ctl.c revision 267639
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: head/sys/cam/ctl/ctl.c 267639 2014-06-19 09:46:43Z 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 * Size and alignment macros needed for Copan-specific HA hardware.  These
95 * can go away when the HA code is re-written, and uses busdma for any
96 * hardware.
97 */
98#define	CTL_ALIGN_8B(target, source, type)				\
99	if (((uint32_t)source & 0x7) != 0)				\
100		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
101	else								\
102		target = (type)source;
103
104#define	CTL_SIZE_8B(target, size)					\
105	if ((size & 0x7) != 0)						\
106		target = size + (0x8 - (size & 0x7));			\
107	else								\
108		target = size;
109
110#define CTL_ALIGN_8B_MARGIN	16
111
112/*
113 * Template mode pages.
114 */
115
116/*
117 * Note that these are default values only.  The actual values will be
118 * filled in when the user does a mode sense.
119 */
120static struct copan_power_subpage power_page_default = {
121	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
122	/*subpage*/ PWR_SUBPAGE_CODE,
123	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
124			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
125	/*page_version*/ PWR_VERSION,
126	/* total_luns */ 26,
127	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
128	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
129		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
130		      0, 0, 0, 0, 0, 0}
131};
132
133static struct copan_power_subpage power_page_changeable = {
134	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
135	/*subpage*/ PWR_SUBPAGE_CODE,
136	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
137			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
138	/*page_version*/ 0,
139	/* total_luns */ 0,
140	/* max_active_luns*/ 0,
141	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
142		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
143		      0, 0, 0, 0, 0, 0}
144};
145
146static struct copan_aps_subpage aps_page_default = {
147	APS_PAGE_CODE | SMPH_SPF, //page_code
148	APS_SUBPAGE_CODE, //subpage
149	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
150	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
151	APS_VERSION, //page_version
152	0, //lock_active
153	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
154	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
155	0, 0, 0, 0, 0} //reserved
156};
157
158static struct copan_aps_subpage aps_page_changeable = {
159	APS_PAGE_CODE | SMPH_SPF, //page_code
160	APS_SUBPAGE_CODE, //subpage
161	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
162	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
163	0, //page_version
164	0, //lock_active
165	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
166	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
167	0, 0, 0, 0, 0} //reserved
168};
169
170static struct copan_debugconf_subpage debugconf_page_default = {
171	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
172	DBGCNF_SUBPAGE_CODE,		/* subpage */
173	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
174	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
175	DBGCNF_VERSION,			/* page_version */
176	{CTL_TIME_IO_DEFAULT_SECS>>8,
177	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
178};
179
180static struct copan_debugconf_subpage debugconf_page_changeable = {
181	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
182	DBGCNF_SUBPAGE_CODE,		/* subpage */
183	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
184	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
185	0,				/* page_version */
186	{0xff,0xff},			/* ctl_time_io_secs */
187};
188
189static struct scsi_format_page format_page_default = {
190	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
191	/*page_length*/sizeof(struct scsi_format_page) - 2,
192	/*tracks_per_zone*/ {0, 0},
193	/*alt_sectors_per_zone*/ {0, 0},
194	/*alt_tracks_per_zone*/ {0, 0},
195	/*alt_tracks_per_lun*/ {0, 0},
196	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
197			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
198	/*bytes_per_sector*/ {0, 0},
199	/*interleave*/ {0, 0},
200	/*track_skew*/ {0, 0},
201	/*cylinder_skew*/ {0, 0},
202	/*flags*/ SFP_HSEC,
203	/*reserved*/ {0, 0, 0}
204};
205
206static struct scsi_format_page format_page_changeable = {
207	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
208	/*page_length*/sizeof(struct scsi_format_page) - 2,
209	/*tracks_per_zone*/ {0, 0},
210	/*alt_sectors_per_zone*/ {0, 0},
211	/*alt_tracks_per_zone*/ {0, 0},
212	/*alt_tracks_per_lun*/ {0, 0},
213	/*sectors_per_track*/ {0, 0},
214	/*bytes_per_sector*/ {0, 0},
215	/*interleave*/ {0, 0},
216	/*track_skew*/ {0, 0},
217	/*cylinder_skew*/ {0, 0},
218	/*flags*/ 0,
219	/*reserved*/ {0, 0, 0}
220};
221
222static struct scsi_rigid_disk_page rigid_disk_page_default = {
223	/*page_code*/SMS_RIGID_DISK_PAGE,
224	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
225	/*cylinders*/ {0, 0, 0},
226	/*heads*/ CTL_DEFAULT_HEADS,
227	/*start_write_precomp*/ {0, 0, 0},
228	/*start_reduced_current*/ {0, 0, 0},
229	/*step_rate*/ {0, 0},
230	/*landing_zone_cylinder*/ {0, 0, 0},
231	/*rpl*/ SRDP_RPL_DISABLED,
232	/*rotational_offset*/ 0,
233	/*reserved1*/ 0,
234	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
235			   CTL_DEFAULT_ROTATION_RATE & 0xff},
236	/*reserved2*/ {0, 0}
237};
238
239static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
240	/*page_code*/SMS_RIGID_DISK_PAGE,
241	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
242	/*cylinders*/ {0, 0, 0},
243	/*heads*/ 0,
244	/*start_write_precomp*/ {0, 0, 0},
245	/*start_reduced_current*/ {0, 0, 0},
246	/*step_rate*/ {0, 0},
247	/*landing_zone_cylinder*/ {0, 0, 0},
248	/*rpl*/ 0,
249	/*rotational_offset*/ 0,
250	/*reserved1*/ 0,
251	/*rotation_rate*/ {0, 0},
252	/*reserved2*/ {0, 0}
253};
254
255static struct scsi_caching_page caching_page_default = {
256	/*page_code*/SMS_CACHING_PAGE,
257	/*page_length*/sizeof(struct scsi_caching_page) - 2,
258	/*flags1*/ SCP_DISC | SCP_WCE,
259	/*ret_priority*/ 0,
260	/*disable_pf_transfer_len*/ {0xff, 0xff},
261	/*min_prefetch*/ {0, 0},
262	/*max_prefetch*/ {0xff, 0xff},
263	/*max_pf_ceiling*/ {0xff, 0xff},
264	/*flags2*/ 0,
265	/*cache_segments*/ 0,
266	/*cache_seg_size*/ {0, 0},
267	/*reserved*/ 0,
268	/*non_cache_seg_size*/ {0, 0, 0}
269};
270
271static struct scsi_caching_page caching_page_changeable = {
272	/*page_code*/SMS_CACHING_PAGE,
273	/*page_length*/sizeof(struct scsi_caching_page) - 2,
274	/*flags1*/ 0,
275	/*ret_priority*/ 0,
276	/*disable_pf_transfer_len*/ {0, 0},
277	/*min_prefetch*/ {0, 0},
278	/*max_prefetch*/ {0, 0},
279	/*max_pf_ceiling*/ {0, 0},
280	/*flags2*/ 0,
281	/*cache_segments*/ 0,
282	/*cache_seg_size*/ {0, 0},
283	/*reserved*/ 0,
284	/*non_cache_seg_size*/ {0, 0, 0}
285};
286
287static struct scsi_control_page control_page_default = {
288	/*page_code*/SMS_CONTROL_MODE_PAGE,
289	/*page_length*/sizeof(struct scsi_control_page) - 2,
290	/*rlec*/0,
291	/*queue_flags*/0,
292	/*eca_and_aen*/0,
293	/*reserved*/0,
294	/*aen_holdoff_period*/{0, 0}
295};
296
297static struct scsi_control_page control_page_changeable = {
298	/*page_code*/SMS_CONTROL_MODE_PAGE,
299	/*page_length*/sizeof(struct scsi_control_page) - 2,
300	/*rlec*/SCP_DSENSE,
301	/*queue_flags*/0,
302	/*eca_and_aen*/0,
303	/*reserved*/0,
304	/*aen_holdoff_period*/{0, 0}
305};
306
307
308/*
309 * XXX KDM move these into the softc.
310 */
311static int rcv_sync_msg;
312static int persis_offset;
313static uint8_t ctl_pause_rtr;
314static int     ctl_is_single = 1;
315static int     index_to_aps_page;
316
317SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
318static int worker_threads = 1;
319TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
320SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
321    &worker_threads, 1, "Number of worker threads");
322static int verbose = 0;
323TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
324SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
325    &verbose, 0, "Show SCSI errors returned to initiator");
326
327/*
328 * Serial number (0x80), device id (0x83), supported pages (0x00),
329 * Block limits (0xB0) and Logical Block Provisioning (0xB2)
330 */
331#define SCSI_EVPD_NUM_SUPPORTED_PAGES	5
332
333static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
334				  int param);
335static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
336static int ctl_init(void);
337void ctl_shutdown(void);
338static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
339static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
340static void ctl_ioctl_online(void *arg);
341static void ctl_ioctl_offline(void *arg);
342static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
343static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
344static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
345static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
346static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
347static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
348static int ctl_ioctl_submit_wait(union ctl_io *io);
349static void ctl_ioctl_datamove(union ctl_io *io);
350static void ctl_ioctl_done(union ctl_io *io);
351static void ctl_ioctl_hard_startstop_callback(void *arg,
352					      struct cfi_metatask *metatask);
353static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
354static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
355			      struct ctl_ooa *ooa_hdr,
356			      struct ctl_ooa_entry *kern_entries);
357static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
358		     struct thread *td);
359uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
360uint32_t ctl_port_idx(int port_num);
361#ifdef unused
362static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
363				   uint32_t targ_target, uint32_t targ_lun,
364				   int can_wait);
365static void ctl_kfree_io(union ctl_io *io);
366#endif /* unused */
367static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
368			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
369static int ctl_free_lun(struct ctl_lun *lun);
370static void ctl_create_lun(struct ctl_be_lun *be_lun);
371/**
372static void ctl_failover_change_pages(struct ctl_softc *softc,
373				      struct ctl_scsiio *ctsio, int master);
374**/
375
376static int ctl_do_mode_select(union ctl_io *io);
377static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
378			   uint64_t res_key, uint64_t sa_res_key,
379			   uint8_t type, uint32_t residx,
380			   struct ctl_scsiio *ctsio,
381			   struct scsi_per_res_out *cdb,
382			   struct scsi_per_res_out_parms* param);
383static void ctl_pro_preempt_other(struct ctl_lun *lun,
384				  union ctl_ha_msg *msg);
385static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
386static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
387static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
388static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
389static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
390					 int alloc_len);
391static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
392static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
393static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
394static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
395static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
396static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
397					 union ctl_io *ooa_io);
398static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
399				union ctl_io *starting_io);
400static int ctl_check_blocked(struct ctl_lun *lun);
401static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
402				struct ctl_lun *lun,
403				struct ctl_cmd_entry *entry,
404				struct ctl_scsiio *ctsio);
405//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
406static void ctl_failover(void);
407static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
408			       struct ctl_scsiio *ctsio);
409static int ctl_scsiio(struct ctl_scsiio *ctsio);
410
411static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
412static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
413			    ctl_ua_type ua_type);
414static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
415			 ctl_ua_type ua_type);
416static int ctl_abort_task(union ctl_io *io);
417static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
418#ifdef CTL_IO_DELAY
419static void ctl_datamove_timer_wakeup(void *arg);
420static void ctl_done_timer_wakeup(void *arg);
421#endif /* CTL_IO_DELAY */
422
423static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
424static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
425static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
426static void ctl_datamove_remote_write(union ctl_io *io);
427static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
428static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
429static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
430static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
431				    ctl_ha_dt_cb callback);
432static void ctl_datamove_remote_read(union ctl_io *io);
433static void ctl_datamove_remote(union ctl_io *io);
434static int ctl_process_done(union ctl_io *io, int have_lock);
435static void ctl_work_thread(void *arg);
436
437/*
438 * Load the serialization table.  This isn't very pretty, but is probably
439 * the easiest way to do it.
440 */
441#include "ctl_ser_table.c"
442
443/*
444 * We only need to define open, close and ioctl routines for this driver.
445 */
446static struct cdevsw ctl_cdevsw = {
447	.d_version =	D_VERSION,
448	.d_flags =	0,
449	.d_open =	ctl_open,
450	.d_close =	ctl_close,
451	.d_ioctl =	ctl_ioctl,
452	.d_name =	"ctl",
453};
454
455
456MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
457
458static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
459
460static moduledata_t ctl_moduledata = {
461	"ctl",
462	ctl_module_event_handler,
463	NULL
464};
465
466DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
467MODULE_VERSION(ctl, 1);
468
469static void
470ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
471			    union ctl_ha_msg *msg_info)
472{
473	struct ctl_scsiio *ctsio;
474
475	if (msg_info->hdr.original_sc == NULL) {
476		printf("%s: original_sc == NULL!\n", __func__);
477		/* XXX KDM now what? */
478		return;
479	}
480
481	ctsio = &msg_info->hdr.original_sc->scsiio;
482	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
483	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
484	ctsio->io_hdr.status = msg_info->hdr.status;
485	ctsio->scsi_status = msg_info->scsi.scsi_status;
486	ctsio->sense_len = msg_info->scsi.sense_len;
487	ctsio->sense_residual = msg_info->scsi.sense_residual;
488	ctsio->residual = msg_info->scsi.residual;
489	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
490	       sizeof(ctsio->sense_data));
491	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
492	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
493	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
494	ctl_wakeup_thread();
495}
496
497static void
498ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
499				union ctl_ha_msg *msg_info)
500{
501	struct ctl_scsiio *ctsio;
502
503	if (msg_info->hdr.serializing_sc == NULL) {
504		printf("%s: serializing_sc == NULL!\n", __func__);
505		/* XXX KDM now what? */
506		return;
507	}
508
509	ctsio = &msg_info->hdr.serializing_sc->scsiio;
510#if 0
511	/*
512	 * Attempt to catch the situation where an I/O has
513	 * been freed, and we're using it again.
514	 */
515	if (ctsio->io_hdr.io_type == 0xff) {
516		union ctl_io *tmp_io;
517		tmp_io = (union ctl_io *)ctsio;
518		printf("%s: %p use after free!\n", __func__,
519		       ctsio);
520		printf("%s: type %d msg %d cdb %x iptl: "
521		       "%d:%d:%d:%d tag 0x%04x "
522		       "flag %#x status %x\n",
523			__func__,
524			tmp_io->io_hdr.io_type,
525			tmp_io->io_hdr.msg_type,
526			tmp_io->scsiio.cdb[0],
527			tmp_io->io_hdr.nexus.initid.id,
528			tmp_io->io_hdr.nexus.targ_port,
529			tmp_io->io_hdr.nexus.targ_target.id,
530			tmp_io->io_hdr.nexus.targ_lun,
531			(tmp_io->io_hdr.io_type ==
532			CTL_IO_TASK) ?
533			tmp_io->taskio.tag_num :
534			tmp_io->scsiio.tag_num,
535		        tmp_io->io_hdr.flags,
536			tmp_io->io_hdr.status);
537	}
538#endif
539	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
540	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
541	ctl_wakeup_thread();
542}
543
544/*
545 * ISC (Inter Shelf Communication) event handler.  Events from the HA
546 * subsystem come in here.
547 */
548static void
549ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
550{
551	struct ctl_softc *ctl_softc;
552	union ctl_io *io;
553	struct ctl_prio *presio;
554	ctl_ha_status isc_status;
555
556	ctl_softc = control_softc;
557	io = NULL;
558
559
560#if 0
561	printf("CTL: Isc Msg event %d\n", event);
562#endif
563	if (event == CTL_HA_EVT_MSG_RECV) {
564		union ctl_ha_msg msg_info;
565
566		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
567					     sizeof(msg_info), /*wait*/ 0);
568#if 0
569		printf("CTL: msg_type %d\n", msg_info.msg_type);
570#endif
571		if (isc_status != 0) {
572			printf("Error receiving message, status = %d\n",
573			       isc_status);
574			return;
575		}
576		mtx_lock(&ctl_softc->ctl_lock);
577
578		switch (msg_info.hdr.msg_type) {
579		case CTL_MSG_SERIALIZE:
580#if 0
581			printf("Serialize\n");
582#endif
583			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
584			if (io == NULL) {
585				printf("ctl_isc_event_handler: can't allocate "
586				       "ctl_io!\n");
587				/* Bad Juju */
588				/* Need to set busy and send msg back */
589				mtx_unlock(&ctl_softc->ctl_lock);
590				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
591				msg_info.hdr.status = CTL_SCSI_ERROR;
592				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
593				msg_info.scsi.sense_len = 0;
594			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
595				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
596				}
597				goto bailout;
598			}
599			ctl_zero_io(io);
600			// populate ctsio from msg_info
601			io->io_hdr.io_type = CTL_IO_SCSI;
602			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
603			io->io_hdr.original_sc = msg_info.hdr.original_sc;
604#if 0
605			printf("pOrig %x\n", (int)msg_info.original_sc);
606#endif
607			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
608					    CTL_FLAG_IO_ACTIVE;
609			/*
610			 * If we're in serialization-only mode, we don't
611			 * want to go through full done processing.  Thus
612			 * the COPY flag.
613			 *
614			 * XXX KDM add another flag that is more specific.
615			 */
616			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
617				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
618			io->io_hdr.nexus = msg_info.hdr.nexus;
619#if 0
620			printf("targ %d, port %d, iid %d, lun %d\n",
621			       io->io_hdr.nexus.targ_target.id,
622			       io->io_hdr.nexus.targ_port,
623			       io->io_hdr.nexus.initid.id,
624			       io->io_hdr.nexus.targ_lun);
625#endif
626			io->scsiio.tag_num = msg_info.scsi.tag_num;
627			io->scsiio.tag_type = msg_info.scsi.tag_type;
628			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
629			       CTL_MAX_CDBLEN);
630			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
631				struct ctl_cmd_entry *entry;
632				uint8_t opcode;
633
634				opcode = io->scsiio.cdb[0];
635				entry = &ctl_cmd_table[opcode];
636				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
637				io->io_hdr.flags |=
638					entry->flags & CTL_FLAG_DATA_MASK;
639			}
640			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
641					   &io->io_hdr, links);
642			ctl_wakeup_thread();
643			break;
644
645		/* Performed on the Originating SC, XFER mode only */
646		case CTL_MSG_DATAMOVE: {
647			struct ctl_sg_entry *sgl;
648			int i, j;
649
650			io = msg_info.hdr.original_sc;
651			if (io == NULL) {
652				printf("%s: original_sc == NULL!\n", __func__);
653				/* XXX KDM do something here */
654				break;
655			}
656			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
657			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
658			/*
659			 * Keep track of this, we need to send it back over
660			 * when the datamove is complete.
661			 */
662			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
663
664			if (msg_info.dt.sg_sequence == 0) {
665				/*
666				 * XXX KDM we use the preallocated S/G list
667				 * here, but we'll need to change this to
668				 * dynamic allocation if we need larger S/G
669				 * lists.
670				 */
671				if (msg_info.dt.kern_sg_entries >
672				    sizeof(io->io_hdr.remote_sglist) /
673				    sizeof(io->io_hdr.remote_sglist[0])) {
674					printf("%s: number of S/G entries "
675					    "needed %u > allocated num %zd\n",
676					    __func__,
677					    msg_info.dt.kern_sg_entries,
678					    sizeof(io->io_hdr.remote_sglist)/
679					    sizeof(io->io_hdr.remote_sglist[0]));
680
681					/*
682					 * XXX KDM send a message back to
683					 * the other side to shut down the
684					 * DMA.  The error will come back
685					 * through via the normal channel.
686					 */
687					break;
688				}
689				sgl = io->io_hdr.remote_sglist;
690				memset(sgl, 0,
691				       sizeof(io->io_hdr.remote_sglist));
692
693				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
694
695				io->scsiio.kern_sg_entries =
696					msg_info.dt.kern_sg_entries;
697				io->scsiio.rem_sg_entries =
698					msg_info.dt.kern_sg_entries;
699				io->scsiio.kern_data_len =
700					msg_info.dt.kern_data_len;
701				io->scsiio.kern_total_len =
702					msg_info.dt.kern_total_len;
703				io->scsiio.kern_data_resid =
704					msg_info.dt.kern_data_resid;
705				io->scsiio.kern_rel_offset =
706					msg_info.dt.kern_rel_offset;
707				/*
708				 * Clear out per-DMA flags.
709				 */
710				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
711				/*
712				 * Add per-DMA flags that are set for this
713				 * particular DMA request.
714				 */
715				io->io_hdr.flags |= msg_info.dt.flags &
716						    CTL_FLAG_RDMA_MASK;
717			} else
718				sgl = (struct ctl_sg_entry *)
719					io->scsiio.kern_data_ptr;
720
721			for (i = msg_info.dt.sent_sg_entries, j = 0;
722			     i < (msg_info.dt.sent_sg_entries +
723			     msg_info.dt.cur_sg_entries); i++, j++) {
724				sgl[i].addr = msg_info.dt.sg_list[j].addr;
725				sgl[i].len = msg_info.dt.sg_list[j].len;
726
727#if 0
728				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
729				       __func__,
730				       msg_info.dt.sg_list[j].addr,
731				       msg_info.dt.sg_list[j].len,
732				       sgl[i].addr, sgl[i].len, j, i);
733#endif
734			}
735#if 0
736			memcpy(&sgl[msg_info.dt.sent_sg_entries],
737			       msg_info.dt.sg_list,
738			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
739#endif
740
741			/*
742			 * If this is the last piece of the I/O, we've got
743			 * the full S/G list.  Queue processing in the thread.
744			 * Otherwise wait for the next piece.
745			 */
746			if (msg_info.dt.sg_last != 0) {
747				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
748						   &io->io_hdr, links);
749				ctl_wakeup_thread();
750			}
751			break;
752		}
753		/* Performed on the Serializing (primary) SC, XFER mode only */
754		case CTL_MSG_DATAMOVE_DONE: {
755			if (msg_info.hdr.serializing_sc == NULL) {
756				printf("%s: serializing_sc == NULL!\n",
757				       __func__);
758				/* XXX KDM now what? */
759				break;
760			}
761			/*
762			 * We grab the sense information here in case
763			 * there was a failure, so we can return status
764			 * back to the initiator.
765			 */
766			io = msg_info.hdr.serializing_sc;
767			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
768			io->io_hdr.status = msg_info.hdr.status;
769			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
770			io->scsiio.sense_len = msg_info.scsi.sense_len;
771			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
772			io->io_hdr.port_status = msg_info.scsi.fetd_status;
773			io->scsiio.residual = msg_info.scsi.residual;
774			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
775			       sizeof(io->scsiio.sense_data));
776
777			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
778					   &io->io_hdr, links);
779			ctl_wakeup_thread();
780			break;
781		}
782
783		/* Preformed on Originating SC, SER_ONLY mode */
784		case CTL_MSG_R2R:
785			io = msg_info.hdr.original_sc;
786			if (io == NULL) {
787				printf("%s: Major Bummer\n", __func__);
788				mtx_unlock(&ctl_softc->ctl_lock);
789				return;
790			} else {
791#if 0
792				printf("pOrig %x\n",(int) ctsio);
793#endif
794			}
795			io->io_hdr.msg_type = CTL_MSG_R2R;
796			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
797			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
798					   &io->io_hdr, links);
799			ctl_wakeup_thread();
800			break;
801
802		/*
803		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
804		 * mode.
805		 * Performed on the Originating (i.e. secondary) SC in XFER
806		 * mode
807		 */
808		case CTL_MSG_FINISH_IO:
809			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
810				ctl_isc_handler_finish_xfer(ctl_softc,
811							    &msg_info);
812			else
813				ctl_isc_handler_finish_ser_only(ctl_softc,
814								&msg_info);
815			break;
816
817		/* Preformed on Originating SC */
818		case CTL_MSG_BAD_JUJU:
819			io = msg_info.hdr.original_sc;
820			if (io == NULL) {
821				printf("%s: Bad JUJU!, original_sc is NULL!\n",
822				       __func__);
823				break;
824			}
825			ctl_copy_sense_data(&msg_info, io);
826			/*
827			 * IO should have already been cleaned up on other
828			 * SC so clear this flag so we won't send a message
829			 * back to finish the IO there.
830			 */
831			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
832			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
833
834			/* io = msg_info.hdr.serializing_sc; */
835			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
836		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
837					   &io->io_hdr, links);
838			ctl_wakeup_thread();
839			break;
840
841		/* Handle resets sent from the other side */
842		case CTL_MSG_MANAGE_TASKS: {
843			struct ctl_taskio *taskio;
844			taskio = (struct ctl_taskio *)ctl_alloc_io(
845				(void *)ctl_softc->othersc_pool);
846			if (taskio == NULL) {
847				printf("ctl_isc_event_handler: can't allocate "
848				       "ctl_io!\n");
849				/* Bad Juju */
850				/* should I just call the proper reset func
851				   here??? */
852				mtx_unlock(&ctl_softc->ctl_lock);
853				goto bailout;
854			}
855			ctl_zero_io((union ctl_io *)taskio);
856			taskio->io_hdr.io_type = CTL_IO_TASK;
857			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
858			taskio->io_hdr.nexus = msg_info.hdr.nexus;
859			taskio->task_action = msg_info.task.task_action;
860			taskio->tag_num = msg_info.task.tag_num;
861			taskio->tag_type = msg_info.task.tag_type;
862#ifdef CTL_TIME_IO
863			taskio->io_hdr.start_time = time_uptime;
864			getbintime(&taskio->io_hdr.start_bt);
865#if 0
866			cs_prof_gettime(&taskio->io_hdr.start_ticks);
867#endif
868#endif /* CTL_TIME_IO */
869		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
870					   &taskio->io_hdr, links);
871			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
872			ctl_wakeup_thread();
873			break;
874		}
875		/* Persistent Reserve action which needs attention */
876		case CTL_MSG_PERS_ACTION:
877			presio = (struct ctl_prio *)ctl_alloc_io(
878				(void *)ctl_softc->othersc_pool);
879			if (presio == NULL) {
880				printf("ctl_isc_event_handler: can't allocate "
881				       "ctl_io!\n");
882				/* Bad Juju */
883				/* Need to set busy and send msg back */
884				mtx_unlock(&ctl_softc->ctl_lock);
885				goto bailout;
886			}
887			ctl_zero_io((union ctl_io *)presio);
888			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
889			presio->pr_msg = msg_info.pr;
890		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
891					   &presio->io_hdr, links);
892			ctl_wakeup_thread();
893			break;
894		case CTL_MSG_SYNC_FE:
895			rcv_sync_msg = 1;
896			break;
897		case CTL_MSG_APS_LOCK: {
898			// It's quicker to execute this then to
899			// queue it.
900			struct ctl_lun *lun;
901			struct ctl_page_index *page_index;
902			struct copan_aps_subpage *current_sp;
903			uint32_t targ_lun;
904
905			targ_lun = msg_info.hdr.nexus.targ_lun;
906			if (msg_info.hdr.nexus.lun_map_fn != NULL)
907				targ_lun = msg_info.hdr.nexus.lun_map_fn(msg_info.hdr.nexus.lun_map_arg, targ_lun);
908
909			lun = ctl_softc->ctl_luns[targ_lun];
910			page_index = &lun->mode_pages.index[index_to_aps_page];
911			current_sp = (struct copan_aps_subpage *)
912				     (page_index->page_data +
913				     (page_index->page_len * CTL_PAGE_CURRENT));
914
915			current_sp->lock_active = msg_info.aps.lock_flag;
916		        break;
917		}
918		default:
919		        printf("How did I get here?\n");
920		}
921		mtx_unlock(&ctl_softc->ctl_lock);
922	} else if (event == CTL_HA_EVT_MSG_SENT) {
923		if (param != CTL_HA_STATUS_SUCCESS) {
924			printf("Bad status from ctl_ha_msg_send status %d\n",
925			       param);
926		}
927		return;
928	} else if (event == CTL_HA_EVT_DISCONNECT) {
929		printf("CTL: Got a disconnect from Isc\n");
930		return;
931	} else {
932		printf("ctl_isc_event_handler: Unknown event %d\n", event);
933		return;
934	}
935
936bailout:
937	return;
938}
939
940static void
941ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
942{
943	struct scsi_sense_data *sense;
944
945	sense = &dest->scsiio.sense_data;
946	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
947	dest->scsiio.scsi_status = src->scsi.scsi_status;
948	dest->scsiio.sense_len = src->scsi.sense_len;
949	dest->io_hdr.status = src->hdr.status;
950}
951
952static int
953ctl_init(void)
954{
955	struct ctl_softc *softc;
956	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
957	struct ctl_frontend *fe;
958        uint8_t sc_id =0;
959	int i, error, retval;
960	//int isc_retval;
961
962	retval = 0;
963	ctl_pause_rtr = 0;
964        rcv_sync_msg = 0;
965
966	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
967			       M_WAITOK | M_ZERO);
968	softc = control_softc;
969
970	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
971			      "cam/ctl");
972
973	softc->dev->si_drv1 = softc;
974
975	/*
976	 * By default, return a "bad LUN" peripheral qualifier for unknown
977	 * LUNs.  The user can override this default using the tunable or
978	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
979	 */
980	softc->inquiry_pq_no_lun = 1;
981	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
982			  &softc->inquiry_pq_no_lun);
983	sysctl_ctx_init(&softc->sysctl_ctx);
984	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
985		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
986		CTLFLAG_RD, 0, "CAM Target Layer");
987
988	if (softc->sysctl_tree == NULL) {
989		printf("%s: unable to allocate sysctl tree\n", __func__);
990		destroy_dev(softc->dev);
991		free(control_softc, M_DEVBUF);
992		control_softc = NULL;
993		return (ENOMEM);
994	}
995
996	SYSCTL_ADD_INT(&softc->sysctl_ctx,
997		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
998		       "inquiry_pq_no_lun", CTLFLAG_RW,
999		       &softc->inquiry_pq_no_lun, 0,
1000		       "Report no lun possible for invalid LUNs");
1001
1002	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1003	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
1004	softc->open_count = 0;
1005
1006	/*
1007	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1008	 * the drive.
1009	 */
1010	softc->flags = CTL_FLAG_REAL_SYNC;
1011
1012	/*
1013	 * In Copan's HA scheme, the "master" and "slave" roles are
1014	 * figured out through the slot the controller is in.  Although it
1015	 * is an active/active system, someone has to be in charge.
1016 	 */
1017#ifdef NEEDTOPORT
1018        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1019#endif
1020
1021        if (sc_id == 0) {
1022		softc->flags |= CTL_FLAG_MASTER_SHELF;
1023		persis_offset = 0;
1024	} else
1025		persis_offset = CTL_MAX_INITIATORS;
1026
1027	/*
1028	 * XXX KDM need to figure out where we want to get our target ID
1029	 * and WWID.  Is it different on each port?
1030	 */
1031	softc->target.id = 0;
1032	softc->target.wwid[0] = 0x12345678;
1033	softc->target.wwid[1] = 0x87654321;
1034	STAILQ_INIT(&softc->lun_list);
1035	STAILQ_INIT(&softc->pending_lun_queue);
1036	STAILQ_INIT(&softc->task_queue);
1037	STAILQ_INIT(&softc->incoming_queue);
1038	STAILQ_INIT(&softc->rtr_queue);
1039	STAILQ_INIT(&softc->done_queue);
1040	STAILQ_INIT(&softc->isc_queue);
1041	STAILQ_INIT(&softc->fe_list);
1042	STAILQ_INIT(&softc->be_list);
1043	STAILQ_INIT(&softc->io_pools);
1044
1045	/*
1046	 * We don't bother calling these with ctl_lock held here, because,
1047	 * in theory, no one else can try to do anything while we're in our
1048	 * module init routine.
1049	 */
1050	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1051			    &internal_pool)!= 0){
1052		printf("ctl: can't allocate %d entry internal pool, "
1053		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1054		return (ENOMEM);
1055	}
1056
1057	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1058			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1059		printf("ctl: can't allocate %d entry emergency pool, "
1060		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1061		ctl_pool_free(internal_pool);
1062		return (ENOMEM);
1063	}
1064
1065	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1066	                    &other_pool) != 0)
1067	{
1068		printf("ctl: can't allocate %d entry other SC pool, "
1069		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1070		ctl_pool_free(internal_pool);
1071		ctl_pool_free(emergency_pool);
1072		return (ENOMEM);
1073	}
1074
1075	softc->internal_pool = internal_pool;
1076	softc->emergency_pool = emergency_pool;
1077	softc->othersc_pool = other_pool;
1078
1079	if (worker_threads > MAXCPU || worker_threads == 0) {
1080		printf("invalid kern.cam.ctl.worker_threads value; "
1081		    "setting to 1");
1082		worker_threads = 1;
1083	} else if (worker_threads < 0) {
1084		if (mp_ncpus > 2) {
1085			/*
1086			 * Using more than two worker threads actually hurts
1087			 * performance due to lock contention.
1088			 */
1089			worker_threads = 2;
1090		} else {
1091			worker_threads = 1;
1092		}
1093	}
1094
1095	for (i = 0; i < worker_threads; i++) {
1096		error = kproc_kthread_add(ctl_work_thread, softc,
1097		    &softc->work_thread, NULL, 0, 0, "ctl", "work%d", i);
1098		if (error != 0) {
1099			printf("error creating CTL work thread!\n");
1100			ctl_pool_free(internal_pool);
1101			ctl_pool_free(emergency_pool);
1102			ctl_pool_free(other_pool);
1103			return (error);
1104		}
1105	}
1106	if (bootverbose)
1107		printf("ctl: CAM Target Layer loaded\n");
1108
1109	/*
1110	 * Initialize the initiator and portname mappings
1111	 */
1112	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1113
1114	/*
1115	 * Initialize the ioctl front end.
1116	 */
1117	fe = &softc->ioctl_info.fe;
1118	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1119	fe->port_type = CTL_PORT_IOCTL;
1120	fe->num_requested_ctl_io = 100;
1121	fe->port_name = softc->ioctl_info.port_name;
1122	fe->port_online = ctl_ioctl_online;
1123	fe->port_offline = ctl_ioctl_offline;
1124	fe->onoff_arg = &softc->ioctl_info;
1125	fe->targ_enable = ctl_ioctl_targ_enable;
1126	fe->targ_disable = ctl_ioctl_targ_disable;
1127	fe->lun_enable = ctl_ioctl_lun_enable;
1128	fe->lun_disable = ctl_ioctl_lun_disable;
1129	fe->targ_lun_arg = &softc->ioctl_info;
1130	fe->fe_datamove = ctl_ioctl_datamove;
1131	fe->fe_done = ctl_ioctl_done;
1132	fe->max_targets = 15;
1133	fe->max_target_id = 15;
1134
1135	if (ctl_frontend_register(&softc->ioctl_info.fe,
1136	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1137		printf("ctl: ioctl front end registration failed, will "
1138		       "continue anyway\n");
1139	}
1140
1141#ifdef CTL_IO_DELAY
1142	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1143		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1144		       sizeof(struct callout), CTL_TIMER_BYTES);
1145		return (EINVAL);
1146	}
1147#endif /* CTL_IO_DELAY */
1148
1149	return (0);
1150}
1151
1152void
1153ctl_shutdown(void)
1154{
1155	struct ctl_softc *softc;
1156	struct ctl_lun *lun, *next_lun;
1157	struct ctl_io_pool *pool;
1158
1159	softc = (struct ctl_softc *)control_softc;
1160
1161	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1162		printf("ctl: ioctl front end deregistration failed\n");
1163
1164	mtx_lock(&softc->ctl_lock);
1165
1166	/*
1167	 * Free up each LUN.
1168	 */
1169	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1170		next_lun = STAILQ_NEXT(lun, links);
1171		ctl_free_lun(lun);
1172	}
1173
1174	mtx_unlock(&softc->ctl_lock);
1175
1176	/*
1177	 * This will rip the rug out from under any FETDs or anyone else
1178	 * that has a pool allocated.  Since we increment our module
1179	 * refcount any time someone outside the main CTL module allocates
1180	 * a pool, we shouldn't have any problems here.  The user won't be
1181	 * able to unload the CTL module until client modules have
1182	 * successfully unloaded.
1183	 */
1184	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1185		ctl_pool_free(pool);
1186
1187#if 0
1188	ctl_shutdown_thread(softc->work_thread);
1189#endif
1190
1191	mtx_destroy(&softc->pool_lock);
1192	mtx_destroy(&softc->ctl_lock);
1193
1194	destroy_dev(softc->dev);
1195
1196	sysctl_ctx_free(&softc->sysctl_ctx);
1197
1198	free(control_softc, M_DEVBUF);
1199	control_softc = NULL;
1200
1201	if (bootverbose)
1202		printf("ctl: CAM Target Layer unloaded\n");
1203}
1204
1205static int
1206ctl_module_event_handler(module_t mod, int what, void *arg)
1207{
1208
1209	switch (what) {
1210	case MOD_LOAD:
1211		return (ctl_init());
1212	case MOD_UNLOAD:
1213		return (EBUSY);
1214	default:
1215		return (EOPNOTSUPP);
1216	}
1217}
1218
1219/*
1220 * XXX KDM should we do some access checks here?  Bump a reference count to
1221 * prevent a CTL module from being unloaded while someone has it open?
1222 */
1223static int
1224ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1225{
1226	return (0);
1227}
1228
1229static int
1230ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1231{
1232	return (0);
1233}
1234
1235int
1236ctl_port_enable(ctl_port_type port_type)
1237{
1238	struct ctl_softc *softc;
1239	struct ctl_frontend *fe;
1240
1241	if (ctl_is_single == 0) {
1242		union ctl_ha_msg msg_info;
1243		int isc_retval;
1244
1245#if 0
1246		printf("%s: HA mode, synchronizing frontend enable\n",
1247		        __func__);
1248#endif
1249		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1250	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1251		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1252			printf("Sync msg send error retval %d\n", isc_retval);
1253		}
1254		if (!rcv_sync_msg) {
1255			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1256			        sizeof(msg_info), 1);
1257		}
1258#if 0
1259        	printf("CTL:Frontend Enable\n");
1260	} else {
1261		printf("%s: single mode, skipping frontend synchronization\n",
1262		        __func__);
1263#endif
1264	}
1265
1266	softc = control_softc;
1267
1268	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1269		if (port_type & fe->port_type)
1270		{
1271#if 0
1272			printf("port %d\n", fe->targ_port);
1273#endif
1274			ctl_frontend_online(fe);
1275		}
1276	}
1277
1278	return (0);
1279}
1280
1281int
1282ctl_port_disable(ctl_port_type port_type)
1283{
1284	struct ctl_softc *softc;
1285	struct ctl_frontend *fe;
1286
1287	softc = control_softc;
1288
1289	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1290		if (port_type & fe->port_type)
1291			ctl_frontend_offline(fe);
1292	}
1293
1294	return (0);
1295}
1296
1297/*
1298 * Returns 0 for success, 1 for failure.
1299 * Currently the only failure mode is if there aren't enough entries
1300 * allocated.  So, in case of a failure, look at num_entries_dropped,
1301 * reallocate and try again.
1302 */
1303int
1304ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1305	      int *num_entries_filled, int *num_entries_dropped,
1306	      ctl_port_type port_type, int no_virtual)
1307{
1308	struct ctl_softc *softc;
1309	struct ctl_frontend *fe;
1310	int entries_dropped, entries_filled;
1311	int retval;
1312	int i;
1313
1314	softc = control_softc;
1315
1316	retval = 0;
1317	entries_filled = 0;
1318	entries_dropped = 0;
1319
1320	i = 0;
1321	mtx_lock(&softc->ctl_lock);
1322	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1323		struct ctl_port_entry *entry;
1324
1325		if ((fe->port_type & port_type) == 0)
1326			continue;
1327
1328		if ((no_virtual != 0)
1329		 && (fe->virtual_port != 0))
1330			continue;
1331
1332		if (entries_filled >= num_entries_alloced) {
1333			entries_dropped++;
1334			continue;
1335		}
1336		entry = &entries[i];
1337
1338		entry->port_type = fe->port_type;
1339		strlcpy(entry->port_name, fe->port_name,
1340			sizeof(entry->port_name));
1341		entry->physical_port = fe->physical_port;
1342		entry->virtual_port = fe->virtual_port;
1343		entry->wwnn = fe->wwnn;
1344		entry->wwpn = fe->wwpn;
1345
1346		i++;
1347		entries_filled++;
1348	}
1349
1350	mtx_unlock(&softc->ctl_lock);
1351
1352	if (entries_dropped > 0)
1353		retval = 1;
1354
1355	*num_entries_dropped = entries_dropped;
1356	*num_entries_filled = entries_filled;
1357
1358	return (retval);
1359}
1360
1361static void
1362ctl_ioctl_online(void *arg)
1363{
1364	struct ctl_ioctl_info *ioctl_info;
1365
1366	ioctl_info = (struct ctl_ioctl_info *)arg;
1367
1368	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1369}
1370
1371static void
1372ctl_ioctl_offline(void *arg)
1373{
1374	struct ctl_ioctl_info *ioctl_info;
1375
1376	ioctl_info = (struct ctl_ioctl_info *)arg;
1377
1378	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1379}
1380
1381/*
1382 * Remove an initiator by port number and initiator ID.
1383 * Returns 0 for success, 1 for failure.
1384 */
1385int
1386ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1387{
1388	struct ctl_softc *softc;
1389
1390	softc = control_softc;
1391
1392	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1393
1394	if ((targ_port < 0)
1395	 || (targ_port > CTL_MAX_PORTS)) {
1396		printf("%s: invalid port number %d\n", __func__, targ_port);
1397		return (1);
1398	}
1399	if (iid > CTL_MAX_INIT_PER_PORT) {
1400		printf("%s: initiator ID %u > maximun %u!\n",
1401		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1402		return (1);
1403	}
1404
1405	mtx_lock(&softc->ctl_lock);
1406
1407	softc->wwpn_iid[targ_port][iid].in_use = 0;
1408
1409	mtx_unlock(&softc->ctl_lock);
1410
1411	return (0);
1412}
1413
1414/*
1415 * Add an initiator to the initiator map.
1416 * Returns 0 for success, 1 for failure.
1417 */
1418int
1419ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1420{
1421	struct ctl_softc *softc;
1422	int retval;
1423
1424	softc = control_softc;
1425
1426	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1427
1428	retval = 0;
1429
1430	if ((targ_port < 0)
1431	 || (targ_port > CTL_MAX_PORTS)) {
1432		printf("%s: invalid port number %d\n", __func__, targ_port);
1433		return (1);
1434	}
1435	if (iid > CTL_MAX_INIT_PER_PORT) {
1436		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1437		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1438		return (1);
1439	}
1440
1441	mtx_lock(&softc->ctl_lock);
1442
1443	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1444		/*
1445		 * We don't treat this as an error.
1446		 */
1447		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1448			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1449			       __func__, targ_port, iid, (uintmax_t)wwpn);
1450			goto bailout;
1451		}
1452
1453		/*
1454		 * This is an error, but what do we do about it?  The
1455		 * driver is telling us we have a new WWPN for this
1456		 * initiator ID, so we pretty much need to use it.
1457		 */
1458		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1459		       "still at that address\n", __func__, targ_port, iid,
1460		       (uintmax_t)wwpn,
1461		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1462
1463		/*
1464		 * XXX KDM clear have_ca and ua_pending on each LUN for
1465		 * this initiator.
1466		 */
1467	}
1468	softc->wwpn_iid[targ_port][iid].in_use = 1;
1469	softc->wwpn_iid[targ_port][iid].iid = iid;
1470	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1471	softc->wwpn_iid[targ_port][iid].port = targ_port;
1472
1473bailout:
1474
1475	mtx_unlock(&softc->ctl_lock);
1476
1477	return (retval);
1478}
1479
1480/*
1481 * XXX KDM should we pretend to do something in the target/lun
1482 * enable/disable functions?
1483 */
1484static int
1485ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1486{
1487	return (0);
1488}
1489
1490static int
1491ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1492{
1493	return (0);
1494}
1495
1496static int
1497ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1498{
1499	return (0);
1500}
1501
1502static int
1503ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1504{
1505	return (0);
1506}
1507
1508/*
1509 * Data movement routine for the CTL ioctl frontend port.
1510 */
1511static int
1512ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1513{
1514	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1515	struct ctl_sg_entry ext_entry, kern_entry;
1516	int ext_sglen, ext_sg_entries, kern_sg_entries;
1517	int ext_sg_start, ext_offset;
1518	int len_to_copy, len_copied;
1519	int kern_watermark, ext_watermark;
1520	int ext_sglist_malloced;
1521	int i, j;
1522
1523	ext_sglist_malloced = 0;
1524	ext_sg_start = 0;
1525	ext_offset = 0;
1526
1527	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1528
1529	/*
1530	 * If this flag is set, fake the data transfer.
1531	 */
1532	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1533		ctsio->ext_data_filled = ctsio->ext_data_len;
1534		goto bailout;
1535	}
1536
1537	/*
1538	 * To simplify things here, if we have a single buffer, stick it in
1539	 * a S/G entry and just make it a single entry S/G list.
1540	 */
1541	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1542		int len_seen;
1543
1544		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1545
1546		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1547							   M_WAITOK);
1548		ext_sglist_malloced = 1;
1549		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1550				   ext_sglen) != 0) {
1551			ctl_set_internal_failure(ctsio,
1552						 /*sks_valid*/ 0,
1553						 /*retry_count*/ 0);
1554			goto bailout;
1555		}
1556		ext_sg_entries = ctsio->ext_sg_entries;
1557		len_seen = 0;
1558		for (i = 0; i < ext_sg_entries; i++) {
1559			if ((len_seen + ext_sglist[i].len) >=
1560			     ctsio->ext_data_filled) {
1561				ext_sg_start = i;
1562				ext_offset = ctsio->ext_data_filled - len_seen;
1563				break;
1564			}
1565			len_seen += ext_sglist[i].len;
1566		}
1567	} else {
1568		ext_sglist = &ext_entry;
1569		ext_sglist->addr = ctsio->ext_data_ptr;
1570		ext_sglist->len = ctsio->ext_data_len;
1571		ext_sg_entries = 1;
1572		ext_sg_start = 0;
1573		ext_offset = ctsio->ext_data_filled;
1574	}
1575
1576	if (ctsio->kern_sg_entries > 0) {
1577		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1578		kern_sg_entries = ctsio->kern_sg_entries;
1579	} else {
1580		kern_sglist = &kern_entry;
1581		kern_sglist->addr = ctsio->kern_data_ptr;
1582		kern_sglist->len = ctsio->kern_data_len;
1583		kern_sg_entries = 1;
1584	}
1585
1586
1587	kern_watermark = 0;
1588	ext_watermark = ext_offset;
1589	len_copied = 0;
1590	for (i = ext_sg_start, j = 0;
1591	     i < ext_sg_entries && j < kern_sg_entries;) {
1592		uint8_t *ext_ptr, *kern_ptr;
1593
1594		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1595				      kern_sglist[j].len - kern_watermark);
1596
1597		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1598		ext_ptr = ext_ptr + ext_watermark;
1599		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1600			/*
1601			 * XXX KDM fix this!
1602			 */
1603			panic("need to implement bus address support");
1604#if 0
1605			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1606#endif
1607		} else
1608			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1609		kern_ptr = kern_ptr + kern_watermark;
1610
1611		kern_watermark += len_to_copy;
1612		ext_watermark += len_to_copy;
1613
1614		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1615		     CTL_FLAG_DATA_IN) {
1616			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1617					 "bytes to user\n", len_to_copy));
1618			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1619					 "to %p\n", kern_ptr, ext_ptr));
1620			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1621				ctl_set_internal_failure(ctsio,
1622							 /*sks_valid*/ 0,
1623							 /*retry_count*/ 0);
1624				goto bailout;
1625			}
1626		} else {
1627			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1628					 "bytes from user\n", len_to_copy));
1629			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1630					 "to %p\n", ext_ptr, kern_ptr));
1631			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1632				ctl_set_internal_failure(ctsio,
1633							 /*sks_valid*/ 0,
1634							 /*retry_count*/0);
1635				goto bailout;
1636			}
1637		}
1638
1639		len_copied += len_to_copy;
1640
1641		if (ext_sglist[i].len == ext_watermark) {
1642			i++;
1643			ext_watermark = 0;
1644		}
1645
1646		if (kern_sglist[j].len == kern_watermark) {
1647			j++;
1648			kern_watermark = 0;
1649		}
1650	}
1651
1652	ctsio->ext_data_filled += len_copied;
1653
1654	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1655			 "kern_sg_entries: %d\n", ext_sg_entries,
1656			 kern_sg_entries));
1657	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1658			 "kern_data_len = %d\n", ctsio->ext_data_len,
1659			 ctsio->kern_data_len));
1660
1661
1662	/* XXX KDM set residual?? */
1663bailout:
1664
1665	if (ext_sglist_malloced != 0)
1666		free(ext_sglist, M_CTL);
1667
1668	return (CTL_RETVAL_COMPLETE);
1669}
1670
1671/*
1672 * Serialize a command that went down the "wrong" side, and so was sent to
1673 * this controller for execution.  The logic is a little different than the
1674 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1675 * sent back to the other side, but in the success case, we execute the
1676 * command on this side (XFER mode) or tell the other side to execute it
1677 * (SER_ONLY mode).
1678 */
1679static int
1680ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1681{
1682	struct ctl_softc *ctl_softc;
1683	union ctl_ha_msg msg_info;
1684	struct ctl_lun *lun;
1685	int retval = 0;
1686	uint32_t targ_lun;
1687
1688	ctl_softc = control_softc;
1689	if (have_lock == 0)
1690		mtx_lock(&ctl_softc->ctl_lock);
1691
1692	targ_lun = ctsio->io_hdr.nexus.targ_lun;
1693	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
1694		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
1695	lun = ctl_softc->ctl_luns[targ_lun];
1696	if (lun==NULL)
1697	{
1698		/*
1699		 * Why isn't LUN defined? The other side wouldn't
1700		 * send a cmd if the LUN is undefined.
1701		 */
1702		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1703
1704		/* "Logical unit not supported" */
1705		ctl_set_sense_data(&msg_info.scsi.sense_data,
1706				   lun,
1707				   /*sense_format*/SSD_TYPE_NONE,
1708				   /*current_error*/ 1,
1709				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1710				   /*asc*/ 0x25,
1711				   /*ascq*/ 0x00,
1712				   SSD_ELEM_NONE);
1713
1714		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1715		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1716		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1717		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1718		msg_info.hdr.serializing_sc = NULL;
1719		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1720	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1721				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1722		}
1723		if (have_lock == 0)
1724			mtx_unlock(&ctl_softc->ctl_lock);
1725		return(1);
1726
1727	}
1728
1729    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1730
1731	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1732		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1733		 ooa_links))) {
1734	case CTL_ACTION_BLOCK:
1735		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1736		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1737				  blocked_links);
1738		break;
1739	case CTL_ACTION_PASS:
1740	case CTL_ACTION_SKIP:
1741		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1742			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1743			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1744					   &ctsio->io_hdr, links);
1745		} else {
1746
1747			/* send msg back to other side */
1748			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1749			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1750			msg_info.hdr.msg_type = CTL_MSG_R2R;
1751#if 0
1752			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1753#endif
1754		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1755			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1756			}
1757		}
1758		break;
1759	case CTL_ACTION_OVERLAP:
1760		/* OVERLAPPED COMMANDS ATTEMPTED */
1761		ctl_set_sense_data(&msg_info.scsi.sense_data,
1762				   lun,
1763				   /*sense_format*/SSD_TYPE_NONE,
1764				   /*current_error*/ 1,
1765				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1766				   /*asc*/ 0x4E,
1767				   /*ascq*/ 0x00,
1768				   SSD_ELEM_NONE);
1769
1770		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1771		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1772		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1773		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1774		msg_info.hdr.serializing_sc = NULL;
1775		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1776#if 0
1777		printf("BAD JUJU:Major Bummer Overlap\n");
1778#endif
1779		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1780		retval = 1;
1781		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1782		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1783		}
1784		break;
1785	case CTL_ACTION_OVERLAP_TAG:
1786		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1787		ctl_set_sense_data(&msg_info.scsi.sense_data,
1788				   lun,
1789				   /*sense_format*/SSD_TYPE_NONE,
1790				   /*current_error*/ 1,
1791				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1792				   /*asc*/ 0x4D,
1793				   /*ascq*/ ctsio->tag_num & 0xff,
1794				   SSD_ELEM_NONE);
1795
1796		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1797		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1798		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1799		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1800		msg_info.hdr.serializing_sc = NULL;
1801		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1802#if 0
1803		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1804#endif
1805		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1806		retval = 1;
1807		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1808		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1809		}
1810		break;
1811	case CTL_ACTION_ERROR:
1812	default:
1813		/* "Internal target failure" */
1814		ctl_set_sense_data(&msg_info.scsi.sense_data,
1815				   lun,
1816				   /*sense_format*/SSD_TYPE_NONE,
1817				   /*current_error*/ 1,
1818				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1819				   /*asc*/ 0x44,
1820				   /*ascq*/ 0x00,
1821				   SSD_ELEM_NONE);
1822
1823		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1824		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1825		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1826		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1827		msg_info.hdr.serializing_sc = NULL;
1828		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1829#if 0
1830		printf("BAD JUJU:Major Bummer HW Error\n");
1831#endif
1832		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1833		retval = 1;
1834		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1835		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1836		}
1837		break;
1838	}
1839	if (have_lock == 0)
1840		mtx_unlock(&ctl_softc->ctl_lock);
1841	return (retval);
1842}
1843
1844static int
1845ctl_ioctl_submit_wait(union ctl_io *io)
1846{
1847	struct ctl_fe_ioctl_params params;
1848	ctl_fe_ioctl_state last_state;
1849	int done, retval;
1850
1851	retval = 0;
1852
1853	bzero(&params, sizeof(params));
1854
1855	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1856	cv_init(&params.sem, "ctlioccv");
1857	params.state = CTL_IOCTL_INPROG;
1858	last_state = params.state;
1859
1860	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1861
1862	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1863
1864	/* This shouldn't happen */
1865	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1866		return (retval);
1867
1868	done = 0;
1869
1870	do {
1871		mtx_lock(&params.ioctl_mtx);
1872		/*
1873		 * Check the state here, and don't sleep if the state has
1874		 * already changed (i.e. wakeup has already occured, but we
1875		 * weren't waiting yet).
