ctl.c revision 275405
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 275405 2014-12-02 12:38:22Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/ctype.h>
50#include <sys/kernel.h>
51#include <sys/types.h>
52#include <sys/kthread.h>
53#include <sys/bio.h>
54#include <sys/fcntl.h>
55#include <sys/lock.h>
56#include <sys/module.h>
57#include <sys/mutex.h>
58#include <sys/condvar.h>
59#include <sys/malloc.h>
60#include <sys/conf.h>
61#include <sys/ioccom.h>
62#include <sys/queue.h>
63#include <sys/sbuf.h>
64#include <sys/smp.h>
65#include <sys/endian.h>
66#include <sys/sysctl.h>
67#include <vm/uma.h>
68
69#include <cam/cam.h>
70#include <cam/scsi/scsi_all.h>
71#include <cam/scsi/scsi_da.h>
72#include <cam/ctl/ctl_io.h>
73#include <cam/ctl/ctl.h>
74#include <cam/ctl/ctl_frontend.h>
75#include <cam/ctl/ctl_frontend_internal.h>
76#include <cam/ctl/ctl_util.h>
77#include <cam/ctl/ctl_backend.h>
78#include <cam/ctl/ctl_ioctl.h>
79#include <cam/ctl/ctl_ha.h>
80#include <cam/ctl/ctl_private.h>
81#include <cam/ctl/ctl_debug.h>
82#include <cam/ctl/ctl_scsi_all.h>
83#include <cam/ctl/ctl_error.h>
84
85struct ctl_softc *control_softc = NULL;
86
87/*
88 * Size and alignment macros needed for Copan-specific HA hardware.  These
89 * can go away when the HA code is re-written, and uses busdma for any
90 * hardware.
91 */
92#define	CTL_ALIGN_8B(target, source, type)				\
93	if (((uint32_t)source & 0x7) != 0)				\
94		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
95	else								\
96		target = (type)source;
97
98#define	CTL_SIZE_8B(target, size)					\
99	if ((size & 0x7) != 0)						\
100		target = size + (0x8 - (size & 0x7));			\
101	else								\
102		target = size;
103
104#define CTL_ALIGN_8B_MARGIN	16
105
106/*
107 * Template mode pages.
108 */
109
110/*
111 * Note that these are default values only.  The actual values will be
112 * filled in when the user does a mode sense.
113 */
114static struct copan_debugconf_subpage debugconf_page_default = {
115	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
116	DBGCNF_SUBPAGE_CODE,		/* subpage */
117	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
118	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
119	DBGCNF_VERSION,			/* page_version */
120	{CTL_TIME_IO_DEFAULT_SECS>>8,
121	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
122};
123
124static struct copan_debugconf_subpage debugconf_page_changeable = {
125	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
126	DBGCNF_SUBPAGE_CODE,		/* subpage */
127	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
128	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
129	0,				/* page_version */
130	{0xff,0xff},			/* ctl_time_io_secs */
131};
132
133static struct scsi_da_rw_recovery_page rw_er_page_default = {
134	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
135	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
136	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
137	/*read_retry_count*/0,
138	/*correction_span*/0,
139	/*head_offset_count*/0,
140	/*data_strobe_offset_cnt*/0,
141	/*byte8*/SMS_RWER_LBPERE,
142	/*write_retry_count*/0,
143	/*reserved2*/0,
144	/*recovery_time_limit*/{0, 0},
145};
146
147static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
148	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
149	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
150	/*byte3*/0,
151	/*read_retry_count*/0,
152	/*correction_span*/0,
153	/*head_offset_count*/0,
154	/*data_strobe_offset_cnt*/0,
155	/*byte8*/0,
156	/*write_retry_count*/0,
157	/*reserved2*/0,
158	/*recovery_time_limit*/{0, 0},
159};
160
161static struct scsi_format_page format_page_default = {
162	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
163	/*page_length*/sizeof(struct scsi_format_page) - 2,
164	/*tracks_per_zone*/ {0, 0},
165	/*alt_sectors_per_zone*/ {0, 0},
166	/*alt_tracks_per_zone*/ {0, 0},
167	/*alt_tracks_per_lun*/ {0, 0},
168	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
169			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
170	/*bytes_per_sector*/ {0, 0},
171	/*interleave*/ {0, 0},
172	/*track_skew*/ {0, 0},
173	/*cylinder_skew*/ {0, 0},
174	/*flags*/ SFP_HSEC,
175	/*reserved*/ {0, 0, 0}
176};
177
178static struct scsi_format_page format_page_changeable = {
179	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
180	/*page_length*/sizeof(struct scsi_format_page) - 2,
181	/*tracks_per_zone*/ {0, 0},
182	/*alt_sectors_per_zone*/ {0, 0},
183	/*alt_tracks_per_zone*/ {0, 0},
184	/*alt_tracks_per_lun*/ {0, 0},
185	/*sectors_per_track*/ {0, 0},
186	/*bytes_per_sector*/ {0, 0},
187	/*interleave*/ {0, 0},
188	/*track_skew*/ {0, 0},
189	/*cylinder_skew*/ {0, 0},
190	/*flags*/ 0,
191	/*reserved*/ {0, 0, 0}
192};
193
194static struct scsi_rigid_disk_page rigid_disk_page_default = {
195	/*page_code*/SMS_RIGID_DISK_PAGE,
196	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
197	/*cylinders*/ {0, 0, 0},
198	/*heads*/ CTL_DEFAULT_HEADS,
199	/*start_write_precomp*/ {0, 0, 0},
200	/*start_reduced_current*/ {0, 0, 0},
201	/*step_rate*/ {0, 0},
202	/*landing_zone_cylinder*/ {0, 0, 0},
203	/*rpl*/ SRDP_RPL_DISABLED,
204	/*rotational_offset*/ 0,
205	/*reserved1*/ 0,
206	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
207			   CTL_DEFAULT_ROTATION_RATE & 0xff},
208	/*reserved2*/ {0, 0}
209};
210
211static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
212	/*page_code*/SMS_RIGID_DISK_PAGE,
213	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
214	/*cylinders*/ {0, 0, 0},
215	/*heads*/ 0,
216	/*start_write_precomp*/ {0, 0, 0},
217	/*start_reduced_current*/ {0, 0, 0},
218	/*step_rate*/ {0, 0},
219	/*landing_zone_cylinder*/ {0, 0, 0},
220	/*rpl*/ 0,
221	/*rotational_offset*/ 0,
222	/*reserved1*/ 0,
223	/*rotation_rate*/ {0, 0},
224	/*reserved2*/ {0, 0}
225};
226
227static struct scsi_caching_page caching_page_default = {
228	/*page_code*/SMS_CACHING_PAGE,
229	/*page_length*/sizeof(struct scsi_caching_page) - 2,
230	/*flags1*/ SCP_DISC | SCP_WCE,
231	/*ret_priority*/ 0,
232	/*disable_pf_transfer_len*/ {0xff, 0xff},
233	/*min_prefetch*/ {0, 0},
234	/*max_prefetch*/ {0xff, 0xff},
235	/*max_pf_ceiling*/ {0xff, 0xff},
236	/*flags2*/ 0,
237	/*cache_segments*/ 0,
238	/*cache_seg_size*/ {0, 0},
239	/*reserved*/ 0,
240	/*non_cache_seg_size*/ {0, 0, 0}
241};
242
243static struct scsi_caching_page caching_page_changeable = {
244	/*page_code*/SMS_CACHING_PAGE,
245	/*page_length*/sizeof(struct scsi_caching_page) - 2,
246	/*flags1*/ SCP_WCE | SCP_RCD,
247	/*ret_priority*/ 0,
248	/*disable_pf_transfer_len*/ {0, 0},
249	/*min_prefetch*/ {0, 0},
250	/*max_prefetch*/ {0, 0},
251	/*max_pf_ceiling*/ {0, 0},
252	/*flags2*/ 0,
253	/*cache_segments*/ 0,
254	/*cache_seg_size*/ {0, 0},
255	/*reserved*/ 0,
256	/*non_cache_seg_size*/ {0, 0, 0}
257};
258
259static struct scsi_control_page control_page_default = {
260	/*page_code*/SMS_CONTROL_MODE_PAGE,
261	/*page_length*/sizeof(struct scsi_control_page) - 2,
262	/*rlec*/0,
263	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
264	/*eca_and_aen*/0,
265	/*flags4*/SCP_TAS,
266	/*aen_holdoff_period*/{0, 0},
267	/*busy_timeout_period*/{0, 0},
268	/*extended_selftest_completion_time*/{0, 0}
269};
270
271static struct scsi_control_page control_page_changeable = {
272	/*page_code*/SMS_CONTROL_MODE_PAGE,
273	/*page_length*/sizeof(struct scsi_control_page) - 2,
274	/*rlec*/SCP_DSENSE,
275	/*queue_flags*/SCP_QUEUE_ALG_MASK,
276	/*eca_and_aen*/SCP_SWP,
277	/*flags4*/0,
278	/*aen_holdoff_period*/{0, 0},
279	/*busy_timeout_period*/{0, 0},
280	/*extended_selftest_completion_time*/{0, 0}
281};
282
283static struct scsi_info_exceptions_page ie_page_default = {
284	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
285	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
286	/*info_flags*/SIEP_FLAGS_DEXCPT,
287	/*mrie*/0,
288	/*interval_timer*/{0, 0, 0, 0},
289	/*report_count*/{0, 0, 0, 0}
290};
291
292static struct scsi_info_exceptions_page ie_page_changeable = {
293	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
294	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
295	/*info_flags*/0,
296	/*mrie*/0,
297	/*interval_timer*/{0, 0, 0, 0},
298	/*report_count*/{0, 0, 0, 0}
299};
300
301#define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
302
303static struct ctl_logical_block_provisioning_page lbp_page_default = {{
304	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
305	/*subpage_code*/0x02,
306	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
307	/*flags*/0,
308	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
309	/*descr*/{}},
310	{{/*flags*/0,
311	  /*resource*/0x01,
312	  /*reserved*/{0, 0},
313	  /*count*/{0, 0, 0, 0}},
314	 {/*flags*/0,
315	  /*resource*/0x02,
316	  /*reserved*/{0, 0},
317	  /*count*/{0, 0, 0, 0}},
318	 {/*flags*/0,
319	  /*resource*/0xf1,
320	  /*reserved*/{0, 0},
321	  /*count*/{0, 0, 0, 0}},
322	 {/*flags*/0,
323	  /*resource*/0xf2,
324	  /*reserved*/{0, 0},
325	  /*count*/{0, 0, 0, 0}}
326	}
327};
328
329static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
330	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
331	/*subpage_code*/0x02,
332	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
333	/*flags*/0,
334	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
335	/*descr*/{}},
336	{{/*flags*/0,
337	  /*resource*/0,
338	  /*reserved*/{0, 0},
339	  /*count*/{0, 0, 0, 0}},
340	 {/*flags*/0,
341	  /*resource*/0,
342	  /*reserved*/{0, 0},
343	  /*count*/{0, 0, 0, 0}},
344	 {/*flags*/0,
345	  /*resource*/0,
346	  /*reserved*/{0, 0},
347	  /*count*/{0, 0, 0, 0}},
348	 {/*flags*/0,
349	  /*resource*/0,
350	  /*reserved*/{0, 0},
351	  /*count*/{0, 0, 0, 0}}
352	}
353};
354
355/*
356 * XXX KDM move these into the softc.
357 */
358static int rcv_sync_msg;
359static uint8_t ctl_pause_rtr;
360
361SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
362static int worker_threads = -1;
363SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
364    &worker_threads, 1, "Number of worker threads");
365static int ctl_debug = CTL_DEBUG_NONE;
366SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
367    &ctl_debug, 0, "Enabled debug flags");
368
369/*
370 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
371 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
372 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
373 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
374 */
375#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
376
377static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
378				  int param);
379static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
380static int ctl_init(void);
381void ctl_shutdown(void);
382static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
383static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
384static void ctl_ioctl_online(void *arg);
385static void ctl_ioctl_offline(void *arg);
386static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
387static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
388static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
389static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
390static int ctl_ioctl_submit_wait(union ctl_io *io);
391static void ctl_ioctl_datamove(union ctl_io *io);
392static void ctl_ioctl_done(union ctl_io *io);
393static void ctl_ioctl_hard_startstop_callback(void *arg,
394					      struct cfi_metatask *metatask);
395static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
396static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
397			      struct ctl_ooa *ooa_hdr,
398			      struct ctl_ooa_entry *kern_entries);
399static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
400		     struct thread *td);
401static uint32_t ctl_map_lun(int port_num, uint32_t lun);
402static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
403#ifdef unused
404static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
405				   uint32_t targ_target, uint32_t targ_lun,
406				   int can_wait);
407static void ctl_kfree_io(union ctl_io *io);
408#endif /* unused */
409static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
410			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
411static int ctl_free_lun(struct ctl_lun *lun);
412static void ctl_create_lun(struct ctl_be_lun *be_lun);
413/**
414static void ctl_failover_change_pages(struct ctl_softc *softc,
415				      struct ctl_scsiio *ctsio, int master);
416**/
417
418static int ctl_do_mode_select(union ctl_io *io);
419static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
420			   uint64_t res_key, uint64_t sa_res_key,
421			   uint8_t type, uint32_t residx,
422			   struct ctl_scsiio *ctsio,
423			   struct scsi_per_res_out *cdb,
424			   struct scsi_per_res_out_parms* param);
425static void ctl_pro_preempt_other(struct ctl_lun *lun,
426				  union ctl_ha_msg *msg);
427static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
428static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
429static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
430static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
431static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
432static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
433static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
434					 int alloc_len);
435static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
436					 int alloc_len);
437static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
438static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
439static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
440static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
441static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
442static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
443static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
444    union ctl_io *pending_io, union ctl_io *ooa_io);
445static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
446				union ctl_io *starting_io);
447static int ctl_check_blocked(struct ctl_lun *lun);
448static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
449				struct ctl_lun *lun,
450				const struct ctl_cmd_entry *entry,
451				struct ctl_scsiio *ctsio);
452//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
453static void ctl_failover(void);
454static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
455			       struct ctl_scsiio *ctsio);
456static int ctl_scsiio(struct ctl_scsiio *ctsio);
457
458static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
459static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
460			    ctl_ua_type ua_type);
461static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
462			 ctl_ua_type ua_type);
463static int ctl_abort_task(union ctl_io *io);
464static int ctl_abort_task_set(union ctl_io *io);
465static int ctl_i_t_nexus_reset(union ctl_io *io);
466static void ctl_run_task(union ctl_io *io);
467#ifdef CTL_IO_DELAY
468static void ctl_datamove_timer_wakeup(void *arg);
469static void ctl_done_timer_wakeup(void *arg);
470#endif /* CTL_IO_DELAY */
471
472static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
473static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
474static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
475static void ctl_datamove_remote_write(union ctl_io *io);
476static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
477static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
478static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
479static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
480				    ctl_ha_dt_cb callback);
481static void ctl_datamove_remote_read(union ctl_io *io);
482static void ctl_datamove_remote(union ctl_io *io);
483static int ctl_process_done(union ctl_io *io);
484static void ctl_lun_thread(void *arg);
485static void ctl_thresh_thread(void *arg);
486static void ctl_work_thread(void *arg);
487static void ctl_enqueue_incoming(union ctl_io *io);
488static void ctl_enqueue_rtr(union ctl_io *io);
489static void ctl_enqueue_done(union ctl_io *io);
490static void ctl_enqueue_isc(union ctl_io *io);
491static const struct ctl_cmd_entry *
492    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
493static const struct ctl_cmd_entry *
494    ctl_validate_command(struct ctl_scsiio *ctsio);
495static int ctl_cmd_applicable(uint8_t lun_type,
496    const struct ctl_cmd_entry *entry);
497
498/*
499 * Load the serialization table.  This isn't very pretty, but is probably
500 * the easiest way to do it.
501 */
502#include "ctl_ser_table.c"
503
504/*
505 * We only need to define open, close and ioctl routines for this driver.
506 */
507static struct cdevsw ctl_cdevsw = {
508	.d_version =	D_VERSION,
509	.d_flags =	0,
510	.d_open =	ctl_open,
511	.d_close =	ctl_close,
512	.d_ioctl =	ctl_ioctl,
513	.d_name =	"ctl",
514};
515
516
517MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
518MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
519
520static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
521
522static moduledata_t ctl_moduledata = {
523	"ctl",
524	ctl_module_event_handler,
525	NULL
526};
527
528DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
529MODULE_VERSION(ctl, 1);
530
531static struct ctl_frontend ioctl_frontend =
532{
533	.name = "ioctl",
534};
535
536static void
537ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
538			    union ctl_ha_msg *msg_info)
539{
540	struct ctl_scsiio *ctsio;
541
542	if (msg_info->hdr.original_sc == NULL) {
543		printf("%s: original_sc == NULL!\n", __func__);
544		/* XXX KDM now what? */
545		return;
546	}
547
548	ctsio = &msg_info->hdr.original_sc->scsiio;
549	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
550	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
551	ctsio->io_hdr.status = msg_info->hdr.status;
552	ctsio->scsi_status = msg_info->scsi.scsi_status;
553	ctsio->sense_len = msg_info->scsi.sense_len;
554	ctsio->sense_residual = msg_info->scsi.sense_residual;
555	ctsio->residual = msg_info->scsi.residual;
556	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
557	       sizeof(ctsio->sense_data));
558	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
559	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
560	ctl_enqueue_isc((union ctl_io *)ctsio);
561}
562
563static void
564ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
565				union ctl_ha_msg *msg_info)
566{
567	struct ctl_scsiio *ctsio;
568
569	if (msg_info->hdr.serializing_sc == NULL) {
570		printf("%s: serializing_sc == NULL!\n", __func__);
571		/* XXX KDM now what? */
572		return;
573	}
574
575	ctsio = &msg_info->hdr.serializing_sc->scsiio;
576#if 0
577	/*
578	 * Attempt to catch the situation where an I/O has
579	 * been freed, and we're using it again.
580	 */
581	if (ctsio->io_hdr.io_type == 0xff) {
582		union ctl_io *tmp_io;
583		tmp_io = (union ctl_io *)ctsio;
584		printf("%s: %p use after free!\n", __func__,
585		       ctsio);
586		printf("%s: type %d msg %d cdb %x iptl: "
587		       "%d:%d:%d:%d tag 0x%04x "
588		       "flag %#x status %x\n",
589			__func__,
590			tmp_io->io_hdr.io_type,
591			tmp_io->io_hdr.msg_type,
592			tmp_io->scsiio.cdb[0],
593			tmp_io->io_hdr.nexus.initid.id,
594			tmp_io->io_hdr.nexus.targ_port,
595			tmp_io->io_hdr.nexus.targ_target.id,
596			tmp_io->io_hdr.nexus.targ_lun,
597			(tmp_io->io_hdr.io_type ==
598			CTL_IO_TASK) ?
599			tmp_io->taskio.tag_num :
600			tmp_io->scsiio.tag_num,
601		        tmp_io->io_hdr.flags,
602			tmp_io->io_hdr.status);
603	}
604#endif
605	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
606	ctl_enqueue_isc((union ctl_io *)ctsio);
607}
608
609/*
610 * ISC (Inter Shelf Communication) event handler.  Events from the HA
611 * subsystem come in here.
612 */
613static void
614ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
615{
616	struct ctl_softc *ctl_softc;
617	union ctl_io *io;
618	struct ctl_prio *presio;
619	ctl_ha_status isc_status;
620
621	ctl_softc = control_softc;
622	io = NULL;
623
624
625#if 0
626	printf("CTL: Isc Msg event %d\n", event);
627#endif
628	if (event == CTL_HA_EVT_MSG_RECV) {
629		union ctl_ha_msg msg_info;
630
631		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
632					     sizeof(msg_info), /*wait*/ 0);
633#if 0
634		printf("CTL: msg_type %d\n", msg_info.msg_type);
635#endif
636		if (isc_status != 0) {
637			printf("Error receiving message, status = %d\n",
638			       isc_status);
639			return;
640		}
641
642		switch (msg_info.hdr.msg_type) {
643		case CTL_MSG_SERIALIZE:
644#if 0
645			printf("Serialize\n");
646#endif
647			io = ctl_alloc_io_nowait(ctl_softc->othersc_pool);
648			if (io == NULL) {
649				printf("ctl_isc_event_handler: can't allocate "
650				       "ctl_io!\n");
651				/* Bad Juju */
652				/* Need to set busy and send msg back */
653				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
654				msg_info.hdr.status = CTL_SCSI_ERROR;
655				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
656				msg_info.scsi.sense_len = 0;
657			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
658				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
659				}
660				goto bailout;
661			}
662			ctl_zero_io(io);
663			// populate ctsio from msg_info
664			io->io_hdr.io_type = CTL_IO_SCSI;
665			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
666			io->io_hdr.original_sc = msg_info.hdr.original_sc;
667#if 0
668			printf("pOrig %x\n", (int)msg_info.original_sc);
669#endif
670			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
671					    CTL_FLAG_IO_ACTIVE;
672			/*
673			 * If we're in serialization-only mode, we don't
674			 * want to go through full done processing.  Thus
675			 * the COPY flag.
676			 *
677			 * XXX KDM add another flag that is more specific.
678			 */
679			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
680				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
681			io->io_hdr.nexus = msg_info.hdr.nexus;
682#if 0
683			printf("targ %d, port %d, iid %d, lun %d\n",
684			       io->io_hdr.nexus.targ_target.id,
685			       io->io_hdr.nexus.targ_port,
686			       io->io_hdr.nexus.initid.id,
687			       io->io_hdr.nexus.targ_lun);
688#endif
689			io->scsiio.tag_num = msg_info.scsi.tag_num;
690			io->scsiio.tag_type = msg_info.scsi.tag_type;
691			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
692			       CTL_MAX_CDBLEN);
693			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
694				const struct ctl_cmd_entry *entry;
695
696				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
697				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
698				io->io_hdr.flags |=
699					entry->flags & CTL_FLAG_DATA_MASK;
700			}
701			ctl_enqueue_isc(io);
702			break;
703
704		/* Performed on the Originating SC, XFER mode only */
705		case CTL_MSG_DATAMOVE: {
706			struct ctl_sg_entry *sgl;
707			int i, j;
708
709			io = msg_info.hdr.original_sc;
710			if (io == NULL) {
711				printf("%s: original_sc == NULL!\n", __func__);
712				/* XXX KDM do something here */
713				break;
714			}
715			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
716			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
717			/*
718			 * Keep track of this, we need to send it back over
719			 * when the datamove is complete.
720			 */
721			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
722
723			if (msg_info.dt.sg_sequence == 0) {
724				/*
725				 * XXX KDM we use the preallocated S/G list
726				 * here, but we'll need to change this to
727				 * dynamic allocation if we need larger S/G
728				 * lists.
729				 */
730				if (msg_info.dt.kern_sg_entries >
731				    sizeof(io->io_hdr.remote_sglist) /
732				    sizeof(io->io_hdr.remote_sglist[0])) {
733					printf("%s: number of S/G entries "
734					    "needed %u > allocated num %zd\n",
735					    __func__,
736					    msg_info.dt.kern_sg_entries,
737					    sizeof(io->io_hdr.remote_sglist)/
738					    sizeof(io->io_hdr.remote_sglist[0]));
739
740					/*
741					 * XXX KDM send a message back to
742					 * the other side to shut down the
743					 * DMA.  The error will come back
744					 * through via the normal channel.
745					 */
746					break;
747				}
748				sgl = io->io_hdr.remote_sglist;
749				memset(sgl, 0,
750				       sizeof(io->io_hdr.remote_sglist));
751
752				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
753
754				io->scsiio.kern_sg_entries =
755					msg_info.dt.kern_sg_entries;
756				io->scsiio.rem_sg_entries =
757					msg_info.dt.kern_sg_entries;
758				io->scsiio.kern_data_len =
759					msg_info.dt.kern_data_len;
760				io->scsiio.kern_total_len =
761					msg_info.dt.kern_total_len;
762				io->scsiio.kern_data_resid =
763					msg_info.dt.kern_data_resid;
764				io->scsiio.kern_rel_offset =
765					msg_info.dt.kern_rel_offset;
766				/*
767				 * Clear out per-DMA flags.
768				 */
769				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
770				/*
771				 * Add per-DMA flags that are set for this
772				 * particular DMA request.
773				 */
774				io->io_hdr.flags |= msg_info.dt.flags &
775						    CTL_FLAG_RDMA_MASK;
776			} else
777				sgl = (struct ctl_sg_entry *)
778					io->scsiio.kern_data_ptr;
779
780			for (i = msg_info.dt.sent_sg_entries, j = 0;
781			     i < (msg_info.dt.sent_sg_entries +
782			     msg_info.dt.cur_sg_entries); i++, j++) {
783				sgl[i].addr = msg_info.dt.sg_list[j].addr;
784				sgl[i].len = msg_info.dt.sg_list[j].len;
785
786#if 0
787				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
788				       __func__,
789				       msg_info.dt.sg_list[j].addr,
790				       msg_info.dt.sg_list[j].len,
791				       sgl[i].addr, sgl[i].len, j, i);
792#endif
793			}
794#if 0
795			memcpy(&sgl[msg_info.dt.sent_sg_entries],
796			       msg_info.dt.sg_list,
797			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
798#endif
799
800			/*
801			 * If this is the last piece of the I/O, we've got
802			 * the full S/G list.  Queue processing in the thread.
803			 * Otherwise wait for the next piece.
804			 */
805			if (msg_info.dt.sg_last != 0)
806				ctl_enqueue_isc(io);
807			break;
808		}
809		/* Performed on the Serializing (primary) SC, XFER mode only */
810		case CTL_MSG_DATAMOVE_DONE: {
811			if (msg_info.hdr.serializing_sc == NULL) {
812				printf("%s: serializing_sc == NULL!\n",
813				       __func__);
814				/* XXX KDM now what? */
815				break;
816			}
817			/*
818			 * We grab the sense information here in case
819			 * there was a failure, so we can return status
820			 * back to the initiator.
821			 */
822			io = msg_info.hdr.serializing_sc;
823			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
824			io->io_hdr.status = msg_info.hdr.status;
825			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
826			io->scsiio.sense_len = msg_info.scsi.sense_len;
827			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
828			io->io_hdr.port_status = msg_info.scsi.fetd_status;
829			io->scsiio.residual = msg_info.scsi.residual;
830			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
831			       sizeof(io->scsiio.sense_data));
832			ctl_enqueue_isc(io);
833			break;
834		}
835
836		/* Preformed on Originating SC, SER_ONLY mode */
837		case CTL_MSG_R2R:
838			io = msg_info.hdr.original_sc;
839			if (io == NULL) {
840				printf("%s: Major Bummer\n", __func__);
841				return;
842			} else {
843#if 0
844				printf("pOrig %x\n",(int) ctsio);
845#endif
846			}
847			io->io_hdr.msg_type = CTL_MSG_R2R;
848			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
849			ctl_enqueue_isc(io);
850			break;
851
852		/*
853		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
854		 * mode.
855		 * Performed on the Originating (i.e. secondary) SC in XFER
856		 * mode
857		 */
858		case CTL_MSG_FINISH_IO:
859			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
860				ctl_isc_handler_finish_xfer(ctl_softc,
861							    &msg_info);
862			else
863				ctl_isc_handler_finish_ser_only(ctl_softc,
864								&msg_info);
865			break;
866
867		/* Preformed on Originating SC */
868		case CTL_MSG_BAD_JUJU:
869			io = msg_info.hdr.original_sc;
870			if (io == NULL) {
871				printf("%s: Bad JUJU!, original_sc is NULL!\n",
872				       __func__);
873				break;
874			}
875			ctl_copy_sense_data(&msg_info, io);
876			/*
877			 * IO should have already been cleaned up on other
878			 * SC so clear this flag so we won't send a message
879			 * back to finish the IO there.
880			 */
881			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
882			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
883
884			/* io = msg_info.hdr.serializing_sc; */
885			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
886			ctl_enqueue_isc(io);
887			break;
888
889		/* Handle resets sent from the other side */
890		case CTL_MSG_MANAGE_TASKS: {
891			struct ctl_taskio *taskio;
892			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
893			    ctl_softc->othersc_pool);
894			if (taskio == NULL) {
895				printf("ctl_isc_event_handler: can't allocate "
896				       "ctl_io!\n");
897				/* Bad Juju */
898				/* should I just call the proper reset func
899				   here??? */
900				goto bailout;
901			}
902			ctl_zero_io((union ctl_io *)taskio);
903			taskio->io_hdr.io_type = CTL_IO_TASK;
904			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
905			taskio->io_hdr.nexus = msg_info.hdr.nexus;
906			taskio->task_action = msg_info.task.task_action;
907			taskio->tag_num = msg_info.task.tag_num;
908			taskio->tag_type = msg_info.task.tag_type;
909#ifdef CTL_TIME_IO
910			taskio->io_hdr.start_time = time_uptime;
911			getbintime(&taskio->io_hdr.start_bt);
912#if 0
913			cs_prof_gettime(&taskio->io_hdr.start_ticks);
914#endif
915#endif /* CTL_TIME_IO */
916			ctl_run_task((union ctl_io *)taskio);
917			break;
918		}
919		/* Persistent Reserve action which needs attention */
920		case CTL_MSG_PERS_ACTION:
921			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
922			    ctl_softc->othersc_pool);
923			if (presio == NULL) {
924				printf("ctl_isc_event_handler: can't allocate "
925				       "ctl_io!\n");
926				/* Bad Juju */
927				/* Need to set busy and send msg back */
928				goto bailout;
929			}
930			ctl_zero_io((union ctl_io *)presio);
931			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
932			presio->pr_msg = msg_info.pr;
933			ctl_enqueue_isc((union ctl_io *)presio);
934			break;
935		case CTL_MSG_SYNC_FE:
936			rcv_sync_msg = 1;
937			break;
938		default:
939		        printf("How did I get here?\n");
940		}
941	} else if (event == CTL_HA_EVT_MSG_SENT) {
942		if (param != CTL_HA_STATUS_SUCCESS) {
943			printf("Bad status from ctl_ha_msg_send status %d\n",
944			       param);
945		}
946		return;
947	} else if (event == CTL_HA_EVT_DISCONNECT) {
948		printf("CTL: Got a disconnect from Isc\n");
949		return;
950	} else {
951		printf("ctl_isc_event_handler: Unknown event %d\n", event);
952		return;
953	}
954
955bailout:
956	return;
957}
958
959static void
960ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
961{
962	struct scsi_sense_data *sense;
963
964	sense = &dest->scsiio.sense_data;
965	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
966	dest->scsiio.scsi_status = src->scsi.scsi_status;
967	dest->scsiio.sense_len = src->scsi.sense_len;
968	dest->io_hdr.status = src->hdr.status;
969}
970
971static int
972ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
973{
974	struct ctl_softc *softc = (struct ctl_softc *)arg1;
975	struct ctl_lun *lun;
976	int error, value, i;
977
978	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
979		value = 0;
980	else
981		value = 1;
982
983	error = sysctl_handle_int(oidp, &value, 0, req);
984	if ((error != 0) || (req->newptr == NULL))
985		return (error);
986
987	mtx_lock(&softc->ctl_lock);
988	if (value == 0)
989		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
990	else
991		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
992	STAILQ_FOREACH(lun, &softc->lun_list, links) {
993		mtx_lock(&lun->lun_lock);
994		for (i = 0; i < CTL_MAX_INITIATORS; i++)
995			lun->pending_ua[i] |= CTL_UA_ASYM_ACC_CHANGE;
996		mtx_unlock(&lun->lun_lock);
997	}
998	mtx_unlock(&softc->ctl_lock);
999	return (0);
1000}
1001
1002static int
1003ctl_init(void)
1004{
1005	struct ctl_softc *softc;
1006	void *other_pool;
1007	struct ctl_port *port;
1008	int i, error, retval;
1009	//int isc_retval;
1010
1011	retval = 0;
1012	ctl_pause_rtr = 0;
1013        rcv_sync_msg = 0;
1014
1015	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1016			       M_WAITOK | M_ZERO);
1017	softc = control_softc;
1018
1019	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1020			      "cam/ctl");
1021
1022	softc->dev->si_drv1 = softc;
1023
1024	/*
1025	 * By default, return a "bad LUN" peripheral qualifier for unknown
1026	 * LUNs.  The user can override this default using the tunable or
1027	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1028	 */
1029	softc->inquiry_pq_no_lun = 1;
1030	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1031			  &softc->inquiry_pq_no_lun);
1032	sysctl_ctx_init(&softc->sysctl_ctx);
1033	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1034		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1035		CTLFLAG_RD, 0, "CAM Target Layer");
1036
1037	if (softc->sysctl_tree == NULL) {
1038		printf("%s: unable to allocate sysctl tree\n", __func__);
1039		destroy_dev(softc->dev);
1040		free(control_softc, M_DEVBUF);
1041		control_softc = NULL;
1042		return (ENOMEM);
1043	}
1044
1045	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1046		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1047		       "inquiry_pq_no_lun", CTLFLAG_RW,
1048		       &softc->inquiry_pq_no_lun, 0,
1049		       "Report no lun possible for invalid LUNs");
1050
1051	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1052	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1053	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1054	softc->open_count = 0;
1055
1056	/*
1057	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1058	 * the drive.
1059	 */
1060	softc->flags = CTL_FLAG_REAL_SYNC;
1061
1062	/*
1063	 * In Copan's HA scheme, the "master" and "slave" roles are
1064	 * figured out through the slot the controller is in.  Although it
1065	 * is an active/active system, someone has to be in charge.
1066	 */
1067	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1068	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1069	    "HA head ID (0 - no HA)");
1070	if (softc->ha_id == 0) {
1071		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1072		softc->is_single = 1;
1073		softc->port_offset = 0;
1074	} else
1075		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1076	softc->persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1077
1078	/*
1079	 * XXX KDM need to figure out where we want to get our target ID
1080	 * and WWID.  Is it different on each port?
1081	 */
1082	softc->target.id = 0;
1083	softc->target.wwid[0] = 0x12345678;
1084	softc->target.wwid[1] = 0x87654321;
1085	STAILQ_INIT(&softc->lun_list);
1086	STAILQ_INIT(&softc->pending_lun_queue);
1087	STAILQ_INIT(&softc->fe_list);
1088	STAILQ_INIT(&softc->port_list);
1089	STAILQ_INIT(&softc->be_list);
1090	ctl_tpc_init(softc);
1091
1092	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1093	                    &other_pool) != 0)
1094	{
1095		printf("ctl: can't allocate %d entry other SC pool, "
1096		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1097		return (ENOMEM);
1098	}
1099	softc->othersc_pool = other_pool;
1100
1101	if (worker_threads <= 0)
1102		worker_threads = max(1, mp_ncpus / 4);
1103	if (worker_threads > CTL_MAX_THREADS)
1104		worker_threads = CTL_MAX_THREADS;
1105
1106	for (i = 0; i < worker_threads; i++) {
1107		struct ctl_thread *thr = &softc->threads[i];
1108
1109		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1110		thr->ctl_softc = softc;
1111		STAILQ_INIT(&thr->incoming_queue);
1112		STAILQ_INIT(&thr->rtr_queue);
1113		STAILQ_INIT(&thr->done_queue);
1114		STAILQ_INIT(&thr->isc_queue);
1115
1116		error = kproc_kthread_add(ctl_work_thread, thr,
1117		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1118		if (error != 0) {
1119			printf("error creating CTL work thread!\n");
1120			ctl_pool_free(other_pool);
1121			return (error);
1122		}
1123	}
1124	error = kproc_kthread_add(ctl_lun_thread, softc,
1125	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1126	if (error != 0) {
1127		printf("error creating CTL lun thread!\n");
1128		ctl_pool_free(other_pool);
1129		return (error);
1130	}
1131	error = kproc_kthread_add(ctl_thresh_thread, softc,
1132	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1133	if (error != 0) {
1134		printf("error creating CTL threshold thread!\n");
1135		ctl_pool_free(other_pool);
1136		return (error);
1137	}
1138	if (bootverbose)
1139		printf("ctl: CAM Target Layer loaded\n");
1140
1141	/*
1142	 * Initialize the ioctl front end.
1143	 */
1144	ctl_frontend_register(&ioctl_frontend);
1145	port = &softc->ioctl_info.port;
1146	port->frontend = &ioctl_frontend;
1147	sprintf(softc->ioctl_info.port_name, "ioctl");
1148	port->port_type = CTL_PORT_IOCTL;
1149	port->num_requested_ctl_io = 100;
1150	port->port_name = softc->ioctl_info.port_name;
1151	port->port_online = ctl_ioctl_online;
1152	port->port_offline = ctl_ioctl_offline;
1153	port->onoff_arg = &softc->ioctl_info;
1154	port->lun_enable = ctl_ioctl_lun_enable;
1155	port->lun_disable = ctl_ioctl_lun_disable;
1156	port->targ_lun_arg = &softc->ioctl_info;
1157	port->fe_datamove = ctl_ioctl_datamove;
1158	port->fe_done = ctl_ioctl_done;
1159	port->max_targets = 15;
1160	port->max_target_id = 15;
1161
1162	if (ctl_port_register(&softc->ioctl_info.port) != 0) {
1163		printf("ctl: ioctl front end registration failed, will "
1164		       "continue anyway\n");
1165	}
1166
1167	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1168	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1169	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1170
1171#ifdef CTL_IO_DELAY
1172	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1173		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1174		       sizeof(struct callout), CTL_TIMER_BYTES);
1175		return (EINVAL);
1176	}
1177#endif /* CTL_IO_DELAY */
1178
1179	return (0);
1180}
1181
1182void
1183ctl_shutdown(void)
1184{
1185	struct ctl_softc *softc;
1186	struct ctl_lun *lun, *next_lun;
1187
1188	softc = (struct ctl_softc *)control_softc;
1189
1190	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1191		printf("ctl: ioctl front end deregistration failed\n");
1192
1193	mtx_lock(&softc->ctl_lock);
1194
1195	/*
1196	 * Free up each LUN.
1197	 */
1198	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1199		next_lun = STAILQ_NEXT(lun, links);
1200		ctl_free_lun(lun);
1201	}
1202
1203	mtx_unlock(&softc->ctl_lock);
1204
1205	ctl_frontend_deregister(&ioctl_frontend);
1206
1207#if 0
1208	ctl_shutdown_thread(softc->work_thread);
1209	mtx_destroy(&softc->queue_lock);
1210#endif
1211
1212	ctl_tpc_shutdown(softc);
1213	uma_zdestroy(softc->io_zone);
1214	mtx_destroy(&softc->ctl_lock);
1215
1216	destroy_dev(softc->dev);
1217
1218	sysctl_ctx_free(&softc->sysctl_ctx);
1219
1220	free(control_softc, M_DEVBUF);
1221	control_softc = NULL;
1222
1223	if (bootverbose)
1224		printf("ctl: CAM Target Layer unloaded\n");
1225}
1226
1227static int
1228ctl_module_event_handler(module_t mod, int what, void *arg)
1229{
1230
1231	switch (what) {
1232	case MOD_LOAD:
1233		return (ctl_init());
1234	case MOD_UNLOAD:
1235		return (EBUSY);
1236	default:
1237		return (EOPNOTSUPP);
1238	}
1239}
1240
1241/*
1242 * XXX KDM should we do some access checks here?  Bump a reference count to
1243 * prevent a CTL module from being unloaded while someone has it open?
1244 */
1245static int
1246ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1247{
1248	return (0);
1249}
1250
1251static int
1252ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1253{
1254	return (0);
1255}
1256
1257int
1258ctl_port_enable(ctl_port_type port_type)
1259{
1260	struct ctl_softc *softc = control_softc;
1261	struct ctl_port *port;
1262
1263	if (softc->is_single == 0) {
1264		union ctl_ha_msg msg_info;
1265		int isc_retval;
1266
1267#if 0
1268		printf("%s: HA mode, synchronizing frontend enable\n",
1269		        __func__);
1270#endif
1271		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1272	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1273		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1274			printf("Sync msg send error retval %d\n", isc_retval);
1275		}
1276		if (!rcv_sync_msg) {
1277			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1278			        sizeof(msg_info), 1);
1279		}
1280#if 0
1281        	printf("CTL:Frontend Enable\n");
1282	} else {
1283		printf("%s: single mode, skipping frontend synchronization\n",
1284		        __func__);
1285#endif
1286	}
1287
1288	STAILQ_FOREACH(port, &softc->port_list, links) {
1289		if (port_type & port->port_type)
1290		{
1291#if 0
1292			printf("port %d\n", port->targ_port);
1293#endif
1294			ctl_port_online(port);
1295		}
1296	}
1297
1298	return (0);
1299}
1300
1301int
1302ctl_port_disable(ctl_port_type port_type)
1303{
1304	struct ctl_softc *softc;
1305	struct ctl_port *port;
1306
1307	softc = control_softc;
1308
1309	STAILQ_FOREACH(port, &softc->port_list, links) {
1310		if (port_type & port->port_type)
1311			ctl_port_offline(port);
1312	}
1313
1314	return (0);
1315}
1316
1317/*
1318 * Returns 0 for success, 1 for failure.
1319 * Currently the only failure mode is if there aren't enough entries
1320 * allocated.  So, in case of a failure, look at num_entries_dropped,
1321 * reallocate and try again.
1322 */
1323int
1324ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1325	      int *num_entries_filled, int *num_entries_dropped,
1326	      ctl_port_type port_type, int no_virtual)
1327{
1328	struct ctl_softc *softc;
1329	struct ctl_port *port;
1330	int entries_dropped, entries_filled;
1331	int retval;
1332	int i;
1333
1334	softc = control_softc;
1335
1336	retval = 0;
1337	entries_filled = 0;
1338	entries_dropped = 0;
1339
1340	i = 0;
1341	mtx_lock(&softc->ctl_lock);
1342	STAILQ_FOREACH(port, &softc->port_list, links) {
1343		struct ctl_port_entry *entry;
1344
1345		if ((port->port_type & port_type) == 0)
1346			continue;
1347
1348		if ((no_virtual != 0)
1349		 && (port->virtual_port != 0))
1350			continue;
1351
1352		if (entries_filled >= num_entries_alloced) {
1353			entries_dropped++;
1354			continue;
1355		}
1356		entry = &entries[i];
1357
1358		entry->port_type = port->port_type;
1359		strlcpy(entry->port_name, port->port_name,
1360			sizeof(entry->port_name));
1361		entry->physical_port = port->physical_port;
1362		entry->virtual_port = port->virtual_port;
1363		entry->wwnn = port->wwnn;
1364		entry->wwpn = port->wwpn;
1365
1366		i++;
1367		entries_filled++;
1368	}
1369
1370	mtx_unlock(&softc->ctl_lock);
1371
1372	if (entries_dropped > 0)
1373		retval = 1;
1374
1375	*num_entries_dropped = entries_dropped;
1376	*num_entries_filled = entries_filled;
1377
1378	return (retval);
1379}
1380
1381static void
1382ctl_ioctl_online(void *arg)
1383{
1384	struct ctl_ioctl_info *ioctl_info;
1385
1386	ioctl_info = (struct ctl_ioctl_info *)arg;
1387
1388	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1389}
1390
1391static void
1392ctl_ioctl_offline(void *arg)
1393{
1394	struct ctl_ioctl_info *ioctl_info;
1395
1396	ioctl_info = (struct ctl_ioctl_info *)arg;
1397
1398	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1399}
1400
1401/*
1402 * Remove an initiator by port number and initiator ID.
1403 * Returns 0 for success, -1 for failure.
1404 */
1405int
1406ctl_remove_initiator(struct ctl_port *port, int iid)
1407{
1408	struct ctl_softc *softc = control_softc;
1409
1410	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1411
1412	if (iid > CTL_MAX_INIT_PER_PORT) {
1413		printf("%s: initiator ID %u > maximun %u!\n",
1414		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1415		return (-1);
1416	}
1417
1418	mtx_lock(&softc->ctl_lock);
1419	port->wwpn_iid[iid].in_use--;
1420	port->wwpn_iid[iid].last_use = time_uptime;
1421	mtx_unlock(&softc->ctl_lock);
1422
1423	return (0);
1424}
1425
1426/*
1427 * Add an initiator to the initiator map.
1428 * Returns iid for success, < 0 for failure.
1429 */
1430int
1431ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1432{
1433	struct ctl_softc *softc = control_softc;
1434	time_t best_time;
1435	int i, best;
1436
1437	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1438
1439	if (iid >= CTL_MAX_INIT_PER_PORT) {
1440		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1441		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1442		free(name, M_CTL);
1443		return (-1);
1444	}
1445
1446	mtx_lock(&softc->ctl_lock);
1447
1448	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1449		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1450			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1451				iid = i;
1452				break;
1453			}
1454			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1455			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1456				iid = i;
1457				break;
1458			}
1459		}
1460	}
1461
1462	if (iid < 0) {
1463		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1464			if (port->wwpn_iid[i].in_use == 0 &&
1465			    port->wwpn_iid[i].wwpn == 0 &&
1466			    port->wwpn_iid[i].name == NULL) {
1467				iid = i;
1468				break;
1469			}
1470		}
1471	}
1472
1473	if (iid < 0) {
1474		best = -1;
1475		best_time = INT32_MAX;
1476		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1477			if (port->wwpn_iid[i].in_use == 0) {
1478				if (port->wwpn_iid[i].last_use < best_time) {
1479					best = i;
1480					best_time = port->wwpn_iid[i].last_use;
1481				}
1482			}
1483		}
1484		iid = best;
1485	}
1486
1487	if (iid < 0) {
1488		mtx_unlock(&softc->ctl_lock);
1489		free(name, M_CTL);
1490		return (-2);
1491	}
1492
1493	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1494		/*
1495		 * This is not an error yet.
1496		 */
1497		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1498#if 0
1499			printf("%s: port %d iid %u WWPN %#jx arrived"
1500			    " again\n", __func__, port->targ_port,
1501			    iid, (uintmax_t)wwpn);
1502#endif
1503			goto take;
1504		}
1505		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1506		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1507#if 0
1508			printf("%s: port %d iid %u name '%s' arrived"
1509			    " again\n", __func__, port->targ_port,
1510			    iid, name);
1511#endif
1512			goto take;
1513		}
1514
1515		/*
1516		 * This is an error, but what do we do about it?  The
1517		 * driver is telling us we have a new WWPN for this
1518		 * initiator ID, so we pretty much need to use it.
1519		 */
1520		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1521		    " but WWPN %#jx '%s' is still at that address\n",
1522		    __func__, port->targ_port, iid, wwpn, name,
1523		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1524		    port->wwpn_iid[iid].name);
1525
1526		/*
1527		 * XXX KDM clear have_ca and ua_pending on each LUN for
1528		 * this initiator.
1529		 */
1530	}
1531take:
1532	free(port->wwpn_iid[iid].name, M_CTL);
1533	port->wwpn_iid[iid].name = name;
1534	port->wwpn_iid[iid].wwpn = wwpn;
1535	port->wwpn_iid[iid].in_use++;
1536	mtx_unlock(&softc->ctl_lock);
1537
1538	return (iid);
1539}
1540
1541static int
1542ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1543{
1544	int len;
1545
1546	switch (port->port_type) {
1547	case CTL_PORT_FC:
1548	{
1549		struct scsi_transportid_fcp *id =
1550		    (struct scsi_transportid_fcp *)buf;
1551		if (port->wwpn_iid[iid].wwpn == 0)
1552			return (0);
1553		memset(id, 0, sizeof(*id));
1554		id->format_protocol = SCSI_PROTO_FC;
1555		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1556		return (sizeof(*id));
1557	}
1558	case CTL_PORT_ISCSI:
1559	{
1560		struct scsi_transportid_iscsi_port *id =
1561		    (struct scsi_transportid_iscsi_port *)buf;
1562		if (port->wwpn_iid[iid].name == NULL)
1563			return (0);
1564		memset(id, 0, 256);
1565		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1566		    SCSI_PROTO_ISCSI;
1567		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1568		len = roundup2(min(len, 252), 4);
1569		scsi_ulto2b(len, id->additional_length);
1570		return (sizeof(*id) + len);
1571	}
1572	case CTL_PORT_SAS:
1573	{
1574		struct scsi_transportid_sas *id =
1575		    (struct scsi_transportid_sas *)buf;
1576		if (port->wwpn_iid[iid].wwpn == 0)
1577			return (0);
1578		memset(id, 0, sizeof(*id));
1579		id->format_protocol = SCSI_PROTO_SAS;
1580		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1581		return (sizeof(*id));
1582	}
1583	default:
1584	{
1585		struct scsi_transportid_spi *id =
1586		    (struct scsi_transportid_spi *)buf;
1587		memset(id, 0, sizeof(*id));
1588		id->format_protocol = SCSI_PROTO_SPI;
1589		scsi_ulto2b(iid, id->scsi_addr);
1590		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1591		return (sizeof(*id));
1592	}
1593	}
1594}
1595
1596static int
1597ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1598{
1599	return (0);
1600}
1601
1602static int
1603ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1604{
1605	return (0);
1606}
1607
1608/*
1609 * Data movement routine for the CTL ioctl frontend port.
1610 */
1611static int
1612ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1613{
1614	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1615	struct ctl_sg_entry ext_entry, kern_entry;
1616	int ext_sglen, ext_sg_entries, kern_sg_entries;
1617	int ext_sg_start, ext_offset;
1618	int len_to_copy, len_copied;
1619	int kern_watermark, ext_watermark;
1620	int ext_sglist_malloced;
1621	int i, j;
1622
1623	ext_sglist_malloced = 0;
1624	ext_sg_start = 0;
1625	ext_offset = 0;
1626
1627	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1628
1629	/*
1630	 * If this flag is set, fake the data transfer.
1631	 */
1632	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1633		ctsio->ext_data_filled = ctsio->ext_data_len;
1634		goto bailout;
1635	}
1636
1637	/*
1638	 * To simplify things here, if we have a single buffer, stick it in
1639	 * a S/G entry and just make it a single entry S/G list.
1640	 */
1641	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1642		int len_seen;
1643
1644		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1645
1646		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1647							   M_WAITOK);
1648		ext_sglist_malloced = 1;
1649		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1650				   ext_sglen) != 0) {
1651			ctl_set_internal_failure(ctsio,
1652						 /*sks_valid*/ 0,
1653						 /*retry_count*/ 0);
1654			goto bailout;
1655		}
1656		ext_sg_entries = ctsio->ext_sg_entries;
1657		len_seen = 0;
1658		for (i = 0; i < ext_sg_entries; i++) {
1659			if ((len_seen + ext_sglist[i].len) >=
1660			     ctsio->ext_data_filled) {
1661				ext_sg_start = i;
1662				ext_offset = ctsio->ext_data_filled - len_seen;
1663				break;
1664			}
1665			len_seen += ext_sglist[i].len;
1666		}
1667	} else {
1668		ext_sglist = &ext_entry;
1669		ext_sglist->addr = ctsio->ext_data_ptr;
1670		ext_sglist->len = ctsio->ext_data_len;
1671		ext_sg_entries = 1;
1672		ext_sg_start = 0;
1673		ext_offset = ctsio->ext_data_filled;
1674	}
1675
1676	if (ctsio->kern_sg_entries > 0) {
1677		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1678		kern_sg_entries = ctsio->kern_sg_entries;
1679	} else {
1680		kern_sglist = &kern_entry;
1681		kern_sglist->addr = ctsio->kern_data_ptr;
1682		kern_sglist->len = ctsio->kern_data_len;
1683		kern_sg_entries = 1;
1684	}
1685
1686
1687	kern_watermark = 0;
1688	ext_watermark = ext_offset;
1689	len_copied = 0;
1690	for (i = ext_sg_start, j = 0;
1691	     i < ext_sg_entries && j < kern_sg_entries;) {
1692		uint8_t *ext_ptr, *kern_ptr;
1693
1694		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1695				      kern_sglist[j].len - kern_watermark);
1696
1697		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1698		ext_ptr = ext_ptr + ext_watermark;
1699		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1700			/*
1701			 * XXX KDM fix this!
1702			 */
1703			panic("need to implement bus address support");
1704#if 0
1705			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1706#endif
1707		} else
1708			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1709		kern_ptr = kern_ptr + kern_watermark;
1710
1711		kern_watermark += len_to_copy;
1712		ext_watermark += len_to_copy;
1713
1714		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1715		     CTL_FLAG_DATA_IN) {
1716			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1717					 "bytes to user\n", len_to_copy));
1718			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1719					 "to %p\n", kern_ptr, ext_ptr));
1720			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1721				ctl_set_internal_failure(ctsio,
1722							 /*sks_valid*/ 0,
1723							 /*retry_count*/ 0);
1724				goto bailout;
1725			}
1726		} else {
1727			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1728					 "bytes from user\n", len_to_copy));
1729			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1730					 "to %p\n", ext_ptr, kern_ptr));
1731			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1732				ctl_set_internal_failure(ctsio,
1733							 /*sks_valid*/ 0,
1734							 /*retry_count*/0);
1735				goto bailout;
1736			}
1737		}
1738
1739		len_copied += len_to_copy;
1740
1741		if (ext_sglist[i].len == ext_watermark) {
1742			i++;
1743			ext_watermark = 0;
1744		}
1745
1746		if (kern_sglist[j].len == kern_watermark) {
1747			j++;
1748			kern_watermark = 0;
1749		}
1750	}
1751
1752	ctsio->ext_data_filled += len_copied;
1753
1754	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1755			 "kern_sg_entries: %d\n", ext_sg_entries,
1756			 kern_sg_entries));
1757	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1758			 "kern_data_len = %d\n", ctsio->ext_data_len,
1759			 ctsio->kern_data_len));
1760
1761
1762	/* XXX KDM set residual?? */
1763bailout:
1764
1765	if (ext_sglist_malloced != 0)
1766		free(ext_sglist, M_CTL);
1767
1768	return (CTL_RETVAL_COMPLETE);
1769}
1770
1771/*
1772 * Serialize a command that went down the "wrong" side, and so was sent to
1773 * this controller for execution.  The logic is a little different than the
1774 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1775 * sent back to the other side, but in the success case, we execute the
1776 * command on this side (XFER mode) or tell the other side to execute it
1777 * (SER_ONLY mode).
