) {
11448		}
11449	}
11450	retval = 0;
11451
11452	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11453		retval += ctl_lun_reset(lun, io, ua_type);
11454
11455	return (retval);
11456}
11457
11458/*
11459 * The LUN should always be set.  The I/O is optional, and is used to
11460 * distinguish between I/Os sent by this initiator, and by other
11461 * initiators.  We set unit attention for initiators other than this one.
11462 * SAM-3 is vague on this point.  It does say that a unit attention should
11463 * be established for other initiators when a LUN is reset (see section
11464 * 5.7.3), but it doesn't specifically say that the unit attention should
11465 * be established for this particular initiator when a LUN is reset.  Here
11466 * is the relevant text, from SAM-3 rev 8:
11467 *
11468 * 5.7.2 When a SCSI initiator port aborts its own tasks
11469 *
11470 * When a SCSI initiator port causes its own task(s) to be aborted, no
11471 * notification that the task(s) have been aborted shall be returned to
11472 * the SCSI initiator port other than the completion response for the
11473 * command or task management function action that caused the task(s) to
11474 * be aborted and notification(s) associated with related effects of the
11475 * action (e.g., a reset unit attention condition).
11476 *
11477 * XXX KDM for now, we're setting unit attention for all initiators.
11478 */
11479static int
11480ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11481{
11482	union ctl_io *xio;
11483#if 0
11484	uint32_t initindex;
11485#endif
11486	int i;
11487
11488	/*
11489	 * Run through the OOA queue and abort each I/O.
11490	 */
11491#if 0
11492	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11493#endif
11494	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11495	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11496		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11497	}
11498
11499	/*
11500	 * This version sets unit attention for every
11501	 */
11502#if 0
11503	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11504	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11505		if (initindex == i)
11506			continue;
11507		lun->pending_sense[i].ua_pending |= ua_type;
11508	}
11509#endif
11510
11511	/*
11512	 * A reset (any kind, really) clears reservations established with
11513	 * RESERVE/RELEASE.  It does not clear reservations established
11514	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11515	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11516	 * reservations made with the RESERVE/RELEASE commands, because
11517	 * those commands are obsolete in SPC-3.
11518	 */
11519	lun->flags &= ~CTL_LUN_RESERVED;
11520
11521	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11522		ctl_clear_mask(lun->have_ca, i);
11523		lun->pending_sense[i].ua_pending |= ua_type;
11524	}
11525
11526	return (0);
11527}
11528
11529static int
11530ctl_abort_task(union ctl_io *io)
11531{
11532	union ctl_io *xio;
11533	struct ctl_lun *lun;
11534	struct ctl_softc *ctl_softc;
11535#if 0
11536	struct sbuf sb;
11537	char printbuf[128];
11538#endif
11539	int found;
11540	uint32_t targ_lun;
11541
11542	ctl_softc = control_softc;
11543	found = 0;
11544
11545	/*
11546	 * Look up the LUN.
11547	 */
11548	targ_lun = io->io_hdr.nexus.targ_lun;
11549	if (io->io_hdr.nexus.lun_map_fn != NULL)
11550		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11551	if ((targ_lun < CTL_MAX_LUNS)
11552	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11553		lun = ctl_softc->ctl_luns[targ_lun];
11554	else
11555		goto bailout;
11556
11557#if 0
11558	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11559	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11560#endif
11561
11562	/*
11563	 * Run through the OOA queue and attempt to find the given I/O.
11564	 * The target port, initiator ID, tag type and tag number have to
11565	 * match the values that we got from the initiator.  If we have an
11566	 * untagged command to abort, simply abort the first untagged command
11567	 * we come to.  We only allow one untagged command at a time of course.
11568	 */
11569#if 0
11570	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11571#endif
11572	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11573	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11574#if 0
11575		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11576
11577		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11578			    lun->lun, xio->scsiio.tag_num,
11579			    xio->scsiio.tag_type,
11580			    (xio->io_hdr.blocked_links.tqe_prev
11581			    == NULL) ? "" : " BLOCKED",
11582			    (xio->io_hdr.flags &
11583			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11584			    (xio->io_hdr.flags &
11585			    CTL_FLAG_ABORT) ? " ABORT" : "",
11586			    (xio->io_hdr.flags &
11587			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11588		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11589		sbuf_finish(&sb);
11590		printf("%s\n", sbuf_data(&sb));
11591#endif
11592
11593		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11594		 && (xio->io_hdr.nexus.initid.id ==
11595		     io->io_hdr.nexus.initid.id)) {
11596			/*
11597			 * If the abort says that the task is untagged, the
11598			 * task in the queue must be untagged.  Otherwise,
11599			 * we just check to see whether the tag numbers
11600			 * match.  This is because the QLogic firmware
11601			 * doesn't pass back the tag type in an abort
11602			 * request.
11603			 */
11604#if 0
11605			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11606			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11607			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11608#endif
11609			/*
11610			 * XXX KDM we've got problems with FC, because it
11611			 * doesn't send down a tag type with aborts.  So we
11612			 * can only really go by the tag number...
11613			 * This may cause problems with parallel SCSI.
11614			 * Need to figure that out!!
11615			 */
11616			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11617				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11618				found = 1;
11619				if ((io->io_hdr.flags &
11620				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11621				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11622					union ctl_ha_msg msg_info;
11623
11624					io->io_hdr.flags |=
11625					                CTL_FLAG_SENT_2OTHER_SC;
11626					msg_info.hdr.nexus = io->io_hdr.nexus;
11627					msg_info.task.task_action =
11628						CTL_TASK_ABORT_TASK;
11629					msg_info.task.tag_num =
11630						io->taskio.tag_num;
11631					msg_info.task.tag_type =
11632						io->taskio.tag_type;
11633					msg_info.hdr.msg_type =
11634						CTL_MSG_MANAGE_TASKS;
11635					msg_info.hdr.original_sc = NULL;
11636					msg_info.hdr.serializing_sc = NULL;
11637#if 0
11638					printf("Sent Abort to other side\n");
11639#endif
11640					if (CTL_HA_STATUS_SUCCESS !=
11641					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11642		    				(void *)&msg_info,
11643						sizeof(msg_info), 0)) {
11644					}
11645				}
11646#if 0
11647				printf("ctl_abort_task: found I/O to abort\n");
11648#endif
11649				break;
11650			}
11651		}
11652	}
11653
11654bailout:
11655
11656	if (found == 0) {
11657		/*
11658		 * This isn't really an error.  It's entirely possible for
11659		 * the abort and command completion to cross on the wire.
11660		 * This is more of an informative/diagnostic error.
11661		 */
11662#if 0
11663		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11664		       "%d:%d:%d:%d tag %d type %d\n",
11665		       io->io_hdr.nexus.initid.id,
11666		       io->io_hdr.nexus.targ_port,
11667		       io->io_hdr.nexus.targ_target.id,
11668		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11669		       io->taskio.tag_type);
11670#endif
11671		return (1);
11672	} else
11673		return (0);
11674}
11675
11676/*
11677 * This routine cannot block!  It must be callable from an interrupt
11678 * handler as well as from the work thread.
11679 */
11680static void
11681ctl_run_task_queue(struct ctl_softc *ctl_softc)
11682{
11683	union ctl_io *io, *next_io;
11684
11685	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
11686
11687	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
11688
11689	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
11690	     io != NULL; io = next_io) {
11691		int retval;
11692		const char *task_desc;
11693
11694		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11695
11696		retval = 0;
11697
11698		switch (io->io_hdr.io_type) {
11699		case CTL_IO_TASK: {
11700			task_desc = ctl_scsi_task_string(&io->taskio);
11701			if (task_desc != NULL) {
11702#ifdef NEEDTOPORT
11703				csevent_log(CSC_CTL | CSC_SHELF_SW |
11704					    CTL_TASK_REPORT,
11705					    csevent_LogType_Trace,
11706					    csevent_Severity_Information,
11707					    csevent_AlertLevel_Green,
11708					    csevent_FRU_Firmware,
11709					    csevent_FRU_Unknown,
11710					    "CTL: received task: %s",task_desc);
11711#endif
11712			} else {
11713#ifdef NEEDTOPORT
11714				csevent_log(CSC_CTL | CSC_SHELF_SW |
11715					    CTL_TASK_REPORT,
11716					    csevent_LogType_Trace,
11717					    csevent_Severity_Information,
11718					    csevent_AlertLevel_Green,
11719					    csevent_FRU_Firmware,
11720					    csevent_FRU_Unknown,
11721					    "CTL: received unknown task "
11722					    "type: %d (%#x)",
11723					    io->taskio.task_action,
11724					    io->taskio.task_action);
11725#endif
11726			}
11727			switch (io->taskio.task_action) {
11728			case CTL_TASK_ABORT_TASK:
11729				retval = ctl_abort_task(io);
11730				break;
11731			case CTL_TASK_ABORT_TASK_SET:
11732				break;
11733			case CTL_TASK_CLEAR_ACA:
11734				break;
11735			case CTL_TASK_CLEAR_TASK_SET:
11736				break;
11737			case CTL_TASK_LUN_RESET: {
11738				struct ctl_lun *lun;
11739				uint32_t targ_lun;
11740				int retval;
11741
11742				targ_lun = io->io_hdr.nexus.targ_lun;
11743				if (io->io_hdr.nexus.lun_map_fn != NULL)
11744					targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11745
11746				if ((targ_lun < CTL_MAX_LUNS)
11747				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11748					lun = ctl_softc->ctl_luns[targ_lun];
11749				else {
11750					retval = 1;
11751					break;
11752				}
11753
11754				if (!(io->io_hdr.flags &
11755				    CTL_FLAG_FROM_OTHER_SC)) {
11756					union ctl_ha_msg msg_info;
11757
11758					io->io_hdr.flags |=
11759						CTL_FLAG_SENT_2OTHER_SC;
11760					msg_info.hdr.msg_type =
11761						CTL_MSG_MANAGE_TASKS;
11762					msg_info.hdr.nexus = io->io_hdr.nexus;
11763					msg_info.task.task_action =
11764						CTL_TASK_LUN_RESET;
11765					msg_info.hdr.original_sc = NULL;
11766					msg_info.hdr.serializing_sc = NULL;
11767					if (CTL_HA_STATUS_SUCCESS !=
11768					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11769					    (void *)&msg_info,
11770					    sizeof(msg_info), 0)) {
11771					}
11772				}
11773
11774				retval = ctl_lun_reset(lun, io,
11775						       CTL_UA_LUN_RESET);
11776				break;
11777			}
11778			case CTL_TASK_TARGET_RESET:
11779				retval = ctl_target_reset(ctl_softc, io,
11780							  CTL_UA_TARG_RESET);
11781				break;
11782			case CTL_TASK_BUS_RESET:
11783				retval = ctl_bus_reset(ctl_softc, io);
11784				break;
11785			case CTL_TASK_PORT_LOGIN:
11786				break;
11787			case CTL_TASK_PORT_LOGOUT:
11788				break;
11789			default:
11790				printf("ctl_run_task_queue: got unknown task "
11791				       "management event %d\n",
11792				       io->taskio.task_action);
11793				break;
11794			}
11795			if (retval == 0)
11796				io->io_hdr.status = CTL_SUCCESS;
11797			else
11798				io->io_hdr.status = CTL_ERROR;
11799
11800			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11801				      ctl_io_hdr, links);
11802			/*
11803			 * This will queue this I/O to the done queue, but the
11804			 * work thread won't be able to process it until we
11805			 * return and the lock is released.