1876		 */
1877		if (params.state == last_state) {
1878			/* XXX KDM cv_wait_sig instead? */
1879			cv_wait(&params.sem, &params.ioctl_mtx);
1880		}
1881		last_state = params.state;
1882
1883		switch (params.state) {
1884		case CTL_IOCTL_INPROG:
1885			/* Why did we wake up? */
1886			/* XXX KDM error here? */
1887			mtx_unlock(&params.ioctl_mtx);
1888			break;
1889		case CTL_IOCTL_DATAMOVE:
1890			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1891
1892			/*
1893			 * change last_state back to INPROG to avoid
1894			 * deadlock on subsequent data moves.
1895			 */
1896			params.state = last_state = CTL_IOCTL_INPROG;
1897
1898			mtx_unlock(&params.ioctl_mtx);
1899			ctl_ioctl_do_datamove(&io->scsiio);
1900			/*
1901			 * Note that in some cases, most notably writes,
1902			 * this will queue the I/O and call us back later.
1903			 * In other cases, generally reads, this routine
1904			 * will immediately call back and wake us up,
1905			 * probably using our own context.
1906			 */
1907			io->scsiio.be_move_done(io);
1908			break;
1909		case CTL_IOCTL_DONE:
1910			mtx_unlock(&params.ioctl_mtx);
1911			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1912			done = 1;
1913			break;
1914		default:
1915			mtx_unlock(&params.ioctl_mtx);
1916			/* XXX KDM error here? */
1917			break;
1918		}
1919	} while (done == 0);
1920
1921	mtx_destroy(&params.ioctl_mtx);
1922	cv_destroy(&params.sem);
1923
1924	return (CTL_RETVAL_COMPLETE);
1925}
1926
1927static void
1928ctl_ioctl_datamove(union ctl_io *io)
1929{
1930	struct ctl_fe_ioctl_params *params;
1931
1932	params = (struct ctl_fe_ioctl_params *)
1933		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1934
1935	mtx_lock(&params->ioctl_mtx);
1936	params->state = CTL_IOCTL_DATAMOVE;
1937	cv_broadcast(&params->sem);
1938	mtx_unlock(&params->ioctl_mtx);
1939}
1940
1941static void
1942ctl_ioctl_done(union ctl_io *io)
1943{
1944	struct ctl_fe_ioctl_params *params;
1945
1946	params = (struct ctl_fe_ioctl_params *)
1947		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1948
1949	mtx_lock(&params->ioctl_mtx);
1950	params->state = CTL_IOCTL_DONE;
1951	cv_broadcast(&params->sem);
1952	mtx_unlock(&params->ioctl_mtx);
1953}
1954
1955static void
1956ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1957{
1958	struct ctl_fe_ioctl_startstop_info *sd_info;
1959
1960	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1961
1962	sd_info->hs_info.status = metatask->status;
1963	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1964	sd_info->hs_info.luns_complete =
1965		metatask->taskinfo.startstop.luns_complete;
1966	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1967
1968	cv_broadcast(&sd_info->sem);
1969}
1970
1971static void
1972ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1973{
1974	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1975
1976	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1977
1978	mtx_lock(fe_bbr_info->lock);
1979	fe_bbr_info->bbr_info->status = metatask->status;
1980	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1981	fe_bbr_info->wakeup_done = 1;
1982	mtx_unlock(fe_bbr_info->lock);
1983
1984	cv_broadcast(&fe_bbr_info->sem);
1985}
1986
1987/*
1988 * Returns 0 for success, errno for failure.
1989 */
1990static int
1991ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1992		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1993{
1994	union ctl_io *io;
1995	int retval;
1996
1997	retval = 0;
1998
1999	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
2000
2001	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2002	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2003	     ooa_links)) {
2004		struct ctl_ooa_entry *entry;
2005
2006		/*
2007		 * If we've got more than we can fit, just count the
2008		 * remaining entries.
2009		 */
2010		if (*cur_fill_num >= ooa_hdr->alloc_num)
2011			continue;
2012
2013		entry = &kern_entries[*cur_fill_num];
2014
2015		entry->tag_num = io->scsiio.tag_num;
2016		entry->lun_num = lun->lun;
2017#ifdef CTL_TIME_IO
2018		entry->start_bt = io->io_hdr.start_bt;
2019#endif
2020		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2021		entry->cdb_len = io->scsiio.cdb_len;
2022		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2023			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2024
2025		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2026			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2027
2028		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2029			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2030
2031		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2032			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2033
2034		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2035			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2036	}
2037
2038	return (retval);
2039}
2040
2041static void *
2042ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2043		 size_t error_str_len)
2044{
2045	void *kptr;
2046
2047	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2048
2049	if (copyin(user_addr, kptr, len) != 0) {
2050		snprintf(error_str, error_str_len, "Error copying %d bytes "
2051			 "from user address %p to kernel address %p", len,
2052			 user_addr, kptr);
2053		free(kptr, M_CTL);
2054		return (NULL);
2055	}
2056
2057	return (kptr);
2058}
2059
2060static void
2061ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2062{
2063	int i;
2064
2065	if (be_args == NULL)
2066		return;
2067
2068	for (i = 0; i < num_be_args; i++) {
2069		free(be_args[i].kname, M_CTL);
2070		free(be_args[i].kvalue, M_CTL);
2071	}
2072
2073	free(be_args, M_CTL);
2074}
2075
2076static struct ctl_be_arg *
2077ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2078		char *error_str, size_t error_str_len)
2079{
2080	struct ctl_be_arg *args;
2081	int i;
2082
2083	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2084				error_str, error_str_len);
2085
2086	if (args == NULL)
2087		goto bailout;
2088
2089	for (i = 0; i < num_be_args; i++) {
2090		args[i].kname = NULL;
2091		args[i].kvalue = NULL;
2092	}
2093
2094	for (i = 0; i < num_be_args; i++) {
2095		uint8_t *tmpptr;
2096
2097		args[i].kname = ctl_copyin_alloc(args[i].name,
2098			args[i].namelen, error_str, error_str_len);
2099		if (args[i].kname == NULL)
2100			goto bailout;
2101
2102		if (args[i].kname[args[i].namelen - 1] != '\0') {
2103			snprintf(error_str, error_str_len, "Argument %d "
2104				 "name is not NUL-terminated", i);
2105			goto bailout;
2106		}
2107
2108		args[i].kvalue = NULL;
2109
2110		tmpptr = ctl_copyin_alloc(args[i].value,
2111			args[i].vallen, error_str, error_str_len);
2112		if (tmpptr == NULL)
2113			goto bailout;
2114
2115		args[i].kvalue = tmpptr;
2116
2117		if ((args[i].flags & CTL_BEARG_ASCII)
2118		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2119			snprintf(error_str, error_str_len, "Argument %d "
2120				 "value is not NUL-terminated", i);
2121			goto bailout;
2122		}
2123	}
2124
2125	return (args);
2126bailout:
2127
2128	ctl_free_args(num_be_args, args);
2129
2130	return (NULL);
2131}
2132
2133/*
2134 * Escape characters that are illegal or not recommended in XML.
2135 */
2136int
2137ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2138{
2139	int retval;
2140
2141	retval = 0;
2142
2143	for (; *str; str++) {
2144		switch (*str) {
2145		case '&':
2146			retval = sbuf_printf(sb, "&amp;");
2147			break;
2148		case '>':
2149			retval = sbuf_printf(sb, "&gt;");
2150			break;
2151		case '<':
2152			retval = sbuf_printf(sb, "&lt;");
2153			break;
2154		default:
2155			retval = sbuf_putc(sb, *str);
2156			break;
2157		}
2158
2159		if (retval != 0)
2160			break;
2161
2162	}
2163
2164	return (retval);
2165}
2166
2167static int
2168ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2169	  struct thread *td)
2170{
2171	struct ctl_softc *softc;
2172	int retval;
2173
2174	softc = control_softc;
2175
2176	retval = 0;
2177
2178	switch (cmd) {
2179	case CTL_IO: {
2180		union ctl_io *io;
2181		void *pool_tmp;
2182
2183		/*
2184		 * If we haven't been "enabled", don't allow any SCSI I/O
2185		 * to this FETD.
2186		 */
2187		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2188			retval = -EPERM;
2189			break;
2190		}
2191
2192		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2193		if (io == NULL) {
2194			printf("ctl_ioctl: can't allocate ctl_io!\n");
2195			retval = -ENOSPC;
2196			break;
2197		}
2198
2199		/*
2200		 * Need to save the pool reference so it doesn't get
2201		 * spammed by the user's ctl_io.
2202		 */
2203		pool_tmp = io->io_hdr.pool;
2204
2205		memcpy(io, (void *)addr, sizeof(*io));
2206
2207		io->io_hdr.pool = pool_tmp;
2208		/*
2209		 * No status yet, so make sure the status is set properly.
2210		 */
2211		io->io_hdr.status = CTL_STATUS_NONE;
2212
2213		/*
2214		 * The user sets the initiator ID, target and LUN IDs.
2215		 */
2216		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2217		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2218		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2219		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2220			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2221
2222		retval = ctl_ioctl_submit_wait(io);
2223
2224		if (retval != 0) {
2225			ctl_free_io(io);
2226			break;
2227		}
2228
2229		memcpy((void *)addr, io, sizeof(*io));
2230
2231		/* return this to our pool */
2232		ctl_free_io(io);
2233
2234		break;
2235	}
2236	case CTL_ENABLE_PORT:
2237	case CTL_DISABLE_PORT:
2238	case CTL_SET_PORT_WWNS: {
2239		struct ctl_frontend *fe;
2240		struct ctl_port_entry *entry;
2241
2242		entry = (struct ctl_port_entry *)addr;
2243
2244		mtx_lock(&softc->ctl_lock);
2245		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2246			int action, done;
2247
2248			action = 0;
2249			done = 0;
2250
2251			if ((entry->port_type == CTL_PORT_NONE)
2252			 && (entry->targ_port == fe->targ_port)) {
2253				/*
2254				 * If the user only wants to enable or
2255				 * disable or set WWNs on a specific port,
2256				 * do the operation and we're done.
2257				 */
2258				action = 1;
2259				done = 1;
2260			} else if (entry->port_type & fe->port_type) {
2261				/*
2262				 * Compare the user's type mask with the
2263				 * particular frontend type to see if we
2264				 * have a match.
2265				 */
2266				action = 1;
2267				done = 0;
2268
2269				/*
2270				 * Make sure the user isn't trying to set
2271				 * WWNs on multiple ports at the same time.
2272				 */
2273				if (cmd == CTL_SET_PORT_WWNS) {
2274					printf("%s: Can't set WWNs on "
2275					       "multiple ports\n", __func__);
2276					retval = EINVAL;
2277					break;
2278				}
2279			}
2280			if (action != 0) {
2281				/*
2282				 * XXX KDM we have to drop the lock here,
2283				 * because the online/offline operations
2284				 * can potentially block.  We need to
2285				 * reference count the frontends so they
2286				 * can't go away,
2287				 */
2288				mtx_unlock(&softc->ctl_lock);
2289
2290				if (cmd == CTL_ENABLE_PORT) {
2291					struct ctl_lun *lun;
2292
2293					STAILQ_FOREACH(lun, &softc->lun_list,
2294						       links) {
2295						fe->lun_enable(fe->targ_lun_arg,
2296						    lun->target,
2297						    lun->lun);
2298					}
2299
2300					ctl_frontend_online(fe);
2301				} else if (cmd == CTL_DISABLE_PORT) {
2302					struct ctl_lun *lun;
2303
2304					ctl_frontend_offline(fe);
2305
2306					STAILQ_FOREACH(lun, &softc->lun_list,
2307						       links) {
2308						fe->lun_disable(
2309						    fe->targ_lun_arg,
2310						    lun->target,
2311						    lun->lun);
2312					}
2313				}
2314
2315				mtx_lock(&softc->ctl_lock);
2316
2317				if (cmd == CTL_SET_PORT_WWNS)
2318					ctl_frontend_set_wwns(fe,
2319					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2320					    1 : 0, entry->wwnn,
2321					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2322					    1 : 0, entry->wwpn);
2323			}
2324			if (done != 0)
2325				break;
2326		}
2327		mtx_unlock(&softc->ctl_lock);
2328		break;
2329	}
2330	case CTL_GET_PORT_LIST: {
2331		struct ctl_frontend *fe;
2332		struct ctl_port_list *list;
2333		int i;
2334
2335		list = (struct ctl_port_list *)addr;
2336
2337		if (list->alloc_len != (list->alloc_num *
2338		    sizeof(struct ctl_port_entry))) {
2339			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2340			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2341			       "%zu\n", __func__, list->alloc_len,
2342			       list->alloc_num, sizeof(struct ctl_port_entry));
2343			retval = EINVAL;
2344			break;
2345		}
2346		list->fill_len = 0;
2347		list->fill_num = 0;
2348		list->dropped_num = 0;
2349		i = 0;
2350		mtx_lock(&softc->ctl_lock);
2351		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2352			struct ctl_port_entry entry, *list_entry;
2353
2354			if (list->fill_num >= list->alloc_num) {
2355				list->dropped_num++;
2356				continue;
2357			}
2358
2359			entry.port_type = fe->port_type;
2360			strlcpy(entry.port_name, fe->port_name,
2361				sizeof(entry.port_name));
2362			entry.targ_port = fe->targ_port;
2363			entry.physical_port = fe->physical_port;
2364			entry.virtual_port = fe->virtual_port;
2365			entry.wwnn = fe->wwnn;
2366			entry.wwpn = fe->wwpn;
2367			if (fe->status & CTL_PORT_STATUS_ONLINE)
2368				entry.online = 1;
2369			else
2370				entry.online = 0;
2371
2372			list_entry = &list->entries[i];
2373
2374			retval = copyout(&entry, list_entry, sizeof(entry));
2375			if (retval != 0) {
2376				printf("%s: CTL_GET_PORT_LIST: copyout "
2377				       "returned %d\n", __func__, retval);
2378				break;
2379			}
2380			i++;
2381			list->fill_num++;
2382			list->fill_len += sizeof(entry);
2383		}
2384		mtx_unlock(&softc->ctl_lock);
2385
2386		/*
2387		 * If this is non-zero, we had a copyout fault, so there's
2388		 * probably no point in attempting to set the status inside
2389		 * the structure.
2390		 */
2391		if (retval != 0)
2392			break;
2393
2394		if (list->dropped_num > 0)
2395			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2396		else
2397			list->status = CTL_PORT_LIST_OK;
2398		break;
2399	}
2400	case CTL_DUMP_OOA: {
2401		struct ctl_lun *lun;
2402		union ctl_io *io;
2403		char printbuf[128];
2404		struct sbuf sb;
2405
2406		mtx_lock(&softc->ctl_lock);
2407		printf("Dumping OOA queues:\n");
2408		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2409			for (io = (union ctl_io *)TAILQ_FIRST(
2410			     &lun->ooa_queue); io != NULL;
2411			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2412			     ooa_links)) {
2413				sbuf_new(&sb, printbuf, sizeof(printbuf),
2414					 SBUF_FIXEDLEN);
2415				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2416					    (intmax_t)lun->lun,
2417					    io->scsiio.tag_num,
2418					    (io->io_hdr.flags &
2419					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2420					    (io->io_hdr.flags &
2421					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2422					    (io->io_hdr.flags &
2423					    CTL_FLAG_ABORT) ? " ABORT" : "",
2424			                    (io->io_hdr.flags &
2425		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2426				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2427				sbuf_finish(&sb);
2428				printf("%s\n", sbuf_data(&sb));
2429			}
2430		}
2431		printf("OOA queues dump done\n");
2432		mtx_unlock(&softc->ctl_lock);
2433		break;
2434	}
2435	case CTL_GET_OOA: {
2436		struct ctl_lun *lun;
2437		struct ctl_ooa *ooa_hdr;
2438		struct ctl_ooa_entry *entries;
2439		uint32_t cur_fill_num;
2440
2441		ooa_hdr = (struct ctl_ooa *)addr;
2442
2443		if ((ooa_hdr->alloc_len == 0)
2444		 || (ooa_hdr->alloc_num == 0)) {
2445			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2446			       "must be non-zero\n", __func__,
2447			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2448			retval = EINVAL;
2449			break;
2450		}
2451
2452		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2453		    sizeof(struct ctl_ooa_entry))) {
2454			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2455			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2456			       __func__, ooa_hdr->alloc_len,
2457			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2458			retval = EINVAL;
2459			break;
2460		}
2461
2462		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2463		if (entries == NULL) {
2464			printf("%s: could not allocate %d bytes for OOA "
2465			       "dump\n", __func__, ooa_hdr->alloc_len);
2466			retval = ENOMEM;
2467			break;
2468		}
2469
2470		mtx_lock(&softc->ctl_lock);
2471		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2472		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2473		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2474			mtx_unlock(&softc->ctl_lock);
2475			free(entries, M_CTL);
2476			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2477			       __func__, (uintmax_t)ooa_hdr->lun_num);
2478			retval = EINVAL;
2479			break;
2480		}
2481
2482		cur_fill_num = 0;
2483
2484		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2485			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2486				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2487					ooa_hdr, entries);
2488				if (retval != 0)
2489					break;
2490			}
2491			if (retval != 0) {
2492				mtx_unlock(&softc->ctl_lock);
2493				free(entries, M_CTL);
2494				break;
2495			}
2496		} else {
2497			lun = softc->ctl_luns[ooa_hdr->lun_num];
2498
2499			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2500						    entries);
2501		}
2502		mtx_unlock(&softc->ctl_lock);
2503
2504		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2505		ooa_hdr->fill_len = ooa_hdr->fill_num *
2506			sizeof(struct ctl_ooa_entry);
2507		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2508		if (retval != 0) {
2509			printf("%s: error copying out %d bytes for OOA dump\n",
2510			       __func__, ooa_hdr->fill_len);
2511		}
2512
2513		getbintime(&ooa_hdr->cur_bt);
2514
2515		if (cur_fill_num > ooa_hdr->alloc_num) {
2516			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2517			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2518		} else {
2519			ooa_hdr->dropped_num = 0;
2520			ooa_hdr->status = CTL_OOA_OK;
2521		}
2522
2523		free(entries, M_CTL);
2524		break;
2525	}
2526	case CTL_CHECK_OOA: {
2527		union ctl_io *io;
2528		struct ctl_lun *lun;
2529		struct ctl_ooa_info *ooa_info;
2530
2531
2532		ooa_info = (struct ctl_ooa_info *)addr;
2533
2534		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2535			ooa_info->status = CTL_OOA_INVALID_LUN;
2536			break;
2537		}
2538		mtx_lock(&softc->ctl_lock);
2539		lun = softc->ctl_luns[ooa_info->lun_id];
2540		if (lun == NULL) {
2541			mtx_unlock(&softc->ctl_lock);
2542			ooa_info->status = CTL_OOA_INVALID_LUN;
2543			break;
2544		}
2545
2546		ooa_info->num_entries = 0;
2547		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2548		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2549		     &io->io_hdr, ooa_links)) {
2550			ooa_info->num_entries++;
2551		}
2552
2553		mtx_unlock(&softc->ctl_lock);
2554		ooa_info->status = CTL_OOA_SUCCESS;
2555
2556		break;
2557	}
2558	case CTL_HARD_START:
2559	case CTL_HARD_STOP: {
2560		struct ctl_fe_ioctl_startstop_info ss_info;
2561		struct cfi_metatask *metatask;
2562		struct mtx hs_mtx;
2563
2564		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2565
2566		cv_init(&ss_info.sem, "hard start/stop cv" );
2567
2568		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2569		if (metatask == NULL) {
2570			retval = ENOMEM;
2571			mtx_destroy(&hs_mtx);
2572			break;
2573		}
2574
2575		if (cmd == CTL_HARD_START)
2576			metatask->tasktype = CFI_TASK_STARTUP;
2577		else
2578			metatask->tasktype = CFI_TASK_SHUTDOWN;
2579
2580		metatask->callback = ctl_ioctl_hard_startstop_callback;
2581		metatask->callback_arg = &ss_info;
2582
2583		cfi_action(metatask);
2584
2585		/* Wait for the callback */
2586		mtx_lock(&hs_mtx);
2587		cv_wait_sig(&ss_info.sem, &hs_mtx);
2588		mtx_unlock(&hs_mtx);
2589
2590		/*
2591		 * All information has been copied from the metatask by the
2592		 * time cv_broadcast() is called, so we free the metatask here.
2593		 */
2594		cfi_free_metatask(metatask);
2595
2596		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2597
2598		mtx_destroy(&hs_mtx);
2599		break;
2600	}
2601	case CTL_BBRREAD: {
2602		struct ctl_bbrread_info *bbr_info;
2603		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2604		struct mtx bbr_mtx;
2605		struct cfi_metatask *metatask;
2606
2607		bbr_info = (struct ctl_bbrread_info *)addr;
2608
2609		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2610
2611		bzero(&bbr_mtx, sizeof(bbr_mtx));
2612		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2613
2614		fe_bbr_info.bbr_info = bbr_info;
2615		fe_bbr_info.lock = &bbr_mtx;
2616
2617		cv_init(&fe_bbr_info.sem, "BBR read cv");
2618		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2619
2620		if (metatask == NULL) {
2621			mtx_destroy(&bbr_mtx);
2622			cv_destroy(&fe_bbr_info.sem);
2623			retval = ENOMEM;
2624			break;
2625		}
2626		metatask->tasktype = CFI_TASK_BBRREAD;
2627		metatask->callback = ctl_ioctl_bbrread_callback;
2628		metatask->callback_arg = &fe_bbr_info;
2629		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2630		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2631		metatask->taskinfo.bbrread.len = bbr_info->len;
2632
2633		cfi_action(metatask);
2634
2635		mtx_lock(&bbr_mtx);
2636		while (fe_bbr_info.wakeup_done == 0)
2637			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2638		mtx_unlock(&bbr_mtx);
2639
2640		bbr_info->status = metatask->status;
2641		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2642		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2643		memcpy(&bbr_info->sense_data,
2644		       &metatask->taskinfo.bbrread.sense_data,
2645		       ctl_min(sizeof(bbr_info->sense_data),
2646			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2647
2648		cfi_free_metatask(metatask);
2649
2650		mtx_destroy(&bbr_mtx);
2651		cv_destroy(&fe_bbr_info.sem);
2652
2653		break;
2654	}
2655	case CTL_DELAY_IO: {
2656		struct ctl_io_delay_info *delay_info;
2657#ifdef CTL_IO_DELAY
2658		struct ctl_lun *lun;
2659#endif /* CTL_IO_DELAY */
2660
2661		delay_info = (struct ctl_io_delay_info *)addr;
2662
2663#ifdef CTL_IO_DELAY
2664		mtx_lock(&softc->ctl_lock);
2665
2666		if ((delay_info->lun_id > CTL_MAX_LUNS)
2667		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2668			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2669		} else {
2670			lun = softc->ctl_luns[delay_info->lun_id];
2671
2672			delay_info->status = CTL_DELAY_STATUS_OK;
2673
2674			switch (delay_info->delay_type) {
2675			case CTL_DELAY_TYPE_CONT:
2676				break;
2677			case CTL_DELAY_TYPE_ONESHOT:
2678				break;
2679			default:
2680				delay_info->status =
2681					CTL_DELAY_STATUS_INVALID_TYPE;
2682				break;
2683			}
2684
2685			switch (delay_info->delay_loc) {
2686			case CTL_DELAY_LOC_DATAMOVE:
2687				lun->delay_info.datamove_type =
2688					delay_info->delay_type;
2689				lun->delay_info.datamove_delay =
2690					delay_info->delay_secs;
2691				break;
2692			case CTL_DELAY_LOC_DONE:
2693				lun->delay_info.done_type =
2694					delay_info->delay_type;
2695				lun->delay_info.done_delay =
2696					delay_info->delay_secs;
2697				break;
2698			default:
2699				delay_info->status =
2700					CTL_DELAY_STATUS_INVALID_LOC;
2701				break;
2702			}
2703		}
2704
2705		mtx_unlock(&softc->ctl_lock);
2706#else
2707		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2708#endif /* CTL_IO_DELAY */
2709		break;
2710	}
2711	case CTL_REALSYNC_SET: {
2712		int *syncstate;
2713
2714		syncstate = (int *)addr;
2715
2716		mtx_lock(&softc->ctl_lock);
2717		switch (*syncstate) {
2718		case 0:
2719			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2720			break;
2721		case 1:
2722			softc->flags |= CTL_FLAG_REAL_SYNC;
2723			break;
2724		default:
2725			retval = -EINVAL;
2726			break;
2727		}
2728		mtx_unlock(&softc->ctl_lock);
2729		break;
2730	}
2731	case CTL_REALSYNC_GET: {
2732		int *syncstate;
2733
2734		syncstate = (int*)addr;
2735
2736		mtx_lock(&softc->ctl_lock);
2737		if (softc->flags & CTL_FLAG_REAL_SYNC)
2738			*syncstate = 1;
2739		else
2740			*syncstate = 0;
2741		mtx_unlock(&softc->ctl_lock);
2742
2743		break;
2744	}
2745	case CTL_SETSYNC:
2746	case CTL_GETSYNC: {
2747		struct ctl_sync_info *sync_info;
2748		struct ctl_lun *lun;
2749
2750		sync_info = (struct ctl_sync_info *)addr;
2751
2752		mtx_lock(&softc->ctl_lock);
2753		lun = softc->ctl_luns[sync_info->lun_id];
2754		if (lun == NULL) {
2755			mtx_unlock(&softc->ctl_lock);
2756			sync_info->status = CTL_GS_SYNC_NO_LUN;
2757		}
2758		/*
2759		 * Get or set the sync interval.  We're not bounds checking
2760		 * in the set case, hopefully the user won't do something
2761		 * silly.
2762		 */
2763		if (cmd == CTL_GETSYNC)
2764			sync_info->sync_interval = lun->sync_interval;
2765		else
2766			lun->sync_interval = sync_info->sync_interval;
2767
2768		mtx_unlock(&softc->ctl_lock);
2769
2770		sync_info->status = CTL_GS_SYNC_OK;
2771
2772		break;
2773	}
2774	case CTL_GETSTATS: {
2775		struct ctl_stats *stats;
2776		struct ctl_lun *lun;
2777		int i;
2778
2779		stats = (struct ctl_stats *)addr;
2780
2781		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2782		     stats->alloc_len) {
2783			stats->status = CTL_SS_NEED_MORE_SPACE;
2784			stats->num_luns = softc->num_luns;
2785			break;
2786		}
2787		/*
2788		 * XXX KDM no locking here.  If the LUN list changes,
2789		 * things can blow up.
2790		 */
2791		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2792		     i++, lun = STAILQ_NEXT(lun, links)) {
2793			retval = copyout(&lun->stats, &stats->lun_stats[i],
2794					 sizeof(lun->stats));
2795			if (retval != 0)
2796				break;
2797		}
2798		stats->num_luns = softc->num_luns;
2799		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2800				 softc->num_luns;
2801		stats->status = CTL_SS_OK;
2802#ifdef CTL_TIME_IO
2803		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2804#else
2805		stats->flags = CTL_STATS_FLAG_NONE;
2806#endif
2807		getnanouptime(&stats->timestamp);
2808		break;
2809	}
2810	case CTL_ERROR_INJECT: {
2811		struct ctl_error_desc *err_desc, *new_err_desc;
2812		struct ctl_lun *lun;
2813
2814		err_desc = (struct ctl_error_desc *)addr;
2815
2816		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2817				      M_WAITOK | M_ZERO);
2818		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2819
2820		mtx_lock(&softc->ctl_lock);
2821		lun = softc->ctl_luns[err_desc->lun_id];
2822		if (lun == NULL) {
2823			mtx_unlock(&softc->ctl_lock);
2824			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2825			       __func__, (uintmax_t)err_desc->lun_id);
2826			retval = EINVAL;
2827			break;
2828		}
2829
2830		/*
2831		 * We could do some checking here to verify the validity
2832		 * of the request, but given the complexity of error
2833		 * injection requests, the checking logic would be fairly
2834		 * complex.
2835		 *
2836		 * For now, if the request is invalid, it just won't get
2837		 * executed and might get deleted.
2838		 */
2839		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2840
2841		/*
2842		 * XXX KDM check to make sure the serial number is unique,
2843		 * in case we somehow manage to wrap.  That shouldn't
2844		 * happen for a very long time, but it's the right thing to
2845		 * do.
2846		 */
2847		new_err_desc->serial = lun->error_serial;
2848		err_desc->serial = lun->error_serial;
2849		lun->error_serial++;
2850
2851		mtx_unlock(&softc->ctl_lock);
2852		break;
2853	}
2854	case CTL_ERROR_INJECT_DELETE: {
2855		struct ctl_error_desc *delete_desc, *desc, *desc2;
2856		struct ctl_lun *lun;
2857		int delete_done;
2858
2859		delete_desc = (struct ctl_error_desc *)addr;
2860		delete_done = 0;
2861
2862		mtx_lock(&softc->ctl_lock);
2863		lun = softc->ctl_luns[delete_desc->lun_id];
2864		if (lun == NULL) {
2865			mtx_unlock(&softc->ctl_lock);
2866			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2867			       __func__, (uintmax_t)delete_desc->lun_id);
2868			retval = EINVAL;
2869			break;
2870		}
2871		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2872			if (desc->serial != delete_desc->serial)
2873				continue;
2874
2875			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2876				      links);
2877			free(desc, M_CTL);
2878			delete_done = 1;
2879		}
2880		mtx_unlock(&softc->ctl_lock);
2881		if (delete_done == 0) {
2882			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2883			       "error serial %ju on LUN %u\n", __func__,
2884			       delete_desc->serial, delete_desc->lun_id);
2885			retval = EINVAL;
2886			break;
2887		}
2888		break;
2889	}
2890	case CTL_DUMP_STRUCTS: {
2891		int i, j, k;
2892		struct ctl_frontend *fe;
2893
2894		printf("CTL IID to WWPN map start:\n");
2895		for (i = 0; i < CTL_MAX_PORTS; i++) {
2896			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2897				if (softc->wwpn_iid[i][j].in_use == 0)
2898					continue;
2899
2900				printf("port %d iid %u WWPN %#jx\n",
2901				       softc->wwpn_iid[i][j].port,
2902				       softc->wwpn_iid[i][j].iid,
2903				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2904			}
2905		}
2906		printf("CTL IID to WWPN map end\n");
2907		printf("CTL Persistent Reservation information start:\n");
2908		for (i = 0; i < CTL_MAX_LUNS; i++) {
2909			struct ctl_lun *lun;
2910
2911			lun = softc->ctl_luns[i];
2912
2913			if ((lun == NULL)
2914			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2915				continue;
2916
2917			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2918				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2919					if (lun->per_res[j+k].registered == 0)
2920						continue;
2921					printf("LUN %d port %d iid %d key "
2922					       "%#jx\n", i, j, k,
2923					       (uintmax_t)scsi_8btou64(
2924					       lun->per_res[j+k].res_key.key));
2925				}
2926			}
2927		}
2928		printf("CTL Persistent Reservation information end\n");
2929		printf("CTL Frontends:\n");
2930		/*
2931		 * XXX KDM calling this without a lock.  We'd likely want
2932		 * to drop the lock before calling the frontend's dump
2933		 * routine anyway.
2934		 */
2935		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2936			printf("Frontend %s Type %u pport %d vport %d WWNN "
2937			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2938			       fe->physical_port, fe->virtual_port,
2939			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2940
2941			/*
2942			 * Frontends are not required to support the dump
2943			 * routine.
2944			 */
2945			if (fe->fe_dump == NULL)
2946				continue;
2947
2948			fe->fe_dump();
2949		}
2950		printf("CTL Frontend information end\n");
2951		break;
2952	}
2953	case CTL_LUN_REQ: {
2954		struct ctl_lun_req *lun_req;
2955		struct ctl_backend_driver *backend;
2956
2957		lun_req = (struct ctl_lun_req *)addr;
2958
2959		backend = ctl_backend_find(lun_req->backend);
2960		if (backend == NULL) {
2961			lun_req->status = CTL_LUN_ERROR;
2962			snprintf(lun_req->error_str,
2963				 sizeof(lun_req->error_str),
2964				 "Backend \"%s\" not found.",
2965				 lun_req->backend);
2966			break;
2967		}
2968		if (lun_req->num_be_args > 0) {
2969			lun_req->kern_be_args = ctl_copyin_args(
2970				lun_req->num_be_args,
2971				lun_req->be_args,
2972				lun_req->error_str,
2973				sizeof(lun_req->error_str));
2974			if (lun_req->kern_be_args == NULL) {
2975				lun_req->status = CTL_LUN_ERROR;
2976				break;
2977			}
2978		}
2979
2980		retval = backend->ioctl(dev, cmd, addr, flag, td);
2981
2982		if (lun_req->num_be_args > 0) {
2983			ctl_free_args(lun_req->num_be_args,
2984				      lun_req->kern_be_args);
2985		}
2986		break;
2987	}
2988	case CTL_LUN_LIST: {
2989		struct sbuf *sb;
2990		struct ctl_lun *lun;
2991		struct ctl_lun_list *list;
2992		struct ctl_be_lun_option *opt;
2993
2994		list = (struct ctl_lun_list *)addr;
2995
2996		/*
2997		 * Allocate a fixed length sbuf here, based on the length
2998		 * of the user's buffer.  We could allocate an auto-extending
2999		 * buffer, and then tell the user how much larger our
3000		 * amount of data is than his buffer, but that presents
3001		 * some problems:
3002		 *
3003		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3004		 *     we can't hold a lock while calling them with an
3005		 *     auto-extending buffer.
3006 		 *
3007		 * 2.  There is not currently a LUN reference counting
3008		 *     mechanism, outside of outstanding transactions on
3009		 *     the LUN's OOA queue.  So a LUN could go away on us
3010		 *     while we're getting the LUN number, backend-specific
3011		 *     information, etc.  Thus, given the way things
3012		 *     currently work, we need to hold the CTL lock while
3013		 *     grabbing LUN information.
3014		 *
3015		 * So, from the user's standpoint, the best thing to do is
3016		 * allocate what he thinks is a reasonable buffer length,
3017		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3018		 * double the buffer length and try again.  (And repeat
3019		 * that until he succeeds.)
3020		 */
3021		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3022		if (sb == NULL) {
3023			list->status = CTL_LUN_LIST_ERROR;
3024			snprintf(list->error_str, sizeof(list->error_str),
3025				 "Unable to allocate %d bytes for LUN list",
3026				 list->alloc_len);
3027			break;
3028		}
3029
3030		sbuf_printf(sb, "<ctllunlist>\n");
3031
3032		mtx_lock(&softc->ctl_lock);
3033
3034		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3035			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3036					     (uintmax_t)lun->lun);
3037
3038			/*
3039			 * Bail out as soon as we see that we've overfilled
3040			 * the buffer.
3041			 */
3042			if (retval != 0)
3043				break;
3044
3045			retval = sbuf_printf(sb, "<backend_type>%s"
3046					     "</backend_type>\n",
3047					     (lun->backend == NULL) ?  "none" :
3048					     lun->backend->name);
3049
3050			if (retval != 0)
3051				break;
3052
3053			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3054					     lun->be_lun->lun_type);
3055
3056			if (retval != 0)
3057				break;
3058
3059			if (lun->backend == NULL) {
3060				retval = sbuf_printf(sb, "</lun>\n");
3061				if (retval != 0)
3062					break;
3063				continue;
3064			}
3065
3066			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3067					     (lun->be_lun->maxlba > 0) ?
3068					     lun->be_lun->maxlba + 1 : 0);
3069
3070			if (retval != 0)
3071				break;
3072
3073			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3074					     lun->be_lun->blocksize);
3075
3076			if (retval != 0)
3077				break;
3078
3079			retval = sbuf_printf(sb, "<serial_number>");
3080
3081			if (retval != 0)
3082				break;
3083
3084			retval = ctl_sbuf_printf_esc(sb,
3085						     lun->be_lun->serial_num);
3086
3087			if (retval != 0)
3088				break;
3089
3090			retval = sbuf_printf(sb, "</serial_number>\n");
3091
3092			if (retval != 0)
3093				break;
3094
3095			retval = sbuf_printf(sb, "<device_id>");
3096
3097			if (retval != 0)
3098				break;
3099
3100			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3101
3102			if (retval != 0)
3103				break;
3104
3105			retval = sbuf_printf(sb, "</device_id>\n");
3106
3107			if (retval != 0)
3108				break;
3109
3110			if (lun->backend->lun_info != NULL) {
3111				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3112				if (retval != 0)
3113					break;
3114			}
3115			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3116				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3117				if (retval != 0)
3118					break;
3119			}
3120
3121			retval = sbuf_printf(sb, "</lun>\n");
3122
3123			if (retval != 0)
3124				break;
3125		}
3126		mtx_unlock(&softc->ctl_lock);
3127
3128		if ((retval != 0)
3129		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3130			retval = 0;
3131			sbuf_delete(sb);
3132			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3133			snprintf(list->error_str, sizeof(list->error_str),
3134				 "Out of space, %d bytes is too small",
3135				 list->alloc_len);
3136			break;
3137		}
3138
3139		sbuf_finish(sb);
3140
3141		retval = copyout(sbuf_data(sb), list->lun_xml,
3142				 sbuf_len(sb) + 1);
3143
3144		list->fill_len = sbuf_len(sb) + 1;
3145		list->status = CTL_LUN_LIST_OK;
3146		sbuf_delete(sb);
3147		break;
3148	}
3149	case CTL_ISCSI: {
3150		struct ctl_iscsi *ci;
3151		struct ctl_frontend *fe;
3152
3153		ci = (struct ctl_iscsi *)addr;
3154
3155		mtx_lock(&softc->ctl_lock);
3156		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3157			if (strcmp(fe->port_name, "iscsi") == 0)
3158				break;
3159		}
3160		mtx_unlock(&softc->ctl_lock);
3161
3162		if (fe == NULL) {
3163			ci->status = CTL_ISCSI_ERROR;
3164			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3165			break;
3166		}
3167
3168		retval = fe->ioctl(dev, cmd, addr, flag, td);
3169		break;
3170	}
3171	default: {
3172		/* XXX KDM should we fix this? */
3173#if 0
3174		struct ctl_backend_driver *backend;
3175		unsigned int type;
3176		int found;
3177
3178		found = 0;
3179
3180		/*
3181		 * We encode the backend type as the ioctl type for backend
3182		 * ioctls.  So parse it out here, and then search for a
3183		 * backend of this type.
3184		 */
3185		type = _IOC_TYPE(cmd);
3186
3187		STAILQ_FOREACH(backend, &softc->be_list, links) {
3188			if (backend->type == type) {
3189				found = 1;
3190				break;
3191			}
3192		}
3193		if (found == 0) {
3194			printf("ctl: unknown ioctl command %#lx or backend "
3195			       "%d\n", cmd, type);
3196			retval = -EINVAL;
3197			break;
3198		}
3199		retval = backend->ioctl(dev, cmd, addr, flag, td);
3200#endif
3201		retval = ENOTTY;
3202		break;
3203	}
3204	}
3205	return (retval);
3206}
3207
3208uint32_t
3209ctl_get_initindex(struct ctl_nexus *nexus)
3210{
3211	if (nexus->targ_port < CTL_MAX_PORTS)
3212		return (nexus->initid.id +
3213			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3214	else
3215		return (nexus->initid.id +
3216		       ((nexus->targ_port - CTL_MAX_PORTS) *
3217			CTL_MAX_INIT_PER_PORT));
3218}
3219
3220uint32_t
3221ctl_get_resindex(struct ctl_nexus *nexus)
3222{
3223	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3224}
3225
3226uint32_t
3227ctl_port_idx(int port_num)
3228{
3229	if (port_num < CTL_MAX_PORTS)
3230		return(port_num);
3231	else
3232		return(port_num - CTL_MAX_PORTS);
3233}
3234
3235/*
3236 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3237 * that are a power of 2.
3238 */
3239int
3240ctl_ffz(uint32_t *mask, uint32_t size)
3241{
3242	uint32_t num_chunks, num_pieces;
3243	int i, j;
3244
3245	num_chunks = (size >> 5);
3246	if (num_chunks == 0)
3247		num_chunks++;
3248	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3249
3250	for (i = 0; i < num_chunks; i++) {
3251		for (j = 0; j < num_pieces; j++) {
3252			if ((mask[i] & (1 << j)) == 0)
3253				return ((i << 5) + j);
3254		}
3255	}
3256
3257	return (-1);
3258}
3259
3260int
3261ctl_set_mask(uint32_t *mask, uint32_t bit)
3262{
3263	uint32_t chunk, piece;
3264
3265	chunk = bit >> 5;
3266	piece = bit % (sizeof(uint32_t) * 8);
3267
3268	if ((mask[chunk] & (1 << piece)) != 0)
3269		return (-1);
3270	else
3271		mask[chunk] |= (1 << piece);
3272
3273	return (0);
3274}
3275
3276int
3277ctl_clear_mask(uint32_t *mask, uint32_t bit)
3278{
3279	uint32_t chunk, piece;
3280
3281	chunk = bit >> 5;
3282	piece = bit % (sizeof(uint32_t) * 8);
3283
3284	if ((mask[chunk] & (1 << piece)) == 0)
3285		return (-1);
3286	else
3287		mask[chunk] &= ~(1 << piece);
3288
3289	return (0);
3290}
3291
3292int
3293ctl_is_set(uint32_t *mask, uint32_t bit)
3294{
3295	uint32_t chunk, piece;
3296
3297	chunk = bit >> 5;
3298	piece = bit % (sizeof(uint32_t) * 8);
3299
3300	if ((mask[chunk] & (1 << piece)) == 0)
3301		return (0);
3302	else
3303		return (1);
3304}
3305
3306#ifdef unused
3307/*
3308 * The bus, target and lun are optional, they can be filled in later.
3309 * can_wait is used to determine whether we can wait on the malloc or not.
3310 */
3311union ctl_io*
3312ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3313	      uint32_t targ_lun, int can_wait)
3314{
3315	union ctl_io *io;
3316
3317	if (can_wait)
3318		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3319	else
3320		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3321
3322	if (io != NULL) {
3323		io->io_hdr.io_type = io_type;
3324		io->io_hdr.targ_port = targ_port;
3325		/*
3326		 * XXX KDM this needs to change/go away.  We need to move
3327		 * to a preallocated pool of ctl_scsiio structures.
3328		 */
3329		io->io_hdr.nexus.targ_target.id = targ_target;
3330		io->io_hdr.nexus.targ_lun = targ_lun;
3331	}
3332
3333	return (io);
3334}
3335
3336void
3337ctl_kfree_io(union ctl_io *io)
3338{
3339	free(io, M_CTL);
3340}
3341#endif /* unused */
3342
3343/*
3344 * ctl_softc, pool_type, total_ctl_io are passed in.
3345 * npool is passed out.
3346 */
3347int
3348ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3349		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3350{
3351	uint32_t i;
3352	union ctl_io *cur_io, *next_io;
3353	struct ctl_io_pool *pool;
3354	int retval;
3355
3356	retval = 0;
3357
3358	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3359					    M_NOWAIT | M_ZERO);
3360	if (pool == NULL) {
3361		retval = -ENOMEM;
3362		goto bailout;
3363	}
3364
3365	pool->type = pool_type;
3366	pool->ctl_softc = ctl_softc;
3367
3368	mtx_lock(&ctl_softc->pool_lock);
3369	pool->id = ctl_softc->cur_pool_id++;
3370	mtx_unlock(&ctl_softc->pool_lock);
3371
3372	pool->flags = CTL_POOL_FLAG_NONE;
3373	pool->refcount = 1;		/* Reference for validity. */
3374	STAILQ_INIT(&pool->free_queue);
3375
3376	/*
3377	 * XXX KDM other options here:
3378	 * - allocate a page at a time
3379	 * - allocate one big chunk of memory.
3380	 * Page allocation might work well, but would take a little more
3381	 * tracking.
3382	 */
3383	for (i = 0; i < total_ctl_io; i++) {
3384		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3385						M_NOWAIT);
3386		if (cur_io == NULL) {
3387			retval = ENOMEM;
3388			break;
3389		}
3390		cur_io->io_hdr.pool = pool;
3391		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3392		pool->total_ctl_io++;
3393		pool->free_ctl_io++;
3394	}
3395
3396	if (retval != 0) {
3397		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3398		     cur_io != NULL; cur_io = next_io) {
3399			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3400							      links);
3401			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3402				      ctl_io_hdr, links);
3403			free(cur_io, M_CTL);
3404		}
3405
3406		free(pool, M_CTL);
3407		goto bailout;
3408	}
3409	mtx_lock(&ctl_softc->pool_lock);
3410	ctl_softc->num_pools++;
3411	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3412	/*
3413	 * Increment our usage count if this is an external consumer, so we
3414	 * can't get unloaded until the external consumer (most likely a
3415	 * FETD) unloads and frees his pool.
3416	 *
3417	 * XXX KDM will this increment the caller's module use count, or
3418	 * mine?
3419	 */
3420#if 0
3421	if ((pool_type != CTL_POOL_EMERGENCY)
3422	 && (pool_type != CTL_POOL_INTERNAL)
3423	 && (pool_type != CTL_POOL_IOCTL)
3424	 && (pool_type != CTL_POOL_4OTHERSC))
3425		MOD_INC_USE_COUNT;
3426#endif
3427
3428	mtx_unlock(&ctl_softc->pool_lock);
3429
3430	*npool = pool;
3431
3432bailout:
3433
3434	return (retval);
3435}
3436
3437static int
3438ctl_pool_acquire(struct ctl_io_pool *pool)
3439{
3440
3441	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3442
3443	if (pool->flags & CTL_POOL_FLAG_INVALID)
3444		return (-EINVAL);
3445
3446	pool->refcount++;
3447
3448	return (0);
3449}
3450
3451static void
3452ctl_pool_release(struct ctl_io_pool *pool)
3453{
3454	struct ctl_softc *ctl_softc = pool->ctl_softc;
3455	union ctl_io *io;
3456
3457	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3458
3459	if (--pool->refcount != 0)
3460		return;
3461
3462	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3463		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3464			      links);
3465		free(io, M_CTL);
3466	}
3467
3468	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3469	ctl_softc->num_pools--;
3470
3471	/*
3472	 * XXX KDM will this decrement the caller's usage count or mine?
3473	 */
3474#if 0
3475	if ((pool->type != CTL_POOL_EMERGENCY)
3476	 && (pool->type != CTL_POOL_INTERNAL)
3477	 && (pool->type != CTL_POOL_IOCTL))
3478		MOD_DEC_USE_COUNT;
3479#endif
3480
3481	free(pool, M_CTL);
3482}
3483
3484void
3485ctl_pool_free(struct ctl_io_pool *pool)
3486{
3487	struct ctl_softc *ctl_softc;
3488
3489	if (pool == NULL)
3490		return;
3491
3492	ctl_softc = pool->ctl_softc;
3493	mtx_lock(&ctl_softc->pool_lock);
3494	pool->flags |= CTL_POOL_FLAG_INVALID;
3495	ctl_pool_release(pool);
3496	mtx_unlock(&ctl_softc->pool_lock);
3497}
3498
3499/*
3500 * This routine does not block (except for spinlocks of course).
3501 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3502 * possible.
3503 */
3504union ctl_io *
3505ctl_alloc_io(void *pool_ref)
3506{
3507	union ctl_io *io;
3508	struct ctl_softc *ctl_softc;
3509	struct ctl_io_pool *pool, *npool;
3510	struct ctl_io_pool *emergency_pool;
3511
3512	pool = (struct ctl_io_pool *)pool_ref;
3513
3514	if (pool == NULL) {
3515		printf("%s: pool is NULL\n", __func__);
3516		return (NULL);
3517	}
3518
3519	emergency_pool = NULL;
3520
3521	ctl_softc = pool->ctl_softc;
3522
3523	mtx_lock(&ctl_softc->pool_lock);
3524	/*
3525	 * First, try to get the io structure from the user's pool.
3526	 */
3527	if (ctl_pool_acquire(pool) == 0) {
3528		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3529		if (io != NULL) {
3530			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3531			pool->total_allocated++;
3532			pool->free_ctl_io--;
3533			mtx_unlock(&ctl_softc->pool_lock);
3534			return (io);
3535		} else
3536			ctl_pool_release(pool);
3537	}
3538	/*
3539	 * If he doesn't have any io structures left, search for an
3540	 * emergency pool and grab one from there.
3541	 */
3542	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3543		if (npool->type != CTL_POOL_EMERGENCY)
3544			continue;
3545
3546		if (ctl_pool_acquire(npool) != 0)
3547			continue;
3548
3549		emergency_pool = npool;
3550
3551		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3552		if (io != NULL) {
3553			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3554			npool->total_allocated++;
3555			npool->free_ctl_io--;
3556			mtx_unlock(&ctl_softc->pool_lock);
3557			return (io);
3558		} else
3559			ctl_pool_release(npool);
3560	}
3561
3562	/* Drop the spinlock before we malloc */
3563	mtx_unlock(&ctl_softc->pool_lock);
3564
3565	/*
3566	 * The emergency pool (if it exists) didn't have one, so try an
3567	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3568	 */
3569	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3570	if (io != NULL) {
3571		/*
3572		 * If the emergency pool exists but is empty, add this
3573		 * ctl_io to its list when it gets freed.
3574		 */
3575		if (emergency_pool != NULL) {
3576			mtx_lock(&ctl_softc->pool_lock);
3577			if (ctl_pool_acquire(emergency_pool) == 0) {
3578				io->io_hdr.pool = emergency_pool;
3579				emergency_pool->total_ctl_io++;
3580				/*
3581				 * Need to bump this, otherwise
3582				 * total_allocated and total_freed won't
3583				 * match when we no longer have anything
3584				 * outstanding.
3585				 */
3586				emergency_pool->total_allocated++;
3587			}
3588			mtx_unlock(&ctl_softc->pool_lock);
3589		} else
3590			io->io_hdr.pool = NULL;
3591	}
3592
3593	return (io);
3594}
3595
3596void
3597ctl_free_io(union ctl_io *io)
3598{
3599	if (io == NULL)
3600		return;
3601
3602	/*
3603	 * If this ctl_io has a pool, return it to that pool.
3604	 */
3605	if (io->io_hdr.pool != NULL) {
3606		struct ctl_io_pool *pool;
3607#if 0
3608		struct ctl_softc *ctl_softc;
3609		union ctl_io *tmp_io;
3610		unsigned long xflags;
3611		int i;
3612
3613		ctl_softc = control_softc;
3614#endif
3615
3616		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3617
3618		mtx_lock(&pool->ctl_softc->pool_lock);
3619#if 0
3620		save_flags(xflags);
3621
3622		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3623		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3624		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3625		     links)) {
3626			if (tmp_io == io) {
3627				printf("%s: %p is still on the task queue!\n",
3628				       __func__, tmp_io);
3629				printf("%s: (%d): type %d "
3630				       "msg %d cdb %x iptl: "
3631				       "%d:%d:%d:%d tag 0x%04x "
3632				       "flg %#lx\n",
3633					__func__, i,
3634					tmp_io->io_hdr.io_type,
3635					tmp_io->io_hdr.msg_type,
3636					tmp_io->scsiio.cdb[0],
3637					tmp_io->io_hdr.nexus.initid.id,
3638					tmp_io->io_hdr.nexus.targ_port,
3639					tmp_io->io_hdr.nexus.targ_target.id,
3640					tmp_io->io_hdr.nexus.targ_lun,
3641					(tmp_io->io_hdr.io_type ==
3642					CTL_IO_TASK) ?
3643					tmp_io->taskio.tag_num :
3644					tmp_io->scsiio.tag_num,
3645					xflags);
3646				panic("I/O still on the task queue!");
3647			}
3648		}
3649#endif
3650		io->io_hdr.io_type = 0xff;
3651		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3652		pool->total_freed++;
3653		pool->free_ctl_io++;
3654		ctl_pool_release(pool);
3655		mtx_unlock(&pool->ctl_softc->pool_lock);
3656	} else {
3657		/*
3658		 * Otherwise, just free it.  We probably malloced it and
3659		 * the emergency pool wasn't available.