1778 */
1779static int
1780ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1781{
1782	struct ctl_softc *ctl_softc;
1783	union ctl_ha_msg msg_info;
1784	struct ctl_lun *lun;
1785	int retval = 0;
1786	uint32_t targ_lun;
1787
1788	ctl_softc = control_softc;
1789
1790	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1791	lun = ctl_softc->ctl_luns[targ_lun];
1792	if (lun==NULL)
1793	{
1794		/*
1795		 * Why isn't LUN defined? The other side wouldn't
1796		 * send a cmd if the LUN is undefined.
1797		 */
1798		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1799
1800		/* "Logical unit not supported" */
1801		ctl_set_sense_data(&msg_info.scsi.sense_data,
1802				   lun,
1803				   /*sense_format*/SSD_TYPE_NONE,
1804				   /*current_error*/ 1,
1805				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1806				   /*asc*/ 0x25,
1807				   /*ascq*/ 0x00,
1808				   SSD_ELEM_NONE);
1809
1810		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1811		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1812		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1813		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1814		msg_info.hdr.serializing_sc = NULL;
1815		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1816	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1817				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1818		}
1819		return(1);
1820
1821	}
1822
1823	mtx_lock(&lun->lun_lock);
1824    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1825
1826	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1827		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1828		 ooa_links))) {
1829	case CTL_ACTION_BLOCK:
1830		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1831		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1832				  blocked_links);
1833		break;
1834	case CTL_ACTION_PASS:
1835	case CTL_ACTION_SKIP:
1836		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1837			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1838			ctl_enqueue_rtr((union ctl_io *)ctsio);
1839		} else {
1840
1841			/* send msg back to other side */
1842			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1843			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1844			msg_info.hdr.msg_type = CTL_MSG_R2R;
1845#if 0
1846			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1847#endif
1848		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1849			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1850			}
1851		}
1852		break;
1853	case CTL_ACTION_OVERLAP:
1854		/* OVERLAPPED COMMANDS ATTEMPTED */
1855		ctl_set_sense_data(&msg_info.scsi.sense_data,
1856				   lun,
1857				   /*sense_format*/SSD_TYPE_NONE,
1858				   /*current_error*/ 1,
1859				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1860				   /*asc*/ 0x4E,
1861				   /*ascq*/ 0x00,
1862				   SSD_ELEM_NONE);
1863
1864		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1865		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1866		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1867		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1868		msg_info.hdr.serializing_sc = NULL;
1869		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1870#if 0
1871		printf("BAD JUJU:Major Bummer Overlap\n");
1872#endif
1873		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1874		retval = 1;
1875		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1876		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1877		}
1878		break;
1879	case CTL_ACTION_OVERLAP_TAG:
1880		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1881		ctl_set_sense_data(&msg_info.scsi.sense_data,
1882				   lun,
1883				   /*sense_format*/SSD_TYPE_NONE,
1884				   /*current_error*/ 1,
1885				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1886				   /*asc*/ 0x4D,
1887				   /*ascq*/ ctsio->tag_num & 0xff,
1888				   SSD_ELEM_NONE);
1889
1890		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1891		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1892		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1893		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1894		msg_info.hdr.serializing_sc = NULL;
1895		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1896#if 0
1897		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1898#endif
1899		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1900		retval = 1;
1901		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1902		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1903		}
1904		break;
1905	case CTL_ACTION_ERROR:
1906	default:
1907		/* "Internal target failure" */
1908		ctl_set_sense_data(&msg_info.scsi.sense_data,
1909				   lun,
1910				   /*sense_format*/SSD_TYPE_NONE,
1911				   /*current_error*/ 1,
1912				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1913				   /*asc*/ 0x44,
1914				   /*ascq*/ 0x00,
1915				   SSD_ELEM_NONE);
1916
1917		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1918		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1919		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1920		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1921		msg_info.hdr.serializing_sc = NULL;
1922		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1923#if 0
1924		printf("BAD JUJU:Major Bummer HW Error\n");
1925#endif
1926		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1927		retval = 1;
1928		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1929		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1930		}
1931		break;
1932	}
1933	mtx_unlock(&lun->lun_lock);
1934	return (retval);
1935}
1936
1937static int
1938ctl_ioctl_submit_wait(union ctl_io *io)
1939{
1940	struct ctl_fe_ioctl_params params;
1941	ctl_fe_ioctl_state last_state;
1942	int done, retval;
1943
1944	retval = 0;
1945
1946	bzero(&params, sizeof(params));
1947
1948	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1949	cv_init(&params.sem, "ctlioccv");
1950	params.state = CTL_IOCTL_INPROG;
1951	last_state = params.state;
1952
1953	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1954
1955	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1956
1957	/* This shouldn't happen */
1958	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1959		return (retval);
1960
1961	done = 0;
1962
1963	do {
1964		mtx_lock(&params.ioctl_mtx);
1965		/*
1966		 * Check the state here, and don't sleep if the state has
1967		 * already changed (i.e. wakeup has already occured, but we
1968		 * weren't waiting yet).
1969		 */
1970		if (params.state == last_state) {
1971			/* XXX KDM cv_wait_sig instead? */
1972			cv_wait(&params.sem, &params.ioctl_mtx);
1973		}
1974		last_state = params.state;
1975
1976		switch (params.state) {
1977		case CTL_IOCTL_INPROG:
1978			/* Why did we wake up? */
1979			/* XXX KDM error here? */
1980			mtx_unlock(&params.ioctl_mtx);
1981			break;
1982		case CTL_IOCTL_DATAMOVE:
1983			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1984
1985			/*
1986			 * change last_state back to INPROG to avoid
1987			 * deadlock on subsequent data moves.
1988			 */
1989			params.state = last_state = CTL_IOCTL_INPROG;
1990
1991			mtx_unlock(&params.ioctl_mtx);
1992			ctl_ioctl_do_datamove(&io->scsiio);
1993			/*
1994			 * Note that in some cases, most notably writes,
1995			 * this will queue the I/O and call us back later.
1996			 * In other cases, generally reads, this routine
1997			 * will immediately call back and wake us up,
1998			 * probably using our own context.
1999			 */
2000			io->scsiio.be_move_done(io);
2001			break;
2002		case CTL_IOCTL_DONE:
2003			mtx_unlock(&params.ioctl_mtx);
2004			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
2005			done = 1;
2006			break;
2007		default:
2008			mtx_unlock(&params.ioctl_mtx);
2009			/* XXX KDM error here? */
2010			break;
2011		}
2012	} while (done == 0);
2013
2014	mtx_destroy(&params.ioctl_mtx);
2015	cv_destroy(&params.sem);
2016
2017	return (CTL_RETVAL_COMPLETE);
2018}
2019
2020static void
2021ctl_ioctl_datamove(union ctl_io *io)
2022{
2023	struct ctl_fe_ioctl_params *params;
2024
2025	params = (struct ctl_fe_ioctl_params *)
2026		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2027
2028	mtx_lock(&params->ioctl_mtx);
2029	params->state = CTL_IOCTL_DATAMOVE;
2030	cv_broadcast(&params->sem);
2031	mtx_unlock(&params->ioctl_mtx);
2032}
2033
2034static void
2035ctl_ioctl_done(union ctl_io *io)
2036{
2037	struct ctl_fe_ioctl_params *params;
2038
2039	params = (struct ctl_fe_ioctl_params *)
2040		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2041
2042	mtx_lock(&params->ioctl_mtx);
2043	params->state = CTL_IOCTL_DONE;
2044	cv_broadcast(&params->sem);
2045	mtx_unlock(&params->ioctl_mtx);
2046}
2047
2048static void
2049ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2050{
2051	struct ctl_fe_ioctl_startstop_info *sd_info;
2052
2053	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2054
2055	sd_info->hs_info.status = metatask->status;
2056	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2057	sd_info->hs_info.luns_complete =
2058		metatask->taskinfo.startstop.luns_complete;
2059	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2060
2061	cv_broadcast(&sd_info->sem);
2062}
2063
2064static void
2065ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2066{
2067	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2068
2069	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2070
2071	mtx_lock(fe_bbr_info->lock);
2072	fe_bbr_info->bbr_info->status = metatask->status;
2073	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2074	fe_bbr_info->wakeup_done = 1;
2075	mtx_unlock(fe_bbr_info->lock);
2076
2077	cv_broadcast(&fe_bbr_info->sem);
2078}
2079
2080/*
2081 * Returns 0 for success, errno for failure.
2082 */
2083static int
2084ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2085		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2086{
2087	union ctl_io *io;
2088	int retval;
2089
2090	retval = 0;
2091
2092	mtx_lock(&lun->lun_lock);
2093	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2094	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2095	     ooa_links)) {
2096		struct ctl_ooa_entry *entry;
2097
2098		/*
2099		 * If we've got more than we can fit, just count the
2100		 * remaining entries.
2101		 */
2102		if (*cur_fill_num >= ooa_hdr->alloc_num)
2103			continue;
2104
2105		entry = &kern_entries[*cur_fill_num];
2106
2107		entry->tag_num = io->scsiio.tag_num;
2108		entry->lun_num = lun->lun;
2109#ifdef CTL_TIME_IO
2110		entry->start_bt = io->io_hdr.start_bt;
2111#endif
2112		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2113		entry->cdb_len = io->scsiio.cdb_len;
2114		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2115			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2116
2117		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2118			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2119
2120		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2121			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2122
2123		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2124			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2125
2126		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2127			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2128	}
2129	mtx_unlock(&lun->lun_lock);
2130
2131	return (retval);
2132}
2133
2134static void *
2135ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2136		 size_t error_str_len)
2137{
2138	void *kptr;
2139
2140	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2141
2142	if (copyin(user_addr, kptr, len) != 0) {
2143		snprintf(error_str, error_str_len, "Error copying %d bytes "
2144			 "from user address %p to kernel address %p", len,
2145			 user_addr, kptr);
2146		free(kptr, M_CTL);
2147		return (NULL);
2148	}
2149
2150	return (kptr);
2151}
2152
2153static void
2154ctl_free_args(int num_args, struct ctl_be_arg *args)
2155{
2156	int i;
2157
2158	if (args == NULL)
2159		return;
2160
2161	for (i = 0; i < num_args; i++) {
2162		free(args[i].kname, M_CTL);
2163		free(args[i].kvalue, M_CTL);
2164	}
2165
2166	free(args, M_CTL);
2167}
2168
2169static struct ctl_be_arg *
2170ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2171		char *error_str, size_t error_str_len)
2172{
2173	struct ctl_be_arg *args;
2174	int i;
2175
2176	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2177				error_str, error_str_len);
2178
2179	if (args == NULL)
2180		goto bailout;
2181
2182	for (i = 0; i < num_args; i++) {
2183		args[i].kname = NULL;
2184		args[i].kvalue = NULL;
2185	}
2186
2187	for (i = 0; i < num_args; i++) {
2188		uint8_t *tmpptr;
2189
2190		args[i].kname = ctl_copyin_alloc(args[i].name,
2191			args[i].namelen, error_str, error_str_len);
2192		if (args[i].kname == NULL)
2193			goto bailout;
2194
2195		if (args[i].kname[args[i].namelen - 1] != '\0') {
2196			snprintf(error_str, error_str_len, "Argument %d "
2197				 "name is not NUL-terminated", i);
2198			goto bailout;
2199		}
2200
2201		if (args[i].flags & CTL_BEARG_RD) {
2202			tmpptr = ctl_copyin_alloc(args[i].value,
2203				args[i].vallen, error_str, error_str_len);
2204			if (tmpptr == NULL)
2205				goto bailout;
2206			if ((args[i].flags & CTL_BEARG_ASCII)
2207			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2208				snprintf(error_str, error_str_len, "Argument "
2209				    "%d value is not NUL-terminated", i);
2210				goto bailout;
2211			}
2212			args[i].kvalue = tmpptr;
2213		} else {
2214			args[i].kvalue = malloc(args[i].vallen,
2215			    M_CTL, M_WAITOK | M_ZERO);
2216		}
2217	}
2218
2219	return (args);
2220bailout:
2221
2222	ctl_free_args(num_args, args);
2223
2224	return (NULL);
2225}
2226
2227static void
2228ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2229{
2230	int i;
2231
2232	for (i = 0; i < num_args; i++) {
2233		if (args[i].flags & CTL_BEARG_WR)
2234			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2235	}
2236}
2237
2238/*
2239 * Escape characters that are illegal or not recommended in XML.
2240 */
2241int
2242ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2243{
2244	char *end = str + size;
2245	int retval;
2246
2247	retval = 0;
2248
2249	for (; *str && str < end; str++) {
2250		switch (*str) {
2251		case '&':
2252			retval = sbuf_printf(sb, "&amp;");
2253			break;
2254		case '>':
2255			retval = sbuf_printf(sb, "&gt;");
2256			break;
2257		case '<':
2258			retval = sbuf_printf(sb, "&lt;");
2259			break;
2260		default:
2261			retval = sbuf_putc(sb, *str);
2262			break;
2263		}
2264
2265		if (retval != 0)
2266			break;
2267
2268	}
2269
2270	return (retval);
2271}
2272
2273static void
2274ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2275{
2276	struct scsi_vpd_id_descriptor *desc;
2277	int i;
2278
2279	if (id == NULL || id->len < 4)
2280		return;
2281	desc = (struct scsi_vpd_id_descriptor *)id->data;
2282	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2283	case SVPD_ID_TYPE_T10:
2284		sbuf_printf(sb, "t10.");
2285		break;
2286	case SVPD_ID_TYPE_EUI64:
2287		sbuf_printf(sb, "eui.");
2288		break;
2289	case SVPD_ID_TYPE_NAA:
2290		sbuf_printf(sb, "naa.");
2291		break;
2292	case SVPD_ID_TYPE_SCSI_NAME:
2293		break;
2294	}
2295	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2296	case SVPD_ID_CODESET_BINARY:
2297		for (i = 0; i < desc->length; i++)
2298			sbuf_printf(sb, "%02x", desc->identifier[i]);
2299		break;
2300	case SVPD_ID_CODESET_ASCII:
2301		sbuf_printf(sb, "%.*s", (int)desc->length,
2302		    (char *)desc->identifier);
2303		break;
2304	case SVPD_ID_CODESET_UTF8:
2305		sbuf_printf(sb, "%s", (char *)desc->identifier);
2306		break;
2307	}
2308}
2309
2310static int
2311ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2312	  struct thread *td)
2313{
2314	struct ctl_softc *softc;
2315	int retval;
2316
2317	softc = control_softc;
2318
2319	retval = 0;
2320
2321	switch (cmd) {
2322	case CTL_IO: {
2323		union ctl_io *io;
2324		void *pool_tmp;
2325
2326		/*
2327		 * If we haven't been "enabled", don't allow any SCSI I/O
2328		 * to this FETD.
2329		 */
2330		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2331			retval = EPERM;
2332			break;
2333		}
2334
2335		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2336
2337		/*
2338		 * Need to save the pool reference so it doesn't get
2339		 * spammed by the user's ctl_io.
2340		 */
2341		pool_tmp = io->io_hdr.pool;
2342		memcpy(io, (void *)addr, sizeof(*io));
2343		io->io_hdr.pool = pool_tmp;
2344
2345		/*
2346		 * No status yet, so make sure the status is set properly.
2347		 */
2348		io->io_hdr.status = CTL_STATUS_NONE;
2349
2350		/*
2351		 * The user sets the initiator ID, target and LUN IDs.
2352		 */
2353		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2354		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2355		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2356		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2357			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2358
2359		retval = ctl_ioctl_submit_wait(io);
2360
2361		if (retval != 0) {
2362			ctl_free_io(io);
2363			break;
2364		}
2365
2366		memcpy((void *)addr, io, sizeof(*io));
2367
2368		/* return this to our pool */
2369		ctl_free_io(io);
2370
2371		break;
2372	}
2373	case CTL_ENABLE_PORT:
2374	case CTL_DISABLE_PORT:
2375	case CTL_SET_PORT_WWNS: {
2376		struct ctl_port *port;
2377		struct ctl_port_entry *entry;
2378
2379		entry = (struct ctl_port_entry *)addr;
2380
2381		mtx_lock(&softc->ctl_lock);
2382		STAILQ_FOREACH(port, &softc->port_list, links) {
2383			int action, done;
2384
2385			action = 0;
2386			done = 0;
2387
2388			if ((entry->port_type == CTL_PORT_NONE)
2389			 && (entry->targ_port == port->targ_port)) {
2390				/*
2391				 * If the user only wants to enable or
2392				 * disable or set WWNs on a specific port,
2393				 * do the operation and we're done.
2394				 */
2395				action = 1;
2396				done = 1;
2397			} else if (entry->port_type & port->port_type) {
2398				/*
2399				 * Compare the user's type mask with the
2400				 * particular frontend type to see if we
2401				 * have a match.
2402				 */
2403				action = 1;
2404				done = 0;
2405
2406				/*
2407				 * Make sure the user isn't trying to set
2408				 * WWNs on multiple ports at the same time.
2409				 */
2410				if (cmd == CTL_SET_PORT_WWNS) {
2411					printf("%s: Can't set WWNs on "
2412					       "multiple ports\n", __func__);
2413					retval = EINVAL;
2414					break;
2415				}
2416			}
2417			if (action != 0) {
2418				/*
2419				 * XXX KDM we have to drop the lock here,
2420				 * because the online/offline operations
2421				 * can potentially block.  We need to
2422				 * reference count the frontends so they
2423				 * can't go away,
2424				 */
2425				mtx_unlock(&softc->ctl_lock);
2426
2427				if (cmd == CTL_ENABLE_PORT) {
2428					struct ctl_lun *lun;
2429
2430					STAILQ_FOREACH(lun, &softc->lun_list,
2431						       links) {
2432						port->lun_enable(port->targ_lun_arg,
2433						    lun->target,
2434						    lun->lun);
2435					}
2436
2437					ctl_port_online(port);
2438				} else if (cmd == CTL_DISABLE_PORT) {
2439					struct ctl_lun *lun;
2440
2441					ctl_port_offline(port);
2442
2443					STAILQ_FOREACH(lun, &softc->lun_list,
2444						       links) {
2445						port->lun_disable(
2446						    port->targ_lun_arg,
2447						    lun->target,
2448						    lun->lun);
2449					}
2450				}
2451
2452				mtx_lock(&softc->ctl_lock);
2453
2454				if (cmd == CTL_SET_PORT_WWNS)
2455					ctl_port_set_wwns(port,
2456					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2457					    1 : 0, entry->wwnn,
2458					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2459					    1 : 0, entry->wwpn);
2460			}
2461			if (done != 0)
2462				break;
2463		}
2464		mtx_unlock(&softc->ctl_lock);
2465		break;
2466	}
2467	case CTL_GET_PORT_LIST: {
2468		struct ctl_port *port;
2469		struct ctl_port_list *list;
2470		int i;
2471
2472		list = (struct ctl_port_list *)addr;
2473
2474		if (list->alloc_len != (list->alloc_num *
2475		    sizeof(struct ctl_port_entry))) {
2476			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2477			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2478			       "%zu\n", __func__, list->alloc_len,
2479			       list->alloc_num, sizeof(struct ctl_port_entry));
2480			retval = EINVAL;
2481			break;
2482		}
2483		list->fill_len = 0;
2484		list->fill_num = 0;
2485		list->dropped_num = 0;
2486		i = 0;
2487		mtx_lock(&softc->ctl_lock);
2488		STAILQ_FOREACH(port, &softc->port_list, links) {
2489			struct ctl_port_entry entry, *list_entry;
2490
2491			if (list->fill_num >= list->alloc_num) {
2492				list->dropped_num++;
2493				continue;
2494			}
2495
2496			entry.port_type = port->port_type;
2497			strlcpy(entry.port_name, port->port_name,
2498				sizeof(entry.port_name));
2499			entry.targ_port = port->targ_port;
2500			entry.physical_port = port->physical_port;
2501			entry.virtual_port = port->virtual_port;
2502			entry.wwnn = port->wwnn;
2503			entry.wwpn = port->wwpn;
2504			if (port->status & CTL_PORT_STATUS_ONLINE)
2505				entry.online = 1;
2506			else
2507				entry.online = 0;
2508
2509			list_entry = &list->entries[i];
2510
2511			retval = copyout(&entry, list_entry, sizeof(entry));
2512			if (retval != 0) {
2513				printf("%s: CTL_GET_PORT_LIST: copyout "
2514				       "returned %d\n", __func__, retval);
2515				break;
2516			}
2517			i++;
2518			list->fill_num++;
2519			list->fill_len += sizeof(entry);
2520		}
2521		mtx_unlock(&softc->ctl_lock);
2522
2523		/*
2524		 * If this is non-zero, we had a copyout fault, so there's
2525		 * probably no point in attempting to set the status inside
2526		 * the structure.
2527		 */
2528		if (retval != 0)
2529			break;
2530
2531		if (list->dropped_num > 0)
2532			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2533		else
2534			list->status = CTL_PORT_LIST_OK;
2535		break;
2536	}
2537	case CTL_DUMP_OOA: {
2538		struct ctl_lun *lun;
2539		union ctl_io *io;
2540		char printbuf[128];
2541		struct sbuf sb;
2542
2543		mtx_lock(&softc->ctl_lock);
2544		printf("Dumping OOA queues:\n");
2545		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2546			mtx_lock(&lun->lun_lock);
2547			for (io = (union ctl_io *)TAILQ_FIRST(
2548			     &lun->ooa_queue); io != NULL;
2549			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2550			     ooa_links)) {
2551				sbuf_new(&sb, printbuf, sizeof(printbuf),
2552					 SBUF_FIXEDLEN);
2553				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2554					    (intmax_t)lun->lun,
2555					    io->scsiio.tag_num,
2556					    (io->io_hdr.flags &
2557					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2558					    (io->io_hdr.flags &
2559					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2560					    (io->io_hdr.flags &
2561					    CTL_FLAG_ABORT) ? " ABORT" : "",
2562			                    (io->io_hdr.flags &
2563		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2564				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2565				sbuf_finish(&sb);
2566				printf("%s\n", sbuf_data(&sb));
2567			}
2568			mtx_unlock(&lun->lun_lock);
2569		}
2570		printf("OOA queues dump done\n");
2571		mtx_unlock(&softc->ctl_lock);
2572		break;
2573	}
2574	case CTL_GET_OOA: {
2575		struct ctl_lun *lun;
2576		struct ctl_ooa *ooa_hdr;
2577		struct ctl_ooa_entry *entries;
2578		uint32_t cur_fill_num;
2579
2580		ooa_hdr = (struct ctl_ooa *)addr;
2581
2582		if ((ooa_hdr->alloc_len == 0)
2583		 || (ooa_hdr->alloc_num == 0)) {
2584			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2585			       "must be non-zero\n", __func__,
2586			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2587			retval = EINVAL;
2588			break;
2589		}
2590
2591		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2592		    sizeof(struct ctl_ooa_entry))) {
2593			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2594			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2595			       __func__, ooa_hdr->alloc_len,
2596			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2597			retval = EINVAL;
2598			break;
2599		}
2600
2601		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2602		if (entries == NULL) {
2603			printf("%s: could not allocate %d bytes for OOA "
2604			       "dump\n", __func__, ooa_hdr->alloc_len);
2605			retval = ENOMEM;
2606			break;
2607		}
2608
2609		mtx_lock(&softc->ctl_lock);
2610		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2611		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2612		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2613			mtx_unlock(&softc->ctl_lock);
2614			free(entries, M_CTL);
2615			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2616			       __func__, (uintmax_t)ooa_hdr->lun_num);
2617			retval = EINVAL;
2618			break;
2619		}
2620
2621		cur_fill_num = 0;
2622
2623		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2624			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2625				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2626					ooa_hdr, entries);
2627				if (retval != 0)
2628					break;
2629			}
2630			if (retval != 0) {
2631				mtx_unlock(&softc->ctl_lock);
2632				free(entries, M_CTL);
2633				break;
2634			}
2635		} else {
2636			lun = softc->ctl_luns[ooa_hdr->lun_num];
2637
2638			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2639						    entries);
2640		}
2641		mtx_unlock(&softc->ctl_lock);
2642
2643		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2644		ooa_hdr->fill_len = ooa_hdr->fill_num *
2645			sizeof(struct ctl_ooa_entry);
2646		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2647		if (retval != 0) {
2648			printf("%s: error copying out %d bytes for OOA dump\n",
2649			       __func__, ooa_hdr->fill_len);
2650		}
2651
2652		getbintime(&ooa_hdr->cur_bt);
2653
2654		if (cur_fill_num > ooa_hdr->alloc_num) {
2655			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2656			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2657		} else {
2658			ooa_hdr->dropped_num = 0;
2659			ooa_hdr->status = CTL_OOA_OK;
2660		}
2661
2662		free(entries, M_CTL);
2663		break;
2664	}
2665	case CTL_CHECK_OOA: {
2666		union ctl_io *io;
2667		struct ctl_lun *lun;
2668		struct ctl_ooa_info *ooa_info;
2669
2670
2671		ooa_info = (struct ctl_ooa_info *)addr;
2672
2673		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2674			ooa_info->status = CTL_OOA_INVALID_LUN;
2675			break;
2676		}
2677		mtx_lock(&softc->ctl_lock);
2678		lun = softc->ctl_luns[ooa_info->lun_id];
2679		if (lun == NULL) {
2680			mtx_unlock(&softc->ctl_lock);
2681			ooa_info->status = CTL_OOA_INVALID_LUN;
2682			break;
2683		}
2684		mtx_lock(&lun->lun_lock);
2685		mtx_unlock(&softc->ctl_lock);
2686		ooa_info->num_entries = 0;
2687		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2688		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2689		     &io->io_hdr, ooa_links)) {
2690			ooa_info->num_entries++;
2691		}
2692		mtx_unlock(&lun->lun_lock);
2693
2694		ooa_info->status = CTL_OOA_SUCCESS;
2695
2696		break;
2697	}
2698	case CTL_HARD_START:
2699	case CTL_HARD_STOP: {
2700		struct ctl_fe_ioctl_startstop_info ss_info;
2701		struct cfi_metatask *metatask;
2702		struct mtx hs_mtx;
2703
2704		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2705
2706		cv_init(&ss_info.sem, "hard start/stop cv" );
2707
2708		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2709		if (metatask == NULL) {
2710			retval = ENOMEM;
2711			mtx_destroy(&hs_mtx);
2712			break;
2713		}
2714
2715		if (cmd == CTL_HARD_START)
2716			metatask->tasktype = CFI_TASK_STARTUP;
2717		else
2718			metatask->tasktype = CFI_TASK_SHUTDOWN;
2719
2720		metatask->callback = ctl_ioctl_hard_startstop_callback;
2721		metatask->callback_arg = &ss_info;
2722
2723		cfi_action(metatask);
2724
2725		/* Wait for the callback */
2726		mtx_lock(&hs_mtx);
2727		cv_wait_sig(&ss_info.sem, &hs_mtx);
2728		mtx_unlock(&hs_mtx);
2729
2730		/*
2731		 * All information has been copied from the metatask by the
2732		 * time cv_broadcast() is called, so we free the metatask here.
2733		 */
2734		cfi_free_metatask(metatask);
2735
2736		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2737
2738		mtx_destroy(&hs_mtx);
2739		break;
2740	}
2741	case CTL_BBRREAD: {
2742		struct ctl_bbrread_info *bbr_info;
2743		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2744		struct mtx bbr_mtx;
2745		struct cfi_metatask *metatask;
2746
2747		bbr_info = (struct ctl_bbrread_info *)addr;
2748
2749		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2750
2751		bzero(&bbr_mtx, sizeof(bbr_mtx));
2752		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2753
2754		fe_bbr_info.bbr_info = bbr_info;
2755		fe_bbr_info.lock = &bbr_mtx;
2756
2757		cv_init(&fe_bbr_info.sem, "BBR read cv");
2758		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2759
2760		if (metatask == NULL) {
2761			mtx_destroy(&bbr_mtx);
2762			cv_destroy(&fe_bbr_info.sem);
2763			retval = ENOMEM;
2764			break;
2765		}
2766		metatask->tasktype = CFI_TASK_BBRREAD;
2767		metatask->callback = ctl_ioctl_bbrread_callback;
2768		metatask->callback_arg = &fe_bbr_info;
2769		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2770		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2771		metatask->taskinfo.bbrread.len = bbr_info->len;
2772
2773		cfi_action(metatask);
2774
2775		mtx_lock(&bbr_mtx);
2776		while (fe_bbr_info.wakeup_done == 0)
2777			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2778		mtx_unlock(&bbr_mtx);
2779
2780		bbr_info->status = metatask->status;
2781		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2782		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2783		memcpy(&bbr_info->sense_data,
2784		       &metatask->taskinfo.bbrread.sense_data,
2785		       ctl_min(sizeof(bbr_info->sense_data),
2786			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2787
2788		cfi_free_metatask(metatask);
2789
2790		mtx_destroy(&bbr_mtx);
2791		cv_destroy(&fe_bbr_info.sem);
2792
2793		break;
2794	}
2795	case CTL_DELAY_IO: {
2796		struct ctl_io_delay_info *delay_info;
2797#ifdef CTL_IO_DELAY
2798		struct ctl_lun *lun;
2799#endif /* CTL_IO_DELAY */
2800
2801		delay_info = (struct ctl_io_delay_info *)addr;
2802
2803#ifdef CTL_IO_DELAY
2804		mtx_lock(&softc->ctl_lock);
2805
2806		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2807		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2808			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2809		} else {
2810			lun = softc->ctl_luns[delay_info->lun_id];
2811			mtx_lock(&lun->lun_lock);
2812
2813			delay_info->status = CTL_DELAY_STATUS_OK;
2814
2815			switch (delay_info->delay_type) {
2816			case CTL_DELAY_TYPE_CONT:
2817				break;
2818			case CTL_DELAY_TYPE_ONESHOT:
2819				break;
2820			default:
2821				delay_info->status =
2822					CTL_DELAY_STATUS_INVALID_TYPE;
2823				break;
2824			}
2825
2826			switch (delay_info->delay_loc) {
2827			case CTL_DELAY_LOC_DATAMOVE:
2828				lun->delay_info.datamove_type =
2829					delay_info->delay_type;
2830				lun->delay_info.datamove_delay =
2831					delay_info->delay_secs;
2832				break;
2833			case CTL_DELAY_LOC_DONE:
2834				lun->delay_info.done_type =
2835					delay_info->delay_type;
2836				lun->delay_info.done_delay =
2837					delay_info->delay_secs;
2838				break;
2839			default:
2840				delay_info->status =
2841					CTL_DELAY_STATUS_INVALID_LOC;
2842				break;
2843			}
2844			mtx_unlock(&lun->lun_lock);
2845		}
2846
2847		mtx_unlock(&softc->ctl_lock);
2848#else
2849		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2850#endif /* CTL_IO_DELAY */
2851		break;
2852	}
2853	case CTL_REALSYNC_SET: {
2854		int *syncstate;
2855
2856		syncstate = (int *)addr;
2857
2858		mtx_lock(&softc->ctl_lock);
2859		switch (*syncstate) {
2860		case 0:
2861			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2862			break;
2863		case 1:
2864			softc->flags |= CTL_FLAG_REAL_SYNC;
2865			break;
2866		default:
2867			retval = EINVAL;
2868			break;
2869		}
2870		mtx_unlock(&softc->ctl_lock);
2871		break;
2872	}
2873	case CTL_REALSYNC_GET: {
2874		int *syncstate;
2875
2876		syncstate = (int*)addr;
2877
2878		mtx_lock(&softc->ctl_lock);
2879		if (softc->flags & CTL_FLAG_REAL_SYNC)
2880			*syncstate = 1;
2881		else
2882			*syncstate = 0;
2883		mtx_unlock(&softc->ctl_lock);
2884
2885		break;
2886	}
2887	case CTL_SETSYNC:
2888	case CTL_GETSYNC: {
2889		struct ctl_sync_info *sync_info;
2890		struct ctl_lun *lun;
2891
2892		sync_info = (struct ctl_sync_info *)addr;
2893
2894		mtx_lock(&softc->ctl_lock);
2895		lun = softc->ctl_luns[sync_info->lun_id];
2896		if (lun == NULL) {
2897			mtx_unlock(&softc->ctl_lock);
2898			sync_info->status = CTL_GS_SYNC_NO_LUN;
2899		}
2900		/*
2901		 * Get or set the sync interval.  We're not bounds checking
2902		 * in the set case, hopefully the user won't do something
2903		 * silly.
2904		 */
2905		mtx_lock(&lun->lun_lock);
2906		mtx_unlock(&softc->ctl_lock);
2907		if (cmd == CTL_GETSYNC)
2908			sync_info->sync_interval = lun->sync_interval;
2909		else
2910			lun->sync_interval = sync_info->sync_interval;
2911		mtx_unlock(&lun->lun_lock);
2912
2913		sync_info->status = CTL_GS_SYNC_OK;
2914
2915		break;
2916	}
2917	case CTL_GETSTATS: {
2918		struct ctl_stats *stats;
2919		struct ctl_lun *lun;
2920		int i;
2921
2922		stats = (struct ctl_stats *)addr;
2923
2924		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2925		     stats->alloc_len) {
2926			stats->status = CTL_SS_NEED_MORE_SPACE;
2927			stats->num_luns = softc->num_luns;
2928			break;
2929		}
2930		/*
2931		 * XXX KDM no locking here.  If the LUN list changes,
2932		 * things can blow up.
2933		 */
2934		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2935		     i++, lun = STAILQ_NEXT(lun, links)) {
2936			retval = copyout(&lun->stats, &stats->lun_stats[i],
2937					 sizeof(lun->stats));
2938			if (retval != 0)
2939				break;
2940		}
2941		stats->num_luns = softc->num_luns;
2942		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2943				 softc->num_luns;
2944		stats->status = CTL_SS_OK;
2945#ifdef CTL_TIME_IO
2946		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2947#else
2948		stats->flags = CTL_STATS_FLAG_NONE;
2949#endif
2950		getnanouptime(&stats->timestamp);
2951		break;
2952	}
2953	case CTL_ERROR_INJECT: {
2954		struct ctl_error_desc *err_desc, *new_err_desc;
2955		struct ctl_lun *lun;
2956
2957		err_desc = (struct ctl_error_desc *)addr;
2958
2959		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2960				      M_WAITOK | M_ZERO);
2961		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2962
2963		mtx_lock(&softc->ctl_lock);
2964		lun = softc->ctl_luns[err_desc->lun_id];
2965		if (lun == NULL) {
2966			mtx_unlock(&softc->ctl_lock);
2967			free(new_err_desc, M_CTL);
2968			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2969			       __func__, (uintmax_t)err_desc->lun_id);
2970			retval = EINVAL;
2971			break;
2972		}
2973		mtx_lock(&lun->lun_lock);
2974		mtx_unlock(&softc->ctl_lock);
2975
2976		/*
2977		 * We could do some checking here to verify the validity
2978		 * of the request, but given the complexity of error
2979		 * injection requests, the checking logic would be fairly
2980		 * complex.
2981		 *
2982		 * For now, if the request is invalid, it just won't get
2983		 * executed and might get deleted.
2984		 */
2985		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2986
2987		/*
2988		 * XXX KDM check to make sure the serial number is unique,
2989		 * in case we somehow manage to wrap.  That shouldn't
2990		 * happen for a very long time, but it's the right thing to
2991		 * do.
2992		 */
2993		new_err_desc->serial = lun->error_serial;
2994		err_desc->serial = lun->error_serial;
2995		lun->error_serial++;
2996
2997		mtx_unlock(&lun->lun_lock);
2998		break;
2999	}
3000	case CTL_ERROR_INJECT_DELETE: {
3001		struct ctl_error_desc *delete_desc, *desc, *desc2;
3002		struct ctl_lun *lun;
3003		int delete_done;
3004
3005		delete_desc = (struct ctl_error_desc *)addr;
3006		delete_done = 0;
3007
3008		mtx_lock(&softc->ctl_lock);
3009		lun = softc->ctl_luns[delete_desc->lun_id];
3010		if (lun == NULL) {
3011			mtx_unlock(&softc->ctl_lock);
3012			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
3013			       __func__, (uintmax_t)delete_desc->lun_id);
3014			retval = EINVAL;
3015			break;
3016		}
3017		mtx_lock(&lun->lun_lock);
3018		mtx_unlock(&softc->ctl_lock);
3019		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
3020			if (desc->serial != delete_desc->serial)
3021				continue;
3022
3023			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
3024				      links);
3025			free(desc, M_CTL);
3026			delete_done = 1;
3027		}
3028		mtx_unlock(&lun->lun_lock);
3029		if (delete_done == 0) {
3030			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3031			       "error serial %ju on LUN %u\n", __func__,
3032			       delete_desc->serial, delete_desc->lun_id);
3033			retval = EINVAL;
3034			break;
3035		}
3036		break;
3037	}
3038	case CTL_DUMP_STRUCTS: {
3039		int i, j, k, idx;
3040		struct ctl_port *port;
3041		struct ctl_frontend *fe;
3042
3043		mtx_lock(&softc->ctl_lock);
3044		printf("CTL Persistent Reservation information start:\n");
3045		for (i = 0; i < CTL_MAX_LUNS; i++) {
3046			struct ctl_lun *lun;
3047
3048			lun = softc->ctl_luns[i];
3049
3050			if ((lun == NULL)
3051			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3052				continue;
3053
3054			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3055				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3056					idx = j * CTL_MAX_INIT_PER_PORT + k;
3057					if (lun->pr_keys[idx] == 0)
3058						continue;
3059					printf("  LUN %d port %d iid %d key "
3060					       "%#jx\n", i, j, k,
3061					       (uintmax_t)lun->pr_keys[idx]);
3062				}
3063			}
3064		}
3065		printf("CTL Persistent Reservation information end\n");
3066		printf("CTL Ports:\n");
3067		STAILQ_FOREACH(port, &softc->port_list, links) {
3068			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3069			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3070			       port->frontend->name, port->port_type,
3071			       port->physical_port, port->virtual_port,
3072			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3073			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3074				if (port->wwpn_iid[j].in_use == 0 &&
3075				    port->wwpn_iid[j].wwpn == 0 &&
3076				    port->wwpn_iid[j].name == NULL)
3077					continue;
3078
3079				printf("    iid %u use %d WWPN %#jx '%s'\n",
3080				    j, port->wwpn_iid[j].in_use,
3081				    (uintmax_t)port->wwpn_iid[j].wwpn,
3082				    port->wwpn_iid[j].name);
3083			}
3084		}
3085		printf("CTL Port information end\n");
3086		mtx_unlock(&softc->ctl_lock);
3087		/*
3088		 * XXX KDM calling this without a lock.  We'd likely want
3089		 * to drop the lock before calling the frontend's dump
3090		 * routine anyway.
3091		 */
3092		printf("CTL Frontends:\n");
3093		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3094			printf("  Frontend '%s'\n", fe->name);
3095			if (fe->fe_dump != NULL)
3096				fe->fe_dump();
3097		}
3098		printf("CTL Frontend information end\n");
3099		break;
3100	}
3101	case CTL_LUN_REQ: {
3102		struct ctl_lun_req *lun_req;
3103		struct ctl_backend_driver *backend;
3104
3105		lun_req = (struct ctl_lun_req *)addr;
3106
3107		backend = ctl_backend_find(lun_req->backend);
3108		if (backend == NULL) {
3109			lun_req->status = CTL_LUN_ERROR;
3110			snprintf(lun_req->error_str,
3111				 sizeof(lun_req->error_str),
3112				 "Backend \"%s\" not found.",
3113				 lun_req->backend);
3114			break;
3115		}
3116		if (lun_req->num_be_args > 0) {
3117			lun_req->kern_be_args = ctl_copyin_args(
3118				lun_req->num_be_args,
3119				lun_req->be_args,
3120				lun_req->error_str,
3121				sizeof(lun_req->error_str));
3122			if (lun_req->kern_be_args == NULL) {
3123				lun_req->status = CTL_LUN_ERROR;
3124				break;
3125			}
3126		}
3127
3128		retval = backend->ioctl(dev, cmd, addr, flag, td);
3129
3130		if (lun_req->num_be_args > 0) {
3131			ctl_copyout_args(lun_req->num_be_args,
3132				      lun_req->kern_be_args);
3133			ctl_free_args(lun_req->num_be_args,
3134				      lun_req->kern_be_args);
3135		}
3136		break;
3137	}
3138	case CTL_LUN_LIST: {
3139		struct sbuf *sb;
3140		struct ctl_lun *lun;
3141		struct ctl_lun_list *list;
3142		struct ctl_option *opt;
3143
3144		list = (struct ctl_lun_list *)addr;
3145
3146		/*
3147		 * Allocate a fixed length sbuf here, based on the length
3148		 * of the user's buffer.  We could allocate an auto-extending
3149		 * buffer, and then tell the user how much larger our
3150		 * amount of data is than his buffer, but that presents
3151		 * some problems:
3152		 *
3153		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3154		 *     we can't hold a lock while calling them with an
3155		 *     auto-extending buffer.
3156 		 *
3157		 * 2.  There is not currently a LUN reference counting
3158		 *     mechanism, outside of outstanding transactions on
3159		 *     the LUN's OOA queue.  So a LUN could go away on us
3160		 *     while we're getting the LUN number, backend-specific
3161		 *     information, etc.  Thus, given the way things
3162		 *     currently work, we need to hold the CTL lock while
3163		 *     grabbing LUN information.
3164		 *
3165		 * So, from the user's standpoint, the best thing to do is
3166		 * allocate what he thinks is a reasonable buffer length,
3167		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3168		 * double the buffer length and try again.  (And repeat
3169		 * that until he succeeds.)
3170		 */
3171		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3172		if (sb == NULL) {
3173			list->status = CTL_LUN_LIST_ERROR;
3174			snprintf(list->error_str, sizeof(list->error_str),
3175				 "Unable to allocate %d bytes for LUN list",
3176				 list->alloc_len);
3177			break;
3178		}
3179
3180		sbuf_printf(sb, "<ctllunlist>\n");
3181
3182		mtx_lock(&softc->ctl_lock);
3183		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3184			mtx_lock(&lun->lun_lock);
3185			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3186					     (uintmax_t)lun->lun);
3187
3188			/*
3189			 * Bail out as soon as we see that we've overfilled
3190			 * the buffer.
3191			 */
3192			if (retval != 0)
3193				break;
3194
3195			retval = sbuf_printf(sb, "\t<backend_type>%s"
3196					     "</backend_type>\n",
3197					     (lun->backend == NULL) ?  "none" :
3198					     lun->backend->name);
3199
3200			if (retval != 0)
3201				break;
3202
3203			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3204					     lun->be_lun->lun_type);
3205
3206			if (retval != 0)
3207				break;
3208
3209			if (lun->backend == NULL) {
3210				retval = sbuf_printf(sb, "</lun>\n");
3211				if (retval != 0)
3212					break;
3213				continue;
3214			}
3215
3216			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3217					     (lun->be_lun->maxlba > 0) ?
3218					     lun->be_lun->maxlba + 1 : 0);
3219
3220			if (retval != 0)
3221				break;
3222
3223			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3224					     lun->be_lun->blocksize);
3225
3226			if (retval != 0)
3227				break;
3228
3229			retval = sbuf_printf(sb, "\t<serial_number>");
3230
3231			if (retval != 0)
3232				break;
3233
3234			retval = ctl_sbuf_printf_esc(sb,
3235			    lun->be_lun->serial_num,
3236			    sizeof(lun->be_lun->serial_num));
3237
3238			if (retval != 0)
3239				break;
3240
3241			retval = sbuf_printf(sb, "</serial_number>\n");
3242
3243			if (retval != 0)
3244				break;
3245
3246			retval = sbuf_printf(sb, "\t<device_id>");
3247
3248			if (retval != 0)
3249				break;
3250
3251			retval = ctl_sbuf_printf_esc(sb,
3252			    lun->be_lun->device_id,
3253			    sizeof(lun->be_lun->device_id));
3254
3255			if (retval != 0)
3256				break;
3257
3258			retval = sbuf_printf(sb, "</device_id>\n");
3259
3260			if (retval != 0)
3261				break;
3262
3263			if (lun->backend->lun_info != NULL) {
3264				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3265				if (retval != 0)
3266					break;
3267			}
3268			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3269				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3270				    opt->name, opt->value, opt->name);
3271				if (retval != 0)
3272					break;
3273			}
3274
3275			retval = sbuf_printf(sb, "</lun>\n");
3276
3277			if (retval != 0)
3278				break;
3279			mtx_unlock(&lun->lun_lock);
3280		}
3281		if (lun != NULL)
3282			mtx_unlock(&lun->lun_lock);
3283		mtx_unlock(&softc->ctl_lock);
3284
3285		if ((retval != 0)
3286		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3287			retval = 0;
3288			sbuf_delete(sb);
3289			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3290			snprintf(list->error_str, sizeof(list->error_str),
3291				 "Out of space, %d bytes is too small",
3292				 list->alloc_len);
3293			break;
3294		}
3295
3296		sbuf_finish(sb);
3297
3298		retval = copyout(sbuf_data(sb), list->lun_xml,
3299				 sbuf_len(sb) + 1);
3300
3301		list->fill_len = sbuf_len(sb) + 1;
3302		list->status = CTL_LUN_LIST_OK;
3303		sbuf_delete(sb);
3304		break;
3305	}
3306	case CTL_ISCSI: {
3307		struct ctl_iscsi *ci;
3308		struct ctl_frontend *fe;
3309
3310		ci = (struct ctl_iscsi *)addr;
3311
3312		fe = ctl_frontend_find("iscsi");
3313		if (fe == NULL) {
3314			ci->status = CTL_ISCSI_ERROR;
3315			snprintf(ci->error_str, sizeof(ci->error_str),
3316			    "Frontend \"iscsi\" not found.");
3317			break;
3318		}
3319
3320		retval = fe->ioctl(dev, cmd, addr, flag, td);
3321		break;
3322	}
3323	case CTL_PORT_REQ: {
3324		struct ctl_req *req;
3325		struct ctl_frontend *fe;
3326
3327		req = (struct ctl_req *)addr;
3328
3329		fe = ctl_frontend_find(req->driver);
3330		if (fe == NULL) {
3331			req->status = CTL_LUN_ERROR;
3332			snprintf(req->error_str, sizeof(req->error_str),
3333			    "Frontend \"%s\" not found.", req->driver);
3334			break;
3335		}
3336		if (req->num_args > 0) {
3337			req->kern_args = ctl_copyin_args(req->num_args,
3338			    req->args, req->error_str, sizeof(req->error_str));
3339			if (req->kern_args == NULL) {
3340				req->status = CTL_LUN_ERROR;
3341				break;
3342			}
3343		}
3344
3345		retval = fe->ioctl(dev, cmd, addr, flag, td);
3346
3347		if (req->num_args > 0) {
3348			ctl_copyout_args(req->num_args, req->kern_args);
3349			ctl_free_args(req->num_args, req->kern_args);
3350		}
3351		break;
3352	}
3353	case CTL_PORT_LIST: {
3354		struct sbuf *sb;
3355		struct ctl_port *port;
3356		struct ctl_lun_list *list;
3357		struct ctl_option *opt;
3358		int j;
3359
3360		list = (struct ctl_lun_list *)addr;
3361
3362		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3363		if (sb == NULL) {
3364			list->status = CTL_LUN_LIST_ERROR;
3365			snprintf(list->error_str, sizeof(list->error_str),
3366				 "Unable to allocate %d bytes for LUN list",
3367				 list->alloc_len);
3368			break;
3369		}
3370
3371		sbuf_printf(sb, "<ctlportlist>\n");
3372
3373		mtx_lock(&softc->ctl_lock);
3374		STAILQ_FOREACH(port, &softc->port_list, links) {
3375			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3376					     (uintmax_t)port->targ_port);
3377
3378			/*
3379			 * Bail out as soon as we see that we've overfilled
3380			 * the buffer.
3381			 */
3382			if (retval != 0)
3383				break;
3384
3385			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3386			    "</frontend_type>\n", port->frontend->name);
3387			if (retval != 0)
3388				break;
3389
3390			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3391					     port->port_type);
3392			if (retval != 0)
3393				break;
3394
3395			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3396			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3397			if (retval != 0)
3398				break;
3399
3400			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3401			    port->port_name);
3402			if (retval != 0)
3403				break;
3404
3405			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3406			    port->physical_port);
3407			if (retval != 0)
3408				break;
3409
3410			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3411			    port->virtual_port);
3412			if (retval != 0)
3413				break;
3414
3415			if (port->target_devid != NULL) {
3416				sbuf_printf(sb, "\t<target>");
3417				ctl_id_sbuf(port->target_devid, sb);
3418				sbuf_printf(sb, "</target>\n");
3419			}
3420
3421			if (port->port_devid != NULL) {
3422				sbuf_printf(sb, "\t<port>");
3423				ctl_id_sbuf(port->port_devid, sb);
3424				sbuf_printf(sb, "</port>\n");
3425			}
3426
3427			if (port->port_info != NULL) {
3428				retval = port->port_info(port->onoff_arg, sb);
3429				if (retval != 0)
3430					break;
3431			}
3432			STAILQ_FOREACH(opt, &port->options, links) {
3433				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3434				    opt->name, opt->value, opt->name);
3435				if (retval != 0)
3436					break;
3437			}
3438
3439			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3440				if (port->wwpn_iid[j].in_use == 0 ||
3441				    (port->wwpn_iid[j].wwpn == 0 &&
3442				     port->wwpn_iid[j].name == NULL))
3443					continue;
3444
3445				if (port->wwpn_iid[j].name != NULL)
3446					retval = sbuf_printf(sb,
3447					    "\t<initiator>%u %s</initiator>\n",
3448					    j, port->wwpn_iid[j].name);
3449				else
3450					retval = sbuf_printf(sb,
3451					    "\t<initiator>%u naa.%08jx</initiator>\n",
3452					    j, port->wwpn_iid[j].wwpn);
3453				if (retval != 0)
3454					break;
3455			}
3456			if (retval != 0)
3457				break;
3458
3459			retval = sbuf_printf(sb, "</targ_port>\n");
3460			if (retval != 0)
3461				break;
3462		}
3463		mtx_unlock(&softc->ctl_lock);
3464
3465		if ((retval != 0)
3466		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3467			retval = 0;
3468			sbuf_delete(sb);
3469			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3470			snprintf(list->error_str, sizeof(list->error_str),
3471				 "Out of space, %d bytes is too small",
3472				 list->alloc_len);
3473			break;
3474		}
3475
3476		sbuf_finish(sb);
3477
3478		retval = copyout(sbuf_data(sb), list->lun_xml,
3479				 sbuf_len(sb) + 1);
3480
3481		list->fill_len = sbuf_len(sb) + 1;
3482		list->status = CTL_LUN_LIST_OK;
3483		sbuf_delete(sb);
3484		break;
3485	}
3486	default: {
3487		/* XXX KDM should we fix this? */
3488#if 0
3489		struct ctl_backend_driver *backend;
3490		unsigned int type;
3491		int found;
3492
3493		found = 0;
3494
3495		/*
3496		 * We encode the backend type as the ioctl type for backend
3497		 * ioctls.  So parse it out here, and then search for a
3498		 * backend of this type.
3499		 */
3500		type = _IOC_TYPE(cmd);
3501
3502		STAILQ_FOREACH(backend, &softc->be_list, links) {
3503			if (backend->type == type) {
3504				found = 1;
3505				break;
3506			}
3507		}
3508		if (found == 0) {
3509			printf("ctl: unknown ioctl command %#lx or backend "
3510			       "%d\n", cmd, type);
3511			retval = EINVAL;
3512			break;
3513		}
3514		retval = backend->ioctl(dev, cmd, addr, flag, td);
3515#endif
3516		retval = ENOTTY;
3517		break;
3518	}
3519	}
3520	return (retval);
3521}
3522
3523uint32_t
3524ctl_get_initindex(struct ctl_nexus *nexus)
3525{
3526	if (nexus->targ_port < CTL_MAX_PORTS)
3527		return (nexus->initid.id +
3528			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3529	else
3530		return (nexus->initid.id +
3531		       ((nexus->targ_port - CTL_MAX_PORTS) *
3532			CTL_MAX_INIT_PER_PORT));
3533}
3534
3535uint32_t
3536ctl_get_resindex(struct ctl_nexus *nexus)
3537{
3538	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3539}
3540
3541uint32_t
3542ctl_port_idx(int port_num)
3543{
3544	if (port_num < CTL_MAX_PORTS)
3545		return(port_num);
3546	else
3547		return(port_num - CTL_MAX_PORTS);
3548}
3549
3550static uint32_t
3551ctl_map_lun(int port_num, uint32_t lun_id)
3552{
3553	struct ctl_port *port;
3554
3555	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3556	if (port == NULL)
3557		return (UINT32_MAX);
3558	if (port->lun_map == NULL)
3559		return (lun_id);
3560	return (port->lun_map(port->targ_lun_arg, lun_id));
3561}
3562
3563static uint32_t
3564ctl_map_lun_back(int port_num, uint32_t lun_id)
3565{
3566	struct ctl_port *port;
3567	uint32_t i;
3568
3569	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3570	if (port->lun_map == NULL)
3571		return (lun_id);
3572	for (i = 0; i < CTL_MAX_LUNS; i++) {
3573		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3574			return (i);
3575	}
3576	return (UINT32_MAX);
3577}
3578
3579/*
3580 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3581 * that are a power of 2.
3582 */
3583int
3584ctl_ffz(uint32_t *mask, uint32_t size)
3585{
3586	uint32_t num_chunks, num_pieces;
3587	int i, j;
3588
3589	num_chunks = (size >> 5);
3590	if (num_chunks == 0)
3591		num_chunks++;
3592	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3593
3594	for (i = 0; i < num_chunks; i++) {
3595		for (j = 0; j < num_pieces; j++) {
3596			if ((mask[i] & (1 << j)) == 0)
3597				return ((i << 5) + j);
3598		}
3599	}
3600
3601	return (-1);
3602}
3603
3604int
3605ctl_set_mask(uint32_t *mask, uint32_t bit)
3606{
3607	uint32_t chunk, piece;
3608
3609	chunk = bit >> 5;
3610	piece = bit % (sizeof(uint32_t) * 8);
3611
3612	if ((mask[chunk] & (1 << piece)) != 0)
3613		return (-1);
3614	else
3615		mask[chunk] |= (1 << piece);
3616
3617	return (0);
3618}
3619
3620int
3621ctl_clear_mask(uint32_t *mask, uint32_t bit)
3622{
3623	uint32_t chunk, piece;
3624
3625	chunk = bit >> 5;
3626	piece = bit % (sizeof(uint32_t) * 8);
3627
3628	if ((mask[chunk] & (1 << piece)) == 0)
3629		return (-1);
3630	else
3631		mask[chunk] &= ~(1 << piece);
3632
3633	return (0);
3634}
3635
3636int
3637ctl_is_set(uint32_t *mask, uint32_t bit)
3638{
3639	uint32_t chunk, piece;
3640
3641	chunk = bit >> 5;
3642	piece = bit % (sizeof(uint32_t) * 8);
3643
3644	if ((mask[chunk] & (1 << piece)) == 0)
3645		return (0);
3646	else
3647		return (1);
3648}
3649
3650#ifdef unused
3651/*
3652 * The bus, target and lun are optional, they can be filled in later.
