ctl.c revision 275920
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 275920 2014-12-18 22:32:22Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/ctype.h>
50#include <sys/kernel.h>
51#include <sys/types.h>
52#include <sys/kthread.h>
53#include <sys/bio.h>
54#include <sys/fcntl.h>
55#include <sys/lock.h>
56#include <sys/module.h>
57#include <sys/mutex.h>
58#include <sys/condvar.h>
59#include <sys/malloc.h>
60#include <sys/conf.h>
61#include <sys/ioccom.h>
62#include <sys/queue.h>
63#include <sys/sbuf.h>
64#include <sys/smp.h>
65#include <sys/endian.h>
66#include <sys/sysctl.h>
67#include <vm/uma.h>
68
69#include <cam/cam.h>
70#include <cam/scsi/scsi_all.h>
71#include <cam/scsi/scsi_da.h>
72#include <cam/ctl/ctl_io.h>
73#include <cam/ctl/ctl.h>
74#include <cam/ctl/ctl_frontend.h>
75#include <cam/ctl/ctl_frontend_internal.h>
76#include <cam/ctl/ctl_util.h>
77#include <cam/ctl/ctl_backend.h>
78#include <cam/ctl/ctl_ioctl.h>
79#include <cam/ctl/ctl_ha.h>
80#include <cam/ctl/ctl_private.h>
81#include <cam/ctl/ctl_debug.h>
82#include <cam/ctl/ctl_scsi_all.h>
83#include <cam/ctl/ctl_error.h>
84
85struct ctl_softc *control_softc = NULL;
86
87/*
88 * Size and alignment macros needed for Copan-specific HA hardware.  These
89 * can go away when the HA code is re-written, and uses busdma for any
90 * hardware.
91 */
92#define	CTL_ALIGN_8B(target, source, type)				\
93	if (((uint32_t)source & 0x7) != 0)				\
94		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
95	else								\
96		target = (type)source;
97
98#define	CTL_SIZE_8B(target, size)					\
99	if ((size & 0x7) != 0)						\
100		target = size + (0x8 - (size & 0x7));			\
101	else								\
102		target = size;
103
104#define CTL_ALIGN_8B_MARGIN	16
105
106/*
107 * Template mode pages.
108 */
109
110/*
111 * Note that these are default values only.  The actual values will be
112 * filled in when the user does a mode sense.
113 */
114static struct copan_debugconf_subpage debugconf_page_default = {
115	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
116	DBGCNF_SUBPAGE_CODE,		/* subpage */
117	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
118	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
119	DBGCNF_VERSION,			/* page_version */
120	{CTL_TIME_IO_DEFAULT_SECS>>8,
121	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
122};
123
124static struct copan_debugconf_subpage debugconf_page_changeable = {
125	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
126	DBGCNF_SUBPAGE_CODE,		/* subpage */
127	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
128	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
129	0,				/* page_version */
130	{0xff,0xff},			/* ctl_time_io_secs */
131};
132
133static struct scsi_da_rw_recovery_page rw_er_page_default = {
134	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
135	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
136	/*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
137	/*read_retry_count*/0,
138	/*correction_span*/0,
139	/*head_offset_count*/0,
140	/*data_strobe_offset_cnt*/0,
141	/*byte8*/SMS_RWER_LBPERE,
142	/*write_retry_count*/0,
143	/*reserved2*/0,
144	/*recovery_time_limit*/{0, 0},
145};
146
147static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
148	/*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
149	/*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
150	/*byte3*/0,
151	/*read_retry_count*/0,
152	/*correction_span*/0,
153	/*head_offset_count*/0,
154	/*data_strobe_offset_cnt*/0,
155	/*byte8*/0,
156	/*write_retry_count*/0,
157	/*reserved2*/0,
158	/*recovery_time_limit*/{0, 0},
159};
160
161static struct scsi_format_page format_page_default = {
162	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
163	/*page_length*/sizeof(struct scsi_format_page) - 2,
164	/*tracks_per_zone*/ {0, 0},
165	/*alt_sectors_per_zone*/ {0, 0},
166	/*alt_tracks_per_zone*/ {0, 0},
167	/*alt_tracks_per_lun*/ {0, 0},
168	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
169			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
170	/*bytes_per_sector*/ {0, 0},
171	/*interleave*/ {0, 0},
172	/*track_skew*/ {0, 0},
173	/*cylinder_skew*/ {0, 0},
174	/*flags*/ SFP_HSEC,
175	/*reserved*/ {0, 0, 0}
176};
177
178static struct scsi_format_page format_page_changeable = {
179	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
180	/*page_length*/sizeof(struct scsi_format_page) - 2,
181	/*tracks_per_zone*/ {0, 0},
182	/*alt_sectors_per_zone*/ {0, 0},
183	/*alt_tracks_per_zone*/ {0, 0},
184	/*alt_tracks_per_lun*/ {0, 0},
185	/*sectors_per_track*/ {0, 0},
186	/*bytes_per_sector*/ {0, 0},
187	/*interleave*/ {0, 0},
188	/*track_skew*/ {0, 0},
189	/*cylinder_skew*/ {0, 0},
190	/*flags*/ 0,
191	/*reserved*/ {0, 0, 0}
192};
193
194static struct scsi_rigid_disk_page rigid_disk_page_default = {
195	/*page_code*/SMS_RIGID_DISK_PAGE,
196	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
197	/*cylinders*/ {0, 0, 0},
198	/*heads*/ CTL_DEFAULT_HEADS,
199	/*start_write_precomp*/ {0, 0, 0},
200	/*start_reduced_current*/ {0, 0, 0},
201	/*step_rate*/ {0, 0},
202	/*landing_zone_cylinder*/ {0, 0, 0},
203	/*rpl*/ SRDP_RPL_DISABLED,
204	/*rotational_offset*/ 0,
205	/*reserved1*/ 0,
206	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
207			   CTL_DEFAULT_ROTATION_RATE & 0xff},
208	/*reserved2*/ {0, 0}
209};
210
211static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
212	/*page_code*/SMS_RIGID_DISK_PAGE,
213	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
214	/*cylinders*/ {0, 0, 0},
215	/*heads*/ 0,
216	/*start_write_precomp*/ {0, 0, 0},
217	/*start_reduced_current*/ {0, 0, 0},
218	/*step_rate*/ {0, 0},
219	/*landing_zone_cylinder*/ {0, 0, 0},
220	/*rpl*/ 0,
221	/*rotational_offset*/ 0,
222	/*reserved1*/ 0,
223	/*rotation_rate*/ {0, 0},
224	/*reserved2*/ {0, 0}
225};
226
227static struct scsi_caching_page caching_page_default = {
228	/*page_code*/SMS_CACHING_PAGE,
229	/*page_length*/sizeof(struct scsi_caching_page) - 2,
230	/*flags1*/ SCP_DISC | SCP_WCE,
231	/*ret_priority*/ 0,
232	/*disable_pf_transfer_len*/ {0xff, 0xff},
233	/*min_prefetch*/ {0, 0},
234	/*max_prefetch*/ {0xff, 0xff},
235	/*max_pf_ceiling*/ {0xff, 0xff},
236	/*flags2*/ 0,
237	/*cache_segments*/ 0,
238	/*cache_seg_size*/ {0, 0},
239	/*reserved*/ 0,
240	/*non_cache_seg_size*/ {0, 0, 0}
241};
242
243static struct scsi_caching_page caching_page_changeable = {
244	/*page_code*/SMS_CACHING_PAGE,
245	/*page_length*/sizeof(struct scsi_caching_page) - 2,
246	/*flags1*/ SCP_WCE | SCP_RCD,
247	/*ret_priority*/ 0,
248	/*disable_pf_transfer_len*/ {0, 0},
249	/*min_prefetch*/ {0, 0},
250	/*max_prefetch*/ {0, 0},
251	/*max_pf_ceiling*/ {0, 0},
252	/*flags2*/ 0,
253	/*cache_segments*/ 0,
254	/*cache_seg_size*/ {0, 0},
255	/*reserved*/ 0,
256	/*non_cache_seg_size*/ {0, 0, 0}
257};
258
259static struct scsi_control_page control_page_default = {
260	/*page_code*/SMS_CONTROL_MODE_PAGE,
261	/*page_length*/sizeof(struct scsi_control_page) - 2,
262	/*rlec*/0,
263	/*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
264	/*eca_and_aen*/0,
265	/*flags4*/SCP_TAS,
266	/*aen_holdoff_period*/{0, 0},
267	/*busy_timeout_period*/{0, 0},
268	/*extended_selftest_completion_time*/{0, 0}
269};
270
271static struct scsi_control_page control_page_changeable = {
272	/*page_code*/SMS_CONTROL_MODE_PAGE,
273	/*page_length*/sizeof(struct scsi_control_page) - 2,
274	/*rlec*/SCP_DSENSE,
275	/*queue_flags*/SCP_QUEUE_ALG_MASK,
276	/*eca_and_aen*/SCP_SWP,
277	/*flags4*/0,
278	/*aen_holdoff_period*/{0, 0},
279	/*busy_timeout_period*/{0, 0},
280	/*extended_selftest_completion_time*/{0, 0}
281};
282
283static struct scsi_info_exceptions_page ie_page_default = {
284	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
285	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
286	/*info_flags*/SIEP_FLAGS_DEXCPT,
287	/*mrie*/0,
288	/*interval_timer*/{0, 0, 0, 0},
289	/*report_count*/{0, 0, 0, 0}
290};
291
292static struct scsi_info_exceptions_page ie_page_changeable = {
293	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
294	/*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
295	/*info_flags*/0,
296	/*mrie*/0,
297	/*interval_timer*/{0, 0, 0, 0},
298	/*report_count*/{0, 0, 0, 0}
299};
300
301#define CTL_LBPM_LEN	(sizeof(struct ctl_logical_block_provisioning_page) - 4)
302
303static struct ctl_logical_block_provisioning_page lbp_page_default = {{
304	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
305	/*subpage_code*/0x02,
306	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
307	/*flags*/0,
308	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
309	/*descr*/{}},
310	{{/*flags*/0,
311	  /*resource*/0x01,
312	  /*reserved*/{0, 0},
313	  /*count*/{0, 0, 0, 0}},
314	 {/*flags*/0,
315	  /*resource*/0x02,
316	  /*reserved*/{0, 0},
317	  /*count*/{0, 0, 0, 0}},
318	 {/*flags*/0,
319	  /*resource*/0xf1,
320	  /*reserved*/{0, 0},
321	  /*count*/{0, 0, 0, 0}},
322	 {/*flags*/0,
323	  /*resource*/0xf2,
324	  /*reserved*/{0, 0},
325	  /*count*/{0, 0, 0, 0}}
326	}
327};
328
329static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
330	/*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
331	/*subpage_code*/0x02,
332	/*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
333	/*flags*/0,
334	/*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
335	/*descr*/{}},
336	{{/*flags*/0,
337	  /*resource*/0,
338	  /*reserved*/{0, 0},
339	  /*count*/{0, 0, 0, 0}},
340	 {/*flags*/0,
341	  /*resource*/0,
342	  /*reserved*/{0, 0},
343	  /*count*/{0, 0, 0, 0}},
344	 {/*flags*/0,
345	  /*resource*/0,
346	  /*reserved*/{0, 0},
347	  /*count*/{0, 0, 0, 0}},
348	 {/*flags*/0,
349	  /*resource*/0,
350	  /*reserved*/{0, 0},
351	  /*count*/{0, 0, 0, 0}}
352	}
353};
354
355/*
356 * XXX KDM move these into the softc.
357 */
358static int rcv_sync_msg;
359static uint8_t ctl_pause_rtr;
360
361SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
362static int worker_threads = -1;
363SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
364    &worker_threads, 1, "Number of worker threads");
365static int ctl_debug = CTL_DEBUG_NONE;
366SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
367    &ctl_debug, 0, "Enabled debug flags");
368
369/*
370 * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
371 * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
372 * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
373 * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
374 */
375#define SCSI_EVPD_NUM_SUPPORTED_PAGES	10
376
377static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
378				  int param);
379static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
380static int ctl_init(void);
381void ctl_shutdown(void);
382static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
383static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
384static void ctl_ioctl_online(void *arg);
385static void ctl_ioctl_offline(void *arg);
386static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
387static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
388static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
389static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
390static int ctl_ioctl_submit_wait(union ctl_io *io);
391static void ctl_ioctl_datamove(union ctl_io *io);
392static void ctl_ioctl_done(union ctl_io *io);
393static void ctl_ioctl_hard_startstop_callback(void *arg,
394					      struct cfi_metatask *metatask);
395static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
396static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
397			      struct ctl_ooa *ooa_hdr,
398			      struct ctl_ooa_entry *kern_entries);
399static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
400		     struct thread *td);
401static uint32_t ctl_map_lun(int port_num, uint32_t lun);
402static uint32_t ctl_map_lun_back(int port_num, uint32_t lun);
403static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
404			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
405static int ctl_free_lun(struct ctl_lun *lun);
406static void ctl_create_lun(struct ctl_be_lun *be_lun);
407/**
408static void ctl_failover_change_pages(struct ctl_softc *softc,
409				      struct ctl_scsiio *ctsio, int master);
410**/
411
412static int ctl_do_mode_select(union ctl_io *io);
413static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
414			   uint64_t res_key, uint64_t sa_res_key,
415			   uint8_t type, uint32_t residx,
416			   struct ctl_scsiio *ctsio,
417			   struct scsi_per_res_out *cdb,
418			   struct scsi_per_res_out_parms* param);
419static void ctl_pro_preempt_other(struct ctl_lun *lun,
420				  union ctl_ha_msg *msg);
421static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
422static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
423static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
424static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
425static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
426static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
427static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
428					 int alloc_len);
429static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
430					 int alloc_len);
431static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
432static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
433static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
434static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
435static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
436static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
437    bool seq);
438static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
439static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
440    union ctl_io *pending_io, union ctl_io *ooa_io);
441static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
442				union ctl_io *starting_io);
443static int ctl_check_blocked(struct ctl_lun *lun);
444static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
445				struct ctl_lun *lun,
446				const struct ctl_cmd_entry *entry,
447				struct ctl_scsiio *ctsio);
448//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
449static void ctl_failover(void);
450static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
451			       struct ctl_scsiio *ctsio);
452static int ctl_scsiio(struct ctl_scsiio *ctsio);
453
454static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
455static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
456			    ctl_ua_type ua_type);
457static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
458			 ctl_ua_type ua_type);
459static int ctl_abort_task(union ctl_io *io);
460static int ctl_abort_task_set(union ctl_io *io);
461static int ctl_i_t_nexus_reset(union ctl_io *io);
462static void ctl_run_task(union ctl_io *io);
463#ifdef CTL_IO_DELAY
464static void ctl_datamove_timer_wakeup(void *arg);
465static void ctl_done_timer_wakeup(void *arg);
466#endif /* CTL_IO_DELAY */
467
468static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
469static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
470static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
471static void ctl_datamove_remote_write(union ctl_io *io);
472static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
473static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
474static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
475static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
476				    ctl_ha_dt_cb callback);
477static void ctl_datamove_remote_read(union ctl_io *io);
478static void ctl_datamove_remote(union ctl_io *io);
479static int ctl_process_done(union ctl_io *io);
480static void ctl_lun_thread(void *arg);
481static void ctl_thresh_thread(void *arg);
482static void ctl_work_thread(void *arg);
483static void ctl_enqueue_incoming(union ctl_io *io);
484static void ctl_enqueue_rtr(union ctl_io *io);
485static void ctl_enqueue_done(union ctl_io *io);
486static void ctl_enqueue_isc(union ctl_io *io);
487static const struct ctl_cmd_entry *
488    ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
489static const struct ctl_cmd_entry *
490    ctl_validate_command(struct ctl_scsiio *ctsio);
491static int ctl_cmd_applicable(uint8_t lun_type,
492    const struct ctl_cmd_entry *entry);
493
494/*
495 * Load the serialization table.  This isn't very pretty, but is probably
496 * the easiest way to do it.
497 */
498#include "ctl_ser_table.c"
499
500/*
501 * We only need to define open, close and ioctl routines for this driver.
502 */
503static struct cdevsw ctl_cdevsw = {
504	.d_version =	D_VERSION,
505	.d_flags =	0,
506	.d_open =	ctl_open,
507	.d_close =	ctl_close,
508	.d_ioctl =	ctl_ioctl,
509	.d_name =	"ctl",
510};
511
512
513MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
514MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests");
515
516static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
517
518static moduledata_t ctl_moduledata = {
519	"ctl",
520	ctl_module_event_handler,
521	NULL
522};
523
524DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
525MODULE_VERSION(ctl, 1);
526
527static struct ctl_frontend ioctl_frontend =
528{
529	.name = "ioctl",
530};
531
532static void
533ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
534			    union ctl_ha_msg *msg_info)
535{
536	struct ctl_scsiio *ctsio;
537
538	if (msg_info->hdr.original_sc == NULL) {
539		printf("%s: original_sc == NULL!\n", __func__);
540		/* XXX KDM now what? */
541		return;
542	}
543
544	ctsio = &msg_info->hdr.original_sc->scsiio;
545	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
546	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
547	ctsio->io_hdr.status = msg_info->hdr.status;
548	ctsio->scsi_status = msg_info->scsi.scsi_status;
549	ctsio->sense_len = msg_info->scsi.sense_len;
550	ctsio->sense_residual = msg_info->scsi.sense_residual;
551	ctsio->residual = msg_info->scsi.residual;
552	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
553	       sizeof(ctsio->sense_data));
554	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
555	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
556	ctl_enqueue_isc((union ctl_io *)ctsio);
557}
558
559static void
560ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
561				union ctl_ha_msg *msg_info)
562{
563	struct ctl_scsiio *ctsio;
564
565	if (msg_info->hdr.serializing_sc == NULL) {
566		printf("%s: serializing_sc == NULL!\n", __func__);
567		/* XXX KDM now what? */
568		return;
569	}
570
571	ctsio = &msg_info->hdr.serializing_sc->scsiio;
572#if 0
573	/*
574	 * Attempt to catch the situation where an I/O has
575	 * been freed, and we're using it again.
576	 */
577	if (ctsio->io_hdr.io_type == 0xff) {
578		union ctl_io *tmp_io;
579		tmp_io = (union ctl_io *)ctsio;
580		printf("%s: %p use after free!\n", __func__,
581		       ctsio);
582		printf("%s: type %d msg %d cdb %x iptl: "
583		       "%d:%d:%d:%d tag 0x%04x "
584		       "flag %#x status %x\n",
585			__func__,
586			tmp_io->io_hdr.io_type,
587			tmp_io->io_hdr.msg_type,
588			tmp_io->scsiio.cdb[0],
589			tmp_io->io_hdr.nexus.initid.id,
590			tmp_io->io_hdr.nexus.targ_port,
591			tmp_io->io_hdr.nexus.targ_target.id,
592			tmp_io->io_hdr.nexus.targ_lun,
593			(tmp_io->io_hdr.io_type ==
594			CTL_IO_TASK) ?
595			tmp_io->taskio.tag_num :
596			tmp_io->scsiio.tag_num,
597		        tmp_io->io_hdr.flags,
598			tmp_io->io_hdr.status);
599	}
600#endif
601	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
602	ctl_enqueue_isc((union ctl_io *)ctsio);
603}
604
605/*
606 * ISC (Inter Shelf Communication) event handler.  Events from the HA
607 * subsystem come in here.