3660		 */
3661		free(io, M_CTL);
3662	}
3663
3664}
3665
3666void
3667ctl_zero_io(union ctl_io *io)
3668{
3669	void *pool_ref;
3670
3671	if (io == NULL)
3672		return;
3673
3674	/*
3675	 * May need to preserve linked list pointers at some point too.
3676	 */
3677	pool_ref = io->io_hdr.pool;
3678
3679	memset(io, 0, sizeof(*io));
3680
3681	io->io_hdr.pool = pool_ref;
3682}
3683
3684/*
3685 * This routine is currently used for internal copies of ctl_ios that need
3686 * to persist for some reason after we've already returned status to the
3687 * FETD.  (Thus the flag set.)
3688 *
3689 * XXX XXX
3690 * Note that this makes a blind copy of all fields in the ctl_io, except
3691 * for the pool reference.  This includes any memory that has been
3692 * allocated!  That memory will no longer be valid after done has been
3693 * called, so this would be VERY DANGEROUS for command that actually does
3694 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3695 * start and stop commands, which don't transfer any data, so this is not a
3696 * problem.  If it is used for anything else, the caller would also need to
3697 * allocate data buffer space and this routine would need to be modified to
3698 * copy the data buffer(s) as well.
3699 */
3700void
3701ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3702{
3703	void *pool_ref;
3704
3705	if ((src == NULL)
3706	 || (dest == NULL))
3707		return;
3708
3709	/*
3710	 * May need to preserve linked list pointers at some point too.
3711	 */
3712	pool_ref = dest->io_hdr.pool;
3713
3714	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3715
3716	dest->io_hdr.pool = pool_ref;
3717	/*
3718	 * We need to know that this is an internal copy, and doesn't need
3719	 * to get passed back to the FETD that allocated it.
3720	 */
3721	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3722}
3723
3724#ifdef NEEDTOPORT
3725static void
3726ctl_update_power_subpage(struct copan_power_subpage *page)
3727{
3728	int num_luns, num_partitions, config_type;
3729	struct ctl_softc *softc;
3730	cs_BOOL_t aor_present, shelf_50pct_power;
3731	cs_raidset_personality_t rs_type;
3732	int max_active_luns;
3733
3734	softc = control_softc;
3735
3736	/* subtract out the processor LUN */
3737	num_luns = softc->num_luns - 1;
3738	/*
3739	 * Default to 7 LUNs active, which was the only number we allowed
3740	 * in the past.
3741	 */
3742	max_active_luns = 7;
3743
3744	num_partitions = config_GetRsPartitionInfo();
3745	config_type = config_GetConfigType();
3746	shelf_50pct_power = config_GetShelfPowerMode();
3747	aor_present = config_IsAorRsPresent();
3748
3749	rs_type = ddb_GetRsRaidType(1);
3750	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3751	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3752		EPRINT(0, "Unsupported RS type %d!", rs_type);
3753	}
3754
3755
3756	page->total_luns = num_luns;
3757
3758	switch (config_type) {
3759	case 40:
3760		/*
3761		 * In a 40 drive configuration, it doesn't matter what DC
3762		 * cards we have, whether we have AOR enabled or not,
3763		 * partitioning or not, or what type of RAIDset we have.
3764		 * In that scenario, we can power up every LUN we present
3765		 * to the user.
3766		 */
3767		max_active_luns = num_luns;
3768
3769		break;
3770	case 64:
3771		if (shelf_50pct_power == CS_FALSE) {
3772			/* 25% power */
3773			if (aor_present == CS_TRUE) {
3774				if (rs_type ==
3775				     CS_RAIDSET_PERSONALITY_RAID5) {
3776					max_active_luns = 7;
3777				} else if (rs_type ==
3778					 CS_RAIDSET_PERSONALITY_RAID1){
3779					max_active_luns = 14;
3780				} else {
3781					/* XXX KDM now what?? */
3782				}
3783			} else {
3784				if (rs_type ==
3785				     CS_RAIDSET_PERSONALITY_RAID5) {
3786					max_active_luns = 8;
3787				} else if (rs_type ==
3788					 CS_RAIDSET_PERSONALITY_RAID1){
3789					max_active_luns = 16;
3790				} else {
3791					/* XXX KDM now what?? */
3792				}
3793			}
3794		} else {
3795			/* 50% power */
3796			/*
3797			 * With 50% power in a 64 drive configuration, we
3798			 * can power all LUNs we present.
3799			 */
3800			max_active_luns = num_luns;
3801		}
3802		break;
3803	case 112:
3804		if (shelf_50pct_power == CS_FALSE) {
3805			/* 25% power */
3806			if (aor_present == CS_TRUE) {
3807				if (rs_type ==
3808				     CS_RAIDSET_PERSONALITY_RAID5) {
3809					max_active_luns = 7;
3810				} else if (rs_type ==
3811					 CS_RAIDSET_PERSONALITY_RAID1){
3812					max_active_luns = 14;
3813				} else {
3814					/* XXX KDM now what?? */
3815				}
3816			} else {
3817				if (rs_type ==
3818				     CS_RAIDSET_PERSONALITY_RAID5) {
3819					max_active_luns = 8;
3820				} else if (rs_type ==
3821					 CS_RAIDSET_PERSONALITY_RAID1){
3822					max_active_luns = 16;
3823				} else {
3824					/* XXX KDM now what?? */
3825				}
3826			}
3827		} else {
3828			/* 50% power */
3829			if (aor_present == CS_TRUE) {
3830				if (rs_type ==
3831				     CS_RAIDSET_PERSONALITY_RAID5) {
3832					max_active_luns = 14;
3833				} else if (rs_type ==
3834					 CS_RAIDSET_PERSONALITY_RAID1){
3835					/*
3836					 * We're assuming here that disk
3837					 * caching is enabled, and so we're
3838					 * able to power up half of each
3839					 * LUN, and cache all writes.
3840					 */
3841					max_active_luns = num_luns;
3842				} else {
3843					/* XXX KDM now what?? */
3844				}
3845			} else {
3846				if (rs_type ==
3847				     CS_RAIDSET_PERSONALITY_RAID5) {
3848					max_active_luns = 15;
3849				} else if (rs_type ==
3850					 CS_RAIDSET_PERSONALITY_RAID1){
3851					max_active_luns = 30;
3852				} else {
3853					/* XXX KDM now what?? */
3854				}
3855			}
3856		}
3857		break;
3858	default:
3859		/*
3860		 * In this case, we have an unknown configuration, so we
3861		 * just use the default from above.
3862		 */
3863		break;
3864	}
3865
3866	page->max_active_luns = max_active_luns;
3867#if 0
3868	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3869	       page->total_luns, page->max_active_luns);
3870#endif
3871}
3872#endif /* NEEDTOPORT */
3873
3874/*
3875 * This routine could be used in the future to load default and/or saved
3876 * mode page parameters for a particuar lun.
3877 */
3878static int
3879ctl_init_page_index(struct ctl_lun *lun)
3880{
3881	int i;
3882	struct ctl_page_index *page_index;
3883	struct ctl_softc *softc;
3884
3885	memcpy(&lun->mode_pages.index, page_index_template,
3886	       sizeof(page_index_template));
3887
3888	softc = lun->ctl_softc;
3889
3890	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3891
3892		page_index = &lun->mode_pages.index[i];
3893		/*
3894		 * If this is a disk-only mode page, there's no point in
3895		 * setting it up.  For some pages, we have to have some
3896		 * basic information about the disk in order to calculate the
3897		 * mode page data.
3898		 */
3899		if ((lun->be_lun->lun_type != T_DIRECT)
3900		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3901			continue;
3902
3903		switch (page_index->page_code & SMPH_PC_MASK) {
3904		case SMS_FORMAT_DEVICE_PAGE: {
3905			struct scsi_format_page *format_page;
3906
3907			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3908				panic("subpage is incorrect!");
3909
3910			/*
3911			 * Sectors per track are set above.  Bytes per
3912			 * sector need to be set here on a per-LUN basis.
3913			 */
3914			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3915			       &format_page_default,
3916			       sizeof(format_page_default));
3917			memcpy(&lun->mode_pages.format_page[
3918			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3919			       sizeof(format_page_changeable));
3920			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3921			       &format_page_default,
3922			       sizeof(format_page_default));
3923			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3924			       &format_page_default,
3925			       sizeof(format_page_default));
3926
3927			format_page = &lun->mode_pages.format_page[
3928				CTL_PAGE_CURRENT];
3929			scsi_ulto2b(lun->be_lun->blocksize,
3930				    format_page->bytes_per_sector);
3931
3932			format_page = &lun->mode_pages.format_page[
3933				CTL_PAGE_DEFAULT];
3934			scsi_ulto2b(lun->be_lun->blocksize,
3935				    format_page->bytes_per_sector);
3936
3937			format_page = &lun->mode_pages.format_page[
3938				CTL_PAGE_SAVED];
3939			scsi_ulto2b(lun->be_lun->blocksize,
3940				    format_page->bytes_per_sector);
3941
3942			page_index->page_data =
3943				(uint8_t *)lun->mode_pages.format_page;
3944			break;
3945		}
3946		case SMS_RIGID_DISK_PAGE: {
3947			struct scsi_rigid_disk_page *rigid_disk_page;
3948			uint32_t sectors_per_cylinder;
3949			uint64_t cylinders;
3950#ifndef	__XSCALE__
3951			int shift;
3952#endif /* !__XSCALE__ */
3953
3954			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3955				panic("invalid subpage value %d",
3956				      page_index->subpage);
3957
3958			/*
3959			 * Rotation rate and sectors per track are set
3960			 * above.  We calculate the cylinders here based on
3961			 * capacity.  Due to the number of heads and
3962			 * sectors per track we're using, smaller arrays
3963			 * may turn out to have 0 cylinders.  Linux and
3964			 * FreeBSD don't pay attention to these mode pages
3965			 * to figure out capacity, but Solaris does.  It
3966			 * seems to deal with 0 cylinders just fine, and
3967			 * works out a fake geometry based on the capacity.
3968			 */
3969			memcpy(&lun->mode_pages.rigid_disk_page[
3970			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3971			       sizeof(rigid_disk_page_default));
3972			memcpy(&lun->mode_pages.rigid_disk_page[
3973			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3974			       sizeof(rigid_disk_page_changeable));
3975			memcpy(&lun->mode_pages.rigid_disk_page[
3976			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3977			       sizeof(rigid_disk_page_default));
3978			memcpy(&lun->mode_pages.rigid_disk_page[
3979			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3980			       sizeof(rigid_disk_page_default));
3981
3982			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3983				CTL_DEFAULT_HEADS;
3984
3985			/*
3986			 * The divide method here will be more accurate,
3987			 * probably, but results in floating point being
3988			 * used in the kernel on i386 (__udivdi3()).  On the
3989			 * XScale, though, __udivdi3() is implemented in
3990			 * software.
3991			 *
3992			 * The shift method for cylinder calculation is
3993			 * accurate if sectors_per_cylinder is a power of
3994			 * 2.  Otherwise it might be slightly off -- you
3995			 * might have a bit of a truncation problem.
3996			 */
3997#ifdef	__XSCALE__
3998			cylinders = (lun->be_lun->maxlba + 1) /
3999				sectors_per_cylinder;
4000#else
4001			for (shift = 31; shift > 0; shift--) {
4002				if (sectors_per_cylinder & (1 << shift))
4003					break;
4004			}
4005			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4006#endif
4007
4008			/*
4009			 * We've basically got 3 bytes, or 24 bits for the
4010			 * cylinder size in the mode page.  If we're over,
4011			 * just round down to 2^24.
4012			 */
4013			if (cylinders > 0xffffff)
4014				cylinders = 0xffffff;
4015
4016			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4017				CTL_PAGE_CURRENT];
4018			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4019
4020			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4021				CTL_PAGE_DEFAULT];
4022			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4023
4024			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4025				CTL_PAGE_SAVED];
4026			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4027
4028			page_index->page_data =
4029				(uint8_t *)lun->mode_pages.rigid_disk_page;
4030			break;
4031		}
4032		case SMS_CACHING_PAGE: {
4033
4034			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4035				panic("invalid subpage value %d",
4036				      page_index->subpage);
4037			/*
4038			 * Defaults should be okay here, no calculations
4039			 * needed.
4040			 */
4041			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4042			       &caching_page_default,
4043			       sizeof(caching_page_default));
4044			memcpy(&lun->mode_pages.caching_page[
4045			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4046			       sizeof(caching_page_changeable));
4047			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4048			       &caching_page_default,
4049			       sizeof(caching_page_default));
4050			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4051			       &caching_page_default,
4052			       sizeof(caching_page_default));
4053			page_index->page_data =
4054				(uint8_t *)lun->mode_pages.caching_page;
4055			break;
4056		}
4057		case SMS_CONTROL_MODE_PAGE: {
4058
4059			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4060				panic("invalid subpage value %d",
4061				      page_index->subpage);
4062
4063			/*
4064			 * Defaults should be okay here, no calculations
4065			 * needed.
4066			 */
4067			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4068			       &control_page_default,
4069			       sizeof(control_page_default));
4070			memcpy(&lun->mode_pages.control_page[
4071			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4072			       sizeof(control_page_changeable));
4073			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4074			       &control_page_default,
4075			       sizeof(control_page_default));
4076			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4077			       &control_page_default,
4078			       sizeof(control_page_default));
4079			page_index->page_data =
4080				(uint8_t *)lun->mode_pages.control_page;
4081			break;
4082
4083		}
4084		case SMS_VENDOR_SPECIFIC_PAGE:{
4085			switch (page_index->subpage) {
4086			case PWR_SUBPAGE_CODE: {
4087				struct copan_power_subpage *current_page,
4088							   *saved_page;
4089
4090				memcpy(&lun->mode_pages.power_subpage[
4091				       CTL_PAGE_CURRENT],
4092				       &power_page_default,
4093				       sizeof(power_page_default));
4094				memcpy(&lun->mode_pages.power_subpage[
4095				       CTL_PAGE_CHANGEABLE],
4096				       &power_page_changeable,
4097				       sizeof(power_page_changeable));
4098				memcpy(&lun->mode_pages.power_subpage[
4099				       CTL_PAGE_DEFAULT],
4100				       &power_page_default,
4101				       sizeof(power_page_default));
4102				memcpy(&lun->mode_pages.power_subpage[
4103				       CTL_PAGE_SAVED],
4104				       &power_page_default,
4105				       sizeof(power_page_default));
4106				page_index->page_data =
4107				    (uint8_t *)lun->mode_pages.power_subpage;
4108
4109				current_page = (struct copan_power_subpage *)
4110					(page_index->page_data +
4111					 (page_index->page_len *
4112					  CTL_PAGE_CURRENT));
4113			        saved_page = (struct copan_power_subpage *)
4114				        (page_index->page_data +
4115					 (page_index->page_len *
4116					  CTL_PAGE_SAVED));
4117				break;
4118			}
4119			case APS_SUBPAGE_CODE: {
4120				struct copan_aps_subpage *current_page,
4121							 *saved_page;
4122
4123				// This gets set multiple times but
4124				// it should always be the same. It's
4125				// only done during init so who cares.
4126				index_to_aps_page = i;
4127
4128				memcpy(&lun->mode_pages.aps_subpage[
4129				       CTL_PAGE_CURRENT],
4130				       &aps_page_default,
4131				       sizeof(aps_page_default));
4132				memcpy(&lun->mode_pages.aps_subpage[
4133				       CTL_PAGE_CHANGEABLE],
4134				       &aps_page_changeable,
4135				       sizeof(aps_page_changeable));
4136				memcpy(&lun->mode_pages.aps_subpage[
4137				       CTL_PAGE_DEFAULT],
4138				       &aps_page_default,
4139				       sizeof(aps_page_default));
4140				memcpy(&lun->mode_pages.aps_subpage[
4141				       CTL_PAGE_SAVED],
4142				       &aps_page_default,
4143				       sizeof(aps_page_default));
4144				page_index->page_data =
4145					(uint8_t *)lun->mode_pages.aps_subpage;
4146
4147				current_page = (struct copan_aps_subpage *)
4148					(page_index->page_data +
4149					 (page_index->page_len *
4150					  CTL_PAGE_CURRENT));
4151				saved_page = (struct copan_aps_subpage *)
4152					(page_index->page_data +
4153					 (page_index->page_len *
4154					  CTL_PAGE_SAVED));
4155				break;
4156			}
4157			case DBGCNF_SUBPAGE_CODE: {
4158				struct copan_debugconf_subpage *current_page,
4159							       *saved_page;
4160
4161				memcpy(&lun->mode_pages.debugconf_subpage[
4162				       CTL_PAGE_CURRENT],
4163				       &debugconf_page_default,
4164				       sizeof(debugconf_page_default));
4165				memcpy(&lun->mode_pages.debugconf_subpage[
4166				       CTL_PAGE_CHANGEABLE],
4167				       &debugconf_page_changeable,
4168				       sizeof(debugconf_page_changeable));
4169				memcpy(&lun->mode_pages.debugconf_subpage[
4170				       CTL_PAGE_DEFAULT],
4171				       &debugconf_page_default,
4172				       sizeof(debugconf_page_default));
4173				memcpy(&lun->mode_pages.debugconf_subpage[
4174				       CTL_PAGE_SAVED],
4175				       &debugconf_page_default,
4176				       sizeof(debugconf_page_default));
4177				page_index->page_data =
4178					(uint8_t *)lun->mode_pages.debugconf_subpage;
4179
4180				current_page = (struct copan_debugconf_subpage *)
4181					(page_index->page_data +
4182					 (page_index->page_len *
4183					  CTL_PAGE_CURRENT));
4184				saved_page = (struct copan_debugconf_subpage *)
4185					(page_index->page_data +
4186					 (page_index->page_len *
4187					  CTL_PAGE_SAVED));
4188				break;
4189			}
4190			default:
4191				panic("invalid subpage value %d",
4192				      page_index->subpage);
4193				break;
4194			}
4195   			break;
4196		}
4197		default:
4198			panic("invalid page value %d",
4199			      page_index->page_code & SMPH_PC_MASK);
4200			break;
4201    	}
4202	}
4203
4204	return (CTL_RETVAL_COMPLETE);
4205}
4206
4207/*
4208 * LUN allocation.
4209 *
4210 * Requirements:
4211 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4212 *   wants us to allocate the LUN and he can block.
4213 * - ctl_softc is always set
4214 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4215 *
4216 * Returns 0 for success, non-zero (errno) for failure.
4217 */
4218static int
4219ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4220	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4221{
4222	struct ctl_lun *nlun, *lun;
4223	struct ctl_frontend *fe;
4224	int lun_number, i, lun_malloced;
4225
4226	if (be_lun == NULL)
4227		return (EINVAL);
4228
4229	/*
4230	 * We currently only support Direct Access or Processor LUN types.
4231	 */
4232	switch (be_lun->lun_type) {
4233	case T_DIRECT:
4234		break;
4235	case T_PROCESSOR:
4236		break;
4237	case T_SEQUENTIAL:
4238	case T_CHANGER:
4239	default:
4240		be_lun->lun_config_status(be_lun->be_lun,
4241					  CTL_LUN_CONFIG_FAILURE);
4242		break;
4243	}
4244	if (ctl_lun == NULL) {
4245		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4246		lun_malloced = 1;
4247	} else {
4248		lun_malloced = 0;
4249		lun = ctl_lun;
4250	}
4251
4252	memset(lun, 0, sizeof(*lun));
4253	if (lun_malloced)
4254		lun->flags = CTL_LUN_MALLOCED;
4255
4256	mtx_lock(&ctl_softc->ctl_lock);
4257	/*
4258	 * See if the caller requested a particular LUN number.  If so, see
4259	 * if it is available.  Otherwise, allocate the first available LUN.
4260	 */
4261	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4262		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4263		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4264			mtx_unlock(&ctl_softc->ctl_lock);
4265			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4266				printf("ctl: requested LUN ID %d is higher "
4267				       "than CTL_MAX_LUNS - 1 (%d)\n",
4268				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4269			} else {
4270				/*
4271				 * XXX KDM return an error, or just assign
4272				 * another LUN ID in this case??
4273				 */
4274				printf("ctl: requested LUN ID %d is already "
4275				       "in use\n", be_lun->req_lun_id);
4276			}
4277			if (lun->flags & CTL_LUN_MALLOCED)
4278				free(lun, M_CTL);
4279			be_lun->lun_config_status(be_lun->be_lun,
4280						  CTL_LUN_CONFIG_FAILURE);
4281			return (ENOSPC);
4282		}
4283		lun_number = be_lun->req_lun_id;
4284	} else {
4285		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4286		if (lun_number == -1) {
4287			mtx_unlock(&ctl_softc->ctl_lock);
4288			printf("ctl: can't allocate LUN on target %ju, out of "
4289			       "LUNs\n", (uintmax_t)target_id.id);
4290			if (lun->flags & CTL_LUN_MALLOCED)
4291				free(lun, M_CTL);
4292			be_lun->lun_config_status(be_lun->be_lun,
4293						  CTL_LUN_CONFIG_FAILURE);
4294			return (ENOSPC);
4295		}
4296	}
4297	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4298
4299	lun->target = target_id;
4300	lun->lun = lun_number;
4301	lun->be_lun = be_lun;
4302	/*
4303	 * The processor LUN is always enabled.  Disk LUNs come on line
4304	 * disabled, and must be enabled by the backend.
4305	 */
4306	lun->flags |= CTL_LUN_DISABLED;
4307	lun->backend = be_lun->be;
4308	be_lun->ctl_lun = lun;
4309	be_lun->lun_id = lun_number;
4310	atomic_add_int(&be_lun->be->num_luns, 1);
4311	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4312		lun->flags |= CTL_LUN_STOPPED;
4313
4314	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4315		lun->flags |= CTL_LUN_INOPERABLE;
4316
4317	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4318		lun->flags |= CTL_LUN_PRIMARY_SC;
4319
4320	lun->ctl_softc = ctl_softc;
4321	TAILQ_INIT(&lun->ooa_queue);
4322	TAILQ_INIT(&lun->blocked_queue);
4323	STAILQ_INIT(&lun->error_list);
4324
4325	/*
4326	 * Initialize the mode page index.
4327	 */
4328	ctl_init_page_index(lun);
4329
4330	/*
4331	 * Set the poweron UA for all initiators on this LUN only.
4332	 */
4333	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4334		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4335
4336	/*
4337	 * Now, before we insert this lun on the lun list, set the lun
4338	 * inventory changed UA for all other luns.
4339	 */
4340	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4341		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4342			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4343		}
4344	}
4345
4346	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4347
4348	ctl_softc->ctl_luns[lun_number] = lun;
4349
4350	ctl_softc->num_luns++;
4351
4352	/* Setup statistics gathering */
4353	lun->stats.device_type = be_lun->lun_type;
4354	lun->stats.lun_number = lun_number;
4355	if (lun->stats.device_type == T_DIRECT)
4356		lun->stats.blocksize = be_lun->blocksize;
4357	else
4358		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4359	for (i = 0;i < CTL_MAX_PORTS;i++)
4360		lun->stats.ports[i].targ_port = i;
4361
4362	mtx_unlock(&ctl_softc->ctl_lock);
4363
4364	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4365
4366	/*
4367	 * Run through each registered FETD and bring it online if it isn't
4368	 * already.  Enable the target ID if it hasn't been enabled, and
4369	 * enable this particular LUN.
4370	 */
4371	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4372		int retval;
4373
4374		/*
4375		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4376		 * to do things differently if we go to a multiple target
4377		 * ID scheme.
4378		 */
4379		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4380
4381			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4382			if (retval != 0) {
4383				printf("ctl_alloc_lun: FETD %s port %d "
4384				       "returned error %d for targ_enable on "
4385				       "target %ju\n", fe->port_name,
4386				       fe->targ_port, retval,
4387				       (uintmax_t)target_id.id);
4388			} else
4389				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4390		}
4391
4392		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4393		if (retval != 0) {
4394			printf("ctl_alloc_lun: FETD %s port %d returned error "
4395			       "%d for lun_enable on target %ju lun %d\n",
4396			       fe->port_name, fe->targ_port, retval,
4397			       (uintmax_t)target_id.id, lun_number);
4398		} else
4399			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4400	}
4401	return (0);
4402}
4403
4404/*
4405 * Delete a LUN.
4406 * Assumptions:
4407 * - LUN has already been marked invalid and any pending I/O has been taken
4408 *   care of.
4409 */
4410static int
4411ctl_free_lun(struct ctl_lun *lun)
4412{
4413	struct ctl_softc *softc;
4414#if 0
4415	struct ctl_frontend *fe;
4416#endif
4417	struct ctl_lun *nlun;
4418	union ctl_io *io, *next_io;
4419	int i;
4420
4421	softc = lun->ctl_softc;
4422
4423	mtx_assert(&softc->ctl_lock, MA_OWNED);
4424
4425	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4426
4427	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4428
4429	softc->ctl_luns[lun->lun] = NULL;
4430
4431	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4432		printf("ctl_free_lun: aieee!! freeing a LUN with "
4433		       "outstanding I/O!!\n");
4434	}
4435
4436	/*
4437	 * If we have anything pending on the RtR queue, remove it.
4438	 */
4439	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4440	     io = next_io) {
4441		uint32_t targ_lun;
4442
4443		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4444		targ_lun = io->io_hdr.nexus.targ_lun;
4445		if (io->io_hdr.nexus.lun_map_fn != NULL)
4446			targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
4447		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4448		 && (targ_lun == lun->lun))
4449			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4450				      ctl_io_hdr, links);
4451	}
4452
4453	/*
4454	 * Then remove everything from the blocked queue.
4455	 */
4456	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4457	     io = next_io) {
4458		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4459		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4460		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4461	}
4462
4463	/*
4464	 * Now clear out the OOA queue, and free all the I/O.
4465	 * XXX KDM should we notify the FETD here?  We probably need to
4466	 * quiesce the LUN before deleting it.
4467	 */
4468	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4469	     io = next_io) {
4470		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4471		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4472		ctl_free_io(io);
4473	}
4474
4475	softc->num_luns--;
4476
4477	/*
4478	 * XXX KDM this scheme only works for a single target/multiple LUN
4479	 * setup.  It needs to be revamped for a multiple target scheme.
4480	 *
4481	 * XXX KDM this results in fe->lun_disable() getting called twice,
4482	 * once when ctl_disable_lun() is called, and a second time here.
4483	 * We really need to re-think the LUN disable semantics.  There
4484	 * should probably be several steps/levels to LUN removal:
4485	 *  - disable
4486	 *  - invalidate
4487	 *  - free
4488 	 *
4489	 * Right now we only have a disable method when communicating to
4490	 * the front end ports, at least for individual LUNs.
4491	 */
4492#if 0
4493	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4494		int retval;
4495
4496		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4497					 lun->lun);
4498		if (retval != 0) {
4499			printf("ctl_free_lun: FETD %s port %d returned error "
4500			       "%d for lun_disable on target %ju lun %jd\n",
4501			       fe->port_name, fe->targ_port, retval,
4502			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4503		}
4504
4505		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4506			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4507
4508			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4509			if (retval != 0) {
4510				printf("ctl_free_lun: FETD %s port %d "
4511				       "returned error %d for targ_disable on "
4512				       "target %ju\n", fe->port_name,
4513				       fe->targ_port, retval,
4514				       (uintmax_t)lun->target.id);
4515			} else
4516				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4517
4518			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4519				continue;
4520
4521#if 0
4522			fe->port_offline(fe->onoff_arg);
4523			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4524#endif
4525		}
4526	}
4527#endif
4528
4529	/*
4530	 * Tell the backend to free resources, if this LUN has a backend.
4531	 */
4532	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4533	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4534
4535	if (lun->flags & CTL_LUN_MALLOCED)
4536		free(lun, M_CTL);
4537
4538	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4539		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4540			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4541		}
4542	}
4543
4544	return (0);
4545}
4546
4547static void
4548ctl_create_lun(struct ctl_be_lun *be_lun)
4549{
4550	struct ctl_softc *ctl_softc;
4551
4552	ctl_softc = control_softc;
4553
4554	/*
4555	 * ctl_alloc_lun() should handle all potential failure cases.
4556	 */
4557	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4558}
4559
4560int
4561ctl_add_lun(struct ctl_be_lun *be_lun)
4562{
4563	struct ctl_softc *ctl_softc;
4564
4565	ctl_softc = control_softc;
4566
4567	mtx_lock(&ctl_softc->ctl_lock);
4568	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4569	mtx_unlock(&ctl_softc->ctl_lock);
4570
4571	ctl_wakeup_thread();
4572
4573	return (0);
4574}
4575
4576int
4577ctl_enable_lun(struct ctl_be_lun *be_lun)
4578{
4579	struct ctl_softc *ctl_softc;
4580	struct ctl_frontend *fe, *nfe;
4581	struct ctl_lun *lun;
4582	int retval;
4583
4584	ctl_softc = control_softc;
4585
4586	lun = (struct ctl_lun *)be_lun->ctl_lun;
4587
4588	mtx_lock(&ctl_softc->ctl_lock);
4589	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4590		/*
4591		 * eh?  Why did we get called if the LUN is already
4592		 * enabled?
4593		 */
4594		mtx_unlock(&ctl_softc->ctl_lock);
4595		return (0);
4596	}
4597	lun->flags &= ~CTL_LUN_DISABLED;
4598
4599	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4600		nfe = STAILQ_NEXT(fe, links);
4601
4602		/*
4603		 * Drop the lock while we call the FETD's enable routine.
4604		 * This can lead to a callback into CTL (at least in the
4605		 * case of the internal initiator frontend.
4606		 */
4607		mtx_unlock(&ctl_softc->ctl_lock);
4608		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4609		mtx_lock(&ctl_softc->ctl_lock);
4610		if (retval != 0) {
4611			printf("%s: FETD %s port %d returned error "
4612			       "%d for lun_enable on target %ju lun %jd\n",
4613			       __func__, fe->port_name, fe->targ_port, retval,
4614			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4615		}
4616#if 0
4617		 else {
4618            /* NOTE:  TODO:  why does lun enable affect port status? */
4619			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4620		}
4621#endif
4622	}
4623
4624	mtx_unlock(&ctl_softc->ctl_lock);
4625
4626	return (0);
4627}
4628
4629int
4630ctl_disable_lun(struct ctl_be_lun *be_lun)
4631{
4632	struct ctl_softc *ctl_softc;
4633	struct ctl_frontend *fe;
4634	struct ctl_lun *lun;
4635	int retval;
4636
4637	ctl_softc = control_softc;
4638
4639	lun = (struct ctl_lun *)be_lun->ctl_lun;
4640
4641	mtx_lock(&ctl_softc->ctl_lock);
4642
4643	if (lun->flags & CTL_LUN_DISABLED) {
4644		mtx_unlock(&ctl_softc->ctl_lock);
4645		return (0);
4646	}
4647	lun->flags |= CTL_LUN_DISABLED;
4648
4649	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4650		mtx_unlock(&ctl_softc->ctl_lock);
4651		/*
4652		 * Drop the lock before we call the frontend's disable
4653		 * routine, to avoid lock order reversals.
4654		 *
4655		 * XXX KDM what happens if the frontend list changes while
4656		 * we're traversing it?  It's unlikely, but should be handled.
4657		 */
4658		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4659					 lun->lun);
4660		mtx_lock(&ctl_softc->ctl_lock);
4661		if (retval != 0) {
4662			printf("ctl_alloc_lun: FETD %s port %d returned error "
4663			       "%d for lun_disable on target %ju lun %jd\n",
4664			       fe->port_name, fe->targ_port, retval,
4665			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4666		}
4667	}
4668
4669	mtx_unlock(&ctl_softc->ctl_lock);
4670
4671	return (0);
4672}
4673
4674int
4675ctl_start_lun(struct ctl_be_lun *be_lun)
4676{
4677	struct ctl_softc *ctl_softc;
4678	struct ctl_lun *lun;
4679
4680	ctl_softc = control_softc;
4681
4682	lun = (struct ctl_lun *)be_lun->ctl_lun;
4683
4684	mtx_lock(&ctl_softc->ctl_lock);
4685	lun->flags &= ~CTL_LUN_STOPPED;
4686	mtx_unlock(&ctl_softc->ctl_lock);
4687
4688	return (0);
4689}
4690
4691int
4692ctl_stop_lun(struct ctl_be_lun *be_lun)
4693{
4694	struct ctl_softc *ctl_softc;
4695	struct ctl_lun *lun;
4696
4697	ctl_softc = control_softc;
4698
4699	lun = (struct ctl_lun *)be_lun->ctl_lun;
4700
4701	mtx_lock(&ctl_softc->ctl_lock);
4702	lun->flags |= CTL_LUN_STOPPED;
4703	mtx_unlock(&ctl_softc->ctl_lock);
4704
4705	return (0);
4706}
4707
4708int
4709ctl_lun_offline(struct ctl_be_lun *be_lun)
4710{
4711	struct ctl_softc *ctl_softc;
4712	struct ctl_lun *lun;
4713
4714	ctl_softc = control_softc;
4715
4716	lun = (struct ctl_lun *)be_lun->ctl_lun;
4717
4718	mtx_lock(&ctl_softc->ctl_lock);
4719	lun->flags |= CTL_LUN_OFFLINE;
4720	mtx_unlock(&ctl_softc->ctl_lock);
4721
4722	return (0);
4723}
4724
4725int
4726ctl_lun_online(struct ctl_be_lun *be_lun)
4727{
4728	struct ctl_softc *ctl_softc;
4729	struct ctl_lun *lun;
4730
4731	ctl_softc = control_softc;
4732
4733	lun = (struct ctl_lun *)be_lun->ctl_lun;
4734
4735	mtx_lock(&ctl_softc->ctl_lock);
4736	lun->flags &= ~CTL_LUN_OFFLINE;
4737	mtx_unlock(&ctl_softc->ctl_lock);
4738
4739	return (0);
4740}
4741
4742int
4743ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4744{
4745	struct ctl_softc *ctl_softc;
4746	struct ctl_lun *lun;
4747
4748	ctl_softc = control_softc;
4749
4750	lun = (struct ctl_lun *)be_lun->ctl_lun;
4751
4752	mtx_lock(&ctl_softc->ctl_lock);
4753
4754	/*
4755	 * The LUN needs to be disabled before it can be marked invalid.
4756	 */
4757	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4758		mtx_unlock(&ctl_softc->ctl_lock);
4759		return (-1);
4760	}
4761	/*
4762	 * Mark the LUN invalid.
4763	 */
4764	lun->flags |= CTL_LUN_INVALID;
4765
4766	/*
4767	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4768	 * If we have something in the OOA queue, we'll free it when the
4769	 * last I/O completes.
4770	 */
4771	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4772		ctl_free_lun(lun);
4773	mtx_unlock(&ctl_softc->ctl_lock);
4774
4775	return (0);
4776}
4777
4778int
4779ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4780{
4781	struct ctl_softc *ctl_softc;
4782	struct ctl_lun *lun;
4783
4784	ctl_softc = control_softc;
4785	lun = (struct ctl_lun *)be_lun->ctl_lun;
4786
4787	mtx_lock(&ctl_softc->ctl_lock);
4788	lun->flags |= CTL_LUN_INOPERABLE;
4789	mtx_unlock(&ctl_softc->ctl_lock);
4790
4791	return (0);
4792}
4793
4794int
4795ctl_lun_operable(struct ctl_be_lun *be_lun)
4796{
4797	struct ctl_softc *ctl_softc;
4798	struct ctl_lun *lun;
4799
4800	ctl_softc = control_softc;
4801	lun = (struct ctl_lun *)be_lun->ctl_lun;
4802
4803	mtx_lock(&ctl_softc->ctl_lock);
4804	lun->flags &= ~CTL_LUN_INOPERABLE;
4805	mtx_unlock(&ctl_softc->ctl_lock);
4806
4807	return (0);
4808}
4809
4810int
4811ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4812		   int lock)
4813{
4814	struct ctl_softc *softc;
4815	struct ctl_lun *lun;
4816	struct copan_aps_subpage *current_sp;
4817	struct ctl_page_index *page_index;
4818	int i;
4819
4820	softc = control_softc;
4821
4822	mtx_lock(&softc->ctl_lock);
4823
4824	lun = (struct ctl_lun *)be_lun->ctl_lun;
4825
4826	page_index = NULL;
4827	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4828		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4829		     APS_PAGE_CODE)
4830			continue;
4831
4832		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4833			continue;
4834		page_index = &lun->mode_pages.index[i];
4835	}
4836
4837	if (page_index == NULL) {
4838		mtx_unlock(&softc->ctl_lock);
4839		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4840		       (uintmax_t)lun->lun);
4841		return (1);
4842	}
4843#if 0
4844	if ((softc->aps_locked_lun != 0)
4845	 && (softc->aps_locked_lun != lun->lun)) {
4846		printf("%s: attempt to lock LUN %llu when %llu is already "
4847		       "locked\n");
4848		mtx_unlock(&softc->ctl_lock);
4849		return (1);
4850	}
4851#endif
4852
4853	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4854		(page_index->page_len * CTL_PAGE_CURRENT));
4855
4856	if (lock != 0) {
4857		current_sp->lock_active = APS_LOCK_ACTIVE;
4858		softc->aps_locked_lun = lun->lun;
4859	} else {
4860		current_sp->lock_active = 0;
4861		softc->aps_locked_lun = 0;
4862	}
4863
4864
4865	/*
4866	 * If we're in HA mode, try to send the lock message to the other
4867	 * side.
4868	 */
4869	if (ctl_is_single == 0) {
4870		int isc_retval;
4871		union ctl_ha_msg lock_msg;
4872
4873		lock_msg.hdr.nexus = *nexus;
4874		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4875		if (lock != 0)
4876			lock_msg.aps.lock_flag = 1;
4877		else
4878			lock_msg.aps.lock_flag = 0;
4879		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4880					 sizeof(lock_msg), 0);
4881		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4882			printf("%s: APS (lock=%d) error returned from "
4883			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4884			mtx_unlock(&softc->ctl_lock);
4885			return (1);
4886		}
4887	}
4888
4889	mtx_unlock(&softc->ctl_lock);
4890
4891	return (0);
4892}
4893
4894void
4895ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4896{
4897	struct ctl_lun *lun;
4898	struct ctl_softc *softc;
4899	int i;
4900
4901	softc = control_softc;
4902
4903	mtx_lock(&softc->ctl_lock);
4904
4905	lun = (struct ctl_lun *)be_lun->ctl_lun;
4906
4907	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4908		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4909
4910	mtx_unlock(&softc->ctl_lock);
4911}
4912
4913/*
4914 * Backend "memory move is complete" callback for requests that never
4915 * make it down to say RAIDCore's configuration code.
4916 */
4917int
4918ctl_config_move_done(union ctl_io *io)
4919{
4920	int retval;
4921
4922	retval = CTL_RETVAL_COMPLETE;
4923
4924
4925	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4926	/*
4927	 * XXX KDM this shouldn't happen, but what if it does?
4928	 */
4929	if (io->io_hdr.io_type != CTL_IO_SCSI)
4930		panic("I/O type isn't CTL_IO_SCSI!");
4931
4932	if ((io->io_hdr.port_status == 0)
4933	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4934	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4935		io->io_hdr.status = CTL_SUCCESS;
4936	else if ((io->io_hdr.port_status != 0)
4937	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4938	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4939		/*
4940		 * For hardware error sense keys, the sense key
4941		 * specific value is defined to be a retry count,
4942		 * but we use it to pass back an internal FETD
4943		 * error code.  XXX KDM  Hopefully the FETD is only
4944		 * using 16 bits for an error code, since that's
4945		 * all the space we have in the sks field.
4946		 */
4947		ctl_set_internal_failure(&io->scsiio,
4948					 /*sks_valid*/ 1,
4949					 /*retry_count*/
4950					 io->io_hdr.port_status);
4951		free(io->scsiio.kern_data_ptr, M_CTL);
4952		ctl_done(io);
4953		goto bailout;
4954	}
4955
4956	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4957	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4958	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4959		/*
4960		 * XXX KDM just assuming a single pointer here, and not a
4961		 * S/G list.  If we start using S/G lists for config data,
4962		 * we'll need to know how to clean them up here as well.
4963		 */
4964		free(io->scsiio.kern_data_ptr, M_CTL);
4965		/* Hopefully the user has already set the status... */
4966		ctl_done(io);
4967	} else {
4968		/*
4969		 * XXX KDM now we need to continue data movement.  Some
4970		 * options:
4971		 * - call ctl_scsiio() again?  We don't do this for data
4972		 *   writes, because for those at least we know ahead of
4973		 *   time where the write will go and how long it is.  For
4974		 *   config writes, though, that information is largely
4975		 *   contained within the write itself, thus we need to
4976		 *   parse out the data again.
4977		 *
4978		 * - Call some other function once the data is in?
4979		 */
4980
4981		/*
4982		 * XXX KDM call ctl_scsiio() again for now, and check flag
4983		 * bits to see whether we're allocated or not.
4984		 */
4985		retval = ctl_scsiio(&io->scsiio);
4986	}
4987bailout:
4988	return (retval);
4989}
4990
4991/*
4992 * This gets called by a backend driver when it is done with a
4993 * data_submit method.
4994 */
4995void
4996ctl_data_submit_done(union ctl_io *io)
4997{
4998	/*
4999	 * If the IO_CONT flag is set, we need to call the supplied
5000	 * function to continue processing the I/O, instead of completing
5001	 * the I/O just yet.
5002	 *
5003	 * If there is an error, though, we don't want to keep processing.
5004	 * Instead, just send status back to the initiator.
5005	 */
5006	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5007	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5008	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5009		io->scsiio.io_cont(io);
5010		return;
5011	}
5012	ctl_done(io);
5013}
5014
5015/*
5016 * This gets called by a backend driver when it is done with a
5017 * configuration write.
5018 */
5019void
5020ctl_config_write_done(union ctl_io *io)
5021{
5022	/*
5023	 * If the IO_CONT flag is set, we need to call the supplied
5024	 * function to continue processing the I/O, instead of completing
5025	 * the I/O just yet.
5026	 *
5027	 * If there is an error, though, we don't want to keep processing.
5028	 * Instead, just send status back to the initiator.
5029	 */
5030	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5031	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5032	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5033		io->scsiio.io_cont(io);
5034		return;
5035	}
5036	/*
5037	 * Since a configuration write can be done for commands that actually
5038	 * have data allocated, like write buffer, and commands that have
5039	 * no data, like start/stop unit, we need to check here.
5040	 */
5041	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5042		free(io->scsiio.kern_data_ptr, M_CTL);
5043	ctl_done(io);
5044}
5045
5046/*
5047 * SCSI release command.
5048 */
5049int
5050ctl_scsi_release(struct ctl_scsiio *ctsio)
5051{
5052	int length, longid, thirdparty_id, resv_id;
5053	struct ctl_softc *ctl_softc;
5054	struct ctl_lun *lun;
5055
5056	length = 0;
5057	resv_id = 0;
5058
5059	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5060
5061	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5062	ctl_softc = control_softc;
5063
5064	switch (ctsio->cdb[0]) {
5065	case RELEASE: {
5066		struct scsi_release *cdb;
5067
5068		cdb = (struct scsi_release *)ctsio->cdb;
5069		if ((cdb->byte2 & 0x1f) != 0) {
5070			ctl_set_invalid_field(ctsio,
5071					      /*sks_valid*/ 1,
5072					      /*command*/ 1,
5073					      /*field*/ 1,
5074					      /*bit_valid*/ 0,
5075					      /*bit*/ 0);
5076			ctl_done((union ctl_io *)ctsio);
5077			return (CTL_RETVAL_COMPLETE);
5078		}
5079		break;
5080	}
5081	case RELEASE_10: {
5082		struct scsi_release_10 *cdb;
5083
5084		cdb = (struct scsi_release_10 *)ctsio->cdb;
5085
5086		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5087			ctl_set_invalid_field(ctsio,
5088					      /*sks_valid*/ 1,
5089					      /*command*/ 1,
5090					      /*field*/ 1,
5091					      /*bit_valid*/ 1,
5092					      /*bit*/ 0);
5093			ctl_done((union ctl_io *)ctsio);
5094			return (CTL_RETVAL_COMPLETE);
5095
5096		}
5097
5098		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5099			ctl_set_invalid_field(ctsio,
5100					      /*sks_valid*/ 1,
5101					      /*command*/ 1,
5102					      /*field*/ 1,
5103					      /*bit_valid*/ 1,
5104					      /*bit*/ 4);
5105			ctl_done((union ctl_io *)ctsio);
5106			return (CTL_RETVAL_COMPLETE);
5107		}
5108
5109		if (cdb->byte2 & SR10_LONGID)
5110			longid = 1;
5111		else
5112			thirdparty_id = cdb->thirdparty_id;
5113
5114		resv_id = cdb->resv_id;
5115		length = scsi_2btoul(cdb->length);
5116		break;
5117	}
5118	}
5119
5120
5121	/*
5122	 * XXX KDM right now, we only support LUN reservation.  We don't
5123	 * support 3rd party reservations, or extent reservations, which
5124	 * might actually need the parameter list.  If we've gotten this
5125	 * far, we've got a LUN reservation.  Anything else got kicked out
5126	 * above.  So, according to SPC, ignore the length.
5127	 */
5128	length = 0;
5129
5130	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5131	 && (length > 0)) {
5132		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5133		ctsio->kern_data_len = length;
5134		ctsio->kern_total_len = length;
5135		ctsio->kern_data_resid = 0;
5136		ctsio->kern_rel_offset = 0;
5137		ctsio->kern_sg_entries = 0;
5138		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5139		ctsio->be_move_done = ctl_config_move_done;
5140		ctl_datamove((union ctl_io *)ctsio);
5141
5142		return (CTL_RETVAL_COMPLETE);
5143	}
5144
5145	if (length > 0)
5146		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5147
5148	mtx_lock(&ctl_softc->ctl_lock);
5149
5150	/*
5151	 * According to SPC, it is not an error for an intiator to attempt
5152	 * to release a reservation on a LUN that isn't reserved, or that
5153	 * is reserved by another initiator.  The reservation can only be
5154	 * released, though, by the initiator who made it or by one of
5155	 * several reset type events.
5156	 */
5157	if (lun->flags & CTL_LUN_RESERVED) {
5158		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5159		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5160		 && (ctsio->io_hdr.nexus.targ_target.id ==
5161		     lun->rsv_nexus.targ_target.id)) {
5162			lun->flags &= ~CTL_LUN_RESERVED;
5163		}
5164	}
5165
5166	ctsio->scsi_status = SCSI_STATUS_OK;
5167	ctsio->io_hdr.status = CTL_SUCCESS;
5168
5169	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5170		free(ctsio->kern_data_ptr, M_CTL);
5171		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5172	}
5173
5174	mtx_unlock(&ctl_softc->ctl_lock);
5175
5176	ctl_done((union ctl_io *)ctsio);
5177	return (CTL_RETVAL_COMPLETE);
5178}
5179
5180int
5181ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5182{
5183	int extent, thirdparty, longid;
5184	int resv_id, length;
5185	uint64_t thirdparty_id;
5186	struct ctl_softc *ctl_softc;
5187	struct ctl_lun *lun;
5188
5189	extent = 0;
5190	thirdparty = 0;
5191	longid = 0;
5192	resv_id = 0;
5193	length = 0;
5194	thirdparty_id = 0;
5195
5196	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5197
5198	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5199	ctl_softc = control_softc;
5200
5201	switch (ctsio->cdb[0]) {
5202	case RESERVE: {
5203		struct scsi_reserve *cdb;
5204
5205		cdb = (struct scsi_reserve *)ctsio->cdb;
5206		if ((cdb->byte2 & 0x1f) != 0) {
5207			ctl_set_invalid_field(ctsio,
5208					      /*sks_valid*/ 1,
5209					      /*command*/ 1,
5210					      /*field*/ 1,
5211					      /*bit_valid*/ 0,
5212					      /*bit*/ 0);
5213			ctl_done((union ctl_io *)ctsio);
5214			return (CTL_RETVAL_COMPLETE);
5215		}
5216		resv_id = cdb->resv_id;
5217		length = scsi_2btoul(cdb->length);
5218		break;
5219	}
5220	case RESERVE_10: {
5221		struct scsi_reserve_10 *cdb;
5222
5223		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5224
5225		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5226			ctl_set_invalid_field(ctsio,
5227					      /*sks_valid*/ 1,
5228					      /*command*/ 1,
5229					      /*field*/ 1,
5230					      /*bit_valid*/ 1,
5231					      /*bit*/ 0);
5232			ctl_done((union ctl_io *)ctsio);
5233			return (CTL_RETVAL_COMPLETE);
5234		}
5235		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5236			ctl_set_invalid_field(ctsio,
5237					      /*sks_valid*/ 1,
5238					      /*command*/ 1,
5239					      /*field*/ 1,
5240					      /*bit_valid*/ 1,
5241					      /*bit*/ 4);
5242			ctl_done((union ctl_io *)ctsio);
5243			return (CTL_RETVAL_COMPLETE);
5244		}
5245		if (cdb->byte2 & SR10_LONGID)
5246			longid = 1;
5247		else
5248			thirdparty_id = cdb->thirdparty_id;
5249
5250		resv_id = cdb->resv_id;
5251		length = scsi_2btoul(cdb->length);
5252		break;
5253	}
5254	}
5255
5256	/*
5257	 * XXX KDM right now, we only support LUN reservation.  We don't
5258	 * support 3rd party reservations, or extent reservations, which
5259	 * might actually need the parameter list.  If we've gotten this
5260	 * far, we've got a LUN reservation.  Anything else got kicked out
5261	 * above.  So, according to SPC, ignore the length.
5262	 */
5263	length = 0;
5264
5265	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5266	 && (length > 0)) {
5267		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5268		ctsio->kern_data_len = length;
5269		ctsio->kern_total_len = length;
5270		ctsio->kern_data_resid = 0;
5271		ctsio->kern_rel_offset = 0;
5272		ctsio->kern_sg_entries = 0;
5273		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5274		ctsio->be_move_done = ctl_config_move_done;
5275		ctl_datamove((union ctl_io *)ctsio);
5276
5277		return (CTL_RETVAL_COMPLETE);
5278	}
5279
5280	if (length > 0)
5281		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5282
5283	mtx_lock(&ctl_softc->ctl_lock);
5284	if (lun->flags & CTL_LUN_RESERVED) {
5285		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5286		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5287		 || (ctsio->io_hdr.nexus.targ_target.id !=
5288		     lun->rsv_nexus.targ_target.id)) {
5289			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5290			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5291			goto bailout;
5292		}
5293	}
5294
5295	lun->flags |= CTL_LUN_RESERVED;
5296	lun->rsv_nexus = ctsio->io_hdr.nexus;
5297
5298	ctsio->scsi_status = SCSI_STATUS_OK;
5299	ctsio->io_hdr.status = CTL_SUCCESS;
5300
5301bailout:
5302	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5303		free(ctsio->kern_data_ptr, M_CTL);
5304		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5305	}
5306
5307	mtx_unlock(&ctl_softc->ctl_lock);
5308
5309	ctl_done((union ctl_io *)ctsio);
5310	return (CTL_RETVAL_COMPLETE);
5311}
5312
5313int
5314ctl_start_stop(struct ctl_scsiio *ctsio)
5315{
5316	struct scsi_start_stop_unit *cdb;
5317	struct ctl_lun *lun;
5318	struct ctl_softc *ctl_softc;
5319	int retval;
5320
5321	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5322
5323	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5324	ctl_softc = control_softc;
5325	retval = 0;
5326
5327	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5328
5329	/*
5330	 * XXX KDM
5331	 * We don't support the immediate bit on a stop unit.  In order to
5332	 * do that, we would need to code up a way to know that a stop is
5333	 * pending, and hold off any new commands until it completes, one
5334	 * way or another.  Then we could accept or reject those commands
5335	 * depending on its status.  We would almost need to do the reverse
5336	 * of what we do below for an immediate start -- return the copy of
5337	 * the ctl_io to the FETD with status to send to the host (and to
5338	 * free the copy!) and then free the original I/O once the stop
5339	 * actually completes.  That way, the OOA queue mechanism can work
5340	 * to block commands that shouldn't proceed.  Another alternative
5341	 * would be to put the copy in the queue in place of the original,
5342	 * and return the original back to the caller.  That could be
5343	 * slightly safer..