11806			 */
11807			ctl_done_lock(io, /*have_lock*/ 1);
11808			break;
11809		}
11810		default: {
11811
11812			printf("%s: invalid I/O type %d msg %d cdb %x"
11813			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11814			       __func__, io->io_hdr.io_type,
11815			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11816			       (uintmax_t)io->io_hdr.nexus.initid.id,
11817			       io->io_hdr.nexus.targ_port,
11818			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11819			       io->io_hdr.nexus.targ_lun /* XXX */,
11820			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11821			       io->taskio.tag_num : io->scsiio.tag_num);
11822			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11823				      ctl_io_hdr, links);
11824			ctl_free_io(io);
11825			break;
11826		}
11827		}
11828	}
11829
11830	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11831}
11832
11833/*
11834 * For HA operation.  Handle commands that come in from the other
11835 * controller.
11836 */
11837static void
11838ctl_handle_isc(union ctl_io *io)
11839{
11840	int free_io;
11841	struct ctl_lun *lun;
11842	struct ctl_softc *ctl_softc;
11843	uint32_t targ_lun;
11844
11845	ctl_softc = control_softc;
11846
11847	targ_lun = io->io_hdr.nexus.targ_lun;
11848	if (io->io_hdr.nexus.lun_map_fn != NULL)
11849		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11850	lun = ctl_softc->ctl_luns[targ_lun];
11851
11852	switch (io->io_hdr.msg_type) {
11853	case CTL_MSG_SERIALIZE:
11854		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11855						     /*have_lock*/ 0);
11856		break;
11857	case CTL_MSG_R2R: {
11858		uint8_t opcode;
11859		struct ctl_cmd_entry *entry;
11860
11861		/*
11862		 * This is only used in SER_ONLY mode.
11863		 */
11864		free_io = 0;
11865		opcode = io->scsiio.cdb[0];
11866		entry = &ctl_cmd_table[opcode];
11867		mtx_lock(&ctl_softc->ctl_lock);
11868		if (ctl_scsiio_lun_check(ctl_softc, lun,
11869		    entry, (struct ctl_scsiio *)io) != 0) {
11870			ctl_done_lock(io, /*have_lock*/ 1);
11871			mtx_unlock(&ctl_softc->ctl_lock);
11872			break;
11873		}
11874		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11875		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11876				   &io->io_hdr, links);
11877		mtx_unlock(&ctl_softc->ctl_lock);
11878		break;
11879	}
11880	case CTL_MSG_FINISH_IO:
11881		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11882			free_io = 0;
11883			ctl_done_lock(io, /*have_lock*/ 0);
11884		} else {
11885			free_io = 1;
11886			mtx_lock(&ctl_softc->ctl_lock);
11887			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11888				     ooa_links);
11889			STAILQ_REMOVE(&ctl_softc->task_queue,
11890				      &io->io_hdr, ctl_io_hdr, links);
11891			ctl_check_blocked(lun);
11892			mtx_unlock(&ctl_softc->ctl_lock);
11893		}
11894		break;
11895	case CTL_MSG_PERS_ACTION:
11896		ctl_hndl_per_res_out_on_other_sc(
11897			(union ctl_ha_msg *)&io->presio.pr_msg);
11898		free_io = 1;
11899		break;
11900	case CTL_MSG_BAD_JUJU:
11901		free_io = 0;
11902		ctl_done_lock(io, /*have_lock*/ 0);
11903		break;
11904	case CTL_MSG_DATAMOVE:
11905		/* Only used in XFER mode */
11906		free_io = 0;
11907		ctl_datamove_remote(io);
11908		break;
11909	case CTL_MSG_DATAMOVE_DONE:
11910		/* Only used in XFER mode */
11911		free_io = 0;
11912		io->scsiio.be_move_done(io);
11913		break;
11914	default:
11915		free_io = 1;
11916		printf("%s: Invalid message type %d\n",
11917		       __func__, io->io_hdr.msg_type);
11918		break;
11919	}
11920	if (free_io)
11921		ctl_free_io(io);
11922
11923}
11924
11925
11926/*
11927 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11928 * there is no match.
11929 */
11930static ctl_lun_error_pattern
11931ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11932{
11933	struct ctl_cmd_entry *entry;
11934	ctl_lun_error_pattern filtered_pattern, pattern;
11935	uint8_t opcode;
11936
11937	pattern = desc->error_pattern;
11938
11939	/*
11940	 * XXX KDM we need more data passed into this function to match a
11941	 * custom pattern, and we actually need to implement custom pattern
11942	 * matching.
11943	 */
11944	if (pattern & CTL_LUN_PAT_CMD)
11945		return (CTL_LUN_PAT_CMD);
11946
11947	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11948		return (CTL_LUN_PAT_ANY);
11949
11950	opcode = ctsio->cdb[0];
11951	entry = &ctl_cmd_table[opcode];
11952
11953	filtered_pattern = entry->pattern & pattern;
11954
11955	/*
11956	 * If the user requested specific flags in the pattern (e.g.
11957	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11958	 * flags.
11959	 *
11960	 * If the user did not specify any flags, it doesn't matter whether
11961	 * or not the command supports the flags.
11962	 */
11963	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11964	     (pattern & ~CTL_LUN_PAT_MASK))
11965		return (CTL_LUN_PAT_NONE);
11966
11967	/*
11968	 * If the user asked for a range check, see if the requested LBA
11969	 * range overlaps with this command's LBA range.
11970	 */
11971	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11972		uint64_t lba1;
11973		uint32_t len1;
11974		ctl_action action;
11975		int retval;
11976
11977		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11978		if (retval != 0)
11979			return (CTL_LUN_PAT_NONE);
11980
11981		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11982					      desc->lba_range.len);
11983		/*
11984		 * A "pass" means that the LBA ranges don't overlap, so
11985		 * this doesn't match the user's range criteria.
11986		 */
11987		if (action == CTL_ACTION_PASS)
11988			return (CTL_LUN_PAT_NONE);
11989	}
11990
11991	return (filtered_pattern);
11992}
11993
11994static void
11995ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11996{
11997	struct ctl_error_desc *desc, *desc2;
11998
11999	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
12000
12001	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12002		ctl_lun_error_pattern pattern;
12003		/*
12004		 * Check to see whether this particular command matches
12005		 * the pattern in the descriptor.
12006		 */
12007		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12008		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12009			continue;
12010
12011		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12012		case CTL_LUN_INJ_ABORTED:
12013			ctl_set_aborted(&io->scsiio);
12014			break;
12015		case CTL_LUN_INJ_MEDIUM_ERR:
12016			ctl_set_medium_error(&io->scsiio);
12017			break;
12018		case CTL_LUN_INJ_UA:
12019			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12020			 * OCCURRED */
12021			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12022			break;
12023		case CTL_LUN_INJ_CUSTOM:
12024			/*
12025			 * We're assuming the user knows what he is doing.
12026			 * Just copy the sense information without doing
12027			 * checks.