3653 * can_wait is used to determine whether we can wait on the malloc or not.
3654 */
3655union ctl_io*
3656ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3657	      uint32_t targ_lun, int can_wait)
3658{
3659	union ctl_io *io;
3660
3661	if (can_wait)
3662		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3663	else
3664		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3665
3666	if (io != NULL) {
3667		io->io_hdr.io_type = io_type;
3668		io->io_hdr.targ_port = targ_port;
3669		/*
3670		 * XXX KDM this needs to change/go away.  We need to move
3671		 * to a preallocated pool of ctl_scsiio structures.
3672		 */
3673		io->io_hdr.nexus.targ_target.id = targ_target;
3674		io->io_hdr.nexus.targ_lun = targ_lun;
3675	}
3676
3677	return (io);
3678}
3679
3680void
3681ctl_kfree_io(union ctl_io *io)
3682{
3683	free(io, M_CTL);
3684}
3685#endif /* unused */
3686
3687/*
3688 * ctl_softc, pool_name, total_ctl_io are passed in.
3689 * npool is passed out.
3690 */
3691int
3692ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3693		uint32_t total_ctl_io, void **npool)
3694{
3695#ifdef IO_POOLS
3696	struct ctl_io_pool *pool;
3697
3698	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3699					    M_NOWAIT | M_ZERO);
3700	if (pool == NULL)
3701		return (ENOMEM);
3702
3703	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3704	pool->ctl_softc = ctl_softc;
3705	pool->zone = uma_zsecond_create(pool->name, NULL,
3706	    NULL, NULL, NULL, ctl_softc->io_zone);
3707	/* uma_prealloc(pool->zone, total_ctl_io); */
3708
3709	*npool = pool;
3710#else
3711	*npool = ctl_softc->io_zone;
3712#endif
3713	return (0);
3714}
3715
3716void
3717ctl_pool_free(struct ctl_io_pool *pool)
3718{
3719
3720	if (pool == NULL)
3721		return;
3722
3723#ifdef IO_POOLS
3724	uma_zdestroy(pool->zone);
3725	free(pool, M_CTL);
3726#endif
3727}
3728
3729union ctl_io *
3730ctl_alloc_io(void *pool_ref)
3731{
3732	union ctl_io *io;
3733#ifdef IO_POOLS
3734	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3735
3736	io = uma_zalloc(pool->zone, M_WAITOK);
3737#else
3738	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3739#endif
3740	if (io != NULL)
3741		io->io_hdr.pool = pool_ref;
3742	return (io);
3743}
3744
3745union ctl_io *
3746ctl_alloc_io_nowait(void *pool_ref)
3747{
3748	union ctl_io *io;
3749#ifdef IO_POOLS
3750	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3751
3752	io = uma_zalloc(pool->zone, M_NOWAIT);
3753#else
3754	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3755#endif
3756	if (io != NULL)
3757		io->io_hdr.pool = pool_ref;
3758	return (io);
3759}
3760
3761void
3762ctl_free_io(union ctl_io *io)
3763{
3764#ifdef IO_POOLS
3765	struct ctl_io_pool *pool;
3766#endif
3767
3768	if (io == NULL)
3769		return;
3770
3771#ifdef IO_POOLS
3772	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3773	uma_zfree(pool->zone, io);
3774#else
3775	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3776#endif
3777}
3778
3779void
3780ctl_zero_io(union ctl_io *io)
3781{
3782	void *pool_ref;
3783
3784	if (io == NULL)
3785		return;
3786
3787	/*
3788	 * May need to preserve linked list pointers at some point too.
3789	 */
3790	pool_ref = io->io_hdr.pool;
3791	memset(io, 0, sizeof(*io));
3792	io->io_hdr.pool = pool_ref;
3793}
3794
3795/*
3796 * This routine is currently used for internal copies of ctl_ios that need
3797 * to persist for some reason after we've already returned status to the
3798 * FETD.  (Thus the flag set.)
3799 *
3800 * XXX XXX
3801 * Note that this makes a blind copy of all fields in the ctl_io, except
3802 * for the pool reference.  This includes any memory that has been
3803 * allocated!  That memory will no longer be valid after done has been
3804 * called, so this would be VERY DANGEROUS for command that actually does
3805 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3806 * start and stop commands, which don't transfer any data, so this is not a
3807 * problem.  If it is used for anything else, the caller would also need to
3808 * allocate data buffer space and this routine would need to be modified to
3809 * copy the data buffer(s) as well.
3810 */
3811void
3812ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3813{
3814	void *pool_ref;
3815
3816	if ((src == NULL)
3817	 || (dest == NULL))
3818		return;
3819
3820	/*
3821	 * May need to preserve linked list pointers at some point too.
3822	 */
3823	pool_ref = dest->io_hdr.pool;
3824
3825	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3826
3827	dest->io_hdr.pool = pool_ref;
3828	/*
3829	 * We need to know that this is an internal copy, and doesn't need
3830	 * to get passed back to the FETD that allocated it.
3831	 */
3832	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3833}
3834
3835static int
3836ctl_expand_number(const char *buf, uint64_t *num)
3837{
3838	char *endptr;
3839	uint64_t number;
3840	unsigned shift;
3841
3842	number = strtoq(buf, &endptr, 0);
3843
3844	switch (tolower((unsigned char)*endptr)) {
3845	case 'e':
3846		shift = 60;
3847		break;
3848	case 'p':
3849		shift = 50;
3850		break;
3851	case 't':
3852		shift = 40;
3853		break;
3854	case 'g':
3855		shift = 30;
3856		break;
3857	case 'm':
3858		shift = 20;
3859		break;
3860	case 'k':
3861		shift = 10;
3862		break;
3863	case 'b':
3864	case '\0': /* No unit. */
3865		*num = number;
3866		return (0);
3867	default:
3868		/* Unrecognized unit. */
3869		return (-1);
3870	}
3871
3872	if ((number << shift) >> shift != number) {
3873		/* Overflow */
3874		return (-1);
3875	}
3876	*num = number << shift;
3877	return (0);
3878}
3879
3880
3881/*
3882 * This routine could be used in the future to load default and/or saved
3883 * mode page parameters for a particuar lun.
3884 */
3885static int
3886ctl_init_page_index(struct ctl_lun *lun)
3887{
3888	int i;
3889	struct ctl_page_index *page_index;
3890	const char *value;
3891	uint64_t ival;
3892
3893	memcpy(&lun->mode_pages.index, page_index_template,
3894	       sizeof(page_index_template));
3895
3896	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3897
3898		page_index = &lun->mode_pages.index[i];
3899		/*
3900		 * If this is a disk-only mode page, there's no point in
3901		 * setting it up.  For some pages, we have to have some
3902		 * basic information about the disk in order to calculate the
3903		 * mode page data.
3904		 */
3905		if ((lun->be_lun->lun_type != T_DIRECT)
3906		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3907			continue;
3908
3909		switch (page_index->page_code & SMPH_PC_MASK) {
3910		case SMS_RW_ERROR_RECOVERY_PAGE: {
3911			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3912				panic("subpage is incorrect!");
3913			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3914			       &rw_er_page_default,
3915			       sizeof(rw_er_page_default));
3916			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3917			       &rw_er_page_changeable,
3918			       sizeof(rw_er_page_changeable));
3919			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3920			       &rw_er_page_default,
3921			       sizeof(rw_er_page_default));
3922			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3923			       &rw_er_page_default,
3924			       sizeof(rw_er_page_default));
3925			page_index->page_data =
3926				(uint8_t *)lun->mode_pages.rw_er_page;
3927			break;
3928		}
3929		case SMS_FORMAT_DEVICE_PAGE: {
3930			struct scsi_format_page *format_page;
3931
3932			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3933				panic("subpage is incorrect!");
3934
3935			/*
3936			 * Sectors per track are set above.  Bytes per
3937			 * sector need to be set here on a per-LUN basis.
3938			 */
3939			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3940			       &format_page_default,
3941			       sizeof(format_page_default));
3942			memcpy(&lun->mode_pages.format_page[
3943			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3944			       sizeof(format_page_changeable));
3945			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3946			       &format_page_default,
3947			       sizeof(format_page_default));
3948			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3949			       &format_page_default,
3950			       sizeof(format_page_default));
3951
3952			format_page = &lun->mode_pages.format_page[
3953				CTL_PAGE_CURRENT];
3954			scsi_ulto2b(lun->be_lun->blocksize,
3955				    format_page->bytes_per_sector);
3956
3957			format_page = &lun->mode_pages.format_page[
3958				CTL_PAGE_DEFAULT];
3959			scsi_ulto2b(lun->be_lun->blocksize,
3960				    format_page->bytes_per_sector);
3961
3962			format_page = &lun->mode_pages.format_page[
3963				CTL_PAGE_SAVED];
3964			scsi_ulto2b(lun->be_lun->blocksize,
3965				    format_page->bytes_per_sector);
3966
3967			page_index->page_data =
3968				(uint8_t *)lun->mode_pages.format_page;
3969			break;
3970		}
3971		case SMS_RIGID_DISK_PAGE: {
3972			struct scsi_rigid_disk_page *rigid_disk_page;
3973			uint32_t sectors_per_cylinder;
3974			uint64_t cylinders;
3975#ifndef	__XSCALE__
3976			int shift;
3977#endif /* !__XSCALE__ */
3978
3979			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3980				panic("invalid subpage value %d",
3981				      page_index->subpage);
3982
3983			/*
3984			 * Rotation rate and sectors per track are set
3985			 * above.  We calculate the cylinders here based on
3986			 * capacity.  Due to the number of heads and
3987			 * sectors per track we're using, smaller arrays
3988			 * may turn out to have 0 cylinders.  Linux and
3989			 * FreeBSD don't pay attention to these mode pages
3990			 * to figure out capacity, but Solaris does.  It
3991			 * seems to deal with 0 cylinders just fine, and
3992			 * works out a fake geometry based on the capacity.
3993			 */
3994			memcpy(&lun->mode_pages.rigid_disk_page[
3995			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3996			       sizeof(rigid_disk_page_default));
3997			memcpy(&lun->mode_pages.rigid_disk_page[
3998			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3999			       sizeof(rigid_disk_page_changeable));
4000
4001			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4002				CTL_DEFAULT_HEADS;
4003
4004			/*
4005			 * The divide method here will be more accurate,
4006			 * probably, but results in floating point being
4007			 * used in the kernel on i386 (__udivdi3()).  On the
4008			 * XScale, though, __udivdi3() is implemented in
4009			 * software.
4010			 *
4011			 * The shift method for cylinder calculation is
4012			 * accurate if sectors_per_cylinder is a power of
4013			 * 2.  Otherwise it might be slightly off -- you
4014			 * might have a bit of a truncation problem.
4015			 */
4016#ifdef	__XSCALE__
4017			cylinders = (lun->be_lun->maxlba + 1) /
4018				sectors_per_cylinder;
4019#else
4020			for (shift = 31; shift > 0; shift--) {
4021				if (sectors_per_cylinder & (1 << shift))
4022					break;
4023			}
4024			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4025#endif
4026
4027			/*
4028			 * We've basically got 3 bytes, or 24 bits for the
4029			 * cylinder size in the mode page.  If we're over,
4030			 * just round down to 2^24.
4031			 */
4032			if (cylinders > 0xffffff)
4033				cylinders = 0xffffff;
4034
4035			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4036				CTL_PAGE_DEFAULT];
4037			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4038
4039			if ((value = ctl_get_opt(&lun->be_lun->options,
4040			    "rpm")) != NULL) {
4041				scsi_ulto2b(strtol(value, NULL, 0),
4042				     rigid_disk_page->rotation_rate);
4043			}
4044
4045			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4046			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4047			       sizeof(rigid_disk_page_default));
4048			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4049			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4050			       sizeof(rigid_disk_page_default));
4051
4052			page_index->page_data =
4053				(uint8_t *)lun->mode_pages.rigid_disk_page;
4054			break;
4055		}
4056		case SMS_CACHING_PAGE: {
4057			struct scsi_caching_page *caching_page;
4058
4059			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4060				panic("invalid subpage value %d",
4061				      page_index->subpage);
4062			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4063			       &caching_page_default,
4064			       sizeof(caching_page_default));
4065			memcpy(&lun->mode_pages.caching_page[
4066			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4067			       sizeof(caching_page_changeable));
4068			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4069			       &caching_page_default,
4070			       sizeof(caching_page_default));
4071			caching_page = &lun->mode_pages.caching_page[
4072			    CTL_PAGE_SAVED];
4073			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4074			if (value != NULL && strcmp(value, "off") == 0)
4075				caching_page->flags1 &= ~SCP_WCE;
4076			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4077			if (value != NULL && strcmp(value, "off") == 0)
4078				caching_page->flags1 |= SCP_RCD;
4079			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4080			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4081			       sizeof(caching_page_default));
4082			page_index->page_data =
4083				(uint8_t *)lun->mode_pages.caching_page;
4084			break;
4085		}
4086		case SMS_CONTROL_MODE_PAGE: {
4087			struct scsi_control_page *control_page;
4088
4089			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4090				panic("invalid subpage value %d",
4091				      page_index->subpage);
4092
4093			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4094			       &control_page_default,
4095			       sizeof(control_page_default));
4096			memcpy(&lun->mode_pages.control_page[
4097			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4098			       sizeof(control_page_changeable));
4099			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4100			       &control_page_default,
4101			       sizeof(control_page_default));
4102			control_page = &lun->mode_pages.control_page[
4103			    CTL_PAGE_SAVED];
4104			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4105			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4106				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4107				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4108			}
4109			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4110			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4111			       sizeof(control_page_default));
4112			page_index->page_data =
4113				(uint8_t *)lun->mode_pages.control_page;
4114			break;
4115
4116		}
4117		case SMS_INFO_EXCEPTIONS_PAGE: {
4118			switch (page_index->subpage) {
4119			case SMS_SUBPAGE_PAGE_0:
4120				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4121				       &ie_page_default,
4122				       sizeof(ie_page_default));
4123				memcpy(&lun->mode_pages.ie_page[
4124				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4125				       sizeof(ie_page_changeable));
4126				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4127				       &ie_page_default,
4128				       sizeof(ie_page_default));
4129				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4130				       &ie_page_default,
4131				       sizeof(ie_page_default));
4132				page_index->page_data =
4133					(uint8_t *)lun->mode_pages.ie_page;
4134				break;
4135			case 0x02: {
4136				struct ctl_logical_block_provisioning_page *page;
4137
4138				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4139				       &lbp_page_default,
4140				       sizeof(lbp_page_default));
4141				memcpy(&lun->mode_pages.lbp_page[
4142				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4143				       sizeof(lbp_page_changeable));
4144				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4145				       &lbp_page_default,
4146				       sizeof(lbp_page_default));
4147				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4148				value = ctl_get_opt(&lun->be_lun->options,
4149				    "avail-threshold");
4150				if (value != NULL &&
4151				    ctl_expand_number(value, &ival) == 0) {
4152					page->descr[0].flags |= SLBPPD_ENABLED |
4153					    SLBPPD_ARMING_DEC;
4154					if (lun->be_lun->blocksize)
4155						ival /= lun->be_lun->blocksize;
4156					else
4157						ival /= 512;
4158					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4159					    page->descr[0].count);
4160				}
4161				value = ctl_get_opt(&lun->be_lun->options,
4162				    "used-threshold");
4163				if (value != NULL &&
4164				    ctl_expand_number(value, &ival) == 0) {
4165					page->descr[1].flags |= SLBPPD_ENABLED |
4166					    SLBPPD_ARMING_INC;
4167					if (lun->be_lun->blocksize)
4168						ival /= lun->be_lun->blocksize;
4169					else
4170						ival /= 512;
4171					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4172					    page->descr[1].count);
4173				}
4174				value = ctl_get_opt(&lun->be_lun->options,
4175				    "pool-avail-threshold");
4176				if (value != NULL &&
4177				    ctl_expand_number(value, &ival) == 0) {
4178					page->descr[2].flags |= SLBPPD_ENABLED |
4179					    SLBPPD_ARMING_DEC;
4180					if (lun->be_lun->blocksize)
4181						ival /= lun->be_lun->blocksize;
4182					else
4183						ival /= 512;
4184					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4185					    page->descr[2].count);
4186				}
4187				value = ctl_get_opt(&lun->be_lun->options,
4188				    "pool-used-threshold");
4189				if (value != NULL &&
4190				    ctl_expand_number(value, &ival) == 0) {
4191					page->descr[3].flags |= SLBPPD_ENABLED |
4192					    SLBPPD_ARMING_INC;
4193					if (lun->be_lun->blocksize)
4194						ival /= lun->be_lun->blocksize;
4195					else
4196						ival /= 512;
4197					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4198					    page->descr[3].count);
4199				}
4200				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4201				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4202				       sizeof(lbp_page_default));
4203				page_index->page_data =
4204					(uint8_t *)lun->mode_pages.lbp_page;
4205			}}
4206			break;
4207		}
4208		case SMS_VENDOR_SPECIFIC_PAGE:{
4209			switch (page_index->subpage) {
4210			case DBGCNF_SUBPAGE_CODE: {
4211				struct copan_debugconf_subpage *current_page,
4212							       *saved_page;
4213
4214				memcpy(&lun->mode_pages.debugconf_subpage[
4215				       CTL_PAGE_CURRENT],
4216				       &debugconf_page_default,
4217				       sizeof(debugconf_page_default));
4218				memcpy(&lun->mode_pages.debugconf_subpage[
4219				       CTL_PAGE_CHANGEABLE],
4220				       &debugconf_page_changeable,
4221				       sizeof(debugconf_page_changeable));
4222				memcpy(&lun->mode_pages.debugconf_subpage[
4223				       CTL_PAGE_DEFAULT],
4224				       &debugconf_page_default,
4225				       sizeof(debugconf_page_default));
4226				memcpy(&lun->mode_pages.debugconf_subpage[
4227				       CTL_PAGE_SAVED],
4228				       &debugconf_page_default,
4229				       sizeof(debugconf_page_default));
4230				page_index->page_data =
4231					(uint8_t *)lun->mode_pages.debugconf_subpage;
4232
4233				current_page = (struct copan_debugconf_subpage *)
4234					(page_index->page_data +
4235					 (page_index->page_len *
4236					  CTL_PAGE_CURRENT));
4237				saved_page = (struct copan_debugconf_subpage *)
4238					(page_index->page_data +
4239					 (page_index->page_len *
4240					  CTL_PAGE_SAVED));
4241				break;
4242			}
4243			default:
4244				panic("invalid subpage value %d",
4245				      page_index->subpage);
4246				break;
4247			}
4248   			break;
4249		}
4250		default:
4251			panic("invalid page value %d",
4252			      page_index->page_code & SMPH_PC_MASK);
4253			break;
4254    	}
4255	}
4256
4257	return (CTL_RETVAL_COMPLETE);
4258}
4259
4260static int
4261ctl_init_log_page_index(struct ctl_lun *lun)
4262{
4263	struct ctl_page_index *page_index;
4264	int i, j, k, prev;
4265
4266	memcpy(&lun->log_pages.index, log_page_index_template,
4267	       sizeof(log_page_index_template));
4268
4269	prev = -1;
4270	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4271
4272		page_index = &lun->log_pages.index[i];
4273		/*
4274		 * If this is a disk-only mode page, there's no point in
4275		 * setting it up.  For some pages, we have to have some
4276		 * basic information about the disk in order to calculate the
4277		 * mode page data.
4278		 */
4279		if ((lun->be_lun->lun_type != T_DIRECT)
4280		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4281			continue;
4282
4283		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4284		    ((lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) == 0 ||
4285		     lun->backend->lun_attr == NULL))
4286			continue;
4287
4288		if (page_index->page_code != prev) {
4289			lun->log_pages.pages_page[j] = page_index->page_code;
4290			prev = page_index->page_code;
4291			j++;
4292		}
4293		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4294		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4295		k++;
4296	}
4297	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4298	lun->log_pages.index[0].page_len = j;
4299	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4300	lun->log_pages.index[1].page_len = k * 2;
4301	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4302	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4303
4304	return (CTL_RETVAL_COMPLETE);
4305}
4306
4307static int
4308hex2bin(const char *str, uint8_t *buf, int buf_size)
4309{
4310	int i;
4311	u_char c;
4312
4313	memset(buf, 0, buf_size);
4314	while (isspace(str[0]))
4315		str++;
4316	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4317		str += 2;
4318	buf_size *= 2;
4319	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4320		c = str[i];
4321		if (isdigit(c))
4322			c -= '0';
4323		else if (isalpha(c))
4324			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4325		else
4326			break;
4327		if (c >= 16)
4328			break;
4329		if ((i & 1) == 0)
4330			buf[i / 2] |= (c << 4);
4331		else
4332			buf[i / 2] |= c;
4333	}
4334	return ((i + 1) / 2);
4335}
4336
4337/*
4338 * LUN allocation.
4339 *
4340 * Requirements:
4341 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4342 *   wants us to allocate the LUN and he can block.
4343 * - ctl_softc is always set
4344 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4345 *
4346 * Returns 0 for success, non-zero (errno) for failure.
4347 */
4348static int
4349ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4350	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4351{
4352	struct ctl_lun *nlun, *lun;
4353	struct ctl_port *port;
4354	struct scsi_vpd_id_descriptor *desc;
4355	struct scsi_vpd_id_t10 *t10id;
4356	const char *eui, *naa, *scsiname, *vendor, *value;
4357	int lun_number, i, lun_malloced;
4358	int devidlen, idlen1, idlen2 = 0, len;
4359
4360	if (be_lun == NULL)
4361		return (EINVAL);
4362
4363	/*
4364	 * We currently only support Direct Access or Processor LUN types.
4365	 */
4366	switch (be_lun->lun_type) {
4367	case T_DIRECT:
4368		break;
4369	case T_PROCESSOR:
4370		break;
4371	case T_SEQUENTIAL:
4372	case T_CHANGER:
4373	default:
4374		be_lun->lun_config_status(be_lun->be_lun,
4375					  CTL_LUN_CONFIG_FAILURE);
4376		break;
4377	}
4378	if (ctl_lun == NULL) {
4379		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4380		lun_malloced = 1;
4381	} else {
4382		lun_malloced = 0;
4383		lun = ctl_lun;
4384	}
4385
4386	memset(lun, 0, sizeof(*lun));
4387	if (lun_malloced)
4388		lun->flags = CTL_LUN_MALLOCED;
4389
4390	/* Generate LUN ID. */
4391	devidlen = max(CTL_DEVID_MIN_LEN,
4392	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4393	idlen1 = sizeof(*t10id) + devidlen;
4394	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4395	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4396	if (scsiname != NULL) {
4397		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4398		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4399	}
4400	eui = ctl_get_opt(&be_lun->options, "eui");
4401	if (eui != NULL) {
4402		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4403	}
4404	naa = ctl_get_opt(&be_lun->options, "naa");
4405	if (naa != NULL) {
4406		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4407	}
4408	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4409	    M_CTL, M_WAITOK | M_ZERO);
4410	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4411	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4412	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4413	desc->length = idlen1;
4414	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4415	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4416	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4417		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4418	} else {
4419		strncpy(t10id->vendor, vendor,
4420		    min(sizeof(t10id->vendor), strlen(vendor)));
4421	}
4422	strncpy((char *)t10id->vendor_spec_id,
4423	    (char *)be_lun->device_id, devidlen);
4424	if (scsiname != NULL) {
4425		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4426		    desc->length);
4427		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4428		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4429		    SVPD_ID_TYPE_SCSI_NAME;
4430		desc->length = idlen2;
4431		strlcpy(desc->identifier, scsiname, idlen2);
4432	}
4433	if (eui != NULL) {
4434		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4435		    desc->length);
4436		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4437		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4438		    SVPD_ID_TYPE_EUI64;
4439		desc->length = hex2bin(eui, desc->identifier, 16);
4440		desc->length = desc->length > 12 ? 16 :
4441		    (desc->length > 8 ? 12 : 8);
4442		len -= 16 - desc->length;
4443	}
4444	if (naa != NULL) {
4445		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4446		    desc->length);
4447		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4448		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4449		    SVPD_ID_TYPE_NAA;
4450		desc->length = hex2bin(naa, desc->identifier, 16);
4451		desc->length = desc->length > 8 ? 16 : 8;
4452		len -= 16 - desc->length;
4453	}
4454	lun->lun_devid->len = len;
4455
4456	mtx_lock(&ctl_softc->ctl_lock);
4457	/*
4458	 * See if the caller requested a particular LUN number.  If so, see
4459	 * if it is available.  Otherwise, allocate the first available LUN.
4460	 */
4461	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4462		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4463		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4464			mtx_unlock(&ctl_softc->ctl_lock);
4465			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4466				printf("ctl: requested LUN ID %d is higher "
4467				       "than CTL_MAX_LUNS - 1 (%d)\n",
4468				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4469			} else {
4470				/*
4471				 * XXX KDM return an error, or just assign
4472				 * another LUN ID in this case??
4473				 */
4474				printf("ctl: requested LUN ID %d is already "
4475				       "in use\n", be_lun->req_lun_id);
4476			}
4477			if (lun->flags & CTL_LUN_MALLOCED)
4478				free(lun, M_CTL);
4479			be_lun->lun_config_status(be_lun->be_lun,
4480						  CTL_LUN_CONFIG_FAILURE);
4481			return (ENOSPC);
4482		}
4483		lun_number = be_lun->req_lun_id;
4484	} else {
4485		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4486		if (lun_number == -1) {
4487			mtx_unlock(&ctl_softc->ctl_lock);
4488			printf("ctl: can't allocate LUN on target %ju, out of "
4489			       "LUNs\n", (uintmax_t)target_id.id);
4490			if (lun->flags & CTL_LUN_MALLOCED)
4491				free(lun, M_CTL);
4492			be_lun->lun_config_status(be_lun->be_lun,
4493						  CTL_LUN_CONFIG_FAILURE);
4494			return (ENOSPC);
4495		}
4496	}
4497	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4498
4499	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4500	lun->target = target_id;
4501	lun->lun = lun_number;
4502	lun->be_lun = be_lun;
4503	/*
4504	 * The processor LUN is always enabled.  Disk LUNs come on line
4505	 * disabled, and must be enabled by the backend.
4506	 */
4507	lun->flags |= CTL_LUN_DISABLED;
4508	lun->backend = be_lun->be;
4509	be_lun->ctl_lun = lun;
4510	be_lun->lun_id = lun_number;
4511	atomic_add_int(&be_lun->be->num_luns, 1);
4512	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4513		lun->flags |= CTL_LUN_OFFLINE;
4514
4515	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4516		lun->flags |= CTL_LUN_STOPPED;
4517
4518	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4519		lun->flags |= CTL_LUN_INOPERABLE;
4520
4521	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4522		lun->flags |= CTL_LUN_PRIMARY_SC;
4523
4524	value = ctl_get_opt(&be_lun->options, "readonly");
4525	if (value != NULL && strcmp(value, "on") == 0)
4526		lun->flags |= CTL_LUN_READONLY;
4527
4528	lun->ctl_softc = ctl_softc;
4529	TAILQ_INIT(&lun->ooa_queue);
4530	TAILQ_INIT(&lun->blocked_queue);
4531	STAILQ_INIT(&lun->error_list);
4532	ctl_tpc_lun_init(lun);
4533
4534	/*
4535	 * Initialize the mode and log page index.
4536	 */
4537	ctl_init_page_index(lun);
4538	ctl_init_log_page_index(lun);
4539
4540	/*
4541	 * Set the poweron UA for all initiators on this LUN only.
4542	 */
4543	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4544		lun->pending_ua[i] = CTL_UA_POWERON;
4545
4546	/*
4547	 * Now, before we insert this lun on the lun list, set the lun
4548	 * inventory changed UA for all other luns.
4549	 */
4550	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4551		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4552			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4553		}
4554	}
4555
4556	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4557
4558	ctl_softc->ctl_luns[lun_number] = lun;
4559
4560	ctl_softc->num_luns++;
4561
4562	/* Setup statistics gathering */
4563	lun->stats.device_type = be_lun->lun_type;
4564	lun->stats.lun_number = lun_number;
4565	if (lun->stats.device_type == T_DIRECT)
4566		lun->stats.blocksize = be_lun->blocksize;
4567	else
4568		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4569	for (i = 0;i < CTL_MAX_PORTS;i++)
4570		lun->stats.ports[i].targ_port = i;
4571
4572	mtx_unlock(&ctl_softc->ctl_lock);
4573
4574	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4575
4576	/*
4577	 * Run through each registered FETD and bring it online if it isn't
4578	 * already.  Enable the target ID if it hasn't been enabled, and
4579	 * enable this particular LUN.
4580	 */
4581	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4582		int retval;
4583
4584		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4585		if (retval != 0) {
4586			printf("ctl_alloc_lun: FETD %s port %d returned error "
4587			       "%d for lun_enable on target %ju lun %d\n",
4588			       port->port_name, port->targ_port, retval,
4589			       (uintmax_t)target_id.id, lun_number);
4590		} else
4591			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4592	}
4593	return (0);
4594}
4595
4596/*
4597 * Delete a LUN.
4598 * Assumptions:
4599 * - LUN has already been marked invalid and any pending I/O has been taken
4600 *   care of.
4601 */
4602static int
4603ctl_free_lun(struct ctl_lun *lun)
4604{
4605	struct ctl_softc *softc;
4606#if 0
4607	struct ctl_port *port;
4608#endif
4609	struct ctl_lun *nlun;
4610	int i;
4611
4612	softc = lun->ctl_softc;
4613
4614	mtx_assert(&softc->ctl_lock, MA_OWNED);
4615
4616	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4617
4618	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4619
4620	softc->ctl_luns[lun->lun] = NULL;
4621
4622	if (!TAILQ_EMPTY(&lun->ooa_queue))
4623		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4624
4625	softc->num_luns--;
4626
4627	/*
4628	 * XXX KDM this scheme only works for a single target/multiple LUN
4629	 * setup.  It needs to be revamped for a multiple target scheme.
4630	 *
4631	 * XXX KDM this results in port->lun_disable() getting called twice,
4632	 * once when ctl_disable_lun() is called, and a second time here.
4633	 * We really need to re-think the LUN disable semantics.  There
4634	 * should probably be several steps/levels to LUN removal:
4635	 *  - disable
4636	 *  - invalidate
4637	 *  - free
4638 	 *
4639	 * Right now we only have a disable method when communicating to
4640	 * the front end ports, at least for individual LUNs.
4641	 */
4642#if 0
4643	STAILQ_FOREACH(port, &softc->port_list, links) {
4644		int retval;
4645
4646		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4647					 lun->lun);
4648		if (retval != 0) {
4649			printf("ctl_free_lun: FETD %s port %d returned error "
4650			       "%d for lun_disable on target %ju lun %jd\n",
4651			       port->port_name, port->targ_port, retval,
4652			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4653		}
4654
4655		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4656			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4657
4658			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4659			if (retval != 0) {
4660				printf("ctl_free_lun: FETD %s port %d "
4661				       "returned error %d for targ_disable on "
4662				       "target %ju\n", port->port_name,
4663				       port->targ_port, retval,
4664				       (uintmax_t)lun->target.id);
4665			} else
4666				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4667
4668			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4669				continue;
4670
4671#if 0
4672			port->port_offline(port->onoff_arg);
4673			port->status &= ~CTL_PORT_STATUS_ONLINE;
4674#endif
4675		}
4676	}
4677#endif
4678
4679	/*
4680	 * Tell the backend to free resources, if this LUN has a backend.
4681	 */
4682	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4683	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4684
4685	ctl_tpc_lun_shutdown(lun);
4686	mtx_destroy(&lun->lun_lock);
4687	free(lun->lun_devid, M_CTL);
4688	free(lun->write_buffer, M_CTL);
4689	if (lun->flags & CTL_LUN_MALLOCED)
4690		free(lun, M_CTL);
4691
4692	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4693		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4694			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4695		}
4696	}
4697
4698	return (0);
4699}
4700
4701static void
4702ctl_create_lun(struct ctl_be_lun *be_lun)
4703{
4704	struct ctl_softc *ctl_softc;
4705
4706	ctl_softc = control_softc;
4707
4708	/*
4709	 * ctl_alloc_lun() should handle all potential failure cases.
4710	 */
4711	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4712}
4713
4714int
4715ctl_add_lun(struct ctl_be_lun *be_lun)
4716{
4717	struct ctl_softc *ctl_softc = control_softc;
4718
4719	mtx_lock(&ctl_softc->ctl_lock);
4720	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4721	mtx_unlock(&ctl_softc->ctl_lock);
4722	wakeup(&ctl_softc->pending_lun_queue);
4723
4724	return (0);
4725}
4726
4727int
4728ctl_enable_lun(struct ctl_be_lun *be_lun)
4729{
4730	struct ctl_softc *ctl_softc;
4731	struct ctl_port *port, *nport;
4732	struct ctl_lun *lun;
4733	int retval;
4734
4735	ctl_softc = control_softc;
4736
4737	lun = (struct ctl_lun *)be_lun->ctl_lun;
4738
4739	mtx_lock(&ctl_softc->ctl_lock);
4740	mtx_lock(&lun->lun_lock);
4741	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4742		/*
4743		 * eh?  Why did we get called if the LUN is already
4744		 * enabled?
4745		 */
4746		mtx_unlock(&lun->lun_lock);
4747		mtx_unlock(&ctl_softc->ctl_lock);
4748		return (0);
4749	}
4750	lun->flags &= ~CTL_LUN_DISABLED;
4751	mtx_unlock(&lun->lun_lock);
4752
4753	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4754		nport = STAILQ_NEXT(port, links);
4755
4756		/*
4757		 * Drop the lock while we call the FETD's enable routine.
4758		 * This can lead to a callback into CTL (at least in the
4759		 * case of the internal initiator frontend.
4760		 */
4761		mtx_unlock(&ctl_softc->ctl_lock);
4762		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4763		mtx_lock(&ctl_softc->ctl_lock);
4764		if (retval != 0) {
4765			printf("%s: FETD %s port %d returned error "
4766			       "%d for lun_enable on target %ju lun %jd\n",
4767			       __func__, port->port_name, port->targ_port, retval,
4768			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4769		}
4770#if 0
4771		 else {
4772            /* NOTE:  TODO:  why does lun enable affect port status? */
4773			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4774		}
4775#endif
4776	}
4777
4778	mtx_unlock(&ctl_softc->ctl_lock);
4779
4780	return (0);
4781}
4782
4783int
4784ctl_disable_lun(struct ctl_be_lun *be_lun)
4785{
4786	struct ctl_softc *ctl_softc;
4787	struct ctl_port *port;
4788	struct ctl_lun *lun;
4789	int retval;
4790
4791	ctl_softc = control_softc;
4792
4793	lun = (struct ctl_lun *)be_lun->ctl_lun;
4794
4795	mtx_lock(&ctl_softc->ctl_lock);
4796	mtx_lock(&lun->lun_lock);
4797	if (lun->flags & CTL_LUN_DISABLED) {
4798		mtx_unlock(&lun->lun_lock);
4799		mtx_unlock(&ctl_softc->ctl_lock);
4800		return (0);
4801	}
4802	lun->flags |= CTL_LUN_DISABLED;
4803	mtx_unlock(&lun->lun_lock);
4804
4805	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4806		mtx_unlock(&ctl_softc->ctl_lock);
4807		/*
4808		 * Drop the lock before we call the frontend's disable
4809		 * routine, to avoid lock order reversals.
4810		 *
4811		 * XXX KDM what happens if the frontend list changes while
4812		 * we're traversing it?  It's unlikely, but should be handled.
4813		 */
4814		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4815					 lun->lun);
4816		mtx_lock(&ctl_softc->ctl_lock);
4817		if (retval != 0) {
4818			printf("ctl_alloc_lun: FETD %s port %d returned error "
4819			       "%d for lun_disable on target %ju lun %jd\n",
4820			       port->port_name, port->targ_port, retval,
4821			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4822		}
4823	}
4824
4825	mtx_unlock(&ctl_softc->ctl_lock);
4826
4827	return (0);
4828}
4829
4830int
4831ctl_start_lun(struct ctl_be_lun *be_lun)
4832{
4833	struct ctl_softc *ctl_softc;
4834	struct ctl_lun *lun;
4835
4836	ctl_softc = control_softc;
4837
4838	lun = (struct ctl_lun *)be_lun->ctl_lun;
4839
4840	mtx_lock(&lun->lun_lock);
4841	lun->flags &= ~CTL_LUN_STOPPED;
4842	mtx_unlock(&lun->lun_lock);
4843
4844	return (0);
4845}
4846
4847int
4848ctl_stop_lun(struct ctl_be_lun *be_lun)
4849{
4850	struct ctl_softc *ctl_softc;
4851	struct ctl_lun *lun;
4852
4853	ctl_softc = control_softc;
4854
4855	lun = (struct ctl_lun *)be_lun->ctl_lun;
4856
4857	mtx_lock(&lun->lun_lock);
4858	lun->flags |= CTL_LUN_STOPPED;
4859	mtx_unlock(&lun->lun_lock);
4860
4861	return (0);
4862}
4863
4864int
4865ctl_lun_offline(struct ctl_be_lun *be_lun)
4866{
4867	struct ctl_softc *ctl_softc;
4868	struct ctl_lun *lun;
4869
4870	ctl_softc = control_softc;
4871
4872	lun = (struct ctl_lun *)be_lun->ctl_lun;
4873
4874	mtx_lock(&lun->lun_lock);
4875	lun->flags |= CTL_LUN_OFFLINE;
4876	mtx_unlock(&lun->lun_lock);
4877
4878	return (0);
4879}
4880
4881int
4882ctl_lun_online(struct ctl_be_lun *be_lun)
4883{
4884	struct ctl_softc *ctl_softc;
4885	struct ctl_lun *lun;
4886
4887	ctl_softc = control_softc;
4888
4889	lun = (struct ctl_lun *)be_lun->ctl_lun;
4890
4891	mtx_lock(&lun->lun_lock);
4892	lun->flags &= ~CTL_LUN_OFFLINE;
4893	mtx_unlock(&lun->lun_lock);
4894
4895	return (0);
4896}
4897
4898int
4899ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4900{
4901	struct ctl_softc *ctl_softc;
4902	struct ctl_lun *lun;
4903
4904	ctl_softc = control_softc;
4905
4906	lun = (struct ctl_lun *)be_lun->ctl_lun;
4907
4908	mtx_lock(&lun->lun_lock);
4909
4910	/*
4911	 * The LUN needs to be disabled before it can be marked invalid.
4912	 */
4913	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4914		mtx_unlock(&lun->lun_lock);
4915		return (-1);
4916	}
4917	/*
4918	 * Mark the LUN invalid.
4919	 */
4920	lun->flags |= CTL_LUN_INVALID;
4921
4922	/*
4923	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4924	 * If we have something in the OOA queue, we'll free it when the
4925	 * last I/O completes.
4926	 */
4927	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4928		mtx_unlock(&lun->lun_lock);
4929		mtx_lock(&ctl_softc->ctl_lock);
4930		ctl_free_lun(lun);
4931		mtx_unlock(&ctl_softc->ctl_lock);
4932	} else
4933		mtx_unlock(&lun->lun_lock);
4934
4935	return (0);
4936}
4937
4938int
4939ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4940{
4941	struct ctl_softc *ctl_softc;
4942	struct ctl_lun *lun;
4943
4944	ctl_softc = control_softc;
4945	lun = (struct ctl_lun *)be_lun->ctl_lun;
4946
4947	mtx_lock(&lun->lun_lock);
4948	lun->flags |= CTL_LUN_INOPERABLE;
4949	mtx_unlock(&lun->lun_lock);
4950
4951	return (0);
4952}
4953
4954int
4955ctl_lun_operable(struct ctl_be_lun *be_lun)
4956{
4957	struct ctl_softc *ctl_softc;
4958	struct ctl_lun *lun;
4959
4960	ctl_softc = control_softc;
4961	lun = (struct ctl_lun *)be_lun->ctl_lun;
4962
4963	mtx_lock(&lun->lun_lock);
4964	lun->flags &= ~CTL_LUN_INOPERABLE;
4965	mtx_unlock(&lun->lun_lock);
4966
4967	return (0);
4968}
4969
4970void
4971ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4972{
4973	struct ctl_lun *lun;
4974	struct ctl_softc *softc;
4975	int i;
4976
4977	softc = control_softc;
4978
4979	lun = (struct ctl_lun *)be_lun->ctl_lun;
4980
4981	mtx_lock(&lun->lun_lock);
4982
4983	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4984		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
4985
4986	mtx_unlock(&lun->lun_lock);
4987}
4988
4989/*
4990 * Backend "memory move is complete" callback for requests that never
4991 * make it down to say RAIDCore's configuration code.
4992 */
4993int
4994ctl_config_move_done(union ctl_io *io)
4995{
4996	int retval;
4997
4998	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4999	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5000	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5001
5002	if ((io->io_hdr.port_status != 0) &&
5003	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5004	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5005		/*
5006		 * For hardware error sense keys, the sense key
5007		 * specific value is defined to be a retry count,
5008		 * but we use it to pass back an internal FETD
5009		 * error code.  XXX KDM  Hopefully the FETD is only
5010		 * using 16 bits for an error code, since that's
5011		 * all the space we have in the sks field.
5012		 */
5013		ctl_set_internal_failure(&io->scsiio,
5014					 /*sks_valid*/ 1,
5015					 /*retry_count*/
5016					 io->io_hdr.port_status);
5017	}
5018
5019	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5020	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5021	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5022	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5023		/*
5024		 * XXX KDM just assuming a single pointer here, and not a
5025		 * S/G list.  If we start using S/G lists for config data,
5026		 * we'll need to know how to clean them up here as well.
5027		 */
5028		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5029			free(io->scsiio.kern_data_ptr, M_CTL);
5030		ctl_done(io);
5031		retval = CTL_RETVAL_COMPLETE;
5032	} else {
5033		/*
5034		 * XXX KDM now we need to continue data movement.  Some
5035		 * options:
5036		 * - call ctl_scsiio() again?  We don't do this for data
5037		 *   writes, because for those at least we know ahead of
5038		 *   time where the write will go and how long it is.  For
5039		 *   config writes, though, that information is largely
5040		 *   contained within the write itself, thus we need to
5041		 *   parse out the data again.
5042		 *
5043		 * - Call some other function once the data is in?
5044		 */
5045		if (ctl_debug & CTL_DEBUG_CDB_DATA)
5046			ctl_data_print(io);
5047
5048		/*
5049		 * XXX KDM call ctl_scsiio() again for now, and check flag
5050		 * bits to see whether we're allocated or not.
5051		 */
5052		retval = ctl_scsiio(&io->scsiio);
5053	}
5054	return (retval);
5055}
5056
5057/*
5058 * This gets called by a backend driver when it is done with a
5059 * data_submit method.
5060 */
5061void
5062ctl_data_submit_done(union ctl_io *io)
5063{
5064	/*
5065	 * If the IO_CONT flag is set, we need to call the supplied
5066	 * function to continue processing the I/O, instead of completing
5067	 * the I/O just yet.
5068	 *
5069	 * If there is an error, though, we don't want to keep processing.
5070	 * Instead, just send status back to the initiator.
5071	 */
5072	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5073	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5074	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5075	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5076		io->scsiio.io_cont(io);
5077		return;
5078	}
5079	ctl_done(io);
5080}
5081
5082/*
5083 * This gets called by a backend driver when it is done with a
5084 * configuration write.
5085 */
5086void
5087ctl_config_write_done(union ctl_io *io)
5088{
5089	uint8_t *buf;
5090
5091	/*
5092	 * If the IO_CONT flag is set, we need to call the supplied
5093	 * function to continue processing the I/O, instead of completing
5094	 * the I/O just yet.
5095	 *
5096	 * If there is an error, though, we don't want to keep processing.
5097	 * Instead, just send status back to the initiator.
5098	 */
5099	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5100	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5101	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5102	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5103		io->scsiio.io_cont(io);
5104		return;
5105	}
5106	/*
5107	 * Since a configuration write can be done for commands that actually
5108	 * have data allocated, like write buffer, and commands that have
5109	 * no data, like start/stop unit, we need to check here.
5110	 */
5111	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5112		buf = io->scsiio.kern_data_ptr;
5113	else
5114		buf = NULL;
5115	ctl_done(io);
5116	if (buf)
5117		free(buf, M_CTL);
5118}
5119
5120/*
5121 * SCSI release command.
5122 */
5123int
5124ctl_scsi_release(struct ctl_scsiio *ctsio)
5125{
5126	int length, longid, thirdparty_id, resv_id;
5127	struct ctl_softc *ctl_softc;
5128	struct ctl_lun *lun;
5129	uint32_t residx;
5130
5131	length = 0;
5132	resv_id = 0;
5133
5134	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5135
5136	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5137	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5138	ctl_softc = control_softc;
5139
5140	switch (ctsio->cdb[0]) {
5141	case RELEASE_10: {
5142		struct scsi_release_10 *cdb;
5143
5144		cdb = (struct scsi_release_10 *)ctsio->cdb;
5145
5146		if (cdb->byte2 & SR10_LONGID)
5147			longid = 1;
5148		else
5149			thirdparty_id = cdb->thirdparty_id;
5150
5151		resv_id = cdb->resv_id;
5152		length = scsi_2btoul(cdb->length);
5153		break;
5154	}
5155	}
5156
5157
5158	/*
5159	 * XXX KDM right now, we only support LUN reservation.  We don't
5160	 * support 3rd party reservations, or extent reservations, which
5161	 * might actually need the parameter list.  If we've gotten this
5162	 * far, we've got a LUN reservation.  Anything else got kicked out
5163	 * above.  So, according to SPC, ignore the length.
5164	 */
5165	length = 0;
5166
5167	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5168	 && (length > 0)) {
5169		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5170		ctsio->kern_data_len = length;
5171		ctsio->kern_total_len = length;
5172		ctsio->kern_data_resid = 0;
5173		ctsio->kern_rel_offset = 0;
5174		ctsio->kern_sg_entries = 0;
5175		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5176		ctsio->be_move_done = ctl_config_move_done;
5177		ctl_datamove((union ctl_io *)ctsio);
5178
5179		return (CTL_RETVAL_COMPLETE);
5180	}
5181
5182	if (length > 0)
5183		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5184
5185	mtx_lock(&lun->lun_lock);
5186
5187	/*
5188	 * According to SPC, it is not an error for an intiator to attempt
5189	 * to release a reservation on a LUN that isn't reserved, or that
5190	 * is reserved by another initiator.  The reservation can only be
5191	 * released, though, by the initiator who made it or by one of
5192	 * several reset type events.
5193	 */
5194	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5195			lun->flags &= ~CTL_LUN_RESERVED;
5196
5197	mtx_unlock(&lun->lun_lock);
5198
5199	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5200		free(ctsio->kern_data_ptr, M_CTL);
5201		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5202	}
5203
5204	ctl_set_success(ctsio);
5205	ctl_done((union ctl_io *)ctsio);
5206	return (CTL_RETVAL_COMPLETE);
5207}
5208
5209int
5210ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5211{
5212	int extent, thirdparty, longid;
5213	int resv_id, length;
5214	uint64_t thirdparty_id;
5215	struct ctl_softc *ctl_softc;
5216	struct ctl_lun *lun;
5217	uint32_t residx;
5218
5219	extent = 0;
5220	thirdparty = 0;
5221	longid = 0;
5222	resv_id = 0;
5223	length = 0;
5224	thirdparty_id = 0;
5225
5226	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5227
5228	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5229	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5230	ctl_softc = control_softc;
5231
5232	switch (ctsio->cdb[0]) {
5233	case RESERVE_10: {
5234		struct scsi_reserve_10 *cdb;
5235
5236		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5237
5238		if (cdb->byte2 & SR10_LONGID)
5239			longid = 1;
5240		else
5241			thirdparty_id = cdb->thirdparty_id;
5242
5243		resv_id = cdb->resv_id;
5244		length = scsi_2btoul(cdb->length);
5245		break;
5246	}
5247	}
5248
5249	/*
5250	 * XXX KDM right now, we only support LUN reservation.  We don't
5251	 * support 3rd party reservations, or extent reservations, which
5252	 * might actually need the parameter list.  If we've gotten this
5253	 * far, we've got a LUN reservation.  Anything else got kicked out
5254	 * above.  So, according to SPC, ignore the length.
5255	 */
5256	length = 0;
5257
5258	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5259	 && (length > 0)) {
5260		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5261		ctsio->kern_data_len = length;
5262		ctsio->kern_total_len = length;
5263		ctsio->kern_data_resid = 0;
5264		ctsio->kern_rel_offset = 0;
5265		ctsio->kern_sg_entries = 0;
5266		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5267		ctsio->be_move_done = ctl_config_move_done;
5268		ctl_datamove((union ctl_io *)ctsio);
5269
5270		return (CTL_RETVAL_COMPLETE);
5271	}
5272
5273	if (length > 0)
5274		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5275
5276	mtx_lock(&lun->lun_lock);
5277	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5278		ctl_set_reservation_conflict(ctsio);
5279		goto bailout;
5280	}
5281
5282	lun->flags |= CTL_LUN_RESERVED;
5283	lun->res_idx = residx;
5284
5285	ctl_set_success(ctsio);
5286
5287bailout:
5288	mtx_unlock(&lun->lun_lock);
5289
5290	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5291		free(ctsio->kern_data_ptr, M_CTL);
5292		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5293	}
5294
5295	ctl_done((union ctl_io *)ctsio);
5296	return (CTL_RETVAL_COMPLETE);
5297}
5298
5299int
5300ctl_start_stop(struct ctl_scsiio *ctsio)
5301{
5302	struct scsi_start_stop_unit *cdb;
5303	struct ctl_lun *lun;
5304	struct ctl_softc *ctl_softc;
5305	int retval;
5306
5307	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5308
5309	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5310	ctl_softc = control_softc;
5311	retval = 0;
5312
5313	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5314
5315	/*
5316	 * XXX KDM
5317	 * We don't support the immediate bit on a stop unit.  In order to
5318	 * do that, we would need to code up a way to know that a stop is
5319	 * pending, and hold off any new commands until it completes, one
5320	 * way or another.  Then we could accept or reject those commands
5321	 * depending on its status.  We would almost need to do the reverse
5322	 * of what we do below for an immediate start -- return the copy of
5323	 * the ctl_io to the FETD with status to send to the host (and to
5324	 * free the copy!) and then free the original I/O once the stop
5325	 * actually completes.  That way, the OOA queue mechanism can work
5326	 * to block commands that shouldn't proceed.  Another alternative
5327	 * would be to put the copy in the queue in place of the original,
5328	 * and return the original back to the caller.  That could be
5329	 * slightly safer..
5330	 */
5331	if ((cdb->byte2 & SSS_IMMED)
5332	 && ((cdb->how & SSS_START) == 0)) {
5333		ctl_set_invalid_field(ctsio,
5334				      /*sks_valid*/ 1,
5335				      /*command*/ 1,
5336				      /*field*/ 1,
5337				      /*bit_valid*/ 1,
5338				      /*bit*/ 0);
5339		ctl_done((union ctl_io *)ctsio);
5340		return (CTL_RETVAL_COMPLETE);
5341	}
5342
5343	if ((lun->flags & CTL_LUN_PR_RESERVED)
5344	 && ((cdb->how & SSS_START)==0)) {
5345		uint32_t residx;
5346
5347		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5348		if (lun->pr_keys[residx] == 0
5349		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5350
5351			ctl_set_reservation_conflict(ctsio);
5352			ctl_done((union ctl_io *)ctsio);
5353			return (CTL_RETVAL_COMPLETE);
5354		}
5355	}
5356
5357	/*
5358	 * If there is no backend on this device, we can't start or stop
5359	 * it.  In theory we shouldn't get any start/stop commands in the
5360	 * first place at this level if the LUN doesn't have a backend.
5361	 * That should get stopped by the command decode code.
5362	 */
5363	if (lun->backend == NULL) {
5364		ctl_set_invalid_opcode(ctsio);
5365		ctl_done((union ctl_io *)ctsio);
5366		return (CTL_RETVAL_COMPLETE);
5367	}
5368
5369	/*
5370	 * XXX KDM Copan-specific offline behavior.
5371	 * Figure out a reasonable way to port this?
5372	 */
5373#ifdef NEEDTOPORT
5374	mtx_lock(&lun->lun_lock);
5375
5376	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5377	 && (lun->flags & CTL_LUN_OFFLINE)) {
5378		/*
5379		 * If the LUN is offline, and the on/offline bit isn't set,
5380		 * reject the start or stop.  Otherwise, let it through.
5381		 */
5382		mtx_unlock(&lun->lun_lock);
5383		ctl_set_lun_not_ready(ctsio);
5384		ctl_done((union ctl_io *)ctsio);
5385	} else {
5386		mtx_unlock(&lun->lun_lock);
5387#endif /* NEEDTOPORT */
5388		/*
5389		 * This could be a start or a stop when we're online,
5390		 * or a stop/offline or start/online.  A start or stop when
5391		 * we're offline is covered in the case above.
5392		 */
5393		/*
5394		 * In the non-immediate case, we send the request to
5395		 * the backend and return status to the user when
5396		 * it is done.
5397		 *
5398		 * In the immediate case, we allocate a new ctl_io
5399		 * to hold a copy of the request, and send that to
5400		 * the backend.  We then set good status on the
5401		 * user's request and return it immediately.