5344	 */
5345	if ((cdb->byte2 & SSS_IMMED)
5346	 && ((cdb->how & SSS_START) == 0)) {
5347		ctl_set_invalid_field(ctsio,
5348				      /*sks_valid*/ 1,
5349				      /*command*/ 1,
5350				      /*field*/ 1,
5351				      /*bit_valid*/ 1,
5352				      /*bit*/ 0);
5353		ctl_done((union ctl_io *)ctsio);
5354		return (CTL_RETVAL_COMPLETE);
5355	}
5356
5357	/*
5358	 * We don't support the power conditions field.  We need to check
5359	 * this prior to checking the load/eject and start/stop bits.
5360	 */
5361	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5362		ctl_set_invalid_field(ctsio,
5363				      /*sks_valid*/ 1,
5364				      /*command*/ 1,
5365				      /*field*/ 4,
5366				      /*bit_valid*/ 1,
5367				      /*bit*/ 4);
5368		ctl_done((union ctl_io *)ctsio);
5369		return (CTL_RETVAL_COMPLETE);
5370	}
5371
5372	/*
5373	 * Media isn't removable, so we can't load or eject it.
5374	 */
5375	if ((cdb->how & SSS_LOEJ) != 0) {
5376		ctl_set_invalid_field(ctsio,
5377				      /*sks_valid*/ 1,
5378				      /*command*/ 1,
5379				      /*field*/ 4,
5380				      /*bit_valid*/ 1,
5381				      /*bit*/ 1);
5382		ctl_done((union ctl_io *)ctsio);
5383		return (CTL_RETVAL_COMPLETE);
5384	}
5385
5386	if ((lun->flags & CTL_LUN_PR_RESERVED)
5387	 && ((cdb->how & SSS_START)==0)) {
5388		uint32_t residx;
5389
5390		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5391		if (!lun->per_res[residx].registered
5392		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5393
5394			ctl_set_reservation_conflict(ctsio);
5395			ctl_done((union ctl_io *)ctsio);
5396			return (CTL_RETVAL_COMPLETE);
5397		}
5398	}
5399
5400	/*
5401	 * If there is no backend on this device, we can't start or stop
5402	 * it.  In theory we shouldn't get any start/stop commands in the
5403	 * first place at this level if the LUN doesn't have a backend.
5404	 * That should get stopped by the command decode code.
5405	 */
5406	if (lun->backend == NULL) {
5407		ctl_set_invalid_opcode(ctsio);
5408		ctl_done((union ctl_io *)ctsio);
5409		return (CTL_RETVAL_COMPLETE);
5410	}
5411
5412	/*
5413	 * XXX KDM Copan-specific offline behavior.
5414	 * Figure out a reasonable way to port this?
5415	 */
5416#ifdef NEEDTOPORT
5417	mtx_lock(&ctl_softc->ctl_lock);
5418
5419	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5420	 && (lun->flags & CTL_LUN_OFFLINE)) {
5421		/*
5422		 * If the LUN is offline, and the on/offline bit isn't set,
5423		 * reject the start or stop.  Otherwise, let it through.
5424		 */
5425		mtx_unlock(&ctl_softc->ctl_lock);
5426		ctl_set_lun_not_ready(ctsio);
5427		ctl_done((union ctl_io *)ctsio);
5428	} else {
5429		mtx_unlock(&ctl_softc->ctl_lock);
5430#endif /* NEEDTOPORT */
5431		/*
5432		 * This could be a start or a stop when we're online,
5433		 * or a stop/offline or start/online.  A start or stop when
5434		 * we're offline is covered in the case above.
5435		 */
5436		/*
5437		 * In the non-immediate case, we send the request to
5438		 * the backend and return status to the user when
5439		 * it is done.
5440		 *
5441		 * In the immediate case, we allocate a new ctl_io
5442		 * to hold a copy of the request, and send that to
5443		 * the backend.  We then set good status on the
5444		 * user's request and return it immediately.
5445		 */
5446		if (cdb->byte2 & SSS_IMMED) {
5447			union ctl_io *new_io;
5448
5449			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5450			if (new_io == NULL) {
5451				ctl_set_busy(ctsio);
5452				ctl_done((union ctl_io *)ctsio);
5453			} else {
5454				ctl_copy_io((union ctl_io *)ctsio,
5455					    new_io);
5456				retval = lun->backend->config_write(new_io);
5457				ctl_set_success(ctsio);
5458				ctl_done((union ctl_io *)ctsio);
5459			}
5460		} else {
5461			retval = lun->backend->config_write(
5462				(union ctl_io *)ctsio);
5463		}
5464#ifdef NEEDTOPORT
5465	}
5466#endif
5467	return (retval);
5468}
5469
5470/*
5471 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5472 * we don't really do anything with the LBA and length fields if the user
5473 * passes them in.  Instead we'll just flush out the cache for the entire
5474 * LUN.
5475 */
5476int
5477ctl_sync_cache(struct ctl_scsiio *ctsio)
5478{
5479	struct ctl_lun *lun;
5480	struct ctl_softc *ctl_softc;
5481	uint64_t starting_lba;
5482	uint32_t block_count;
5483	int reladr, immed;
5484	int retval;
5485
5486	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5487
5488	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5489	ctl_softc = control_softc;
5490	retval = 0;
5491	reladr = 0;
5492	immed = 0;
5493
5494	switch (ctsio->cdb[0]) {
5495	case SYNCHRONIZE_CACHE: {
5496		struct scsi_sync_cache *cdb;
5497		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5498
5499		if (cdb->byte2 & SSC_RELADR)
5500			reladr = 1;
5501
5502		if (cdb->byte2 & SSC_IMMED)
5503			immed = 1;
5504
5505		starting_lba = scsi_4btoul(cdb->begin_lba);
5506		block_count = scsi_2btoul(cdb->lb_count);
5507		break;
5508	}
5509	case SYNCHRONIZE_CACHE_16: {
5510		struct scsi_sync_cache_16 *cdb;
5511		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5512
5513		if (cdb->byte2 & SSC_RELADR)
5514			reladr = 1;
5515
5516		if (cdb->byte2 & SSC_IMMED)
5517			immed = 1;
5518
5519		starting_lba = scsi_8btou64(cdb->begin_lba);
5520		block_count = scsi_4btoul(cdb->lb_count);
5521		break;
5522	}
5523	default:
5524		ctl_set_invalid_opcode(ctsio);
5525		ctl_done((union ctl_io *)ctsio);
5526		goto bailout;
5527		break; /* NOTREACHED */
5528	}
5529
5530	if (immed) {
5531		/*
5532		 * We don't support the immediate bit.  Since it's in the
5533		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5534		 * commands, we can just return the same error in either
5535		 * case.
5536		 */
5537		ctl_set_invalid_field(ctsio,
5538				      /*sks_valid*/ 1,
5539				      /*command*/ 1,
5540				      /*field*/ 1,
5541				      /*bit_valid*/ 1,
5542				      /*bit*/ 1);
5543		ctl_done((union ctl_io *)ctsio);
5544		goto bailout;
5545	}
5546
5547	if (reladr) {
5548		/*
5549		 * We don't support the reladr bit either.  It can only be
5550		 * used with linked commands, and we don't support linked
5551		 * commands.  Since the bit is in the same place for the
5552		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5553		 * just return the same error in either case.
5554		 */
5555		ctl_set_invalid_field(ctsio,
5556				      /*sks_valid*/ 1,
5557				      /*command*/ 1,
5558				      /*field*/ 1,
5559				      /*bit_valid*/ 1,
5560				      /*bit*/ 0);
5561		ctl_done((union ctl_io *)ctsio);
5562		goto bailout;
5563	}
5564
5565	/*
5566	 * We check the LBA and length, but don't do anything with them.
5567	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5568	 * get flushed.  This check will just help satisfy anyone who wants
5569	 * to see an error for an out of range LBA.
5570	 */
5571	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5572		ctl_set_lba_out_of_range(ctsio);
5573		ctl_done((union ctl_io *)ctsio);
5574		goto bailout;
5575	}
5576
5577	/*
5578	 * If this LUN has no backend, we can't flush the cache anyway.
5579	 */
5580	if (lun->backend == NULL) {
5581		ctl_set_invalid_opcode(ctsio);
5582		ctl_done((union ctl_io *)ctsio);
5583		goto bailout;
5584	}
5585
5586	/*
5587	 * Check to see whether we're configured to send the SYNCHRONIZE
5588	 * CACHE command directly to the back end.
5589	 */
5590	mtx_lock(&ctl_softc->ctl_lock);
5591	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5592	 && (++(lun->sync_count) >= lun->sync_interval)) {
5593		lun->sync_count = 0;
5594		mtx_unlock(&ctl_softc->ctl_lock);
5595		retval = lun->backend->config_write((union ctl_io *)ctsio);
5596	} else {
5597		mtx_unlock(&ctl_softc->ctl_lock);
5598		ctl_set_success(ctsio);
5599		ctl_done((union ctl_io *)ctsio);
5600	}
5601
5602bailout:
5603
5604	return (retval);
5605}
5606
5607int
5608ctl_format(struct ctl_scsiio *ctsio)
5609{
5610	struct scsi_format *cdb;
5611	struct ctl_lun *lun;
5612	struct ctl_softc *ctl_softc;
5613	int length, defect_list_len;
5614
5615	CTL_DEBUG_PRINT(("ctl_format\n"));
5616
5617	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5618	ctl_softc = control_softc;
5619
5620	cdb = (struct scsi_format *)ctsio->cdb;
5621
5622	length = 0;
5623	if (cdb->byte2 & SF_FMTDATA) {
5624		if (cdb->byte2 & SF_LONGLIST)
5625			length = sizeof(struct scsi_format_header_long);
5626		else
5627			length = sizeof(struct scsi_format_header_short);
5628	}
5629
5630	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5631	 && (length > 0)) {
5632		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5633		ctsio->kern_data_len = length;
5634		ctsio->kern_total_len = length;
5635		ctsio->kern_data_resid = 0;
5636		ctsio->kern_rel_offset = 0;
5637		ctsio->kern_sg_entries = 0;
5638		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5639		ctsio->be_move_done = ctl_config_move_done;
5640		ctl_datamove((union ctl_io *)ctsio);
5641
5642		return (CTL_RETVAL_COMPLETE);
5643	}
5644
5645	defect_list_len = 0;
5646
5647	if (cdb->byte2 & SF_FMTDATA) {
5648		if (cdb->byte2 & SF_LONGLIST) {
5649			struct scsi_format_header_long *header;
5650
5651			header = (struct scsi_format_header_long *)
5652				ctsio->kern_data_ptr;
5653
5654			defect_list_len = scsi_4btoul(header->defect_list_len);
5655			if (defect_list_len != 0) {
5656				ctl_set_invalid_field(ctsio,
5657						      /*sks_valid*/ 1,
5658						      /*command*/ 0,
5659						      /*field*/ 2,
5660						      /*bit_valid*/ 0,
5661						      /*bit*/ 0);
5662				goto bailout;
5663			}
5664		} else {
5665			struct scsi_format_header_short *header;
5666
5667			header = (struct scsi_format_header_short *)
5668				ctsio->kern_data_ptr;
5669
5670			defect_list_len = scsi_2btoul(header->defect_list_len);
5671			if (defect_list_len != 0) {
5672				ctl_set_invalid_field(ctsio,
5673						      /*sks_valid*/ 1,
5674						      /*command*/ 0,
5675						      /*field*/ 2,
5676						      /*bit_valid*/ 0,
5677						      /*bit*/ 0);
5678				goto bailout;
5679			}
5680		}
5681	}
5682
5683	/*
5684	 * The format command will clear out the "Medium format corrupted"
5685	 * status if set by the configuration code.  That status is really
5686	 * just a way to notify the host that we have lost the media, and
5687	 * get them to issue a command that will basically make them think
5688	 * they're blowing away the media.
5689	 */
5690	mtx_lock(&ctl_softc->ctl_lock);
5691	lun->flags &= ~CTL_LUN_INOPERABLE;
5692	mtx_unlock(&ctl_softc->ctl_lock);
5693
5694	ctsio->scsi_status = SCSI_STATUS_OK;
5695	ctsio->io_hdr.status = CTL_SUCCESS;
5696bailout:
5697
5698	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5699		free(ctsio->kern_data_ptr, M_CTL);
5700		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5701	}
5702
5703	ctl_done((union ctl_io *)ctsio);
5704	return (CTL_RETVAL_COMPLETE);
5705}
5706
5707int
5708ctl_write_buffer(struct ctl_scsiio *ctsio)
5709{
5710	struct scsi_write_buffer *cdb;
5711	struct copan_page_header *header;
5712	struct ctl_lun *lun;
5713	struct ctl_softc *ctl_softc;
5714	int buffer_offset, len;
5715	int retval;
5716
5717	header = NULL;
5718
5719	retval = CTL_RETVAL_COMPLETE;
5720
5721	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5722
5723	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5724	ctl_softc = control_softc;
5725	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5726
5727	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5728		ctl_set_invalid_field(ctsio,
5729				      /*sks_valid*/ 1,
5730				      /*command*/ 1,
5731				      /*field*/ 1,
5732				      /*bit_valid*/ 1,
5733				      /*bit*/ 4);
5734		ctl_done((union ctl_io *)ctsio);
5735		return (CTL_RETVAL_COMPLETE);
5736	}
5737	if (cdb->buffer_id != 0) {
5738		ctl_set_invalid_field(ctsio,
5739				      /*sks_valid*/ 1,
5740				      /*command*/ 1,
5741				      /*field*/ 2,
5742				      /*bit_valid*/ 0,
5743				      /*bit*/ 0);
5744		ctl_done((union ctl_io *)ctsio);
5745		return (CTL_RETVAL_COMPLETE);
5746	}
5747
5748	len = scsi_3btoul(cdb->length);
5749	buffer_offset = scsi_3btoul(cdb->offset);
5750
5751	if (len > sizeof(lun->write_buffer)) {
5752		ctl_set_invalid_field(ctsio,
5753				      /*sks_valid*/ 1,
5754				      /*command*/ 1,
5755				      /*field*/ 6,
5756				      /*bit_valid*/ 0,
5757				      /*bit*/ 0);
5758		ctl_done((union ctl_io *)ctsio);
5759		return (CTL_RETVAL_COMPLETE);
5760	}
5761
5762	if (buffer_offset != 0) {
5763		ctl_set_invalid_field(ctsio,
5764				      /*sks_valid*/ 1,
5765				      /*command*/ 1,
5766				      /*field*/ 3,
5767				      /*bit_valid*/ 0,
5768				      /*bit*/ 0);
5769		ctl_done((union ctl_io *)ctsio);
5770		return (CTL_RETVAL_COMPLETE);
5771	}
5772
5773	/*
5774	 * If we've got a kernel request that hasn't been malloced yet,
5775	 * malloc it and tell the caller the data buffer is here.
5776	 */
5777	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5778		ctsio->kern_data_ptr = lun->write_buffer;
5779		ctsio->kern_data_len = len;
5780		ctsio->kern_total_len = len;
5781		ctsio->kern_data_resid = 0;
5782		ctsio->kern_rel_offset = 0;
5783		ctsio->kern_sg_entries = 0;
5784		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5785		ctsio->be_move_done = ctl_config_move_done;
5786		ctl_datamove((union ctl_io *)ctsio);
5787
5788		return (CTL_RETVAL_COMPLETE);
5789	}
5790
5791	ctl_done((union ctl_io *)ctsio);
5792
5793	return (CTL_RETVAL_COMPLETE);
5794}
5795
5796int
5797ctl_write_same(struct ctl_scsiio *ctsio)
5798{
5799	struct ctl_lun *lun;
5800	struct ctl_lba_len_flags *lbalen;
5801	uint64_t lba;
5802	uint32_t num_blocks;
5803	int len, retval;
5804	uint8_t byte2;
5805
5806	retval = CTL_RETVAL_COMPLETE;
5807
5808	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5809
5810	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5811
5812	switch (ctsio->cdb[0]) {
5813	case WRITE_SAME_10: {
5814		struct scsi_write_same_10 *cdb;
5815
5816		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5817
5818		lba = scsi_4btoul(cdb->addr);
5819		num_blocks = scsi_2btoul(cdb->length);
5820		byte2 = cdb->byte2;
5821		break;
5822	}
5823	case WRITE_SAME_16: {
5824		struct scsi_write_same_16 *cdb;
5825
5826		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5827
5828		lba = scsi_8btou64(cdb->addr);
5829		num_blocks = scsi_4btoul(cdb->length);
5830		byte2 = cdb->byte2;
5831		break;
5832	}
5833	default:
5834		/*
5835		 * We got a command we don't support.  This shouldn't
5836		 * happen, commands should be filtered out above us.
5837		 */
5838		ctl_set_invalid_opcode(ctsio);
5839		ctl_done((union ctl_io *)ctsio);
5840
5841		return (CTL_RETVAL_COMPLETE);
5842		break; /* NOTREACHED */
5843	}
5844
5845	/*
5846	 * The first check is to make sure we're in bounds, the second
5847	 * check is to catch wrap-around problems.  If the lba + num blocks
5848	 * is less than the lba, then we've wrapped around and the block
5849	 * range is invalid anyway.
5850	 */
5851	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5852	 || ((lba + num_blocks) < lba)) {
5853		ctl_set_lba_out_of_range(ctsio);
5854		ctl_done((union ctl_io *)ctsio);
5855		return (CTL_RETVAL_COMPLETE);
5856	}
5857
5858	/* Zero number of blocks means "to the last logical block" */
5859	if (num_blocks == 0) {
5860		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5861			ctl_set_invalid_field(ctsio,
5862					      /*sks_valid*/ 0,
5863					      /*command*/ 1,
5864					      /*field*/ 0,
5865					      /*bit_valid*/ 0,
5866					      /*bit*/ 0);
5867			ctl_done((union ctl_io *)ctsio);
5868			return (CTL_RETVAL_COMPLETE);
5869		}
5870		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5871	}
5872
5873	len = lun->be_lun->blocksize;
5874
5875	/*
5876	 * If we've got a kernel request that hasn't been malloced yet,
5877	 * malloc it and tell the caller the data buffer is here.
5878	 */
5879	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5880		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5881		ctsio->kern_data_len = len;
5882		ctsio->kern_total_len = len;
5883		ctsio->kern_data_resid = 0;
5884		ctsio->kern_rel_offset = 0;
5885		ctsio->kern_sg_entries = 0;
5886		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5887		ctsio->be_move_done = ctl_config_move_done;
5888		ctl_datamove((union ctl_io *)ctsio);
5889
5890		return (CTL_RETVAL_COMPLETE);
5891	}
5892
5893	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5894	lbalen->lba = lba;
5895	lbalen->len = num_blocks;
5896	lbalen->flags = byte2;
5897	retval = lun->backend->config_write((union ctl_io *)ctsio);
5898
5899	return (retval);
5900}
5901
5902int
5903ctl_unmap(struct ctl_scsiio *ctsio)
5904{
5905	struct ctl_lun *lun;
5906	struct scsi_unmap *cdb;
5907	struct ctl_ptr_len_flags *ptrlen;
5908	struct scsi_unmap_header *hdr;
5909	struct scsi_unmap_desc *buf, *end;
5910	uint64_t lba;
5911	uint32_t num_blocks;
5912	int len, retval;
5913	uint8_t byte2;
5914
5915	retval = CTL_RETVAL_COMPLETE;
5916
5917	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5918
5919	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5920	cdb = (struct scsi_unmap *)ctsio->cdb;
5921
5922	len = scsi_2btoul(cdb->length);
5923	byte2 = cdb->byte2;
5924
5925	/*
5926	 * If we've got a kernel request that hasn't been malloced yet,
5927	 * malloc it and tell the caller the data buffer is here.
5928	 */
5929	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5930		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5931		ctsio->kern_data_len = len;
5932		ctsio->kern_total_len = len;
5933		ctsio->kern_data_resid = 0;
5934		ctsio->kern_rel_offset = 0;
5935		ctsio->kern_sg_entries = 0;
5936		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5937		ctsio->be_move_done = ctl_config_move_done;
5938		ctl_datamove((union ctl_io *)ctsio);
5939
5940		return (CTL_RETVAL_COMPLETE);
5941	}
5942
5943	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5944	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5945	if (len < sizeof (*hdr) ||
5946	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5947	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5948	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5949		ctl_set_invalid_field(ctsio,
5950				      /*sks_valid*/ 0,
5951				      /*command*/ 0,
5952				      /*field*/ 0,
5953				      /*bit_valid*/ 0,
5954				      /*bit*/ 0);
5955		ctl_done((union ctl_io *)ctsio);
5956		return (CTL_RETVAL_COMPLETE);
5957	}
5958	len = scsi_2btoul(hdr->desc_length);
5959	buf = (struct scsi_unmap_desc *)(hdr + 1);
5960	end = buf + len / sizeof(*buf);
5961
5962	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5963	ptrlen->ptr = (void *)buf;
5964	ptrlen->len = len;
5965	ptrlen->flags = byte2;
5966
5967	for (; buf < end; buf++) {
5968		lba = scsi_8btou64(buf->lba);
5969		num_blocks = scsi_4btoul(buf->length);
5970		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5971		 || ((lba + num_blocks) < lba)) {
5972			ctl_set_lba_out_of_range(ctsio);
5973			ctl_done((union ctl_io *)ctsio);
5974			return (CTL_RETVAL_COMPLETE);
5975		}
5976	}
5977
5978	retval = lun->backend->config_write((union ctl_io *)ctsio);
5979
5980	return (retval);
5981}
5982
5983/*
5984 * Note that this function currently doesn't actually do anything inside
5985 * CTL to enforce things if the DQue bit is turned on.
5986 *
5987 * Also note that this function can't be used in the default case, because
5988 * the DQue bit isn't set in the changeable mask for the control mode page
5989 * anyway.  This is just here as an example for how to implement a page
5990 * handler, and a placeholder in case we want to allow the user to turn
5991 * tagged queueing on and off.
5992 *
5993 * The D_SENSE bit handling is functional, however, and will turn
5994 * descriptor sense on and off for a given LUN.
5995 */
5996int
5997ctl_control_page_handler(struct ctl_scsiio *ctsio,
5998			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5999{
6000	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6001	struct ctl_lun *lun;
6002	struct ctl_softc *softc;
6003	int set_ua;
6004	uint32_t initidx;
6005
6006	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6007	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6008	set_ua = 0;
6009
6010	user_cp = (struct scsi_control_page *)page_ptr;
6011	current_cp = (struct scsi_control_page *)
6012		(page_index->page_data + (page_index->page_len *
6013		CTL_PAGE_CURRENT));
6014	saved_cp = (struct scsi_control_page *)
6015		(page_index->page_data + (page_index->page_len *
6016		CTL_PAGE_SAVED));
6017
6018	softc = control_softc;
6019
6020	mtx_lock(&softc->ctl_lock);
6021	if (((current_cp->rlec & SCP_DSENSE) == 0)
6022	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6023		/*
6024		 * Descriptor sense is currently turned off and the user
6025		 * wants to turn it on.
6026		 */
6027		current_cp->rlec |= SCP_DSENSE;
6028		saved_cp->rlec |= SCP_DSENSE;
6029		lun->flags |= CTL_LUN_SENSE_DESC;
6030		set_ua = 1;
6031	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6032		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6033		/*
6034		 * Descriptor sense is currently turned on, and the user
6035		 * wants to turn it off.
6036		 */
6037		current_cp->rlec &= ~SCP_DSENSE;
6038		saved_cp->rlec &= ~SCP_DSENSE;
6039		lun->flags &= ~CTL_LUN_SENSE_DESC;
6040		set_ua = 1;
6041	}
6042	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6043		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6044#ifdef NEEDTOPORT
6045			csevent_log(CSC_CTL | CSC_SHELF_SW |
6046				    CTL_UNTAG_TO_UNTAG,
6047				    csevent_LogType_Trace,
6048				    csevent_Severity_Information,
6049				    csevent_AlertLevel_Green,
6050				    csevent_FRU_Firmware,
6051				    csevent_FRU_Unknown,
6052				    "Received untagged to untagged transition");
6053#endif /* NEEDTOPORT */
6054		} else {
6055#ifdef NEEDTOPORT
6056			csevent_log(CSC_CTL | CSC_SHELF_SW |
6057				    CTL_UNTAG_TO_TAG,
6058				    csevent_LogType_ConfigChange,
6059				    csevent_Severity_Information,
6060				    csevent_AlertLevel_Green,
6061				    csevent_FRU_Firmware,
6062				    csevent_FRU_Unknown,
6063				    "Received untagged to tagged "
6064				    "queueing transition");
6065#endif /* NEEDTOPORT */
6066
6067			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6068			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6069			set_ua = 1;
6070		}
6071	} else {
6072		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6073#ifdef NEEDTOPORT
6074			csevent_log(CSC_CTL | CSC_SHELF_SW |
6075				    CTL_TAG_TO_UNTAG,
6076				    csevent_LogType_ConfigChange,
6077				    csevent_Severity_Warning,
6078				    csevent_AlertLevel_Yellow,
6079				    csevent_FRU_Firmware,
6080				    csevent_FRU_Unknown,
6081				    "Received tagged queueing to untagged "
6082				    "transition");
6083#endif /* NEEDTOPORT */
6084
6085			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6086			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6087			set_ua = 1;
6088		} else {
6089#ifdef NEEDTOPORT
6090			csevent_log(CSC_CTL | CSC_SHELF_SW |
6091				    CTL_TAG_TO_TAG,
6092				    csevent_LogType_Trace,
6093				    csevent_Severity_Information,
6094				    csevent_AlertLevel_Green,
6095				    csevent_FRU_Firmware,
6096				    csevent_FRU_Unknown,
6097				    "Received tagged queueing to tagged "
6098				    "queueing transition");
6099#endif /* NEEDTOPORT */
6100		}
6101	}
6102	if (set_ua != 0) {
6103		int i;
6104		/*
6105		 * Let other initiators know that the mode
6106		 * parameters for this LUN have changed.
6107		 */
6108		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6109			if (i == initidx)
6110				continue;
6111
6112			lun->pending_sense[i].ua_pending |=
6113				CTL_UA_MODE_CHANGE;
6114		}
6115	}
6116	mtx_unlock(&softc->ctl_lock);
6117
6118	return (0);
6119}
6120
6121int
6122ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6123		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6124{
6125	return (0);
6126}
6127
6128int
6129ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6130			   struct ctl_page_index *page_index, int pc)
6131{
6132	struct copan_power_subpage *page;
6133
6134	page = (struct copan_power_subpage *)page_index->page_data +
6135		(page_index->page_len * pc);
6136
6137	switch (pc) {
6138	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6139		/*
6140		 * We don't update the changable bits for this page.
6141		 */
6142		break;
6143	case SMS_PAGE_CTRL_CURRENT >> 6:
6144	case SMS_PAGE_CTRL_DEFAULT >> 6:
6145	case SMS_PAGE_CTRL_SAVED >> 6:
6146#ifdef NEEDTOPORT
6147		ctl_update_power_subpage(page);
6148#endif
6149		break;
6150	default:
6151#ifdef NEEDTOPORT
6152		EPRINT(0, "Invalid PC %d!!", pc);
6153#endif
6154		break;
6155	}
6156	return (0);
6157}
6158
6159
6160int
6161ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6162		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6163{
6164	struct copan_aps_subpage *user_sp;
6165	struct copan_aps_subpage *current_sp;
6166	union ctl_modepage_info *modepage_info;
6167	struct ctl_softc *softc;
6168	struct ctl_lun *lun;
6169	int retval;
6170
6171	retval = CTL_RETVAL_COMPLETE;
6172	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6173		     (page_index->page_len * CTL_PAGE_CURRENT));
6174	softc = control_softc;
6175	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6176
6177	user_sp = (struct copan_aps_subpage *)page_ptr;
6178
6179	modepage_info = (union ctl_modepage_info *)
6180		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6181
6182	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6183	modepage_info->header.subpage = page_index->subpage;
6184	modepage_info->aps.lock_active = user_sp->lock_active;
6185
6186	mtx_lock(&softc->ctl_lock);
6187
6188	/*
6189	 * If there is a request to lock the LUN and another LUN is locked
6190	 * this is an error. If the requested LUN is already locked ignore
6191	 * the request. If no LUN is locked attempt to lock it.
6192	 * if there is a request to unlock the LUN and the LUN is currently
6193	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6194	 * if another LUN is locked or no LUN is locked.
6195	 */
6196	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6197		if (softc->aps_locked_lun == lun->lun) {
6198			/*
6199			 * This LUN is already locked, so we're done.
6200			 */
6201			retval = CTL_RETVAL_COMPLETE;
6202		} else if (softc->aps_locked_lun == 0) {
6203			/*
6204			 * No one has the lock, pass the request to the
6205			 * backend.
6206			 */
6207			retval = lun->backend->config_write(
6208				(union ctl_io *)ctsio);
6209		} else {
6210			/*
6211			 * Someone else has the lock, throw out the request.
6212			 */
6213			ctl_set_already_locked(ctsio);
6214			free(ctsio->kern_data_ptr, M_CTL);
6215			ctl_done((union ctl_io *)ctsio);
6216
6217			/*
6218			 * Set the return value so that ctl_do_mode_select()
6219			 * won't try to complete the command.  We already
6220			 * completed it here.
6221			 */
6222			retval = CTL_RETVAL_ERROR;
6223		}
6224	} else if (softc->aps_locked_lun == lun->lun) {
6225		/*
6226		 * This LUN is locked, so pass the unlock request to the
6227		 * backend.
6228		 */
6229		retval = lun->backend->config_write((union ctl_io *)ctsio);
6230	}
6231	mtx_unlock(&softc->ctl_lock);
6232
6233	return (retval);
6234}
6235
6236int
6237ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6238				struct ctl_page_index *page_index,
6239				uint8_t *page_ptr)
6240{
6241	uint8_t *c;
6242	int i;
6243
6244	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6245	ctl_time_io_secs =
6246		(c[0] << 8) |
6247		(c[1] << 0) |
6248		0;
6249	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6250	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6251	printf("page data:");
6252	for (i=0; i<8; i++)
6253		printf(" %.2x",page_ptr[i]);
6254	printf("\n");
6255	return (0);
6256}
6257
6258int
6259ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6260			       struct ctl_page_index *page_index,
6261			       int pc)
6262{
6263	struct copan_debugconf_subpage *page;
6264
6265	page = (struct copan_debugconf_subpage *)page_index->page_data +
6266		(page_index->page_len * pc);
6267
6268	switch (pc) {
6269	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6270	case SMS_PAGE_CTRL_DEFAULT >> 6:
6271	case SMS_PAGE_CTRL_SAVED >> 6:
6272		/*
6273		 * We don't update the changable or default bits for this page.
6274		 */
6275		break;
6276	case SMS_PAGE_CTRL_CURRENT >> 6:
6277		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6278		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6279		break;
6280	default:
6281#ifdef NEEDTOPORT
6282		EPRINT(0, "Invalid PC %d!!", pc);
6283#endif /* NEEDTOPORT */
6284		break;
6285	}
6286	return (0);
6287}
6288
6289
6290static int
6291ctl_do_mode_select(union ctl_io *io)
6292{
6293	struct scsi_mode_page_header *page_header;
6294	struct ctl_page_index *page_index;
6295	struct ctl_scsiio *ctsio;
6296	int control_dev, page_len;
6297	int page_len_offset, page_len_size;
6298	union ctl_modepage_info *modepage_info;
6299	struct ctl_lun *lun;
6300	int *len_left, *len_used;
6301	int retval, i;
6302
6303	ctsio = &io->scsiio;
6304	page_index = NULL;
6305	page_len = 0;
6306	retval = CTL_RETVAL_COMPLETE;
6307
6308	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6309
6310	if (lun->be_lun->lun_type != T_DIRECT)
6311		control_dev = 1;
6312	else
6313		control_dev = 0;
6314
6315	modepage_info = (union ctl_modepage_info *)
6316		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6317	len_left = &modepage_info->header.len_left;
6318	len_used = &modepage_info->header.len_used;
6319
6320do_next_page:
6321
6322	page_header = (struct scsi_mode_page_header *)
6323		(ctsio->kern_data_ptr + *len_used);
6324
6325	if (*len_left == 0) {
6326		free(ctsio->kern_data_ptr, M_CTL);
6327		ctl_set_success(ctsio);
6328		ctl_done((union ctl_io *)ctsio);
6329		return (CTL_RETVAL_COMPLETE);
6330	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6331
6332		free(ctsio->kern_data_ptr, M_CTL);
6333		ctl_set_param_len_error(ctsio);
6334		ctl_done((union ctl_io *)ctsio);
6335		return (CTL_RETVAL_COMPLETE);
6336
6337	} else if ((page_header->page_code & SMPH_SPF)
6338		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6339
6340		free(ctsio->kern_data_ptr, M_CTL);
6341		ctl_set_param_len_error(ctsio);
6342		ctl_done((union ctl_io *)ctsio);
6343		return (CTL_RETVAL_COMPLETE);
6344	}
6345
6346
6347	/*
6348	 * XXX KDM should we do something with the block descriptor?
6349	 */
6350	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6351
6352		if ((control_dev != 0)
6353		 && (lun->mode_pages.index[i].page_flags &
6354		     CTL_PAGE_FLAG_DISK_ONLY))
6355			continue;
6356
6357		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6358		    (page_header->page_code & SMPH_PC_MASK))
6359			continue;
6360
6361		/*
6362		 * If neither page has a subpage code, then we've got a
6363		 * match.
6364		 */
6365		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6366		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6367			page_index = &lun->mode_pages.index[i];
6368			page_len = page_header->page_length;
6369			break;
6370		}
6371
6372		/*
6373		 * If both pages have subpages, then the subpage numbers
6374		 * have to match.
6375		 */
6376		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6377		  && (page_header->page_code & SMPH_SPF)) {
6378			struct scsi_mode_page_header_sp *sph;
6379
6380			sph = (struct scsi_mode_page_header_sp *)page_header;
6381
6382			if (lun->mode_pages.index[i].subpage ==
6383			    sph->subpage) {
6384				page_index = &lun->mode_pages.index[i];
6385				page_len = scsi_2btoul(sph->page_length);
6386				break;
6387			}
6388		}
6389	}
6390
6391	/*
6392	 * If we couldn't find the page, or if we don't have a mode select
6393	 * handler for it, send back an error to the user.
6394	 */
6395	if ((page_index == NULL)
6396	 || (page_index->select_handler == NULL)) {
6397		ctl_set_invalid_field(ctsio,
6398				      /*sks_valid*/ 1,
6399				      /*command*/ 0,
6400				      /*field*/ *len_used,
6401				      /*bit_valid*/ 0,
6402				      /*bit*/ 0);
6403		free(ctsio->kern_data_ptr, M_CTL);
6404		ctl_done((union ctl_io *)ctsio);
6405		return (CTL_RETVAL_COMPLETE);
6406	}
6407
6408	if (page_index->page_code & SMPH_SPF) {
6409		page_len_offset = 2;
6410		page_len_size = 2;
6411	} else {
6412		page_len_size = 1;
6413		page_len_offset = 1;
6414	}
6415
6416	/*
6417	 * If the length the initiator gives us isn't the one we specify in
6418	 * the mode page header, or if they didn't specify enough data in
6419	 * the CDB to avoid truncating this page, kick out the request.
6420	 */
6421	if ((page_len != (page_index->page_len - page_len_offset -
6422			  page_len_size))
6423	 || (*len_left < page_index->page_len)) {
6424
6425
6426		ctl_set_invalid_field(ctsio,
6427				      /*sks_valid*/ 1,
6428				      /*command*/ 0,
6429				      /*field*/ *len_used + page_len_offset,
6430				      /*bit_valid*/ 0,
6431				      /*bit*/ 0);
6432		free(ctsio->kern_data_ptr, M_CTL);
6433		ctl_done((union ctl_io *)ctsio);
6434		return (CTL_RETVAL_COMPLETE);
6435	}
6436
6437	/*
6438	 * Run through the mode page, checking to make sure that the bits
6439	 * the user changed are actually legal for him to change.
6440	 */
6441	for (i = 0; i < page_index->page_len; i++) {
6442		uint8_t *user_byte, *change_mask, *current_byte;
6443		int bad_bit;
6444		int j;
6445
6446		user_byte = (uint8_t *)page_header + i;
6447		change_mask = page_index->page_data +
6448			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6449		current_byte = page_index->page_data +
6450			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6451
6452		/*
6453		 * Check to see whether the user set any bits in this byte
6454		 * that he is not allowed to set.
6455		 */
6456		if ((*user_byte & ~(*change_mask)) ==
6457		    (*current_byte & ~(*change_mask)))
6458			continue;
6459
6460		/*
6461		 * Go through bit by bit to determine which one is illegal.
6462		 */
6463		bad_bit = 0;
6464		for (j = 7; j >= 0; j--) {
6465			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6466			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6467				bad_bit = i;
6468				break;
6469			}
6470		}
6471		ctl_set_invalid_field(ctsio,
6472				      /*sks_valid*/ 1,
6473				      /*command*/ 0,
6474				      /*field*/ *len_used + i,
6475				      /*bit_valid*/ 1,
6476				      /*bit*/ bad_bit);
6477		free(ctsio->kern_data_ptr, M_CTL);
6478		ctl_done((union ctl_io *)ctsio);
6479		return (CTL_RETVAL_COMPLETE);
6480	}
6481
6482	/*
6483	 * Decrement these before we call the page handler, since we may
6484	 * end up getting called back one way or another before the handler
6485	 * returns to this context.
6486	 */
6487	*len_left -= page_index->page_len;
6488	*len_used += page_index->page_len;
6489
6490	retval = page_index->select_handler(ctsio, page_index,
6491					    (uint8_t *)page_header);
6492
6493	/*
6494	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6495	 * wait until this queued command completes to finish processing
6496	 * the mode page.  If it returns anything other than
6497	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6498	 * already set the sense information, freed the data pointer, and
6499	 * completed the io for us.
6500	 */
6501	if (retval != CTL_RETVAL_COMPLETE)
6502		goto bailout_no_done;
6503
6504	/*
6505	 * If the initiator sent us more than one page, parse the next one.
6506	 */
6507	if (*len_left > 0)
6508		goto do_next_page;
6509
6510	ctl_set_success(ctsio);
6511	free(ctsio->kern_data_ptr, M_CTL);
6512	ctl_done((union ctl_io *)ctsio);
6513
6514bailout_no_done:
6515
6516	return (CTL_RETVAL_COMPLETE);
6517
6518}
6519
6520int
6521ctl_mode_select(struct ctl_scsiio *ctsio)
6522{
6523	int param_len, pf, sp;
6524	int header_size, bd_len;
6525	int len_left, len_used;
6526	struct ctl_page_index *page_index;
6527	struct ctl_lun *lun;
6528	int control_dev, page_len;
6529	union ctl_modepage_info *modepage_info;
6530	int retval;
6531
6532	pf = 0;
6533	sp = 0;
6534	page_len = 0;
6535	len_used = 0;
6536	len_left = 0;
6537	retval = 0;
6538	bd_len = 0;
6539	page_index = NULL;
6540
6541	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6542
6543	if (lun->be_lun->lun_type != T_DIRECT)
6544		control_dev = 1;
6545	else
6546		control_dev = 0;
6547
6548	switch (ctsio->cdb[0]) {
6549	case MODE_SELECT_6: {
6550		struct scsi_mode_select_6 *cdb;
6551
6552		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6553
6554		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6555		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6556
6557		param_len = cdb->length;
6558		header_size = sizeof(struct scsi_mode_header_6);
6559		break;
6560	}
6561	case MODE_SELECT_10: {
6562		struct scsi_mode_select_10 *cdb;
6563
6564		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6565
6566		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6567		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6568
6569		param_len = scsi_2btoul(cdb->length);
6570		header_size = sizeof(struct scsi_mode_header_10);
6571		break;
6572	}
6573	default:
6574		ctl_set_invalid_opcode(ctsio);
6575		ctl_done((union ctl_io *)ctsio);
6576		return (CTL_RETVAL_COMPLETE);
6577		break; /* NOTREACHED */
6578	}
6579
6580	/*
6581	 * From SPC-3:
6582	 * "A parameter list length of zero indicates that the Data-Out Buffer
6583	 * shall be empty. This condition shall not be considered as an error."
6584	 */
6585	if (param_len == 0) {
6586		ctl_set_success(ctsio);
6587		ctl_done((union ctl_io *)ctsio);
6588		return (CTL_RETVAL_COMPLETE);
6589	}
6590
6591	/*
6592	 * Since we'll hit this the first time through, prior to
6593	 * allocation, we don't need to free a data buffer here.
6594	 */
6595	if (param_len < header_size) {
6596		ctl_set_param_len_error(ctsio);
6597		ctl_done((union ctl_io *)ctsio);
6598		return (CTL_RETVAL_COMPLETE);
6599	}
6600
6601	/*
6602	 * Allocate the data buffer and grab the user's data.  In theory,
6603	 * we shouldn't have to sanity check the parameter list length here
6604	 * because the maximum size is 64K.  We should be able to malloc
6605	 * that much without too many problems.
6606	 */
6607	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6608		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6609		ctsio->kern_data_len = param_len;
6610		ctsio->kern_total_len = param_len;
6611		ctsio->kern_data_resid = 0;
6612		ctsio->kern_rel_offset = 0;
6613		ctsio->kern_sg_entries = 0;
6614		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6615		ctsio->be_move_done = ctl_config_move_done;
6616		ctl_datamove((union ctl_io *)ctsio);
6617
6618		return (CTL_RETVAL_COMPLETE);
6619	}
6620
6621	switch (ctsio->cdb[0]) {
6622	case MODE_SELECT_6: {
6623		struct scsi_mode_header_6 *mh6;
6624
6625		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6626		bd_len = mh6->blk_desc_len;
6627		break;
6628	}
6629	case MODE_SELECT_10: {
6630		struct scsi_mode_header_10 *mh10;
6631
6632		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6633		bd_len = scsi_2btoul(mh10->blk_desc_len);
6634		break;
6635	}
6636	default:
6637		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6638		break;
6639	}
6640
6641	if (param_len < (header_size + bd_len)) {
6642		free(ctsio->kern_data_ptr, M_CTL);
6643		ctl_set_param_len_error(ctsio);
6644		ctl_done((union ctl_io *)ctsio);
6645		return (CTL_RETVAL_COMPLETE);
6646	}
6647
6648	/*
6649	 * Set the IO_CONT flag, so that if this I/O gets passed to
6650	 * ctl_config_write_done(), it'll get passed back to
6651	 * ctl_do_mode_select() for further processing, or completion if
6652	 * we're all done.
6653	 */
6654	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6655	ctsio->io_cont = ctl_do_mode_select;
6656
6657	modepage_info = (union ctl_modepage_info *)
6658		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6659
6660	memset(modepage_info, 0, sizeof(*modepage_info));
6661
6662	len_left = param_len - header_size - bd_len;
6663	len_used = header_size + bd_len;
6664
6665	modepage_info->header.len_left = len_left;
6666	modepage_info->header.len_used = len_used;
6667
6668	return (ctl_do_mode_select((union ctl_io *)ctsio));
6669}
6670
6671int
6672ctl_mode_sense(struct ctl_scsiio *ctsio)
6673{
6674	struct ctl_lun *lun;
6675	int pc, page_code, dbd, llba, subpage;
6676	int alloc_len, page_len, header_len, total_len;
6677	struct scsi_mode_block_descr *block_desc;
6678	struct ctl_page_index *page_index;
6679	int control_dev;
6680
6681	dbd = 0;
6682	llba = 0;
6683	block_desc = NULL;
6684	page_index = NULL;
6685
6686	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6687
6688	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6689
6690	if (lun->be_lun->lun_type != T_DIRECT)
6691		control_dev = 1;
6692	else
6693		control_dev = 0;
6694
6695	switch (ctsio->cdb[0]) {
6696	case MODE_SENSE_6: {
6697		struct scsi_mode_sense_6 *cdb;
6698
6699		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6700
6701		header_len = sizeof(struct scsi_mode_hdr_6);
6702		if (cdb->byte2 & SMS_DBD)
6703			dbd = 1;
6704		else
6705			header_len += sizeof(struct scsi_mode_block_descr);
6706
6707		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6708		page_code = cdb->page & SMS_PAGE_CODE;
6709		subpage = cdb->subpage;
6710		alloc_len = cdb->length;
6711		break;
6712	}
6713	case MODE_SENSE_10: {
6714		struct scsi_mode_sense_10 *cdb;
6715
6716		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6717
6718		header_len = sizeof(struct scsi_mode_hdr_10);
6719
6720		if (cdb->byte2 & SMS_DBD)
6721			dbd = 1;
6722		else
6723			header_len += sizeof(struct scsi_mode_block_descr);
6724		if (cdb->byte2 & SMS10_LLBAA)
6725			llba = 1;
6726		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6727		page_code = cdb->page & SMS_PAGE_CODE;
6728		subpage = cdb->subpage;
6729		alloc_len = scsi_2btoul(cdb->length);
6730		break;
6731	}
6732	default:
6733		ctl_set_invalid_opcode(ctsio);
6734		ctl_done((union ctl_io *)ctsio);
6735		return (CTL_RETVAL_COMPLETE);
6736		break; /* NOTREACHED */
6737	}
6738
6739	/*
6740	 * We have to make a first pass through to calculate the size of
6741	 * the pages that match the user's query.  Then we allocate enough
6742	 * memory to hold it, and actually copy the data into the buffer.
6743	 */
6744	switch (page_code) {
6745	case SMS_ALL_PAGES_PAGE: {
6746		int i;
6747
6748		page_len = 0;
6749
6750		/*
6751		 * At the moment, values other than 0 and 0xff here are
6752		 * reserved according to SPC-3.
6753		 */
6754		if ((subpage != SMS_SUBPAGE_PAGE_0)
6755		 && (subpage != SMS_SUBPAGE_ALL)) {
6756			ctl_set_invalid_field(ctsio,
6757					      /*sks_valid*/ 1,
6758					      /*command*/ 1,
6759					      /*field*/ 3,
6760					      /*bit_valid*/ 0,
6761					      /*bit*/ 0);
6762			ctl_done((union ctl_io *)ctsio);
6763			return (CTL_RETVAL_COMPLETE);
6764		}
6765
6766		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6767			if ((control_dev != 0)
6768			 && (lun->mode_pages.index[i].page_flags &
6769			     CTL_PAGE_FLAG_DISK_ONLY))
6770				continue;
6771
6772			/*
6773			 * We don't use this subpage if the user didn't
6774			 * request all subpages.
6775			 */
6776			if ((lun->mode_pages.index[i].subpage != 0)
6777			 && (subpage == SMS_SUBPAGE_PAGE_0))
6778				continue;
6779
6780#if 0
6781			printf("found page %#x len %d\n",
6782			       lun->mode_pages.index[i].page_code &
6783			       SMPH_PC_MASK,
6784			       lun->mode_pages.index[i].page_len);
6785#endif
6786			page_len += lun->mode_pages.index[i].page_len;
6787		}
6788		break;
6789	}
6790	default: {
6791		int i;
6792
6793		page_len = 0;
6794
6795		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6796			/* Look for the right page code */
6797			if ((lun->mode_pages.index[i].page_code &
6798			     SMPH_PC_MASK) != page_code)
6799				continue;
6800
6801			/* Look for the right subpage or the subpage wildcard*/
6802			if ((lun->mode_pages.index[i].subpage != subpage)
6803			 && (subpage != SMS_SUBPAGE_ALL))
6804				continue;
6805
6806			/* Make sure the page is supported for this dev type */
6807			if ((control_dev != 0)
6808			 && (lun->mode_pages.index[i].page_flags &
6809			     CTL_PAGE_FLAG_DISK_ONLY))
6810				continue;
6811
6812#if 0
6813			printf("found page %#x len %d\n",
6814			       lun->mode_pages.index[i].page_code &
6815			       SMPH_PC_MASK,
6816			       lun->mode_pages.index[i].page_len);
6817#endif
6818
6819			page_len += lun->mode_pages.index[i].page_len;
6820		}
6821
6822		if (page_len == 0) {
6823			ctl_set_invalid_field(ctsio,
6824					      /*sks_valid*/ 1,
6825					      /*command*/ 1,
6826					      /*field*/ 2,
6827					      /*bit_valid*/ 1,
6828					      /*bit*/ 5);
6829			ctl_done((union ctl_io *)ctsio);
6830			return (CTL_RETVAL_COMPLETE);
6831		}
6832		break;
6833	}
6834	}
6835
6836	total_len = header_len + page_len;
6837#if 0
6838	printf("header_len = %d, page_len = %d, total_len = %d\n",
6839	       header_len, page_len, total_len);
6840#endif
6841
6842	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6843	ctsio->kern_sg_entries = 0;
6844	ctsio->kern_data_resid = 0;
6845	ctsio->kern_rel_offset = 0;
6846	if (total_len < alloc_len) {
6847		ctsio->residual = alloc_len - total_len;
6848		ctsio->kern_data_len = total_len;
6849		ctsio->kern_total_len = total_len;
6850	} else {
6851		ctsio->residual = 0;
6852		ctsio->kern_data_len = alloc_len;
6853		ctsio->kern_total_len = alloc_len;
6854	}
6855
6856	switch (ctsio->cdb[0]) {
6857	case MODE_SENSE_6: {
6858		struct scsi_mode_hdr_6 *header;
6859
6860		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6861
6862		header->datalen = ctl_min(total_len - 1, 254);
6863
6864		if (dbd)
6865			header->block_descr_len = 0;
6866		else
6867			header->block_descr_len =
6868				sizeof(struct scsi_mode_block_descr);
6869		block_desc = (struct scsi_mode_block_descr *)&header[1];
6870		break;
6871	}
6872	case MODE_SENSE_10: {
6873		struct scsi_mode_hdr_10 *header;
6874		int datalen;
6875
6876		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6877
6878		datalen = ctl_min(total_len - 2, 65533);
6879		scsi_ulto2b(datalen, header->datalen);
6880		if (dbd)
6881			scsi_ulto2b(0, header->block_descr_len);
6882		else
6883			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6884				    header->block_descr_len);
6885		block_desc = (struct scsi_mode_block_descr *)&header[1];
6886		break;
6887	}
6888	default:
6889		panic("invalid CDB type %#x", ctsio->cdb[0]);
6890		break; /* NOTREACHED */
6891	}
6892
6893	/*
6894	 * If we've got a disk, use its blocksize in the block
6895	 * descriptor.  Otherwise, just set it to 0.
6896	 */
6897	if (dbd == 0) {
6898		if (control_dev != 0)
6899			scsi_ulto3b(lun->be_lun->blocksize,
6900				    block_desc->block_len);
6901		else
6902			scsi_ulto3b(0, block_desc->block_len);
6903	}
6904
6905	switch (page_code) {
6906	case SMS_ALL_PAGES_PAGE: {
6907		int i, data_used;
6908
6909		data_used = header_len;
6910		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6911			struct ctl_page_index *page_index;
6912
6913			page_index = &lun->mode_pages.index[i];
6914
6915			if ((control_dev != 0)
6916			 && (page_index->page_flags &
6917			    CTL_PAGE_FLAG_DISK_ONLY))
6918				continue;
6919
6920			/*
6921			 * We don't use this subpage if the user didn't
6922			 * request all subpages.  We already checked (above)
6923			 * to make sure the user only specified a subpage
6924			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6925			 */
6926			if ((page_index->subpage != 0)
6927			 && (subpage == SMS_SUBPAGE_PAGE_0))
6928				continue;
6929
6930			/*
6931			 * Call the handler, if it exists, to update the
6932			 * page to the latest values.
6933			 */
6934			if (page_index->sense_handler != NULL)
6935				page_index->sense_handler(ctsio, page_index,pc);
6936
6937			memcpy(ctsio->kern_data_ptr + data_used,
6938			       page_index->page_data +
6939			       (page_index->page_len * pc),
6940			       page_index->page_len);
6941			data_used += page_index->page_len;
6942		}
6943		break;
6944	}
6945	default: {
6946		int i, data_used;
6947
6948		data_used = header_len;
6949
6950		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6951			struct ctl_page_index *page_index;
6952
6953			page_index = &lun->mode_pages.index[i];
6954
6955			/* Look for the right page code */
6956			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6957				continue;
6958
6959			/* Look for the right subpage or the subpage wildcard*/
6960			if ((page_index->subpage != subpage)
6961			 && (subpage != SMS_SUBPAGE_ALL))
6962				continue;
6963
6964			/* Make sure the page is supported for this dev type */
6965			if ((control_dev != 0)
6966			 && (page_index->page_flags &
6967			     CTL_PAGE_FLAG_DISK_ONLY))
6968				continue;
6969
6970			/*
6971			 * Call the handler, if it exists, to update the
6972			 * page to the latest values.