12028			 */
12029			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12030			      ctl_min(sizeof(desc->custom_sense),
12031				      sizeof(io->scsiio.sense_data)));
12032			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12033			io->scsiio.sense_len = SSD_FULL_SIZE;
12034			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12035			break;
12036		case CTL_LUN_INJ_NONE:
12037		default:
12038			/*
12039			 * If this is an error injection type we don't know
12040			 * about, clear the continuous flag (if it is set)
12041			 * so it will get deleted below.
12042			 */
12043			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12044			break;
12045		}
12046		/*
12047		 * By default, each error injection action is a one-shot
12048		 */
12049		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12050			continue;
12051
12052		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12053
12054		free(desc, M_CTL);
12055	}
12056}
12057
12058#ifdef CTL_IO_DELAY
12059static void
12060ctl_datamove_timer_wakeup(void *arg)
12061{
12062	union ctl_io *io;
12063
12064	io = (union ctl_io *)arg;
12065
12066	ctl_datamove(io);
12067}
12068#endif /* CTL_IO_DELAY */
12069
12070void
12071ctl_datamove(union ctl_io *io)
12072{
12073	void (*fe_datamove)(union ctl_io *io);
12074
12075	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12076
12077	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12078
12079#ifdef CTL_TIME_IO
12080	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12081		char str[256];
12082		char path_str[64];
12083		struct sbuf sb;
12084
12085		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12086		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12087
12088		sbuf_cat(&sb, path_str);
12089		switch (io->io_hdr.io_type) {
12090		case CTL_IO_SCSI:
12091			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12092			sbuf_printf(&sb, "\n");
12093			sbuf_cat(&sb, path_str);
12094			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12095				    io->scsiio.tag_num, io->scsiio.tag_type);
12096			break;
12097		case CTL_IO_TASK:
12098			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12099				    "Tag Type: %d\n", io->taskio.task_action,
12100				    io->taskio.tag_num, io->taskio.tag_type);
12101			break;
12102		default:
12103			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12104			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12105			break;
12106		}
12107		sbuf_cat(&sb, path_str);
12108		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12109			    (intmax_t)time_uptime - io->io_hdr.start_time);
12110		sbuf_finish(&sb);
12111		printf("%s", sbuf_data(&sb));
12112	}
12113#endif /* CTL_TIME_IO */
12114
12115	mtx_lock(&control_softc->ctl_lock);
12116#ifdef CTL_IO_DELAY
12117	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12118		struct ctl_lun *lun;
12119
12120		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12121
12122		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12123	} else {
12124		struct ctl_lun *lun;
12125
12126		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12127		if ((lun != NULL)
12128		 && (lun->delay_info.datamove_delay > 0)) {
12129			struct callout *callout;
12130
12131			callout = (struct callout *)&io->io_hdr.timer_bytes;
12132			callout_init(callout, /*mpsafe*/ 1);
12133			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12134			callout_reset(callout,
12135				      lun->delay_info.datamove_delay * hz,
12136				      ctl_datamove_timer_wakeup, io);
12137			if (lun->delay_info.datamove_type ==
12138			    CTL_DELAY_TYPE_ONESHOT)
12139				lun->delay_info.datamove_delay = 0;
12140			mtx_unlock(&control_softc->ctl_lock);
12141			return;
12142		}
12143	}
12144#endif
12145	/*
12146	 * If we have any pending task management commands, process them
12147	 * first.  This is necessary to eliminate a race condition with the
12148	 * FETD:
12149	 *
12150	 * - FETD submits a task management command, like an abort.
12151	 * - Back end calls fe_datamove() to move the data for the aborted
12152	 *   command.  The FETD can't really accept it, but if it did, it
12153	 *   would end up transmitting data for a command that the initiator
12154	 *   told us to abort.
12155	 *
12156	 * We close the race by processing all pending task management
12157	 * commands here (we can't block!), and then check this I/O to see
12158	 * if it has been aborted.  If so, return it to the back end with
12159	 * bad status, so the back end can say return an error to the back end
12160	 * and then when the back end returns an error, we can return the
12161	 * aborted command to the FETD, so it can clean up its resources.
12162	 */
12163	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
12164		ctl_run_task_queue(control_softc);
12165
12166	/*
12167	 * This command has been aborted.  Set the port status, so we fail
12168	 * the data move.
12169	 */
12170	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12171		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12172		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12173		       io->io_hdr.nexus.targ_port,
12174		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12175		       io->io_hdr.nexus.targ_lun);
12176		io->io_hdr.status = CTL_CMD_ABORTED;
12177		io->io_hdr.port_status = 31337;
12178		mtx_unlock(&control_softc->ctl_lock);
12179		/*
12180		 * Note that the backend, in this case, will get the
12181		 * callback in its context.  In other cases it may get
12182		 * called in the frontend's interrupt thread context.
12183		 */
12184		io->scsiio.be_move_done(io);
12185		return;
12186	}
12187
12188	/*
12189	 * If we're in XFER mode and this I/O is from the other shelf
12190	 * controller, we need to send the DMA to the other side to
12191	 * actually transfer the data to/from the host.  In serialize only
12192	 * mode the transfer happens below CTL and ctl_datamove() is only
12193	 * called on the machine that originally received the I/O.
12194	 */
12195	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12196	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12197		union ctl_ha_msg msg;
12198		uint32_t sg_entries_sent;
12199		int do_sg_copy;
12200		int i;
12201
12202		memset(&msg, 0, sizeof(msg));
12203		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12204		msg.hdr.original_sc = io->io_hdr.original_sc;
12205		msg.hdr.serializing_sc = io;
12206		msg.hdr.nexus = io->io_hdr.nexus;
12207		msg.dt.flags = io->io_hdr.flags;
12208		/*
12209		 * We convert everything into a S/G list here.  We can't
12210		 * pass by reference, only by value between controllers.
12211		 * So we can't pass a pointer to the S/G list, only as many
12212		 * S/G entries as we can fit in here.  If it's possible for
12213		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12214		 * then we need to break this up into multiple transfers.
12215		 */
12216		if (io->scsiio.kern_sg_entries == 0) {
12217			msg.dt.kern_sg_entries = 1;
12218			/*
12219			 * If this is in cached memory, flush the cache
12220			 * before we send the DMA request to the other
12221			 * controller.  We want to do this in either the
12222			 * read or the write case.  The read case is
12223			 * straightforward.  In the write case, we want to
12224			 * make sure nothing is in the local cache that
12225			 * could overwrite the DMAed data.
12226			 */
12227			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12228				/*
12229				 * XXX KDM use bus_dmamap_sync() here.
12230				 */
12231			}
12232
12233			/*
12234			 * Convert to a physical address if this is a
12235			 * virtual address.
12236			 */
12237			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12238				msg.dt.sg_list[0].addr =
12239					io->scsiio.kern_data_ptr;
12240			} else {
12241				/*
12242				 * XXX KDM use busdma here!
12243				 */
12244#if 0
12245				msg.dt.sg_list[0].addr = (void *)
12246					vtophys(io->scsiio.kern_data_ptr);
12247#endif
12248			}
12249
12250			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12251			do_sg_copy = 0;
12252		} else {
12253			struct ctl_sg_entry *sgl;
12254
12255			do_sg_copy = 1;
12256			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12257			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12258			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12259				/*
12260				 * XXX KDM use bus_dmamap_sync() here.
12261				 */
12262			}
12263		}
12264
12265		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12266		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12267		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12268		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12269		msg.dt.sg_sequence = 0;
12270
12271		/*
12272		 * Loop until we've sent all of the S/G entries.  On the
12273		 * other end, we'll recompose these S/G entries into one
12274		 * contiguous list before passing it to the
12275		 */
12276		for (sg_entries_sent = 0; sg_entries_sent <
12277		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12278			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12279				sizeof(msg.dt.sg_list[0])),
12280				msg.dt.kern_sg_entries - sg_entries_sent);
12281
12282			if (do_sg_copy != 0) {
12283				struct ctl_sg_entry *sgl;
12284				int j;
12285
12286				sgl = (struct ctl_sg_entry *)
12287					io->scsiio.kern_data_ptr;
12288				/*
12289				 * If this is in cached memory, flush the cache
12290				 * before we send the DMA request to the other
12291				 * controller.  We want to do this in either
12292				 * the * read or the write case.  The read
12293				 * case is straightforward.  In the write
12294				 * case, we want to make sure nothing is
12295				 * in the local cache that could overwrite
12296				 * the DMAed data.
12297				 */
12298
12299				for (i = sg_entries_sent, j = 0;
12300				     i < msg.dt.cur_sg_entries; i++, j++) {
12301					if ((io->io_hdr.flags &
12302					     CTL_FLAG_NO_DATASYNC) == 0) {
12303						/*
12304						 * XXX KDM use bus_dmamap_sync()
12305						 */
12306					}
12307					if ((io->io_hdr.flags &
12308					     CTL_FLAG_BUS_ADDR) == 0) {
12309						/*
12310						 * XXX KDM use busdma.
12311						 */
12312#if 0
12313						msg.dt.sg_list[j].addr =(void *)
12314						       vtophys(sgl[i].addr);
12315#endif
12316					} else {
12317						msg.dt.sg_list[j].addr =
12318							sgl[i].addr;
12319					}
12320					msg.dt.sg_list[j].len = sgl[i].len;
12321				}
12322			}
12323
12324			sg_entries_sent += msg.dt.cur_sg_entries;
12325			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12326				msg.dt.sg_last = 1;
12327			else
12328				msg.dt.sg_last = 0;
12329
12330			/*
12331			 * XXX KDM drop and reacquire the lock here?