608 */
609static void
610ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
611{
612	struct ctl_softc *ctl_softc;
613	union ctl_io *io;
614	struct ctl_prio *presio;
615	ctl_ha_status isc_status;
616
617	ctl_softc = control_softc;
618	io = NULL;
619
620
621#if 0
622	printf("CTL: Isc Msg event %d\n", event);
623#endif
624	if (event == CTL_HA_EVT_MSG_RECV) {
625		union ctl_ha_msg msg_info;
626
627		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
628					     sizeof(msg_info), /*wait*/ 0);
629#if 0
630		printf("CTL: msg_type %d\n", msg_info.msg_type);
631#endif
632		if (isc_status != 0) {
633			printf("Error receiving message, status = %d\n",
634			       isc_status);
635			return;
636		}
637
638		switch (msg_info.hdr.msg_type) {
639		case CTL_MSG_SERIALIZE:
640#if 0
641			printf("Serialize\n");
642#endif
643			io = ctl_alloc_io_nowait(ctl_softc->othersc_pool);
644			if (io == NULL) {
645				printf("ctl_isc_event_handler: can't allocate "
646				       "ctl_io!\n");
647				/* Bad Juju */
648				/* Need to set busy and send msg back */
649				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
650				msg_info.hdr.status = CTL_SCSI_ERROR;
651				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
652				msg_info.scsi.sense_len = 0;
653			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
654				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
655				}
656				goto bailout;
657			}
658			ctl_zero_io(io);
659			// populate ctsio from msg_info
660			io->io_hdr.io_type = CTL_IO_SCSI;
661			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
662			io->io_hdr.original_sc = msg_info.hdr.original_sc;
663#if 0
664			printf("pOrig %x\n", (int)msg_info.original_sc);
665#endif
666			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
667					    CTL_FLAG_IO_ACTIVE;
668			/*
669			 * If we're in serialization-only mode, we don't
670			 * want to go through full done processing.  Thus
671			 * the COPY flag.
672			 *
673			 * XXX KDM add another flag that is more specific.
674			 */
675			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
676				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
677			io->io_hdr.nexus = msg_info.hdr.nexus;
678#if 0
679			printf("targ %d, port %d, iid %d, lun %d\n",
680			       io->io_hdr.nexus.targ_target.id,
681			       io->io_hdr.nexus.targ_port,
682			       io->io_hdr.nexus.initid.id,
683			       io->io_hdr.nexus.targ_lun);
684#endif
685			io->scsiio.tag_num = msg_info.scsi.tag_num;
686			io->scsiio.tag_type = msg_info.scsi.tag_type;
687			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
688			       CTL_MAX_CDBLEN);
689			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
690				const struct ctl_cmd_entry *entry;
691
692				entry = ctl_get_cmd_entry(&io->scsiio, NULL);
693				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
694				io->io_hdr.flags |=
695					entry->flags & CTL_FLAG_DATA_MASK;
696			}
697			ctl_enqueue_isc(io);
698			break;
699
700		/* Performed on the Originating SC, XFER mode only */
701		case CTL_MSG_DATAMOVE: {
702			struct ctl_sg_entry *sgl;
703			int i, j;
704
705			io = msg_info.hdr.original_sc;
706			if (io == NULL) {
707				printf("%s: original_sc == NULL!\n", __func__);
708				/* XXX KDM do something here */
709				break;
710			}
711			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
712			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
713			/*
714			 * Keep track of this, we need to send it back over
715			 * when the datamove is complete.
716			 */
717			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
718
719			if (msg_info.dt.sg_sequence == 0) {
720				/*
721				 * XXX KDM we use the preallocated S/G list
722				 * here, but we'll need to change this to
723				 * dynamic allocation if we need larger S/G
724				 * lists.
725				 */
726				if (msg_info.dt.kern_sg_entries >
727				    sizeof(io->io_hdr.remote_sglist) /
728				    sizeof(io->io_hdr.remote_sglist[0])) {
729					printf("%s: number of S/G entries "
730					    "needed %u > allocated num %zd\n",
731					    __func__,
732					    msg_info.dt.kern_sg_entries,
733					    sizeof(io->io_hdr.remote_sglist)/
734					    sizeof(io->io_hdr.remote_sglist[0]));
735
736					/*
737					 * XXX KDM send a message back to
738					 * the other side to shut down the
739					 * DMA.  The error will come back
740					 * through via the normal channel.
741					 */
742					break;
743				}
744				sgl = io->io_hdr.remote_sglist;
745				memset(sgl, 0,
746				       sizeof(io->io_hdr.remote_sglist));
747
748				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
749
750				io->scsiio.kern_sg_entries =
751					msg_info.dt.kern_sg_entries;
752				io->scsiio.rem_sg_entries =
753					msg_info.dt.kern_sg_entries;
754				io->scsiio.kern_data_len =
755					msg_info.dt.kern_data_len;
756				io->scsiio.kern_total_len =
757					msg_info.dt.kern_total_len;
758				io->scsiio.kern_data_resid =
759					msg_info.dt.kern_data_resid;
760				io->scsiio.kern_rel_offset =
761					msg_info.dt.kern_rel_offset;
762				/*
763				 * Clear out per-DMA flags.
764				 */
765				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
766				/*
767				 * Add per-DMA flags that are set for this
768				 * particular DMA request.
769				 */
770				io->io_hdr.flags |= msg_info.dt.flags &
771						    CTL_FLAG_RDMA_MASK;
772			} else
773				sgl = (struct ctl_sg_entry *)
774					io->scsiio.kern_data_ptr;
775
776			for (i = msg_info.dt.sent_sg_entries, j = 0;
777			     i < (msg_info.dt.sent_sg_entries +
778			     msg_info.dt.cur_sg_entries); i++, j++) {
779				sgl[i].addr = msg_info.dt.sg_list[j].addr;
780				sgl[i].len = msg_info.dt.sg_list[j].len;
781
782#if 0
783				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
784				       __func__,
785				       msg_info.dt.sg_list[j].addr,
786				       msg_info.dt.sg_list[j].len,
787				       sgl[i].addr, sgl[i].len, j, i);
788#endif
789			}
790#if 0
791			memcpy(&sgl[msg_info.dt.sent_sg_entries],
792			       msg_info.dt.sg_list,
793			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
794#endif
795
796			/*
797			 * If this is the last piece of the I/O, we've got
798			 * the full S/G list.  Queue processing in the thread.
799			 * Otherwise wait for the next piece.
800			 */
801			if (msg_info.dt.sg_last != 0)
802				ctl_enqueue_isc(io);
803			break;
804		}
805		/* Performed on the Serializing (primary) SC, XFER mode only */
806		case CTL_MSG_DATAMOVE_DONE: {
807			if (msg_info.hdr.serializing_sc == NULL) {
808				printf("%s: serializing_sc == NULL!\n",
809				       __func__);
810				/* XXX KDM now what? */
811				break;
812			}
813			/*
814			 * We grab the sense information here in case
815			 * there was a failure, so we can return status
816			 * back to the initiator.
817			 */
818			io = msg_info.hdr.serializing_sc;
819			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
820			io->io_hdr.status = msg_info.hdr.status;
821			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
822			io->scsiio.sense_len = msg_info.scsi.sense_len;
823			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
824			io->io_hdr.port_status = msg_info.scsi.fetd_status;
825			io->scsiio.residual = msg_info.scsi.residual;
826			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
827			       sizeof(io->scsiio.sense_data));
828			ctl_enqueue_isc(io);
829			break;
830		}
831
832		/* Preformed on Originating SC, SER_ONLY mode */
833		case CTL_MSG_R2R:
834			io = msg_info.hdr.original_sc;
835			if (io == NULL) {
836				printf("%s: Major Bummer\n", __func__);
837				return;
838			} else {
839#if 0
840				printf("pOrig %x\n",(int) ctsio);
841#endif
842			}
843			io->io_hdr.msg_type = CTL_MSG_R2R;
844			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
845			ctl_enqueue_isc(io);
846			break;
847
848		/*
849		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
850		 * mode.
851		 * Performed on the Originating (i.e. secondary) SC in XFER
852		 * mode
853		 */
854		case CTL_MSG_FINISH_IO:
855			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
856				ctl_isc_handler_finish_xfer(ctl_softc,
857							    &msg_info);
858			else
859				ctl_isc_handler_finish_ser_only(ctl_softc,
860								&msg_info);
861			break;
862
863		/* Preformed on Originating SC */
864		case CTL_MSG_BAD_JUJU:
865			io = msg_info.hdr.original_sc;
866			if (io == NULL) {
867				printf("%s: Bad JUJU!, original_sc is NULL!\n",
868				       __func__);
869				break;
870			}
871			ctl_copy_sense_data(&msg_info, io);
872			/*
873			 * IO should have already been cleaned up on other
874			 * SC so clear this flag so we won't send a message
875			 * back to finish the IO there.
876			 */
877			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
878			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
879
880			/* io = msg_info.hdr.serializing_sc; */
881			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
882			ctl_enqueue_isc(io);
883			break;
884
885		/* Handle resets sent from the other side */
886		case CTL_MSG_MANAGE_TASKS: {
887			struct ctl_taskio *taskio;
888			taskio = (struct ctl_taskio *)ctl_alloc_io_nowait(
889			    ctl_softc->othersc_pool);
890			if (taskio == NULL) {
891				printf("ctl_isc_event_handler: can't allocate "
892				       "ctl_io!\n");
893				/* Bad Juju */
894				/* should I just call the proper reset func
895				   here??? */
896				goto bailout;
897			}
898			ctl_zero_io((union ctl_io *)taskio);
899			taskio->io_hdr.io_type = CTL_IO_TASK;
900			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
901			taskio->io_hdr.nexus = msg_info.hdr.nexus;
902			taskio->task_action = msg_info.task.task_action;
903			taskio->tag_num = msg_info.task.tag_num;
904			taskio->tag_type = msg_info.task.tag_type;
905#ifdef CTL_TIME_IO
906			taskio->io_hdr.start_time = time_uptime;
907			getbintime(&taskio->io_hdr.start_bt);
908#if 0
909			cs_prof_gettime(&taskio->io_hdr.start_ticks);
910#endif
911#endif /* CTL_TIME_IO */
912			ctl_run_task((union ctl_io *)taskio);
913			break;
914		}
915		/* Persistent Reserve action which needs attention */
916		case CTL_MSG_PERS_ACTION:
917			presio = (struct ctl_prio *)ctl_alloc_io_nowait(
918			    ctl_softc->othersc_pool);
919			if (presio == NULL) {
920				printf("ctl_isc_event_handler: can't allocate "
921				       "ctl_io!\n");
922				/* Bad Juju */
923				/* Need to set busy and send msg back */
924				goto bailout;
925			}
926			ctl_zero_io((union ctl_io *)presio);
927			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
928			presio->pr_msg = msg_info.pr;
929			ctl_enqueue_isc((union ctl_io *)presio);
930			break;
931		case CTL_MSG_SYNC_FE:
932			rcv_sync_msg = 1;
933			break;
934		default:
935		        printf("How did I get here?\n");
936		}
937	} else if (event == CTL_HA_EVT_MSG_SENT) {
938		if (param != CTL_HA_STATUS_SUCCESS) {
939			printf("Bad status from ctl_ha_msg_send status %d\n",
940			       param);
941		}
942		return;
943	} else if (event == CTL_HA_EVT_DISCONNECT) {
944		printf("CTL: Got a disconnect from Isc\n");
945		return;
946	} else {
947		printf("ctl_isc_event_handler: Unknown event %d\n", event);
948		return;
949	}
950
951bailout:
952	return;
953}
954
955static void
956ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
957{
958	struct scsi_sense_data *sense;
959
960	sense = &dest->scsiio.sense_data;
961	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
962	dest->scsiio.scsi_status = src->scsi.scsi_status;
963	dest->scsiio.sense_len = src->scsi.sense_len;
964	dest->io_hdr.status = src->hdr.status;
965}
966
967static void
968ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
969{
970	ctl_ua_type *pu;
971
972	mtx_assert(&lun->lun_lock, MA_OWNED);
973	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
974	if (pu == NULL)
975		return;
976	pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
977}
978
979static void
980ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
981{
982	int i, j;
983
984	mtx_assert(&lun->lun_lock, MA_OWNED);
985	for (i = 0; i < CTL_MAX_PORTS; i++) {
986		if (lun->pending_ua[i] == NULL)
987			continue;
988		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
989			if (i * CTL_MAX_INIT_PER_PORT + j == except)
990				continue;
991			lun->pending_ua[i][j] |= ua;
992		}
993	}
994}
995
996static void
997ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
998{
999	ctl_ua_type *pu;
1000
1001	mtx_assert(&lun->lun_lock, MA_OWNED);
1002	pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1003	if (pu == NULL)
1004		return;
1005	pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1006}
1007
1008static void
1009ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1010{
1011	int i, j;
1012
1013	mtx_assert(&lun->lun_lock, MA_OWNED);
1014	for (i = 0; i < CTL_MAX_PORTS; i++) {
1015		if (lun->pending_ua[i] == NULL)
1016			continue;
1017		for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1018			if (i * CTL_MAX_INIT_PER_PORT + j == except)
1019				continue;
1020			lun->pending_ua[i][j] &= ~ua;
1021		}
1022	}
1023}
1024
1025static int
1026ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS)
1027{
1028	struct ctl_softc *softc = (struct ctl_softc *)arg1;
1029	struct ctl_lun *lun;
1030	int error, value;
1031
1032	if (softc->flags & CTL_FLAG_ACTIVE_SHELF)
1033		value = 0;
1034	else
1035		value = 1;
1036
1037	error = sysctl_handle_int(oidp, &value, 0, req);
1038	if ((error != 0) || (req->newptr == NULL))
1039		return (error);
1040
1041	mtx_lock(&softc->ctl_lock);
1042	if (value == 0)
1043		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1044	else
1045		softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1046	STAILQ_FOREACH(lun, &softc->lun_list, links) {
1047		mtx_lock(&lun->lun_lock);
1048		ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1049		mtx_unlock(&lun->lun_lock);
1050	}
1051	mtx_unlock(&softc->ctl_lock);
1052	return (0);
1053}
1054
1055static int
1056ctl_init(void)
1057{
1058	struct ctl_softc *softc;
1059	void *other_pool;
1060	struct ctl_port *port;
1061	int i, error, retval;
1062	//int isc_retval;
1063
1064	retval = 0;
1065	ctl_pause_rtr = 0;
1066        rcv_sync_msg = 0;
1067
1068	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1069			       M_WAITOK | M_ZERO);
1070	softc = control_softc;
1071
1072	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1073			      "cam/ctl");
1074
1075	softc->dev->si_drv1 = softc;
1076
1077	/*
1078	 * By default, return a "bad LUN" peripheral qualifier for unknown
1079	 * LUNs.  The user can override this default using the tunable or
1080	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
1081	 */
1082	softc->inquiry_pq_no_lun = 1;
1083	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
1084			  &softc->inquiry_pq_no_lun);
1085	sysctl_ctx_init(&softc->sysctl_ctx);
1086	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1087		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1088		CTLFLAG_RD, 0, "CAM Target Layer");
1089
1090	if (softc->sysctl_tree == NULL) {
1091		printf("%s: unable to allocate sysctl tree\n", __func__);
1092		destroy_dev(softc->dev);
1093		free(control_softc, M_DEVBUF);
1094		control_softc = NULL;
1095		return (ENOMEM);
1096	}
1097
1098	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1099		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1100		       "inquiry_pq_no_lun", CTLFLAG_RW,
1101		       &softc->inquiry_pq_no_lun, 0,
1102		       "Report no lun possible for invalid LUNs");
1103
1104	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1105	softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1106	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1107	softc->open_count = 0;
1108
1109	/*
1110	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1111	 * the drive.
1112	 */
1113	softc->flags = CTL_FLAG_REAL_SYNC;
1114
1115	/*
1116	 * In Copan's HA scheme, the "master" and "slave" roles are
1117	 * figured out through the slot the controller is in.  Although it
1118	 * is an active/active system, someone has to be in charge.
1119	 */
1120	SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1121	    OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1122	    "HA head ID (0 - no HA)");
1123	if (softc->ha_id == 0) {
1124		softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1125		softc->is_single = 1;
1126		softc->port_offset = 0;
1127	} else
1128		softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS;
1129	softc->persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT;
1130
1131	/*
1132	 * XXX KDM need to figure out where we want to get our target ID
1133	 * and WWID.  Is it different on each port?
1134	 */
1135	softc->target.id = 0;
1136	softc->target.wwid[0] = 0x12345678;
1137	softc->target.wwid[1] = 0x87654321;
1138	STAILQ_INIT(&softc->lun_list);
1139	STAILQ_INIT(&softc->pending_lun_queue);
1140	STAILQ_INIT(&softc->fe_list);
1141	STAILQ_INIT(&softc->port_list);
1142	STAILQ_INIT(&softc->be_list);
1143	ctl_tpc_init(softc);
1144
1145	if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1146	                    &other_pool) != 0)
1147	{
1148		printf("ctl: can't allocate %d entry other SC pool, "
1149		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1150		return (ENOMEM);
1151	}
1152	softc->othersc_pool = other_pool;
1153
1154	if (worker_threads <= 0)
1155		worker_threads = max(1, mp_ncpus / 4);
1156	if (worker_threads > CTL_MAX_THREADS)
1157		worker_threads = CTL_MAX_THREADS;
1158
1159	for (i = 0; i < worker_threads; i++) {
1160		struct ctl_thread *thr = &softc->threads[i];
1161
1162		mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1163		thr->ctl_softc = softc;
1164		STAILQ_INIT(&thr->incoming_queue);
1165		STAILQ_INIT(&thr->rtr_queue);
1166		STAILQ_INIT(&thr->done_queue);
1167		STAILQ_INIT(&thr->isc_queue);
1168
1169		error = kproc_kthread_add(ctl_work_thread, thr,
1170		    &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1171		if (error != 0) {
1172			printf("error creating CTL work thread!\n");
1173			ctl_pool_free(other_pool);
1174			return (error);
1175		}
1176	}
1177	error = kproc_kthread_add(ctl_lun_thread, softc,
1178	    &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1179	if (error != 0) {
1180		printf("error creating CTL lun thread!\n");
1181		ctl_pool_free(other_pool);
1182		return (error);
1183	}
1184	error = kproc_kthread_add(ctl_thresh_thread, softc,
1185	    &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1186	if (error != 0) {
1187		printf("error creating CTL threshold thread!\n");
1188		ctl_pool_free(other_pool);
1189		return (error);
1190	}
1191	if (bootverbose)
1192		printf("ctl: CAM Target Layer loaded\n");
1193
1194	/*
1195	 * Initialize the ioctl front end.