5402		 */
5403		if (cdb->byte2 & SSS_IMMED) {
5404			union ctl_io *new_io;
5405
5406			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5407			ctl_copy_io((union ctl_io *)ctsio, new_io);
5408			retval = lun->backend->config_write(new_io);
5409			ctl_set_success(ctsio);
5410			ctl_done((union ctl_io *)ctsio);
5411		} else {
5412			retval = lun->backend->config_write(
5413				(union ctl_io *)ctsio);
5414		}
5415#ifdef NEEDTOPORT
5416	}
5417#endif
5418	return (retval);
5419}
5420
5421/*
5422 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5423 * we don't really do anything with the LBA and length fields if the user
5424 * passes them in.  Instead we'll just flush out the cache for the entire
5425 * LUN.
5426 */
5427int
5428ctl_sync_cache(struct ctl_scsiio *ctsio)
5429{
5430	struct ctl_lun *lun;
5431	struct ctl_softc *ctl_softc;
5432	uint64_t starting_lba;
5433	uint32_t block_count;
5434	int retval;
5435
5436	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5437
5438	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5439	ctl_softc = control_softc;
5440	retval = 0;
5441
5442	switch (ctsio->cdb[0]) {
5443	case SYNCHRONIZE_CACHE: {
5444		struct scsi_sync_cache *cdb;
5445		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5446
5447		starting_lba = scsi_4btoul(cdb->begin_lba);
5448		block_count = scsi_2btoul(cdb->lb_count);
5449		break;
5450	}
5451	case SYNCHRONIZE_CACHE_16: {
5452		struct scsi_sync_cache_16 *cdb;
5453		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5454
5455		starting_lba = scsi_8btou64(cdb->begin_lba);
5456		block_count = scsi_4btoul(cdb->lb_count);
5457		break;
5458	}
5459	default:
5460		ctl_set_invalid_opcode(ctsio);
5461		ctl_done((union ctl_io *)ctsio);
5462		goto bailout;
5463		break; /* NOTREACHED */
5464	}
5465
5466	/*
5467	 * We check the LBA and length, but don't do anything with them.
5468	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5469	 * get flushed.  This check will just help satisfy anyone who wants
5470	 * to see an error for an out of range LBA.
5471	 */
5472	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5473		ctl_set_lba_out_of_range(ctsio);
5474		ctl_done((union ctl_io *)ctsio);
5475		goto bailout;
5476	}
5477
5478	/*
5479	 * If this LUN has no backend, we can't flush the cache anyway.
5480	 */
5481	if (lun->backend == NULL) {
5482		ctl_set_invalid_opcode(ctsio);
5483		ctl_done((union ctl_io *)ctsio);
5484		goto bailout;
5485	}
5486
5487	/*
5488	 * Check to see whether we're configured to send the SYNCHRONIZE
5489	 * CACHE command directly to the back end.
5490	 */
5491	mtx_lock(&lun->lun_lock);
5492	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5493	 && (++(lun->sync_count) >= lun->sync_interval)) {
5494		lun->sync_count = 0;
5495		mtx_unlock(&lun->lun_lock);
5496		retval = lun->backend->config_write((union ctl_io *)ctsio);
5497	} else {
5498		mtx_unlock(&lun->lun_lock);
5499		ctl_set_success(ctsio);
5500		ctl_done((union ctl_io *)ctsio);
5501	}
5502
5503bailout:
5504
5505	return (retval);
5506}
5507
5508int
5509ctl_format(struct ctl_scsiio *ctsio)
5510{
5511	struct scsi_format *cdb;
5512	struct ctl_lun *lun;
5513	struct ctl_softc *ctl_softc;
5514	int length, defect_list_len;
5515
5516	CTL_DEBUG_PRINT(("ctl_format\n"));
5517
5518	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5519	ctl_softc = control_softc;
5520
5521	cdb = (struct scsi_format *)ctsio->cdb;
5522
5523	length = 0;
5524	if (cdb->byte2 & SF_FMTDATA) {
5525		if (cdb->byte2 & SF_LONGLIST)
5526			length = sizeof(struct scsi_format_header_long);
5527		else
5528			length = sizeof(struct scsi_format_header_short);
5529	}
5530
5531	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5532	 && (length > 0)) {
5533		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5534		ctsio->kern_data_len = length;
5535		ctsio->kern_total_len = length;
5536		ctsio->kern_data_resid = 0;
5537		ctsio->kern_rel_offset = 0;
5538		ctsio->kern_sg_entries = 0;
5539		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5540		ctsio->be_move_done = ctl_config_move_done;
5541		ctl_datamove((union ctl_io *)ctsio);
5542
5543		return (CTL_RETVAL_COMPLETE);
5544	}
5545
5546	defect_list_len = 0;
5547
5548	if (cdb->byte2 & SF_FMTDATA) {
5549		if (cdb->byte2 & SF_LONGLIST) {
5550			struct scsi_format_header_long *header;
5551
5552			header = (struct scsi_format_header_long *)
5553				ctsio->kern_data_ptr;
5554
5555			defect_list_len = scsi_4btoul(header->defect_list_len);
5556			if (defect_list_len != 0) {
5557				ctl_set_invalid_field(ctsio,
5558						      /*sks_valid*/ 1,
5559						      /*command*/ 0,
5560						      /*field*/ 2,
5561						      /*bit_valid*/ 0,
5562						      /*bit*/ 0);
5563				goto bailout;
5564			}
5565		} else {
5566			struct scsi_format_header_short *header;
5567
5568			header = (struct scsi_format_header_short *)
5569				ctsio->kern_data_ptr;
5570
5571			defect_list_len = scsi_2btoul(header->defect_list_len);
5572			if (defect_list_len != 0) {
5573				ctl_set_invalid_field(ctsio,
5574						      /*sks_valid*/ 1,
5575						      /*command*/ 0,
5576						      /*field*/ 2,
5577						      /*bit_valid*/ 0,
5578						      /*bit*/ 0);
5579				goto bailout;
5580			}
5581		}
5582	}
5583
5584	/*
5585	 * The format command will clear out the "Medium format corrupted"
5586	 * status if set by the configuration code.  That status is really
5587	 * just a way to notify the host that we have lost the media, and
5588	 * get them to issue a command that will basically make them think
5589	 * they're blowing away the media.
5590	 */
5591	mtx_lock(&lun->lun_lock);
5592	lun->flags &= ~CTL_LUN_INOPERABLE;
5593	mtx_unlock(&lun->lun_lock);
5594
5595	ctl_set_success(ctsio);
5596bailout:
5597
5598	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5599		free(ctsio->kern_data_ptr, M_CTL);
5600		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5601	}
5602
5603	ctl_done((union ctl_io *)ctsio);
5604	return (CTL_RETVAL_COMPLETE);
5605}
5606
5607int
5608ctl_read_buffer(struct ctl_scsiio *ctsio)
5609{
5610	struct scsi_read_buffer *cdb;
5611	struct ctl_lun *lun;
5612	int buffer_offset, len;
5613	static uint8_t descr[4];
5614	static uint8_t echo_descr[4] = { 0 };
5615
5616	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5617
5618	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5619	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5620
5621	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5622	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5623	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5624		ctl_set_invalid_field(ctsio,
5625				      /*sks_valid*/ 1,
5626				      /*command*/ 1,
5627				      /*field*/ 1,
5628				      /*bit_valid*/ 1,
5629				      /*bit*/ 4);
5630		ctl_done((union ctl_io *)ctsio);
5631		return (CTL_RETVAL_COMPLETE);
5632	}
5633
5634	len = scsi_3btoul(cdb->length);
5635	buffer_offset = scsi_3btoul(cdb->offset);
5636
5637	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5638		ctl_set_invalid_field(ctsio,
5639				      /*sks_valid*/ 1,
5640				      /*command*/ 1,
5641				      /*field*/ 6,
5642				      /*bit_valid*/ 0,
5643				      /*bit*/ 0);
5644		ctl_done((union ctl_io *)ctsio);
5645		return (CTL_RETVAL_COMPLETE);
5646	}
5647
5648	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5649		descr[0] = 0;
5650		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5651		ctsio->kern_data_ptr = descr;
5652		len = min(len, sizeof(descr));
5653	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5654		ctsio->kern_data_ptr = echo_descr;
5655		len = min(len, sizeof(echo_descr));
5656	} else {
5657		if (lun->write_buffer == NULL) {
5658			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5659			    M_CTL, M_WAITOK);
5660		}
5661		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5662	}
5663	ctsio->kern_data_len = len;
5664	ctsio->kern_total_len = len;
5665	ctsio->kern_data_resid = 0;
5666	ctsio->kern_rel_offset = 0;
5667	ctsio->kern_sg_entries = 0;
5668	ctl_set_success(ctsio);
5669	ctsio->be_move_done = ctl_config_move_done;
5670	ctl_datamove((union ctl_io *)ctsio);
5671	return (CTL_RETVAL_COMPLETE);
5672}
5673
5674int
5675ctl_write_buffer(struct ctl_scsiio *ctsio)
5676{
5677	struct scsi_write_buffer *cdb;
5678	struct ctl_lun *lun;
5679	int buffer_offset, len;
5680
5681	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5682
5683	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5684	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5685
5686	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5687		ctl_set_invalid_field(ctsio,
5688				      /*sks_valid*/ 1,
5689				      /*command*/ 1,
5690				      /*field*/ 1,
5691				      /*bit_valid*/ 1,
5692				      /*bit*/ 4);
5693		ctl_done((union ctl_io *)ctsio);
5694		return (CTL_RETVAL_COMPLETE);
5695	}
5696
5697	len = scsi_3btoul(cdb->length);
5698	buffer_offset = scsi_3btoul(cdb->offset);
5699
5700	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5701		ctl_set_invalid_field(ctsio,
5702				      /*sks_valid*/ 1,
5703				      /*command*/ 1,
5704				      /*field*/ 6,
5705				      /*bit_valid*/ 0,
5706				      /*bit*/ 0);
5707		ctl_done((union ctl_io *)ctsio);
5708		return (CTL_RETVAL_COMPLETE);
5709	}
5710
5711	/*
5712	 * If we've got a kernel request that hasn't been malloced yet,
5713	 * malloc it and tell the caller the data buffer is here.
5714	 */
5715	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5716		if (lun->write_buffer == NULL) {
5717			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5718			    M_CTL, M_WAITOK);
5719		}
5720		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5721		ctsio->kern_data_len = len;
5722		ctsio->kern_total_len = len;
5723		ctsio->kern_data_resid = 0;
5724		ctsio->kern_rel_offset = 0;
5725		ctsio->kern_sg_entries = 0;
5726		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5727		ctsio->be_move_done = ctl_config_move_done;
5728		ctl_datamove((union ctl_io *)ctsio);
5729
5730		return (CTL_RETVAL_COMPLETE);
5731	}
5732
5733	ctl_set_success(ctsio);
5734	ctl_done((union ctl_io *)ctsio);
5735	return (CTL_RETVAL_COMPLETE);
5736}
5737
5738int
5739ctl_write_same(struct ctl_scsiio *ctsio)
5740{
5741	struct ctl_lun *lun;
5742	struct ctl_lba_len_flags *lbalen;
5743	uint64_t lba;
5744	uint32_t num_blocks;
5745	int len, retval;
5746	uint8_t byte2;
5747
5748	retval = CTL_RETVAL_COMPLETE;
5749
5750	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5751
5752	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5753
5754	switch (ctsio->cdb[0]) {
5755	case WRITE_SAME_10: {
5756		struct scsi_write_same_10 *cdb;
5757
5758		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5759
5760		lba = scsi_4btoul(cdb->addr);
5761		num_blocks = scsi_2btoul(cdb->length);
5762		byte2 = cdb->byte2;
5763		break;
5764	}
5765	case WRITE_SAME_16: {
5766		struct scsi_write_same_16 *cdb;
5767
5768		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5769
5770		lba = scsi_8btou64(cdb->addr);
5771		num_blocks = scsi_4btoul(cdb->length);
5772		byte2 = cdb->byte2;
5773		break;
5774	}
5775	default:
5776		/*
5777		 * We got a command we don't support.  This shouldn't
5778		 * happen, commands should be filtered out above us.
5779		 */
5780		ctl_set_invalid_opcode(ctsio);
5781		ctl_done((union ctl_io *)ctsio);
5782
5783		return (CTL_RETVAL_COMPLETE);
5784		break; /* NOTREACHED */
5785	}
5786
5787	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5788	if ((byte2 & SWS_UNMAP) == 0 &&
5789	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5790		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5791		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5792		ctl_done((union ctl_io *)ctsio);
5793		return (CTL_RETVAL_COMPLETE);
5794	}
5795
5796	/*
5797	 * The first check is to make sure we're in bounds, the second
5798	 * check is to catch wrap-around problems.  If the lba + num blocks
5799	 * is less than the lba, then we've wrapped around and the block
5800	 * range is invalid anyway.
5801	 */
5802	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5803	 || ((lba + num_blocks) < lba)) {
5804		ctl_set_lba_out_of_range(ctsio);
5805		ctl_done((union ctl_io *)ctsio);
5806		return (CTL_RETVAL_COMPLETE);
5807	}
5808
5809	/* Zero number of blocks means "to the last logical block" */
5810	if (num_blocks == 0) {
5811		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5812			ctl_set_invalid_field(ctsio,
5813					      /*sks_valid*/ 0,
5814					      /*command*/ 1,
5815					      /*field*/ 0,
5816					      /*bit_valid*/ 0,
5817					      /*bit*/ 0);
5818			ctl_done((union ctl_io *)ctsio);
5819			return (CTL_RETVAL_COMPLETE);
5820		}
5821		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5822	}
5823
5824	len = lun->be_lun->blocksize;
5825
5826	/*
5827	 * If we've got a kernel request that hasn't been malloced yet,
5828	 * malloc it and tell the caller the data buffer is here.
5829	 */
5830	if ((byte2 & SWS_NDOB) == 0 &&
5831	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5832		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5833		ctsio->kern_data_len = len;
5834		ctsio->kern_total_len = len;
5835		ctsio->kern_data_resid = 0;
5836		ctsio->kern_rel_offset = 0;
5837		ctsio->kern_sg_entries = 0;
5838		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5839		ctsio->be_move_done = ctl_config_move_done;
5840		ctl_datamove((union ctl_io *)ctsio);
5841
5842		return (CTL_RETVAL_COMPLETE);
5843	}
5844
5845	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5846	lbalen->lba = lba;
5847	lbalen->len = num_blocks;
5848	lbalen->flags = byte2;
5849	retval = lun->backend->config_write((union ctl_io *)ctsio);
5850
5851	return (retval);
5852}
5853
5854int
5855ctl_unmap(struct ctl_scsiio *ctsio)
5856{
5857	struct ctl_lun *lun;
5858	struct scsi_unmap *cdb;
5859	struct ctl_ptr_len_flags *ptrlen;
5860	struct scsi_unmap_header *hdr;
5861	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5862	uint64_t lba;
5863	uint32_t num_blocks;
5864	int len, retval;
5865	uint8_t byte2;
5866
5867	retval = CTL_RETVAL_COMPLETE;
5868
5869	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5870
5871	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5872	cdb = (struct scsi_unmap *)ctsio->cdb;
5873
5874	len = scsi_2btoul(cdb->length);
5875	byte2 = cdb->byte2;
5876
5877	/*
5878	 * If we've got a kernel request that hasn't been malloced yet,
5879	 * malloc it and tell the caller the data buffer is here.
5880	 */
5881	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5882		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5883		ctsio->kern_data_len = len;
5884		ctsio->kern_total_len = len;
5885		ctsio->kern_data_resid = 0;
5886		ctsio->kern_rel_offset = 0;
5887		ctsio->kern_sg_entries = 0;
5888		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5889		ctsio->be_move_done = ctl_config_move_done;
5890		ctl_datamove((union ctl_io *)ctsio);
5891
5892		return (CTL_RETVAL_COMPLETE);
5893	}
5894
5895	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5896	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5897	if (len < sizeof (*hdr) ||
5898	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5899	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5900	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5901		ctl_set_invalid_field(ctsio,
5902				      /*sks_valid*/ 0,
5903				      /*command*/ 0,
5904				      /*field*/ 0,
5905				      /*bit_valid*/ 0,
5906				      /*bit*/ 0);
5907		ctl_done((union ctl_io *)ctsio);
5908		return (CTL_RETVAL_COMPLETE);
5909	}
5910	len = scsi_2btoul(hdr->desc_length);
5911	buf = (struct scsi_unmap_desc *)(hdr + 1);
5912	end = buf + len / sizeof(*buf);
5913
5914	endnz = buf;
5915	for (range = buf; range < end; range++) {
5916		lba = scsi_8btou64(range->lba);
5917		num_blocks = scsi_4btoul(range->length);
5918		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5919		 || ((lba + num_blocks) < lba)) {
5920			ctl_set_lba_out_of_range(ctsio);
5921			ctl_done((union ctl_io *)ctsio);
5922			return (CTL_RETVAL_COMPLETE);
5923		}
5924		if (num_blocks != 0)
5925			endnz = range + 1;
5926	}
5927
5928	/*
5929	 * Block backend can not handle zero last range.
5930	 * Filter it out and return if there is nothing left.
5931	 */
5932	len = (uint8_t *)endnz - (uint8_t *)buf;
5933	if (len == 0) {
5934		ctl_set_success(ctsio);
5935		ctl_done((union ctl_io *)ctsio);
5936		return (CTL_RETVAL_COMPLETE);
5937	}
5938
5939	mtx_lock(&lun->lun_lock);
5940	ptrlen = (struct ctl_ptr_len_flags *)
5941	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5942	ptrlen->ptr = (void *)buf;
5943	ptrlen->len = len;
5944	ptrlen->flags = byte2;
5945	ctl_check_blocked(lun);
5946	mtx_unlock(&lun->lun_lock);
5947
5948	retval = lun->backend->config_write((union ctl_io *)ctsio);
5949	return (retval);
5950}
5951
5952/*
5953 * Note that this function currently doesn't actually do anything inside
5954 * CTL to enforce things if the DQue bit is turned on.
5955 *
5956 * Also note that this function can't be used in the default case, because
5957 * the DQue bit isn't set in the changeable mask for the control mode page
5958 * anyway.  This is just here as an example for how to implement a page
5959 * handler, and a placeholder in case we want to allow the user to turn
5960 * tagged queueing on and off.
5961 *
5962 * The D_SENSE bit handling is functional, however, and will turn
5963 * descriptor sense on and off for a given LUN.
5964 */
5965int
5966ctl_control_page_handler(struct ctl_scsiio *ctsio,
5967			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5968{
5969	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5970	struct ctl_lun *lun;
5971	struct ctl_softc *softc;
5972	int set_ua;
5973	uint32_t initidx;
5974
5975	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5976	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5977	set_ua = 0;
5978
5979	user_cp = (struct scsi_control_page *)page_ptr;
5980	current_cp = (struct scsi_control_page *)
5981		(page_index->page_data + (page_index->page_len *
5982		CTL_PAGE_CURRENT));
5983	saved_cp = (struct scsi_control_page *)
5984		(page_index->page_data + (page_index->page_len *
5985		CTL_PAGE_SAVED));
5986
5987	softc = control_softc;
5988
5989	mtx_lock(&lun->lun_lock);
5990	if (((current_cp->rlec & SCP_DSENSE) == 0)
5991	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5992		/*
5993		 * Descriptor sense is currently turned off and the user
5994		 * wants to turn it on.
5995		 */
5996		current_cp->rlec |= SCP_DSENSE;
5997		saved_cp->rlec |= SCP_DSENSE;
5998		lun->flags |= CTL_LUN_SENSE_DESC;
5999		set_ua = 1;
6000	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6001		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6002		/*
6003		 * Descriptor sense is currently turned on, and the user
6004		 * wants to turn it off.
6005		 */
6006		current_cp->rlec &= ~SCP_DSENSE;
6007		saved_cp->rlec &= ~SCP_DSENSE;
6008		lun->flags &= ~CTL_LUN_SENSE_DESC;
6009		set_ua = 1;
6010	}
6011	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6012	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6013		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6014		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6015		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6016		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6017		set_ua = 1;
6018	}
6019	if ((current_cp->eca_and_aen & SCP_SWP) !=
6020	    (user_cp->eca_and_aen & SCP_SWP)) {
6021		current_cp->eca_and_aen &= ~SCP_SWP;
6022		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6023		saved_cp->eca_and_aen &= ~SCP_SWP;
6024		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6025		set_ua = 1;
6026	}
6027	if (set_ua != 0) {
6028		int i;
6029		/*
6030		 * Let other initiators know that the mode
6031		 * parameters for this LUN have changed.
6032		 */
6033		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6034			if (i == initidx)
6035				continue;
6036
6037			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6038		}
6039	}
6040	mtx_unlock(&lun->lun_lock);
6041
6042	return (0);
6043}
6044
6045int
6046ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6047		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6048{
6049	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6050	struct ctl_lun *lun;
6051	int set_ua;
6052	uint32_t initidx;
6053
6054	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6055	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6056	set_ua = 0;
6057
6058	user_cp = (struct scsi_caching_page *)page_ptr;
6059	current_cp = (struct scsi_caching_page *)
6060		(page_index->page_data + (page_index->page_len *
6061		CTL_PAGE_CURRENT));
6062	saved_cp = (struct scsi_caching_page *)
6063		(page_index->page_data + (page_index->page_len *
6064		CTL_PAGE_SAVED));
6065
6066	mtx_lock(&lun->lun_lock);
6067	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6068	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6069		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6070		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6071		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6072		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6073		set_ua = 1;
6074	}
6075	if (set_ua != 0) {
6076		int i;
6077		/*
6078		 * Let other initiators know that the mode
6079		 * parameters for this LUN have changed.
6080		 */
6081		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6082			if (i == initidx)
6083				continue;
6084
6085			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6086		}
6087	}
6088	mtx_unlock(&lun->lun_lock);
6089
6090	return (0);
6091}
6092
6093int
6094ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6095				struct ctl_page_index *page_index,
6096				uint8_t *page_ptr)
6097{
6098	uint8_t *c;
6099	int i;
6100
6101	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6102	ctl_time_io_secs =
6103		(c[0] << 8) |
6104		(c[1] << 0) |
6105		0;
6106	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6107	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6108	printf("page data:");
6109	for (i=0; i<8; i++)
6110		printf(" %.2x",page_ptr[i]);
6111	printf("\n");
6112	return (0);
6113}
6114
6115int
6116ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6117			       struct ctl_page_index *page_index,
6118			       int pc)
6119{
6120	struct copan_debugconf_subpage *page;
6121
6122	page = (struct copan_debugconf_subpage *)page_index->page_data +
6123		(page_index->page_len * pc);
6124
6125	switch (pc) {
6126	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6127	case SMS_PAGE_CTRL_DEFAULT >> 6:
6128	case SMS_PAGE_CTRL_SAVED >> 6:
6129		/*
6130		 * We don't update the changable or default bits for this page.
6131		 */
6132		break;
6133	case SMS_PAGE_CTRL_CURRENT >> 6:
6134		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6135		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6136		break;
6137	default:
6138#ifdef NEEDTOPORT
6139		EPRINT(0, "Invalid PC %d!!", pc);
6140#endif /* NEEDTOPORT */
6141		break;
6142	}
6143	return (0);
6144}
6145
6146
6147static int
6148ctl_do_mode_select(union ctl_io *io)
6149{
6150	struct scsi_mode_page_header *page_header;
6151	struct ctl_page_index *page_index;
6152	struct ctl_scsiio *ctsio;
6153	int control_dev, page_len;
6154	int page_len_offset, page_len_size;
6155	union ctl_modepage_info *modepage_info;
6156	struct ctl_lun *lun;
6157	int *len_left, *len_used;
6158	int retval, i;
6159
6160	ctsio = &io->scsiio;
6161	page_index = NULL;
6162	page_len = 0;
6163	retval = CTL_RETVAL_COMPLETE;
6164
6165	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6166
6167	if (lun->be_lun->lun_type != T_DIRECT)
6168		control_dev = 1;
6169	else
6170		control_dev = 0;
6171
6172	modepage_info = (union ctl_modepage_info *)
6173		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6174	len_left = &modepage_info->header.len_left;
6175	len_used = &modepage_info->header.len_used;
6176
6177do_next_page:
6178
6179	page_header = (struct scsi_mode_page_header *)
6180		(ctsio->kern_data_ptr + *len_used);
6181
6182	if (*len_left == 0) {
6183		free(ctsio->kern_data_ptr, M_CTL);
6184		ctl_set_success(ctsio);
6185		ctl_done((union ctl_io *)ctsio);
6186		return (CTL_RETVAL_COMPLETE);
6187	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6188
6189		free(ctsio->kern_data_ptr, M_CTL);
6190		ctl_set_param_len_error(ctsio);
6191		ctl_done((union ctl_io *)ctsio);
6192		return (CTL_RETVAL_COMPLETE);
6193
6194	} else if ((page_header->page_code & SMPH_SPF)
6195		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6196
6197		free(ctsio->kern_data_ptr, M_CTL);
6198		ctl_set_param_len_error(ctsio);
6199		ctl_done((union ctl_io *)ctsio);
6200		return (CTL_RETVAL_COMPLETE);
6201	}
6202
6203
6204	/*
6205	 * XXX KDM should we do something with the block descriptor?
6206	 */
6207	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6208
6209		if ((control_dev != 0)
6210		 && (lun->mode_pages.index[i].page_flags &
6211		     CTL_PAGE_FLAG_DISK_ONLY))
6212			continue;
6213
6214		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6215		    (page_header->page_code & SMPH_PC_MASK))
6216			continue;
6217
6218		/*
6219		 * If neither page has a subpage code, then we've got a
6220		 * match.
6221		 */
6222		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6223		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6224			page_index = &lun->mode_pages.index[i];
6225			page_len = page_header->page_length;
6226			break;
6227		}
6228
6229		/*
6230		 * If both pages have subpages, then the subpage numbers
6231		 * have to match.
6232		 */
6233		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6234		  && (page_header->page_code & SMPH_SPF)) {
6235			struct scsi_mode_page_header_sp *sph;
6236
6237			sph = (struct scsi_mode_page_header_sp *)page_header;
6238
6239			if (lun->mode_pages.index[i].subpage ==
6240			    sph->subpage) {
6241				page_index = &lun->mode_pages.index[i];
6242				page_len = scsi_2btoul(sph->page_length);
6243				break;
6244			}
6245		}
6246	}
6247
6248	/*
6249	 * If we couldn't find the page, or if we don't have a mode select
6250	 * handler for it, send back an error to the user.
6251	 */
6252	if ((page_index == NULL)
6253	 || (page_index->select_handler == NULL)) {
6254		ctl_set_invalid_field(ctsio,
6255				      /*sks_valid*/ 1,
6256				      /*command*/ 0,
6257				      /*field*/ *len_used,
6258				      /*bit_valid*/ 0,
6259				      /*bit*/ 0);
6260		free(ctsio->kern_data_ptr, M_CTL);
6261		ctl_done((union ctl_io *)ctsio);
6262		return (CTL_RETVAL_COMPLETE);
6263	}
6264
6265	if (page_index->page_code & SMPH_SPF) {
6266		page_len_offset = 2;
6267		page_len_size = 2;
6268	} else {
6269		page_len_size = 1;
6270		page_len_offset = 1;
6271	}
6272
6273	/*
6274	 * If the length the initiator gives us isn't the one we specify in
6275	 * the mode page header, or if they didn't specify enough data in
6276	 * the CDB to avoid truncating this page, kick out the request.
6277	 */
6278	if ((page_len != (page_index->page_len - page_len_offset -
6279			  page_len_size))
6280	 || (*len_left < page_index->page_len)) {
6281
6282
6283		ctl_set_invalid_field(ctsio,
6284				      /*sks_valid*/ 1,
6285				      /*command*/ 0,
6286				      /*field*/ *len_used + page_len_offset,
6287				      /*bit_valid*/ 0,
6288				      /*bit*/ 0);
6289		free(ctsio->kern_data_ptr, M_CTL);
6290		ctl_done((union ctl_io *)ctsio);
6291		return (CTL_RETVAL_COMPLETE);
6292	}
6293
6294	/*
6295	 * Run through the mode page, checking to make sure that the bits
6296	 * the user changed are actually legal for him to change.
6297	 */
6298	for (i = 0; i < page_index->page_len; i++) {
6299		uint8_t *user_byte, *change_mask, *current_byte;
6300		int bad_bit;
6301		int j;
6302
6303		user_byte = (uint8_t *)page_header + i;
6304		change_mask = page_index->page_data +
6305			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6306		current_byte = page_index->page_data +
6307			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6308
6309		/*
6310		 * Check to see whether the user set any bits in this byte
6311		 * that he is not allowed to set.
6312		 */
6313		if ((*user_byte & ~(*change_mask)) ==
6314		    (*current_byte & ~(*change_mask)))
6315			continue;
6316
6317		/*
6318		 * Go through bit by bit to determine which one is illegal.
6319		 */
6320		bad_bit = 0;
6321		for (j = 7; j >= 0; j--) {
6322			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6323			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6324				bad_bit = i;
6325				break;
6326			}
6327		}
6328		ctl_set_invalid_field(ctsio,
6329				      /*sks_valid*/ 1,
6330				      /*command*/ 0,
6331				      /*field*/ *len_used + i,
6332				      /*bit_valid*/ 1,
6333				      /*bit*/ bad_bit);
6334		free(ctsio->kern_data_ptr, M_CTL);
6335		ctl_done((union ctl_io *)ctsio);
6336		return (CTL_RETVAL_COMPLETE);
6337	}
6338
6339	/*
6340	 * Decrement these before we call the page handler, since we may
6341	 * end up getting called back one way or another before the handler
6342	 * returns to this context.
6343	 */
6344	*len_left -= page_index->page_len;
6345	*len_used += page_index->page_len;
6346
6347	retval = page_index->select_handler(ctsio, page_index,
6348					    (uint8_t *)page_header);
6349
6350	/*
6351	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6352	 * wait until this queued command completes to finish processing
6353	 * the mode page.  If it returns anything other than
6354	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6355	 * already set the sense information, freed the data pointer, and
6356	 * completed the io for us.
6357	 */
6358	if (retval != CTL_RETVAL_COMPLETE)
6359		goto bailout_no_done;
6360
6361	/*
6362	 * If the initiator sent us more than one page, parse the next one.
6363	 */
6364	if (*len_left > 0)
6365		goto do_next_page;
6366
6367	ctl_set_success(ctsio);
6368	free(ctsio->kern_data_ptr, M_CTL);
6369	ctl_done((union ctl_io *)ctsio);
6370
6371bailout_no_done:
6372
6373	return (CTL_RETVAL_COMPLETE);
6374
6375}
6376
6377int
6378ctl_mode_select(struct ctl_scsiio *ctsio)
6379{
6380	int param_len, pf, sp;
6381	int header_size, bd_len;
6382	int len_left, len_used;
6383	struct ctl_page_index *page_index;
6384	struct ctl_lun *lun;
6385	int control_dev, page_len;
6386	union ctl_modepage_info *modepage_info;
6387	int retval;
6388
6389	pf = 0;
6390	sp = 0;
6391	page_len = 0;
6392	len_used = 0;
6393	len_left = 0;
6394	retval = 0;
6395	bd_len = 0;
6396	page_index = NULL;
6397
6398	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6399
6400	if (lun->be_lun->lun_type != T_DIRECT)
6401		control_dev = 1;
6402	else
6403		control_dev = 0;
6404
6405	switch (ctsio->cdb[0]) {
6406	case MODE_SELECT_6: {
6407		struct scsi_mode_select_6 *cdb;
6408
6409		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6410
6411		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6412		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6413
6414		param_len = cdb->length;
6415		header_size = sizeof(struct scsi_mode_header_6);
6416		break;
6417	}
6418	case MODE_SELECT_10: {
6419		struct scsi_mode_select_10 *cdb;
6420
6421		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6422
6423		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6424		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6425
6426		param_len = scsi_2btoul(cdb->length);
6427		header_size = sizeof(struct scsi_mode_header_10);
6428		break;
6429	}
6430	default:
6431		ctl_set_invalid_opcode(ctsio);
6432		ctl_done((union ctl_io *)ctsio);
6433		return (CTL_RETVAL_COMPLETE);
6434		break; /* NOTREACHED */
6435	}
6436
6437	/*
6438	 * From SPC-3:
6439	 * "A parameter list length of zero indicates that the Data-Out Buffer
6440	 * shall be empty. This condition shall not be considered as an error."
6441	 */
6442	if (param_len == 0) {
6443		ctl_set_success(ctsio);
6444		ctl_done((union ctl_io *)ctsio);
6445		return (CTL_RETVAL_COMPLETE);
6446	}
6447
6448	/*
6449	 * Since we'll hit this the first time through, prior to
6450	 * allocation, we don't need to free a data buffer here.
6451	 */
6452	if (param_len < header_size) {
6453		ctl_set_param_len_error(ctsio);
6454		ctl_done((union ctl_io *)ctsio);
6455		return (CTL_RETVAL_COMPLETE);
6456	}
6457
6458	/*
6459	 * Allocate the data buffer and grab the user's data.  In theory,
6460	 * we shouldn't have to sanity check the parameter list length here
6461	 * because the maximum size is 64K.  We should be able to malloc
6462	 * that much without too many problems.
6463	 */
6464	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6465		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6466		ctsio->kern_data_len = param_len;
6467		ctsio->kern_total_len = param_len;
6468		ctsio->kern_data_resid = 0;
6469		ctsio->kern_rel_offset = 0;
6470		ctsio->kern_sg_entries = 0;
6471		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6472		ctsio->be_move_done = ctl_config_move_done;
6473		ctl_datamove((union ctl_io *)ctsio);
6474
6475		return (CTL_RETVAL_COMPLETE);
6476	}
6477
6478	switch (ctsio->cdb[0]) {
6479	case MODE_SELECT_6: {
6480		struct scsi_mode_header_6 *mh6;
6481
6482		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6483		bd_len = mh6->blk_desc_len;
6484		break;
6485	}
6486	case MODE_SELECT_10: {
6487		struct scsi_mode_header_10 *mh10;
6488
6489		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6490		bd_len = scsi_2btoul(mh10->blk_desc_len);
6491		break;
6492	}
6493	default:
6494		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6495		break;
6496	}
6497
6498	if (param_len < (header_size + bd_len)) {
6499		free(ctsio->kern_data_ptr, M_CTL);
6500		ctl_set_param_len_error(ctsio);
6501		ctl_done((union ctl_io *)ctsio);
6502		return (CTL_RETVAL_COMPLETE);
6503	}
6504
6505	/*
6506	 * Set the IO_CONT flag, so that if this I/O gets passed to
6507	 * ctl_config_write_done(), it'll get passed back to
6508	 * ctl_do_mode_select() for further processing, or completion if
6509	 * we're all done.
6510	 */
6511	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6512	ctsio->io_cont = ctl_do_mode_select;
6513
6514	modepage_info = (union ctl_modepage_info *)
6515		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6516
6517	memset(modepage_info, 0, sizeof(*modepage_info));
6518
6519	len_left = param_len - header_size - bd_len;
6520	len_used = header_size + bd_len;
6521
6522	modepage_info->header.len_left = len_left;
6523	modepage_info->header.len_used = len_used;
6524
6525	return (ctl_do_mode_select((union ctl_io *)ctsio));
6526}
6527
6528int
6529ctl_mode_sense(struct ctl_scsiio *ctsio)
6530{
6531	struct ctl_lun *lun;
6532	int pc, page_code, dbd, llba, subpage;
6533	int alloc_len, page_len, header_len, total_len;
6534	struct scsi_mode_block_descr *block_desc;
6535	struct ctl_page_index *page_index;
6536	int control_dev;
6537
6538	dbd = 0;
6539	llba = 0;
6540	block_desc = NULL;
6541	page_index = NULL;
6542
6543	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6544
6545	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6546
6547	if (lun->be_lun->lun_type != T_DIRECT)
6548		control_dev = 1;
6549	else
6550		control_dev = 0;
6551
6552	switch (ctsio->cdb[0]) {
6553	case MODE_SENSE_6: {
6554		struct scsi_mode_sense_6 *cdb;
6555
6556		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6557
6558		header_len = sizeof(struct scsi_mode_hdr_6);
6559		if (cdb->byte2 & SMS_DBD)
6560			dbd = 1;
6561		else
6562			header_len += sizeof(struct scsi_mode_block_descr);
6563
6564		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6565		page_code = cdb->page & SMS_PAGE_CODE;
6566		subpage = cdb->subpage;
6567		alloc_len = cdb->length;
6568		break;
6569	}
6570	case MODE_SENSE_10: {
6571		struct scsi_mode_sense_10 *cdb;
6572
6573		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6574
6575		header_len = sizeof(struct scsi_mode_hdr_10);
6576
6577		if (cdb->byte2 & SMS_DBD)
6578			dbd = 1;
6579		else
6580			header_len += sizeof(struct scsi_mode_block_descr);
6581		if (cdb->byte2 & SMS10_LLBAA)
6582			llba = 1;
6583		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6584		page_code = cdb->page & SMS_PAGE_CODE;
6585		subpage = cdb->subpage;
6586		alloc_len = scsi_2btoul(cdb->length);
6587		break;
6588	}
6589	default:
6590		ctl_set_invalid_opcode(ctsio);
6591		ctl_done((union ctl_io *)ctsio);
6592		return (CTL_RETVAL_COMPLETE);
6593		break; /* NOTREACHED */
6594	}
6595
6596	/*
6597	 * We have to make a first pass through to calculate the size of
6598	 * the pages that match the user's query.  Then we allocate enough
6599	 * memory to hold it, and actually copy the data into the buffer.
6600	 */
6601	switch (page_code) {
6602	case SMS_ALL_PAGES_PAGE: {
6603		int i;
6604
6605		page_len = 0;
6606
6607		/*
6608		 * At the moment, values other than 0 and 0xff here are
6609		 * reserved according to SPC-3.
6610		 */
6611		if ((subpage != SMS_SUBPAGE_PAGE_0)
6612		 && (subpage != SMS_SUBPAGE_ALL)) {
6613			ctl_set_invalid_field(ctsio,
6614					      /*sks_valid*/ 1,
6615					      /*command*/ 1,
6616					      /*field*/ 3,
6617					      /*bit_valid*/ 0,
6618					      /*bit*/ 0);
6619			ctl_done((union ctl_io *)ctsio);
6620			return (CTL_RETVAL_COMPLETE);
6621		}
6622
6623		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6624			if ((control_dev != 0)
6625			 && (lun->mode_pages.index[i].page_flags &
6626			     CTL_PAGE_FLAG_DISK_ONLY))
6627				continue;
6628
6629			/*
6630			 * We don't use this subpage if the user didn't
6631			 * request all subpages.
6632			 */
6633			if ((lun->mode_pages.index[i].subpage != 0)
6634			 && (subpage == SMS_SUBPAGE_PAGE_0))
6635				continue;
6636
6637#if 0
6638			printf("found page %#x len %d\n",
6639			       lun->mode_pages.index[i].page_code &
6640			       SMPH_PC_MASK,
6641			       lun->mode_pages.index[i].page_len);
6642#endif
6643			page_len += lun->mode_pages.index[i].page_len;
6644		}
6645		break;
6646	}
6647	default: {
6648		int i;
6649
6650		page_len = 0;
6651
6652		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6653			/* Look for the right page code */
6654			if ((lun->mode_pages.index[i].page_code &
6655			     SMPH_PC_MASK) != page_code)
6656				continue;
6657
6658			/* Look for the right subpage or the subpage wildcard*/
6659			if ((lun->mode_pages.index[i].subpage != subpage)
6660			 && (subpage != SMS_SUBPAGE_ALL))
6661				continue;
6662
6663			/* Make sure the page is supported for this dev type */
6664			if ((control_dev != 0)
6665			 && (lun->mode_pages.index[i].page_flags &
6666			     CTL_PAGE_FLAG_DISK_ONLY))
6667				continue;
6668
6669#if 0
6670			printf("found page %#x len %d\n",
6671			       lun->mode_pages.index[i].page_code &
6672			       SMPH_PC_MASK,
6673			       lun->mode_pages.index[i].page_len);
6674#endif
6675
6676			page_len += lun->mode_pages.index[i].page_len;
6677		}
6678
6679		if (page_len == 0) {
6680			ctl_set_invalid_field(ctsio,
6681					      /*sks_valid*/ 1,
6682					      /*command*/ 1,
6683					      /*field*/ 2,
6684					      /*bit_valid*/ 1,
6685					      /*bit*/ 5);
6686			ctl_done((union ctl_io *)ctsio);
6687			return (CTL_RETVAL_COMPLETE);
6688		}
6689		break;
6690	}
6691	}
6692
6693	total_len = header_len + page_len;
6694#if 0
6695	printf("header_len = %d, page_len = %d, total_len = %d\n",
6696	       header_len, page_len, total_len);
6697#endif
6698
6699	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6700	ctsio->kern_sg_entries = 0;
6701	ctsio->kern_data_resid = 0;
6702	ctsio->kern_rel_offset = 0;
6703	if (total_len < alloc_len) {
6704		ctsio->residual = alloc_len - total_len;
6705		ctsio->kern_data_len = total_len;
6706		ctsio->kern_total_len = total_len;
6707	} else {
6708		ctsio->residual = 0;
6709		ctsio->kern_data_len = alloc_len;
6710		ctsio->kern_total_len = alloc_len;
6711	}
6712
6713	switch (ctsio->cdb[0]) {
6714	case MODE_SENSE_6: {
6715		struct scsi_mode_hdr_6 *header;
6716
6717		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6718
6719		header->datalen = ctl_min(total_len - 1, 254);
6720		if (control_dev == 0) {
6721			header->dev_specific = 0x10; /* DPOFUA */
6722			if ((lun->flags & CTL_LUN_READONLY) ||
6723			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6724			    .eca_and_aen & SCP_SWP) != 0)
6725				    header->dev_specific |= 0x80; /* WP */
6726		}
6727		if (dbd)
6728			header->block_descr_len = 0;
6729		else
6730			header->block_descr_len =
6731				sizeof(struct scsi_mode_block_descr);
6732		block_desc = (struct scsi_mode_block_descr *)&header[1];
6733		break;
6734	}
6735	case MODE_SENSE_10: {
6736		struct scsi_mode_hdr_10 *header;
6737		int datalen;
6738
6739		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6740
6741		datalen = ctl_min(total_len - 2, 65533);
6742		scsi_ulto2b(datalen, header->datalen);
6743		if (control_dev == 0) {
6744			header->dev_specific = 0x10; /* DPOFUA */
6745			if ((lun->flags & CTL_LUN_READONLY) ||
6746			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6747			    .eca_and_aen & SCP_SWP) != 0)
6748				    header->dev_specific |= 0x80; /* WP */
6749		}
6750		if (dbd)
6751			scsi_ulto2b(0, header->block_descr_len);
6752		else
6753			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6754				    header->block_descr_len);
6755		block_desc = (struct scsi_mode_block_descr *)&header[1];
6756		break;
6757	}
6758	default:
6759		panic("invalid CDB type %#x", ctsio->cdb[0]);
6760		break; /* NOTREACHED */
6761	}
6762
6763	/*
6764	 * If we've got a disk, use its blocksize in the block
6765	 * descriptor.  Otherwise, just set it to 0.
6766	 */
6767	if (dbd == 0) {
6768		if (control_dev == 0)
6769			scsi_ulto3b(lun->be_lun->blocksize,
6770				    block_desc->block_len);
6771		else
6772			scsi_ulto3b(0, block_desc->block_len);
6773	}
6774
6775	switch (page_code) {
6776	case SMS_ALL_PAGES_PAGE: {
6777		int i, data_used;
6778
6779		data_used = header_len;
6780		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6781			struct ctl_page_index *page_index;
6782
6783			page_index = &lun->mode_pages.index[i];
6784
6785			if ((control_dev != 0)
6786			 && (page_index->page_flags &
6787			    CTL_PAGE_FLAG_DISK_ONLY))
6788				continue;
6789
6790			/*
6791			 * We don't use this subpage if the user didn't
6792			 * request all subpages.  We already checked (above)
6793			 * to make sure the user only specified a subpage
6794			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6795			 */
6796			if ((page_index->subpage != 0)
6797			 && (subpage == SMS_SUBPAGE_PAGE_0))
6798				continue;
6799
6800			/*
6801			 * Call the handler, if it exists, to update the
6802			 * page to the latest values.
6803			 */
6804			if (page_index->sense_handler != NULL)
6805				page_index->sense_handler(ctsio, page_index,pc);
6806
6807			memcpy(ctsio->kern_data_ptr + data_used,
6808			       page_index->page_data +
6809			       (page_index->page_len * pc),
6810			       page_index->page_len);
6811			data_used += page_index->page_len;
6812		}
6813		break;
6814	}
6815	default: {
6816		int i, data_used;
6817
6818		data_used = header_len;
6819
6820		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6821			struct ctl_page_index *page_index;
6822
6823			page_index = &lun->mode_pages.index[i];
6824
6825			/* Look for the right page code */
6826			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6827				continue;
6828
6829			/* Look for the right subpage or the subpage wildcard*/
6830			if ((page_index->subpage != subpage)
6831			 && (subpage != SMS_SUBPAGE_ALL))
6832				continue;
6833
6834			/* Make sure the page is supported for this dev type */
6835			if ((control_dev != 0)
6836			 && (page_index->page_flags &
6837			     CTL_PAGE_FLAG_DISK_ONLY))
6838				continue;
6839
6840			/*
6841			 * Call the handler, if it exists, to update the
6842			 * page to the latest values.
6843			 */
6844			if (page_index->sense_handler != NULL)
6845				page_index->sense_handler(ctsio, page_index,pc);
6846
6847			memcpy(ctsio->kern_data_ptr + data_used,
6848			       page_index->page_data +
6849			       (page_index->page_len * pc),
6850			       page_index->page_len);
6851			data_used += page_index->page_len;
6852		}
6853		break;
6854	}
6855	}
6856
6857	ctl_set_success(ctsio);
6858	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6859	ctsio->be_move_done = ctl_config_move_done;
6860	ctl_datamove((union ctl_io *)ctsio);
6861	return (CTL_RETVAL_COMPLETE);
6862}
6863
6864int
6865ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6866			       struct ctl_page_index *page_index,
6867			       int pc)
6868{
6869	struct ctl_lun *lun;
6870	struct scsi_log_param_header *phdr;
6871	uint8_t *data;
6872	uint64_t val;
6873
6874	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6875	data = page_index->page_data;
6876
6877	if (lun->backend->lun_attr != NULL &&
6878	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6879	     != UINT64_MAX) {
6880		phdr = (struct scsi_log_param_header *)data;
6881		scsi_ulto2b(0x0001, phdr->param_code);
6882		phdr->param_control = SLP_LBIN | SLP_LP;
6883		phdr->param_len = 8;
6884		data = (uint8_t *)(phdr + 1);
6885		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6886		data[4] = 0x01; /* per-LUN */
6887		data += phdr->param_len;
6888	}
6889
6890	if (lun->backend->lun_attr != NULL &&
6891	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6892	     != UINT64_MAX) {
6893		phdr = (struct scsi_log_param_header *)data;
6894		scsi_ulto2b(0x0002, phdr->param_code);
6895		phdr->param_control = SLP_LBIN | SLP_LP;
6896		phdr->param_len = 8;
6897		data = (uint8_t *)(phdr + 1);
6898		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6899		data[4] = 0x02; /* per-pool */
6900		data += phdr->param_len;
6901	}
6902
6903	if (lun->backend->lun_attr != NULL &&
6904	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6905	     != UINT64_MAX) {
6906		phdr = (struct scsi_log_param_header *)data;
6907		scsi_ulto2b(0x00f1, phdr->param_code);
6908		phdr->param_control = SLP_LBIN | SLP_LP;
6909		phdr->param_len = 8;
6910		data = (uint8_t *)(phdr + 1);
6911		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6912		data[4] = 0x02; /* per-pool */
6913		data += phdr->param_len;
6914	}
6915
6916	if (lun->backend->lun_attr != NULL &&
6917	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6918	     != UINT64_MAX) {
6919		phdr = (struct scsi_log_param_header *)data;
6920		scsi_ulto2b(0x00f2, phdr->param_code);
6921		phdr->param_control = SLP_LBIN | SLP_LP;
6922		phdr->param_len = 8;
6923		data = (uint8_t *)(phdr + 1);
6924		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6925		data[4] = 0x02; /* per-pool */
6926		data += phdr->param_len;
6927	}
6928
6929	page_index->page_len = data - page_index->page_data;
6930	return (0);
6931}
6932
6933int
6934ctl_log_sense(struct ctl_scsiio *ctsio)
6935{
6936	struct ctl_lun *lun;
6937	int i, pc, page_code, subpage;
6938	int alloc_len, total_len;
6939	struct ctl_page_index *page_index;
6940	struct scsi_log_sense *cdb;
6941	struct scsi_log_header *header;
6942
6943	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6944
6945	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6946	cdb = (struct scsi_log_sense *)ctsio->cdb;
6947	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6948	page_code = cdb->page & SLS_PAGE_CODE;
6949	subpage = cdb->subpage;
6950	alloc_len = scsi_2btoul(cdb->length);
6951
6952	page_index = NULL;
6953	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6954		page_index = &lun->log_pages.index[i];
6955
6956		/* Look for the right page code */
6957		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6958			continue;
6959
6960		/* Look for the right subpage or the subpage wildcard*/
6961		if (page_index->subpage != subpage)
6962			continue;
6963
6964		break;
6965	}
6966	if (i >= CTL_NUM_LOG_PAGES) {
6967		ctl_set_invalid_field(ctsio,
6968				      /*sks_valid*/ 1,
6969				      /*command*/ 1,
6970				      /*field*/ 2,
6971				      /*bit_valid*/ 0,
6972				      /*bit*/ 0);
6973		ctl_done((union ctl_io *)ctsio);
6974		return (CTL_RETVAL_COMPLETE);
6975	}
6976
6977	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6978
6979	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6980	ctsio->kern_sg_entries = 0;
6981	ctsio->kern_data_resid = 0;
6982	ctsio->kern_rel_offset = 0;
6983	if (total_len < alloc_len) {
6984		ctsio->residual = alloc_len - total_len;
6985		ctsio->kern_data_len = total_len;
6986		ctsio->kern_total_len = total_len;
6987	} else {
6988		ctsio->residual = 0;
6989		ctsio->kern_data_len = alloc_len;
6990		ctsio->kern_total_len = alloc_len;
6991	}
6992
6993	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6994	header->page = page_index->page_code;
6995	if (page_index->subpage) {
6996		header->page |= SL_SPF;
6997		header->subpage = page_index->subpage;
6998	}
6999	scsi_ulto2b(page_index->page_len, header->datalen);
7000
7001	/*
7002	 * Call the handler, if it exists, to update the
7003	 * page to the latest values.
7004	 */
7005	if (page_index->sense_handler != NULL)
7006		page_index->sense_handler(ctsio, page_index, pc);
7007
7008	memcpy(header + 1, page_index->page_data, page_index->page_len);
7009
7010	ctl_set_success(ctsio);
7011	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7012	ctsio->be_move_done = ctl_config_move_done;
7013	ctl_datamove((union ctl_io *)ctsio);
7014	return (CTL_RETVAL_COMPLETE);
7015}
7016
7017int
7018ctl_read_capacity(struct ctl_scsiio *ctsio)
7019{
7020	struct scsi_read_capacity *cdb;
7021	struct scsi_read_capacity_data *data;
7022	struct ctl_lun *lun;
7023	uint32_t lba;
7024
7025	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7026
7027	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7028
7029	lba = scsi_4btoul(cdb->addr);
7030	if (((cdb->pmi & SRC_PMI) == 0)
7031	 && (lba != 0)) {
7032		ctl_set_invalid_field(/*ctsio*/ ctsio,
7033				      /*sks_valid*/ 1,
7034				      /*command*/ 1,
7035				      /*field*/ 2,
7036				      /*bit_valid*/ 0,
7037				      /*bit*/ 0);
7038		ctl_done((union ctl_io *)ctsio);
7039		return (CTL_RETVAL_COMPLETE);
7040	}
7041
7042	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7043
7044	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7045	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7046	ctsio->residual = 0;
7047	ctsio->kern_data_len = sizeof(*data);
7048	ctsio->kern_total_len = sizeof(*data);
7049	ctsio->kern_data_resid = 0;
7050	ctsio->kern_rel_offset = 0;
7051	ctsio->kern_sg_entries = 0;
7052
7053	/*
7054	 * If the maximum LBA is greater than 0xfffffffe, the user must
7055	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7056	 * serivce action set.
7057	 */
7058	if (lun->be_lun->maxlba > 0xfffffffe)
7059		scsi_ulto4b(0xffffffff, data->addr);
7060	else
7061		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7062
7063	/*
7064	 * XXX KDM this may not be 512 bytes...