12332			 */
12333			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12334			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12335				/*
12336				 * XXX do something here.
12337				 */
12338			}
12339
12340			msg.dt.sent_sg_entries = sg_entries_sent;
12341		}
12342		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12343		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12344			ctl_failover_io(io, /*have_lock*/ 1);
12345
12346	} else {
12347
12348		/*
12349		 * Lookup the fe_datamove() function for this particular
12350		 * front end.
12351		 */
12352		fe_datamove =
12353		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12354		mtx_unlock(&control_softc->ctl_lock);
12355
12356		fe_datamove(io);
12357	}
12358}
12359
12360static void
12361ctl_send_datamove_done(union ctl_io *io, int have_lock)
12362{
12363	union ctl_ha_msg msg;
12364	int isc_status;
12365
12366	memset(&msg, 0, sizeof(msg));
12367
12368	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12369	msg.hdr.original_sc = io;
12370	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12371	msg.hdr.nexus = io->io_hdr.nexus;
12372	msg.hdr.status = io->io_hdr.status;
12373	msg.scsi.tag_num = io->scsiio.tag_num;
12374	msg.scsi.tag_type = io->scsiio.tag_type;
12375	msg.scsi.scsi_status = io->scsiio.scsi_status;
12376	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12377	       sizeof(io->scsiio.sense_data));
12378	msg.scsi.sense_len = io->scsiio.sense_len;
12379	msg.scsi.sense_residual = io->scsiio.sense_residual;
12380	msg.scsi.fetd_status = io->io_hdr.port_status;
12381	msg.scsi.residual = io->scsiio.residual;
12382	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12383
12384	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12385		ctl_failover_io(io, /*have_lock*/ have_lock);
12386		return;
12387	}
12388
12389	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12390	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12391		/* XXX do something if this fails */
12392	}
12393
12394}
12395
12396/*
12397 * The DMA to the remote side is done, now we need to tell the other side
12398 * we're done so it can continue with its data movement.
12399 */
12400static void
12401ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12402{
12403	union ctl_io *io;
12404
12405	io = rq->context;
12406
12407	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12408		printf("%s: ISC DMA write failed with error %d", __func__,
12409		       rq->ret);
12410		ctl_set_internal_failure(&io->scsiio,
12411					 /*sks_valid*/ 1,
12412					 /*retry_count*/ rq->ret);
12413	}
12414
12415	ctl_dt_req_free(rq);
12416
12417	/*
12418	 * In this case, we had to malloc the memory locally.  Free it.
12419	 */
12420	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12421		int i;
12422		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12423			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12424	}
12425	/*
12426	 * The data is in local and remote memory, so now we need to send
12427	 * status (good or back) back to the other side.
12428	 */
12429	ctl_send_datamove_done(io, /*have_lock*/ 0);
12430}
12431
12432/*
12433 * We've moved the data from the host/controller into local memory.  Now we
12434 * need to push it over to the remote controller's memory.
12435 */
12436static int
12437ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12438{
12439	int retval;
12440
12441	retval = 0;
12442
12443	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12444					  ctl_datamove_remote_write_cb);
12445
12446	return (retval);
12447}
12448
12449static void
12450ctl_datamove_remote_write(union ctl_io *io)
12451{
12452	int retval;
12453	void (*fe_datamove)(union ctl_io *io);
12454
12455	/*
12456	 * - Get the data from the host/HBA into local memory.
12457	 * - DMA memory from the local controller to the remote controller.
12458	 * - Send status back to the remote controller.
12459	 */
12460
12461	retval = ctl_datamove_remote_sgl_setup(io);
12462	if (retval != 0)
12463		return;
12464
12465	/* Switch the pointer over so the FETD knows what to do */
12466	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12467
12468	/*
12469	 * Use a custom move done callback, since we need to send completion
12470	 * back to the other controller, not to the backend on this side.
12471	 */
12472	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12473
12474	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12475
12476	fe_datamove(io);
12477
12478	return;
12479
12480}
12481
12482static int
12483ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12484{
12485#if 0
12486	char str[256];
12487	char path_str[64];
12488	struct sbuf sb;
12489#endif
12490
12491	/*
12492	 * In this case, we had to malloc the memory locally.  Free it.
12493	 */
12494	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12495		int i;
12496		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12497			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12498	}
12499
12500#if 0
12501	scsi_path_string(io, path_str, sizeof(path_str));
12502	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12503	sbuf_cat(&sb, path_str);
12504	scsi_command_string(&io->scsiio, NULL, &sb);
12505	sbuf_printf(&sb, "\n");
12506	sbuf_cat(&sb, path_str);
12507	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12508		    io->scsiio.tag_num, io->scsiio.tag_type);
12509	sbuf_cat(&sb, path_str);
12510	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12511		    io->io_hdr.flags, io->io_hdr.status);
12512	sbuf_finish(&sb);
12513	printk("%s", sbuf_data(&sb));
12514#endif
12515
12516
12517	/*
12518	 * The read is done, now we need to send status (good or bad) back
12519	 * to the other side.
12520	 */
12521	ctl_send_datamove_done(io, /*have_lock*/ 0);
12522
12523	return (0);
12524}
12525
12526static void
12527ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12528{
12529	union ctl_io *io;
12530	void (*fe_datamove)(union ctl_io *io);
12531
12532	io = rq->context;
12533
12534	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12535		printf("%s: ISC DMA read failed with error %d", __func__,
12536		       rq->ret);
12537		ctl_set_internal_failure(&io->scsiio,
12538					 /*sks_valid*/ 1,
12539					 /*retry_count*/ rq->ret);
12540	}
12541
12542	ctl_dt_req_free(rq);
12543
12544	/* Switch the pointer over so the FETD knows what to do */
12545	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12546
12547	/*
12548	 * Use a custom move done callback, since we need to send completion
12549	 * back to the other controller, not to the backend on this side.
12550	 */
12551	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12552
12553	/* XXX KDM add checks like the ones in ctl_datamove? */
12554
12555	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12556
12557	fe_datamove(io);
12558}
12559
12560static int
12561ctl_datamove_remote_sgl_setup(union ctl_io *io)
12562{
12563	struct ctl_sg_entry *local_sglist, *remote_sglist;
12564	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12565	struct ctl_softc *softc;
12566	int retval;
12567	int i;
12568
12569	retval = 0;
12570	softc = control_softc;
12571
12572	local_sglist = io->io_hdr.local_sglist;
12573	local_dma_sglist = io->io_hdr.local_dma_sglist;
12574	remote_sglist = io->io_hdr.remote_sglist;
12575	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12576
12577	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12578		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12579			local_sglist[i].len = remote_sglist[i].len;
12580
12581			/*
12582			 * XXX Detect the situation where the RS-level I/O
12583			 * redirector on the other side has already read the
12584			 * data off of the AOR RS on this side, and
12585			 * transferred it to remote (mirror) memory on the
12586			 * other side.  Since we already have the data in
12587			 * memory here, we just need to use it.
12588			 *
12589			 * XXX KDM this can probably be removed once we
12590			 * get the cache device code in and take the
12591			 * current AOR implementation out.
12592			 */
12593#ifdef NEEDTOPORT
12594			if ((remote_sglist[i].addr >=
12595			     (void *)vtophys(softc->mirr->addr))
12596			 && (remote_sglist[i].addr <
12597			     ((void *)vtophys(softc->mirr->addr) +
12598			     CacheMirrorOffset))) {
12599				local_sglist[i].addr = remote_sglist[i].addr -
12600					CacheMirrorOffset;
12601				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12602				     CTL_FLAG_DATA_IN)
12603					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12604			} else {
12605				local_sglist[i].addr = remote_sglist[i].addr +
12606					CacheMirrorOffset;
12607			}
12608#endif
12609#if 0
12610			printf("%s: local %p, remote %p, len %d\n",
12611			       __func__, local_sglist[i].addr,
12612			       remote_sglist[i].addr, local_sglist[i].len);
12613#endif
12614		}
12615	} else {
12616		uint32_t len_to_go;
12617
12618		/*
12619		 * In this case, we don't have automatically allocated
12620		 * memory for this I/O on this controller.  This typically
12621		 * happens with internal CTL I/O -- e.g. inquiry, mode
12622		 * sense, etc.  Anything coming from RAIDCore will have
12623		 * a mirror area available.
12624		 */
12625		len_to_go = io->scsiio.kern_data_len;
12626
12627		/*
12628		 * Clear the no datasync flag, we have to use malloced
12629		 * buffers.
12630		 */
12631		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12632
12633		/*
12634		 * The difficult thing here is that the size of the various
12635		 * S/G segments may be different than the size from the
12636		 * remote controller.  That'll make it harder when DMAing
12637		 * the data back to the other side.
