ctl.c revision 275058
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 275058 2014-11-25 17:53:35Z 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 int persis_offset;
360static uint8_t ctl_pause_rtr;
361
362SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
363static int worker_threads = -1;
364SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
365    &worker_threads, 1, "Number of worker threads");
366static int ctl_debug = CTL_DEBUG_NONE;
367SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
368    &ctl_debug, 0, "Enabled debug flags");
369
370/*
371 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
372 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
373 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
374 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
375 */
376#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
377
378static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
379				  int param);
380static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
381static int ctl_init(void);
382void ctl_shutdown(void);
383static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
384static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
385static void ctl_ioctl_online(void *arg);
386static void ctl_ioctl_offline(void *arg);
387static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
388static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
389static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
390static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
391static int ctl_ioctl_submit_wait(union ctl_io *io);
392static void ctl_ioctl_datamove(union ctl_io *io);
393static void ctl_ioctl_done(union ctl_io *io);
394static void ctl_ioctl_hard_startstop_callback(void *arg,
395					      struct cfi_metatask *metatask);
396static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
397static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
398			      struct ctl_ooa *ooa_hdr,
399			      struct ctl_ooa_entry *kern_entries);
400static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
401		     struct thread *td);
402static uint32_t ctl_map_lun(int port_num, uint32_t lun);
403static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
404#ifdef unused
405static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
406				   uint32_t targ_target, uint32_t targ_lun,
407				   int can_wait);
408static void ctl_kfree_io(union ctl_io *io);
409#endif /* unused */
410static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
411			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
412static int ctl_free_lun(struct ctl_lun *lun);
413static void ctl_create_lun(struct ctl_be_lun *be_lun);
414/**
415static void ctl_failover_change_pages(struct ctl_softc *softc,
416				      struct ctl_scsiio *ctsio, int master);
417**/
418
419static int ctl_do_mode_select(union ctl_io *io);
420static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
421			   uint64_t res_key, uint64_t sa_res_key,
422			   uint8_t type, uint32_t residx,
423			   struct ctl_scsiio *ctsio,
424			   struct scsi_per_res_out *cdb,
425			   struct scsi_per_res_out_parms* param);
426static void ctl_pro_preempt_other(struct ctl_lun *lun,
427				  union ctl_ha_msg *msg);
428static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
429static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
430static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
431static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
432static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
433static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
434static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
435					 int alloc_len);
436static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
437					 int alloc_len);
438static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
439static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
440static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
441static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
442static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
443static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
444static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
445    union ctl_io *pending_io, union ctl_io *ooa_io);
446static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
447				union ctl_io *starting_io);
448static int ctl_check_blocked(struct ctl_lun *lun);
449static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
450				struct ctl_lun *lun,
451				const struct ctl_cmd_entry *entry,
452				struct ctl_scsiio *ctsio);
453//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
454static void ctl_failover(void);
455static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
456			       struct ctl_scsiio *ctsio);
457static int ctl_scsiio(struct ctl_scsiio *ctsio);
458
459static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
460static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
461			    ctl_ua_type ua_type);
462static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
463			 ctl_ua_type ua_type);
464static int ctl_abort_task(union ctl_io *io);
465static int ctl_abort_task_set(union ctl_io *io);
466static int ctl_i_t_nexus_reset(union ctl_io *io);
467static void ctl_run_task(union ctl_io *io);
468#ifdef CTL_IO_DELAY
469static void ctl_datamove_timer_wakeup(void *arg);
470static void ctl_done_timer_wakeup(void *arg);
471#endif /* CTL_IO_DELAY */
472
473static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
474static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
475static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
476static void ctl_datamove_remote_write(union ctl_io *io);
477static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
478static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
479static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
480static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
481				    ctl_ha_dt_cb callback);
482static void ctl_datamove_remote_read(union ctl_io *io);
483static void ctl_datamove_remote(union ctl_io *io);
484static int ctl_process_done(union ctl_io *io);
485static void ctl_lun_thread(void *arg);
486static void ctl_thresh_thread(void *arg);
487static void ctl_work_thread(void *arg);
488static void ctl_enqueue_incoming(union ctl_io *io);
489static void ctl_enqueue_rtr(union ctl_io *io);
490static void ctl_enqueue_done(union ctl_io *io);
491static void ctl_enqueue_isc(union ctl_io *io);
492static const struct ctl_cmd_entry *
493    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
494static const struct ctl_cmd_entry *
495    ctl_validate_command(struct ctl_scsiio *ctsio);
496static int ctl_cmd_applicable(uint8_t lun_type,
497    const struct ctl_cmd_entry *entry);
498
499/*
500 * Load the serialization table.  This isn't very pretty, but is probably
501 * the easiest way to do it.
502 */
503#include "ctl_ser_table.c"
504
505/*
506 * We only need to define open, close and ioctl routines for this driver.
507 */
508static struct cdevsw ctl_cdevsw = {
509	.d_version =	D_VERSION,
510	.d_flags =	0,
511	.d_open =	ctl_open,
512	.d_close =	ctl_close,
513	.d_ioctl =	ctl_ioctl,
514	.d_name =	"ctl",
515};
516
517
518MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
519MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
520
521static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
522
523static moduledata_t ctl_moduledata = {
524	"ctl",
525	ctl_module_event_handler,
526	NULL
527};
528
529DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
530MODULE_VERSION(ctl, 1);
531
532static struct ctl_frontend ioctl_frontend =
533{
534	.name = "ioctl",
535};
536
537static void
538ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
539			    union ctl_ha_msg *msg_info)
540{
541	struct ctl_scsiio *ctsio;
542
543	if (msg_info->hdr.original_sc == NULL) {
544		printf("%s: original_sc == NULL!\n", __func__);
545		/* XXX KDM now what? */
546		return;
547	}
548
549	ctsio = &msg_info->hdr.original_sc->scsiio;
550	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
551	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
552	ctsio->io_hdr.status = msg_info->hdr.status;
553	ctsio->scsi_status = msg_info->scsi.scsi_status;
554	ctsio->sense_len = msg_info->scsi.sense_len;
555	ctsio->sense_residual = msg_info->scsi.sense_residual;
556	ctsio->residual = msg_info->scsi.residual;
557	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
558	       sizeof(ctsio->sense_data));
559	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
560	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
561	ctl_enqueue_isc((union ctl_io *)ctsio);
562}
563
564static void
565ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
566				union ctl_ha_msg *msg_info)
567{
568	struct ctl_scsiio *ctsio;
569
570	if (msg_info->hdr.serializing_sc == NULL) {
571		printf("%s: serializing_sc == NULL!\n", __func__);
572		/* XXX KDM now what? */
573		return;
574	}
575
576	ctsio = &msg_info->hdr.serializing_sc->scsiio;
577#if 0
578	/*
579	 * Attempt to catch the situation where an I/O has
580	 * been freed, and we're using it again.
581	 */
582	if (ctsio->io_hdr.io_type == 0xff) {
583		union ctl_io *tmp_io;
584		tmp_io = (union ctl_io *)ctsio;
585		printf("%s: %p use after free!\n", __func__,
586		       ctsio);
587		printf("%s: type %d msg %d cdb %x iptl: "
588		       "%d:%d:%d:%d tag 0x%04x "
589		       "flag %#x status %x\n",
590			__func__,
591			tmp_io->io_hdr.io_type,
592			tmp_io->io_hdr.msg_type,
593			tmp_io->scsiio.cdb[0],
594			tmp_io->io_hdr.nexus.initid.id,
595			tmp_io->io_hdr.nexus.targ_port,
596			tmp_io->io_hdr.nexus.targ_target.id,
597			tmp_io->io_hdr.nexus.targ_lun,
598			(tmp_io->io_hdr.io_type ==
599			CTL_IO_TASK) ?
600			tmp_io->taskio.tag_num :
601			tmp_io->scsiio.tag_num,
602		        tmp_io->io_hdr.flags,
603			tmp_io->io_hdr.status);
604	}
605#endif
606	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
607	ctl_enqueue_isc((union ctl_io *)ctsio);
608}
609
610/*
611 * ISC (Inter Shelf Communication) event handler.  Events from the HA
612 * subsystem come in here.
613 */
614static void
615ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
616{
617	struct ctl_softc *ctl_softc;
618	union ctl_io *io;
619	struct ctl_prio *presio;
620	ctl_ha_status isc_status;
621
622	ctl_softc = control_softc;
623	io = NULL;
624
625
626#if 0
627	printf("CTL: Isc Msg event %d\n", event);
628#endif
629	if (event == CTL_HA_EVT_MSG_RECV) {
630		union ctl_ha_msg msg_info;
631
632		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
633					     sizeof(msg_info), /*wait*/ 0);
634#if 0
635		printf("CTL: msg_type %d\n", msg_info.msg_type);
636#endif
637		if (isc_status != 0) {
638			printf("Error receiving message, status = %d\n",
639			       isc_status);
640			return;
641		}
642
643		switch (msg_info.hdr.msg_type) {
644		case CTL_MSG_SERIALIZE:
645#if 0
646			printf("Serialize\n");
647#endif
648			io = ctl_alloc_io_nowait(ctl_softc->othersc_pool);
649			if (io == NULL) {
650				printf("ctl_isc_event_handler: can't allocate "
651				       "ctl_io!\n");
652				/* Bad Juju */
653				/* Need to set busy and send msg back */
654				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
655				msg_info.hdr.status = CTL_SCSI_ERROR;
656				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
657				msg_info.scsi.sense_len = 0;
658			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
659				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
660				}
661				goto bailout;
662			}
663			ctl_zero_io(io);
664			// populate ctsio from msg_info
665			io->io_hdr.io_type = CTL_IO_SCSI;
666			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
667			io->io_hdr.original_sc = msg_info.hdr.original_sc;
668#if 0
669			printf("pOrig %x\n", (int)msg_info.original_sc);
670#endif
671			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
672					    CTL_FLAG_IO_ACTIVE;
673			/*
674			 * If we're in serialization-only mode, we don't
675			 * want to go through full done processing.  Thus
676			 * the COPY flag.
677			 *
678			 * XXX KDM add another flag that is more specific.
679			 */
680			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
681				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
682			io->io_hdr.nexus = msg_info.hdr.nexus;
683#if 0
684			printf("targ %d, port %d, iid %d, lun %d\n",
685			       io->io_hdr.nexus.targ_target.id,
686			       io->io_hdr.nexus.targ_port,
687			       io->io_hdr.nexus.initid.id,
688			       io->io_hdr.nexus.targ_lun);
689#endif
690			io->scsiio.tag_num = msg_info.scsi.tag_num;
691			io->scsiio.tag_type = msg_info.scsi.tag_type;
692			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
693			       CTL_MAX_CDBLEN);
694			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
695				const struct ctl_cmd_entry *entry;
696
697				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
698				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
699				io->io_hdr.flags |=
700					entry->flags & CTL_FLAG_DATA_MASK;
701			}
702			ctl_enqueue_isc(io);
703			break;
704
705		/* Performed on the Originating SC, XFER mode only */
706		case CTL_MSG_DATAMOVE: {
707			struct ctl_sg_entry *sgl;
708			int i, j;
709
710			io = msg_info.hdr.original_sc;
711			if (io == NULL) {
712				printf("%s: original_sc == NULL!\n", __func__);
713				/* XXX KDM do something here */
714				break;
715			}
716			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
717			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
718			/*
719			 * Keep track of this, we need to send it back over
720			 * when the datamove is complete.
721			 */
722			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
723
724			if (msg_info.dt.sg_sequence == 0) {
725				/*
726				 * XXX KDM we use the preallocated S/G list
727				 * here, but we'll need to change this to
728				 * dynamic allocation if we need larger S/G
729				 * lists.
730				 */
731				if (msg_info.dt.kern_sg_entries >
732				    sizeof(io->io_hdr.remote_sglist) /
733				    sizeof(io->io_hdr.remote_sglist[0])) {
734					printf("%s: number of S/G entries "
735					    "needed %u > allocated num %zd\n",
736					    __func__,
737					    msg_info.dt.kern_sg_entries,
738					    sizeof(io->io_hdr.remote_sglist)/
739					    sizeof(io->io_hdr.remote_sglist[0]));
740
741					/*
742					 * XXX KDM send a message back to
743					 * the other side to shut down the
744					 * DMA.  The error will come back
745					 * through via the normal channel.
746					 */
747					break;
748				}
749				sgl = io->io_hdr.remote_sglist;
750				memset(sgl, 0,
751				       sizeof(io->io_hdr.remote_sglist));
752
753				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
754
755				io->scsiio.kern_sg_entries =
756					msg_info.dt.kern_sg_entries;
757				io->scsiio.rem_sg_entries =
758					msg_info.dt.kern_sg_entries;
759				io->scsiio.kern_data_len =
760					msg_info.dt.kern_data_len;
761				io->scsiio.kern_total_len =
762					msg_info.dt.kern_total_len;
763				io->scsiio.kern_data_resid =
764					msg_info.dt.kern_data_resid;
765				io->scsiio.kern_rel_offset =
766					msg_info.dt.kern_rel_offset;
767				/*
768				 * Clear out per-DMA flags.
769				 */
770				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
771				/*
772				 * Add per-DMA flags that are set for this
773				 * particular DMA request.
774				 */
775				io->io_hdr.flags |= msg_info.dt.flags &
776						    CTL_FLAG_RDMA_MASK;
777			} else
778				sgl = (struct ctl_sg_entry *)
779					io->scsiio.kern_data_ptr;
780
781			for (i = msg_info.dt.sent_sg_entries, j = 0;
782			     i < (msg_info.dt.sent_sg_entries +
783			     msg_info.dt.cur_sg_entries); i++, j++) {
784				sgl[i].addr = msg_info.dt.sg_list[j].addr;
785				sgl[i].len = msg_info.dt.sg_list[j].len;
786
787#if 0
788				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
789				       __func__,
790				       msg_info.dt.sg_list[j].addr,
791				       msg_info.dt.sg_list[j].len,
792				       sgl[i].addr, sgl[i].len, j, i);
793#endif
794			}
795#if 0
796			memcpy(&sgl[msg_info.dt.sent_sg_entries],
797			       msg_info.dt.sg_list,
798			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
799#endif
800
801			/*
802			 * If this is the last piece of the I/O, we've got
803			 * the full S/G list.  Queue processing in the thread.
804			 * Otherwise wait for the next piece.
805			 */
806			if (msg_info.dt.sg_last != 0)
807				ctl_enqueue_isc(io);
808			break;
809		}
810		/* Performed on the Serializing (primary) SC, XFER mode only */
811		case CTL_MSG_DATAMOVE_DONE: {
812			if (msg_info.hdr.serializing_sc == NULL) {
813				printf("%s: serializing_sc == NULL!\n",
814				       __func__);
815				/* XXX KDM now what? */
816				break;
817			}
818			/*
819			 * We grab the sense information here in case
820			 * there was a failure, so we can return status
821			 * back to the initiator.
822			 */
823			io = msg_info.hdr.serializing_sc;
824			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
825			io->io_hdr.status = msg_info.hdr.status;
826			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
827			io->scsiio.sense_len = msg_info.scsi.sense_len;
828			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
829			io->io_hdr.port_status = msg_info.scsi.fetd_status;
830			io->scsiio.residual = msg_info.scsi.residual;
831			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
832			       sizeof(io->scsiio.sense_data));
833			ctl_enqueue_isc(io);
834			break;
835		}
836
837		/* Preformed on Originating SC, SER_ONLY mode */
838		case CTL_MSG_R2R:
839			io = msg_info.hdr.original_sc;
840			if (io == NULL) {
841				printf("%s: Major Bummer\n", __func__);
842				return;
843			} else {
844#if 0
845				printf("pOrig %x\n",(int) ctsio);
846#endif
847			}
848			io->io_hdr.msg_type = CTL_MSG_R2R;
849			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
850			ctl_enqueue_isc(io);
851			break;
852
853		/*
854		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
855		 * mode.
856		 * Performed on the Originating (i.e. secondary) SC in XFER
857		 * mode
858		 */
859		case CTL_MSG_FINISH_IO:
860			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
861				ctl_isc_handler_finish_xfer(ctl_softc,
862							    &msg_info);
863			else
864				ctl_isc_handler_finish_ser_only(ctl_softc,
865								&msg_info);
866			break;
867
868		/* Preformed on Originating SC */
869		case CTL_MSG_BAD_JUJU:
870			io = msg_info.hdr.original_sc;
871			if (io == NULL) {
872				printf("%s: Bad JUJU!, original_sc is NULL!\n",
873				       __func__);
874				break;
875			}
876			ctl_copy_sense_data(&msg_info, io);
877			/*
878			 * IO should have already been cleaned up on other
879			 * SC so clear this flag so we won't send a message
880			 * back to finish the IO there.
881			 */
882			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
883			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
884
885			/* io = msg_info.hdr.serializing_sc; */
886			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
887			ctl_enqueue_isc(io);
888			break;
889
890		/* Handle resets sent from the other side */
891		case CTL_MSG_MANAGE_TASKS: {
892			struct ctl_taskio *taskio;
893			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
894			    ctl_softc->othersc_pool);
895			if (taskio == NULL) {
896				printf("ctl_isc_event_handler: can't allocate "
897				       "ctl_io!\n");
898				/* Bad Juju */
899				/* should I just call the proper reset func
900				   here??? */
901				goto bailout;
902			}
903			ctl_zero_io((union ctl_io *)taskio);
904			taskio->io_hdr.io_type = CTL_IO_TASK;
905			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
906			taskio->io_hdr.nexus = msg_info.hdr.nexus;
907			taskio->task_action = msg_info.task.task_action;
908			taskio->tag_num = msg_info.task.tag_num;
909			taskio->tag_type = msg_info.task.tag_type;
910#ifdef CTL_TIME_IO
911			taskio->io_hdr.start_time = time_uptime;
912			getbintime(&taskio->io_hdr.start_bt);
913#if 0
914			cs_prof_gettime(&taskio->io_hdr.start_ticks);
915#endif
916#endif /* CTL_TIME_IO */
917			ctl_run_task((union ctl_io *)taskio);
918			break;
919		}
920		/* Persistent Reserve action which needs attention */
921		case CTL_MSG_PERS_ACTION:
922			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
923			    ctl_softc->othersc_pool);
924			if (presio == NULL) {
925				printf("ctl_isc_event_handler: can't allocate "
926				       "ctl_io!\n");
927				/* Bad Juju */
928				/* Need to set busy and send msg back */
929				goto bailout;
930			}
931			ctl_zero_io((union ctl_io *)presio);
932			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
933			presio->pr_msg = msg_info.pr;
934			ctl_enqueue_isc((union ctl_io *)presio);
935			break;
936		case CTL_MSG_SYNC_FE:
937			rcv_sync_msg = 1;
938			break;
939		default:
940		        printf("How did I get here?\n");
941		}
942	} else if (event == CTL_HA_EVT_MSG_SENT) {
943		if (param != CTL_HA_STATUS_SUCCESS) {
944			printf("Bad status from ctl_ha_msg_send status %d\n",
945			       param);
946		}
947		return;
948	} else if (event == CTL_HA_EVT_DISCONNECT) {
949		printf("CTL: Got a disconnect from Isc\n");
950		return;
951	} else {
952		printf("ctl_isc_event_handler: Unknown event %d\n", event);
953		return;
954	}
955
956bailout:
957	return;
958}
959
960static void
961ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
962{
963	struct scsi_sense_data *sense;
964
965	sense = &dest->scsiio.sense_data;
966	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
967	dest->scsiio.scsi_status = src->scsi.scsi_status;
968	dest->scsiio.sense_len = src->scsi.sense_len;
969	dest->io_hdr.status = src->hdr.status;
970}
971
972static int
973ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
974{
975	struct ctl_softc *softc = (struct ctl_softc *)arg1;
976	struct ctl_lun *lun;
977	int error, value, i;
978
979	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
980		value = 0;
981	else
982		value = 1;
983
984	error = sysctl_handle_int(oidp, &value, 0, req);
985	if ((error != 0) || (req->newptr == NULL))
986		return (error);
987
988	mtx_lock(&softc->ctl_lock);
989	if (value == 0)
990		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
991	else
992		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
993	STAILQ_FOREACH(lun, &softc->lun_list, links) {
994		mtx_lock(&lun->lun_lock);
995		for (i = 0; i < CTL_MAX_INITIATORS; i++)
996			lun->pending_ua[i] |= CTL_UA_ASYM_ACC_CHANGE;
997		mtx_unlock(&lun->lun_lock);
998	}
999	mtx_unlock(&softc->ctl_lock);
1000	return (0);
1001}
1002
1003static int
1004ctl_init(void)
1005{
1006	struct ctl_softc *softc;
1007	void *other_pool;
1008	struct ctl_port *port;
1009	int i, error, retval;
1010	//int isc_retval;
1011
1012	retval = 0;
1013	ctl_pause_rtr = 0;
1014        rcv_sync_msg = 0;
1015
1016	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1017			       M_WAITOK | M_ZERO);
1018	softc = control_softc;
1019
1020	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1021			      "cam/ctl");
1022
1023	softc->dev->si_drv1 = softc;
1024
1025	/*
1026	 * By default, return a "bad LUN" peripheral qualifier for unknown
1027	 * LUNs.  The user can override this default using the tunable or
1028	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1029	 */
1030	softc->inquiry_pq_no_lun = 1;
1031	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1032			  &softc->inquiry_pq_no_lun);
1033	sysctl_ctx_init(&softc->sysctl_ctx);
1034	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1035		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1036		CTLFLAG_RD, 0, "CAM Target Layer");
1037
1038	if (softc->sysctl_tree == NULL) {
1039		printf("%s: unable to allocate sysctl tree\n", __func__);
1040		destroy_dev(softc->dev);
1041		free(control_softc, M_DEVBUF);
1042		control_softc = NULL;
1043		return (ENOMEM);
1044	}
1045
1046	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1047		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1048		       "inquiry_pq_no_lun", CTLFLAG_RW,
1049		       &softc->inquiry_pq_no_lun, 0,
1050		       "Report no lun possible for invalid LUNs");
1051
1052	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1053	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1054	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1055	softc->open_count = 0;
1056
1057	/*
1058	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1059	 * the drive.
1060	 */
1061	softc->flags = CTL_FLAG_REAL_SYNC;
1062
1063	/*
1064	 * In Copan's HA scheme, the "master" and "slave" roles are
1065	 * figured out through the slot the controller is in.  Although it
1066	 * is an active/active system, someone has to be in charge.
1067	 */
1068	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1069	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1070	    "HA head ID (0 - no HA)");
1071	if (softc->ha_id == 0) {
1072		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1073		softc->is_single = 1;
1074		softc->port_offset = 0;
1075	} else
1076		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1077	persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1078
1079	/*
1080	 * XXX KDM need to figure out where we want to get our target ID
1081	 * and WWID.  Is it different on each port?
1082	 */
1083	softc->target.id = 0;
1084	softc->target.wwid[0] = 0x12345678;
1085	softc->target.wwid[1] = 0x87654321;
1086	STAILQ_INIT(&softc->lun_list);
1087	STAILQ_INIT(&softc->pending_lun_queue);
1088	STAILQ_INIT(&softc->fe_list);
1089	STAILQ_INIT(&softc->port_list);
1090	STAILQ_INIT(&softc->be_list);
1091	ctl_tpc_init(softc);
1092
1093	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1094	                    &other_pool) != 0)
1095	{
1096		printf("ctl: can't allocate %d entry other SC pool, "
1097		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1098		return (ENOMEM);
1099	}
1100	softc->othersc_pool = other_pool;
1101
1102	if (worker_threads <= 0)
1103		worker_threads = max(1, mp_ncpus / 4);
1104	if (worker_threads > CTL_MAX_THREADS)
1105		worker_threads = CTL_MAX_THREADS;
1106
1107	for (i = 0; i < worker_threads; i++) {
1108		struct ctl_thread *thr = &softc->threads[i];
1109
1110		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1111		thr->ctl_softc = softc;
1112		STAILQ_INIT(&thr->incoming_queue);
1113		STAILQ_INIT(&thr->rtr_queue);
1114		STAILQ_INIT(&thr->done_queue);
1115		STAILQ_INIT(&thr->isc_queue);
1116
1117		error = kproc_kthread_add(ctl_work_thread, thr,
1118		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1119		if (error != 0) {
1120			printf("error creating CTL work thread!\n");
1121			ctl_pool_free(other_pool);
1122			return (error);
1123		}
1124	}
1125	error = kproc_kthread_add(ctl_lun_thread, softc,
1126	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1127	if (error != 0) {
1128		printf("error creating CTL lun thread!\n");
1129		ctl_pool_free(other_pool);
1130		return (error);
1131	}
1132	error = kproc_kthread_add(ctl_thresh_thread, softc,
1133	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1134	if (error != 0) {
1135		printf("error creating CTL threshold thread!\n");
1136		ctl_pool_free(other_pool);
1137		return (error);
1138	}
1139	if (bootverbose)
1140		printf("ctl: CAM Target Layer loaded\n");
1141
1142	/*
1143	 * Initialize the ioctl front end.
1144	 */
1145	ctl_frontend_register(&ioctl_frontend);
1146	port = &softc->ioctl_info.port;
1147	port->frontend = &ioctl_frontend;
1148	sprintf(softc->ioctl_info.port_name, "ioctl");
1149	port->port_type = CTL_PORT_IOCTL;
1150	port->num_requested_ctl_io = 100;
1151	port->port_name = softc->ioctl_info.port_name;
1152	port->port_online = ctl_ioctl_online;
1153	port->port_offline = ctl_ioctl_offline;
1154	port->onoff_arg = &softc->ioctl_info;
1155	port->lun_enable = ctl_ioctl_lun_enable;
1156	port->lun_disable = ctl_ioctl_lun_disable;
1157	port->targ_lun_arg = &softc->ioctl_info;
1158	port->fe_datamove = ctl_ioctl_datamove;
1159	port->fe_done = ctl_ioctl_done;
1160	port->max_targets = 15;
1161	port->max_target_id = 15;
1162
1163	if (ctl_port_register(&softc->ioctl_info.port) != 0) {
1164		printf("ctl: ioctl front end registration failed, will "
1165		       "continue anyway\n");
1166	}
1167
1168	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1169	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1170	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1171
1172#ifdef CTL_IO_DELAY
1173	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1174		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1175		       sizeof(struct callout), CTL_TIMER_BYTES);
1176		return (EINVAL);
1177	}
1178#endif /* CTL_IO_DELAY */
1179
1180	return (0);
1181}
1182
1183void
1184ctl_shutdown(void)
1185{
1186	struct ctl_softc *softc;
1187	struct ctl_lun *lun, *next_lun;
1188
1189	softc = (struct ctl_softc *)control_softc;
1190
1191	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1192		printf("ctl: ioctl front end deregistration failed\n");
1193
1194	mtx_lock(&softc->ctl_lock);
1195
1196	/*
1197	 * Free up each LUN.
1198	 */
1199	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1200		next_lun = STAILQ_NEXT(lun, links);
1201		ctl_free_lun(lun);
1202	}
1203
1204	mtx_unlock(&softc->ctl_lock);
1205
1206	ctl_frontend_deregister(&ioctl_frontend);
1207
1208#if 0
1209	ctl_shutdown_thread(softc->work_thread);
1210	mtx_destroy(&softc->queue_lock);
1211#endif
1212
1213	ctl_tpc_shutdown(softc);
1214	uma_zdestroy(softc->io_zone);
1215	mtx_destroy(&softc->ctl_lock);
1216
1217	destroy_dev(softc->dev);
1218
1219	sysctl_ctx_free(&softc->sysctl_ctx);
1220
1221	free(control_softc, M_DEVBUF);
1222	control_softc = NULL;
1223
1224	if (bootverbose)
1225		printf("ctl: CAM Target Layer unloaded\n");
1226}
1227
1228static int
1229ctl_module_event_handler(module_t mod, int what, void *arg)
1230{
1231
1232	switch (what) {
1233	case MOD_LOAD:
1234		return (ctl_init());
1235	case MOD_UNLOAD:
1236		return (EBUSY);
1237	default:
1238		return (EOPNOTSUPP);
1239	}
1240}
1241
1242/*
1243 * XXX KDM should we do some access checks here?  Bump a reference count to
1244 * prevent a CTL module from being unloaded while someone has it open?
1245 */
1246static int
1247ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1248{
1249	return (0);
1250}
1251
1252static int
1253ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1254{
1255	return (0);
1256}
1257
1258int
1259ctl_port_enable(ctl_port_type port_type)
1260{
1261	struct ctl_softc *softc = control_softc;
1262	struct ctl_port *port;
1263
1264	if (softc->is_single == 0) {
1265		union ctl_ha_msg msg_info;
1266		int isc_retval;
1267
1268#if 0
1269		printf("%s: HA mode, synchronizing frontend enable\n",
1270		        __func__);
1271#endif
1272		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1273	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1274		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1275			printf("Sync msg send error retval %d\n", isc_retval);
1276		}
1277		if (!rcv_sync_msg) {
1278			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1279			        sizeof(msg_info), 1);
1280		}
1281#if 0
1282        	printf("CTL:Frontend Enable\n");
1283	} else {
1284		printf("%s: single mode, skipping frontend synchronization\n",
1285		        __func__);
1286#endif
1287	}
1288
1289	STAILQ_FOREACH(port, &softc->port_list, links) {
1290		if (port_type & port->port_type)
1291		{
1292#if 0
1293			printf("port %d\n", port->targ_port);
1294#endif
1295			ctl_port_online(port);
1296		}
1297	}
1298
1299	return (0);
1300}
1301
1302int
1303ctl_port_disable(ctl_port_type port_type)
1304{
1305	struct ctl_softc *softc;
1306	struct ctl_port *port;
1307
1308	softc = control_softc;
1309
1310	STAILQ_FOREACH(port, &softc->port_list, links) {
1311		if (port_type & port->port_type)
1312			ctl_port_offline(port);
1313	}
1314
1315	return (0);
1316}
1317
1318/*
1319 * Returns 0 for success, 1 for failure.
1320 * Currently the only failure mode is if there aren't enough entries
1321 * allocated.  So, in case of a failure, look at num_entries_dropped,
1322 * reallocate and try again.
1323 */
1324int
1325ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1326	      int *num_entries_filled, int *num_entries_dropped,
1327	      ctl_port_type port_type, int no_virtual)
1328{
1329	struct ctl_softc *softc;
1330	struct ctl_port *port;
1331	int entries_dropped, entries_filled;
1332	int retval;
1333	int i;
1334
1335	softc = control_softc;
1336
1337	retval = 0;
1338	entries_filled = 0;
1339	entries_dropped = 0;
1340
1341	i = 0;
1342	mtx_lock(&softc->ctl_lock);
1343	STAILQ_FOREACH(port, &softc->port_list, links) {
1344		struct ctl_port_entry *entry;
1345
1346		if ((port->port_type & port_type) == 0)
1347			continue;
1348
1349		if ((no_virtual != 0)
1350		 && (port->virtual_port != 0))
1351			continue;
1352
1353		if (entries_filled >= num_entries_alloced) {
1354			entries_dropped++;
1355			continue;
1356		}
1357		entry = &entries[i];
1358
1359		entry->port_type = port->port_type;
1360		strlcpy(entry->port_name, port->port_name,
1361			sizeof(entry->port_name));
1362		entry->physical_port = port->physical_port;
1363		entry->virtual_port = port->virtual_port;
1364		entry->wwnn = port->wwnn;
1365		entry->wwpn = port->wwpn;
1366
1367		i++;
1368		entries_filled++;
1369	}
1370
1371	mtx_unlock(&softc->ctl_lock);
1372
1373	if (entries_dropped > 0)
1374		retval = 1;
1375
1376	*num_entries_dropped = entries_dropped;
1377	*num_entries_filled = entries_filled;
1378
1379	return (retval);
1380}
1381
1382static void
1383ctl_ioctl_online(void *arg)
1384{
1385	struct ctl_ioctl_info *ioctl_info;
1386
1387	ioctl_info = (struct ctl_ioctl_info *)arg;
1388
1389	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1390}
1391
1392static void
1393ctl_ioctl_offline(void *arg)
1394{
1395	struct ctl_ioctl_info *ioctl_info;
1396
1397	ioctl_info = (struct ctl_ioctl_info *)arg;
1398
1399	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1400}
1401
1402/*
1403 * Remove an initiator by port number and initiator ID.
1404 * Returns 0 for success, -1 for failure.
1405 */
1406int
1407ctl_remove_initiator(struct ctl_port *port, int iid)
1408{
1409	struct ctl_softc *softc = control_softc;
1410
1411	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1412
1413	if (iid > CTL_MAX_INIT_PER_PORT) {
1414		printf("%s: initiator ID %u > maximun %u!\n",
1415		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1416		return (-1);
1417	}
1418
1419	mtx_lock(&softc->ctl_lock);
1420	port->wwpn_iid[iid].in_use--;
1421	port->wwpn_iid[iid].last_use = time_uptime;
1422	mtx_unlock(&softc->ctl_lock);
1423
1424	return (0);
1425}
1426
1427/*
1428 * Add an initiator to the initiator map.
1429 * Returns iid for success, < 0 for failure.
1430 */
1431int
1432ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1433{
1434	struct ctl_softc *softc = control_softc;
1435	time_t best_time;
1436	int i, best;
1437
1438	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1439
1440	if (iid >= CTL_MAX_INIT_PER_PORT) {
1441		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1442		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1443		free(name, M_CTL);
1444		return (-1);
1445	}
1446
1447	mtx_lock(&softc->ctl_lock);
1448
1449	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1450		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1451			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1452				iid = i;
1453				break;
1454			}
1455			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1456			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1457				iid = i;
1458				break;
1459			}
1460		}
1461	}
1462
1463	if (iid < 0) {
1464		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1465			if (port->wwpn_iid[i].in_use == 0 &&
1466			    port->wwpn_iid[i].wwpn == 0 &&
1467			    port->wwpn_iid[i].name == NULL) {
1468				iid = i;
1469				break;
1470			}
1471		}
1472	}
1473
1474	if (iid < 0) {
1475		best = -1;
1476		best_time = INT32_MAX;
1477		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1478			if (port->wwpn_iid[i].in_use == 0) {
1479				if (port->wwpn_iid[i].last_use < best_time) {
1480					best = i;
1481					best_time = port->wwpn_iid[i].last_use;
1482				}
1483			}
1484		}
1485		iid = best;
1486	}
1487
1488	if (iid < 0) {
1489		mtx_unlock(&softc->ctl_lock);
1490		free(name, M_CTL);
1491		return (-2);
1492	}
1493
1494	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1495		/*
1496		 * This is not an error yet.
1497		 */
1498		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1499#if 0
1500			printf("%s: port %d iid %u WWPN %#jx arrived"
1501			    " again\n", __func__, port->targ_port,
1502			    iid, (uintmax_t)wwpn);
1503#endif
1504			goto take;
1505		}
1506		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1507		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1508#if 0
1509			printf("%s: port %d iid %u name '%s' arrived"
1510			    " again\n", __func__, port->targ_port,
1511			    iid, name);
1512#endif
1513			goto take;
1514		}
1515
1516		/*
1517		 * This is an error, but what do we do about it?  The
1518		 * driver is telling us we have a new WWPN for this
1519		 * initiator ID, so we pretty much need to use it.
1520		 */
1521		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1522		    " but WWPN %#jx '%s' is still at that address\n",
1523		    __func__, port->targ_port, iid, wwpn, name,
1524		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1525		    port->wwpn_iid[iid].name);
1526
1527		/*
1528		 * XXX KDM clear have_ca and ua_pending on each LUN for
1529		 * this initiator.
1530		 */
1531	}
1532take:
1533	free(port->wwpn_iid[iid].name, M_CTL);
1534	port->wwpn_iid[iid].name = name;
1535	port->wwpn_iid[iid].wwpn = wwpn;
1536	port->wwpn_iid[iid].in_use++;
1537	mtx_unlock(&softc->ctl_lock);
1538
1539	return (iid);
1540}
1541
1542static int
1543ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1544{
1545	int len;
1546
1547	switch (port->port_type) {
1548	case CTL_PORT_FC:
1549	{
1550		struct scsi_transportid_fcp *id =
1551		    (struct scsi_transportid_fcp *)buf;
1552		if (port->wwpn_iid[iid].wwpn == 0)
1553			return (0);
1554		memset(id, 0, sizeof(*id));
1555		id->format_protocol = SCSI_PROTO_FC;
1556		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1557		return (sizeof(*id));
1558	}
1559	case CTL_PORT_ISCSI:
1560	{
1561		struct scsi_transportid_iscsi_port *id =
1562		    (struct scsi_transportid_iscsi_port *)buf;
1563		if (port->wwpn_iid[iid].name == NULL)
1564			return (0);
1565		memset(id, 0, 256);
1566		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1567		    SCSI_PROTO_ISCSI;
1568		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1569		len = roundup2(min(len, 252), 4);
1570		scsi_ulto2b(len, id->additional_length);
1571		return (sizeof(*id) + len);
1572	}
1573	case CTL_PORT_SAS:
1574	{
1575		struct scsi_transportid_sas *id =
1576		    (struct scsi_transportid_sas *)buf;
1577		if (port->wwpn_iid[iid].wwpn == 0)
1578			return (0);
1579		memset(id, 0, sizeof(*id));
1580		id->format_protocol = SCSI_PROTO_SAS;
1581		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1582		return (sizeof(*id));
1583	}
1584	default:
1585	{
1586		struct scsi_transportid_spi *id =
1587		    (struct scsi_transportid_spi *)buf;
1588		memset(id, 0, sizeof(*id));
1589		id->format_protocol = SCSI_PROTO_SPI;
1590		scsi_ulto2b(iid, id->scsi_addr);
1591		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1592		return (sizeof(*id));
1593	}
1594	}
1595}
1596
1597static int
1598ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1599{
1600	return (0);
1601}
1602
1603static int
1604ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1605{
1606	return (0);
1607}
1608
1609/*
1610 * Data movement routine for the CTL ioctl frontend port.
1611 */
1612static int
1613ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1614{
1615	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1616	struct ctl_sg_entry ext_entry, kern_entry;
1617	int ext_sglen, ext_sg_entries, kern_sg_entries;
1618	int ext_sg_start, ext_offset;
1619	int len_to_copy, len_copied;
1620	int kern_watermark, ext_watermark;
1621	int ext_sglist_malloced;
1622	int i, j;
1623
1624	ext_sglist_malloced = 0;
1625	ext_sg_start = 0;
1626	ext_offset = 0;
1627
1628	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1629
1630	/*
1631	 * If this flag is set, fake the data transfer.
1632	 */
1633	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1634		ctsio->ext_data_filled = ctsio->ext_data_len;
1635		goto bailout;
1636	}
1637
1638	/*
1639	 * To simplify things here, if we have a single buffer, stick it in
1640	 * a S/G entry and just make it a single entry S/G list.
1641	 */
1642	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1643		int len_seen;
1644
1645		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1646
1647		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1648							   M_WAITOK);
1649		ext_sglist_malloced = 1;
1650		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1651				   ext_sglen) != 0) {
1652			ctl_set_internal_failure(ctsio,
1653						 /*sks_valid*/ 0,
1654						 /*retry_count*/ 0);
1655			goto bailout;
1656		}
1657		ext_sg_entries = ctsio->ext_sg_entries;
1658		len_seen = 0;
1659		for (i = 0; i < ext_sg_entries; i++) {
1660			if ((len_seen + ext_sglist[i].len) >=
1661			     ctsio->ext_data_filled) {
1662				ext_sg_start = i;
1663				ext_offset = ctsio->ext_data_filled - len_seen;
1664				break;
1665			}
1666			len_seen += ext_sglist[i].len;
1667		}
1668	} else {
1669		ext_sglist = &ext_entry;
1670		ext_sglist->addr = ctsio->ext_data_ptr;
1671		ext_sglist->len = ctsio->ext_data_len;
1672		ext_sg_entries = 1;
1673		ext_sg_start = 0;
1674		ext_offset = ctsio->ext_data_filled;
1675	}
1676
1677	if (ctsio->kern_sg_entries > 0) {
1678		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1679		kern_sg_entries = ctsio->kern_sg_entries;
1680	} else {
1681		kern_sglist = &kern_entry;
1682		kern_sglist->addr = ctsio->kern_data_ptr;
1683		kern_sglist->len = ctsio->kern_data_len;
1684		kern_sg_entries = 1;
1685	}
1686
1687
1688	kern_watermark = 0;
1689	ext_watermark = ext_offset;
1690	len_copied = 0;
1691	for (i = ext_sg_start, j = 0;
1692	     i < ext_sg_entries && j < kern_sg_entries;) {
1693		uint8_t *ext_ptr, *kern_ptr;
1694
1695		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1696				      kern_sglist[j].len - kern_watermark);
1697
1698		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1699		ext_ptr = ext_ptr + ext_watermark;
1700		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1701			/*
1702			 * XXX KDM fix this!
1703			 */
1704			panic("need to implement bus address support");
1705#if 0
1706			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1707#endif
1708		} else
1709			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1710		kern_ptr = kern_ptr + kern_watermark;
1711
1712		kern_watermark += len_to_copy;
1713		ext_watermark += len_to_copy;
1714
1715		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1716		     CTL_FLAG_DATA_IN) {
1717			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1718					 "bytes to user\n", len_to_copy));
1719			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1720					 "to %p\n", kern_ptr, ext_ptr));
1721			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1722				ctl_set_internal_failure(ctsio,
1723							 /*sks_valid*/ 0,
1724							 /*retry_count*/ 0);
1725				goto bailout;
1726			}
1727		} else {
1728			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1729					 "bytes from user\n", len_to_copy));
1730			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1731					 "to %p\n", ext_ptr, kern_ptr));
1732			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1733				ctl_set_internal_failure(ctsio,
1734							 /*sks_valid*/ 0,
1735							 /*retry_count*/0);
1736				goto bailout;
1737			}
1738		}
1739
1740		len_copied += len_to_copy;
1741
1742		if (ext_sglist[i].len == ext_watermark) {
1743			i++;
1744			ext_watermark = 0;
1745		}
1746
1747		if (kern_sglist[j].len == kern_watermark) {
1748			j++;
1749			kern_watermark = 0;
1750		}
1751	}
1752
1753	ctsio->ext_data_filled += len_copied;
1754
1755	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1756			 "kern_sg_entries: %d\n", ext_sg_entries,
1757			 kern_sg_entries));
1758	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1759			 "kern_data_len = %d\n", ctsio->ext_data_len,
1760			 ctsio->kern_data_len));
1761
1762
1763	/* XXX KDM set residual?? */
1764bailout:
1765
1766	if (ext_sglist_malloced != 0)
1767		free(ext_sglist, M_CTL);
1768
1769	return (CTL_RETVAL_COMPLETE);
1770}
1771
1772/*
1773 * Serialize a command that went down the "wrong" side, and so was sent to
1774 * this controller for execution.  The logic is a little different than the
1775 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1776 * sent back to the other side, but in the success case, we execute the
1777 * command on this side (XFER mode) or tell the other side to execute it
1778 * (SER_ONLY mode).
1779 */
1780static int
1781ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1782{
1783	struct ctl_softc *ctl_softc;
1784	union ctl_ha_msg msg_info;
1785	struct ctl_lun *lun;
1786	int retval = 0;
1787	uint32_t targ_lun;
1788
1789	ctl_softc = control_softc;
1790
1791	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1792	lun = ctl_softc->ctl_luns[targ_lun];
1793	if (lun==NULL)
1794	{
1795		/*
1796		 * Why isn't LUN defined? The other side wouldn't
1797		 * send a cmd if the LUN is undefined.
1798		 */
1799		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1800
1801		/* "Logical unit not supported" */
1802		ctl_set_sense_data(&msg_info.scsi.sense_data,
1803				   lun,
1804				   /*sense_format*/SSD_TYPE_NONE,
1805				   /*current_error*/ 1,
1806				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1807				   /*asc*/ 0x25,
1808				   /*ascq*/ 0x00,
1809				   SSD_ELEM_NONE);
1810
1811		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1812		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1813		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1814		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1815		msg_info.hdr.serializing_sc = NULL;
1816		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1817	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1818				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1819		}
1820		return(1);
1821
1822	}
1823
1824	mtx_lock(&lun->lun_lock);
1825    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1826
1827	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1828		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1829		 ooa_links))) {
1830	case CTL_ACTION_BLOCK:
1831		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1832		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1833				  blocked_links);
1834		break;
1835	case CTL_ACTION_PASS:
1836	case CTL_ACTION_SKIP:
1837		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1838			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1839			ctl_enqueue_rtr((union ctl_io *)ctsio);
1840		} else {
1841
1842			/* send msg back to other side */
1843			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1844			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1845			msg_info.hdr.msg_type = CTL_MSG_R2R;
1846#if 0
1847			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1848#endif
1849		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1850			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1851			}
1852		}
1853		break;
1854	case CTL_ACTION_OVERLAP:
1855		/* OVERLAPPED COMMANDS ATTEMPTED */
1856		ctl_set_sense_data(&msg_info.scsi.sense_data,
1857				   lun,
1858				   /*sense_format*/SSD_TYPE_NONE,
1859				   /*current_error*/ 1,
1860				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1861				   /*asc*/ 0x4E,
1862				   /*ascq*/ 0x00,
1863				   SSD_ELEM_NONE);
1864
1865		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1866		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1867		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1868		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1869		msg_info.hdr.serializing_sc = NULL;
1870		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1871#if 0
1872		printf("BAD JUJU:Major Bummer Overlap\n");
1873#endif
1874		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1875		retval = 1;
1876		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1877		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1878		}
1879		break;
1880	case CTL_ACTION_OVERLAP_TAG:
1881		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1882		ctl_set_sense_data(&msg_info.scsi.sense_data,
1883				   lun,
1884				   /*sense_format*/SSD_TYPE_NONE,
1885				   /*current_error*/ 1,
1886				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1887				   /*asc*/ 0x4D,
1888				   /*ascq*/ ctsio->tag_num & 0xff,
1889				   SSD_ELEM_NONE);
1890
1891		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1892		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1893		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1894		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1895		msg_info.hdr.serializing_sc = NULL;
1896		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1897#if 0
1898		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1899#endif
1900		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1901		retval = 1;
1902		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1903		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1904		}
1905		break;
1906	case CTL_ACTION_ERROR:
1907	default:
1908		/* "Internal target failure" */
1909		ctl_set_sense_data(&msg_info.scsi.sense_data,
1910				   lun,
1911				   /*sense_format*/SSD_TYPE_NONE,
1912				   /*current_error*/ 1,
1913				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1914				   /*asc*/ 0x44,
1915				   /*ascq*/ 0x00,
1916				   SSD_ELEM_NONE);
1917
1918		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1919		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1920		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1921		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1922		msg_info.hdr.serializing_sc = NULL;
1923		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1924#if 0
1925		printf("BAD JUJU:Major Bummer HW Error\n");
1926#endif
1927		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1928		retval = 1;
1929		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1930		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1931		}
1932		break;
1933	}
1934	mtx_unlock(&lun->lun_lock);
1935	return (retval);
1936}
1937
1938static int
1939ctl_ioctl_submit_wait(union ctl_io *io)
1940{
1941	struct ctl_fe_ioctl_params params;
1942	ctl_fe_ioctl_state last_state;
1943	int done, retval;
1944
1945	retval = 0;
1946
1947	bzero(&params, sizeof(params));
1948
1949	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1950	cv_init(&params.sem, "ctlioccv");
1951	params.state = CTL_IOCTL_INPROG;
1952	last_state = params.state;
1953
1954	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1955
1956	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1957
1958	/* This shouldn't happen */
1959	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1960		return (retval);
1961
1962	done = 0;
1963
1964	do {
1965		mtx_lock(&params.ioctl_mtx);
1966		/*
1967		 * Check the state here, and don't sleep if the state has
1968		 * already changed (i.e. wakeup has already occured, but we
1969		 * weren't waiting yet).
1970		 */
1971		if (params.state == last_state) {
1972			/* XXX KDM cv_wait_sig instead? */
1973			cv_wait(&params.sem, &params.ioctl_mtx);
1974		}
1975		last_state = params.state;
1976
1977		switch (params.state) {
1978		case CTL_IOCTL_INPROG:
1979			/* Why did we wake up? */
1980			/* XXX KDM error here? */
1981			mtx_unlock(&params.ioctl_mtx);
1982			break;
1983		case CTL_IOCTL_DATAMOVE:
1984			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1985
1986			/*
1987			 * change last_state back to INPROG to avoid
1988			 * deadlock on subsequent data moves.
1989			 */
1990			params.state = last_state = CTL_IOCTL_INPROG;
1991
1992			mtx_unlock(&params.ioctl_mtx);
1993			ctl_ioctl_do_datamove(&io->scsiio);
1994			/*
1995			 * Note that in some cases, most notably writes,
1996			 * this will queue the I/O and call us back later.
1997			 * In other cases, generally reads, this routine
1998			 * will immediately call back and wake us up,
1999			 * probably using our own context.
2000			 */
2001			io->scsiio.be_move_done(io);
2002			break;
2003		case CTL_IOCTL_DONE:
2004			mtx_unlock(&params.ioctl_mtx);
2005			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
2006			done = 1;
2007			break;
2008		default:
2009			mtx_unlock(&params.ioctl_mtx);
2010			/* XXX KDM error here? */
2011			break;
2012		}
2013	} while (done == 0);
2014
2015	mtx_destroy(&params.ioctl_mtx);
2016	cv_destroy(&params.sem);
2017
2018	return (CTL_RETVAL_COMPLETE);
2019}
2020
2021static void
2022ctl_ioctl_datamove(union ctl_io *io)
2023{
2024	struct ctl_fe_ioctl_params *params;
2025
2026	params = (struct ctl_fe_ioctl_params *)
2027		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2028
2029	mtx_lock(&params->ioctl_mtx);
2030	params->state = CTL_IOCTL_DATAMOVE;
2031	cv_broadcast(&params->sem);
2032	mtx_unlock(&params->ioctl_mtx);
2033}
2034
2035static void
2036ctl_ioctl_done(union ctl_io *io)
2037{
2038	struct ctl_fe_ioctl_params *params;
2039
2040	params = (struct ctl_fe_ioctl_params *)
2041		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2042
2043	mtx_lock(&params->ioctl_mtx);
2044	params->state = CTL_IOCTL_DONE;
2045	cv_broadcast(&params->sem);
2046	mtx_unlock(&params->ioctl_mtx);
2047}
2048
2049static void
2050ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2051{
2052	struct ctl_fe_ioctl_startstop_info *sd_info;
2053
2054	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2055
2056	sd_info->hs_info.status = metatask->status;
2057	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2058	sd_info->hs_info.luns_complete =
2059		metatask->taskinfo.startstop.luns_complete;
2060	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2061
2062	cv_broadcast(&sd_info->sem);
2063}
2064
2065static void
2066ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2067{
2068	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2069
2070	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2071
2072	mtx_lock(fe_bbr_info->lock);
2073	fe_bbr_info->bbr_info->status = metatask->status;
2074	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2075	fe_bbr_info->wakeup_done = 1;
2076	mtx_unlock(fe_bbr_info->lock);
2077
2078	cv_broadcast(&fe_bbr_info->sem);
2079}
2080
2081/*
2082 * Returns 0 for success, errno for failure.
2083 */
2084static int
2085ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2086		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2087{
2088	union ctl_io *io;
2089	int retval;
2090
2091	retval = 0;
2092
2093	mtx_lock(&lun->lun_lock);
2094	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2095	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2096	     ooa_links)) {
2097		struct ctl_ooa_entry *entry;
2098
2099		/*
2100		 * If we've got more than we can fit, just count the
2101		 * remaining entries.