7065	 */
7066	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7067
7068	ctl_set_success(ctsio);
7069	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7070	ctsio->be_move_done = ctl_config_move_done;
7071	ctl_datamove((union ctl_io *)ctsio);
7072	return (CTL_RETVAL_COMPLETE);
7073}
7074
7075int
7076ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7077{
7078	struct scsi_read_capacity_16 *cdb;
7079	struct scsi_read_capacity_data_long *data;
7080	struct ctl_lun *lun;
7081	uint64_t lba;
7082	uint32_t alloc_len;
7083
7084	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7085
7086	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7087
7088	alloc_len = scsi_4btoul(cdb->alloc_len);
7089	lba = scsi_8btou64(cdb->addr);
7090
7091	if ((cdb->reladr & SRC16_PMI)
7092	 && (lba != 0)) {
7093		ctl_set_invalid_field(/*ctsio*/ ctsio,
7094				      /*sks_valid*/ 1,
7095				      /*command*/ 1,
7096				      /*field*/ 2,
7097				      /*bit_valid*/ 0,
7098				      /*bit*/ 0);
7099		ctl_done((union ctl_io *)ctsio);
7100		return (CTL_RETVAL_COMPLETE);
7101	}
7102
7103	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7104
7105	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7106	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7107
7108	if (sizeof(*data) < alloc_len) {
7109		ctsio->residual = alloc_len - sizeof(*data);
7110		ctsio->kern_data_len = sizeof(*data);
7111		ctsio->kern_total_len = sizeof(*data);
7112	} else {
7113		ctsio->residual = 0;
7114		ctsio->kern_data_len = alloc_len;
7115		ctsio->kern_total_len = alloc_len;
7116	}
7117	ctsio->kern_data_resid = 0;
7118	ctsio->kern_rel_offset = 0;
7119	ctsio->kern_sg_entries = 0;
7120
7121	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7122	/* XXX KDM this may not be 512 bytes... */
7123	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7124	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7125	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7126	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7127		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7128
7129	ctl_set_success(ctsio);
7130	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7131	ctsio->be_move_done = ctl_config_move_done;
7132	ctl_datamove((union ctl_io *)ctsio);
7133	return (CTL_RETVAL_COMPLETE);
7134}
7135
7136int
7137ctl_read_defect(struct ctl_scsiio *ctsio)
7138{
7139	struct scsi_read_defect_data_10 *ccb10;
7140	struct scsi_read_defect_data_12 *ccb12;
7141	struct scsi_read_defect_data_hdr_10 *data10;
7142	struct scsi_read_defect_data_hdr_12 *data12;
7143	uint32_t alloc_len, data_len;
7144	uint8_t format;
7145
7146	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7147
7148	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7149		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7150		format = ccb10->format;
7151		alloc_len = scsi_2btoul(ccb10->alloc_length);
7152		data_len = sizeof(*data10);
7153	} else {
7154		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7155		format = ccb12->format;
7156		alloc_len = scsi_4btoul(ccb12->alloc_length);
7157		data_len = sizeof(*data12);
7158	}
7159	if (alloc_len == 0) {
7160		ctl_set_success(ctsio);
7161		ctl_done((union ctl_io *)ctsio);
7162		return (CTL_RETVAL_COMPLETE);
7163	}
7164
7165	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7166	if (data_len < alloc_len) {
7167		ctsio->residual = alloc_len - data_len;
7168		ctsio->kern_data_len = data_len;
7169		ctsio->kern_total_len = data_len;
7170	} else {
7171		ctsio->residual = 0;
7172		ctsio->kern_data_len = alloc_len;
7173		ctsio->kern_total_len = alloc_len;
7174	}
7175	ctsio->kern_data_resid = 0;
7176	ctsio->kern_rel_offset = 0;
7177	ctsio->kern_sg_entries = 0;
7178
7179	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7180		data10 = (struct scsi_read_defect_data_hdr_10 *)
7181		    ctsio->kern_data_ptr;
7182		data10->format = format;
7183		scsi_ulto2b(0, data10->length);
7184	} else {
7185		data12 = (struct scsi_read_defect_data_hdr_12 *)
7186		    ctsio->kern_data_ptr;
7187		data12->format = format;
7188		scsi_ulto2b(0, data12->generation);
7189		scsi_ulto4b(0, data12->length);
7190	}
7191
7192	ctl_set_success(ctsio);
7193	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7194	ctsio->be_move_done = ctl_config_move_done;
7195	ctl_datamove((union ctl_io *)ctsio);
7196	return (CTL_RETVAL_COMPLETE);
7197}
7198
7199int
7200ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7201{
7202	struct scsi_maintenance_in *cdb;
7203	int retval;
7204	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
7205	int num_target_port_groups, num_target_ports;
7206	struct ctl_lun *lun;
7207	struct ctl_softc *softc;
7208	struct ctl_port *port;
7209	struct scsi_target_group_data *rtg_ptr;
7210	struct scsi_target_group_data_extended *rtg_ext_ptr;
7211	struct scsi_target_port_group_descriptor *tpg_desc;
7212
7213	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7214
7215	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7216	softc = control_softc;
7217	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7218
7219	retval = CTL_RETVAL_COMPLETE;
7220
7221	switch (cdb->byte2 & STG_PDF_MASK) {
7222	case STG_PDF_LENGTH:
7223		ext = 0;
7224		break;
7225	case STG_PDF_EXTENDED:
7226		ext = 1;
7227		break;
7228	default:
7229		ctl_set_invalid_field(/*ctsio*/ ctsio,
7230				      /*sks_valid*/ 1,
7231				      /*command*/ 1,
7232				      /*field*/ 2,
7233				      /*bit_valid*/ 1,
7234				      /*bit*/ 5);
7235		ctl_done((union ctl_io *)ctsio);
7236		return(retval);
7237	}
7238
7239	if (softc->is_single)
7240		num_target_port_groups = 1;
7241	else
7242		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7243	num_target_ports = 0;
7244	mtx_lock(&softc->ctl_lock);
7245	STAILQ_FOREACH(port, &softc->port_list, links) {
7246		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7247			continue;
7248		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7249			continue;
7250		num_target_ports++;
7251	}
7252	mtx_unlock(&softc->ctl_lock);
7253
7254	if (ext)
7255		total_len = sizeof(struct scsi_target_group_data_extended);
7256	else
7257		total_len = sizeof(struct scsi_target_group_data);
7258	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7259		num_target_port_groups +
7260	    sizeof(struct scsi_target_port_descriptor) *
7261		num_target_ports * num_target_port_groups;
7262
7263	alloc_len = scsi_4btoul(cdb->length);
7264
7265	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7266
7267	ctsio->kern_sg_entries = 0;
7268
7269	if (total_len < alloc_len) {
7270		ctsio->residual = alloc_len - total_len;
7271		ctsio->kern_data_len = total_len;
7272		ctsio->kern_total_len = total_len;
7273	} else {
7274		ctsio->residual = 0;
7275		ctsio->kern_data_len = alloc_len;
7276		ctsio->kern_total_len = alloc_len;
7277	}
7278	ctsio->kern_data_resid = 0;
7279	ctsio->kern_rel_offset = 0;
7280
7281	if (ext) {
7282		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7283		    ctsio->kern_data_ptr;
7284		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7285		rtg_ext_ptr->format_type = 0x10;
7286		rtg_ext_ptr->implicit_transition_time = 0;
7287		tpg_desc = &rtg_ext_ptr->groups[0];
7288	} else {
7289		rtg_ptr = (struct scsi_target_group_data *)
7290		    ctsio->kern_data_ptr;
7291		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7292		tpg_desc = &rtg_ptr->groups[0];
7293	}
7294
7295	mtx_lock(&softc->ctl_lock);
7296	pg = softc->port_offset / CTL_MAX_PORTS;
7297	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
7298		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
7299			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7300			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7301		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
7302			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7303			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7304		} else {
7305			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7306			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7307		}
7308	} else {
7309		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7310		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7311	}
7312	for (g = 0; g < num_target_port_groups; g++) {
7313		tpg_desc->pref_state = (g == pg) ? gs : os;
7314		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
7315		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7316		tpg_desc->status = TPG_IMPLICIT;
7317		pc = 0;
7318		STAILQ_FOREACH(port, &softc->port_list, links) {
7319			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7320				continue;
7321			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7322			    CTL_MAX_LUNS)
7323				continue;
7324			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7325			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7326			    relative_target_port_identifier);
7327			pc++;
7328		}
7329		tpg_desc->target_port_count = pc;
7330		tpg_desc = (struct scsi_target_port_group_descriptor *)
7331		    &tpg_desc->descriptors[pc];
7332	}
7333	mtx_unlock(&softc->ctl_lock);
7334
7335	ctl_set_success(ctsio);
7336	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7337	ctsio->be_move_done = ctl_config_move_done;
7338	ctl_datamove((union ctl_io *)ctsio);
7339	return(retval);
7340}
7341
7342int
7343ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7344{
7345	struct ctl_lun *lun;
7346	struct scsi_report_supported_opcodes *cdb;
7347	const struct ctl_cmd_entry *entry, *sentry;
7348	struct scsi_report_supported_opcodes_all *all;
7349	struct scsi_report_supported_opcodes_descr *descr;
7350	struct scsi_report_supported_opcodes_one *one;
7351	int retval;
7352	int alloc_len, total_len;
7353	int opcode, service_action, i, j, num;
7354
7355	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7356
7357	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7358	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7359
7360	retval = CTL_RETVAL_COMPLETE;
7361
7362	opcode = cdb->requested_opcode;
7363	service_action = scsi_2btoul(cdb->requested_service_action);
7364	switch (cdb->options & RSO_OPTIONS_MASK) {
7365	case RSO_OPTIONS_ALL:
7366		num = 0;
7367		for (i = 0; i < 256; i++) {
7368			entry = &ctl_cmd_table[i];
7369			if (entry->flags & CTL_CMD_FLAG_SA5) {
7370				for (j = 0; j < 32; j++) {
7371					sentry = &((const struct ctl_cmd_entry *)
7372					    entry->execute)[j];
7373					if (ctl_cmd_applicable(
7374					    lun->be_lun->lun_type, sentry))
7375						num++;
7376				}
7377			} else {
7378				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7379				    entry))
7380					num++;
7381			}
7382		}
7383		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7384		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7385		break;
7386	case RSO_OPTIONS_OC:
7387		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7388			ctl_set_invalid_field(/*ctsio*/ ctsio,
7389					      /*sks_valid*/ 1,
7390					      /*command*/ 1,
7391					      /*field*/ 2,
7392					      /*bit_valid*/ 1,
7393					      /*bit*/ 2);
7394			ctl_done((union ctl_io *)ctsio);
7395			return (CTL_RETVAL_COMPLETE);
7396		}
7397		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7398		break;
7399	case RSO_OPTIONS_OC_SA:
7400		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7401		    service_action >= 32) {
7402			ctl_set_invalid_field(/*ctsio*/ ctsio,
7403					      /*sks_valid*/ 1,
7404					      /*command*/ 1,
7405					      /*field*/ 2,
7406					      /*bit_valid*/ 1,
7407					      /*bit*/ 2);
7408			ctl_done((union ctl_io *)ctsio);
7409			return (CTL_RETVAL_COMPLETE);
7410		}
7411		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7412		break;
7413	default:
7414		ctl_set_invalid_field(/*ctsio*/ ctsio,
7415				      /*sks_valid*/ 1,
7416				      /*command*/ 1,
7417				      /*field*/ 2,
7418				      /*bit_valid*/ 1,
7419				      /*bit*/ 2);
7420		ctl_done((union ctl_io *)ctsio);
7421		return (CTL_RETVAL_COMPLETE);
7422	}
7423
7424	alloc_len = scsi_4btoul(cdb->length);
7425
7426	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7427
7428	ctsio->kern_sg_entries = 0;
7429
7430	if (total_len < alloc_len) {
7431		ctsio->residual = alloc_len - total_len;
7432		ctsio->kern_data_len = total_len;
7433		ctsio->kern_total_len = total_len;
7434	} else {
7435		ctsio->residual = 0;
7436		ctsio->kern_data_len = alloc_len;
7437		ctsio->kern_total_len = alloc_len;
7438	}
7439	ctsio->kern_data_resid = 0;
7440	ctsio->kern_rel_offset = 0;
7441
7442	switch (cdb->options & RSO_OPTIONS_MASK) {
7443	case RSO_OPTIONS_ALL:
7444		all = (struct scsi_report_supported_opcodes_all *)
7445		    ctsio->kern_data_ptr;
7446		num = 0;
7447		for (i = 0; i < 256; i++) {
7448			entry = &ctl_cmd_table[i];
7449			if (entry->flags & CTL_CMD_FLAG_SA5) {
7450				for (j = 0; j < 32; j++) {
7451					sentry = &((const struct ctl_cmd_entry *)
7452					    entry->execute)[j];
7453					if (!ctl_cmd_applicable(
7454					    lun->be_lun->lun_type, sentry))
7455						continue;
7456					descr = &all->descr[num++];
7457					descr->opcode = i;
7458					scsi_ulto2b(j, descr->service_action);
7459					descr->flags = RSO_SERVACTV;
7460					scsi_ulto2b(sentry->length,
7461					    descr->cdb_length);
7462				}
7463			} else {
7464				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7465				    entry))
7466					continue;
7467				descr = &all->descr[num++];
7468				descr->opcode = i;
7469				scsi_ulto2b(0, descr->service_action);
7470				descr->flags = 0;
7471				scsi_ulto2b(entry->length, descr->cdb_length);
7472			}
7473		}
7474		scsi_ulto4b(
7475		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7476		    all->length);
7477		break;
7478	case RSO_OPTIONS_OC:
7479		one = (struct scsi_report_supported_opcodes_one *)
7480		    ctsio->kern_data_ptr;
7481		entry = &ctl_cmd_table[opcode];
7482		goto fill_one;
7483	case RSO_OPTIONS_OC_SA:
7484		one = (struct scsi_report_supported_opcodes_one *)
7485		    ctsio->kern_data_ptr;
7486		entry = &ctl_cmd_table[opcode];
7487		entry = &((const struct ctl_cmd_entry *)
7488		    entry->execute)[service_action];
7489fill_one:
7490		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7491			one->support = 3;
7492			scsi_ulto2b(entry->length, one->cdb_length);
7493			one->cdb_usage[0] = opcode;
7494			memcpy(&one->cdb_usage[1], entry->usage,
7495			    entry->length - 1);
7496		} else
7497			one->support = 1;
7498		break;
7499	}
7500
7501	ctl_set_success(ctsio);
7502	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7503	ctsio->be_move_done = ctl_config_move_done;
7504	ctl_datamove((union ctl_io *)ctsio);
7505	return(retval);
7506}
7507
7508int
7509ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7510{
7511	struct scsi_report_supported_tmf *cdb;
7512	struct scsi_report_supported_tmf_data *data;
7513	int retval;
7514	int alloc_len, total_len;
7515
7516	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7517
7518	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7519
7520	retval = CTL_RETVAL_COMPLETE;
7521
7522	total_len = sizeof(struct scsi_report_supported_tmf_data);
7523	alloc_len = scsi_4btoul(cdb->length);
7524
7525	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7526
7527	ctsio->kern_sg_entries = 0;
7528
7529	if (total_len < alloc_len) {
7530		ctsio->residual = alloc_len - total_len;
7531		ctsio->kern_data_len = total_len;
7532		ctsio->kern_total_len = total_len;
7533	} else {
7534		ctsio->residual = 0;
7535		ctsio->kern_data_len = alloc_len;
7536		ctsio->kern_total_len = alloc_len;
7537	}
7538	ctsio->kern_data_resid = 0;
7539	ctsio->kern_rel_offset = 0;
7540
7541	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7542	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7543	data->byte2 |= RST_ITNRS;
7544
7545	ctl_set_success(ctsio);
7546	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7547	ctsio->be_move_done = ctl_config_move_done;
7548	ctl_datamove((union ctl_io *)ctsio);
7549	return (retval);
7550}
7551
7552int
7553ctl_report_timestamp(struct ctl_scsiio *ctsio)
7554{
7555	struct scsi_report_timestamp *cdb;
7556	struct scsi_report_timestamp_data *data;
7557	struct timeval tv;
7558	int64_t timestamp;
7559	int retval;
7560	int alloc_len, total_len;
7561
7562	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7563
7564	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7565
7566	retval = CTL_RETVAL_COMPLETE;
7567
7568	total_len = sizeof(struct scsi_report_timestamp_data);
7569	alloc_len = scsi_4btoul(cdb->length);
7570
7571	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7572
7573	ctsio->kern_sg_entries = 0;
7574
7575	if (total_len < alloc_len) {
7576		ctsio->residual = alloc_len - total_len;
7577		ctsio->kern_data_len = total_len;
7578		ctsio->kern_total_len = total_len;
7579	} else {
7580		ctsio->residual = 0;
7581		ctsio->kern_data_len = alloc_len;
7582		ctsio->kern_total_len = alloc_len;
7583	}
7584	ctsio->kern_data_resid = 0;
7585	ctsio->kern_rel_offset = 0;
7586
7587	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7588	scsi_ulto2b(sizeof(*data) - 2, data->length);
7589	data->origin = RTS_ORIG_OUTSIDE;
7590	getmicrotime(&tv);
7591	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7592	scsi_ulto4b(timestamp >> 16, data->timestamp);
7593	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7594
7595	ctl_set_success(ctsio);
7596	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7597	ctsio->be_move_done = ctl_config_move_done;
7598	ctl_datamove((union ctl_io *)ctsio);
7599	return (retval);
7600}
7601
7602int
7603ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7604{
7605	struct scsi_per_res_in *cdb;
7606	int alloc_len, total_len = 0;
7607	/* struct scsi_per_res_in_rsrv in_data; */
7608	struct ctl_lun *lun;
7609	struct ctl_softc *softc;
7610
7611	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7612
7613	softc = control_softc;
7614
7615	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7616
7617	alloc_len = scsi_2btoul(cdb->length);
7618
7619	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7620
7621retry:
7622	mtx_lock(&lun->lun_lock);
7623	switch (cdb->action) {
7624	case SPRI_RK: /* read keys */
7625		total_len = sizeof(struct scsi_per_res_in_keys) +
7626			lun->pr_key_count *
7627			sizeof(struct scsi_per_res_key);
7628		break;
7629	case SPRI_RR: /* read reservation */
7630		if (lun->flags & CTL_LUN_PR_RESERVED)
7631			total_len = sizeof(struct scsi_per_res_in_rsrv);
7632		else
7633			total_len = sizeof(struct scsi_per_res_in_header);
7634		break;
7635	case SPRI_RC: /* report capabilities */
7636		total_len = sizeof(struct scsi_per_res_cap);
7637		break;
7638	case SPRI_RS: /* read full status */
7639		total_len = sizeof(struct scsi_per_res_in_header) +
7640		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7641		    lun->pr_key_count;
7642		break;
7643	default:
7644		panic("Invalid PR type %x", cdb->action);
7645	}
7646	mtx_unlock(&lun->lun_lock);
7647
7648	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7649
7650	if (total_len < alloc_len) {
7651		ctsio->residual = alloc_len - total_len;
7652		ctsio->kern_data_len = total_len;
7653		ctsio->kern_total_len = total_len;
7654	} else {
7655		ctsio->residual = 0;
7656		ctsio->kern_data_len = alloc_len;
7657		ctsio->kern_total_len = alloc_len;
7658	}
7659
7660	ctsio->kern_data_resid = 0;
7661	ctsio->kern_rel_offset = 0;
7662	ctsio->kern_sg_entries = 0;
7663
7664	mtx_lock(&lun->lun_lock);
7665	switch (cdb->action) {
7666	case SPRI_RK: { // read keys
7667        struct scsi_per_res_in_keys *res_keys;
7668		int i, key_count;
7669
7670		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7671
7672		/*
7673		 * We had to drop the lock to allocate our buffer, which
7674		 * leaves time for someone to come in with another
7675		 * persistent reservation.  (That is unlikely, though,
7676		 * since this should be the only persistent reservation
7677		 * command active right now.)
7678		 */
7679		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7680		    (lun->pr_key_count *
7681		     sizeof(struct scsi_per_res_key)))){
7682			mtx_unlock(&lun->lun_lock);
7683			free(ctsio->kern_data_ptr, M_CTL);
7684			printf("%s: reservation length changed, retrying\n",
7685			       __func__);
7686			goto retry;
7687		}
7688
7689		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7690
7691		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7692			     lun->pr_key_count, res_keys->header.length);
7693
7694		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7695			if (lun->pr_keys[i] == 0)
7696				continue;
7697
7698			/*
7699			 * We used lun->pr_key_count to calculate the
7700			 * size to allocate.  If it turns out the number of
7701			 * initiators with the registered flag set is
7702			 * larger than that (i.e. they haven't been kept in
7703			 * sync), we've got a problem.
7704			 */
7705			if (key_count >= lun->pr_key_count) {
7706#ifdef NEEDTOPORT
7707				csevent_log(CSC_CTL | CSC_SHELF_SW |
7708					    CTL_PR_ERROR,
7709					    csevent_LogType_Fault,
7710					    csevent_AlertLevel_Yellow,
7711					    csevent_FRU_ShelfController,
7712					    csevent_FRU_Firmware,
7713				        csevent_FRU_Unknown,
7714					    "registered keys %d >= key "
7715					    "count %d", key_count,
7716					    lun->pr_key_count);
7717#endif
7718				key_count++;
7719				continue;
7720			}
7721			scsi_u64to8b(lun->pr_keys[i],
7722			    res_keys->keys[key_count].key);
7723			key_count++;
7724		}
7725		break;
7726	}
7727	case SPRI_RR: { // read reservation
7728		struct scsi_per_res_in_rsrv *res;
7729		int tmp_len, header_only;
7730
7731		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7732
7733		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7734
7735		if (lun->flags & CTL_LUN_PR_RESERVED)
7736		{
7737			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7738			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7739				    res->header.length);
7740			header_only = 0;
7741		} else {
7742			tmp_len = sizeof(struct scsi_per_res_in_header);
7743			scsi_ulto4b(0, res->header.length);
7744			header_only = 1;
7745		}
7746
7747		/*
7748		 * We had to drop the lock to allocate our buffer, which
7749		 * leaves time for someone to come in with another
7750		 * persistent reservation.  (That is unlikely, though,
7751		 * since this should be the only persistent reservation
7752		 * command active right now.)
7753		 */
7754		if (tmp_len != total_len) {
7755			mtx_unlock(&lun->lun_lock);
7756			free(ctsio->kern_data_ptr, M_CTL);
7757			printf("%s: reservation status changed, retrying\n",
7758			       __func__);
7759			goto retry;
7760		}
7761
7762		/*
7763		 * No reservation held, so we're done.
7764		 */
7765		if (header_only != 0)
7766			break;
7767
7768		/*
7769		 * If the registration is an All Registrants type, the key
7770		 * is 0, since it doesn't really matter.
7771		 */
7772		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7773			scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7774			    res->data.reservation);
7775		}
7776		res->data.scopetype = lun->res_type;
7777		break;
7778	}
7779	case SPRI_RC:     //report capabilities
7780	{
7781		struct scsi_per_res_cap *res_cap;
7782		uint16_t type_mask;
7783
7784		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7785		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7786		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7787		type_mask = SPRI_TM_WR_EX_AR |
7788			    SPRI_TM_EX_AC_RO |
7789			    SPRI_TM_WR_EX_RO |
7790			    SPRI_TM_EX_AC |
7791			    SPRI_TM_WR_EX |
7792			    SPRI_TM_EX_AC_AR;
7793		scsi_ulto2b(type_mask, res_cap->type_mask);
7794		break;
7795	}
7796	case SPRI_RS: { // read full status
7797		struct scsi_per_res_in_full *res_status;
7798		struct scsi_per_res_in_full_desc *res_desc;
7799		struct ctl_port *port;
7800		int i, len;
7801
7802		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7803
7804		/*
7805		 * We had to drop the lock to allocate our buffer, which
7806		 * leaves time for someone to come in with another
7807		 * persistent reservation.  (That is unlikely, though,
7808		 * since this should be the only persistent reservation
7809		 * command active right now.)
7810		 */
7811		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7812		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7813		     lun->pr_key_count)){
7814			mtx_unlock(&lun->lun_lock);
7815			free(ctsio->kern_data_ptr, M_CTL);
7816			printf("%s: reservation length changed, retrying\n",
7817			       __func__);
7818			goto retry;
7819		}
7820
7821		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7822
7823		res_desc = &res_status->desc[0];
7824		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7825			if (lun->pr_keys[i] == 0)
7826				continue;
7827
7828			scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
7829			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7830			    (lun->pr_res_idx == i ||
7831			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7832				res_desc->flags = SPRI_FULL_R_HOLDER;
7833				res_desc->scopetype = lun->res_type;
7834			}
7835			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7836			    res_desc->rel_trgt_port_id);
7837			len = 0;
7838			port = softc->ctl_ports[
7839			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7840			if (port != NULL)
7841				len = ctl_create_iid(port,
7842				    i % CTL_MAX_INIT_PER_PORT,
7843				    res_desc->transport_id);
7844			scsi_ulto4b(len, res_desc->additional_length);
7845			res_desc = (struct scsi_per_res_in_full_desc *)
7846			    &res_desc->transport_id[len];
7847		}
7848		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7849		    res_status->header.length);
7850		break;
7851	}
7852	default:
7853		/*
7854		 * This is a bug, because we just checked for this above,
7855		 * and should have returned an error.
7856		 */
7857		panic("Invalid PR type %x", cdb->action);
7858		break; /* NOTREACHED */
7859	}
7860	mtx_unlock(&lun->lun_lock);
7861
7862	ctl_set_success(ctsio);
7863	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7864	ctsio->be_move_done = ctl_config_move_done;
7865	ctl_datamove((union ctl_io *)ctsio);
7866	return (CTL_RETVAL_COMPLETE);
7867}
7868
7869static void
7870ctl_set_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
7871{
7872	int off = lun->ctl_softc->persis_offset;
7873
7874	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
7875		lun->pending_ua[residx - off] |= ua;
7876}
7877
7878/*
7879 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7880 * it should return.
7881 */
7882static int
7883ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7884		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7885		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7886		struct scsi_per_res_out_parms* param)
7887{
7888	union ctl_ha_msg persis_io;
7889	int retval, i;
7890	int isc_retval;
7891
7892	retval = 0;
7893
7894	mtx_lock(&lun->lun_lock);
7895	if (sa_res_key == 0) {
7896		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7897			/* validate scope and type */
7898			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7899			     SPR_LU_SCOPE) {
7900				mtx_unlock(&lun->lun_lock);
7901				ctl_set_invalid_field(/*ctsio*/ ctsio,
7902						      /*sks_valid*/ 1,
7903						      /*command*/ 1,
7904						      /*field*/ 2,
7905						      /*bit_valid*/ 1,
7906						      /*bit*/ 4);
7907				ctl_done((union ctl_io *)ctsio);
7908				return (1);
7909			}
7910
7911		        if (type>8 || type==2 || type==4 || type==0) {
7912				mtx_unlock(&lun->lun_lock);
7913				ctl_set_invalid_field(/*ctsio*/ ctsio,
7914       	           				      /*sks_valid*/ 1,
7915						      /*command*/ 1,
7916						      /*field*/ 2,
7917						      /*bit_valid*/ 1,
7918						      /*bit*/ 0);
7919				ctl_done((union ctl_io *)ctsio);
7920				return (1);
7921		        }
7922
7923			/*
7924			 * Unregister everybody else and build UA for
7925			 * them
7926			 */
7927			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7928				if (i == residx || lun->pr_keys[i] == 0)
7929					continue;
7930
7931				lun->pr_keys[i] = 0;
7932				ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
7933			}
7934			lun->pr_key_count = 1;
7935			lun->res_type = type;
7936			if (lun->res_type != SPR_TYPE_WR_EX_AR
7937			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7938				lun->pr_res_idx = residx;
7939
7940			/* send msg to other side */
7941			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7942			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7943			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7944			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7945			persis_io.pr.pr_info.res_type = type;
7946			memcpy(persis_io.pr.pr_info.sa_res_key,
7947			       param->serv_act_res_key,
7948			       sizeof(param->serv_act_res_key));
7949			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7950			     &persis_io, sizeof(persis_io), 0)) >
7951			     CTL_HA_STATUS_SUCCESS) {
7952				printf("CTL:Persis Out error returned "
7953				       "from ctl_ha_msg_send %d\n",
7954				       isc_retval);
7955			}
7956		} else {
7957			/* not all registrants */
7958			mtx_unlock(&lun->lun_lock);
7959			free(ctsio->kern_data_ptr, M_CTL);
7960			ctl_set_invalid_field(ctsio,
7961					      /*sks_valid*/ 1,
7962					      /*command*/ 0,
7963					      /*field*/ 8,
7964					      /*bit_valid*/ 0,
7965					      /*bit*/ 0);
7966			ctl_done((union ctl_io *)ctsio);
7967			return (1);
7968		}
7969	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7970		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7971		int found = 0;
7972
7973		if (res_key == sa_res_key) {
7974			/* special case */
7975			/*
7976			 * The spec implies this is not good but doesn't
7977			 * say what to do. There are two choices either
7978			 * generate a res conflict or check condition
7979			 * with illegal field in parameter data. Since
7980			 * that is what is done when the sa_res_key is
7981			 * zero I'll take that approach since this has
7982			 * to do with the sa_res_key.
7983			 */
7984			mtx_unlock(&lun->lun_lock);
7985			free(ctsio->kern_data_ptr, M_CTL);
7986			ctl_set_invalid_field(ctsio,
7987					      /*sks_valid*/ 1,
7988					      /*command*/ 0,
7989					      /*field*/ 8,
7990					      /*bit_valid*/ 0,
7991					      /*bit*/ 0);
7992			ctl_done((union ctl_io *)ctsio);
7993			return (1);
7994		}
7995
7996		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7997			if (lun->pr_keys[i] != sa_res_key)
7998				continue;
7999
8000			found = 1;
8001			lun->pr_keys[i] = 0;
8002			lun->pr_key_count--;
8003			ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8004		}
8005		if (!found) {
8006			mtx_unlock(&lun->lun_lock);
8007			free(ctsio->kern_data_ptr, M_CTL);
8008			ctl_set_reservation_conflict(ctsio);
8009			ctl_done((union ctl_io *)ctsio);
8010			return (CTL_RETVAL_COMPLETE);
8011		}
8012		/* send msg to other side */
8013		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8014		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8015		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8016		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8017		persis_io.pr.pr_info.res_type = type;
8018		memcpy(persis_io.pr.pr_info.sa_res_key,
8019		       param->serv_act_res_key,
8020		       sizeof(param->serv_act_res_key));
8021		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8022		     &persis_io, sizeof(persis_io), 0)) >
8023		     CTL_HA_STATUS_SUCCESS) {
8024			printf("CTL:Persis Out error returned from "
8025			       "ctl_ha_msg_send %d\n", isc_retval);
8026		}
8027	} else {
8028		/* Reserved but not all registrants */
8029		/* sa_res_key is res holder */
8030		if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8031			/* validate scope and type */
8032			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8033			     SPR_LU_SCOPE) {
8034				mtx_unlock(&lun->lun_lock);
8035				ctl_set_invalid_field(/*ctsio*/ ctsio,
8036						      /*sks_valid*/ 1,
8037						      /*command*/ 1,
8038						      /*field*/ 2,
8039						      /*bit_valid*/ 1,
8040						      /*bit*/ 4);
8041				ctl_done((union ctl_io *)ctsio);
8042				return (1);
8043			}
8044
8045			if (type>8 || type==2 || type==4 || type==0) {
8046				mtx_unlock(&lun->lun_lock);
8047				ctl_set_invalid_field(/*ctsio*/ ctsio,
8048						      /*sks_valid*/ 1,
8049						      /*command*/ 1,
8050						      /*field*/ 2,
8051						      /*bit_valid*/ 1,
8052						      /*bit*/ 0);
8053				ctl_done((union ctl_io *)ctsio);
8054				return (1);
8055			}
8056
8057			/*
8058			 * Do the following:
8059			 * if sa_res_key != res_key remove all
8060			 * registrants w/sa_res_key and generate UA
8061			 * for these registrants(Registrations
8062			 * Preempted) if it wasn't an exclusive
8063			 * reservation generate UA(Reservations
8064			 * Preempted) for all other registered nexuses
8065			 * if the type has changed. Establish the new
8066			 * reservation and holder. If res_key and
8067			 * sa_res_key are the same do the above
8068			 * except don't unregister the res holder.
8069			 */
8070
8071			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8072				if (i == residx || lun->pr_keys[i] == 0)
8073					continue;
8074
8075				if (sa_res_key == lun->pr_keys[i]) {
8076					lun->pr_keys[i] = 0;
8077					lun->pr_key_count--;
8078					ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8079				} else if (type != lun->res_type
8080					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8081					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8082					ctl_set_res_ua(lun, i, CTL_UA_RES_RELEASE);
8083				}
8084			}
8085			lun->res_type = type;
8086			if (lun->res_type != SPR_TYPE_WR_EX_AR
8087			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8088				lun->pr_res_idx = residx;
8089			else
8090				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8091
8092			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8093			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8094			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8095			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8096			persis_io.pr.pr_info.res_type = type;
8097			memcpy(persis_io.pr.pr_info.sa_res_key,
8098			       param->serv_act_res_key,
8099			       sizeof(param->serv_act_res_key));
8100			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8101			     &persis_io, sizeof(persis_io), 0)) >
8102			     CTL_HA_STATUS_SUCCESS) {
8103				printf("CTL:Persis Out error returned "
8104				       "from ctl_ha_msg_send %d\n",
8105				       isc_retval);
8106			}
8107		} else {
8108			/*
8109			 * sa_res_key is not the res holder just
8110			 * remove registrants
8111			 */
8112			int found=0;
8113
8114			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8115				if (sa_res_key != lun->pr_keys[i])
8116					continue;
8117
8118				found = 1;
8119				lun->pr_keys[i] = 0;
8120				lun->pr_key_count--;
8121				ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8122			}
8123
8124			if (!found) {
8125				mtx_unlock(&lun->lun_lock);
8126				free(ctsio->kern_data_ptr, M_CTL);
8127				ctl_set_reservation_conflict(ctsio);
8128				ctl_done((union ctl_io *)ctsio);
8129		        	return (1);
8130			}
8131			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8132			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8133			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8134			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8135			persis_io.pr.pr_info.res_type = type;
8136			memcpy(persis_io.pr.pr_info.sa_res_key,
8137			       param->serv_act_res_key,
8138			       sizeof(param->serv_act_res_key));
8139			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8140			     &persis_io, sizeof(persis_io), 0)) >
8141			     CTL_HA_STATUS_SUCCESS) {
8142				printf("CTL:Persis Out error returned "
8143				       "from ctl_ha_msg_send %d\n",
8144				isc_retval);
8145			}
8146		}
8147	}
8148
8149	lun->PRGeneration++;
8150	mtx_unlock(&lun->lun_lock);
8151
8152	return (retval);
8153}
8154
8155static void
8156ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8157{
8158	uint64_t sa_res_key;
8159	int i;
8160
8161	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8162
8163	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8164	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8165	 || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8166		if (sa_res_key == 0) {
8167			/*
8168			 * Unregister everybody else and build UA for
8169			 * them
8170			 */
8171			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8172				if (i == msg->pr.pr_info.residx ||
8173				    lun->pr_keys[i] == 0)
8174					continue;
8175
8176				lun->pr_keys[i] = 0;
8177				ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8178			}
8179
8180			lun->pr_key_count = 1;
8181			lun->res_type = msg->pr.pr_info.res_type;
8182			if (lun->res_type != SPR_TYPE_WR_EX_AR
8183			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8184				lun->pr_res_idx = msg->pr.pr_info.residx;
8185		} else {
8186		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8187				if (sa_res_key == lun->pr_keys[i])
8188					continue;
8189
8190				lun->pr_keys[i] = 0;
8191				lun->pr_key_count--;
8192				ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8193			}
8194		}
8195	} else {
8196		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8197			if (i == msg->pr.pr_info.residx ||
8198			    lun->pr_keys[i] == 0)
8199				continue;
8200
8201			if (sa_res_key == lun->pr_keys[i]) {
8202				lun->pr_keys[i] = 0;
8203				lun->pr_key_count--;
8204				ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8205			} else if (msg->pr.pr_info.res_type != lun->res_type
8206				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8207				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8208				ctl_set_res_ua(lun, i, CTL_UA_RES_RELEASE);
8209			}
8210		}
8211		lun->res_type = msg->pr.pr_info.res_type;
8212		if (lun->res_type != SPR_TYPE_WR_EX_AR
8213		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8214			lun->pr_res_idx = msg->pr.pr_info.residx;
8215		else
8216			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8217	}
8218	lun->PRGeneration++;
8219
8220}
8221
8222
8223int
8224ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8225{
8226	int retval;
8227	int isc_retval;
8228	u_int32_t param_len;
8229	struct scsi_per_res_out *cdb;
8230	struct ctl_lun *lun;
8231	struct scsi_per_res_out_parms* param;
8232	struct ctl_softc *softc;
8233	uint32_t residx;
8234	uint64_t res_key, sa_res_key;
8235	uint8_t type;
8236	union ctl_ha_msg persis_io;
8237	int    i;
8238
8239	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8240
8241	retval = CTL_RETVAL_COMPLETE;
8242
8243	softc = control_softc;
8244
8245	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8246	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8247
8248	/*
8249	 * We only support whole-LUN scope.  The scope & type are ignored for
8250	 * register, register and ignore existing key and clear.
8251	 * We sometimes ignore scope and type on preempts too!!
8252	 * Verify reservation type here as well.
8253	 */
8254	type = cdb->scope_type & SPR_TYPE_MASK;
8255	if ((cdb->action == SPRO_RESERVE)
8256	 || (cdb->action == SPRO_RELEASE)) {
8257		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8258			ctl_set_invalid_field(/*ctsio*/ ctsio,
8259					      /*sks_valid*/ 1,
8260					      /*command*/ 1,
8261					      /*field*/ 2,
8262					      /*bit_valid*/ 1,
8263					      /*bit*/ 4);
8264			ctl_done((union ctl_io *)ctsio);
8265			return (CTL_RETVAL_COMPLETE);
8266		}
8267
8268		if (type>8 || type==2 || type==4 || type==0) {
8269			ctl_set_invalid_field(/*ctsio*/ ctsio,
8270					      /*sks_valid*/ 1,
8271					      /*command*/ 1,
8272					      /*field*/ 2,
8273					      /*bit_valid*/ 1,
8274					      /*bit*/ 0);
8275			ctl_done((union ctl_io *)ctsio);
8276			return (CTL_RETVAL_COMPLETE);
8277		}
8278	}
8279
8280	param_len = scsi_4btoul(cdb->length);
8281
8282	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8283		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8284		ctsio->kern_data_len = param_len;
8285		ctsio->kern_total_len = param_len;
8286		ctsio->kern_data_resid = 0;
8287		ctsio->kern_rel_offset = 0;
8288		ctsio->kern_sg_entries = 0;
8289		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8290		ctsio->be_move_done = ctl_config_move_done;
8291		ctl_datamove((union ctl_io *)ctsio);
8292
8293		return (CTL_RETVAL_COMPLETE);
8294	}
8295
8296	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8297
8298	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8299	res_key = scsi_8btou64(param->res_key.key);
8300	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8301
8302	/*
8303	 * Validate the reservation key here except for SPRO_REG_IGNO
8304	 * This must be done for all other service actions
8305	 */
8306	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8307		mtx_lock(&lun->lun_lock);
8308		if (lun->pr_keys[residx] != 0) {
8309		    if (res_key != lun->pr_keys[residx]) {
8310				/*
8311				 * The current key passed in doesn't match
8312				 * the one the initiator previously
8313				 * registered.
8314				 */
8315				mtx_unlock(&lun->lun_lock);
8316				free(ctsio->kern_data_ptr, M_CTL);
8317				ctl_set_reservation_conflict(ctsio);
8318				ctl_done((union ctl_io *)ctsio);
8319				return (CTL_RETVAL_COMPLETE);
8320			}
8321		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8322			/*
8323			 * We are not registered
8324			 */
8325			mtx_unlock(&lun->lun_lock);
8326			free(ctsio->kern_data_ptr, M_CTL);
8327			ctl_set_reservation_conflict(ctsio);
8328			ctl_done((union ctl_io *)ctsio);
8329			return (CTL_RETVAL_COMPLETE);
8330		} else if (res_key != 0) {
8331			/*
8332			 * We are not registered and trying to register but
8333			 * the register key isn't zero.
8334			 */
8335			mtx_unlock(&lun->lun_lock);
8336			free(ctsio->kern_data_ptr, M_CTL);
8337			ctl_set_reservation_conflict(ctsio);
8338			ctl_done((union ctl_io *)ctsio);
8339			return (CTL_RETVAL_COMPLETE);
8340		}
8341		mtx_unlock(&lun->lun_lock);
8342	}
8343
8344	switch (cdb->action & SPRO_ACTION_MASK) {
8345	case SPRO_REGISTER:
8346	case SPRO_REG_IGNO: {
8347
8348#if 0
8349		printf("Registration received\n");
8350#endif
8351
8352		/*
8353		 * We don't support any of these options, as we report in
8354		 * the read capabilities request (see
8355		 * ctl_persistent_reserve_in(), above).
8356		 */
8357		if ((param->flags & SPR_SPEC_I_PT)
8358		 || (param->flags & SPR_ALL_TG_PT)
8359		 || (param->flags & SPR_APTPL)) {
8360			int bit_ptr;
8361
8362			if (param->flags & SPR_APTPL)
8363				bit_ptr = 0;
8364			else if (param->flags & SPR_ALL_TG_PT)
8365				bit_ptr = 2;
8366			else /* SPR_SPEC_I_PT */
8367				bit_ptr = 3;
8368
8369			free(ctsio->kern_data_ptr, M_CTL);
8370			ctl_set_invalid_field(ctsio,
8371					      /*sks_valid*/ 1,
8372					      /*command*/ 0,
8373					      /*field*/ 20,
8374					      /*bit_valid*/ 1,
8375					      /*bit*/ bit_ptr);
8376			ctl_done((union ctl_io *)ctsio);
8377			return (CTL_RETVAL_COMPLETE);
8378		}
8379
8380		mtx_lock(&lun->lun_lock);
8381
8382		/*
8383		 * The initiator wants to clear the
8384		 * key/unregister.
8385		 */
8386		if (sa_res_key == 0) {
8387			if ((res_key == 0
8388			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8389			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8390			  && lun->pr_keys[residx] == 0)) {
8391				mtx_unlock(&lun->lun_lock);
8392				goto done;
8393			}
8394
8395			lun->pr_keys[residx] = 0;
8396			lun->pr_key_count--;
8397
8398			if (residx == lun->pr_res_idx) {
8399				lun->flags &= ~CTL_LUN_PR_RESERVED;
8400				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8401
8402				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8403				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8404				 && lun->pr_key_count) {
8405					/*
8406					 * If the reservation is a registrants
8407					 * only type we need to generate a UA
8408					 * for other registered inits.  The
8409					 * sense code should be RESERVATIONS
8410					 * RELEASED
8411					 */
8412
8413					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8414						if (lun->pr_keys[i +
8415						    softc->persis_offset] == 0)
8416							continue;
8417						lun->pending_ua[i] |=
8418							CTL_UA_RES_RELEASE;
8419					}
8420				}
8421				lun->res_type = 0;
8422			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8423				if (lun->pr_key_count==0) {
8424					lun->flags &= ~CTL_LUN_PR_RESERVED;
8425					lun->res_type = 0;
8426					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8427				}
8428			}
8429			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8430			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8431			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8432			persis_io.pr.pr_info.residx = residx;
8433			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8434			     &persis_io, sizeof(persis_io), 0 )) >
8435			     CTL_HA_STATUS_SUCCESS) {
8436				printf("CTL:Persis Out error returned from "
8437				       "ctl_ha_msg_send %d\n", isc_retval);
8438			}
8439		} else /* sa_res_key != 0 */ {
8440
8441			/*
8442			 * If we aren't registered currently then increment
8443			 * the key count and set the registered flag.
8444			 */
8445			if (lun->pr_keys[residx] == 0)
8446				lun->pr_key_count++;
8447			lun->pr_keys[residx] = sa_res_key;
8448
8449			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8450			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8451			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8452			persis_io.pr.pr_info.residx = residx;
8453			memcpy(persis_io.pr.pr_info.sa_res_key,
8454			       param->serv_act_res_key,
8455			       sizeof(param->serv_act_res_key));
8456			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8457			     &persis_io, sizeof(persis_io), 0)) >
8458			     CTL_HA_STATUS_SUCCESS) {
8459				printf("CTL:Persis Out error returned from "
8460				       "ctl_ha_msg_send %d\n", isc_retval);
8461			}
8462		}
8463		lun->PRGeneration++;
8464		mtx_unlock(&lun->lun_lock);
8465
8466		break;
8467	}
8468	case SPRO_RESERVE:
8469#if 0
8470                printf("Reserve executed type %d\n", type);
8471#endif
8472		mtx_lock(&lun->lun_lock);
8473		if (lun->flags & CTL_LUN_PR_RESERVED) {
8474			/*
8475			 * if this isn't the reservation holder and it's
8476			 * not a "all registrants" type or if the type is
8477			 * different then we have a conflict
8478			 */
8479			if ((lun->pr_res_idx != residx
8480			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8481			 || lun->res_type != type) {
8482				mtx_unlock(&lun->lun_lock);
8483				free(ctsio->kern_data_ptr, M_CTL);
8484				ctl_set_reservation_conflict(ctsio);
8485				ctl_done((union ctl_io *)ctsio);
8486				return (CTL_RETVAL_COMPLETE);
8487			}
8488			mtx_unlock(&lun->lun_lock);
8489		} else /* create a reservation */ {
8490			/*
8491			 * If it's not an "all registrants" type record
8492			 * reservation holder
8493			 */
8494			if (type != SPR_TYPE_WR_EX_AR
8495			 && type != SPR_TYPE_EX_AC_AR)
8496				lun->pr_res_idx = residx; /* Res holder */
8497			else
8498				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8499
8500			lun->flags |= CTL_LUN_PR_RESERVED;
8501			lun->res_type = type;
8502
8503			mtx_unlock(&lun->lun_lock);
8504
8505			/* send msg to other side */
8506			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8507			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8508			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8509			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8510			persis_io.pr.pr_info.res_type = type;
8511			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8512			     &persis_io, sizeof(persis_io), 0)) >
8513			     CTL_HA_STATUS_SUCCESS) {
8514				printf("CTL:Persis Out error returned from "
8515				       "ctl_ha_msg_send %d\n", isc_retval);
8516			}
8517		}
8518		break;
8519
8520	case SPRO_RELEASE:
8521		mtx_lock(&lun->lun_lock);
8522		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8523			/* No reservation exists return good status */
8524			mtx_unlock(&lun->lun_lock);
8525			goto done;
8526		}
8527		/*
8528		 * Is this nexus a reservation holder?
8529		 */
8530		if (lun->pr_res_idx != residx
8531		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8532			/*
8533			 * not a res holder return good status but
8534			 * do nothing
8535			 */
8536			mtx_unlock(&lun->lun_lock);
8537			goto done;
8538		}
8539
8540		if (lun->res_type != type) {
8541			mtx_unlock(&lun->lun_lock);
8542			free(ctsio->kern_data_ptr, M_CTL);
8543			ctl_set_illegal_pr_release(ctsio);
8544			ctl_done((union ctl_io *)ctsio);
8545			return (CTL_RETVAL_COMPLETE);
8546		}
8547
8548		/* okay to release */
8549		lun->flags &= ~CTL_LUN_PR_RESERVED;
8550		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8551		lun->res_type = 0;
8552
8553		/*
8554		 * if this isn't an exclusive access
8555		 * res generate UA for all other
8556		 * registrants.
8557		 */
8558		if (type != SPR_TYPE_EX_AC
8559		 && type != SPR_TYPE_WR_EX) {
8560			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8561				if (i == residx ||
8562				    lun->pr_keys[i + softc->persis_offset] == 0)
8563					continue;
8564				lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8565			}
8566		}
8567		mtx_unlock(&lun->lun_lock);
8568		/* Send msg to other side */
8569		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8570		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8571		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8572		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8573		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8574			printf("CTL:Persis Out error returned from "
8575			       "ctl_ha_msg_send %d\n", isc_retval);
8576		}
8577		break;
8578
8579	case SPRO_CLEAR:
8580		/* send msg to other side */
8581
8582		mtx_lock(&lun->lun_lock);
8583		lun->flags &= ~CTL_LUN_PR_RESERVED;
8584		lun->res_type = 0;
8585		lun->pr_key_count = 0;
8586		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8587
8588		lun->pr_keys[residx] = 0;
8589
8590		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8591			if (lun->pr_keys[i] != 0) {
8592				lun->pr_keys[i] = 0;
8593				ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8594			}
8595		lun->PRGeneration++;
8596		mtx_unlock(&lun->lun_lock);
8597		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8598		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8599		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8600		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8601		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8602			printf("CTL:Persis Out error returned from "
8603			       "ctl_ha_msg_send %d\n", isc_retval);
8604		}
8605		break;
8606
8607	case SPRO_PREEMPT:
8608	case SPRO_PRE_ABO: {
8609		int nretval;
8610
8611		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8612					  residx, ctsio, cdb, param);
8613		if (nretval != 0)
8614			return (CTL_RETVAL_COMPLETE);
8615		break;
8616	}
8617	default:
8618		panic("Invalid PR type %x", cdb->action);
8619	}
8620
8621done:
8622	free(ctsio->kern_data_ptr, M_CTL);
8623	ctl_set_success(ctsio);
8624	ctl_done((union ctl_io *)ctsio);
8625
8626	return (retval);
8627}
8628
8629/*
8630 * This routine is for handling a message from the other SC pertaining to
8631 * persistent reserve out. All the error checking will have been done
8632 * so only perorming the action need be done here to keep the two
8633 * in sync.
8634 */
8635static void
8636ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8637{
8638	struct ctl_lun *lun;
8639	struct ctl_softc *softc;
8640	int i;
8641	uint32_t targ_lun;
8642
8643	softc = control_softc;
8644
8645	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8646	lun = softc->ctl_luns[targ_lun];
8647	mtx_lock(&lun->lun_lock);
8648	switch(msg->pr.pr_info.action) {
8649	case CTL_PR_REG_KEY:
8650		if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8651			lun->pr_key_count++;
8652		lun->pr_keys[msg->pr.pr_info.residx] =
8653		    scsi_8btou64(msg->pr.pr_info.sa_res_key);
8654		lun->PRGeneration++;
8655		break;
8656
8657	case CTL_PR_UNREG_KEY:
8658		lun->pr_keys[msg->pr.pr_info.residx] = 0;
8659		lun->pr_key_count--;
8660
8661		/* XXX Need to see if the reservation has been released */
8662		/* if so do we need to generate UA? */
8663		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8664			lun->flags &= ~CTL_LUN_PR_RESERVED;
8665			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8666
8667			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8668			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8669			 && lun->pr_key_count) {
8670				/*
8671				 * If the reservation is a registrants
8672				 * only type we need to generate a UA
8673				 * for other registered inits.  The
8674				 * sense code should be RESERVATIONS
8675				 * RELEASED
8676				 */
8677
8678				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8679					if (lun->pr_keys[i +
8680					    softc->persis_offset] == 0)
8681						continue;
8682
8683					lun->pending_ua[i] |=
8684						CTL_UA_RES_RELEASE;
8685				}
8686			}
8687			lun->res_type = 0;
8688		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8689			if (lun->pr_key_count==0) {
8690				lun->flags &= ~CTL_LUN_PR_RESERVED;
8691				lun->res_type = 0;
8692				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8693			}
8694		}
8695		lun->PRGeneration++;
8696		break;
8697
8698	case CTL_PR_RESERVE:
8699		lun->flags |= CTL_LUN_PR_RESERVED;
8700		lun->res_type = msg->pr.pr_info.res_type;
8701		lun->pr_res_idx = msg->pr.pr_info.residx;
8702
8703		break;
8704
8705	case CTL_PR_RELEASE:
8706		/*
8707		 * if this isn't an exclusive access res generate UA for all
8708		 * other registrants.
8709		 */
8710		if (lun->res_type != SPR_TYPE_EX_AC
8711		 && lun->res_type != SPR_TYPE_WR_EX) {
8712			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8713				if (lun->pr_keys[i + softc->persis_offset] != 0)
8714					lun->pending_ua[i] |=
8715						CTL_UA_RES_RELEASE;
8716		}
8717
8718		lun->flags &= ~CTL_LUN_PR_RESERVED;
8719		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8720		lun->res_type = 0;
8721		break;
8722
8723	case CTL_PR_PREEMPT:
8724		ctl_pro_preempt_other(lun, msg);
8725		break;
8726	case CTL_PR_CLEAR:
8727		lun->flags &= ~CTL_LUN_PR_RESERVED;
8728		lun->res_type = 0;
8729		lun->pr_key_count = 0;
8730		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8731
8732		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8733			if (lun->pr_keys[i] == 0)
8734				continue;
8735			lun->pr_keys[i] = 0;
8736			ctl_set_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8737		}
8738		lun->PRGeneration++;
8739		break;
8740	}
8741
8742	mtx_unlock(&lun->lun_lock);
8743}
8744
8745int
8746ctl_read_write(struct ctl_scsiio *ctsio)
8747{
8748	struct ctl_lun *lun;
8749	struct ctl_lba_len_flags *lbalen;
8750	uint64_t lba;
8751	uint32_t num_blocks;
8752	int flags, retval;
8753	int isread;
8754
8755	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8756
8757	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8758
8759	flags = 0;
8760	retval = CTL_RETVAL_COMPLETE;
8761
8762	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8763	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8764	switch (ctsio->cdb[0]) {
8765	case READ_6:
8766	case WRITE_6: {
8767		struct scsi_rw_6 *cdb;
8768
8769		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8770
8771		lba = scsi_3btoul(cdb->addr);
8772		/* only 5 bits are valid in the most significant address byte */
8773		lba &= 0x1fffff;
8774		num_blocks = cdb->length;
8775		/*
8776		 * This is correct according to SBC-2.