12638		 */
12639		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12640		     sizeof(io->io_hdr.remote_sglist[0])) &&
12641		     (len_to_go > 0); i++) {
12642			local_sglist[i].len = ctl_min(len_to_go, 131072);
12643			CTL_SIZE_8B(local_dma_sglist[i].len,
12644				    local_sglist[i].len);
12645			local_sglist[i].addr =
12646				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12647
12648			local_dma_sglist[i].addr = local_sglist[i].addr;
12649
12650			if (local_sglist[i].addr == NULL) {
12651				int j;
12652
12653				printf("malloc failed for %zd bytes!",
12654				       local_dma_sglist[i].len);
12655				for (j = 0; j < i; j++) {
12656					free(local_sglist[j].addr, M_CTL);
12657				}
12658				ctl_set_internal_failure(&io->scsiio,
12659							 /*sks_valid*/ 1,
12660							 /*retry_count*/ 4857);
12661				retval = 1;
12662				goto bailout_error;
12663
12664			}
12665			/* XXX KDM do we need a sync here? */
12666
12667			len_to_go -= local_sglist[i].len;
12668		}
12669		/*
12670		 * Reset the number of S/G entries accordingly.  The
12671		 * original number of S/G entries is available in
12672		 * rem_sg_entries.
12673		 */
12674		io->scsiio.kern_sg_entries = i;
12675
12676#if 0
12677		printf("%s: kern_sg_entries = %d\n", __func__,
12678		       io->scsiio.kern_sg_entries);
12679		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12680			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12681			       local_sglist[i].addr, local_sglist[i].len,
12682			       local_dma_sglist[i].len);
12683#endif
12684	}
12685
12686
12687	return (retval);
12688
12689bailout_error:
12690
12691	ctl_send_datamove_done(io, /*have_lock*/ 0);
12692
12693	return (retval);
12694}
12695
12696static int
12697ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12698			 ctl_ha_dt_cb callback)
12699{
12700	struct ctl_ha_dt_req *rq;
12701	struct ctl_sg_entry *remote_sglist, *local_sglist;
12702	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12703	uint32_t local_used, remote_used, total_used;
12704	int retval;
12705	int i, j;
12706
12707	retval = 0;
12708
12709	rq = ctl_dt_req_alloc();
12710
12711	/*
12712	 * If we failed to allocate the request, and if the DMA didn't fail
12713	 * anyway, set busy status.  This is just a resource allocation
12714	 * failure.
12715	 */
12716	if ((rq == NULL)
12717	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12718		ctl_set_busy(&io->scsiio);
12719
12720	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12721
12722		if (rq != NULL)
12723			ctl_dt_req_free(rq);
12724
12725		/*
12726		 * The data move failed.  We need to return status back
12727		 * to the other controller.  No point in trying to DMA
12728		 * data to the remote controller.
12729		 */
12730
12731		ctl_send_datamove_done(io, /*have_lock*/ 0);
12732
12733		retval = 1;
12734
12735		goto bailout;
12736	}
12737
12738	local_sglist = io->io_hdr.local_sglist;
12739	local_dma_sglist = io->io_hdr.local_dma_sglist;
12740	remote_sglist = io->io_hdr.remote_sglist;
12741	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12742	local_used = 0;
12743	remote_used = 0;
12744	total_used = 0;
12745
12746	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12747		rq->ret = CTL_HA_STATUS_SUCCESS;
12748		rq->context = io;
12749		callback(rq);
12750		goto bailout;
12751	}
12752
12753	/*
12754	 * Pull/push the data over the wire from/to the other controller.
12755	 * This takes into account the possibility that the local and
12756	 * remote sglists may not be identical in terms of the size of
12757	 * the elements and the number of elements.
12758	 *
12759	 * One fundamental assumption here is that the length allocated for
12760	 * both the local and remote sglists is identical.  Otherwise, we've
12761	 * essentially got a coding error of some sort.
12762	 */
12763	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12764		int isc_ret;
12765		uint32_t cur_len, dma_length;
12766		uint8_t *tmp_ptr;
12767
12768		rq->id = CTL_HA_DATA_CTL;
12769		rq->command = command;
12770		rq->context = io;
12771
12772		/*
12773		 * Both pointers should be aligned.  But it is possible
12774		 * that the allocation length is not.  They should both
12775		 * also have enough slack left over at the end, though,
12776		 * to round up to the next 8 byte boundary.
12777		 */
12778		cur_len = ctl_min(local_sglist[i].len - local_used,
12779				  remote_sglist[j].len - remote_used);
12780
12781		/*
12782		 * In this case, we have a size issue and need to decrease
12783		 * the size, except in the case where we actually have less
12784		 * than 8 bytes left.  In that case, we need to increase
12785		 * the DMA length to get the last bit.
12786		 */
12787		if ((cur_len & 0x7) != 0) {
12788			if (cur_len > 0x7) {
12789				cur_len = cur_len - (cur_len & 0x7);
12790				dma_length = cur_len;
12791			} else {
12792				CTL_SIZE_8B(dma_length, cur_len);
12793			}
12794
12795		} else
12796			dma_length = cur_len;
12797
12798		/*
12799		 * If we had to allocate memory for this I/O, instead of using
12800		 * the non-cached mirror memory, we'll need to flush the cache
12801		 * before trying to DMA to the other controller.
12802		 *
12803		 * We could end up doing this multiple times for the same
12804		 * segment if we have a larger local segment than remote
12805		 * segment.  That shouldn't be an issue.
12806		 */
12807		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12808			/*
12809			 * XXX KDM use bus_dmamap_sync() here.
12810			 */
12811		}
12812
12813		rq->size = dma_length;
12814
12815		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12816		tmp_ptr += local_used;
12817
12818		/* Use physical addresses when talking to ISC hardware */
12819		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12820			/* XXX KDM use busdma */
12821#if 0
12822			rq->local = vtophys(tmp_ptr);
12823#endif
12824		} else
12825			rq->local = tmp_ptr;
12826
12827		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12828		tmp_ptr += remote_used;
12829		rq->remote = tmp_ptr;
12830
12831		rq->callback = NULL;
12832
12833		local_used += cur_len;
12834		if (local_used >= local_sglist[i].len) {
12835			i++;
12836			local_used = 0;
12837		}
12838
12839		remote_used += cur_len;
12840		if (remote_used >= remote_sglist[j].len) {
12841			j++;
12842			remote_used = 0;
12843		}
12844		total_used += cur_len;
12845
12846		if (total_used >= io->scsiio.kern_data_len)
12847			rq->callback = callback;
12848
12849		if ((rq->size & 0x7) != 0) {
12850			printf("%s: warning: size %d is not on 8b boundary\n",
12851			       __func__, rq->size);
12852		}
12853		if (((uintptr_t)rq->local & 0x7) != 0) {
12854			printf("%s: warning: local %p not on 8b boundary\n",
12855			       __func__, rq->local);
12856		}
12857		if (((uintptr_t)rq->remote & 0x7) != 0) {
12858			printf("%s: warning: remote %p not on 8b boundary\n",
12859			       __func__, rq->local);
12860		}
12861#if 0
12862		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12863		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12864		       rq->local, rq->remote, rq->size);
12865#endif
12866
12867		isc_ret = ctl_dt_single(rq);
12868		if (isc_ret == CTL_HA_STATUS_WAIT)
12869			continue;
12870
12871		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12872			rq->ret = CTL_HA_STATUS_SUCCESS;
12873		} else {
12874			rq->ret = isc_ret;
12875		}
12876		callback(rq);
12877		goto bailout;
12878	}
12879
12880bailout:
12881	return (retval);
12882
12883}
12884
12885static void
12886ctl_datamove_remote_read(union ctl_io *io)
12887{
12888	int retval;
12889	int i;
12890
12891	/*
12892	 * This will send an error to the other controller in the case of a
12893	 * failure.
12894	 */
12895	retval = ctl_datamove_remote_sgl_setup(io);
12896	if (retval != 0)
12897		return;
12898
12899	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12900					  ctl_datamove_remote_read_cb);
12901	if ((retval != 0)
12902	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12903		/*
12904		 * Make sure we free memory if there was an error..  The
12905		 * ctl_datamove_remote_xfer() function will send the
12906		 * datamove done message, or call the callback with an
12907		 * error if there is a problem.
12908		 */
12909		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12910			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12911	}
12912
12913	return;
12914}
12915
12916/*
12917 * Process a datamove request from the other controller.  This is used for
12918 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12919 * first.  Once that is complete, the data gets DMAed into the remote
12920 * controller's memory.  For reads, we DMA from the remote controller's
12921 * memory into our memory first, and then move it out to the FETD.
12922 */
12923static void
12924ctl_datamove_remote(union ctl_io *io)
12925{
12926	struct ctl_softc *softc;
12927
12928	softc = control_softc;
12929
12930	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12931
12932	/*
12933	 * Note that we look for an aborted I/O here, but don't do some of
12934	 * the other checks that ctl_datamove() normally does.  We don't
12935	 * need to run the task queue, because this I/O is on the ISC
12936	 * queue, which is executed by the work thread after the task queue.
12937	 * We don't need to run the datamove delay code, since that should
12938	 * have been done if need be on the other controller.