6973			 */
6974			if (page_index->sense_handler != NULL)
6975				page_index->sense_handler(ctsio, page_index,pc);
6976
6977			memcpy(ctsio->kern_data_ptr + data_used,
6978			       page_index->page_data +
6979			       (page_index->page_len * pc),
6980			       page_index->page_len);
6981			data_used += page_index->page_len;
6982		}
6983		break;
6984	}
6985	}
6986
6987	ctsio->scsi_status = SCSI_STATUS_OK;
6988
6989	ctsio->be_move_done = ctl_config_move_done;
6990	ctl_datamove((union ctl_io *)ctsio);
6991
6992	return (CTL_RETVAL_COMPLETE);
6993}
6994
6995int
6996ctl_read_capacity(struct ctl_scsiio *ctsio)
6997{
6998	struct scsi_read_capacity *cdb;
6999	struct scsi_read_capacity_data *data;
7000	struct ctl_lun *lun;
7001	uint32_t lba;
7002
7003	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7004
7005	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7006
7007	lba = scsi_4btoul(cdb->addr);
7008	if (((cdb->pmi & SRC_PMI) == 0)
7009	 && (lba != 0)) {
7010		ctl_set_invalid_field(/*ctsio*/ ctsio,
7011				      /*sks_valid*/ 1,
7012				      /*command*/ 1,
7013				      /*field*/ 2,
7014				      /*bit_valid*/ 0,
7015				      /*bit*/ 0);
7016		ctl_done((union ctl_io *)ctsio);
7017		return (CTL_RETVAL_COMPLETE);
7018	}
7019
7020	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7021
7022	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7023	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7024	ctsio->residual = 0;
7025	ctsio->kern_data_len = sizeof(*data);
7026	ctsio->kern_total_len = sizeof(*data);
7027	ctsio->kern_data_resid = 0;
7028	ctsio->kern_rel_offset = 0;
7029	ctsio->kern_sg_entries = 0;
7030
7031	/*
7032	 * If the maximum LBA is greater than 0xfffffffe, the user must
7033	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7034	 * serivce action set.
7035	 */
7036	if (lun->be_lun->maxlba > 0xfffffffe)
7037		scsi_ulto4b(0xffffffff, data->addr);
7038	else
7039		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7040
7041	/*
7042	 * XXX KDM this may not be 512 bytes...
7043	 */
7044	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7045
7046	ctsio->scsi_status = SCSI_STATUS_OK;
7047
7048	ctsio->be_move_done = ctl_config_move_done;
7049	ctl_datamove((union ctl_io *)ctsio);
7050
7051	return (CTL_RETVAL_COMPLETE);
7052}
7053
7054static int
7055ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7056{
7057	struct scsi_read_capacity_16 *cdb;
7058	struct scsi_read_capacity_data_long *data;
7059	struct ctl_lun *lun;
7060	uint64_t lba;
7061	uint32_t alloc_len;
7062
7063	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7064
7065	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7066
7067	alloc_len = scsi_4btoul(cdb->alloc_len);
7068	lba = scsi_8btou64(cdb->addr);
7069
7070	if ((cdb->reladr & SRC16_PMI)
7071	 && (lba != 0)) {
7072		ctl_set_invalid_field(/*ctsio*/ ctsio,
7073				      /*sks_valid*/ 1,
7074				      /*command*/ 1,
7075				      /*field*/ 2,
7076				      /*bit_valid*/ 0,
7077				      /*bit*/ 0);
7078		ctl_done((union ctl_io *)ctsio);
7079		return (CTL_RETVAL_COMPLETE);
7080	}
7081
7082	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7083
7084	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7085	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7086
7087	if (sizeof(*data) < alloc_len) {
7088		ctsio->residual = alloc_len - sizeof(*data);
7089		ctsio->kern_data_len = sizeof(*data);
7090		ctsio->kern_total_len = sizeof(*data);
7091	} else {
7092		ctsio->residual = 0;
7093		ctsio->kern_data_len = alloc_len;
7094		ctsio->kern_total_len = alloc_len;
7095	}
7096	ctsio->kern_data_resid = 0;
7097	ctsio->kern_rel_offset = 0;
7098	ctsio->kern_sg_entries = 0;
7099
7100	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7101	/* XXX KDM this may not be 512 bytes... */
7102	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7103	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7104	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7105	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7106		data->lalba_lbp[0] |= SRC16_LBPME;
7107
7108	ctsio->scsi_status = SCSI_STATUS_OK;
7109
7110	ctsio->be_move_done = ctl_config_move_done;
7111	ctl_datamove((union ctl_io *)ctsio);
7112
7113	return (CTL_RETVAL_COMPLETE);
7114}
7115
7116int
7117ctl_service_action_in(struct ctl_scsiio *ctsio)
7118{
7119	struct scsi_service_action_in *cdb;
7120	int retval;
7121
7122	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
7123
7124	cdb = (struct scsi_service_action_in *)ctsio->cdb;
7125
7126	retval = CTL_RETVAL_COMPLETE;
7127
7128	switch (cdb->service_action) {
7129	case SRC16_SERVICE_ACTION:
7130		retval = ctl_read_capacity_16(ctsio);
7131		break;
7132	default:
7133		ctl_set_invalid_field(/*ctsio*/ ctsio,
7134				      /*sks_valid*/ 1,
7135				      /*command*/ 1,
7136				      /*field*/ 1,
7137				      /*bit_valid*/ 1,
7138				      /*bit*/ 4);
7139		ctl_done((union ctl_io *)ctsio);
7140		break;
7141	}
7142
7143	return (retval);
7144}
7145
7146int
7147ctl_maintenance_in(struct ctl_scsiio *ctsio)
7148{
7149	struct scsi_maintenance_in *cdb;
7150	int retval;
7151	int alloc_len, total_len = 0;
7152	int num_target_port_groups, single;
7153	struct ctl_lun *lun;
7154	struct ctl_softc *softc;
7155	struct scsi_target_group_data *rtg_ptr;
7156	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
7157	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
7158	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
7159
7160	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
7161
7162	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7163	softc = control_softc;
7164	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7165
7166	retval = CTL_RETVAL_COMPLETE;
7167
7168	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
7169		ctl_set_invalid_field(/*ctsio*/ ctsio,
7170				      /*sks_valid*/ 1,
7171				      /*command*/ 1,
7172				      /*field*/ 1,
7173				      /*bit_valid*/ 1,
7174				      /*bit*/ 4);
7175		ctl_done((union ctl_io *)ctsio);
7176		return(retval);
7177	}
7178
7179	mtx_lock(&softc->ctl_lock);
7180	single = ctl_is_single;
7181	mtx_unlock(&softc->ctl_lock);
7182
7183	if (single)
7184        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
7185	else
7186        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7187
7188	total_len = sizeof(struct scsi_target_group_data) +
7189		sizeof(struct scsi_target_port_group_descriptor) *
7190		num_target_port_groups +
7191		sizeof(struct scsi_target_port_descriptor) *
7192		NUM_PORTS_PER_GRP * num_target_port_groups;
7193
7194	alloc_len = scsi_4btoul(cdb->length);
7195
7196	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7197
7198	ctsio->kern_sg_entries = 0;
7199
7200	if (total_len < alloc_len) {
7201		ctsio->residual = alloc_len - total_len;
7202		ctsio->kern_data_len = total_len;
7203		ctsio->kern_total_len = total_len;
7204	} else {
7205		ctsio->residual = 0;
7206		ctsio->kern_data_len = alloc_len;
7207		ctsio->kern_total_len = alloc_len;
7208	}
7209	ctsio->kern_data_resid = 0;
7210	ctsio->kern_rel_offset = 0;
7211
7212	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7213
7214	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7215	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7216	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7217	        &tp_desc_ptr1_1->desc_list[0];
7218
7219	if (single == 0) {
7220		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7221	                &tp_desc_ptr1_2->desc_list[0];
7222		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7223		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7224	        	&tp_desc_ptr2_1->desc_list[0];
7225        } else {
7226		tpg_desc_ptr2 = NULL;
7227		tp_desc_ptr2_1 = NULL;
7228		tp_desc_ptr2_2 = NULL;
7229	}
7230
7231	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7232	if (single == 0) {
7233        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7234			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7235				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7236				tpg_desc_ptr2->pref_state =
7237					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7238			} else {
7239				tpg_desc_ptr1->pref_state =
7240					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7241				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7242			}
7243		} else {
7244			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7245				tpg_desc_ptr1->pref_state =
7246					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7247				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7248			} else {
7249				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7250				tpg_desc_ptr2->pref_state =
7251					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7252			}
7253		}
7254	} else {
7255		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7256	}
7257	tpg_desc_ptr1->support = 0;
7258	tpg_desc_ptr1->target_port_group[1] = 1;
7259	tpg_desc_ptr1->status = TPG_IMPLICIT;
7260	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7261
7262	if (single == 0) {
7263		tpg_desc_ptr2->support = 0;
7264		tpg_desc_ptr2->target_port_group[1] = 2;
7265		tpg_desc_ptr2->status = TPG_IMPLICIT;
7266		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7267
7268		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7269		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7270
7271		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7272		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7273	} else {
7274        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7275			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7276			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7277		} else {
7278			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7279			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7280		}
7281	}
7282
7283	ctsio->be_move_done = ctl_config_move_done;
7284
7285	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7286			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7287			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7288			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7289			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7290
7291	ctl_datamove((union ctl_io *)ctsio);
7292	return(retval);
7293}
7294
7295int
7296ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7297{
7298	struct scsi_per_res_in *cdb;
7299	int alloc_len, total_len = 0;
7300	/* struct scsi_per_res_in_rsrv in_data; */
7301	struct ctl_lun *lun;
7302	struct ctl_softc *softc;
7303
7304	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7305
7306	softc = control_softc;
7307
7308	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7309
7310	alloc_len = scsi_2btoul(cdb->length);
7311
7312	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7313
7314retry:
7315	mtx_lock(&softc->ctl_lock);
7316	switch (cdb->action) {
7317	case SPRI_RK: /* read keys */
7318		total_len = sizeof(struct scsi_per_res_in_keys) +
7319			lun->pr_key_count *
7320			sizeof(struct scsi_per_res_key);
7321		break;
7322	case SPRI_RR: /* read reservation */
7323		if (lun->flags & CTL_LUN_PR_RESERVED)
7324			total_len = sizeof(struct scsi_per_res_in_rsrv);
7325		else
7326			total_len = sizeof(struct scsi_per_res_in_header);
7327		break;
7328	case SPRI_RC: /* report capabilities */
7329		total_len = sizeof(struct scsi_per_res_cap);
7330		break;
7331	case SPRI_RS: /* read full status */
7332	default:
7333		mtx_unlock(&softc->ctl_lock);
7334		ctl_set_invalid_field(ctsio,
7335				      /*sks_valid*/ 1,
7336				      /*command*/ 1,
7337				      /*field*/ 1,
7338				      /*bit_valid*/ 1,
7339				      /*bit*/ 0);
7340		ctl_done((union ctl_io *)ctsio);
7341		return (CTL_RETVAL_COMPLETE);
7342		break; /* NOTREACHED */
7343	}
7344	mtx_unlock(&softc->ctl_lock);
7345
7346	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7347
7348	if (total_len < alloc_len) {
7349		ctsio->residual = alloc_len - total_len;
7350		ctsio->kern_data_len = total_len;
7351		ctsio->kern_total_len = total_len;
7352	} else {
7353		ctsio->residual = 0;
7354		ctsio->kern_data_len = alloc_len;
7355		ctsio->kern_total_len = alloc_len;
7356	}
7357
7358	ctsio->kern_data_resid = 0;
7359	ctsio->kern_rel_offset = 0;
7360	ctsio->kern_sg_entries = 0;
7361
7362	mtx_lock(&softc->ctl_lock);
7363	switch (cdb->action) {
7364	case SPRI_RK: { // read keys
7365        struct scsi_per_res_in_keys *res_keys;
7366		int i, key_count;
7367
7368		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7369
7370		/*
7371		 * We had to drop the lock to allocate our buffer, which
7372		 * leaves time for someone to come in with another
7373		 * persistent reservation.  (That is unlikely, though,
7374		 * since this should be the only persistent reservation
7375		 * command active right now.)
7376		 */
7377		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7378		    (lun->pr_key_count *
7379		     sizeof(struct scsi_per_res_key)))){
7380			mtx_unlock(&softc->ctl_lock);
7381			free(ctsio->kern_data_ptr, M_CTL);
7382			printf("%s: reservation length changed, retrying\n",
7383			       __func__);
7384			goto retry;
7385		}
7386
7387		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7388
7389		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7390			     lun->pr_key_count, res_keys->header.length);
7391
7392		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7393			if (!lun->per_res[i].registered)
7394				continue;
7395
7396			/*
7397			 * We used lun->pr_key_count to calculate the
7398			 * size to allocate.  If it turns out the number of
7399			 * initiators with the registered flag set is
7400			 * larger than that (i.e. they haven't been kept in
7401			 * sync), we've got a problem.
7402			 */
7403			if (key_count >= lun->pr_key_count) {
7404#ifdef NEEDTOPORT
7405				csevent_log(CSC_CTL | CSC_SHELF_SW |
7406					    CTL_PR_ERROR,
7407					    csevent_LogType_Fault,
7408					    csevent_AlertLevel_Yellow,
7409					    csevent_FRU_ShelfController,
7410					    csevent_FRU_Firmware,
7411				        csevent_FRU_Unknown,
7412					    "registered keys %d >= key "
7413					    "count %d", key_count,
7414					    lun->pr_key_count);
7415#endif
7416				key_count++;
7417				continue;
7418			}
7419			memcpy(res_keys->keys[key_count].key,
7420			       lun->per_res[i].res_key.key,
7421			       ctl_min(sizeof(res_keys->keys[key_count].key),
7422			       sizeof(lun->per_res[i].res_key)));
7423			key_count++;
7424		}
7425		break;
7426	}
7427	case SPRI_RR: { // read reservation
7428		struct scsi_per_res_in_rsrv *res;
7429		int tmp_len, header_only;
7430
7431		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7432
7433		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7434
7435		if (lun->flags & CTL_LUN_PR_RESERVED)
7436		{
7437			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7438			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7439				    res->header.length);
7440			header_only = 0;
7441		} else {
7442			tmp_len = sizeof(struct scsi_per_res_in_header);
7443			scsi_ulto4b(0, res->header.length);
7444			header_only = 1;
7445		}
7446
7447		/*
7448		 * We had to drop the lock to allocate our buffer, which
7449		 * leaves time for someone to come in with another
7450		 * persistent reservation.  (That is unlikely, though,
7451		 * since this should be the only persistent reservation
7452		 * command active right now.)
7453		 */
7454		if (tmp_len != total_len) {
7455			mtx_unlock(&softc->ctl_lock);
7456			free(ctsio->kern_data_ptr, M_CTL);
7457			printf("%s: reservation status changed, retrying\n",
7458			       __func__);
7459			goto retry;
7460		}
7461
7462		/*
7463		 * No reservation held, so we're done.
7464		 */
7465		if (header_only != 0)
7466			break;
7467
7468		/*
7469		 * If the registration is an All Registrants type, the key
7470		 * is 0, since it doesn't really matter.
7471		 */
7472		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7473			memcpy(res->data.reservation,
7474			       &lun->per_res[lun->pr_res_idx].res_key,
7475			       sizeof(struct scsi_per_res_key));
7476		}
7477		res->data.scopetype = lun->res_type;
7478		break;
7479	}
7480	case SPRI_RC:     //report capabilities
7481	{
7482		struct scsi_per_res_cap *res_cap;
7483		uint16_t type_mask;
7484
7485		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7486		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7487		res_cap->flags2 |= SPRI_TMV;
7488		type_mask = SPRI_TM_WR_EX_AR |
7489			    SPRI_TM_EX_AC_RO |
7490			    SPRI_TM_WR_EX_RO |
7491			    SPRI_TM_EX_AC |
7492			    SPRI_TM_WR_EX |
7493			    SPRI_TM_EX_AC_AR;
7494		scsi_ulto2b(type_mask, res_cap->type_mask);
7495		break;
7496	}
7497	case SPRI_RS: //read full status
7498	default:
7499		/*
7500		 * This is a bug, because we just checked for this above,
7501		 * and should have returned an error.
7502		 */
7503		panic("Invalid PR type %x", cdb->action);
7504		break; /* NOTREACHED */
7505	}
7506	mtx_unlock(&softc->ctl_lock);
7507
7508	ctsio->be_move_done = ctl_config_move_done;
7509
7510	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7511			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7512			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7513			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7514			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7515
7516	ctl_datamove((union ctl_io *)ctsio);
7517
7518	return (CTL_RETVAL_COMPLETE);
7519}
7520
7521/*
7522 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7523 * it should return.
7524 */
7525static int
7526ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7527		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7528		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7529		struct scsi_per_res_out_parms* param)
7530{
7531	union ctl_ha_msg persis_io;
7532	int retval, i;
7533	int isc_retval;
7534
7535	retval = 0;
7536
7537	if (sa_res_key == 0) {
7538		mtx_lock(&softc->ctl_lock);
7539		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7540			/* validate scope and type */
7541			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7542			     SPR_LU_SCOPE) {
7543				mtx_unlock(&softc->ctl_lock);
7544				ctl_set_invalid_field(/*ctsio*/ ctsio,
7545						      /*sks_valid*/ 1,
7546						      /*command*/ 1,
7547						      /*field*/ 2,
7548						      /*bit_valid*/ 1,
7549						      /*bit*/ 4);
7550				ctl_done((union ctl_io *)ctsio);
7551				return (1);
7552			}
7553
7554		        if (type>8 || type==2 || type==4 || type==0) {
7555				mtx_unlock(&softc->ctl_lock);
7556				ctl_set_invalid_field(/*ctsio*/ ctsio,
7557       	           				      /*sks_valid*/ 1,
7558						      /*command*/ 1,
7559						      /*field*/ 2,
7560						      /*bit_valid*/ 1,
7561						      /*bit*/ 0);
7562				ctl_done((union ctl_io *)ctsio);
7563				return (1);
7564		        }
7565
7566			/* temporarily unregister this nexus */
7567			lun->per_res[residx].registered = 0;
7568
7569			/*
7570			 * Unregister everybody else and build UA for
7571			 * them
7572			 */
7573			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7574				if (lun->per_res[i].registered == 0)
7575					continue;
7576
7577				if (!persis_offset
7578				 && i <CTL_MAX_INITIATORS)
7579					lun->pending_sense[i].ua_pending |=
7580						CTL_UA_REG_PREEMPT;
7581				else if (persis_offset
7582				      && i >= persis_offset)
7583					lun->pending_sense[i-persis_offset
7584						].ua_pending |=
7585						CTL_UA_REG_PREEMPT;
7586				lun->per_res[i].registered = 0;
7587				memset(&lun->per_res[i].res_key, 0,
7588				       sizeof(struct scsi_per_res_key));
7589			}
7590			lun->per_res[residx].registered = 1;
7591			lun->pr_key_count = 1;
7592			lun->res_type = type;
7593			if (lun->res_type != SPR_TYPE_WR_EX_AR
7594			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7595				lun->pr_res_idx = residx;
7596
7597			mtx_unlock(&softc->ctl_lock);
7598			/* send msg to other side */
7599			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7600			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7601			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7602			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7603			persis_io.pr.pr_info.res_type = type;
7604			memcpy(persis_io.pr.pr_info.sa_res_key,
7605			       param->serv_act_res_key,
7606			       sizeof(param->serv_act_res_key));
7607			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7608			     &persis_io, sizeof(persis_io), 0)) >
7609			     CTL_HA_STATUS_SUCCESS) {
7610				printf("CTL:Persis Out error returned "
7611				       "from ctl_ha_msg_send %d\n",
7612				       isc_retval);
7613			}
7614		} else {
7615			/* not all registrants */
7616			mtx_unlock(&softc->ctl_lock);
7617			free(ctsio->kern_data_ptr, M_CTL);
7618			ctl_set_invalid_field(ctsio,
7619					      /*sks_valid*/ 1,
7620					      /*command*/ 0,
7621					      /*field*/ 8,
7622					      /*bit_valid*/ 0,
7623					      /*bit*/ 0);
7624			ctl_done((union ctl_io *)ctsio);
7625			return (1);
7626		}
7627	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7628		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7629		int found = 0;
7630
7631		mtx_lock(&softc->ctl_lock);
7632		if (res_key == sa_res_key) {
7633			/* special case */
7634			/*
7635			 * The spec implies this is not good but doesn't
7636			 * say what to do. There are two choices either
7637			 * generate a res conflict or check condition
7638			 * with illegal field in parameter data. Since
7639			 * that is what is done when the sa_res_key is
7640			 * zero I'll take that approach since this has
7641			 * to do with the sa_res_key.
7642			 */
7643			mtx_unlock(&softc->ctl_lock);
7644			free(ctsio->kern_data_ptr, M_CTL);
7645			ctl_set_invalid_field(ctsio,
7646					      /*sks_valid*/ 1,
7647					      /*command*/ 0,
7648					      /*field*/ 8,
7649					      /*bit_valid*/ 0,
7650					      /*bit*/ 0);
7651			ctl_done((union ctl_io *)ctsio);
7652			return (1);
7653		}
7654
7655		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7656			if (lun->per_res[i].registered
7657			 && memcmp(param->serv_act_res_key,
7658			    lun->per_res[i].res_key.key,
7659			    sizeof(struct scsi_per_res_key)) != 0)
7660				continue;
7661
7662			found = 1;
7663			lun->per_res[i].registered = 0;
7664			memset(&lun->per_res[i].res_key, 0,
7665			       sizeof(struct scsi_per_res_key));
7666			lun->pr_key_count--;
7667
7668			if (!persis_offset
7669			 && i < CTL_MAX_INITIATORS)
7670				lun->pending_sense[i].ua_pending |=
7671					CTL_UA_REG_PREEMPT;
7672			else if (persis_offset
7673			      && i >= persis_offset)
7674				lun->pending_sense[i-persis_offset].ua_pending|=
7675					CTL_UA_REG_PREEMPT;
7676		}
7677		mtx_unlock(&softc->ctl_lock);
7678		if (!found) {
7679			free(ctsio->kern_data_ptr, M_CTL);
7680			ctl_set_reservation_conflict(ctsio);
7681			ctl_done((union ctl_io *)ctsio);
7682			return (CTL_RETVAL_COMPLETE);
7683		}
7684		/* send msg to other side */
7685		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7686		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7687		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7688		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7689		persis_io.pr.pr_info.res_type = type;
7690		memcpy(persis_io.pr.pr_info.sa_res_key,
7691		       param->serv_act_res_key,
7692		       sizeof(param->serv_act_res_key));
7693		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7694		     &persis_io, sizeof(persis_io), 0)) >
7695		     CTL_HA_STATUS_SUCCESS) {
7696			printf("CTL:Persis Out error returned from "
7697			       "ctl_ha_msg_send %d\n", isc_retval);
7698		}
7699	} else {
7700		/* Reserved but not all registrants */
7701		/* sa_res_key is res holder */
7702		if (memcmp(param->serv_act_res_key,
7703                   lun->per_res[lun->pr_res_idx].res_key.key,
7704                   sizeof(struct scsi_per_res_key)) == 0) {
7705			/* validate scope and type */
7706			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7707			     SPR_LU_SCOPE) {
7708				ctl_set_invalid_field(/*ctsio*/ ctsio,
7709						      /*sks_valid*/ 1,
7710						      /*command*/ 1,
7711						      /*field*/ 2,
7712						      /*bit_valid*/ 1,
7713						      /*bit*/ 4);
7714				ctl_done((union ctl_io *)ctsio);
7715				return (1);
7716			}
7717
7718			if (type>8 || type==2 || type==4 || type==0) {
7719				ctl_set_invalid_field(/*ctsio*/ ctsio,
7720						      /*sks_valid*/ 1,
7721						      /*command*/ 1,
7722						      /*field*/ 2,
7723						      /*bit_valid*/ 1,
7724						      /*bit*/ 0);
7725				ctl_done((union ctl_io *)ctsio);
7726				return (1);
7727			}
7728
7729			/*
7730			 * Do the following:
7731			 * if sa_res_key != res_key remove all
7732			 * registrants w/sa_res_key and generate UA
7733			 * for these registrants(Registrations
7734			 * Preempted) if it wasn't an exclusive
7735			 * reservation generate UA(Reservations
7736			 * Preempted) for all other registered nexuses
7737			 * if the type has changed. Establish the new
7738			 * reservation and holder. If res_key and
7739			 * sa_res_key are the same do the above
7740			 * except don't unregister the res holder.
7741			 */
7742
7743			/*
7744			 * Temporarily unregister so it won't get
7745			 * removed or UA generated
7746			 */
7747			lun->per_res[residx].registered = 0;
7748			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7749				if (lun->per_res[i].registered == 0)
7750					continue;
7751
7752				if (memcmp(param->serv_act_res_key,
7753				    lun->per_res[i].res_key.key,
7754				    sizeof(struct scsi_per_res_key)) == 0) {
7755					lun->per_res[i].registered = 0;
7756					memset(&lun->per_res[i].res_key,
7757					       0,
7758					       sizeof(struct scsi_per_res_key));
7759					lun->pr_key_count--;
7760
7761					if (!persis_offset
7762					 && i < CTL_MAX_INITIATORS)
7763						lun->pending_sense[i
7764							].ua_pending |=
7765							CTL_UA_REG_PREEMPT;
7766					else if (persis_offset
7767					      && i >= persis_offset)
7768						lun->pending_sense[
7769						  i-persis_offset].ua_pending |=
7770						  CTL_UA_REG_PREEMPT;
7771				} else if (type != lun->res_type
7772					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7773					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7774						if (!persis_offset
7775						 && i < CTL_MAX_INITIATORS)
7776							lun->pending_sense[i
7777							].ua_pending |=
7778							CTL_UA_RES_RELEASE;
7779						else if (persis_offset
7780						      && i >= persis_offset)
7781							lun->pending_sense[
7782							i-persis_offset
7783							].ua_pending |=
7784							CTL_UA_RES_RELEASE;
7785				}
7786			}
7787			lun->per_res[residx].registered = 1;
7788			lun->res_type = type;
7789			if (lun->res_type != SPR_TYPE_WR_EX_AR
7790			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7791				lun->pr_res_idx = residx;
7792			else
7793				lun->pr_res_idx =
7794					CTL_PR_ALL_REGISTRANTS;
7795
7796			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7797			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7798			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7799			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7800			persis_io.pr.pr_info.res_type = type;
7801			memcpy(persis_io.pr.pr_info.sa_res_key,
7802			       param->serv_act_res_key,
7803			       sizeof(param->serv_act_res_key));
7804			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7805			     &persis_io, sizeof(persis_io), 0)) >
7806			     CTL_HA_STATUS_SUCCESS) {
7807				printf("CTL:Persis Out error returned "
7808				       "from ctl_ha_msg_send %d\n",
7809				       isc_retval);
7810			}
7811		} else {
7812			/*
7813			 * sa_res_key is not the res holder just
7814			 * remove registrants
7815			 */
7816			int found=0;
7817			mtx_lock(&softc->ctl_lock);
7818
7819			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7820				if (memcmp(param->serv_act_res_key,
7821				    lun->per_res[i].res_key.key,
7822				    sizeof(struct scsi_per_res_key)) != 0)
7823					continue;
7824
7825				found = 1;
7826				lun->per_res[i].registered = 0;
7827				memset(&lun->per_res[i].res_key, 0,
7828				       sizeof(struct scsi_per_res_key));
7829				lun->pr_key_count--;
7830
7831				if (!persis_offset
7832				 && i < CTL_MAX_INITIATORS)
7833					lun->pending_sense[i].ua_pending |=
7834						CTL_UA_REG_PREEMPT;
7835				else if (persis_offset
7836				      && i >= persis_offset)
7837					lun->pending_sense[
7838						i-persis_offset].ua_pending |=
7839						CTL_UA_REG_PREEMPT;
7840			}
7841
7842			if (!found) {
7843				mtx_unlock(&softc->ctl_lock);
7844				free(ctsio->kern_data_ptr, M_CTL);
7845				ctl_set_reservation_conflict(ctsio);
7846				ctl_done((union ctl_io *)ctsio);
7847		        	return (1);
7848			}
7849			mtx_unlock(&softc->ctl_lock);
7850			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7851			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7852			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7853			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7854			persis_io.pr.pr_info.res_type = type;
7855			memcpy(persis_io.pr.pr_info.sa_res_key,
7856			       param->serv_act_res_key,
7857			       sizeof(param->serv_act_res_key));
7858			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7859			     &persis_io, sizeof(persis_io), 0)) >
7860			     CTL_HA_STATUS_SUCCESS) {
7861				printf("CTL:Persis Out error returned "
7862				       "from ctl_ha_msg_send %d\n",
7863				isc_retval);
7864			}
7865		}
7866	}
7867
7868	lun->PRGeneration++;
7869
7870	return (retval);
7871}
7872
7873static void
7874ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7875{
7876	int i;
7877
7878	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7879	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7880	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7881		   msg->pr.pr_info.sa_res_key,
7882		   sizeof(struct scsi_per_res_key)) != 0) {
7883		uint64_t sa_res_key;
7884		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7885
7886		if (sa_res_key == 0) {
7887			/* temporarily unregister this nexus */
7888			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7889
7890			/*
7891			 * Unregister everybody else and build UA for
7892			 * them
7893			 */
7894			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7895				if (lun->per_res[i].registered == 0)
7896					continue;
7897
7898				if (!persis_offset
7899				 && i < CTL_MAX_INITIATORS)
7900					lun->pending_sense[i].ua_pending |=
7901						CTL_UA_REG_PREEMPT;
7902				else if (persis_offset && i >= persis_offset)
7903					lun->pending_sense[i -
7904						persis_offset].ua_pending |=
7905						CTL_UA_REG_PREEMPT;
7906				lun->per_res[i].registered = 0;
7907				memset(&lun->per_res[i].res_key, 0,
7908				       sizeof(struct scsi_per_res_key));
7909			}
7910
7911			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7912			lun->pr_key_count = 1;
7913			lun->res_type = msg->pr.pr_info.res_type;
7914			if (lun->res_type != SPR_TYPE_WR_EX_AR
7915			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7916				lun->pr_res_idx = msg->pr.pr_info.residx;
7917		} else {
7918		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7919				if (memcmp(msg->pr.pr_info.sa_res_key,
7920		                   lun->per_res[i].res_key.key,
7921		                   sizeof(struct scsi_per_res_key)) != 0)
7922					continue;
7923
7924				lun->per_res[i].registered = 0;
7925				memset(&lun->per_res[i].res_key, 0,
7926				       sizeof(struct scsi_per_res_key));
7927				lun->pr_key_count--;
7928
7929				if (!persis_offset
7930				 && i < persis_offset)
7931					lun->pending_sense[i].ua_pending |=
7932						CTL_UA_REG_PREEMPT;
7933				else if (persis_offset
7934				      && i >= persis_offset)
7935					lun->pending_sense[i -
7936						persis_offset].ua_pending |=
7937						CTL_UA_REG_PREEMPT;
7938			}
7939		}
7940	} else {
7941		/*
7942		 * Temporarily unregister so it won't get removed
7943		 * or UA generated
7944		 */
7945		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7946		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7947			if (lun->per_res[i].registered == 0)
7948				continue;
7949
7950			if (memcmp(msg->pr.pr_info.sa_res_key,
7951	                   lun->per_res[i].res_key.key,
7952	                   sizeof(struct scsi_per_res_key)) == 0) {
7953				lun->per_res[i].registered = 0;
7954				memset(&lun->per_res[i].res_key, 0,
7955				       sizeof(struct scsi_per_res_key));
7956				lun->pr_key_count--;
7957				if (!persis_offset
7958				 && i < CTL_MAX_INITIATORS)
7959					lun->pending_sense[i].ua_pending |=
7960						CTL_UA_REG_PREEMPT;
7961				else if (persis_offset
7962				      && i >= persis_offset)
7963					lun->pending_sense[i -
7964						persis_offset].ua_pending |=
7965						CTL_UA_REG_PREEMPT;
7966			} else if (msg->pr.pr_info.res_type != lun->res_type
7967				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7968				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7969					if (!persis_offset
7970					 && i < persis_offset)
7971						lun->pending_sense[i
7972							].ua_pending |=
7973							CTL_UA_RES_RELEASE;
7974					else if (persis_offset
7975					      && i >= persis_offset)
7976					lun->pending_sense[i -
7977						persis_offset].ua_pending |=
7978						CTL_UA_RES_RELEASE;
7979			}
7980		}
7981		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7982		lun->res_type = msg->pr.pr_info.res_type;
7983		if (lun->res_type != SPR_TYPE_WR_EX_AR
7984		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7985			lun->pr_res_idx = msg->pr.pr_info.residx;
7986		else
7987			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7988	}
7989	lun->PRGeneration++;
7990
7991}
7992
7993
7994int
7995ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
7996{
7997	int retval;
7998	int isc_retval;
7999	u_int32_t param_len;
8000	struct scsi_per_res_out *cdb;
8001	struct ctl_lun *lun;
8002	struct scsi_per_res_out_parms* param;
8003	struct ctl_softc *softc;
8004	uint32_t residx;
8005	uint64_t res_key, sa_res_key;
8006	uint8_t type;
8007	union ctl_ha_msg persis_io;
8008	int    i;
8009
8010	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8011
8012	retval = CTL_RETVAL_COMPLETE;
8013
8014	softc = control_softc;
8015
8016	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8017	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8018
8019	/*
8020	 * We only support whole-LUN scope.  The scope & type are ignored for
8021	 * register, register and ignore existing key and clear.
8022	 * We sometimes ignore scope and type on preempts too!!
8023	 * Verify reservation type here as well.
8024	 */
8025	type = cdb->scope_type & SPR_TYPE_MASK;
8026	if ((cdb->action == SPRO_RESERVE)
8027	 || (cdb->action == SPRO_RELEASE)) {
8028		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8029			ctl_set_invalid_field(/*ctsio*/ ctsio,
8030					      /*sks_valid*/ 1,
8031					      /*command*/ 1,
8032					      /*field*/ 2,
8033					      /*bit_valid*/ 1,
8034					      /*bit*/ 4);
8035			ctl_done((union ctl_io *)ctsio);
8036			return (CTL_RETVAL_COMPLETE);
8037		}
8038
8039		if (type>8 || type==2 || type==4 || type==0) {
8040			ctl_set_invalid_field(/*ctsio*/ ctsio,
8041					      /*sks_valid*/ 1,
8042					      /*command*/ 1,
8043					      /*field*/ 2,
8044					      /*bit_valid*/ 1,
8045					      /*bit*/ 0);
8046			ctl_done((union ctl_io *)ctsio);
8047			return (CTL_RETVAL_COMPLETE);
8048		}
8049	}
8050
8051	switch (cdb->action & SPRO_ACTION_MASK) {
8052	case SPRO_REGISTER:
8053	case SPRO_RESERVE:
8054	case SPRO_RELEASE:
8055	case SPRO_CLEAR:
8056	case SPRO_PREEMPT:
8057	case SPRO_REG_IGNO:
8058		break;
8059	case SPRO_REG_MOVE:
8060	case SPRO_PRE_ABO:
8061	default:
8062		ctl_set_invalid_field(/*ctsio*/ ctsio,
8063				      /*sks_valid*/ 1,
8064				      /*command*/ 1,
8065				      /*field*/ 1,
8066				      /*bit_valid*/ 1,
8067				      /*bit*/ 0);
8068		ctl_done((union ctl_io *)ctsio);
8069		return (CTL_RETVAL_COMPLETE);
8070		break; /* NOTREACHED */
8071	}
8072
8073	param_len = scsi_4btoul(cdb->length);
8074
8075	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8076		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8077		ctsio->kern_data_len = param_len;
8078		ctsio->kern_total_len = param_len;
8079		ctsio->kern_data_resid = 0;
8080		ctsio->kern_rel_offset = 0;
8081		ctsio->kern_sg_entries = 0;
8082		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8083		ctsio->be_move_done = ctl_config_move_done;
8084		ctl_datamove((union ctl_io *)ctsio);
8085
8086		return (CTL_RETVAL_COMPLETE);
8087	}
8088
8089	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8090
8091	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8092	res_key = scsi_8btou64(param->res_key.key);
8093	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8094
8095	/*
8096	 * Validate the reservation key here except for SPRO_REG_IGNO
8097	 * This must be done for all other service actions
8098	 */
8099	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8100		mtx_lock(&softc->ctl_lock);
8101		if (lun->per_res[residx].registered) {
8102		    if (memcmp(param->res_key.key,
8103			       lun->per_res[residx].res_key.key,
8104			       ctl_min(sizeof(param->res_key),
8105			       sizeof(lun->per_res[residx].res_key))) != 0) {
8106				/*
8107				 * The current key passed in doesn't match
8108				 * the one the initiator previously
8109				 * registered.
8110				 */
8111				mtx_unlock(&softc->ctl_lock);
8112				free(ctsio->kern_data_ptr, M_CTL);
8113				ctl_set_reservation_conflict(ctsio);
8114				ctl_done((union ctl_io *)ctsio);
8115				return (CTL_RETVAL_COMPLETE);
8116			}
8117		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8118			/*
8119			 * We are not registered
8120			 */
8121			mtx_unlock(&softc->ctl_lock);
8122			free(ctsio->kern_data_ptr, M_CTL);
8123			ctl_set_reservation_conflict(ctsio);
8124			ctl_done((union ctl_io *)ctsio);
8125			return (CTL_RETVAL_COMPLETE);
8126		} else if (res_key != 0) {
8127			/*
8128			 * We are not registered and trying to register but
8129			 * the register key isn't zero.
8130			 */
8131			mtx_unlock(&softc->ctl_lock);
8132			free(ctsio->kern_data_ptr, M_CTL);
8133			ctl_set_reservation_conflict(ctsio);
8134			ctl_done((union ctl_io *)ctsio);
8135			return (CTL_RETVAL_COMPLETE);
8136		}
8137		mtx_unlock(&softc->ctl_lock);
8138	}
8139
8140	switch (cdb->action & SPRO_ACTION_MASK) {
8141	case SPRO_REGISTER:
8142	case SPRO_REG_IGNO: {
8143
8144#if 0
8145		printf("Registration received\n");
8146#endif
8147
8148		/*
8149		 * We don't support any of these options, as we report in
8150		 * the read capabilities request (see
8151		 * ctl_persistent_reserve_in(), above).
8152		 */
8153		if ((param->flags & SPR_SPEC_I_PT)
8154		 || (param->flags & SPR_ALL_TG_PT)
8155		 || (param->flags & SPR_APTPL)) {
8156			int bit_ptr;
8157
8158			if (param->flags & SPR_APTPL)
8159				bit_ptr = 0;
8160			else if (param->flags & SPR_ALL_TG_PT)
8161				bit_ptr = 2;
8162			else /* SPR_SPEC_I_PT */
8163				bit_ptr = 3;
8164
8165			free(ctsio->kern_data_ptr, M_CTL);
8166			ctl_set_invalid_field(ctsio,
8167					      /*sks_valid*/ 1,
8168					      /*command*/ 0,
8169					      /*field*/ 20,
8170					      /*bit_valid*/ 1,
8171					      /*bit*/ bit_ptr);
8172			ctl_done((union ctl_io *)ctsio);
8173			return (CTL_RETVAL_COMPLETE);
8174		}
8175
8176		mtx_lock(&softc->ctl_lock);
8177
8178		/*
8179		 * The initiator wants to clear the
8180		 * key/unregister.
8181		 */
8182		if (sa_res_key == 0) {
8183			if ((res_key == 0
8184			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8185			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8186			  && !lun->per_res[residx].registered)) {
8187				mtx_unlock(&softc->ctl_lock);
8188				goto done;
8189			}
8190
8191			lun->per_res[residx].registered = 0;
8192			memset(&lun->per_res[residx].res_key,
8193			       0, sizeof(lun->per_res[residx].res_key));
8194			lun->pr_key_count--;
8195
8196			if (residx == lun->pr_res_idx) {
8197				lun->flags &= ~CTL_LUN_PR_RESERVED;
8198				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8199
8200				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8201				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8202				 && lun->pr_key_count) {
8203					/*
8204					 * If the reservation is a registrants
8205					 * only type we need to generate a UA
8206					 * for other registered inits.  The
8207					 * sense code should be RESERVATIONS
8208					 * RELEASED
8209					 */
8210
8211					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8212						if (lun->per_res[
8213						    i+persis_offset].registered
8214						    == 0)
8215							continue;
8216						lun->pending_sense[i
8217							].ua_pending |=
8218							CTL_UA_RES_RELEASE;
8219					}
8220				}
8221				lun->res_type = 0;
8222			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8223				if (lun->pr_key_count==0) {
8224					lun->flags &= ~CTL_LUN_PR_RESERVED;
8225					lun->res_type = 0;
8226					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8227				}
8228			}
8229			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8230			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8231			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8232			persis_io.pr.pr_info.residx = residx;
8233			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8234			     &persis_io, sizeof(persis_io), 0 )) >
8235			     CTL_HA_STATUS_SUCCESS) {
8236				printf("CTL:Persis Out error returned from "
8237				       "ctl_ha_msg_send %d\n", isc_retval);
8238			}
8239			mtx_unlock(&softc->ctl_lock);
8240		} else /* sa_res_key != 0 */ {
8241
8242			/*
8243			 * If we aren't registered currently then increment
8244			 * the key count and set the registered flag.
8245			 */
8246			if (!lun->per_res[residx].registered) {
8247				lun->pr_key_count++;
8248				lun->per_res[residx].registered = 1;
8249			}
8250
8251			memcpy(&lun->per_res[residx].res_key,
8252			       param->serv_act_res_key,
8253			       ctl_min(sizeof(param->serv_act_res_key),
8254			       sizeof(lun->per_res[residx].res_key)));
8255
8256			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8257			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8258			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8259			persis_io.pr.pr_info.residx = residx;
8260			memcpy(persis_io.pr.pr_info.sa_res_key,
8261			       param->serv_act_res_key,
8262			       sizeof(param->serv_act_res_key));
8263			mtx_unlock(&softc->ctl_lock);
8264			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8265			     &persis_io, sizeof(persis_io), 0)) >
8266			     CTL_HA_STATUS_SUCCESS) {
8267				printf("CTL:Persis Out error returned from "
8268				       "ctl_ha_msg_send %d\n", isc_retval);
8269			}
8270		}
8271		lun->PRGeneration++;
8272
8273		break;
8274	}
8275	case SPRO_RESERVE:
8276#if 0
8277                printf("Reserve executed type %d\n", type);
8278#endif
8279		mtx_lock(&softc->ctl_lock);
8280		if (lun->flags & CTL_LUN_PR_RESERVED) {
8281			/*
8282			 * if this isn't the reservation holder and it's
8283			 * not a "all registrants" type or if the type is
8284			 * different then we have a conflict
8285			 */
8286			if ((lun->pr_res_idx != residx
8287			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8288			 || lun->res_type != type) {
8289				mtx_unlock(&softc->ctl_lock);
8290				free(ctsio->kern_data_ptr, M_CTL);
8291				ctl_set_reservation_conflict(ctsio);
8292				ctl_done((union ctl_io *)ctsio);
8293				return (CTL_RETVAL_COMPLETE);
8294			}
8295			mtx_unlock(&softc->ctl_lock);
8296		} else /* create a reservation */ {
8297			/*
8298			 * If it's not an "all registrants" type record
8299			 * reservation holder
8300			 */
8301			if (type != SPR_TYPE_WR_EX_AR
8302			 && type != SPR_TYPE_EX_AC_AR)
8303				lun->pr_res_idx = residx; /* Res holder */
8304			else
8305				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8306
8307			lun->flags |= CTL_LUN_PR_RESERVED;
8308			lun->res_type = type;
8309
8310			mtx_unlock(&softc->ctl_lock);
8311
8312			/* send msg to other side */
8313			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8314			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8315			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8316			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8317			persis_io.pr.pr_info.res_type = type;
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		}
8325		break;
8326
8327	case SPRO_RELEASE:
8328		mtx_lock(&softc->ctl_lock);
8329		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8330			/* No reservation exists return good status */
8331			mtx_unlock(&softc->ctl_lock);
8332			goto done;
8333		}
8334		/*
8335		 * Is this nexus a reservation holder?
8336		 */
8337		if (lun->pr_res_idx != residx
8338		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8339			/*
8340			 * not a res holder return good status but
8341			 * do nothing
8342			 */
8343			mtx_unlock(&softc->ctl_lock);
8344			goto done;
8345		}
8346
8347		if (lun->res_type != type) {
8348			mtx_unlock(&softc->ctl_lock);
8349			free(ctsio->kern_data_ptr, M_CTL);
8350			ctl_set_illegal_pr_release(ctsio);
8351			ctl_done((union ctl_io *)ctsio);
8352			return (CTL_RETVAL_COMPLETE);
8353		}
8354
8355		/* okay to release */
8356		lun->flags &= ~CTL_LUN_PR_RESERVED;
8357		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8358		lun->res_type = 0;
8359
8360		/*
8361		 * if this isn't an exclusive access
8362		 * res generate UA for all other
8363		 * registrants.
8364		 */
8365		if (type != SPR_TYPE_EX_AC
8366		 && type != SPR_TYPE_WR_EX) {
8367			/*
8368			 * temporarily unregister so we don't generate UA
8369			 */
8370			lun->per_res[residx].registered = 0;
8371
8372			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8373				if (lun->per_res[i+persis_offset].registered
8374				    == 0)
8375					continue;
8376				lun->pending_sense[i].ua_pending |=
8377					CTL_UA_RES_RELEASE;
8378			}
8379
8380			lun->per_res[residx].registered = 1;
8381		}
8382		mtx_unlock(&softc->ctl_lock);
8383		/* Send msg to other side */
8384		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8385		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8386		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8387		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8388		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8389			printf("CTL:Persis Out error returned from "
8390			       "ctl_ha_msg_send %d\n", isc_retval);
8391		}
8392		break;
8393
8394	case SPRO_CLEAR:
8395		/* send msg to other side */
8396
8397		mtx_lock(&softc->ctl_lock);
8398		lun->flags &= ~CTL_LUN_PR_RESERVED;
8399		lun->res_type = 0;
8400		lun->pr_key_count = 0;
8401		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8402
8403
8404		memset(&lun->per_res[residx].res_key,
8405		       0, sizeof(lun->per_res[residx].res_key));
8406		lun->per_res[residx].registered = 0;
8407
8408		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8409			if (lun->per_res[i].registered) {
8410				if (!persis_offset && i < CTL_MAX_INITIATORS)
8411					lun->pending_sense[i].ua_pending |=
8412						CTL_UA_RES_PREEMPT;
8413				else if (persis_offset && i >= persis_offset)
8414					lun->pending_sense[i-persis_offset
8415					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8416
8417				memset(&lun->per_res[i].res_key,
8418				       0, sizeof(struct scsi_per_res_key));
8419				lun->per_res[i].registered = 0;
8420			}
8421		lun->PRGeneration++;
8422		mtx_unlock(&softc->ctl_lock);
8423		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8424		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8425		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8426		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8427		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8428			printf("CTL:Persis Out error returned from "
8429			       "ctl_ha_msg_send %d\n", isc_retval);
8430		}
8431		break;
8432
8433	case SPRO_PREEMPT: {
8434		int nretval;
8435
8436		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8437					  residx, ctsio, cdb, param);
8438		if (nretval != 0)
8439			return (CTL_RETVAL_COMPLETE);
8440		break;
8441	}
8442	case SPRO_REG_MOVE:
8443	case SPRO_PRE_ABO:
8444	default:
8445		free(ctsio->kern_data_ptr, M_CTL);
8446		ctl_set_invalid_field(/*ctsio*/ ctsio,
8447				      /*sks_valid*/ 1,
8448				      /*command*/ 1,
8449				      /*field*/ 1,
8450				      /*bit_valid*/ 1,
8451				      /*bit*/ 0);
8452		ctl_done((union ctl_io *)ctsio);
8453		return (CTL_RETVAL_COMPLETE);
8454		break; /* NOTREACHED */
8455	}
8456
8457done:
8458	free(ctsio->kern_data_ptr, M_CTL);
8459	ctl_set_success(ctsio);
8460	ctl_done((union ctl_io *)ctsio);
8461
8462	return (retval);
8463}
8464
8465/*
8466 * This routine is for handling a message from the other SC pertaining to
8467 * persistent reserve out. All the error checking will have been done
8468 * so only perorming the action need be done here to keep the two
8469 * in sync.
8470 */
8471static void
8472ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8473{
8474	struct ctl_lun *lun;
8475	struct ctl_softc *softc;
8476	int i;
8477	uint32_t targ_lun;
8478
8479	softc = control_softc;
8480
8481	mtx_lock(&softc->ctl_lock);
8482
8483	targ_lun = msg->hdr.nexus.targ_lun;
8484	if (msg->hdr.nexus.lun_map_fn != NULL)
8485		targ_lun = msg->hdr.nexus.lun_map_fn(msg->hdr.nexus.lun_map_arg, targ_lun);
8486	lun = softc->ctl_luns[targ_lun];
8487	switch(msg->pr.pr_info.action) {
8488	case CTL_PR_REG_KEY:
8489		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8490			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8491			lun->pr_key_count++;
8492		}
8493		lun->PRGeneration++;
8494		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8495		       msg->pr.pr_info.sa_res_key,
8496		       sizeof(struct scsi_per_res_key));
8497		break;
8498
8499	case CTL_PR_UNREG_KEY:
8500		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8501		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8502		       0, sizeof(struct scsi_per_res_key));
8503		lun->pr_key_count--;
8504
8505		/* XXX Need to see if the reservation has been released */
8506		/* if so do we need to generate UA? */
8507		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8508			lun->flags &= ~CTL_LUN_PR_RESERVED;
8509			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8510
8511			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8512			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8513			 && lun->pr_key_count) {
8514				/*
8515				 * If the reservation is a registrants
8516				 * only type we need to generate a UA
8517				 * for other registered inits.  The
8518				 * sense code should be RESERVATIONS
8519				 * RELEASED
8520				 */
8521
8522				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8523					if (lun->per_res[i+
8524					    persis_offset].registered == 0)
8525						continue;
8526
8527					lun->pending_sense[i
8528						].ua_pending |=
8529						CTL_UA_RES_RELEASE;
8530				}
8531			}
8532			lun->res_type = 0;
8533		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8534			if (lun->pr_key_count==0) {
8535				lun->flags &= ~CTL_LUN_PR_RESERVED;
8536				lun->res_type = 0;
8537				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8538			}
8539		}
8540		lun->PRGeneration++;
8541		break;
8542
8543	case CTL_PR_RESERVE:
8544		lun->flags |= CTL_LUN_PR_RESERVED;
8545		lun->res_type = msg->pr.pr_info.res_type;
8546		lun->pr_res_idx = msg->pr.pr_info.residx;
8547
8548		break;
8549
8550	case CTL_PR_RELEASE:
8551		/*
8552		 * if this isn't an exclusive access res generate UA for all
8553		 * other registrants.
8554		 */
8555		if (lun->res_type != SPR_TYPE_EX_AC
8556		 && lun->res_type != SPR_TYPE_WR_EX) {
8557			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8558				if (lun->per_res[i+persis_offset].registered)
8559					lun->pending_sense[i].ua_pending |=
8560						CTL_UA_RES_RELEASE;
8561		}
8562
8563		lun->flags &= ~CTL_LUN_PR_RESERVED;
8564		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8565		lun->res_type = 0;
8566		break;
8567
8568	case CTL_PR_PREEMPT:
8569		ctl_pro_preempt_other(lun, msg);
8570		break;
8571	case CTL_PR_CLEAR:
8572		lun->flags &= ~CTL_LUN_PR_RESERVED;
8573		lun->res_type = 0;
8574		lun->pr_key_count = 0;
8575		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8576
8577		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8578			if (lun->per_res[i].registered == 0)
8579				continue;
8580			if (!persis_offset
8581			 && i < CTL_MAX_INITIATORS)
8582				lun->pending_sense[i].ua_pending |=
8583					CTL_UA_RES_PREEMPT;
8584			else if (persis_offset
8585			      && i >= persis_offset)
8586   				lun->pending_sense[i-persis_offset].ua_pending|=
8587					CTL_UA_RES_PREEMPT;
8588			memset(&lun->per_res[i].res_key, 0,
8589			       sizeof(struct scsi_per_res_key));
8590			lun->per_res[i].registered = 0;
8591		}
8592		lun->PRGeneration++;
8593		break;
8594	}
8595
8596	mtx_unlock(&softc->ctl_lock);
8597}
8598
8599int
8600ctl_read_write(struct ctl_scsiio *ctsio)
8601{
8602	struct ctl_lun *lun;
8603	struct ctl_lba_len_flags *lbalen;
8604	uint64_t lba;
8605	uint32_t num_blocks;
8606	int reladdr, fua, dpo, ebp;
8607	int retval;
8608	int isread;
8609
8610	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8611
8612	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8613
8614	reladdr = 0;
8615	fua = 0;
8616	dpo = 0;
8617	ebp = 0;
8618
8619	retval = CTL_RETVAL_COMPLETE;
8620
8621	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8622	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8623	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8624		uint32_t residx;
8625
8626		/*
8627		 * XXX KDM need a lock here.