8777		 */
8778		if (num_blocks == 0)
8779			num_blocks = 256;
8780		break;
8781	}
8782	case READ_10:
8783	case WRITE_10: {
8784		struct scsi_rw_10 *cdb;
8785
8786		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8787		if (cdb->byte2 & SRW10_FUA)
8788			flags |= CTL_LLF_FUA;
8789		if (cdb->byte2 & SRW10_DPO)
8790			flags |= CTL_LLF_DPO;
8791		lba = scsi_4btoul(cdb->addr);
8792		num_blocks = scsi_2btoul(cdb->length);
8793		break;
8794	}
8795	case WRITE_VERIFY_10: {
8796		struct scsi_write_verify_10 *cdb;
8797
8798		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8799		flags |= CTL_LLF_FUA;
8800		if (cdb->byte2 & SWV_DPO)
8801			flags |= CTL_LLF_DPO;
8802		lba = scsi_4btoul(cdb->addr);
8803		num_blocks = scsi_2btoul(cdb->length);
8804		break;
8805	}
8806	case READ_12:
8807	case WRITE_12: {
8808		struct scsi_rw_12 *cdb;
8809
8810		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8811		if (cdb->byte2 & SRW12_FUA)
8812			flags |= CTL_LLF_FUA;
8813		if (cdb->byte2 & SRW12_DPO)
8814			flags |= CTL_LLF_DPO;
8815		lba = scsi_4btoul(cdb->addr);
8816		num_blocks = scsi_4btoul(cdb->length);
8817		break;
8818	}
8819	case WRITE_VERIFY_12: {
8820		struct scsi_write_verify_12 *cdb;
8821
8822		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8823		flags |= CTL_LLF_FUA;
8824		if (cdb->byte2 & SWV_DPO)
8825			flags |= CTL_LLF_DPO;
8826		lba = scsi_4btoul(cdb->addr);
8827		num_blocks = scsi_4btoul(cdb->length);
8828		break;
8829	}
8830	case READ_16:
8831	case WRITE_16: {
8832		struct scsi_rw_16 *cdb;
8833
8834		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8835		if (cdb->byte2 & SRW12_FUA)
8836			flags |= CTL_LLF_FUA;
8837		if (cdb->byte2 & SRW12_DPO)
8838			flags |= CTL_LLF_DPO;
8839		lba = scsi_8btou64(cdb->addr);
8840		num_blocks = scsi_4btoul(cdb->length);
8841		break;
8842	}
8843	case WRITE_ATOMIC_16: {
8844		struct scsi_rw_16 *cdb;
8845
8846		if (lun->be_lun->atomicblock == 0) {
8847			ctl_set_invalid_opcode(ctsio);
8848			ctl_done((union ctl_io *)ctsio);
8849			return (CTL_RETVAL_COMPLETE);
8850		}
8851
8852		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8853		if (cdb->byte2 & SRW12_FUA)
8854			flags |= CTL_LLF_FUA;
8855		if (cdb->byte2 & SRW12_DPO)
8856			flags |= CTL_LLF_DPO;
8857		lba = scsi_8btou64(cdb->addr);
8858		num_blocks = scsi_4btoul(cdb->length);
8859		if (num_blocks > lun->be_lun->atomicblock) {
8860			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8861			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8862			    /*bit*/ 0);
8863			ctl_done((union ctl_io *)ctsio);
8864			return (CTL_RETVAL_COMPLETE);
8865		}
8866		break;
8867	}
8868	case WRITE_VERIFY_16: {
8869		struct scsi_write_verify_16 *cdb;
8870
8871		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8872		flags |= CTL_LLF_FUA;
8873		if (cdb->byte2 & SWV_DPO)
8874			flags |= CTL_LLF_DPO;
8875		lba = scsi_8btou64(cdb->addr);
8876		num_blocks = scsi_4btoul(cdb->length);
8877		break;
8878	}
8879	default:
8880		/*
8881		 * We got a command we don't support.  This shouldn't
8882		 * happen, commands should be filtered out above us.
8883		 */
8884		ctl_set_invalid_opcode(ctsio);
8885		ctl_done((union ctl_io *)ctsio);
8886
8887		return (CTL_RETVAL_COMPLETE);
8888		break; /* NOTREACHED */
8889	}
8890
8891	/*
8892	 * The first check is to make sure we're in bounds, the second
8893	 * check is to catch wrap-around problems.  If the lba + num blocks
8894	 * is less than the lba, then we've wrapped around and the block
8895	 * range is invalid anyway.
8896	 */
8897	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8898	 || ((lba + num_blocks) < lba)) {
8899		ctl_set_lba_out_of_range(ctsio);
8900		ctl_done((union ctl_io *)ctsio);
8901		return (CTL_RETVAL_COMPLETE);
8902	}
8903
8904	/*
8905	 * According to SBC-3, a transfer length of 0 is not an error.
8906	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8907	 * translates to 256 blocks for those commands.
8908	 */
8909	if (num_blocks == 0) {
8910		ctl_set_success(ctsio);
8911		ctl_done((union ctl_io *)ctsio);
8912		return (CTL_RETVAL_COMPLETE);
8913	}
8914
8915	/* Set FUA and/or DPO if caches are disabled. */
8916	if (isread) {
8917		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8918		    SCP_RCD) != 0)
8919			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8920	} else {
8921		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8922		    SCP_WCE) == 0)
8923			flags |= CTL_LLF_FUA;
8924	}
8925
8926	lbalen = (struct ctl_lba_len_flags *)
8927	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8928	lbalen->lba = lba;
8929	lbalen->len = num_blocks;
8930	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8931
8932	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8933	ctsio->kern_rel_offset = 0;
8934
8935	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8936
8937	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8938
8939	return (retval);
8940}
8941
8942static int
8943ctl_cnw_cont(union ctl_io *io)
8944{
8945	struct ctl_scsiio *ctsio;
8946	struct ctl_lun *lun;
8947	struct ctl_lba_len_flags *lbalen;
8948	int retval;
8949
8950	ctsio = &io->scsiio;
8951	ctsio->io_hdr.status = CTL_STATUS_NONE;
8952	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8953	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8954	lbalen = (struct ctl_lba_len_flags *)
8955	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8956	lbalen->flags &= ~CTL_LLF_COMPARE;
8957	lbalen->flags |= CTL_LLF_WRITE;
8958
8959	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8960	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8961	return (retval);
8962}
8963
8964int
8965ctl_cnw(struct ctl_scsiio *ctsio)
8966{
8967	struct ctl_lun *lun;
8968	struct ctl_lba_len_flags *lbalen;
8969	uint64_t lba;
8970	uint32_t num_blocks;
8971	int flags, retval;
8972
8973	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8974
8975	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8976
8977	flags = 0;
8978	retval = CTL_RETVAL_COMPLETE;
8979
8980	switch (ctsio->cdb[0]) {
8981	case COMPARE_AND_WRITE: {
8982		struct scsi_compare_and_write *cdb;
8983
8984		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8985		if (cdb->byte2 & SRW10_FUA)
8986			flags |= CTL_LLF_FUA;
8987		if (cdb->byte2 & SRW10_DPO)
8988			flags |= CTL_LLF_DPO;
8989		lba = scsi_8btou64(cdb->addr);
8990		num_blocks = cdb->length;
8991		break;
8992	}
8993	default:
8994		/*
8995		 * We got a command we don't support.  This shouldn't
8996		 * happen, commands should be filtered out above us.
8997		 */
8998		ctl_set_invalid_opcode(ctsio);
8999		ctl_done((union ctl_io *)ctsio);
9000
9001		return (CTL_RETVAL_COMPLETE);
9002		break; /* NOTREACHED */
9003	}
9004
9005	/*
9006	 * The first check is to make sure we're in bounds, the second
9007	 * check is to catch wrap-around problems.  If the lba + num blocks
9008	 * is less than the lba, then we've wrapped around and the block
9009	 * range is invalid anyway.
9010	 */
9011	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9012	 || ((lba + num_blocks) < lba)) {
9013		ctl_set_lba_out_of_range(ctsio);
9014		ctl_done((union ctl_io *)ctsio);
9015		return (CTL_RETVAL_COMPLETE);
9016	}
9017
9018	/*
9019	 * According to SBC-3, a transfer length of 0 is not an error.
9020	 */
9021	if (num_blocks == 0) {
9022		ctl_set_success(ctsio);
9023		ctl_done((union ctl_io *)ctsio);
9024		return (CTL_RETVAL_COMPLETE);
9025	}
9026
9027	/* Set FUA if write cache is disabled. */
9028	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9029	    SCP_WCE) == 0)
9030		flags |= CTL_LLF_FUA;
9031
9032	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9033	ctsio->kern_rel_offset = 0;
9034
9035	/*
9036	 * Set the IO_CONT flag, so that if this I/O gets passed to
9037	 * ctl_data_submit_done(), it'll get passed back to
9038	 * ctl_ctl_cnw_cont() for further processing.
9039	 */
9040	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9041	ctsio->io_cont = ctl_cnw_cont;
9042
9043	lbalen = (struct ctl_lba_len_flags *)
9044	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9045	lbalen->lba = lba;
9046	lbalen->len = num_blocks;
9047	lbalen->flags = CTL_LLF_COMPARE | flags;
9048
9049	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9050	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9051	return (retval);
9052}
9053
9054int
9055ctl_verify(struct ctl_scsiio *ctsio)
9056{
9057	struct ctl_lun *lun;
9058	struct ctl_lba_len_flags *lbalen;
9059	uint64_t lba;
9060	uint32_t num_blocks;
9061	int bytchk, flags;
9062	int retval;
9063
9064	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9065
9066	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9067
9068	bytchk = 0;
9069	flags = CTL_LLF_FUA;
9070	retval = CTL_RETVAL_COMPLETE;
9071
9072	switch (ctsio->cdb[0]) {
9073	case VERIFY_10: {
9074		struct scsi_verify_10 *cdb;
9075
9076		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9077		if (cdb->byte2 & SVFY_BYTCHK)
9078			bytchk = 1;
9079		if (cdb->byte2 & SVFY_DPO)
9080			flags |= CTL_LLF_DPO;
9081		lba = scsi_4btoul(cdb->addr);
9082		num_blocks = scsi_2btoul(cdb->length);
9083		break;
9084	}
9085	case VERIFY_12: {
9086		struct scsi_verify_12 *cdb;
9087
9088		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9089		if (cdb->byte2 & SVFY_BYTCHK)
9090			bytchk = 1;
9091		if (cdb->byte2 & SVFY_DPO)
9092			flags |= CTL_LLF_DPO;
9093		lba = scsi_4btoul(cdb->addr);
9094		num_blocks = scsi_4btoul(cdb->length);
9095		break;
9096	}
9097	case VERIFY_16: {
9098		struct scsi_rw_16 *cdb;
9099
9100		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9101		if (cdb->byte2 & SVFY_BYTCHK)
9102			bytchk = 1;
9103		if (cdb->byte2 & SVFY_DPO)
9104			flags |= CTL_LLF_DPO;
9105		lba = scsi_8btou64(cdb->addr);
9106		num_blocks = scsi_4btoul(cdb->length);
9107		break;
9108	}
9109	default:
9110		/*
9111		 * We got a command we don't support.  This shouldn't
9112		 * happen, commands should be filtered out above us.
9113		 */
9114		ctl_set_invalid_opcode(ctsio);
9115		ctl_done((union ctl_io *)ctsio);
9116		return (CTL_RETVAL_COMPLETE);
9117	}
9118
9119	/*
9120	 * The first check is to make sure we're in bounds, the second
9121	 * check is to catch wrap-around problems.  If the lba + num blocks
9122	 * is less than the lba, then we've wrapped around and the block
9123	 * range is invalid anyway.
9124	 */
9125	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9126	 || ((lba + num_blocks) < lba)) {
9127		ctl_set_lba_out_of_range(ctsio);
9128		ctl_done((union ctl_io *)ctsio);
9129		return (CTL_RETVAL_COMPLETE);
9130	}
9131
9132	/*
9133	 * According to SBC-3, a transfer length of 0 is not an error.
9134	 */
9135	if (num_blocks == 0) {
9136		ctl_set_success(ctsio);
9137		ctl_done((union ctl_io *)ctsio);
9138		return (CTL_RETVAL_COMPLETE);
9139	}
9140
9141	lbalen = (struct ctl_lba_len_flags *)
9142	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9143	lbalen->lba = lba;
9144	lbalen->len = num_blocks;
9145	if (bytchk) {
9146		lbalen->flags = CTL_LLF_COMPARE | flags;
9147		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9148	} else {
9149		lbalen->flags = CTL_LLF_VERIFY | flags;
9150		ctsio->kern_total_len = 0;
9151	}
9152	ctsio->kern_rel_offset = 0;
9153
9154	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9155	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9156	return (retval);
9157}
9158
9159int
9160ctl_report_luns(struct ctl_scsiio *ctsio)
9161{
9162	struct scsi_report_luns *cdb;
9163	struct scsi_report_luns_data *lun_data;
9164	struct ctl_lun *lun, *request_lun;
9165	int num_luns, retval;
9166	uint32_t alloc_len, lun_datalen;
9167	int num_filled, well_known;
9168	uint32_t initidx, targ_lun_id, lun_id;
9169
9170	retval = CTL_RETVAL_COMPLETE;
9171	well_known = 0;
9172
9173	cdb = (struct scsi_report_luns *)ctsio->cdb;
9174
9175	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9176
9177	mtx_lock(&control_softc->ctl_lock);
9178	num_luns = control_softc->num_luns;
9179	mtx_unlock(&control_softc->ctl_lock);
9180
9181	switch (cdb->select_report) {
9182	case RPL_REPORT_DEFAULT:
9183	case RPL_REPORT_ALL:
9184		break;
9185	case RPL_REPORT_WELLKNOWN:
9186		well_known = 1;
9187		num_luns = 0;
9188		break;
9189	default:
9190		ctl_set_invalid_field(ctsio,
9191				      /*sks_valid*/ 1,
9192				      /*command*/ 1,
9193				      /*field*/ 2,
9194				      /*bit_valid*/ 0,
9195				      /*bit*/ 0);
9196		ctl_done((union ctl_io *)ctsio);
9197		return (retval);
9198		break; /* NOTREACHED */
9199	}
9200
9201	alloc_len = scsi_4btoul(cdb->length);
9202	/*
9203	 * The initiator has to allocate at least 16 bytes for this request,
9204	 * so he can at least get the header and the first LUN.  Otherwise
9205	 * we reject the request (per SPC-3 rev 14, section 6.21).
9206	 */
9207	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9208	    sizeof(struct scsi_report_luns_lundata))) {
9209		ctl_set_invalid_field(ctsio,
9210				      /*sks_valid*/ 1,
9211				      /*command*/ 1,
9212				      /*field*/ 6,
9213				      /*bit_valid*/ 0,
9214				      /*bit*/ 0);
9215		ctl_done((union ctl_io *)ctsio);
9216		return (retval);
9217	}
9218
9219	request_lun = (struct ctl_lun *)
9220		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9221
9222	lun_datalen = sizeof(*lun_data) +
9223		(num_luns * sizeof(struct scsi_report_luns_lundata));
9224
9225	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9226	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9227	ctsio->kern_sg_entries = 0;
9228
9229	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9230
9231	mtx_lock(&control_softc->ctl_lock);
9232	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9233		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9234		if (lun_id >= CTL_MAX_LUNS)
9235			continue;
9236		lun = control_softc->ctl_luns[lun_id];
9237		if (lun == NULL)
9238			continue;
9239
9240		if (targ_lun_id <= 0xff) {
9241			/*
9242			 * Peripheral addressing method, bus number 0.
9243			 */
9244			lun_data->luns[num_filled].lundata[0] =
9245				RPL_LUNDATA_ATYP_PERIPH;
9246			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9247			num_filled++;
9248		} else if (targ_lun_id <= 0x3fff) {
9249			/*
9250			 * Flat addressing method.
9251			 */
9252			lun_data->luns[num_filled].lundata[0] =
9253				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9254			lun_data->luns[num_filled].lundata[1] =
9255				(targ_lun_id & 0xff);
9256			num_filled++;
9257		} else if (targ_lun_id <= 0xffffff) {
9258			/*
9259			 * Extended flat addressing method.
9260			 */
9261			lun_data->luns[num_filled].lundata[0] =
9262			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9263			scsi_ulto3b(targ_lun_id,
9264			    &lun_data->luns[num_filled].lundata[1]);
9265			num_filled++;
9266		} else {
9267			printf("ctl_report_luns: bogus LUN number %jd, "
9268			       "skipping\n", (intmax_t)targ_lun_id);
9269		}
9270		/*
9271		 * According to SPC-3, rev 14 section 6.21:
9272		 *
9273		 * "The execution of a REPORT LUNS command to any valid and
9274		 * installed logical unit shall clear the REPORTED LUNS DATA
9275		 * HAS CHANGED unit attention condition for all logical
9276		 * units of that target with respect to the requesting
9277		 * initiator. A valid and installed logical unit is one
9278		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9279		 * INQUIRY data (see 6.4.2)."
9280		 *
9281		 * If request_lun is NULL, the LUN this report luns command
9282		 * was issued to is either disabled or doesn't exist. In that
9283		 * case, we shouldn't clear any pending lun change unit
9284		 * attention.
9285		 */
9286		if (request_lun != NULL) {
9287			mtx_lock(&lun->lun_lock);
9288			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9289			mtx_unlock(&lun->lun_lock);
9290		}
9291	}
9292	mtx_unlock(&control_softc->ctl_lock);
9293
9294	/*
9295	 * It's quite possible that we've returned fewer LUNs than we allocated
9296	 * space for.  Trim it.
9297	 */
9298	lun_datalen = sizeof(*lun_data) +
9299		(num_filled * sizeof(struct scsi_report_luns_lundata));
9300
9301	if (lun_datalen < alloc_len) {
9302		ctsio->residual = alloc_len - lun_datalen;
9303		ctsio->kern_data_len = lun_datalen;
9304		ctsio->kern_total_len = lun_datalen;
9305	} else {
9306		ctsio->residual = 0;
9307		ctsio->kern_data_len = alloc_len;
9308		ctsio->kern_total_len = alloc_len;
9309	}
9310	ctsio->kern_data_resid = 0;
9311	ctsio->kern_rel_offset = 0;
9312	ctsio->kern_sg_entries = 0;
9313
9314	/*
9315	 * We set this to the actual data length, regardless of how much
9316	 * space we actually have to return results.  If the user looks at
9317	 * this value, he'll know whether or not he allocated enough space
9318	 * and reissue the command if necessary.  We don't support well
9319	 * known logical units, so if the user asks for that, return none.
9320	 */
9321	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9322
9323	/*
9324	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9325	 * this request.
9326	 */
9327	ctl_set_success(ctsio);
9328	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9329	ctsio->be_move_done = ctl_config_move_done;
9330	ctl_datamove((union ctl_io *)ctsio);
9331	return (retval);
9332}
9333
9334int
9335ctl_request_sense(struct ctl_scsiio *ctsio)
9336{
9337	struct scsi_request_sense *cdb;
9338	struct scsi_sense_data *sense_ptr;
9339	struct ctl_lun *lun;
9340	uint32_t initidx;
9341	int have_error;
9342	scsi_sense_data_type sense_format;
9343
9344	cdb = (struct scsi_request_sense *)ctsio->cdb;
9345
9346	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9347
9348	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9349
9350	/*
9351	 * Determine which sense format the user wants.
9352	 */
9353	if (cdb->byte2 & SRS_DESC)
9354		sense_format = SSD_TYPE_DESC;
9355	else
9356		sense_format = SSD_TYPE_FIXED;
9357
9358	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9359	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9360	ctsio->kern_sg_entries = 0;
9361
9362	/*
9363	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9364	 * larger than the largest allowed value for the length field in the
9365	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9366	 */
9367	ctsio->residual = 0;
9368	ctsio->kern_data_len = cdb->length;
9369	ctsio->kern_total_len = cdb->length;
9370
9371	ctsio->kern_data_resid = 0;
9372	ctsio->kern_rel_offset = 0;
9373	ctsio->kern_sg_entries = 0;
9374
9375	/*
9376	 * If we don't have a LUN, we don't have any pending sense.
9377	 */
9378	if (lun == NULL)
9379		goto no_sense;
9380
9381	have_error = 0;
9382	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9383	/*
9384	 * Check for pending sense, and then for pending unit attentions.
9385	 * Pending sense gets returned first, then pending unit attentions.
9386	 */
9387	mtx_lock(&lun->lun_lock);
9388#ifdef CTL_WITH_CA
9389	if (ctl_is_set(lun->have_ca, initidx)) {
9390		scsi_sense_data_type stored_format;
9391
9392		/*
9393		 * Check to see which sense format was used for the stored
9394		 * sense data.
9395		 */
9396		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9397
9398		/*
9399		 * If the user requested a different sense format than the
9400		 * one we stored, then we need to convert it to the other
9401		 * format.  If we're going from descriptor to fixed format
9402		 * sense data, we may lose things in translation, depending
9403		 * on what options were used.
9404		 *
9405		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9406		 * for some reason we'll just copy it out as-is.
9407		 */
9408		if ((stored_format == SSD_TYPE_FIXED)
9409		 && (sense_format == SSD_TYPE_DESC))
9410			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9411			    &lun->pending_sense[initidx],
9412			    (struct scsi_sense_data_desc *)sense_ptr);
9413		else if ((stored_format == SSD_TYPE_DESC)
9414		      && (sense_format == SSD_TYPE_FIXED))
9415			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9416			    &lun->pending_sense[initidx],
9417			    (struct scsi_sense_data_fixed *)sense_ptr);
9418		else
9419			memcpy(sense_ptr, &lun->pending_sense[initidx],
9420			       ctl_min(sizeof(*sense_ptr),
9421			       sizeof(lun->pending_sense[initidx])));
9422
9423		ctl_clear_mask(lun->have_ca, initidx);
9424		have_error = 1;
9425	} else
9426#endif
9427	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9428		ctl_ua_type ua_type;
9429
9430		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9431				       sense_ptr, sense_format);
9432		if (ua_type != CTL_UA_NONE)
9433			have_error = 1;
9434	}
9435	mtx_unlock(&lun->lun_lock);
9436
9437	/*
9438	 * We already have a pending error, return it.
9439	 */
9440	if (have_error != 0) {
9441		/*
9442		 * We report the SCSI status as OK, since the status of the
9443		 * request sense command itself is OK.
9444		 * We report 0 for the sense length, because we aren't doing
9445		 * autosense in this case.  We're reporting sense as
9446		 * parameter data.
9447		 */
9448		ctl_set_success(ctsio);
9449		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9450		ctsio->be_move_done = ctl_config_move_done;
9451		ctl_datamove((union ctl_io *)ctsio);
9452		return (CTL_RETVAL_COMPLETE);
9453	}
9454
9455no_sense:
9456
9457	/*
9458	 * No sense information to report, so we report that everything is
9459	 * okay.
9460	 */
9461	ctl_set_sense_data(sense_ptr,
9462			   lun,
9463			   sense_format,
9464			   /*current_error*/ 1,
9465			   /*sense_key*/ SSD_KEY_NO_SENSE,
9466			   /*asc*/ 0x00,
9467			   /*ascq*/ 0x00,
9468			   SSD_ELEM_NONE);
9469
9470	/*
9471	 * We report 0 for the sense length, because we aren't doing
9472	 * autosense in this case.  We're reporting sense as parameter data.
9473	 */
9474	ctl_set_success(ctsio);
9475	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9476	ctsio->be_move_done = ctl_config_move_done;
9477	ctl_datamove((union ctl_io *)ctsio);
9478	return (CTL_RETVAL_COMPLETE);
9479}
9480
9481int
9482ctl_tur(struct ctl_scsiio *ctsio)
9483{
9484
9485	CTL_DEBUG_PRINT(("ctl_tur\n"));
9486
9487	ctl_set_success(ctsio);
9488	ctl_done((union ctl_io *)ctsio);
9489
9490	return (CTL_RETVAL_COMPLETE);
9491}
9492
9493#ifdef notyet
9494static int
9495ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9496{
9497
9498}
9499#endif
9500
9501static int
9502ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9503{
9504	struct scsi_vpd_supported_pages *pages;
9505	int sup_page_size;
9506	struct ctl_lun *lun;
9507
9508	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9509
9510	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9511	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9512	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9513	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9514	ctsio->kern_sg_entries = 0;
9515
9516	if (sup_page_size < alloc_len) {
9517		ctsio->residual = alloc_len - sup_page_size;
9518		ctsio->kern_data_len = sup_page_size;
9519		ctsio->kern_total_len = sup_page_size;
9520	} else {
9521		ctsio->residual = 0;
9522		ctsio->kern_data_len = alloc_len;
9523		ctsio->kern_total_len = alloc_len;
9524	}
9525	ctsio->kern_data_resid = 0;
9526	ctsio->kern_rel_offset = 0;
9527	ctsio->kern_sg_entries = 0;
9528
9529	/*
9530	 * The control device is always connected.  The disk device, on the
9531	 * other hand, may not be online all the time.  Need to change this
9532	 * to figure out whether the disk device is actually online or not.
9533	 */
9534	if (lun != NULL)
9535		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9536				lun->be_lun->lun_type;
9537	else
9538		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9539
9540	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9541	/* Supported VPD pages */
9542	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9543	/* Serial Number */
9544	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9545	/* Device Identification */
9546	pages->page_list[2] = SVPD_DEVICE_ID;
9547	/* Extended INQUIRY Data */
9548	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9549	/* Mode Page Policy */
9550	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9551	/* SCSI Ports */
9552	pages->page_list[5] = SVPD_SCSI_PORTS;
9553	/* Third-party Copy */
9554	pages->page_list[6] = SVPD_SCSI_TPC;
9555	/* Block limits */
9556	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9557	/* Block Device Characteristics */
9558	pages->page_list[8] = SVPD_BDC;
9559	/* Logical Block Provisioning */
9560	pages->page_list[9] = SVPD_LBP;
9561
9562	ctl_set_success(ctsio);
9563	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9564	ctsio->be_move_done = ctl_config_move_done;
9565	ctl_datamove((union ctl_io *)ctsio);
9566	return (CTL_RETVAL_COMPLETE);
9567}
9568
9569static int
9570ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9571{
9572	struct scsi_vpd_unit_serial_number *sn_ptr;
9573	struct ctl_lun *lun;
9574	int data_len;
9575
9576	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9577
9578	data_len = 4 + CTL_SN_LEN;
9579	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9580	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9581	if (data_len < alloc_len) {
9582		ctsio->residual = alloc_len - data_len;
9583		ctsio->kern_data_len = data_len;
9584		ctsio->kern_total_len = data_len;
9585	} else {
9586		ctsio->residual = 0;
9587		ctsio->kern_data_len = alloc_len;
9588		ctsio->kern_total_len = alloc_len;
9589	}
9590	ctsio->kern_data_resid = 0;
9591	ctsio->kern_rel_offset = 0;
9592	ctsio->kern_sg_entries = 0;
9593
9594	/*
9595	 * The control device is always connected.  The disk device, on the
9596	 * other hand, may not be online all the time.  Need to change this
9597	 * to figure out whether the disk device is actually online or not.
9598	 */
9599	if (lun != NULL)
9600		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9601				  lun->be_lun->lun_type;
9602	else
9603		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9604
9605	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9606	sn_ptr->length = CTL_SN_LEN;
9607	/*
9608	 * If we don't have a LUN, we just leave the serial number as
9609	 * all spaces.
9610	 */
9611	if (lun != NULL) {
9612		strncpy((char *)sn_ptr->serial_num,
9613			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9614	} else
9615		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9616
9617	ctl_set_success(ctsio);
9618	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9619	ctsio->be_move_done = ctl_config_move_done;
9620	ctl_datamove((union ctl_io *)ctsio);
9621	return (CTL_RETVAL_COMPLETE);
9622}
9623
9624
9625static int
9626ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9627{
9628	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9629	struct ctl_lun *lun;
9630	int data_len;
9631
9632	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9633
9634	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9635	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9636	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9637	ctsio->kern_sg_entries = 0;
9638
9639	if (data_len < alloc_len) {
9640		ctsio->residual = alloc_len - data_len;
9641		ctsio->kern_data_len = data_len;
9642		ctsio->kern_total_len = data_len;
9643	} else {
9644		ctsio->residual = 0;
9645		ctsio->kern_data_len = alloc_len;
9646		ctsio->kern_total_len = alloc_len;
9647	}
9648	ctsio->kern_data_resid = 0;
9649	ctsio->kern_rel_offset = 0;
9650	ctsio->kern_sg_entries = 0;
9651
9652	/*
9653	 * The control device is always connected.  The disk device, on the
9654	 * other hand, may not be online all the time.
9655	 */
9656	if (lun != NULL)
9657		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9658				     lun->be_lun->lun_type;
9659	else
9660		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9661	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9662	eid_ptr->page_length = data_len - 4;
9663	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9664	eid_ptr->flags3 = SVPD_EID_V_SUP;
9665
9666	ctl_set_success(ctsio);
9667	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9668	ctsio->be_move_done = ctl_config_move_done;
9669	ctl_datamove((union ctl_io *)ctsio);
9670	return (CTL_RETVAL_COMPLETE);
9671}
9672
9673static int
9674ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9675{
9676	struct scsi_vpd_mode_page_policy *mpp_ptr;
9677	struct ctl_lun *lun;
9678	int data_len;
9679
9680	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9681
9682	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9683	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9684
9685	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9686	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9687	ctsio->kern_sg_entries = 0;
9688
9689	if (data_len < alloc_len) {
9690		ctsio->residual = alloc_len - data_len;
9691		ctsio->kern_data_len = data_len;
9692		ctsio->kern_total_len = data_len;
9693	} else {
9694		ctsio->residual = 0;
9695		ctsio->kern_data_len = alloc_len;
9696		ctsio->kern_total_len = alloc_len;
9697	}
9698	ctsio->kern_data_resid = 0;
9699	ctsio->kern_rel_offset = 0;
9700	ctsio->kern_sg_entries = 0;
9701
9702	/*
9703	 * The control device is always connected.  The disk device, on the
9704	 * other hand, may not be online all the time.
9705	 */
9706	if (lun != NULL)
9707		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9708				     lun->be_lun->lun_type;
9709	else
9710		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9711	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9712	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9713	mpp_ptr->descr[0].page_code = 0x3f;
9714	mpp_ptr->descr[0].subpage_code = 0xff;
9715	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9716
9717	ctl_set_success(ctsio);
9718	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9719	ctsio->be_move_done = ctl_config_move_done;
9720	ctl_datamove((union ctl_io *)ctsio);
9721	return (CTL_RETVAL_COMPLETE);
9722}
9723
9724static int
9725ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9726{
9727	struct scsi_vpd_device_id *devid_ptr;
9728	struct scsi_vpd_id_descriptor *desc;
9729	struct ctl_softc *ctl_softc;
9730	struct ctl_lun *lun;
9731	struct ctl_port *port;
9732	int data_len;
9733	uint8_t proto;
9734
9735	ctl_softc = control_softc;
9736
9737	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9738	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9739
9740	data_len = sizeof(struct scsi_vpd_device_id) +
9741	    sizeof(struct scsi_vpd_id_descriptor) +
9742		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9743	    sizeof(struct scsi_vpd_id_descriptor) +
9744		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9745	if (lun && lun->lun_devid)
9746		data_len += lun->lun_devid->len;
9747	if (port->port_devid)
9748		data_len += port->port_devid->len;
9749	if (port->target_devid)
9750		data_len += port->target_devid->len;
9751
9752	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9753	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9754	ctsio->kern_sg_entries = 0;
9755
9756	if (data_len < alloc_len) {
9757		ctsio->residual = alloc_len - data_len;
9758		ctsio->kern_data_len = data_len;
9759		ctsio->kern_total_len = data_len;
9760	} else {
9761		ctsio->residual = 0;
9762		ctsio->kern_data_len = alloc_len;
9763		ctsio->kern_total_len = alloc_len;
9764	}
9765	ctsio->kern_data_resid = 0;
9766	ctsio->kern_rel_offset = 0;
9767	ctsio->kern_sg_entries = 0;
9768
9769	/*
9770	 * The control device is always connected.  The disk device, on the
9771	 * other hand, may not be online all the time.
9772	 */
9773	if (lun != NULL)
9774		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9775				     lun->be_lun->lun_type;
9776	else
9777		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9778	devid_ptr->page_code = SVPD_DEVICE_ID;
9779	scsi_ulto2b(data_len - 4, devid_ptr->length);
9780
9781	if (port->port_type == CTL_PORT_FC)
9782		proto = SCSI_PROTO_FC << 4;
9783	else if (port->port_type == CTL_PORT_ISCSI)
9784		proto = SCSI_PROTO_ISCSI << 4;
9785	else
9786		proto = SCSI_PROTO_SPI << 4;
9787	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9788
9789	/*
9790	 * We're using a LUN association here.  i.e., this device ID is a
9791	 * per-LUN identifier.
9792	 */
9793	if (lun && lun->lun_devid) {
9794		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9795		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9796		    lun->lun_devid->len);
9797	}
9798
9799	/*
9800	 * This is for the WWPN which is a port association.
9801	 */
9802	if (port->port_devid) {
9803		memcpy(desc, port->port_devid->data, port->port_devid->len);
9804		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9805		    port->port_devid->len);
9806	}
9807
9808	/*
9809	 * This is for the Relative Target Port(type 4h) identifier
9810	 */
9811	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9812	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9813	    SVPD_ID_TYPE_RELTARG;
9814	desc->length = 4;
9815	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9816	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9817	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9818
9819	/*
9820	 * This is for the Target Port Group(type 5h) identifier
9821	 */
9822	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9823	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9824	    SVPD_ID_TYPE_TPORTGRP;
9825	desc->length = 4;
9826	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9827	    &desc->identifier[2]);
9828	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9829	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9830
9831	/*
9832	 * This is for the Target identifier
9833	 */
9834	if (port->target_devid) {
9835		memcpy(desc, port->target_devid->data, port->target_devid->len);
9836	}
9837
9838	ctl_set_success(ctsio);
9839	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9840	ctsio->be_move_done = ctl_config_move_done;
9841	ctl_datamove((union ctl_io *)ctsio);
9842	return (CTL_RETVAL_COMPLETE);
9843}
9844
9845static int
9846ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9847{
9848	struct ctl_softc *softc = control_softc;
9849	struct scsi_vpd_scsi_ports *sp;
9850	struct scsi_vpd_port_designation *pd;
9851	struct scsi_vpd_port_designation_cont *pdc;
9852	struct ctl_lun *lun;
9853	struct ctl_port *port;
9854	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9855	int num_target_port_groups;
9856
9857	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9858
9859	if (softc->is_single)
9860		num_target_port_groups = 1;
9861	else
9862		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9863	num_target_ports = 0;
9864	iid_len = 0;
9865	id_len = 0;
9866	mtx_lock(&softc->ctl_lock);
9867	STAILQ_FOREACH(port, &softc->port_list, links) {
9868		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9869			continue;
9870		if (lun != NULL &&
9871		    ctl_map_lun_back(port->targ_port, lun->lun) >=
9872		    CTL_MAX_LUNS)
9873			continue;
9874		num_target_ports++;
9875		if (port->init_devid)
9876			iid_len += port->init_devid->len;
9877		if (port->port_devid)
9878			id_len += port->port_devid->len;
9879	}
9880	mtx_unlock(&softc->ctl_lock);
9881
9882	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9883	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9884	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9885	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9886	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9887	ctsio->kern_sg_entries = 0;
9888
9889	if (data_len < alloc_len) {
9890		ctsio->residual = alloc_len - data_len;
9891		ctsio->kern_data_len = data_len;
9892		ctsio->kern_total_len = data_len;
9893	} else {
9894		ctsio->residual = 0;
9895		ctsio->kern_data_len = alloc_len;
9896		ctsio->kern_total_len = alloc_len;
9897	}
9898	ctsio->kern_data_resid = 0;
9899	ctsio->kern_rel_offset = 0;
9900	ctsio->kern_sg_entries = 0;
9901
9902	/*
9903	 * The control device is always connected.  The disk device, on the
9904	 * other hand, may not be online all the time.  Need to change this
9905	 * to figure out whether the disk device is actually online or not.
9906	 */
9907	if (lun != NULL)
9908		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9909				  lun->be_lun->lun_type;
9910	else
9911		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9912
9913	sp->page_code = SVPD_SCSI_PORTS;
9914	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9915	    sp->page_length);
9916	pd = &sp->design[0];
9917
9918	mtx_lock(&softc->ctl_lock);
9919	pg = softc->port_offset / CTL_MAX_PORTS;
9920	for (g = 0; g < num_target_port_groups; g++) {
9921		STAILQ_FOREACH(port, &softc->port_list, links) {
9922			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9923				continue;
9924			if (lun != NULL &&
9925			    ctl_map_lun_back(port->targ_port, lun->lun) >=
9926			    CTL_MAX_LUNS)
9927				continue;
9928			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
9929			scsi_ulto2b(p, pd->relative_port_id);
9930			if (port->init_devid && g == pg) {
9931				iid_len = port->init_devid->len;
9932				memcpy(pd->initiator_transportid,
9933				    port->init_devid->data, port->init_devid->len);
9934			} else
9935				iid_len = 0;
9936			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9937			pdc = (struct scsi_vpd_port_designation_cont *)
9938			    (&pd->initiator_transportid[iid_len]);
9939			if (port->port_devid && g == pg) {
9940				id_len = port->port_devid->len;
9941				memcpy(pdc->target_port_descriptors,
9942				    port->port_devid->data, port->port_devid->len);
9943			} else
9944				id_len = 0;
9945			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9946			pd = (struct scsi_vpd_port_designation *)
9947			    ((uint8_t *)pdc->target_port_descriptors + id_len);
9948		}
9949	}
9950	mtx_unlock(&softc->ctl_lock);
9951
9952	ctl_set_success(ctsio);
9953	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9954	ctsio->be_move_done = ctl_config_move_done;
9955	ctl_datamove((union ctl_io *)ctsio);
9956	return (CTL_RETVAL_COMPLETE);
9957}
9958
9959static int
9960ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9961{
9962	struct scsi_vpd_block_limits *bl_ptr;
9963	struct ctl_lun *lun;
9964	int bs;
9965
9966	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9967
9968	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9969	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9970	ctsio->kern_sg_entries = 0;
9971
9972	if (sizeof(*bl_ptr) < alloc_len) {
9973		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9974		ctsio->kern_data_len = sizeof(*bl_ptr);
9975		ctsio->kern_total_len = sizeof(*bl_ptr);
9976	} else {
9977		ctsio->residual = 0;
9978		ctsio->kern_data_len = alloc_len;
9979		ctsio->kern_total_len = alloc_len;
9980	}
9981	ctsio->kern_data_resid = 0;
9982	ctsio->kern_rel_offset = 0;
9983	ctsio->kern_sg_entries = 0;
9984
9985	/*
9986	 * The control device is always connected.  The disk device, on the
9987	 * other hand, may not be online all the time.  Need to change this
9988	 * to figure out whether the disk device is actually online or not.
9989	 */
9990	if (lun != NULL)
9991		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9992				  lun->be_lun->lun_type;
9993	else
9994		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9995
9996	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9997	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9998	bl_ptr->max_cmp_write_len = 0xff;
9999	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10000	if (lun != NULL) {
10001		bs = lun->be_lun->blocksize;
10002		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10003		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10004			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10005			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10006			if (lun->be_lun->pblockexp != 0) {
10007				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10008				    bl_ptr->opt_unmap_grain);
10009				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10010				    bl_ptr->unmap_grain_align);
10011			}
10012		}
10013		scsi_ulto4b(lun->be_lun->atomicblock,
10014		    bl_ptr->max_atomic_transfer_length);
10015		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10016		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10017	}
10018	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10019
10020	ctl_set_success(ctsio);
10021	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10022	ctsio->be_move_done = ctl_config_move_done;
10023	ctl_datamove((union ctl_io *)ctsio);
10024	return (CTL_RETVAL_COMPLETE);
10025}
10026
10027static int
10028ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10029{
10030	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10031	struct ctl_lun *lun;
10032	const char *value;
10033	u_int i;
10034
10035	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10036
10037	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10038	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10039	ctsio->kern_sg_entries = 0;
10040
10041	if (sizeof(*bdc_ptr) < alloc_len) {
10042		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10043		ctsio->kern_data_len = sizeof(*bdc_ptr);
10044		ctsio->kern_total_len = sizeof(*bdc_ptr);
10045	} else {
10046		ctsio->residual = 0;
10047		ctsio->kern_data_len = alloc_len;
10048		ctsio->kern_total_len = alloc_len;
10049	}
10050	ctsio->kern_data_resid = 0;
10051	ctsio->kern_rel_offset = 0;
10052	ctsio->kern_sg_entries = 0;
10053
10054	/*
10055	 * The control device is always connected.  The disk device, on the
10056	 * other hand, may not be online all the time.  Need to change this
10057	 * to figure out whether the disk device is actually online or not.
10058	 */
10059	if (lun != NULL)
10060		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10061				  lun->be_lun->lun_type;
10062	else
10063		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10064	bdc_ptr->page_code = SVPD_BDC;
10065	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10066	if (lun != NULL &&
10067	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10068		i = strtol(value, NULL, 0);
10069	else
10070		i = CTL_DEFAULT_ROTATION_RATE;
10071	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10072	if (lun != NULL &&
10073	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10074		i = strtol(value, NULL, 0);
10075	else
10076		i = 0;
10077	bdc_ptr->wab_wac_ff = (i & 0x0f);
10078	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10079
10080	ctl_set_success(ctsio);
10081	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10082	ctsio->be_move_done = ctl_config_move_done;
10083	ctl_datamove((union ctl_io *)ctsio);
10084	return (CTL_RETVAL_COMPLETE);
10085}
10086
10087static int
10088ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10089{
10090	struct scsi_vpd_logical_block_prov *lbp_ptr;
10091	struct ctl_lun *lun;
10092
10093	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10094
10095	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10096	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10097	ctsio->kern_sg_entries = 0;
10098
10099	if (sizeof(*lbp_ptr) < alloc_len) {
10100		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10101		ctsio->kern_data_len = sizeof(*lbp_ptr);
10102		ctsio->kern_total_len = sizeof(*lbp_ptr);
10103	} else {
10104		ctsio->residual = 0;
10105		ctsio->kern_data_len = alloc_len;
10106		ctsio->kern_total_len = alloc_len;
10107	}
10108	ctsio->kern_data_resid = 0;
10109	ctsio->kern_rel_offset = 0;
10110	ctsio->kern_sg_entries = 0;
10111
10112	/*
10113	 * The control device is always connected.  The disk device, on the
10114	 * other hand, may not be online all the time.  Need to change this
10115	 * to figure out whether the disk device is actually online or not.
10116	 */
10117	if (lun != NULL)
10118		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10119				  lun->be_lun->lun_type;
10120	else
10121		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10122
10123	lbp_ptr->page_code = SVPD_LBP;
10124	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10125	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10126		lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10127		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10128		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10129		lbp_ptr->prov_type = SVPD_LBP_THIN;
10130	}
10131
10132	ctl_set_success(ctsio);
10133	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10134	ctsio->be_move_done = ctl_config_move_done;
10135	ctl_datamove((union ctl_io *)ctsio);
10136	return (CTL_RETVAL_COMPLETE);
10137}
10138
10139static int
10140ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10141{
10142	struct scsi_inquiry *cdb;
10143	int alloc_len, retval;
10144
10145	cdb = (struct scsi_inquiry *)ctsio->cdb;
10146
10147	retval = CTL_RETVAL_COMPLETE;
10148
10149	alloc_len = scsi_2btoul(cdb->length);
10150
10151	switch (cdb->page_code) {
10152	case SVPD_SUPPORTED_PAGES:
10153		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10154		break;
10155	case SVPD_UNIT_SERIAL_NUMBER:
10156		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10157		break;
10158	case SVPD_DEVICE_ID:
10159		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10160		break;
10161	case SVPD_EXTENDED_INQUIRY_DATA:
10162		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10163		break;
10164	case SVPD_MODE_PAGE_POLICY:
10165		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10166		break;
10167	case SVPD_SCSI_PORTS:
10168		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10169		break;
10170	case SVPD_SCSI_TPC:
10171		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10172		break;
10173	case SVPD_BLOCK_LIMITS:
10174		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10175		break;
10176	case SVPD_BDC:
10177		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10178		break;
10179	case SVPD_LBP:
10180		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10181		break;
10182	default:
10183		ctl_set_invalid_field(ctsio,
10184				      /*sks_valid*/ 1,
10185				      /*command*/ 1,
10186				      /*field*/ 2,
10187				      /*bit_valid*/ 0,
10188				      /*bit*/ 0);
10189		ctl_done((union ctl_io *)ctsio);
10190		retval = CTL_RETVAL_COMPLETE;
10191		break;
10192	}
10193
10194	return (retval);
10195}
10196
10197static int
10198ctl_inquiry_std(struct ctl_scsiio *ctsio)
10199{
10200	struct scsi_inquiry_data *inq_ptr;
10201	struct scsi_inquiry *cdb;
10202	struct ctl_softc *ctl_softc;
10203	struct ctl_lun *lun;
10204	char *val;
10205	uint32_t alloc_len, data_len;
10206	ctl_port_type port_type;
10207
10208	ctl_softc = control_softc;
10209
10210	/*
10211	 * Figure out whether we're talking to a Fibre Channel port or not.
10212	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10213	 * SCSI front ends.
10214	 */
10215	port_type = ctl_softc->ctl_ports[
10216	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10217	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10218		port_type = CTL_PORT_SCSI;
10219
10220	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10221	cdb = (struct scsi_inquiry *)ctsio->cdb;
10222	alloc_len = scsi_2btoul(cdb->length);
10223
10224	/*
10225	 * We malloc the full inquiry data size here and fill it
10226	 * in.  If the user only asks for less, we'll give him
10227	 * that much.
10228	 */
10229	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10230	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10231	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10232	ctsio->kern_sg_entries = 0;
10233	ctsio->kern_data_resid = 0;
10234	ctsio->kern_rel_offset = 0;
10235
10236	if (data_len < alloc_len) {
10237		ctsio->residual = alloc_len - data_len;
10238		ctsio->kern_data_len = data_len;
10239		ctsio->kern_total_len = data_len;
10240	} else {
10241		ctsio->residual = 0;
10242		ctsio->kern_data_len = alloc_len;
10243		ctsio->kern_total_len = alloc_len;
10244	}
10245
10246	/*
10247	 * If we have a LUN configured, report it as connected.  Otherwise,
10248	 * report that it is offline or no device is supported, depending
10249	 * on the value of inquiry_pq_no_lun.
10250	 *
10251	 * According to the spec (SPC-4 r34), the peripheral qualifier
10252	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10253	 *
10254	 * "A peripheral device having the specified peripheral device type
10255	 * is not connected to this logical unit. However, the device
10256	 * server is capable of supporting the specified peripheral device
10257	 * type on this logical unit."
10258	 *
10259	 * According to the same spec, the peripheral qualifier
10260	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10261	 *
10262	 * "The device server is not capable of supporting a peripheral
10263	 * device on this logical unit. For this peripheral qualifier the
10264	 * peripheral device type shall be set to 1Fh. All other peripheral
10265	 * device type values are reserved for this peripheral qualifier."
10266	 *
10267	 * Given the text, it would seem that we probably want to report that
10268	 * the LUN is offline here.  There is no LUN connected, but we can
10269	 * support a LUN at the given LUN number.
10270	 *
10271	 * In the real world, though, it sounds like things are a little
10272	 * different:
10273	 *
10274	 * - Linux, when presented with a LUN with the offline peripheral
10275	 *   qualifier, will create an sg driver instance for it.  So when
10276	 *   you attach it to CTL, you wind up with a ton of sg driver
10277	 *   instances.  (One for every LUN that Linux bothered to probe.)
10278	 *   Linux does this despite the fact that it issues a REPORT LUNs
10279	 *   to LUN 0 to get the inventory of supported LUNs.
10280	 *
10281	 * - There is other anecdotal evidence (from Emulex folks) about
10282	 *   arrays that use the offline peripheral qualifier for LUNs that
10283	 *   are on the "passive" path in an active/passive array.
10284	 *
10285	 * So the solution is provide a hopefully reasonable default
10286	 * (return bad/no LUN) and allow the user to change the behavior
10287	 * with a tunable/sysctl variable.
10288	 */
10289	if (lun != NULL)
10290		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10291				  lun->be_lun->lun_type;
10292	else if (ctl_softc->inquiry_pq_no_lun == 0)
10293		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10294	else
10295		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10296
10297	/* RMB in byte 2 is 0 */
10298	inq_ptr->version = SCSI_REV_SPC4;
10299
10300	/*
10301	 * According to SAM-3, even if a device only supports a single
10302	 * level of LUN addressing, it should still set the HISUP bit:
10303	 *
10304	 * 4.9.1 Logical unit numbers overview
10305	 *
10306	 * All logical unit number formats described in this standard are
10307	 * hierarchical in structure even when only a single level in that
10308	 * hierarchy is used. The HISUP bit shall be set to one in the
10309	 * standard INQUIRY data (see SPC-2) when any logical unit number
10310	 * format described in this standard is used.  Non-hierarchical
10311	 * formats are outside the scope of this standard.
10312	 *
10313	 * Therefore we set the HiSup bit here.
10314	 *
10315	 * The reponse format is 2, per SPC-3.
10316	 */
10317	inq_ptr->response_format = SID_HiSup | 2;
10318
10319	inq_ptr->additional_length = data_len -
10320	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10321	CTL_DEBUG_PRINT(("additional_length = %d\n",
10322			 inq_ptr->additional_length));
10323
10324	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10325	/* 16 bit addressing */
10326	if (port_type == CTL_PORT_SCSI)
10327		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10328	/* XXX set the SID_MultiP bit here if we're actually going to
10329	   respond on multiple ports */
10330	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10331
10332	/* 16 bit data bus, synchronous transfers */
10333	if (port_type == CTL_PORT_SCSI)
10334		inq_ptr->flags = SID_WBus16 | SID_Sync;
10335	/*
10336	 * XXX KDM do we want to support tagged queueing on the control
10337	 * device at all?
10338	 */
10339	if ((lun == NULL)
10340	 || (lun->be_lun->lun_type != T_PROCESSOR))
10341		inq_ptr->flags |= SID_CmdQue;
10342	/*
10343	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10344	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10345	 * name and 4 bytes for the revision.
10346	 */
10347	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10348	    "vendor")) == NULL) {
10349		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10350	} else {
10351		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10352		strncpy(inq_ptr->vendor, val,
10353		    min(sizeof(inq_ptr->vendor), strlen(val)));
10354	}
10355	if (lun == NULL) {
10356		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10357		    sizeof(inq_ptr->product));
10358	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10359		switch (lun->be_lun->lun_type) {
10360		case T_DIRECT:
10361			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10362			    sizeof(inq_ptr->product));
10363			break;
10364		case T_PROCESSOR:
10365			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10366			    sizeof(inq_ptr->product));
10367			break;
10368		default:
10369			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10370			    sizeof(inq_ptr->product));
10371			break;
10372		}
10373	} else {
10374		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10375		strncpy(inq_ptr->product, val,
10376		    min(sizeof(inq_ptr->product), strlen(val)));
10377	}
10378
10379	/*
10380	 * XXX make this a macro somewhere so it automatically gets
10381	 * incremented when we make changes.
10382	 */
10383	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10384	    "revision")) == NULL) {
10385		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10386	} else {
10387		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10388		strncpy(inq_ptr->revision, val,
10389		    min(sizeof(inq_ptr->revision), strlen(val)));
10390	}
10391
10392	/*
10393	 * For parallel SCSI, we support double transition and single
10394	 * transition clocking.  We also support QAS (Quick Arbitration
10395	 * and Selection) and Information Unit transfers on both the
10396	 * control and array devices.
10397	 */
10398	if (port_type == CTL_PORT_SCSI)
10399		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10400				    SID_SPI_IUS;
10401
10402	/* SAM-5 (no version claimed) */
10403	scsi_ulto2b(0x00A0, inq_ptr->version1);
10404	/* SPC-4 (no version claimed) */
10405	scsi_ulto2b(0x0460, inq_ptr->version2);
10406	if (port_type == CTL_PORT_FC) {
10407		/* FCP-2 ANSI INCITS.350:2003 */
10408		scsi_ulto2b(0x0917, inq_ptr->version3);
10409	} else if (port_type == CTL_PORT_SCSI) {
10410		/* SPI-4 ANSI INCITS.362:200x */
10411		scsi_ulto2b(0x0B56, inq_ptr->version3);
10412	} else if (port_type == CTL_PORT_ISCSI) {
10413		/* iSCSI (no version claimed) */
10414		scsi_ulto2b(0x0960, inq_ptr->version3);
10415	} else if (port_type == CTL_PORT_SAS) {
10416		/* SAS (no version claimed) */
10417		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10418	}
10419
10420	if (lun == NULL) {
10421		/* SBC-4 (no version claimed) */
10422		scsi_ulto2b(0x0600, inq_ptr->version4);
10423	} else {
10424		switch (lun->be_lun->lun_type) {
10425		case T_DIRECT:
10426			/* SBC-4 (no version claimed) */
10427			scsi_ulto2b(0x0600, inq_ptr->version4);
10428			break;
10429		case T_PROCESSOR:
10430		default:
10431			break;
10432		}
10433	}
10434
10435	ctl_set_success(ctsio);
10436	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10437	ctsio->be_move_done = ctl_config_move_done;
10438	ctl_datamove((union ctl_io *)ctsio);
10439	return (CTL_RETVAL_COMPLETE);
10440}
10441
10442int
10443ctl_inquiry(struct ctl_scsiio *ctsio)
10444{
10445	struct scsi_inquiry *cdb;
10446	int retval;
10447
10448	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10449
10450	cdb = (struct scsi_inquiry *)ctsio->cdb;
10451	if (cdb->byte2 & SI_EVPD)
10452		retval = ctl_inquiry_evpd(ctsio);
10453	else if (cdb->page_code == 0)
10454		retval = ctl_inquiry_std(ctsio);
10455	else {
10456		ctl_set_invalid_field(ctsio,
10457				      /*sks_valid*/ 1,
10458				      /*command*/ 1,
10459				      /*field*/ 2,
10460				      /*bit_valid*/ 0,
10461				      /*bit*/ 0);
10462		ctl_done((union ctl_io *)ctsio);
10463		return (CTL_RETVAL_COMPLETE);
10464	}
10465
10466	return (retval);
10467}
10468
10469/*
10470 * For known CDB types, parse the LBA and length.