1196	 */
1197	ctl_frontend_register(&ioctl_frontend);
1198	port = &softc->ioctl_info.port;
1199	port->frontend = &ioctl_frontend;
1200	sprintf(softc->ioctl_info.port_name, "ioctl");
1201	port->port_type = CTL_PORT_IOCTL;
1202	port->num_requested_ctl_io = 100;
1203	port->port_name = softc->ioctl_info.port_name;
1204	port->port_online = ctl_ioctl_online;
1205	port->port_offline = ctl_ioctl_offline;
1206	port->onoff_arg = &softc->ioctl_info;
1207	port->lun_enable = ctl_ioctl_lun_enable;
1208	port->lun_disable = ctl_ioctl_lun_disable;
1209	port->targ_lun_arg = &softc->ioctl_info;
1210	port->fe_datamove = ctl_ioctl_datamove;
1211	port->fe_done = ctl_ioctl_done;
1212	port->max_targets = 15;
1213	port->max_target_id = 15;
1214
1215	if (ctl_port_register(&softc->ioctl_info.port) != 0) {
1216		printf("ctl: ioctl front end registration failed, will "
1217		       "continue anyway\n");
1218	}
1219
1220	SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1221	    OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN,
1222	    softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head");
1223
1224#ifdef CTL_IO_DELAY
1225	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1226		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1227		       sizeof(struct callout), CTL_TIMER_BYTES);
1228		return (EINVAL);
1229	}
1230#endif /* CTL_IO_DELAY */
1231
1232	return (0);
1233}
1234
1235void
1236ctl_shutdown(void)
1237{
1238	struct ctl_softc *softc;
1239	struct ctl_lun *lun, *next_lun;
1240
1241	softc = (struct ctl_softc *)control_softc;
1242
1243	if (ctl_port_deregister(&softc->ioctl_info.port) != 0)
1244		printf("ctl: ioctl front end deregistration failed\n");
1245
1246	mtx_lock(&softc->ctl_lock);
1247
1248	/*
1249	 * Free up each LUN.
1250	 */
1251	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1252		next_lun = STAILQ_NEXT(lun, links);
1253		ctl_free_lun(lun);
1254	}
1255
1256	mtx_unlock(&softc->ctl_lock);
1257
1258	ctl_frontend_deregister(&ioctl_frontend);
1259
1260#if 0
1261	ctl_shutdown_thread(softc->work_thread);
1262	mtx_destroy(&softc->queue_lock);
1263#endif
1264
1265	ctl_tpc_shutdown(softc);
1266	uma_zdestroy(softc->io_zone);
1267	mtx_destroy(&softc->ctl_lock);
1268
1269	destroy_dev(softc->dev);
1270
1271	sysctl_ctx_free(&softc->sysctl_ctx);
1272
1273	free(control_softc, M_DEVBUF);
1274	control_softc = NULL;
1275
1276	if (bootverbose)
1277		printf("ctl: CAM Target Layer unloaded\n");
1278}
1279
1280static int
1281ctl_module_event_handler(module_t mod, int what, void *arg)
1282{
1283
1284	switch (what) {
1285	case MOD_LOAD:
1286		return (ctl_init());
1287	case MOD_UNLOAD:
1288		return (EBUSY);
1289	default:
1290		return (EOPNOTSUPP);
1291	}
1292}
1293
1294/*
1295 * XXX KDM should we do some access checks here?  Bump a reference count to
1296 * prevent a CTL module from being unloaded while someone has it open?
1297 */
1298static int
1299ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1300{
1301	return (0);
1302}
1303
1304static int
1305ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1306{
1307	return (0);
1308}
1309
1310int
1311ctl_port_enable(ctl_port_type port_type)
1312{
1313	struct ctl_softc *softc = control_softc;
1314	struct ctl_port *port;
1315
1316	if (softc->is_single == 0) {
1317		union ctl_ha_msg msg_info;
1318		int isc_retval;
1319
1320#if 0
1321		printf("%s: HA mode, synchronizing frontend enable\n",
1322		        __func__);
1323#endif
1324		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1325	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1326		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1327			printf("Sync msg send error retval %d\n", isc_retval);
1328		}
1329		if (!rcv_sync_msg) {
1330			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1331			        sizeof(msg_info), 1);
1332		}
1333#if 0
1334        	printf("CTL:Frontend Enable\n");
1335	} else {
1336		printf("%s: single mode, skipping frontend synchronization\n",
1337		        __func__);
1338#endif
1339	}
1340
1341	STAILQ_FOREACH(port, &softc->port_list, links) {
1342		if (port_type & port->port_type)
1343		{
1344#if 0
1345			printf("port %d\n", port->targ_port);
1346#endif
1347			ctl_port_online(port);
1348		}
1349	}
1350
1351	return (0);
1352}
1353
1354int
1355ctl_port_disable(ctl_port_type port_type)
1356{
1357	struct ctl_softc *softc;
1358	struct ctl_port *port;
1359
1360	softc = control_softc;
1361
1362	STAILQ_FOREACH(port, &softc->port_list, links) {
1363		if (port_type & port->port_type)
1364			ctl_port_offline(port);
1365	}
1366
1367	return (0);
1368}
1369
1370/*
1371 * Returns 0 for success, 1 for failure.
1372 * Currently the only failure mode is if there aren't enough entries
1373 * allocated.  So, in case of a failure, look at num_entries_dropped,
1374 * reallocate and try again.
1375 */
1376int
1377ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1378	      int *num_entries_filled, int *num_entries_dropped,
1379	      ctl_port_type port_type, int no_virtual)
1380{
1381	struct ctl_softc *softc;
1382	struct ctl_port *port;
1383	int entries_dropped, entries_filled;
1384	int retval;
1385	int i;
1386
1387	softc = control_softc;
1388
1389	retval = 0;
1390	entries_filled = 0;
1391	entries_dropped = 0;
1392
1393	i = 0;
1394	mtx_lock(&softc->ctl_lock);
1395	STAILQ_FOREACH(port, &softc->port_list, links) {
1396		struct ctl_port_entry *entry;
1397
1398		if ((port->port_type & port_type) == 0)
1399			continue;
1400
1401		if ((no_virtual != 0)
1402		 && (port->virtual_port != 0))
1403			continue;
1404
1405		if (entries_filled >= num_entries_alloced) {
1406			entries_dropped++;
1407			continue;
1408		}
1409		entry = &entries[i];
1410
1411		entry->port_type = port->port_type;
1412		strlcpy(entry->port_name, port->port_name,
1413			sizeof(entry->port_name));
1414		entry->physical_port = port->physical_port;
1415		entry->virtual_port = port->virtual_port;
1416		entry->wwnn = port->wwnn;
1417		entry->wwpn = port->wwpn;
1418
1419		i++;
1420		entries_filled++;
1421	}
1422
1423	mtx_unlock(&softc->ctl_lock);
1424
1425	if (entries_dropped > 0)
1426		retval = 1;
1427
1428	*num_entries_dropped = entries_dropped;
1429	*num_entries_filled = entries_filled;
1430
1431	return (retval);
1432}
1433
1434static void
1435ctl_ioctl_online(void *arg)
1436{
1437	struct ctl_ioctl_info *ioctl_info;
1438
1439	ioctl_info = (struct ctl_ioctl_info *)arg;
1440
1441	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1442}
1443
1444static void
1445ctl_ioctl_offline(void *arg)
1446{
1447	struct ctl_ioctl_info *ioctl_info;
1448
1449	ioctl_info = (struct ctl_ioctl_info *)arg;
1450
1451	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1452}
1453
1454/*
1455 * Remove an initiator by port number and initiator ID.
1456 * Returns 0 for success, -1 for failure.
1457 */
1458int
1459ctl_remove_initiator(struct ctl_port *port, int iid)
1460{
1461	struct ctl_softc *softc = control_softc;
1462
1463	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1464
1465	if (iid > CTL_MAX_INIT_PER_PORT) {
1466		printf("%s: initiator ID %u > maximun %u!\n",
1467		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1468		return (-1);
1469	}
1470
1471	mtx_lock(&softc->ctl_lock);
1472	port->wwpn_iid[iid].in_use--;
1473	port->wwpn_iid[iid].last_use = time_uptime;
1474	mtx_unlock(&softc->ctl_lock);
1475
1476	return (0);
1477}
1478
1479/*
1480 * Add an initiator to the initiator map.
1481 * Returns iid for success, < 0 for failure.
1482 */
1483int
1484ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1485{
1486	struct ctl_softc *softc = control_softc;
1487	time_t best_time;
1488	int i, best;
1489
1490	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1491
1492	if (iid >= CTL_MAX_INIT_PER_PORT) {
1493		printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1494		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1495		free(name, M_CTL);
1496		return (-1);
1497	}
1498
1499	mtx_lock(&softc->ctl_lock);
1500
1501	if (iid < 0 && (wwpn != 0 || name != NULL)) {
1502		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1503			if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1504				iid = i;
1505				break;
1506			}
1507			if (name != NULL && port->wwpn_iid[i].name != NULL &&
1508			    strcmp(name, port->wwpn_iid[i].name) == 0) {
1509				iid = i;
1510				break;
1511			}
1512		}
1513	}
1514
1515	if (iid < 0) {
1516		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1517			if (port->wwpn_iid[i].in_use == 0 &&
1518			    port->wwpn_iid[i].wwpn == 0 &&
1519			    port->wwpn_iid[i].name == NULL) {
1520				iid = i;
1521				break;
1522			}
1523		}
1524	}
1525
1526	if (iid < 0) {
1527		best = -1;
1528		best_time = INT32_MAX;
1529		for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1530			if (port->wwpn_iid[i].in_use == 0) {
1531				if (port->wwpn_iid[i].last_use < best_time) {
1532					best = i;
1533					best_time = port->wwpn_iid[i].last_use;
1534				}
1535			}
1536		}
1537		iid = best;
1538	}
1539
1540	if (iid < 0) {
1541		mtx_unlock(&softc->ctl_lock);
1542		free(name, M_CTL);
1543		return (-2);
1544	}
1545
1546	if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1547		/*
1548		 * This is not an error yet.
1549		 */
1550		if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1551#if 0
1552			printf("%s: port %d iid %u WWPN %#jx arrived"
1553			    " again\n", __func__, port->targ_port,
1554			    iid, (uintmax_t)wwpn);
1555#endif
1556			goto take;
1557		}
1558		if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1559		    strcmp(name, port->wwpn_iid[iid].name) == 0) {
1560#if 0
1561			printf("%s: port %d iid %u name '%s' arrived"
1562			    " again\n", __func__, port->targ_port,
1563			    iid, name);
1564#endif
1565			goto take;
1566		}
1567
1568		/*
1569		 * This is an error, but what do we do about it?  The
1570		 * driver is telling us we have a new WWPN for this
1571		 * initiator ID, so we pretty much need to use it.
1572		 */
1573		printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1574		    " but WWPN %#jx '%s' is still at that address\n",
1575		    __func__, port->targ_port, iid, wwpn, name,
1576		    (uintmax_t)port->wwpn_iid[iid].wwpn,
1577		    port->wwpn_iid[iid].name);
1578
1579		/*
1580		 * XXX KDM clear have_ca and ua_pending on each LUN for
1581		 * this initiator.
1582		 */
1583	}
1584take:
1585	free(port->wwpn_iid[iid].name, M_CTL);
1586	port->wwpn_iid[iid].name = name;
1587	port->wwpn_iid[iid].wwpn = wwpn;
1588	port->wwpn_iid[iid].in_use++;
1589	mtx_unlock(&softc->ctl_lock);
1590
1591	return (iid);
1592}
1593
1594static int
1595ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1596{
1597	int len;
1598
1599	switch (port->port_type) {
1600	case CTL_PORT_FC:
1601	{
1602		struct scsi_transportid_fcp *id =
1603		    (struct scsi_transportid_fcp *)buf;
1604		if (port->wwpn_iid[iid].wwpn == 0)
1605			return (0);
1606		memset(id, 0, sizeof(*id));
1607		id->format_protocol = SCSI_PROTO_FC;
1608		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1609		return (sizeof(*id));
1610	}
1611	case CTL_PORT_ISCSI:
1612	{
1613		struct scsi_transportid_iscsi_port *id =
1614		    (struct scsi_transportid_iscsi_port *)buf;
1615		if (port->wwpn_iid[iid].name == NULL)
1616			return (0);
1617		memset(id, 0, 256);
1618		id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1619		    SCSI_PROTO_ISCSI;
1620		len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1621		len = roundup2(min(len, 252), 4);
1622		scsi_ulto2b(len, id->additional_length);
1623		return (sizeof(*id) + len);
1624	}
1625	case CTL_PORT_SAS:
1626	{
1627		struct scsi_transportid_sas *id =
1628		    (struct scsi_transportid_sas *)buf;
1629		if (port->wwpn_iid[iid].wwpn == 0)
1630			return (0);
1631		memset(id, 0, sizeof(*id));
1632		id->format_protocol = SCSI_PROTO_SAS;
1633		scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1634		return (sizeof(*id));
1635	}
1636	default:
1637	{
1638		struct scsi_transportid_spi *id =
1639		    (struct scsi_transportid_spi *)buf;
1640		memset(id, 0, sizeof(*id));
1641		id->format_protocol = SCSI_PROTO_SPI;
1642		scsi_ulto2b(iid, id->scsi_addr);
1643		scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1644		return (sizeof(*id));
1645	}
1646	}
1647}
1648
1649static int
1650ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1651{
1652	return (0);
1653}
1654
1655static int
1656ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1657{
1658	return (0);
1659}
1660
1661/*
1662 * Data movement routine for the CTL ioctl frontend port.
1663 */
1664static int
1665ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1666{
1667	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1668	struct ctl_sg_entry ext_entry, kern_entry;
1669	int ext_sglen, ext_sg_entries, kern_sg_entries;
1670	int ext_sg_start, ext_offset;
1671	int len_to_copy, len_copied;
1672	int kern_watermark, ext_watermark;
1673	int ext_sglist_malloced;
1674	int i, j;
1675
1676	ext_sglist_malloced = 0;
1677	ext_sg_start = 0;
1678	ext_offset = 0;
1679
1680	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1681
1682	/*
1683	 * If this flag is set, fake the data transfer.
1684	 */
1685	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1686		ctsio->ext_data_filled = ctsio->ext_data_len;
1687		goto bailout;
1688	}
1689
1690	/*
1691	 * To simplify things here, if we have a single buffer, stick it in
1692	 * a S/G entry and just make it a single entry S/G list.
1693	 */
1694	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1695		int len_seen;
1696
1697		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1698
1699		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1700							   M_WAITOK);
1701		ext_sglist_malloced = 1;
1702		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1703				   ext_sglen) != 0) {
1704			ctl_set_internal_failure(ctsio,
1705						 /*sks_valid*/ 0,
1706						 /*retry_count*/ 0);
1707			goto bailout;
1708		}
1709		ext_sg_entries = ctsio->ext_sg_entries;
1710		len_seen = 0;
1711		for (i = 0; i < ext_sg_entries; i++) {
1712			if ((len_seen + ext_sglist[i].len) >=
1713			     ctsio->ext_data_filled) {
1714				ext_sg_start = i;
1715				ext_offset = ctsio->ext_data_filled - len_seen;
1716				break;
1717			}
1718			len_seen += ext_sglist[i].len;
1719		}
1720	} else {
1721		ext_sglist = &ext_entry;
1722		ext_sglist->addr = ctsio->ext_data_ptr;
1723		ext_sglist->len = ctsio->ext_data_len;
1724		ext_sg_entries = 1;
1725		ext_sg_start = 0;
1726		ext_offset = ctsio->ext_data_filled;
1727	}
1728
1729	if (ctsio->kern_sg_entries > 0) {
1730		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1731		kern_sg_entries = ctsio->kern_sg_entries;
1732	} else {
1733		kern_sglist = &kern_entry;
1734		kern_sglist->addr = ctsio->kern_data_ptr;
1735		kern_sglist->len = ctsio->kern_data_len;
1736		kern_sg_entries = 1;
1737	}
1738
1739
1740	kern_watermark = 0;
1741	ext_watermark = ext_offset;
1742	len_copied = 0;
1743	for (i = ext_sg_start, j = 0;
1744	     i < ext_sg_entries && j < kern_sg_entries;) {
1745		uint8_t *ext_ptr, *kern_ptr;
1746
1747		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1748				      kern_sglist[j].len - kern_watermark);
1749
1750		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1751		ext_ptr = ext_ptr + ext_watermark;
1752		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1753			/*
1754			 * XXX KDM fix this!
1755			 */
1756			panic("need to implement bus address support");
1757#if 0
1758			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1759#endif
1760		} else
1761			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1762		kern_ptr = kern_ptr + kern_watermark;
1763
1764		kern_watermark += len_to_copy;
1765		ext_watermark += len_to_copy;
1766
1767		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1768		     CTL_FLAG_DATA_IN) {
1769			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1770					 "bytes to user\n", len_to_copy));
1771			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1772					 "to %p\n", kern_ptr, ext_ptr));
1773			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1774				ctl_set_internal_failure(ctsio,
1775							 /*sks_valid*/ 0,
1776							 /*retry_count*/ 0);
1777				goto bailout;
1778			}
1779		} else {
1780			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1781					 "bytes from user\n", len_to_copy));
1782			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1783					 "to %p\n", ext_ptr, kern_ptr));
1784			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1785				ctl_set_internal_failure(ctsio,
1786							 /*sks_valid*/ 0,
1787							 /*retry_count*/0);
1788				goto bailout;
1789			}
1790		}
1791
1792		len_copied += len_to_copy;
1793
1794		if (ext_sglist[i].len == ext_watermark) {
1795			i++;
1796			ext_watermark = 0;
1797		}
1798
1799		if (kern_sglist[j].len == kern_watermark) {
1800			j++;
1801			kern_watermark = 0;
1802		}
1803	}
1804
1805	ctsio->ext_data_filled += len_copied;
1806
1807	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1808			 "kern_sg_entries: %d\n", ext_sg_entries,
1809			 kern_sg_entries));
1810	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1811			 "kern_data_len = %d\n", ctsio->ext_data_len,
1812			 ctsio->kern_data_len));
1813
1814
1815	/* XXX KDM set residual?? */
1816bailout:
1817
1818	if (ext_sglist_malloced != 0)
1819		free(ext_sglist, M_CTL);
1820
1821	return (CTL_RETVAL_COMPLETE);
1822}
1823
1824/*
1825 * Serialize a command that went down the "wrong" side, and so was sent to
1826 * this controller for execution.  The logic is a little different than the
1827 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1828 * sent back to the other side, but in the success case, we execute the
1829 * command on this side (XFER mode) or tell the other side to execute it
1830 * (SER_ONLY mode).
1831 */
1832static int
1833ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1834{
1835	struct ctl_softc *ctl_softc;
1836	union ctl_ha_msg msg_info;
1837	struct ctl_lun *lun;
1838	int retval = 0;
1839	uint32_t targ_lun;
1840
1841	ctl_softc = control_softc;
1842
1843	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1844	lun = ctl_softc->ctl_luns[targ_lun];
1845	if (lun==NULL)
1846	{
1847		/*
1848		 * Why isn't LUN defined? The other side wouldn't
1849		 * send a cmd if the LUN is undefined.