8628		 */
8629		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8630		if ((lun->res_type == SPR_TYPE_EX_AC
8631		  && residx != lun->pr_res_idx)
8632		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8633		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8634		  && !lun->per_res[residx].registered)) {
8635			ctl_set_reservation_conflict(ctsio);
8636			ctl_done((union ctl_io *)ctsio);
8637			return (CTL_RETVAL_COMPLETE);
8638	        }
8639	}
8640
8641	switch (ctsio->cdb[0]) {
8642	case READ_6:
8643	case WRITE_6: {
8644		struct scsi_rw_6 *cdb;
8645
8646		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8647
8648		lba = scsi_3btoul(cdb->addr);
8649		/* only 5 bits are valid in the most significant address byte */
8650		lba &= 0x1fffff;
8651		num_blocks = cdb->length;
8652		/*
8653		 * This is correct according to SBC-2.
8654		 */
8655		if (num_blocks == 0)
8656			num_blocks = 256;
8657		break;
8658	}
8659	case READ_10:
8660	case WRITE_10: {
8661		struct scsi_rw_10 *cdb;
8662
8663		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8664
8665		if (cdb->byte2 & SRW10_RELADDR)
8666			reladdr = 1;
8667		if (cdb->byte2 & SRW10_FUA)
8668			fua = 1;
8669		if (cdb->byte2 & SRW10_DPO)
8670			dpo = 1;
8671
8672		if ((cdb->opcode == WRITE_10)
8673		 && (cdb->byte2 & SRW10_EBP))
8674			ebp = 1;
8675
8676		lba = scsi_4btoul(cdb->addr);
8677		num_blocks = scsi_2btoul(cdb->length);
8678		break;
8679	}
8680	case WRITE_VERIFY_10: {
8681		struct scsi_write_verify_10 *cdb;
8682
8683		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8684
8685		/*
8686		 * XXX KDM we should do actual write verify support at some
8687		 * point.  This is obviously fake, we're just translating
8688		 * things to a write.  So we don't even bother checking the
8689		 * BYTCHK field, since we don't do any verification.  If
8690		 * the user asks for it, we'll just pretend we did it.
8691		 */
8692		if (cdb->byte2 & SWV_DPO)
8693			dpo = 1;
8694
8695		lba = scsi_4btoul(cdb->addr);
8696		num_blocks = scsi_2btoul(cdb->length);
8697		break;
8698	}
8699	case READ_12:
8700	case WRITE_12: {
8701		struct scsi_rw_12 *cdb;
8702
8703		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8704
8705		if (cdb->byte2 & SRW12_RELADDR)
8706			reladdr = 1;
8707		if (cdb->byte2 & SRW12_FUA)
8708			fua = 1;
8709		if (cdb->byte2 & SRW12_DPO)
8710			dpo = 1;
8711		lba = scsi_4btoul(cdb->addr);
8712		num_blocks = scsi_4btoul(cdb->length);
8713		break;
8714	}
8715	case WRITE_VERIFY_12: {
8716		struct scsi_write_verify_12 *cdb;
8717
8718		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8719
8720		if (cdb->byte2 & SWV_DPO)
8721			dpo = 1;
8722
8723		lba = scsi_4btoul(cdb->addr);
8724		num_blocks = scsi_4btoul(cdb->length);
8725
8726		break;
8727	}
8728	case READ_16:
8729	case WRITE_16: {
8730		struct scsi_rw_16 *cdb;
8731
8732		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8733
8734		if (cdb->byte2 & SRW12_RELADDR)
8735			reladdr = 1;
8736		if (cdb->byte2 & SRW12_FUA)
8737			fua = 1;
8738		if (cdb->byte2 & SRW12_DPO)
8739			dpo = 1;
8740
8741		lba = scsi_8btou64(cdb->addr);
8742		num_blocks = scsi_4btoul(cdb->length);
8743		break;
8744	}
8745	case WRITE_VERIFY_16: {
8746		struct scsi_write_verify_16 *cdb;
8747
8748		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8749
8750		if (cdb->byte2 & SWV_DPO)
8751			dpo = 1;
8752
8753		lba = scsi_8btou64(cdb->addr);
8754		num_blocks = scsi_4btoul(cdb->length);
8755		break;
8756	}
8757	default:
8758		/*
8759		 * We got a command we don't support.  This shouldn't
8760		 * happen, commands should be filtered out above us.
8761		 */
8762		ctl_set_invalid_opcode(ctsio);
8763		ctl_done((union ctl_io *)ctsio);
8764
8765		return (CTL_RETVAL_COMPLETE);
8766		break; /* NOTREACHED */
8767	}
8768
8769	/*
8770	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8771	 * interesting for us, but if RAIDCore is in write-back mode,
8772	 * getting it to do write-through for a particular transaction may
8773	 * not be possible.
8774	 */
8775	/*
8776	 * We don't support relative addressing.  That also requires
8777	 * supporting linked commands, which we don't do.
8778	 */
8779	if (reladdr != 0) {
8780		ctl_set_invalid_field(ctsio,
8781				      /*sks_valid*/ 1,
8782				      /*command*/ 1,
8783				      /*field*/ 1,
8784				      /*bit_valid*/ 1,
8785				      /*bit*/ 0);
8786		ctl_done((union ctl_io *)ctsio);
8787		return (CTL_RETVAL_COMPLETE);
8788	}
8789
8790	/*
8791	 * The first check is to make sure we're in bounds, the second
8792	 * check is to catch wrap-around problems.  If the lba + num blocks
8793	 * is less than the lba, then we've wrapped around and the block
8794	 * range is invalid anyway.
8795	 */
8796	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8797	 || ((lba + num_blocks) < lba)) {
8798		ctl_set_lba_out_of_range(ctsio);
8799		ctl_done((union ctl_io *)ctsio);
8800		return (CTL_RETVAL_COMPLETE);
8801	}
8802
8803	/*
8804	 * According to SBC-3, a transfer length of 0 is not an error.
8805	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8806	 * translates to 256 blocks for those commands.
8807	 */
8808	if (num_blocks == 0) {
8809		ctl_set_success(ctsio);
8810		ctl_done((union ctl_io *)ctsio);
8811		return (CTL_RETVAL_COMPLETE);
8812	}
8813
8814	lbalen = (struct ctl_lba_len_flags *)
8815	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8816	lbalen->lba = lba;
8817	lbalen->len = num_blocks;
8818	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8819
8820	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8821	ctsio->kern_rel_offset = 0;
8822
8823	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8824
8825	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8826
8827	return (retval);
8828}
8829
8830static int
8831ctl_cnw_cont(union ctl_io *io)
8832{
8833	struct ctl_scsiio *ctsio;
8834	struct ctl_lun *lun;
8835	struct ctl_lba_len_flags *lbalen;
8836	int retval;
8837
8838	ctsio = &io->scsiio;
8839	ctsio->io_hdr.status = CTL_STATUS_NONE;
8840	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8841	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8842	lbalen = (struct ctl_lba_len_flags *)
8843	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8844	lbalen->flags = CTL_LLF_WRITE;
8845
8846	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8847	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8848	return (retval);
8849}
8850
8851int
8852ctl_cnw(struct ctl_scsiio *ctsio)
8853{
8854	struct ctl_lun *lun;
8855	struct ctl_lba_len_flags *lbalen;
8856	uint64_t lba;
8857	uint32_t num_blocks;
8858	int fua, dpo;
8859	int retval;
8860
8861	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8862
8863	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8864
8865	fua = 0;
8866	dpo = 0;
8867
8868	retval = CTL_RETVAL_COMPLETE;
8869
8870	switch (ctsio->cdb[0]) {
8871	case COMPARE_AND_WRITE: {
8872		struct scsi_compare_and_write *cdb;
8873
8874		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8875
8876		if (cdb->byte2 & SRW10_FUA)
8877			fua = 1;
8878		if (cdb->byte2 & SRW10_DPO)
8879			dpo = 1;
8880		lba = scsi_8btou64(cdb->addr);
8881		num_blocks = cdb->length;
8882		break;
8883	}
8884	default:
8885		/*
8886		 * We got a command we don't support.  This shouldn't
8887		 * happen, commands should be filtered out above us.
8888		 */
8889		ctl_set_invalid_opcode(ctsio);
8890		ctl_done((union ctl_io *)ctsio);
8891
8892		return (CTL_RETVAL_COMPLETE);
8893		break; /* NOTREACHED */
8894	}
8895
8896	/*
8897	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8898	 * interesting for us, but if RAIDCore is in write-back mode,
8899	 * getting it to do write-through for a particular transaction may
8900	 * not be possible.
8901	 */
8902
8903	/*
8904	 * The first check is to make sure we're in bounds, the second
8905	 * check is to catch wrap-around problems.  If the lba + num blocks
8906	 * is less than the lba, then we've wrapped around and the block
8907	 * range is invalid anyway.
8908	 */
8909	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8910	 || ((lba + num_blocks) < lba)) {
8911		ctl_set_lba_out_of_range(ctsio);
8912		ctl_done((union ctl_io *)ctsio);
8913		return (CTL_RETVAL_COMPLETE);
8914	}
8915
8916	/*
8917	 * According to SBC-3, a transfer length of 0 is not an error.
8918	 */
8919	if (num_blocks == 0) {
8920		ctl_set_success(ctsio);
8921		ctl_done((union ctl_io *)ctsio);
8922		return (CTL_RETVAL_COMPLETE);
8923	}
8924
8925	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8926	ctsio->kern_rel_offset = 0;
8927
8928	/*
8929	 * Set the IO_CONT flag, so that if this I/O gets passed to
8930	 * ctl_data_submit_done(), it'll get passed back to
8931	 * ctl_ctl_cnw_cont() for further processing.
8932	 */
8933	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8934	ctsio->io_cont = ctl_cnw_cont;
8935
8936	lbalen = (struct ctl_lba_len_flags *)
8937	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8938	lbalen->lba = lba;
8939	lbalen->len = num_blocks;
8940	lbalen->flags = CTL_LLF_COMPARE;
8941
8942	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8943	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8944	return (retval);
8945}
8946
8947int
8948ctl_verify(struct ctl_scsiio *ctsio)
8949{
8950	struct ctl_lun *lun;
8951	struct ctl_lba_len_flags *lbalen;
8952	uint64_t lba;
8953	uint32_t num_blocks;
8954	int bytchk, dpo;
8955	int retval;
8956
8957	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8958
8959	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8960
8961	bytchk = 0;
8962	dpo = 0;
8963	retval = CTL_RETVAL_COMPLETE;
8964
8965	switch (ctsio->cdb[0]) {
8966	case VERIFY_10: {
8967		struct scsi_verify_10 *cdb;
8968
8969		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8970		if (cdb->byte2 & SVFY_BYTCHK)
8971			bytchk = 1;
8972		if (cdb->byte2 & SVFY_DPO)
8973			dpo = 1;
8974		lba = scsi_4btoul(cdb->addr);
8975		num_blocks = scsi_2btoul(cdb->length);
8976		break;
8977	}
8978	case VERIFY_12: {
8979		struct scsi_verify_12 *cdb;
8980
8981		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8982		if (cdb->byte2 & SVFY_BYTCHK)
8983			bytchk = 1;
8984		if (cdb->byte2 & SVFY_DPO)
8985			dpo = 1;
8986		lba = scsi_4btoul(cdb->addr);
8987		num_blocks = scsi_4btoul(cdb->length);
8988		break;
8989	}
8990	case VERIFY_16: {
8991		struct scsi_rw_16 *cdb;
8992
8993		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8994		if (cdb->byte2 & SVFY_BYTCHK)
8995			bytchk = 1;
8996		if (cdb->byte2 & SVFY_DPO)
8997			dpo = 1;
8998		lba = scsi_8btou64(cdb->addr);
8999		num_blocks = scsi_4btoul(cdb->length);
9000		break;
9001	}
9002	default:
9003		/*
9004		 * We got a command we don't support.  This shouldn't
9005		 * happen, commands should be filtered out above us.
9006		 */
9007		ctl_set_invalid_opcode(ctsio);
9008		ctl_done((union ctl_io *)ctsio);
9009		return (CTL_RETVAL_COMPLETE);
9010	}
9011
9012	/*
9013	 * The first check is to make sure we're in bounds, the second
9014	 * check is to catch wrap-around problems.  If the lba + num blocks
9015	 * is less than the lba, then we've wrapped around and the block
9016	 * range is invalid anyway.
9017	 */
9018	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9019	 || ((lba + num_blocks) < lba)) {
9020		ctl_set_lba_out_of_range(ctsio);
9021		ctl_done((union ctl_io *)ctsio);
9022		return (CTL_RETVAL_COMPLETE);
9023	}
9024
9025	/*
9026	 * According to SBC-3, a transfer length of 0 is not an error.
9027	 */
9028	if (num_blocks == 0) {
9029		ctl_set_success(ctsio);
9030		ctl_done((union ctl_io *)ctsio);
9031		return (CTL_RETVAL_COMPLETE);
9032	}
9033
9034	lbalen = (struct ctl_lba_len_flags *)
9035	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9036	lbalen->lba = lba;
9037	lbalen->len = num_blocks;
9038	if (bytchk) {
9039		lbalen->flags = CTL_LLF_COMPARE;
9040		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9041	} else {
9042		lbalen->flags = CTL_LLF_VERIFY;
9043		ctsio->kern_total_len = 0;
9044	}
9045	ctsio->kern_rel_offset = 0;
9046
9047	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9048	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9049	return (retval);
9050}
9051
9052int
9053ctl_report_luns(struct ctl_scsiio *ctsio)
9054{
9055	struct scsi_report_luns *cdb;
9056	struct scsi_report_luns_data *lun_data;
9057	struct ctl_lun *lun, *request_lun;
9058	int num_luns, retval;
9059	uint32_t alloc_len, lun_datalen;
9060	int num_filled, well_known;
9061	uint32_t initidx, targ_lun_id, lun_id;
9062
9063	retval = CTL_RETVAL_COMPLETE;
9064	well_known = 0;
9065
9066	cdb = (struct scsi_report_luns *)ctsio->cdb;
9067
9068	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9069
9070	mtx_lock(&control_softc->ctl_lock);
9071	num_luns = control_softc->num_luns;
9072	mtx_unlock(&control_softc->ctl_lock);
9073
9074	switch (cdb->select_report) {
9075	case RPL_REPORT_DEFAULT:
9076	case RPL_REPORT_ALL:
9077		break;
9078	case RPL_REPORT_WELLKNOWN:
9079		well_known = 1;
9080		num_luns = 0;
9081		break;
9082	default:
9083		ctl_set_invalid_field(ctsio,
9084				      /*sks_valid*/ 1,
9085				      /*command*/ 1,
9086				      /*field*/ 2,
9087				      /*bit_valid*/ 0,
9088				      /*bit*/ 0);
9089		ctl_done((union ctl_io *)ctsio);
9090		return (retval);
9091		break; /* NOTREACHED */
9092	}
9093
9094	alloc_len = scsi_4btoul(cdb->length);
9095	/*
9096	 * The initiator has to allocate at least 16 bytes for this request,
9097	 * so he can at least get the header and the first LUN.  Otherwise
9098	 * we reject the request (per SPC-3 rev 14, section 6.21).
9099	 */
9100	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9101	    sizeof(struct scsi_report_luns_lundata))) {
9102		ctl_set_invalid_field(ctsio,
9103				      /*sks_valid*/ 1,
9104				      /*command*/ 1,
9105				      /*field*/ 6,
9106				      /*bit_valid*/ 0,
9107				      /*bit*/ 0);
9108		ctl_done((union ctl_io *)ctsio);
9109		return (retval);
9110	}
9111
9112	request_lun = (struct ctl_lun *)
9113		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9114
9115	lun_datalen = sizeof(*lun_data) +
9116		(num_luns * sizeof(struct scsi_report_luns_lundata));
9117
9118	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9119	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9120	ctsio->kern_sg_entries = 0;
9121
9122	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9123
9124	mtx_lock(&control_softc->ctl_lock);
9125	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9126		lun_id = targ_lun_id;
9127		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9128			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9129		if (lun_id >= CTL_MAX_LUNS)
9130			continue;
9131		lun = control_softc->ctl_luns[lun_id];
9132		if (lun == NULL)
9133			continue;
9134
9135		if (targ_lun_id <= 0xff) {
9136			/*
9137			 * Peripheral addressing method, bus number 0.
9138			 */
9139			lun_data->luns[num_filled].lundata[0] =
9140				RPL_LUNDATA_ATYP_PERIPH;
9141			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9142			num_filled++;
9143		} else if (targ_lun_id <= 0x3fff) {
9144			/*
9145			 * Flat addressing method.
9146			 */
9147			lun_data->luns[num_filled].lundata[0] =
9148				RPL_LUNDATA_ATYP_FLAT |
9149				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9150#ifdef OLDCTLHEADERS
9151				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9152				(targ_lun_id & SRLD_BUS_LUN_MASK);
9153#endif
9154			lun_data->luns[num_filled].lundata[1] =
9155#ifdef OLDCTLHEADERS
9156				targ_lun_id >> SRLD_BUS_LUN_BITS;
9157#endif
9158				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9159			num_filled++;
9160		} else {
9161			printf("ctl_report_luns: bogus LUN number %jd, "
9162			       "skipping\n", (intmax_t)targ_lun_id);
9163		}
9164		/*
9165		 * According to SPC-3, rev 14 section 6.21:
9166		 *
9167		 * "The execution of a REPORT LUNS command to any valid and
9168		 * installed logical unit shall clear the REPORTED LUNS DATA
9169		 * HAS CHANGED unit attention condition for all logical
9170		 * units of that target with respect to the requesting
9171		 * initiator. A valid and installed logical unit is one
9172		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9173		 * INQUIRY data (see 6.4.2)."
9174		 *
9175		 * If request_lun is NULL, the LUN this report luns command
9176		 * was issued to is either disabled or doesn't exist. In that
9177		 * case, we shouldn't clear any pending lun change unit
9178		 * attention.
9179		 */
9180		if (request_lun != NULL)
9181			lun->pending_sense[initidx].ua_pending &=
9182				~CTL_UA_LUN_CHANGE;
9183	}
9184	mtx_unlock(&control_softc->ctl_lock);
9185
9186	/*
9187	 * It's quite possible that we've returned fewer LUNs than we allocated
9188	 * space for.  Trim it.
9189	 */
9190	lun_datalen = sizeof(*lun_data) +
9191		(num_filled * sizeof(struct scsi_report_luns_lundata));
9192
9193	if (lun_datalen < alloc_len) {
9194		ctsio->residual = alloc_len - lun_datalen;
9195		ctsio->kern_data_len = lun_datalen;
9196		ctsio->kern_total_len = lun_datalen;
9197	} else {
9198		ctsio->residual = 0;
9199		ctsio->kern_data_len = alloc_len;
9200		ctsio->kern_total_len = alloc_len;
9201	}
9202	ctsio->kern_data_resid = 0;
9203	ctsio->kern_rel_offset = 0;
9204	ctsio->kern_sg_entries = 0;
9205
9206	/*
9207	 * We set this to the actual data length, regardless of how much
9208	 * space we actually have to return results.  If the user looks at
9209	 * this value, he'll know whether or not he allocated enough space
9210	 * and reissue the command if necessary.  We don't support well
9211	 * known logical units, so if the user asks for that, return none.
9212	 */
9213	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9214
9215	/*
9216	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9217	 * this request.
9218	 */
9219	ctsio->scsi_status = SCSI_STATUS_OK;
9220
9221	ctsio->be_move_done = ctl_config_move_done;
9222	ctl_datamove((union ctl_io *)ctsio);
9223
9224	return (retval);
9225}
9226
9227int
9228ctl_request_sense(struct ctl_scsiio *ctsio)
9229{
9230	struct scsi_request_sense *cdb;
9231	struct scsi_sense_data *sense_ptr;
9232	struct ctl_lun *lun;
9233	uint32_t initidx;
9234	int have_error;
9235	scsi_sense_data_type sense_format;
9236
9237	cdb = (struct scsi_request_sense *)ctsio->cdb;
9238
9239	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9240
9241	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9242
9243	/*
9244	 * Determine which sense format the user wants.
9245	 */
9246	if (cdb->byte2 & SRS_DESC)
9247		sense_format = SSD_TYPE_DESC;
9248	else
9249		sense_format = SSD_TYPE_FIXED;
9250
9251	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9252	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9253	ctsio->kern_sg_entries = 0;
9254
9255	/*
9256	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9257	 * larger than the largest allowed value for the length field in the
9258	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9259	 */
9260	ctsio->residual = 0;
9261	ctsio->kern_data_len = cdb->length;
9262	ctsio->kern_total_len = cdb->length;
9263
9264	ctsio->kern_data_resid = 0;
9265	ctsio->kern_rel_offset = 0;
9266	ctsio->kern_sg_entries = 0;
9267
9268	/*
9269	 * If we don't have a LUN, we don't have any pending sense.
9270	 */
9271	if (lun == NULL)
9272		goto no_sense;
9273
9274	have_error = 0;
9275	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9276	/*
9277	 * Check for pending sense, and then for pending unit attentions.
9278	 * Pending sense gets returned first, then pending unit attentions.
9279	 */
9280	mtx_lock(&lun->ctl_softc->ctl_lock);
9281	if (ctl_is_set(lun->have_ca, initidx)) {
9282		scsi_sense_data_type stored_format;
9283
9284		/*
9285		 * Check to see which sense format was used for the stored
9286		 * sense data.
9287		 */
9288		stored_format = scsi_sense_type(
9289		    &lun->pending_sense[initidx].sense);
9290
9291		/*
9292		 * If the user requested a different sense format than the
9293		 * one we stored, then we need to convert it to the other
9294		 * format.  If we're going from descriptor to fixed format
9295		 * sense data, we may lose things in translation, depending
9296		 * on what options were used.
9297		 *
9298		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9299		 * for some reason we'll just copy it out as-is.
9300		 */
9301		if ((stored_format == SSD_TYPE_FIXED)
9302		 && (sense_format == SSD_TYPE_DESC))
9303			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9304			    &lun->pending_sense[initidx].sense,
9305			    (struct scsi_sense_data_desc *)sense_ptr);
9306		else if ((stored_format == SSD_TYPE_DESC)
9307		      && (sense_format == SSD_TYPE_FIXED))
9308			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9309			    &lun->pending_sense[initidx].sense,
9310			    (struct scsi_sense_data_fixed *)sense_ptr);
9311		else
9312			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9313			       ctl_min(sizeof(*sense_ptr),
9314			       sizeof(lun->pending_sense[initidx].sense)));
9315
9316		ctl_clear_mask(lun->have_ca, initidx);
9317		have_error = 1;
9318	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9319		ctl_ua_type ua_type;
9320
9321		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9322				       sense_ptr, sense_format);
9323		if (ua_type != CTL_UA_NONE) {
9324			have_error = 1;
9325			/* We're reporting this UA, so clear it */
9326			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9327		}
9328	}
9329	mtx_unlock(&lun->ctl_softc->ctl_lock);
9330
9331	/*
9332	 * We already have a pending error, return it.
9333	 */
9334	if (have_error != 0) {
9335		/*
9336		 * We report the SCSI status as OK, since the status of the
9337		 * request sense command itself is OK.
9338		 */
9339		ctsio->scsi_status = SCSI_STATUS_OK;
9340
9341		/*
9342		 * We report 0 for the sense length, because we aren't doing
9343		 * autosense in this case.  We're reporting sense as
9344		 * parameter data.
9345		 */
9346		ctsio->sense_len = 0;
9347
9348		ctsio->be_move_done = ctl_config_move_done;
9349		ctl_datamove((union ctl_io *)ctsio);
9350
9351		return (CTL_RETVAL_COMPLETE);
9352	}
9353
9354no_sense:
9355
9356	/*
9357	 * No sense information to report, so we report that everything is
9358	 * okay.
9359	 */
9360	ctl_set_sense_data(sense_ptr,
9361			   lun,
9362			   sense_format,
9363			   /*current_error*/ 1,
9364			   /*sense_key*/ SSD_KEY_NO_SENSE,
9365			   /*asc*/ 0x00,
9366			   /*ascq*/ 0x00,
9367			   SSD_ELEM_NONE);
9368
9369	ctsio->scsi_status = SCSI_STATUS_OK;
9370
9371	/*
9372	 * We report 0 for the sense length, because we aren't doing
9373	 * autosense in this case.  We're reporting sense as parameter data.
9374	 */
9375	ctsio->sense_len = 0;
9376	ctsio->be_move_done = ctl_config_move_done;
9377	ctl_datamove((union ctl_io *)ctsio);
9378
9379	return (CTL_RETVAL_COMPLETE);
9380}
9381
9382int
9383ctl_tur(struct ctl_scsiio *ctsio)
9384{
9385	struct ctl_lun *lun;
9386
9387	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9388
9389	CTL_DEBUG_PRINT(("ctl_tur\n"));
9390
9391	if (lun == NULL)
9392		return (-EINVAL);
9393
9394	ctsio->scsi_status = SCSI_STATUS_OK;
9395	ctsio->io_hdr.status = CTL_SUCCESS;
9396
9397	ctl_done((union ctl_io *)ctsio);
9398
9399	return (CTL_RETVAL_COMPLETE);
9400}
9401
9402#ifdef notyet
9403static int
9404ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9405{
9406
9407}
9408#endif
9409
9410static int
9411ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9412{
9413	struct scsi_vpd_supported_pages *pages;
9414	int sup_page_size;
9415	struct ctl_lun *lun;
9416
9417	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9418
9419	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9420	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9421	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9422	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9423	ctsio->kern_sg_entries = 0;
9424
9425	if (sup_page_size < alloc_len) {
9426		ctsio->residual = alloc_len - sup_page_size;
9427		ctsio->kern_data_len = sup_page_size;
9428		ctsio->kern_total_len = sup_page_size;
9429	} else {
9430		ctsio->residual = 0;
9431		ctsio->kern_data_len = alloc_len;
9432		ctsio->kern_total_len = alloc_len;
9433	}
9434	ctsio->kern_data_resid = 0;
9435	ctsio->kern_rel_offset = 0;
9436	ctsio->kern_sg_entries = 0;
9437
9438	/*
9439	 * The control device is always connected.  The disk device, on the
9440	 * other hand, may not be online all the time.  Need to change this
9441	 * to figure out whether the disk device is actually online or not.
9442	 */
9443	if (lun != NULL)
9444		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9445				lun->be_lun->lun_type;
9446	else
9447		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9448
9449	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9450	/* Supported VPD pages */
9451	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9452	/* Serial Number */
9453	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9454	/* Device Identification */
9455	pages->page_list[2] = SVPD_DEVICE_ID;
9456	/* Block limits */
9457	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9458	/* Logical Block Provisioning */
9459	pages->page_list[4] = SVPD_LBP;
9460
9461	ctsio->scsi_status = SCSI_STATUS_OK;
9462
9463	ctsio->be_move_done = ctl_config_move_done;
9464	ctl_datamove((union ctl_io *)ctsio);
9465
9466	return (CTL_RETVAL_COMPLETE);
9467}
9468
9469static int
9470ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9471{
9472	struct scsi_vpd_unit_serial_number *sn_ptr;
9473	struct ctl_lun *lun;
9474
9475	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9476
9477	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9478	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9479	ctsio->kern_sg_entries = 0;
9480
9481	if (sizeof(*sn_ptr) < alloc_len) {
9482		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9483		ctsio->kern_data_len = sizeof(*sn_ptr);
9484		ctsio->kern_total_len = sizeof(*sn_ptr);
9485	} else {
9486		ctsio->residual = 0;
9487		ctsio->kern_data_len = alloc_len;
9488		ctsio->kern_total_len = alloc_len;
9489	}
9490	ctsio->kern_data_resid = 0;
9491	ctsio->kern_rel_offset = 0;
9492	ctsio->kern_sg_entries = 0;
9493
9494	/*
9495	 * The control device is always connected.  The disk device, on the
9496	 * other hand, may not be online all the time.  Need to change this
9497	 * to figure out whether the disk device is actually online or not.
9498	 */
9499	if (lun != NULL)
9500		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9501				  lun->be_lun->lun_type;
9502	else
9503		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9504
9505	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9506	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9507	/*
9508	 * If we don't have a LUN, we just leave the serial number as
9509	 * all spaces.
9510	 */
9511	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9512	if (lun != NULL) {
9513		strncpy((char *)sn_ptr->serial_num,
9514			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9515	}
9516	ctsio->scsi_status = SCSI_STATUS_OK;
9517
9518	ctsio->be_move_done = ctl_config_move_done;
9519	ctl_datamove((union ctl_io *)ctsio);
9520
9521	return (CTL_RETVAL_COMPLETE);
9522}
9523
9524
9525static int
9526ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9527{
9528	struct scsi_vpd_device_id *devid_ptr;
9529	struct scsi_vpd_id_descriptor *desc, *desc1;
9530	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9531	struct scsi_vpd_id_t10 *t10id;
9532	struct ctl_softc *ctl_softc;
9533	struct ctl_lun *lun;
9534	struct ctl_frontend *fe;
9535	char *val;
9536	int data_len, devid_len;
9537
9538	ctl_softc = control_softc;
9539
9540	mtx_lock(&ctl_softc->ctl_lock);
9541	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9542	mtx_unlock(&ctl_softc->ctl_lock);
9543
9544	if (fe->devid != NULL)
9545		return ((fe->devid)(ctsio, alloc_len));
9546
9547	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9548
9549	if (lun == NULL) {
9550		devid_len = CTL_DEVID_MIN_LEN;
9551	} else {
9552		devid_len = max(CTL_DEVID_MIN_LEN,
9553		    strnlen(lun->be_lun->device_id, CTL_DEVID_LEN));
9554	}
9555
9556	data_len = sizeof(struct scsi_vpd_device_id) +
9557		sizeof(struct scsi_vpd_id_descriptor) +
9558		sizeof(struct scsi_vpd_id_t10) + devid_len +
9559		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9560		sizeof(struct scsi_vpd_id_descriptor) +
9561		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9562		sizeof(struct scsi_vpd_id_descriptor) +
9563		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9564
9565	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9566	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9567	ctsio->kern_sg_entries = 0;
9568
9569	if (data_len < alloc_len) {
9570		ctsio->residual = alloc_len - data_len;
9571		ctsio->kern_data_len = data_len;
9572		ctsio->kern_total_len = data_len;
9573	} else {
9574		ctsio->residual = 0;
9575		ctsio->kern_data_len = alloc_len;
9576		ctsio->kern_total_len = alloc_len;
9577	}
9578	ctsio->kern_data_resid = 0;
9579	ctsio->kern_rel_offset = 0;
9580	ctsio->kern_sg_entries = 0;
9581
9582	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9583	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9584	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9585		sizeof(struct scsi_vpd_id_t10) + devid_len);
9586	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9587	          CTL_WWPN_LEN);
9588	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9589	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9590
9591	/*
9592	 * The control device is always connected.  The disk device, on the
9593	 * other hand, may not be online all the time.
9594	 */
9595	if (lun != NULL)
9596		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9597				     lun->be_lun->lun_type;
9598	else
9599		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9600
9601	devid_ptr->page_code = SVPD_DEVICE_ID;
9602
9603	scsi_ulto2b(data_len - 4, devid_ptr->length);
9604
9605	mtx_lock(&ctl_softc->ctl_lock);
9606
9607	/*
9608	 * For Fibre channel,
9609	 */
9610	if (fe->port_type == CTL_PORT_FC)
9611	{
9612		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9613				      SVPD_ID_CODESET_ASCII;
9614        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9615		              SVPD_ID_CODESET_BINARY;
9616	}
9617	else
9618	{
9619		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9620				      SVPD_ID_CODESET_ASCII;
9621        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9622		              SVPD_ID_CODESET_BINARY;
9623	}
9624	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9625	mtx_unlock(&ctl_softc->ctl_lock);
9626
9627	/*
9628	 * We're using a LUN association here.  i.e., this device ID is a
9629	 * per-LUN identifier.
9630	 */
9631	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9632	desc->length = sizeof(*t10id) + devid_len;
9633	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9634		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9635	} else {
9636		memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9637		strncpy(t10id->vendor, val,
9638		    min(sizeof(t10id->vendor), strlen(val)));
9639	}
9640
9641	/*
9642	 * desc1 is for the WWPN which is a port asscociation.
9643	 */
9644	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9645	desc1->length = CTL_WWPN_LEN;
9646	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9647	/* For testing just create the WWPN */
9648#if 0
9649	ddb_GetWWNN((char *)desc1->identifier);
9650
9651	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9652	/* This is so Copancontrol will return something sane */
9653	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9654	    ctsio->io_hdr.nexus.targ_port!=8)
9655		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9656	else
9657		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9658#endif
9659
9660	be64enc(desc1->identifier, fe->wwpn);
9661
9662	/*
9663	 * desc2 is for the Relative Target Port(type 4h) identifier
9664	 */
9665	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9666	                 | SVPD_ID_TYPE_RELTARG;
9667	desc2->length = 4;
9668//#if 0
9669	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9670	/* This is so Copancontrol will return something sane */
9671	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9672	    ctsio->io_hdr.nexus.targ_port!=8)
9673		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9674	else
9675	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9676//#endif
9677
9678	/*
9679	 * desc3 is for the Target Port Group(type 5h) identifier
9680	 */
9681	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9682	                 | SVPD_ID_TYPE_TPORTGRP;
9683	desc3->length = 4;
9684	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9685		desc3->identifier[3] = 1;
9686	else
9687		desc3->identifier[3] = 2;
9688
9689	/*
9690	 * If we've actually got a backend, copy the device id from the
9691	 * per-LUN data.  Otherwise, set it to all spaces.
9692	 */
9693	if (lun != NULL) {
9694		/*
9695		 * Copy the backend's LUN ID.
9696		 */
9697		strncpy((char *)t10id->vendor_spec_id,
9698			(char *)lun->be_lun->device_id, devid_len);
9699	} else {
9700		/*
9701		 * No backend, set this to spaces.
9702		 */
9703		memset(t10id->vendor_spec_id, 0x20, devid_len);
9704	}
9705
9706	ctsio->scsi_status = SCSI_STATUS_OK;
9707
9708	ctsio->be_move_done = ctl_config_move_done;
9709	ctl_datamove((union ctl_io *)ctsio);
9710
9711	return (CTL_RETVAL_COMPLETE);
9712}
9713
9714static int
9715ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9716{
9717	struct scsi_vpd_block_limits *bl_ptr;
9718	struct ctl_lun *lun;
9719	int bs;
9720
9721	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9722	bs = lun->be_lun->blocksize;
9723
9724	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9725	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9726	ctsio->kern_sg_entries = 0;
9727
9728	if (sizeof(*bl_ptr) < alloc_len) {
9729		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9730		ctsio->kern_data_len = sizeof(*bl_ptr);
9731		ctsio->kern_total_len = sizeof(*bl_ptr);
9732	} else {
9733		ctsio->residual = 0;
9734		ctsio->kern_data_len = alloc_len;
9735		ctsio->kern_total_len = alloc_len;
9736	}
9737	ctsio->kern_data_resid = 0;
9738	ctsio->kern_rel_offset = 0;
9739	ctsio->kern_sg_entries = 0;
9740
9741	/*
9742	 * The control device is always connected.  The disk device, on the
9743	 * other hand, may not be online all the time.  Need to change this
9744	 * to figure out whether the disk device is actually online or not.
9745	 */
9746	if (lun != NULL)
9747		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9748				  lun->be_lun->lun_type;
9749	else
9750		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9751
9752	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9753	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9754	bl_ptr->max_cmp_write_len = 0xff;
9755	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9756	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9757	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9758		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9759		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9760	}
9761	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9762
9763	ctsio->scsi_status = SCSI_STATUS_OK;
9764	ctsio->be_move_done = ctl_config_move_done;
9765	ctl_datamove((union ctl_io *)ctsio);
9766
9767	return (CTL_RETVAL_COMPLETE);
9768}
9769
9770static int
9771ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9772{
9773	struct scsi_vpd_logical_block_prov *lbp_ptr;
9774	struct ctl_lun *lun;
9775	int bs;
9776
9777	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9778	bs = lun->be_lun->blocksize;
9779
9780	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9781	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9782	ctsio->kern_sg_entries = 0;
9783
9784	if (sizeof(*lbp_ptr) < alloc_len) {
9785		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9786		ctsio->kern_data_len = sizeof(*lbp_ptr);
9787		ctsio->kern_total_len = sizeof(*lbp_ptr);
9788	} else {
9789		ctsio->residual = 0;
9790		ctsio->kern_data_len = alloc_len;
9791		ctsio->kern_total_len = alloc_len;
9792	}
9793	ctsio->kern_data_resid = 0;
9794	ctsio->kern_rel_offset = 0;
9795	ctsio->kern_sg_entries = 0;
9796
9797	/*
9798	 * The control device is always connected.  The disk device, on the
9799	 * other hand, may not be online all the time.  Need to change this
9800	 * to figure out whether the disk device is actually online or not.
9801	 */
9802	if (lun != NULL)
9803		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9804				  lun->be_lun->lun_type;
9805	else
9806		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9807
9808	lbp_ptr->page_code = SVPD_LBP;
9809	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9810		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9811
9812	ctsio->scsi_status = SCSI_STATUS_OK;
9813	ctsio->be_move_done = ctl_config_move_done;
9814	ctl_datamove((union ctl_io *)ctsio);
9815
9816	return (CTL_RETVAL_COMPLETE);
9817}
9818
9819static int
9820ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9821{
9822	struct scsi_inquiry *cdb;
9823	struct ctl_lun *lun;
9824	int alloc_len, retval;
9825
9826	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9827	cdb = (struct scsi_inquiry *)ctsio->cdb;
9828
9829	retval = CTL_RETVAL_COMPLETE;
9830
9831	alloc_len = scsi_2btoul(cdb->length);
9832
9833	switch (cdb->page_code) {
9834	case SVPD_SUPPORTED_PAGES:
9835		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9836		break;
9837	case SVPD_UNIT_SERIAL_NUMBER:
9838		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9839		break;
9840	case SVPD_DEVICE_ID:
9841		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9842		break;
9843	case SVPD_BLOCK_LIMITS:
9844		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9845		break;
9846	case SVPD_LBP:
9847		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9848		break;
9849	default:
9850		ctl_set_invalid_field(ctsio,
9851				      /*sks_valid*/ 1,
9852				      /*command*/ 1,
9853				      /*field*/ 2,
9854				      /*bit_valid*/ 0,
9855				      /*bit*/ 0);
9856		ctl_done((union ctl_io *)ctsio);
9857		retval = CTL_RETVAL_COMPLETE;
9858		break;
9859	}
9860
9861	return (retval);
9862}
9863
9864static int
9865ctl_inquiry_std(struct ctl_scsiio *ctsio)
9866{
9867	struct scsi_inquiry_data *inq_ptr;
9868	struct scsi_inquiry *cdb;
9869	struct ctl_softc *ctl_softc;
9870	struct ctl_lun *lun;
9871	char *val;
9872	uint32_t alloc_len;
9873	int is_fc;
9874
9875	ctl_softc = control_softc;
9876
9877	/*
9878	 * Figure out whether we're talking to a Fibre Channel port or not.
9879	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9880	 * SCSI front ends.
9881	 */
9882	mtx_lock(&ctl_softc->ctl_lock);
9883	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9884	    CTL_PORT_FC)
9885		is_fc = 0;
9886	else
9887		is_fc = 1;
9888	mtx_unlock(&ctl_softc->ctl_lock);
9889
9890	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9891	cdb = (struct scsi_inquiry *)ctsio->cdb;
9892	alloc_len = scsi_2btoul(cdb->length);
9893
9894	/*
9895	 * We malloc the full inquiry data size here and fill it
9896	 * in.  If the user only asks for less, we'll give him
9897	 * that much.
9898	 */
9899	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9900	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9901	ctsio->kern_sg_entries = 0;
9902	ctsio->kern_data_resid = 0;
9903	ctsio->kern_rel_offset = 0;
9904
9905	if (sizeof(*inq_ptr) < alloc_len) {
9906		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9907		ctsio->kern_data_len = sizeof(*inq_ptr);
9908		ctsio->kern_total_len = sizeof(*inq_ptr);
9909	} else {
9910		ctsio->residual = 0;
9911		ctsio->kern_data_len = alloc_len;
9912		ctsio->kern_total_len = alloc_len;
9913	}
9914
9915	/*
9916	 * If we have a LUN configured, report it as connected.  Otherwise,
9917	 * report that it is offline or no device is supported, depending
9918	 * on the value of inquiry_pq_no_lun.
9919	 *
9920	 * According to the spec (SPC-4 r34), the peripheral qualifier
9921	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9922	 *
9923	 * "A peripheral device having the specified peripheral device type
9924	 * is not connected to this logical unit. However, the device
9925	 * server is capable of supporting the specified peripheral device
9926	 * type on this logical unit."
9927	 *
9928	 * According to the same spec, the peripheral qualifier
9929	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9930	 *
9931	 * "The device server is not capable of supporting a peripheral
9932	 * device on this logical unit. For this peripheral qualifier the
9933	 * peripheral device type shall be set to 1Fh. All other peripheral
9934	 * device type values are reserved for this peripheral qualifier."
9935	 *
9936	 * Given the text, it would seem that we probably want to report that
9937	 * the LUN is offline here.  There is no LUN connected, but we can
9938	 * support a LUN at the given LUN number.
9939	 *
9940	 * In the real world, though, it sounds like things are a little
9941	 * different:
9942	 *
9943	 * - Linux, when presented with a LUN with the offline peripheral
9944	 *   qualifier, will create an sg driver instance for it.  So when
9945	 *   you attach it to CTL, you wind up with a ton of sg driver
9946	 *   instances.  (One for every LUN that Linux bothered to probe.)
9947	 *   Linux does this despite the fact that it issues a REPORT LUNs
9948	 *   to LUN 0 to get the inventory of supported LUNs.
9949	 *
9950	 * - There is other anecdotal evidence (from Emulex folks) about
9951	 *   arrays that use the offline peripheral qualifier for LUNs that
9952	 *   are on the "passive" path in an active/passive array.
9953	 *
9954	 * So the solution is provide a hopefully reasonable default
9955	 * (return bad/no LUN) and allow the user to change the behavior
9956	 * with a tunable/sysctl variable.
9957	 */
9958	if (lun != NULL)
9959		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9960				  lun->be_lun->lun_type;
9961	else if (ctl_softc->inquiry_pq_no_lun == 0)
9962		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9963	else
9964		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9965
9966	/* RMB in byte 2 is 0 */
9967	inq_ptr->version = SCSI_REV_SPC3;
9968
9969	/*
9970	 * According to SAM-3, even if a device only supports a single
9971	 * level of LUN addressing, it should still set the HISUP bit:
9972	 *
9973	 * 4.9.1 Logical unit numbers overview
9974	 *
9975	 * All logical unit number formats described in this standard are
9976	 * hierarchical in structure even when only a single level in that
9977	 * hierarchy is used. The HISUP bit shall be set to one in the
9978	 * standard INQUIRY data (see SPC-2) when any logical unit number
9979	 * format described in this standard is used.  Non-hierarchical
9980	 * formats are outside the scope of this standard.
9981	 *
9982	 * Therefore we set the HiSup bit here.
9983	 *
9984	 * The reponse format is 2, per SPC-3.
9985	 */
9986	inq_ptr->response_format = SID_HiSup | 2;
9987
9988	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
9989	CTL_DEBUG_PRINT(("additional_length = %d\n",
9990			 inq_ptr->additional_length));
9991
9992	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
9993	/* 16 bit addressing */
9994	if (is_fc == 0)
9995		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
9996	/* XXX set the SID_MultiP bit here if we're actually going to
9997	   respond on multiple ports */
9998	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
9999
10000	/* 16 bit data bus, synchronous transfers */
10001	/* XXX these flags don't apply for FC */
10002	if (is_fc == 0)
10003		inq_ptr->flags = SID_WBus16 | SID_Sync;
10004	/*
10005	 * XXX KDM do we want to support tagged queueing on the control
10006	 * device at all?
10007	 */
10008	if ((lun == NULL)
10009	 || (lun->be_lun->lun_type != T_PROCESSOR))
10010		inq_ptr->flags |= SID_CmdQue;
10011	/*
10012	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10013	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10014	 * name and 4 bytes for the revision.
10015	 */
10016	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
10017		strcpy(inq_ptr->vendor, CTL_VENDOR);
10018	} else {
10019		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10020		strncpy(inq_ptr->vendor, val,
10021		    min(sizeof(inq_ptr->vendor), strlen(val)));
10022	}
10023	if (lun == NULL) {
10024		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10025	} else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
10026		switch (lun->be_lun->lun_type) {
10027		case T_DIRECT:
10028			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10029			break;
10030		case T_PROCESSOR:
10031			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10032			break;
10033		default:
10034			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10035			break;
10036		}
10037	} else {
10038		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10039		strncpy(inq_ptr->product, val,
10040		    min(sizeof(inq_ptr->product), strlen(val)));
10041	}
10042
10043	/*
10044	 * XXX make this a macro somewhere so it automatically gets
10045	 * incremented when we make changes.
10046	 */
10047	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
10048		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10049	} else {
10050		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10051		strncpy(inq_ptr->revision, val,
10052		    min(sizeof(inq_ptr->revision), strlen(val)));
10053	}
10054
10055	/*
10056	 * For parallel SCSI, we support double transition and single
10057	 * transition clocking.  We also support QAS (Quick Arbitration
10058	 * and Selection) and Information Unit transfers on both the
10059	 * control and array devices.
10060	 */
10061	if (is_fc == 0)
10062		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10063				    SID_SPI_IUS;
10064
10065	/* SAM-3 */
10066	scsi_ulto2b(0x0060, inq_ptr->version1);
10067	/* SPC-3 (no version claimed) XXX should we claim a version? */
10068	scsi_ulto2b(0x0300, inq_ptr->version2);
10069	if (is_fc) {
10070		/* FCP-2 ANSI INCITS.350:2003 */
10071		scsi_ulto2b(0x0917, inq_ptr->version3);
10072	} else {
10073		/* SPI-4 ANSI INCITS.362:200x */
10074		scsi_ulto2b(0x0B56, inq_ptr->version3);
10075	}
10076
10077	if (lun == NULL) {
10078		/* SBC-2 (no version claimed) XXX should we claim a version? */
10079		scsi_ulto2b(0x0320, inq_ptr->version4);
10080	} else {
10081		switch (lun->be_lun->lun_type) {
10082		case T_DIRECT:
10083			/*
10084			 * SBC-2 (no version claimed) XXX should we claim a
10085			 * version?
10086			 */
10087			scsi_ulto2b(0x0320, inq_ptr->version4);
10088			break;
10089		case T_PROCESSOR:
10090		default:
10091			break;
10092		}
10093	}
10094
10095	ctsio->scsi_status = SCSI_STATUS_OK;
10096	if (ctsio->kern_data_len > 0) {
10097		ctsio->be_move_done = ctl_config_move_done;
10098		ctl_datamove((union ctl_io *)ctsio);
10099	} else {
10100		ctsio->io_hdr.status = CTL_SUCCESS;
10101		ctl_done((union ctl_io *)ctsio);
10102	}
10103
10104	return (CTL_RETVAL_COMPLETE);
10105}
10106
10107int
10108ctl_inquiry(struct ctl_scsiio *ctsio)
10109{
10110	struct scsi_inquiry *cdb;
10111	int retval;
10112
10113	cdb = (struct scsi_inquiry *)ctsio->cdb;
10114
10115	retval = 0;
10116
10117	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10118
10119	/*
10120	 * Right now, we don't support the CmdDt inquiry information.
10121	 * This would be nice to support in the future.  When we do
10122	 * support it, we should change this test so that it checks to make
10123	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10124	 */
10125#ifdef notyet
10126	if (((cdb->byte2 & SI_EVPD)
10127	 && (cdb->byte2 & SI_CMDDT)))
10128#endif
10129	if (cdb->byte2 & SI_CMDDT) {
10130		/*
10131		 * Point to the SI_CMDDT bit.  We might change this
10132		 * when we support SI_CMDDT, but since both bits would be
10133		 * "wrong", this should probably just stay as-is then.
10134		 */
10135		ctl_set_invalid_field(ctsio,
10136				      /*sks_valid*/ 1,
10137				      /*command*/ 1,
10138				      /*field*/ 1,
10139				      /*bit_valid*/ 1,
10140				      /*bit*/ 1);
10141		ctl_done((union ctl_io *)ctsio);
10142		return (CTL_RETVAL_COMPLETE);
10143	}
10144	if (cdb->byte2 & SI_EVPD)
10145		retval = ctl_inquiry_evpd(ctsio);
10146#ifdef notyet
10147	else if (cdb->byte2 & SI_CMDDT)
10148		retval = ctl_inquiry_cmddt(ctsio);
10149#endif
10150	else
10151		retval = ctl_inquiry_std(ctsio);
10152
10153	return (retval);
10154}
10155
10156/*
10157 * For known CDB types, parse the LBA and length.
10158 */
10159static int
10160ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10161{
10162	if (io->io_hdr.io_type != CTL_IO_SCSI)
10163		return (1);
10164
10165	switch (io->scsiio.cdb[0]) {
10166	case COMPARE_AND_WRITE: {
10167		struct scsi_compare_and_write *cdb;
10168
10169		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10170
10171		*lba = scsi_8btou64(cdb->addr);
10172		*len = cdb->length;
10173		break;
10174	}
10175	case READ_6:
10176	case WRITE_6: {
10177		struct scsi_rw_6 *cdb;
10178
10179		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10180
10181		*lba = scsi_3btoul(cdb->addr);
10182		/* only 5 bits are valid in the most significant address byte */
10183		*lba &= 0x1fffff;
10184		*len = cdb->length;
10185		break;
10186	}
10187	case READ_10:
10188	case WRITE_10: {
10189		struct scsi_rw_10 *cdb;
10190
10191		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10192
10193		*lba = scsi_4btoul(cdb->addr);
10194		*len = scsi_2btoul(cdb->length);
10195		break;
10196	}
10197	case WRITE_VERIFY_10: {
10198		struct scsi_write_verify_10 *cdb;
10199
10200		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10201
10202		*lba = scsi_4btoul(cdb->addr);
10203		*len = scsi_2btoul(cdb->length);
10204		break;
10205	}
10206	case READ_12:
10207	case WRITE_12: {
10208		struct scsi_rw_12 *cdb;
10209
10210		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10211
10212		*lba = scsi_4btoul(cdb->addr);
10213		*len = scsi_4btoul(cdb->length);
10214		break;
10215	}
10216	case WRITE_VERIFY_12: {
10217		struct scsi_write_verify_12 *cdb;
10218
10219		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10220
10221		*lba = scsi_4btoul(cdb->addr);
10222		*len = scsi_4btoul(cdb->length);
10223		break;
10224	}
10225	case READ_16:
10226	case WRITE_16: {
10227		struct scsi_rw_16 *cdb;
10228
10229		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10230
10231		*lba = scsi_8btou64(cdb->addr);
10232		*len = scsi_4btoul(cdb->length);
10233		break;
10234	}
10235	case WRITE_VERIFY_16: {
10236		struct scsi_write_verify_16 *cdb;
10237
10238		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10239
10240
10241		*lba = scsi_8btou64(cdb->addr);
10242		*len = scsi_4btoul(cdb->length);
10243		break;
10244	}
10245	case WRITE_SAME_10: {
10246		struct scsi_write_same_10 *cdb;
10247
10248		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10249
10250		*lba = scsi_4btoul(cdb->addr);
10251		*len = scsi_2btoul(cdb->length);
10252		break;
10253	}
10254	case WRITE_SAME_16: {
10255		struct scsi_write_same_16 *cdb;
10256
10257		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10258
10259		*lba = scsi_8btou64(cdb->addr);
10260		*len = scsi_4btoul(cdb->length);
10261		break;
10262	}
10263	case VERIFY_10: {
10264		struct scsi_verify_10 *cdb;
10265
10266		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10267
10268		*lba = scsi_4btoul(cdb->addr);
10269		*len = scsi_2btoul(cdb->length);
10270		break;
10271	}
10272	case VERIFY_12: {
10273		struct scsi_verify_12 *cdb;
10274
10275		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10276
10277		*lba = scsi_4btoul(cdb->addr);
10278		*len = scsi_4btoul(cdb->length);
10279		break;
10280	}
10281	case VERIFY_16: {
10282		struct scsi_verify_16 *cdb;
10283
10284		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10285
10286		*lba = scsi_8btou64(cdb->addr);
10287		*len = scsi_4btoul(cdb->length);
10288		break;
10289	}
10290	default:
10291		return (1);
10292		break; /* NOTREACHED */
10293	}
10294
10295	return (0);
10296}
10297
10298static ctl_action
10299ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10300{
10301	uint64_t endlba1, endlba2;
10302
10303	endlba1 = lba1 + len1 - 1;
10304	endlba2 = lba2 + len2 - 1;
10305
10306	if ((endlba1 < lba2)
10307	 || (endlba2 < lba1))
10308		return (CTL_ACTION_PASS);
10309	else
10310		return (CTL_ACTION_BLOCK);
10311}
10312
10313static ctl_action
10314ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10315{
10316	uint64_t lba1, lba2;
10317	uint32_t len1, len2;
10318	int retval;
10319
10320	retval = ctl_get_lba_len(io1, &lba1, &len1);
10321	if (retval != 0)
10322		return (CTL_ACTION_ERROR);
10323
10324	retval = ctl_get_lba_len(io2, &lba2, &len2);
10325	if (retval != 0)
10326		return (CTL_ACTION_ERROR);
10327
10328	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10329}
10330
10331static ctl_action
10332ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10333{
10334	struct ctl_cmd_entry *pending_entry, *ooa_entry;
10335	ctl_serialize_action *serialize_row;
10336
10337	/*
10338	 * The initiator attempted multiple untagged commands at the same
10339	 * time.  Can't do that.