10471 */
10472static int
10473ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10474{
10475	if (io->io_hdr.io_type != CTL_IO_SCSI)
10476		return (1);
10477
10478	switch (io->scsiio.cdb[0]) {
10479	case COMPARE_AND_WRITE: {
10480		struct scsi_compare_and_write *cdb;
10481
10482		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10483
10484		*lba = scsi_8btou64(cdb->addr);
10485		*len = cdb->length;
10486		break;
10487	}
10488	case READ_6:
10489	case WRITE_6: {
10490		struct scsi_rw_6 *cdb;
10491
10492		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10493
10494		*lba = scsi_3btoul(cdb->addr);
10495		/* only 5 bits are valid in the most significant address byte */
10496		*lba &= 0x1fffff;
10497		*len = cdb->length;
10498		break;
10499	}
10500	case READ_10:
10501	case WRITE_10: {
10502		struct scsi_rw_10 *cdb;
10503
10504		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10505
10506		*lba = scsi_4btoul(cdb->addr);
10507		*len = scsi_2btoul(cdb->length);
10508		break;
10509	}
10510	case WRITE_VERIFY_10: {
10511		struct scsi_write_verify_10 *cdb;
10512
10513		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10514
10515		*lba = scsi_4btoul(cdb->addr);
10516		*len = scsi_2btoul(cdb->length);
10517		break;
10518	}
10519	case READ_12:
10520	case WRITE_12: {
10521		struct scsi_rw_12 *cdb;
10522
10523		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10524
10525		*lba = scsi_4btoul(cdb->addr);
10526		*len = scsi_4btoul(cdb->length);
10527		break;
10528	}
10529	case WRITE_VERIFY_12: {
10530		struct scsi_write_verify_12 *cdb;
10531
10532		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10533
10534		*lba = scsi_4btoul(cdb->addr);
10535		*len = scsi_4btoul(cdb->length);
10536		break;
10537	}
10538	case READ_16:
10539	case WRITE_16:
10540	case WRITE_ATOMIC_16: {
10541		struct scsi_rw_16 *cdb;
10542
10543		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10544
10545		*lba = scsi_8btou64(cdb->addr);
10546		*len = scsi_4btoul(cdb->length);
10547		break;
10548	}
10549	case WRITE_VERIFY_16: {
10550		struct scsi_write_verify_16 *cdb;
10551
10552		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10553
10554		*lba = scsi_8btou64(cdb->addr);
10555		*len = scsi_4btoul(cdb->length);
10556		break;
10557	}
10558	case WRITE_SAME_10: {
10559		struct scsi_write_same_10 *cdb;
10560
10561		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10562
10563		*lba = scsi_4btoul(cdb->addr);
10564		*len = scsi_2btoul(cdb->length);
10565		break;
10566	}
10567	case WRITE_SAME_16: {
10568		struct scsi_write_same_16 *cdb;
10569
10570		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10571
10572		*lba = scsi_8btou64(cdb->addr);
10573		*len = scsi_4btoul(cdb->length);
10574		break;
10575	}
10576	case VERIFY_10: {
10577		struct scsi_verify_10 *cdb;
10578
10579		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10580
10581		*lba = scsi_4btoul(cdb->addr);
10582		*len = scsi_2btoul(cdb->length);
10583		break;
10584	}
10585	case VERIFY_12: {
10586		struct scsi_verify_12 *cdb;
10587
10588		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10589
10590		*lba = scsi_4btoul(cdb->addr);
10591		*len = scsi_4btoul(cdb->length);
10592		break;
10593	}
10594	case VERIFY_16: {
10595		struct scsi_verify_16 *cdb;
10596
10597		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10598
10599		*lba = scsi_8btou64(cdb->addr);
10600		*len = scsi_4btoul(cdb->length);
10601		break;
10602	}
10603	case UNMAP: {
10604		*lba = 0;
10605		*len = UINT64_MAX;
10606		break;
10607	}
10608	default:
10609		return (1);
10610		break; /* NOTREACHED */
10611	}
10612
10613	return (0);
10614}
10615
10616static ctl_action
10617ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10618{
10619	uint64_t endlba1, endlba2;
10620
10621	endlba1 = lba1 + len1 - 1;
10622	endlba2 = lba2 + len2 - 1;
10623
10624	if ((endlba1 < lba2)
10625	 || (endlba2 < lba1))
10626		return (CTL_ACTION_PASS);
10627	else
10628		return (CTL_ACTION_BLOCK);
10629}
10630
10631static int
10632ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10633{
10634	struct ctl_ptr_len_flags *ptrlen;
10635	struct scsi_unmap_desc *buf, *end, *range;
10636	uint64_t lba;
10637	uint32_t len;
10638
10639	/* If not UNMAP -- go other way. */
10640	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10641	    io->scsiio.cdb[0] != UNMAP)
10642		return (CTL_ACTION_ERROR);
10643
10644	/* If UNMAP without data -- block and wait for data. */
10645	ptrlen = (struct ctl_ptr_len_flags *)
10646	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10647	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10648	    ptrlen->ptr == NULL)
10649		return (CTL_ACTION_BLOCK);
10650
10651	/* UNMAP with data -- check for collision. */
10652	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10653	end = buf + ptrlen->len / sizeof(*buf);
10654	for (range = buf; range < end; range++) {
10655		lba = scsi_8btou64(range->lba);
10656		len = scsi_4btoul(range->length);
10657		if ((lba < lba2 + len2) && (lba + len > lba2))
10658			return (CTL_ACTION_BLOCK);
10659	}
10660	return (CTL_ACTION_PASS);
10661}
10662
10663static ctl_action
10664ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10665{
10666	uint64_t lba1, lba2;
10667	uint64_t len1, len2;
10668	int retval;
10669
10670	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10671		return (CTL_ACTION_ERROR);
10672
10673	retval = ctl_extent_check_unmap(io2, lba1, len1);
10674	if (retval != CTL_ACTION_ERROR)
10675		return (retval);
10676
10677	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10678		return (CTL_ACTION_ERROR);
10679
10680	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10681}
10682
10683static ctl_action
10684ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10685    union ctl_io *ooa_io)
10686{
10687	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10688	ctl_serialize_action *serialize_row;
10689
10690	/*
10691	 * The initiator attempted multiple untagged commands at the same
10692	 * time.  Can't do that.
10693	 */
10694	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10695	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10696	 && ((pending_io->io_hdr.nexus.targ_port ==
10697	      ooa_io->io_hdr.nexus.targ_port)
10698	  && (pending_io->io_hdr.nexus.initid.id ==
10699	      ooa_io->io_hdr.nexus.initid.id))
10700	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10701		return (CTL_ACTION_OVERLAP);
10702
10703	/*
10704	 * The initiator attempted to send multiple tagged commands with
10705	 * the same ID.  (It's fine if different initiators have the same
10706	 * tag ID.)
10707	 *
10708	 * Even if all of those conditions are true, we don't kill the I/O
10709	 * if the command ahead of us has been aborted.  We won't end up
10710	 * sending it to the FETD, and it's perfectly legal to resend a
10711	 * command with the same tag number as long as the previous
10712	 * instance of this tag number has been aborted somehow.
10713	 */
10714	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10715	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10716	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10717	 && ((pending_io->io_hdr.nexus.targ_port ==
10718	      ooa_io->io_hdr.nexus.targ_port)
10719	  && (pending_io->io_hdr.nexus.initid.id ==
10720	      ooa_io->io_hdr.nexus.initid.id))
10721	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10722		return (CTL_ACTION_OVERLAP_TAG);
10723
10724	/*
10725	 * If we get a head of queue tag, SAM-3 says that we should
10726	 * immediately execute it.
10727	 *
10728	 * What happens if this command would normally block for some other
10729	 * reason?  e.g. a request sense with a head of queue tag
10730	 * immediately after a write.  Normally that would block, but this
10731	 * will result in its getting executed immediately...
10732	 *
10733	 * We currently return "pass" instead of "skip", so we'll end up
10734	 * going through the rest of the queue to check for overlapped tags.
10735	 *
10736	 * XXX KDM check for other types of blockage first??
10737	 */
10738	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10739		return (CTL_ACTION_PASS);
10740
10741	/*
10742	 * Ordered tags have to block until all items ahead of them
10743	 * have completed.  If we get called with an ordered tag, we always
10744	 * block, if something else is ahead of us in the queue.
10745	 */
10746	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10747		return (CTL_ACTION_BLOCK);
10748
10749	/*
10750	 * Simple tags get blocked until all head of queue and ordered tags
10751	 * ahead of them have completed.  I'm lumping untagged commands in
10752	 * with simple tags here.  XXX KDM is that the right thing to do?
10753	 */
10754	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10755	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10756	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10757	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10758		return (CTL_ACTION_BLOCK);
10759
10760	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10761	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10762
10763	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10764
10765	switch (serialize_row[pending_entry->seridx]) {
10766	case CTL_SER_BLOCK:
10767		return (CTL_ACTION_BLOCK);
10768	case CTL_SER_EXTENT:
10769		return (ctl_extent_check(pending_io, ooa_io));
10770	case CTL_SER_EXTENTOPT:
10771		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10772		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10773			return (ctl_extent_check(pending_io, ooa_io));
10774		/* FALLTHROUGH */
10775	case CTL_SER_PASS:
10776		return (CTL_ACTION_PASS);
10777	case CTL_SER_BLOCKOPT:
10778		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10779		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10780			return (CTL_ACTION_BLOCK);
10781		return (CTL_ACTION_PASS);
10782	case CTL_SER_SKIP:
10783		return (CTL_ACTION_SKIP);
10784	default:
10785		panic("invalid serialization value %d",
10786		      serialize_row[pending_entry->seridx]);
10787	}
10788
10789	return (CTL_ACTION_ERROR);
10790}
10791
10792/*
10793 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10794 * Assumptions:
10795 * - pending_io is generally either incoming, or on the blocked queue
10796 * - starting I/O is the I/O we want to start the check with.
10797 */
10798static ctl_action
10799ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10800	      union ctl_io *starting_io)
10801{
10802	union ctl_io *ooa_io;
10803	ctl_action action;
10804
10805	mtx_assert(&lun->lun_lock, MA_OWNED);
10806
10807	/*
10808	 * Run back along the OOA queue, starting with the current
10809	 * blocked I/O and going through every I/O before it on the
10810	 * queue.  If starting_io is NULL, we'll just end up returning
10811	 * CTL_ACTION_PASS.
10812	 */
10813	for (ooa_io = starting_io; ooa_io != NULL;
10814	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10815	     ooa_links)){
10816
10817		/*
10818		 * This routine just checks to see whether
10819		 * cur_blocked is blocked by ooa_io, which is ahead
10820		 * of it in the queue.  It doesn't queue/dequeue
10821		 * cur_blocked.
10822		 */
10823		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10824		switch (action) {
10825		case CTL_ACTION_BLOCK:
10826		case CTL_ACTION_OVERLAP:
10827		case CTL_ACTION_OVERLAP_TAG:
10828		case CTL_ACTION_SKIP:
10829		case CTL_ACTION_ERROR:
10830			return (action);
10831			break; /* NOTREACHED */
10832		case CTL_ACTION_PASS:
10833			break;
10834		default:
10835			panic("invalid action %d", action);
10836			break;  /* NOTREACHED */
10837		}
10838	}
10839
10840	return (CTL_ACTION_PASS);
10841}
10842
10843/*
10844 * Assumptions:
10845 * - An I/O has just completed, and has been removed from the per-LUN OOA
10846 *   queue, so some items on the blocked queue may now be unblocked.
10847 */
10848static int
10849ctl_check_blocked(struct ctl_lun *lun)
10850{
10851	union ctl_io *cur_blocked, *next_blocked;
10852
10853	mtx_assert(&lun->lun_lock, MA_OWNED);
10854
10855	/*
10856	 * Run forward from the head of the blocked queue, checking each
10857	 * entry against the I/Os prior to it on the OOA queue to see if
10858	 * there is still any blockage.
10859	 *
10860	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10861	 * with our removing a variable on it while it is traversing the
10862	 * list.
10863	 */
10864	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10865	     cur_blocked != NULL; cur_blocked = next_blocked) {
10866		union ctl_io *prev_ooa;
10867		ctl_action action;
10868
10869		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10870							  blocked_links);
10871
10872		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10873						      ctl_ooaq, ooa_links);
10874
10875		/*
10876		 * If cur_blocked happens to be the first item in the OOA
10877		 * queue now, prev_ooa will be NULL, and the action
10878		 * returned will just be CTL_ACTION_PASS.
10879		 */
10880		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10881
10882		switch (action) {
10883		case CTL_ACTION_BLOCK:
10884			/* Nothing to do here, still blocked */
10885			break;
10886		case CTL_ACTION_OVERLAP:
10887		case CTL_ACTION_OVERLAP_TAG:
10888			/*
10889			 * This shouldn't happen!  In theory we've already
10890			 * checked this command for overlap...
10891			 */
10892			break;
10893		case CTL_ACTION_PASS:
10894		case CTL_ACTION_SKIP: {
10895			struct ctl_softc *softc;
10896			const struct ctl_cmd_entry *entry;
10897			int isc_retval;
10898
10899			/*
10900			 * The skip case shouldn't happen, this transaction
10901			 * should have never made it onto the blocked queue.
10902			 */
10903			/*
10904			 * This I/O is no longer blocked, we can remove it
10905			 * from the blocked queue.  Since this is a TAILQ
10906			 * (doubly linked list), we can do O(1) removals
10907			 * from any place on the list.
10908			 */
10909			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10910				     blocked_links);
10911			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10912
10913			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10914				/*
10915				 * Need to send IO back to original side to
10916				 * run
10917				 */
10918				union ctl_ha_msg msg_info;
10919
10920				msg_info.hdr.original_sc =
10921					cur_blocked->io_hdr.original_sc;
10922				msg_info.hdr.serializing_sc = cur_blocked;
10923				msg_info.hdr.msg_type = CTL_MSG_R2R;
10924				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10925				     &msg_info, sizeof(msg_info), 0)) >
10926				     CTL_HA_STATUS_SUCCESS) {
10927					printf("CTL:Check Blocked error from "
10928					       "ctl_ha_msg_send %d\n",
10929					       isc_retval);
10930				}
10931				break;
10932			}
10933			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10934			softc = control_softc;
10935
10936			/*
10937			 * Check this I/O for LUN state changes that may
10938			 * have happened while this command was blocked.
10939			 * The LUN state may have been changed by a command
10940			 * ahead of us in the queue, so we need to re-check
10941			 * for any states that can be caused by SCSI
10942			 * commands.
10943			 */
10944			if (ctl_scsiio_lun_check(softc, lun, entry,
10945						 &cur_blocked->scsiio) == 0) {
10946				cur_blocked->io_hdr.flags |=
10947				                      CTL_FLAG_IS_WAS_ON_RTR;
10948				ctl_enqueue_rtr(cur_blocked);
10949			} else
10950				ctl_done(cur_blocked);
10951			break;
10952		}
10953		default:
10954			/*
10955			 * This probably shouldn't happen -- we shouldn't
10956			 * get CTL_ACTION_ERROR, or anything else.
10957			 */
10958			break;
10959		}
10960	}
10961
10962	return (CTL_RETVAL_COMPLETE);
10963}
10964
10965/*
10966 * This routine (with one exception) checks LUN flags that can be set by
10967 * commands ahead of us in the OOA queue.  These flags have to be checked
10968 * when a command initially comes in, and when we pull a command off the
10969 * blocked queue and are preparing to execute it.  The reason we have to
10970 * check these flags for commands on the blocked queue is that the LUN
10971 * state may have been changed by a command ahead of us while we're on the
10972 * blocked queue.
10973 *
10974 * Ordering is somewhat important with these checks, so please pay
10975 * careful attention to the placement of any new checks.
10976 */
10977static int
10978ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10979    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10980{
10981	int retval;
10982	uint32_t residx;
10983
10984	retval = 0;
10985
10986	mtx_assert(&lun->lun_lock, MA_OWNED);
10987
10988	/*
10989	 * If this shelf is a secondary shelf controller, we have to reject
10990	 * any media access commands.
10991	 */
10992	if ((ctl_softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
10993	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
10994		ctl_set_lun_standby(ctsio);
10995		retval = 1;
10996		goto bailout;
10997	}
10998
10999	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11000		if (lun->flags & CTL_LUN_READONLY) {
11001			ctl_set_sense(ctsio, /*current_error*/ 1,
11002			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11003			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11004			retval = 1;
11005			goto bailout;
11006		}
11007		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11008		    .eca_and_aen & SCP_SWP) != 0) {
11009			ctl_set_sense(ctsio, /*current_error*/ 1,
11010			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11011			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11012			retval = 1;
11013			goto bailout;
11014		}
11015	}
11016
11017	/*
11018	 * Check for a reservation conflict.  If this command isn't allowed
11019	 * even on reserved LUNs, and if this initiator isn't the one who
11020	 * reserved us, reject the command with a reservation conflict.
11021	 */
11022	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11023	if ((lun->flags & CTL_LUN_RESERVED)
11024	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11025		if (lun->res_idx != residx) {
11026			ctl_set_reservation_conflict(ctsio);
11027			retval = 1;
11028			goto bailout;
11029		}
11030	}
11031
11032	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11033	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11034		/* No reservation or command is allowed. */;
11035	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11036	    (lun->res_type == SPR_TYPE_WR_EX ||
11037	     lun->res_type == SPR_TYPE_WR_EX_RO ||
11038	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
11039		/* The command is allowed for Write Exclusive resv. */;
11040	} else {
11041		/*
11042		 * if we aren't registered or it's a res holder type
11043		 * reservation and this isn't the res holder then set a
11044		 * conflict.
11045		 */
11046		if (lun->pr_keys[residx] == 0
11047		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11048			ctl_set_reservation_conflict(ctsio);
11049			retval = 1;
11050			goto bailout;
11051		}
11052
11053	}
11054
11055	if ((lun->flags & CTL_LUN_OFFLINE)
11056	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11057		ctl_set_lun_not_ready(ctsio);
11058		retval = 1;
11059		goto bailout;
11060	}
11061
11062	/*
11063	 * If the LUN is stopped, see if this particular command is allowed
11064	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11065	 */
11066	if ((lun->flags & CTL_LUN_STOPPED)
11067	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11068		/* "Logical unit not ready, initializing cmd. required" */
11069		ctl_set_lun_stopped(ctsio);
11070		retval = 1;
11071		goto bailout;
11072	}
11073
11074	if ((lun->flags & CTL_LUN_INOPERABLE)
11075	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11076		/* "Medium format corrupted" */
11077		ctl_set_medium_format_corrupted(ctsio);
11078		retval = 1;
11079		goto bailout;
11080	}
11081
11082bailout:
11083	return (retval);
11084
11085}
11086
11087static void
11088ctl_failover_io(union ctl_io *io, int have_lock)
11089{
11090	ctl_set_busy(&io->scsiio);
11091	ctl_done(io);
11092}
11093
11094static void
11095ctl_failover(void)
11096{
11097	struct ctl_lun *lun;
11098	struct ctl_softc *ctl_softc;
11099	union ctl_io *next_io, *pending_io;
11100	union ctl_io *io;
11101	int lun_idx;
11102	int i;
11103
11104	ctl_softc = control_softc;
11105
11106	mtx_lock(&ctl_softc->ctl_lock);
11107	/*
11108	 * Remove any cmds from the other SC from the rtr queue.  These
11109	 * will obviously only be for LUNs for which we're the primary.
11110	 * We can't send status or get/send data for these commands.
11111	 * Since they haven't been executed yet, we can just remove them.
11112	 * We'll either abort them or delete them below, depending on
11113	 * which HA mode we're in.
11114	 */
11115#ifdef notyet
11116	mtx_lock(&ctl_softc->queue_lock);
11117	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11118	     io != NULL; io = next_io) {
11119		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11120		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11121			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11122				      ctl_io_hdr, links);
11123	}
11124	mtx_unlock(&ctl_softc->queue_lock);
11125#endif
11126
11127	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11128		lun = ctl_softc->ctl_luns[lun_idx];
11129		if (lun==NULL)
11130			continue;
11131
11132		/*
11133		 * Processor LUNs are primary on both sides.
11134		 * XXX will this always be true?
11135		 */
11136		if (lun->be_lun->lun_type == T_PROCESSOR)
11137			continue;
11138
11139		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11140		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11141			printf("FAILOVER: primary lun %d\n", lun_idx);
11142		        /*
11143			 * Remove all commands from the other SC. First from the
11144			 * blocked queue then from the ooa queue. Once we have
11145			 * removed them. Call ctl_check_blocked to see if there
11146			 * is anything that can run.
11147			 */
11148			for (io = (union ctl_io *)TAILQ_FIRST(
11149			     &lun->blocked_queue); io != NULL; io = next_io) {
11150
11151		        	next_io = (union ctl_io *)TAILQ_NEXT(
11152				    &io->io_hdr, blocked_links);
11153
11154				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11155					TAILQ_REMOVE(&lun->blocked_queue,
11156						     &io->io_hdr,blocked_links);
11157					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11158					TAILQ_REMOVE(&lun->ooa_queue,
11159						     &io->io_hdr, ooa_links);
11160
11161					ctl_free_io(io);
11162				}
11163			}
11164
11165			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11166	     		     io != NULL; io = next_io) {
11167
11168		        	next_io = (union ctl_io *)TAILQ_NEXT(
11169				    &io->io_hdr, ooa_links);
11170
11171				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11172
11173					TAILQ_REMOVE(&lun->ooa_queue,
11174						&io->io_hdr,
11175					     	ooa_links);
11176
11177					ctl_free_io(io);
11178				}
11179			}
11180			ctl_check_blocked(lun);
11181		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11182			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11183
11184			printf("FAILOVER: primary lun %d\n", lun_idx);
11185			/*
11186			 * Abort all commands from the other SC.  We can't
11187			 * send status back for them now.  These should get
11188			 * cleaned up when they are completed or come out
11189			 * for a datamove operation.
11190			 */
11191			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11192	     		     io != NULL; io = next_io) {
11193		        	next_io = (union ctl_io *)TAILQ_NEXT(
11194					&io->io_hdr, ooa_links);
11195
11196				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11197					io->io_hdr.flags |= CTL_FLAG_ABORT;
11198			}
11199		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11200			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11201
11202			printf("FAILOVER: secondary lun %d\n", lun_idx);
11203
11204			lun->flags |= CTL_LUN_PRIMARY_SC;
11205
11206			/*
11207			 * We send all I/O that was sent to this controller
11208			 * and redirected to the other side back with
11209			 * busy status, and have the initiator retry it.
11210			 * Figuring out how much data has been transferred,
11211			 * etc. and picking up where we left off would be
11212			 * very tricky.
11213			 *
11214			 * XXX KDM need to remove I/O from the blocked
11215			 * queue as well!
11216			 */
11217			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11218			     &lun->ooa_queue); pending_io != NULL;
11219			     pending_io = next_io) {
11220
11221				next_io =  (union ctl_io *)TAILQ_NEXT(
11222					&pending_io->io_hdr, ooa_links);
11223
11224				pending_io->io_hdr.flags &=
11225					~CTL_FLAG_SENT_2OTHER_SC;
11226
11227				if (pending_io->io_hdr.flags &
11228				    CTL_FLAG_IO_ACTIVE) {
11229					pending_io->io_hdr.flags |=
11230						CTL_FLAG_FAILOVER;
11231				} else {
11232					ctl_set_busy(&pending_io->scsiio);
11233					ctl_done(pending_io);
11234				}
11235			}
11236
11237			/*
11238			 * Build Unit Attention
11239			 */
11240			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11241				lun->pending_ua[i] |=
11242				                     CTL_UA_ASYM_ACC_CHANGE;
11243			}
11244		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11245			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11246			printf("FAILOVER: secondary lun %d\n", lun_idx);
11247			/*
11248			 * if the first io on the OOA is not on the RtR queue
11249			 * add it.
11250			 */
11251			lun->flags |= CTL_LUN_PRIMARY_SC;
11252
11253			pending_io = (union ctl_io *)TAILQ_FIRST(
11254			    &lun->ooa_queue);
11255			if (pending_io==NULL) {
11256				printf("Nothing on OOA queue\n");
11257				continue;
11258			}
11259
11260			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11261			if ((pending_io->io_hdr.flags &
11262			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11263				pending_io->io_hdr.flags |=
11264				    CTL_FLAG_IS_WAS_ON_RTR;
11265				ctl_enqueue_rtr(pending_io);
11266			}
11267#if 0
11268			else
11269			{
11270				printf("Tag 0x%04x is running\n",
11271				      pending_io->scsiio.tag_num);
11272			}
11273#endif
11274
11275			next_io = (union ctl_io *)TAILQ_NEXT(
11276			    &pending_io->io_hdr, ooa_links);
11277			for (pending_io=next_io; pending_io != NULL;
11278			     pending_io = next_io) {
11279				pending_io->io_hdr.flags &=
11280				    ~CTL_FLAG_SENT_2OTHER_SC;
11281				next_io = (union ctl_io *)TAILQ_NEXT(
11282					&pending_io->io_hdr, ooa_links);
11283				if (pending_io->io_hdr.flags &
11284				    CTL_FLAG_IS_WAS_ON_RTR) {
11285#if 0
11286				        printf("Tag 0x%04x is running\n",
11287				      		pending_io->scsiio.tag_num);
11288#endif
11289					continue;
11290				}
11291
11292				switch (ctl_check_ooa(lun, pending_io,
11293			            (union ctl_io *)TAILQ_PREV(
11294				    &pending_io->io_hdr, ctl_ooaq,
11295				    ooa_links))) {
11296
11297				case CTL_ACTION_BLOCK:
11298					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11299							  &pending_io->io_hdr,
11300							  blocked_links);
11301					pending_io->io_hdr.flags |=
11302					    CTL_FLAG_BLOCKED;
11303					break;
11304				case CTL_ACTION_PASS:
11305				case CTL_ACTION_SKIP:
11306					pending_io->io_hdr.flags |=
11307					    CTL_FLAG_IS_WAS_ON_RTR;
11308					ctl_enqueue_rtr(pending_io);
11309					break;
11310				case CTL_ACTION_OVERLAP:
11311					ctl_set_overlapped_cmd(
11312					    (struct ctl_scsiio *)pending_io);
11313					ctl_done(pending_io);
11314					break;
11315				case CTL_ACTION_OVERLAP_TAG:
11316					ctl_set_overlapped_tag(
11317					    (struct ctl_scsiio *)pending_io,
11318					    pending_io->scsiio.tag_num & 0xff);
11319					ctl_done(pending_io);
11320					break;
11321				case CTL_ACTION_ERROR:
11322				default:
11323					ctl_set_internal_failure(
11324						(struct ctl_scsiio *)pending_io,
11325						0,  // sks_valid
11326						0); //retry count
11327					ctl_done(pending_io);
11328					break;
11329				}
11330			}
11331
11332			/*
11333			 * Build Unit Attention
11334			 */
11335			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11336				lun->pending_ua[i] |=
11337				                     CTL_UA_ASYM_ACC_CHANGE;
11338			}
11339		} else {
11340			panic("Unhandled HA mode failover, LUN flags = %#x, "
11341			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11342		}
11343	}
11344	ctl_pause_rtr = 0;
11345	mtx_unlock(&ctl_softc->ctl_lock);
11346}
11347
11348static int
11349ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11350{
11351	struct ctl_lun *lun;
11352	const struct ctl_cmd_entry *entry;
11353	uint32_t initidx, targ_lun;
11354	int retval;
11355
11356	retval = 0;
11357
11358	lun = NULL;
11359
11360	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11361	if ((targ_lun < CTL_MAX_LUNS)
11362	 && ((lun = ctl_softc->ctl_luns[targ_lun]) != NULL)) {
11363		/*
11364		 * If the LUN is invalid, pretend that it doesn't exist.
11365		 * It will go away as soon as all pending I/O has been
11366		 * completed.
11367		 */
11368		mtx_lock(&lun->lun_lock);
11369		if (lun->flags & CTL_LUN_DISABLED) {
11370			mtx_unlock(&lun->lun_lock);
11371			lun = NULL;
11372			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11373			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11374		} else {
11375			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11376			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11377				lun->be_lun;
11378			if (lun->be_lun->lun_type == T_PROCESSOR) {
11379				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11380			}
11381
11382			/*
11383			 * Every I/O goes into the OOA queue for a
11384			 * particular LUN, and stays there until completion.
11385			 */
11386			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11387			    ooa_links);
11388		}
11389	} else {
11390		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11391		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11392	}
11393
11394	/* Get command entry and return error if it is unsuppotyed. */
11395	entry = ctl_validate_command(ctsio);
11396	if (entry == NULL) {
11397		if (lun)
11398			mtx_unlock(&lun->lun_lock);
11399		return (retval);
11400	}
11401
11402	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11403	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11404
11405	/*
11406	 * Check to see whether we can send this command to LUNs that don't
11407	 * exist.  This should pretty much only be the case for inquiry
11408	 * and request sense.  Further checks, below, really require having
11409	 * a LUN, so we can't really check the command anymore.  Just put
11410	 * it on the rtr queue.
11411	 */
11412	if (lun == NULL) {
11413		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11414			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11415			ctl_enqueue_rtr((union ctl_io *)ctsio);
11416			return (retval);
11417		}
11418
11419		ctl_set_unsupported_lun(ctsio);
11420		ctl_done((union ctl_io *)ctsio);
11421		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11422		return (retval);
11423	} else {
11424		/*
11425		 * Make sure we support this particular command on this LUN.
11426		 * e.g., we don't support writes to the control LUN.
11427		 */
11428		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11429			mtx_unlock(&lun->lun_lock);
11430			ctl_set_invalid_opcode(ctsio);
11431			ctl_done((union ctl_io *)ctsio);
11432			return (retval);
11433		}
11434	}
11435
11436	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11437
11438#ifdef CTL_WITH_CA
11439	/*
11440	 * If we've got a request sense, it'll clear the contingent
11441	 * allegiance condition.  Otherwise, if we have a CA condition for
11442	 * this initiator, clear it, because it sent down a command other
11443	 * than request sense.
11444	 */
11445	if ((ctsio->cdb[0] != REQUEST_SENSE)
11446	 && (ctl_is_set(lun->have_ca, initidx)))
11447		ctl_clear_mask(lun->have_ca, initidx);
11448#endif
11449
11450	/*
11451	 * If the command has this flag set, it handles its own unit
11452	 * attention reporting, we shouldn't do anything.  Otherwise we
11453	 * check for any pending unit attentions, and send them back to the
11454	 * initiator.  We only do this when a command initially comes in,
11455	 * not when we pull it off the blocked queue.
11456	 *
11457	 * According to SAM-3, section 5.3.2, the order that things get
11458	 * presented back to the host is basically unit attentions caused
11459	 * by some sort of reset event, busy status, reservation conflicts
11460	 * or task set full, and finally any other status.
11461	 *
11462	 * One issue here is that some of the unit attentions we report
11463	 * don't fall into the "reset" category (e.g. "reported luns data
11464	 * has changed").  So reporting it here, before the reservation
11465	 * check, may be technically wrong.  I guess the only thing to do
11466	 * would be to check for and report the reset events here, and then
11467	 * check for the other unit attention types after we check for a
11468	 * reservation conflict.
11469	 *
11470	 * XXX KDM need to fix this
11471	 */
11472	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11473		ctl_ua_type ua_type;
11474
11475		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11476			scsi_sense_data_type sense_format;
11477
11478			if (lun != NULL)
11479				sense_format = (lun->flags &
11480				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11481				    SSD_TYPE_FIXED;
11482			else
11483				sense_format = SSD_TYPE_FIXED;
11484
11485			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11486			    &ctsio->sense_data, sense_format);
11487			if (ua_type != CTL_UA_NONE) {
11488				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11489				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11490						       CTL_AUTOSENSE;
11491				ctsio->sense_len = SSD_FULL_SIZE;
11492				mtx_unlock(&lun->lun_lock);
11493				ctl_done((union ctl_io *)ctsio);
11494				return (retval);
11495			}
11496		}
11497	}
11498
11499
11500	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11501		mtx_unlock(&lun->lun_lock);
11502		ctl_done((union ctl_io *)ctsio);
11503		return (retval);
11504	}
11505
11506	/*
11507	 * XXX CHD this is where we want to send IO to other side if
11508	 * this LUN is secondary on this SC. We will need to make a copy
11509	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11510	 * the copy we send as FROM_OTHER.
11511	 * We also need to stuff the address of the original IO so we can
11512	 * find it easily. Something similar will need be done on the other
11513	 * side so when we are done we can find the copy.
11514	 */
11515	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11516		union ctl_ha_msg msg_info;
11517		int isc_retval;
11518
11519		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11520
11521		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11522		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11523#if 0
11524		printf("1. ctsio %p\n", ctsio);
11525#endif
11526		msg_info.hdr.serializing_sc = NULL;
11527		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11528		msg_info.scsi.tag_num = ctsio->tag_num;
11529		msg_info.scsi.tag_type = ctsio->tag_type;
11530		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11531
11532		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11533
11534		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11535		    (void *)&msg_info, sizeof(msg_info), 0)) >
11536		    CTL_HA_STATUS_SUCCESS) {
11537			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11538			       isc_retval);
11539			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11540		} else {
11541#if 0
11542			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11543#endif
11544		}
11545
11546		/*
11547		 * XXX KDM this I/O is off the incoming queue, but hasn't
11548		 * been inserted on any other queue.  We may need to come
11549		 * up with a holding queue while we wait for serialization
11550		 * so that we have an idea of what we're waiting for from
11551		 * the other side.
11552		 */
11553		mtx_unlock(&lun->lun_lock);
11554		return (retval);
11555	}
11556
11557	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11558			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11559			      ctl_ooaq, ooa_links))) {
11560	case CTL_ACTION_BLOCK:
11561		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11562		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11563				  blocked_links);
11564		mtx_unlock(&lun->lun_lock);
11565		return (retval);
11566	case CTL_ACTION_PASS:
11567	case CTL_ACTION_SKIP:
11568		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11569		mtx_unlock(&lun->lun_lock);
11570		ctl_enqueue_rtr((union ctl_io *)ctsio);
11571		break;
11572	case CTL_ACTION_OVERLAP:
11573		mtx_unlock(&lun->lun_lock);
11574		ctl_set_overlapped_cmd(ctsio);
11575		ctl_done((union ctl_io *)ctsio);
11576		break;
11577	case CTL_ACTION_OVERLAP_TAG:
11578		mtx_unlock(&lun->lun_lock);
11579		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11580		ctl_done((union ctl_io *)ctsio);
11581		break;
11582	case CTL_ACTION_ERROR:
11583	default:
11584		mtx_unlock(&lun->lun_lock);
11585		ctl_set_internal_failure(ctsio,
11586					 /*sks_valid*/ 0,
11587					 /*retry_count*/ 0);
11588		ctl_done((union ctl_io *)ctsio);
11589		break;
11590	}
11591	return (retval);
11592}
11593
11594const struct ctl_cmd_entry *
11595ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11596{
11597	const struct ctl_cmd_entry *entry;
11598	int service_action;
11599
11600	entry = &ctl_cmd_table[ctsio->cdb[0]];
11601	if (sa)
11602		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11603	if (entry->flags & CTL_CMD_FLAG_SA5) {
11604		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11605		entry = &((const struct ctl_cmd_entry *)
11606		    entry->execute)[service_action];
11607	}
11608	return (entry);
11609}
11610
11611const struct ctl_cmd_entry *
11612ctl_validate_command(struct ctl_scsiio *ctsio)
11613{
11614	const struct ctl_cmd_entry *entry;
11615	int i, sa;
11616	uint8_t diff;
11617
11618	entry = ctl_get_cmd_entry(ctsio, &sa);
11619	if (entry->execute == NULL) {
11620		if (sa)
11621			ctl_set_invalid_field(ctsio,
11622					      /*sks_valid*/ 1,
11623					      /*command*/ 1,
11624					      /*field*/ 1,
11625					      /*bit_valid*/ 1,
11626					      /*bit*/ 4);
11627		else
11628			ctl_set_invalid_opcode(ctsio);
11629		ctl_done((union ctl_io *)ctsio);
11630		return (NULL);
11631	}
11632	KASSERT(entry->length > 0,
11633	    ("Not defined length for command 0x%02x/0x%02x",
11634	     ctsio->cdb[0], ctsio->cdb[1]));
11635	for (i = 1; i < entry->length; i++) {
11636		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11637		if (diff == 0)
11638			continue;
11639		ctl_set_invalid_field(ctsio,
11640				      /*sks_valid*/ 1,
11641				      /*command*/ 1,
11642				      /*field*/ i,
11643				      /*bit_valid*/ 1,
11644				      /*bit*/ fls(diff) - 1);
11645		ctl_done((union ctl_io *)ctsio);
11646		return (NULL);
11647	}
11648	return (entry);
11649}
11650
11651static int
11652ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11653{
11654
11655	switch (lun_type) {
11656	case T_PROCESSOR:
11657		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11658		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11659			return (0);
11660		break;
11661	case T_DIRECT:
11662		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11663		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11664			return (0);
11665		break;
11666	default:
11667		return (0);
11668	}
11669	return (1);
11670}
11671
11672static int
11673ctl_scsiio(struct ctl_scsiio *ctsio)
11674{
11675	int retval;
11676	const struct ctl_cmd_entry *entry;
11677
11678	retval = CTL_RETVAL_COMPLETE;
11679
11680	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11681
11682	entry = ctl_get_cmd_entry(ctsio, NULL);
11683
11684	/*
11685	 * If this I/O has been aborted, just send it straight to
11686	 * ctl_done() without executing it.
11687	 */
11688	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11689		ctl_done((union ctl_io *)ctsio);
11690		goto bailout;
11691	}
11692
11693	/*
11694	 * All the checks should have been handled by ctl_scsiio_precheck().
11695	 * We should be clear now to just execute the I/O.
11696	 */
11697	retval = entry->execute(ctsio);
11698
11699bailout:
11700	return (retval);
11701}
11702
11703/*
11704 * Since we only implement one target right now, a bus reset simply resets
11705 * our single target.
11706 */
11707static int
11708ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11709{
11710	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11711}
11712
11713static int
11714ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11715		 ctl_ua_type ua_type)
11716{
11717	struct ctl_lun *lun;
11718	int retval;
11719
11720	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11721		union ctl_ha_msg msg_info;
11722
11723		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11724		msg_info.hdr.nexus = io->io_hdr.nexus;
11725		if (ua_type==CTL_UA_TARG_RESET)
11726			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11727		else
11728			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11729		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11730		msg_info.hdr.original_sc = NULL;
11731		msg_info.hdr.serializing_sc = NULL;
11732		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11733		    (void *)&msg_info, sizeof(msg_info), 0)) {
11734		}
11735	}
11736	retval = 0;
11737
11738	mtx_lock(&ctl_softc->ctl_lock);
11739	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11740		retval += ctl_lun_reset(lun, io, ua_type);
11741	mtx_unlock(&ctl_softc->ctl_lock);
11742
11743	return (retval);
11744}
11745
11746/*
11747 * The LUN should always be set.  The I/O is optional, and is used to
11748 * distinguish between I/Os sent by this initiator, and by other
11749 * initiators.  We set unit attention for initiators other than this one.
11750 * SAM-3 is vague on this point.  It does say that a unit attention should
11751 * be established for other initiators when a LUN is reset (see section
11752 * 5.7.3), but it doesn't specifically say that the unit attention should
11753 * be established for this particular initiator when a LUN is reset.  Here
11754 * is the relevant text, from SAM-3 rev 8:
11755 *
11756 * 5.7.2 When a SCSI initiator port aborts its own tasks
11757 *
11758 * When a SCSI initiator port causes its own task(s) to be aborted, no
11759 * notification that the task(s) have been aborted shall be returned to
11760 * the SCSI initiator port other than the completion response for the
11761 * command or task management function action that caused the task(s) to
11762 * be aborted and notification(s) associated with related effects of the
11763 * action (e.g., a reset unit attention condition).
11764 *
11765 * XXX KDM for now, we're setting unit attention for all initiators.
11766 */
11767static int
11768ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11769{
11770	union ctl_io *xio;
11771#if 0
11772	uint32_t initidx;
11773#endif
11774	int i;
11775
11776	mtx_lock(&lun->lun_lock);
11777	/*
11778	 * Run through the OOA queue and abort each I/O.
11779	 */
11780#if 0
11781	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11782#endif
11783	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11784	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11785		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11786	}
11787
11788	/*
11789	 * This version sets unit attention for every
11790	 */
11791#if 0
11792	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11793	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11794		if (initidx == i)
11795			continue;
11796		lun->pending_ua[i] |= ua_type;
11797	}
11798#endif
11799
11800	/*
11801	 * A reset (any kind, really) clears reservations established with
11802	 * RESERVE/RELEASE.  It does not clear reservations established
11803	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11804	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11805	 * reservations made with the RESERVE/RELEASE commands, because
11806	 * those commands are obsolete in SPC-3.
11807	 */
11808	lun->flags &= ~CTL_LUN_RESERVED;
11809
11810	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11811#ifdef CTL_WITH_CA
11812		ctl_clear_mask(lun->have_ca, i);
11813#endif
11814		lun->pending_ua[i] |= ua_type;
11815	}
11816	mtx_unlock(&lun->lun_lock);
11817
11818	return (0);
11819}
11820
11821static void
11822ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11823    int other_sc)
11824{
11825	union ctl_io *xio;
11826
11827	mtx_assert(&lun->lun_lock, MA_OWNED);
11828
11829	/*
11830	 * Run through the OOA queue and attempt to find the given I/O.
11831	 * The target port, initiator ID, tag type and tag number have to
11832	 * match the values that we got from the initiator.  If we have an
11833	 * untagged command to abort, simply abort the first untagged command
11834	 * we come to.  We only allow one untagged command at a time of course.
11835	 */
11836	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11837	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11838
11839		if ((targ_port == UINT32_MAX ||
11840		     targ_port == xio->io_hdr.nexus.targ_port) &&
11841		    (init_id == UINT32_MAX ||
11842		     init_id == xio->io_hdr.nexus.initid.id)) {
11843			if (targ_port != xio->io_hdr.nexus.targ_port ||
11844			    init_id != xio->io_hdr.nexus.initid.id)
11845				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11846			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11847			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11848				union ctl_ha_msg msg_info;
11849
11850				msg_info.hdr.nexus = xio->io_hdr.nexus;
11851				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11852				msg_info.task.tag_num = xio->scsiio.tag_num;
11853				msg_info.task.tag_type = xio->scsiio.tag_type;
11854				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11855				msg_info.hdr.original_sc = NULL;
11856				msg_info.hdr.serializing_sc = NULL;
11857				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11858				    (void *)&msg_info, sizeof(msg_info), 0);
11859			}
11860		}
11861	}
11862}
11863
11864static int
11865ctl_abort_task_set(union ctl_io *io)
11866{
11867	struct ctl_softc *softc = control_softc;
11868	struct ctl_lun *lun;
11869	uint32_t targ_lun;
11870
11871	/*
11872	 * Look up the LUN.
11873	 */
11874	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11875	mtx_lock(&softc->ctl_lock);
11876	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11877		lun = softc->ctl_luns[targ_lun];
11878	else {
11879		mtx_unlock(&softc->ctl_lock);
11880		return (1);
11881	}
11882
11883	mtx_lock(&lun->lun_lock);
11884	mtx_unlock(&softc->ctl_lock);
11885	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11886		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11887		    io->io_hdr.nexus.initid.id,
11888		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11889	} else { /* CTL_TASK_CLEAR_TASK_SET */
11890		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11891		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11892	}
11893	mtx_unlock(&lun->lun_lock);
11894	return (0);
11895}
11896
11897static int
11898ctl_i_t_nexus_reset(union ctl_io *io)
11899{
11900	struct ctl_softc *softc = control_softc;
11901	struct ctl_lun *lun;
11902	uint32_t initidx, residx;
11903
11904	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11905	residx = ctl_get_resindex(&io->io_hdr.nexus);
11906	mtx_lock(&softc->ctl_lock);
11907	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11908		mtx_lock(&lun->lun_lock);
11909		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11910		    io->io_hdr.nexus.initid.id,
11911		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11912#ifdef CTL_WITH_CA
11913		ctl_clear_mask(lun->have_ca, initidx);
11914#endif
11915		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
11916			lun->flags &= ~CTL_LUN_RESERVED;
11917		lun->pending_ua[initidx] |= CTL_UA_I_T_NEXUS_LOSS;
11918		mtx_unlock(&lun->lun_lock);
11919	}
11920	mtx_unlock(&softc->ctl_lock);
11921	return (0);
11922}
11923
11924static int
11925ctl_abort_task(union ctl_io *io)
11926{
11927	union ctl_io *xio;
11928	struct ctl_lun *lun;
11929	struct ctl_softc *ctl_softc;
11930#if 0
11931	struct sbuf sb;
11932	char printbuf[128];
11933#endif
11934	int found;
11935	uint32_t targ_lun;
11936
11937	ctl_softc = control_softc;
11938	found = 0;
11939
11940	/*
11941	 * Look up the LUN.
11942	 */
11943	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11944	mtx_lock(&ctl_softc->ctl_lock);
11945	if ((targ_lun < CTL_MAX_LUNS)
11946	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11947		lun = ctl_softc->ctl_luns[targ_lun];
11948	else {
11949		mtx_unlock(&ctl_softc->ctl_lock);
11950		return (1);
11951	}
11952
11953#if 0
11954	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11955	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11956#endif
11957
11958	mtx_lock(&lun->lun_lock);
11959	mtx_unlock(&ctl_softc->ctl_lock);
11960	/*
11961	 * Run through the OOA queue and attempt to find the given I/O.
11962	 * The target port, initiator ID, tag type and tag number have to
11963	 * match the values that we got from the initiator.  If we have an
11964	 * untagged command to abort, simply abort the first untagged command
11965	 * we come to.  We only allow one untagged command at a time of course.
11966	 */
11967#if 0
11968	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11969#endif
11970	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11971	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11972#if 0
11973		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11974
11975		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11976			    lun->lun, xio->scsiio.tag_num,
11977			    xio->scsiio.tag_type,
11978			    (xio->io_hdr.blocked_links.tqe_prev
11979			    == NULL) ? "" : " BLOCKED",
11980			    (xio->io_hdr.flags &
11981			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11982			    (xio->io_hdr.flags &
11983			    CTL_FLAG_ABORT) ? " ABORT" : "",
11984			    (xio->io_hdr.flags &
11985			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11986		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11987		sbuf_finish(&sb);
11988		printf("%s\n", sbuf_data(&sb));
11989#endif
11990
11991		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11992		 && (xio->io_hdr.nexus.initid.id ==
11993		     io->io_hdr.nexus.initid.id)) {
11994			/*
11995			 * If the abort says that the task is untagged, the
11996			 * task in the queue must be untagged.  Otherwise,
11997			 * we just check to see whether the tag numbers
11998			 * match.  This is because the QLogic firmware
11999			 * doesn't pass back the tag type in an abort
12000			 * request.
12001			 */
12002#if 0
12003			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12004			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12005			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12006#endif
12007			/*
12008			 * XXX KDM we've got problems with FC, because it
12009			 * doesn't send down a tag type with aborts.  So we
12010			 * can only really go by the tag number...
12011			 * This may cause problems with parallel SCSI.
12012			 * Need to figure that out!!
12013			 */
12014			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12015				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12016				found = 1;
12017				if ((io->io_hdr.flags &
12018				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12019				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12020					union ctl_ha_msg msg_info;
12021
12022					io->io_hdr.flags |=
12023					                CTL_FLAG_SENT_2OTHER_SC;
12024					msg_info.hdr.nexus = io->io_hdr.nexus;
12025					msg_info.task.task_action =
12026						CTL_TASK_ABORT_TASK;
12027					msg_info.task.tag_num =
12028						io->taskio.tag_num;
12029					msg_info.task.tag_type =
12030						io->taskio.tag_type;
12031					msg_info.hdr.msg_type =
12032						CTL_MSG_MANAGE_TASKS;
12033					msg_info.hdr.original_sc = NULL;
12034					msg_info.hdr.serializing_sc = NULL;
12035#if 0
12036					printf("Sent Abort to other side\n");
12037#endif
12038					if (CTL_HA_STATUS_SUCCESS !=
12039					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12040		    				(void *)&msg_info,
12041						sizeof(msg_info), 0)) {
12042					}
12043				}
12044#if 0
12045				printf("ctl_abort_task: found I/O to abort\n");
12046#endif
12047				break;
12048			}
12049		}
12050	}
12051	mtx_unlock(&lun->lun_lock);
12052
12053	if (found == 0) {
12054		/*
12055		 * This isn't really an error.  It's entirely possible for
12056		 * the abort and command completion to cross on the wire.
12057		 * This is more of an informative/diagnostic error.
12058		 */
12059#if 0
12060		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12061		       "%d:%d:%d:%d tag %d type %d\n",
12062		       io->io_hdr.nexus.initid.id,
12063		       io->io_hdr.nexus.targ_port,
12064		       io->io_hdr.nexus.targ_target.id,
12065		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12066		       io->taskio.tag_type);
12067#endif
12068	}
12069	return (0);
12070}
12071
12072static void
12073ctl_run_task(union ctl_io *io)
12074{
12075	struct ctl_softc *ctl_softc = control_softc;
12076	int retval = 1;
12077	const char *task_desc;
12078
12079	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12080
12081	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12082	    ("ctl_run_task: Unextected io_type %d\n",
12083	     io->io_hdr.io_type));
12084
12085	task_desc = ctl_scsi_task_string(&io->taskio);
12086	if (task_desc != NULL) {
12087#ifdef NEEDTOPORT
12088		csevent_log(CSC_CTL | CSC_SHELF_SW |
12089			    CTL_TASK_REPORT,
12090			    csevent_LogType_Trace,
12091			    csevent_Severity_Information,
12092			    csevent_AlertLevel_Green,
12093			    csevent_FRU_Firmware,
12094			    csevent_FRU_Unknown,
12095			    "CTL: received task: %s",task_desc);
12096#endif
12097	} else {
12098#ifdef NEEDTOPORT
12099		csevent_log(CSC_CTL | CSC_SHELF_SW |
12100			    CTL_TASK_REPORT,
12101			    csevent_LogType_Trace,
12102			    csevent_Severity_Information,
12103			    csevent_AlertLevel_Green,
12104			    csevent_FRU_Firmware,
12105			    csevent_FRU_Unknown,
12106			    "CTL: received unknown task "
12107			    "type: %d (%#x)",
12108			    io->taskio.task_action,
12109			    io->taskio.task_action);
12110#endif
12111	}
12112	switch (io->taskio.task_action) {
12113	case CTL_TASK_ABORT_TASK:
12114		retval = ctl_abort_task(io);
12115		break;
12116	case CTL_TASK_ABORT_TASK_SET:
12117	case CTL_TASK_CLEAR_TASK_SET:
12118		retval = ctl_abort_task_set(io);
12119		break;
12120	case CTL_TASK_CLEAR_ACA:
12121		break;
12122	case CTL_TASK_I_T_NEXUS_RESET:
12123		retval = ctl_i_t_nexus_reset(io);
12124		break;
12125	case CTL_TASK_LUN_RESET: {
12126		struct ctl_lun *lun;
12127		uint32_t targ_lun;
12128
12129		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12130		mtx_lock(&ctl_softc->ctl_lock);
12131		if ((targ_lun < CTL_MAX_LUNS)
12132		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12133			lun = ctl_softc->ctl_luns[targ_lun];
12134		else {
12135			mtx_unlock(&ctl_softc->ctl_lock);
12136			retval = 1;
12137			break;
12138		}
12139
12140		if (!(io->io_hdr.flags &
12141		    CTL_FLAG_FROM_OTHER_SC)) {
12142			union ctl_ha_msg msg_info;
12143
12144			io->io_hdr.flags |=
12145				CTL_FLAG_SENT_2OTHER_SC;
12146			msg_info.hdr.msg_type =
12147				CTL_MSG_MANAGE_TASKS;
12148			msg_info.hdr.nexus = io->io_hdr.nexus;
12149			msg_info.task.task_action =
12150				CTL_TASK_LUN_RESET;
12151			msg_info.hdr.original_sc = NULL;
12152			msg_info.hdr.serializing_sc = NULL;
12153			if (CTL_HA_STATUS_SUCCESS !=
12154			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12155			    (void *)&msg_info,
12156			    sizeof(msg_info), 0)) {
12157			}
12158		}
12159
12160		retval = ctl_lun_reset(lun, io,
12161				       CTL_UA_LUN_RESET);
12162		mtx_unlock(&ctl_softc->ctl_lock);
12163		break;
12164	}
12165	case CTL_TASK_TARGET_RESET:
12166		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12167		break;
12168	case CTL_TASK_BUS_RESET:
12169		retval = ctl_bus_reset(ctl_softc, io);
12170		break;
12171	case CTL_TASK_PORT_LOGIN:
12172		break;
12173	case CTL_TASK_PORT_LOGOUT:
12174		break;
12175	default:
12176		printf("ctl_run_task: got unknown task management event %d\n",
12177		       io->taskio.task_action);
12178		break;
12179	}
12180	if (retval == 0)
12181		io->io_hdr.status = CTL_SUCCESS;
12182	else
12183		io->io_hdr.status = CTL_ERROR;
12184	ctl_done(io);
12185}
12186
12187/*
12188 * For HA operation.  Handle commands that come in from the other
12189 * controller.
12190 */
12191static void
12192ctl_handle_isc(union ctl_io *io)
12193{
12194	int free_io;
12195	struct ctl_lun *lun;
12196	struct ctl_softc *ctl_softc;
12197	uint32_t targ_lun;
12198
12199	ctl_softc = control_softc;
12200
12201	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12202	lun = ctl_softc->ctl_luns[targ_lun];
12203
12204	switch (io->io_hdr.msg_type) {
12205	case CTL_MSG_SERIALIZE:
12206		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12207		break;
12208	case CTL_MSG_R2R: {
12209		const struct ctl_cmd_entry *entry;
12210
12211		/*
12212		 * This is only used in SER_ONLY mode.
12213		 */
12214		free_io = 0;
12215		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12216		mtx_lock(&lun->lun_lock);
12217		if (ctl_scsiio_lun_check(ctl_softc, lun,
12218		    entry, (struct ctl_scsiio *)io) != 0) {
12219			mtx_unlock(&lun->lun_lock);
12220			ctl_done(io);
12221			break;
12222		}
12223		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12224		mtx_unlock(&lun->lun_lock);
12225		ctl_enqueue_rtr(io);
12226		break;
12227	}
12228	case CTL_MSG_FINISH_IO:
12229		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12230			free_io = 0;
12231			ctl_done(io);
12232		} else {
12233			free_io = 1;
12234			mtx_lock(&lun->lun_lock);
12235			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12236				     ooa_links);
12237			ctl_check_blocked(lun);
12238			mtx_unlock(&lun->lun_lock);
12239		}
12240		break;
12241	case CTL_MSG_PERS_ACTION:
12242		ctl_hndl_per_res_out_on_other_sc(
12243			(union ctl_ha_msg *)&io->presio.pr_msg);
12244		free_io = 1;
12245		break;
12246	case CTL_MSG_BAD_JUJU:
12247		free_io = 0;
12248		ctl_done(io);
12249		break;
12250	case CTL_MSG_DATAMOVE:
12251		/* Only used in XFER mode */
12252		free_io = 0;
12253		ctl_datamove_remote(io);
12254		break;
12255	case CTL_MSG_DATAMOVE_DONE:
12256		/* Only used in XFER mode */
12257		free_io = 0;
12258		io->scsiio.be_move_done(io);
12259		break;
12260	default:
12261		free_io = 1;
12262		printf("%s: Invalid message type %d\n",
12263		       __func__, io->io_hdr.msg_type);
12264		break;
12265	}
12266	if (free_io)
12267		ctl_free_io(io);
12268
12269}
12270
12271
12272/*
12273 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12274 * there is no match.