1850		 */
1851		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1852
1853		/* "Logical unit not supported" */
1854		ctl_set_sense_data(&msg_info.scsi.sense_data,
1855				   lun,
1856				   /*sense_format*/SSD_TYPE_NONE,
1857				   /*current_error*/ 1,
1858				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1859				   /*asc*/ 0x25,
1860				   /*ascq*/ 0x00,
1861				   SSD_ELEM_NONE);
1862
1863		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1864		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1865		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1866		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1867		msg_info.hdr.serializing_sc = NULL;
1868		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1869	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1870				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1871		}
1872		return(1);
1873
1874	}
1875
1876	mtx_lock(&lun->lun_lock);
1877    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1878
1879	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1880		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1881		 ooa_links))) {
1882	case CTL_ACTION_BLOCK:
1883		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1884		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1885				  blocked_links);
1886		break;
1887	case CTL_ACTION_PASS:
1888	case CTL_ACTION_SKIP:
1889		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1890			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1891			ctl_enqueue_rtr((union ctl_io *)ctsio);
1892		} else {
1893
1894			/* send msg back to other side */
1895			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1896			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1897			msg_info.hdr.msg_type = CTL_MSG_R2R;
1898#if 0
1899			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1900#endif
1901		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1902			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1903			}
1904		}
1905		break;
1906	case CTL_ACTION_OVERLAP:
1907		/* OVERLAPPED COMMANDS ATTEMPTED */
1908		ctl_set_sense_data(&msg_info.scsi.sense_data,
1909				   lun,
1910				   /*sense_format*/SSD_TYPE_NONE,
1911				   /*current_error*/ 1,
1912				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1913				   /*asc*/ 0x4E,
1914				   /*ascq*/ 0x00,
1915				   SSD_ELEM_NONE);
1916
1917		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1918		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1919		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1920		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1921		msg_info.hdr.serializing_sc = NULL;
1922		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1923#if 0
1924		printf("BAD JUJU:Major Bummer Overlap\n");
1925#endif
1926		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1927		retval = 1;
1928		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1929		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1930		}
1931		break;
1932	case CTL_ACTION_OVERLAP_TAG:
1933		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1934		ctl_set_sense_data(&msg_info.scsi.sense_data,
1935				   lun,
1936				   /*sense_format*/SSD_TYPE_NONE,
1937				   /*current_error*/ 1,
1938				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1939				   /*asc*/ 0x4D,
1940				   /*ascq*/ ctsio->tag_num & 0xff,
1941				   SSD_ELEM_NONE);
1942
1943		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1944		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1945		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1946		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1947		msg_info.hdr.serializing_sc = NULL;
1948		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1949#if 0
1950		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1951#endif
1952		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1953		retval = 1;
1954		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1955		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1956		}
1957		break;
1958	case CTL_ACTION_ERROR:
1959	default:
1960		/* "Internal target failure" */
1961		ctl_set_sense_data(&msg_info.scsi.sense_data,
1962				   lun,
1963				   /*sense_format*/SSD_TYPE_NONE,
1964				   /*current_error*/ 1,
1965				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1966				   /*asc*/ 0x44,
1967				   /*ascq*/ 0x00,
1968				   SSD_ELEM_NONE);
1969
1970		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1971		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1972		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1973		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1974		msg_info.hdr.serializing_sc = NULL;
1975		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1976#if 0
1977		printf("BAD JUJU:Major Bummer HW Error\n");
1978#endif
1979		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1980		retval = 1;
1981		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1982		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1983		}
1984		break;
1985	}
1986	mtx_unlock(&lun->lun_lock);
1987	return (retval);
1988}
1989
1990static int
1991ctl_ioctl_submit_wait(union ctl_io *io)
1992{
1993	struct ctl_fe_ioctl_params params;
1994	ctl_fe_ioctl_state last_state;
1995	int done, retval;
1996
1997	retval = 0;
1998
1999	bzero(&params, sizeof(params));
2000
2001	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
2002	cv_init(&params.sem, "ctlioccv");
2003	params.state = CTL_IOCTL_INPROG;
2004	last_state = params.state;
2005
2006	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
2007
2008	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
2009
2010	/* This shouldn't happen */
2011	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
2012		return (retval);
2013
2014	done = 0;
2015
2016	do {
2017		mtx_lock(&params.ioctl_mtx);
2018		/*
2019		 * Check the state here, and don't sleep if the state has
2020		 * already changed (i.e. wakeup has already occured, but we
2021		 * weren't waiting yet).
2022		 */
2023		if (params.state == last_state) {
2024			/* XXX KDM cv_wait_sig instead? */
2025			cv_wait(&params.sem, &params.ioctl_mtx);
2026		}
2027		last_state = params.state;
2028
2029		switch (params.state) {
2030		case CTL_IOCTL_INPROG:
2031			/* Why did we wake up? */
2032			/* XXX KDM error here? */
2033			mtx_unlock(&params.ioctl_mtx);
2034			break;
2035		case CTL_IOCTL_DATAMOVE:
2036			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
2037
2038			/*
2039			 * change last_state back to INPROG to avoid
2040			 * deadlock on subsequent data moves.
2041			 */
2042			params.state = last_state = CTL_IOCTL_INPROG;
2043
2044			mtx_unlock(&params.ioctl_mtx);
2045			ctl_ioctl_do_datamove(&io->scsiio);
2046			/*
2047			 * Note that in some cases, most notably writes,
2048			 * this will queue the I/O and call us back later.
2049			 * In other cases, generally reads, this routine
2050			 * will immediately call back and wake us up,
2051			 * probably using our own context.
2052			 */
2053			io->scsiio.be_move_done(io);
2054			break;
2055		case CTL_IOCTL_DONE:
2056			mtx_unlock(&params.ioctl_mtx);
2057			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
2058			done = 1;
2059			break;
2060		default:
2061			mtx_unlock(&params.ioctl_mtx);
2062			/* XXX KDM error here? */
2063			break;
2064		}
2065	} while (done == 0);
2066
2067	mtx_destroy(&params.ioctl_mtx);
2068	cv_destroy(&params.sem);
2069
2070	return (CTL_RETVAL_COMPLETE);
2071}
2072
2073static void
2074ctl_ioctl_datamove(union ctl_io *io)
2075{
2076	struct ctl_fe_ioctl_params *params;
2077
2078	params = (struct ctl_fe_ioctl_params *)
2079		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2080
2081	mtx_lock(&params->ioctl_mtx);
2082	params->state = CTL_IOCTL_DATAMOVE;
2083	cv_broadcast(&params->sem);
2084	mtx_unlock(&params->ioctl_mtx);
2085}
2086
2087static void
2088ctl_ioctl_done(union ctl_io *io)
2089{
2090	struct ctl_fe_ioctl_params *params;
2091
2092	params = (struct ctl_fe_ioctl_params *)
2093		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
2094
2095	mtx_lock(&params->ioctl_mtx);
2096	params->state = CTL_IOCTL_DONE;
2097	cv_broadcast(&params->sem);
2098	mtx_unlock(&params->ioctl_mtx);
2099}
2100
2101static void
2102ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
2103{
2104	struct ctl_fe_ioctl_startstop_info *sd_info;
2105
2106	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
2107
2108	sd_info->hs_info.status = metatask->status;
2109	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
2110	sd_info->hs_info.luns_complete =
2111		metatask->taskinfo.startstop.luns_complete;
2112	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
2113
2114	cv_broadcast(&sd_info->sem);
2115}
2116
2117static void
2118ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
2119{
2120	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
2121
2122	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
2123
2124	mtx_lock(fe_bbr_info->lock);
2125	fe_bbr_info->bbr_info->status = metatask->status;
2126	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2127	fe_bbr_info->wakeup_done = 1;
2128	mtx_unlock(fe_bbr_info->lock);
2129
2130	cv_broadcast(&fe_bbr_info->sem);
2131}
2132
2133/*
2134 * Returns 0 for success, errno for failure.
2135 */
2136static int
2137ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2138		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2139{
2140	union ctl_io *io;
2141	int retval;
2142
2143	retval = 0;
2144
2145	mtx_lock(&lun->lun_lock);
2146	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2147	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2148	     ooa_links)) {
2149		struct ctl_ooa_entry *entry;
2150
2151		/*
2152		 * If we've got more than we can fit, just count the
2153		 * remaining entries.
2154		 */
2155		if (*cur_fill_num >= ooa_hdr->alloc_num)
2156			continue;
2157
2158		entry = &kern_entries[*cur_fill_num];
2159
2160		entry->tag_num = io->scsiio.tag_num;
2161		entry->lun_num = lun->lun;
2162#ifdef CTL_TIME_IO
2163		entry->start_bt = io->io_hdr.start_bt;
2164#endif
2165		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2166		entry->cdb_len = io->scsiio.cdb_len;
2167		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2168			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2169
2170		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2171			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2172
2173		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2174			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2175
2176		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2177			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2178
2179		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2180			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2181	}
2182	mtx_unlock(&lun->lun_lock);
2183
2184	return (retval);
2185}
2186
2187static void *
2188ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2189		 size_t error_str_len)
2190{
2191	void *kptr;
2192
2193	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2194
2195	if (copyin(user_addr, kptr, len) != 0) {
2196		snprintf(error_str, error_str_len, "Error copying %d bytes "
2197			 "from user address %p to kernel address %p", len,
2198			 user_addr, kptr);
2199		free(kptr, M_CTL);
2200		return (NULL);
2201	}
2202
2203	return (kptr);
2204}
2205
2206static void
2207ctl_free_args(int num_args, struct ctl_be_arg *args)
2208{
2209	int i;
2210
2211	if (args == NULL)
2212		return;
2213
2214	for (i = 0; i < num_args; i++) {
2215		free(args[i].kname, M_CTL);
2216		free(args[i].kvalue, M_CTL);
2217	}
2218
2219	free(args, M_CTL);
2220}
2221
2222static struct ctl_be_arg *
2223ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2224		char *error_str, size_t error_str_len)
2225{
2226	struct ctl_be_arg *args;
2227	int i;
2228
2229	args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2230				error_str, error_str_len);
2231
2232	if (args == NULL)
2233		goto bailout;
2234
2235	for (i = 0; i < num_args; i++) {
2236		args[i].kname = NULL;
2237		args[i].kvalue = NULL;
2238	}
2239
2240	for (i = 0; i < num_args; i++) {
2241		uint8_t *tmpptr;
2242
2243		args[i].kname = ctl_copyin_alloc(args[i].name,
2244			args[i].namelen, error_str, error_str_len);
2245		if (args[i].kname == NULL)
2246			goto bailout;
2247
2248		if (args[i].kname[args[i].namelen - 1] != '\0') {
2249			snprintf(error_str, error_str_len, "Argument %d "
2250				 "name is not NUL-terminated", i);
2251			goto bailout;
2252		}
2253
2254		if (args[i].flags & CTL_BEARG_RD) {
2255			tmpptr = ctl_copyin_alloc(args[i].value,
2256				args[i].vallen, error_str, error_str_len);
2257			if (tmpptr == NULL)
2258				goto bailout;
2259			if ((args[i].flags & CTL_BEARG_ASCII)
2260			 && (tmpptr[args[i].vallen - 1] != '\0')) {
2261				snprintf(error_str, error_str_len, "Argument "
2262				    "%d value is not NUL-terminated", i);
2263				goto bailout;
2264			}
2265			args[i].kvalue = tmpptr;
2266		} else {
2267			args[i].kvalue = malloc(args[i].vallen,
2268			    M_CTL, M_WAITOK | M_ZERO);
2269		}
2270	}
2271
2272	return (args);
2273bailout:
2274
2275	ctl_free_args(num_args, args);
2276
2277	return (NULL);
2278}
2279
2280static void
2281ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2282{
2283	int i;
2284
2285	for (i = 0; i < num_args; i++) {
2286		if (args[i].flags & CTL_BEARG_WR)
2287			copyout(args[i].kvalue, args[i].value, args[i].vallen);
2288	}
2289}
2290
2291/*
2292 * Escape characters that are illegal or not recommended in XML.
2293 */
2294int
2295ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2296{
2297	char *end = str + size;
2298	int retval;
2299
2300	retval = 0;
2301
2302	for (; *str && str < end; str++) {
2303		switch (*str) {
2304		case '&':
2305			retval = sbuf_printf(sb, "&amp;");
2306			break;
2307		case '>':
2308			retval = sbuf_printf(sb, "&gt;");
2309			break;
2310		case '<':
2311			retval = sbuf_printf(sb, "&lt;");
2312			break;
2313		default:
2314			retval = sbuf_putc(sb, *str);
2315			break;
2316		}
2317
2318		if (retval != 0)
2319			break;
2320
2321	}
2322
2323	return (retval);
2324}
2325
2326static void
2327ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2328{
2329	struct scsi_vpd_id_descriptor *desc;
2330	int i;
2331
2332	if (id == NULL || id->len < 4)
2333		return;
2334	desc = (struct scsi_vpd_id_descriptor *)id->data;
2335	switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2336	case SVPD_ID_TYPE_T10:
2337		sbuf_printf(sb, "t10.");
2338		break;
2339	case SVPD_ID_TYPE_EUI64:
2340		sbuf_printf(sb, "eui.");
2341		break;
2342	case SVPD_ID_TYPE_NAA:
2343		sbuf_printf(sb, "naa.");
2344		break;
2345	case SVPD_ID_TYPE_SCSI_NAME:
2346		break;
2347	}
2348	switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2349	case SVPD_ID_CODESET_BINARY:
2350		for (i = 0; i < desc->length; i++)
2351			sbuf_printf(sb, "%02x", desc->identifier[i]);
2352		break;
2353	case SVPD_ID_CODESET_ASCII:
2354		sbuf_printf(sb, "%.*s", (int)desc->length,
2355		    (char *)desc->identifier);
2356		break;
2357	case SVPD_ID_CODESET_UTF8:
2358		sbuf_printf(sb, "%s", (char *)desc->identifier);
2359		break;
2360	}
2361}
2362
2363static int
2364ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2365	  struct thread *td)
2366{
2367	struct ctl_softc *softc;
2368	int retval;
2369
2370	softc = control_softc;
2371
2372	retval = 0;
2373
2374	switch (cmd) {
2375	case CTL_IO: {
2376		union ctl_io *io;
2377		void *pool_tmp;
2378
2379		/*
2380		 * If we haven't been "enabled", don't allow any SCSI I/O
2381		 * to this FETD.
2382		 */
2383		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2384			retval = EPERM;
2385			break;
2386		}
2387
2388		io = ctl_alloc_io(softc->ioctl_info.port.ctl_pool_ref);
2389
2390		/*
2391		 * Need to save the pool reference so it doesn't get
2392		 * spammed by the user's ctl_io.
2393		 */
2394		pool_tmp = io->io_hdr.pool;
2395		memcpy(io, (void *)addr, sizeof(*io));
2396		io->io_hdr.pool = pool_tmp;
2397
2398		/*
2399		 * No status yet, so make sure the status is set properly.
2400		 */
2401		io->io_hdr.status = CTL_STATUS_NONE;
2402
2403		/*
2404		 * The user sets the initiator ID, target and LUN IDs.
2405		 */
2406		io->io_hdr.nexus.targ_port = softc->ioctl_info.port.targ_port;
2407		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2408		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2409		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2410			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2411
2412		retval = ctl_ioctl_submit_wait(io);
2413
2414		if (retval != 0) {
2415			ctl_free_io(io);
2416			break;
2417		}
2418
2419		memcpy((void *)addr, io, sizeof(*io));
2420
2421		/* return this to our pool */
2422		ctl_free_io(io);
2423
2424		break;
2425	}
2426	case CTL_ENABLE_PORT:
2427	case CTL_DISABLE_PORT:
2428	case CTL_SET_PORT_WWNS: {
2429		struct ctl_port *port;
2430		struct ctl_port_entry *entry;
2431
2432		entry = (struct ctl_port_entry *)addr;
2433
2434		mtx_lock(&softc->ctl_lock);
2435		STAILQ_FOREACH(port, &softc->port_list, links) {
2436			int action, done;
2437
2438			action = 0;
2439			done = 0;
2440
2441			if ((entry->port_type == CTL_PORT_NONE)
2442			 && (entry->targ_port == port->targ_port)) {
2443				/*
2444				 * If the user only wants to enable or
2445				 * disable or set WWNs on a specific port,
2446				 * do the operation and we're done.
2447				 */
2448				action = 1;
2449				done = 1;
2450			} else if (entry->port_type & port->port_type) {
2451				/*
2452				 * Compare the user's type mask with the
2453				 * particular frontend type to see if we
2454				 * have a match.
2455				 */
2456				action = 1;
2457				done = 0;
2458
2459				/*
2460				 * Make sure the user isn't trying to set
2461				 * WWNs on multiple ports at the same time.
2462				 */
2463				if (cmd == CTL_SET_PORT_WWNS) {
2464					printf("%s: Can't set WWNs on "
2465					       "multiple ports\n", __func__);
2466					retval = EINVAL;
2467					break;
2468				}
2469			}
2470			if (action != 0) {
2471				/*
2472				 * XXX KDM we have to drop the lock here,
2473				 * because the online/offline operations
2474				 * can potentially block.  We need to
2475				 * reference count the frontends so they
2476				 * can't go away,
2477				 */
2478				mtx_unlock(&softc->ctl_lock);
2479
2480				if (cmd == CTL_ENABLE_PORT) {
2481					struct ctl_lun *lun;
2482
2483					STAILQ_FOREACH(lun, &softc->lun_list,
2484						       links) {
2485						port->lun_enable(port->targ_lun_arg,
2486						    lun->target,
2487						    lun->lun);
2488					}
2489
2490					ctl_port_online(port);
2491				} else if (cmd == CTL_DISABLE_PORT) {
2492					struct ctl_lun *lun;
2493
2494					ctl_port_offline(port);
2495
2496					STAILQ_FOREACH(lun, &softc->lun_list,
2497						       links) {
2498						port->lun_disable(
2499						    port->targ_lun_arg,
2500						    lun->target,
2501						    lun->lun);
2502					}
2503				}
2504
2505				mtx_lock(&softc->ctl_lock);
2506
2507				if (cmd == CTL_SET_PORT_WWNS)
2508					ctl_port_set_wwns(port,
2509					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2510					    1 : 0, entry->wwnn,
2511					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2512					    1 : 0, entry->wwpn);
2513			}
2514			if (done != 0)
2515				break;
2516		}
2517		mtx_unlock(&softc->ctl_lock);
2518		break;
2519	}
2520	case CTL_GET_PORT_LIST: {
2521		struct ctl_port *port;
2522		struct ctl_port_list *list;
2523		int i;
2524
2525		list = (struct ctl_port_list *)addr;
2526
2527		if (list->alloc_len != (list->alloc_num *
2528		    sizeof(struct ctl_port_entry))) {
2529			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2530			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2531			       "%zu\n", __func__, list->alloc_len,
2532			       list->alloc_num, sizeof(struct ctl_port_entry));
2533			retval = EINVAL;
2534			break;
2535		}
2536		list->fill_len = 0;
2537		list->fill_num = 0;
2538		list->dropped_num = 0;
2539		i = 0;
2540		mtx_lock(&softc->ctl_lock);
2541		STAILQ_FOREACH(port, &softc->port_list, links) {
2542			struct ctl_port_entry entry, *list_entry;
2543
2544			if (list->fill_num >= list->alloc_num) {
2545				list->dropped_num++;
2546				continue;
2547			}
2548
2549			entry.port_type = port->port_type;
2550			strlcpy(entry.port_name, port->port_name,
2551				sizeof(entry.port_name));
2552			entry.targ_port = port->targ_port;
2553			entry.physical_port = port->physical_port;
2554			entry.virtual_port = port->virtual_port;
2555			entry.wwnn = port->wwnn;
2556			entry.wwpn = port->wwpn;
2557			if (port->status & CTL_PORT_STATUS_ONLINE)
2558				entry.online = 1;
2559			else
2560				entry.online = 0;
2561
2562			list_entry = &list->entries[i];
2563
2564			retval = copyout(&entry, list_entry, sizeof(entry));
2565			if (retval != 0) {
2566				printf("%s: CTL_GET_PORT_LIST: copyout "
2567				       "returned %d\n", __func__, retval);
2568				break;
2569			}
2570			i++;
2571			list->fill_num++;
2572			list->fill_len += sizeof(entry);
2573		}
2574		mtx_unlock(&softc->ctl_lock);
2575
2576		/*
2577		 * If this is non-zero, we had a copyout fault, so there's
2578		 * probably no point in attempting to set the status inside
2579		 * the structure.