2102		 */
2103		if (*cur_fill_num >= ooa_hdr->alloc_num)
2104			continue;
2105
2106		entry = &kern_entries[*cur_fill_num];
2107
2108		entry->tag_num = io->scsiio.tag_num;
2109		entry->lun_num = lun->lun;
2110#ifdef CTL_TIME_IO
2111		entry->start_bt = io->io_hdr.start_bt;
2112#endif
2113		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2114		entry->cdb_len = io->scsiio.cdb_len;
2115		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2116			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2117
2118		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2119			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2120
2121		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2122			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2123
2124		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2125			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2126
2127		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2128			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2129	}
2130	mtx_unlock(&lun->lun_lock);
2131
2132	return (retval);
2133}
2134
2135static void *
2136ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2137		 size_t error_str_len)
2138{
2139	void *kptr;
2140
2141	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2142
2143	if (copyin(user_addr, kptr, len) != 0) {
2144		snprintf(error_str, error_str_len, "Error copying %d bytes "
2145			 "from user address %p to kernel address %p", len,
2146			 user_addr, kptr);
2147		free(kptr, M_CTL);
2148		return (NULL);
2149	}
2150
2151	return (kptr);
2152}
2153
2154static void
2155ctl_free_args(int num_args, struct ctl_be_arg *args)
2156{
2157	int i;
2158
2159	if (args == NULL)
2160		return;
2161
2162	for (i = 0; i < num_args; i++) {
2163		free(args[i].kname, M_CTL);
2164		free(args[i].kvalue, M_CTL);
2165	}
2166
2167	free(args, M_CTL);
2168}
2169
2170static struct ctl_be_arg *
2171ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2172		char *error_str, size_t error_str_len)
2173{
2174	struct ctl_be_arg *args;
2175	int i;
2176
2177	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2178				error_str, error_str_len);
2179
2180	if (args == NULL)
2181		goto bailout;
2182
2183	for (i = 0; i < num_args; i++) {
2184		args[i].kname = NULL;
2185		args[i].kvalue = NULL;
2186	}
2187
2188	for (i = 0; i < num_args; i++) {
2189		uint8_t *tmpptr;
2190
2191		args[i].kname = ctl_copyin_alloc(args[i].name,
2192			args[i].namelen, error_str, error_str_len);
2193		if (args[i].kname == NULL)
2194			goto bailout;
2195
2196		if (args[i].kname[args[i].namelen - 1] != '\0') {
2197			snprintf(error_str, error_str_len, "Argument %d "
2198				 "name is not NUL-terminated", i);
2199			goto bailout;
2200		}
2201
2202		if (args[i].flags & CTL_BEARG_RD) {
2203			tmpptr = ctl_copyin_alloc(args[i].value,
2204				args[i].vallen, error_str, error_str_len);
2205			if (tmpptr == NULL)
2206				goto bailout;
2207			if ((args[i].flags & CTL_BEARG_ASCII)
2208			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2209				snprintf(error_str, error_str_len, "Argument "
2210				    "%d value is not NUL-terminated", i);
2211				goto bailout;
2212			}
2213			args[i].kvalue = tmpptr;
2214		} else {
2215			args[i].kvalue = malloc(args[i].vallen,
2216			    M_CTL, M_WAITOK | M_ZERO);
2217		}
2218	}
2219
2220	return (args);
2221bailout:
2222
2223	ctl_free_args(num_args, args);
2224
2225	return (NULL);
2226}
2227
2228static void
2229ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2230{
2231	int i;
2232
2233	for (i = 0; i < num_args; i++) {
2234		if (args[i].flags & CTL_BEARG_WR)
2235			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2236	}
2237}
2238
2239/*
2240 * Escape characters that are illegal or not recommended in XML.
2241 */
2242int
2243ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2244{
2245	char *end = str + size;
2246	int retval;
2247
2248	retval = 0;
2249
2250	for (; *str && str < end; str++) {
2251		switch (*str) {
2252		case '&':
2253			retval = sbuf_printf(sb, "&amp;");
2254			break;
2255		case '>':
2256			retval = sbuf_printf(sb, "&gt;");
2257			break;
2258		case '<':
2259			retval = sbuf_printf(sb, "&lt;");
2260			break;
2261		default:
2262			retval = sbuf_putc(sb, *str);
2263			break;
2264		}
2265
2266		if (retval != 0)
2267			break;
2268
2269	}
2270
2271	return (retval);
2272}
2273
2274static void
2275ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2276{
2277	struct scsi_vpd_id_descriptor *desc;
2278	int i;
2279
2280	if (id == NULL || id->len < 4)
2281		return;
2282	desc = (struct scsi_vpd_id_descriptor *)id->data;
2283	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2284	case SVPD_ID_TYPE_T10:
2285		sbuf_printf(sb, "t10.");
2286		break;
2287	case SVPD_ID_TYPE_EUI64:
2288		sbuf_printf(sb, "eui.");
2289		break;
2290	case SVPD_ID_TYPE_NAA:
2291		sbuf_printf(sb, "naa.");
2292		break;
2293	case SVPD_ID_TYPE_SCSI_NAME:
2294		break;
2295	}
2296	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2297	case SVPD_ID_CODESET_BINARY:
2298		for (i = 0; i < desc->length; i++)
2299			sbuf_printf(sb, "%02x", desc->identifier[i]);
2300		break;
2301	case SVPD_ID_CODESET_ASCII:
2302		sbuf_printf(sb, "%.*s", (int)desc->length,
2303		    (char *)desc->identifier);
2304		break;
2305	case SVPD_ID_CODESET_UTF8:
2306		sbuf_printf(sb, "%s", (char *)desc->identifier);
2307		break;
2308	}
2309}
2310
2311static int
2312ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2313	  struct thread *td)
2314{
2315	struct ctl_softc *softc;
2316	int retval;
2317
2318	softc = control_softc;
2319
2320	retval = 0;
2321
2322	switch (cmd) {
2323	case CTL_IO: {
2324		union ctl_io *io;
2325		void *pool_tmp;
2326
2327		/*
2328		 * If we haven't been "enabled", don't allow any SCSI I/O
2329		 * to this FETD.
2330		 */
2331		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2332			retval = EPERM;
2333			break;
2334		}
2335
2336		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2337
2338		/*
2339		 * Need to save the pool reference so it doesn't get
2340		 * spammed by the user's ctl_io.
2341		 */
2342		pool_tmp = io->io_hdr.pool;
2343		memcpy(io, (void *)addr, sizeof(*io));
2344		io->io_hdr.pool = pool_tmp;
2345
2346		/*
2347		 * No status yet, so make sure the status is set properly.
2348		 */
2349		io->io_hdr.status = CTL_STATUS_NONE;
2350
2351		/*
2352		 * The user sets the initiator ID, target and LUN IDs.
2353		 */
2354		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2355		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2356		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2357		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2358			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2359
2360		retval = ctl_ioctl_submit_wait(io);
2361
2362		if (retval != 0) {
2363			ctl_free_io(io);
2364			break;
2365		}
2366
2367		memcpy((void *)addr, io, sizeof(*io));
2368
2369		/* return this to our pool */
2370		ctl_free_io(io);
2371
2372		break;
2373	}
2374	case CTL_ENABLE_PORT:
2375	case CTL_DISABLE_PORT:
2376	case CTL_SET_PORT_WWNS: {
2377		struct ctl_port *port;
2378		struct ctl_port_entry *entry;
2379
2380		entry = (struct ctl_port_entry *)addr;
2381
2382		mtx_lock(&softc->ctl_lock);
2383		STAILQ_FOREACH(port, &softc->port_list, links) {
2384			int action, done;
2385
2386			action = 0;
2387			done = 0;
2388
2389			if ((entry->port_type == CTL_PORT_NONE)
2390			 && (entry->targ_port == port->targ_port)) {
2391				/*
2392				 * If the user only wants to enable or
2393				 * disable or set WWNs on a specific port,
2394				 * do the operation and we're done.
2395				 */
2396				action = 1;
2397				done = 1;
2398			} else if (entry->port_type & port->port_type) {
2399				/*
2400				 * Compare the user's type mask with the
2401				 * particular frontend type to see if we
2402				 * have a match.
2403				 */
2404				action = 1;
2405				done = 0;
2406
2407				/*
2408				 * Make sure the user isn't trying to set
2409				 * WWNs on multiple ports at the same time.
2410				 */
2411				if (cmd == CTL_SET_PORT_WWNS) {
2412					printf("%s: Can't set WWNs on "
2413					       "multiple ports\n", __func__);
2414					retval = EINVAL;
2415					break;
2416				}
2417			}
2418			if (action != 0) {
2419				/*
2420				 * XXX KDM we have to drop the lock here,
2421				 * because the online/offline operations
2422				 * can potentially block.  We need to
2423				 * reference count the frontends so they
2424				 * can't go away,
2425				 */
2426				mtx_unlock(&softc->ctl_lock);
2427
2428				if (cmd == CTL_ENABLE_PORT) {
2429					struct ctl_lun *lun;
2430
2431					STAILQ_FOREACH(lun, &softc->lun_list,
2432						       links) {
2433						port->lun_enable(port->targ_lun_arg,
2434						    lun->target,
2435						    lun->lun);
2436					}
2437
2438					ctl_port_online(port);
2439				} else if (cmd == CTL_DISABLE_PORT) {
2440					struct ctl_lun *lun;
2441
2442					ctl_port_offline(port);
2443
2444					STAILQ_FOREACH(lun, &softc->lun_list,
2445						       links) {
2446						port->lun_disable(
2447						    port->targ_lun_arg,
2448						    lun->target,
2449						    lun->lun);
2450					}
2451				}
2452
2453				mtx_lock(&softc->ctl_lock);
2454
2455				if (cmd == CTL_SET_PORT_WWNS)
2456					ctl_port_set_wwns(port,
2457					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2458					    1 : 0, entry->wwnn,
2459					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2460					    1 : 0, entry->wwpn);
2461			}
2462			if (done != 0)
2463				break;
2464		}
2465		mtx_unlock(&softc->ctl_lock);
2466		break;
2467	}
2468	case CTL_GET_PORT_LIST: {
2469		struct ctl_port *port;
2470		struct ctl_port_list *list;
2471		int i;
2472
2473		list = (struct ctl_port_list *)addr;
2474
2475		if (list->alloc_len != (list->alloc_num *
2476		    sizeof(struct ctl_port_entry))) {
2477			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2478			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2479			       "%zu\n", __func__, list->alloc_len,
2480			       list->alloc_num, sizeof(struct ctl_port_entry));
2481			retval = EINVAL;
2482			break;
2483		}
2484		list->fill_len = 0;
2485		list->fill_num = 0;
2486		list->dropped_num = 0;
2487		i = 0;
2488		mtx_lock(&softc->ctl_lock);
2489		STAILQ_FOREACH(port, &softc->port_list, links) {
2490			struct ctl_port_entry entry, *list_entry;
2491
2492			if (list->fill_num >= list->alloc_num) {
2493				list->dropped_num++;
2494				continue;
2495			}
2496
2497			entry.port_type = port->port_type;
2498			strlcpy(entry.port_name, port->port_name,
2499				sizeof(entry.port_name));
2500			entry.targ_port = port->targ_port;
2501			entry.physical_port = port->physical_port;
2502			entry.virtual_port = port->virtual_port;
2503			entry.wwnn = port->wwnn;
2504			entry.wwpn = port->wwpn;
2505			if (port->status & CTL_PORT_STATUS_ONLINE)
2506				entry.online = 1;
2507			else
2508				entry.online = 0;
2509
2510			list_entry = &list->entries[i];
2511
2512			retval = copyout(&entry, list_entry, sizeof(entry));
2513			if (retval != 0) {
2514				printf("%s: CTL_GET_PORT_LIST: copyout "
2515				       "returned %d\n", __func__, retval);
2516				break;
2517			}
2518			i++;
2519			list->fill_num++;
2520			list->fill_len += sizeof(entry);
2521		}
2522		mtx_unlock(&softc->ctl_lock);
2523
2524		/*
2525		 * If this is non-zero, we had a copyout fault, so there's
2526		 * probably no point in attempting to set the status inside
2527		 * the structure.
2528		 */
2529		if (retval != 0)
2530			break;
2531
2532		if (list->dropped_num > 0)
2533			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2534		else
2535			list->status = CTL_PORT_LIST_OK;
2536		break;
2537	}
2538	case CTL_DUMP_OOA: {
2539		struct ctl_lun *lun;
2540		union ctl_io *io;
2541		char printbuf[128];
2542		struct sbuf sb;
2543
2544		mtx_lock(&softc->ctl_lock);
2545		printf("Dumping OOA queues:\n");
2546		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2547			mtx_lock(&lun->lun_lock);
2548			for (io = (union ctl_io *)TAILQ_FIRST(
2549			     &lun->ooa_queue); io != NULL;
2550			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2551			     ooa_links)) {
2552				sbuf_new(&sb, printbuf, sizeof(printbuf),
2553					 SBUF_FIXEDLEN);
2554				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2555					    (intmax_t)lun->lun,
2556					    io->scsiio.tag_num,
2557					    (io->io_hdr.flags &
2558					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2559					    (io->io_hdr.flags &
2560					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2561					    (io->io_hdr.flags &
2562					    CTL_FLAG_ABORT) ? " ABORT" : "",
2563			                    (io->io_hdr.flags &
2564		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2565				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2566				sbuf_finish(&sb);
2567				printf("%s\n", sbuf_data(&sb));
2568			}
2569			mtx_unlock(&lun->lun_lock);
2570		}
2571		printf("OOA queues dump done\n");
2572		mtx_unlock(&softc->ctl_lock);
2573		break;
2574	}
2575	case CTL_GET_OOA: {
2576		struct ctl_lun *lun;
2577		struct ctl_ooa *ooa_hdr;
2578		struct ctl_ooa_entry *entries;
2579		uint32_t cur_fill_num;
2580
2581		ooa_hdr = (struct ctl_ooa *)addr;
2582
2583		if ((ooa_hdr->alloc_len == 0)
2584		 || (ooa_hdr->alloc_num == 0)) {
2585			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2586			       "must be non-zero\n", __func__,
2587			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2588			retval = EINVAL;
2589			break;
2590		}
2591
2592		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2593		    sizeof(struct ctl_ooa_entry))) {
2594			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2595			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2596			       __func__, ooa_hdr->alloc_len,
2597			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2598			retval = EINVAL;
2599			break;
2600		}
2601
2602		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2603		if (entries == NULL) {
2604			printf("%s: could not allocate %d bytes for OOA "
2605			       "dump\n", __func__, ooa_hdr->alloc_len);
2606			retval = ENOMEM;
2607			break;
2608		}
2609
2610		mtx_lock(&softc->ctl_lock);
2611		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2612		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2613		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2614			mtx_unlock(&softc->ctl_lock);
2615			free(entries, M_CTL);
2616			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2617			       __func__, (uintmax_t)ooa_hdr->lun_num);
2618			retval = EINVAL;
2619			break;
2620		}
2621
2622		cur_fill_num = 0;
2623
2624		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2625			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2626				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2627					ooa_hdr, entries);
2628				if (retval != 0)
2629					break;
2630			}
2631			if (retval != 0) {
2632				mtx_unlock(&softc->ctl_lock);
2633				free(entries, M_CTL);
2634				break;
2635			}
2636		} else {
2637			lun = softc->ctl_luns[ooa_hdr->lun_num];
2638
2639			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2640						    entries);
2641		}
2642		mtx_unlock(&softc->ctl_lock);
2643
2644		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2645		ooa_hdr->fill_len = ooa_hdr->fill_num *
2646			sizeof(struct ctl_ooa_entry);
2647		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2648		if (retval != 0) {
2649			printf("%s: error copying out %d bytes for OOA dump\n",
2650			       __func__, ooa_hdr->fill_len);
2651		}
2652
2653		getbintime(&ooa_hdr->cur_bt);
2654
2655		if (cur_fill_num > ooa_hdr->alloc_num) {
2656			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2657			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2658		} else {
2659			ooa_hdr->dropped_num = 0;
2660			ooa_hdr->status = CTL_OOA_OK;
2661		}
2662
2663		free(entries, M_CTL);
2664		break;
2665	}
2666	case CTL_CHECK_OOA: {
2667		union ctl_io *io;
2668		struct ctl_lun *lun;
2669		struct ctl_ooa_info *ooa_info;
2670
2671
2672		ooa_info = (struct ctl_ooa_info *)addr;
2673
2674		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2675			ooa_info->status = CTL_OOA_INVALID_LUN;
2676			break;
2677		}
2678		mtx_lock(&softc->ctl_lock);
2679		lun = softc->ctl_luns[ooa_info->lun_id];
2680		if (lun == NULL) {
2681			mtx_unlock(&softc->ctl_lock);
2682			ooa_info->status = CTL_OOA_INVALID_LUN;
2683			break;
2684		}
2685		mtx_lock(&lun->lun_lock);
2686		mtx_unlock(&softc->ctl_lock);
2687		ooa_info->num_entries = 0;
2688		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2689		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2690		     &io->io_hdr, ooa_links)) {
2691			ooa_info->num_entries++;
2692		}
2693		mtx_unlock(&lun->lun_lock);
2694
2695		ooa_info->status = CTL_OOA_SUCCESS;
2696
2697		break;
2698	}
2699	case CTL_HARD_START:
2700	case CTL_HARD_STOP: {
2701		struct ctl_fe_ioctl_startstop_info ss_info;
2702		struct cfi_metatask *metatask;
2703		struct mtx hs_mtx;
2704
2705		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2706
2707		cv_init(&ss_info.sem, "hard start/stop cv" );
2708
2709		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2710		if (metatask == NULL) {
2711			retval = ENOMEM;
2712			mtx_destroy(&hs_mtx);
2713			break;
2714		}
2715
2716		if (cmd == CTL_HARD_START)
2717			metatask->tasktype = CFI_TASK_STARTUP;
2718		else
2719			metatask->tasktype = CFI_TASK_SHUTDOWN;
2720
2721		metatask->callback = ctl_ioctl_hard_startstop_callback;
2722		metatask->callback_arg = &ss_info;
2723
2724		cfi_action(metatask);
2725
2726		/* Wait for the callback */
2727		mtx_lock(&hs_mtx);
2728		cv_wait_sig(&ss_info.sem, &hs_mtx);
2729		mtx_unlock(&hs_mtx);
2730
2731		/*
2732		 * All information has been copied from the metatask by the
2733		 * time cv_broadcast() is called, so we free the metatask here.
2734		 */
2735		cfi_free_metatask(metatask);
2736
2737		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2738
2739		mtx_destroy(&hs_mtx);
2740		break;
2741	}
2742	case CTL_BBRREAD: {
2743		struct ctl_bbrread_info *bbr_info;
2744		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2745		struct mtx bbr_mtx;
2746		struct cfi_metatask *metatask;
2747
2748		bbr_info = (struct ctl_bbrread_info *)addr;
2749
2750		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2751
2752		bzero(&bbr_mtx, sizeof(bbr_mtx));
2753		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2754
2755		fe_bbr_info.bbr_info = bbr_info;
2756		fe_bbr_info.lock = &bbr_mtx;
2757
2758		cv_init(&fe_bbr_info.sem, "BBR read cv");
2759		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2760
2761		if (metatask == NULL) {
2762			mtx_destroy(&bbr_mtx);
2763			cv_destroy(&fe_bbr_info.sem);
2764			retval = ENOMEM;
2765			break;
2766		}
2767		metatask->tasktype = CFI_TASK_BBRREAD;
2768		metatask->callback = ctl_ioctl_bbrread_callback;
2769		metatask->callback_arg = &fe_bbr_info;
2770		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2771		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2772		metatask->taskinfo.bbrread.len = bbr_info->len;
2773
2774		cfi_action(metatask);
2775
2776		mtx_lock(&bbr_mtx);
2777		while (fe_bbr_info.wakeup_done == 0)
2778			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2779		mtx_unlock(&bbr_mtx);
2780
2781		bbr_info->status = metatask->status;
2782		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2783		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2784		memcpy(&bbr_info->sense_data,
2785		       &metatask->taskinfo.bbrread.sense_data,
2786		       ctl_min(sizeof(bbr_info->sense_data),
2787			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2788
2789		cfi_free_metatask(metatask);
2790
2791		mtx_destroy(&bbr_mtx);
2792		cv_destroy(&fe_bbr_info.sem);
2793
2794		break;
2795	}
2796	case CTL_DELAY_IO: {
2797		struct ctl_io_delay_info *delay_info;
2798#ifdef CTL_IO_DELAY
2799		struct ctl_lun *lun;
2800#endif /* CTL_IO_DELAY */
2801
2802		delay_info = (struct ctl_io_delay_info *)addr;
2803
2804#ifdef CTL_IO_DELAY
2805		mtx_lock(&softc->ctl_lock);
2806
2807		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2808		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2809			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2810		} else {
2811			lun = softc->ctl_luns[delay_info->lun_id];
2812			mtx_lock(&lun->lun_lock);
2813
2814			delay_info->status = CTL_DELAY_STATUS_OK;
2815
2816			switch (delay_info->delay_type) {
2817			case CTL_DELAY_TYPE_CONT:
2818				break;
2819			case CTL_DELAY_TYPE_ONESHOT:
2820				break;
2821			default:
2822				delay_info->status =
2823					CTL_DELAY_STATUS_INVALID_TYPE;
2824				break;
2825			}
2826
2827			switch (delay_info->delay_loc) {
2828			case CTL_DELAY_LOC_DATAMOVE:
2829				lun->delay_info.datamove_type =
2830					delay_info->delay_type;
2831				lun->delay_info.datamove_delay =
2832					delay_info->delay_secs;
2833				break;
2834			case CTL_DELAY_LOC_DONE:
2835				lun->delay_info.done_type =
2836					delay_info->delay_type;
2837				lun->delay_info.done_delay =
2838					delay_info->delay_secs;
2839				break;
2840			default:
2841				delay_info->status =
2842					CTL_DELAY_STATUS_INVALID_LOC;
2843				break;
2844			}
2845			mtx_unlock(&lun->lun_lock);
2846		}
2847
2848		mtx_unlock(&softc->ctl_lock);
2849#else
2850		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2851#endif /* CTL_IO_DELAY */
2852		break;
2853	}
2854	case CTL_REALSYNC_SET: {
2855		int *syncstate;
2856
2857		syncstate = (int *)addr;
2858
2859		mtx_lock(&softc->ctl_lock);
2860		switch (*syncstate) {
2861		case 0:
2862			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2863			break;
2864		case 1:
2865			softc->flags |= CTL_FLAG_REAL_SYNC;
2866			break;
2867		default:
2868			retval = EINVAL;
2869			break;
2870		}
2871		mtx_unlock(&softc->ctl_lock);
2872		break;
2873	}
2874	case CTL_REALSYNC_GET: {
2875		int *syncstate;
2876
2877		syncstate = (int*)addr;
2878
2879		mtx_lock(&softc->ctl_lock);
2880		if (softc->flags & CTL_FLAG_REAL_SYNC)
2881			*syncstate = 1;
2882		else
2883			*syncstate = 0;
2884		mtx_unlock(&softc->ctl_lock);
2885
2886		break;
2887	}
2888	case CTL_SETSYNC:
2889	case CTL_GETSYNC: {
2890		struct ctl_sync_info *sync_info;
2891		struct ctl_lun *lun;
2892
2893		sync_info = (struct ctl_sync_info *)addr;
2894
2895		mtx_lock(&softc->ctl_lock);
2896		lun = softc->ctl_luns[sync_info->lun_id];
2897		if (lun == NULL) {
2898			mtx_unlock(&softc->ctl_lock);
2899			sync_info->status = CTL_GS_SYNC_NO_LUN;
2900		}
2901		/*
2902		 * Get or set the sync interval.  We're not bounds checking
2903		 * in the set case, hopefully the user won't do something
2904		 * silly.
2905		 */
2906		mtx_lock(&lun->lun_lock);
2907		mtx_unlock(&softc->ctl_lock);
2908		if (cmd == CTL_GETSYNC)
2909			sync_info->sync_interval = lun->sync_interval;
2910		else
2911			lun->sync_interval = sync_info->sync_interval;
2912		mtx_unlock(&lun->lun_lock);
2913
2914		sync_info->status = CTL_GS_SYNC_OK;
2915
2916		break;
2917	}
2918	case CTL_GETSTATS: {
2919		struct ctl_stats *stats;
2920		struct ctl_lun *lun;
2921		int i;
2922
2923		stats = (struct ctl_stats *)addr;
2924
2925		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2926		     stats->alloc_len) {
2927			stats->status = CTL_SS_NEED_MORE_SPACE;
2928			stats->num_luns = softc->num_luns;
2929			break;
2930		}
2931		/*
2932		 * XXX KDM no locking here.  If the LUN list changes,
2933		 * things can blow up.
2934		 */
2935		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2936		     i++, lun = STAILQ_NEXT(lun, links)) {
2937			retval = copyout(&lun->stats, &stats->lun_stats[i],
2938					 sizeof(lun->stats));
2939			if (retval != 0)
2940				break;
2941		}
2942		stats->num_luns = softc->num_luns;
2943		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2944				 softc->num_luns;
2945		stats->status = CTL_SS_OK;
2946#ifdef CTL_TIME_IO
2947		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2948#else
2949		stats->flags = CTL_STATS_FLAG_NONE;
2950#endif
2951		getnanouptime(&stats->timestamp);
2952		break;
2953	}
2954	case CTL_ERROR_INJECT: {
2955		struct ctl_error_desc *err_desc, *new_err_desc;
2956		struct ctl_lun *lun;
2957
2958		err_desc = (struct ctl_error_desc *)addr;
2959
2960		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2961				      M_WAITOK | M_ZERO);
2962		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2963
2964		mtx_lock(&softc->ctl_lock);
2965		lun = softc->ctl_luns[err_desc->lun_id];
2966		if (lun == NULL) {
2967			mtx_unlock(&softc->ctl_lock);
2968			free(new_err_desc, M_CTL);
2969			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2970			       __func__, (uintmax_t)err_desc->lun_id);
2971			retval = EINVAL;
2972			break;
2973		}
2974		mtx_lock(&lun->lun_lock);
2975		mtx_unlock(&softc->ctl_lock);
2976
2977		/*
2978		 * We could do some checking here to verify the validity
2979		 * of the request, but given the complexity of error
2980		 * injection requests, the checking logic would be fairly
2981		 * complex.
2982		 *
2983		 * For now, if the request is invalid, it just won't get
2984		 * executed and might get deleted.
2985		 */
2986		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2987
2988		/*
2989		 * XXX KDM check to make sure the serial number is unique,
2990		 * in case we somehow manage to wrap.  That shouldn't
2991		 * happen for a very long time, but it's the right thing to
2992		 * do.
2993		 */
2994		new_err_desc->serial = lun->error_serial;
2995		err_desc->serial = lun->error_serial;
2996		lun->error_serial++;
2997
2998		mtx_unlock(&lun->lun_lock);
2999		break;
3000	}
3001	case CTL_ERROR_INJECT_DELETE: {
3002		struct ctl_error_desc *delete_desc, *desc, *desc2;
3003		struct ctl_lun *lun;
3004		int delete_done;
3005
3006		delete_desc = (struct ctl_error_desc *)addr;
3007		delete_done = 0;
3008
3009		mtx_lock(&softc->ctl_lock);
3010		lun = softc->ctl_luns[delete_desc->lun_id];
3011		if (lun == NULL) {
3012			mtx_unlock(&softc->ctl_lock);
3013			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
3014			       __func__, (uintmax_t)delete_desc->lun_id);
3015			retval = EINVAL;
3016			break;
3017		}
3018		mtx_lock(&lun->lun_lock);
3019		mtx_unlock(&softc->ctl_lock);
3020		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
3021			if (desc->serial != delete_desc->serial)
3022				continue;
3023
3024			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
3025				      links);
3026			free(desc, M_CTL);
3027			delete_done = 1;
3028		}
3029		mtx_unlock(&lun->lun_lock);
3030		if (delete_done == 0) {
3031			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3032			       "error serial %ju on LUN %u\n", __func__,
3033			       delete_desc->serial, delete_desc->lun_id);
3034			retval = EINVAL;
3035			break;
3036		}
3037		break;
3038	}
3039	case CTL_DUMP_STRUCTS: {
3040		int i, j, k, idx;
3041		struct ctl_port *port;
3042		struct ctl_frontend *fe;
3043
3044		mtx_lock(&softc->ctl_lock);
3045		printf("CTL Persistent Reservation information start:\n");
3046		for (i = 0; i < CTL_MAX_LUNS; i++) {
3047			struct ctl_lun *lun;
3048
3049			lun = softc->ctl_luns[i];
3050
3051			if ((lun == NULL)
3052			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3053				continue;
3054
3055			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3056				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3057					idx = j * CTL_MAX_INIT_PER_PORT + k;
3058					if (lun->pr_keys[idx] == 0)
3059						continue;
3060					printf("  LUN %d port %d iid %d key "
3061					       "%#jx\n", i, j, k,
3062					       (uintmax_t)lun->pr_keys[idx]);
3063				}
3064			}
3065		}
3066		printf("CTL Persistent Reservation information end\n");
3067		printf("CTL Ports:\n");
3068		STAILQ_FOREACH(port, &softc->port_list, links) {
3069			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3070			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3071			       port->frontend->name, port->port_type,
3072			       port->physical_port, port->virtual_port,
3073			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3074			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3075				if (port->wwpn_iid[j].in_use == 0 &&
3076				    port->wwpn_iid[j].wwpn == 0 &&
3077				    port->wwpn_iid[j].name == NULL)
3078					continue;
3079
3080				printf("    iid %u use %d WWPN %#jx '%s'\n",
3081				    j, port->wwpn_iid[j].in_use,
3082				    (uintmax_t)port->wwpn_iid[j].wwpn,
3083				    port->wwpn_iid[j].name);
3084			}
3085		}
3086		printf("CTL Port information end\n");
3087		mtx_unlock(&softc->ctl_lock);
3088		/*
3089		 * XXX KDM calling this without a lock.  We'd likely want
3090		 * to drop the lock before calling the frontend's dump
3091		 * routine anyway.
3092		 */
3093		printf("CTL Frontends:\n");
3094		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3095			printf("  Frontend '%s'\n", fe->name);
3096			if (fe->fe_dump != NULL)
3097				fe->fe_dump();
3098		}
3099		printf("CTL Frontend information end\n");
3100		break;
3101	}
3102	case CTL_LUN_REQ: {
3103		struct ctl_lun_req *lun_req;
3104		struct ctl_backend_driver *backend;
3105
3106		lun_req = (struct ctl_lun_req *)addr;
3107
3108		backend = ctl_backend_find(lun_req->backend);
3109		if (backend == NULL) {
3110			lun_req->status = CTL_LUN_ERROR;
3111			snprintf(lun_req->error_str,
3112				 sizeof(lun_req->error_str),
3113				 "Backend \"%s\" not found.",
3114				 lun_req->backend);
3115			break;
3116		}
3117		if (lun_req->num_be_args > 0) {
3118			lun_req->kern_be_args = ctl_copyin_args(
3119				lun_req->num_be_args,
3120				lun_req->be_args,
3121				lun_req->error_str,
3122				sizeof(lun_req->error_str));
3123			if (lun_req->kern_be_args == NULL) {
3124				lun_req->status = CTL_LUN_ERROR;
3125				break;
3126			}
3127		}
3128
3129		retval = backend->ioctl(dev, cmd, addr, flag, td);
3130
3131		if (lun_req->num_be_args > 0) {
3132			ctl_copyout_args(lun_req->num_be_args,
3133				      lun_req->kern_be_args);
3134			ctl_free_args(lun_req->num_be_args,
3135				      lun_req->kern_be_args);
3136		}
3137		break;
3138	}
3139	case CTL_LUN_LIST: {
3140		struct sbuf *sb;
3141		struct ctl_lun *lun;
3142		struct ctl_lun_list *list;
3143		struct ctl_option *opt;
3144
3145		list = (struct ctl_lun_list *)addr;
3146
3147		/*
3148		 * Allocate a fixed length sbuf here, based on the length
3149		 * of the user's buffer.  We could allocate an auto-extending
3150		 * buffer, and then tell the user how much larger our
3151		 * amount of data is than his buffer, but that presents
3152		 * some problems:
3153		 *
3154		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3155		 *     we can't hold a lock while calling them with an
3156		 *     auto-extending buffer.
3157 		 *
3158		 * 2.  There is not currently a LUN reference counting
3159		 *     mechanism, outside of outstanding transactions on
3160		 *     the LUN's OOA queue.  So a LUN could go away on us
3161		 *     while we're getting the LUN number, backend-specific
3162		 *     information, etc.  Thus, given the way things
3163		 *     currently work, we need to hold the CTL lock while
3164		 *     grabbing LUN information.
3165		 *
3166		 * So, from the user's standpoint, the best thing to do is
3167		 * allocate what he thinks is a reasonable buffer length,
3168		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3169		 * double the buffer length and try again.  (And repeat
3170		 * that until he succeeds.)
3171		 */
3172		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3173		if (sb == NULL) {
3174			list->status = CTL_LUN_LIST_ERROR;
3175			snprintf(list->error_str, sizeof(list->error_str),
3176				 "Unable to allocate %d bytes for LUN list",
3177				 list->alloc_len);
3178			break;
3179		}
3180
3181		sbuf_printf(sb, "<ctllunlist>\n");
3182
3183		mtx_lock(&softc->ctl_lock);
3184		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3185			mtx_lock(&lun->lun_lock);
3186			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3187					     (uintmax_t)lun->lun);
3188
3189			/*
3190			 * Bail out as soon as we see that we've overfilled
3191			 * the buffer.
3192			 */
3193			if (retval != 0)
3194				break;
3195
3196			retval = sbuf_printf(sb, "\t<backend_type>%s"
3197					     "</backend_type>\n",
3198					     (lun->backend == NULL) ?  "none" :
3199					     lun->backend->name);
3200
3201			if (retval != 0)
3202				break;
3203
3204			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3205					     lun->be_lun->lun_type);
3206
3207			if (retval != 0)
3208				break;
3209
3210			if (lun->backend == NULL) {
3211				retval = sbuf_printf(sb, "</lun>\n");
3212				if (retval != 0)
3213					break;
3214				continue;
3215			}
3216
3217			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3218					     (lun->be_lun->maxlba > 0) ?
3219					     lun->be_lun->maxlba + 1 : 0);
3220
3221			if (retval != 0)
3222				break;
3223
3224			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3225					     lun->be_lun->blocksize);
3226
3227			if (retval != 0)
3228				break;
3229
3230			retval = sbuf_printf(sb, "\t<serial_number>");
3231
3232			if (retval != 0)
3233				break;
3234
3235			retval = ctl_sbuf_printf_esc(sb,
3236			    lun->be_lun->serial_num,
3237			    sizeof(lun->be_lun->serial_num));
3238
3239			if (retval != 0)
3240				break;
3241
3242			retval = sbuf_printf(sb, "</serial_number>\n");
3243
3244			if (retval != 0)
3245				break;
3246
3247			retval = sbuf_printf(sb, "\t<device_id>");
3248
3249			if (retval != 0)
3250				break;
3251
3252			retval = ctl_sbuf_printf_esc(sb,
3253			    lun->be_lun->device_id,
3254			    sizeof(lun->be_lun->device_id));
3255
3256			if (retval != 0)
3257				break;
3258
3259			retval = sbuf_printf(sb, "</device_id>\n");
3260
3261			if (retval != 0)
3262				break;
3263
3264			if (lun->backend->lun_info != NULL) {
3265				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3266				if (retval != 0)
3267					break;
3268			}
3269			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3270				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3271				    opt->name, opt->value, opt->name);
3272				if (retval != 0)
3273					break;
3274			}
3275
3276			retval = sbuf_printf(sb, "</lun>\n");
3277
3278			if (retval != 0)
3279				break;
3280			mtx_unlock(&lun->lun_lock);
3281		}
3282		if (lun != NULL)
3283			mtx_unlock(&lun->lun_lock);
3284		mtx_unlock(&softc->ctl_lock);
3285
3286		if ((retval != 0)
3287		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3288			retval = 0;
3289			sbuf_delete(sb);
3290			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3291			snprintf(list->error_str, sizeof(list->error_str),
3292				 "Out of space, %d bytes is too small",
3293				 list->alloc_len);
3294			break;
3295		}
3296
3297		sbuf_finish(sb);
3298
3299		retval = copyout(sbuf_data(sb), list->lun_xml,
3300				 sbuf_len(sb) + 1);
3301
3302		list->fill_len = sbuf_len(sb) + 1;
3303		list->status = CTL_LUN_LIST_OK;
3304		sbuf_delete(sb);
3305		break;
3306	}
3307	case CTL_ISCSI: {
3308		struct ctl_iscsi *ci;
3309		struct ctl_frontend *fe;
3310
3311		ci = (struct ctl_iscsi *)addr;
3312
3313		fe = ctl_frontend_find("iscsi");
3314		if (fe == NULL) {
3315			ci->status = CTL_ISCSI_ERROR;
3316			snprintf(ci->error_str, sizeof(ci->error_str),
3317			    "Frontend \"iscsi\" not found.");
3318			break;
3319		}
3320
3321		retval = fe->ioctl(dev, cmd, addr, flag, td);
3322		break;
3323	}
3324	case CTL_PORT_REQ: {
3325		struct ctl_req *req;
3326		struct ctl_frontend *fe;
3327
3328		req = (struct ctl_req *)addr;
3329
3330		fe = ctl_frontend_find(req->driver);
3331		if (fe == NULL) {
3332			req->status = CTL_LUN_ERROR;
3333			snprintf(req->error_str, sizeof(req->error_str),
3334			    "Frontend \"%s\" not found.", req->driver);
3335			break;
3336		}
3337		if (req->num_args > 0) {
3338			req->kern_args = ctl_copyin_args(req->num_args,
3339			    req->args, req->error_str, sizeof(req->error_str));
3340			if (req->kern_args == NULL) {
3341				req->status = CTL_LUN_ERROR;
3342				break;
3343			}
3344		}
3345
3346		retval = fe->ioctl(dev, cmd, addr, flag, td);
3347
3348		if (req->num_args > 0) {
3349			ctl_copyout_args(req->num_args, req->kern_args);
3350			ctl_free_args(req->num_args, req->kern_args);
3351		}
3352		break;
3353	}
3354	case CTL_PORT_LIST: {
3355		struct sbuf *sb;
3356		struct ctl_port *port;
3357		struct ctl_lun_list *list;
3358		struct ctl_option *opt;
3359		int j;
3360
3361		list = (struct ctl_lun_list *)addr;
3362
3363		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3364		if (sb == NULL) {
3365			list->status = CTL_LUN_LIST_ERROR;
3366			snprintf(list->error_str, sizeof(list->error_str),
3367				 "Unable to allocate %d bytes for LUN list",
3368				 list->alloc_len);
3369			break;
3370		}
3371
3372		sbuf_printf(sb, "<ctlportlist>\n");
3373
3374		mtx_lock(&softc->ctl_lock);
3375		STAILQ_FOREACH(port, &softc->port_list, links) {
3376			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3377					     (uintmax_t)port->targ_port);
3378
3379			/*
3380			 * Bail out as soon as we see that we've overfilled
3381			 * the buffer.
3382			 */
3383			if (retval != 0)
3384				break;
3385
3386			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3387			    "</frontend_type>\n", port->frontend->name);
3388			if (retval != 0)
3389				break;
3390
3391			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3392					     port->port_type);
3393			if (retval != 0)
3394				break;
3395
3396			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3397			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3398			if (retval != 0)
3399				break;
3400
3401			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3402			    port->port_name);
3403			if (retval != 0)
3404				break;
3405
3406			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3407			    port->physical_port);
3408			if (retval != 0)
3409				break;
3410
3411			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3412			    port->virtual_port);
3413			if (retval != 0)
3414				break;
3415
3416			if (port->target_devid != NULL) {
3417				sbuf_printf(sb, "\t<target>");
3418				ctl_id_sbuf(port->target_devid, sb);
3419				sbuf_printf(sb, "</target>\n");
3420			}
3421
3422			if (port->port_devid != NULL) {
3423				sbuf_printf(sb, "\t<port>");
3424				ctl_id_sbuf(port->port_devid, sb);
3425				sbuf_printf(sb, "</port>\n");
3426			}
3427
3428			if (port->port_info != NULL) {
3429				retval = port->port_info(port->onoff_arg, sb);
3430				if (retval != 0)
3431					break;
3432			}
3433			STAILQ_FOREACH(opt, &port->options, links) {
3434				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3435				    opt->name, opt->value, opt->name);
3436				if (retval != 0)
3437					break;
3438			}
3439
3440			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3441				if (port->wwpn_iid[j].in_use == 0 ||
3442				    (port->wwpn_iid[j].wwpn == 0 &&
3443				     port->wwpn_iid[j].name == NULL))
3444					continue;
3445
3446				if (port->wwpn_iid[j].name != NULL)
3447					retval = sbuf_printf(sb,
3448					    "\t<initiator>%u %s</initiator>\n",
3449					    j, port->wwpn_iid[j].name);
3450				else
3451					retval = sbuf_printf(sb,
3452					    "\t<initiator>%u naa.%08jx</initiator>\n",
3453					    j, port->wwpn_iid[j].wwpn);
3454				if (retval != 0)
3455					break;
3456			}
3457			if (retval != 0)
3458				break;
3459
3460			retval = sbuf_printf(sb, "</targ_port>\n");
3461			if (retval != 0)
3462				break;
3463		}
3464		mtx_unlock(&softc->ctl_lock);
3465
3466		if ((retval != 0)
3467		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3468			retval = 0;
3469			sbuf_delete(sb);
3470			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3471			snprintf(list->error_str, sizeof(list->error_str),
3472				 "Out of space, %d bytes is too small",
3473				 list->alloc_len);
3474			break;
3475		}
3476
3477		sbuf_finish(sb);
3478
3479		retval = copyout(sbuf_data(sb), list->lun_xml,
3480				 sbuf_len(sb) + 1);
3481
3482		list->fill_len = sbuf_len(sb) + 1;
3483		list->status = CTL_LUN_LIST_OK;
3484		sbuf_delete(sb);
3485		break;
3486	}
3487	default: {
3488		/* XXX KDM should we fix this? */
3489#if 0
3490		struct ctl_backend_driver *backend;
3491		unsigned int type;
3492		int found;
3493
3494		found = 0;
3495
3496		/*
3497		 * We encode the backend type as the ioctl type for backend
3498		 * ioctls.  So parse it out here, and then search for a
3499		 * backend of this type.
3500		 */
3501		type = _IOC_TYPE(cmd);
3502
3503		STAILQ_FOREACH(backend, &softc->be_list, links) {
3504			if (backend->type == type) {
3505				found = 1;
3506				break;
3507			}
3508		}
3509		if (found == 0) {
3510			printf("ctl: unknown ioctl command %#lx or backend "
3511			       "%d\n", cmd, type);
3512			retval = EINVAL;
3513			break;
3514		}
3515		retval = backend->ioctl(dev, cmd, addr, flag, td);
3516#endif
3517		retval = ENOTTY;
3518		break;
3519	}
3520	}
3521	return (retval);
3522}
3523
3524uint32_t
3525ctl_get_initindex(struct ctl_nexus *nexus)
3526{
3527	if (nexus->targ_port < CTL_MAX_PORTS)
3528		return (nexus->initid.id +
3529			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3530	else
3531		return (nexus->initid.id +
3532		       ((nexus->targ_port - CTL_MAX_PORTS) *
3533			CTL_MAX_INIT_PER_PORT));
3534}
3535
3536uint32_t
3537ctl_get_resindex(struct ctl_nexus *nexus)
3538{
3539	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3540}
3541
3542uint32_t
3543ctl_port_idx(int port_num)
3544{
3545	if (port_num < CTL_MAX_PORTS)
3546		return(port_num);
3547	else
3548		return(port_num - CTL_MAX_PORTS);
3549}
3550
3551static uint32_t
3552ctl_map_lun(int port_num, uint32_t lun_id)
3553{
3554	struct ctl_port *port;
3555
3556	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3557	if (port == NULL)
3558		return (UINT32_MAX);
3559	if (port->lun_map == NULL)
3560		return (lun_id);
3561	return (port->lun_map(port->targ_lun_arg, lun_id));
3562}
3563
3564static uint32_t
3565ctl_map_lun_back(int port_num, uint32_t lun_id)
3566{
3567	struct ctl_port *port;
3568	uint32_t i;
3569
3570	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3571	if (port->lun_map == NULL)
3572		return (lun_id);
3573	for (i = 0; i < CTL_MAX_LUNS; i++) {
3574		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3575			return (i);
3576	}
3577	return (UINT32_MAX);
3578}
3579
3580/*
3581 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3582 * that are a power of 2.
3583 */
3584int
3585ctl_ffz(uint32_t *mask, uint32_t size)
3586{
3587	uint32_t num_chunks, num_pieces;
3588	int i, j;
3589
3590	num_chunks = (size >> 5);
3591	if (num_chunks == 0)
3592		num_chunks++;
3593	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3594
3595	for (i = 0; i < num_chunks; i++) {
3596		for (j = 0; j < num_pieces; j++) {
3597			if ((mask[i] & (1 << j)) == 0)
3598				return ((i << 5) + j);
3599		}
3600	}
3601
3602	return (-1);
3603}
3604
3605int
3606ctl_set_mask(uint32_t *mask, uint32_t bit)
3607{
3608	uint32_t chunk, piece;
3609
3610	chunk = bit >> 5;
3611	piece = bit % (sizeof(uint32_t) * 8);
3612
3613	if ((mask[chunk] & (1 << piece)) != 0)
3614		return (-1);
3615	else
3616		mask[chunk] |= (1 << piece);
3617
3618	return (0);
3619}
3620
3621int
3622ctl_clear_mask(uint32_t *mask, uint32_t bit)
3623{
3624	uint32_t chunk, piece;
3625
3626	chunk = bit >> 5;
3627	piece = bit % (sizeof(uint32_t) * 8);
3628
3629	if ((mask[chunk] & (1 << piece)) == 0)
3630		return (-1);
3631	else
3632		mask[chunk] &= ~(1 << piece);
3633
3634	return (0);
3635}
3636
3637int
3638ctl_is_set(uint32_t *mask, uint32_t bit)
3639{
3640	uint32_t chunk, piece;
3641
3642	chunk = bit >> 5;
3643	piece = bit % (sizeof(uint32_t) * 8);
3644
3645	if ((mask[chunk] & (1 << piece)) == 0)
3646		return (0);
3647	else
3648		return (1);
3649}
3650
3651#ifdef unused
3652/*
3653 * The bus, target and lun are optional, they can be filled in later.
3654 * can_wait is used to determine whether we can wait on the malloc or not.
3655 */
3656union ctl_io*
3657ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3658	      uint32_t targ_lun, int can_wait)
3659{
3660	union ctl_io *io;
3661
3662	if (can_wait)
3663		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3664	else
3665		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3666
3667	if (io != NULL) {
3668		io->io_hdr.io_type = io_type;
3669		io->io_hdr.targ_port = targ_port;
3670		/*
3671		 * XXX KDM this needs to change/go away.  We need to move
3672		 * to a preallocated pool of ctl_scsiio structures.
3673		 */
3674		io->io_hdr.nexus.targ_target.id = targ_target;
3675		io->io_hdr.nexus.targ_lun = targ_lun;
3676	}
3677
3678	return (io);
3679}
3680
3681void
3682ctl_kfree_io(union ctl_io *io)
3683{
3684	free(io, M_CTL);
3685}
3686#endif /* unused */
3687
3688/*
3689 * ctl_softc, pool_name, total_ctl_io are passed in.
3690 * npool is passed out.
3691 */
3692int
3693ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3694		uint32_t total_ctl_io, void **npool)
3695{
3696#ifdef IO_POOLS
3697	struct ctl_io_pool *pool;
3698
3699	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3700					    M_NOWAIT | M_ZERO);
3701	if (pool == NULL)
3702		return (ENOMEM);
3703
3704	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3705	pool->ctl_softc = ctl_softc;
3706	pool->zone = uma_zsecond_create(pool->name, NULL,
3707	    NULL, NULL, NULL, ctl_softc->io_zone);
3708	/* uma_prealloc(pool->zone, total_ctl_io); */
3709
3710	*npool = pool;
3711#else
3712	*npool = ctl_softc->io_zone;
3713#endif
3714	return (0);
3715}
3716
3717void
3718ctl_pool_free(struct ctl_io_pool *pool)
3719{
3720
3721	if (pool == NULL)
3722		return;
3723
3724#ifdef IO_POOLS
3725	uma_zdestroy(pool->zone);
3726	free(pool, M_CTL);
3727#endif
3728}
3729
3730union ctl_io *
3731ctl_alloc_io(void *pool_ref)
3732{
3733	union ctl_io *io;
3734#ifdef IO_POOLS
3735	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3736
3737	io = uma_zalloc(pool->zone, M_WAITOK);
3738#else
3739	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3740#endif
3741	if (io != NULL)
3742		io->io_hdr.pool = pool_ref;
3743	return (io);
3744}
3745
3746union ctl_io *
3747ctl_alloc_io_nowait(void *pool_ref)
3748{
3749	union ctl_io *io;
3750#ifdef IO_POOLS
3751	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3752
3753	io = uma_zalloc(pool->zone, M_NOWAIT);
3754#else
3755	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3756#endif
3757	if (io != NULL)
3758		io->io_hdr.pool = pool_ref;
3759	return (io);
3760}
3761
3762void
3763ctl_free_io(union ctl_io *io)
3764{
3765#ifdef IO_POOLS
3766	struct ctl_io_pool *pool;
3767#endif
3768
3769	if (io == NULL)
3770		return;
3771
3772#ifdef IO_POOLS
3773	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3774	uma_zfree(pool->zone, io);
3775#else
3776	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3777#endif
3778}
3779
3780void
3781ctl_zero_io(union ctl_io *io)
3782{
3783	void *pool_ref;
3784
3785	if (io == NULL)
3786		return;
3787
3788	/*
3789	 * May need to preserve linked list pointers at some point too.
3790	 */
3791	pool_ref = io->io_hdr.pool;
3792	memset(io, 0, sizeof(*io));
3793	io->io_hdr.pool = pool_ref;
3794}
3795
3796/*
3797 * This routine is currently used for internal copies of ctl_ios that need
3798 * to persist for some reason after we've already returned status to the
3799 * FETD.  (Thus the flag set.)
3800 *
3801 * XXX XXX
3802 * Note that this makes a blind copy of all fields in the ctl_io, except
3803 * for the pool reference.  This includes any memory that has been
3804 * allocated!  That memory will no longer be valid after done has been
3805 * called, so this would be VERY DANGEROUS for command that actually does
3806 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3807 * start and stop commands, which don't transfer any data, so this is not a
3808 * problem.  If it is used for anything else, the caller would also need to
3809 * allocate data buffer space and this routine would need to be modified to
3810 * copy the data buffer(s) as well.
3811 */
3812void
3813ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3814{
3815	void *pool_ref;
3816
3817	if ((src == NULL)
3818	 || (dest == NULL))
3819		return;
3820
3821	/*
3822	 * May need to preserve linked list pointers at some point too.
3823	 */
3824	pool_ref = dest->io_hdr.pool;
3825
3826	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3827
3828	dest->io_hdr.pool = pool_ref;
3829	/*
3830	 * We need to know that this is an internal copy, and doesn't need
3831	 * to get passed back to the FETD that allocated it.
3832	 */
3833	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3834}
3835
3836static int
3837ctl_expand_number(const char *buf, uint64_t *num)
3838{
3839	char *endptr;
3840	uint64_t number;
3841	unsigned shift;
3842
3843	number = strtoq(buf, &endptr, 0);
3844
3845	switch (tolower((unsigned char)*endptr)) {
3846	case 'e':
3847		shift = 60;
3848		break;
3849	case 'p':
3850		shift = 50;
3851		break;
3852	case 't':
3853		shift = 40;
3854		break;
3855	case 'g':
3856		shift = 30;
3857		break;
3858	case 'm':
3859		shift = 20;
3860		break;
3861	case 'k':
3862		shift = 10;
3863		break;
3864	case 'b':
3865	case '\0': /* No unit. */
3866		*num = number;
3867		return (0);
3868	default:
3869		/* Unrecognized unit. */
3870		return (-1);
3871	}
3872
3873	if ((number << shift) >> shift != number) {
3874		/* Overflow */
3875		return (-1);
3876	}
3877	*num = number << shift;
3878	return (0);
3879}
3880
3881
3882/*
3883 * This routine could be used in the future to load default and/or saved
3884 * mode page parameters for a particuar lun.
3885 */
3886static int
3887ctl_init_page_index(struct ctl_lun *lun)
3888{
3889	int i;
3890	struct ctl_page_index *page_index;
3891	const char *value;
3892	uint64_t ival;
3893
3894	memcpy(&lun->mode_pages.index, page_index_template,
3895	       sizeof(page_index_template));
3896
3897	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3898
3899		page_index = &lun->mode_pages.index[i];
3900		/*
3901		 * If this is a disk-only mode page, there's no point in
3902		 * setting it up.  For some pages, we have to have some
3903		 * basic information about the disk in order to calculate the
3904		 * mode page data.
3905		 */
3906		if ((lun->be_lun->lun_type != T_DIRECT)
3907		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3908			continue;
3909
3910		switch (page_index->page_code & SMPH_PC_MASK) {
3911		case SMS_RW_ERROR_RECOVERY_PAGE: {
3912			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3913				panic("subpage is incorrect!");
3914			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3915			       &rw_er_page_default,
3916			       sizeof(rw_er_page_default));
3917			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3918			       &rw_er_page_changeable,
3919			       sizeof(rw_er_page_changeable));
3920			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3921			       &rw_er_page_default,
3922			       sizeof(rw_er_page_default));
3923			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3924			       &rw_er_page_default,
3925			       sizeof(rw_er_page_default));
3926			page_index->page_data =
3927				(uint8_t *)lun->mode_pages.rw_er_page;
3928			break;
3929		}
3930		case SMS_FORMAT_DEVICE_PAGE: {
3931			struct scsi_format_page *format_page;
3932
3933			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3934				panic("subpage is incorrect!");
3935
3936			/*
3937			 * Sectors per track are set above.  Bytes per
3938			 * sector need to be set here on a per-LUN basis.