12939	 */
12940	mtx_lock(&softc->ctl_lock);
12941
12942	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12943
12944		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12945		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12946		       io->io_hdr.nexus.targ_port,
12947		       io->io_hdr.nexus.targ_target.id,
12948		       io->io_hdr.nexus.targ_lun);
12949		io->io_hdr.status = CTL_CMD_ABORTED;
12950		io->io_hdr.port_status = 31338;
12951
12952		mtx_unlock(&softc->ctl_lock);
12953
12954		ctl_send_datamove_done(io, /*have_lock*/ 0);
12955
12956		return;
12957	}
12958
12959	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12960		mtx_unlock(&softc->ctl_lock);
12961		ctl_datamove_remote_write(io);
12962	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12963		mtx_unlock(&softc->ctl_lock);
12964		ctl_datamove_remote_read(io);
12965	} else {
12966		union ctl_ha_msg msg;
12967		struct scsi_sense_data *sense;
12968		uint8_t sks[3];
12969		int retry_count;
12970
12971		memset(&msg, 0, sizeof(msg));
12972
12973		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12974		msg.hdr.status = CTL_SCSI_ERROR;
12975		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12976
12977		retry_count = 4243;
12978
12979		sense = &msg.scsi.sense_data;
12980		sks[0] = SSD_SCS_VALID;
12981		sks[1] = (retry_count >> 8) & 0xff;
12982		sks[2] = retry_count & 0xff;
12983
12984		/* "Internal target failure" */
12985		scsi_set_sense_data(sense,
12986				    /*sense_format*/ SSD_TYPE_NONE,
12987				    /*current_error*/ 1,
12988				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12989				    /*asc*/ 0x44,
12990				    /*ascq*/ 0x00,
12991				    /*type*/ SSD_ELEM_SKS,
12992				    /*size*/ sizeof(sks),
12993				    /*data*/ sks,
12994				    SSD_ELEM_NONE);
12995
12996		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12997		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12998			ctl_failover_io(io, /*have_lock*/ 1);
12999			mtx_unlock(&softc->ctl_lock);
13000			return;
13001		}
13002
13003		mtx_unlock(&softc->ctl_lock);
13004
13005		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13006		    CTL_HA_STATUS_SUCCESS) {
13007			/* XXX KDM what to do if this fails? */
13008		}
13009		return;
13010	}
13011
13012}
13013
13014static int
13015ctl_process_done(union ctl_io *io, int have_lock)
13016{
13017	struct ctl_lun *lun;
13018	struct ctl_softc *ctl_softc;
13019	void (*fe_done)(union ctl_io *io);
13020	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13021
13022	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13023
13024	fe_done =
13025	    control_softc->ctl_ports[targ_port]->fe_done;
13026
13027#ifdef CTL_TIME_IO
13028	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13029		char str[256];
13030		char path_str[64];
13031		struct sbuf sb;
13032
13033		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13034		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13035
13036		sbuf_cat(&sb, path_str);
13037		switch (io->io_hdr.io_type) {
13038		case CTL_IO_SCSI:
13039			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13040			sbuf_printf(&sb, "\n");
13041			sbuf_cat(&sb, path_str);
13042			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13043				    io->scsiio.tag_num, io->scsiio.tag_type);
13044			break;
13045		case CTL_IO_TASK:
13046			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13047				    "Tag Type: %d\n", io->taskio.task_action,
13048				    io->taskio.tag_num, io->taskio.tag_type);
13049			break;
13050		default:
13051			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13052			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13053			break;
13054		}
13055		sbuf_cat(&sb, path_str);
13056		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13057			    (intmax_t)time_uptime - io->io_hdr.start_time);
13058		sbuf_finish(&sb);
13059		printf("%s", sbuf_data(&sb));
13060	}
13061#endif /* CTL_TIME_IO */
13062
13063	switch (io->io_hdr.io_type) {
13064	case CTL_IO_SCSI:
13065		break;
13066	case CTL_IO_TASK:
13067		if (bootverbose || verbose > 0)
13068			ctl_io_error_print(io, NULL);
13069		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13070			ctl_free_io(io);
13071		else
13072			fe_done(io);
13073		return (CTL_RETVAL_COMPLETE);
13074		break;
13075	default:
13076		printf("ctl_process_done: invalid io type %d\n",
13077		       io->io_hdr.io_type);
13078		panic("ctl_process_done: invalid io type %d\n",
13079		      io->io_hdr.io_type);
13080		break; /* NOTREACHED */
13081	}
13082
13083	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13084	if (lun == NULL) {
13085		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13086				 io->io_hdr.nexus.targ_lun));
13087		fe_done(io);
13088		goto bailout;
13089	}
13090	ctl_softc = lun->ctl_softc;
13091
13092	/*
13093	 * Remove this from the OOA queue.
13094	 */
13095	if (have_lock == 0)
13096		mtx_lock(&ctl_softc->ctl_lock);
13097
13098	/*
13099	 * Check to see if we have any errors to inject here.  We only
13100	 * inject errors for commands that don't already have errors set.
13101	 */
13102	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13103	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13104		ctl_inject_error(lun, io);
13105
13106	/*
13107	 * XXX KDM how do we treat commands that aren't completed
13108	 * successfully?
13109	 *
13110	 * XXX KDM should we also track I/O latency?
13111	 */
13112	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
13113		uint32_t blocksize;
13114#ifdef CTL_TIME_IO
13115		struct bintime cur_bt;
13116#endif
13117
13118		if ((lun->be_lun != NULL)
13119		 && (lun->be_lun->blocksize != 0))
13120			blocksize = lun->be_lun->blocksize;
13121		else
13122			blocksize = 512;
13123
13124		switch (io->io_hdr.io_type) {
13125		case CTL_IO_SCSI: {
13126			int isread;
13127			struct ctl_lba_len_flags *lbalen;
13128
13129			isread = 0;
13130			switch (io->scsiio.cdb[0]) {
13131			case READ_6:
13132			case READ_10:
13133			case READ_12:
13134			case READ_16:
13135				isread = 1;
13136				/* FALLTHROUGH */
13137			case WRITE_6:
13138			case WRITE_10:
13139			case WRITE_12:
13140			case WRITE_16:
13141			case WRITE_VERIFY_10:
13142			case WRITE_VERIFY_12:
13143			case WRITE_VERIFY_16:
13144				lbalen = (struct ctl_lba_len_flags *)
13145				    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
13146
13147				if (isread) {
13148					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
13149					    lbalen->len * blocksize;
13150					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
13151
13152#ifdef CTL_TIME_IO
13153					bintime_add(
13154					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
13155					   &io->io_hdr.dma_bt);
13156					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
13157						io->io_hdr.num_dmas;
13158					getbintime(&cur_bt);
13159					bintime_sub(&cur_bt,
13160						    &io->io_hdr.start_bt);
13161
13162					bintime_add(
13163					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
13164					    &cur_bt);
13165
13166#if 0
13167					cs_prof_gettime(&cur_ticks);
13168					lun->stats.time[CTL_STATS_READ] +=
13169						cur_ticks -
13170						io->io_hdr.start_ticks;
13171#endif
13172#if 0
13173					lun->stats.time[CTL_STATS_READ] +=
13174						jiffies - io->io_hdr.start_time;
13175#endif
13176#endif /* CTL_TIME_IO */
13177				} else {
13178					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
13179					    lbalen->len * blocksize;
13180					lun->stats.ports[targ_port].operations[
13181						CTL_STATS_WRITE]++;
13182
13183#ifdef CTL_TIME_IO
13184					bintime_add(
13185					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
13186					  &io->io_hdr.dma_bt);
13187					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
13188						io->io_hdr.num_dmas;
13189					getbintime(&cur_bt);
13190					bintime_sub(&cur_bt,
13191						    &io->io_hdr.start_bt);
13192
13193					bintime_add(
13194					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
13195					    &cur_bt);
13196#if 0
13197					cs_prof_gettime(&cur_ticks);
13198					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13199						cur_ticks -
13200						io->io_hdr.start_ticks;
13201					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13202						jiffies - io->io_hdr.start_time;
13203#endif
13204#endif /* CTL_TIME_IO */
13205				}
13206				break;
13207			default:
13208				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
13209
13210#ifdef CTL_TIME_IO
13211				bintime_add(
13212				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
13213				  &io->io_hdr.dma_bt);
13214				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
13215					io->io_hdr.num_dmas;
13216				getbintime(&cur_bt);
13217				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13218
13219				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
13220					    &cur_bt);
13221
13222#if 0
13223				cs_prof_gettime(&cur_ticks);
13224				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13225					cur_ticks -
13226					io->io_hdr.start_ticks;
13227				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13228					jiffies - io->io_hdr.start_time;
13229#endif
13230#endif /* CTL_TIME_IO */
13231				break;
13232			}
13233			break;
13234		}
13235		default:
13236			break;
13237		}
13238	}
13239
13240	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13241
13242	/*
13243	 * Run through the blocked queue on this LUN and see if anything
13244	 * has become unblocked, now that this transaction is done.
13245	 */
13246	ctl_check_blocked(lun);
13247
13248	/*
13249	 * If the LUN has been invalidated, free it if there is nothing
13250	 * left on its OOA queue.
13251	 */
13252	if ((lun->flags & CTL_LUN_INVALID)
13253	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
13254		ctl_free_lun(lun);
13255
13256	/*
13257	 * If this command has been aborted, make sure we set the status
13258	 * properly.  The FETD is responsible for freeing the I/O and doing
13259	 * whatever it needs to do to clean up its state.
13260	 */
13261	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13262		io->io_hdr.status = CTL_CMD_ABORTED;
13263
13264	/*
13265	 * We print out status for every task management command.  For SCSI
13266	 * commands, we filter out any unit attention errors; they happen
13267	 * on every boot, and would clutter up the log.  Note:  task
13268	 * management commands aren't printed here, they are printed above,
13269	 * since they should never even make it down here.