10340	 */
10341	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10342	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10343	 && ((pending_io->io_hdr.nexus.targ_port ==
10344	      ooa_io->io_hdr.nexus.targ_port)
10345	  && (pending_io->io_hdr.nexus.initid.id ==
10346	      ooa_io->io_hdr.nexus.initid.id))
10347	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10348		return (CTL_ACTION_OVERLAP);
10349
10350	/*
10351	 * The initiator attempted to send multiple tagged commands with
10352	 * the same ID.  (It's fine if different initiators have the same
10353	 * tag ID.)
10354	 *
10355	 * Even if all of those conditions are true, we don't kill the I/O
10356	 * if the command ahead of us has been aborted.  We won't end up
10357	 * sending it to the FETD, and it's perfectly legal to resend a
10358	 * command with the same tag number as long as the previous
10359	 * instance of this tag number has been aborted somehow.
10360	 */
10361	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10362	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10363	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10364	 && ((pending_io->io_hdr.nexus.targ_port ==
10365	      ooa_io->io_hdr.nexus.targ_port)
10366	  && (pending_io->io_hdr.nexus.initid.id ==
10367	      ooa_io->io_hdr.nexus.initid.id))
10368	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10369		return (CTL_ACTION_OVERLAP_TAG);
10370
10371	/*
10372	 * If we get a head of queue tag, SAM-3 says that we should
10373	 * immediately execute it.
10374	 *
10375	 * What happens if this command would normally block for some other
10376	 * reason?  e.g. a request sense with a head of queue tag
10377	 * immediately after a write.  Normally that would block, but this
10378	 * will result in its getting executed immediately...
10379	 *
10380	 * We currently return "pass" instead of "skip", so we'll end up
10381	 * going through the rest of the queue to check for overlapped tags.
10382	 *
10383	 * XXX KDM check for other types of blockage first??
10384	 */
10385	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10386		return (CTL_ACTION_PASS);
10387
10388	/*
10389	 * Ordered tags have to block until all items ahead of them
10390	 * have completed.  If we get called with an ordered tag, we always
10391	 * block, if something else is ahead of us in the queue.
10392	 */
10393	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10394		return (CTL_ACTION_BLOCK);
10395
10396	/*
10397	 * Simple tags get blocked until all head of queue and ordered tags
10398	 * ahead of them have completed.  I'm lumping untagged commands in
10399	 * with simple tags here.  XXX KDM is that the right thing to do?
10400	 */
10401	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10402	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10403	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10404	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10405		return (CTL_ACTION_BLOCK);
10406
10407	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
10408	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
10409
10410	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10411
10412	switch (serialize_row[pending_entry->seridx]) {
10413	case CTL_SER_BLOCK:
10414		return (CTL_ACTION_BLOCK);
10415		break; /* NOTREACHED */
10416	case CTL_SER_EXTENT:
10417		return (ctl_extent_check(pending_io, ooa_io));
10418		break; /* NOTREACHED */
10419	case CTL_SER_PASS:
10420		return (CTL_ACTION_PASS);
10421		break; /* NOTREACHED */
10422	case CTL_SER_SKIP:
10423		return (CTL_ACTION_SKIP);
10424		break;
10425	default:
10426		panic("invalid serialization value %d",
10427		      serialize_row[pending_entry->seridx]);
10428		break; /* NOTREACHED */
10429	}
10430
10431	return (CTL_ACTION_ERROR);
10432}
10433
10434/*
10435 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10436 * Assumptions:
10437 * - pending_io is generally either incoming, or on the blocked queue
10438 * - starting I/O is the I/O we want to start the check with.
10439 */
10440static ctl_action
10441ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10442	      union ctl_io *starting_io)
10443{
10444	union ctl_io *ooa_io;
10445	ctl_action action;
10446
10447	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10448
10449	/*
10450	 * Run back along the OOA queue, starting with the current
10451	 * blocked I/O and going through every I/O before it on the
10452	 * queue.  If starting_io is NULL, we'll just end up returning
10453	 * CTL_ACTION_PASS.
10454	 */
10455	for (ooa_io = starting_io; ooa_io != NULL;
10456	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10457	     ooa_links)){
10458
10459		/*
10460		 * This routine just checks to see whether
10461		 * cur_blocked is blocked by ooa_io, which is ahead
10462		 * of it in the queue.  It doesn't queue/dequeue
10463		 * cur_blocked.
10464		 */
10465		action = ctl_check_for_blockage(pending_io, ooa_io);
10466		switch (action) {
10467		case CTL_ACTION_BLOCK:
10468		case CTL_ACTION_OVERLAP:
10469		case CTL_ACTION_OVERLAP_TAG:
10470		case CTL_ACTION_SKIP:
10471		case CTL_ACTION_ERROR:
10472			return (action);
10473			break; /* NOTREACHED */
10474		case CTL_ACTION_PASS:
10475			break;
10476		default:
10477			panic("invalid action %d", action);
10478			break;  /* NOTREACHED */
10479		}
10480	}
10481
10482	return (CTL_ACTION_PASS);
10483}
10484
10485/*
10486 * Assumptions:
10487 * - An I/O has just completed, and has been removed from the per-LUN OOA
10488 *   queue, so some items on the blocked queue may now be unblocked.
10489 */
10490static int
10491ctl_check_blocked(struct ctl_lun *lun)
10492{
10493	union ctl_io *cur_blocked, *next_blocked;
10494
10495	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10496
10497	/*
10498	 * Run forward from the head of the blocked queue, checking each
10499	 * entry against the I/Os prior to it on the OOA queue to see if
10500	 * there is still any blockage.
10501	 *
10502	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10503	 * with our removing a variable on it while it is traversing the
10504	 * list.
10505	 */
10506	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10507	     cur_blocked != NULL; cur_blocked = next_blocked) {
10508		union ctl_io *prev_ooa;
10509		ctl_action action;
10510
10511		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10512							  blocked_links);
10513
10514		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10515						      ctl_ooaq, ooa_links);
10516
10517		/*
10518		 * If cur_blocked happens to be the first item in the OOA
10519		 * queue now, prev_ooa will be NULL, and the action
10520		 * returned will just be CTL_ACTION_PASS.
10521		 */
10522		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10523
10524		switch (action) {
10525		case CTL_ACTION_BLOCK:
10526			/* Nothing to do here, still blocked */
10527			break;
10528		case CTL_ACTION_OVERLAP:
10529		case CTL_ACTION_OVERLAP_TAG:
10530			/*
10531			 * This shouldn't happen!  In theory we've already
10532			 * checked this command for overlap...
10533			 */
10534			break;
10535		case CTL_ACTION_PASS:
10536		case CTL_ACTION_SKIP: {
10537			struct ctl_softc *softc;
10538			struct ctl_cmd_entry *entry;
10539			uint32_t initidx;
10540			uint8_t opcode;
10541			int isc_retval;
10542
10543			/*
10544			 * The skip case shouldn't happen, this transaction
10545			 * should have never made it onto the blocked queue.
10546			 */
10547			/*
10548			 * This I/O is no longer blocked, we can remove it
10549			 * from the blocked queue.  Since this is a TAILQ
10550			 * (doubly linked list), we can do O(1) removals
10551			 * from any place on the list.
10552			 */
10553			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10554				     blocked_links);
10555			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10556
10557			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10558				/*
10559				 * Need to send IO back to original side to
10560				 * run
10561				 */
10562				union ctl_ha_msg msg_info;
10563
10564				msg_info.hdr.original_sc =
10565					cur_blocked->io_hdr.original_sc;
10566				msg_info.hdr.serializing_sc = cur_blocked;
10567				msg_info.hdr.msg_type = CTL_MSG_R2R;
10568				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10569				     &msg_info, sizeof(msg_info), 0)) >
10570				     CTL_HA_STATUS_SUCCESS) {
10571					printf("CTL:Check Blocked error from "
10572					       "ctl_ha_msg_send %d\n",
10573					       isc_retval);
10574				}
10575				break;
10576			}
10577			opcode = cur_blocked->scsiio.cdb[0];
10578			entry = &ctl_cmd_table[opcode];
10579			softc = control_softc;
10580
10581			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10582
10583			/*
10584			 * Check this I/O for LUN state changes that may
10585			 * have happened while this command was blocked.
10586			 * The LUN state may have been changed by a command
10587			 * ahead of us in the queue, so we need to re-check
10588			 * for any states that can be caused by SCSI
10589			 * commands.
10590			 */
10591			if (ctl_scsiio_lun_check(softc, lun, entry,
10592						 &cur_blocked->scsiio) == 0) {
10593				cur_blocked->io_hdr.flags |=
10594				                      CTL_FLAG_IS_WAS_ON_RTR;
10595				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
10596						   &cur_blocked->io_hdr, links);
10597				/*
10598				 * In the non CTL_DONE_THREAD case, we need
10599				 * to wake up the work thread here.  When
10600				 * we're processing completed requests from
10601				 * the work thread context, we'll pop back
10602				 * around and end up pulling things off the
10603				 * RtR queue.  When we aren't processing
10604				 * things from the work thread context,
10605				 * though, we won't ever check the RtR queue.
10606				 * So we need to wake up the thread to clear
10607				 * things off the queue.  Otherwise this
10608				 * transaction will just sit on the RtR queue
10609				 * until a new I/O comes in.  (Which may or
10610				 * may not happen...)
10611				 */
10612#ifndef CTL_DONE_THREAD
10613				ctl_wakeup_thread();
10614#endif
10615			} else
10616				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
10617			break;
10618		}
10619		default:
10620			/*
10621			 * This probably shouldn't happen -- we shouldn't
10622			 * get CTL_ACTION_ERROR, or anything else.
10623			 */
10624			break;
10625		}
10626	}
10627
10628	return (CTL_RETVAL_COMPLETE);
10629}
10630
10631/*
10632 * This routine (with one exception) checks LUN flags that can be set by
10633 * commands ahead of us in the OOA queue.  These flags have to be checked
10634 * when a command initially comes in, and when we pull a command off the
10635 * blocked queue and are preparing to execute it.  The reason we have to
10636 * check these flags for commands on the blocked queue is that the LUN
10637 * state may have been changed by a command ahead of us while we're on the
10638 * blocked queue.
10639 *
10640 * Ordering is somewhat important with these checks, so please pay
10641 * careful attention to the placement of any new checks.
10642 */
10643static int
10644ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10645		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10646{
10647	int retval;
10648
10649	retval = 0;
10650
10651	/*
10652	 * If this shelf is a secondary shelf controller, we have to reject
10653	 * any media access commands.
10654	 */
10655#if 0
10656	/* No longer needed for HA */
10657	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10658	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10659		ctl_set_lun_standby(ctsio);
10660		retval = 1;
10661		goto bailout;
10662	}
10663#endif
10664
10665	/*
10666	 * Check for a reservation conflict.  If this command isn't allowed
10667	 * even on reserved LUNs, and if this initiator isn't the one who
10668	 * reserved us, reject the command with a reservation conflict.
10669	 */
10670	if ((lun->flags & CTL_LUN_RESERVED)
10671	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10672		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10673		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10674		 || (ctsio->io_hdr.nexus.targ_target.id !=
10675		     lun->rsv_nexus.targ_target.id)) {
10676			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10677			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10678			retval = 1;
10679			goto bailout;
10680		}
10681	}
10682
10683	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10684	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10685		uint32_t residx;
10686
10687		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10688		/*
10689		 * if we aren't registered or it's a res holder type
10690		 * reservation and this isn't the res holder then set a
10691		 * conflict.
10692		 * NOTE: Commands which might be allowed on write exclusive
10693		 * type reservations are checked in the particular command
10694		 * for a conflict. Read and SSU are the only ones.
10695		 */
10696		if (!lun->per_res[residx].registered
10697		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10698			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10699			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10700			retval = 1;
10701			goto bailout;
10702		}
10703
10704	}
10705
10706	if ((lun->flags & CTL_LUN_OFFLINE)
10707	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10708		ctl_set_lun_not_ready(ctsio);
10709		retval = 1;
10710		goto bailout;
10711	}
10712
10713	/*
10714	 * If the LUN is stopped, see if this particular command is allowed
10715	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10716	 */
10717	if ((lun->flags & CTL_LUN_STOPPED)
10718	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10719		/* "Logical unit not ready, initializing cmd. required" */
10720		ctl_set_lun_stopped(ctsio);
10721		retval = 1;
10722		goto bailout;
10723	}
10724
10725	if ((lun->flags & CTL_LUN_INOPERABLE)
10726	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10727		/* "Medium format corrupted" */
10728		ctl_set_medium_format_corrupted(ctsio);
10729		retval = 1;
10730		goto bailout;
10731	}
10732
10733bailout:
10734	return (retval);
10735
10736}
10737
10738static void
10739ctl_failover_io(union ctl_io *io, int have_lock)
10740{
10741	ctl_set_busy(&io->scsiio);
10742	ctl_done_lock(io, have_lock);
10743}
10744
10745static void
10746ctl_failover(void)
10747{
10748	struct ctl_lun *lun;
10749	struct ctl_softc *ctl_softc;
10750	union ctl_io *next_io, *pending_io;
10751	union ctl_io *io;
10752	int lun_idx;
10753	int i;
10754
10755	ctl_softc = control_softc;
10756
10757	mtx_lock(&ctl_softc->ctl_lock);
10758	/*
10759	 * Remove any cmds from the other SC from the rtr queue.  These
10760	 * will obviously only be for LUNs for which we're the primary.
10761	 * We can't send status or get/send data for these commands.
10762	 * Since they haven't been executed yet, we can just remove them.
10763	 * We'll either abort them or delete them below, depending on
10764	 * which HA mode we're in.
10765	 */
10766	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10767	     io != NULL; io = next_io) {
10768		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10769		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10770			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10771				      ctl_io_hdr, links);
10772	}
10773
10774	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10775		lun = ctl_softc->ctl_luns[lun_idx];
10776		if (lun==NULL)
10777			continue;
10778
10779		/*
10780		 * Processor LUNs are primary on both sides.
10781		 * XXX will this always be true?
10782		 */
10783		if (lun->be_lun->lun_type == T_PROCESSOR)
10784			continue;
10785
10786		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10787		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10788			printf("FAILOVER: primary lun %d\n", lun_idx);
10789		        /*
10790			 * Remove all commands from the other SC. First from the
10791			 * blocked queue then from the ooa queue. Once we have
10792			 * removed them. Call ctl_check_blocked to see if there
10793			 * is anything that can run.
10794			 */
10795			for (io = (union ctl_io *)TAILQ_FIRST(
10796			     &lun->blocked_queue); io != NULL; io = next_io) {
10797
10798		        	next_io = (union ctl_io *)TAILQ_NEXT(
10799				    &io->io_hdr, blocked_links);
10800
10801				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10802					TAILQ_REMOVE(&lun->blocked_queue,
10803						     &io->io_hdr,blocked_links);
10804					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10805					TAILQ_REMOVE(&lun->ooa_queue,
10806						     &io->io_hdr, ooa_links);
10807
10808					ctl_free_io(io);
10809				}
10810			}
10811
10812			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10813	     		     io != NULL; io = next_io) {
10814
10815		        	next_io = (union ctl_io *)TAILQ_NEXT(
10816				    &io->io_hdr, ooa_links);
10817
10818				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10819
10820					TAILQ_REMOVE(&lun->ooa_queue,
10821						&io->io_hdr,
10822					     	ooa_links);
10823
10824					ctl_free_io(io);
10825				}
10826			}
10827			ctl_check_blocked(lun);
10828		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10829			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10830
10831			printf("FAILOVER: primary lun %d\n", lun_idx);
10832			/*
10833			 * Abort all commands from the other SC.  We can't
10834			 * send status back for them now.  These should get
10835			 * cleaned up when they are completed or come out
10836			 * for a datamove operation.
10837			 */
10838			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10839	     		     io != NULL; io = next_io) {
10840		        	next_io = (union ctl_io *)TAILQ_NEXT(
10841					&io->io_hdr, ooa_links);
10842
10843				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10844					io->io_hdr.flags |= CTL_FLAG_ABORT;
10845			}
10846		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10847			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10848
10849			printf("FAILOVER: secondary lun %d\n", lun_idx);
10850
10851			lun->flags |= CTL_LUN_PRIMARY_SC;
10852
10853			/*
10854			 * We send all I/O that was sent to this controller
10855			 * and redirected to the other side back with
10856			 * busy status, and have the initiator retry it.
10857			 * Figuring out how much data has been transferred,
10858			 * etc. and picking up where we left off would be
10859			 * very tricky.
10860			 *
10861			 * XXX KDM need to remove I/O from the blocked
10862			 * queue as well!
10863			 */
10864			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10865			     &lun->ooa_queue); pending_io != NULL;
10866			     pending_io = next_io) {
10867
10868				next_io =  (union ctl_io *)TAILQ_NEXT(
10869					&pending_io->io_hdr, ooa_links);
10870
10871				pending_io->io_hdr.flags &=
10872					~CTL_FLAG_SENT_2OTHER_SC;
10873
10874				if (pending_io->io_hdr.flags &
10875				    CTL_FLAG_IO_ACTIVE) {
10876					pending_io->io_hdr.flags |=
10877						CTL_FLAG_FAILOVER;
10878				} else {
10879					ctl_set_busy(&pending_io->scsiio);
10880					ctl_done_lock(pending_io,
10881						      /*have_lock*/1);
10882				}
10883			}
10884
10885			/*
10886			 * Build Unit Attention
10887			 */
10888			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10889				lun->pending_sense[i].ua_pending |=
10890				                     CTL_UA_ASYM_ACC_CHANGE;
10891			}
10892		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10893			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10894			printf("FAILOVER: secondary lun %d\n", lun_idx);
10895			/*
10896			 * if the first io on the OOA is not on the RtR queue
10897			 * add it.
10898			 */
10899			lun->flags |= CTL_LUN_PRIMARY_SC;
10900
10901			pending_io = (union ctl_io *)TAILQ_FIRST(
10902			    &lun->ooa_queue);
10903			if (pending_io==NULL) {
10904				printf("Nothing on OOA queue\n");
10905				continue;
10906			}
10907
10908			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10909			if ((pending_io->io_hdr.flags &
10910			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10911				pending_io->io_hdr.flags |=
10912				    CTL_FLAG_IS_WAS_ON_RTR;
10913				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10914						   &pending_io->io_hdr, links);
10915			}
10916#if 0
10917			else
10918			{
10919				printf("Tag 0x%04x is running\n",
10920				      pending_io->scsiio.tag_num);
10921			}
10922#endif
10923
10924			next_io = (union ctl_io *)TAILQ_NEXT(
10925			    &pending_io->io_hdr, ooa_links);
10926			for (pending_io=next_io; pending_io != NULL;
10927			     pending_io = next_io) {
10928				pending_io->io_hdr.flags &=
10929				    ~CTL_FLAG_SENT_2OTHER_SC;
10930				next_io = (union ctl_io *)TAILQ_NEXT(
10931					&pending_io->io_hdr, ooa_links);
10932				if (pending_io->io_hdr.flags &
10933				    CTL_FLAG_IS_WAS_ON_RTR) {
10934#if 0
10935				        printf("Tag 0x%04x is running\n",
10936				      		pending_io->scsiio.tag_num);
10937#endif
10938					continue;
10939				}
10940
10941				switch (ctl_check_ooa(lun, pending_io,
10942			            (union ctl_io *)TAILQ_PREV(
10943				    &pending_io->io_hdr, ctl_ooaq,
10944				    ooa_links))) {
10945
10946				case CTL_ACTION_BLOCK:
10947					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10948							  &pending_io->io_hdr,
10949							  blocked_links);
10950					pending_io->io_hdr.flags |=
10951					    CTL_FLAG_BLOCKED;
10952					break;
10953				case CTL_ACTION_PASS:
10954				case CTL_ACTION_SKIP:
10955					pending_io->io_hdr.flags |=
10956					    CTL_FLAG_IS_WAS_ON_RTR;
10957					STAILQ_INSERT_TAIL(
10958					    &ctl_softc->rtr_queue,
10959					    &pending_io->io_hdr, links);
10960					break;
10961				case CTL_ACTION_OVERLAP:
10962					ctl_set_overlapped_cmd(
10963					    (struct ctl_scsiio *)pending_io);
10964					ctl_done_lock(pending_io,
10965						      /*have_lock*/ 1);
10966					break;
10967				case CTL_ACTION_OVERLAP_TAG:
10968					ctl_set_overlapped_tag(
10969					    (struct ctl_scsiio *)pending_io,
10970					    pending_io->scsiio.tag_num & 0xff);
10971					ctl_done_lock(pending_io,
10972						      /*have_lock*/ 1);
10973					break;
10974				case CTL_ACTION_ERROR:
10975				default:
10976					ctl_set_internal_failure(
10977						(struct ctl_scsiio *)pending_io,
10978						0,  // sks_valid
10979						0); //retry count
10980					ctl_done_lock(pending_io,
10981						      /*have_lock*/ 1);
10982					break;
10983				}
10984			}
10985
10986			/*
10987			 * Build Unit Attention
10988			 */
10989			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10990				lun->pending_sense[i].ua_pending |=
10991				                     CTL_UA_ASYM_ACC_CHANGE;
10992			}
10993		} else {
10994			panic("Unhandled HA mode failover, LUN flags = %#x, "
10995			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
10996		}
10997	}
10998	ctl_pause_rtr = 0;
10999	mtx_unlock(&ctl_softc->ctl_lock);
11000}
11001
11002static int
11003ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11004{
11005	struct ctl_lun *lun;
11006	struct ctl_cmd_entry *entry;
11007	uint8_t opcode;
11008	uint32_t initidx, targ_lun;
11009	int retval;
11010
11011	retval = 0;
11012
11013	lun = NULL;
11014
11015	opcode = ctsio->cdb[0];
11016
11017	mtx_lock(&ctl_softc->ctl_lock);
11018
11019	targ_lun = ctsio->io_hdr.nexus.targ_lun;
11020	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
11021		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
11022	if ((targ_lun < CTL_MAX_LUNS)
11023	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11024		lun = ctl_softc->ctl_luns[targ_lun];
11025		/*
11026		 * If the LUN is invalid, pretend that it doesn't exist.
11027		 * It will go away as soon as all pending I/O has been
11028		 * completed.
11029		 */
11030		if (lun->flags & CTL_LUN_DISABLED) {
11031			lun = NULL;
11032		} else {
11033			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11034			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11035				lun->be_lun;
11036			if (lun->be_lun->lun_type == T_PROCESSOR) {
11037				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11038			}
11039		}
11040	} else {
11041		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11042		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11043	}
11044
11045	entry = &ctl_cmd_table[opcode];
11046
11047	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11048	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11049
11050	/*
11051	 * Check to see whether we can send this command to LUNs that don't
11052	 * exist.  This should pretty much only be the case for inquiry
11053	 * and request sense.  Further checks, below, really require having
11054	 * a LUN, so we can't really check the command anymore.  Just put
11055	 * it on the rtr queue.
11056	 */
11057	if (lun == NULL) {
11058		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11059			goto queue_rtr;
11060
11061		ctl_set_unsupported_lun(ctsio);
11062		mtx_unlock(&ctl_softc->ctl_lock);
11063		ctl_done((union ctl_io *)ctsio);
11064		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11065		goto bailout;
11066	} else {
11067		/*
11068		 * Every I/O goes into the OOA queue for a particular LUN, and
11069		 * stays there until completion.
11070		 */
11071		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11072
11073		/*
11074		 * Make sure we support this particular command on this LUN.
11075		 * e.g., we don't support writes to the control LUN.
11076		 */
11077		switch (lun->be_lun->lun_type) {
11078		case T_PROCESSOR:
11079		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11080			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11081			      == 0)) {
11082				ctl_set_invalid_opcode(ctsio);
11083				mtx_unlock(&ctl_softc->ctl_lock);
11084				ctl_done((union ctl_io *)ctsio);
11085				goto bailout;
11086			}
11087			break;
11088		case T_DIRECT:
11089			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11090			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11091			      == 0)){
11092				ctl_set_invalid_opcode(ctsio);
11093				mtx_unlock(&ctl_softc->ctl_lock);
11094				ctl_done((union ctl_io *)ctsio);
11095				goto bailout;
11096			}
11097			break;
11098		default:
11099			printf("Unsupported CTL LUN type %d\n",
11100			       lun->be_lun->lun_type);
11101			panic("Unsupported CTL LUN type %d\n",
11102			      lun->be_lun->lun_type);
11103			break; /* NOTREACHED */
11104		}
11105	}
11106
11107	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11108
11109	/*
11110	 * If we've got a request sense, it'll clear the contingent
11111	 * allegiance condition.  Otherwise, if we have a CA condition for
11112	 * this initiator, clear it, because it sent down a command other
11113	 * than request sense.
11114	 */
11115	if ((opcode != REQUEST_SENSE)
11116	 && (ctl_is_set(lun->have_ca, initidx)))
11117		ctl_clear_mask(lun->have_ca, initidx);
11118
11119	/*
11120	 * If the command has this flag set, it handles its own unit
11121	 * attention reporting, we shouldn't do anything.  Otherwise we
11122	 * check for any pending unit attentions, and send them back to the
11123	 * initiator.  We only do this when a command initially comes in,
11124	 * not when we pull it off the blocked queue.
11125	 *
11126	 * According to SAM-3, section 5.3.2, the order that things get
11127	 * presented back to the host is basically unit attentions caused
11128	 * by some sort of reset event, busy status, reservation conflicts
11129	 * or task set full, and finally any other status.
11130	 *
11131	 * One issue here is that some of the unit attentions we report
11132	 * don't fall into the "reset" category (e.g. "reported luns data
11133	 * has changed").  So reporting it here, before the reservation
11134	 * check, may be technically wrong.  I guess the only thing to do
11135	 * would be to check for and report the reset events here, and then
11136	 * check for the other unit attention types after we check for a
11137	 * reservation conflict.
11138	 *
11139	 * XXX KDM need to fix this
11140	 */
11141	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11142		ctl_ua_type ua_type;
11143
11144		ua_type = lun->pending_sense[initidx].ua_pending;
11145		if (ua_type != CTL_UA_NONE) {
11146			scsi_sense_data_type sense_format;
11147
11148			if (lun != NULL)
11149				sense_format = (lun->flags &
11150				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11151				    SSD_TYPE_FIXED;
11152			else
11153				sense_format = SSD_TYPE_FIXED;
11154
11155			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11156					       sense_format);
11157			if (ua_type != CTL_UA_NONE) {
11158				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11159				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11160						       CTL_AUTOSENSE;
11161				ctsio->sense_len = SSD_FULL_SIZE;
11162				lun->pending_sense[initidx].ua_pending &=
11163					~ua_type;
11164				mtx_unlock(&ctl_softc->ctl_lock);
11165				ctl_done((union ctl_io *)ctsio);
11166				goto bailout;
11167			}
11168		}
11169	}
11170
11171
11172	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11173		mtx_unlock(&ctl_softc->ctl_lock);
11174		ctl_done((union ctl_io *)ctsio);
11175		goto bailout;
11176	}
11177
11178	/*
11179	 * XXX CHD this is where we want to send IO to other side if
11180	 * this LUN is secondary on this SC. We will need to make a copy
11181	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11182	 * the copy we send as FROM_OTHER.
11183	 * We also need to stuff the address of the original IO so we can
11184	 * find it easily. Something similar will need be done on the other
11185	 * side so when we are done we can find the copy.
11186	 */
11187	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11188		union ctl_ha_msg msg_info;
11189		int isc_retval;
11190
11191		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11192
11193		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11194		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11195#if 0
11196		printf("1. ctsio %p\n", ctsio);
11197#endif
11198		msg_info.hdr.serializing_sc = NULL;
11199		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11200		msg_info.scsi.tag_num = ctsio->tag_num;
11201		msg_info.scsi.tag_type = ctsio->tag_type;
11202		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11203
11204		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11205
11206		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11207		    (void *)&msg_info, sizeof(msg_info), 0)) >
11208		    CTL_HA_STATUS_SUCCESS) {
11209			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11210			       isc_retval);
11211			printf("CTL:opcode is %x\n",opcode);
11212		} else {
11213#if 0
11214			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11215#endif
11216		}
11217
11218		/*
11219		 * XXX KDM this I/O is off the incoming queue, but hasn't
11220		 * been inserted on any other queue.  We may need to come
11221		 * up with a holding queue while we wait for serialization
11222		 * so that we have an idea of what we're waiting for from
11223		 * the other side.
11224		 */
11225		goto bailout_unlock;
11226	}
11227
11228	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11229			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11230			      ctl_ooaq, ooa_links))) {
11231	case CTL_ACTION_BLOCK:
11232		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11233		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11234				  blocked_links);
11235		goto bailout_unlock;
11236		break; /* NOTREACHED */
11237	case CTL_ACTION_PASS:
11238	case CTL_ACTION_SKIP:
11239		goto queue_rtr;
11240		break; /* NOTREACHED */
11241	case CTL_ACTION_OVERLAP:
11242		ctl_set_overlapped_cmd(ctsio);
11243		mtx_unlock(&ctl_softc->ctl_lock);
11244		ctl_done((union ctl_io *)ctsio);
11245		goto bailout;
11246		break; /* NOTREACHED */
11247	case CTL_ACTION_OVERLAP_TAG:
11248		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11249		mtx_unlock(&ctl_softc->ctl_lock);
11250		ctl_done((union ctl_io *)ctsio);
11251		goto bailout;
11252		break; /* NOTREACHED */
11253	case CTL_ACTION_ERROR:
11254	default:
11255		ctl_set_internal_failure(ctsio,
11256					 /*sks_valid*/ 0,
11257					 /*retry_count*/ 0);
11258		mtx_unlock(&ctl_softc->ctl_lock);
11259		ctl_done((union ctl_io *)ctsio);
11260		goto bailout;
11261		break; /* NOTREACHED */
11262	}
11263
11264	goto bailout_unlock;
11265
11266queue_rtr:
11267	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11268	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
11269
11270bailout_unlock:
11271	mtx_unlock(&ctl_softc->ctl_lock);
11272
11273bailout:
11274	return (retval);
11275}
11276
11277static int
11278ctl_scsiio(struct ctl_scsiio *ctsio)
11279{
11280	int retval;
11281	struct ctl_cmd_entry *entry;
11282
11283	retval = CTL_RETVAL_COMPLETE;
11284
11285	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11286
11287	entry = &ctl_cmd_table[ctsio->cdb[0]];
11288
11289	/*
11290	 * If this I/O has been aborted, just send it straight to
11291	 * ctl_done() without executing it.
11292	 */
11293	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11294		ctl_done((union ctl_io *)ctsio);
11295		goto bailout;
11296	}
11297
11298	/*
11299	 * All the checks should have been handled by ctl_scsiio_precheck().
11300	 * We should be clear now to just execute the I/O.
11301	 */
11302	retval = entry->execute(ctsio);
11303
11304bailout:
11305	return (retval);
11306}
11307
11308/*
11309 * Since we only implement one target right now, a bus reset simply resets
11310 * our single target.
11311 */
11312static int
11313ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11314{
11315	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11316}
11317
11318static int
11319ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11320		 ctl_ua_type ua_type)
11321{
11322	struct ctl_lun *lun;
11323	int retval;
11324
11325	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11326		union ctl_ha_msg msg_info;
11327
11328		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11329		msg_info.hdr.nexus = io->io_hdr.nexus;
11330		if (ua_type==CTL_UA_TARG_RESET)
11331			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11332		else
11333			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11334		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11335		msg_info.hdr.original_sc = NULL;
11336		msg_info.hdr.serializing_sc = NULL;
11337		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11338		    (void *)&msg_info, sizeof(msg_info), 0)) {
11339		}
11340	}
11341	retval = 0;
11342
11343	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11344		retval += ctl_lun_reset(lun, io, ua_type);
11345
11346	return (retval);
11347}
11348
11349/*
11350 * The LUN should always be set.  The I/O is optional, and is used to
11351 * distinguish between I/Os sent by this initiator, and by other
11352 * initiators.  We set unit attention for initiators other than this one.
11353 * SAM-3 is vague on this point.  It does say that a unit attention should
11354 * be established for other initiators when a LUN is reset (see section
11355 * 5.7.3), but it doesn't specifically say that the unit attention should
11356 * be established for this particular initiator when a LUN is reset.  Here
11357 * is the relevant text, from SAM-3 rev 8:
11358 *
11359 * 5.7.2 When a SCSI initiator port aborts its own tasks
11360 *
11361 * When a SCSI initiator port causes its own task(s) to be aborted, no
11362 * notification that the task(s) have been aborted shall be returned to
11363 * the SCSI initiator port other than the completion response for the
11364 * command or task management function action that caused the task(s) to
11365 * be aborted and notification(s) associated with related effects of the
11366 * action (e.g., a reset unit attention condition).
11367 *
11368 * XXX KDM for now, we're setting unit attention for all initiators.
11369 */
11370static int
11371ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11372{
11373	union ctl_io *xio;
11374#if 0
11375	uint32_t initindex;
11376#endif
11377	int i;
11378
11379	/*
11380	 * Run through the OOA queue and abort each I/O.
11381	 */
11382#if 0
11383	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11384#endif
11385	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11386	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11387		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11388	}
11389
11390	/*
11391	 * This version sets unit attention for every
11392	 */
11393#if 0
11394	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11395	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11396		if (initindex == i)
11397			continue;
11398		lun->pending_sense[i].ua_pending |= ua_type;
11399	}
11400#endif
11401
11402	/*
11403	 * A reset (any kind, really) clears reservations established with
11404	 * RESERVE/RELEASE.  It does not clear reservations established
11405	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11406	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11407	 * reservations made with the RESERVE/RELEASE commands, because
11408	 * those commands are obsolete in SPC-3.
11409	 */
11410	lun->flags &= ~CTL_LUN_RESERVED;
11411
11412	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11413		ctl_clear_mask(lun->have_ca, i);
11414		lun->pending_sense[i].ua_pending |= ua_type;
11415	}
11416
11417	return (0);
11418}
11419
11420static int
11421ctl_abort_task(union ctl_io *io)
11422{
11423	union ctl_io *xio;
11424	struct ctl_lun *lun;
11425	struct ctl_softc *ctl_softc;
11426#if 0
11427	struct sbuf sb;
11428	char printbuf[128];
11429#endif
11430	int found;
11431	uint32_t targ_lun;
11432
11433	ctl_softc = control_softc;
11434	found = 0;
11435
11436	/*
11437	 * Look up the LUN.
11438	 */
11439	targ_lun = io->io_hdr.nexus.targ_lun;
11440	if (io->io_hdr.nexus.lun_map_fn != NULL)
11441		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11442	if ((targ_lun < CTL_MAX_LUNS)
11443	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11444		lun = ctl_softc->ctl_luns[targ_lun];
11445	else
11446		goto bailout;
11447
11448#if 0
11449	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11450	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11451#endif
11452
11453	/*
11454	 * Run through the OOA queue and attempt to find the given I/O.
11455	 * The target port, initiator ID, tag type and tag number have to
11456	 * match the values that we got from the initiator.  If we have an
11457	 * untagged command to abort, simply abort the first untagged command
11458	 * we come to.  We only allow one untagged command at a time of course.
11459	 */
11460#if 0
11461	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11462#endif
11463	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11464	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11465#if 0
11466		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11467
11468		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11469			    lun->lun, xio->scsiio.tag_num,
11470			    xio->scsiio.tag_type,
11471			    (xio->io_hdr.blocked_links.tqe_prev
11472			    == NULL) ? "" : " BLOCKED",
11473			    (xio->io_hdr.flags &
11474			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11475			    (xio->io_hdr.flags &
11476			    CTL_FLAG_ABORT) ? " ABORT" : "",
11477			    (xio->io_hdr.flags &
11478			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11479		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11480		sbuf_finish(&sb);
11481		printf("%s\n", sbuf_data(&sb));
11482#endif
11483
11484		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11485		 && (xio->io_hdr.nexus.initid.id ==
11486		     io->io_hdr.nexus.initid.id)) {
11487			/*
11488			 * If the abort says that the task is untagged, the
11489			 * task in the queue must be untagged.  Otherwise,
11490			 * we just check to see whether the tag numbers
11491			 * match.  This is because the QLogic firmware
11492			 * doesn't pass back the tag type in an abort
11493			 * request.
11494			 */
11495#if 0
11496			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11497			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11498			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11499#endif
11500			/*
11501			 * XXX KDM we've got problems with FC, because it
11502			 * doesn't send down a tag type with aborts.  So we
11503			 * can only really go by the tag number...
11504			 * This may cause problems with parallel SCSI.
11505			 * Need to figure that out!!
11506			 */
11507			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11508				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11509				found = 1;
11510				if ((io->io_hdr.flags &
11511				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11512				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11513					union ctl_ha_msg msg_info;
11514
11515					io->io_hdr.flags |=
11516					                CTL_FLAG_SENT_2OTHER_SC;
11517					msg_info.hdr.nexus = io->io_hdr.nexus;
11518					msg_info.task.task_action =
11519						CTL_TASK_ABORT_TASK;
11520					msg_info.task.tag_num =
11521						io->taskio.tag_num;
11522					msg_info.task.tag_type =
11523						io->taskio.tag_type;
11524					msg_info.hdr.msg_type =
11525						CTL_MSG_MANAGE_TASKS;
11526					msg_info.hdr.original_sc = NULL;
11527					msg_info.hdr.serializing_sc = NULL;
11528#if 0
11529					printf("Sent Abort to other side\n");
11530#endif
11531					if (CTL_HA_STATUS_SUCCESS !=
11532					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11533		    				(void *)&msg_info,
11534						sizeof(msg_info), 0)) {
11535					}
11536				}
11537#if 0
11538				printf("ctl_abort_task: found I/O to abort\n");
11539#endif
11540				break;
11541			}
11542		}
11543	}
11544
11545bailout:
11546
11547	if (found == 0) {
11548		/*
11549		 * This isn't really an error.  It's entirely possible for
11550		 * the abort and command completion to cross on the wire.
11551		 * This is more of an informative/diagnostic error.
11552		 */
11553#if 0
11554		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11555		       "%d:%d:%d:%d tag %d type %d\n",
11556		       io->io_hdr.nexus.initid.id,
11557		       io->io_hdr.nexus.targ_port,
11558		       io->io_hdr.nexus.targ_target.id,
11559		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11560		       io->taskio.tag_type);
11561#endif
11562		return (1);
11563	} else
11564		return (0);
11565}
11566
11567/*
11568 * This routine cannot block!  It must be callable from an interrupt
11569 * handler as well as from the work thread.
11570 */
11571static void
11572ctl_run_task_queue(struct ctl_softc *ctl_softc)
11573{
11574	union ctl_io *io, *next_io;
11575
11576	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
11577
11578	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
11579
11580	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
11581	     io != NULL; io = next_io) {
11582		int retval;
11583		const char *task_desc;
11584
11585		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11586
11587		retval = 0;
11588
11589		switch (io->io_hdr.io_type) {
11590		case CTL_IO_TASK: {
11591			task_desc = ctl_scsi_task_string(&io->taskio);
11592			if (task_desc != NULL) {
11593#ifdef NEEDTOPORT
11594				csevent_log(CSC_CTL | CSC_SHELF_SW |
11595					    CTL_TASK_REPORT,
11596					    csevent_LogType_Trace,
11597					    csevent_Severity_Information,
11598					    csevent_AlertLevel_Green,
11599					    csevent_FRU_Firmware,
11600					    csevent_FRU_Unknown,
11601					    "CTL: received task: %s",task_desc);
11602#endif
11603			} else {
11604#ifdef NEEDTOPORT
11605				csevent_log(CSC_CTL | CSC_SHELF_SW |
11606					    CTL_TASK_REPORT,
11607					    csevent_LogType_Trace,
11608					    csevent_Severity_Information,
11609					    csevent_AlertLevel_Green,
11610					    csevent_FRU_Firmware,
11611					    csevent_FRU_Unknown,
11612					    "CTL: received unknown task "
11613					    "type: %d (%#x)",
11614					    io->taskio.task_action,
11615					    io->taskio.task_action);
11616#endif
11617			}
11618			switch (io->taskio.task_action) {
11619			case CTL_TASK_ABORT_TASK:
11620				retval = ctl_abort_task(io);
11621				break;
11622			case CTL_TASK_ABORT_TASK_SET:
11623				break;
11624			case CTL_TASK_CLEAR_ACA:
11625				break;
11626			case CTL_TASK_CLEAR_TASK_SET:
11627				break;
11628			case CTL_TASK_LUN_RESET: {
11629				struct ctl_lun *lun;
11630				uint32_t targ_lun;
11631				int retval;
11632
11633				targ_lun = io->io_hdr.nexus.targ_lun;
11634				if (io->io_hdr.nexus.lun_map_fn != NULL)
11635					targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11636
11637				if ((targ_lun < CTL_MAX_LUNS)
11638				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11639					lun = ctl_softc->ctl_luns[targ_lun];
11640				else {
11641					retval = 1;
11642					break;
11643				}
11644
11645				if (!(io->io_hdr.flags &
11646				    CTL_FLAG_FROM_OTHER_SC)) {
11647					union ctl_ha_msg msg_info;
11648
11649					io->io_hdr.flags |=
11650						CTL_FLAG_SENT_2OTHER_SC;
11651					msg_info.hdr.msg_type =
11652						CTL_MSG_MANAGE_TASKS;
11653					msg_info.hdr.nexus = io->io_hdr.nexus;
11654					msg_info.task.task_action =
11655						CTL_TASK_LUN_RESET;
11656					msg_info.hdr.original_sc = NULL;
11657					msg_info.hdr.serializing_sc = NULL;
11658					if (CTL_HA_STATUS_SUCCESS !=
11659					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11660					    (void *)&msg_info,
11661					    sizeof(msg_info), 0)) {
11662					}
11663				}
11664
11665				retval = ctl_lun_reset(lun, io,
11666						       CTL_UA_LUN_RESET);
11667				break;
11668			}
11669			case CTL_TASK_TARGET_RESET:
11670				retval = ctl_target_reset(ctl_softc, io,
11671							  CTL_UA_TARG_RESET);
11672				break;
11673			case CTL_TASK_BUS_RESET:
11674				retval = ctl_bus_reset(ctl_softc, io);
11675				break;
11676			case CTL_TASK_PORT_LOGIN:
11677				break;
11678			case CTL_TASK_PORT_LOGOUT:
11679				break;
11680			default:
11681				printf("ctl_run_task_queue: got unknown task "
11682				       "management event %d\n",
11683				       io->taskio.task_action);
11684				break;
11685			}
11686			if (retval == 0)
11687				io->io_hdr.status = CTL_SUCCESS;
11688			else
11689				io->io_hdr.status = CTL_ERROR;
11690
11691			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11692				      ctl_io_hdr, links);
11693			/*
11694			 * This will queue this I/O to the done queue, but the
11695			 * work thread won't be able to process it until we
11696			 * return and the lock is released.
11697			 */
11698			ctl_done_lock(io, /*have_lock*/ 1);
11699			break;
11700		}
11701		default: {
11702
11703			printf("%s: invalid I/O type %d msg %d cdb %x"
11704			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11705			       __func__, io->io_hdr.io_type,
11706			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11707			       (uintmax_t)io->io_hdr.nexus.initid.id,
11708			       io->io_hdr.nexus.targ_port,
11709			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11710			       io->io_hdr.nexus.targ_lun /* XXX */,
11711			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11712			       io->taskio.tag_num : io->scsiio.tag_num);
11713			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11714				      ctl_io_hdr, links);
11715			ctl_free_io(io);
11716			break;
11717		}
11718		}
11719	}
11720
11721	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11722}
11723
11724/*
11725 * For HA operation.  Handle commands that come in from the other
11726 * controller.
11727 */
11728static void
11729ctl_handle_isc(union ctl_io *io)
11730{
11731	int free_io;
11732	struct ctl_lun *lun;
11733	struct ctl_softc *ctl_softc;
11734	uint32_t targ_lun;
11735
11736	ctl_softc = control_softc;
11737
11738	targ_lun = io->io_hdr.nexus.targ_lun;
11739	if (io->io_hdr.nexus.lun_map_fn != NULL)
11740		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11741	lun = ctl_softc->ctl_luns[targ_lun];
11742
11743	switch (io->io_hdr.msg_type) {
11744	case CTL_MSG_SERIALIZE:
11745		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11746						     /*have_lock*/ 0);
11747		break;
11748	case CTL_MSG_R2R: {
11749		uint8_t opcode;
11750		struct ctl_cmd_entry *entry;
11751
11752		/*
11753		 * This is only used in SER_ONLY mode.
11754		 */
11755		free_io = 0;
11756		opcode = io->scsiio.cdb[0];
11757		entry = &ctl_cmd_table[opcode];
11758		mtx_lock(&ctl_softc->ctl_lock);
11759		if (ctl_scsiio_lun_check(ctl_softc, lun,
11760		    entry, (struct ctl_scsiio *)io) != 0) {
11761			ctl_done_lock(io, /*have_lock*/ 1);
11762			mtx_unlock(&ctl_softc->ctl_lock);
11763			break;
11764		}
11765		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11766		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11767				   &io->io_hdr, links);
11768		mtx_unlock(&ctl_softc->ctl_lock);
11769		break;
11770	}
11771	case CTL_MSG_FINISH_IO:
11772		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11773			free_io = 0;
11774			ctl_done_lock(io, /*have_lock*/ 0);
11775		} else {
11776			free_io = 1;
11777			mtx_lock(&ctl_softc->ctl_lock);
11778			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11779				     ooa_links);
11780			STAILQ_REMOVE(&ctl_softc->task_queue,
11781				      &io->io_hdr, ctl_io_hdr, links);
11782			ctl_check_blocked(lun);
11783			mtx_unlock(&ctl_softc->ctl_lock);
11784		}
11785		break;
11786	case CTL_MSG_PERS_ACTION:
11787		ctl_hndl_per_res_out_on_other_sc(
11788			(union ctl_ha_msg *)&io->presio.pr_msg);
11789		free_io = 1;
11790		break;
11791	case CTL_MSG_BAD_JUJU:
11792		free_io = 0;
11793		ctl_done_lock(io, /*have_lock*/ 0);
11794		break;
11795	case CTL_MSG_DATAMOVE:
11796		/* Only used in XFER mode */
11797		free_io = 0;
11798		ctl_datamove_remote(io);
11799		break;
11800	case CTL_MSG_DATAMOVE_DONE:
11801		/* Only used in XFER mode */
11802		free_io = 0;
11803		io->scsiio.be_move_done(io);
11804		break;
11805	default:
11806		free_io = 1;
11807		printf("%s: Invalid message type %d\n",
11808		       __func__, io->io_hdr.msg_type);
11809		break;
11810	}
11811	if (free_io)
11812		ctl_free_io(io);
11813
11814}
11815
11816
11817/*
11818 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11819 * there is no match.
11820 */
11821static ctl_lun_error_pattern
11822ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11823{
11824	struct ctl_cmd_entry *entry;
11825	ctl_lun_error_pattern filtered_pattern, pattern;
11826	uint8_t opcode;
11827
11828	pattern = desc->error_pattern;
11829
11830	/*
11831	 * XXX KDM we need more data passed into this function to match a
11832	 * custom pattern, and we actually need to implement custom pattern
11833	 * matching.
11834	 */
11835	if (pattern & CTL_LUN_PAT_CMD)
11836		return (CTL_LUN_PAT_CMD);
11837
11838	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11839		return (CTL_LUN_PAT_ANY);
11840
11841	opcode = ctsio->cdb[0];
11842	entry = &ctl_cmd_table[opcode];
11843
11844	filtered_pattern = entry->pattern & pattern;
11845
11846	/*
11847	 * If the user requested specific flags in the pattern (e.g.
11848	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11849	 * flags.
11850	 *
11851	 * If the user did not specify any flags, it doesn't matter whether
11852	 * or not the command supports the flags.
11853	 */
11854	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11855	     (pattern & ~CTL_LUN_PAT_MASK))
11856		return (CTL_LUN_PAT_NONE);
11857
11858	/*
11859	 * If the user asked for a range check, see if the requested LBA
11860	 * range overlaps with this command's LBA range.
11861	 */
11862	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11863		uint64_t lba1;
11864		uint32_t len1;
11865		ctl_action action;
11866		int retval;
11867
11868		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11869		if (retval != 0)
11870			return (CTL_LUN_PAT_NONE);
11871
11872		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11873					      desc->lba_range.len);
11874		/*
11875		 * A "pass" means that the LBA ranges don't overlap, so
11876		 * this doesn't match the user's range criteria.
11877		 */
11878		if (action == CTL_ACTION_PASS)
11879			return (CTL_LUN_PAT_NONE);
11880	}
11881
11882	return (filtered_pattern);
11883}
11884
11885static void
11886ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11887{
11888	struct ctl_error_desc *desc, *desc2;
11889
11890	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
11891
11892	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11893		ctl_lun_error_pattern pattern;
11894		/*
11895		 * Check to see whether this particular command matches
11896		 * the pattern in the descriptor.
11897		 */
11898		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11899		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11900			continue;
11901
11902		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11903		case CTL_LUN_INJ_ABORTED:
11904			ctl_set_aborted(&io->scsiio);
11905			break;
11906		case CTL_LUN_INJ_MEDIUM_ERR:
11907			ctl_set_medium_error(&io->scsiio);
11908			break;
11909		case CTL_LUN_INJ_UA:
11910			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11911			 * OCCURRED */
11912			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11913			break;
11914		case CTL_LUN_INJ_CUSTOM:
11915			/*
11916			 * We're assuming the user knows what he is doing.
11917			 * Just copy the sense information without doing
11918			 * checks.
11919			 */
11920			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11921			      ctl_min(sizeof(desc->custom_sense),
11922				      sizeof(io->scsiio.sense_data)));
11923			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11924			io->scsiio.sense_len = SSD_FULL_SIZE;
11925			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11926			break;
11927		case CTL_LUN_INJ_NONE:
11928		default:
11929			/*
11930			 * If this is an error injection type we don't know
11931			 * about, clear the continuous flag (if it is set)
11932			 * so it will get deleted below.
11933			 */
11934			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11935			break;
11936		}
11937		/*
11938		 * By default, each error injection action is a one-shot
11939		 */
11940		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11941			continue;
11942
11943		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11944
11945		free(desc, M_CTL);
11946	}
11947}
11948
11949#ifdef CTL_IO_DELAY
11950static void
11951ctl_datamove_timer_wakeup(void *arg)
11952{
11953	union ctl_io *io;
11954
11955	io = (union ctl_io *)arg;
11956
11957	ctl_datamove(io);
11958}
11959#endif /* CTL_IO_DELAY */
11960
11961void
11962ctl_datamove(union ctl_io *io)
11963{
11964	void (*fe_datamove)(union ctl_io *io);
11965
11966	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11967
11968	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11969
11970#ifdef CTL_TIME_IO
11971	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11972		char str[256];
11973		char path_str[64];
11974		struct sbuf sb;
11975
11976		ctl_scsi_path_string(io, path_str, sizeof(path_str));
11977		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
11978
11979		sbuf_cat(&sb, path_str);
11980		switch (io->io_hdr.io_type) {
11981		case CTL_IO_SCSI:
11982			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
11983			sbuf_printf(&sb, "\n");
11984			sbuf_cat(&sb, path_str);
11985			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
11986				    io->scsiio.tag_num, io->scsiio.tag_type);
11987			break;
11988		case CTL_IO_TASK:
11989			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
11990				    "Tag Type: %d\n", io->taskio.task_action,
11991				    io->taskio.tag_num, io->taskio.tag_type);
11992			break;
11993		default:
11994			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11995			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
11996			break;
11997		}
11998		sbuf_cat(&sb, path_str);
11999		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12000			    (intmax_t)time_uptime - io->io_hdr.start_time);
12001		sbuf_finish(&sb);
12002		printf("%s", sbuf_data(&sb));
12003	}
12004#endif /* CTL_TIME_IO */
12005
12006	mtx_lock(&control_softc->ctl_lock);
12007#ifdef CTL_IO_DELAY
12008	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12009		struct ctl_lun *lun;
12010
12011		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12012
12013		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12014	} else {
12015		struct ctl_lun *lun;
12016
12017		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12018		if ((lun != NULL)
12019		 && (lun->delay_info.datamove_delay > 0)) {
12020			struct callout *callout;
12021
12022			callout = (struct callout *)&io->io_hdr.timer_bytes;
12023			callout_init(callout, /*mpsafe*/ 1);
12024			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12025			callout_reset(callout,
12026				      lun->delay_info.datamove_delay * hz,
12027				      ctl_datamove_timer_wakeup, io);
12028			if (lun->delay_info.datamove_type ==
12029			    CTL_DELAY_TYPE_ONESHOT)
12030				lun->delay_info.datamove_delay = 0;
12031			mtx_unlock(&control_softc->ctl_lock);
12032			return;
12033		}
12034	}
12035#endif
12036	/*
12037	 * If we have any pending task management commands, process them
12038	 * first.  This is necessary to eliminate a race condition with the
12039	 * FETD:
12040	 *
12041	 * - FETD submits a task management command, like an abort.