2580		 */
2581		if (retval != 0)
2582			break;
2583
2584		if (list->dropped_num > 0)
2585			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2586		else
2587			list->status = CTL_PORT_LIST_OK;
2588		break;
2589	}
2590	case CTL_DUMP_OOA: {
2591		struct ctl_lun *lun;
2592		union ctl_io *io;
2593		char printbuf[128];
2594		struct sbuf sb;
2595
2596		mtx_lock(&softc->ctl_lock);
2597		printf("Dumping OOA queues:\n");
2598		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2599			mtx_lock(&lun->lun_lock);
2600			for (io = (union ctl_io *)TAILQ_FIRST(
2601			     &lun->ooa_queue); io != NULL;
2602			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2603			     ooa_links)) {
2604				sbuf_new(&sb, printbuf, sizeof(printbuf),
2605					 SBUF_FIXEDLEN);
2606				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2607					    (intmax_t)lun->lun,
2608					    io->scsiio.tag_num,
2609					    (io->io_hdr.flags &
2610					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2611					    (io->io_hdr.flags &
2612					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2613					    (io->io_hdr.flags &
2614					    CTL_FLAG_ABORT) ? " ABORT" : "",
2615			                    (io->io_hdr.flags &
2616		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2617				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2618				sbuf_finish(&sb);
2619				printf("%s\n", sbuf_data(&sb));
2620			}
2621			mtx_unlock(&lun->lun_lock);
2622		}
2623		printf("OOA queues dump done\n");
2624		mtx_unlock(&softc->ctl_lock);
2625		break;
2626	}
2627	case CTL_GET_OOA: {
2628		struct ctl_lun *lun;
2629		struct ctl_ooa *ooa_hdr;
2630		struct ctl_ooa_entry *entries;
2631		uint32_t cur_fill_num;
2632
2633		ooa_hdr = (struct ctl_ooa *)addr;
2634
2635		if ((ooa_hdr->alloc_len == 0)
2636		 || (ooa_hdr->alloc_num == 0)) {
2637			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2638			       "must be non-zero\n", __func__,
2639			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2640			retval = EINVAL;
2641			break;
2642		}
2643
2644		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2645		    sizeof(struct ctl_ooa_entry))) {
2646			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2647			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2648			       __func__, ooa_hdr->alloc_len,
2649			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2650			retval = EINVAL;
2651			break;
2652		}
2653
2654		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2655		if (entries == NULL) {
2656			printf("%s: could not allocate %d bytes for OOA "
2657			       "dump\n", __func__, ooa_hdr->alloc_len);
2658			retval = ENOMEM;
2659			break;
2660		}
2661
2662		mtx_lock(&softc->ctl_lock);
2663		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2664		 && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2665		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2666			mtx_unlock(&softc->ctl_lock);
2667			free(entries, M_CTL);
2668			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2669			       __func__, (uintmax_t)ooa_hdr->lun_num);
2670			retval = EINVAL;
2671			break;
2672		}
2673
2674		cur_fill_num = 0;
2675
2676		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2677			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2678				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2679					ooa_hdr, entries);
2680				if (retval != 0)
2681					break;
2682			}
2683			if (retval != 0) {
2684				mtx_unlock(&softc->ctl_lock);
2685				free(entries, M_CTL);
2686				break;
2687			}
2688		} else {
2689			lun = softc->ctl_luns[ooa_hdr->lun_num];
2690
2691			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2692						    entries);
2693		}
2694		mtx_unlock(&softc->ctl_lock);
2695
2696		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2697		ooa_hdr->fill_len = ooa_hdr->fill_num *
2698			sizeof(struct ctl_ooa_entry);
2699		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2700		if (retval != 0) {
2701			printf("%s: error copying out %d bytes for OOA dump\n",
2702			       __func__, ooa_hdr->fill_len);
2703		}
2704
2705		getbintime(&ooa_hdr->cur_bt);
2706
2707		if (cur_fill_num > ooa_hdr->alloc_num) {
2708			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2709			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2710		} else {
2711			ooa_hdr->dropped_num = 0;
2712			ooa_hdr->status = CTL_OOA_OK;
2713		}
2714
2715		free(entries, M_CTL);
2716		break;
2717	}
2718	case CTL_CHECK_OOA: {
2719		union ctl_io *io;
2720		struct ctl_lun *lun;
2721		struct ctl_ooa_info *ooa_info;
2722
2723
2724		ooa_info = (struct ctl_ooa_info *)addr;
2725
2726		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2727			ooa_info->status = CTL_OOA_INVALID_LUN;
2728			break;
2729		}
2730		mtx_lock(&softc->ctl_lock);
2731		lun = softc->ctl_luns[ooa_info->lun_id];
2732		if (lun == NULL) {
2733			mtx_unlock(&softc->ctl_lock);
2734			ooa_info->status = CTL_OOA_INVALID_LUN;
2735			break;
2736		}
2737		mtx_lock(&lun->lun_lock);
2738		mtx_unlock(&softc->ctl_lock);
2739		ooa_info->num_entries = 0;
2740		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2741		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2742		     &io->io_hdr, ooa_links)) {
2743			ooa_info->num_entries++;
2744		}
2745		mtx_unlock(&lun->lun_lock);
2746
2747		ooa_info->status = CTL_OOA_SUCCESS;
2748
2749		break;
2750	}
2751	case CTL_HARD_START:
2752	case CTL_HARD_STOP: {
2753		struct ctl_fe_ioctl_startstop_info ss_info;
2754		struct cfi_metatask *metatask;
2755		struct mtx hs_mtx;
2756
2757		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2758
2759		cv_init(&ss_info.sem, "hard start/stop cv" );
2760
2761		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2762		if (metatask == NULL) {
2763			retval = ENOMEM;
2764			mtx_destroy(&hs_mtx);
2765			break;
2766		}
2767
2768		if (cmd == CTL_HARD_START)
2769			metatask->tasktype = CFI_TASK_STARTUP;
2770		else
2771			metatask->tasktype = CFI_TASK_SHUTDOWN;
2772
2773		metatask->callback = ctl_ioctl_hard_startstop_callback;
2774		metatask->callback_arg = &ss_info;
2775
2776		cfi_action(metatask);
2777
2778		/* Wait for the callback */
2779		mtx_lock(&hs_mtx);
2780		cv_wait_sig(&ss_info.sem, &hs_mtx);
2781		mtx_unlock(&hs_mtx);
2782
2783		/*
2784		 * All information has been copied from the metatask by the
2785		 * time cv_broadcast() is called, so we free the metatask here.
2786		 */
2787		cfi_free_metatask(metatask);
2788
2789		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2790
2791		mtx_destroy(&hs_mtx);
2792		break;
2793	}
2794	case CTL_BBRREAD: {
2795		struct ctl_bbrread_info *bbr_info;
2796		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2797		struct mtx bbr_mtx;
2798		struct cfi_metatask *metatask;
2799
2800		bbr_info = (struct ctl_bbrread_info *)addr;
2801
2802		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2803
2804		bzero(&bbr_mtx, sizeof(bbr_mtx));
2805		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2806
2807		fe_bbr_info.bbr_info = bbr_info;
2808		fe_bbr_info.lock = &bbr_mtx;
2809
2810		cv_init(&fe_bbr_info.sem, "BBR read cv");
2811		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2812
2813		if (metatask == NULL) {
2814			mtx_destroy(&bbr_mtx);
2815			cv_destroy(&fe_bbr_info.sem);
2816			retval = ENOMEM;
2817			break;
2818		}
2819		metatask->tasktype = CFI_TASK_BBRREAD;
2820		metatask->callback = ctl_ioctl_bbrread_callback;
2821		metatask->callback_arg = &fe_bbr_info;
2822		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2823		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2824		metatask->taskinfo.bbrread.len = bbr_info->len;
2825
2826		cfi_action(metatask);
2827
2828		mtx_lock(&bbr_mtx);
2829		while (fe_bbr_info.wakeup_done == 0)
2830			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2831		mtx_unlock(&bbr_mtx);
2832
2833		bbr_info->status = metatask->status;
2834		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2835		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2836		memcpy(&bbr_info->sense_data,
2837		       &metatask->taskinfo.bbrread.sense_data,
2838		       ctl_min(sizeof(bbr_info->sense_data),
2839			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2840
2841		cfi_free_metatask(metatask);
2842
2843		mtx_destroy(&bbr_mtx);
2844		cv_destroy(&fe_bbr_info.sem);
2845
2846		break;
2847	}
2848	case CTL_DELAY_IO: {
2849		struct ctl_io_delay_info *delay_info;
2850#ifdef CTL_IO_DELAY
2851		struct ctl_lun *lun;
2852#endif /* CTL_IO_DELAY */
2853
2854		delay_info = (struct ctl_io_delay_info *)addr;
2855
2856#ifdef CTL_IO_DELAY
2857		mtx_lock(&softc->ctl_lock);
2858
2859		if ((delay_info->lun_id >= CTL_MAX_LUNS)
2860		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2861			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2862		} else {
2863			lun = softc->ctl_luns[delay_info->lun_id];
2864			mtx_lock(&lun->lun_lock);
2865
2866			delay_info->status = CTL_DELAY_STATUS_OK;
2867
2868			switch (delay_info->delay_type) {
2869			case CTL_DELAY_TYPE_CONT:
2870				break;
2871			case CTL_DELAY_TYPE_ONESHOT:
2872				break;
2873			default:
2874				delay_info->status =
2875					CTL_DELAY_STATUS_INVALID_TYPE;
2876				break;
2877			}
2878
2879			switch (delay_info->delay_loc) {
2880			case CTL_DELAY_LOC_DATAMOVE:
2881				lun->delay_info.datamove_type =
2882					delay_info->delay_type;
2883				lun->delay_info.datamove_delay =
2884					delay_info->delay_secs;
2885				break;
2886			case CTL_DELAY_LOC_DONE:
2887				lun->delay_info.done_type =
2888					delay_info->delay_type;
2889				lun->delay_info.done_delay =
2890					delay_info->delay_secs;
2891				break;
2892			default:
2893				delay_info->status =
2894					CTL_DELAY_STATUS_INVALID_LOC;
2895				break;
2896			}
2897			mtx_unlock(&lun->lun_lock);
2898		}
2899
2900		mtx_unlock(&softc->ctl_lock);
2901#else
2902		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2903#endif /* CTL_IO_DELAY */
2904		break;
2905	}
2906	case CTL_REALSYNC_SET: {
2907		int *syncstate;
2908
2909		syncstate = (int *)addr;
2910
2911		mtx_lock(&softc->ctl_lock);
2912		switch (*syncstate) {
2913		case 0:
2914			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2915			break;
2916		case 1:
2917			softc->flags |= CTL_FLAG_REAL_SYNC;
2918			break;
2919		default:
2920			retval = EINVAL;
2921			break;
2922		}
2923		mtx_unlock(&softc->ctl_lock);
2924		break;
2925	}
2926	case CTL_REALSYNC_GET: {
2927		int *syncstate;
2928
2929		syncstate = (int*)addr;
2930
2931		mtx_lock(&softc->ctl_lock);
2932		if (softc->flags & CTL_FLAG_REAL_SYNC)
2933			*syncstate = 1;
2934		else
2935			*syncstate = 0;
2936		mtx_unlock(&softc->ctl_lock);
2937
2938		break;
2939	}
2940	case CTL_SETSYNC:
2941	case CTL_GETSYNC: {
2942		struct ctl_sync_info *sync_info;
2943		struct ctl_lun *lun;
2944
2945		sync_info = (struct ctl_sync_info *)addr;
2946
2947		mtx_lock(&softc->ctl_lock);
2948		lun = softc->ctl_luns[sync_info->lun_id];
2949		if (lun == NULL) {
2950			mtx_unlock(&softc->ctl_lock);
2951			sync_info->status = CTL_GS_SYNC_NO_LUN;
2952		}
2953		/*
2954		 * Get or set the sync interval.  We're not bounds checking
2955		 * in the set case, hopefully the user won't do something
2956		 * silly.
2957		 */
2958		mtx_lock(&lun->lun_lock);
2959		mtx_unlock(&softc->ctl_lock);
2960		if (cmd == CTL_GETSYNC)
2961			sync_info->sync_interval = lun->sync_interval;
2962		else
2963			lun->sync_interval = sync_info->sync_interval;
2964		mtx_unlock(&lun->lun_lock);
2965
2966		sync_info->status = CTL_GS_SYNC_OK;
2967
2968		break;
2969	}
2970	case CTL_GETSTATS: {
2971		struct ctl_stats *stats;
2972		struct ctl_lun *lun;
2973		int i;
2974
2975		stats = (struct ctl_stats *)addr;
2976
2977		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2978		     stats->alloc_len) {
2979			stats->status = CTL_SS_NEED_MORE_SPACE;
2980			stats->num_luns = softc->num_luns;
2981			break;
2982		}
2983		/*
2984		 * XXX KDM no locking here.  If the LUN list changes,
2985		 * things can blow up.
2986		 */
2987		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2988		     i++, lun = STAILQ_NEXT(lun, links)) {
2989			retval = copyout(&lun->stats, &stats->lun_stats[i],
2990					 sizeof(lun->stats));
2991			if (retval != 0)
2992				break;
2993		}
2994		stats->num_luns = softc->num_luns;
2995		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2996				 softc->num_luns;
2997		stats->status = CTL_SS_OK;
2998#ifdef CTL_TIME_IO
2999		stats->flags = CTL_STATS_FLAG_TIME_VALID;
3000#else
3001		stats->flags = CTL_STATS_FLAG_NONE;
3002#endif
3003		getnanouptime(&stats->timestamp);
3004		break;
3005	}
3006	case CTL_ERROR_INJECT: {
3007		struct ctl_error_desc *err_desc, *new_err_desc;
3008		struct ctl_lun *lun;
3009
3010		err_desc = (struct ctl_error_desc *)addr;
3011
3012		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
3013				      M_WAITOK | M_ZERO);
3014		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
3015
3016		mtx_lock(&softc->ctl_lock);
3017		lun = softc->ctl_luns[err_desc->lun_id];
3018		if (lun == NULL) {
3019			mtx_unlock(&softc->ctl_lock);
3020			free(new_err_desc, M_CTL);
3021			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
3022			       __func__, (uintmax_t)err_desc->lun_id);
3023			retval = EINVAL;
3024			break;
3025		}
3026		mtx_lock(&lun->lun_lock);
3027		mtx_unlock(&softc->ctl_lock);
3028
3029		/*
3030		 * We could do some checking here to verify the validity
3031		 * of the request, but given the complexity of error
3032		 * injection requests, the checking logic would be fairly
3033		 * complex.
3034		 *
3035		 * For now, if the request is invalid, it just won't get
3036		 * executed and might get deleted.
3037		 */
3038		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
3039
3040		/*
3041		 * XXX KDM check to make sure the serial number is unique,
3042		 * in case we somehow manage to wrap.  That shouldn't
3043		 * happen for a very long time, but it's the right thing to
3044		 * do.
3045		 */
3046		new_err_desc->serial = lun->error_serial;
3047		err_desc->serial = lun->error_serial;
3048		lun->error_serial++;
3049
3050		mtx_unlock(&lun->lun_lock);
3051		break;
3052	}
3053	case CTL_ERROR_INJECT_DELETE: {
3054		struct ctl_error_desc *delete_desc, *desc, *desc2;
3055		struct ctl_lun *lun;
3056		int delete_done;
3057
3058		delete_desc = (struct ctl_error_desc *)addr;
3059		delete_done = 0;
3060
3061		mtx_lock(&softc->ctl_lock);
3062		lun = softc->ctl_luns[delete_desc->lun_id];
3063		if (lun == NULL) {
3064			mtx_unlock(&softc->ctl_lock);
3065			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
3066			       __func__, (uintmax_t)delete_desc->lun_id);
3067			retval = EINVAL;
3068			break;
3069		}
3070		mtx_lock(&lun->lun_lock);
3071		mtx_unlock(&softc->ctl_lock);
3072		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
3073			if (desc->serial != delete_desc->serial)
3074				continue;
3075
3076			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
3077				      links);
3078			free(desc, M_CTL);
3079			delete_done = 1;
3080		}
3081		mtx_unlock(&lun->lun_lock);
3082		if (delete_done == 0) {
3083			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
3084			       "error serial %ju on LUN %u\n", __func__,
3085			       delete_desc->serial, delete_desc->lun_id);
3086			retval = EINVAL;
3087			break;
3088		}
3089		break;
3090	}
3091	case CTL_DUMP_STRUCTS: {
3092		int i, j, k;
3093		struct ctl_port *port;
3094		struct ctl_frontend *fe;
3095
3096		mtx_lock(&softc->ctl_lock);
3097		printf("CTL Persistent Reservation information start:\n");
3098		for (i = 0; i < CTL_MAX_LUNS; i++) {
3099			struct ctl_lun *lun;
3100
3101			lun = softc->ctl_luns[i];
3102
3103			if ((lun == NULL)
3104			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
3105				continue;
3106
3107			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
3108				if (lun->pr_keys[j] == NULL)
3109					continue;
3110				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
3111					if (lun->pr_keys[j][k] == 0)
3112						continue;
3113					printf("  LUN %d port %d iid %d key "
3114					       "%#jx\n", i, j, k,
3115					       (uintmax_t)lun->pr_keys[j][k]);
3116				}
3117			}
3118		}
3119		printf("CTL Persistent Reservation information end\n");
3120		printf("CTL Ports:\n");
3121		STAILQ_FOREACH(port, &softc->port_list, links) {
3122			printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3123			       "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3124			       port->frontend->name, port->port_type,
3125			       port->physical_port, port->virtual_port,
3126			       (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3127			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3128				if (port->wwpn_iid[j].in_use == 0 &&
3129				    port->wwpn_iid[j].wwpn == 0 &&
3130				    port->wwpn_iid[j].name == NULL)
3131					continue;
3132
3133				printf("    iid %u use %d WWPN %#jx '%s'\n",
3134				    j, port->wwpn_iid[j].in_use,
3135				    (uintmax_t)port->wwpn_iid[j].wwpn,
3136				    port->wwpn_iid[j].name);
3137			}
3138		}
3139		printf("CTL Port information end\n");
3140		mtx_unlock(&softc->ctl_lock);
3141		/*
3142		 * XXX KDM calling this without a lock.  We'd likely want
3143		 * to drop the lock before calling the frontend's dump
3144		 * routine anyway.