3939			 */
3940			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3941			       &format_page_default,
3942			       sizeof(format_page_default));
3943			memcpy(&lun->mode_pages.format_page[
3944			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3945			       sizeof(format_page_changeable));
3946			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3947			       &format_page_default,
3948			       sizeof(format_page_default));
3949			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3950			       &format_page_default,
3951			       sizeof(format_page_default));
3952
3953			format_page = &lun->mode_pages.format_page[
3954				CTL_PAGE_CURRENT];
3955			scsi_ulto2b(lun->be_lun->blocksize,
3956				    format_page->bytes_per_sector);
3957
3958			format_page = &lun->mode_pages.format_page[
3959				CTL_PAGE_DEFAULT];
3960			scsi_ulto2b(lun->be_lun->blocksize,
3961				    format_page->bytes_per_sector);
3962
3963			format_page = &lun->mode_pages.format_page[
3964				CTL_PAGE_SAVED];
3965			scsi_ulto2b(lun->be_lun->blocksize,
3966				    format_page->bytes_per_sector);
3967
3968			page_index->page_data =
3969				(uint8_t *)lun->mode_pages.format_page;
3970			break;
3971		}
3972		case SMS_RIGID_DISK_PAGE: {
3973			struct scsi_rigid_disk_page *rigid_disk_page;
3974			uint32_t sectors_per_cylinder;
3975			uint64_t cylinders;
3976#ifndef	__XSCALE__
3977			int shift;
3978#endif /* !__XSCALE__ */
3979
3980			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3981				panic("invalid subpage value %d",
3982				      page_index->subpage);
3983
3984			/*
3985			 * Rotation rate and sectors per track are set
3986			 * above.  We calculate the cylinders here based on
3987			 * capacity.  Due to the number of heads and
3988			 * sectors per track we're using, smaller arrays
3989			 * may turn out to have 0 cylinders.  Linux and
3990			 * FreeBSD don't pay attention to these mode pages
3991			 * to figure out capacity, but Solaris does.  It
3992			 * seems to deal with 0 cylinders just fine, and
3993			 * works out a fake geometry based on the capacity.
3994			 */
3995			memcpy(&lun->mode_pages.rigid_disk_page[
3996			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3997			       sizeof(rigid_disk_page_default));
3998			memcpy(&lun->mode_pages.rigid_disk_page[
3999			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4000			       sizeof(rigid_disk_page_changeable));
4001
4002			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4003				CTL_DEFAULT_HEADS;
4004
4005			/*
4006			 * The divide method here will be more accurate,
4007			 * probably, but results in floating point being
4008			 * used in the kernel on i386 (__udivdi3()).  On the
4009			 * XScale, though, __udivdi3() is implemented in
4010			 * software.
4011			 *
4012			 * The shift method for cylinder calculation is
4013			 * accurate if sectors_per_cylinder is a power of
4014			 * 2.  Otherwise it might be slightly off -- you
4015			 * might have a bit of a truncation problem.
4016			 */
4017#ifdef	__XSCALE__
4018			cylinders = (lun->be_lun->maxlba + 1) /
4019				sectors_per_cylinder;
4020#else
4021			for (shift = 31; shift > 0; shift--) {
4022				if (sectors_per_cylinder & (1 << shift))
4023					break;
4024			}
4025			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4026#endif
4027
4028			/*
4029			 * We've basically got 3 bytes, or 24 bits for the
4030			 * cylinder size in the mode page.  If we're over,
4031			 * just round down to 2^24.
4032			 */
4033			if (cylinders > 0xffffff)
4034				cylinders = 0xffffff;
4035
4036			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4037				CTL_PAGE_DEFAULT];
4038			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4039
4040			if ((value = ctl_get_opt(&lun->be_lun->options,
4041			    "rpm")) != NULL) {
4042				scsi_ulto2b(strtol(value, NULL, 0),
4043				     rigid_disk_page->rotation_rate);
4044			}
4045
4046			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4047			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4048			       sizeof(rigid_disk_page_default));
4049			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4050			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4051			       sizeof(rigid_disk_page_default));
4052
4053			page_index->page_data =
4054				(uint8_t *)lun->mode_pages.rigid_disk_page;
4055			break;
4056		}
4057		case SMS_CACHING_PAGE: {
4058			struct scsi_caching_page *caching_page;
4059
4060			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4061				panic("invalid subpage value %d",
4062				      page_index->subpage);
4063			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4064			       &caching_page_default,
4065			       sizeof(caching_page_default));
4066			memcpy(&lun->mode_pages.caching_page[
4067			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4068			       sizeof(caching_page_changeable));
4069			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4070			       &caching_page_default,
4071			       sizeof(caching_page_default));
4072			caching_page = &lun->mode_pages.caching_page[
4073			    CTL_PAGE_SAVED];
4074			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4075			if (value != NULL && strcmp(value, "off") == 0)
4076				caching_page->flags1 &= ~SCP_WCE;
4077			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4078			if (value != NULL && strcmp(value, "off") == 0)
4079				caching_page->flags1 |= SCP_RCD;
4080			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4081			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4082			       sizeof(caching_page_default));
4083			page_index->page_data =
4084				(uint8_t *)lun->mode_pages.caching_page;
4085			break;
4086		}
4087		case SMS_CONTROL_MODE_PAGE: {
4088			struct scsi_control_page *control_page;
4089
4090			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4091				panic("invalid subpage value %d",
4092				      page_index->subpage);
4093
4094			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4095			       &control_page_default,
4096			       sizeof(control_page_default));
4097			memcpy(&lun->mode_pages.control_page[
4098			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4099			       sizeof(control_page_changeable));
4100			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4101			       &control_page_default,
4102			       sizeof(control_page_default));
4103			control_page = &lun->mode_pages.control_page[
4104			    CTL_PAGE_SAVED];
4105			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4106			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4107				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4108				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4109			}
4110			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4111			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4112			       sizeof(control_page_default));
4113			page_index->page_data =
4114				(uint8_t *)lun->mode_pages.control_page;
4115			break;
4116
4117		}
4118		case SMS_INFO_EXCEPTIONS_PAGE: {
4119			switch (page_index->subpage) {
4120			case SMS_SUBPAGE_PAGE_0:
4121				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4122				       &ie_page_default,
4123				       sizeof(ie_page_default));
4124				memcpy(&lun->mode_pages.ie_page[
4125				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4126				       sizeof(ie_page_changeable));
4127				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4128				       &ie_page_default,
4129				       sizeof(ie_page_default));
4130				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4131				       &ie_page_default,
4132				       sizeof(ie_page_default));
4133				page_index->page_data =
4134					(uint8_t *)lun->mode_pages.ie_page;
4135				break;
4136			case 0x02: {
4137				struct ctl_logical_block_provisioning_page *page;
4138
4139				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4140				       &lbp_page_default,
4141				       sizeof(lbp_page_default));
4142				memcpy(&lun->mode_pages.lbp_page[
4143				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4144				       sizeof(lbp_page_changeable));
4145				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4146				       &lbp_page_default,
4147				       sizeof(lbp_page_default));
4148				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4149				value = ctl_get_opt(&lun->be_lun->options,
4150				    "avail-threshold");
4151				if (value != NULL &&
4152				    ctl_expand_number(value, &ival) == 0) {
4153					page->descr[0].flags |= SLBPPD_ENABLED |
4154					    SLBPPD_ARMING_DEC;
4155					if (lun->be_lun->blocksize)
4156						ival /= lun->be_lun->blocksize;
4157					else
4158						ival /= 512;
4159					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4160					    page->descr[0].count);
4161				}
4162				value = ctl_get_opt(&lun->be_lun->options,
4163				    "used-threshold");
4164				if (value != NULL &&
4165				    ctl_expand_number(value, &ival) == 0) {
4166					page->descr[1].flags |= SLBPPD_ENABLED |
4167					    SLBPPD_ARMING_INC;
4168					if (lun->be_lun->blocksize)
4169						ival /= lun->be_lun->blocksize;
4170					else
4171						ival /= 512;
4172					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4173					    page->descr[1].count);
4174				}
4175				value = ctl_get_opt(&lun->be_lun->options,
4176				    "pool-avail-threshold");
4177				if (value != NULL &&
4178				    ctl_expand_number(value, &ival) == 0) {
4179					page->descr[2].flags |= SLBPPD_ENABLED |
4180					    SLBPPD_ARMING_DEC;
4181					if (lun->be_lun->blocksize)
4182						ival /= lun->be_lun->blocksize;
4183					else
4184						ival /= 512;
4185					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4186					    page->descr[2].count);
4187				}
4188				value = ctl_get_opt(&lun->be_lun->options,
4189				    "pool-used-threshold");
4190				if (value != NULL &&
4191				    ctl_expand_number(value, &ival) == 0) {
4192					page->descr[3].flags |= SLBPPD_ENABLED |
4193					    SLBPPD_ARMING_INC;
4194					if (lun->be_lun->blocksize)
4195						ival /= lun->be_lun->blocksize;
4196					else
4197						ival /= 512;
4198					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4199					    page->descr[3].count);
4200				}
4201				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4202				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4203				       sizeof(lbp_page_default));
4204				page_index->page_data =
4205					(uint8_t *)lun->mode_pages.lbp_page;
4206			}}
4207			break;
4208		}
4209		case SMS_VENDOR_SPECIFIC_PAGE:{
4210			switch (page_index->subpage) {
4211			case DBGCNF_SUBPAGE_CODE: {
4212				struct copan_debugconf_subpage *current_page,
4213							       *saved_page;
4214
4215				memcpy(&lun->mode_pages.debugconf_subpage[
4216				       CTL_PAGE_CURRENT],
4217				       &debugconf_page_default,
4218				       sizeof(debugconf_page_default));
4219				memcpy(&lun->mode_pages.debugconf_subpage[
4220				       CTL_PAGE_CHANGEABLE],
4221				       &debugconf_page_changeable,
4222				       sizeof(debugconf_page_changeable));
4223				memcpy(&lun->mode_pages.debugconf_subpage[
4224				       CTL_PAGE_DEFAULT],
4225				       &debugconf_page_default,
4226				       sizeof(debugconf_page_default));
4227				memcpy(&lun->mode_pages.debugconf_subpage[
4228				       CTL_PAGE_SAVED],
4229				       &debugconf_page_default,
4230				       sizeof(debugconf_page_default));
4231				page_index->page_data =
4232					(uint8_t *)lun->mode_pages.debugconf_subpage;
4233
4234				current_page = (struct copan_debugconf_subpage *)
4235					(page_index->page_data +
4236					 (page_index->page_len *
4237					  CTL_PAGE_CURRENT));
4238				saved_page = (struct copan_debugconf_subpage *)
4239					(page_index->page_data +
4240					 (page_index->page_len *
4241					  CTL_PAGE_SAVED));
4242				break;
4243			}
4244			default:
4245				panic("invalid subpage value %d",
4246				      page_index->subpage);
4247				break;
4248			}
4249   			break;
4250		}
4251		default:
4252			panic("invalid page value %d",
4253			      page_index->page_code & SMPH_PC_MASK);
4254			break;
4255    	}
4256	}
4257
4258	return (CTL_RETVAL_COMPLETE);
4259}
4260
4261static int
4262ctl_init_log_page_index(struct ctl_lun *lun)
4263{
4264	struct ctl_page_index *page_index;
4265	int i, j, k, prev;
4266
4267	memcpy(&lun->log_pages.index, log_page_index_template,
4268	       sizeof(log_page_index_template));
4269
4270	prev = -1;
4271	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4272
4273		page_index = &lun->log_pages.index[i];
4274		/*
4275		 * If this is a disk-only mode page, there's no point in
4276		 * setting it up.  For some pages, we have to have some
4277		 * basic information about the disk in order to calculate the
4278		 * mode page data.
4279		 */
4280		if ((lun->be_lun->lun_type != T_DIRECT)
4281		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4282			continue;
4283
4284		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4285		    ((lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) == 0 ||
4286		     lun->backend->lun_attr == NULL))
4287			continue;
4288
4289		if (page_index->page_code != prev) {
4290			lun->log_pages.pages_page[j] = page_index->page_code;
4291			prev = page_index->page_code;
4292			j++;
4293		}
4294		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4295		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4296		k++;
4297	}
4298	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4299	lun->log_pages.index[0].page_len = j;
4300	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4301	lun->log_pages.index[1].page_len = k * 2;
4302	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4303	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4304
4305	return (CTL_RETVAL_COMPLETE);
4306}
4307
4308static int
4309hex2bin(const char *str, uint8_t *buf, int buf_size)
4310{
4311	int i;
4312	u_char c;
4313
4314	memset(buf, 0, buf_size);
4315	while (isspace(str[0]))
4316		str++;
4317	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4318		str += 2;
4319	buf_size *= 2;
4320	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4321		c = str[i];
4322		if (isdigit(c))
4323			c -= '0';
4324		else if (isalpha(c))
4325			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4326		else
4327			break;
4328		if (c >= 16)
4329			break;
4330		if ((i & 1) == 0)
4331			buf[i / 2] |= (c << 4);
4332		else
4333			buf[i / 2] |= c;
4334	}
4335	return ((i + 1) / 2);
4336}
4337
4338/*
4339 * LUN allocation.
4340 *
4341 * Requirements:
4342 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4343 *   wants us to allocate the LUN and he can block.
4344 * - ctl_softc is always set
4345 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4346 *
4347 * Returns 0 for success, non-zero (errno) for failure.
4348 */
4349static int
4350ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4351	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4352{
4353	struct ctl_lun *nlun, *lun;
4354	struct ctl_port *port;
4355	struct scsi_vpd_id_descriptor *desc;
4356	struct scsi_vpd_id_t10 *t10id;
4357	const char *eui, *naa, *scsiname, *vendor, *value;
4358	int lun_number, i, lun_malloced;
4359	int devidlen, idlen1, idlen2 = 0, len;
4360
4361	if (be_lun == NULL)
4362		return (EINVAL);
4363
4364	/*
4365	 * We currently only support Direct Access or Processor LUN types.
4366	 */
4367	switch (be_lun->lun_type) {
4368	case T_DIRECT:
4369		break;
4370	case T_PROCESSOR:
4371		break;
4372	case T_SEQUENTIAL:
4373	case T_CHANGER:
4374	default:
4375		be_lun->lun_config_status(be_lun->be_lun,
4376					  CTL_LUN_CONFIG_FAILURE);
4377		break;
4378	}
4379	if (ctl_lun == NULL) {
4380		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4381		lun_malloced = 1;
4382	} else {
4383		lun_malloced = 0;
4384		lun = ctl_lun;
4385	}
4386
4387	memset(lun, 0, sizeof(*lun));
4388	if (lun_malloced)
4389		lun->flags = CTL_LUN_MALLOCED;
4390
4391	/* Generate LUN ID. */
4392	devidlen = max(CTL_DEVID_MIN_LEN,
4393	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4394	idlen1 = sizeof(*t10id) + devidlen;
4395	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4396	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4397	if (scsiname != NULL) {
4398		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4399		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4400	}
4401	eui = ctl_get_opt(&be_lun->options, "eui");
4402	if (eui != NULL) {
4403		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4404	}
4405	naa = ctl_get_opt(&be_lun->options, "naa");
4406	if (naa != NULL) {
4407		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4408	}
4409	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4410	    M_CTL, M_WAITOK | M_ZERO);
4411	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4412	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4413	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4414	desc->length = idlen1;
4415	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4416	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4417	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4418		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4419	} else {
4420		strncpy(t10id->vendor, vendor,
4421		    min(sizeof(t10id->vendor), strlen(vendor)));
4422	}
4423	strncpy((char *)t10id->vendor_spec_id,
4424	    (char *)be_lun->device_id, devidlen);
4425	if (scsiname != NULL) {
4426		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4427		    desc->length);
4428		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4429		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4430		    SVPD_ID_TYPE_SCSI_NAME;
4431		desc->length = idlen2;
4432		strlcpy(desc->identifier, scsiname, idlen2);
4433	}
4434	if (eui != NULL) {
4435		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4436		    desc->length);
4437		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4438		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4439		    SVPD_ID_TYPE_EUI64;
4440		desc->length = hex2bin(eui, desc->identifier, 16);
4441		desc->length = desc->length > 12 ? 16 :
4442		    (desc->length > 8 ? 12 : 8);
4443		len -= 16 - desc->length;
4444	}
4445	if (naa != NULL) {
4446		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4447		    desc->length);
4448		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4449		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4450		    SVPD_ID_TYPE_NAA;
4451		desc->length = hex2bin(naa, desc->identifier, 16);
4452		desc->length = desc->length > 8 ? 16 : 8;
4453		len -= 16 - desc->length;
4454	}
4455	lun->lun_devid->len = len;
4456
4457	mtx_lock(&ctl_softc->ctl_lock);
4458	/*
4459	 * See if the caller requested a particular LUN number.  If so, see
4460	 * if it is available.  Otherwise, allocate the first available LUN.
4461	 */
4462	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4463		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4464		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4465			mtx_unlock(&ctl_softc->ctl_lock);
4466			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4467				printf("ctl: requested LUN ID %d is higher "
4468				       "than CTL_MAX_LUNS - 1 (%d)\n",
4469				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4470			} else {
4471				/*
4472				 * XXX KDM return an error, or just assign
4473				 * another LUN ID in this case??
4474				 */
4475				printf("ctl: requested LUN ID %d is already "
4476				       "in use\n", be_lun->req_lun_id);
4477			}
4478			if (lun->flags & CTL_LUN_MALLOCED)
4479				free(lun, M_CTL);
4480			be_lun->lun_config_status(be_lun->be_lun,
4481						  CTL_LUN_CONFIG_FAILURE);
4482			return (ENOSPC);
4483		}
4484		lun_number = be_lun->req_lun_id;
4485	} else {
4486		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4487		if (lun_number == -1) {
4488			mtx_unlock(&ctl_softc->ctl_lock);
4489			printf("ctl: can't allocate LUN on target %ju, out of "
4490			       "LUNs\n", (uintmax_t)target_id.id);
4491			if (lun->flags & CTL_LUN_MALLOCED)
4492				free(lun, M_CTL);
4493			be_lun->lun_config_status(be_lun->be_lun,
4494						  CTL_LUN_CONFIG_FAILURE);
4495			return (ENOSPC);
4496		}
4497	}
4498	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4499
4500	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4501	lun->target = target_id;
4502	lun->lun = lun_number;
4503	lun->be_lun = be_lun;
4504	/*
4505	 * The processor LUN is always enabled.  Disk LUNs come on line
4506	 * disabled, and must be enabled by the backend.
4507	 */
4508	lun->flags |= CTL_LUN_DISABLED;
4509	lun->backend = be_lun->be;
4510	be_lun->ctl_lun = lun;
4511	be_lun->lun_id = lun_number;
4512	atomic_add_int(&be_lun->be->num_luns, 1);
4513	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4514		lun->flags |= CTL_LUN_OFFLINE;
4515
4516	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4517		lun->flags |= CTL_LUN_STOPPED;
4518
4519	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4520		lun->flags |= CTL_LUN_INOPERABLE;
4521
4522	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4523		lun->flags |= CTL_LUN_PRIMARY_SC;
4524
4525	value = ctl_get_opt(&be_lun->options, "readonly");
4526	if (value != NULL && strcmp(value, "on") == 0)
4527		lun->flags |= CTL_LUN_READONLY;
4528
4529	lun->ctl_softc = ctl_softc;
4530	TAILQ_INIT(&lun->ooa_queue);
4531	TAILQ_INIT(&lun->blocked_queue);
4532	STAILQ_INIT(&lun->error_list);
4533	ctl_tpc_lun_init(lun);
4534
4535	/*
4536	 * Initialize the mode and log page index.
4537	 */
4538	ctl_init_page_index(lun);
4539	ctl_init_log_page_index(lun);
4540
4541	/*
4542	 * Set the poweron UA for all initiators on this LUN only.
4543	 */
4544	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4545		lun->pending_ua[i] = CTL_UA_POWERON;
4546
4547	/*
4548	 * Now, before we insert this lun on the lun list, set the lun
4549	 * inventory changed UA for all other luns.
4550	 */
4551	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4552		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4553			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4554		}
4555	}
4556
4557	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4558
4559	ctl_softc->ctl_luns[lun_number] = lun;
4560
4561	ctl_softc->num_luns++;
4562
4563	/* Setup statistics gathering */
4564	lun->stats.device_type = be_lun->lun_type;
4565	lun->stats.lun_number = lun_number;
4566	if (lun->stats.device_type == T_DIRECT)
4567		lun->stats.blocksize = be_lun->blocksize;
4568	else
4569		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4570	for (i = 0;i < CTL_MAX_PORTS;i++)
4571		lun->stats.ports[i].targ_port = i;
4572
4573	mtx_unlock(&ctl_softc->ctl_lock);
4574
4575	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4576
4577	/*
4578	 * Run through each registered FETD and bring it online if it isn't
4579	 * already.  Enable the target ID if it hasn't been enabled, and
4580	 * enable this particular LUN.
4581	 */
4582	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4583		int retval;
4584
4585		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4586		if (retval != 0) {
4587			printf("ctl_alloc_lun: FETD %s port %d returned error "
4588			       "%d for lun_enable on target %ju lun %d\n",
4589			       port->port_name, port->targ_port, retval,
4590			       (uintmax_t)target_id.id, lun_number);
4591		} else
4592			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4593	}
4594	return (0);
4595}
4596
4597/*
4598 * Delete a LUN.
4599 * Assumptions:
4600 * - LUN has already been marked invalid and any pending I/O has been taken
4601 *   care of.
4602 */
4603static int
4604ctl_free_lun(struct ctl_lun *lun)
4605{
4606	struct ctl_softc *softc;
4607#if 0
4608	struct ctl_port *port;
4609#endif
4610	struct ctl_lun *nlun;
4611	int i;
4612
4613	softc = lun->ctl_softc;
4614
4615	mtx_assert(&softc->ctl_lock, MA_OWNED);
4616
4617	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4618
4619	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4620
4621	softc->ctl_luns[lun->lun] = NULL;
4622
4623	if (!TAILQ_EMPTY(&lun->ooa_queue))
4624		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4625
4626	softc->num_luns--;
4627
4628	/*
4629	 * XXX KDM this scheme only works for a single target/multiple LUN
4630	 * setup.  It needs to be revamped for a multiple target scheme.
4631	 *
4632	 * XXX KDM this results in port->lun_disable() getting called twice,
4633	 * once when ctl_disable_lun() is called, and a second time here.
4634	 * We really need to re-think the LUN disable semantics.  There
4635	 * should probably be several steps/levels to LUN removal:
4636	 *  - disable
4637	 *  - invalidate
4638	 *  - free
4639 	 *
4640	 * Right now we only have a disable method when communicating to
4641	 * the front end ports, at least for individual LUNs.
4642	 */
4643#if 0
4644	STAILQ_FOREACH(port, &softc->port_list, links) {
4645		int retval;
4646
4647		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4648					 lun->lun);
4649		if (retval != 0) {
4650			printf("ctl_free_lun: FETD %s port %d returned error "
4651			       "%d for lun_disable on target %ju lun %jd\n",
4652			       port->port_name, port->targ_port, retval,
4653			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4654		}
4655
4656		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4657			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4658
4659			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4660			if (retval != 0) {
4661				printf("ctl_free_lun: FETD %s port %d "
4662				       "returned error %d for targ_disable on "
4663				       "target %ju\n", port->port_name,
4664				       port->targ_port, retval,
4665				       (uintmax_t)lun->target.id);
4666			} else
4667				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4668
4669			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4670				continue;
4671
4672#if 0
4673			port->port_offline(port->onoff_arg);
4674			port->status &= ~CTL_PORT_STATUS_ONLINE;
4675#endif
4676		}
4677	}
4678#endif
4679
4680	/*
4681	 * Tell the backend to free resources, if this LUN has a backend.
4682	 */
4683	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4684	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4685
4686	ctl_tpc_lun_shutdown(lun);
4687	mtx_destroy(&lun->lun_lock);
4688	free(lun->lun_devid, M_CTL);
4689	free(lun->write_buffer, M_CTL);
4690	if (lun->flags & CTL_LUN_MALLOCED)
4691		free(lun, M_CTL);
4692
4693	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4694		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4695			nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE;
4696		}
4697	}
4698
4699	return (0);
4700}
4701
4702static void
4703ctl_create_lun(struct ctl_be_lun *be_lun)
4704{
4705	struct ctl_softc *ctl_softc;
4706
4707	ctl_softc = control_softc;
4708
4709	/*
4710	 * ctl_alloc_lun() should handle all potential failure cases.
4711	 */
4712	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4713}
4714
4715int
4716ctl_add_lun(struct ctl_be_lun *be_lun)
4717{
4718	struct ctl_softc *ctl_softc = control_softc;
4719
4720	mtx_lock(&ctl_softc->ctl_lock);
4721	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4722	mtx_unlock(&ctl_softc->ctl_lock);
4723	wakeup(&ctl_softc->pending_lun_queue);
4724
4725	return (0);
4726}
4727
4728int
4729ctl_enable_lun(struct ctl_be_lun *be_lun)
4730{
4731	struct ctl_softc *ctl_softc;
4732	struct ctl_port *port, *nport;
4733	struct ctl_lun *lun;
4734	int retval;
4735
4736	ctl_softc = control_softc;
4737
4738	lun = (struct ctl_lun *)be_lun->ctl_lun;
4739
4740	mtx_lock(&ctl_softc->ctl_lock);
4741	mtx_lock(&lun->lun_lock);
4742	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4743		/*
4744		 * eh?  Why did we get called if the LUN is already
4745		 * enabled?
4746		 */
4747		mtx_unlock(&lun->lun_lock);
4748		mtx_unlock(&ctl_softc->ctl_lock);
4749		return (0);
4750	}
4751	lun->flags &= ~CTL_LUN_DISABLED;
4752	mtx_unlock(&lun->lun_lock);
4753
4754	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4755		nport = STAILQ_NEXT(port, links);
4756
4757		/*
4758		 * Drop the lock while we call the FETD's enable routine.
4759		 * This can lead to a callback into CTL (at least in the
4760		 * case of the internal initiator frontend.
4761		 */
4762		mtx_unlock(&ctl_softc->ctl_lock);
4763		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4764		mtx_lock(&ctl_softc->ctl_lock);
4765		if (retval != 0) {
4766			printf("%s: FETD %s port %d returned error "
4767			       "%d for lun_enable on target %ju lun %jd\n",
4768			       __func__, port->port_name, port->targ_port, retval,
4769			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4770		}
4771#if 0
4772		 else {
4773            /* NOTE:  TODO:  why does lun enable affect port status? */
4774			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4775		}
4776#endif
4777	}
4778
4779	mtx_unlock(&ctl_softc->ctl_lock);
4780
4781	return (0);
4782}
4783
4784int
4785ctl_disable_lun(struct ctl_be_lun *be_lun)
4786{
4787	struct ctl_softc *ctl_softc;
4788	struct ctl_port *port;
4789	struct ctl_lun *lun;
4790	int retval;
4791
4792	ctl_softc = control_softc;
4793
4794	lun = (struct ctl_lun *)be_lun->ctl_lun;
4795
4796	mtx_lock(&ctl_softc->ctl_lock);
4797	mtx_lock(&lun->lun_lock);
4798	if (lun->flags & CTL_LUN_DISABLED) {
4799		mtx_unlock(&lun->lun_lock);
4800		mtx_unlock(&ctl_softc->ctl_lock);
4801		return (0);
4802	}
4803	lun->flags |= CTL_LUN_DISABLED;
4804	mtx_unlock(&lun->lun_lock);
4805
4806	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4807		mtx_unlock(&ctl_softc->ctl_lock);
4808		/*
4809		 * Drop the lock before we call the frontend's disable
4810		 * routine, to avoid lock order reversals.
4811		 *
4812		 * XXX KDM what happens if the frontend list changes while
4813		 * we're traversing it?  It's unlikely, but should be handled.
4814		 */
4815		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4816					 lun->lun);
4817		mtx_lock(&ctl_softc->ctl_lock);
4818		if (retval != 0) {
4819			printf("ctl_alloc_lun: FETD %s port %d returned error "
4820			       "%d for lun_disable on target %ju lun %jd\n",
4821			       port->port_name, port->targ_port, retval,
4822			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4823		}
4824	}
4825
4826	mtx_unlock(&ctl_softc->ctl_lock);
4827
4828	return (0);
4829}
4830
4831int
4832ctl_start_lun(struct ctl_be_lun *be_lun)
4833{
4834	struct ctl_softc *ctl_softc;
4835	struct ctl_lun *lun;
4836
4837	ctl_softc = control_softc;
4838
4839	lun = (struct ctl_lun *)be_lun->ctl_lun;
4840
4841	mtx_lock(&lun->lun_lock);
4842	lun->flags &= ~CTL_LUN_STOPPED;
4843	mtx_unlock(&lun->lun_lock);
4844
4845	return (0);
4846}
4847
4848int
4849ctl_stop_lun(struct ctl_be_lun *be_lun)
4850{
4851	struct ctl_softc *ctl_softc;
4852	struct ctl_lun *lun;
4853
4854	ctl_softc = control_softc;
4855
4856	lun = (struct ctl_lun *)be_lun->ctl_lun;
4857
4858	mtx_lock(&lun->lun_lock);
4859	lun->flags |= CTL_LUN_STOPPED;
4860	mtx_unlock(&lun->lun_lock);
4861
4862	return (0);
4863}
4864
4865int
4866ctl_lun_offline(struct ctl_be_lun *be_lun)
4867{
4868	struct ctl_softc *ctl_softc;
4869	struct ctl_lun *lun;
4870
4871	ctl_softc = control_softc;
4872
4873	lun = (struct ctl_lun *)be_lun->ctl_lun;
4874
4875	mtx_lock(&lun->lun_lock);
4876	lun->flags |= CTL_LUN_OFFLINE;
4877	mtx_unlock(&lun->lun_lock);
4878
4879	return (0);
4880}
4881
4882int
4883ctl_lun_online(struct ctl_be_lun *be_lun)
4884{
4885	struct ctl_softc *ctl_softc;
4886	struct ctl_lun *lun;
4887
4888	ctl_softc = control_softc;
4889
4890	lun = (struct ctl_lun *)be_lun->ctl_lun;
4891
4892	mtx_lock(&lun->lun_lock);
4893	lun->flags &= ~CTL_LUN_OFFLINE;
4894	mtx_unlock(&lun->lun_lock);
4895
4896	return (0);
4897}
4898
4899int
4900ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4901{
4902	struct ctl_softc *ctl_softc;
4903	struct ctl_lun *lun;
4904
4905	ctl_softc = control_softc;
4906
4907	lun = (struct ctl_lun *)be_lun->ctl_lun;
4908
4909	mtx_lock(&lun->lun_lock);
4910
4911	/*
4912	 * The LUN needs to be disabled before it can be marked invalid.
4913	 */
4914	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4915		mtx_unlock(&lun->lun_lock);
4916		return (-1);
4917	}
4918	/*
4919	 * Mark the LUN invalid.
4920	 */
4921	lun->flags |= CTL_LUN_INVALID;
4922
4923	/*
4924	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4925	 * If we have something in the OOA queue, we'll free it when the
4926	 * last I/O completes.
4927	 */
4928	if (TAILQ_EMPTY(&lun->ooa_queue)) {
4929		mtx_unlock(&lun->lun_lock);
4930		mtx_lock(&ctl_softc->ctl_lock);
4931		ctl_free_lun(lun);
4932		mtx_unlock(&ctl_softc->ctl_lock);
4933	} else
4934		mtx_unlock(&lun->lun_lock);
4935
4936	return (0);
4937}
4938
4939int
4940ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4941{
4942	struct ctl_softc *ctl_softc;
4943	struct ctl_lun *lun;
4944
4945	ctl_softc = control_softc;
4946	lun = (struct ctl_lun *)be_lun->ctl_lun;
4947
4948	mtx_lock(&lun->lun_lock);
4949	lun->flags |= CTL_LUN_INOPERABLE;
4950	mtx_unlock(&lun->lun_lock);
4951
4952	return (0);
4953}
4954
4955int
4956ctl_lun_operable(struct ctl_be_lun *be_lun)
4957{
4958	struct ctl_softc *ctl_softc;
4959	struct ctl_lun *lun;
4960
4961	ctl_softc = control_softc;
4962	lun = (struct ctl_lun *)be_lun->ctl_lun;
4963
4964	mtx_lock(&lun->lun_lock);
4965	lun->flags &= ~CTL_LUN_INOPERABLE;
4966	mtx_unlock(&lun->lun_lock);
4967
4968	return (0);
4969}
4970
4971void
4972ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4973{
4974	struct ctl_lun *lun;
4975	struct ctl_softc *softc;
4976	int i;
4977
4978	softc = control_softc;
4979
4980	lun = (struct ctl_lun *)be_lun->ctl_lun;
4981
4982	mtx_lock(&lun->lun_lock);
4983
4984	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4985		lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED;
4986
4987	mtx_unlock(&lun->lun_lock);
4988}
4989
4990/*
4991 * Backend "memory move is complete" callback for requests that never
4992 * make it down to say RAIDCore's configuration code.
4993 */
4994int
4995ctl_config_move_done(union ctl_io *io)
4996{
4997	int retval;
4998
4999	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5000	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5001	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5002
5003	if ((io->io_hdr.port_status != 0) &&
5004	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5005	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5006		/*
5007		 * For hardware error sense keys, the sense key
5008		 * specific value is defined to be a retry count,
5009		 * but we use it to pass back an internal FETD
5010		 * error code.  XXX KDM  Hopefully the FETD is only
5011		 * using 16 bits for an error code, since that's
5012		 * all the space we have in the sks field.
5013		 */
5014		ctl_set_internal_failure(&io->scsiio,
5015					 /*sks_valid*/ 1,
5016					 /*retry_count*/
5017					 io->io_hdr.port_status);
5018	}
5019
5020	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5021	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5022	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5023	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5024		/*
5025		 * XXX KDM just assuming a single pointer here, and not a
5026		 * S/G list.  If we start using S/G lists for config data,
5027		 * we'll need to know how to clean them up here as well.
5028		 */
5029		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5030			free(io->scsiio.kern_data_ptr, M_CTL);
5031		ctl_done(io);
5032		retval = CTL_RETVAL_COMPLETE;
5033	} else {
5034		/*
5035		 * XXX KDM now we need to continue data movement.  Some
5036		 * options:
5037		 * - call ctl_scsiio() again?  We don't do this for data
5038		 *   writes, because for those at least we know ahead of
5039		 *   time where the write will go and how long it is.  For
5040		 *   config writes, though, that information is largely
5041		 *   contained within the write itself, thus we need to
5042		 *   parse out the data again.
5043		 *
5044		 * - Call some other function once the data is in?
5045		 */
5046		if (ctl_debug & CTL_DEBUG_CDB_DATA)
5047			ctl_data_print(io);
5048
5049		/*
5050		 * XXX KDM call ctl_scsiio() again for now, and check flag
5051		 * bits to see whether we're allocated or not.
5052		 */
5053		retval = ctl_scsiio(&io->scsiio);
5054	}
5055	return (retval);
5056}
5057
5058/*
5059 * This gets called by a backend driver when it is done with a
5060 * data_submit method.
5061 */
5062void
5063ctl_data_submit_done(union ctl_io *io)
5064{
5065	/*
5066	 * If the IO_CONT flag is set, we need to call the supplied
5067	 * function to continue processing the I/O, instead of completing
5068	 * the I/O just yet.
5069	 *
5070	 * If there is an error, though, we don't want to keep processing.
5071	 * Instead, just send status back to the initiator.
5072	 */
5073	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5074	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5075	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5076	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5077		io->scsiio.io_cont(io);
5078		return;
5079	}
5080	ctl_done(io);
5081}
5082
5083/*
5084 * This gets called by a backend driver when it is done with a
5085 * configuration write.
5086 */
5087void
5088ctl_config_write_done(union ctl_io *io)
5089{
5090	uint8_t *buf;
5091
5092	/*
5093	 * If the IO_CONT flag is set, we need to call the supplied
5094	 * function to continue processing the I/O, instead of completing
5095	 * the I/O just yet.
5096	 *
5097	 * If there is an error, though, we don't want to keep processing.
5098	 * Instead, just send status back to the initiator.
5099	 */
5100	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5101	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5102	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5103	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5104		io->scsiio.io_cont(io);
5105		return;
5106	}
5107	/*
5108	 * Since a configuration write can be done for commands that actually
5109	 * have data allocated, like write buffer, and commands that have
5110	 * no data, like start/stop unit, we need to check here.
5111	 */
5112	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5113		buf = io->scsiio.kern_data_ptr;
5114	else
5115		buf = NULL;
5116	ctl_done(io);
5117	if (buf)
5118		free(buf, M_CTL);
5119}
5120
5121/*
5122 * SCSI release command.
5123 */
5124int
5125ctl_scsi_release(struct ctl_scsiio *ctsio)
5126{
5127	int length, longid, thirdparty_id, resv_id;
5128	struct ctl_softc *ctl_softc;
5129	struct ctl_lun *lun;
5130	uint32_t residx;
5131
5132	length = 0;
5133	resv_id = 0;
5134
5135	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5136
5137	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5138	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5139	ctl_softc = control_softc;
5140
5141	switch (ctsio->cdb[0]) {
5142	case RELEASE_10: {
5143		struct scsi_release_10 *cdb;
5144
5145		cdb = (struct scsi_release_10 *)ctsio->cdb;
5146
5147		if (cdb->byte2 & SR10_LONGID)
5148			longid = 1;
5149		else
5150			thirdparty_id = cdb->thirdparty_id;
5151
5152		resv_id = cdb->resv_id;
5153		length = scsi_2btoul(cdb->length);
5154		break;
5155	}
5156	}
5157
5158
5159	/*
5160	 * XXX KDM right now, we only support LUN reservation.  We don't
5161	 * support 3rd party reservations, or extent reservations, which
5162	 * might actually need the parameter list.  If we've gotten this
5163	 * far, we've got a LUN reservation.  Anything else got kicked out
5164	 * above.  So, according to SPC, ignore the length.
5165	 */
5166	length = 0;
5167
5168	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5169	 && (length > 0)) {
5170		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5171		ctsio->kern_data_len = length;
5172		ctsio->kern_total_len = length;
5173		ctsio->kern_data_resid = 0;
5174		ctsio->kern_rel_offset = 0;
5175		ctsio->kern_sg_entries = 0;
5176		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5177		ctsio->be_move_done = ctl_config_move_done;
5178		ctl_datamove((union ctl_io *)ctsio);
5179
5180		return (CTL_RETVAL_COMPLETE);
5181	}
5182
5183	if (length > 0)
5184		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5185
5186	mtx_lock(&lun->lun_lock);
5187
5188	/*
5189	 * According to SPC, it is not an error for an intiator to attempt
5190	 * to release a reservation on a LUN that isn't reserved, or that
5191	 * is reserved by another initiator.  The reservation can only be
5192	 * released, though, by the initiator who made it or by one of
5193	 * several reset type events.
5194	 */
5195	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5196			lun->flags &= ~CTL_LUN_RESERVED;
5197
5198	mtx_unlock(&lun->lun_lock);
5199
5200	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5201		free(ctsio->kern_data_ptr, M_CTL);
5202		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5203	}
5204
5205	ctl_set_success(ctsio);
5206	ctl_done((union ctl_io *)ctsio);
5207	return (CTL_RETVAL_COMPLETE);
5208}
5209
5210int
5211ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5212{
5213	int extent, thirdparty, longid;
5214	int resv_id, length;
5215	uint64_t thirdparty_id;
5216	struct ctl_softc *ctl_softc;
5217	struct ctl_lun *lun;
5218	uint32_t residx;
5219
5220	extent = 0;
5221	thirdparty = 0;
5222	longid = 0;
5223	resv_id = 0;
5224	length = 0;
5225	thirdparty_id = 0;
5226
5227	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5228
5229	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5230	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5231	ctl_softc = control_softc;
5232
5233	switch (ctsio->cdb[0]) {
5234	case RESERVE_10: {
5235		struct scsi_reserve_10 *cdb;
5236
5237		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5238
5239		if (cdb->byte2 & SR10_LONGID)
5240			longid = 1;
5241		else
5242			thirdparty_id = cdb->thirdparty_id;
5243
5244		resv_id = cdb->resv_id;
5245		length = scsi_2btoul(cdb->length);
5246		break;
5247	}
5248	}
5249
5250	/*
5251	 * XXX KDM right now, we only support LUN reservation.  We don't
5252	 * support 3rd party reservations, or extent reservations, which
5253	 * might actually need the parameter list.  If we've gotten this
5254	 * far, we've got a LUN reservation.  Anything else got kicked out
5255	 * above.  So, according to SPC, ignore the length.
5256	 */
5257	length = 0;
5258
5259	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5260	 && (length > 0)) {
5261		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5262		ctsio->kern_data_len = length;
5263		ctsio->kern_total_len = length;
5264		ctsio->kern_data_resid = 0;
5265		ctsio->kern_rel_offset = 0;
5266		ctsio->kern_sg_entries = 0;
5267		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5268		ctsio->be_move_done = ctl_config_move_done;
5269		ctl_datamove((union ctl_io *)ctsio);
5270
5271		return (CTL_RETVAL_COMPLETE);
5272	}
5273
5274	if (length > 0)
5275		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5276
5277	mtx_lock(&lun->lun_lock);
5278	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5279		ctl_set_reservation_conflict(ctsio);
5280		goto bailout;
5281	}
5282
5283	lun->flags |= CTL_LUN_RESERVED;
5284	lun->res_idx = residx;
5285
5286	ctl_set_success(ctsio);
5287
5288bailout:
5289	mtx_unlock(&lun->lun_lock);
5290
5291	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5292		free(ctsio->kern_data_ptr, M_CTL);
5293		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5294	}
5295
5296	ctl_done((union ctl_io *)ctsio);
5297	return (CTL_RETVAL_COMPLETE);
5298}
5299
5300int
5301ctl_start_stop(struct ctl_scsiio *ctsio)
5302{
5303	struct scsi_start_stop_unit *cdb;
5304	struct ctl_lun *lun;
5305	struct ctl_softc *ctl_softc;
5306	int retval;
5307
5308	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5309
5310	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5311	ctl_softc = control_softc;
5312	retval = 0;
5313
5314	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5315
5316	/*
5317	 * XXX KDM
5318	 * We don't support the immediate bit on a stop unit.  In order to
5319	 * do that, we would need to code up a way to know that a stop is
5320	 * pending, and hold off any new commands until it completes, one
5321	 * way or another.  Then we could accept or reject those commands
5322	 * depending on its status.  We would almost need to do the reverse
5323	 * of what we do below for an immediate start -- return the copy of
5324	 * the ctl_io to the FETD with status to send to the host (and to
5325	 * free the copy!) and then free the original I/O once the stop
5326	 * actually completes.  That way, the OOA queue mechanism can work
5327	 * to block commands that shouldn't proceed.  Another alternative
5328	 * would be to put the copy in the queue in place of the original,
5329	 * and return the original back to the caller.  That could be
5330	 * slightly safer..
5331	 */
5332	if ((cdb->byte2 & SSS_IMMED)
5333	 && ((cdb->how & SSS_START) == 0)) {
5334		ctl_set_invalid_field(ctsio,
5335				      /*sks_valid*/ 1,
5336				      /*command*/ 1,
5337				      /*field*/ 1,
5338				      /*bit_valid*/ 1,
5339				      /*bit*/ 0);
5340		ctl_done((union ctl_io *)ctsio);
5341		return (CTL_RETVAL_COMPLETE);
5342	}
5343
5344	if ((lun->flags & CTL_LUN_PR_RESERVED)
5345	 && ((cdb->how & SSS_START)==0)) {
5346		uint32_t residx;
5347
5348		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5349		if (lun->pr_keys[residx] == 0
5350		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5351
5352			ctl_set_reservation_conflict(ctsio);
5353			ctl_done((union ctl_io *)ctsio);
5354			return (CTL_RETVAL_COMPLETE);
5355		}
5356	}
5357
5358	/*
5359	 * If there is no backend on this device, we can't start or stop
5360	 * it.  In theory we shouldn't get any start/stop commands in the
5361	 * first place at this level if the LUN doesn't have a backend.
5362	 * That should get stopped by the command decode code.
5363	 */
5364	if (lun->backend == NULL) {
5365		ctl_set_invalid_opcode(ctsio);
5366		ctl_done((union ctl_io *)ctsio);
5367		return (CTL_RETVAL_COMPLETE);
5368	}
5369
5370	/*
5371	 * XXX KDM Copan-specific offline behavior.
5372	 * Figure out a reasonable way to port this?
5373	 */
5374#ifdef NEEDTOPORT
5375	mtx_lock(&lun->lun_lock);
5376
5377	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5378	 && (lun->flags & CTL_LUN_OFFLINE)) {
5379		/*
5380		 * If the LUN is offline, and the on/offline bit isn't set,
5381		 * reject the start or stop.  Otherwise, let it through.
5382		 */
5383		mtx_unlock(&lun->lun_lock);
5384		ctl_set_lun_not_ready(ctsio);
5385		ctl_done((union ctl_io *)ctsio);
5386	} else {
5387		mtx_unlock(&lun->lun_lock);
5388#endif /* NEEDTOPORT */
5389		/*
5390		 * This could be a start or a stop when we're online,
5391		 * or a stop/offline or start/online.  A start or stop when
5392		 * we're offline is covered in the case above.
5393		 */
5394		/*
5395		 * In the non-immediate case, we send the request to
5396		 * the backend and return status to the user when
5397		 * it is done.
5398		 *
5399		 * In the immediate case, we allocate a new ctl_io
5400		 * to hold a copy of the request, and send that to
5401		 * the backend.  We then set good status on the
5402		 * user's request and return it immediately.
5403		 */
5404		if (cdb->byte2 & SSS_IMMED) {
5405			union ctl_io *new_io;
5406
5407			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5408			ctl_copy_io((union ctl_io *)ctsio, new_io);
5409			retval = lun->backend->config_write(new_io);
5410			ctl_set_success(ctsio);
5411			ctl_done((union ctl_io *)ctsio);
5412		} else {
5413			retval = lun->backend->config_write(
5414				(union ctl_io *)ctsio);
5415		}
5416#ifdef NEEDTOPORT
5417	}
5418#endif
5419	return (retval);
5420}
5421
5422/*
5423 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5424 * we don't really do anything with the LBA and length fields if the user
5425 * passes them in.  Instead we'll just flush out the cache for the entire
5426 * LUN.
5427 */
5428int
5429ctl_sync_cache(struct ctl_scsiio *ctsio)
5430{
5431	struct ctl_lun *lun;
5432	struct ctl_softc *ctl_softc;
5433	uint64_t starting_lba;
5434	uint32_t block_count;
5435	int retval;
5436
5437	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5438
5439	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5440	ctl_softc = control_softc;
5441	retval = 0;
5442
5443	switch (ctsio->cdb[0]) {
5444	case SYNCHRONIZE_CACHE: {
5445		struct scsi_sync_cache *cdb;
5446		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5447
5448		starting_lba = scsi_4btoul(cdb->begin_lba);
5449		block_count = scsi_2btoul(cdb->lb_count);
5450		break;
5451	}
5452	case SYNCHRONIZE_CACHE_16: {
5453		struct scsi_sync_cache_16 *cdb;
5454		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5455
5456		starting_lba = scsi_8btou64(cdb->begin_lba);
5457		block_count = scsi_4btoul(cdb->lb_count);
5458		break;
5459	}
5460	default:
5461		ctl_set_invalid_opcode(ctsio);
5462		ctl_done((union ctl_io *)ctsio);
5463		goto bailout;
5464		break; /* NOTREACHED */
5465	}
5466
5467	/*
5468	 * We check the LBA and length, but don't do anything with them.
5469	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5470	 * get flushed.  This check will just help satisfy anyone who wants
5471	 * to see an error for an out of range LBA.
5472	 */
5473	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5474		ctl_set_lba_out_of_range(ctsio);
5475		ctl_done((union ctl_io *)ctsio);
5476		goto bailout;
5477	}
5478
5479	/*
5480	 * If this LUN has no backend, we can't flush the cache anyway.
5481	 */
5482	if (lun->backend == NULL) {
5483		ctl_set_invalid_opcode(ctsio);
5484		ctl_done((union ctl_io *)ctsio);
5485		goto bailout;
5486	}
5487
5488	/*
5489	 * Check to see whether we're configured to send the SYNCHRONIZE
5490	 * CACHE command directly to the back end.
5491	 */
5492	mtx_lock(&lun->lun_lock);
5493	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5494	 && (++(lun->sync_count) >= lun->sync_interval)) {
5495		lun->sync_count = 0;
5496		mtx_unlock(&lun->lun_lock);
5497		retval = lun->backend->config_write((union ctl_io *)ctsio);
5498	} else {
5499		mtx_unlock(&lun->lun_lock);
5500		ctl_set_success(ctsio);
5501		ctl_done((union ctl_io *)ctsio);
5502	}
5503
5504bailout:
5505
5506	return (retval);
5507}
5508
5509int
5510ctl_format(struct ctl_scsiio *ctsio)
5511{
5512	struct scsi_format *cdb;
5513	struct ctl_lun *lun;
5514	struct ctl_softc *ctl_softc;
5515	int length, defect_list_len;
5516
5517	CTL_DEBUG_PRINT(("ctl_format\n"));
5518
5519	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5520	ctl_softc = control_softc;
5521
5522	cdb = (struct scsi_format *)ctsio->cdb;
5523
5524	length = 0;
5525	if (cdb->byte2 & SF_FMTDATA) {
5526		if (cdb->byte2 & SF_LONGLIST)
5527			length = sizeof(struct scsi_format_header_long);
5528		else
5529			length = sizeof(struct scsi_format_header_short);
5530	}
5531
5532	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5533	 && (length > 0)) {
5534		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5535		ctsio->kern_data_len = length;
5536		ctsio->kern_total_len = length;
5537		ctsio->kern_data_resid = 0;
5538		ctsio->kern_rel_offset = 0;
5539		ctsio->kern_sg_entries = 0;
5540		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5541		ctsio->be_move_done = ctl_config_move_done;
5542		ctl_datamove((union ctl_io *)ctsio);
5543
5544		return (CTL_RETVAL_COMPLETE);
5545	}
5546
5547	defect_list_len = 0;
5548
5549	if (cdb->byte2 & SF_FMTDATA) {
5550		if (cdb->byte2 & SF_LONGLIST) {
5551			struct scsi_format_header_long *header;
5552
5553			header = (struct scsi_format_header_long *)
5554				ctsio->kern_data_ptr;
5555
5556			defect_list_len = scsi_4btoul(header->defect_list_len);
5557			if (defect_list_len != 0) {
5558				ctl_set_invalid_field(ctsio,
5559						      /*sks_valid*/ 1,
5560						      /*command*/ 0,
5561						      /*field*/ 2,
5562						      /*bit_valid*/ 0,
5563						      /*bit*/ 0);
5564				goto bailout;
5565			}
5566		} else {
5567			struct scsi_format_header_short *header;
5568
5569			header = (struct scsi_format_header_short *)
5570				ctsio->kern_data_ptr;
5571
5572			defect_list_len = scsi_2btoul(header->defect_list_len);
5573			if (defect_list_len != 0) {
5574				ctl_set_invalid_field(ctsio,
5575						      /*sks_valid*/ 1,
5576						      /*command*/ 0,
5577						      /*field*/ 2,
5578						      /*bit_valid*/ 0,
5579						      /*bit*/ 0);
5580				goto bailout;
5581			}
5582		}
5583	}
5584
5585	/*
5586	 * The format command will clear out the "Medium format corrupted"
5587	 * status if set by the configuration code.  That status is really
5588	 * just a way to notify the host that we have lost the media, and
5589	 * get them to issue a command that will basically make them think
5590	 * they're blowing away the media.
5591	 */
5592	mtx_lock(&lun->lun_lock);
5593	lun->flags &= ~CTL_LUN_INOPERABLE;
5594	mtx_unlock(&lun->lun_lock);
5595
5596	ctl_set_success(ctsio);
5597bailout:
5598
5599	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5600		free(ctsio->kern_data_ptr, M_CTL);
5601		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5602	}
5603
5604	ctl_done((union ctl_io *)ctsio);
5605	return (CTL_RETVAL_COMPLETE);
5606}
5607
5608int
5609ctl_read_buffer(struct ctl_scsiio *ctsio)
5610{
5611	struct scsi_read_buffer *cdb;
5612	struct ctl_lun *lun;
5613	int buffer_offset, len;
5614	static uint8_t descr[4];
5615	static uint8_t echo_descr[4] = { 0 };
5616
5617	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5618
5619	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5620	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5621
5622	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5623	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5624	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5625		ctl_set_invalid_field(ctsio,
5626				      /*sks_valid*/ 1,
5627				      /*command*/ 1,
5628				      /*field*/ 1,
5629				      /*bit_valid*/ 1,
5630				      /*bit*/ 4);
5631		ctl_done((union ctl_io *)ctsio);
5632		return (CTL_RETVAL_COMPLETE);
5633	}
5634
5635	len = scsi_3btoul(cdb->length);
5636	buffer_offset = scsi_3btoul(cdb->offset);
5637
5638	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5639		ctl_set_invalid_field(ctsio,
5640				      /*sks_valid*/ 1,
5641				      /*command*/ 1,
5642				      /*field*/ 6,
5643				      /*bit_valid*/ 0,
5644				      /*bit*/ 0);
5645		ctl_done((union ctl_io *)ctsio);
5646		return (CTL_RETVAL_COMPLETE);
5647	}
5648
5649	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5650		descr[0] = 0;
5651		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5652		ctsio->kern_data_ptr = descr;
5653		len = min(len, sizeof(descr));
5654	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5655		ctsio->kern_data_ptr = echo_descr;
5656		len = min(len, sizeof(echo_descr));
5657	} else {
5658		if (lun->write_buffer == NULL) {
5659			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5660			    M_CTL, M_WAITOK);
5661		}
5662		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5663	}
5664	ctsio->kern_data_len = len;
5665	ctsio->kern_total_len = len;
5666	ctsio->kern_data_resid = 0;
5667	ctsio->kern_rel_offset = 0;
5668	ctsio->kern_sg_entries = 0;
5669	ctl_set_success(ctsio);
5670	ctsio->be_move_done = ctl_config_move_done;
5671	ctl_datamove((union ctl_io *)ctsio);
5672	return (CTL_RETVAL_COMPLETE);
5673}
5674
5675int
5676ctl_write_buffer(struct ctl_scsiio *ctsio)
5677{
5678	struct scsi_write_buffer *cdb;
5679	struct ctl_lun *lun;
5680	int buffer_offset, len;
5681
5682	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5683
5684	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5685	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5686
5687	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5688		ctl_set_invalid_field(ctsio,
5689				      /*sks_valid*/ 1,
5690				      /*command*/ 1,
5691				      /*field*/ 1,
5692				      /*bit_valid*/ 1,
5693				      /*bit*/ 4);
5694		ctl_done((union ctl_io *)ctsio);
5695		return (CTL_RETVAL_COMPLETE);
5696	}
5697
5698	len = scsi_3btoul(cdb->length);
5699	buffer_offset = scsi_3btoul(cdb->offset);
5700
5701	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5702		ctl_set_invalid_field(ctsio,
5703				      /*sks_valid*/ 1,
5704				      /*command*/ 1,
5705				      /*field*/ 6,
5706				      /*bit_valid*/ 0,
5707				      /*bit*/ 0);
5708		ctl_done((union ctl_io *)ctsio);
5709		return (CTL_RETVAL_COMPLETE);
5710	}
5711
5712	/*
5713	 * If we've got a kernel request that hasn't been malloced yet,
5714	 * malloc it and tell the caller the data buffer is here.