12042	 * - Back end calls fe_datamove() to move the data for the aborted
12043	 *   command.  The FETD can't really accept it, but if it did, it
12044	 *   would end up transmitting data for a command that the initiator
12045	 *   told us to abort.
12046	 *
12047	 * We close the race by processing all pending task management
12048	 * commands here (we can't block!), and then check this I/O to see
12049	 * if it has been aborted.  If so, return it to the back end with
12050	 * bad status, so the back end can say return an error to the back end
12051	 * and then when the back end returns an error, we can return the
12052	 * aborted command to the FETD, so it can clean up its resources.
12053	 */
12054	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
12055		ctl_run_task_queue(control_softc);
12056
12057	/*
12058	 * This command has been aborted.  Set the port status, so we fail
12059	 * the data move.
12060	 */
12061	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12062		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12063		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12064		       io->io_hdr.nexus.targ_port,
12065		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12066		       io->io_hdr.nexus.targ_lun);
12067		io->io_hdr.status = CTL_CMD_ABORTED;
12068		io->io_hdr.port_status = 31337;
12069		mtx_unlock(&control_softc->ctl_lock);
12070		/*
12071		 * Note that the backend, in this case, will get the
12072		 * callback in its context.  In other cases it may get
12073		 * called in the frontend's interrupt thread context.
12074		 */
12075		io->scsiio.be_move_done(io);
12076		return;
12077	}
12078
12079	/*
12080	 * If we're in XFER mode and this I/O is from the other shelf
12081	 * controller, we need to send the DMA to the other side to
12082	 * actually transfer the data to/from the host.  In serialize only
12083	 * mode the transfer happens below CTL and ctl_datamove() is only
12084	 * called on the machine that originally received the I/O.
12085	 */
12086	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12087	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12088		union ctl_ha_msg msg;
12089		uint32_t sg_entries_sent;
12090		int do_sg_copy;
12091		int i;
12092
12093		memset(&msg, 0, sizeof(msg));
12094		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12095		msg.hdr.original_sc = io->io_hdr.original_sc;
12096		msg.hdr.serializing_sc = io;
12097		msg.hdr.nexus = io->io_hdr.nexus;
12098		msg.dt.flags = io->io_hdr.flags;
12099		/*
12100		 * We convert everything into a S/G list here.  We can't
12101		 * pass by reference, only by value between controllers.
12102		 * So we can't pass a pointer to the S/G list, only as many
12103		 * S/G entries as we can fit in here.  If it's possible for
12104		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12105		 * then we need to break this up into multiple transfers.
12106		 */
12107		if (io->scsiio.kern_sg_entries == 0) {
12108			msg.dt.kern_sg_entries = 1;
12109			/*
12110			 * If this is in cached memory, flush the cache
12111			 * before we send the DMA request to the other
12112			 * controller.  We want to do this in either the
12113			 * read or the write case.  The read case is
12114			 * straightforward.  In the write case, we want to
12115			 * make sure nothing is in the local cache that
12116			 * could overwrite the DMAed data.
12117			 */
12118			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12119				/*
12120				 * XXX KDM use bus_dmamap_sync() here.
12121				 */
12122			}
12123
12124			/*
12125			 * Convert to a physical address if this is a
12126			 * virtual address.
12127			 */
12128			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12129				msg.dt.sg_list[0].addr =
12130					io->scsiio.kern_data_ptr;
12131			} else {
12132				/*
12133				 * XXX KDM use busdma here!
12134				 */
12135#if 0
12136				msg.dt.sg_list[0].addr = (void *)
12137					vtophys(io->scsiio.kern_data_ptr);
12138#endif
12139			}
12140
12141			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12142			do_sg_copy = 0;
12143		} else {
12144			struct ctl_sg_entry *sgl;
12145
12146			do_sg_copy = 1;
12147			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12148			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12149			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12150				/*
12151				 * XXX KDM use bus_dmamap_sync() here.
12152				 */
12153			}
12154		}
12155
12156		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12157		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12158		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12159		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12160		msg.dt.sg_sequence = 0;
12161
12162		/*
12163		 * Loop until we've sent all of the S/G entries.  On the
12164		 * other end, we'll recompose these S/G entries into one
12165		 * contiguous list before passing it to the
12166		 */
12167		for (sg_entries_sent = 0; sg_entries_sent <
12168		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12169			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12170				sizeof(msg.dt.sg_list[0])),
12171				msg.dt.kern_sg_entries - sg_entries_sent);
12172
12173			if (do_sg_copy != 0) {
12174				struct ctl_sg_entry *sgl;
12175				int j;
12176
12177				sgl = (struct ctl_sg_entry *)
12178					io->scsiio.kern_data_ptr;
12179				/*
12180				 * If this is in cached memory, flush the cache
12181				 * before we send the DMA request to the other
12182				 * controller.  We want to do this in either
12183				 * the * read or the write case.  The read
12184				 * case is straightforward.  In the write
12185				 * case, we want to make sure nothing is
12186				 * in the local cache that could overwrite
12187				 * the DMAed data.
12188				 */
12189
12190				for (i = sg_entries_sent, j = 0;
12191				     i < msg.dt.cur_sg_entries; i++, j++) {
12192					if ((io->io_hdr.flags &
12193					     CTL_FLAG_NO_DATASYNC) == 0) {
12194						/*
12195						 * XXX KDM use bus_dmamap_sync()
12196						 */
12197					}
12198					if ((io->io_hdr.flags &
12199					     CTL_FLAG_BUS_ADDR) == 0) {
12200						/*
12201						 * XXX KDM use busdma.
12202						 */
12203#if 0
12204						msg.dt.sg_list[j].addr =(void *)
12205						       vtophys(sgl[i].addr);
12206#endif
12207					} else {
12208						msg.dt.sg_list[j].addr =
12209							sgl[i].addr;
12210					}
12211					msg.dt.sg_list[j].len = sgl[i].len;
12212				}
12213			}
12214
12215			sg_entries_sent += msg.dt.cur_sg_entries;
12216			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12217				msg.dt.sg_last = 1;
12218			else
12219				msg.dt.sg_last = 0;
12220
12221			/*
12222			 * XXX KDM drop and reacquire the lock here?
12223			 */
12224			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12225			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12226				/*
12227				 * XXX do something here.
12228				 */
12229			}
12230
12231			msg.dt.sent_sg_entries = sg_entries_sent;
12232		}
12233		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12234		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12235			ctl_failover_io(io, /*have_lock*/ 1);
12236
12237	} else {
12238
12239		/*
12240		 * Lookup the fe_datamove() function for this particular
12241		 * front end.
12242		 */
12243		fe_datamove =
12244		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12245		mtx_unlock(&control_softc->ctl_lock);
12246
12247		fe_datamove(io);
12248	}
12249}
12250
12251static void
12252ctl_send_datamove_done(union ctl_io *io, int have_lock)
12253{
12254	union ctl_ha_msg msg;
12255	int isc_status;
12256
12257	memset(&msg, 0, sizeof(msg));
12258
12259	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12260	msg.hdr.original_sc = io;
12261	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12262	msg.hdr.nexus = io->io_hdr.nexus;
12263	msg.hdr.status = io->io_hdr.status;
12264	msg.scsi.tag_num = io->scsiio.tag_num;
12265	msg.scsi.tag_type = io->scsiio.tag_type;
12266	msg.scsi.scsi_status = io->scsiio.scsi_status;
12267	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12268	       sizeof(io->scsiio.sense_data));
12269	msg.scsi.sense_len = io->scsiio.sense_len;
12270	msg.scsi.sense_residual = io->scsiio.sense_residual;
12271	msg.scsi.fetd_status = io->io_hdr.port_status;
12272	msg.scsi.residual = io->scsiio.residual;
12273	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12274
12275	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12276		ctl_failover_io(io, /*have_lock*/ have_lock);
12277		return;
12278	}
12279
12280	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12281	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12282		/* XXX do something if this fails */
12283	}
12284
12285}
12286
12287/*
12288 * The DMA to the remote side is done, now we need to tell the other side
12289 * we're done so it can continue with its data movement.
12290 */
12291static void
12292ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12293{
12294	union ctl_io *io;
12295
12296	io = rq->context;
12297
12298	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12299		printf("%s: ISC DMA write failed with error %d", __func__,
12300		       rq->ret);
12301		ctl_set_internal_failure(&io->scsiio,
12302					 /*sks_valid*/ 1,
12303					 /*retry_count*/ rq->ret);
12304	}
12305
12306	ctl_dt_req_free(rq);
12307
12308	/*
12309	 * In this case, we had to malloc the memory locally.  Free it.
12310	 */
12311	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12312		int i;
12313		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12314			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12315	}
12316	/*
12317	 * The data is in local and remote memory, so now we need to send
12318	 * status (good or back) back to the other side.
12319	 */
12320	ctl_send_datamove_done(io, /*have_lock*/ 0);
12321}
12322
12323/*
12324 * We've moved the data from the host/controller into local memory.  Now we
12325 * need to push it over to the remote controller's memory.
12326 */
12327static int
12328ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12329{
12330	int retval;
12331
12332	retval = 0;
12333
12334	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12335					  ctl_datamove_remote_write_cb);
12336
12337	return (retval);
12338}
12339
12340static void
12341ctl_datamove_remote_write(union ctl_io *io)
12342{
12343	int retval;
12344	void (*fe_datamove)(union ctl_io *io);
12345
12346	/*
12347	 * - Get the data from the host/HBA into local memory.
12348	 * - DMA memory from the local controller to the remote controller.
12349	 * - Send status back to the remote controller.
12350	 */
12351
12352	retval = ctl_datamove_remote_sgl_setup(io);
12353	if (retval != 0)
12354		return;
12355
12356	/* Switch the pointer over so the FETD knows what to do */
12357	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12358
12359	/*
12360	 * Use a custom move done callback, since we need to send completion
12361	 * back to the other controller, not to the backend on this side.
12362	 */
12363	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12364
12365	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12366
12367	fe_datamove(io);
12368
12369	return;
12370
12371}
12372
12373static int
12374ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12375{
12376#if 0
12377	char str[256];
12378	char path_str[64];
12379	struct sbuf sb;
12380#endif
12381
12382	/*
12383	 * In this case, we had to malloc the memory locally.  Free it.
12384	 */
12385	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12386		int i;
12387		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12388			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12389	}
12390
12391#if 0
12392	scsi_path_string(io, path_str, sizeof(path_str));
12393	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12394	sbuf_cat(&sb, path_str);
12395	scsi_command_string(&io->scsiio, NULL, &sb);
12396	sbuf_printf(&sb, "\n");
12397	sbuf_cat(&sb, path_str);
12398	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12399		    io->scsiio.tag_num, io->scsiio.tag_type);
12400	sbuf_cat(&sb, path_str);
12401	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12402		    io->io_hdr.flags, io->io_hdr.status);
12403	sbuf_finish(&sb);
12404	printk("%s", sbuf_data(&sb));
12405#endif
12406
12407
12408	/*
12409	 * The read is done, now we need to send status (good or bad) back
12410	 * to the other side.
12411	 */
12412	ctl_send_datamove_done(io, /*have_lock*/ 0);
12413
12414	return (0);
12415}
12416
12417static void
12418ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12419{
12420	union ctl_io *io;
12421	void (*fe_datamove)(union ctl_io *io);
12422
12423	io = rq->context;
12424
12425	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12426		printf("%s: ISC DMA read failed with error %d", __func__,
12427		       rq->ret);
12428		ctl_set_internal_failure(&io->scsiio,
12429					 /*sks_valid*/ 1,
12430					 /*retry_count*/ rq->ret);
12431	}
12432
12433	ctl_dt_req_free(rq);
12434
12435	/* Switch the pointer over so the FETD knows what to do */
12436	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12437
12438	/*
12439	 * Use a custom move done callback, since we need to send completion
12440	 * back to the other controller, not to the backend on this side.
12441	 */
12442	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12443
12444	/* XXX KDM add checks like the ones in ctl_datamove? */
12445
12446	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12447
12448	fe_datamove(io);
12449}
12450
12451static int
12452ctl_datamove_remote_sgl_setup(union ctl_io *io)
12453{
12454	struct ctl_sg_entry *local_sglist, *remote_sglist;
12455	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12456	struct ctl_softc *softc;
12457	int retval;
12458	int i;
12459
12460	retval = 0;
12461	softc = control_softc;
12462
12463	local_sglist = io->io_hdr.local_sglist;
12464	local_dma_sglist = io->io_hdr.local_dma_sglist;
12465	remote_sglist = io->io_hdr.remote_sglist;
12466	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12467
12468	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12469		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12470			local_sglist[i].len = remote_sglist[i].len;
12471
12472			/*
12473			 * XXX Detect the situation where the RS-level I/O
12474			 * redirector on the other side has already read the
12475			 * data off of the AOR RS on this side, and
12476			 * transferred it to remote (mirror) memory on the
12477			 * other side.  Since we already have the data in
12478			 * memory here, we just need to use it.
12479			 *
12480			 * XXX KDM this can probably be removed once we
12481			 * get the cache device code in and take the
12482			 * current AOR implementation out.
12483			 */
12484#ifdef NEEDTOPORT
12485			if ((remote_sglist[i].addr >=
12486			     (void *)vtophys(softc->mirr->addr))
12487			 && (remote_sglist[i].addr <
12488			     ((void *)vtophys(softc->mirr->addr) +
12489			     CacheMirrorOffset))) {
12490				local_sglist[i].addr = remote_sglist[i].addr -
12491					CacheMirrorOffset;
12492				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12493				     CTL_FLAG_DATA_IN)
12494					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12495			} else {
12496				local_sglist[i].addr = remote_sglist[i].addr +
12497					CacheMirrorOffset;
12498			}
12499#endif
12500#if 0
12501			printf("%s: local %p, remote %p, len %d\n",
12502			       __func__, local_sglist[i].addr,
12503			       remote_sglist[i].addr, local_sglist[i].len);
12504#endif
12505		}
12506	} else {
12507		uint32_t len_to_go;
12508
12509		/*
12510		 * In this case, we don't have automatically allocated
12511		 * memory for this I/O on this controller.  This typically
12512		 * happens with internal CTL I/O -- e.g. inquiry, mode
12513		 * sense, etc.  Anything coming from RAIDCore will have
12514		 * a mirror area available.
12515		 */
12516		len_to_go = io->scsiio.kern_data_len;
12517
12518		/*
12519		 * Clear the no datasync flag, we have to use malloced
12520		 * buffers.
12521		 */
12522		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12523
12524		/*
12525		 * The difficult thing here is that the size of the various
12526		 * S/G segments may be different than the size from the
12527		 * remote controller.  That'll make it harder when DMAing
12528		 * the data back to the other side.
12529		 */
12530		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12531		     sizeof(io->io_hdr.remote_sglist[0])) &&
12532		     (len_to_go > 0); i++) {
12533			local_sglist[i].len = ctl_min(len_to_go, 131072);
12534			CTL_SIZE_8B(local_dma_sglist[i].len,
12535				    local_sglist[i].len);
12536			local_sglist[i].addr =
12537				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12538
12539			local_dma_sglist[i].addr = local_sglist[i].addr;
12540
12541			if (local_sglist[i].addr == NULL) {
12542				int j;
12543
12544				printf("malloc failed for %zd bytes!",
12545				       local_dma_sglist[i].len);
12546				for (j = 0; j < i; j++) {
12547					free(local_sglist[j].addr, M_CTL);
12548				}
12549				ctl_set_internal_failure(&io->scsiio,
12550							 /*sks_valid*/ 1,
12551							 /*retry_count*/ 4857);
12552				retval = 1;
12553				goto bailout_error;
12554
12555			}
12556			/* XXX KDM do we need a sync here? */
12557
12558			len_to_go -= local_sglist[i].len;
12559		}
12560		/*
12561		 * Reset the number of S/G entries accordingly.  The
12562		 * original number of S/G entries is available in
12563		 * rem_sg_entries.
12564		 */
12565		io->scsiio.kern_sg_entries = i;
12566
12567#if 0
12568		printf("%s: kern_sg_entries = %d\n", __func__,
12569		       io->scsiio.kern_sg_entries);
12570		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12571			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12572			       local_sglist[i].addr, local_sglist[i].len,
12573			       local_dma_sglist[i].len);
12574#endif
12575	}
12576
12577
12578	return (retval);
12579
12580bailout_error:
12581
12582	ctl_send_datamove_done(io, /*have_lock*/ 0);
12583
12584	return (retval);
12585}
12586
12587static int
12588ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12589			 ctl_ha_dt_cb callback)
12590{
12591	struct ctl_ha_dt_req *rq;
12592	struct ctl_sg_entry *remote_sglist, *local_sglist;
12593	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12594	uint32_t local_used, remote_used, total_used;
12595	int retval;
12596	int i, j;
12597
12598	retval = 0;
12599
12600	rq = ctl_dt_req_alloc();
12601
12602	/*
12603	 * If we failed to allocate the request, and if the DMA didn't fail
12604	 * anyway, set busy status.  This is just a resource allocation
12605	 * failure.
12606	 */
12607	if ((rq == NULL)
12608	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12609		ctl_set_busy(&io->scsiio);
12610
12611	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12612
12613		if (rq != NULL)
12614			ctl_dt_req_free(rq);
12615
12616		/*
12617		 * The data move failed.  We need to return status back
12618		 * to the other controller.  No point in trying to DMA
12619		 * data to the remote controller.
12620		 */
12621
12622		ctl_send_datamove_done(io, /*have_lock*/ 0);
12623
12624		retval = 1;
12625
12626		goto bailout;
12627	}
12628
12629	local_sglist = io->io_hdr.local_sglist;
12630	local_dma_sglist = io->io_hdr.local_dma_sglist;
12631	remote_sglist = io->io_hdr.remote_sglist;
12632	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12633	local_used = 0;
12634	remote_used = 0;
12635	total_used = 0;
12636
12637	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12638		rq->ret = CTL_HA_STATUS_SUCCESS;
12639		rq->context = io;
12640		callback(rq);
12641		goto bailout;
12642	}
12643
12644	/*
12645	 * Pull/push the data over the wire from/to the other controller.
12646	 * This takes into account the possibility that the local and
12647	 * remote sglists may not be identical in terms of the size of
12648	 * the elements and the number of elements.
12649	 *
12650	 * One fundamental assumption here is that the length allocated for
12651	 * both the local and remote sglists is identical.  Otherwise, we've
12652	 * essentially got a coding error of some sort.
12653	 */
12654	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12655		int isc_ret;
12656		uint32_t cur_len, dma_length;
12657		uint8_t *tmp_ptr;
12658
12659		rq->id = CTL_HA_DATA_CTL;
12660		rq->command = command;
12661		rq->context = io;
12662
12663		/*
12664		 * Both pointers should be aligned.  But it is possible
12665		 * that the allocation length is not.  They should both
12666		 * also have enough slack left over at the end, though,
12667		 * to round up to the next 8 byte boundary.
12668		 */
12669		cur_len = ctl_min(local_sglist[i].len - local_used,
12670				  remote_sglist[j].len - remote_used);
12671
12672		/*
12673		 * In this case, we have a size issue and need to decrease
12674		 * the size, except in the case where we actually have less
12675		 * than 8 bytes left.  In that case, we need to increase
12676		 * the DMA length to get the last bit.
12677		 */
12678		if ((cur_len & 0x7) != 0) {
12679			if (cur_len > 0x7) {
12680				cur_len = cur_len - (cur_len & 0x7);
12681				dma_length = cur_len;
12682			} else {
12683				CTL_SIZE_8B(dma_length, cur_len);
12684			}
12685
12686		} else
12687			dma_length = cur_len;
12688
12689		/*
12690		 * If we had to allocate memory for this I/O, instead of using
12691		 * the non-cached mirror memory, we'll need to flush the cache
12692		 * before trying to DMA to the other controller.
12693		 *
12694		 * We could end up doing this multiple times for the same
12695		 * segment if we have a larger local segment than remote
12696		 * segment.  That shouldn't be an issue.
12697		 */
12698		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12699			/*
12700			 * XXX KDM use bus_dmamap_sync() here.
12701			 */
12702		}
12703
12704		rq->size = dma_length;
12705
12706		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12707		tmp_ptr += local_used;
12708
12709		/* Use physical addresses when talking to ISC hardware */
12710		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12711			/* XXX KDM use busdma */
12712#if 0
12713			rq->local = vtophys(tmp_ptr);
12714#endif
12715		} else
12716			rq->local = tmp_ptr;
12717
12718		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12719		tmp_ptr += remote_used;
12720		rq->remote = tmp_ptr;
12721
12722		rq->callback = NULL;
12723
12724		local_used += cur_len;
12725		if (local_used >= local_sglist[i].len) {
12726			i++;
12727			local_used = 0;
12728		}
12729
12730		remote_used += cur_len;
12731		if (remote_used >= remote_sglist[j].len) {
12732			j++;
12733			remote_used = 0;
12734		}
12735		total_used += cur_len;
12736
12737		if (total_used >= io->scsiio.kern_data_len)
12738			rq->callback = callback;
12739
12740		if ((rq->size & 0x7) != 0) {
12741			printf("%s: warning: size %d is not on 8b boundary\n",
12742			       __func__, rq->size);
12743		}
12744		if (((uintptr_t)rq->local & 0x7) != 0) {
12745			printf("%s: warning: local %p not on 8b boundary\n",
12746			       __func__, rq->local);
12747		}
12748		if (((uintptr_t)rq->remote & 0x7) != 0) {
12749			printf("%s: warning: remote %p not on 8b boundary\n",
12750			       __func__, rq->local);
12751		}
12752#if 0
12753		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12754		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12755		       rq->local, rq->remote, rq->size);
12756#endif
12757
12758		isc_ret = ctl_dt_single(rq);
12759		if (isc_ret == CTL_HA_STATUS_WAIT)
12760			continue;
12761
12762		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12763			rq->ret = CTL_HA_STATUS_SUCCESS;
12764		} else {
12765			rq->ret = isc_ret;
12766		}
12767		callback(rq);
12768		goto bailout;
12769	}
12770
12771bailout:
12772	return (retval);
12773
12774}
12775
12776static void
12777ctl_datamove_remote_read(union ctl_io *io)
12778{
12779	int retval;
12780	int i;
12781
12782	/*
12783	 * This will send an error to the other controller in the case of a
12784	 * failure.
12785	 */
12786	retval = ctl_datamove_remote_sgl_setup(io);
12787	if (retval != 0)
12788		return;
12789
12790	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12791					  ctl_datamove_remote_read_cb);
12792	if ((retval != 0)
12793	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12794		/*
12795		 * Make sure we free memory if there was an error..  The
12796		 * ctl_datamove_remote_xfer() function will send the
12797		 * datamove done message, or call the callback with an
12798		 * error if there is a problem.
12799		 */
12800		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12801			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12802	}
12803
12804	return;
12805}
12806
12807/*
12808 * Process a datamove request from the other controller.  This is used for
12809 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12810 * first.  Once that is complete, the data gets DMAed into the remote
12811 * controller's memory.  For reads, we DMA from the remote controller's
12812 * memory into our memory first, and then move it out to the FETD.
12813 */
12814static void
12815ctl_datamove_remote(union ctl_io *io)
12816{
12817	struct ctl_softc *softc;
12818
12819	softc = control_softc;
12820
12821	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12822
12823	/*
12824	 * Note that we look for an aborted I/O here, but don't do some of
12825	 * the other checks that ctl_datamove() normally does.  We don't
12826	 * need to run the task queue, because this I/O is on the ISC
12827	 * queue, which is executed by the work thread after the task queue.
12828	 * We don't need to run the datamove delay code, since that should
12829	 * have been done if need be on the other controller.
12830	 */
12831	mtx_lock(&softc->ctl_lock);
12832
12833	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12834
12835		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12836		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12837		       io->io_hdr.nexus.targ_port,
12838		       io->io_hdr.nexus.targ_target.id,
12839		       io->io_hdr.nexus.targ_lun);
12840		io->io_hdr.status = CTL_CMD_ABORTED;
12841		io->io_hdr.port_status = 31338;
12842
12843		mtx_unlock(&softc->ctl_lock);
12844
12845		ctl_send_datamove_done(io, /*have_lock*/ 0);
12846
12847		return;
12848	}
12849
12850	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12851		mtx_unlock(&softc->ctl_lock);
12852		ctl_datamove_remote_write(io);
12853	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12854		mtx_unlock(&softc->ctl_lock);
12855		ctl_datamove_remote_read(io);
12856	} else {
12857		union ctl_ha_msg msg;
12858		struct scsi_sense_data *sense;
12859		uint8_t sks[3];
12860		int retry_count;
12861
12862		memset(&msg, 0, sizeof(msg));
12863
12864		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12865		msg.hdr.status = CTL_SCSI_ERROR;
12866		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12867
12868		retry_count = 4243;
12869
12870		sense = &msg.scsi.sense_data;
12871		sks[0] = SSD_SCS_VALID;
12872		sks[1] = (retry_count >> 8) & 0xff;
12873		sks[2] = retry_count & 0xff;
12874
12875		/* "Internal target failure" */
12876		scsi_set_sense_data(sense,
12877				    /*sense_format*/ SSD_TYPE_NONE,
12878				    /*current_error*/ 1,
12879				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12880				    /*asc*/ 0x44,
12881				    /*ascq*/ 0x00,
12882				    /*type*/ SSD_ELEM_SKS,
12883				    /*size*/ sizeof(sks),
12884				    /*data*/ sks,
12885				    SSD_ELEM_NONE);
12886
12887		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12888		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12889			ctl_failover_io(io, /*have_lock*/ 1);
12890			mtx_unlock(&softc->ctl_lock);
12891			return;
12892		}
12893
12894		mtx_unlock(&softc->ctl_lock);
12895
12896		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12897		    CTL_HA_STATUS_SUCCESS) {
12898			/* XXX KDM what to do if this fails? */
12899		}
12900		return;
12901	}
12902
12903}
12904
12905static int
12906ctl_process_done(union ctl_io *io, int have_lock)
12907{
12908	struct ctl_lun *lun;
12909	struct ctl_softc *ctl_softc;
12910	void (*fe_done)(union ctl_io *io);
12911	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12912
12913	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12914
12915	fe_done =
12916	    control_softc->ctl_ports[targ_port]->fe_done;
12917
12918#ifdef CTL_TIME_IO
12919	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12920		char str[256];
12921		char path_str[64];
12922		struct sbuf sb;
12923
12924		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12925		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12926
12927		sbuf_cat(&sb, path_str);
12928		switch (io->io_hdr.io_type) {
12929		case CTL_IO_SCSI:
12930			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12931			sbuf_printf(&sb, "\n");
12932			sbuf_cat(&sb, path_str);
12933			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12934				    io->scsiio.tag_num, io->scsiio.tag_type);
12935			break;
12936		case CTL_IO_TASK:
12937			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12938				    "Tag Type: %d\n", io->taskio.task_action,
12939				    io->taskio.tag_num, io->taskio.tag_type);
12940			break;
12941		default:
12942			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12943			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12944			break;
12945		}
12946		sbuf_cat(&sb, path_str);
12947		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12948			    (intmax_t)time_uptime - io->io_hdr.start_time);
12949		sbuf_finish(&sb);
12950		printf("%s", sbuf_data(&sb));
12951	}
12952#endif /* CTL_TIME_IO */
12953
12954	switch (io->io_hdr.io_type) {
12955	case CTL_IO_SCSI:
12956		break;
12957	case CTL_IO_TASK:
12958		if (bootverbose || verbose > 0)
12959			ctl_io_error_print(io, NULL);
12960		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12961			ctl_free_io(io);
12962		else
12963			fe_done(io);
12964		return (CTL_RETVAL_COMPLETE);
12965		break;
12966	default:
12967		printf("ctl_process_done: invalid io type %d\n",
12968		       io->io_hdr.io_type);
12969		panic("ctl_process_done: invalid io type %d\n",
12970		      io->io_hdr.io_type);
12971		break; /* NOTREACHED */
12972	}
12973
12974	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12975	if (lun == NULL) {
12976		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12977				 io->io_hdr.nexus.targ_lun));
12978		fe_done(io);
12979		goto bailout;
12980	}
12981	ctl_softc = lun->ctl_softc;
12982
12983	/*
12984	 * Remove this from the OOA queue.
12985	 */
12986	if (have_lock == 0)
12987		mtx_lock(&ctl_softc->ctl_lock);
12988
12989	/*
12990	 * Check to see if we have any errors to inject here.  We only
12991	 * inject errors for commands that don't already have errors set.
12992	 */
12993	if ((STAILQ_FIRST(&lun->error_list) != NULL)
12994	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
12995		ctl_inject_error(lun, io);
12996
12997	/*
12998	 * XXX KDM how do we treat commands that aren't completed
12999	 * successfully?
13000	 *
13001	 * XXX KDM should we also track I/O latency?
13002	 */
13003	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
13004		uint32_t blocksize;
13005#ifdef CTL_TIME_IO
13006		struct bintime cur_bt;
13007#endif
13008
13009		if ((lun->be_lun != NULL)
13010		 && (lun->be_lun->blocksize != 0))
13011			blocksize = lun->be_lun->blocksize;
13012		else
13013			blocksize = 512;
13014
13015		switch (io->io_hdr.io_type) {
13016		case CTL_IO_SCSI: {
13017			int isread;
13018			struct ctl_lba_len_flags *lbalen;
13019
13020			isread = 0;
13021			switch (io->scsiio.cdb[0]) {
13022			case READ_6:
13023			case READ_10:
13024			case READ_12:
13025			case READ_16:
13026				isread = 1;
13027				/* FALLTHROUGH */
13028			case WRITE_6:
13029			case WRITE_10:
13030			case WRITE_12:
13031			case WRITE_16:
13032			case WRITE_VERIFY_10:
13033			case WRITE_VERIFY_12:
13034			case WRITE_VERIFY_16:
13035				lbalen = (struct ctl_lba_len_flags *)
13036				    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
13037
13038				if (isread) {
13039					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
13040					    lbalen->len * blocksize;
13041					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
13042
13043#ifdef CTL_TIME_IO
13044					bintime_add(
13045					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
13046					   &io->io_hdr.dma_bt);
13047					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
13048						io->io_hdr.num_dmas;
13049					getbintime(&cur_bt);
13050					bintime_sub(&cur_bt,
13051						    &io->io_hdr.start_bt);
13052
13053					bintime_add(
13054					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
13055					    &cur_bt);
13056
13057#if 0
13058					cs_prof_gettime(&cur_ticks);
13059					lun->stats.time[CTL_STATS_READ] +=
13060						cur_ticks -
13061						io->io_hdr.start_ticks;
13062#endif
13063#if 0
13064					lun->stats.time[CTL_STATS_READ] +=
13065						jiffies - io->io_hdr.start_time;
13066#endif
13067#endif /* CTL_TIME_IO */
13068				} else {
13069					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
13070					    lbalen->len * blocksize;
13071					lun->stats.ports[targ_port].operations[
13072						CTL_STATS_WRITE]++;
13073
13074#ifdef CTL_TIME_IO
13075					bintime_add(
13076					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
13077					  &io->io_hdr.dma_bt);
13078					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
13079						io->io_hdr.num_dmas;
13080					getbintime(&cur_bt);
13081					bintime_sub(&cur_bt,
13082						    &io->io_hdr.start_bt);
13083
13084					bintime_add(
13085					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
13086					    &cur_bt);
13087#if 0
13088					cs_prof_gettime(&cur_ticks);
13089					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13090						cur_ticks -
13091						io->io_hdr.start_ticks;
13092					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13093						jiffies - io->io_hdr.start_time;
13094#endif
13095#endif /* CTL_TIME_IO */
13096				}
13097				break;
13098			default:
13099				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
13100
13101#ifdef CTL_TIME_IO
13102				bintime_add(
13103				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
13104				  &io->io_hdr.dma_bt);
13105				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
13106					io->io_hdr.num_dmas;
13107				getbintime(&cur_bt);
13108				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13109
13110				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
13111					    &cur_bt);
13112
13113#if 0
13114				cs_prof_gettime(&cur_ticks);
13115				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13116					cur_ticks -
13117					io->io_hdr.start_ticks;
13118				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13119					jiffies - io->io_hdr.start_time;
13120#endif
13121#endif /* CTL_TIME_IO */
13122				break;
13123			}
13124			break;
13125		}
13126		default:
13127			break;
13128		}
13129	}
13130
13131	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13132
13133	/*
13134	 * Run through the blocked queue on this LUN and see if anything
13135	 * has become unblocked, now that this transaction is done.
13136	 */
13137	ctl_check_blocked(lun);
13138
13139	/*
13140	 * If the LUN has been invalidated, free it if there is nothing
13141	 * left on its OOA queue.
13142	 */
13143	if ((lun->flags & CTL_LUN_INVALID)
13144	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
13145		ctl_free_lun(lun);
13146
13147	/*
13148	 * If this command has been aborted, make sure we set the status
13149	 * properly.  The FETD is responsible for freeing the I/O and doing
13150	 * whatever it needs to do to clean up its state.
13151	 */
13152	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13153		io->io_hdr.status = CTL_CMD_ABORTED;
13154
13155	/*
13156	 * We print out status for every task management command.  For SCSI
13157	 * commands, we filter out any unit attention errors; they happen
13158	 * on every boot, and would clutter up the log.  Note:  task
13159	 * management commands aren't printed here, they are printed above,
13160	 * since they should never even make it down here.
13161	 */
13162	switch (io->io_hdr.io_type) {
13163	case CTL_IO_SCSI: {
13164		int error_code, sense_key, asc, ascq;
13165
13166		sense_key = 0;
13167
13168		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13169		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13170			/*
13171			 * Since this is just for printing, no need to
13172			 * show errors here.
13173			 */
13174			scsi_extract_sense_len(&io->scsiio.sense_data,
13175					       io->scsiio.sense_len,
13176					       &error_code,
13177					       &sense_key,
13178					       &asc,
13179					       &ascq,
13180					       /*show_errors*/ 0);
13181		}
13182
13183		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13184		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13185		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13186		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13187
13188			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13189				ctl_softc->skipped_prints++;
13190				if (have_lock == 0)
13191					mtx_unlock(&ctl_softc->ctl_lock);
13192			} else {
13193				uint32_t skipped_prints;
13194
13195				skipped_prints = ctl_softc->skipped_prints;
13196
13197				ctl_softc->skipped_prints = 0;
13198				ctl_softc->last_print_jiffies = time_uptime;
13199
13200				if (have_lock == 0)
13201					mtx_unlock(&ctl_softc->ctl_lock);
13202				if (skipped_prints > 0) {
13203#ifdef NEEDTOPORT
13204					csevent_log(CSC_CTL | CSC_SHELF_SW |
13205					    CTL_ERROR_REPORT,
13206					    csevent_LogType_Trace,
13207					    csevent_Severity_Information,
13208					    csevent_AlertLevel_Green,
13209					    csevent_FRU_Firmware,
13210					    csevent_FRU_Unknown,
13211					    "High CTL error volume, %d prints "
13212					    "skipped", skipped_prints);
13213#endif
13214				}
13215				if (bootverbose || verbose > 0)
13216					ctl_io_error_print(io, NULL);
13217			}
13218		} else {
13219			if (have_lock == 0)
13220				mtx_unlock(&ctl_softc->ctl_lock);
13221		}
13222		break;
13223	}
13224	case CTL_IO_TASK:
13225		if (have_lock == 0)
13226			mtx_unlock(&ctl_softc->ctl_lock);
13227		if (bootverbose || verbose > 0)
13228			ctl_io_error_print(io, NULL);
13229		break;
13230	default:
13231		if (have_lock == 0)
13232			mtx_unlock(&ctl_softc->ctl_lock);
13233		break;
13234	}
13235
13236	/*
13237	 * Tell the FETD or the other shelf controller we're done with this
13238	 * command.  Note that only SCSI commands get to this point.  Task
13239	 * management commands are completed above.
13240	 *
13241	 * We only send status to the other controller if we're in XFER
13242	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13243	 * received the I/O (from CTL's perspective), and so the status is
13244	 * generated there.
13245	 *
13246	 * XXX KDM if we hold the lock here, we could cause a deadlock
13247	 * if the frontend comes back in in this context to queue
13248	 * something.
13249	 */
13250	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13251	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13252		union ctl_ha_msg msg;
13253
13254		memset(&msg, 0, sizeof(msg));
13255		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13256		msg.hdr.original_sc = io->io_hdr.original_sc;
13257		msg.hdr.nexus = io->io_hdr.nexus;
13258		msg.hdr.status = io->io_hdr.status;
13259		msg.scsi.scsi_status = io->scsiio.scsi_status;
13260		msg.scsi.tag_num = io->scsiio.tag_num;
13261		msg.scsi.tag_type = io->scsiio.tag_type;
13262		msg.scsi.sense_len = io->scsiio.sense_len;
13263		msg.scsi.sense_residual = io->scsiio.sense_residual;
13264		msg.scsi.residual = io->scsiio.residual;
13265		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13266		       sizeof(io->scsiio.sense_data));
13267		/*
13268		 * We copy this whether or not this is an I/O-related
13269		 * command.  Otherwise, we'd have to go and check to see
13270		 * whether it's a read/write command, and it really isn't
13271		 * worth it.
13272		 */
13273		memcpy(&msg.scsi.lbalen,
13274		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13275		       sizeof(msg.scsi.lbalen));
13276
13277		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13278				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13279			/* XXX do something here */
13280		}
13281
13282		ctl_free_io(io);
13283	} else
13284		fe_done(io);
13285
13286bailout:
13287
13288	return (CTL_RETVAL_COMPLETE);
13289}
13290
13291/*
13292 * Front end should call this if it doesn't do autosense.  When the request
13293 * sense comes back in from the initiator, we'll dequeue this and send it.
13294 */
13295int
13296ctl_queue_sense(union ctl_io *io)
13297{
13298	struct ctl_lun *lun;
13299	struct ctl_softc *ctl_softc;
13300	uint32_t initidx, targ_lun;
13301
13302	ctl_softc = control_softc;
13303
13304	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13305
13306	/*
13307	 * LUN lookup will likely move to the ctl_work_thread() once we
13308	 * have our new queueing infrastructure (that doesn't put things on
13309	 * a per-LUN queue initially).  That is so that we can handle
13310	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13311	 * can't deal with that right now.
13312	 */
13313	mtx_lock(&ctl_softc->ctl_lock);
13314
13315	/*
13316	 * If we don't have a LUN for this, just toss the sense
13317	 * information.
13318	 */
13319	targ_lun = io->io_hdr.nexus.targ_lun;
13320	if (io->io_hdr.nexus.lun_map_fn != NULL)
13321		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13322	if ((targ_lun < CTL_MAX_LUNS)
13323	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13324		lun = ctl_softc->ctl_luns[targ_lun];
13325	else
13326		goto bailout;
13327
13328	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13329
13330	/*
13331	 * Already have CA set for this LUN...toss the sense information.
13332	 */
13333	if (ctl_is_set(lun->have_ca, initidx))
13334		goto bailout;
13335
13336	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13337	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13338	       sizeof(io->scsiio.sense_data)));
13339	ctl_set_mask(lun->have_ca, initidx);
13340
13341bailout:
13342	mtx_unlock(&ctl_softc->ctl_lock);
13343
13344	ctl_free_io(io);
13345
13346	return (CTL_RETVAL_COMPLETE);
13347}
13348
13349/*
13350 * Primary command inlet from frontend ports.  All SCSI and task I/O
13351 * requests must go through this function.
13352 */
13353int
13354ctl_queue(union ctl_io *io)
13355{
13356	struct ctl_softc *ctl_softc;
13357
13358	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13359
13360	ctl_softc = control_softc;
13361
13362#ifdef CTL_TIME_IO
13363	io->io_hdr.start_time = time_uptime;
13364	getbintime(&io->io_hdr.start_bt);
13365#endif /* CTL_TIME_IO */
13366
13367	mtx_lock(&ctl_softc->ctl_lock);
13368
13369	switch (io->io_hdr.io_type) {
13370	case CTL_IO_SCSI:
13371		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
13372				   links);
13373		break;
13374	case CTL_IO_TASK:
13375		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
13376		/*
13377		 * Set the task pending flag.  This is necessary to close a
13378		 * race condition with the FETD:
13379		 *
13380		 * - FETD submits a task management command, like an abort.
13381		 * - Back end calls fe_datamove() to move the data for the
13382		 *   aborted command.  The FETD can't really accept it, but
13383		 *   if it did, it would end up transmitting data for a
13384		 *   command that the initiator told us to abort.
13385		 *
13386		 * We close the race condition by setting the flag here,
13387		 * and checking it in ctl_datamove(), before calling the
13388		 * FETD's fe_datamove routine.  If we've got a task
13389		 * pending, we run the task queue and then check to see
13390		 * whether our particular I/O has been aborted.
13391		 */
13392		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
13393		break;
13394	default:
13395		mtx_unlock(&ctl_softc->ctl_lock);
13396		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13397		return (-EINVAL);
13398		break; /* NOTREACHED */
13399	}
13400	mtx_unlock(&ctl_softc->ctl_lock);
13401
13402	ctl_wakeup_thread();
13403
13404	return (CTL_RETVAL_COMPLETE);
13405}
13406
13407#ifdef CTL_IO_DELAY
13408static void
13409ctl_done_timer_wakeup(void *arg)
13410{
13411	union ctl_io *io;
13412
13413	io = (union ctl_io *)arg;
13414	ctl_done_lock(io, /*have_lock*/ 0);
13415}
13416#endif /* CTL_IO_DELAY */
13417
13418void
13419ctl_done_lock(union ctl_io *io, int have_lock)
13420{
13421	struct ctl_softc *ctl_softc;
13422#ifndef CTL_DONE_THREAD
13423	union ctl_io *xio;
13424#endif /* !CTL_DONE_THREAD */
13425
13426	ctl_softc = control_softc;
13427
13428	if (have_lock == 0)
13429		mtx_lock(&ctl_softc->ctl_lock);
13430
13431	/*
13432	 * Enable this to catch duplicate completion issues.
13433	 */
13434#if 0
13435	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13436		printf("%s: type %d msg %d cdb %x iptl: "
13437		       "%d:%d:%d:%d tag 0x%04x "
13438		       "flag %#x status %x\n",
13439			__func__,
13440			io->io_hdr.io_type,
13441			io->io_hdr.msg_type,
13442			io->scsiio.cdb[0],
13443			io->io_hdr.nexus.initid.id,
13444			io->io_hdr.nexus.targ_port,
13445			io->io_hdr.nexus.targ_target.id,
13446			io->io_hdr.nexus.targ_lun,
13447			(io->io_hdr.io_type ==
13448			CTL_IO_TASK) ?
13449			io->taskio.tag_num :
13450			io->scsiio.tag_num,
13451		        io->io_hdr.flags,
13452			io->io_hdr.status);
13453	} else
13454		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13455#endif
13456
13457	/*
13458	 * This is an internal copy of an I/O, and should not go through
13459	 * the normal done processing logic.
13460	 */
13461	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
13462		if (have_lock == 0)
13463			mtx_unlock(&ctl_softc->ctl_lock);
13464		return;
13465	}
13466
13467	/*
13468	 * We need to send a msg to the serializing shelf to finish the IO
13469	 * as well.  We don't send a finish message to the other shelf if
13470	 * this is a task management command.  Task management commands
13471	 * aren't serialized in the OOA queue, but rather just executed on
13472	 * both shelf controllers for commands that originated on that
13473	 * controller.
13474	 */
13475	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13476	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13477		union ctl_ha_msg msg_io;
13478
13479		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13480		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13481		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13482		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13483		}
13484		/* continue on to finish IO */
13485	}
13486#ifdef CTL_IO_DELAY
13487	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13488		struct ctl_lun *lun;
13489
13490		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13491
13492		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13493	} else {
13494		struct ctl_lun *lun;
13495
13496		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13497
13498		if ((lun != NULL)
13499		 && (lun->delay_info.done_delay > 0)) {
13500			struct callout *callout;
13501
13502			callout = (struct callout *)&io->io_hdr.timer_bytes;
13503			callout_init(callout, /*mpsafe*/ 1);
13504			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13505			callout_reset(callout,
13506				      lun->delay_info.done_delay * hz,
13507				      ctl_done_timer_wakeup, io);
13508			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13509				lun->delay_info.done_delay = 0;
13510			if (have_lock == 0)
13511				mtx_unlock(&ctl_softc->ctl_lock);
13512			return;
13513		}
13514	}
13515#endif /* CTL_IO_DELAY */
13516
13517	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
13518
13519#ifdef CTL_DONE_THREAD
13520	if (have_lock == 0)
13521		mtx_unlock(&ctl_softc->ctl_lock);
13522
13523	ctl_wakeup_thread();
13524#else /* CTL_DONE_THREAD */
13525	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
13526	     xio != NULL;
13527	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
13528
13529		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
13530
13531		ctl_process_done(xio, /*have_lock*/ 1);
13532	}
13533	if (have_lock == 0)
13534		mtx_unlock(&ctl_softc->ctl_lock);
13535#endif /* CTL_DONE_THREAD */
13536}
13537
13538void
13539ctl_done(union ctl_io *io)
13540{
13541	ctl_done_lock(io, /*have_lock*/ 0);
13542}
13543
13544int
13545ctl_isc(struct ctl_scsiio *ctsio)
13546{
13547	struct ctl_lun *lun;
13548	int retval;
13549
13550	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13551
13552	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13553
13554	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13555
13556	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13557
13558	return (retval);
13559}
13560
13561
13562static void
13563ctl_work_thread(void *arg)
13564{
13565	struct ctl_softc *softc;
13566	union ctl_io *io;
13567	struct ctl_be_lun *be_lun;
13568	int retval;
13569
13570	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13571
13572	softc = (struct ctl_softc *)arg;
13573	if (softc == NULL)
13574		return;
13575
13576	mtx_lock(&softc->ctl_lock);
13577	for (;;) {
13578		retval = 0;
13579
13580		/*
13581		 * We handle the queues in this order:
13582		 * - task management
13583		 * - ISC
13584		 * - done queue (to free up resources, unblock other commands)
13585		 * - RtR queue
13586		 * - incoming queue
13587		 *
13588		 * If those queues are empty, we break out of the loop and
13589		 * go to sleep.
13590		 */
13591		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
13592		if (io != NULL) {
13593			ctl_run_task_queue(softc);
13594			continue;
13595		}
13596		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
13597		if (io != NULL) {
13598			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
13599			ctl_handle_isc(io);
13600			continue;
13601		}
13602		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
13603		if (io != NULL) {
13604			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
13605			/* clear any blocked commands, call fe_done */
13606			mtx_unlock(&softc->ctl_lock);
13607			/*
13608			 * XXX KDM
13609			 * Call this without a lock for now.  This will
13610			 * depend on whether there is any way the FETD can
13611			 * sleep or deadlock if called with the CTL lock
13612			 * held.
13613			 */
13614			retval = ctl_process_done(io, /*have_lock*/ 0);
13615			mtx_lock(&softc->ctl_lock);
13616			continue;
13617		}
13618		if (!ctl_pause_rtr) {
13619			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
13620			if (io != NULL) {
13621				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
13622				mtx_unlock(&softc->ctl_lock);
13623				retval = ctl_scsiio(&io->scsiio);
13624				if (retval != CTL_RETVAL_COMPLETE)
13625					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13626				mtx_lock(&softc->ctl_lock);
13627				continue;
13628			}
13629		}
13630		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
13631		if (io != NULL) {
13632			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
13633			mtx_unlock(&softc->ctl_lock);
13634			ctl_scsiio_precheck(softc, &io->scsiio);
13635			mtx_lock(&softc->ctl_lock);
13636			continue;
13637		}
13638		/*
13639		 * We might want to move this to a separate thread, so that
13640		 * configuration requests (in this case LUN creations)
13641		 * won't impact the I/O path.
13642		 */
13643		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13644		if (be_lun != NULL) {
13645			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13646			mtx_unlock(&softc->ctl_lock);
13647			ctl_create_lun(be_lun);
13648			mtx_lock(&softc->ctl_lock);
13649			continue;
13650		}
13651
13652		/* XXX KDM use the PDROP flag?? */
13653		/* Sleep until we have something to do. */
13654		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "-", 0);
13655
13656		/* Back to the top of the loop to see what woke us up. */
13657		continue;
13658	}
13659}
13660
13661void
13662ctl_wakeup_thread()
13663{
13664	struct ctl_softc *softc;
13665
13666	softc = control_softc;
13667
13668	wakeup_one(softc);
13669}
13670
13671/* Initialization and failover */
13672
13673void
13674ctl_init_isc_msg(void)
13675{
13676	printf("CTL: Still calling this thing\n");
13677}
13678
13679/*
13680 * Init component
13681 * 	Initializes component into configuration defined by bootMode
13682 *	(see hasc-sv.c)
13683 *  	returns hasc_Status:
13684 * 		OK
13685 *		ERROR - fatal error
13686 */
13687static ctl_ha_comp_status
13688ctl_isc_init(struct ctl_ha_component *c)
13689{
13690	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13691
13692	c->status = ret;
13693	return ret;
13694}
13695
13696/* Start component
13697 * 	Starts component in state requested. If component starts successfully,
13698 *	it must set its own state to the requestrd state
13699 *	When requested state is HASC_STATE_HA, the component may refine it
13700 * 	by adding _SLAVE or _MASTER flags.
13701 *	Currently allowed state transitions are:
13702 *	UNKNOWN->HA		- initial startup
13703 *	UNKNOWN->SINGLE - initial startup when no parter detected
13704 *	HA->SINGLE		- failover
13705 * returns ctl_ha_comp_status:
13706 * 		OK	- component successfully started in requested state
13707 *		FAILED  - could not start the requested state, failover may
13708 * 			  be possible
13709 *		ERROR	- fatal error detected, no future startup possible
13710 */
13711static ctl_ha_comp_status
13712ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13713{
13714	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13715
13716	printf("%s: go\n", __func__);
13717
13718	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13719	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13720		ctl_is_single = 0;
13721		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13722		    != CTL_HA_STATUS_SUCCESS) {
13723			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13724			ret = CTL_HA_COMP_STATUS_ERROR;
13725		}
13726	} else if (CTL_HA_STATE_IS_HA(c->state)
13727		&& CTL_HA_STATE_IS_SINGLE(state)){
13728		// HA->SINGLE transition
13729	        ctl_failover();
13730		ctl_is_single = 1;
13731	} else {
13732		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13733		       c->state, state);
13734		ret = CTL_HA_COMP_STATUS_ERROR;
13735	}
13736	if (CTL_HA_STATE_IS_SINGLE(state))
13737		ctl_is_single = 1;
13738
13739	c->state = state;
13740	c->status = ret;
13741	return ret;
13742}
13743
13744/*
13745 * Quiesce component
13746 * The component must clear any error conditions (set status to OK) and
13747 * prepare itself to another Start call
13748 * returns ctl_ha_comp_status:
13749 * 	OK
13750 *	ERROR
13751 */
13752static ctl_ha_comp_status
13753ctl_isc_quiesce(struct ctl_ha_component *c)
13754{
13755	int ret = CTL_HA_COMP_STATUS_OK;
13756
13757	ctl_pause_rtr = 1;
13758	c->status = ret;
13759	return ret;
13760}
13761
13762struct ctl_ha_component ctl_ha_component_ctlisc =
13763{
13764	.name = "CTL ISC",
13765	.state = CTL_HA_STATE_UNKNOWN,
13766	.init = ctl_isc_init,
13767	.start = ctl_isc_start,
13768	.quiesce = ctl_isc_quiesce
13769};
13770
13771/*
13772 *  vim: ts=8
13773 */
13774