3145		 */
3146		printf("CTL Frontends:\n");
3147		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3148			printf("  Frontend '%s'\n", fe->name);
3149			if (fe->fe_dump != NULL)
3150				fe->fe_dump();
3151		}
3152		printf("CTL Frontend information end\n");
3153		break;
3154	}
3155	case CTL_LUN_REQ: {
3156		struct ctl_lun_req *lun_req;
3157		struct ctl_backend_driver *backend;
3158
3159		lun_req = (struct ctl_lun_req *)addr;
3160
3161		backend = ctl_backend_find(lun_req->backend);
3162		if (backend == NULL) {
3163			lun_req->status = CTL_LUN_ERROR;
3164			snprintf(lun_req->error_str,
3165				 sizeof(lun_req->error_str),
3166				 "Backend \"%s\" not found.",
3167				 lun_req->backend);
3168			break;
3169		}
3170		if (lun_req->num_be_args > 0) {
3171			lun_req->kern_be_args = ctl_copyin_args(
3172				lun_req->num_be_args,
3173				lun_req->be_args,
3174				lun_req->error_str,
3175				sizeof(lun_req->error_str));
3176			if (lun_req->kern_be_args == NULL) {
3177				lun_req->status = CTL_LUN_ERROR;
3178				break;
3179			}
3180		}
3181
3182		retval = backend->ioctl(dev, cmd, addr, flag, td);
3183
3184		if (lun_req->num_be_args > 0) {
3185			ctl_copyout_args(lun_req->num_be_args,
3186				      lun_req->kern_be_args);
3187			ctl_free_args(lun_req->num_be_args,
3188				      lun_req->kern_be_args);
3189		}
3190		break;
3191	}
3192	case CTL_LUN_LIST: {
3193		struct sbuf *sb;
3194		struct ctl_lun *lun;
3195		struct ctl_lun_list *list;
3196		struct ctl_option *opt;
3197
3198		list = (struct ctl_lun_list *)addr;
3199
3200		/*
3201		 * Allocate a fixed length sbuf here, based on the length
3202		 * of the user's buffer.  We could allocate an auto-extending
3203		 * buffer, and then tell the user how much larger our
3204		 * amount of data is than his buffer, but that presents
3205		 * some problems:
3206		 *
3207		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3208		 *     we can't hold a lock while calling them with an
3209		 *     auto-extending buffer.
3210 		 *
3211		 * 2.  There is not currently a LUN reference counting
3212		 *     mechanism, outside of outstanding transactions on
3213		 *     the LUN's OOA queue.  So a LUN could go away on us
3214		 *     while we're getting the LUN number, backend-specific
3215		 *     information, etc.  Thus, given the way things
3216		 *     currently work, we need to hold the CTL lock while
3217		 *     grabbing LUN information.
3218		 *
3219		 * So, from the user's standpoint, the best thing to do is
3220		 * allocate what he thinks is a reasonable buffer length,
3221		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3222		 * double the buffer length and try again.  (And repeat
3223		 * that until he succeeds.)
3224		 */
3225		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3226		if (sb == NULL) {
3227			list->status = CTL_LUN_LIST_ERROR;
3228			snprintf(list->error_str, sizeof(list->error_str),
3229				 "Unable to allocate %d bytes for LUN list",
3230				 list->alloc_len);
3231			break;
3232		}
3233
3234		sbuf_printf(sb, "<ctllunlist>\n");
3235
3236		mtx_lock(&softc->ctl_lock);
3237		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3238			mtx_lock(&lun->lun_lock);
3239			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3240					     (uintmax_t)lun->lun);
3241
3242			/*
3243			 * Bail out as soon as we see that we've overfilled
3244			 * the buffer.
3245			 */
3246			if (retval != 0)
3247				break;
3248
3249			retval = sbuf_printf(sb, "\t<backend_type>%s"
3250					     "</backend_type>\n",
3251					     (lun->backend == NULL) ?  "none" :
3252					     lun->backend->name);
3253
3254			if (retval != 0)
3255				break;
3256
3257			retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3258					     lun->be_lun->lun_type);
3259
3260			if (retval != 0)
3261				break;
3262
3263			if (lun->backend == NULL) {
3264				retval = sbuf_printf(sb, "</lun>\n");
3265				if (retval != 0)
3266					break;
3267				continue;
3268			}
3269
3270			retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3271					     (lun->be_lun->maxlba > 0) ?
3272					     lun->be_lun->maxlba + 1 : 0);
3273
3274			if (retval != 0)
3275				break;
3276
3277			retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3278					     lun->be_lun->blocksize);
3279
3280			if (retval != 0)
3281				break;
3282
3283			retval = sbuf_printf(sb, "\t<serial_number>");
3284
3285			if (retval != 0)
3286				break;
3287
3288			retval = ctl_sbuf_printf_esc(sb,
3289			    lun->be_lun->serial_num,
3290			    sizeof(lun->be_lun->serial_num));
3291
3292			if (retval != 0)
3293				break;
3294
3295			retval = sbuf_printf(sb, "</serial_number>\n");
3296
3297			if (retval != 0)
3298				break;
3299
3300			retval = sbuf_printf(sb, "\t<device_id>");
3301
3302			if (retval != 0)
3303				break;
3304
3305			retval = ctl_sbuf_printf_esc(sb,
3306			    lun->be_lun->device_id,
3307			    sizeof(lun->be_lun->device_id));
3308
3309			if (retval != 0)
3310				break;
3311
3312			retval = sbuf_printf(sb, "</device_id>\n");
3313
3314			if (retval != 0)
3315				break;
3316
3317			if (lun->backend->lun_info != NULL) {
3318				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3319				if (retval != 0)
3320					break;
3321			}
3322			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3323				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3324				    opt->name, opt->value, opt->name);
3325				if (retval != 0)
3326					break;
3327			}
3328
3329			retval = sbuf_printf(sb, "</lun>\n");
3330
3331			if (retval != 0)
3332				break;
3333			mtx_unlock(&lun->lun_lock);
3334		}
3335		if (lun != NULL)
3336			mtx_unlock(&lun->lun_lock);
3337		mtx_unlock(&softc->ctl_lock);
3338
3339		if ((retval != 0)
3340		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3341			retval = 0;
3342			sbuf_delete(sb);
3343			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3344			snprintf(list->error_str, sizeof(list->error_str),
3345				 "Out of space, %d bytes is too small",
3346				 list->alloc_len);
3347			break;
3348		}
3349
3350		sbuf_finish(sb);
3351
3352		retval = copyout(sbuf_data(sb), list->lun_xml,
3353				 sbuf_len(sb) + 1);
3354
3355		list->fill_len = sbuf_len(sb) + 1;
3356		list->status = CTL_LUN_LIST_OK;
3357		sbuf_delete(sb);
3358		break;
3359	}
3360	case CTL_ISCSI: {
3361		struct ctl_iscsi *ci;
3362		struct ctl_frontend *fe;
3363
3364		ci = (struct ctl_iscsi *)addr;
3365
3366		fe = ctl_frontend_find("iscsi");
3367		if (fe == NULL) {
3368			ci->status = CTL_ISCSI_ERROR;
3369			snprintf(ci->error_str, sizeof(ci->error_str),
3370			    "Frontend \"iscsi\" not found.");
3371			break;
3372		}
3373
3374		retval = fe->ioctl(dev, cmd, addr, flag, td);
3375		break;
3376	}
3377	case CTL_PORT_REQ: {
3378		struct ctl_req *req;
3379		struct ctl_frontend *fe;
3380
3381		req = (struct ctl_req *)addr;
3382
3383		fe = ctl_frontend_find(req->driver);
3384		if (fe == NULL) {
3385			req->status = CTL_LUN_ERROR;
3386			snprintf(req->error_str, sizeof(req->error_str),
3387			    "Frontend \"%s\" not found.", req->driver);
3388			break;
3389		}
3390		if (req->num_args > 0) {
3391			req->kern_args = ctl_copyin_args(req->num_args,
3392			    req->args, req->error_str, sizeof(req->error_str));
3393			if (req->kern_args == NULL) {
3394				req->status = CTL_LUN_ERROR;
3395				break;
3396			}
3397		}
3398
3399		retval = fe->ioctl(dev, cmd, addr, flag, td);
3400
3401		if (req->num_args > 0) {
3402			ctl_copyout_args(req->num_args, req->kern_args);
3403			ctl_free_args(req->num_args, req->kern_args);
3404		}
3405		break;
3406	}
3407	case CTL_PORT_LIST: {
3408		struct sbuf *sb;
3409		struct ctl_port *port;
3410		struct ctl_lun_list *list;
3411		struct ctl_option *opt;
3412		int j;
3413
3414		list = (struct ctl_lun_list *)addr;
3415
3416		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3417		if (sb == NULL) {
3418			list->status = CTL_LUN_LIST_ERROR;
3419			snprintf(list->error_str, sizeof(list->error_str),
3420				 "Unable to allocate %d bytes for LUN list",
3421				 list->alloc_len);
3422			break;
3423		}
3424
3425		sbuf_printf(sb, "<ctlportlist>\n");
3426
3427		mtx_lock(&softc->ctl_lock);
3428		STAILQ_FOREACH(port, &softc->port_list, links) {
3429			retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3430					     (uintmax_t)port->targ_port);
3431
3432			/*
3433			 * Bail out as soon as we see that we've overfilled
3434			 * the buffer.
3435			 */
3436			if (retval != 0)
3437				break;
3438
3439			retval = sbuf_printf(sb, "\t<frontend_type>%s"
3440			    "</frontend_type>\n", port->frontend->name);
3441			if (retval != 0)
3442				break;
3443
3444			retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3445					     port->port_type);
3446			if (retval != 0)
3447				break;
3448
3449			retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3450			    (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3451			if (retval != 0)
3452				break;
3453
3454			retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3455			    port->port_name);
3456			if (retval != 0)
3457				break;
3458
3459			retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3460			    port->physical_port);
3461			if (retval != 0)
3462				break;
3463
3464			retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3465			    port->virtual_port);
3466			if (retval != 0)
3467				break;
3468
3469			if (port->target_devid != NULL) {
3470				sbuf_printf(sb, "\t<target>");
3471				ctl_id_sbuf(port->target_devid, sb);
3472				sbuf_printf(sb, "</target>\n");
3473			}
3474
3475			if (port->port_devid != NULL) {
3476				sbuf_printf(sb, "\t<port>");
3477				ctl_id_sbuf(port->port_devid, sb);
3478				sbuf_printf(sb, "</port>\n");
3479			}
3480
3481			if (port->port_info != NULL) {
3482				retval = port->port_info(port->onoff_arg, sb);
3483				if (retval != 0)
3484					break;
3485			}
3486			STAILQ_FOREACH(opt, &port->options, links) {
3487				retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3488				    opt->name, opt->value, opt->name);
3489				if (retval != 0)
3490					break;
3491			}
3492
3493			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3494				if (port->wwpn_iid[j].in_use == 0 ||
3495				    (port->wwpn_iid[j].wwpn == 0 &&
3496				     port->wwpn_iid[j].name == NULL))
3497					continue;
3498
3499				if (port->wwpn_iid[j].name != NULL)
3500					retval = sbuf_printf(sb,
3501					    "\t<initiator>%u %s</initiator>\n",
3502					    j, port->wwpn_iid[j].name);
3503				else
3504					retval = sbuf_printf(sb,
3505					    "\t<initiator>%u naa.%08jx</initiator>\n",
3506					    j, port->wwpn_iid[j].wwpn);
3507				if (retval != 0)
3508					break;
3509			}
3510			if (retval != 0)
3511				break;
3512
3513			retval = sbuf_printf(sb, "</targ_port>\n");
3514			if (retval != 0)
3515				break;
3516		}
3517		mtx_unlock(&softc->ctl_lock);
3518
3519		if ((retval != 0)
3520		 || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3521			retval = 0;
3522			sbuf_delete(sb);
3523			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3524			snprintf(list->error_str, sizeof(list->error_str),
3525				 "Out of space, %d bytes is too small",
3526				 list->alloc_len);
3527			break;
3528		}
3529
3530		sbuf_finish(sb);
3531
3532		retval = copyout(sbuf_data(sb), list->lun_xml,
3533				 sbuf_len(sb) + 1);
3534
3535		list->fill_len = sbuf_len(sb) + 1;
3536		list->status = CTL_LUN_LIST_OK;
3537		sbuf_delete(sb);
3538		break;
3539	}
3540	default: {
3541		/* XXX KDM should we fix this? */
3542#if 0
3543		struct ctl_backend_driver *backend;
3544		unsigned int type;
3545		int found;
3546
3547		found = 0;
3548
3549		/*
3550		 * We encode the backend type as the ioctl type for backend
3551		 * ioctls.  So parse it out here, and then search for a
3552		 * backend of this type.
3553		 */
3554		type = _IOC_TYPE(cmd);
3555
3556		STAILQ_FOREACH(backend, &softc->be_list, links) {
3557			if (backend->type == type) {
3558				found = 1;
3559				break;
3560			}
3561		}
3562		if (found == 0) {
3563			printf("ctl: unknown ioctl command %#lx or backend "
3564			       "%d\n", cmd, type);
3565			retval = EINVAL;
3566			break;
3567		}
3568		retval = backend->ioctl(dev, cmd, addr, flag, td);
3569#endif
3570		retval = ENOTTY;
3571		break;
3572	}
3573	}
3574	return (retval);
3575}
3576
3577uint32_t
3578ctl_get_initindex(struct ctl_nexus *nexus)
3579{
3580	if (nexus->targ_port < CTL_MAX_PORTS)
3581		return (nexus->initid.id +
3582			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3583	else
3584		return (nexus->initid.id +
3585		       ((nexus->targ_port - CTL_MAX_PORTS) *
3586			CTL_MAX_INIT_PER_PORT));
3587}
3588
3589uint32_t
3590ctl_get_resindex(struct ctl_nexus *nexus)
3591{
3592	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3593}
3594
3595uint32_t
3596ctl_port_idx(int port_num)
3597{
3598	if (port_num < CTL_MAX_PORTS)
3599		return(port_num);
3600	else
3601		return(port_num - CTL_MAX_PORTS);
3602}
3603
3604static uint32_t
3605ctl_map_lun(int port_num, uint32_t lun_id)
3606{
3607	struct ctl_port *port;
3608
3609	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3610	if (port == NULL)
3611		return (UINT32_MAX);
3612	if (port->lun_map == NULL)
3613		return (lun_id);
3614	return (port->lun_map(port->targ_lun_arg, lun_id));
3615}
3616
3617static uint32_t
3618ctl_map_lun_back(int port_num, uint32_t lun_id)
3619{
3620	struct ctl_port *port;
3621	uint32_t i;
3622
3623	port = control_softc->ctl_ports[ctl_port_idx(port_num)];
3624	if (port->lun_map == NULL)
3625		return (lun_id);
3626	for (i = 0; i < CTL_MAX_LUNS; i++) {
3627		if (port->lun_map(port->targ_lun_arg, i) == lun_id)
3628			return (i);
3629	}
3630	return (UINT32_MAX);
3631}
3632
3633/*
3634 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3635 * that are a power of 2.
3636 */
3637int
3638ctl_ffz(uint32_t *mask, uint32_t size)
3639{
3640	uint32_t num_chunks, num_pieces;
3641	int i, j;
3642
3643	num_chunks = (size >> 5);
3644	if (num_chunks == 0)
3645		num_chunks++;
3646	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3647
3648	for (i = 0; i < num_chunks; i++) {
3649		for (j = 0; j < num_pieces; j++) {
3650			if ((mask[i] & (1 << j)) == 0)
3651				return ((i << 5) + j);
3652		}
3653	}
3654
3655	return (-1);
3656}
3657
3658int
3659ctl_set_mask(uint32_t *mask, uint32_t bit)
3660{
3661	uint32_t chunk, piece;
3662
3663	chunk = bit >> 5;
3664	piece = bit % (sizeof(uint32_t) * 8);
3665
3666	if ((mask[chunk] & (1 << piece)) != 0)
3667		return (-1);
3668	else
3669		mask[chunk] |= (1 << piece);
3670
3671	return (0);
3672}
3673
3674int
3675ctl_clear_mask(uint32_t *mask, uint32_t bit)
3676{
3677	uint32_t chunk, piece;
3678
3679	chunk = bit >> 5;
3680	piece = bit % (sizeof(uint32_t) * 8);
3681
3682	if ((mask[chunk] & (1 << piece)) == 0)
3683		return (-1);
3684	else
3685		mask[chunk] &= ~(1 << piece);
3686
3687	return (0);
3688}
3689
3690int
3691ctl_is_set(uint32_t *mask, uint32_t bit)
3692{
3693	uint32_t chunk, piece;
3694
3695	chunk = bit >> 5;
3696	piece = bit % (sizeof(uint32_t) * 8);
3697
3698	if ((mask[chunk] & (1 << piece)) == 0)
3699		return (0);
3700	else
3701		return (1);
3702}
3703
3704static uint64_t
3705ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3706{
3707	uint64_t *t;
3708
3709	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3710	if (t == NULL)
3711		return (0);
3712	return (t[residx % CTL_MAX_INIT_PER_PORT]);
3713}
3714
3715static void
3716ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3717{
3718	uint64_t *t;
3719
3720	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3721	if (t == NULL)
3722		return;
3723	t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3724}
3725
3726static void
3727ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3728{
3729	uint64_t *p;
3730	u_int i;
3731
3732	i = residx/CTL_MAX_INIT_PER_PORT;
3733	if (lun->pr_keys[i] != NULL)
3734		return;
3735	mtx_unlock(&lun->lun_lock);
3736	p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3737	    M_WAITOK | M_ZERO);
3738	mtx_lock(&lun->lun_lock);
3739	if (lun->pr_keys[i] == NULL)
3740		lun->pr_keys[i] = p;
3741	else
3742		free(p, M_CTL);
3743}
3744
3745static void
3746ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3747{
3748	uint64_t *t;
3749
3750	t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3751	KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3752	t[residx % CTL_MAX_INIT_PER_PORT] = key;
3753}
3754
3755/*
3756 * ctl_softc, pool_name, total_ctl_io are passed in.
3757 * npool is passed out.
3758 */
3759int
3760ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3761		uint32_t total_ctl_io, void **npool)
3762{
3763#ifdef IO_POOLS
3764	struct ctl_io_pool *pool;
3765
3766	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3767					    M_NOWAIT | M_ZERO);
3768	if (pool == NULL)
3769		return (ENOMEM);
3770
3771	snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3772	pool->ctl_softc = ctl_softc;
3773	pool->zone = uma_zsecond_create(pool->name, NULL,
3774	    NULL, NULL, NULL, ctl_softc->io_zone);
3775	/* uma_prealloc(pool->zone, total_ctl_io); */
3776
3777	*npool = pool;
3778#else
3779	*npool = ctl_softc->io_zone;
3780#endif
3781	return (0);
3782}
3783
3784void
3785ctl_pool_free(struct ctl_io_pool *pool)
3786{
3787
3788	if (pool == NULL)
3789		return;
3790
3791#ifdef IO_POOLS
3792	uma_zdestroy(pool->zone);
3793	free(pool, M_CTL);
3794#endif
3795}
3796
3797union ctl_io *
3798ctl_alloc_io(void *pool_ref)
3799{
3800	union ctl_io *io;
3801#ifdef IO_POOLS
3802	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3803
3804	io = uma_zalloc(pool->zone, M_WAITOK);
3805#else
3806	io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3807#endif
3808	if (io != NULL)
3809		io->io_hdr.pool = pool_ref;
3810	return (io);
3811}
3812
3813union ctl_io *
3814ctl_alloc_io_nowait(void *pool_ref)
3815{
3816	union ctl_io *io;
3817#ifdef IO_POOLS
3818	struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3819
3820	io = uma_zalloc(pool->zone, M_NOWAIT);
3821#else
3822	io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3823#endif
3824	if (io != NULL)
3825		io->io_hdr.pool = pool_ref;
3826	return (io);
3827}
3828
3829void
3830ctl_free_io(union ctl_io *io)
3831{
3832#ifdef IO_POOLS
3833	struct ctl_io_pool *pool;
3834#endif
3835
3836	if (io == NULL)
3837		return;
3838
3839#ifdef IO_POOLS
3840	pool = (struct ctl_io_pool *)io->io_hdr.pool;
3841	uma_zfree(pool->zone, io);
3842#else
3843	uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3844#endif
3845}
3846
3847void
3848ctl_zero_io(union ctl_io *io)
3849{
3850	void *pool_ref;
3851
3852	if (io == NULL)
3853		return;
3854
3855	/*
3856	 * May need to preserve linked list pointers at some point too.