5715	 */
5716	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5717		if (lun->write_buffer == NULL) {
5718			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5719			    M_CTL, M_WAITOK);
5720		}
5721		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5722		ctsio->kern_data_len = len;
5723		ctsio->kern_total_len = len;
5724		ctsio->kern_data_resid = 0;
5725		ctsio->kern_rel_offset = 0;
5726		ctsio->kern_sg_entries = 0;
5727		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5728		ctsio->be_move_done = ctl_config_move_done;
5729		ctl_datamove((union ctl_io *)ctsio);
5730
5731		return (CTL_RETVAL_COMPLETE);
5732	}
5733
5734	ctl_set_success(ctsio);
5735	ctl_done((union ctl_io *)ctsio);
5736	return (CTL_RETVAL_COMPLETE);
5737}
5738
5739int
5740ctl_write_same(struct ctl_scsiio *ctsio)
5741{
5742	struct ctl_lun *lun;
5743	struct ctl_lba_len_flags *lbalen;
5744	uint64_t lba;
5745	uint32_t num_blocks;
5746	int len, retval;
5747	uint8_t byte2;
5748
5749	retval = CTL_RETVAL_COMPLETE;
5750
5751	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5752
5753	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5754
5755	switch (ctsio->cdb[0]) {
5756	case WRITE_SAME_10: {
5757		struct scsi_write_same_10 *cdb;
5758
5759		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5760
5761		lba = scsi_4btoul(cdb->addr);
5762		num_blocks = scsi_2btoul(cdb->length);
5763		byte2 = cdb->byte2;
5764		break;
5765	}
5766	case WRITE_SAME_16: {
5767		struct scsi_write_same_16 *cdb;
5768
5769		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5770
5771		lba = scsi_8btou64(cdb->addr);
5772		num_blocks = scsi_4btoul(cdb->length);
5773		byte2 = cdb->byte2;
5774		break;
5775	}
5776	default:
5777		/*
5778		 * We got a command we don't support.  This shouldn't
5779		 * happen, commands should be filtered out above us.
5780		 */
5781		ctl_set_invalid_opcode(ctsio);
5782		ctl_done((union ctl_io *)ctsio);
5783
5784		return (CTL_RETVAL_COMPLETE);
5785		break; /* NOTREACHED */
5786	}
5787
5788	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5789	if ((byte2 & SWS_UNMAP) == 0 &&
5790	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5791		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5792		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5793		ctl_done((union ctl_io *)ctsio);
5794		return (CTL_RETVAL_COMPLETE);
5795	}
5796
5797	/*
5798	 * The first check is to make sure we're in bounds, the second
5799	 * check is to catch wrap-around problems.  If the lba + num blocks
5800	 * is less than the lba, then we've wrapped around and the block
5801	 * range is invalid anyway.
5802	 */
5803	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5804	 || ((lba + num_blocks) < lba)) {
5805		ctl_set_lba_out_of_range(ctsio);
5806		ctl_done((union ctl_io *)ctsio);
5807		return (CTL_RETVAL_COMPLETE);
5808	}
5809
5810	/* Zero number of blocks means "to the last logical block" */
5811	if (num_blocks == 0) {
5812		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5813			ctl_set_invalid_field(ctsio,
5814					      /*sks_valid*/ 0,
5815					      /*command*/ 1,
5816					      /*field*/ 0,
5817					      /*bit_valid*/ 0,
5818					      /*bit*/ 0);
5819			ctl_done((union ctl_io *)ctsio);
5820			return (CTL_RETVAL_COMPLETE);
5821		}
5822		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5823	}
5824
5825	len = lun->be_lun->blocksize;
5826
5827	/*
5828	 * If we've got a kernel request that hasn't been malloced yet,
5829	 * malloc it and tell the caller the data buffer is here.
5830	 */
5831	if ((byte2 & SWS_NDOB) == 0 &&
5832	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5833		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5834		ctsio->kern_data_len = len;
5835		ctsio->kern_total_len = len;
5836		ctsio->kern_data_resid = 0;
5837		ctsio->kern_rel_offset = 0;
5838		ctsio->kern_sg_entries = 0;
5839		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5840		ctsio->be_move_done = ctl_config_move_done;
5841		ctl_datamove((union ctl_io *)ctsio);
5842
5843		return (CTL_RETVAL_COMPLETE);
5844	}
5845
5846	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5847	lbalen->lba = lba;
5848	lbalen->len = num_blocks;
5849	lbalen->flags = byte2;
5850	retval = lun->backend->config_write((union ctl_io *)ctsio);
5851
5852	return (retval);
5853}
5854
5855int
5856ctl_unmap(struct ctl_scsiio *ctsio)
5857{
5858	struct ctl_lun *lun;
5859	struct scsi_unmap *cdb;
5860	struct ctl_ptr_len_flags *ptrlen;
5861	struct scsi_unmap_header *hdr;
5862	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5863	uint64_t lba;
5864	uint32_t num_blocks;
5865	int len, retval;
5866	uint8_t byte2;
5867
5868	retval = CTL_RETVAL_COMPLETE;
5869
5870	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5871
5872	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5873	cdb = (struct scsi_unmap *)ctsio->cdb;
5874
5875	len = scsi_2btoul(cdb->length);
5876	byte2 = cdb->byte2;
5877
5878	/*
5879	 * If we've got a kernel request that hasn't been malloced yet,
5880	 * malloc it and tell the caller the data buffer is here.
5881	 */
5882	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5883		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5884		ctsio->kern_data_len = len;
5885		ctsio->kern_total_len = len;
5886		ctsio->kern_data_resid = 0;
5887		ctsio->kern_rel_offset = 0;
5888		ctsio->kern_sg_entries = 0;
5889		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5890		ctsio->be_move_done = ctl_config_move_done;
5891		ctl_datamove((union ctl_io *)ctsio);
5892
5893		return (CTL_RETVAL_COMPLETE);
5894	}
5895
5896	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5897	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5898	if (len < sizeof (*hdr) ||
5899	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5900	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5901	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5902		ctl_set_invalid_field(ctsio,
5903				      /*sks_valid*/ 0,
5904				      /*command*/ 0,
5905				      /*field*/ 0,
5906				      /*bit_valid*/ 0,
5907				      /*bit*/ 0);
5908		ctl_done((union ctl_io *)ctsio);
5909		return (CTL_RETVAL_COMPLETE);
5910	}
5911	len = scsi_2btoul(hdr->desc_length);
5912	buf = (struct scsi_unmap_desc *)(hdr + 1);
5913	end = buf + len / sizeof(*buf);
5914
5915	endnz = buf;
5916	for (range = buf; range < end; range++) {
5917		lba = scsi_8btou64(range->lba);
5918		num_blocks = scsi_4btoul(range->length);
5919		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5920		 || ((lba + num_blocks) < lba)) {
5921			ctl_set_lba_out_of_range(ctsio);
5922			ctl_done((union ctl_io *)ctsio);
5923			return (CTL_RETVAL_COMPLETE);
5924		}
5925		if (num_blocks != 0)
5926			endnz = range + 1;
5927	}
5928
5929	/*
5930	 * Block backend can not handle zero last range.
5931	 * Filter it out and return if there is nothing left.
5932	 */
5933	len = (uint8_t *)endnz - (uint8_t *)buf;
5934	if (len == 0) {
5935		ctl_set_success(ctsio);
5936		ctl_done((union ctl_io *)ctsio);
5937		return (CTL_RETVAL_COMPLETE);
5938	}
5939
5940	mtx_lock(&lun->lun_lock);
5941	ptrlen = (struct ctl_ptr_len_flags *)
5942	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5943	ptrlen->ptr = (void *)buf;
5944	ptrlen->len = len;
5945	ptrlen->flags = byte2;
5946	ctl_check_blocked(lun);
5947	mtx_unlock(&lun->lun_lock);
5948
5949	retval = lun->backend->config_write((union ctl_io *)ctsio);
5950	return (retval);
5951}
5952
5953/*
5954 * Note that this function currently doesn't actually do anything inside
5955 * CTL to enforce things if the DQue bit is turned on.
5956 *
5957 * Also note that this function can't be used in the default case, because
5958 * the DQue bit isn't set in the changeable mask for the control mode page
5959 * anyway.  This is just here as an example for how to implement a page
5960 * handler, and a placeholder in case we want to allow the user to turn
5961 * tagged queueing on and off.
5962 *
5963 * The D_SENSE bit handling is functional, however, and will turn
5964 * descriptor sense on and off for a given LUN.
5965 */
5966int
5967ctl_control_page_handler(struct ctl_scsiio *ctsio,
5968			 struct ctl_page_index *page_index, uint8_t *page_ptr)
5969{
5970	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5971	struct ctl_lun *lun;
5972	struct ctl_softc *softc;
5973	int set_ua;
5974	uint32_t initidx;
5975
5976	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5977	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5978	set_ua = 0;
5979
5980	user_cp = (struct scsi_control_page *)page_ptr;
5981	current_cp = (struct scsi_control_page *)
5982		(page_index->page_data + (page_index->page_len *
5983		CTL_PAGE_CURRENT));
5984	saved_cp = (struct scsi_control_page *)
5985		(page_index->page_data + (page_index->page_len *
5986		CTL_PAGE_SAVED));
5987
5988	softc = control_softc;
5989
5990	mtx_lock(&lun->lun_lock);
5991	if (((current_cp->rlec & SCP_DSENSE) == 0)
5992	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5993		/*
5994		 * Descriptor sense is currently turned off and the user
5995		 * wants to turn it on.
5996		 */
5997		current_cp->rlec |= SCP_DSENSE;
5998		saved_cp->rlec |= SCP_DSENSE;
5999		lun->flags |= CTL_LUN_SENSE_DESC;
6000		set_ua = 1;
6001	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6002		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6003		/*
6004		 * Descriptor sense is currently turned on, and the user
6005		 * wants to turn it off.
6006		 */
6007		current_cp->rlec &= ~SCP_DSENSE;
6008		saved_cp->rlec &= ~SCP_DSENSE;
6009		lun->flags &= ~CTL_LUN_SENSE_DESC;
6010		set_ua = 1;
6011	}
6012	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6013	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6014		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6015		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6016		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6017		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6018		set_ua = 1;
6019	}
6020	if ((current_cp->eca_and_aen & SCP_SWP) !=
6021	    (user_cp->eca_and_aen & SCP_SWP)) {
6022		current_cp->eca_and_aen &= ~SCP_SWP;
6023		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6024		saved_cp->eca_and_aen &= ~SCP_SWP;
6025		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6026		set_ua = 1;
6027	}
6028	if (set_ua != 0) {
6029		int i;
6030		/*
6031		 * Let other initiators know that the mode
6032		 * parameters for this LUN have changed.
6033		 */
6034		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6035			if (i == initidx)
6036				continue;
6037
6038			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6039		}
6040	}
6041	mtx_unlock(&lun->lun_lock);
6042
6043	return (0);
6044}
6045
6046int
6047ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6048		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6049{
6050	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6051	struct ctl_lun *lun;
6052	int set_ua;
6053	uint32_t initidx;
6054
6055	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6056	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6057	set_ua = 0;
6058
6059	user_cp = (struct scsi_caching_page *)page_ptr;
6060	current_cp = (struct scsi_caching_page *)
6061		(page_index->page_data + (page_index->page_len *
6062		CTL_PAGE_CURRENT));
6063	saved_cp = (struct scsi_caching_page *)
6064		(page_index->page_data + (page_index->page_len *
6065		CTL_PAGE_SAVED));
6066
6067	mtx_lock(&lun->lun_lock);
6068	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6069	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6070		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6071		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6072		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6073		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6074		set_ua = 1;
6075	}
6076	if (set_ua != 0) {
6077		int i;
6078		/*
6079		 * Let other initiators know that the mode
6080		 * parameters for this LUN have changed.
6081		 */
6082		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6083			if (i == initidx)
6084				continue;
6085
6086			lun->pending_ua[i] |= CTL_UA_MODE_CHANGE;
6087		}
6088	}
6089	mtx_unlock(&lun->lun_lock);
6090
6091	return (0);
6092}
6093
6094int
6095ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6096				struct ctl_page_index *page_index,
6097				uint8_t *page_ptr)
6098{
6099	uint8_t *c;
6100	int i;
6101
6102	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6103	ctl_time_io_secs =
6104		(c[0] << 8) |
6105		(c[1] << 0) |
6106		0;
6107	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6108	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6109	printf("page data:");
6110	for (i=0; i<8; i++)
6111		printf(" %.2x",page_ptr[i]);
6112	printf("\n");
6113	return (0);
6114}
6115
6116int
6117ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6118			       struct ctl_page_index *page_index,
6119			       int pc)
6120{
6121	struct copan_debugconf_subpage *page;
6122
6123	page = (struct copan_debugconf_subpage *)page_index->page_data +
6124		(page_index->page_len * pc);
6125
6126	switch (pc) {
6127	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6128	case SMS_PAGE_CTRL_DEFAULT >> 6:
6129	case SMS_PAGE_CTRL_SAVED >> 6:
6130		/*
6131		 * We don't update the changable or default bits for this page.
6132		 */
6133		break;
6134	case SMS_PAGE_CTRL_CURRENT >> 6:
6135		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6136		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6137		break;
6138	default:
6139#ifdef NEEDTOPORT
6140		EPRINT(0, "Invalid PC %d!!", pc);
6141#endif /* NEEDTOPORT */
6142		break;
6143	}
6144	return (0);
6145}
6146
6147
6148static int
6149ctl_do_mode_select(union ctl_io *io)
6150{
6151	struct scsi_mode_page_header *page_header;
6152	struct ctl_page_index *page_index;
6153	struct ctl_scsiio *ctsio;
6154	int control_dev, page_len;
6155	int page_len_offset, page_len_size;
6156	union ctl_modepage_info *modepage_info;
6157	struct ctl_lun *lun;
6158	int *len_left, *len_used;
6159	int retval, i;
6160
6161	ctsio = &io->scsiio;
6162	page_index = NULL;
6163	page_len = 0;
6164	retval = CTL_RETVAL_COMPLETE;
6165
6166	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6167
6168	if (lun->be_lun->lun_type != T_DIRECT)
6169		control_dev = 1;
6170	else
6171		control_dev = 0;
6172
6173	modepage_info = (union ctl_modepage_info *)
6174		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6175	len_left = &modepage_info->header.len_left;
6176	len_used = &modepage_info->header.len_used;
6177
6178do_next_page:
6179
6180	page_header = (struct scsi_mode_page_header *)
6181		(ctsio->kern_data_ptr + *len_used);
6182
6183	if (*len_left == 0) {
6184		free(ctsio->kern_data_ptr, M_CTL);
6185		ctl_set_success(ctsio);
6186		ctl_done((union ctl_io *)ctsio);
6187		return (CTL_RETVAL_COMPLETE);
6188	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6189
6190		free(ctsio->kern_data_ptr, M_CTL);
6191		ctl_set_param_len_error(ctsio);
6192		ctl_done((union ctl_io *)ctsio);
6193		return (CTL_RETVAL_COMPLETE);
6194
6195	} else if ((page_header->page_code & SMPH_SPF)
6196		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6197
6198		free(ctsio->kern_data_ptr, M_CTL);
6199		ctl_set_param_len_error(ctsio);
6200		ctl_done((union ctl_io *)ctsio);
6201		return (CTL_RETVAL_COMPLETE);
6202	}
6203
6204
6205	/*
6206	 * XXX KDM should we do something with the block descriptor?
6207	 */
6208	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6209
6210		if ((control_dev != 0)
6211		 && (lun->mode_pages.index[i].page_flags &
6212		     CTL_PAGE_FLAG_DISK_ONLY))
6213			continue;
6214
6215		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6216		    (page_header->page_code & SMPH_PC_MASK))
6217			continue;
6218
6219		/*
6220		 * If neither page has a subpage code, then we've got a
6221		 * match.
6222		 */
6223		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6224		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6225			page_index = &lun->mode_pages.index[i];
6226			page_len = page_header->page_length;
6227			break;
6228		}
6229
6230		/*
6231		 * If both pages have subpages, then the subpage numbers
6232		 * have to match.
6233		 */
6234		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6235		  && (page_header->page_code & SMPH_SPF)) {
6236			struct scsi_mode_page_header_sp *sph;
6237
6238			sph = (struct scsi_mode_page_header_sp *)page_header;
6239
6240			if (lun->mode_pages.index[i].subpage ==
6241			    sph->subpage) {
6242				page_index = &lun->mode_pages.index[i];
6243				page_len = scsi_2btoul(sph->page_length);
6244				break;
6245			}
6246		}
6247	}
6248
6249	/*
6250	 * If we couldn't find the page, or if we don't have a mode select
6251	 * handler for it, send back an error to the user.
6252	 */
6253	if ((page_index == NULL)
6254	 || (page_index->select_handler == NULL)) {
6255		ctl_set_invalid_field(ctsio,
6256				      /*sks_valid*/ 1,
6257				      /*command*/ 0,
6258				      /*field*/ *len_used,
6259				      /*bit_valid*/ 0,
6260				      /*bit*/ 0);
6261		free(ctsio->kern_data_ptr, M_CTL);
6262		ctl_done((union ctl_io *)ctsio);
6263		return (CTL_RETVAL_COMPLETE);
6264	}
6265
6266	if (page_index->page_code & SMPH_SPF) {
6267		page_len_offset = 2;
6268		page_len_size = 2;
6269	} else {
6270		page_len_size = 1;
6271		page_len_offset = 1;
6272	}
6273
6274	/*
6275	 * If the length the initiator gives us isn't the one we specify in
6276	 * the mode page header, or if they didn't specify enough data in
6277	 * the CDB to avoid truncating this page, kick out the request.
6278	 */
6279	if ((page_len != (page_index->page_len - page_len_offset -
6280			  page_len_size))
6281	 || (*len_left < page_index->page_len)) {
6282
6283
6284		ctl_set_invalid_field(ctsio,
6285				      /*sks_valid*/ 1,
6286				      /*command*/ 0,
6287				      /*field*/ *len_used + page_len_offset,
6288				      /*bit_valid*/ 0,
6289				      /*bit*/ 0);
6290		free(ctsio->kern_data_ptr, M_CTL);
6291		ctl_done((union ctl_io *)ctsio);
6292		return (CTL_RETVAL_COMPLETE);
6293	}
6294
6295	/*
6296	 * Run through the mode page, checking to make sure that the bits
6297	 * the user changed are actually legal for him to change.
6298	 */
6299	for (i = 0; i < page_index->page_len; i++) {
6300		uint8_t *user_byte, *change_mask, *current_byte;
6301		int bad_bit;
6302		int j;
6303
6304		user_byte = (uint8_t *)page_header + i;
6305		change_mask = page_index->page_data +
6306			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6307		current_byte = page_index->page_data +
6308			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6309
6310		/*
6311		 * Check to see whether the user set any bits in this byte
6312		 * that he is not allowed to set.
6313		 */
6314		if ((*user_byte & ~(*change_mask)) ==
6315		    (*current_byte & ~(*change_mask)))
6316			continue;
6317
6318		/*
6319		 * Go through bit by bit to determine which one is illegal.
6320		 */
6321		bad_bit = 0;
6322		for (j = 7; j >= 0; j--) {
6323			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6324			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6325				bad_bit = i;
6326				break;
6327			}
6328		}
6329		ctl_set_invalid_field(ctsio,
6330				      /*sks_valid*/ 1,
6331				      /*command*/ 0,
6332				      /*field*/ *len_used + i,
6333				      /*bit_valid*/ 1,
6334				      /*bit*/ bad_bit);
6335		free(ctsio->kern_data_ptr, M_CTL);
6336		ctl_done((union ctl_io *)ctsio);
6337		return (CTL_RETVAL_COMPLETE);
6338	}
6339
6340	/*
6341	 * Decrement these before we call the page handler, since we may
6342	 * end up getting called back one way or another before the handler
6343	 * returns to this context.
6344	 */
6345	*len_left -= page_index->page_len;
6346	*len_used += page_index->page_len;
6347
6348	retval = page_index->select_handler(ctsio, page_index,
6349					    (uint8_t *)page_header);
6350
6351	/*
6352	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6353	 * wait until this queued command completes to finish processing
6354	 * the mode page.  If it returns anything other than
6355	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6356	 * already set the sense information, freed the data pointer, and
6357	 * completed the io for us.
6358	 */
6359	if (retval != CTL_RETVAL_COMPLETE)
6360		goto bailout_no_done;
6361
6362	/*
6363	 * If the initiator sent us more than one page, parse the next one.
6364	 */
6365	if (*len_left > 0)
6366		goto do_next_page;
6367
6368	ctl_set_success(ctsio);
6369	free(ctsio->kern_data_ptr, M_CTL);
6370	ctl_done((union ctl_io *)ctsio);
6371
6372bailout_no_done:
6373
6374	return (CTL_RETVAL_COMPLETE);
6375
6376}
6377
6378int
6379ctl_mode_select(struct ctl_scsiio *ctsio)
6380{
6381	int param_len, pf, sp;
6382	int header_size, bd_len;
6383	int len_left, len_used;
6384	struct ctl_page_index *page_index;
6385	struct ctl_lun *lun;
6386	int control_dev, page_len;
6387	union ctl_modepage_info *modepage_info;
6388	int retval;
6389
6390	pf = 0;
6391	sp = 0;
6392	page_len = 0;
6393	len_used = 0;
6394	len_left = 0;
6395	retval = 0;
6396	bd_len = 0;
6397	page_index = NULL;
6398
6399	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6400
6401	if (lun->be_lun->lun_type != T_DIRECT)
6402		control_dev = 1;
6403	else
6404		control_dev = 0;
6405
6406	switch (ctsio->cdb[0]) {
6407	case MODE_SELECT_6: {
6408		struct scsi_mode_select_6 *cdb;
6409
6410		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6411
6412		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6413		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6414
6415		param_len = cdb->length;
6416		header_size = sizeof(struct scsi_mode_header_6);
6417		break;
6418	}
6419	case MODE_SELECT_10: {
6420		struct scsi_mode_select_10 *cdb;
6421
6422		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6423
6424		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6425		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6426
6427		param_len = scsi_2btoul(cdb->length);
6428		header_size = sizeof(struct scsi_mode_header_10);
6429		break;
6430	}
6431	default:
6432		ctl_set_invalid_opcode(ctsio);
6433		ctl_done((union ctl_io *)ctsio);
6434		return (CTL_RETVAL_COMPLETE);
6435		break; /* NOTREACHED */
6436	}
6437
6438	/*
6439	 * From SPC-3:
6440	 * "A parameter list length of zero indicates that the Data-Out Buffer
6441	 * shall be empty. This condition shall not be considered as an error."
6442	 */
6443	if (param_len == 0) {
6444		ctl_set_success(ctsio);
6445		ctl_done((union ctl_io *)ctsio);
6446		return (CTL_RETVAL_COMPLETE);
6447	}
6448
6449	/*
6450	 * Since we'll hit this the first time through, prior to
6451	 * allocation, we don't need to free a data buffer here.
6452	 */
6453	if (param_len < header_size) {
6454		ctl_set_param_len_error(ctsio);
6455		ctl_done((union ctl_io *)ctsio);
6456		return (CTL_RETVAL_COMPLETE);
6457	}
6458
6459	/*
6460	 * Allocate the data buffer and grab the user's data.  In theory,
6461	 * we shouldn't have to sanity check the parameter list length here
6462	 * because the maximum size is 64K.  We should be able to malloc
6463	 * that much without too many problems.
6464	 */
6465	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6466		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6467		ctsio->kern_data_len = param_len;
6468		ctsio->kern_total_len = param_len;
6469		ctsio->kern_data_resid = 0;
6470		ctsio->kern_rel_offset = 0;
6471		ctsio->kern_sg_entries = 0;
6472		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6473		ctsio->be_move_done = ctl_config_move_done;
6474		ctl_datamove((union ctl_io *)ctsio);
6475
6476		return (CTL_RETVAL_COMPLETE);
6477	}
6478
6479	switch (ctsio->cdb[0]) {
6480	case MODE_SELECT_6: {
6481		struct scsi_mode_header_6 *mh6;
6482
6483		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6484		bd_len = mh6->blk_desc_len;
6485		break;
6486	}
6487	case MODE_SELECT_10: {
6488		struct scsi_mode_header_10 *mh10;
6489
6490		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6491		bd_len = scsi_2btoul(mh10->blk_desc_len);
6492		break;
6493	}
6494	default:
6495		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6496		break;
6497	}
6498
6499	if (param_len < (header_size + bd_len)) {
6500		free(ctsio->kern_data_ptr, M_CTL);
6501		ctl_set_param_len_error(ctsio);
6502		ctl_done((union ctl_io *)ctsio);
6503		return (CTL_RETVAL_COMPLETE);
6504	}
6505
6506	/*
6507	 * Set the IO_CONT flag, so that if this I/O gets passed to
6508	 * ctl_config_write_done(), it'll get passed back to
6509	 * ctl_do_mode_select() for further processing, or completion if
6510	 * we're all done.
6511	 */
6512	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6513	ctsio->io_cont = ctl_do_mode_select;
6514
6515	modepage_info = (union ctl_modepage_info *)
6516		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6517
6518	memset(modepage_info, 0, sizeof(*modepage_info));
6519
6520	len_left = param_len - header_size - bd_len;
6521	len_used = header_size + bd_len;
6522
6523	modepage_info->header.len_left = len_left;
6524	modepage_info->header.len_used = len_used;
6525
6526	return (ctl_do_mode_select((union ctl_io *)ctsio));
6527}
6528
6529int
6530ctl_mode_sense(struct ctl_scsiio *ctsio)
6531{
6532	struct ctl_lun *lun;
6533	int pc, page_code, dbd, llba, subpage;
6534	int alloc_len, page_len, header_len, total_len;
6535	struct scsi_mode_block_descr *block_desc;
6536	struct ctl_page_index *page_index;
6537	int control_dev;
6538
6539	dbd = 0;
6540	llba = 0;
6541	block_desc = NULL;
6542	page_index = NULL;
6543
6544	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6545
6546	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6547
6548	if (lun->be_lun->lun_type != T_DIRECT)
6549		control_dev = 1;
6550	else
6551		control_dev = 0;
6552
6553	switch (ctsio->cdb[0]) {
6554	case MODE_SENSE_6: {
6555		struct scsi_mode_sense_6 *cdb;
6556
6557		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6558
6559		header_len = sizeof(struct scsi_mode_hdr_6);
6560		if (cdb->byte2 & SMS_DBD)
6561			dbd = 1;
6562		else
6563			header_len += sizeof(struct scsi_mode_block_descr);
6564
6565		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6566		page_code = cdb->page & SMS_PAGE_CODE;
6567		subpage = cdb->subpage;
6568		alloc_len = cdb->length;
6569		break;
6570	}
6571	case MODE_SENSE_10: {
6572		struct scsi_mode_sense_10 *cdb;
6573
6574		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6575
6576		header_len = sizeof(struct scsi_mode_hdr_10);
6577
6578		if (cdb->byte2 & SMS_DBD)
6579			dbd = 1;
6580		else
6581			header_len += sizeof(struct scsi_mode_block_descr);
6582		if (cdb->byte2 & SMS10_LLBAA)
6583			llba = 1;
6584		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6585		page_code = cdb->page & SMS_PAGE_CODE;
6586		subpage = cdb->subpage;
6587		alloc_len = scsi_2btoul(cdb->length);
6588		break;
6589	}
6590	default:
6591		ctl_set_invalid_opcode(ctsio);
6592		ctl_done((union ctl_io *)ctsio);
6593		return (CTL_RETVAL_COMPLETE);
6594		break; /* NOTREACHED */
6595	}
6596
6597	/*
6598	 * We have to make a first pass through to calculate the size of
6599	 * the pages that match the user's query.  Then we allocate enough
6600	 * memory to hold it, and actually copy the data into the buffer.
6601	 */
6602	switch (page_code) {
6603	case SMS_ALL_PAGES_PAGE: {
6604		int i;
6605
6606		page_len = 0;
6607
6608		/*
6609		 * At the moment, values other than 0 and 0xff here are
6610		 * reserved according to SPC-3.
6611		 */
6612		if ((subpage != SMS_SUBPAGE_PAGE_0)
6613		 && (subpage != SMS_SUBPAGE_ALL)) {
6614			ctl_set_invalid_field(ctsio,
6615					      /*sks_valid*/ 1,
6616					      /*command*/ 1,
6617					      /*field*/ 3,
6618					      /*bit_valid*/ 0,
6619					      /*bit*/ 0);
6620			ctl_done((union ctl_io *)ctsio);
6621			return (CTL_RETVAL_COMPLETE);
6622		}
6623
6624		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6625			if ((control_dev != 0)
6626			 && (lun->mode_pages.index[i].page_flags &
6627			     CTL_PAGE_FLAG_DISK_ONLY))
6628				continue;
6629
6630			/*
6631			 * We don't use this subpage if the user didn't
6632			 * request all subpages.
6633			 */
6634			if ((lun->mode_pages.index[i].subpage != 0)
6635			 && (subpage == SMS_SUBPAGE_PAGE_0))
6636				continue;
6637
6638#if 0
6639			printf("found page %#x len %d\n",
6640			       lun->mode_pages.index[i].page_code &
6641			       SMPH_PC_MASK,
6642			       lun->mode_pages.index[i].page_len);
6643#endif
6644			page_len += lun->mode_pages.index[i].page_len;
6645		}
6646		break;
6647	}
6648	default: {
6649		int i;
6650
6651		page_len = 0;
6652
6653		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6654			/* Look for the right page code */
6655			if ((lun->mode_pages.index[i].page_code &
6656			     SMPH_PC_MASK) != page_code)
6657				continue;
6658
6659			/* Look for the right subpage or the subpage wildcard*/
6660			if ((lun->mode_pages.index[i].subpage != subpage)
6661			 && (subpage != SMS_SUBPAGE_ALL))
6662				continue;
6663
6664			/* Make sure the page is supported for this dev type */
6665			if ((control_dev != 0)
6666			 && (lun->mode_pages.index[i].page_flags &
6667			     CTL_PAGE_FLAG_DISK_ONLY))
6668				continue;
6669
6670#if 0
6671			printf("found page %#x len %d\n",
6672			       lun->mode_pages.index[i].page_code &
6673			       SMPH_PC_MASK,
6674			       lun->mode_pages.index[i].page_len);
6675#endif
6676
6677			page_len += lun->mode_pages.index[i].page_len;
6678		}
6679
6680		if (page_len == 0) {
6681			ctl_set_invalid_field(ctsio,
6682					      /*sks_valid*/ 1,
6683					      /*command*/ 1,
6684					      /*field*/ 2,
6685					      /*bit_valid*/ 1,
6686					      /*bit*/ 5);
6687			ctl_done((union ctl_io *)ctsio);
6688			return (CTL_RETVAL_COMPLETE);
6689		}
6690		break;
6691	}
6692	}
6693
6694	total_len = header_len + page_len;
6695#if 0
6696	printf("header_len = %d, page_len = %d, total_len = %d\n",
6697	       header_len, page_len, total_len);
6698#endif
6699
6700	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6701	ctsio->kern_sg_entries = 0;
6702	ctsio->kern_data_resid = 0;
6703	ctsio->kern_rel_offset = 0;
6704	if (total_len < alloc_len) {
6705		ctsio->residual = alloc_len - total_len;
6706		ctsio->kern_data_len = total_len;
6707		ctsio->kern_total_len = total_len;
6708	} else {
6709		ctsio->residual = 0;
6710		ctsio->kern_data_len = alloc_len;
6711		ctsio->kern_total_len = alloc_len;
6712	}
6713
6714	switch (ctsio->cdb[0]) {
6715	case MODE_SENSE_6: {
6716		struct scsi_mode_hdr_6 *header;
6717
6718		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6719
6720		header->datalen = ctl_min(total_len - 1, 254);
6721		if (control_dev == 0) {
6722			header->dev_specific = 0x10; /* DPOFUA */
6723			if ((lun->flags & CTL_LUN_READONLY) ||
6724			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6725			    .eca_and_aen & SCP_SWP) != 0)
6726				    header->dev_specific |= 0x80; /* WP */
6727		}
6728		if (dbd)
6729			header->block_descr_len = 0;
6730		else
6731			header->block_descr_len =
6732				sizeof(struct scsi_mode_block_descr);
6733		block_desc = (struct scsi_mode_block_descr *)&header[1];
6734		break;
6735	}
6736	case MODE_SENSE_10: {
6737		struct scsi_mode_hdr_10 *header;
6738		int datalen;
6739
6740		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6741
6742		datalen = ctl_min(total_len - 2, 65533);
6743		scsi_ulto2b(datalen, header->datalen);
6744		if (control_dev == 0) {
6745			header->dev_specific = 0x10; /* DPOFUA */
6746			if ((lun->flags & CTL_LUN_READONLY) ||
6747			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6748			    .eca_and_aen & SCP_SWP) != 0)
6749				    header->dev_specific |= 0x80; /* WP */
6750		}
6751		if (dbd)
6752			scsi_ulto2b(0, header->block_descr_len);
6753		else
6754			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6755				    header->block_descr_len);
6756		block_desc = (struct scsi_mode_block_descr *)&header[1];
6757		break;
6758	}
6759	default:
6760		panic("invalid CDB type %#x", ctsio->cdb[0]);
6761		break; /* NOTREACHED */
6762	}
6763
6764	/*
6765	 * If we've got a disk, use its blocksize in the block
6766	 * descriptor.  Otherwise, just set it to 0.
6767	 */
6768	if (dbd == 0) {
6769		if (control_dev == 0)
6770			scsi_ulto3b(lun->be_lun->blocksize,
6771				    block_desc->block_len);
6772		else
6773			scsi_ulto3b(0, block_desc->block_len);
6774	}
6775
6776	switch (page_code) {
6777	case SMS_ALL_PAGES_PAGE: {
6778		int i, data_used;
6779
6780		data_used = header_len;
6781		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6782			struct ctl_page_index *page_index;
6783
6784			page_index = &lun->mode_pages.index[i];
6785
6786			if ((control_dev != 0)
6787			 && (page_index->page_flags &
6788			    CTL_PAGE_FLAG_DISK_ONLY))
6789				continue;
6790
6791			/*
6792			 * We don't use this subpage if the user didn't
6793			 * request all subpages.  We already checked (above)
6794			 * to make sure the user only specified a subpage
6795			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6796			 */
6797			if ((page_index->subpage != 0)
6798			 && (subpage == SMS_SUBPAGE_PAGE_0))
6799				continue;
6800
6801			/*
6802			 * Call the handler, if it exists, to update the
6803			 * page to the latest values.
6804			 */
6805			if (page_index->sense_handler != NULL)
6806				page_index->sense_handler(ctsio, page_index,pc);
6807
6808			memcpy(ctsio->kern_data_ptr + data_used,
6809			       page_index->page_data +
6810			       (page_index->page_len * pc),
6811			       page_index->page_len);
6812			data_used += page_index->page_len;
6813		}
6814		break;
6815	}
6816	default: {
6817		int i, data_used;
6818
6819		data_used = header_len;
6820
6821		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6822			struct ctl_page_index *page_index;
6823
6824			page_index = &lun->mode_pages.index[i];
6825
6826			/* Look for the right page code */
6827			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6828				continue;
6829
6830			/* Look for the right subpage or the subpage wildcard*/
6831			if ((page_index->subpage != subpage)
6832			 && (subpage != SMS_SUBPAGE_ALL))
6833				continue;
6834
6835			/* Make sure the page is supported for this dev type */
6836			if ((control_dev != 0)
6837			 && (page_index->page_flags &
6838			     CTL_PAGE_FLAG_DISK_ONLY))
6839				continue;
6840
6841			/*
6842			 * Call the handler, if it exists, to update the
6843			 * page to the latest values.
6844			 */
6845			if (page_index->sense_handler != NULL)
6846				page_index->sense_handler(ctsio, page_index,pc);
6847
6848			memcpy(ctsio->kern_data_ptr + data_used,
6849			       page_index->page_data +
6850			       (page_index->page_len * pc),
6851			       page_index->page_len);
6852			data_used += page_index->page_len;
6853		}
6854		break;
6855	}
6856	}
6857
6858	ctl_set_success(ctsio);
6859	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6860	ctsio->be_move_done = ctl_config_move_done;
6861	ctl_datamove((union ctl_io *)ctsio);
6862	return (CTL_RETVAL_COMPLETE);
6863}
6864
6865int
6866ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6867			       struct ctl_page_index *page_index,
6868			       int pc)
6869{
6870	struct ctl_lun *lun;
6871	struct scsi_log_param_header *phdr;
6872	uint8_t *data;
6873	uint64_t val;
6874
6875	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6876	data = page_index->page_data;
6877
6878	if (lun->backend->lun_attr != NULL &&
6879	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6880	     != UINT64_MAX) {
6881		phdr = (struct scsi_log_param_header *)data;
6882		scsi_ulto2b(0x0001, phdr->param_code);
6883		phdr->param_control = SLP_LBIN | SLP_LP;
6884		phdr->param_len = 8;
6885		data = (uint8_t *)(phdr + 1);
6886		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6887		data[4] = 0x01; /* per-LUN */
6888		data += phdr->param_len;
6889	}
6890
6891	if (lun->backend->lun_attr != NULL &&
6892	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6893	     != UINT64_MAX) {
6894		phdr = (struct scsi_log_param_header *)data;
6895		scsi_ulto2b(0x0002, phdr->param_code);
6896		phdr->param_control = SLP_LBIN | SLP_LP;
6897		phdr->param_len = 8;
6898		data = (uint8_t *)(phdr + 1);
6899		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6900		data[4] = 0x02; /* per-pool */
6901		data += phdr->param_len;
6902	}
6903
6904	if (lun->backend->lun_attr != NULL &&
6905	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6906	     != UINT64_MAX) {
6907		phdr = (struct scsi_log_param_header *)data;
6908		scsi_ulto2b(0x00f1, phdr->param_code);
6909		phdr->param_control = SLP_LBIN | SLP_LP;
6910		phdr->param_len = 8;
6911		data = (uint8_t *)(phdr + 1);
6912		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6913		data[4] = 0x02; /* per-pool */
6914		data += phdr->param_len;
6915	}
6916
6917	if (lun->backend->lun_attr != NULL &&
6918	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6919	     != UINT64_MAX) {
6920		phdr = (struct scsi_log_param_header *)data;
6921		scsi_ulto2b(0x00f2, phdr->param_code);
6922		phdr->param_control = SLP_LBIN | SLP_LP;
6923		phdr->param_len = 8;
6924		data = (uint8_t *)(phdr + 1);
6925		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6926		data[4] = 0x02; /* per-pool */
6927		data += phdr->param_len;
6928	}
6929
6930	page_index->page_len = data - page_index->page_data;
6931	return (0);
6932}
6933
6934int
6935ctl_log_sense(struct ctl_scsiio *ctsio)
6936{
6937	struct ctl_lun *lun;
6938	int i, pc, page_code, subpage;
6939	int alloc_len, total_len;
6940	struct ctl_page_index *page_index;
6941	struct scsi_log_sense *cdb;
6942	struct scsi_log_header *header;
6943
6944	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6945
6946	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6947	cdb = (struct scsi_log_sense *)ctsio->cdb;
6948	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6949	page_code = cdb->page & SLS_PAGE_CODE;
6950	subpage = cdb->subpage;
6951	alloc_len = scsi_2btoul(cdb->length);
6952
6953	page_index = NULL;
6954	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6955		page_index = &lun->log_pages.index[i];
6956
6957		/* Look for the right page code */
6958		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6959			continue;
6960
6961		/* Look for the right subpage or the subpage wildcard*/
6962		if (page_index->subpage != subpage)
6963			continue;
6964
6965		break;
6966	}
6967	if (i >= CTL_NUM_LOG_PAGES) {
6968		ctl_set_invalid_field(ctsio,
6969				      /*sks_valid*/ 1,
6970				      /*command*/ 1,
6971				      /*field*/ 2,
6972				      /*bit_valid*/ 0,
6973				      /*bit*/ 0);
6974		ctl_done((union ctl_io *)ctsio);
6975		return (CTL_RETVAL_COMPLETE);
6976	}
6977
6978	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6979
6980	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6981	ctsio->kern_sg_entries = 0;
6982	ctsio->kern_data_resid = 0;
6983	ctsio->kern_rel_offset = 0;
6984	if (total_len < alloc_len) {
6985		ctsio->residual = alloc_len - total_len;
6986		ctsio->kern_data_len = total_len;
6987		ctsio->kern_total_len = total_len;
6988	} else {
6989		ctsio->residual = 0;
6990		ctsio->kern_data_len = alloc_len;
6991		ctsio->kern_total_len = alloc_len;
6992	}
6993
6994	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6995	header->page = page_index->page_code;
6996	if (page_index->subpage) {
6997		header->page |= SL_SPF;
6998		header->subpage = page_index->subpage;
6999	}
7000	scsi_ulto2b(page_index->page_len, header->datalen);
7001
7002	/*
7003	 * Call the handler, if it exists, to update the
7004	 * page to the latest values.
7005	 */
7006	if (page_index->sense_handler != NULL)
7007		page_index->sense_handler(ctsio, page_index, pc);
7008
7009	memcpy(header + 1, page_index->page_data, page_index->page_len);
7010
7011	ctl_set_success(ctsio);
7012	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7013	ctsio->be_move_done = ctl_config_move_done;
7014	ctl_datamove((union ctl_io *)ctsio);
7015	return (CTL_RETVAL_COMPLETE);
7016}
7017
7018int
7019ctl_read_capacity(struct ctl_scsiio *ctsio)
7020{
7021	struct scsi_read_capacity *cdb;
7022	struct scsi_read_capacity_data *data;
7023	struct ctl_lun *lun;
7024	uint32_t lba;
7025
7026	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7027
7028	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7029
7030	lba = scsi_4btoul(cdb->addr);
7031	if (((cdb->pmi & SRC_PMI) == 0)
7032	 && (lba != 0)) {
7033		ctl_set_invalid_field(/*ctsio*/ ctsio,
7034				      /*sks_valid*/ 1,
7035				      /*command*/ 1,
7036				      /*field*/ 2,
7037				      /*bit_valid*/ 0,
7038				      /*bit*/ 0);
7039		ctl_done((union ctl_io *)ctsio);
7040		return (CTL_RETVAL_COMPLETE);
7041	}
7042
7043	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7044
7045	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7046	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7047	ctsio->residual = 0;
7048	ctsio->kern_data_len = sizeof(*data);
7049	ctsio->kern_total_len = sizeof(*data);
7050	ctsio->kern_data_resid = 0;
7051	ctsio->kern_rel_offset = 0;
7052	ctsio->kern_sg_entries = 0;
7053
7054	/*
7055	 * If the maximum LBA is greater than 0xfffffffe, the user must
7056	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7057	 * serivce action set.
7058	 */
7059	if (lun->be_lun->maxlba > 0xfffffffe)
7060		scsi_ulto4b(0xffffffff, data->addr);
7061	else
7062		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7063
7064	/*
7065	 * XXX KDM this may not be 512 bytes...