12275 */
12276static ctl_lun_error_pattern
12277ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12278{
12279	const struct ctl_cmd_entry *entry;
12280	ctl_lun_error_pattern filtered_pattern, pattern;
12281
12282	pattern = desc->error_pattern;
12283
12284	/*
12285	 * XXX KDM we need more data passed into this function to match a
12286	 * custom pattern, and we actually need to implement custom pattern
12287	 * matching.
12288	 */
12289	if (pattern & CTL_LUN_PAT_CMD)
12290		return (CTL_LUN_PAT_CMD);
12291
12292	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12293		return (CTL_LUN_PAT_ANY);
12294
12295	entry = ctl_get_cmd_entry(ctsio, NULL);
12296
12297	filtered_pattern = entry->pattern & pattern;
12298
12299	/*
12300	 * If the user requested specific flags in the pattern (e.g.
12301	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12302	 * flags.
12303	 *
12304	 * If the user did not specify any flags, it doesn't matter whether
12305	 * or not the command supports the flags.
12306	 */
12307	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12308	     (pattern & ~CTL_LUN_PAT_MASK))
12309		return (CTL_LUN_PAT_NONE);
12310
12311	/*
12312	 * If the user asked for a range check, see if the requested LBA
12313	 * range overlaps with this command's LBA range.
12314	 */
12315	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12316		uint64_t lba1;
12317		uint64_t len1;
12318		ctl_action action;
12319		int retval;
12320
12321		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12322		if (retval != 0)
12323			return (CTL_LUN_PAT_NONE);
12324
12325		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12326					      desc->lba_range.len);
12327		/*
12328		 * A "pass" means that the LBA ranges don't overlap, so
12329		 * this doesn't match the user's range criteria.
12330		 */
12331		if (action == CTL_ACTION_PASS)
12332			return (CTL_LUN_PAT_NONE);
12333	}
12334
12335	return (filtered_pattern);
12336}
12337
12338static void
12339ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12340{
12341	struct ctl_error_desc *desc, *desc2;
12342
12343	mtx_assert(&lun->lun_lock, MA_OWNED);
12344
12345	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12346		ctl_lun_error_pattern pattern;
12347		/*
12348		 * Check to see whether this particular command matches
12349		 * the pattern in the descriptor.
12350		 */
12351		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12352		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12353			continue;
12354
12355		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12356		case CTL_LUN_INJ_ABORTED:
12357			ctl_set_aborted(&io->scsiio);
12358			break;
12359		case CTL_LUN_INJ_MEDIUM_ERR:
12360			ctl_set_medium_error(&io->scsiio);
12361			break;
12362		case CTL_LUN_INJ_UA:
12363			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12364			 * OCCURRED */
12365			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12366			break;
12367		case CTL_LUN_INJ_CUSTOM:
12368			/*
12369			 * We're assuming the user knows what he is doing.
12370			 * Just copy the sense information without doing
12371			 * checks.
12372			 */
12373			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12374			      ctl_min(sizeof(desc->custom_sense),
12375				      sizeof(io->scsiio.sense_data)));
12376			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12377			io->scsiio.sense_len = SSD_FULL_SIZE;
12378			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12379			break;
12380		case CTL_LUN_INJ_NONE:
12381		default:
12382			/*
12383			 * If this is an error injection type we don't know
12384			 * about, clear the continuous flag (if it is set)
12385			 * so it will get deleted below.
12386			 */
12387			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12388			break;
12389		}
12390		/*
12391		 * By default, each error injection action is a one-shot
12392		 */
12393		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12394			continue;
12395
12396		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12397
12398		free(desc, M_CTL);
12399	}
12400}
12401
12402#ifdef CTL_IO_DELAY
12403static void
12404ctl_datamove_timer_wakeup(void *arg)
12405{
12406	union ctl_io *io;
12407
12408	io = (union ctl_io *)arg;
12409
12410	ctl_datamove(io);
12411}
12412#endif /* CTL_IO_DELAY */
12413
12414void
12415ctl_datamove(union ctl_io *io)
12416{
12417	void (*fe_datamove)(union ctl_io *io);
12418
12419	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12420
12421	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12422
12423#ifdef CTL_TIME_IO
12424	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12425		char str[256];
12426		char path_str[64];
12427		struct sbuf sb;
12428
12429		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12430		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12431
12432		sbuf_cat(&sb, path_str);
12433		switch (io->io_hdr.io_type) {
12434		case CTL_IO_SCSI:
12435			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12436			sbuf_printf(&sb, "\n");
12437			sbuf_cat(&sb, path_str);
12438			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12439				    io->scsiio.tag_num, io->scsiio.tag_type);
12440			break;
12441		case CTL_IO_TASK:
12442			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12443				    "Tag Type: %d\n", io->taskio.task_action,
12444				    io->taskio.tag_num, io->taskio.tag_type);
12445			break;
12446		default:
12447			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12448			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12449			break;
12450		}
12451		sbuf_cat(&sb, path_str);
12452		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12453			    (intmax_t)time_uptime - io->io_hdr.start_time);
12454		sbuf_finish(&sb);
12455		printf("%s", sbuf_data(&sb));
12456	}
12457#endif /* CTL_TIME_IO */
12458
12459#ifdef CTL_IO_DELAY
12460	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12461		struct ctl_lun *lun;
12462
12463		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12464
12465		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12466	} else {
12467		struct ctl_lun *lun;
12468
12469		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12470		if ((lun != NULL)
12471		 && (lun->delay_info.datamove_delay > 0)) {
12472			struct callout *callout;
12473
12474			callout = (struct callout *)&io->io_hdr.timer_bytes;
12475			callout_init(callout, /*mpsafe*/ 1);
12476			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12477			callout_reset(callout,
12478				      lun->delay_info.datamove_delay * hz,
12479				      ctl_datamove_timer_wakeup, io);
12480			if (lun->delay_info.datamove_type ==
12481			    CTL_DELAY_TYPE_ONESHOT)
12482				lun->delay_info.datamove_delay = 0;
12483			return;
12484		}
12485	}
12486#endif
12487
12488	/*
12489	 * This command has been aborted.  Set the port status, so we fail
12490	 * the data move.
12491	 */
12492	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12493		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12494		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12495		       io->io_hdr.nexus.targ_port,
12496		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12497		       io->io_hdr.nexus.targ_lun);
12498		io->io_hdr.port_status = 31337;
12499		/*
12500		 * Note that the backend, in this case, will get the
12501		 * callback in its context.  In other cases it may get
12502		 * called in the frontend's interrupt thread context.
12503		 */
12504		io->scsiio.be_move_done(io);
12505		return;
12506	}
12507
12508	/* Don't confuse frontend with zero length data move. */
12509	if (io->scsiio.kern_data_len == 0) {
12510		io->scsiio.be_move_done(io);
12511		return;
12512	}
12513
12514	/*
12515	 * If we're in XFER mode and this I/O is from the other shelf
12516	 * controller, we need to send the DMA to the other side to
12517	 * actually transfer the data to/from the host.  In serialize only
12518	 * mode the transfer happens below CTL and ctl_datamove() is only
12519	 * called on the machine that originally received the I/O.
12520	 */
12521	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12522	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12523		union ctl_ha_msg msg;
12524		uint32_t sg_entries_sent;
12525		int do_sg_copy;
12526		int i;
12527
12528		memset(&msg, 0, sizeof(msg));
12529		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12530		msg.hdr.original_sc = io->io_hdr.original_sc;
12531		msg.hdr.serializing_sc = io;
12532		msg.hdr.nexus = io->io_hdr.nexus;
12533		msg.dt.flags = io->io_hdr.flags;
12534		/*
12535		 * We convert everything into a S/G list here.  We can't
12536		 * pass by reference, only by value between controllers.
12537		 * So we can't pass a pointer to the S/G list, only as many
12538		 * S/G entries as we can fit in here.  If it's possible for
12539		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12540		 * then we need to break this up into multiple transfers.
12541		 */
12542		if (io->scsiio.kern_sg_entries == 0) {
12543			msg.dt.kern_sg_entries = 1;
12544			/*
12545			 * If this is in cached memory, flush the cache
12546			 * before we send the DMA request to the other
12547			 * controller.  We want to do this in either the
12548			 * read or the write case.  The read case is
12549			 * straightforward.  In the write case, we want to
12550			 * make sure nothing is in the local cache that
12551			 * could overwrite the DMAed data.
12552			 */
12553			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12554				/*
12555				 * XXX KDM use bus_dmamap_sync() here.
12556				 */
12557			}
12558
12559			/*
12560			 * Convert to a physical address if this is a
12561			 * virtual address.
12562			 */
12563			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12564				msg.dt.sg_list[0].addr =
12565					io->scsiio.kern_data_ptr;
12566			} else {
12567				/*
12568				 * XXX KDM use busdma here!
12569				 */
12570#if 0
12571				msg.dt.sg_list[0].addr = (void *)
12572					vtophys(io->scsiio.kern_data_ptr);
12573#endif
12574			}
12575
12576			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12577			do_sg_copy = 0;
12578		} else {
12579			struct ctl_sg_entry *sgl;
12580
12581			do_sg_copy = 1;
12582			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12583			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12584			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12585				/*
12586				 * XXX KDM use bus_dmamap_sync() here.
12587				 */
12588			}
12589		}
12590
12591		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12592		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12593		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12594		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12595		msg.dt.sg_sequence = 0;
12596
12597		/*
12598		 * Loop until we've sent all of the S/G entries.  On the
12599		 * other end, we'll recompose these S/G entries into one
12600		 * contiguous list before passing it to the
12601		 */
12602		for (sg_entries_sent = 0; sg_entries_sent <
12603		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12604			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12605				sizeof(msg.dt.sg_list[0])),
12606				msg.dt.kern_sg_entries - sg_entries_sent);
12607
12608			if (do_sg_copy != 0) {
12609				struct ctl_sg_entry *sgl;
12610				int j;
12611
12612				sgl = (struct ctl_sg_entry *)
12613					io->scsiio.kern_data_ptr;
12614				/*
12615				 * If this is in cached memory, flush the cache
12616				 * before we send the DMA request to the other
12617				 * controller.  We want to do this in either
12618				 * the * read or the write case.  The read
12619				 * case is straightforward.  In the write
12620				 * case, we want to make sure nothing is
12621				 * in the local cache that could overwrite
12622				 * the DMAed data.
12623				 */
12624
12625				for (i = sg_entries_sent, j = 0;
12626				     i < msg.dt.cur_sg_entries; i++, j++) {
12627					if ((io->io_hdr.flags &
12628					     CTL_FLAG_NO_DATASYNC) == 0) {
12629						/*
12630						 * XXX KDM use bus_dmamap_sync()
12631						 */
12632					}
12633					if ((io->io_hdr.flags &
12634					     CTL_FLAG_BUS_ADDR) == 0) {
12635						/*
12636						 * XXX KDM use busdma.
12637						 */
12638#if 0
12639						msg.dt.sg_list[j].addr =(void *)
12640						       vtophys(sgl[i].addr);
12641#endif
12642					} else {
12643						msg.dt.sg_list[j].addr =
12644							sgl[i].addr;
12645					}
12646					msg.dt.sg_list[j].len = sgl[i].len;
12647				}
12648			}
12649
12650			sg_entries_sent += msg.dt.cur_sg_entries;
12651			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12652				msg.dt.sg_last = 1;
12653			else
12654				msg.dt.sg_last = 0;
12655
12656			/*
12657			 * XXX KDM drop and reacquire the lock here?
12658			 */
12659			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12660			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12661				/*
12662				 * XXX do something here.
12663				 */
12664			}
12665
12666			msg.dt.sent_sg_entries = sg_entries_sent;
12667		}
12668		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12669		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12670			ctl_failover_io(io, /*have_lock*/ 0);
12671
12672	} else {
12673
12674		/*
12675		 * Lookup the fe_datamove() function for this particular
12676		 * front end.
12677		 */
12678		fe_datamove =
12679		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12680
12681		fe_datamove(io);
12682	}
12683}
12684
12685static void
12686ctl_send_datamove_done(union ctl_io *io, int have_lock)
12687{
12688	union ctl_ha_msg msg;
12689	int isc_status;
12690
12691	memset(&msg, 0, sizeof(msg));
12692
12693	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12694	msg.hdr.original_sc = io;
12695	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12696	msg.hdr.nexus = io->io_hdr.nexus;
12697	msg.hdr.status = io->io_hdr.status;
12698	msg.scsi.tag_num = io->scsiio.tag_num;
12699	msg.scsi.tag_type = io->scsiio.tag_type;
12700	msg.scsi.scsi_status = io->scsiio.scsi_status;
12701	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12702	       sizeof(io->scsiio.sense_data));
12703	msg.scsi.sense_len = io->scsiio.sense_len;
12704	msg.scsi.sense_residual = io->scsiio.sense_residual;
12705	msg.scsi.fetd_status = io->io_hdr.port_status;
12706	msg.scsi.residual = io->scsiio.residual;
12707	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12708
12709	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12710		ctl_failover_io(io, /*have_lock*/ have_lock);
12711		return;
12712	}
12713
12714	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12715	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12716		/* XXX do something if this fails */
12717	}
12718
12719}
12720
12721/*
12722 * The DMA to the remote side is done, now we need to tell the other side
12723 * we're done so it can continue with its data movement.
12724 */
12725static void
12726ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12727{
12728	union ctl_io *io;
12729
12730	io = rq->context;
12731
12732	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12733		printf("%s: ISC DMA write failed with error %d", __func__,
12734		       rq->ret);
12735		ctl_set_internal_failure(&io->scsiio,
12736					 /*sks_valid*/ 1,
12737					 /*retry_count*/ rq->ret);
12738	}
12739
12740	ctl_dt_req_free(rq);
12741
12742	/*
12743	 * In this case, we had to malloc the memory locally.  Free it.
12744	 */
12745	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12746		int i;
12747		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12748			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12749	}
12750	/*
12751	 * The data is in local and remote memory, so now we need to send
12752	 * status (good or back) back to the other side.
12753	 */
12754	ctl_send_datamove_done(io, /*have_lock*/ 0);
12755}
12756
12757/*
12758 * We've moved the data from the host/controller into local memory.  Now we
12759 * need to push it over to the remote controller's memory.
12760 */
12761static int
12762ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12763{
12764	int retval;
12765
12766	retval = 0;
12767
12768	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12769					  ctl_datamove_remote_write_cb);
12770
12771	return (retval);
12772}
12773
12774static void
12775ctl_datamove_remote_write(union ctl_io *io)
12776{
12777	int retval;
12778	void (*fe_datamove)(union ctl_io *io);
12779
12780	/*
12781	 * - Get the data from the host/HBA into local memory.
12782	 * - DMA memory from the local controller to the remote controller.
12783	 * - Send status back to the remote controller.
12784	 */
12785
12786	retval = ctl_datamove_remote_sgl_setup(io);
12787	if (retval != 0)
12788		return;
12789
12790	/* Switch the pointer over so the FETD knows what to do */
12791	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12792
12793	/*
12794	 * Use a custom move done callback, since we need to send completion
12795	 * back to the other controller, not to the backend on this side.
12796	 */
12797	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12798
12799	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12800
12801	fe_datamove(io);
12802
12803	return;
12804
12805}
12806
12807static int
12808ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12809{
12810#if 0
12811	char str[256];
12812	char path_str[64];
12813	struct sbuf sb;
12814#endif
12815
12816	/*
12817	 * In this case, we had to malloc the memory locally.  Free it.
12818	 */
12819	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12820		int i;
12821		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12822			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12823	}
12824
12825#if 0
12826	scsi_path_string(io, path_str, sizeof(path_str));
12827	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12828	sbuf_cat(&sb, path_str);
12829	scsi_command_string(&io->scsiio, NULL, &sb);
12830	sbuf_printf(&sb, "\n");
12831	sbuf_cat(&sb, path_str);
12832	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12833		    io->scsiio.tag_num, io->scsiio.tag_type);
12834	sbuf_cat(&sb, path_str);
12835	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12836		    io->io_hdr.flags, io->io_hdr.status);
12837	sbuf_finish(&sb);
12838	printk("%s", sbuf_data(&sb));
12839#endif
12840
12841
12842	/*
12843	 * The read is done, now we need to send status (good or bad) back
12844	 * to the other side.
12845	 */
12846	ctl_send_datamove_done(io, /*have_lock*/ 0);
12847
12848	return (0);
12849}
12850
12851static void
12852ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12853{
12854	union ctl_io *io;
12855	void (*fe_datamove)(union ctl_io *io);
12856
12857	io = rq->context;
12858
12859	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12860		printf("%s: ISC DMA read failed with error %d", __func__,
12861		       rq->ret);
12862		ctl_set_internal_failure(&io->scsiio,
12863					 /*sks_valid*/ 1,
12864					 /*retry_count*/ rq->ret);
12865	}
12866
12867	ctl_dt_req_free(rq);
12868
12869	/* Switch the pointer over so the FETD knows what to do */
12870	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12871
12872	/*
12873	 * Use a custom move done callback, since we need to send completion
12874	 * back to the other controller, not to the backend on this side.
12875	 */
12876	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12877
12878	/* XXX KDM add checks like the ones in ctl_datamove? */
12879
12880	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12881
12882	fe_datamove(io);
12883}
12884
12885static int
12886ctl_datamove_remote_sgl_setup(union ctl_io *io)
12887{
12888	struct ctl_sg_entry *local_sglist, *remote_sglist;
12889	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12890	struct ctl_softc *softc;
12891	int retval;
12892	int i;
12893
12894	retval = 0;
12895	softc = control_softc;
12896
12897	local_sglist = io->io_hdr.local_sglist;
12898	local_dma_sglist = io->io_hdr.local_dma_sglist;
12899	remote_sglist = io->io_hdr.remote_sglist;
12900	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12901
12902	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12903		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12904			local_sglist[i].len = remote_sglist[i].len;
12905
12906			/*
12907			 * XXX Detect the situation where the RS-level I/O
12908			 * redirector on the other side has already read the
12909			 * data off of the AOR RS on this side, and
12910			 * transferred it to remote (mirror) memory on the
12911			 * other side.  Since we already have the data in
12912			 * memory here, we just need to use it.
12913			 *
12914			 * XXX KDM this can probably be removed once we
12915			 * get the cache device code in and take the
12916			 * current AOR implementation out.
12917			 */
12918#ifdef NEEDTOPORT
12919			if ((remote_sglist[i].addr >=
12920			     (void *)vtophys(softc->mirr->addr))
12921			 && (remote_sglist[i].addr <
12922			     ((void *)vtophys(softc->mirr->addr) +
12923			     CacheMirrorOffset))) {
12924				local_sglist[i].addr = remote_sglist[i].addr -
12925					CacheMirrorOffset;
12926				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12927				     CTL_FLAG_DATA_IN)
12928					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12929			} else {
12930				local_sglist[i].addr = remote_sglist[i].addr +
12931					CacheMirrorOffset;
12932			}
12933#endif
12934#if 0
12935			printf("%s: local %p, remote %p, len %d\n",
12936			       __func__, local_sglist[i].addr,
12937			       remote_sglist[i].addr, local_sglist[i].len);
12938#endif
12939		}
12940	} else {
12941		uint32_t len_to_go;
12942
12943		/*
12944		 * In this case, we don't have automatically allocated
12945		 * memory for this I/O on this controller.  This typically
12946		 * happens with internal CTL I/O -- e.g. inquiry, mode
12947		 * sense, etc.  Anything coming from RAIDCore will have
12948		 * a mirror area available.
12949		 */
12950		len_to_go = io->scsiio.kern_data_len;
12951
12952		/*
12953		 * Clear the no datasync flag, we have to use malloced
12954		 * buffers.
12955		 */
12956		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12957
12958		/*
12959		 * The difficult thing here is that the size of the various
12960		 * S/G segments may be different than the size from the
12961		 * remote controller.  That'll make it harder when DMAing
12962		 * the data back to the other side.
12963		 */
12964		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12965		     sizeof(io->io_hdr.remote_sglist[0])) &&
12966		     (len_to_go > 0); i++) {
12967			local_sglist[i].len = ctl_min(len_to_go, 131072);
12968			CTL_SIZE_8B(local_dma_sglist[i].len,
12969				    local_sglist[i].len);
12970			local_sglist[i].addr =
12971				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12972
12973			local_dma_sglist[i].addr = local_sglist[i].addr;
12974
12975			if (local_sglist[i].addr == NULL) {
12976				int j;
12977
12978				printf("malloc failed for %zd bytes!",
12979				       local_dma_sglist[i].len);
12980				for (j = 0; j < i; j++) {
12981					free(local_sglist[j].addr, M_CTL);
12982				}
12983				ctl_set_internal_failure(&io->scsiio,
12984							 /*sks_valid*/ 1,
12985							 /*retry_count*/ 4857);
12986				retval = 1;
12987				goto bailout_error;
12988
12989			}
12990			/* XXX KDM do we need a sync here? */
12991
12992			len_to_go -= local_sglist[i].len;
12993		}
12994		/*
12995		 * Reset the number of S/G entries accordingly.  The
12996		 * original number of S/G entries is available in
12997		 * rem_sg_entries.
12998		 */
12999		io->scsiio.kern_sg_entries = i;
13000
13001#if 0
13002		printf("%s: kern_sg_entries = %d\n", __func__,
13003		       io->scsiio.kern_sg_entries);
13004		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13005			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13006			       local_sglist[i].addr, local_sglist[i].len,
13007			       local_dma_sglist[i].len);
13008#endif
13009	}
13010
13011
13012	return (retval);
13013
13014bailout_error:
13015
13016	ctl_send_datamove_done(io, /*have_lock*/ 0);
13017
13018	return (retval);
13019}
13020
13021static int
13022ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13023			 ctl_ha_dt_cb callback)
13024{
13025	struct ctl_ha_dt_req *rq;
13026	struct ctl_sg_entry *remote_sglist, *local_sglist;
13027	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13028	uint32_t local_used, remote_used, total_used;
13029	int retval;
13030	int i, j;
13031
13032	retval = 0;
13033
13034	rq = ctl_dt_req_alloc();
13035
13036	/*
13037	 * If we failed to allocate the request, and if the DMA didn't fail
13038	 * anyway, set busy status.  This is just a resource allocation
13039	 * failure.
13040	 */
13041	if ((rq == NULL)
13042	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13043		ctl_set_busy(&io->scsiio);
13044
13045	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13046
13047		if (rq != NULL)
13048			ctl_dt_req_free(rq);
13049
13050		/*
13051		 * The data move failed.  We need to return status back
13052		 * to the other controller.  No point in trying to DMA
13053		 * data to the remote controller.
13054		 */
13055
13056		ctl_send_datamove_done(io, /*have_lock*/ 0);
13057
13058		retval = 1;
13059
13060		goto bailout;
13061	}
13062
13063	local_sglist = io->io_hdr.local_sglist;
13064	local_dma_sglist = io->io_hdr.local_dma_sglist;
13065	remote_sglist = io->io_hdr.remote_sglist;
13066	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13067	local_used = 0;
13068	remote_used = 0;
13069	total_used = 0;
13070
13071	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13072		rq->ret = CTL_HA_STATUS_SUCCESS;
13073		rq->context = io;
13074		callback(rq);
13075		goto bailout;
13076	}
13077
13078	/*
13079	 * Pull/push the data over the wire from/to the other controller.
13080	 * This takes into account the possibility that the local and
13081	 * remote sglists may not be identical in terms of the size of
13082	 * the elements and the number of elements.
13083	 *
13084	 * One fundamental assumption here is that the length allocated for
13085	 * both the local and remote sglists is identical.  Otherwise, we've
13086	 * essentially got a coding error of some sort.
13087	 */
13088	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13089		int isc_ret;
13090		uint32_t cur_len, dma_length;
13091		uint8_t *tmp_ptr;
13092
13093		rq->id = CTL_HA_DATA_CTL;
13094		rq->command = command;
13095		rq->context = io;
13096
13097		/*
13098		 * Both pointers should be aligned.  But it is possible
13099		 * that the allocation length is not.  They should both
13100		 * also have enough slack left over at the end, though,
13101		 * to round up to the next 8 byte boundary.
13102		 */
13103		cur_len = ctl_min(local_sglist[i].len - local_used,
13104				  remote_sglist[j].len - remote_used);
13105
13106		/*
13107		 * In this case, we have a size issue and need to decrease
13108		 * the size, except in the case where we actually have less
13109		 * than 8 bytes left.  In that case, we need to increase
13110		 * the DMA length to get the last bit.
13111		 */
13112		if ((cur_len & 0x7) != 0) {
13113			if (cur_len > 0x7) {
13114				cur_len = cur_len - (cur_len & 0x7);
13115				dma_length = cur_len;
13116			} else {
13117				CTL_SIZE_8B(dma_length, cur_len);
13118			}
13119
13120		} else
13121			dma_length = cur_len;
13122
13123		/*
13124		 * If we had to allocate memory for this I/O, instead of using
13125		 * the non-cached mirror memory, we'll need to flush the cache
13126		 * before trying to DMA to the other controller.
13127		 *
13128		 * We could end up doing this multiple times for the same
13129		 * segment if we have a larger local segment than remote
13130		 * segment.  That shouldn't be an issue.
13131		 */
13132		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13133			/*
13134			 * XXX KDM use bus_dmamap_sync() here.
13135			 */
13136		}
13137
13138		rq->size = dma_length;
13139
13140		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13141		tmp_ptr += local_used;
13142
13143		/* Use physical addresses when talking to ISC hardware */
13144		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13145			/* XXX KDM use busdma */
13146#if 0
13147			rq->local = vtophys(tmp_ptr);
13148#endif
13149		} else
13150			rq->local = tmp_ptr;
13151
13152		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13153		tmp_ptr += remote_used;
13154		rq->remote = tmp_ptr;
13155
13156		rq->callback = NULL;
13157
13158		local_used += cur_len;
13159		if (local_used >= local_sglist[i].len) {
13160			i++;
13161			local_used = 0;
13162		}
13163
13164		remote_used += cur_len;
13165		if (remote_used >= remote_sglist[j].len) {
13166			j++;
13167			remote_used = 0;
13168		}
13169		total_used += cur_len;
13170
13171		if (total_used >= io->scsiio.kern_data_len)
13172			rq->callback = callback;
13173
13174		if ((rq->size & 0x7) != 0) {
13175			printf("%s: warning: size %d is not on 8b boundary\n",
13176			       __func__, rq->size);
13177		}
13178		if (((uintptr_t)rq->local & 0x7) != 0) {
13179			printf("%s: warning: local %p not on 8b boundary\n",
13180			       __func__, rq->local);
13181		}
13182		if (((uintptr_t)rq->remote & 0x7) != 0) {
13183			printf("%s: warning: remote %p not on 8b boundary\n",
13184			       __func__, rq->local);
13185		}
13186#if 0
13187		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13188		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13189		       rq->local, rq->remote, rq->size);
13190#endif
13191
13192		isc_ret = ctl_dt_single(rq);
13193		if (isc_ret == CTL_HA_STATUS_WAIT)
13194			continue;
13195
13196		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13197			rq->ret = CTL_HA_STATUS_SUCCESS;
13198		} else {
13199			rq->ret = isc_ret;
13200		}
13201		callback(rq);
13202		goto bailout;
13203	}
13204
13205bailout:
13206	return (retval);
13207
13208}
13209
13210static void
13211ctl_datamove_remote_read(union ctl_io *io)
13212{
13213	int retval;
13214	int i;
13215
13216	/*
13217	 * This will send an error to the other controller in the case of a
13218	 * failure.
13219	 */
13220	retval = ctl_datamove_remote_sgl_setup(io);
13221	if (retval != 0)
13222		return;
13223
13224	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13225					  ctl_datamove_remote_read_cb);
13226	if ((retval != 0)
13227	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13228		/*
13229		 * Make sure we free memory if there was an error..  The
13230		 * ctl_datamove_remote_xfer() function will send the
13231		 * datamove done message, or call the callback with an
13232		 * error if there is a problem.
13233		 */
13234		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13235			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13236	}
13237
13238	return;
13239}
13240
13241/*
13242 * Process a datamove request from the other controller.  This is used for
13243 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13244 * first.  Once that is complete, the data gets DMAed into the remote
13245 * controller's memory.  For reads, we DMA from the remote controller's
13246 * memory into our memory first, and then move it out to the FETD.
13247 */
13248static void
13249ctl_datamove_remote(union ctl_io *io)
13250{
13251	struct ctl_softc *softc;
13252
13253	softc = control_softc;
13254
13255	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13256
13257	/*
13258	 * Note that we look for an aborted I/O here, but don't do some of
13259	 * the other checks that ctl_datamove() normally does.
13260	 * We don't need to run the datamove delay code, since that should
13261	 * have been done if need be on the other controller.
13262	 */
13263	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13264		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13265		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13266		       io->io_hdr.nexus.targ_port,
13267		       io->io_hdr.nexus.targ_target.id,
13268		       io->io_hdr.nexus.targ_lun);
13269		io->io_hdr.port_status = 31338;
13270		ctl_send_datamove_done(io, /*have_lock*/ 0);
13271		return;
13272	}
13273
13274	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13275		ctl_datamove_remote_write(io);
13276	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13277		ctl_datamove_remote_read(io);
13278	} else {
13279		union ctl_ha_msg msg;
13280		struct scsi_sense_data *sense;
13281		uint8_t sks[3];
13282		int retry_count;
13283
13284		memset(&msg, 0, sizeof(msg));
13285
13286		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13287		msg.hdr.status = CTL_SCSI_ERROR;
13288		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13289
13290		retry_count = 4243;
13291
13292		sense = &msg.scsi.sense_data;
13293		sks[0] = SSD_SCS_VALID;
13294		sks[1] = (retry_count >> 8) & 0xff;
13295		sks[2] = retry_count & 0xff;
13296
13297		/* "Internal target failure" */
13298		scsi_set_sense_data(sense,
13299				    /*sense_format*/ SSD_TYPE_NONE,
13300				    /*current_error*/ 1,
13301				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13302				    /*asc*/ 0x44,
13303				    /*ascq*/ 0x00,
13304				    /*type*/ SSD_ELEM_SKS,
13305				    /*size*/ sizeof(sks),
13306				    /*data*/ sks,
13307				    SSD_ELEM_NONE);
13308
13309		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13310		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13311			ctl_failover_io(io, /*have_lock*/ 1);
13312			return;
13313		}
13314
13315		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13316		    CTL_HA_STATUS_SUCCESS) {
13317			/* XXX KDM what to do if this fails? */
13318		}
13319		return;
13320	}
13321
13322}
13323
13324static int
13325ctl_process_done(union ctl_io *io)
13326{
13327	struct ctl_lun *lun;
13328	struct ctl_softc *ctl_softc = control_softc;
13329	void (*fe_done)(union ctl_io *io);
13330	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13331
13332	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13333
13334	fe_done =
13335	    control_softc->ctl_ports[targ_port]->fe_done;
13336
13337#ifdef CTL_TIME_IO
13338	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13339		char str[256];
13340		char path_str[64];
13341		struct sbuf sb;
13342
13343		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13344		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13345
13346		sbuf_cat(&sb, path_str);
13347		switch (io->io_hdr.io_type) {
13348		case CTL_IO_SCSI:
13349			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13350			sbuf_printf(&sb, "\n");
13351			sbuf_cat(&sb, path_str);
13352			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13353				    io->scsiio.tag_num, io->scsiio.tag_type);
13354			break;
13355		case CTL_IO_TASK:
13356			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13357				    "Tag Type: %d\n", io->taskio.task_action,
13358				    io->taskio.tag_num, io->taskio.tag_type);
13359			break;
13360		default:
13361			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13362			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13363			break;
13364		}
13365		sbuf_cat(&sb, path_str);
13366		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13367			    (intmax_t)time_uptime - io->io_hdr.start_time);
13368		sbuf_finish(&sb);
13369		printf("%s", sbuf_data(&sb));
13370	}
13371#endif /* CTL_TIME_IO */
13372
13373	switch (io->io_hdr.io_type) {
13374	case CTL_IO_SCSI:
13375		break;
13376	case CTL_IO_TASK:
13377		if (bootverbose || (ctl_debug & CTL_DEBUG_INFO))
13378			ctl_io_error_print(io, NULL);
13379		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13380			ctl_free_io(io);
13381		else
13382			fe_done(io);
13383		return (CTL_RETVAL_COMPLETE);
13384	default:
13385		panic("ctl_process_done: invalid io type %d\n",
13386		      io->io_hdr.io_type);
13387		break; /* NOTREACHED */
13388	}
13389
13390	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13391	if (lun == NULL) {
13392		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13393				 io->io_hdr.nexus.targ_mapped_lun));
13394		goto bailout;
13395	}
13396
13397	mtx_lock(&lun->lun_lock);
13398
13399	/*
13400	 * Check to see if we have any errors to inject here.  We only
13401	 * inject errors for commands that don't already have errors set.
13402	 */
13403	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13404	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13405	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13406		ctl_inject_error(lun, io);
13407
13408	/*
13409	 * XXX KDM how do we treat commands that aren't completed
13410	 * successfully?
13411	 *
13412	 * XXX KDM should we also track I/O latency?
13413	 */
13414	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13415	    io->io_hdr.io_type == CTL_IO_SCSI) {
13416#ifdef CTL_TIME_IO
13417		struct bintime cur_bt;
13418#endif
13419		int type;
13420
13421		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13422		    CTL_FLAG_DATA_IN)
13423			type = CTL_STATS_READ;
13424		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13425		    CTL_FLAG_DATA_OUT)
13426			type = CTL_STATS_WRITE;
13427		else
13428			type = CTL_STATS_NO_IO;
13429
13430		lun->stats.ports[targ_port].bytes[type] +=
13431		    io->scsiio.kern_total_len;
13432		lun->stats.ports[targ_port].operations[type]++;
13433#ifdef CTL_TIME_IO
13434		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13435		   &io->io_hdr.dma_bt);
13436		lun->stats.ports[targ_port].num_dmas[type] +=
13437		    io->io_hdr.num_dmas;
13438		getbintime(&cur_bt);
13439		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13440		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13441#endif
13442	}
13443
13444	/*
13445	 * Remove this from the OOA queue.
13446	 */
13447	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13448
13449	/*
13450	 * Run through the blocked queue on this LUN and see if anything
13451	 * has become unblocked, now that this transaction is done.
13452	 */
13453	ctl_check_blocked(lun);
13454
13455	/*
13456	 * If the LUN has been invalidated, free it if there is nothing
13457	 * left on its OOA queue.
13458	 */
13459	if ((lun->flags & CTL_LUN_INVALID)
13460	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13461		mtx_unlock(&lun->lun_lock);
13462		mtx_lock(&ctl_softc->ctl_lock);
13463		ctl_free_lun(lun);
13464		mtx_unlock(&ctl_softc->ctl_lock);
13465	} else
13466		mtx_unlock(&lun->lun_lock);
13467
13468bailout:
13469
13470	/*
13471	 * If this command has been aborted, make sure we set the status
13472	 * properly.  The FETD is responsible for freeing the I/O and doing
13473	 * whatever it needs to do to clean up its state.
13474	 */
13475	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13476		ctl_set_task_aborted(&io->scsiio);
13477
13478	/*
13479	 * If enabled, print command error status.
13480	 * We don't print UAs unless debugging was enabled explicitly.
13481	 */
13482	do {
13483		if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
13484			break;
13485		if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0)
13486			break;
13487		if ((ctl_debug & CTL_DEBUG_INFO) == 0 &&
13488		    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) &&
13489		     (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13490			int error_code, sense_key, asc, ascq;
13491
13492			scsi_extract_sense_len(&io->scsiio.sense_data,
13493			    io->scsiio.sense_len, &error_code, &sense_key,
13494			    &asc, &ascq, /*show_errors*/ 0);
13495			if (sense_key == SSD_KEY_UNIT_ATTENTION)
13496				break;
13497		}
13498
13499		ctl_io_error_print(io, NULL);
13500	} while (0);
13501
13502	/*
13503	 * Tell the FETD or the other shelf controller we're done with this
13504	 * command.  Note that only SCSI commands get to this point.  Task
13505	 * management commands are completed above.
13506	 *
13507	 * We only send status to the other controller if we're in XFER
13508	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13509	 * received the I/O (from CTL's perspective), and so the status is
13510	 * generated there.
13511	 *
13512	 * XXX KDM if we hold the lock here, we could cause a deadlock
13513	 * if the frontend comes back in in this context to queue
13514	 * something.
13515	 */
13516	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13517	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13518		union ctl_ha_msg msg;
13519
13520		memset(&msg, 0, sizeof(msg));
13521		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13522		msg.hdr.original_sc = io->io_hdr.original_sc;
13523		msg.hdr.nexus = io->io_hdr.nexus;
13524		msg.hdr.status = io->io_hdr.status;
13525		msg.scsi.scsi_status = io->scsiio.scsi_status;
13526		msg.scsi.tag_num = io->scsiio.tag_num;
13527		msg.scsi.tag_type = io->scsiio.tag_type;
13528		msg.scsi.sense_len = io->scsiio.sense_len;
13529		msg.scsi.sense_residual = io->scsiio.sense_residual;
13530		msg.scsi.residual = io->scsiio.residual;
13531		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13532		       sizeof(io->scsiio.sense_data));
13533		/*
13534		 * We copy this whether or not this is an I/O-related
13535		 * command.  Otherwise, we'd have to go and check to see
13536		 * whether it's a read/write command, and it really isn't
13537		 * worth it.
13538		 */
13539		memcpy(&msg.scsi.lbalen,
13540		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13541		       sizeof(msg.scsi.lbalen));
13542
13543		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13544				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13545			/* XXX do something here */
13546		}
13547
13548		ctl_free_io(io);
13549	} else
13550		fe_done(io);
13551
13552	return (CTL_RETVAL_COMPLETE);
13553}
13554
13555#ifdef CTL_WITH_CA
13556/*
13557 * Front end should call this if it doesn't do autosense.  When the request
13558 * sense comes back in from the initiator, we'll dequeue this and send it.
13559 */
13560int
13561ctl_queue_sense(union ctl_io *io)
13562{
13563	struct ctl_lun *lun;
13564	struct ctl_softc *ctl_softc;
13565	uint32_t initidx, targ_lun;
13566
13567	ctl_softc = control_softc;
13568
13569	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13570
13571	/*
13572	 * LUN lookup will likely move to the ctl_work_thread() once we
13573	 * have our new queueing infrastructure (that doesn't put things on
13574	 * a per-LUN queue initially).  That is so that we can handle
13575	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13576	 * can't deal with that right now.
13577	 */
13578	mtx_lock(&ctl_softc->ctl_lock);
13579
13580	/*
13581	 * If we don't have a LUN for this, just toss the sense
13582	 * information.
13583	 */
13584	targ_lun = io->io_hdr.nexus.targ_lun;
13585	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13586	if ((targ_lun < CTL_MAX_LUNS)
13587	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13588		lun = ctl_softc->ctl_luns[targ_lun];
13589	else
13590		goto bailout;
13591
13592	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13593
13594	mtx_lock(&lun->lun_lock);
13595	/*
13596	 * Already have CA set for this LUN...toss the sense information.
13597	 */
13598	if (ctl_is_set(lun->have_ca, initidx)) {
13599		mtx_unlock(&lun->lun_lock);
13600		goto bailout;
13601	}
13602
13603	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13604	       ctl_min(sizeof(lun->pending_sense[initidx]),
13605	       sizeof(io->scsiio.sense_data)));
13606	ctl_set_mask(lun->have_ca, initidx);
13607	mtx_unlock(&lun->lun_lock);
13608
13609bailout:
13610	mtx_unlock(&ctl_softc->ctl_lock);
13611
13612	ctl_free_io(io);
13613
13614	return (CTL_RETVAL_COMPLETE);
13615}
13616#endif
13617
13618/*
13619 * Primary command inlet from frontend ports.  All SCSI and task I/O
13620 * requests must go through this function.
13621 */
13622int
13623ctl_queue(union ctl_io *io)
13624{
13625	struct ctl_softc *ctl_softc;
13626
13627	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13628
13629	ctl_softc = control_softc;
13630
13631#ifdef CTL_TIME_IO
13632	io->io_hdr.start_time = time_uptime;
13633	getbintime(&io->io_hdr.start_bt);
13634#endif /* CTL_TIME_IO */
13635
13636	/* Map FE-specific LUN ID into global one. */
13637	io->io_hdr.nexus.targ_mapped_lun =
13638	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13639
13640	switch (io->io_hdr.io_type) {
13641	case CTL_IO_SCSI:
13642	case CTL_IO_TASK:
13643		if (ctl_debug & CTL_DEBUG_CDB)
13644			ctl_io_print(io);
13645		ctl_enqueue_incoming(io);
13646		break;
13647	default:
13648		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13649		return (EINVAL);
13650	}
13651
13652	return (CTL_RETVAL_COMPLETE);
13653}
13654
13655#ifdef CTL_IO_DELAY
13656static void
13657ctl_done_timer_wakeup(void *arg)
13658{
13659	union ctl_io *io;
13660
13661	io = (union ctl_io *)arg;
13662	ctl_done(io);
13663}
13664#endif /* CTL_IO_DELAY */
13665
13666void
13667ctl_done(union ctl_io *io)
13668{
13669	struct ctl_softc *ctl_softc;
13670
13671	ctl_softc = control_softc;
13672
13673	/*
13674	 * Enable this to catch duplicate completion issues.
13675	 */
13676#if 0
13677	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13678		printf("%s: type %d msg %d cdb %x iptl: "
13679		       "%d:%d:%d:%d tag 0x%04x "
13680		       "flag %#x status %x\n",
13681			__func__,
13682			io->io_hdr.io_type,
13683			io->io_hdr.msg_type,
13684			io->scsiio.cdb[0],
13685			io->io_hdr.nexus.initid.id,
13686			io->io_hdr.nexus.targ_port,
13687			io->io_hdr.nexus.targ_target.id,
13688			io->io_hdr.nexus.targ_lun,
13689			(io->io_hdr.io_type ==
13690			CTL_IO_TASK) ?
13691			io->taskio.tag_num :
13692			io->scsiio.tag_num,
13693		        io->io_hdr.flags,
13694			io->io_hdr.status);
13695	} else
13696		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13697#endif
13698
13699	/*
13700	 * This is an internal copy of an I/O, and should not go through
13701	 * the normal done processing logic.
13702	 */
13703	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13704		return;
13705
13706	/*
13707	 * We need to send a msg to the serializing shelf to finish the IO
13708	 * as well.  We don't send a finish message to the other shelf if
13709	 * this is a task management command.  Task management commands
13710	 * aren't serialized in the OOA queue, but rather just executed on
13711	 * both shelf controllers for commands that originated on that
13712	 * controller.
13713	 */
13714	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13715	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13716		union ctl_ha_msg msg_io;
13717
13718		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13719		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13720		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13721		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13722		}
13723		/* continue on to finish IO */
13724	}
13725#ifdef CTL_IO_DELAY
13726	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13727		struct ctl_lun *lun;
13728
13729		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13730
13731		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13732	} else {
13733		struct ctl_lun *lun;
13734
13735		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13736
13737		if ((lun != NULL)
13738		 && (lun->delay_info.done_delay > 0)) {
13739			struct callout *callout;
13740
13741			callout = (struct callout *)&io->io_hdr.timer_bytes;
13742			callout_init(callout, /*mpsafe*/ 1);
13743			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13744			callout_reset(callout,
13745				      lun->delay_info.done_delay * hz,
13746				      ctl_done_timer_wakeup, io);
13747			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13748				lun->delay_info.done_delay = 0;
13749			return;
13750		}
13751	}
13752#endif /* CTL_IO_DELAY */
13753
13754	ctl_enqueue_done(io);
13755}
13756
13757int
13758ctl_isc(struct ctl_scsiio *ctsio)
13759{
13760	struct ctl_lun *lun;
13761	int retval;
13762
13763	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13764
13765	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13766
13767	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13768
13769	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13770
13771	return (retval);
13772}
13773
13774
13775static void
13776ctl_work_thread(void *arg)
13777{
13778	struct ctl_thread *thr = (struct ctl_thread *)arg;
13779	struct ctl_softc *softc = thr->ctl_softc;
13780	union ctl_io *io;
13781	int retval;
13782
13783	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13784
13785	for (;;) {
13786		retval = 0;
13787
13788		/*
13789		 * We handle the queues in this order:
13790		 * - ISC
13791		 * - done queue (to free up resources, unblock other commands)
13792		 * - RtR queue
13793		 * - incoming queue
13794		 *
13795		 * If those queues are empty, we break out of the loop and
13796		 * go to sleep.
13797		 */
13798		mtx_lock(&thr->queue_lock);
13799		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13800		if (io != NULL) {
13801			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13802			mtx_unlock(&thr->queue_lock);
13803			ctl_handle_isc(io);
13804			continue;
13805		}
13806		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13807		if (io != NULL) {
13808			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13809			/* clear any blocked commands, call fe_done */
13810			mtx_unlock(&thr->queue_lock);
13811			retval = ctl_process_done(io);
13812			continue;
13813		}
13814		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13815		if (io != NULL) {
13816			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13817			mtx_unlock(&thr->queue_lock);
13818			if (io->io_hdr.io_type == CTL_IO_TASK)
13819				ctl_run_task(io);
13820			else
13821				ctl_scsiio_precheck(softc, &io->scsiio);
13822			continue;
13823		}
13824		if (!ctl_pause_rtr) {
13825			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13826			if (io != NULL) {
13827				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13828				mtx_unlock(&thr->queue_lock);
13829				retval = ctl_scsiio(&io->scsiio);
13830				if (retval != CTL_RETVAL_COMPLETE)
13831					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13832				continue;
13833			}
13834		}
13835
13836		/* Sleep until we have something to do. */
13837		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13838	}
13839}
13840
13841static void
13842ctl_lun_thread(void *arg)
13843{
13844	struct ctl_softc *softc = (struct ctl_softc *)arg;
13845	struct ctl_be_lun *be_lun;
13846	int retval;
13847
13848	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13849
13850	for (;;) {
13851		retval = 0;
13852		mtx_lock(&softc->ctl_lock);
13853		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13854		if (be_lun != NULL) {
13855			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13856			mtx_unlock(&softc->ctl_lock);
13857			ctl_create_lun(be_lun);
13858			continue;
13859		}
13860
13861		/* Sleep until we have something to do. */
13862		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13863		    PDROP | PRIBIO, "-", 0);
13864	}
13865}
13866
13867static void
13868ctl_thresh_thread(void *arg)
13869{
13870	struct ctl_softc *softc = (struct ctl_softc *)arg;
13871	struct ctl_lun *lun;
13872	struct ctl_be_lun *be_lun;
13873	struct scsi_da_rw_recovery_page *rwpage;
13874	struct ctl_logical_block_provisioning_page *page;
13875	const char *attr;
13876	uint64_t thres, val;
13877	int i, e;
13878
13879	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13880
13881	for (;;) {
13882		mtx_lock(&softc->ctl_lock);
13883		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13884			be_lun = lun->be_lun;
13885			if ((lun->flags & CTL_LUN_DISABLED) ||
13886			    (lun->flags & CTL_LUN_OFFLINE) ||
13887			    (be_lun->flags & CTL_LUN_FLAG_UNMAP) == 0 ||
13888			    lun->backend->lun_attr == NULL)
13889				continue;
13890			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13891			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13892				continue;
13893			e = 0;
13894			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13895			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13896				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13897					continue;
13898				thres = scsi_4btoul(page->descr[i].count);
13899				thres <<= CTL_LBP_EXPONENT;
13900				switch (page->descr[i].resource) {
13901				case 0x01:
13902					attr = "blocksavail";
13903					break;
13904				case 0x02:
13905					attr = "blocksused";
13906					break;
13907				case 0xf1:
13908					attr = "poolblocksavail";
13909					break;
13910				case 0xf2:
13911					attr = "poolblocksused";
13912					break;
13913				default:
13914					continue;
13915				}
13916				mtx_unlock(&softc->ctl_lock); // XXX
13917				val = lun->backend->lun_attr(
13918				    lun->be_lun->be_lun, attr);
13919				mtx_lock(&softc->ctl_lock);
13920				if (val == UINT64_MAX)
13921					continue;
13922				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13923				    == SLBPPD_ARMING_INC)
13924					e |= (val >= thres);
13925				else
13926					e |= (val <= thres);
13927			}
13928			mtx_lock(&lun->lun_lock);
13929			if (e) {
13930				if (lun->lasttpt == 0 ||
13931				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13932					lun->lasttpt = time_uptime;
13933					for (i = 0; i < CTL_MAX_INITIATORS; i++)
13934						lun->pending_ua[i] |=
13935						    CTL_UA_THIN_PROV_THRES;
13936				}
13937			} else {
13938				lun->lasttpt = 0;
13939				for (i = 0; i < CTL_MAX_INITIATORS; i++)
13940					lun->pending_ua[i] &= ~CTL_UA_THIN_PROV_THRES;
13941			}
13942			mtx_unlock(&lun->lun_lock);
13943		}
13944		mtx_unlock(&softc->ctl_lock);
13945		pause("-", CTL_LBP_PERIOD * hz);
13946	}
13947}
13948
13949static void
13950ctl_enqueue_incoming(union ctl_io *io)
13951{
13952	struct ctl_softc *softc = control_softc;
13953	struct ctl_thread *thr;
13954	u_int idx;
13955
13956	idx = (io->io_hdr.nexus.targ_port * 127 +
13957	       io->io_hdr.nexus.initid.id) % worker_threads;
13958	thr = &softc->threads[idx];
13959	mtx_lock(&thr->queue_lock);
13960	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13961	mtx_unlock(&thr->queue_lock);
13962	wakeup(thr);
13963}
13964
13965static void
13966ctl_enqueue_rtr(union ctl_io *io)
13967{
13968	struct ctl_softc *softc = control_softc;
13969	struct ctl_thread *thr;
13970
13971	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13972	mtx_lock(&thr->queue_lock);
13973	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13974	mtx_unlock(&thr->queue_lock);
13975	wakeup(thr);
13976}
13977
13978static void
13979ctl_enqueue_done(union ctl_io *io)
13980{
13981	struct ctl_softc *softc = control_softc;
13982	struct ctl_thread *thr;
13983
13984	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13985	mtx_lock(&thr->queue_lock);
13986	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13987	mtx_unlock(&thr->queue_lock);
13988	wakeup(thr);
13989}
13990
13991static void
13992ctl_enqueue_isc(union ctl_io *io)
13993{
13994	struct ctl_softc *softc = control_softc;
13995	struct ctl_thread *thr;
13996
13997	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13998	mtx_lock(&thr->queue_lock);
13999	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14000	mtx_unlock(&thr->queue_lock);
14001	wakeup(thr);
14002}
14003
14004/* Initialization and failover */
14005
14006void
14007ctl_init_isc_msg(void)
14008{
14009	printf("CTL: Still calling this thing\n");
14010}
14011
14012/*
14013 * Init component
14014 * 	Initializes component into configuration defined by bootMode
14015 *	(see hasc-sv.c)
14016 *  	returns hasc_Status:
14017 * 		OK
14018 *		ERROR - fatal error
14019 */
14020static ctl_ha_comp_status
14021ctl_isc_init(struct ctl_ha_component *c)
14022{
14023	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14024
14025	c->status = ret;
14026	return ret;
14027}
14028
14029/* Start component
14030 * 	Starts component in state requested. If component starts successfully,
14031 *	it must set its own state to the requestrd state
14032 *	When requested state is HASC_STATE_HA, the component may refine it
14033 * 	by adding _SLAVE or _MASTER flags.
14034 *	Currently allowed state transitions are:
14035 *	UNKNOWN->HA		- initial startup
14036 *	UNKNOWN->SINGLE - initial startup when no parter detected
14037 *	HA->SINGLE		- failover
14038 * returns ctl_ha_comp_status:
14039 * 		OK	- component successfully started in requested state
14040 *		FAILED  - could not start the requested state, failover may
14041 * 			  be possible
14042 *		ERROR	- fatal error detected, no future startup possible
14043 */
14044static ctl_ha_comp_status
14045ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14046{
14047	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14048
14049	printf("%s: go\n", __func__);
14050
14051	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14052	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14053		control_softc->is_single = 0;
14054		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14055		    != CTL_HA_STATUS_SUCCESS) {
14056			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14057			ret = CTL_HA_COMP_STATUS_ERROR;
14058		}
14059	} else if (CTL_HA_STATE_IS_HA(c->state)
14060		&& CTL_HA_STATE_IS_SINGLE(state)){
14061		// HA->SINGLE transition
14062	        ctl_failover();
14063		control_softc->is_single = 1;
14064	} else {
14065		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14066		       c->state, state);
14067		ret = CTL_HA_COMP_STATUS_ERROR;
14068	}
14069	if (CTL_HA_STATE_IS_SINGLE(state))
14070		control_softc->is_single = 1;
14071
14072	c->state = state;
14073	c->status = ret;
14074	return ret;
14075}
14076
14077/*
14078 * Quiesce component
14079 * The component must clear any error conditions (set status to OK) and
14080 * prepare itself to another Start call
14081 * returns ctl_ha_comp_status:
14082 * 	OK
14083 *	ERROR
14084 */
14085static ctl_ha_comp_status
14086ctl_isc_quiesce(struct ctl_ha_component *c)
14087{
14088	int ret = CTL_HA_COMP_STATUS_OK;
14089
14090	ctl_pause_rtr = 1;
14091	c->status = ret;
14092	return ret;
14093}
14094
14095struct ctl_ha_component ctl_ha_component_ctlisc =
14096{
14097	.name = "CTL ISC",
14098	.state = CTL_HA_STATE_UNKNOWN,
14099	.init = ctl_isc_init,
14100	.start = ctl_isc_start,
14101	.quiesce = ctl_isc_quiesce
14102};
14103
14104/*
14105 *  vim: ts=8
14106 */
14107