3857	 */
3858	pool_ref = io->io_hdr.pool;
3859	memset(io, 0, sizeof(*io));
3860	io->io_hdr.pool = pool_ref;
3861}
3862
3863/*
3864 * This routine is currently used for internal copies of ctl_ios that need
3865 * to persist for some reason after we've already returned status to the
3866 * FETD.  (Thus the flag set.)
3867 *
3868 * XXX XXX
3869 * Note that this makes a blind copy of all fields in the ctl_io, except
3870 * for the pool reference.  This includes any memory that has been
3871 * allocated!  That memory will no longer be valid after done has been
3872 * called, so this would be VERY DANGEROUS for command that actually does
3873 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3874 * start and stop commands, which don't transfer any data, so this is not a
3875 * problem.  If it is used for anything else, the caller would also need to
3876 * allocate data buffer space and this routine would need to be modified to
3877 * copy the data buffer(s) as well.
3878 */
3879void
3880ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3881{
3882	void *pool_ref;
3883
3884	if ((src == NULL)
3885	 || (dest == NULL))
3886		return;
3887
3888	/*
3889	 * May need to preserve linked list pointers at some point too.
3890	 */
3891	pool_ref = dest->io_hdr.pool;
3892
3893	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3894
3895	dest->io_hdr.pool = pool_ref;
3896	/*
3897	 * We need to know that this is an internal copy, and doesn't need
3898	 * to get passed back to the FETD that allocated it.
3899	 */
3900	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3901}
3902
3903int
3904ctl_expand_number(const char *buf, uint64_t *num)
3905{
3906	char *endptr;
3907	uint64_t number;
3908	unsigned shift;
3909
3910	number = strtoq(buf, &endptr, 0);
3911
3912	switch (tolower((unsigned char)*endptr)) {
3913	case 'e':
3914		shift = 60;
3915		break;
3916	case 'p':
3917		shift = 50;
3918		break;
3919	case 't':
3920		shift = 40;
3921		break;
3922	case 'g':
3923		shift = 30;
3924		break;
3925	case 'm':
3926		shift = 20;
3927		break;
3928	case 'k':
3929		shift = 10;
3930		break;
3931	case 'b':
3932	case '\0': /* No unit. */
3933		*num = number;
3934		return (0);
3935	default:
3936		/* Unrecognized unit. */
3937		return (-1);
3938	}
3939
3940	if ((number << shift) >> shift != number) {
3941		/* Overflow */
3942		return (-1);
3943	}
3944	*num = number << shift;
3945	return (0);
3946}
3947
3948
3949/*
3950 * This routine could be used in the future to load default and/or saved
3951 * mode page parameters for a particuar lun.
3952 */
3953static int
3954ctl_init_page_index(struct ctl_lun *lun)
3955{
3956	int i;
3957	struct ctl_page_index *page_index;
3958	const char *value;
3959	uint64_t ival;
3960
3961	memcpy(&lun->mode_pages.index, page_index_template,
3962	       sizeof(page_index_template));
3963
3964	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3965
3966		page_index = &lun->mode_pages.index[i];
3967		/*
3968		 * If this is a disk-only mode page, there's no point in
3969		 * setting it up.  For some pages, we have to have some
3970		 * basic information about the disk in order to calculate the
3971		 * mode page data.
3972		 */
3973		if ((lun->be_lun->lun_type != T_DIRECT)
3974		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3975			continue;
3976
3977		switch (page_index->page_code & SMPH_PC_MASK) {
3978		case SMS_RW_ERROR_RECOVERY_PAGE: {
3979			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3980				panic("subpage is incorrect!");
3981			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3982			       &rw_er_page_default,
3983			       sizeof(rw_er_page_default));
3984			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3985			       &rw_er_page_changeable,
3986			       sizeof(rw_er_page_changeable));
3987			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3988			       &rw_er_page_default,
3989			       sizeof(rw_er_page_default));
3990			memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3991			       &rw_er_page_default,
3992			       sizeof(rw_er_page_default));
3993			page_index->page_data =
3994				(uint8_t *)lun->mode_pages.rw_er_page;
3995			break;
3996		}
3997		case SMS_FORMAT_DEVICE_PAGE: {
3998			struct scsi_format_page *format_page;
3999
4000			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4001				panic("subpage is incorrect!");
4002
4003			/*
4004			 * Sectors per track are set above.  Bytes per
4005			 * sector need to be set here on a per-LUN basis.
4006			 */
4007			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4008			       &format_page_default,
4009			       sizeof(format_page_default));
4010			memcpy(&lun->mode_pages.format_page[
4011			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
4012			       sizeof(format_page_changeable));
4013			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4014			       &format_page_default,
4015			       sizeof(format_page_default));
4016			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4017			       &format_page_default,
4018			       sizeof(format_page_default));
4019
4020			format_page = &lun->mode_pages.format_page[
4021				CTL_PAGE_CURRENT];
4022			scsi_ulto2b(lun->be_lun->blocksize,
4023				    format_page->bytes_per_sector);
4024
4025			format_page = &lun->mode_pages.format_page[
4026				CTL_PAGE_DEFAULT];
4027			scsi_ulto2b(lun->be_lun->blocksize,
4028				    format_page->bytes_per_sector);
4029
4030			format_page = &lun->mode_pages.format_page[
4031				CTL_PAGE_SAVED];
4032			scsi_ulto2b(lun->be_lun->blocksize,
4033				    format_page->bytes_per_sector);
4034
4035			page_index->page_data =
4036				(uint8_t *)lun->mode_pages.format_page;
4037			break;
4038		}
4039		case SMS_RIGID_DISK_PAGE: {
4040			struct scsi_rigid_disk_page *rigid_disk_page;
4041			uint32_t sectors_per_cylinder;
4042			uint64_t cylinders;
4043#ifndef	__XSCALE__
4044			int shift;
4045#endif /* !__XSCALE__ */
4046
4047			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4048				panic("invalid subpage value %d",
4049				      page_index->subpage);
4050
4051			/*
4052			 * Rotation rate and sectors per track are set
4053			 * above.  We calculate the cylinders here based on
4054			 * capacity.  Due to the number of heads and
4055			 * sectors per track we're using, smaller arrays
4056			 * may turn out to have 0 cylinders.  Linux and
4057			 * FreeBSD don't pay attention to these mode pages
4058			 * to figure out capacity, but Solaris does.  It
4059			 * seems to deal with 0 cylinders just fine, and
4060			 * works out a fake geometry based on the capacity.
4061			 */
4062			memcpy(&lun->mode_pages.rigid_disk_page[
4063			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4064			       sizeof(rigid_disk_page_default));
4065			memcpy(&lun->mode_pages.rigid_disk_page[
4066			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4067			       sizeof(rigid_disk_page_changeable));
4068
4069			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4070				CTL_DEFAULT_HEADS;
4071
4072			/*
4073			 * The divide method here will be more accurate,
4074			 * probably, but results in floating point being
4075			 * used in the kernel on i386 (__udivdi3()).  On the
4076			 * XScale, though, __udivdi3() is implemented in
4077			 * software.
4078			 *
4079			 * The shift method for cylinder calculation is
4080			 * accurate if sectors_per_cylinder is a power of
4081			 * 2.  Otherwise it might be slightly off -- you
4082			 * might have a bit of a truncation problem.
4083			 */
4084#ifdef	__XSCALE__
4085			cylinders = (lun->be_lun->maxlba + 1) /
4086				sectors_per_cylinder;
4087#else
4088			for (shift = 31; shift > 0; shift--) {
4089				if (sectors_per_cylinder & (1 << shift))
4090					break;
4091			}
4092			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4093#endif
4094
4095			/*
4096			 * We've basically got 3 bytes, or 24 bits for the
4097			 * cylinder size in the mode page.  If we're over,
4098			 * just round down to 2^24.
4099			 */
4100			if (cylinders > 0xffffff)
4101				cylinders = 0xffffff;
4102
4103			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4104				CTL_PAGE_DEFAULT];
4105			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4106
4107			if ((value = ctl_get_opt(&lun->be_lun->options,
4108			    "rpm")) != NULL) {
4109				scsi_ulto2b(strtol(value, NULL, 0),
4110				     rigid_disk_page->rotation_rate);
4111			}
4112
4113			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4114			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4115			       sizeof(rigid_disk_page_default));
4116			memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4117			       &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4118			       sizeof(rigid_disk_page_default));
4119
4120			page_index->page_data =
4121				(uint8_t *)lun->mode_pages.rigid_disk_page;
4122			break;
4123		}
4124		case SMS_CACHING_PAGE: {
4125			struct scsi_caching_page *caching_page;
4126
4127			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4128				panic("invalid subpage value %d",
4129				      page_index->subpage);
4130			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4131			       &caching_page_default,
4132			       sizeof(caching_page_default));
4133			memcpy(&lun->mode_pages.caching_page[
4134			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4135			       sizeof(caching_page_changeable));
4136			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4137			       &caching_page_default,
4138			       sizeof(caching_page_default));
4139			caching_page = &lun->mode_pages.caching_page[
4140			    CTL_PAGE_SAVED];
4141			value = ctl_get_opt(&lun->be_lun->options, "writecache");
4142			if (value != NULL && strcmp(value, "off") == 0)
4143				caching_page->flags1 &= ~SCP_WCE;
4144			value = ctl_get_opt(&lun->be_lun->options, "readcache");
4145			if (value != NULL && strcmp(value, "off") == 0)
4146				caching_page->flags1 |= SCP_RCD;
4147			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4148			       &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4149			       sizeof(caching_page_default));
4150			page_index->page_data =
4151				(uint8_t *)lun->mode_pages.caching_page;
4152			break;
4153		}
4154		case SMS_CONTROL_MODE_PAGE: {
4155			struct scsi_control_page *control_page;
4156
4157			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4158				panic("invalid subpage value %d",
4159				      page_index->subpage);
4160
4161			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4162			       &control_page_default,
4163			       sizeof(control_page_default));
4164			memcpy(&lun->mode_pages.control_page[
4165			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4166			       sizeof(control_page_changeable));
4167			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4168			       &control_page_default,
4169			       sizeof(control_page_default));
4170			control_page = &lun->mode_pages.control_page[
4171			    CTL_PAGE_SAVED];
4172			value = ctl_get_opt(&lun->be_lun->options, "reordering");
4173			if (value != NULL && strcmp(value, "unrestricted") == 0) {
4174				control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
4175				control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
4176			}
4177			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4178			       &lun->mode_pages.control_page[CTL_PAGE_SAVED],
4179			       sizeof(control_page_default));
4180			page_index->page_data =
4181				(uint8_t *)lun->mode_pages.control_page;
4182			break;
4183
4184		}
4185		case SMS_INFO_EXCEPTIONS_PAGE: {
4186			switch (page_index->subpage) {
4187			case SMS_SUBPAGE_PAGE_0:
4188				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4189				       &ie_page_default,
4190				       sizeof(ie_page_default));
4191				memcpy(&lun->mode_pages.ie_page[
4192				       CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4193				       sizeof(ie_page_changeable));
4194				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4195				       &ie_page_default,
4196				       sizeof(ie_page_default));
4197				memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4198				       &ie_page_default,
4199				       sizeof(ie_page_default));
4200				page_index->page_data =
4201					(uint8_t *)lun->mode_pages.ie_page;
4202				break;
4203			case 0x02: {
4204				struct ctl_logical_block_provisioning_page *page;
4205
4206				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4207				       &lbp_page_default,
4208				       sizeof(lbp_page_default));
4209				memcpy(&lun->mode_pages.lbp_page[
4210				       CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4211				       sizeof(lbp_page_changeable));
4212				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4213				       &lbp_page_default,
4214				       sizeof(lbp_page_default));
4215				page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4216				value = ctl_get_opt(&lun->be_lun->options,
4217				    "avail-threshold");
4218				if (value != NULL &&
4219				    ctl_expand_number(value, &ival) == 0) {
4220					page->descr[0].flags |= SLBPPD_ENABLED |
4221					    SLBPPD_ARMING_DEC;
4222					if (lun->be_lun->blocksize)
4223						ival /= lun->be_lun->blocksize;
4224					else
4225						ival /= 512;
4226					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4227					    page->descr[0].count);
4228				}
4229				value = ctl_get_opt(&lun->be_lun->options,
4230				    "used-threshold");
4231				if (value != NULL &&
4232				    ctl_expand_number(value, &ival) == 0) {
4233					page->descr[1].flags |= SLBPPD_ENABLED |
4234					    SLBPPD_ARMING_INC;
4235					if (lun->be_lun->blocksize)
4236						ival /= lun->be_lun->blocksize;
4237					else
4238						ival /= 512;
4239					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4240					    page->descr[1].count);
4241				}
4242				value = ctl_get_opt(&lun->be_lun->options,
4243				    "pool-avail-threshold");
4244				if (value != NULL &&
4245				    ctl_expand_number(value, &ival) == 0) {
4246					page->descr[2].flags |= SLBPPD_ENABLED |
4247					    SLBPPD_ARMING_DEC;
4248					if (lun->be_lun->blocksize)
4249						ival /= lun->be_lun->blocksize;
4250					else
4251						ival /= 512;
4252					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4253					    page->descr[2].count);
4254				}
4255				value = ctl_get_opt(&lun->be_lun->options,
4256				    "pool-used-threshold");
4257				if (value != NULL &&
4258				    ctl_expand_number(value, &ival) == 0) {
4259					page->descr[3].flags |= SLBPPD_ENABLED |
4260					    SLBPPD_ARMING_INC;
4261					if (lun->be_lun->blocksize)
4262						ival /= lun->be_lun->blocksize;
4263					else
4264						ival /= 512;
4265					scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4266					    page->descr[3].count);
4267				}
4268				memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4269				       &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4270				       sizeof(lbp_page_default));
4271				page_index->page_data =
4272					(uint8_t *)lun->mode_pages.lbp_page;
4273			}}
4274			break;
4275		}
4276		case SMS_VENDOR_SPECIFIC_PAGE:{
4277			switch (page_index->subpage) {
4278			case DBGCNF_SUBPAGE_CODE: {
4279				struct copan_debugconf_subpage *current_page,
4280							       *saved_page;
4281
4282				memcpy(&lun->mode_pages.debugconf_subpage[
4283				       CTL_PAGE_CURRENT],
4284				       &debugconf_page_default,
4285				       sizeof(debugconf_page_default));
4286				memcpy(&lun->mode_pages.debugconf_subpage[
4287				       CTL_PAGE_CHANGEABLE],
4288				       &debugconf_page_changeable,
4289				       sizeof(debugconf_page_changeable));
4290				memcpy(&lun->mode_pages.debugconf_subpage[
4291				       CTL_PAGE_DEFAULT],
4292				       &debugconf_page_default,
4293				       sizeof(debugconf_page_default));
4294				memcpy(&lun->mode_pages.debugconf_subpage[
4295				       CTL_PAGE_SAVED],
4296				       &debugconf_page_default,
4297				       sizeof(debugconf_page_default));
4298				page_index->page_data =
4299					(uint8_t *)lun->mode_pages.debugconf_subpage;
4300
4301				current_page = (struct copan_debugconf_subpage *)
4302					(page_index->page_data +
4303					 (page_index->page_len *
4304					  CTL_PAGE_CURRENT));
4305				saved_page = (struct copan_debugconf_subpage *)
4306					(page_index->page_data +
4307					 (page_index->page_len *
4308					  CTL_PAGE_SAVED));
4309				break;
4310			}
4311			default:
4312				panic("invalid subpage value %d",
4313				      page_index->subpage);
4314				break;
4315			}
4316   			break;
4317		}
4318		default:
4319			panic("invalid page value %d",
4320			      page_index->page_code & SMPH_PC_MASK);
4321			break;
4322    	}
4323	}
4324
4325	return (CTL_RETVAL_COMPLETE);
4326}
4327
4328static int
4329ctl_init_log_page_index(struct ctl_lun *lun)
4330{
4331	struct ctl_page_index *page_index;
4332	int i, j, k, prev;
4333
4334	memcpy(&lun->log_pages.index, log_page_index_template,
4335	       sizeof(log_page_index_template));
4336
4337	prev = -1;
4338	for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4339
4340		page_index = &lun->log_pages.index[i];
4341		/*
4342		 * If this is a disk-only mode page, there's no point in
4343		 * setting it up.  For some pages, we have to have some
4344		 * basic information about the disk in order to calculate the
4345		 * mode page data.
4346		 */
4347		if ((lun->be_lun->lun_type != T_DIRECT)
4348		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4349			continue;
4350
4351		if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4352		     lun->backend->lun_attr == NULL)
4353			continue;
4354
4355		if (page_index->page_code != prev) {
4356			lun->log_pages.pages_page[j] = page_index->page_code;
4357			prev = page_index->page_code;
4358			j++;
4359		}
4360		lun->log_pages.subpages_page[k*2] = page_index->page_code;
4361		lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4362		k++;
4363	}
4364	lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4365	lun->log_pages.index[0].page_len = j;
4366	lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4367	lun->log_pages.index[1].page_len = k * 2;
4368	lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4369	lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4370
4371	return (CTL_RETVAL_COMPLETE);
4372}
4373
4374static int
4375hex2bin(const char *str, uint8_t *buf, int buf_size)
4376{
4377	int i;
4378	u_char c;
4379
4380	memset(buf, 0, buf_size);
4381	while (isspace(str[0]))
4382		str++;
4383	if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4384		str += 2;
4385	buf_size *= 2;
4386	for (i = 0; str[i] != 0 && i < buf_size; i++) {
4387		c = str[i];
4388		if (isdigit(c))
4389			c -= '0';
4390		else if (isalpha(c))
4391			c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4392		else
4393			break;
4394		if (c >= 16)
4395			break;
4396		if ((i & 1) == 0)
4397			buf[i / 2] |= (c << 4);
4398		else
4399			buf[i / 2] |= c;
4400	}
4401	return ((i + 1) / 2);
4402}
4403
4404/*
4405 * LUN allocation.
4406 *
4407 * Requirements:
4408 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4409 *   wants us to allocate the LUN and he can block.
4410 * - ctl_softc is always set
4411 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4412 *
4413 * Returns 0 for success, non-zero (errno) for failure.