7066	 */
7067	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7068
7069	ctl_set_success(ctsio);
7070	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7071	ctsio->be_move_done = ctl_config_move_done;
7072	ctl_datamove((union ctl_io *)ctsio);
7073	return (CTL_RETVAL_COMPLETE);
7074}
7075
7076int
7077ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7078{
7079	struct scsi_read_capacity_16 *cdb;
7080	struct scsi_read_capacity_data_long *data;
7081	struct ctl_lun *lun;
7082	uint64_t lba;
7083	uint32_t alloc_len;
7084
7085	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7086
7087	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7088
7089	alloc_len = scsi_4btoul(cdb->alloc_len);
7090	lba = scsi_8btou64(cdb->addr);
7091
7092	if ((cdb->reladr & SRC16_PMI)
7093	 && (lba != 0)) {
7094		ctl_set_invalid_field(/*ctsio*/ ctsio,
7095				      /*sks_valid*/ 1,
7096				      /*command*/ 1,
7097				      /*field*/ 2,
7098				      /*bit_valid*/ 0,
7099				      /*bit*/ 0);
7100		ctl_done((union ctl_io *)ctsio);
7101		return (CTL_RETVAL_COMPLETE);
7102	}
7103
7104	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7105
7106	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7107	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7108
7109	if (sizeof(*data) < alloc_len) {
7110		ctsio->residual = alloc_len - sizeof(*data);
7111		ctsio->kern_data_len = sizeof(*data);
7112		ctsio->kern_total_len = sizeof(*data);
7113	} else {
7114		ctsio->residual = 0;
7115		ctsio->kern_data_len = alloc_len;
7116		ctsio->kern_total_len = alloc_len;
7117	}
7118	ctsio->kern_data_resid = 0;
7119	ctsio->kern_rel_offset = 0;
7120	ctsio->kern_sg_entries = 0;
7121
7122	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7123	/* XXX KDM this may not be 512 bytes... */
7124	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7125	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7126	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7127	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7128		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7129
7130	ctl_set_success(ctsio);
7131	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7132	ctsio->be_move_done = ctl_config_move_done;
7133	ctl_datamove((union ctl_io *)ctsio);
7134	return (CTL_RETVAL_COMPLETE);
7135}
7136
7137int
7138ctl_read_defect(struct ctl_scsiio *ctsio)
7139{
7140	struct scsi_read_defect_data_10 *ccb10;
7141	struct scsi_read_defect_data_12 *ccb12;
7142	struct scsi_read_defect_data_hdr_10 *data10;
7143	struct scsi_read_defect_data_hdr_12 *data12;
7144	uint32_t alloc_len, data_len;
7145	uint8_t format;
7146
7147	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7148
7149	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7150		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7151		format = ccb10->format;
7152		alloc_len = scsi_2btoul(ccb10->alloc_length);
7153		data_len = sizeof(*data10);
7154	} else {
7155		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7156		format = ccb12->format;
7157		alloc_len = scsi_4btoul(ccb12->alloc_length);
7158		data_len = sizeof(*data12);
7159	}
7160	if (alloc_len == 0) {
7161		ctl_set_success(ctsio);
7162		ctl_done((union ctl_io *)ctsio);
7163		return (CTL_RETVAL_COMPLETE);
7164	}
7165
7166	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7167	if (data_len < alloc_len) {
7168		ctsio->residual = alloc_len - data_len;
7169		ctsio->kern_data_len = data_len;
7170		ctsio->kern_total_len = data_len;
7171	} else {
7172		ctsio->residual = 0;
7173		ctsio->kern_data_len = alloc_len;
7174		ctsio->kern_total_len = alloc_len;
7175	}
7176	ctsio->kern_data_resid = 0;
7177	ctsio->kern_rel_offset = 0;
7178	ctsio->kern_sg_entries = 0;
7179
7180	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7181		data10 = (struct scsi_read_defect_data_hdr_10 *)
7182		    ctsio->kern_data_ptr;
7183		data10->format = format;
7184		scsi_ulto2b(0, data10->length);
7185	} else {
7186		data12 = (struct scsi_read_defect_data_hdr_12 *)
7187		    ctsio->kern_data_ptr;
7188		data12->format = format;
7189		scsi_ulto2b(0, data12->generation);
7190		scsi_ulto4b(0, data12->length);
7191	}
7192
7193	ctl_set_success(ctsio);
7194	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7195	ctsio->be_move_done = ctl_config_move_done;
7196	ctl_datamove((union ctl_io *)ctsio);
7197	return (CTL_RETVAL_COMPLETE);
7198}
7199
7200int
7201ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7202{
7203	struct scsi_maintenance_in *cdb;
7204	int retval;
7205	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
7206	int num_target_port_groups, num_target_ports;
7207	struct ctl_lun *lun;
7208	struct ctl_softc *softc;
7209	struct ctl_port *port;
7210	struct scsi_target_group_data *rtg_ptr;
7211	struct scsi_target_group_data_extended *rtg_ext_ptr;
7212	struct scsi_target_port_group_descriptor *tpg_desc;
7213
7214	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7215
7216	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7217	softc = control_softc;
7218	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7219
7220	retval = CTL_RETVAL_COMPLETE;
7221
7222	switch (cdb->byte2 & STG_PDF_MASK) {
7223	case STG_PDF_LENGTH:
7224		ext = 0;
7225		break;
7226	case STG_PDF_EXTENDED:
7227		ext = 1;
7228		break;
7229	default:
7230		ctl_set_invalid_field(/*ctsio*/ ctsio,
7231				      /*sks_valid*/ 1,
7232				      /*command*/ 1,
7233				      /*field*/ 2,
7234				      /*bit_valid*/ 1,
7235				      /*bit*/ 5);
7236		ctl_done((union ctl_io *)ctsio);
7237		return(retval);
7238	}
7239
7240	if (softc->is_single)
7241		num_target_port_groups = 1;
7242	else
7243		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7244	num_target_ports = 0;
7245	mtx_lock(&softc->ctl_lock);
7246	STAILQ_FOREACH(port, &softc->port_list, links) {
7247		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7248			continue;
7249		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7250			continue;
7251		num_target_ports++;
7252	}
7253	mtx_unlock(&softc->ctl_lock);
7254
7255	if (ext)
7256		total_len = sizeof(struct scsi_target_group_data_extended);
7257	else
7258		total_len = sizeof(struct scsi_target_group_data);
7259	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7260		num_target_port_groups +
7261	    sizeof(struct scsi_target_port_descriptor) *
7262		num_target_ports * num_target_port_groups;
7263
7264	alloc_len = scsi_4btoul(cdb->length);
7265
7266	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7267
7268	ctsio->kern_sg_entries = 0;
7269
7270	if (total_len < alloc_len) {
7271		ctsio->residual = alloc_len - total_len;
7272		ctsio->kern_data_len = total_len;
7273		ctsio->kern_total_len = total_len;
7274	} else {
7275		ctsio->residual = 0;
7276		ctsio->kern_data_len = alloc_len;
7277		ctsio->kern_total_len = alloc_len;
7278	}
7279	ctsio->kern_data_resid = 0;
7280	ctsio->kern_rel_offset = 0;
7281
7282	if (ext) {
7283		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7284		    ctsio->kern_data_ptr;
7285		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7286		rtg_ext_ptr->format_type = 0x10;
7287		rtg_ext_ptr->implicit_transition_time = 0;
7288		tpg_desc = &rtg_ext_ptr->groups[0];
7289	} else {
7290		rtg_ptr = (struct scsi_target_group_data *)
7291		    ctsio->kern_data_ptr;
7292		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7293		tpg_desc = &rtg_ptr->groups[0];
7294	}
7295
7296	mtx_lock(&softc->ctl_lock);
7297	pg = softc->port_offset / CTL_MAX_PORTS;
7298	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
7299		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
7300			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7301			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7302		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
7303			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7304			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7305		} else {
7306			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7307			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7308		}
7309	} else {
7310		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7311		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7312	}
7313	for (g = 0; g < num_target_port_groups; g++) {
7314		tpg_desc->pref_state = (g == pg) ? gs : os;
7315		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
7316		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7317		tpg_desc->status = TPG_IMPLICIT;
7318		pc = 0;
7319		STAILQ_FOREACH(port, &softc->port_list, links) {
7320			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7321				continue;
7322			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7323			    CTL_MAX_LUNS)
7324				continue;
7325			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7326			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7327			    relative_target_port_identifier);
7328			pc++;
7329		}
7330		tpg_desc->target_port_count = pc;
7331		tpg_desc = (struct scsi_target_port_group_descriptor *)
7332		    &tpg_desc->descriptors[pc];
7333	}
7334	mtx_unlock(&softc->ctl_lock);
7335
7336	ctl_set_success(ctsio);
7337	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7338	ctsio->be_move_done = ctl_config_move_done;
7339	ctl_datamove((union ctl_io *)ctsio);
7340	return(retval);
7341}
7342
7343int
7344ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7345{
7346	struct ctl_lun *lun;
7347	struct scsi_report_supported_opcodes *cdb;
7348	const struct ctl_cmd_entry *entry, *sentry;
7349	struct scsi_report_supported_opcodes_all *all;
7350	struct scsi_report_supported_opcodes_descr *descr;
7351	struct scsi_report_supported_opcodes_one *one;
7352	int retval;
7353	int alloc_len, total_len;
7354	int opcode, service_action, i, j, num;
7355
7356	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7357
7358	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7359	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7360
7361	retval = CTL_RETVAL_COMPLETE;
7362
7363	opcode = cdb->requested_opcode;
7364	service_action = scsi_2btoul(cdb->requested_service_action);
7365	switch (cdb->options & RSO_OPTIONS_MASK) {
7366	case RSO_OPTIONS_ALL:
7367		num = 0;
7368		for (i = 0; i < 256; i++) {
7369			entry = &ctl_cmd_table[i];
7370			if (entry->flags & CTL_CMD_FLAG_SA5) {
7371				for (j = 0; j < 32; j++) {
7372					sentry = &((const struct ctl_cmd_entry *)
7373					    entry->execute)[j];
7374					if (ctl_cmd_applicable(
7375					    lun->be_lun->lun_type, sentry))
7376						num++;
7377				}
7378			} else {
7379				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7380				    entry))
7381					num++;
7382			}
7383		}
7384		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7385		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7386		break;
7387	case RSO_OPTIONS_OC:
7388		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7389			ctl_set_invalid_field(/*ctsio*/ ctsio,
7390					      /*sks_valid*/ 1,
7391					      /*command*/ 1,
7392					      /*field*/ 2,
7393					      /*bit_valid*/ 1,
7394					      /*bit*/ 2);
7395			ctl_done((union ctl_io *)ctsio);
7396			return (CTL_RETVAL_COMPLETE);
7397		}
7398		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7399		break;
7400	case RSO_OPTIONS_OC_SA:
7401		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7402		    service_action >= 32) {
7403			ctl_set_invalid_field(/*ctsio*/ ctsio,
7404					      /*sks_valid*/ 1,
7405					      /*command*/ 1,
7406					      /*field*/ 2,
7407					      /*bit_valid*/ 1,
7408					      /*bit*/ 2);
7409			ctl_done((union ctl_io *)ctsio);
7410			return (CTL_RETVAL_COMPLETE);
7411		}
7412		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7413		break;
7414	default:
7415		ctl_set_invalid_field(/*ctsio*/ ctsio,
7416				      /*sks_valid*/ 1,
7417				      /*command*/ 1,
7418				      /*field*/ 2,
7419				      /*bit_valid*/ 1,
7420				      /*bit*/ 2);
7421		ctl_done((union ctl_io *)ctsio);
7422		return (CTL_RETVAL_COMPLETE);
7423	}
7424
7425	alloc_len = scsi_4btoul(cdb->length);
7426
7427	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7428
7429	ctsio->kern_sg_entries = 0;
7430
7431	if (total_len < alloc_len) {
7432		ctsio->residual = alloc_len - total_len;
7433		ctsio->kern_data_len = total_len;
7434		ctsio->kern_total_len = total_len;
7435	} else {
7436		ctsio->residual = 0;
7437		ctsio->kern_data_len = alloc_len;
7438		ctsio->kern_total_len = alloc_len;
7439	}
7440	ctsio->kern_data_resid = 0;
7441	ctsio->kern_rel_offset = 0;
7442
7443	switch (cdb->options & RSO_OPTIONS_MASK) {
7444	case RSO_OPTIONS_ALL:
7445		all = (struct scsi_report_supported_opcodes_all *)
7446		    ctsio->kern_data_ptr;
7447		num = 0;
7448		for (i = 0; i < 256; i++) {
7449			entry = &ctl_cmd_table[i];
7450			if (entry->flags & CTL_CMD_FLAG_SA5) {
7451				for (j = 0; j < 32; j++) {
7452					sentry = &((const struct ctl_cmd_entry *)
7453					    entry->execute)[j];
7454					if (!ctl_cmd_applicable(
7455					    lun->be_lun->lun_type, sentry))
7456						continue;
7457					descr = &all->descr[num++];
7458					descr->opcode = i;
7459					scsi_ulto2b(j, descr->service_action);
7460					descr->flags = RSO_SERVACTV;
7461					scsi_ulto2b(sentry->length,
7462					    descr->cdb_length);
7463				}
7464			} else {
7465				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7466				    entry))
7467					continue;
7468				descr = &all->descr[num++];
7469				descr->opcode = i;
7470				scsi_ulto2b(0, descr->service_action);
7471				descr->flags = 0;
7472				scsi_ulto2b(entry->length, descr->cdb_length);
7473			}
7474		}
7475		scsi_ulto4b(
7476		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7477		    all->length);
7478		break;
7479	case RSO_OPTIONS_OC:
7480		one = (struct scsi_report_supported_opcodes_one *)
7481		    ctsio->kern_data_ptr;
7482		entry = &ctl_cmd_table[opcode];
7483		goto fill_one;
7484	case RSO_OPTIONS_OC_SA:
7485		one = (struct scsi_report_supported_opcodes_one *)
7486		    ctsio->kern_data_ptr;
7487		entry = &ctl_cmd_table[opcode];
7488		entry = &((const struct ctl_cmd_entry *)
7489		    entry->execute)[service_action];
7490fill_one:
7491		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7492			one->support = 3;
7493			scsi_ulto2b(entry->length, one->cdb_length);
7494			one->cdb_usage[0] = opcode;
7495			memcpy(&one->cdb_usage[1], entry->usage,
7496			    entry->length - 1);
7497		} else
7498			one->support = 1;
7499		break;
7500	}
7501
7502	ctl_set_success(ctsio);
7503	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7504	ctsio->be_move_done = ctl_config_move_done;
7505	ctl_datamove((union ctl_io *)ctsio);
7506	return(retval);
7507}
7508
7509int
7510ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7511{
7512	struct scsi_report_supported_tmf *cdb;
7513	struct scsi_report_supported_tmf_data *data;
7514	int retval;
7515	int alloc_len, total_len;
7516
7517	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7518
7519	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7520
7521	retval = CTL_RETVAL_COMPLETE;
7522
7523	total_len = sizeof(struct scsi_report_supported_tmf_data);
7524	alloc_len = scsi_4btoul(cdb->length);
7525
7526	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7527
7528	ctsio->kern_sg_entries = 0;
7529
7530	if (total_len < alloc_len) {
7531		ctsio->residual = alloc_len - total_len;
7532		ctsio->kern_data_len = total_len;
7533		ctsio->kern_total_len = total_len;
7534	} else {
7535		ctsio->residual = 0;
7536		ctsio->kern_data_len = alloc_len;
7537		ctsio->kern_total_len = alloc_len;
7538	}
7539	ctsio->kern_data_resid = 0;
7540	ctsio->kern_rel_offset = 0;
7541
7542	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7543	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7544	data->byte2 |= RST_ITNRS;
7545
7546	ctl_set_success(ctsio);
7547	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7548	ctsio->be_move_done = ctl_config_move_done;
7549	ctl_datamove((union ctl_io *)ctsio);
7550	return (retval);
7551}
7552
7553int
7554ctl_report_timestamp(struct ctl_scsiio *ctsio)
7555{
7556	struct scsi_report_timestamp *cdb;
7557	struct scsi_report_timestamp_data *data;
7558	struct timeval tv;
7559	int64_t timestamp;
7560	int retval;
7561	int alloc_len, total_len;
7562
7563	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7564
7565	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7566
7567	retval = CTL_RETVAL_COMPLETE;
7568
7569	total_len = sizeof(struct scsi_report_timestamp_data);
7570	alloc_len = scsi_4btoul(cdb->length);
7571
7572	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7573
7574	ctsio->kern_sg_entries = 0;
7575
7576	if (total_len < alloc_len) {
7577		ctsio->residual = alloc_len - total_len;
7578		ctsio->kern_data_len = total_len;
7579		ctsio->kern_total_len = total_len;
7580	} else {
7581		ctsio->residual = 0;
7582		ctsio->kern_data_len = alloc_len;
7583		ctsio->kern_total_len = alloc_len;
7584	}
7585	ctsio->kern_data_resid = 0;
7586	ctsio->kern_rel_offset = 0;
7587
7588	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7589	scsi_ulto2b(sizeof(*data) - 2, data->length);
7590	data->origin = RTS_ORIG_OUTSIDE;
7591	getmicrotime(&tv);
7592	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7593	scsi_ulto4b(timestamp >> 16, data->timestamp);
7594	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7595
7596	ctl_set_success(ctsio);
7597	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7598	ctsio->be_move_done = ctl_config_move_done;
7599	ctl_datamove((union ctl_io *)ctsio);
7600	return (retval);
7601}
7602
7603int
7604ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7605{
7606	struct scsi_per_res_in *cdb;
7607	int alloc_len, total_len = 0;
7608	/* struct scsi_per_res_in_rsrv in_data; */
7609	struct ctl_lun *lun;
7610	struct ctl_softc *softc;
7611
7612	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7613
7614	softc = control_softc;
7615
7616	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7617
7618	alloc_len = scsi_2btoul(cdb->length);
7619
7620	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7621
7622retry:
7623	mtx_lock(&lun->lun_lock);
7624	switch (cdb->action) {
7625	case SPRI_RK: /* read keys */
7626		total_len = sizeof(struct scsi_per_res_in_keys) +
7627			lun->pr_key_count *
7628			sizeof(struct scsi_per_res_key);
7629		break;
7630	case SPRI_RR: /* read reservation */
7631		if (lun->flags & CTL_LUN_PR_RESERVED)
7632			total_len = sizeof(struct scsi_per_res_in_rsrv);
7633		else
7634			total_len = sizeof(struct scsi_per_res_in_header);
7635		break;
7636	case SPRI_RC: /* report capabilities */
7637		total_len = sizeof(struct scsi_per_res_cap);
7638		break;
7639	case SPRI_RS: /* read full status */
7640		total_len = sizeof(struct scsi_per_res_in_header) +
7641		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7642		    lun->pr_key_count;
7643		break;
7644	default:
7645		panic("Invalid PR type %x", cdb->action);
7646	}
7647	mtx_unlock(&lun->lun_lock);
7648
7649	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7650
7651	if (total_len < alloc_len) {
7652		ctsio->residual = alloc_len - total_len;
7653		ctsio->kern_data_len = total_len;
7654		ctsio->kern_total_len = total_len;
7655	} else {
7656		ctsio->residual = 0;
7657		ctsio->kern_data_len = alloc_len;
7658		ctsio->kern_total_len = alloc_len;
7659	}
7660
7661	ctsio->kern_data_resid = 0;
7662	ctsio->kern_rel_offset = 0;
7663	ctsio->kern_sg_entries = 0;
7664
7665	mtx_lock(&lun->lun_lock);
7666	switch (cdb->action) {
7667	case SPRI_RK: { // read keys
7668        struct scsi_per_res_in_keys *res_keys;
7669		int i, key_count;
7670
7671		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7672
7673		/*
7674		 * We had to drop the lock to allocate our buffer, which
7675		 * leaves time for someone to come in with another
7676		 * persistent reservation.  (That is unlikely, though,
7677		 * since this should be the only persistent reservation
7678		 * command active right now.)
7679		 */
7680		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7681		    (lun->pr_key_count *
7682		     sizeof(struct scsi_per_res_key)))){
7683			mtx_unlock(&lun->lun_lock);
7684			free(ctsio->kern_data_ptr, M_CTL);
7685			printf("%s: reservation length changed, retrying\n",
7686			       __func__);
7687			goto retry;
7688		}
7689
7690		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7691
7692		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7693			     lun->pr_key_count, res_keys->header.length);
7694
7695		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7696			if (lun->pr_keys[i] == 0)
7697				continue;
7698
7699			/*
7700			 * We used lun->pr_key_count to calculate the
7701			 * size to allocate.  If it turns out the number of
7702			 * initiators with the registered flag set is
7703			 * larger than that (i.e. they haven't been kept in
7704			 * sync), we've got a problem.
7705			 */
7706			if (key_count >= lun->pr_key_count) {
7707#ifdef NEEDTOPORT
7708				csevent_log(CSC_CTL | CSC_SHELF_SW |
7709					    CTL_PR_ERROR,
7710					    csevent_LogType_Fault,
7711					    csevent_AlertLevel_Yellow,
7712					    csevent_FRU_ShelfController,
7713					    csevent_FRU_Firmware,
7714				        csevent_FRU_Unknown,
7715					    "registered keys %d >= key "
7716					    "count %d", key_count,
7717					    lun->pr_key_count);
7718#endif
7719				key_count++;
7720				continue;
7721			}
7722			scsi_u64to8b(lun->pr_keys[i],
7723			    res_keys->keys[key_count].key);
7724			key_count++;
7725		}
7726		break;
7727	}
7728	case SPRI_RR: { // read reservation
7729		struct scsi_per_res_in_rsrv *res;
7730		int tmp_len, header_only;
7731
7732		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7733
7734		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7735
7736		if (lun->flags & CTL_LUN_PR_RESERVED)
7737		{
7738			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7739			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7740				    res->header.length);
7741			header_only = 0;
7742		} else {
7743			tmp_len = sizeof(struct scsi_per_res_in_header);
7744			scsi_ulto4b(0, res->header.length);
7745			header_only = 1;
7746		}
7747
7748		/*
7749		 * We had to drop the lock to allocate our buffer, which
7750		 * leaves time for someone to come in with another
7751		 * persistent reservation.  (That is unlikely, though,
7752		 * since this should be the only persistent reservation
7753		 * command active right now.)
7754		 */
7755		if (tmp_len != total_len) {
7756			mtx_unlock(&lun->lun_lock);
7757			free(ctsio->kern_data_ptr, M_CTL);
7758			printf("%s: reservation status changed, retrying\n",
7759			       __func__);
7760			goto retry;
7761		}
7762
7763		/*
7764		 * No reservation held, so we're done.
7765		 */
7766		if (header_only != 0)
7767			break;
7768
7769		/*
7770		 * If the registration is an All Registrants type, the key
7771		 * is 0, since it doesn't really matter.
7772		 */
7773		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7774			scsi_u64to8b(lun->pr_keys[lun->pr_res_idx],
7775			    res->data.reservation);
7776		}
7777		res->data.scopetype = lun->res_type;
7778		break;
7779	}
7780	case SPRI_RC:     //report capabilities
7781	{
7782		struct scsi_per_res_cap *res_cap;
7783		uint16_t type_mask;
7784
7785		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7786		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7787		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7788		type_mask = SPRI_TM_WR_EX_AR |
7789			    SPRI_TM_EX_AC_RO |
7790			    SPRI_TM_WR_EX_RO |
7791			    SPRI_TM_EX_AC |
7792			    SPRI_TM_WR_EX |
7793			    SPRI_TM_EX_AC_AR;
7794		scsi_ulto2b(type_mask, res_cap->type_mask);
7795		break;
7796	}
7797	case SPRI_RS: { // read full status
7798		struct scsi_per_res_in_full *res_status;
7799		struct scsi_per_res_in_full_desc *res_desc;
7800		struct ctl_port *port;
7801		int i, len;
7802
7803		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7804
7805		/*
7806		 * We had to drop the lock to allocate our buffer, which
7807		 * leaves time for someone to come in with another
7808		 * persistent reservation.  (That is unlikely, though,
7809		 * since this should be the only persistent reservation
7810		 * command active right now.)
7811		 */
7812		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7813		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7814		     lun->pr_key_count)){
7815			mtx_unlock(&lun->lun_lock);
7816			free(ctsio->kern_data_ptr, M_CTL);
7817			printf("%s: reservation length changed, retrying\n",
7818			       __func__);
7819			goto retry;
7820		}
7821
7822		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7823
7824		res_desc = &res_status->desc[0];
7825		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7826			if (lun->pr_keys[i] == 0)
7827				continue;
7828
7829			scsi_u64to8b(lun->pr_keys[i], res_desc->res_key.key);
7830			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7831			    (lun->pr_res_idx == i ||
7832			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7833				res_desc->flags = SPRI_FULL_R_HOLDER;
7834				res_desc->scopetype = lun->res_type;
7835			}
7836			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7837			    res_desc->rel_trgt_port_id);
7838			len = 0;
7839			port = softc->ctl_ports[
7840			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7841			if (port != NULL)
7842				len = ctl_create_iid(port,
7843				    i % CTL_MAX_INIT_PER_PORT,
7844				    res_desc->transport_id);
7845			scsi_ulto4b(len, res_desc->additional_length);
7846			res_desc = (struct scsi_per_res_in_full_desc *)
7847			    &res_desc->transport_id[len];
7848		}
7849		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7850		    res_status->header.length);
7851		break;
7852	}
7853	default:
7854		/*
7855		 * This is a bug, because we just checked for this above,
7856		 * and should have returned an error.
7857		 */
7858		panic("Invalid PR type %x", cdb->action);
7859		break; /* NOTREACHED */
7860	}
7861	mtx_unlock(&lun->lun_lock);
7862
7863	ctl_set_success(ctsio);
7864	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7865	ctsio->be_move_done = ctl_config_move_done;
7866	ctl_datamove((union ctl_io *)ctsio);
7867	return (CTL_RETVAL_COMPLETE);
7868}
7869
7870/*
7871 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7872 * it should return.
7873 */
7874static int
7875ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7876		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7877		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7878		struct scsi_per_res_out_parms* param)
7879{
7880	union ctl_ha_msg persis_io;
7881	int retval, i;
7882	int isc_retval;
7883
7884	retval = 0;
7885
7886	mtx_lock(&lun->lun_lock);
7887	if (sa_res_key == 0) {
7888		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7889			/* validate scope and type */
7890			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7891			     SPR_LU_SCOPE) {
7892				mtx_unlock(&lun->lun_lock);
7893				ctl_set_invalid_field(/*ctsio*/ ctsio,
7894						      /*sks_valid*/ 1,
7895						      /*command*/ 1,
7896						      /*field*/ 2,
7897						      /*bit_valid*/ 1,
7898						      /*bit*/ 4);
7899				ctl_done((union ctl_io *)ctsio);
7900				return (1);
7901			}
7902
7903		        if (type>8 || type==2 || type==4 || type==0) {
7904				mtx_unlock(&lun->lun_lock);
7905				ctl_set_invalid_field(/*ctsio*/ ctsio,
7906       	           				      /*sks_valid*/ 1,
7907						      /*command*/ 1,
7908						      /*field*/ 2,
7909						      /*bit_valid*/ 1,
7910						      /*bit*/ 0);
7911				ctl_done((union ctl_io *)ctsio);
7912				return (1);
7913		        }
7914
7915			/*
7916			 * Unregister everybody else and build UA for
7917			 * them
7918			 */
7919			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7920				if (i == residx || lun->pr_keys[i] == 0)
7921					continue;
7922
7923				if (!persis_offset
7924				 && i <CTL_MAX_INITIATORS)
7925					lun->pending_ua[i] |=
7926						CTL_UA_REG_PREEMPT;
7927				else if (persis_offset
7928				      && i >= persis_offset)
7929					lun->pending_ua[i-persis_offset] |=
7930						CTL_UA_REG_PREEMPT;
7931				lun->pr_keys[i] = 0;
7932			}
7933			lun->pr_key_count = 1;
7934			lun->res_type = type;
7935			if (lun->res_type != SPR_TYPE_WR_EX_AR
7936			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7937				lun->pr_res_idx = residx;
7938
7939			/* send msg to other side */
7940			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7941			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7942			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7943			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7944			persis_io.pr.pr_info.res_type = type;
7945			memcpy(persis_io.pr.pr_info.sa_res_key,
7946			       param->serv_act_res_key,
7947			       sizeof(param->serv_act_res_key));
7948			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7949			     &persis_io, sizeof(persis_io), 0)) >
7950			     CTL_HA_STATUS_SUCCESS) {
7951				printf("CTL:Persis Out error returned "
7952				       "from ctl_ha_msg_send %d\n",
7953				       isc_retval);
7954			}
7955		} else {
7956			/* not all registrants */
7957			mtx_unlock(&lun->lun_lock);
7958			free(ctsio->kern_data_ptr, M_CTL);
7959			ctl_set_invalid_field(ctsio,
7960					      /*sks_valid*/ 1,
7961					      /*command*/ 0,
7962					      /*field*/ 8,
7963					      /*bit_valid*/ 0,
7964					      /*bit*/ 0);
7965			ctl_done((union ctl_io *)ctsio);
7966			return (1);
7967		}
7968	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7969		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7970		int found = 0;
7971
7972		if (res_key == sa_res_key) {
7973			/* special case */
7974			/*
7975			 * The spec implies this is not good but doesn't
7976			 * say what to do. There are two choices either
7977			 * generate a res conflict or check condition
7978			 * with illegal field in parameter data. Since
7979			 * that is what is done when the sa_res_key is
7980			 * zero I'll take that approach since this has
7981			 * to do with the sa_res_key.
7982			 */
7983			mtx_unlock(&lun->lun_lock);
7984			free(ctsio->kern_data_ptr, M_CTL);
7985			ctl_set_invalid_field(ctsio,
7986					      /*sks_valid*/ 1,
7987					      /*command*/ 0,
7988					      /*field*/ 8,
7989					      /*bit_valid*/ 0,
7990					      /*bit*/ 0);
7991			ctl_done((union ctl_io *)ctsio);
7992			return (1);
7993		}
7994
7995		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7996			if (lun->pr_keys[i] != sa_res_key)
7997				continue;
7998
7999			found = 1;
8000			lun->pr_keys[i] = 0;
8001			lun->pr_key_count--;
8002
8003			if (!persis_offset && i < CTL_MAX_INITIATORS)
8004				lun->pending_ua[i] |= CTL_UA_REG_PREEMPT;
8005			else if (persis_offset && i >= persis_offset)
8006				lun->pending_ua[i-persis_offset] |=
8007					CTL_UA_REG_PREEMPT;
8008		}
8009		if (!found) {
8010			mtx_unlock(&lun->lun_lock);
8011			free(ctsio->kern_data_ptr, M_CTL);
8012			ctl_set_reservation_conflict(ctsio);
8013			ctl_done((union ctl_io *)ctsio);
8014			return (CTL_RETVAL_COMPLETE);
8015		}
8016		/* send msg to other side */
8017		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8018		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8019		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8020		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8021		persis_io.pr.pr_info.res_type = type;
8022		memcpy(persis_io.pr.pr_info.sa_res_key,
8023		       param->serv_act_res_key,
8024		       sizeof(param->serv_act_res_key));
8025		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8026		     &persis_io, sizeof(persis_io), 0)) >
8027		     CTL_HA_STATUS_SUCCESS) {
8028			printf("CTL:Persis Out error returned from "
8029			       "ctl_ha_msg_send %d\n", isc_retval);
8030		}
8031	} else {
8032		/* Reserved but not all registrants */
8033		/* sa_res_key is res holder */
8034		if (sa_res_key == lun->pr_keys[lun->pr_res_idx]) {
8035			/* validate scope and type */
8036			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8037			     SPR_LU_SCOPE) {
8038				mtx_unlock(&lun->lun_lock);
8039				ctl_set_invalid_field(/*ctsio*/ ctsio,
8040						      /*sks_valid*/ 1,
8041						      /*command*/ 1,
8042						      /*field*/ 2,
8043						      /*bit_valid*/ 1,
8044						      /*bit*/ 4);
8045				ctl_done((union ctl_io *)ctsio);
8046				return (1);
8047			}
8048
8049			if (type>8 || type==2 || type==4 || type==0) {
8050				mtx_unlock(&lun->lun_lock);
8051				ctl_set_invalid_field(/*ctsio*/ ctsio,
8052						      /*sks_valid*/ 1,
8053						      /*command*/ 1,
8054						      /*field*/ 2,
8055						      /*bit_valid*/ 1,
8056						      /*bit*/ 0);
8057				ctl_done((union ctl_io *)ctsio);
8058				return (1);
8059			}
8060
8061			/*
8062			 * Do the following:
8063			 * if sa_res_key != res_key remove all
8064			 * registrants w/sa_res_key and generate UA
8065			 * for these registrants(Registrations
8066			 * Preempted) if it wasn't an exclusive
8067			 * reservation generate UA(Reservations
8068			 * Preempted) for all other registered nexuses
8069			 * if the type has changed. Establish the new
8070			 * reservation and holder. If res_key and
8071			 * sa_res_key are the same do the above
8072			 * except don't unregister the res holder.
8073			 */
8074
8075			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8076				if (i == residx || lun->pr_keys[i] == 0)
8077					continue;
8078
8079				if (sa_res_key == lun->pr_keys[i]) {
8080					lun->pr_keys[i] = 0;
8081					lun->pr_key_count--;
8082
8083					if (!persis_offset
8084					 && i < CTL_MAX_INITIATORS)
8085						lun->pending_ua[i] |=
8086							CTL_UA_REG_PREEMPT;
8087					else if (persis_offset
8088					      && i >= persis_offset)
8089						lun->pending_ua[i-persis_offset] |=
8090						  CTL_UA_REG_PREEMPT;
8091				} else if (type != lun->res_type
8092					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8093					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8094						if (!persis_offset
8095						 && i < CTL_MAX_INITIATORS)
8096							lun->pending_ua[i] |=
8097							CTL_UA_RES_RELEASE;
8098						else if (persis_offset
8099						      && i >= persis_offset)
8100							lun->pending_ua[
8101							i-persis_offset] |=
8102							CTL_UA_RES_RELEASE;
8103				}
8104			}
8105			lun->res_type = type;
8106			if (lun->res_type != SPR_TYPE_WR_EX_AR
8107			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8108				lun->pr_res_idx = residx;
8109			else
8110				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8111
8112			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8113			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8114			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8115			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8116			persis_io.pr.pr_info.res_type = type;
8117			memcpy(persis_io.pr.pr_info.sa_res_key,
8118			       param->serv_act_res_key,
8119			       sizeof(param->serv_act_res_key));
8120			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8121			     &persis_io, sizeof(persis_io), 0)) >
8122			     CTL_HA_STATUS_SUCCESS) {
8123				printf("CTL:Persis Out error returned "
8124				       "from ctl_ha_msg_send %d\n",
8125				       isc_retval);
8126			}
8127		} else {
8128			/*
8129			 * sa_res_key is not the res holder just
8130			 * remove registrants
8131			 */
8132			int found=0;
8133
8134			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8135				if (sa_res_key != lun->pr_keys[i])
8136					continue;
8137
8138				found = 1;
8139				lun->pr_keys[i] = 0;
8140				lun->pr_key_count--;
8141
8142				if (!persis_offset
8143				 && i < CTL_MAX_INITIATORS)
8144					lun->pending_ua[i] |=
8145						CTL_UA_REG_PREEMPT;
8146				else if (persis_offset
8147				      && i >= persis_offset)
8148					lun->pending_ua[i-persis_offset] |=
8149						CTL_UA_REG_PREEMPT;
8150			}
8151
8152			if (!found) {
8153				mtx_unlock(&lun->lun_lock);
8154				free(ctsio->kern_data_ptr, M_CTL);
8155				ctl_set_reservation_conflict(ctsio);
8156				ctl_done((union ctl_io *)ctsio);
8157		        	return (1);
8158			}
8159			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8160			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8161			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8162			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8163			persis_io.pr.pr_info.res_type = type;
8164			memcpy(persis_io.pr.pr_info.sa_res_key,
8165			       param->serv_act_res_key,
8166			       sizeof(param->serv_act_res_key));
8167			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8168			     &persis_io, sizeof(persis_io), 0)) >
8169			     CTL_HA_STATUS_SUCCESS) {
8170				printf("CTL:Persis Out error returned "
8171				       "from ctl_ha_msg_send %d\n",
8172				isc_retval);
8173			}
8174		}
8175	}
8176
8177	lun->PRGeneration++;
8178	mtx_unlock(&lun->lun_lock);
8179
8180	return (retval);
8181}
8182
8183static void
8184ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8185{
8186	uint64_t sa_res_key;
8187	int i;
8188
8189	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8190
8191	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8192	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8193	 || sa_res_key != lun->pr_keys[lun->pr_res_idx]) {
8194		if (sa_res_key == 0) {
8195			/*
8196			 * Unregister everybody else and build UA for
8197			 * them
8198			 */
8199			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8200				if (i == msg->pr.pr_info.residx ||
8201				    lun->pr_keys[i] == 0)
8202					continue;
8203
8204				if (!persis_offset
8205				 && i < CTL_MAX_INITIATORS)
8206					lun->pending_ua[i] |=
8207						CTL_UA_REG_PREEMPT;
8208				else if (persis_offset && i >= persis_offset)
8209					lun->pending_ua[i - persis_offset] |=
8210						CTL_UA_REG_PREEMPT;
8211				lun->pr_keys[i] = 0;
8212			}
8213
8214			lun->pr_key_count = 1;
8215			lun->res_type = msg->pr.pr_info.res_type;
8216			if (lun->res_type != SPR_TYPE_WR_EX_AR
8217			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8218				lun->pr_res_idx = msg->pr.pr_info.residx;
8219		} else {
8220		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8221				if (sa_res_key == lun->pr_keys[i])
8222					continue;
8223
8224				lun->pr_keys[i] = 0;
8225				lun->pr_key_count--;
8226
8227				if (!persis_offset
8228				 && i < persis_offset)
8229					lun->pending_ua[i] |=
8230						CTL_UA_REG_PREEMPT;
8231				else if (persis_offset
8232				      && i >= persis_offset)
8233					lun->pending_ua[i - persis_offset] |=
8234						CTL_UA_REG_PREEMPT;
8235			}
8236		}
8237	} else {
8238		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8239			if (i == msg->pr.pr_info.residx ||
8240			    lun->pr_keys[i] == 0)
8241				continue;
8242
8243			if (sa_res_key == lun->pr_keys[i]) {
8244				lun->pr_keys[i] = 0;
8245				lun->pr_key_count--;
8246				if (!persis_offset
8247				 && i < CTL_MAX_INITIATORS)
8248					lun->pending_ua[i] |=
8249						CTL_UA_REG_PREEMPT;
8250				else if (persis_offset
8251				      && i >= persis_offset)
8252					lun->pending_ua[i - persis_offset] |=
8253						CTL_UA_REG_PREEMPT;
8254			} else if (msg->pr.pr_info.res_type != lun->res_type
8255				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8256				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8257					if (!persis_offset
8258					 && i < persis_offset)
8259						lun->pending_ua[i] |=
8260							CTL_UA_RES_RELEASE;
8261					else if (persis_offset
8262					      && i >= persis_offset)
8263					lun->pending_ua[i - persis_offset] |=
8264						CTL_UA_RES_RELEASE;
8265			}
8266		}
8267		lun->res_type = msg->pr.pr_info.res_type;
8268		if (lun->res_type != SPR_TYPE_WR_EX_AR
8269		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8270			lun->pr_res_idx = msg->pr.pr_info.residx;
8271		else
8272			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8273	}
8274	lun->PRGeneration++;
8275
8276}
8277
8278
8279int
8280ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8281{
8282	int retval;
8283	int isc_retval;
8284	u_int32_t param_len;
8285	struct scsi_per_res_out *cdb;
8286	struct ctl_lun *lun;
8287	struct scsi_per_res_out_parms* param;
8288	struct ctl_softc *softc;
8289	uint32_t residx;
8290	uint64_t res_key, sa_res_key;
8291	uint8_t type;
8292	union ctl_ha_msg persis_io;
8293	int    i;
8294
8295	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8296
8297	retval = CTL_RETVAL_COMPLETE;
8298
8299	softc = control_softc;
8300
8301	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8302	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8303
8304	/*
8305	 * We only support whole-LUN scope.  The scope & type are ignored for
8306	 * register, register and ignore existing key and clear.
8307	 * We sometimes ignore scope and type on preempts too!!
8308	 * Verify reservation type here as well.
8309	 */
8310	type = cdb->scope_type & SPR_TYPE_MASK;
8311	if ((cdb->action == SPRO_RESERVE)
8312	 || (cdb->action == SPRO_RELEASE)) {
8313		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8314			ctl_set_invalid_field(/*ctsio*/ ctsio,
8315					      /*sks_valid*/ 1,
8316					      /*command*/ 1,
8317					      /*field*/ 2,
8318					      /*bit_valid*/ 1,
8319					      /*bit*/ 4);
8320			ctl_done((union ctl_io *)ctsio);
8321			return (CTL_RETVAL_COMPLETE);
8322		}
8323
8324		if (type>8 || type==2 || type==4 || type==0) {
8325			ctl_set_invalid_field(/*ctsio*/ ctsio,
8326					      /*sks_valid*/ 1,
8327					      /*command*/ 1,
8328					      /*field*/ 2,
8329					      /*bit_valid*/ 1,
8330					      /*bit*/ 0);
8331			ctl_done((union ctl_io *)ctsio);
8332			return (CTL_RETVAL_COMPLETE);
8333		}
8334	}
8335
8336	param_len = scsi_4btoul(cdb->length);
8337
8338	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8339		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8340		ctsio->kern_data_len = param_len;
8341		ctsio->kern_total_len = param_len;
8342		ctsio->kern_data_resid = 0;
8343		ctsio->kern_rel_offset = 0;
8344		ctsio->kern_sg_entries = 0;
8345		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8346		ctsio->be_move_done = ctl_config_move_done;
8347		ctl_datamove((union ctl_io *)ctsio);
8348
8349		return (CTL_RETVAL_COMPLETE);
8350	}
8351
8352	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8353
8354	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8355	res_key = scsi_8btou64(param->res_key.key);
8356	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8357
8358	/*
8359	 * Validate the reservation key here except for SPRO_REG_IGNO
8360	 * This must be done for all other service actions
8361	 */
8362	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8363		mtx_lock(&lun->lun_lock);
8364		if (lun->pr_keys[residx] != 0) {
8365		    if (res_key != lun->pr_keys[residx]) {
8366				/*
8367				 * The current key passed in doesn't match
8368				 * the one the initiator previously
8369				 * registered.
8370				 */
8371				mtx_unlock(&lun->lun_lock);
8372				free(ctsio->kern_data_ptr, M_CTL);
8373				ctl_set_reservation_conflict(ctsio);
8374				ctl_done((union ctl_io *)ctsio);
8375				return (CTL_RETVAL_COMPLETE);
8376			}
8377		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8378			/*
8379			 * We are not registered
8380			 */
8381			mtx_unlock(&lun->lun_lock);
8382			free(ctsio->kern_data_ptr, M_CTL);
8383			ctl_set_reservation_conflict(ctsio);
8384			ctl_done((union ctl_io *)ctsio);
8385			return (CTL_RETVAL_COMPLETE);
8386		} else if (res_key != 0) {
8387			/*
8388			 * We are not registered and trying to register but
8389			 * the register key isn't zero.
8390			 */
8391			mtx_unlock(&lun->lun_lock);
8392			free(ctsio->kern_data_ptr, M_CTL);
8393			ctl_set_reservation_conflict(ctsio);
8394			ctl_done((union ctl_io *)ctsio);
8395			return (CTL_RETVAL_COMPLETE);
8396		}
8397		mtx_unlock(&lun->lun_lock);
8398	}
8399
8400	switch (cdb->action & SPRO_ACTION_MASK) {
8401	case SPRO_REGISTER:
8402	case SPRO_REG_IGNO: {
8403
8404#if 0
8405		printf("Registration received\n");
8406#endif
8407
8408		/*
8409		 * We don't support any of these options, as we report in
8410		 * the read capabilities request (see
8411		 * ctl_persistent_reserve_in(), above).
8412		 */
8413		if ((param->flags & SPR_SPEC_I_PT)
8414		 || (param->flags & SPR_ALL_TG_PT)
8415		 || (param->flags & SPR_APTPL)) {
8416			int bit_ptr;
8417
8418			if (param->flags & SPR_APTPL)
8419				bit_ptr = 0;
8420			else if (param->flags & SPR_ALL_TG_PT)
8421				bit_ptr = 2;
8422			else /* SPR_SPEC_I_PT */
8423				bit_ptr = 3;
8424
8425			free(ctsio->kern_data_ptr, M_CTL);
8426			ctl_set_invalid_field(ctsio,
8427					      /*sks_valid*/ 1,
8428					      /*command*/ 0,
8429					      /*field*/ 20,
8430					      /*bit_valid*/ 1,
8431					      /*bit*/ bit_ptr);
8432			ctl_done((union ctl_io *)ctsio);
8433			return (CTL_RETVAL_COMPLETE);
8434		}
8435
8436		mtx_lock(&lun->lun_lock);
8437
8438		/*
8439		 * The initiator wants to clear the
8440		 * key/unregister.
8441		 */
8442		if (sa_res_key == 0) {
8443			if ((res_key == 0
8444			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8445			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8446			  && lun->pr_keys[residx] == 0)) {
8447				mtx_unlock(&lun->lun_lock);
8448				goto done;
8449			}
8450
8451			lun->pr_keys[residx] = 0;
8452			lun->pr_key_count--;
8453
8454			if (residx == lun->pr_res_idx) {
8455				lun->flags &= ~CTL_LUN_PR_RESERVED;
8456				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8457
8458				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8459				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8460				 && lun->pr_key_count) {
8461					/*
8462					 * If the reservation is a registrants
8463					 * only type we need to generate a UA
8464					 * for other registered inits.  The
8465					 * sense code should be RESERVATIONS
8466					 * RELEASED
8467					 */
8468
8469					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8470						if (lun->pr_keys[
8471						    i + persis_offset] == 0)
8472							continue;
8473						lun->pending_ua[i] |=
8474							CTL_UA_RES_RELEASE;
8475					}
8476				}
8477				lun->res_type = 0;
8478			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8479				if (lun->pr_key_count==0) {
8480					lun->flags &= ~CTL_LUN_PR_RESERVED;
8481					lun->res_type = 0;
8482					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8483				}
8484			}
8485			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8486			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8487			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8488			persis_io.pr.pr_info.residx = residx;
8489			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8490			     &persis_io, sizeof(persis_io), 0 )) >
8491			     CTL_HA_STATUS_SUCCESS) {
8492				printf("CTL:Persis Out error returned from "
8493				       "ctl_ha_msg_send %d\n", isc_retval);
8494			}
8495		} else /* sa_res_key != 0 */ {
8496
8497			/*
8498			 * If we aren't registered currently then increment
8499			 * the key count and set the registered flag.
8500			 */
8501			if (lun->pr_keys[residx] == 0)
8502				lun->pr_key_count++;
8503			lun->pr_keys[residx] = sa_res_key;
8504
8505			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8506			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8507			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8508			persis_io.pr.pr_info.residx = residx;
8509			memcpy(persis_io.pr.pr_info.sa_res_key,
8510			       param->serv_act_res_key,
8511			       sizeof(param->serv_act_res_key));
8512			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8513			     &persis_io, sizeof(persis_io), 0)) >
8514			     CTL_HA_STATUS_SUCCESS) {
8515				printf("CTL:Persis Out error returned from "
8516				       "ctl_ha_msg_send %d\n", isc_retval);
8517			}
8518		}
8519		lun->PRGeneration++;
8520		mtx_unlock(&lun->lun_lock);
8521
8522		break;
8523	}
8524	case SPRO_RESERVE:
8525#if 0
8526                printf("Reserve executed type %d\n", type);
8527#endif
8528		mtx_lock(&lun->lun_lock);
8529		if (lun->flags & CTL_LUN_PR_RESERVED) {
8530			/*
8531			 * if this isn't the reservation holder and it's
8532			 * not a "all registrants" type or if the type is
8533			 * different then we have a conflict
8534			 */
8535			if ((lun->pr_res_idx != residx
8536			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8537			 || lun->res_type != type) {
8538				mtx_unlock(&lun->lun_lock);
8539				free(ctsio->kern_data_ptr, M_CTL);
8540				ctl_set_reservation_conflict(ctsio);
8541				ctl_done((union ctl_io *)ctsio);
8542				return (CTL_RETVAL_COMPLETE);
8543			}
8544			mtx_unlock(&lun->lun_lock);
8545		} else /* create a reservation */ {
8546			/*
8547			 * If it's not an "all registrants" type record
8548			 * reservation holder
8549			 */
8550			if (type != SPR_TYPE_WR_EX_AR
8551			 && type != SPR_TYPE_EX_AC_AR)
8552				lun->pr_res_idx = residx; /* Res holder */
8553			else
8554				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8555
8556			lun->flags |= CTL_LUN_PR_RESERVED;
8557			lun->res_type = type;
8558
8559			mtx_unlock(&lun->lun_lock);
8560
8561			/* send msg to other side */
8562			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8563			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8564			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8565			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8566			persis_io.pr.pr_info.res_type = type;
8567			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8568			     &persis_io, sizeof(persis_io), 0)) >
8569			     CTL_HA_STATUS_SUCCESS) {
8570				printf("CTL:Persis Out error returned from "
8571				       "ctl_ha_msg_send %d\n", isc_retval);
8572			}
8573		}
8574		break;
8575
8576	case SPRO_RELEASE:
8577		mtx_lock(&lun->lun_lock);
8578		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8579			/* No reservation exists return good status */
8580			mtx_unlock(&lun->lun_lock);
8581			goto done;
8582		}
8583		/*
8584		 * Is this nexus a reservation holder?
8585		 */
8586		if (lun->pr_res_idx != residx
8587		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8588			/*
8589			 * not a res holder return good status but
8590			 * do nothing
8591			 */
8592			mtx_unlock(&lun->lun_lock);
8593			goto done;
8594		}
8595
8596		if (lun->res_type != type) {
8597			mtx_unlock(&lun->lun_lock);
8598			free(ctsio->kern_data_ptr, M_CTL);
8599			ctl_set_illegal_pr_release(ctsio);
8600			ctl_done((union ctl_io *)ctsio);
8601			return (CTL_RETVAL_COMPLETE);
8602		}
8603
8604		/* okay to release */
8605		lun->flags &= ~CTL_LUN_PR_RESERVED;
8606		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8607		lun->res_type = 0;
8608
8609		/*
8610		 * if this isn't an exclusive access
8611		 * res generate UA for all other
8612		 * registrants.
8613		 */
8614		if (type != SPR_TYPE_EX_AC
8615		 && type != SPR_TYPE_WR_EX) {
8616			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8617				if (i == residx ||
8618				    lun->pr_keys[i + persis_offset] == 0)
8619					continue;
8620				lun->pending_ua[i] |= CTL_UA_RES_RELEASE;
8621			}
8622		}
8623		mtx_unlock(&lun->lun_lock);
8624		/* Send msg to other side */
8625		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8626		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8627		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8628		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8629		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8630			printf("CTL:Persis Out error returned from "
8631			       "ctl_ha_msg_send %d\n", isc_retval);
8632		}
8633		break;
8634
8635	case SPRO_CLEAR:
8636		/* send msg to other side */
8637
8638		mtx_lock(&lun->lun_lock);
8639		lun->flags &= ~CTL_LUN_PR_RESERVED;
8640		lun->res_type = 0;
8641		lun->pr_key_count = 0;
8642		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8643
8644		lun->pr_keys[residx] = 0;
8645
8646		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8647			if (lun->pr_keys[i] != 0) {
8648				if (!persis_offset && i < CTL_MAX_INITIATORS)
8649					lun->pending_ua[i] |=
8650						CTL_UA_RES_PREEMPT;
8651				else if (persis_offset && i >= persis_offset)
8652					lun->pending_ua[i-persis_offset] |=
8653					    CTL_UA_RES_PREEMPT;
8654
8655				lun->pr_keys[i] = 0;
8656			}
8657		lun->PRGeneration++;
8658		mtx_unlock(&lun->lun_lock);
8659		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8660		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8661		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8662		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8663		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8664			printf("CTL:Persis Out error returned from "
8665			       "ctl_ha_msg_send %d\n", isc_retval);
8666		}
8667		break;
8668
8669	case SPRO_PREEMPT:
8670	case SPRO_PRE_ABO: {
8671		int nretval;
8672
8673		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8674					  residx, ctsio, cdb, param);
8675		if (nretval != 0)
8676			return (CTL_RETVAL_COMPLETE);
8677		break;
8678	}
8679	default:
8680		panic("Invalid PR type %x", cdb->action);
8681	}
8682
8683done:
8684	free(ctsio->kern_data_ptr, M_CTL);
8685	ctl_set_success(ctsio);
8686	ctl_done((union ctl_io *)ctsio);
8687
8688	return (retval);
8689}
8690
8691/*
8692 * This routine is for handling a message from the other SC pertaining to
8693 * persistent reserve out. All the error checking will have been done
8694 * so only perorming the action need be done here to keep the two
8695 * in sync.
8696 */
8697static void
8698ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8699{
8700	struct ctl_lun *lun;
8701	struct ctl_softc *softc;
8702	int i;
8703	uint32_t targ_lun;
8704
8705	softc = control_softc;
8706
8707	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8708	lun = softc->ctl_luns[targ_lun];
8709	mtx_lock(&lun->lun_lock);
8710	switch(msg->pr.pr_info.action) {
8711	case CTL_PR_REG_KEY:
8712		if (lun->pr_keys[msg->pr.pr_info.residx] == 0)
8713			lun->pr_key_count++;
8714		lun->pr_keys[msg->pr.pr_info.residx] =
8715		    scsi_8btou64(msg->pr.pr_info.sa_res_key);
8716		lun->PRGeneration++;
8717		break;
8718
8719	case CTL_PR_UNREG_KEY:
8720		lun->pr_keys[msg->pr.pr_info.residx] = 0;
8721		lun->pr_key_count--;
8722
8723		/* XXX Need to see if the reservation has been released */
8724		/* if so do we need to generate UA? */
8725		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8726			lun->flags &= ~CTL_LUN_PR_RESERVED;
8727			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8728
8729			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8730			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8731			 && lun->pr_key_count) {
8732				/*
8733				 * If the reservation is a registrants
8734				 * only type we need to generate a UA
8735				 * for other registered inits.  The
8736				 * sense code should be RESERVATIONS
8737				 * RELEASED
8738				 */
8739
8740				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8741					if (lun->pr_keys[i+
8742					    persis_offset] == 0)
8743						continue;
8744
8745					lun->pending_ua[i] |=
8746						CTL_UA_RES_RELEASE;
8747				}
8748			}
8749			lun->res_type = 0;
8750		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8751			if (lun->pr_key_count==0) {
8752				lun->flags &= ~CTL_LUN_PR_RESERVED;
8753				lun->res_type = 0;
8754				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8755			}
8756		}
8757		lun->PRGeneration++;
8758		break;
8759
8760	case CTL_PR_RESERVE:
8761		lun->flags |= CTL_LUN_PR_RESERVED;
8762		lun->res_type = msg->pr.pr_info.res_type;
8763		lun->pr_res_idx = msg->pr.pr_info.residx;
8764
8765		break;
8766
8767	case CTL_PR_RELEASE:
8768		/*
8769		 * if this isn't an exclusive access res generate UA for all
8770		 * other registrants.
8771		 */
8772		if (lun->res_type != SPR_TYPE_EX_AC
8773		 && lun->res_type != SPR_TYPE_WR_EX) {
8774			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8775				if (lun->pr_keys[i+persis_offset] != 0)
8776					lun->pending_ua[i] |=
8777						CTL_UA_RES_RELEASE;
8778		}
8779
8780		lun->flags &= ~CTL_LUN_PR_RESERVED;
8781		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8782		lun->res_type = 0;
8783		break;
8784
8785	case CTL_PR_PREEMPT:
8786		ctl_pro_preempt_other(lun, msg);
8787		break;
8788	case CTL_PR_CLEAR:
8789		lun->flags &= ~CTL_LUN_PR_RESERVED;
8790		lun->res_type = 0;
8791		lun->pr_key_count = 0;
8792		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8793
8794		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8795			if (lun->pr_keys[i] == 0)
8796				continue;
8797			if (!persis_offset
8798			 && i < CTL_MAX_INITIATORS)
8799				lun->pending_ua[i] |= CTL_UA_RES_PREEMPT;
8800			else if (persis_offset
8801			      && i >= persis_offset)
8802				lun->pending_ua[i-persis_offset] |=
8803					CTL_UA_RES_PREEMPT;
8804			lun->pr_keys[i] = 0;
8805		}
8806		lun->PRGeneration++;
8807		break;
8808	}
8809
8810	mtx_unlock(&lun->lun_lock);
8811}
8812
8813int
8814ctl_read_write(struct ctl_scsiio *ctsio)
8815{
8816	struct ctl_lun *lun;
8817	struct ctl_lba_len_flags *lbalen;
8818	uint64_t lba;
8819	uint32_t num_blocks;
8820	int flags, retval;
8821	int isread;
8822
8823	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8824
8825	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8826
8827	flags = 0;
8828	retval = CTL_RETVAL_COMPLETE;
8829
8830	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8831	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8832	switch (ctsio->cdb[0]) {
8833	case READ_6:
8834	case WRITE_6: {
8835		struct scsi_rw_6 *cdb;
8836
8837		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8838
8839		lba = scsi_3btoul(cdb->addr);
8840		/* only 5 bits are valid in the most significant address byte */
8841		lba &= 0x1fffff;
8842		num_blocks = cdb->length;
8843		/*
8844		 * This is correct according to SBC-2.
8845		 */
8846		if (num_blocks == 0)
8847			num_blocks = 256;
8848		break;
8849	}
8850	case READ_10:
8851	case WRITE_10: {
8852		struct scsi_rw_10 *cdb;
8853
8854		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8855		if (cdb->byte2 & SRW10_FUA)
8856			flags |= CTL_LLF_FUA;
8857		if (cdb->byte2 & SRW10_DPO)
8858			flags |= CTL_LLF_DPO;
8859		lba = scsi_4btoul(cdb->addr);
8860		num_blocks = scsi_2btoul(cdb->length);
8861		break;
8862	}
8863	case WRITE_VERIFY_10: {
8864		struct scsi_write_verify_10 *cdb;
8865
8866		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8867		flags |= CTL_LLF_FUA;
8868		if (cdb->byte2 & SWV_DPO)
8869			flags |= CTL_LLF_DPO;
8870		lba = scsi_4btoul(cdb->addr);
8871		num_blocks = scsi_2btoul(cdb->length);
8872		break;
8873	}
8874	case READ_12:
8875	case WRITE_12: {
8876		struct scsi_rw_12 *cdb;
8877
8878		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8879		if (cdb->byte2 & SRW12_FUA)
8880			flags |= CTL_LLF_FUA;
8881		if (cdb->byte2 & SRW12_DPO)
8882			flags |= CTL_LLF_DPO;
8883		lba = scsi_4btoul(cdb->addr);
8884		num_blocks = scsi_4btoul(cdb->length);
8885		break;
8886	}
8887	case WRITE_VERIFY_12: {
8888		struct scsi_write_verify_12 *cdb;
8889
8890		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8891		flags |= CTL_LLF_FUA;
8892		if (cdb->byte2 & SWV_DPO)
8893			flags |= CTL_LLF_DPO;
8894		lba = scsi_4btoul(cdb->addr);
8895		num_blocks = scsi_4btoul(cdb->length);
8896		break;
8897	}
8898	case READ_16:
8899	case WRITE_16: {
8900		struct scsi_rw_16 *cdb;
8901
8902		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8903		if (cdb->byte2 & SRW12_FUA)
8904			flags |= CTL_LLF_FUA;
8905		if (cdb->byte2 & SRW12_DPO)
8906			flags |= CTL_LLF_DPO;
8907		lba = scsi_8btou64(cdb->addr);
8908		num_blocks = scsi_4btoul(cdb->length);
8909		break;
8910	}
8911	case WRITE_ATOMIC_16: {
8912		struct scsi_rw_16 *cdb;
8913
8914		if (lun->be_lun->atomicblock == 0) {
8915			ctl_set_invalid_opcode(ctsio);
8916			ctl_done((union ctl_io *)ctsio);
8917			return (CTL_RETVAL_COMPLETE);
8918		}
8919
8920		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8921		if (cdb->byte2 & SRW12_FUA)
8922			flags |= CTL_LLF_FUA;
8923		if (cdb->byte2 & SRW12_DPO)
8924			flags |= CTL_LLF_DPO;
8925		lba = scsi_8btou64(cdb->addr);
8926		num_blocks = scsi_4btoul(cdb->length);
8927		if (num_blocks > lun->be_lun->atomicblock) {
8928			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8929			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8930			    /*bit*/ 0);
8931			ctl_done((union ctl_io *)ctsio);
8932			return (CTL_RETVAL_COMPLETE);
8933		}
8934		break;
8935	}
8936	case WRITE_VERIFY_16: {
8937		struct scsi_write_verify_16 *cdb;
8938
8939		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8940		flags |= CTL_LLF_FUA;
8941		if (cdb->byte2 & SWV_DPO)
8942			flags |= CTL_LLF_DPO;
8943		lba = scsi_8btou64(cdb->addr);
8944		num_blocks = scsi_4btoul(cdb->length);
8945		break;
8946	}
8947	default:
8948		/*
8949		 * We got a command we don't support.  This shouldn't
8950		 * happen, commands should be filtered out above us.
8951		 */
8952		ctl_set_invalid_opcode(ctsio);
8953		ctl_done((union ctl_io *)ctsio);
8954
8955		return (CTL_RETVAL_COMPLETE);
8956		break; /* NOTREACHED */
8957	}
8958
8959	/*
8960	 * The first check is to make sure we're in bounds, the second
8961	 * check is to catch wrap-around problems.  If the lba + num blocks
8962	 * is less than the lba, then we've wrapped around and the block
8963	 * range is invalid anyway.