13270	 */
13271	switch (io->io_hdr.io_type) {
13272	case CTL_IO_SCSI: {
13273		int error_code, sense_key, asc, ascq;
13274
13275		sense_key = 0;
13276
13277		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13278		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13279			/*
13280			 * Since this is just for printing, no need to
13281			 * show errors here.
13282			 */
13283			scsi_extract_sense_len(&io->scsiio.sense_data,
13284					       io->scsiio.sense_len,
13285					       &error_code,
13286					       &sense_key,
13287					       &asc,
13288					       &ascq,
13289					       /*show_errors*/ 0);
13290		}
13291
13292		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13293		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13294		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13295		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13296
13297			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13298				ctl_softc->skipped_prints++;
13299				if (have_lock == 0)
13300					mtx_unlock(&ctl_softc->ctl_lock);
13301			} else {
13302				uint32_t skipped_prints;
13303
13304				skipped_prints = ctl_softc->skipped_prints;
13305
13306				ctl_softc->skipped_prints = 0;
13307				ctl_softc->last_print_jiffies = time_uptime;
13308
13309				if (have_lock == 0)
13310					mtx_unlock(&ctl_softc->ctl_lock);
13311				if (skipped_prints > 0) {
13312#ifdef NEEDTOPORT
13313					csevent_log(CSC_CTL | CSC_SHELF_SW |
13314					    CTL_ERROR_REPORT,
13315					    csevent_LogType_Trace,
13316					    csevent_Severity_Information,
13317					    csevent_AlertLevel_Green,
13318					    csevent_FRU_Firmware,
13319					    csevent_FRU_Unknown,
13320					    "High CTL error volume, %d prints "
13321					    "skipped", skipped_prints);
13322#endif
13323				}
13324				if (bootverbose || verbose > 0)
13325					ctl_io_error_print(io, NULL);
13326			}
13327		} else {
13328			if (have_lock == 0)
13329				mtx_unlock(&ctl_softc->ctl_lock);
13330		}
13331		break;
13332	}
13333	case CTL_IO_TASK:
13334		if (have_lock == 0)
13335			mtx_unlock(&ctl_softc->ctl_lock);
13336		if (bootverbose || verbose > 0)
13337			ctl_io_error_print(io, NULL);
13338		break;
13339	default:
13340		if (have_lock == 0)
13341			mtx_unlock(&ctl_softc->ctl_lock);
13342		break;
13343	}
13344
13345	/*
13346	 * Tell the FETD or the other shelf controller we're done with this
13347	 * command.  Note that only SCSI commands get to this point.  Task
13348	 * management commands are completed above.
13349	 *
13350	 * We only send status to the other controller if we're in XFER
13351	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13352	 * received the I/O (from CTL's perspective), and so the status is
13353	 * generated there.
13354	 *
13355	 * XXX KDM if we hold the lock here, we could cause a deadlock
13356	 * if the frontend comes back in in this context to queue
13357	 * something.
13358	 */
13359	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13360	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13361		union ctl_ha_msg msg;
13362
13363		memset(&msg, 0, sizeof(msg));
13364		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13365		msg.hdr.original_sc = io->io_hdr.original_sc;
13366		msg.hdr.nexus = io->io_hdr.nexus;
13367		msg.hdr.status = io->io_hdr.status;
13368		msg.scsi.scsi_status = io->scsiio.scsi_status;
13369		msg.scsi.tag_num = io->scsiio.tag_num;
13370		msg.scsi.tag_type = io->scsiio.tag_type;
13371		msg.scsi.sense_len = io->scsiio.sense_len;
13372		msg.scsi.sense_residual = io->scsiio.sense_residual;
13373		msg.scsi.residual = io->scsiio.residual;
13374		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13375		       sizeof(io->scsiio.sense_data));
13376		/*
13377		 * We copy this whether or not this is an I/O-related
13378		 * command.  Otherwise, we'd have to go and check to see
13379		 * whether it's a read/write command, and it really isn't
13380		 * worth it.
13381		 */
13382		memcpy(&msg.scsi.lbalen,
13383		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13384		       sizeof(msg.scsi.lbalen));
13385
13386		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13387				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13388			/* XXX do something here */
13389		}
13390
13391		ctl_free_io(io);
13392	} else
13393		fe_done(io);
13394
13395bailout:
13396
13397	return (CTL_RETVAL_COMPLETE);
13398}
13399
13400/*
13401 * Front end should call this if it doesn't do autosense.  When the request
13402 * sense comes back in from the initiator, we'll dequeue this and send it.
13403 */
13404int
13405ctl_queue_sense(union ctl_io *io)
13406{
13407	struct ctl_lun *lun;
13408	struct ctl_softc *ctl_softc;
13409	uint32_t initidx, targ_lun;
13410
13411	ctl_softc = control_softc;
13412
13413	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13414
13415	/*
13416	 * LUN lookup will likely move to the ctl_work_thread() once we
13417	 * have our new queueing infrastructure (that doesn't put things on
13418	 * a per-LUN queue initially).  That is so that we can handle
13419	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13420	 * can't deal with that right now.
13421	 */
13422	mtx_lock(&ctl_softc->ctl_lock);
13423
13424	/*
13425	 * If we don't have a LUN for this, just toss the sense
13426	 * information.
13427	 */
13428	targ_lun = io->io_hdr.nexus.targ_lun;
13429	if (io->io_hdr.nexus.lun_map_fn != NULL)
13430		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13431	if ((targ_lun < CTL_MAX_LUNS)
13432	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13433		lun = ctl_softc->ctl_luns[targ_lun];
13434	else
13435		goto bailout;
13436
13437	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13438
13439	/*
13440	 * Already have CA set for this LUN...toss the sense information.
13441	 */
13442	if (ctl_is_set(lun->have_ca, initidx))
13443		goto bailout;
13444
13445	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13446	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13447	       sizeof(io->scsiio.sense_data)));
13448	ctl_set_mask(lun->have_ca, initidx);
13449
13450bailout:
13451	mtx_unlock(&ctl_softc->ctl_lock);
13452
13453	ctl_free_io(io);
13454
13455	return (CTL_RETVAL_COMPLETE);
13456}
13457
13458/*
13459 * Primary command inlet from frontend ports.  All SCSI and task I/O
13460 * requests must go through this function.
13461 */
13462int
13463ctl_queue(union ctl_io *io)
13464{
13465	struct ctl_softc *ctl_softc;
13466
13467	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13468
13469	ctl_softc = control_softc;
13470
13471#ifdef CTL_TIME_IO
13472	io->io_hdr.start_time = time_uptime;
13473	getbintime(&io->io_hdr.start_bt);
13474#endif /* CTL_TIME_IO */
13475
13476	mtx_lock(&ctl_softc->ctl_lock);
13477
13478	switch (io->io_hdr.io_type) {
13479	case CTL_IO_SCSI:
13480		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
13481				   links);
13482		break;
13483	case CTL_IO_TASK:
13484		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
13485		/*
13486		 * Set the task pending flag.  This is necessary to close a
13487		 * race condition with the FETD:
13488		 *
13489		 * - FETD submits a task management command, like an abort.
13490		 * - Back end calls fe_datamove() to move the data for the
13491		 *   aborted command.  The FETD can't really accept it, but
13492		 *   if it did, it would end up transmitting data for a
13493		 *   command that the initiator told us to abort.
13494		 *
13495		 * We close the race condition by setting the flag here,
13496		 * and checking it in ctl_datamove(), before calling the
13497		 * FETD's fe_datamove routine.  If we've got a task
13498		 * pending, we run the task queue and then check to see
13499		 * whether our particular I/O has been aborted.
13500		 */
13501		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
13502		break;
13503	default:
13504		mtx_unlock(&ctl_softc->ctl_lock);
13505		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13506		return (EINVAL);
13507		break; /* NOTREACHED */
13508	}
13509	mtx_unlock(&ctl_softc->ctl_lock);
13510
13511	ctl_wakeup_thread();
13512
13513	return (CTL_RETVAL_COMPLETE);
13514}
13515
13516#ifdef CTL_IO_DELAY
13517static void
13518ctl_done_timer_wakeup(void *arg)
13519{
13520	union ctl_io *io;
13521
13522	io = (union ctl_io *)arg;
13523	ctl_done_lock(io, /*have_lock*/ 0);
13524}
13525#endif /* CTL_IO_DELAY */
13526
13527void
13528ctl_done_lock(union ctl_io *io, int have_lock)
13529{
13530	struct ctl_softc *ctl_softc;
13531#ifndef CTL_DONE_THREAD
13532	union ctl_io *xio;
13533#endif /* !CTL_DONE_THREAD */
13534
13535	ctl_softc = control_softc;
13536
13537	if (have_lock == 0)
13538		mtx_lock(&ctl_softc->ctl_lock);
13539
13540	/*
13541	 * Enable this to catch duplicate completion issues.
13542	 */
13543#if 0
13544	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13545		printf("%s: type %d msg %d cdb %x iptl: "
13546		       "%d:%d:%d:%d tag 0x%04x "
13547		       "flag %#x status %x\n",
13548			__func__,
13549			io->io_hdr.io_type,
13550			io->io_hdr.msg_type,
13551			io->scsiio.cdb[0],
13552			io->io_hdr.nexus.initid.id,
13553			io->io_hdr.nexus.targ_port,
13554			io->io_hdr.nexus.targ_target.id,
13555			io->io_hdr.nexus.targ_lun,
13556			(io->io_hdr.io_type ==
13557			CTL_IO_TASK) ?
13558			io->taskio.tag_num :
13559			io->scsiio.tag_num,
13560		        io->io_hdr.flags,
13561			io->io_hdr.status);
13562	} else
13563		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13564#endif
13565
13566	/*
13567	 * This is an internal copy of an I/O, and should not go through
13568	 * the normal done processing logic.