8964	 */
8965	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8966	 || ((lba + num_blocks) < lba)) {
8967		ctl_set_lba_out_of_range(ctsio);
8968		ctl_done((union ctl_io *)ctsio);
8969		return (CTL_RETVAL_COMPLETE);
8970	}
8971
8972	/*
8973	 * According to SBC-3, a transfer length of 0 is not an error.
8974	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8975	 * translates to 256 blocks for those commands.
8976	 */
8977	if (num_blocks == 0) {
8978		ctl_set_success(ctsio);
8979		ctl_done((union ctl_io *)ctsio);
8980		return (CTL_RETVAL_COMPLETE);
8981	}
8982
8983	/* Set FUA and/or DPO if caches are disabled. */
8984	if (isread) {
8985		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8986		    SCP_RCD) != 0)
8987			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8988	} else {
8989		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8990		    SCP_WCE) == 0)
8991			flags |= CTL_LLF_FUA;
8992	}
8993
8994	lbalen = (struct ctl_lba_len_flags *)
8995	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8996	lbalen->lba = lba;
8997	lbalen->len = num_blocks;
8998	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8999
9000	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9001	ctsio->kern_rel_offset = 0;
9002
9003	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9004
9005	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9006
9007	return (retval);
9008}
9009
9010static int
9011ctl_cnw_cont(union ctl_io *io)
9012{
9013	struct ctl_scsiio *ctsio;
9014	struct ctl_lun *lun;
9015	struct ctl_lba_len_flags *lbalen;
9016	int retval;
9017
9018	ctsio = &io->scsiio;
9019	ctsio->io_hdr.status = CTL_STATUS_NONE;
9020	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9021	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9022	lbalen = (struct ctl_lba_len_flags *)
9023	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9024	lbalen->flags &= ~CTL_LLF_COMPARE;
9025	lbalen->flags |= CTL_LLF_WRITE;
9026
9027	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9028	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9029	return (retval);
9030}
9031
9032int
9033ctl_cnw(struct ctl_scsiio *ctsio)
9034{
9035	struct ctl_lun *lun;
9036	struct ctl_lba_len_flags *lbalen;
9037	uint64_t lba;
9038	uint32_t num_blocks;
9039	int flags, retval;
9040
9041	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9042
9043	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9044
9045	flags = 0;
9046	retval = CTL_RETVAL_COMPLETE;
9047
9048	switch (ctsio->cdb[0]) {
9049	case COMPARE_AND_WRITE: {
9050		struct scsi_compare_and_write *cdb;
9051
9052		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9053		if (cdb->byte2 & SRW10_FUA)
9054			flags |= CTL_LLF_FUA;
9055		if (cdb->byte2 & SRW10_DPO)
9056			flags |= CTL_LLF_DPO;
9057		lba = scsi_8btou64(cdb->addr);
9058		num_blocks = cdb->length;
9059		break;
9060	}
9061	default:
9062		/*
9063		 * We got a command we don't support.  This shouldn't
9064		 * happen, commands should be filtered out above us.
9065		 */
9066		ctl_set_invalid_opcode(ctsio);
9067		ctl_done((union ctl_io *)ctsio);
9068
9069		return (CTL_RETVAL_COMPLETE);
9070		break; /* NOTREACHED */
9071	}
9072
9073	/*
9074	 * The first check is to make sure we're in bounds, the second
9075	 * check is to catch wrap-around problems.  If the lba + num blocks
9076	 * is less than the lba, then we've wrapped around and the block
9077	 * range is invalid anyway.
9078	 */
9079	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9080	 || ((lba + num_blocks) < lba)) {
9081		ctl_set_lba_out_of_range(ctsio);
9082		ctl_done((union ctl_io *)ctsio);
9083		return (CTL_RETVAL_COMPLETE);
9084	}
9085
9086	/*
9087	 * According to SBC-3, a transfer length of 0 is not an error.
9088	 */
9089	if (num_blocks == 0) {
9090		ctl_set_success(ctsio);
9091		ctl_done((union ctl_io *)ctsio);
9092		return (CTL_RETVAL_COMPLETE);
9093	}
9094
9095	/* Set FUA if write cache is disabled. */
9096	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9097	    SCP_WCE) == 0)
9098		flags |= CTL_LLF_FUA;
9099
9100	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9101	ctsio->kern_rel_offset = 0;
9102
9103	/*
9104	 * Set the IO_CONT flag, so that if this I/O gets passed to
9105	 * ctl_data_submit_done(), it'll get passed back to
9106	 * ctl_ctl_cnw_cont() for further processing.
9107	 */
9108	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9109	ctsio->io_cont = ctl_cnw_cont;
9110
9111	lbalen = (struct ctl_lba_len_flags *)
9112	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9113	lbalen->lba = lba;
9114	lbalen->len = num_blocks;
9115	lbalen->flags = CTL_LLF_COMPARE | flags;
9116
9117	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9118	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9119	return (retval);
9120}
9121
9122int
9123ctl_verify(struct ctl_scsiio *ctsio)
9124{
9125	struct ctl_lun *lun;
9126	struct ctl_lba_len_flags *lbalen;
9127	uint64_t lba;
9128	uint32_t num_blocks;
9129	int bytchk, flags;
9130	int retval;
9131
9132	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9133
9134	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9135
9136	bytchk = 0;
9137	flags = CTL_LLF_FUA;
9138	retval = CTL_RETVAL_COMPLETE;
9139
9140	switch (ctsio->cdb[0]) {
9141	case VERIFY_10: {
9142		struct scsi_verify_10 *cdb;
9143
9144		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9145		if (cdb->byte2 & SVFY_BYTCHK)
9146			bytchk = 1;
9147		if (cdb->byte2 & SVFY_DPO)
9148			flags |= CTL_LLF_DPO;
9149		lba = scsi_4btoul(cdb->addr);
9150		num_blocks = scsi_2btoul(cdb->length);
9151		break;
9152	}
9153	case VERIFY_12: {
9154		struct scsi_verify_12 *cdb;
9155
9156		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9157		if (cdb->byte2 & SVFY_BYTCHK)
9158			bytchk = 1;
9159		if (cdb->byte2 & SVFY_DPO)
9160			flags |= CTL_LLF_DPO;
9161		lba = scsi_4btoul(cdb->addr);
9162		num_blocks = scsi_4btoul(cdb->length);
9163		break;
9164	}
9165	case VERIFY_16: {
9166		struct scsi_rw_16 *cdb;
9167
9168		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9169		if (cdb->byte2 & SVFY_BYTCHK)
9170			bytchk = 1;
9171		if (cdb->byte2 & SVFY_DPO)
9172			flags |= CTL_LLF_DPO;
9173		lba = scsi_8btou64(cdb->addr);
9174		num_blocks = scsi_4btoul(cdb->length);
9175		break;
9176	}
9177	default:
9178		/*
9179		 * We got a command we don't support.  This shouldn't
9180		 * happen, commands should be filtered out above us.
9181		 */
9182		ctl_set_invalid_opcode(ctsio);
9183		ctl_done((union ctl_io *)ctsio);
9184		return (CTL_RETVAL_COMPLETE);
9185	}
9186
9187	/*
9188	 * The first check is to make sure we're in bounds, the second
9189	 * check is to catch wrap-around problems.  If the lba + num blocks
9190	 * is less than the lba, then we've wrapped around and the block
9191	 * range is invalid anyway.
9192	 */
9193	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9194	 || ((lba + num_blocks) < lba)) {
9195		ctl_set_lba_out_of_range(ctsio);
9196		ctl_done((union ctl_io *)ctsio);
9197		return (CTL_RETVAL_COMPLETE);
9198	}
9199
9200	/*
9201	 * According to SBC-3, a transfer length of 0 is not an error.
9202	 */
9203	if (num_blocks == 0) {
9204		ctl_set_success(ctsio);
9205		ctl_done((union ctl_io *)ctsio);
9206		return (CTL_RETVAL_COMPLETE);
9207	}
9208
9209	lbalen = (struct ctl_lba_len_flags *)
9210	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9211	lbalen->lba = lba;
9212	lbalen->len = num_blocks;
9213	if (bytchk) {
9214		lbalen->flags = CTL_LLF_COMPARE | flags;
9215		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9216	} else {
9217		lbalen->flags = CTL_LLF_VERIFY | flags;
9218		ctsio->kern_total_len = 0;
9219	}
9220	ctsio->kern_rel_offset = 0;
9221
9222	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9223	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9224	return (retval);
9225}
9226
9227int
9228ctl_report_luns(struct ctl_scsiio *ctsio)
9229{
9230	struct scsi_report_luns *cdb;
9231	struct scsi_report_luns_data *lun_data;
9232	struct ctl_lun *lun, *request_lun;
9233	int num_luns, retval;
9234	uint32_t alloc_len, lun_datalen;
9235	int num_filled, well_known;
9236	uint32_t initidx, targ_lun_id, lun_id;
9237
9238	retval = CTL_RETVAL_COMPLETE;
9239	well_known = 0;
9240
9241	cdb = (struct scsi_report_luns *)ctsio->cdb;
9242
9243	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9244
9245	mtx_lock(&control_softc->ctl_lock);
9246	num_luns = control_softc->num_luns;
9247	mtx_unlock(&control_softc->ctl_lock);
9248
9249	switch (cdb->select_report) {
9250	case RPL_REPORT_DEFAULT:
9251	case RPL_REPORT_ALL:
9252		break;
9253	case RPL_REPORT_WELLKNOWN:
9254		well_known = 1;
9255		num_luns = 0;
9256		break;
9257	default:
9258		ctl_set_invalid_field(ctsio,
9259				      /*sks_valid*/ 1,
9260				      /*command*/ 1,
9261				      /*field*/ 2,
9262				      /*bit_valid*/ 0,
9263				      /*bit*/ 0);
9264		ctl_done((union ctl_io *)ctsio);
9265		return (retval);
9266		break; /* NOTREACHED */
9267	}
9268
9269	alloc_len = scsi_4btoul(cdb->length);
9270	/*
9271	 * The initiator has to allocate at least 16 bytes for this request,
9272	 * so he can at least get the header and the first LUN.  Otherwise
9273	 * we reject the request (per SPC-3 rev 14, section 6.21).
9274	 */
9275	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9276	    sizeof(struct scsi_report_luns_lundata))) {
9277		ctl_set_invalid_field(ctsio,
9278				      /*sks_valid*/ 1,
9279				      /*command*/ 1,
9280				      /*field*/ 6,
9281				      /*bit_valid*/ 0,
9282				      /*bit*/ 0);
9283		ctl_done((union ctl_io *)ctsio);
9284		return (retval);
9285	}
9286
9287	request_lun = (struct ctl_lun *)
9288		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9289
9290	lun_datalen = sizeof(*lun_data) +
9291		(num_luns * sizeof(struct scsi_report_luns_lundata));
9292
9293	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9294	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9295	ctsio->kern_sg_entries = 0;
9296
9297	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9298
9299	mtx_lock(&control_softc->ctl_lock);
9300	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9301		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9302		if (lun_id >= CTL_MAX_LUNS)
9303			continue;
9304		lun = control_softc->ctl_luns[lun_id];
9305		if (lun == NULL)
9306			continue;
9307
9308		if (targ_lun_id <= 0xff) {
9309			/*
9310			 * Peripheral addressing method, bus number 0.
9311			 */
9312			lun_data->luns[num_filled].lundata[0] =
9313				RPL_LUNDATA_ATYP_PERIPH;
9314			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9315			num_filled++;
9316		} else if (targ_lun_id <= 0x3fff) {
9317			/*
9318			 * Flat addressing method.
9319			 */
9320			lun_data->luns[num_filled].lundata[0] =
9321				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9322			lun_data->luns[num_filled].lundata[1] =
9323				(targ_lun_id & 0xff);
9324			num_filled++;
9325		} else if (targ_lun_id <= 0xffffff) {
9326			/*
9327			 * Extended flat addressing method.
9328			 */
9329			lun_data->luns[num_filled].lundata[0] =
9330			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9331			scsi_ulto3b(targ_lun_id,
9332			    &lun_data->luns[num_filled].lundata[1]);
9333			num_filled++;
9334		} else {
9335			printf("ctl_report_luns: bogus LUN number %jd, "
9336			       "skipping\n", (intmax_t)targ_lun_id);
9337		}
9338		/*
9339		 * According to SPC-3, rev 14 section 6.21:
9340		 *
9341		 * "The execution of a REPORT LUNS command to any valid and
9342		 * installed logical unit shall clear the REPORTED LUNS DATA
9343		 * HAS CHANGED unit attention condition for all logical
9344		 * units of that target with respect to the requesting
9345		 * initiator. A valid and installed logical unit is one
9346		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9347		 * INQUIRY data (see 6.4.2)."
9348		 *
9349		 * If request_lun is NULL, the LUN this report luns command
9350		 * was issued to is either disabled or doesn't exist. In that
9351		 * case, we shouldn't clear any pending lun change unit
9352		 * attention.
9353		 */
9354		if (request_lun != NULL) {
9355			mtx_lock(&lun->lun_lock);
9356			lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE;
9357			mtx_unlock(&lun->lun_lock);
9358		}
9359	}
9360	mtx_unlock(&control_softc->ctl_lock);
9361
9362	/*
9363	 * It's quite possible that we've returned fewer LUNs than we allocated
9364	 * space for.  Trim it.
9365	 */
9366	lun_datalen = sizeof(*lun_data) +
9367		(num_filled * sizeof(struct scsi_report_luns_lundata));
9368
9369	if (lun_datalen < alloc_len) {
9370		ctsio->residual = alloc_len - lun_datalen;
9371		ctsio->kern_data_len = lun_datalen;
9372		ctsio->kern_total_len = lun_datalen;
9373	} else {
9374		ctsio->residual = 0;
9375		ctsio->kern_data_len = alloc_len;
9376		ctsio->kern_total_len = alloc_len;
9377	}
9378	ctsio->kern_data_resid = 0;
9379	ctsio->kern_rel_offset = 0;
9380	ctsio->kern_sg_entries = 0;
9381
9382	/*
9383	 * We set this to the actual data length, regardless of how much
9384	 * space we actually have to return results.  If the user looks at
9385	 * this value, he'll know whether or not he allocated enough space
9386	 * and reissue the command if necessary.  We don't support well
9387	 * known logical units, so if the user asks for that, return none.
9388	 */
9389	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9390
9391	/*
9392	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9393	 * this request.
9394	 */
9395	ctl_set_success(ctsio);
9396	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9397	ctsio->be_move_done = ctl_config_move_done;
9398	ctl_datamove((union ctl_io *)ctsio);
9399	return (retval);
9400}
9401
9402int
9403ctl_request_sense(struct ctl_scsiio *ctsio)
9404{
9405	struct scsi_request_sense *cdb;
9406	struct scsi_sense_data *sense_ptr;
9407	struct ctl_lun *lun;
9408	uint32_t initidx;
9409	int have_error;
9410	scsi_sense_data_type sense_format;
9411
9412	cdb = (struct scsi_request_sense *)ctsio->cdb;
9413
9414	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9415
9416	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9417
9418	/*
9419	 * Determine which sense format the user wants.
9420	 */
9421	if (cdb->byte2 & SRS_DESC)
9422		sense_format = SSD_TYPE_DESC;
9423	else
9424		sense_format = SSD_TYPE_FIXED;
9425
9426	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9427	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9428	ctsio->kern_sg_entries = 0;
9429
9430	/*
9431	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9432	 * larger than the largest allowed value for the length field in the
9433	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9434	 */
9435	ctsio->residual = 0;
9436	ctsio->kern_data_len = cdb->length;
9437	ctsio->kern_total_len = cdb->length;
9438
9439	ctsio->kern_data_resid = 0;
9440	ctsio->kern_rel_offset = 0;
9441	ctsio->kern_sg_entries = 0;
9442
9443	/*
9444	 * If we don't have a LUN, we don't have any pending sense.
9445	 */
9446	if (lun == NULL)
9447		goto no_sense;
9448
9449	have_error = 0;
9450	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9451	/*
9452	 * Check for pending sense, and then for pending unit attentions.
9453	 * Pending sense gets returned first, then pending unit attentions.
9454	 */
9455	mtx_lock(&lun->lun_lock);
9456#ifdef CTL_WITH_CA
9457	if (ctl_is_set(lun->have_ca, initidx)) {
9458		scsi_sense_data_type stored_format;
9459
9460		/*
9461		 * Check to see which sense format was used for the stored
9462		 * sense data.
9463		 */
9464		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9465
9466		/*
9467		 * If the user requested a different sense format than the
9468		 * one we stored, then we need to convert it to the other
9469		 * format.  If we're going from descriptor to fixed format
9470		 * sense data, we may lose things in translation, depending
9471		 * on what options were used.
9472		 *
9473		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9474		 * for some reason we'll just copy it out as-is.
9475		 */
9476		if ((stored_format == SSD_TYPE_FIXED)
9477		 && (sense_format == SSD_TYPE_DESC))
9478			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9479			    &lun->pending_sense[initidx],
9480			    (struct scsi_sense_data_desc *)sense_ptr);
9481		else if ((stored_format == SSD_TYPE_DESC)
9482		      && (sense_format == SSD_TYPE_FIXED))
9483			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9484			    &lun->pending_sense[initidx],
9485			    (struct scsi_sense_data_fixed *)sense_ptr);
9486		else
9487			memcpy(sense_ptr, &lun->pending_sense[initidx],
9488			       ctl_min(sizeof(*sense_ptr),
9489			       sizeof(lun->pending_sense[initidx])));
9490
9491		ctl_clear_mask(lun->have_ca, initidx);
9492		have_error = 1;
9493	} else
9494#endif
9495	if (lun->pending_ua[initidx] != CTL_UA_NONE) {
9496		ctl_ua_type ua_type;
9497
9498		ua_type = ctl_build_ua(&lun->pending_ua[initidx],
9499				       sense_ptr, sense_format);
9500		if (ua_type != CTL_UA_NONE)
9501			have_error = 1;
9502	}
9503	mtx_unlock(&lun->lun_lock);
9504
9505	/*
9506	 * We already have a pending error, return it.
9507	 */
9508	if (have_error != 0) {
9509		/*
9510		 * We report the SCSI status as OK, since the status of the
9511		 * request sense command itself is OK.
9512		 * We report 0 for the sense length, because we aren't doing
9513		 * autosense in this case.  We're reporting sense as
9514		 * parameter data.
9515		 */
9516		ctl_set_success(ctsio);
9517		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9518		ctsio->be_move_done = ctl_config_move_done;
9519		ctl_datamove((union ctl_io *)ctsio);
9520		return (CTL_RETVAL_COMPLETE);
9521	}
9522
9523no_sense:
9524
9525	/*
9526	 * No sense information to report, so we report that everything is
9527	 * okay.
9528	 */
9529	ctl_set_sense_data(sense_ptr,
9530			   lun,
9531			   sense_format,
9532			   /*current_error*/ 1,
9533			   /*sense_key*/ SSD_KEY_NO_SENSE,
9534			   /*asc*/ 0x00,
9535			   /*ascq*/ 0x00,
9536			   SSD_ELEM_NONE);
9537
9538	/*
9539	 * We report 0 for the sense length, because we aren't doing
9540	 * autosense in this case.  We're reporting sense as parameter data.
9541	 */
9542	ctl_set_success(ctsio);
9543	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9544	ctsio->be_move_done = ctl_config_move_done;
9545	ctl_datamove((union ctl_io *)ctsio);
9546	return (CTL_RETVAL_COMPLETE);
9547}
9548
9549int
9550ctl_tur(struct ctl_scsiio *ctsio)
9551{
9552
9553	CTL_DEBUG_PRINT(("ctl_tur\n"));
9554
9555	ctl_set_success(ctsio);
9556	ctl_done((union ctl_io *)ctsio);
9557
9558	return (CTL_RETVAL_COMPLETE);
9559}
9560
9561#ifdef notyet
9562static int
9563ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9564{
9565
9566}
9567#endif
9568
9569static int
9570ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9571{
9572	struct scsi_vpd_supported_pages *pages;
9573	int sup_page_size;
9574	struct ctl_lun *lun;
9575
9576	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9577
9578	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9579	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9580	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9581	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9582	ctsio->kern_sg_entries = 0;
9583
9584	if (sup_page_size < alloc_len) {
9585		ctsio->residual = alloc_len - sup_page_size;
9586		ctsio->kern_data_len = sup_page_size;
9587		ctsio->kern_total_len = sup_page_size;
9588	} else {
9589		ctsio->residual = 0;
9590		ctsio->kern_data_len = alloc_len;
9591		ctsio->kern_total_len = alloc_len;
9592	}
9593	ctsio->kern_data_resid = 0;
9594	ctsio->kern_rel_offset = 0;
9595	ctsio->kern_sg_entries = 0;
9596
9597	/*
9598	 * The control device is always connected.  The disk device, on the
9599	 * other hand, may not be online all the time.  Need to change this
9600	 * to figure out whether the disk device is actually online or not.
9601	 */
9602	if (lun != NULL)
9603		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9604				lun->be_lun->lun_type;
9605	else
9606		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9607
9608	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9609	/* Supported VPD pages */
9610	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9611	/* Serial Number */
9612	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9613	/* Device Identification */
9614	pages->page_list[2] = SVPD_DEVICE_ID;
9615	/* Extended INQUIRY Data */
9616	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9617	/* Mode Page Policy */
9618	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9619	/* SCSI Ports */
9620	pages->page_list[5] = SVPD_SCSI_PORTS;
9621	/* Third-party Copy */
9622	pages->page_list[6] = SVPD_SCSI_TPC;
9623	/* Block limits */
9624	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9625	/* Block Device Characteristics */
9626	pages->page_list[8] = SVPD_BDC;
9627	/* Logical Block Provisioning */
9628	pages->page_list[9] = SVPD_LBP;
9629
9630	ctl_set_success(ctsio);
9631	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9632	ctsio->be_move_done = ctl_config_move_done;
9633	ctl_datamove((union ctl_io *)ctsio);
9634	return (CTL_RETVAL_COMPLETE);
9635}
9636
9637static int
9638ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9639{
9640	struct scsi_vpd_unit_serial_number *sn_ptr;
9641	struct ctl_lun *lun;
9642	int data_len;
9643
9644	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9645
9646	data_len = 4 + CTL_SN_LEN;
9647	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9648	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9649	if (data_len < alloc_len) {
9650		ctsio->residual = alloc_len - data_len;
9651		ctsio->kern_data_len = data_len;
9652		ctsio->kern_total_len = data_len;
9653	} else {
9654		ctsio->residual = 0;
9655		ctsio->kern_data_len = alloc_len;
9656		ctsio->kern_total_len = alloc_len;
9657	}
9658	ctsio->kern_data_resid = 0;
9659	ctsio->kern_rel_offset = 0;
9660	ctsio->kern_sg_entries = 0;
9661
9662	/*
9663	 * The control device is always connected.  The disk device, on the
9664	 * other hand, may not be online all the time.  Need to change this
9665	 * to figure out whether the disk device is actually online or not.
9666	 */
9667	if (lun != NULL)
9668		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9669				  lun->be_lun->lun_type;
9670	else
9671		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9672
9673	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9674	sn_ptr->length = CTL_SN_LEN;
9675	/*
9676	 * If we don't have a LUN, we just leave the serial number as
9677	 * all spaces.
9678	 */
9679	if (lun != NULL) {
9680		strncpy((char *)sn_ptr->serial_num,
9681			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9682	} else
9683		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9684
9685	ctl_set_success(ctsio);
9686	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9687	ctsio->be_move_done = ctl_config_move_done;
9688	ctl_datamove((union ctl_io *)ctsio);
9689	return (CTL_RETVAL_COMPLETE);
9690}
9691
9692
9693static int
9694ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9695{
9696	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9697	struct ctl_lun *lun;
9698	int data_len;
9699
9700	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9701
9702	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9703	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9704	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9705	ctsio->kern_sg_entries = 0;
9706
9707	if (data_len < alloc_len) {
9708		ctsio->residual = alloc_len - data_len;
9709		ctsio->kern_data_len = data_len;
9710		ctsio->kern_total_len = data_len;
9711	} else {
9712		ctsio->residual = 0;
9713		ctsio->kern_data_len = alloc_len;
9714		ctsio->kern_total_len = alloc_len;
9715	}
9716	ctsio->kern_data_resid = 0;
9717	ctsio->kern_rel_offset = 0;
9718	ctsio->kern_sg_entries = 0;
9719
9720	/*
9721	 * The control device is always connected.  The disk device, on the
9722	 * other hand, may not be online all the time.
9723	 */
9724	if (lun != NULL)
9725		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9726				     lun->be_lun->lun_type;
9727	else
9728		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9729	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9730	eid_ptr->page_length = data_len - 4;
9731	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9732	eid_ptr->flags3 = SVPD_EID_V_SUP;
9733
9734	ctl_set_success(ctsio);
9735	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9736	ctsio->be_move_done = ctl_config_move_done;
9737	ctl_datamove((union ctl_io *)ctsio);
9738	return (CTL_RETVAL_COMPLETE);
9739}
9740
9741static int
9742ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9743{
9744	struct scsi_vpd_mode_page_policy *mpp_ptr;
9745	struct ctl_lun *lun;
9746	int data_len;
9747
9748	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9749
9750	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9751	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9752
9753	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9754	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9755	ctsio->kern_sg_entries = 0;
9756
9757	if (data_len < alloc_len) {
9758		ctsio->residual = alloc_len - data_len;
9759		ctsio->kern_data_len = data_len;
9760		ctsio->kern_total_len = data_len;
9761	} else {
9762		ctsio->residual = 0;
9763		ctsio->kern_data_len = alloc_len;
9764		ctsio->kern_total_len = alloc_len;
9765	}
9766	ctsio->kern_data_resid = 0;
9767	ctsio->kern_rel_offset = 0;
9768	ctsio->kern_sg_entries = 0;
9769
9770	/*
9771	 * The control device is always connected.  The disk device, on the
9772	 * other hand, may not be online all the time.
9773	 */
9774	if (lun != NULL)
9775		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9776				     lun->be_lun->lun_type;
9777	else
9778		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9779	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9780	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9781	mpp_ptr->descr[0].page_code = 0x3f;
9782	mpp_ptr->descr[0].subpage_code = 0xff;
9783	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9784
9785	ctl_set_success(ctsio);
9786	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9787	ctsio->be_move_done = ctl_config_move_done;
9788	ctl_datamove((union ctl_io *)ctsio);
9789	return (CTL_RETVAL_COMPLETE);
9790}
9791
9792static int
9793ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9794{
9795	struct scsi_vpd_device_id *devid_ptr;
9796	struct scsi_vpd_id_descriptor *desc;
9797	struct ctl_softc *ctl_softc;
9798	struct ctl_lun *lun;
9799	struct ctl_port *port;
9800	int data_len;
9801	uint8_t proto;
9802
9803	ctl_softc = control_softc;
9804
9805	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9806	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9807
9808	data_len = sizeof(struct scsi_vpd_device_id) +
9809	    sizeof(struct scsi_vpd_id_descriptor) +
9810		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9811	    sizeof(struct scsi_vpd_id_descriptor) +
9812		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9813	if (lun && lun->lun_devid)
9814		data_len += lun->lun_devid->len;
9815	if (port->port_devid)
9816		data_len += port->port_devid->len;
9817	if (port->target_devid)
9818		data_len += port->target_devid->len;
9819
9820	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9821	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9822	ctsio->kern_sg_entries = 0;
9823
9824	if (data_len < alloc_len) {
9825		ctsio->residual = alloc_len - data_len;
9826		ctsio->kern_data_len = data_len;
9827		ctsio->kern_total_len = data_len;
9828	} else {
9829		ctsio->residual = 0;
9830		ctsio->kern_data_len = alloc_len;
9831		ctsio->kern_total_len = alloc_len;
9832	}
9833	ctsio->kern_data_resid = 0;
9834	ctsio->kern_rel_offset = 0;
9835	ctsio->kern_sg_entries = 0;
9836
9837	/*
9838	 * The control device is always connected.  The disk device, on the
9839	 * other hand, may not be online all the time.
9840	 */
9841	if (lun != NULL)
9842		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9843				     lun->be_lun->lun_type;
9844	else
9845		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9846	devid_ptr->page_code = SVPD_DEVICE_ID;
9847	scsi_ulto2b(data_len - 4, devid_ptr->length);
9848
9849	if (port->port_type == CTL_PORT_FC)
9850		proto = SCSI_PROTO_FC << 4;
9851	else if (port->port_type == CTL_PORT_ISCSI)
9852		proto = SCSI_PROTO_ISCSI << 4;
9853	else
9854		proto = SCSI_PROTO_SPI << 4;
9855	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9856
9857	/*
9858	 * We're using a LUN association here.  i.e., this device ID is a
9859	 * per-LUN identifier.
9860	 */
9861	if (lun && lun->lun_devid) {
9862		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9863		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9864		    lun->lun_devid->len);
9865	}
9866
9867	/*
9868	 * This is for the WWPN which is a port association.
9869	 */
9870	if (port->port_devid) {
9871		memcpy(desc, port->port_devid->data, port->port_devid->len);
9872		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9873		    port->port_devid->len);
9874	}
9875
9876	/*
9877	 * This is for the Relative Target Port(type 4h) identifier
9878	 */
9879	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9880	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9881	    SVPD_ID_TYPE_RELTARG;
9882	desc->length = 4;
9883	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9884	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9885	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9886
9887	/*
9888	 * This is for the Target Port Group(type 5h) identifier
9889	 */
9890	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9891	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9892	    SVPD_ID_TYPE_TPORTGRP;
9893	desc->length = 4;
9894	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9895	    &desc->identifier[2]);
9896	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9897	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9898
9899	/*
9900	 * This is for the Target identifier
9901	 */
9902	if (port->target_devid) {
9903		memcpy(desc, port->target_devid->data, port->target_devid->len);
9904	}
9905
9906	ctl_set_success(ctsio);
9907	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9908	ctsio->be_move_done = ctl_config_move_done;
9909	ctl_datamove((union ctl_io *)ctsio);
9910	return (CTL_RETVAL_COMPLETE);
9911}
9912
9913static int
9914ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9915{
9916	struct ctl_softc *softc = control_softc;
9917	struct scsi_vpd_scsi_ports *sp;
9918	struct scsi_vpd_port_designation *pd;
9919	struct scsi_vpd_port_designation_cont *pdc;
9920	struct ctl_lun *lun;
9921	struct ctl_port *port;
9922	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9923	int num_target_port_groups;
9924
9925	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9926
9927	if (softc->is_single)
9928		num_target_port_groups = 1;
9929	else
9930		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
9931	num_target_ports = 0;
9932	iid_len = 0;
9933	id_len = 0;
9934	mtx_lock(&softc->ctl_lock);
9935	STAILQ_FOREACH(port, &softc->port_list, links) {
9936		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9937			continue;
9938		if (lun != NULL &&
9939		    ctl_map_lun_back(port->targ_port, lun->lun) >=
9940		    CTL_MAX_LUNS)
9941			continue;
9942		num_target_ports++;
9943		if (port->init_devid)
9944			iid_len += port->init_devid->len;
9945		if (port->port_devid)
9946			id_len += port->port_devid->len;
9947	}
9948	mtx_unlock(&softc->ctl_lock);
9949
9950	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
9951	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9952	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9953	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9954	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9955	ctsio->kern_sg_entries = 0;
9956
9957	if (data_len < alloc_len) {
9958		ctsio->residual = alloc_len - data_len;
9959		ctsio->kern_data_len = data_len;
9960		ctsio->kern_total_len = data_len;
9961	} else {
9962		ctsio->residual = 0;
9963		ctsio->kern_data_len = alloc_len;
9964		ctsio->kern_total_len = alloc_len;
9965	}
9966	ctsio->kern_data_resid = 0;
9967	ctsio->kern_rel_offset = 0;
9968	ctsio->kern_sg_entries = 0;
9969
9970	/*
9971	 * The control device is always connected.  The disk device, on the
9972	 * other hand, may not be online all the time.  Need to change this
9973	 * to figure out whether the disk device is actually online or not.
9974	 */
9975	if (lun != NULL)
9976		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9977				  lun->be_lun->lun_type;
9978	else
9979		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9980
9981	sp->page_code = SVPD_SCSI_PORTS;
9982	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9983	    sp->page_length);
9984	pd = &sp->design[0];
9985
9986	mtx_lock(&softc->ctl_lock);
9987	pg = softc->port_offset / CTL_MAX_PORTS;
9988	for (g = 0; g < num_target_port_groups; g++) {
9989		STAILQ_FOREACH(port, &softc->port_list, links) {
9990			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9991				continue;
9992			if (lun != NULL &&
9993			    ctl_map_lun_back(port->targ_port, lun->lun) >=
9994			    CTL_MAX_LUNS)
9995				continue;
9996			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
9997			scsi_ulto2b(p, pd->relative_port_id);
9998			if (port->init_devid && g == pg) {
9999				iid_len = port->init_devid->len;
10000				memcpy(pd->initiator_transportid,
10001				    port->init_devid->data, port->init_devid->len);
10002			} else
10003				iid_len = 0;
10004			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10005			pdc = (struct scsi_vpd_port_designation_cont *)
10006			    (&pd->initiator_transportid[iid_len]);
10007			if (port->port_devid && g == pg) {
10008				id_len = port->port_devid->len;
10009				memcpy(pdc->target_port_descriptors,
10010				    port->port_devid->data, port->port_devid->len);
10011			} else
10012				id_len = 0;
10013			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10014			pd = (struct scsi_vpd_port_designation *)
10015			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10016		}
10017	}
10018	mtx_unlock(&softc->ctl_lock);
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_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10029{
10030	struct scsi_vpd_block_limits *bl_ptr;
10031	struct ctl_lun *lun;
10032	int bs;
10033
10034	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10035
10036	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10037	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10038	ctsio->kern_sg_entries = 0;
10039
10040	if (sizeof(*bl_ptr) < alloc_len) {
10041		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10042		ctsio->kern_data_len = sizeof(*bl_ptr);
10043		ctsio->kern_total_len = sizeof(*bl_ptr);
10044	} else {
10045		ctsio->residual = 0;
10046		ctsio->kern_data_len = alloc_len;
10047		ctsio->kern_total_len = alloc_len;
10048	}
10049	ctsio->kern_data_resid = 0;
10050	ctsio->kern_rel_offset = 0;
10051	ctsio->kern_sg_entries = 0;
10052
10053	/*
10054	 * The control device is always connected.  The disk device, on the
10055	 * other hand, may not be online all the time.  Need to change this
10056	 * to figure out whether the disk device is actually online or not.
10057	 */
10058	if (lun != NULL)
10059		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10060				  lun->be_lun->lun_type;
10061	else
10062		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10063
10064	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10065	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10066	bl_ptr->max_cmp_write_len = 0xff;
10067	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10068	if (lun != NULL) {
10069		bs = lun->be_lun->blocksize;
10070		scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
10071		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10072			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10073			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10074			if (lun->be_lun->pblockexp != 0) {
10075				scsi_ulto4b((1 << lun->be_lun->pblockexp),
10076				    bl_ptr->opt_unmap_grain);
10077				scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff,
10078				    bl_ptr->unmap_grain_align);
10079			}
10080		}
10081		scsi_ulto4b(lun->be_lun->atomicblock,
10082		    bl_ptr->max_atomic_transfer_length);
10083		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10084		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10085	}
10086	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10087
10088	ctl_set_success(ctsio);
10089	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10090	ctsio->be_move_done = ctl_config_move_done;
10091	ctl_datamove((union ctl_io *)ctsio);
10092	return (CTL_RETVAL_COMPLETE);
10093}
10094
10095static int
10096ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10097{
10098	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10099	struct ctl_lun *lun;
10100	const char *value;
10101	u_int i;
10102
10103	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10104
10105	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10106	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10107	ctsio->kern_sg_entries = 0;
10108
10109	if (sizeof(*bdc_ptr) < alloc_len) {
10110		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10111		ctsio->kern_data_len = sizeof(*bdc_ptr);
10112		ctsio->kern_total_len = sizeof(*bdc_ptr);
10113	} else {
10114		ctsio->residual = 0;
10115		ctsio->kern_data_len = alloc_len;
10116		ctsio->kern_total_len = alloc_len;
10117	}
10118	ctsio->kern_data_resid = 0;
10119	ctsio->kern_rel_offset = 0;
10120	ctsio->kern_sg_entries = 0;
10121
10122	/*
10123	 * The control device is always connected.  The disk device, on the
10124	 * other hand, may not be online all the time.  Need to change this
10125	 * to figure out whether the disk device is actually online or not.
10126	 */
10127	if (lun != NULL)
10128		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10129				  lun->be_lun->lun_type;
10130	else
10131		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10132	bdc_ptr->page_code = SVPD_BDC;
10133	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10134	if (lun != NULL &&
10135	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10136		i = strtol(value, NULL, 0);
10137	else
10138		i = CTL_DEFAULT_ROTATION_RATE;
10139	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10140	if (lun != NULL &&
10141	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10142		i = strtol(value, NULL, 0);
10143	else
10144		i = 0;
10145	bdc_ptr->wab_wac_ff = (i & 0x0f);
10146	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10147
10148	ctl_set_success(ctsio);
10149	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10150	ctsio->be_move_done = ctl_config_move_done;
10151	ctl_datamove((union ctl_io *)ctsio);
10152	return (CTL_RETVAL_COMPLETE);
10153}
10154
10155static int
10156ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10157{
10158	struct scsi_vpd_logical_block_prov *lbp_ptr;
10159	struct ctl_lun *lun;
10160
10161	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10162
10163	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10164	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10165	ctsio->kern_sg_entries = 0;
10166
10167	if (sizeof(*lbp_ptr) < alloc_len) {
10168		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10169		ctsio->kern_data_len = sizeof(*lbp_ptr);
10170		ctsio->kern_total_len = sizeof(*lbp_ptr);
10171	} else {
10172		ctsio->residual = 0;
10173		ctsio->kern_data_len = alloc_len;
10174		ctsio->kern_total_len = alloc_len;
10175	}
10176	ctsio->kern_data_resid = 0;
10177	ctsio->kern_rel_offset = 0;
10178	ctsio->kern_sg_entries = 0;
10179
10180	/*
10181	 * The control device is always connected.  The disk device, on the
10182	 * other hand, may not be online all the time.  Need to change this
10183	 * to figure out whether the disk device is actually online or not.
10184	 */
10185	if (lun != NULL)
10186		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10187				  lun->be_lun->lun_type;
10188	else
10189		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10190
10191	lbp_ptr->page_code = SVPD_LBP;
10192	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10193	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10194		lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10195		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10196		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10197		lbp_ptr->prov_type = SVPD_LBP_THIN;
10198	}
10199
10200	ctl_set_success(ctsio);
10201	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10202	ctsio->be_move_done = ctl_config_move_done;
10203	ctl_datamove((union ctl_io *)ctsio);
10204	return (CTL_RETVAL_COMPLETE);
10205}
10206
10207static int
10208ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10209{
10210	struct scsi_inquiry *cdb;
10211	int alloc_len, retval;
10212
10213	cdb = (struct scsi_inquiry *)ctsio->cdb;
10214
10215	retval = CTL_RETVAL_COMPLETE;
10216
10217	alloc_len = scsi_2btoul(cdb->length);
10218
10219	switch (cdb->page_code) {
10220	case SVPD_SUPPORTED_PAGES:
10221		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10222		break;
10223	case SVPD_UNIT_SERIAL_NUMBER:
10224		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10225		break;
10226	case SVPD_DEVICE_ID:
10227		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10228		break;
10229	case SVPD_EXTENDED_INQUIRY_DATA:
10230		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10231		break;
10232	case SVPD_MODE_PAGE_POLICY:
10233		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10234		break;
10235	case SVPD_SCSI_PORTS:
10236		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10237		break;
10238	case SVPD_SCSI_TPC:
10239		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10240		break;
10241	case SVPD_BLOCK_LIMITS:
10242		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10243		break;
10244	case SVPD_BDC:
10245		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10246		break;
10247	case SVPD_LBP:
10248		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10249		break;
10250	default:
10251		ctl_set_invalid_field(ctsio,
10252				      /*sks_valid*/ 1,
10253				      /*command*/ 1,
10254				      /*field*/ 2,
10255				      /*bit_valid*/ 0,
10256				      /*bit*/ 0);
10257		ctl_done((union ctl_io *)ctsio);
10258		retval = CTL_RETVAL_COMPLETE;
10259		break;
10260	}
10261
10262	return (retval);
10263}
10264
10265static int
10266ctl_inquiry_std(struct ctl_scsiio *ctsio)
10267{
10268	struct scsi_inquiry_data *inq_ptr;
10269	struct scsi_inquiry *cdb;
10270	struct ctl_softc *ctl_softc;
10271	struct ctl_lun *lun;
10272	char *val;
10273	uint32_t alloc_len, data_len;
10274	ctl_port_type port_type;
10275
10276	ctl_softc = control_softc;
10277
10278	/*
10279	 * Figure out whether we're talking to a Fibre Channel port or not.
10280	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10281	 * SCSI front ends.
10282	 */
10283	port_type = ctl_softc->ctl_ports[
10284	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10285	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10286		port_type = CTL_PORT_SCSI;
10287
10288	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10289	cdb = (struct scsi_inquiry *)ctsio->cdb;
10290	alloc_len = scsi_2btoul(cdb->length);
10291
10292	/*
10293	 * We malloc the full inquiry data size here and fill it
10294	 * in.  If the user only asks for less, we'll give him
10295	 * that much.
10296	 */
10297	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10298	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10299	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10300	ctsio->kern_sg_entries = 0;
10301	ctsio->kern_data_resid = 0;
10302	ctsio->kern_rel_offset = 0;
10303
10304	if (data_len < alloc_len) {
10305		ctsio->residual = alloc_len - data_len;
10306		ctsio->kern_data_len = data_len;
10307		ctsio->kern_total_len = data_len;
10308	} else {
10309		ctsio->residual = 0;
10310		ctsio->kern_data_len = alloc_len;
10311		ctsio->kern_total_len = alloc_len;
10312	}
10313
10314	/*
10315	 * If we have a LUN configured, report it as connected.  Otherwise,
10316	 * report that it is offline or no device is supported, depending
10317	 * on the value of inquiry_pq_no_lun.
10318	 *
10319	 * According to the spec (SPC-4 r34), the peripheral qualifier
10320	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10321	 *
10322	 * "A peripheral device having the specified peripheral device type
10323	 * is not connected to this logical unit. However, the device
10324	 * server is capable of supporting the specified peripheral device
10325	 * type on this logical unit."
10326	 *
10327	 * According to the same spec, the peripheral qualifier
10328	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10329	 *
10330	 * "The device server is not capable of supporting a peripheral
10331	 * device on this logical unit. For this peripheral qualifier the
10332	 * peripheral device type shall be set to 1Fh. All other peripheral
10333	 * device type values are reserved for this peripheral qualifier."
10334	 *
10335	 * Given the text, it would seem that we probably want to report that
10336	 * the LUN is offline here.  There is no LUN connected, but we can
10337	 * support a LUN at the given LUN number.
10338	 *
10339	 * In the real world, though, it sounds like things are a little
10340	 * different:
10341	 *
10342	 * - Linux, when presented with a LUN with the offline peripheral
10343	 *   qualifier, will create an sg driver instance for it.  So when
10344	 *   you attach it to CTL, you wind up with a ton of sg driver
10345	 *   instances.  (One for every LUN that Linux bothered to probe.)
10346	 *   Linux does this despite the fact that it issues a REPORT LUNs
10347	 *   to LUN 0 to get the inventory of supported LUNs.
10348	 *
10349	 * - There is other anecdotal evidence (from Emulex folks) about
10350	 *   arrays that use the offline peripheral qualifier for LUNs that
10351	 *   are on the "passive" path in an active/passive array.
10352	 *
10353	 * So the solution is provide a hopefully reasonable default
10354	 * (return bad/no LUN) and allow the user to change the behavior
10355	 * with a tunable/sysctl variable.
10356	 */
10357	if (lun != NULL)
10358		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10359				  lun->be_lun->lun_type;
10360	else if (ctl_softc->inquiry_pq_no_lun == 0)
10361		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10362	else
10363		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10364
10365	/* RMB in byte 2 is 0 */
10366	inq_ptr->version = SCSI_REV_SPC4;
10367
10368	/*
10369	 * According to SAM-3, even if a device only supports a single
10370	 * level of LUN addressing, it should still set the HISUP bit:
10371	 *
10372	 * 4.9.1 Logical unit numbers overview
10373	 *
10374	 * All logical unit number formats described in this standard are
10375	 * hierarchical in structure even when only a single level in that
10376	 * hierarchy is used. The HISUP bit shall be set to one in the
10377	 * standard INQUIRY data (see SPC-2) when any logical unit number
10378	 * format described in this standard is used.  Non-hierarchical
10379	 * formats are outside the scope of this standard.
10380	 *
10381	 * Therefore we set the HiSup bit here.
10382	 *
10383	 * The reponse format is 2, per SPC-3.
10384	 */
10385	inq_ptr->response_format = SID_HiSup | 2;
10386
10387	inq_ptr->additional_length = data_len -
10388	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10389	CTL_DEBUG_PRINT(("additional_length = %d\n",
10390			 inq_ptr->additional_length));
10391
10392	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10393	/* 16 bit addressing */
10394	if (port_type == CTL_PORT_SCSI)
10395		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10396	/* XXX set the SID_MultiP bit here if we're actually going to
10397	   respond on multiple ports */
10398	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10399
10400	/* 16 bit data bus, synchronous transfers */
10401	if (port_type == CTL_PORT_SCSI)
10402		inq_ptr->flags = SID_WBus16 | SID_Sync;
10403	/*
10404	 * XXX KDM do we want to support tagged queueing on the control
10405	 * device at all?
10406	 */
10407	if ((lun == NULL)
10408	 || (lun->be_lun->lun_type != T_PROCESSOR))
10409		inq_ptr->flags |= SID_CmdQue;
10410	/*
10411	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10412	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10413	 * name and 4 bytes for the revision.
10414	 */
10415	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10416	    "vendor")) == NULL) {
10417		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10418	} else {
10419		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10420		strncpy(inq_ptr->vendor, val,
10421		    min(sizeof(inq_ptr->vendor), strlen(val)));
10422	}
10423	if (lun == NULL) {
10424		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10425		    sizeof(inq_ptr->product));
10426	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10427		switch (lun->be_lun->lun_type) {
10428		case T_DIRECT:
10429			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10430			    sizeof(inq_ptr->product));
10431			break;
10432		case T_PROCESSOR:
10433			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10434			    sizeof(inq_ptr->product));
10435			break;
10436		default:
10437			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10438			    sizeof(inq_ptr->product));
10439			break;
10440		}
10441	} else {
10442		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10443		strncpy(inq_ptr->product, val,
10444		    min(sizeof(inq_ptr->product), strlen(val)));
10445	}
10446
10447	/*
10448	 * XXX make this a macro somewhere so it automatically gets
10449	 * incremented when we make changes.
10450	 */
10451	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10452	    "revision")) == NULL) {
10453		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10454	} else {
10455		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10456		strncpy(inq_ptr->revision, val,
10457		    min(sizeof(inq_ptr->revision), strlen(val)));
10458	}
10459
10460	/*
10461	 * For parallel SCSI, we support double transition and single
10462	 * transition clocking.  We also support QAS (Quick Arbitration
10463	 * and Selection) and Information Unit transfers on both the
10464	 * control and array devices.
10465	 */
10466	if (port_type == CTL_PORT_SCSI)
10467		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10468				    SID_SPI_IUS;
10469
10470	/* SAM-5 (no version claimed) */
10471	scsi_ulto2b(0x00A0, inq_ptr->version1);
10472	/* SPC-4 (no version claimed) */
10473	scsi_ulto2b(0x0460, inq_ptr->version2);
10474	if (port_type == CTL_PORT_FC) {
10475		/* FCP-2 ANSI INCITS.350:2003 */
10476		scsi_ulto2b(0x0917, inq_ptr->version3);
10477	} else if (port_type == CTL_PORT_SCSI) {
10478		/* SPI-4 ANSI INCITS.362:200x */
10479		scsi_ulto2b(0x0B56, inq_ptr->version3);
10480	} else if (port_type == CTL_PORT_ISCSI) {
10481		/* iSCSI (no version claimed) */
10482		scsi_ulto2b(0x0960, inq_ptr->version3);
10483	} else if (port_type == CTL_PORT_SAS) {
10484		/* SAS (no version claimed) */
10485		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10486	}
10487
10488	if (lun == NULL) {
10489		/* SBC-4 (no version claimed) */
10490		scsi_ulto2b(0x0600, inq_ptr->version4);
10491	} else {
10492		switch (lun->be_lun->lun_type) {
10493		case T_DIRECT:
10494			/* SBC-4 (no version claimed) */
10495			scsi_ulto2b(0x0600, inq_ptr->version4);
10496			break;
10497		case T_PROCESSOR:
10498		default:
10499			break;
10500		}
10501	}
10502
10503	ctl_set_success(ctsio);
10504	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10505	ctsio->be_move_done = ctl_config_move_done;
10506	ctl_datamove((union ctl_io *)ctsio);
10507	return (CTL_RETVAL_COMPLETE);
10508}
10509
10510int
10511ctl_inquiry(struct ctl_scsiio *ctsio)
10512{
10513	struct scsi_inquiry *cdb;
10514	int retval;
10515
10516	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10517
10518	cdb = (struct scsi_inquiry *)ctsio->cdb;
10519	if (cdb->byte2 & SI_EVPD)
10520		retval = ctl_inquiry_evpd(ctsio);
10521	else if (cdb->page_code == 0)
10522		retval = ctl_inquiry_std(ctsio);
10523	else {
10524		ctl_set_invalid_field(ctsio,
10525				      /*sks_valid*/ 1,
10526				      /*command*/ 1,
10527				      /*field*/ 2,
10528				      /*bit_valid*/ 0,
10529				      /*bit*/ 0);
10530		ctl_done((union ctl_io *)ctsio);
10531		return (CTL_RETVAL_COMPLETE);
10532	}
10533
10534	return (retval);
10535}
10536
10537/*
10538 * For known CDB types, parse the LBA and length.
10539 */
10540static int
10541ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10542{
10543	if (io->io_hdr.io_type != CTL_IO_SCSI)
10544		return (1);
10545
10546	switch (io->scsiio.cdb[0]) {
10547	case COMPARE_AND_WRITE: {
10548		struct scsi_compare_and_write *cdb;
10549
10550		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10551
10552		*lba = scsi_8btou64(cdb->addr);
10553		*len = cdb->length;
10554		break;
10555	}
10556	case READ_6:
10557	case WRITE_6: {
10558		struct scsi_rw_6 *cdb;
10559
10560		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10561
10562		*lba = scsi_3btoul(cdb->addr);
10563		/* only 5 bits are valid in the most significant address byte */
10564		*lba &= 0x1fffff;
10565		*len = cdb->length;
10566		break;
10567	}
10568	case READ_10:
10569	case WRITE_10: {
10570		struct scsi_rw_10 *cdb;
10571
10572		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10573
10574		*lba = scsi_4btoul(cdb->addr);
10575		*len = scsi_2btoul(cdb->length);
10576		break;
10577	}
10578	case WRITE_VERIFY_10: {
10579		struct scsi_write_verify_10 *cdb;
10580
10581		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10582
10583		*lba = scsi_4btoul(cdb->addr);
10584		*len = scsi_2btoul(cdb->length);
10585		break;
10586	}
10587	case READ_12:
10588	case WRITE_12: {
10589		struct scsi_rw_12 *cdb;
10590
10591		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10592
10593		*lba = scsi_4btoul(cdb->addr);
10594		*len = scsi_4btoul(cdb->length);
10595		break;
10596	}
10597	case WRITE_VERIFY_12: {
10598		struct scsi_write_verify_12 *cdb;
10599
10600		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10601
10602		*lba = scsi_4btoul(cdb->addr);
10603		*len = scsi_4btoul(cdb->length);
10604		break;
10605	}
10606	case READ_16:
10607	case WRITE_16:
10608	case WRITE_ATOMIC_16: {
10609		struct scsi_rw_16 *cdb;
10610
10611		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10612
10613		*lba = scsi_8btou64(cdb->addr);
10614		*len = scsi_4btoul(cdb->length);
10615		break;
10616	}
10617	case WRITE_VERIFY_16: {
10618		struct scsi_write_verify_16 *cdb;
10619
10620		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10621
10622		*lba = scsi_8btou64(cdb->addr);
10623		*len = scsi_4btoul(cdb->length);
10624		break;
10625	}
10626	case WRITE_SAME_10: {
10627		struct scsi_write_same_10 *cdb;
10628
10629		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10630
10631		*lba = scsi_4btoul(cdb->addr);
10632		*len = scsi_2btoul(cdb->length);
10633		break;
10634	}
10635	case WRITE_SAME_16: {
10636		struct scsi_write_same_16 *cdb;
10637
10638		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10639
10640		*lba = scsi_8btou64(cdb->addr);
10641		*len = scsi_4btoul(cdb->length);
10642		break;
10643	}
10644	case VERIFY_10: {
10645		struct scsi_verify_10 *cdb;
10646
10647		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10648
10649		*lba = scsi_4btoul(cdb->addr);
10650		*len = scsi_2btoul(cdb->length);
10651		break;
10652	}
10653	case VERIFY_12: {
10654		struct scsi_verify_12 *cdb;
10655
10656		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10657
10658		*lba = scsi_4btoul(cdb->addr);
10659		*len = scsi_4btoul(cdb->length);
10660		break;
10661	}
10662	case VERIFY_16: {
10663		struct scsi_verify_16 *cdb;
10664
10665		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10666
10667		*lba = scsi_8btou64(cdb->addr);
10668		*len = scsi_4btoul(cdb->length);
10669		break;
10670	}
10671	case UNMAP: {
10672		*lba = 0;
10673		*len = UINT64_MAX;
10674		break;
10675	}
10676	default:
10677		return (1);
10678		break; /* NOTREACHED */
10679	}
10680
10681	return (0);
10682}
10683
10684static ctl_action
10685ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2)
10686{
10687	uint64_t endlba1, endlba2;
10688
10689	endlba1 = lba1 + len1 - 1;
10690	endlba2 = lba2 + len2 - 1;
10691
10692	if ((endlba1 < lba2)
10693	 || (endlba2 < lba1))
10694		return (CTL_ACTION_PASS);
10695	else
10696		return (CTL_ACTION_BLOCK);
10697}
10698
10699static int
10700ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10701{
10702	struct ctl_ptr_len_flags *ptrlen;
10703	struct scsi_unmap_desc *buf, *end, *range;
10704	uint64_t lba;
10705	uint32_t len;
10706
10707	/* If not UNMAP -- go other way. */
10708	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10709	    io->scsiio.cdb[0] != UNMAP)
10710		return (CTL_ACTION_ERROR);
10711
10712	/* If UNMAP without data -- block and wait for data. */
10713	ptrlen = (struct ctl_ptr_len_flags *)
10714	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10715	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10716	    ptrlen->ptr == NULL)
10717		return (CTL_ACTION_BLOCK);
10718
10719	/* UNMAP with data -- check for collision. */
10720	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10721	end = buf + ptrlen->len / sizeof(*buf);
10722	for (range = buf; range < end; range++) {
10723		lba = scsi_8btou64(range->lba);
10724		len = scsi_4btoul(range->length);
10725		if ((lba < lba2 + len2) && (lba + len > lba2))
10726			return (CTL_ACTION_BLOCK);
10727	}
10728	return (CTL_ACTION_PASS);
10729}
10730
10731static ctl_action
10732ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10733{
10734	uint64_t lba1, lba2;
10735	uint64_t len1, len2;
10736	int retval;
10737
10738	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10739		return (CTL_ACTION_ERROR);
10740
10741	retval = ctl_extent_check_unmap(io2, lba1, len1);
10742	if (retval != CTL_ACTION_ERROR)
10743		return (retval);
10744
10745	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10746		return (CTL_ACTION_ERROR);
10747
10748	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10749}
10750
10751static ctl_action
10752ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10753    union ctl_io *ooa_io)
10754{
10755	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10756	ctl_serialize_action *serialize_row;
10757
10758	/*
10759	 * The initiator attempted multiple untagged commands at the same
10760	 * time.  Can't do that.
10761	 */
10762	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10763	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10764	 && ((pending_io->io_hdr.nexus.targ_port ==
10765	      ooa_io->io_hdr.nexus.targ_port)
10766	  && (pending_io->io_hdr.nexus.initid.id ==
10767	      ooa_io->io_hdr.nexus.initid.id))
10768	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10769		return (CTL_ACTION_OVERLAP);
10770
10771	/*
10772	 * The initiator attempted to send multiple tagged commands with
10773	 * the same ID.  (It's fine if different initiators have the same
10774	 * tag ID.)
10775	 *
10776	 * Even if all of those conditions are true, we don't kill the I/O
10777	 * if the command ahead of us has been aborted.  We won't end up
10778	 * sending it to the FETD, and it's perfectly legal to resend a
10779	 * command with the same tag number as long as the previous
10780	 * instance of this tag number has been aborted somehow.
10781	 */
10782	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10783	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10784	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10785	 && ((pending_io->io_hdr.nexus.targ_port ==
10786	      ooa_io->io_hdr.nexus.targ_port)
10787	  && (pending_io->io_hdr.nexus.initid.id ==
10788	      ooa_io->io_hdr.nexus.initid.id))
10789	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10790		return (CTL_ACTION_OVERLAP_TAG);
10791
10792	/*
10793	 * If we get a head of queue tag, SAM-3 says that we should
10794	 * immediately execute it.
10795	 *
10796	 * What happens if this command would normally block for some other
10797	 * reason?  e.g. a request sense with a head of queue tag
10798	 * immediately after a write.  Normally that would block, but this
10799	 * will result in its getting executed immediately...
10800	 *
10801	 * We currently return "pass" instead of "skip", so we'll end up
10802	 * going through the rest of the queue to check for overlapped tags.
10803	 *
10804	 * XXX KDM check for other types of blockage first??
10805	 */
10806	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10807		return (CTL_ACTION_PASS);
10808
10809	/*
10810	 * Ordered tags have to block until all items ahead of them
10811	 * have completed.  If we get called with an ordered tag, we always
10812	 * block, if something else is ahead of us in the queue.
10813	 */
10814	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10815		return (CTL_ACTION_BLOCK);
10816
10817	/*
10818	 * Simple tags get blocked until all head of queue and ordered tags
10819	 * ahead of them have completed.  I'm lumping untagged commands in
10820	 * with simple tags here.  XXX KDM is that the right thing to do?
10821	 */
10822	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10823	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10824	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10825	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10826		return (CTL_ACTION_BLOCK);
10827
10828	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10829	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10830
10831	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10832
10833	switch (serialize_row[pending_entry->seridx]) {
10834	case CTL_SER_BLOCK:
10835		return (CTL_ACTION_BLOCK);
10836	case CTL_SER_EXTENT:
10837		return (ctl_extent_check(pending_io, ooa_io));
10838	case CTL_SER_EXTENTOPT:
10839		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10840		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10841			return (ctl_extent_check(pending_io, ooa_io));
10842		/* FALLTHROUGH */
10843	case CTL_SER_PASS:
10844		return (CTL_ACTION_PASS);
10845	case CTL_SER_BLOCKOPT:
10846		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10847		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10848			return (CTL_ACTION_BLOCK);
10849		return (CTL_ACTION_PASS);
10850	case CTL_SER_SKIP:
10851		return (CTL_ACTION_SKIP);
10852	default:
10853		panic("invalid serialization value %d",
10854		      serialize_row[pending_entry->seridx]);
10855	}
10856
10857	return (CTL_ACTION_ERROR);
10858}
10859
10860/*
10861 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10862 * Assumptions:
10863 * - pending_io is generally either incoming, or on the blocked queue
10864 * - starting I/O is the I/O we want to start the check with.
10865 */
10866static ctl_action
10867ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10868	      union ctl_io *starting_io)
10869{
10870	union ctl_io *ooa_io;
10871	ctl_action action;
10872
10873	mtx_assert(&lun->lun_lock, MA_OWNED);
10874
10875	/*
10876	 * Run back along the OOA queue, starting with the current
10877	 * blocked I/O and going through every I/O before it on the
10878	 * queue.  If starting_io is NULL, we'll just end up returning
10879	 * CTL_ACTION_PASS.
10880	 */
10881	for (ooa_io = starting_io; ooa_io != NULL;
10882	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10883	     ooa_links)){
10884
10885		/*
10886		 * This routine just checks to see whether
10887		 * cur_blocked is blocked by ooa_io, which is ahead
10888		 * of it in the queue.  It doesn't queue/dequeue
10889		 * cur_blocked.
10890		 */
10891		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10892		switch (action) {
10893		case CTL_ACTION_BLOCK:
10894		case CTL_ACTION_OVERLAP:
10895		case CTL_ACTION_OVERLAP_TAG:
10896		case CTL_ACTION_SKIP:
10897		case CTL_ACTION_ERROR:
10898			return (action);
10899			break; /* NOTREACHED */
10900		case CTL_ACTION_PASS:
10901			break;
10902		default:
10903			panic("invalid action %d", action);
10904			break;  /* NOTREACHED */
10905		}
10906	}
10907
10908	return (CTL_ACTION_PASS);
10909}
10910
10911/*
10912 * Assumptions:
10913 * - An I/O has just completed, and has been removed from the per-LUN OOA
10914 *   queue, so some items on the blocked queue may now be unblocked.
10915 */
10916static int
10917ctl_check_blocked(struct ctl_lun *lun)
10918{
10919	union ctl_io *cur_blocked, *next_blocked;
10920
10921	mtx_assert(&lun->lun_lock, MA_OWNED);
10922
10923	/*
10924	 * Run forward from the head of the blocked queue, checking each
10925	 * entry against the I/Os prior to it on the OOA queue to see if
10926	 * there is still any blockage.
10927	 *
10928	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10929	 * with our removing a variable on it while it is traversing the
10930	 * list.
10931	 */
10932	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10933	     cur_blocked != NULL; cur_blocked = next_blocked) {
10934		union ctl_io *prev_ooa;
10935		ctl_action action;
10936
10937		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10938							  blocked_links);
10939
10940		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10941						      ctl_ooaq, ooa_links);
10942
10943		/*
10944		 * If cur_blocked happens to be the first item in the OOA
10945		 * queue now, prev_ooa will be NULL, and the action
10946		 * returned will just be CTL_ACTION_PASS.
10947		 */
10948		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10949
10950		switch (action) {
10951		case CTL_ACTION_BLOCK:
10952			/* Nothing to do here, still blocked */
10953			break;
10954		case CTL_ACTION_OVERLAP:
10955		case CTL_ACTION_OVERLAP_TAG:
10956			/*
10957			 * This shouldn't happen!  In theory we've already
10958			 * checked this command for overlap...
10959			 */
10960			break;
10961		case CTL_ACTION_PASS:
10962		case CTL_ACTION_SKIP: {
10963			struct ctl_softc *softc;
10964			const struct ctl_cmd_entry *entry;
10965			uint32_t initidx;
10966			int isc_retval;
10967
10968			/*
10969			 * The skip case shouldn't happen, this transaction
10970			 * should have never made it onto the blocked queue.
10971			 */
10972			/*
10973			 * This I/O is no longer blocked, we can remove it
10974			 * from the blocked queue.  Since this is a TAILQ
10975			 * (doubly linked list), we can do O(1) removals
10976			 * from any place on the list.
10977			 */
10978			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10979				     blocked_links);
10980			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10981
10982			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10983				/*
10984				 * Need to send IO back to original side to
10985				 * run
10986				 */
10987				union ctl_ha_msg msg_info;
10988
10989				msg_info.hdr.original_sc =
10990					cur_blocked->io_hdr.original_sc;
10991				msg_info.hdr.serializing_sc = cur_blocked;
10992				msg_info.hdr.msg_type = CTL_MSG_R2R;
10993				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10994				     &msg_info, sizeof(msg_info), 0)) >
10995				     CTL_HA_STATUS_SUCCESS) {
10996					printf("CTL:Check Blocked error from "
10997					       "ctl_ha_msg_send %d\n",
10998					       isc_retval);
10999				}
11000				break;
11001			}
11002			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11003			softc = control_softc;
11004
11005			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
11006
11007			/*
11008			 * Check this I/O for LUN state changes that may
11009			 * have happened while this command was blocked.
11010			 * The LUN state may have been changed by a command
11011			 * ahead of us in the queue, so we need to re-check
11012			 * for any states that can be caused by SCSI
11013			 * commands.
11014			 */
11015			if (ctl_scsiio_lun_check(softc, lun, entry,
11016						 &cur_blocked->scsiio) == 0) {
11017				cur_blocked->io_hdr.flags |=
11018				                      CTL_FLAG_IS_WAS_ON_RTR;
11019				ctl_enqueue_rtr(cur_blocked);
11020			} else
11021				ctl_done(cur_blocked);
11022			break;
11023		}
11024		default:
11025			/*
11026			 * This probably shouldn't happen -- we shouldn't
11027			 * get CTL_ACTION_ERROR, or anything else.
11028			 */
11029			break;
11030		}
11031	}
11032
11033	return (CTL_RETVAL_COMPLETE);
11034}
11035
11036/*
11037 * This routine (with one exception) checks LUN flags that can be set by
11038 * commands ahead of us in the OOA queue.  These flags have to be checked
11039 * when a command initially comes in, and when we pull a command off the
11040 * blocked queue and are preparing to execute it.  The reason we have to
11041 * check these flags for commands on the blocked queue is that the LUN
11042 * state may have been changed by a command ahead of us while we're on the
11043 * blocked queue.
11044 *
11045 * Ordering is somewhat important with these checks, so please pay
11046 * careful attention to the placement of any new checks.
11047 */
11048static int
11049ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11050    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11051{
11052	int retval;
11053	uint32_t residx;
11054
11055	retval = 0;
11056
11057	mtx_assert(&lun->lun_lock, MA_OWNED);
11058
11059	/*
11060	 * If this shelf is a secondary shelf controller, we have to reject
11061	 * any media access commands.
11062	 */
11063	if ((ctl_softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
11064	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
11065		ctl_set_lun_standby(ctsio);
11066		retval = 1;
11067		goto bailout;
11068	}
11069
11070	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11071		if (lun->flags & CTL_LUN_READONLY) {
11072			ctl_set_sense(ctsio, /*current_error*/ 1,
11073			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11074			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11075			retval = 1;
11076			goto bailout;
11077		}
11078		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11079		    .eca_and_aen & SCP_SWP) != 0) {
11080			ctl_set_sense(ctsio, /*current_error*/ 1,
11081			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11082			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11083			retval = 1;
11084			goto bailout;
11085		}
11086	}
11087
11088	/*
11089	 * Check for a reservation conflict.  If this command isn't allowed
11090	 * even on reserved LUNs, and if this initiator isn't the one who
11091	 * reserved us, reject the command with a reservation conflict.
11092	 */
11093	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11094	if ((lun->flags & CTL_LUN_RESERVED)
11095	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11096		if (lun->res_idx != residx) {
11097			ctl_set_reservation_conflict(ctsio);
11098			retval = 1;
11099			goto bailout;
11100		}
11101	}
11102
11103	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11104	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11105		/* No reservation or command is allowed. */;
11106	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11107	    (lun->res_type == SPR_TYPE_WR_EX ||
11108	     lun->res_type == SPR_TYPE_WR_EX_RO ||
11109	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
11110		/* The command is allowed for Write Exclusive resv. */;
11111	} else {
11112		/*
11113		 * if we aren't registered or it's a res holder type
11114		 * reservation and this isn't the res holder then set a
11115		 * conflict.
11116		 */
11117		if (lun->pr_keys[residx] == 0
11118		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11119			ctl_set_reservation_conflict(ctsio);
11120			retval = 1;
11121			goto bailout;
11122		}
11123
11124	}
11125
11126	if ((lun->flags & CTL_LUN_OFFLINE)
11127	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11128		ctl_set_lun_not_ready(ctsio);
11129		retval = 1;
11130		goto bailout;
11131	}
11132
11133	/*
11134	 * If the LUN is stopped, see if this particular command is allowed
11135	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11136	 */
11137	if ((lun->flags & CTL_LUN_STOPPED)
11138	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11139		/* "Logical unit not ready, initializing cmd. required" */
11140		ctl_set_lun_stopped(ctsio);
11141		retval = 1;
11142		goto bailout;
11143	}
11144
11145	if ((lun->flags & CTL_LUN_INOPERABLE)
11146	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11147		/* "Medium format corrupted" */
11148		ctl_set_medium_format_corrupted(ctsio);
11149		retval = 1;
11150		goto bailout;
11151	}
11152
11153bailout:
11154	return (retval);
11155
11156}
11157
11158static void
11159ctl_failover_io(union ctl_io *io, int have_lock)
11160{
11161	ctl_set_busy(&io->scsiio);
11162	ctl_done(io);
11163}
11164
11165static void
11166ctl_failover(void)
11167{
11168	struct ctl_lun *lun;
11169	struct ctl_softc *ctl_softc;
11170	union ctl_io *next_io, *pending_io;
11171	union ctl_io *io;
11172	int lun_idx;
11173	int i;
11174
11175	ctl_softc = control_softc;
11176
11177	mtx_lock(&ctl_softc->ctl_lock);
11178	/*
11179	 * Remove any cmds from the other SC from the rtr queue.  These
11180	 * will obviously only be for LUNs for which we're the primary.
11181	 * We can't send status or get/send data for these commands.
11182	 * Since they haven't been executed yet, we can just remove them.
11183	 * We'll either abort them or delete them below, depending on
11184	 * which HA mode we're in.
11185	 */
11186#ifdef notyet
11187	mtx_lock(&ctl_softc->queue_lock);
11188	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11189	     io != NULL; io = next_io) {
11190		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11191		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11192			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11193				      ctl_io_hdr, links);
11194	}
11195	mtx_unlock(&ctl_softc->queue_lock);
11196#endif
11197
11198	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11199		lun = ctl_softc->ctl_luns[lun_idx];
11200		if (lun==NULL)
11201			continue;
11202
11203		/*
11204		 * Processor LUNs are primary on both sides.
11205		 * XXX will this always be true?
11206		 */
11207		if (lun->be_lun->lun_type == T_PROCESSOR)
11208			continue;
11209
11210		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11211		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11212			printf("FAILOVER: primary lun %d\n", lun_idx);
11213		        /*
11214			 * Remove all commands from the other SC. First from the
11215			 * blocked queue then from the ooa queue. Once we have
11216			 * removed them. Call ctl_check_blocked to see if there
11217			 * is anything that can run.
11218			 */
11219			for (io = (union ctl_io *)TAILQ_FIRST(
11220			     &lun->blocked_queue); io != NULL; io = next_io) {
11221
11222		        	next_io = (union ctl_io *)TAILQ_NEXT(
11223				    &io->io_hdr, blocked_links);
11224
11225				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11226					TAILQ_REMOVE(&lun->blocked_queue,
11227						     &io->io_hdr,blocked_links);
11228					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11229					TAILQ_REMOVE(&lun->ooa_queue,
11230						     &io->io_hdr, ooa_links);
11231
11232					ctl_free_io(io);
11233				}
11234			}
11235
11236			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11237	     		     io != NULL; io = next_io) {
11238
11239		        	next_io = (union ctl_io *)TAILQ_NEXT(
11240				    &io->io_hdr, ooa_links);
11241
11242				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11243
11244					TAILQ_REMOVE(&lun->ooa_queue,
11245						&io->io_hdr,
11246					     	ooa_links);
11247
11248					ctl_free_io(io);
11249				}
11250			}
11251			ctl_check_blocked(lun);
11252		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11253			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11254
11255			printf("FAILOVER: primary lun %d\n", lun_idx);
11256			/*
11257			 * Abort all commands from the other SC.  We can't
11258			 * send status back for them now.  These should get
11259			 * cleaned up when they are completed or come out
11260			 * for a datamove operation.
11261			 */
11262			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11263	     		     io != NULL; io = next_io) {
11264		        	next_io = (union ctl_io *)TAILQ_NEXT(
11265					&io->io_hdr, ooa_links);
11266
11267				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11268					io->io_hdr.flags |= CTL_FLAG_ABORT;
11269			}
11270		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11271			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11272
11273			printf("FAILOVER: secondary lun %d\n", lun_idx);
11274
11275			lun->flags |= CTL_LUN_PRIMARY_SC;
11276
11277			/*
11278			 * We send all I/O that was sent to this controller
11279			 * and redirected to the other side back with
11280			 * busy status, and have the initiator retry it.
11281			 * Figuring out how much data has been transferred,
11282			 * etc. and picking up where we left off would be
11283			 * very tricky.
11284			 *
11285			 * XXX KDM need to remove I/O from the blocked
11286			 * queue as well!
11287			 */
11288			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11289			     &lun->ooa_queue); pending_io != NULL;
11290			     pending_io = next_io) {
11291
11292				next_io =  (union ctl_io *)TAILQ_NEXT(
11293					&pending_io->io_hdr, ooa_links);
11294
11295				pending_io->io_hdr.flags &=
11296					~CTL_FLAG_SENT_2OTHER_SC;
11297
11298				if (pending_io->io_hdr.flags &
11299				    CTL_FLAG_IO_ACTIVE) {
11300					pending_io->io_hdr.flags |=
11301						CTL_FLAG_FAILOVER;
11302				} else {
11303					ctl_set_busy(&pending_io->scsiio);
11304					ctl_done(pending_io);
11305				}
11306			}
11307
11308			/*
11309			 * Build Unit Attention
11310			 */
11311			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11312				lun->pending_ua[i] |=
11313				                     CTL_UA_ASYM_ACC_CHANGE;
11314			}
11315		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11316			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11317			printf("FAILOVER: secondary lun %d\n", lun_idx);
11318			/*
11319			 * if the first io on the OOA is not on the RtR queue
11320			 * add it.
11321			 */
11322			lun->flags |= CTL_LUN_PRIMARY_SC;
11323
11324			pending_io = (union ctl_io *)TAILQ_FIRST(
11325			    &lun->ooa_queue);
11326			if (pending_io==NULL) {
11327				printf("Nothing on OOA queue\n");
11328				continue;
11329			}
11330
11331			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11332			if ((pending_io->io_hdr.flags &
11333			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11334				pending_io->io_hdr.flags |=
11335				    CTL_FLAG_IS_WAS_ON_RTR;
11336				ctl_enqueue_rtr(pending_io);
11337			}
11338#if 0
11339			else
11340			{
11341				printf("Tag 0x%04x is running\n",
11342				      pending_io->scsiio.tag_num);
11343			}
11344#endif
11345
11346			next_io = (union ctl_io *)TAILQ_NEXT(
11347			    &pending_io->io_hdr, ooa_links);
11348			for (pending_io=next_io; pending_io != NULL;
11349			     pending_io = next_io) {
11350				pending_io->io_hdr.flags &=
11351				    ~CTL_FLAG_SENT_2OTHER_SC;
11352				next_io = (union ctl_io *)TAILQ_NEXT(
11353					&pending_io->io_hdr, ooa_links);
11354				if (pending_io->io_hdr.flags &
11355				    CTL_FLAG_IS_WAS_ON_RTR) {
11356#if 0
11357				        printf("Tag 0x%04x is running\n",
11358				      		pending_io->scsiio.tag_num);
11359#endif
11360					continue;
11361				}
11362
11363				switch (ctl_check_ooa(lun, pending_io,
11364			            (union ctl_io *)TAILQ_PREV(
11365				    &pending_io->io_hdr, ctl_ooaq,
11366				    ooa_links))) {
11367
11368				case CTL_ACTION_BLOCK:
11369					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11370							  &pending_io->io_hdr,
11371							  blocked_links);
11372					pending_io->io_hdr.flags |=
11373					    CTL_FLAG_BLOCKED;
11374					break;
11375				case CTL_ACTION_PASS:
11376				case CTL_ACTION_SKIP:
11377					pending_io->io_hdr.flags |=
11378					    CTL_FLAG_IS_WAS_ON_RTR;
11379					ctl_enqueue_rtr(pending_io);
11380					break;
11381				case CTL_ACTION_OVERLAP:
11382					ctl_set_overlapped_cmd(
11383					    (struct ctl_scsiio *)pending_io);
11384					ctl_done(pending_io);
11385					break;
11386				case CTL_ACTION_OVERLAP_TAG:
11387					ctl_set_overlapped_tag(
11388					    (struct ctl_scsiio *)pending_io,
11389					    pending_io->scsiio.tag_num & 0xff);
11390					ctl_done(pending_io);
11391					break;
11392				case CTL_ACTION_ERROR:
11393				default:
11394					ctl_set_internal_failure(
11395						(struct ctl_scsiio *)pending_io,
11396						0,  // sks_valid
11397						0); //retry count
11398					ctl_done(pending_io);
11399					break;
11400				}
11401			}
11402
11403			/*
11404			 * Build Unit Attention
11405			 */
11406			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11407				lun->pending_ua[i] |=
11408				                     CTL_UA_ASYM_ACC_CHANGE;
11409			}
11410		} else {
11411			panic("Unhandled HA mode failover, LUN flags = %#x, "
11412			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11413		}
11414	}
11415	ctl_pause_rtr = 0;
11416	mtx_unlock(&ctl_softc->ctl_lock);
11417}
11418
11419static int
11420ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11421{
11422	struct ctl_lun *lun;
11423	const struct ctl_cmd_entry *entry;
11424	uint32_t initidx, targ_lun;
11425	int retval;
11426
11427	retval = 0;
11428
11429	lun = NULL;
11430
11431	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11432	if ((targ_lun < CTL_MAX_LUNS)
11433	 && ((lun = ctl_softc->ctl_luns[targ_lun]) != NULL)) {
11434		/*
11435		 * If the LUN is invalid, pretend that it doesn't exist.
11436		 * It will go away as soon as all pending I/O has been
11437		 * completed.
11438		 */
11439		mtx_lock(&lun->lun_lock);
11440		if (lun->flags & CTL_LUN_DISABLED) {
11441			mtx_unlock(&lun->lun_lock);
11442			lun = NULL;
11443			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11444			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11445		} else {
11446			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11447			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11448				lun->be_lun;
11449			if (lun->be_lun->lun_type == T_PROCESSOR) {
11450				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11451			}
11452
11453			/*
11454			 * Every I/O goes into the OOA queue for a
11455			 * particular LUN, and stays there until completion.
11456			 */
11457			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11458			    ooa_links);
11459		}
11460	} else {
11461		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11462		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11463	}
11464
11465	/* Get command entry and return error if it is unsuppotyed. */
11466	entry = ctl_validate_command(ctsio);
11467	if (entry == NULL) {
11468		if (lun)
11469			mtx_unlock(&lun->lun_lock);
11470		return (retval);
11471	}
11472
11473	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11474	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11475
11476	/*
11477	 * Check to see whether we can send this command to LUNs that don't
11478	 * exist.  This should pretty much only be the case for inquiry
11479	 * and request sense.  Further checks, below, really require having
11480	 * a LUN, so we can't really check the command anymore.  Just put
11481	 * it on the rtr queue.
11482	 */
11483	if (lun == NULL) {
11484		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11485			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11486			ctl_enqueue_rtr((union ctl_io *)ctsio);
11487			return (retval);
11488		}
11489
11490		ctl_set_unsupported_lun(ctsio);
11491		ctl_done((union ctl_io *)ctsio);
11492		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11493		return (retval);
11494	} else {
11495		/*
11496		 * Make sure we support this particular command on this LUN.
11497		 * e.g., we don't support writes to the control LUN.
11498		 */
11499		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11500			mtx_unlock(&lun->lun_lock);
11501			ctl_set_invalid_opcode(ctsio);
11502			ctl_done((union ctl_io *)ctsio);
11503			return (retval);
11504		}
11505	}
11506
11507	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11508
11509#ifdef CTL_WITH_CA
11510	/*
11511	 * If we've got a request sense, it'll clear the contingent
11512	 * allegiance condition.  Otherwise, if we have a CA condition for
11513	 * this initiator, clear it, because it sent down a command other
11514	 * than request sense.
11515	 */
11516	if ((ctsio->cdb[0] != REQUEST_SENSE)
11517	 && (ctl_is_set(lun->have_ca, initidx)))
11518		ctl_clear_mask(lun->have_ca, initidx);
11519#endif
11520
11521	/*
11522	 * If the command has this flag set, it handles its own unit
11523	 * attention reporting, we shouldn't do anything.  Otherwise we
11524	 * check for any pending unit attentions, and send them back to the
11525	 * initiator.  We only do this when a command initially comes in,
11526	 * not when we pull it off the blocked queue.
11527	 *
11528	 * According to SAM-3, section 5.3.2, the order that things get
11529	 * presented back to the host is basically unit attentions caused
11530	 * by some sort of reset event, busy status, reservation conflicts
11531	 * or task set full, and finally any other status.
11532	 *
11533	 * One issue here is that some of the unit attentions we report
11534	 * don't fall into the "reset" category (e.g. "reported luns data
11535	 * has changed").  So reporting it here, before the reservation
11536	 * check, may be technically wrong.  I guess the only thing to do
11537	 * would be to check for and report the reset events here, and then
11538	 * check for the other unit attention types after we check for a
11539	 * reservation conflict.
11540	 *
11541	 * XXX KDM need to fix this
11542	 */
11543	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11544		ctl_ua_type ua_type;
11545
11546		if (lun->pending_ua[initidx] != CTL_UA_NONE) {
11547			scsi_sense_data_type sense_format;
11548
11549			if (lun != NULL)
11550				sense_format = (lun->flags &
11551				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11552				    SSD_TYPE_FIXED;
11553			else
11554				sense_format = SSD_TYPE_FIXED;
11555
11556			ua_type = ctl_build_ua(&lun->pending_ua[initidx],
11557			    &ctsio->sense_data, sense_format);
11558			if (ua_type != CTL_UA_NONE) {
11559				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11560				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11561						       CTL_AUTOSENSE;
11562				ctsio->sense_len = SSD_FULL_SIZE;
11563				mtx_unlock(&lun->lun_lock);
11564				ctl_done((union ctl_io *)ctsio);
11565				return (retval);
11566			}
11567		}
11568	}
11569
11570
11571	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11572		mtx_unlock(&lun->lun_lock);
11573		ctl_done((union ctl_io *)ctsio);
11574		return (retval);
11575	}
11576
11577	/*
11578	 * XXX CHD this is where we want to send IO to other side if
11579	 * this LUN is secondary on this SC. We will need to make a copy
11580	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11581	 * the copy we send as FROM_OTHER.
11582	 * We also need to stuff the address of the original IO so we can
11583	 * find it easily. Something similar will need be done on the other
11584	 * side so when we are done we can find the copy.
11585	 */
11586	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11587		union ctl_ha_msg msg_info;
11588		int isc_retval;
11589
11590		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11591
11592		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11593		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11594#if 0
11595		printf("1. ctsio %p\n", ctsio);
11596#endif
11597		msg_info.hdr.serializing_sc = NULL;
11598		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11599		msg_info.scsi.tag_num = ctsio->tag_num;
11600		msg_info.scsi.tag_type = ctsio->tag_type;
11601		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11602
11603		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11604
11605		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11606		    (void *)&msg_info, sizeof(msg_info), 0)) >
11607		    CTL_HA_STATUS_SUCCESS) {
11608			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11609			       isc_retval);
11610			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11611		} else {
11612#if 0
11613			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11614#endif
11615		}
11616
11617		/*
11618		 * XXX KDM this I/O is off the incoming queue, but hasn't
11619		 * been inserted on any other queue.  We may need to come
11620		 * up with a holding queue while we wait for serialization
11621		 * so that we have an idea of what we're waiting for from
11622		 * the other side.
11623		 */
11624		mtx_unlock(&lun->lun_lock);
11625		return (retval);
11626	}
11627
11628	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11629			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11630			      ctl_ooaq, ooa_links))) {
11631	case CTL_ACTION_BLOCK:
11632		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11633		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11634				  blocked_links);
11635		mtx_unlock(&lun->lun_lock);
11636		return (retval);
11637	case CTL_ACTION_PASS:
11638	case CTL_ACTION_SKIP:
11639		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11640		mtx_unlock(&lun->lun_lock);
11641		ctl_enqueue_rtr((union ctl_io *)ctsio);
11642		break;
11643	case CTL_ACTION_OVERLAP:
11644		mtx_unlock(&lun->lun_lock);
11645		ctl_set_overlapped_cmd(ctsio);
11646		ctl_done((union ctl_io *)ctsio);
11647		break;
11648	case CTL_ACTION_OVERLAP_TAG:
11649		mtx_unlock(&lun->lun_lock);
11650		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11651		ctl_done((union ctl_io *)ctsio);
11652		break;
11653	case CTL_ACTION_ERROR:
11654	default:
11655		mtx_unlock(&lun->lun_lock);
11656		ctl_set_internal_failure(ctsio,
11657					 /*sks_valid*/ 0,
11658					 /*retry_count*/ 0);
11659		ctl_done((union ctl_io *)ctsio);
11660		break;
11661	}
11662	return (retval);
11663}
11664
11665const struct ctl_cmd_entry *
11666ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11667{
11668	const struct ctl_cmd_entry *entry;
11669	int service_action;
11670
11671	entry = &ctl_cmd_table[ctsio->cdb[0]];
11672	if (sa)
11673		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11674	if (entry->flags & CTL_CMD_FLAG_SA5) {
11675		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11676		entry = &((const struct ctl_cmd_entry *)
11677		    entry->execute)[service_action];
11678	}
11679	return (entry);
11680}
11681
11682const struct ctl_cmd_entry *
11683ctl_validate_command(struct ctl_scsiio *ctsio)
11684{
11685	const struct ctl_cmd_entry *entry;
11686	int i, sa;
11687	uint8_t diff;
11688
11689	entry = ctl_get_cmd_entry(ctsio, &sa);
11690	if (entry->execute == NULL) {
11691		if (sa)
11692			ctl_set_invalid_field(ctsio,
11693					      /*sks_valid*/ 1,
11694					      /*command*/ 1,
11695					      /*field*/ 1,
11696					      /*bit_valid*/ 1,
11697					      /*bit*/ 4);
11698		else
11699			ctl_set_invalid_opcode(ctsio);
11700		ctl_done((union ctl_io *)ctsio);
11701		return (NULL);
11702	}
11703	KASSERT(entry->length > 0,
11704	    ("Not defined length for command 0x%02x/0x%02x",
11705	     ctsio->cdb[0], ctsio->cdb[1]));
11706	for (i = 1; i < entry->length; i++) {
11707		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11708		if (diff == 0)
11709			continue;
11710		ctl_set_invalid_field(ctsio,
11711				      /*sks_valid*/ 1,
11712				      /*command*/ 1,
11713				      /*field*/ i,
11714				      /*bit_valid*/ 1,
11715				      /*bit*/ fls(diff) - 1);
11716		ctl_done((union ctl_io *)ctsio);
11717		return (NULL);
11718	}
11719	return (entry);
11720}
11721
11722static int
11723ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11724{
11725
11726	switch (lun_type) {
11727	case T_PROCESSOR:
11728		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11729		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11730			return (0);
11731		break;
11732	case T_DIRECT:
11733		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11734		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11735			return (0);
11736		break;
11737	default:
11738		return (0);
11739	}
11740	return (1);
11741}
11742
11743static int
11744ctl_scsiio(struct ctl_scsiio *ctsio)
11745{
11746	int retval;
11747	const struct ctl_cmd_entry *entry;
11748
11749	retval = CTL_RETVAL_COMPLETE;
11750
11751	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11752
11753	entry = ctl_get_cmd_entry(ctsio, NULL);
11754
11755	/*
11756	 * If this I/O has been aborted, just send it straight to
11757	 * ctl_done() without executing it.
11758	 */
11759	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11760		ctl_done((union ctl_io *)ctsio);
11761		goto bailout;
11762	}
11763
11764	/*
11765	 * All the checks should have been handled by ctl_scsiio_precheck().
11766	 * We should be clear now to just execute the I/O.
11767	 */
11768	retval = entry->execute(ctsio);
11769
11770bailout:
11771	return (retval);
11772}
11773
11774/*
11775 * Since we only implement one target right now, a bus reset simply resets
11776 * our single target.
11777 */
11778static int
11779ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11780{
11781	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11782}
11783
11784static int
11785ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11786		 ctl_ua_type ua_type)
11787{
11788	struct ctl_lun *lun;
11789	int retval;
11790
11791	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11792		union ctl_ha_msg msg_info;
11793
11794		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11795		msg_info.hdr.nexus = io->io_hdr.nexus;
11796		if (ua_type==CTL_UA_TARG_RESET)
11797			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11798		else
11799			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11800		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11801		msg_info.hdr.original_sc = NULL;
11802		msg_info.hdr.serializing_sc = NULL;
11803		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11804		    (void *)&msg_info, sizeof(msg_info), 0)) {
11805		}
11806	}
11807	retval = 0;
11808
11809	mtx_lock(&ctl_softc->ctl_lock);
11810	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11811		retval += ctl_lun_reset(lun, io, ua_type);
11812	mtx_unlock(&ctl_softc->ctl_lock);
11813
11814	return (retval);
11815}
11816
11817/*
11818 * The LUN should always be set.  The I/O is optional, and is used to
11819 * distinguish between I/Os sent by this initiator, and by other
11820 * initiators.  We set unit attention for initiators other than this one.
11821 * SAM-3 is vague on this point.  It does say that a unit attention should
11822 * be established for other initiators when a LUN is reset (see section
11823 * 5.7.3), but it doesn't specifically say that the unit attention should
11824 * be established for this particular initiator when a LUN is reset.  Here
11825 * is the relevant text, from SAM-3 rev 8:
11826 *
11827 * 5.7.2 When a SCSI initiator port aborts its own tasks
11828 *
11829 * When a SCSI initiator port causes its own task(s) to be aborted, no
11830 * notification that the task(s) have been aborted shall be returned to
11831 * the SCSI initiator port other than the completion response for the
11832 * command or task management function action that caused the task(s) to
11833 * be aborted and notification(s) associated with related effects of the
11834 * action (e.g., a reset unit attention condition).
11835 *
11836 * XXX KDM for now, we're setting unit attention for all initiators.
11837 */
11838static int
11839ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11840{
11841	union ctl_io *xio;
11842#if 0
11843	uint32_t initindex;
11844#endif
11845	int i;
11846
11847	mtx_lock(&lun->lun_lock);
11848	/*
11849	 * Run through the OOA queue and abort each I/O.
11850	 */
11851#if 0
11852	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11853#endif
11854	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11855	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11856		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11857	}
11858
11859	/*
11860	 * This version sets unit attention for every
11861	 */
11862#if 0
11863	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11864	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11865		if (initindex == i)
11866			continue;
11867		lun->pending_ua[i] |= ua_type;
11868	}
11869#endif
11870
11871	/*
11872	 * A reset (any kind, really) clears reservations established with
11873	 * RESERVE/RELEASE.  It does not clear reservations established
11874	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11875	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11876	 * reservations made with the RESERVE/RELEASE commands, because
11877	 * those commands are obsolete in SPC-3.
11878	 */
11879	lun->flags &= ~CTL_LUN_RESERVED;
11880
11881	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11882#ifdef CTL_WITH_CA
11883		ctl_clear_mask(lun->have_ca, i);
11884#endif
11885		lun->pending_ua[i] |= ua_type;
11886	}
11887	mtx_unlock(&lun->lun_lock);
11888
11889	return (0);
11890}
11891
11892static void
11893ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11894    int other_sc)
11895{
11896	union ctl_io *xio;
11897
11898	mtx_assert(&lun->lun_lock, MA_OWNED);
11899
11900	/*
11901	 * Run through the OOA queue and attempt to find the given I/O.
11902	 * The target port, initiator ID, tag type and tag number have to
11903	 * match the values that we got from the initiator.  If we have an
11904	 * untagged command to abort, simply abort the first untagged command
11905	 * we come to.  We only allow one untagged command at a time of course.
11906	 */
11907	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11908	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11909
11910		if ((targ_port == UINT32_MAX ||
11911		     targ_port == xio->io_hdr.nexus.targ_port) &&
11912		    (init_id == UINT32_MAX ||
11913		     init_id == xio->io_hdr.nexus.initid.id)) {
11914			if (targ_port != xio->io_hdr.nexus.targ_port ||
11915			    init_id != xio->io_hdr.nexus.initid.id)
11916				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11917			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11918			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11919				union ctl_ha_msg msg_info;
11920
11921				msg_info.hdr.nexus = xio->io_hdr.nexus;
11922				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11923				msg_info.task.tag_num = xio->scsiio.tag_num;
11924				msg_info.task.tag_type = xio->scsiio.tag_type;
11925				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11926				msg_info.hdr.original_sc = NULL;
11927				msg_info.hdr.serializing_sc = NULL;
11928				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11929				    (void *)&msg_info, sizeof(msg_info), 0);
11930			}
11931		}
11932	}
11933}
11934
11935static int
11936ctl_abort_task_set(union ctl_io *io)
11937{
11938	struct ctl_softc *softc = control_softc;
11939	struct ctl_lun *lun;
11940	uint32_t targ_lun;
11941
11942	/*
11943	 * Look up the LUN.
11944	 */
11945	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11946	mtx_lock(&softc->ctl_lock);
11947	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11948		lun = softc->ctl_luns[targ_lun];
11949	else {
11950		mtx_unlock(&softc->ctl_lock);
11951		return (1);
11952	}
11953
11954	mtx_lock(&lun->lun_lock);
11955	mtx_unlock(&softc->ctl_lock);
11956	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11957		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11958		    io->io_hdr.nexus.initid.id,
11959		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11960	} else { /* CTL_TASK_CLEAR_TASK_SET */
11961		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11962		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11963	}
11964	mtx_unlock(&lun->lun_lock);
11965	return (0);
11966}
11967
11968static int
11969ctl_i_t_nexus_reset(union ctl_io *io)
11970{
11971	struct ctl_softc *softc = control_softc;
11972	struct ctl_lun *lun;
11973	uint32_t initindex, residx;
11974
11975	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11976	residx = ctl_get_resindex(&io->io_hdr.nexus);
11977	mtx_lock(&softc->ctl_lock);
11978	STAILQ_FOREACH(lun, &softc->lun_list, links) {
11979		mtx_lock(&lun->lun_lock);
11980		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11981		    io->io_hdr.nexus.initid.id,
11982		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11983#ifdef CTL_WITH_CA
11984		ctl_clear_mask(lun->have_ca, initindex);
11985#endif
11986		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
11987			lun->flags &= ~CTL_LUN_RESERVED;
11988		lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS;
11989		mtx_unlock(&lun->lun_lock);
11990	}
11991	mtx_unlock(&softc->ctl_lock);
11992	return (0);
11993}
11994
11995static int
11996ctl_abort_task(union ctl_io *io)
11997{
11998	union ctl_io *xio;
11999	struct ctl_lun *lun;
12000	struct ctl_softc *ctl_softc;
12001#if 0
12002	struct sbuf sb;
12003	char printbuf[128];
12004#endif
12005	int found;
12006	uint32_t targ_lun;
12007
12008	ctl_softc = control_softc;
12009	found = 0;
12010
12011	/*
12012	 * Look up the LUN.
12013	 */
12014	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12015	mtx_lock(&ctl_softc->ctl_lock);
12016	if ((targ_lun < CTL_MAX_LUNS)
12017	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12018		lun = ctl_softc->ctl_luns[targ_lun];
12019	else {
12020		mtx_unlock(&ctl_softc->ctl_lock);
12021		return (1);
12022	}
12023
12024#if 0
12025	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12026	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12027#endif
12028
12029	mtx_lock(&lun->lun_lock);
12030	mtx_unlock(&ctl_softc->ctl_lock);
12031	/*
12032	 * Run through the OOA queue and attempt to find the given I/O.
12033	 * The target port, initiator ID, tag type and tag number have to
12034	 * match the values that we got from the initiator.  If we have an
12035	 * untagged command to abort, simply abort the first untagged command
12036	 * we come to.  We only allow one untagged command at a time of course.
12037	 */
12038#if 0
12039	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12040#endif
12041	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12042	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12043#if 0
12044		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12045
12046		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12047			    lun->lun, xio->scsiio.tag_num,
12048			    xio->scsiio.tag_type,
12049			    (xio->io_hdr.blocked_links.tqe_prev
12050			    == NULL) ? "" : " BLOCKED",
12051			    (xio->io_hdr.flags &
12052			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12053			    (xio->io_hdr.flags &
12054			    CTL_FLAG_ABORT) ? " ABORT" : "",
12055			    (xio->io_hdr.flags &
12056			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12057		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12058		sbuf_finish(&sb);
12059		printf("%s\n", sbuf_data(&sb));
12060#endif
12061
12062		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12063		 && (xio->io_hdr.nexus.initid.id ==
12064		     io->io_hdr.nexus.initid.id)) {
12065			/*
12066			 * If the abort says that the task is untagged, the
12067			 * task in the queue must be untagged.  Otherwise,
12068			 * we just check to see whether the tag numbers
12069			 * match.  This is because the QLogic firmware
12070			 * doesn't pass back the tag type in an abort
12071			 * request.
12072			 */
12073#if 0
12074			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12075			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12076			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12077#endif
12078			/*
12079			 * XXX KDM we've got problems with FC, because it
12080			 * doesn't send down a tag type with aborts.  So we
12081			 * can only really go by the tag number...
12082			 * This may cause problems with parallel SCSI.
12083			 * Need to figure that out!!
12084			 */
12085			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12086				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12087				found = 1;
12088				if ((io->io_hdr.flags &
12089				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12090				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12091					union ctl_ha_msg msg_info;
12092
12093					io->io_hdr.flags |=
12094					                CTL_FLAG_SENT_2OTHER_SC;
12095					msg_info.hdr.nexus = io->io_hdr.nexus;
12096					msg_info.task.task_action =
12097						CTL_TASK_ABORT_TASK;
12098					msg_info.task.tag_num =
12099						io->taskio.tag_num;
12100					msg_info.task.tag_type =
12101						io->taskio.tag_type;
12102					msg_info.hdr.msg_type =
12103						CTL_MSG_MANAGE_TASKS;
12104					msg_info.hdr.original_sc = NULL;
12105					msg_info.hdr.serializing_sc = NULL;
12106#if 0
12107					printf("Sent Abort to other side\n");
12108#endif
12109					if (CTL_HA_STATUS_SUCCESS !=
12110					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12111		    				(void *)&msg_info,
12112						sizeof(msg_info), 0)) {
12113					}
12114				}
12115#if 0
12116				printf("ctl_abort_task: found I/O to abort\n");
12117#endif
12118				break;
12119			}
12120		}
12121	}
12122	mtx_unlock(&lun->lun_lock);
12123
12124	if (found == 0) {
12125		/*
12126		 * This isn't really an error.  It's entirely possible for
12127		 * the abort and command completion to cross on the wire.
12128		 * This is more of an informative/diagnostic error.
12129		 */
12130#if 0
12131		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12132		       "%d:%d:%d:%d tag %d type %d\n",
12133		       io->io_hdr.nexus.initid.id,
12134		       io->io_hdr.nexus.targ_port,
12135		       io->io_hdr.nexus.targ_target.id,
12136		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12137		       io->taskio.tag_type);
12138#endif
12139	}
12140	return (0);
12141}
12142
12143static void
12144ctl_run_task(union ctl_io *io)
12145{
12146	struct ctl_softc *ctl_softc = control_softc;
12147	int retval = 1;
12148	const char *task_desc;
12149
12150	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12151
12152	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12153	    ("ctl_run_task: Unextected io_type %d\n",
12154	     io->io_hdr.io_type));
12155
12156	task_desc = ctl_scsi_task_string(&io->taskio);
12157	if (task_desc != NULL) {
12158#ifdef NEEDTOPORT
12159		csevent_log(CSC_CTL | CSC_SHELF_SW |
12160			    CTL_TASK_REPORT,
12161			    csevent_LogType_Trace,
12162			    csevent_Severity_Information,
12163			    csevent_AlertLevel_Green,
12164			    csevent_FRU_Firmware,
12165			    csevent_FRU_Unknown,
12166			    "CTL: received task: %s",task_desc);
12167#endif
12168	} else {
12169#ifdef NEEDTOPORT
12170		csevent_log(CSC_CTL | CSC_SHELF_SW |
12171			    CTL_TASK_REPORT,
12172			    csevent_LogType_Trace,
12173			    csevent_Severity_Information,
12174			    csevent_AlertLevel_Green,
12175			    csevent_FRU_Firmware,
12176			    csevent_FRU_Unknown,
12177			    "CTL: received unknown task "
12178			    "type: %d (%#x)",
12179			    io->taskio.task_action,
12180			    io->taskio.task_action);
12181#endif
12182	}
12183	switch (io->taskio.task_action) {
12184	case CTL_TASK_ABORT_TASK:
12185		retval = ctl_abort_task(io);
12186		break;
12187	case CTL_TASK_ABORT_TASK_SET:
12188	case CTL_TASK_CLEAR_TASK_SET:
12189		retval = ctl_abort_task_set(io);
12190		break;
12191	case CTL_TASK_CLEAR_ACA:
12192		break;
12193	case CTL_TASK_I_T_NEXUS_RESET:
12194		retval = ctl_i_t_nexus_reset(io);
12195		break;
12196	case CTL_TASK_LUN_RESET: {
12197		struct ctl_lun *lun;
12198		uint32_t targ_lun;
12199
12200		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12201		mtx_lock(&ctl_softc->ctl_lock);
12202		if ((targ_lun < CTL_MAX_LUNS)
12203		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12204			lun = ctl_softc->ctl_luns[targ_lun];
12205		else {
12206			mtx_unlock(&ctl_softc->ctl_lock);
12207			retval = 1;
12208			break;
12209		}
12210
12211		if (!(io->io_hdr.flags &
12212		    CTL_FLAG_FROM_OTHER_SC)) {
12213			union ctl_ha_msg msg_info;
12214
12215			io->io_hdr.flags |=
12216				CTL_FLAG_SENT_2OTHER_SC;
12217			msg_info.hdr.msg_type =
12218				CTL_MSG_MANAGE_TASKS;
12219			msg_info.hdr.nexus = io->io_hdr.nexus;
12220			msg_info.task.task_action =
12221				CTL_TASK_LUN_RESET;
12222			msg_info.hdr.original_sc = NULL;
12223			msg_info.hdr.serializing_sc = NULL;
12224			if (CTL_HA_STATUS_SUCCESS !=
12225			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12226			    (void *)&msg_info,
12227			    sizeof(msg_info), 0)) {
12228			}
12229		}
12230
12231		retval = ctl_lun_reset(lun, io,
12232				       CTL_UA_LUN_RESET);
12233		mtx_unlock(&ctl_softc->ctl_lock);
12234		break;
12235	}
12236	case CTL_TASK_TARGET_RESET:
12237		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12238		break;
12239	case CTL_TASK_BUS_RESET:
12240		retval = ctl_bus_reset(ctl_softc, io);
12241		break;
12242	case CTL_TASK_PORT_LOGIN:
12243		break;
12244	case CTL_TASK_PORT_LOGOUT:
12245		break;
12246	default:
12247		printf("ctl_run_task: got unknown task management event %d\n",
12248		       io->taskio.task_action);
12249		break;
12250	}
12251	if (retval == 0)
12252		io->io_hdr.status = CTL_SUCCESS;
12253	else
12254		io->io_hdr.status = CTL_ERROR;
12255	ctl_done(io);
12256}
12257
12258/*
12259 * For HA operation.  Handle commands that come in from the other
12260 * controller.
12261 */
12262static void
12263ctl_handle_isc(union ctl_io *io)
12264{
12265	int free_io;
12266	struct ctl_lun *lun;
12267	struct ctl_softc *ctl_softc;
12268	uint32_t targ_lun;
12269
12270	ctl_softc = control_softc;
12271
12272	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12273	lun = ctl_softc->ctl_luns[targ_lun];
12274
12275	switch (io->io_hdr.msg_type) {
12276	case CTL_MSG_SERIALIZE:
12277		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12278		break;
12279	case CTL_MSG_R2R: {
12280		const struct ctl_cmd_entry *entry;
12281
12282		/*
12283		 * This is only used in SER_ONLY mode.
12284		 */
12285		free_io = 0;
12286		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12287		mtx_lock(&lun->lun_lock);
12288		if (ctl_scsiio_lun_check(ctl_softc, lun,
12289		    entry, (struct ctl_scsiio *)io) != 0) {
12290			mtx_unlock(&lun->lun_lock);
12291			ctl_done(io);
12292			break;
12293		}
12294		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12295		mtx_unlock(&lun->lun_lock);
12296		ctl_enqueue_rtr(io);
12297		break;
12298	}
12299	case CTL_MSG_FINISH_IO:
12300		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12301			free_io = 0;
12302			ctl_done(io);
12303		} else {
12304			free_io = 1;
12305			mtx_lock(&lun->lun_lock);
12306			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12307				     ooa_links);
12308			ctl_check_blocked(lun);
12309			mtx_unlock(&lun->lun_lock);
12310		}
12311		break;
12312	case CTL_MSG_PERS_ACTION:
12313		ctl_hndl_per_res_out_on_other_sc(
12314			(union ctl_ha_msg *)&io->presio.pr_msg);
12315		free_io = 1;
12316		break;
12317	case CTL_MSG_BAD_JUJU:
12318		free_io = 0;
12319		ctl_done(io);
12320		break;
12321	case CTL_MSG_DATAMOVE:
12322		/* Only used in XFER mode */
12323		free_io = 0;
12324		ctl_datamove_remote(io);
12325		break;
12326	case CTL_MSG_DATAMOVE_DONE:
12327		/* Only used in XFER mode */
12328		free_io = 0;
12329		io->scsiio.be_move_done(io);
12330		break;
12331	default:
12332		free_io = 1;
12333		printf("%s: Invalid message type %d\n",
12334		       __func__, io->io_hdr.msg_type);
12335		break;
12336	}
12337	if (free_io)
12338		ctl_free_io(io);
12339
12340}
12341
12342
12343/*
12344 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12345 * there is no match.
12346 */
12347static ctl_lun_error_pattern
12348ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12349{
12350	const struct ctl_cmd_entry *entry;
12351	ctl_lun_error_pattern filtered_pattern, pattern;
12352
12353	pattern = desc->error_pattern;
12354
12355	/*
12356	 * XXX KDM we need more data passed into this function to match a
12357	 * custom pattern, and we actually need to implement custom pattern
12358	 * matching.