13569	 */
13570	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
13571		if (have_lock == 0)
13572			mtx_unlock(&ctl_softc->ctl_lock);
13573		return;
13574	}
13575
13576	/*
13577	 * We need to send a msg to the serializing shelf to finish the IO
13578	 * as well.  We don't send a finish message to the other shelf if
13579	 * this is a task management command.  Task management commands
13580	 * aren't serialized in the OOA queue, but rather just executed on
13581	 * both shelf controllers for commands that originated on that
13582	 * controller.
13583	 */
13584	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13585	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13586		union ctl_ha_msg msg_io;
13587
13588		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13589		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13590		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13591		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13592		}
13593		/* continue on to finish IO */
13594	}
13595#ifdef CTL_IO_DELAY
13596	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13597		struct ctl_lun *lun;
13598
13599		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13600
13601		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13602	} else {
13603		struct ctl_lun *lun;
13604
13605		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13606
13607		if ((lun != NULL)
13608		 && (lun->delay_info.done_delay > 0)) {
13609			struct callout *callout;
13610
13611			callout = (struct callout *)&io->io_hdr.timer_bytes;
13612			callout_init(callout, /*mpsafe*/ 1);
13613			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13614			callout_reset(callout,
13615				      lun->delay_info.done_delay * hz,
13616				      ctl_done_timer_wakeup, io);
13617			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13618				lun->delay_info.done_delay = 0;
13619			if (have_lock == 0)
13620				mtx_unlock(&ctl_softc->ctl_lock);
13621			return;
13622		}
13623	}
13624#endif /* CTL_IO_DELAY */
13625
13626	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
13627
13628#ifdef CTL_DONE_THREAD
13629	if (have_lock == 0)
13630		mtx_unlock(&ctl_softc->ctl_lock);
13631
13632	ctl_wakeup_thread();
13633#else /* CTL_DONE_THREAD */
13634	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
13635	     xio != NULL;
13636	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
13637
13638		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
13639
13640		ctl_process_done(xio, /*have_lock*/ 1);
13641	}
13642	if (have_lock == 0)
13643		mtx_unlock(&ctl_softc->ctl_lock);
13644#endif /* CTL_DONE_THREAD */
13645}
13646
13647void
13648ctl_done(union ctl_io *io)
13649{
13650	ctl_done_lock(io, /*have_lock*/ 0);
13651}
13652
13653int
13654ctl_isc(struct ctl_scsiio *ctsio)
13655{
13656	struct ctl_lun *lun;
13657	int retval;
13658
13659	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13660
13661	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13662
13663	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13664
13665	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13666
13667	return (retval);
13668}
13669
13670
13671static void
13672ctl_work_thread(void *arg)
13673{
13674	struct ctl_softc *softc;
13675	union ctl_io *io;
13676	struct ctl_be_lun *be_lun;
13677	int retval;
13678
13679	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13680
13681	softc = (struct ctl_softc *)arg;
13682	if (softc == NULL)
13683		return;
13684
13685	mtx_lock(&softc->ctl_lock);
13686	for (;;) {
13687		retval = 0;
13688
13689		/*
13690		 * We handle the queues in this order:
13691		 * - task management
13692		 * - ISC
13693		 * - done queue (to free up resources, unblock other commands)
13694		 * - RtR queue
13695		 * - incoming queue
13696		 *
13697		 * If those queues are empty, we break out of the loop and
13698		 * go to sleep.
13699		 */
13700		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
13701		if (io != NULL) {
13702			ctl_run_task_queue(softc);
13703			continue;
13704		}
13705		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
13706		if (io != NULL) {
13707			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
13708			ctl_handle_isc(io);
13709			continue;
13710		}
13711		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
13712		if (io != NULL) {
13713			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
13714			/* clear any blocked commands, call fe_done */
13715			mtx_unlock(&softc->ctl_lock);
13716			/*
13717			 * XXX KDM
13718			 * Call this without a lock for now.  This will
13719			 * depend on whether there is any way the FETD can
13720			 * sleep or deadlock if called with the CTL lock
13721			 * held.
13722			 */
13723			retval = ctl_process_done(io, /*have_lock*/ 0);
13724			mtx_lock(&softc->ctl_lock);
13725			continue;
13726		}
13727		if (!ctl_pause_rtr) {
13728			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
13729			if (io != NULL) {
13730				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
13731				mtx_unlock(&softc->ctl_lock);
13732				retval = ctl_scsiio(&io->scsiio);
13733				if (retval != CTL_RETVAL_COMPLETE)
13734					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13735				mtx_lock(&softc->ctl_lock);
13736				continue;
13737			}
13738		}
13739		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
13740		if (io != NULL) {
13741			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
13742			mtx_unlock(&softc->ctl_lock);
13743			ctl_scsiio_precheck(softc, &io->scsiio);
13744			mtx_lock(&softc->ctl_lock);
13745			continue;
13746		}
13747		/*
13748		 * We might want to move this to a separate thread, so that
13749		 * configuration requests (in this case LUN creations)
13750		 * won't impact the I/O path.
13751		 */
13752		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13753		if (be_lun != NULL) {
13754			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13755			mtx_unlock(&softc->ctl_lock);
13756			ctl_create_lun(be_lun);
13757			mtx_lock(&softc->ctl_lock);
13758			continue;
13759		}
13760
13761		/* XXX KDM use the PDROP flag?? */
13762		/* Sleep until we have something to do. */
13763		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "-", 0);
13764
13765		/* Back to the top of the loop to see what woke us up. */
13766		continue;
13767	}
13768}
13769
13770void
13771ctl_wakeup_thread()
13772{
13773	struct ctl_softc *softc;
13774
13775	softc = control_softc;
13776
13777	wakeup_one(softc);
13778}
13779
13780/* Initialization and failover */
13781
13782void
13783ctl_init_isc_msg(void)
13784{
13785	printf("CTL: Still calling this thing\n");
13786}
13787
13788/*
13789 * Init component
13790 * 	Initializes component into configuration defined by bootMode
13791 *	(see hasc-sv.c)
13792 *  	returns hasc_Status:
13793 * 		OK
13794 *		ERROR - fatal error
13795 */
13796static ctl_ha_comp_status
13797ctl_isc_init(struct ctl_ha_component *c)
13798{
13799	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13800
13801	c->status = ret;
13802	return ret;
13803}
13804
13805/* Start component
13806 * 	Starts component in state requested. If component starts successfully,
13807 *	it must set its own state to the requestrd state
13808 *	When requested state is HASC_STATE_HA, the component may refine it
13809 * 	by adding _SLAVE or _MASTER flags.
13810 *	Currently allowed state transitions are:
13811 *	UNKNOWN->HA		- initial startup
13812 *	UNKNOWN->SINGLE - initial startup when no parter detected
13813 *	HA->SINGLE		- failover
13814 * returns ctl_ha_comp_status:
13815 * 		OK	- component successfully started in requested state
13816 *		FAILED  - could not start the requested state, failover may
13817 * 			  be possible
13818 *		ERROR	- fatal error detected, no future startup possible
13819 */
13820static ctl_ha_comp_status
13821ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13822{
13823	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13824
13825	printf("%s: go\n", __func__);
13826
13827	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13828	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13829		ctl_is_single = 0;
13830		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13831		    != CTL_HA_STATUS_SUCCESS) {
13832			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13833			ret = CTL_HA_COMP_STATUS_ERROR;
13834		}
13835	} else if (CTL_HA_STATE_IS_HA(c->state)
13836		&& CTL_HA_STATE_IS_SINGLE(state)){
13837		// HA->SINGLE transition
13838	        ctl_failover();
13839		ctl_is_single = 1;
13840	} else {
13841		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13842		       c->state, state);
13843		ret = CTL_HA_COMP_STATUS_ERROR;
13844	}
13845	if (CTL_HA_STATE_IS_SINGLE(state))
13846		ctl_is_single = 1;
13847
13848	c->state = state;
13849	c->status = ret;
13850	return ret;
13851}
13852
13853/*
13854 * Quiesce component
13855 * The component must clear any error conditions (set status to OK) and
13856 * prepare itself to another Start call
13857 * returns ctl_ha_comp_status:
13858 * 	OK
13859 *	ERROR
13860 */
13861static ctl_ha_comp_status
13862ctl_isc_quiesce(struct ctl_ha_component *c)
13863{
13864	int ret = CTL_HA_COMP_STATUS_OK;
13865
13866	ctl_pause_rtr = 1;
13867	c->status = ret;
13868	return ret;
13869}
13870
13871struct ctl_ha_component ctl_ha_component_ctlisc =
13872{
13873	.name = "CTL ISC",
13874	.state = CTL_HA_STATE_UNKNOWN,
13875	.init = ctl_isc_init,
13876	.start = ctl_isc_start,
13877	.quiesce = ctl_isc_quiesce
13878};
13879
13880/*
13881 *  vim: ts=8
13882 */
13883