12359	 */
12360	if (pattern & CTL_LUN_PAT_CMD)
12361		return (CTL_LUN_PAT_CMD);
12362
12363	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12364		return (CTL_LUN_PAT_ANY);
12365
12366	entry = ctl_get_cmd_entry(ctsio, NULL);
12367
12368	filtered_pattern = entry->pattern & pattern;
12369
12370	/*
12371	 * If the user requested specific flags in the pattern (e.g.
12372	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12373	 * flags.
12374	 *
12375	 * If the user did not specify any flags, it doesn't matter whether
12376	 * or not the command supports the flags.
12377	 */
12378	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12379	     (pattern & ~CTL_LUN_PAT_MASK))
12380		return (CTL_LUN_PAT_NONE);
12381
12382	/*
12383	 * If the user asked for a range check, see if the requested LBA
12384	 * range overlaps with this command's LBA range.
12385	 */
12386	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12387		uint64_t lba1;
12388		uint64_t len1;
12389		ctl_action action;
12390		int retval;
12391
12392		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12393		if (retval != 0)
12394			return (CTL_LUN_PAT_NONE);
12395
12396		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12397					      desc->lba_range.len);
12398		/*
12399		 * A "pass" means that the LBA ranges don't overlap, so
12400		 * this doesn't match the user's range criteria.
12401		 */
12402		if (action == CTL_ACTION_PASS)
12403			return (CTL_LUN_PAT_NONE);
12404	}
12405
12406	return (filtered_pattern);
12407}
12408
12409static void
12410ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12411{
12412	struct ctl_error_desc *desc, *desc2;
12413
12414	mtx_assert(&lun->lun_lock, MA_OWNED);
12415
12416	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12417		ctl_lun_error_pattern pattern;
12418		/*
12419		 * Check to see whether this particular command matches
12420		 * the pattern in the descriptor.
12421		 */
12422		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12423		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12424			continue;
12425
12426		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12427		case CTL_LUN_INJ_ABORTED:
12428			ctl_set_aborted(&io->scsiio);
12429			break;
12430		case CTL_LUN_INJ_MEDIUM_ERR:
12431			ctl_set_medium_error(&io->scsiio);
12432			break;
12433		case CTL_LUN_INJ_UA:
12434			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12435			 * OCCURRED */
12436			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12437			break;
12438		case CTL_LUN_INJ_CUSTOM:
12439			/*
12440			 * We're assuming the user knows what he is doing.
12441			 * Just copy the sense information without doing
12442			 * checks.
12443			 */
12444			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12445			      ctl_min(sizeof(desc->custom_sense),
12446				      sizeof(io->scsiio.sense_data)));
12447			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12448			io->scsiio.sense_len = SSD_FULL_SIZE;
12449			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12450			break;
12451		case CTL_LUN_INJ_NONE:
12452		default:
12453			/*
12454			 * If this is an error injection type we don't know
12455			 * about, clear the continuous flag (if it is set)
12456			 * so it will get deleted below.
12457			 */
12458			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12459			break;
12460		}
12461		/*
12462		 * By default, each error injection action is a one-shot
12463		 */
12464		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12465			continue;
12466
12467		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12468
12469		free(desc, M_CTL);
12470	}
12471}
12472
12473#ifdef CTL_IO_DELAY
12474static void
12475ctl_datamove_timer_wakeup(void *arg)
12476{
12477	union ctl_io *io;
12478
12479	io = (union ctl_io *)arg;
12480
12481	ctl_datamove(io);
12482}
12483#endif /* CTL_IO_DELAY */
12484
12485void
12486ctl_datamove(union ctl_io *io)
12487{
12488	void (*fe_datamove)(union ctl_io *io);
12489
12490	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12491
12492	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12493
12494#ifdef CTL_TIME_IO
12495	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12496		char str[256];
12497		char path_str[64];
12498		struct sbuf sb;
12499
12500		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12501		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12502
12503		sbuf_cat(&sb, path_str);
12504		switch (io->io_hdr.io_type) {
12505		case CTL_IO_SCSI:
12506			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12507			sbuf_printf(&sb, "\n");
12508			sbuf_cat(&sb, path_str);
12509			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12510				    io->scsiio.tag_num, io->scsiio.tag_type);
12511			break;
12512		case CTL_IO_TASK:
12513			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12514				    "Tag Type: %d\n", io->taskio.task_action,
12515				    io->taskio.tag_num, io->taskio.tag_type);
12516			break;
12517		default:
12518			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12519			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12520			break;
12521		}
12522		sbuf_cat(&sb, path_str);
12523		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12524			    (intmax_t)time_uptime - io->io_hdr.start_time);
12525		sbuf_finish(&sb);
12526		printf("%s", sbuf_data(&sb));
12527	}
12528#endif /* CTL_TIME_IO */
12529
12530#ifdef CTL_IO_DELAY
12531	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12532		struct ctl_lun *lun;
12533
12534		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12535
12536		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12537	} else {
12538		struct ctl_lun *lun;
12539
12540		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12541		if ((lun != NULL)
12542		 && (lun->delay_info.datamove_delay > 0)) {
12543			struct callout *callout;
12544
12545			callout = (struct callout *)&io->io_hdr.timer_bytes;
12546			callout_init(callout, /*mpsafe*/ 1);
12547			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12548			callout_reset(callout,
12549				      lun->delay_info.datamove_delay * hz,
12550				      ctl_datamove_timer_wakeup, io);
12551			if (lun->delay_info.datamove_type ==
12552			    CTL_DELAY_TYPE_ONESHOT)
12553				lun->delay_info.datamove_delay = 0;
12554			return;
12555		}
12556	}
12557#endif
12558
12559	/*
12560	 * This command has been aborted.  Set the port status, so we fail
12561	 * the data move.
12562	 */
12563	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12564		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12565		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12566		       io->io_hdr.nexus.targ_port,
12567		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12568		       io->io_hdr.nexus.targ_lun);
12569		io->io_hdr.port_status = 31337;
12570		/*
12571		 * Note that the backend, in this case, will get the
12572		 * callback in its context.  In other cases it may get
12573		 * called in the frontend's interrupt thread context.
12574		 */
12575		io->scsiio.be_move_done(io);
12576		return;
12577	}
12578
12579	/* Don't confuse frontend with zero length data move. */
12580	if (io->scsiio.kern_data_len == 0) {
12581		io->scsiio.be_move_done(io);
12582		return;
12583	}
12584
12585	/*
12586	 * If we're in XFER mode and this I/O is from the other shelf
12587	 * controller, we need to send the DMA to the other side to
12588	 * actually transfer the data to/from the host.  In serialize only
12589	 * mode the transfer happens below CTL and ctl_datamove() is only
12590	 * called on the machine that originally received the I/O.
12591	 */
12592	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12593	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12594		union ctl_ha_msg msg;
12595		uint32_t sg_entries_sent;
12596		int do_sg_copy;
12597		int i;
12598
12599		memset(&msg, 0, sizeof(msg));
12600		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12601		msg.hdr.original_sc = io->io_hdr.original_sc;
12602		msg.hdr.serializing_sc = io;
12603		msg.hdr.nexus = io->io_hdr.nexus;
12604		msg.dt.flags = io->io_hdr.flags;
12605		/*
12606		 * We convert everything into a S/G list here.  We can't
12607		 * pass by reference, only by value between controllers.
12608		 * So we can't pass a pointer to the S/G list, only as many
12609		 * S/G entries as we can fit in here.  If it's possible for
12610		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12611		 * then we need to break this up into multiple transfers.
12612		 */
12613		if (io->scsiio.kern_sg_entries == 0) {
12614			msg.dt.kern_sg_entries = 1;
12615			/*
12616			 * If this is in cached memory, flush the cache
12617			 * before we send the DMA request to the other
12618			 * controller.  We want to do this in either the
12619			 * read or the write case.  The read case is
12620			 * straightforward.  In the write case, we want to
12621			 * make sure nothing is in the local cache that
12622			 * could overwrite the DMAed data.
12623			 */
12624			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12625				/*
12626				 * XXX KDM use bus_dmamap_sync() here.
12627				 */
12628			}
12629
12630			/*
12631			 * Convert to a physical address if this is a
12632			 * virtual address.
12633			 */
12634			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12635				msg.dt.sg_list[0].addr =
12636					io->scsiio.kern_data_ptr;
12637			} else {
12638				/*
12639				 * XXX KDM use busdma here!
12640				 */
12641#if 0
12642				msg.dt.sg_list[0].addr = (void *)
12643					vtophys(io->scsiio.kern_data_ptr);
12644#endif
12645			}
12646
12647			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12648			do_sg_copy = 0;
12649		} else {
12650			struct ctl_sg_entry *sgl;
12651
12652			do_sg_copy = 1;
12653			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12654			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12655			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12656				/*
12657				 * XXX KDM use bus_dmamap_sync() here.
12658				 */
12659			}
12660		}
12661
12662		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12663		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12664		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12665		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12666		msg.dt.sg_sequence = 0;
12667
12668		/*
12669		 * Loop until we've sent all of the S/G entries.  On the
12670		 * other end, we'll recompose these S/G entries into one
12671		 * contiguous list before passing it to the
12672		 */
12673		for (sg_entries_sent = 0; sg_entries_sent <
12674		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12675			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12676				sizeof(msg.dt.sg_list[0])),
12677				msg.dt.kern_sg_entries - sg_entries_sent);
12678
12679			if (do_sg_copy != 0) {
12680				struct ctl_sg_entry *sgl;
12681				int j;
12682
12683				sgl = (struct ctl_sg_entry *)
12684					io->scsiio.kern_data_ptr;
12685				/*
12686				 * If this is in cached memory, flush the cache
12687				 * before we send the DMA request to the other
12688				 * controller.  We want to do this in either
12689				 * the * read or the write case.  The read
12690				 * case is straightforward.  In the write
12691				 * case, we want to make sure nothing is
12692				 * in the local cache that could overwrite
12693				 * the DMAed data.
12694				 */
12695
12696				for (i = sg_entries_sent, j = 0;
12697				     i < msg.dt.cur_sg_entries; i++, j++) {
12698					if ((io->io_hdr.flags &
12699					     CTL_FLAG_NO_DATASYNC) == 0) {
12700						/*
12701						 * XXX KDM use bus_dmamap_sync()
12702						 */
12703					}
12704					if ((io->io_hdr.flags &
12705					     CTL_FLAG_BUS_ADDR) == 0) {
12706						/*
12707						 * XXX KDM use busdma.
12708						 */
12709#if 0
12710						msg.dt.sg_list[j].addr =(void *)
12711						       vtophys(sgl[i].addr);
12712#endif
12713					} else {
12714						msg.dt.sg_list[j].addr =
12715							sgl[i].addr;
12716					}
12717					msg.dt.sg_list[j].len = sgl[i].len;
12718				}
12719			}
12720
12721			sg_entries_sent += msg.dt.cur_sg_entries;
12722			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12723				msg.dt.sg_last = 1;
12724			else
12725				msg.dt.sg_last = 0;
12726
12727			/*
12728			 * XXX KDM drop and reacquire the lock here?
12729			 */
12730			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12731			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12732				/*
12733				 * XXX do something here.
12734				 */
12735			}
12736
12737			msg.dt.sent_sg_entries = sg_entries_sent;
12738		}
12739		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12740		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12741			ctl_failover_io(io, /*have_lock*/ 0);
12742
12743	} else {
12744
12745		/*
12746		 * Lookup the fe_datamove() function for this particular
12747		 * front end.
12748		 */
12749		fe_datamove =
12750		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12751
12752		fe_datamove(io);
12753	}
12754}
12755
12756static void
12757ctl_send_datamove_done(union ctl_io *io, int have_lock)
12758{
12759	union ctl_ha_msg msg;
12760	int isc_status;
12761
12762	memset(&msg, 0, sizeof(msg));
12763
12764	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12765	msg.hdr.original_sc = io;
12766	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12767	msg.hdr.nexus = io->io_hdr.nexus;
12768	msg.hdr.status = io->io_hdr.status;
12769	msg.scsi.tag_num = io->scsiio.tag_num;
12770	msg.scsi.tag_type = io->scsiio.tag_type;
12771	msg.scsi.scsi_status = io->scsiio.scsi_status;
12772	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12773	       sizeof(io->scsiio.sense_data));
12774	msg.scsi.sense_len = io->scsiio.sense_len;
12775	msg.scsi.sense_residual = io->scsiio.sense_residual;
12776	msg.scsi.fetd_status = io->io_hdr.port_status;
12777	msg.scsi.residual = io->scsiio.residual;
12778	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12779
12780	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12781		ctl_failover_io(io, /*have_lock*/ have_lock);
12782		return;
12783	}
12784
12785	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12786	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12787		/* XXX do something if this fails */
12788	}
12789
12790}
12791
12792/*
12793 * The DMA to the remote side is done, now we need to tell the other side
12794 * we're done so it can continue with its data movement.
12795 */
12796static void
12797ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12798{
12799	union ctl_io *io;
12800
12801	io = rq->context;
12802
12803	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12804		printf("%s: ISC DMA write failed with error %d", __func__,
12805		       rq->ret);
12806		ctl_set_internal_failure(&io->scsiio,
12807					 /*sks_valid*/ 1,
12808					 /*retry_count*/ rq->ret);
12809	}
12810
12811	ctl_dt_req_free(rq);
12812
12813	/*
12814	 * In this case, we had to malloc the memory locally.  Free it.
12815	 */
12816	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12817		int i;
12818		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12819			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12820	}
12821	/*
12822	 * The data is in local and remote memory, so now we need to send
12823	 * status (good or back) back to the other side.
12824	 */
12825	ctl_send_datamove_done(io, /*have_lock*/ 0);
12826}
12827
12828/*
12829 * We've moved the data from the host/controller into local memory.  Now we
12830 * need to push it over to the remote controller's memory.
12831 */
12832static int
12833ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12834{
12835	int retval;
12836
12837	retval = 0;
12838
12839	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12840					  ctl_datamove_remote_write_cb);
12841
12842	return (retval);
12843}
12844
12845static void
12846ctl_datamove_remote_write(union ctl_io *io)
12847{
12848	int retval;
12849	void (*fe_datamove)(union ctl_io *io);
12850
12851	/*
12852	 * - Get the data from the host/HBA into local memory.
12853	 * - DMA memory from the local controller to the remote controller.
12854	 * - Send status back to the remote controller.
12855	 */
12856
12857	retval = ctl_datamove_remote_sgl_setup(io);
12858	if (retval != 0)
12859		return;
12860
12861	/* Switch the pointer over so the FETD knows what to do */
12862	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12863
12864	/*
12865	 * Use a custom move done callback, since we need to send completion
12866	 * back to the other controller, not to the backend on this side.
12867	 */
12868	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12869
12870	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12871
12872	fe_datamove(io);
12873
12874	return;
12875
12876}
12877
12878static int
12879ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12880{
12881#if 0
12882	char str[256];
12883	char path_str[64];
12884	struct sbuf sb;
12885#endif
12886
12887	/*
12888	 * In this case, we had to malloc the memory locally.  Free it.
12889	 */
12890	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12891		int i;
12892		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12893			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12894	}
12895
12896#if 0
12897	scsi_path_string(io, path_str, sizeof(path_str));
12898	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12899	sbuf_cat(&sb, path_str);
12900	scsi_command_string(&io->scsiio, NULL, &sb);
12901	sbuf_printf(&sb, "\n");
12902	sbuf_cat(&sb, path_str);
12903	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12904		    io->scsiio.tag_num, io->scsiio.tag_type);
12905	sbuf_cat(&sb, path_str);
12906	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12907		    io->io_hdr.flags, io->io_hdr.status);
12908	sbuf_finish(&sb);
12909	printk("%s", sbuf_data(&sb));
12910#endif
12911
12912
12913	/*
12914	 * The read is done, now we need to send status (good or bad) back
12915	 * to the other side.
12916	 */
12917	ctl_send_datamove_done(io, /*have_lock*/ 0);
12918
12919	return (0);
12920}
12921
12922static void
12923ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12924{
12925	union ctl_io *io;
12926	void (*fe_datamove)(union ctl_io *io);
12927
12928	io = rq->context;
12929
12930	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12931		printf("%s: ISC DMA read failed with error %d", __func__,
12932		       rq->ret);
12933		ctl_set_internal_failure(&io->scsiio,
12934					 /*sks_valid*/ 1,
12935					 /*retry_count*/ rq->ret);
12936	}
12937
12938	ctl_dt_req_free(rq);
12939
12940	/* Switch the pointer over so the FETD knows what to do */
12941	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12942
12943	/*
12944	 * Use a custom move done callback, since we need to send completion
12945	 * back to the other controller, not to the backend on this side.
12946	 */
12947	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12948
12949	/* XXX KDM add checks like the ones in ctl_datamove? */
12950
12951	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12952
12953	fe_datamove(io);
12954}
12955
12956static int
12957ctl_datamove_remote_sgl_setup(union ctl_io *io)
12958{
12959	struct ctl_sg_entry *local_sglist, *remote_sglist;
12960	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12961	struct ctl_softc *softc;
12962	int retval;
12963	int i;
12964
12965	retval = 0;
12966	softc = control_softc;
12967
12968	local_sglist = io->io_hdr.local_sglist;
12969	local_dma_sglist = io->io_hdr.local_dma_sglist;
12970	remote_sglist = io->io_hdr.remote_sglist;
12971	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12972
12973	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12974		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12975			local_sglist[i].len = remote_sglist[i].len;
12976
12977			/*
12978			 * XXX Detect the situation where the RS-level I/O
12979			 * redirector on the other side has already read the
12980			 * data off of the AOR RS on this side, and
12981			 * transferred it to remote (mirror) memory on the
12982			 * other side.  Since we already have the data in
12983			 * memory here, we just need to use it.
12984			 *
12985			 * XXX KDM this can probably be removed once we
12986			 * get the cache device code in and take the
12987			 * current AOR implementation out.
12988			 */
12989#ifdef NEEDTOPORT
12990			if ((remote_sglist[i].addr >=
12991			     (void *)vtophys(softc->mirr->addr))
12992			 && (remote_sglist[i].addr <
12993			     ((void *)vtophys(softc->mirr->addr) +
12994			     CacheMirrorOffset))) {
12995				local_sglist[i].addr = remote_sglist[i].addr -
12996					CacheMirrorOffset;
12997				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12998				     CTL_FLAG_DATA_IN)
12999					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13000			} else {
13001				local_sglist[i].addr = remote_sglist[i].addr +
13002					CacheMirrorOffset;
13003			}
13004#endif
13005#if 0
13006			printf("%s: local %p, remote %p, len %d\n",
13007			       __func__, local_sglist[i].addr,
13008			       remote_sglist[i].addr, local_sglist[i].len);
13009#endif
13010		}
13011	} else {
13012		uint32_t len_to_go;
13013
13014		/*
13015		 * In this case, we don't have automatically allocated
13016		 * memory for this I/O on this controller.  This typically
13017		 * happens with internal CTL I/O -- e.g. inquiry, mode
13018		 * sense, etc.  Anything coming from RAIDCore will have
13019		 * a mirror area available.
13020		 */
13021		len_to_go = io->scsiio.kern_data_len;
13022
13023		/*
13024		 * Clear the no datasync flag, we have to use malloced
13025		 * buffers.
13026		 */
13027		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13028
13029		/*
13030		 * The difficult thing here is that the size of the various
13031		 * S/G segments may be different than the size from the
13032		 * remote controller.  That'll make it harder when DMAing
13033		 * the data back to the other side.
13034		 */
13035		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13036		     sizeof(io->io_hdr.remote_sglist[0])) &&
13037		     (len_to_go > 0); i++) {
13038			local_sglist[i].len = ctl_min(len_to_go, 131072);
13039			CTL_SIZE_8B(local_dma_sglist[i].len,
13040				    local_sglist[i].len);
13041			local_sglist[i].addr =
13042				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13043
13044			local_dma_sglist[i].addr = local_sglist[i].addr;
13045
13046			if (local_sglist[i].addr == NULL) {
13047				int j;
13048
13049				printf("malloc failed for %zd bytes!",
13050				       local_dma_sglist[i].len);
13051				for (j = 0; j < i; j++) {
13052					free(local_sglist[j].addr, M_CTL);
13053				}
13054				ctl_set_internal_failure(&io->scsiio,
13055							 /*sks_valid*/ 1,
13056							 /*retry_count*/ 4857);
13057				retval = 1;
13058				goto bailout_error;
13059
13060			}
13061			/* XXX KDM do we need a sync here? */
13062
13063			len_to_go -= local_sglist[i].len;
13064		}
13065		/*
13066		 * Reset the number of S/G entries accordingly.  The
13067		 * original number of S/G entries is available in
13068		 * rem_sg_entries.
13069		 */
13070		io->scsiio.kern_sg_entries = i;
13071
13072#if 0
13073		printf("%s: kern_sg_entries = %d\n", __func__,
13074		       io->scsiio.kern_sg_entries);
13075		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13076			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13077			       local_sglist[i].addr, local_sglist[i].len,
13078			       local_dma_sglist[i].len);
13079#endif
13080	}
13081
13082
13083	return (retval);
13084
13085bailout_error:
13086
13087	ctl_send_datamove_done(io, /*have_lock*/ 0);
13088
13089	return (retval);
13090}
13091
13092static int
13093ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13094			 ctl_ha_dt_cb callback)
13095{
13096	struct ctl_ha_dt_req *rq;
13097	struct ctl_sg_entry *remote_sglist, *local_sglist;
13098	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13099	uint32_t local_used, remote_used, total_used;
13100	int retval;
13101	int i, j;
13102
13103	retval = 0;
13104
13105	rq = ctl_dt_req_alloc();
13106
13107	/*
13108	 * If we failed to allocate the request, and if the DMA didn't fail
13109	 * anyway, set busy status.  This is just a resource allocation
13110	 * failure.
13111	 */
13112	if ((rq == NULL)
13113	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13114		ctl_set_busy(&io->scsiio);
13115
13116	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13117
13118		if (rq != NULL)
13119			ctl_dt_req_free(rq);
13120
13121		/*
13122		 * The data move failed.  We need to return status back
13123		 * to the other controller.  No point in trying to DMA
13124		 * data to the remote controller.
13125		 */
13126
13127		ctl_send_datamove_done(io, /*have_lock*/ 0);
13128
13129		retval = 1;
13130
13131		goto bailout;
13132	}
13133
13134	local_sglist = io->io_hdr.local_sglist;
13135	local_dma_sglist = io->io_hdr.local_dma_sglist;
13136	remote_sglist = io->io_hdr.remote_sglist;
13137	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13138	local_used = 0;
13139	remote_used = 0;
13140	total_used = 0;
13141
13142	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13143		rq->ret = CTL_HA_STATUS_SUCCESS;
13144		rq->context = io;
13145		callback(rq);
13146		goto bailout;
13147	}
13148
13149	/*
13150	 * Pull/push the data over the wire from/to the other controller.
13151	 * This takes into account the possibility that the local and
13152	 * remote sglists may not be identical in terms of the size of
13153	 * the elements and the number of elements.
13154	 *
13155	 * One fundamental assumption here is that the length allocated for
13156	 * both the local and remote sglists is identical.  Otherwise, we've
13157	 * essentially got a coding error of some sort.
13158	 */
13159	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13160		int isc_ret;
13161		uint32_t cur_len, dma_length;
13162		uint8_t *tmp_ptr;
13163
13164		rq->id = CTL_HA_DATA_CTL;
13165		rq->command = command;
13166		rq->context = io;
13167
13168		/*
13169		 * Both pointers should be aligned.  But it is possible
13170		 * that the allocation length is not.  They should both
13171		 * also have enough slack left over at the end, though,
13172		 * to round up to the next 8 byte boundary.
13173		 */
13174		cur_len = ctl_min(local_sglist[i].len - local_used,
13175				  remote_sglist[j].len - remote_used);
13176
13177		/*
13178		 * In this case, we have a size issue and need to decrease
13179		 * the size, except in the case where we actually have less
13180		 * than 8 bytes left.  In that case, we need to increase
13181		 * the DMA length to get the last bit.
13182		 */
13183		if ((cur_len & 0x7) != 0) {
13184			if (cur_len > 0x7) {
13185				cur_len = cur_len - (cur_len & 0x7);
13186				dma_length = cur_len;
13187			} else {
13188				CTL_SIZE_8B(dma_length, cur_len);
13189			}
13190
13191		} else
13192			dma_length = cur_len;
13193
13194		/*
13195		 * If we had to allocate memory for this I/O, instead of using
13196		 * the non-cached mirror memory, we'll need to flush the cache
13197		 * before trying to DMA to the other controller.
13198		 *
13199		 * We could end up doing this multiple times for the same
13200		 * segment if we have a larger local segment than remote
13201		 * segment.  That shouldn't be an issue.
13202		 */
13203		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13204			/*
13205			 * XXX KDM use bus_dmamap_sync() here.
13206			 */
13207		}
13208
13209		rq->size = dma_length;
13210
13211		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13212		tmp_ptr += local_used;
13213
13214		/* Use physical addresses when talking to ISC hardware */
13215		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13216			/* XXX KDM use busdma */
13217#if 0
13218			rq->local = vtophys(tmp_ptr);
13219#endif
13220		} else
13221			rq->local = tmp_ptr;
13222
13223		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13224		tmp_ptr += remote_used;
13225		rq->remote = tmp_ptr;
13226
13227		rq->callback = NULL;
13228
13229		local_used += cur_len;
13230		if (local_used >= local_sglist[i].len) {
13231			i++;
13232			local_used = 0;
13233		}
13234
13235		remote_used += cur_len;
13236		if (remote_used >= remote_sglist[j].len) {
13237			j++;
13238			remote_used = 0;
13239		}
13240		total_used += cur_len;
13241
13242		if (total_used >= io->scsiio.kern_data_len)
13243			rq->callback = callback;
13244
13245		if ((rq->size & 0x7) != 0) {
13246			printf("%s: warning: size %d is not on 8b boundary\n",
13247			       __func__, rq->size);
13248		}
13249		if (((uintptr_t)rq->local & 0x7) != 0) {
13250			printf("%s: warning: local %p not on 8b boundary\n",
13251			       __func__, rq->local);
13252		}
13253		if (((uintptr_t)rq->remote & 0x7) != 0) {
13254			printf("%s: warning: remote %p not on 8b boundary\n",
13255			       __func__, rq->local);
13256		}
13257#if 0
13258		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13259		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13260		       rq->local, rq->remote, rq->size);
13261#endif
13262
13263		isc_ret = ctl_dt_single(rq);
13264		if (isc_ret == CTL_HA_STATUS_WAIT)
13265			continue;
13266
13267		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13268			rq->ret = CTL_HA_STATUS_SUCCESS;
13269		} else {
13270			rq->ret = isc_ret;
13271		}
13272		callback(rq);
13273		goto bailout;
13274	}
13275
13276bailout:
13277	return (retval);
13278
13279}
13280
13281static void
13282ctl_datamove_remote_read(union ctl_io *io)
13283{
13284	int retval;
13285	int i;
13286
13287	/*
13288	 * This will send an error to the other controller in the case of a
13289	 * failure.
13290	 */
13291	retval = ctl_datamove_remote_sgl_setup(io);
13292	if (retval != 0)
13293		return;
13294
13295	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13296					  ctl_datamove_remote_read_cb);
13297	if ((retval != 0)
13298	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13299		/*
13300		 * Make sure we free memory if there was an error..  The
13301		 * ctl_datamove_remote_xfer() function will send the
13302		 * datamove done message, or call the callback with an
13303		 * error if there is a problem.
13304		 */
13305		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13306			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13307	}
13308
13309	return;
13310}
13311
13312/*
13313 * Process a datamove request from the other controller.  This is used for
13314 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13315 * first.  Once that is complete, the data gets DMAed into the remote
13316 * controller's memory.  For reads, we DMA from the remote controller's
13317 * memory into our memory first, and then move it out to the FETD.
13318 */
13319static void
13320ctl_datamove_remote(union ctl_io *io)
13321{
13322	struct ctl_softc *softc;
13323
13324	softc = control_softc;
13325
13326	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13327
13328	/*
13329	 * Note that we look for an aborted I/O here, but don't do some of
13330	 * the other checks that ctl_datamove() normally does.
13331	 * We don't need to run the datamove delay code, since that should
13332	 * have been done if need be on the other controller.
13333	 */
13334	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13335		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13336		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13337		       io->io_hdr.nexus.targ_port,
13338		       io->io_hdr.nexus.targ_target.id,
13339		       io->io_hdr.nexus.targ_lun);
13340		io->io_hdr.port_status = 31338;
13341		ctl_send_datamove_done(io, /*have_lock*/ 0);
13342		return;
13343	}
13344
13345	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13346		ctl_datamove_remote_write(io);
13347	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13348		ctl_datamove_remote_read(io);
13349	} else {
13350		union ctl_ha_msg msg;
13351		struct scsi_sense_data *sense;
13352		uint8_t sks[3];
13353		int retry_count;
13354
13355		memset(&msg, 0, sizeof(msg));
13356
13357		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13358		msg.hdr.status = CTL_SCSI_ERROR;
13359		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13360
13361		retry_count = 4243;
13362
13363		sense = &msg.scsi.sense_data;
13364		sks[0] = SSD_SCS_VALID;
13365		sks[1] = (retry_count >> 8) & 0xff;
13366		sks[2] = retry_count & 0xff;
13367
13368		/* "Internal target failure" */
13369		scsi_set_sense_data(sense,
13370				    /*sense_format*/ SSD_TYPE_NONE,
13371				    /*current_error*/ 1,
13372				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13373				    /*asc*/ 0x44,
13374				    /*ascq*/ 0x00,
13375				    /*type*/ SSD_ELEM_SKS,
13376				    /*size*/ sizeof(sks),
13377				    /*data*/ sks,
13378				    SSD_ELEM_NONE);
13379
13380		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13381		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13382			ctl_failover_io(io, /*have_lock*/ 1);
13383			return;
13384		}
13385
13386		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13387		    CTL_HA_STATUS_SUCCESS) {
13388			/* XXX KDM what to do if this fails? */
13389		}
13390		return;
13391	}
13392
13393}
13394
13395static int
13396ctl_process_done(union ctl_io *io)
13397{
13398	struct ctl_lun *lun;
13399	struct ctl_softc *ctl_softc = control_softc;
13400	void (*fe_done)(union ctl_io *io);
13401	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13402
13403	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13404
13405	fe_done =
13406	    control_softc->ctl_ports[targ_port]->fe_done;
13407
13408#ifdef CTL_TIME_IO
13409	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13410		char str[256];
13411		char path_str[64];
13412		struct sbuf sb;
13413
13414		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13415		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13416
13417		sbuf_cat(&sb, path_str);
13418		switch (io->io_hdr.io_type) {
13419		case CTL_IO_SCSI:
13420			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13421			sbuf_printf(&sb, "\n");
13422			sbuf_cat(&sb, path_str);
13423			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13424				    io->scsiio.tag_num, io->scsiio.tag_type);
13425			break;
13426		case CTL_IO_TASK:
13427			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13428				    "Tag Type: %d\n", io->taskio.task_action,
13429				    io->taskio.tag_num, io->taskio.tag_type);
13430			break;
13431		default:
13432			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13433			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13434			break;
13435		}
13436		sbuf_cat(&sb, path_str);
13437		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13438			    (intmax_t)time_uptime - io->io_hdr.start_time);
13439		sbuf_finish(&sb);
13440		printf("%s", sbuf_data(&sb));
13441	}
13442#endif /* CTL_TIME_IO */
13443
13444	switch (io->io_hdr.io_type) {
13445	case CTL_IO_SCSI:
13446		break;
13447	case CTL_IO_TASK:
13448		if (bootverbose || (ctl_debug & CTL_DEBUG_INFO))
13449			ctl_io_error_print(io, NULL);
13450		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13451			ctl_free_io(io);
13452		else
13453			fe_done(io);
13454		return (CTL_RETVAL_COMPLETE);
13455	default:
13456		panic("ctl_process_done: invalid io type %d\n",
13457		      io->io_hdr.io_type);
13458		break; /* NOTREACHED */
13459	}
13460
13461	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13462	if (lun == NULL) {
13463		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13464				 io->io_hdr.nexus.targ_mapped_lun));
13465		goto bailout;
13466	}
13467
13468	mtx_lock(&lun->lun_lock);
13469
13470	/*
13471	 * Check to see if we have any errors to inject here.  We only
13472	 * inject errors for commands that don't already have errors set.
13473	 */
13474	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13475	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13476	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13477		ctl_inject_error(lun, io);
13478
13479	/*
13480	 * XXX KDM how do we treat commands that aren't completed
13481	 * successfully?
13482	 *
13483	 * XXX KDM should we also track I/O latency?
13484	 */
13485	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13486	    io->io_hdr.io_type == CTL_IO_SCSI) {
13487#ifdef CTL_TIME_IO
13488		struct bintime cur_bt;
13489#endif
13490		int type;
13491
13492		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13493		    CTL_FLAG_DATA_IN)
13494			type = CTL_STATS_READ;
13495		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13496		    CTL_FLAG_DATA_OUT)
13497			type = CTL_STATS_WRITE;
13498		else
13499			type = CTL_STATS_NO_IO;
13500
13501		lun->stats.ports[targ_port].bytes[type] +=
13502		    io->scsiio.kern_total_len;
13503		lun->stats.ports[targ_port].operations[type]++;
13504#ifdef CTL_TIME_IO
13505		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13506		   &io->io_hdr.dma_bt);
13507		lun->stats.ports[targ_port].num_dmas[type] +=
13508		    io->io_hdr.num_dmas;
13509		getbintime(&cur_bt);
13510		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13511		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13512#endif
13513	}
13514
13515	/*
13516	 * Remove this from the OOA queue.
13517	 */
13518	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13519
13520	/*
13521	 * Run through the blocked queue on this LUN and see if anything
13522	 * has become unblocked, now that this transaction is done.
13523	 */
13524	ctl_check_blocked(lun);
13525
13526	/*
13527	 * If the LUN has been invalidated, free it if there is nothing
13528	 * left on its OOA queue.
13529	 */
13530	if ((lun->flags & CTL_LUN_INVALID)
13531	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13532		mtx_unlock(&lun->lun_lock);
13533		mtx_lock(&ctl_softc->ctl_lock);
13534		ctl_free_lun(lun);
13535		mtx_unlock(&ctl_softc->ctl_lock);
13536	} else
13537		mtx_unlock(&lun->lun_lock);
13538
13539bailout:
13540
13541	/*
13542	 * If this command has been aborted, make sure we set the status
13543	 * properly.  The FETD is responsible for freeing the I/O and doing
13544	 * whatever it needs to do to clean up its state.
13545	 */
13546	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13547		ctl_set_task_aborted(&io->scsiio);
13548
13549	/*
13550	 * If enabled, print command error status.
13551	 * We don't print UAs unless debugging was enabled explicitly.
13552	 */
13553	do {
13554		if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
13555			break;
13556		if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0)
13557			break;
13558		if ((ctl_debug & CTL_DEBUG_INFO) == 0 &&
13559		    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) &&
13560		     (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13561			int error_code, sense_key, asc, ascq;
13562
13563			scsi_extract_sense_len(&io->scsiio.sense_data,
13564			    io->scsiio.sense_len, &error_code, &sense_key,
13565			    &asc, &ascq, /*show_errors*/ 0);
13566			if (sense_key == SSD_KEY_UNIT_ATTENTION)
13567				break;
13568		}
13569
13570		ctl_io_error_print(io, NULL);
13571	} while (0);
13572
13573	/*
13574	 * Tell the FETD or the other shelf controller we're done with this
13575	 * command.  Note that only SCSI commands get to this point.  Task
13576	 * management commands are completed above.
13577	 *
13578	 * We only send status to the other controller if we're in XFER
13579	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13580	 * received the I/O (from CTL's perspective), and so the status is
13581	 * generated there.
13582	 *
13583	 * XXX KDM if we hold the lock here, we could cause a deadlock
13584	 * if the frontend comes back in in this context to queue
13585	 * something.
13586	 */
13587	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13588	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13589		union ctl_ha_msg msg;
13590
13591		memset(&msg, 0, sizeof(msg));
13592		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13593		msg.hdr.original_sc = io->io_hdr.original_sc;
13594		msg.hdr.nexus = io->io_hdr.nexus;
13595		msg.hdr.status = io->io_hdr.status;
13596		msg.scsi.scsi_status = io->scsiio.scsi_status;
13597		msg.scsi.tag_num = io->scsiio.tag_num;
13598		msg.scsi.tag_type = io->scsiio.tag_type;
13599		msg.scsi.sense_len = io->scsiio.sense_len;
13600		msg.scsi.sense_residual = io->scsiio.sense_residual;
13601		msg.scsi.residual = io->scsiio.residual;
13602		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13603		       sizeof(io->scsiio.sense_data));
13604		/*
13605		 * We copy this whether or not this is an I/O-related
13606		 * command.  Otherwise, we'd have to go and check to see
13607		 * whether it's a read/write command, and it really isn't
13608		 * worth it.
13609		 */
13610		memcpy(&msg.scsi.lbalen,
13611		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13612		       sizeof(msg.scsi.lbalen));
13613
13614		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13615				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13616			/* XXX do something here */
13617		}
13618
13619		ctl_free_io(io);
13620	} else
13621		fe_done(io);
13622
13623	return (CTL_RETVAL_COMPLETE);
13624}
13625
13626#ifdef CTL_WITH_CA
13627/*
13628 * Front end should call this if it doesn't do autosense.  When the request
13629 * sense comes back in from the initiator, we'll dequeue this and send it.
13630 */
13631int
13632ctl_queue_sense(union ctl_io *io)
13633{
13634	struct ctl_lun *lun;
13635	struct ctl_softc *ctl_softc;
13636	uint32_t initidx, targ_lun;
13637
13638	ctl_softc = control_softc;
13639
13640	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13641
13642	/*
13643	 * LUN lookup will likely move to the ctl_work_thread() once we
13644	 * have our new queueing infrastructure (that doesn't put things on
13645	 * a per-LUN queue initially).  That is so that we can handle
13646	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13647	 * can't deal with that right now.
13648	 */
13649	mtx_lock(&ctl_softc->ctl_lock);
13650
13651	/*
13652	 * If we don't have a LUN for this, just toss the sense
13653	 * information.
13654	 */
13655	targ_lun = io->io_hdr.nexus.targ_lun;
13656	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13657	if ((targ_lun < CTL_MAX_LUNS)
13658	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13659		lun = ctl_softc->ctl_luns[targ_lun];
13660	else
13661		goto bailout;
13662
13663	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13664
13665	mtx_lock(&lun->lun_lock);
13666	/*
13667	 * Already have CA set for this LUN...toss the sense information.
13668	 */
13669	if (ctl_is_set(lun->have_ca, initidx)) {
13670		mtx_unlock(&lun->lun_lock);
13671		goto bailout;
13672	}
13673
13674	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13675	       ctl_min(sizeof(lun->pending_sense[initidx]),
13676	       sizeof(io->scsiio.sense_data)));
13677	ctl_set_mask(lun->have_ca, initidx);
13678	mtx_unlock(&lun->lun_lock);
13679
13680bailout:
13681	mtx_unlock(&ctl_softc->ctl_lock);
13682
13683	ctl_free_io(io);
13684
13685	return (CTL_RETVAL_COMPLETE);
13686}
13687#endif
13688
13689/*
13690 * Primary command inlet from frontend ports.  All SCSI and task I/O
13691 * requests must go through this function.
13692 */
13693int
13694ctl_queue(union ctl_io *io)
13695{
13696	struct ctl_softc *ctl_softc;
13697
13698	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13699
13700	ctl_softc = control_softc;
13701
13702#ifdef CTL_TIME_IO
13703	io->io_hdr.start_time = time_uptime;
13704	getbintime(&io->io_hdr.start_bt);
13705#endif /* CTL_TIME_IO */
13706
13707	/* Map FE-specific LUN ID into global one. */
13708	io->io_hdr.nexus.targ_mapped_lun =
13709	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13710
13711	switch (io->io_hdr.io_type) {
13712	case CTL_IO_SCSI:
13713	case CTL_IO_TASK:
13714		if (ctl_debug & CTL_DEBUG_CDB)
13715			ctl_io_print(io);
13716		ctl_enqueue_incoming(io);
13717		break;
13718	default:
13719		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13720		return (EINVAL);
13721	}
13722
13723	return (CTL_RETVAL_COMPLETE);
13724}
13725
13726#ifdef CTL_IO_DELAY
13727static void
13728ctl_done_timer_wakeup(void *arg)
13729{
13730	union ctl_io *io;
13731
13732	io = (union ctl_io *)arg;
13733	ctl_done(io);
13734}
13735#endif /* CTL_IO_DELAY */
13736
13737void
13738ctl_done(union ctl_io *io)
13739{
13740	struct ctl_softc *ctl_softc;
13741
13742	ctl_softc = control_softc;
13743
13744	/*
13745	 * Enable this to catch duplicate completion issues.
13746	 */
13747#if 0
13748	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13749		printf("%s: type %d msg %d cdb %x iptl: "
13750		       "%d:%d:%d:%d tag 0x%04x "
13751		       "flag %#x status %x\n",
13752			__func__,
13753			io->io_hdr.io_type,
13754			io->io_hdr.msg_type,
13755			io->scsiio.cdb[0],
13756			io->io_hdr.nexus.initid.id,
13757			io->io_hdr.nexus.targ_port,
13758			io->io_hdr.nexus.targ_target.id,
13759			io->io_hdr.nexus.targ_lun,
13760			(io->io_hdr.io_type ==
13761			CTL_IO_TASK) ?
13762			io->taskio.tag_num :
13763			io->scsiio.tag_num,
13764		        io->io_hdr.flags,
13765			io->io_hdr.status);
13766	} else
13767		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13768#endif
13769
13770	/*
13771	 * This is an internal copy of an I/O, and should not go through
13772	 * the normal done processing logic.
13773	 */
13774	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13775		return;
13776
13777	/*
13778	 * We need to send a msg to the serializing shelf to finish the IO
13779	 * as well.  We don't send a finish message to the other shelf if
13780	 * this is a task management command.  Task management commands
13781	 * aren't serialized in the OOA queue, but rather just executed on
13782	 * both shelf controllers for commands that originated on that
13783	 * controller.
13784	 */
13785	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13786	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13787		union ctl_ha_msg msg_io;
13788
13789		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13790		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13791		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13792		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13793		}
13794		/* continue on to finish IO */
13795	}
13796#ifdef CTL_IO_DELAY
13797	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13798		struct ctl_lun *lun;
13799
13800		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13801
13802		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13803	} else {
13804		struct ctl_lun *lun;
13805
13806		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13807
13808		if ((lun != NULL)
13809		 && (lun->delay_info.done_delay > 0)) {
13810			struct callout *callout;
13811
13812			callout = (struct callout *)&io->io_hdr.timer_bytes;
13813			callout_init(callout, /*mpsafe*/ 1);
13814			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13815			callout_reset(callout,
13816				      lun->delay_info.done_delay * hz,
13817				      ctl_done_timer_wakeup, io);
13818			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13819				lun->delay_info.done_delay = 0;
13820			return;
13821		}
13822	}
13823#endif /* CTL_IO_DELAY */
13824
13825	ctl_enqueue_done(io);
13826}
13827
13828int
13829ctl_isc(struct ctl_scsiio *ctsio)
13830{
13831	struct ctl_lun *lun;
13832	int retval;
13833
13834	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13835
13836	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13837
13838	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13839
13840	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13841
13842	return (retval);
13843}
13844
13845
13846static void
13847ctl_work_thread(void *arg)
13848{
13849	struct ctl_thread *thr = (struct ctl_thread *)arg;
13850	struct ctl_softc *softc = thr->ctl_softc;
13851	union ctl_io *io;
13852	int retval;
13853
13854	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13855
13856	for (;;) {
13857		retval = 0;
13858
13859		/*
13860		 * We handle the queues in this order:
13861		 * - ISC
13862		 * - done queue (to free up resources, unblock other commands)
13863		 * - RtR queue
13864		 * - incoming queue
13865		 *
13866		 * If those queues are empty, we break out of the loop and
13867		 * go to sleep.
13868		 */
13869		mtx_lock(&thr->queue_lock);
13870		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13871		if (io != NULL) {
13872			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13873			mtx_unlock(&thr->queue_lock);
13874			ctl_handle_isc(io);
13875			continue;
13876		}
13877		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13878		if (io != NULL) {
13879			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13880			/* clear any blocked commands, call fe_done */
13881			mtx_unlock(&thr->queue_lock);
13882			retval = ctl_process_done(io);
13883			continue;
13884		}
13885		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13886		if (io != NULL) {
13887			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13888			mtx_unlock(&thr->queue_lock);
13889			if (io->io_hdr.io_type == CTL_IO_TASK)
13890				ctl_run_task(io);
13891			else
13892				ctl_scsiio_precheck(softc, &io->scsiio);
13893			continue;
13894		}
13895		if (!ctl_pause_rtr) {
13896			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13897			if (io != NULL) {
13898				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13899				mtx_unlock(&thr->queue_lock);
13900				retval = ctl_scsiio(&io->scsiio);
13901				if (retval != CTL_RETVAL_COMPLETE)
13902					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13903				continue;
13904			}
13905		}
13906
13907		/* Sleep until we have something to do. */
13908		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13909	}
13910}
13911
13912static void
13913ctl_lun_thread(void *arg)
13914{
13915	struct ctl_softc *softc = (struct ctl_softc *)arg;
13916	struct ctl_be_lun *be_lun;
13917	int retval;
13918
13919	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13920
13921	for (;;) {
13922		retval = 0;
13923		mtx_lock(&softc->ctl_lock);
13924		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13925		if (be_lun != NULL) {
13926			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13927			mtx_unlock(&softc->ctl_lock);
13928			ctl_create_lun(be_lun);
13929			continue;
13930		}
13931
13932		/* Sleep until we have something to do. */
13933		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13934		    PDROP | PRIBIO, "-", 0);
13935	}
13936}
13937
13938static void
13939ctl_thresh_thread(void *arg)
13940{
13941	struct ctl_softc *softc = (struct ctl_softc *)arg;
13942	struct ctl_lun *lun;
13943	struct ctl_be_lun *be_lun;
13944	struct scsi_da_rw_recovery_page *rwpage;
13945	struct ctl_logical_block_provisioning_page *page;
13946	const char *attr;
13947	uint64_t thres, val;
13948	int i, e;
13949
13950	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13951
13952	for (;;) {
13953		mtx_lock(&softc->ctl_lock);
13954		STAILQ_FOREACH(lun, &softc->lun_list, links) {
13955			be_lun = lun->be_lun;
13956			if ((lun->flags & CTL_LUN_DISABLED) ||
13957			    (lun->flags & CTL_LUN_OFFLINE) ||
13958			    (be_lun->flags & CTL_LUN_FLAG_UNMAP) == 0 ||
13959			    lun->backend->lun_attr == NULL)
13960				continue;
13961			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13962			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13963				continue;
13964			e = 0;
13965			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13966			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13967				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13968					continue;
13969				thres = scsi_4btoul(page->descr[i].count);
13970				thres <<= CTL_LBP_EXPONENT;
13971				switch (page->descr[i].resource) {
13972				case 0x01:
13973					attr = "blocksavail";
13974					break;
13975				case 0x02:
13976					attr = "blocksused";
13977					break;
13978				case 0xf1:
13979					attr = "poolblocksavail";
13980					break;
13981				case 0xf2:
13982					attr = "poolblocksused";
13983					break;
13984				default:
13985					continue;
13986				}
13987				mtx_unlock(&softc->ctl_lock); // XXX
13988				val = lun->backend->lun_attr(
13989				    lun->be_lun->be_lun, attr);
13990				mtx_lock(&softc->ctl_lock);
13991				if (val == UINT64_MAX)
13992					continue;
13993				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13994				    == SLBPPD_ARMING_INC)
13995					e |= (val >= thres);
13996				else
13997					e |= (val <= thres);
13998			}
13999			mtx_lock(&lun->lun_lock);
14000			if (e) {
14001				if (lun->lasttpt == 0 ||
14002				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
14003					lun->lasttpt = time_uptime;
14004					for (i = 0; i < CTL_MAX_INITIATORS; i++)
14005						lun->pending_ua[i] |=
14006						    CTL_UA_THIN_PROV_THRES;
14007				}
14008			} else {
14009				lun->lasttpt = 0;
14010				for (i = 0; i < CTL_MAX_INITIATORS; i++)
14011					lun->pending_ua[i] &= ~CTL_UA_THIN_PROV_THRES;
14012			}
14013			mtx_unlock(&lun->lun_lock);
14014		}
14015		mtx_unlock(&softc->ctl_lock);
14016		pause("-", CTL_LBP_PERIOD * hz);
14017	}
14018}
14019
14020static void
14021ctl_enqueue_incoming(union ctl_io *io)
14022{
14023	struct ctl_softc *softc = control_softc;
14024	struct ctl_thread *thr;
14025	u_int idx;
14026
14027	idx = (io->io_hdr.nexus.targ_port * 127 +
14028	       io->io_hdr.nexus.initid.id) % worker_threads;
14029	thr = &softc->threads[idx];
14030	mtx_lock(&thr->queue_lock);
14031	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14032	mtx_unlock(&thr->queue_lock);
14033	wakeup(thr);
14034}
14035
14036static void
14037ctl_enqueue_rtr(union ctl_io *io)
14038{
14039	struct ctl_softc *softc = control_softc;
14040	struct ctl_thread *thr;
14041
14042	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14043	mtx_lock(&thr->queue_lock);
14044	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14045	mtx_unlock(&thr->queue_lock);
14046	wakeup(thr);
14047}
14048
14049static void
14050ctl_enqueue_done(union ctl_io *io)
14051{
14052	struct ctl_softc *softc = control_softc;
14053	struct ctl_thread *thr;
14054
14055	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14056	mtx_lock(&thr->queue_lock);
14057	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14058	mtx_unlock(&thr->queue_lock);
14059	wakeup(thr);
14060}
14061
14062static void
14063ctl_enqueue_isc(union ctl_io *io)
14064{
14065	struct ctl_softc *softc = control_softc;
14066	struct ctl_thread *thr;
14067
14068	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14069	mtx_lock(&thr->queue_lock);
14070	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14071	mtx_unlock(&thr->queue_lock);
14072	wakeup(thr);
14073}
14074
14075/* Initialization and failover */
14076
14077void
14078ctl_init_isc_msg(void)
14079{
14080	printf("CTL: Still calling this thing\n");
14081}
14082
14083/*
14084 * Init component
14085 * 	Initializes component into configuration defined by bootMode
14086 *	(see hasc-sv.c)
14087 *  	returns hasc_Status:
14088 * 		OK
14089 *		ERROR - fatal error
14090 */
14091static ctl_ha_comp_status
14092ctl_isc_init(struct ctl_ha_component *c)
14093{
14094	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14095
14096	c->status = ret;
14097	return ret;
14098}
14099
14100/* Start component
14101 * 	Starts component in state requested. If component starts successfully,
14102 *	it must set its own state to the requestrd state
14103 *	When requested state is HASC_STATE_HA, the component may refine it
14104 * 	by adding _SLAVE or _MASTER flags.
14105 *	Currently allowed state transitions are:
14106 *	UNKNOWN->HA		- initial startup
14107 *	UNKNOWN->SINGLE - initial startup when no parter detected
14108 *	HA->SINGLE		- failover
14109 * returns ctl_ha_comp_status:
14110 * 		OK	- component successfully started in requested state
14111 *		FAILED  - could not start the requested state, failover may
14112 * 			  be possible
14113 *		ERROR	- fatal error detected, no future startup possible
14114 */
14115static ctl_ha_comp_status
14116ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14117{
14118	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14119
14120	printf("%s: go\n", __func__);
14121
14122	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14123	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14124		control_softc->is_single = 0;
14125		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14126		    != CTL_HA_STATUS_SUCCESS) {
14127			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14128			ret = CTL_HA_COMP_STATUS_ERROR;
14129		}
14130	} else if (CTL_HA_STATE_IS_HA(c->state)
14131		&& CTL_HA_STATE_IS_SINGLE(state)){
14132		// HA->SINGLE transition
14133	        ctl_failover();
14134		control_softc->is_single = 1;
14135	} else {
14136		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14137		       c->state, state);
14138		ret = CTL_HA_COMP_STATUS_ERROR;
14139	}
14140	if (CTL_HA_STATE_IS_SINGLE(state))
14141		control_softc->is_single = 1;
14142
14143	c->state = state;
14144	c->status = ret;
14145	return ret;
14146}
14147
14148/*
14149 * Quiesce component
14150 * The component must clear any error conditions (set status to OK) and
14151 * prepare itself to another Start call
14152 * returns ctl_ha_comp_status:
14153 * 	OK
14154 *	ERROR
14155 */
14156static ctl_ha_comp_status
14157ctl_isc_quiesce(struct ctl_ha_component *c)
14158{
14159	int ret = CTL_HA_COMP_STATUS_OK;
14160
14161	ctl_pause_rtr = 1;
14162	c->status = ret;
14163	return ret;
14164}
14165
14166struct ctl_ha_component ctl_ha_component_ctlisc =
14167{
14168	.name = "CTL ISC",
14169	.state = CTL_HA_STATE_UNKNOWN,
14170	.init = ctl_isc_init,
14171	.start = ctl_isc_start,
14172	.quiesce = ctl_isc_quiesce
14173};
14174
14175/*
14176 *  vim: ts=8
14177 */
14178