4414 */
4415static int
4416ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4417	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4418{
4419	struct ctl_lun *nlun, *lun;
4420	struct ctl_port *port;
4421	struct scsi_vpd_id_descriptor *desc;
4422	struct scsi_vpd_id_t10 *t10id;
4423	const char *eui, *naa, *scsiname, *vendor, *value;
4424	int lun_number, i, lun_malloced;
4425	int devidlen, idlen1, idlen2 = 0, len;
4426
4427	if (be_lun == NULL)
4428		return (EINVAL);
4429
4430	/*
4431	 * We currently only support Direct Access or Processor LUN types.
4432	 */
4433	switch (be_lun->lun_type) {
4434	case T_DIRECT:
4435		break;
4436	case T_PROCESSOR:
4437		break;
4438	case T_SEQUENTIAL:
4439	case T_CHANGER:
4440	default:
4441		be_lun->lun_config_status(be_lun->be_lun,
4442					  CTL_LUN_CONFIG_FAILURE);
4443		break;
4444	}
4445	if (ctl_lun == NULL) {
4446		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4447		lun_malloced = 1;
4448	} else {
4449		lun_malloced = 0;
4450		lun = ctl_lun;
4451	}
4452
4453	memset(lun, 0, sizeof(*lun));
4454	if (lun_malloced)
4455		lun->flags = CTL_LUN_MALLOCED;
4456
4457	/* Generate LUN ID. */
4458	devidlen = max(CTL_DEVID_MIN_LEN,
4459	    strnlen(be_lun->device_id, CTL_DEVID_LEN));
4460	idlen1 = sizeof(*t10id) + devidlen;
4461	len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4462	scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4463	if (scsiname != NULL) {
4464		idlen2 = roundup2(strlen(scsiname) + 1, 4);
4465		len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4466	}
4467	eui = ctl_get_opt(&be_lun->options, "eui");
4468	if (eui != NULL) {
4469		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4470	}
4471	naa = ctl_get_opt(&be_lun->options, "naa");
4472	if (naa != NULL) {
4473		len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4474	}
4475	lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4476	    M_CTL, M_WAITOK | M_ZERO);
4477	desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4478	desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4479	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4480	desc->length = idlen1;
4481	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4482	memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4483	if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4484		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4485	} else {
4486		strncpy(t10id->vendor, vendor,
4487		    min(sizeof(t10id->vendor), strlen(vendor)));
4488	}
4489	strncpy((char *)t10id->vendor_spec_id,
4490	    (char *)be_lun->device_id, devidlen);
4491	if (scsiname != NULL) {
4492		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4493		    desc->length);
4494		desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4495		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4496		    SVPD_ID_TYPE_SCSI_NAME;
4497		desc->length = idlen2;
4498		strlcpy(desc->identifier, scsiname, idlen2);
4499	}
4500	if (eui != NULL) {
4501		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4502		    desc->length);
4503		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4504		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4505		    SVPD_ID_TYPE_EUI64;
4506		desc->length = hex2bin(eui, desc->identifier, 16);
4507		desc->length = desc->length > 12 ? 16 :
4508		    (desc->length > 8 ? 12 : 8);
4509		len -= 16 - desc->length;
4510	}
4511	if (naa != NULL) {
4512		desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4513		    desc->length);
4514		desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4515		desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4516		    SVPD_ID_TYPE_NAA;
4517		desc->length = hex2bin(naa, desc->identifier, 16);
4518		desc->length = desc->length > 8 ? 16 : 8;
4519		len -= 16 - desc->length;
4520	}
4521	lun->lun_devid->len = len;
4522
4523	mtx_lock(&ctl_softc->ctl_lock);
4524	/*
4525	 * See if the caller requested a particular LUN number.  If so, see
4526	 * if it is available.  Otherwise, allocate the first available LUN.
4527	 */
4528	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4529		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4530		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4531			mtx_unlock(&ctl_softc->ctl_lock);
4532			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4533				printf("ctl: requested LUN ID %d is higher "
4534				       "than CTL_MAX_LUNS - 1 (%d)\n",
4535				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4536			} else {
4537				/*
4538				 * XXX KDM return an error, or just assign
4539				 * another LUN ID in this case??
4540				 */
4541				printf("ctl: requested LUN ID %d is already "
4542				       "in use\n", be_lun->req_lun_id);
4543			}
4544			if (lun->flags & CTL_LUN_MALLOCED)
4545				free(lun, M_CTL);
4546			be_lun->lun_config_status(be_lun->be_lun,
4547						  CTL_LUN_CONFIG_FAILURE);
4548			return (ENOSPC);
4549		}
4550		lun_number = be_lun->req_lun_id;
4551	} else {
4552		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4553		if (lun_number == -1) {
4554			mtx_unlock(&ctl_softc->ctl_lock);
4555			printf("ctl: can't allocate LUN on target %ju, out of "
4556			       "LUNs\n", (uintmax_t)target_id.id);
4557			if (lun->flags & CTL_LUN_MALLOCED)
4558				free(lun, M_CTL);
4559			be_lun->lun_config_status(be_lun->be_lun,
4560						  CTL_LUN_CONFIG_FAILURE);
4561			return (ENOSPC);
4562		}
4563	}
4564	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4565
4566	mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4567	lun->target = target_id;
4568	lun->lun = lun_number;
4569	lun->be_lun = be_lun;
4570	/*
4571	 * The processor LUN is always enabled.  Disk LUNs come on line
4572	 * disabled, and must be enabled by the backend.
4573	 */
4574	lun->flags |= CTL_LUN_DISABLED;
4575	lun->backend = be_lun->be;
4576	be_lun->ctl_lun = lun;
4577	be_lun->lun_id = lun_number;
4578	atomic_add_int(&be_lun->be->num_luns, 1);
4579	if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4580		lun->flags |= CTL_LUN_OFFLINE;
4581
4582	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4583		lun->flags |= CTL_LUN_STOPPED;
4584
4585	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4586		lun->flags |= CTL_LUN_INOPERABLE;
4587
4588	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4589		lun->flags |= CTL_LUN_PRIMARY_SC;
4590
4591	value = ctl_get_opt(&be_lun->options, "readonly");
4592	if (value != NULL && strcmp(value, "on") == 0)
4593		lun->flags |= CTL_LUN_READONLY;
4594
4595	lun->serseq = CTL_LUN_SERSEQ_OFF;
4596	if (be_lun->flags & CTL_LUN_FLAG_SERSEQ_READ)
4597		lun->serseq = CTL_LUN_SERSEQ_READ;
4598	value = ctl_get_opt(&be_lun->options, "serseq");
4599	if (value != NULL && strcmp(value, "on") == 0)
4600		lun->serseq = CTL_LUN_SERSEQ_ON;
4601	else if (value != NULL && strcmp(value, "read") == 0)
4602		lun->serseq = CTL_LUN_SERSEQ_READ;
4603	else if (value != NULL && strcmp(value, "off") == 0)
4604		lun->serseq = CTL_LUN_SERSEQ_OFF;
4605
4606	lun->ctl_softc = ctl_softc;
4607	TAILQ_INIT(&lun->ooa_queue);
4608	TAILQ_INIT(&lun->blocked_queue);
4609	STAILQ_INIT(&lun->error_list);
4610	ctl_tpc_lun_init(lun);
4611
4612	/*
4613	 * Initialize the mode and log page index.
4614	 */
4615	ctl_init_page_index(lun);
4616	ctl_init_log_page_index(lun);
4617
4618	/*
4619	 * Now, before we insert this lun on the lun list, set the lun
4620	 * inventory changed UA for all other luns.
4621	 */
4622	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4623		mtx_lock(&nlun->lun_lock);
4624		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4625		mtx_unlock(&nlun->lun_lock);
4626	}
4627
4628	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4629
4630	ctl_softc->ctl_luns[lun_number] = lun;
4631
4632	ctl_softc->num_luns++;
4633
4634	/* Setup statistics gathering */
4635	lun->stats.device_type = be_lun->lun_type;
4636	lun->stats.lun_number = lun_number;
4637	if (lun->stats.device_type == T_DIRECT)
4638		lun->stats.blocksize = be_lun->blocksize;
4639	else
4640		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4641	for (i = 0;i < CTL_MAX_PORTS;i++)
4642		lun->stats.ports[i].targ_port = i;
4643
4644	mtx_unlock(&ctl_softc->ctl_lock);
4645
4646	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4647
4648	/*
4649	 * Run through each registered FETD and bring it online if it isn't
4650	 * already.  Enable the target ID if it hasn't been enabled, and
4651	 * enable this particular LUN.
4652	 */
4653	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4654		int retval;
4655
4656		retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number);
4657		if (retval != 0) {
4658			printf("ctl_alloc_lun: FETD %s port %d returned error "
4659			       "%d for lun_enable on target %ju lun %d\n",
4660			       port->port_name, port->targ_port, retval,
4661			       (uintmax_t)target_id.id, lun_number);
4662		} else
4663			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4664	}
4665	return (0);
4666}
4667
4668/*
4669 * Delete a LUN.
4670 * Assumptions:
4671 * - LUN has already been marked invalid and any pending I/O has been taken
4672 *   care of.
4673 */
4674static int
4675ctl_free_lun(struct ctl_lun *lun)
4676{
4677	struct ctl_softc *softc;
4678#if 0
4679	struct ctl_port *port;
4680#endif
4681	struct ctl_lun *nlun;
4682	int i;
4683
4684	softc = lun->ctl_softc;
4685
4686	mtx_assert(&softc->ctl_lock, MA_OWNED);
4687
4688	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4689
4690	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4691
4692	softc->ctl_luns[lun->lun] = NULL;
4693
4694	if (!TAILQ_EMPTY(&lun->ooa_queue))
4695		panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4696
4697	softc->num_luns--;
4698
4699	/*
4700	 * XXX KDM this scheme only works for a single target/multiple LUN
4701	 * setup.  It needs to be revamped for a multiple target scheme.
4702	 *
4703	 * XXX KDM this results in port->lun_disable() getting called twice,
4704	 * once when ctl_disable_lun() is called, and a second time here.
4705	 * We really need to re-think the LUN disable semantics.  There
4706	 * should probably be several steps/levels to LUN removal:
4707	 *  - disable
4708	 *  - invalidate
4709	 *  - free
4710 	 *
4711	 * Right now we only have a disable method when communicating to
4712	 * the front end ports, at least for individual LUNs.
4713	 */
4714#if 0
4715	STAILQ_FOREACH(port, &softc->port_list, links) {
4716		int retval;
4717
4718		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4719					 lun->lun);
4720		if (retval != 0) {
4721			printf("ctl_free_lun: FETD %s port %d returned error "
4722			       "%d for lun_disable on target %ju lun %jd\n",
4723			       port->port_name, port->targ_port, retval,
4724			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4725		}
4726
4727		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4728			port->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4729
4730			retval = port->targ_disable(port->targ_lun_arg,lun->target);
4731			if (retval != 0) {
4732				printf("ctl_free_lun: FETD %s port %d "
4733				       "returned error %d for targ_disable on "
4734				       "target %ju\n", port->port_name,
4735				       port->targ_port, retval,
4736				       (uintmax_t)lun->target.id);
4737			} else
4738				port->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4739
4740			if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4741				continue;
4742
4743#if 0
4744			port->port_offline(port->onoff_arg);
4745			port->status &= ~CTL_PORT_STATUS_ONLINE;
4746#endif
4747		}
4748	}
4749#endif
4750
4751	/*
4752	 * Tell the backend to free resources, if this LUN has a backend.
4753	 */
4754	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4755	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4756
4757	ctl_tpc_lun_shutdown(lun);
4758	mtx_destroy(&lun->lun_lock);
4759	free(lun->lun_devid, M_CTL);
4760	for (i = 0; i < CTL_MAX_PORTS; i++)
4761		free(lun->pending_ua[i], M_CTL);
4762	for (i = 0; i < 2 * CTL_MAX_PORTS; i++)
4763		free(lun->pr_keys[i], M_CTL);
4764	free(lun->write_buffer, M_CTL);
4765	if (lun->flags & CTL_LUN_MALLOCED)
4766		free(lun, M_CTL);
4767
4768	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4769		mtx_lock(&nlun->lun_lock);
4770		ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4771		mtx_unlock(&nlun->lun_lock);
4772	}
4773
4774	return (0);
4775}
4776
4777static void
4778ctl_create_lun(struct ctl_be_lun *be_lun)
4779{
4780	struct ctl_softc *ctl_softc;
4781
4782	ctl_softc = control_softc;
4783
4784	/*
4785	 * ctl_alloc_lun() should handle all potential failure cases.
4786	 */
4787	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4788}
4789
4790int
4791ctl_add_lun(struct ctl_be_lun *be_lun)
4792{
4793	struct ctl_softc *ctl_softc = control_softc;
4794
4795	mtx_lock(&ctl_softc->ctl_lock);
4796	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4797	mtx_unlock(&ctl_softc->ctl_lock);
4798	wakeup(&ctl_softc->pending_lun_queue);
4799
4800	return (0);
4801}
4802
4803int
4804ctl_enable_lun(struct ctl_be_lun *be_lun)
4805{
4806	struct ctl_softc *ctl_softc;
4807	struct ctl_port *port, *nport;
4808	struct ctl_lun *lun;
4809	int retval;
4810
4811	ctl_softc = control_softc;
4812
4813	lun = (struct ctl_lun *)be_lun->ctl_lun;
4814
4815	mtx_lock(&ctl_softc->ctl_lock);
4816	mtx_lock(&lun->lun_lock);
4817	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4818		/*
4819		 * eh?  Why did we get called if the LUN is already
4820		 * enabled?
4821		 */
4822		mtx_unlock(&lun->lun_lock);
4823		mtx_unlock(&ctl_softc->ctl_lock);
4824		return (0);
4825	}
4826	lun->flags &= ~CTL_LUN_DISABLED;
4827	mtx_unlock(&lun->lun_lock);
4828
4829	for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) {
4830		nport = STAILQ_NEXT(port, links);
4831
4832		/*
4833		 * Drop the lock while we call the FETD's enable routine.
4834		 * This can lead to a callback into CTL (at least in the
4835		 * case of the internal initiator frontend.
4836		 */
4837		mtx_unlock(&ctl_softc->ctl_lock);
4838		retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun);
4839		mtx_lock(&ctl_softc->ctl_lock);
4840		if (retval != 0) {
4841			printf("%s: FETD %s port %d returned error "
4842			       "%d for lun_enable on target %ju lun %jd\n",
4843			       __func__, port->port_name, port->targ_port, retval,
4844			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4845		}
4846#if 0
4847		 else {
4848            /* NOTE:  TODO:  why does lun enable affect port status? */
4849			port->status |= CTL_PORT_STATUS_LUN_ONLINE;
4850		}
4851#endif
4852	}
4853
4854	mtx_unlock(&ctl_softc->ctl_lock);
4855
4856	return (0);
4857}
4858
4859int
4860ctl_disable_lun(struct ctl_be_lun *be_lun)
4861{
4862	struct ctl_softc *ctl_softc;
4863	struct ctl_port *port;
4864	struct ctl_lun *lun;
4865	int retval;
4866
4867	ctl_softc = control_softc;
4868
4869	lun = (struct ctl_lun *)be_lun->ctl_lun;
4870
4871	mtx_lock(&ctl_softc->ctl_lock);
4872	mtx_lock(&lun->lun_lock);
4873	if (lun->flags & CTL_LUN_DISABLED) {
4874		mtx_unlock(&lun->lun_lock);
4875		mtx_unlock(&ctl_softc->ctl_lock);
4876		return (0);
4877	}
4878	lun->flags |= CTL_LUN_DISABLED;
4879	mtx_unlock(&lun->lun_lock);
4880
4881	STAILQ_FOREACH(port, &ctl_softc->port_list, links) {
4882		mtx_unlock(&ctl_softc->ctl_lock);
4883		/*
4884		 * Drop the lock before we call the frontend's disable
4885		 * routine, to avoid lock order reversals.
4886		 *
4887		 * XXX KDM what happens if the frontend list changes while
4888		 * we're traversing it?  It's unlikely, but should be handled.
4889		 */
4890		retval = port->lun_disable(port->targ_lun_arg, lun->target,
4891					 lun->lun);
4892		mtx_lock(&ctl_softc->ctl_lock);
4893		if (retval != 0) {
4894			printf("ctl_alloc_lun: FETD %s port %d returned error "
4895			       "%d for lun_disable on target %ju lun %jd\n",
4896			       port->port_name, port->targ_port, retval,
4897			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4898		}
4899	}
4900
4901	mtx_unlock(&ctl_softc->ctl_lock);
4902
4903	return (0);
4904}
4905
4906int
4907ctl_start_lun(struct ctl_be_lun *be_lun)
4908{
4909	struct ctl_softc *ctl_softc;
4910	struct ctl_lun *lun;
4911
4912	ctl_softc = control_softc;
4913
4914	lun = (struct ctl_lun *)be_lun->ctl_lun;
4915
4916	mtx_lock(&lun->lun_lock);
4917	lun->flags &= ~CTL_LUN_STOPPED;
4918	mtx_unlock(&lun->lun_lock);
4919
4920	return (0);
4921}
4922
4923int
4924ctl_stop_lun(struct ctl_be_lun *be_lun)
4925{
4926	struct ctl_softc *ctl_softc;
4927	struct ctl_lun *lun;
4928
4929	ctl_softc = control_softc;
4930
4931	lun = (struct ctl_lun *)be_lun->ctl_lun;
4932
4933	mtx_lock(&lun->lun_lock);
4934	lun->flags |= CTL_LUN_STOPPED;
4935	mtx_unlock(&lun->lun_lock);
4936
4937	return (0);
4938}
4939
4940int
4941ctl_lun_offline(struct ctl_be_lun *be_lun)
4942{
4943	struct ctl_softc *ctl_softc;
4944	struct ctl_lun *lun;
4945
4946	ctl_softc = control_softc;
4947
4948	lun = (struct ctl_lun *)be_lun->ctl_lun;
4949
4950	mtx_lock(&lun->lun_lock);
4951	lun->flags |= CTL_LUN_OFFLINE;
4952	mtx_unlock(&lun->lun_lock);
4953
4954	return (0);
4955}
4956
4957int
4958ctl_lun_online(struct ctl_be_lun *be_lun)
4959{
4960	struct ctl_softc *ctl_softc;
4961	struct ctl_lun *lun;
4962
4963	ctl_softc = control_softc;
4964
4965	lun = (struct ctl_lun *)be_lun->ctl_lun;
4966
4967	mtx_lock(&lun->lun_lock);
4968	lun->flags &= ~CTL_LUN_OFFLINE;
4969	mtx_unlock(&lun->lun_lock);
4970
4971	return (0);
4972}
4973
4974int
4975ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4976{
4977	struct ctl_softc *ctl_softc;
4978	struct ctl_lun *lun;
4979
4980	ctl_softc = control_softc;
4981
4982	lun = (struct ctl_lun *)be_lun->ctl_lun;
4983
4984	mtx_lock(&lun->lun_lock);
4985
4986	/*
4987	 * The LUN needs to be disabled before it can be marked invalid.
4988	 */
4989	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4990		mtx_unlock(&lun->lun_lock);
4991		return (-1);
4992	}
4993	/*
4994	 * Mark the LUN invalid.
4995	 */
4996	lun->flags |= CTL_LUN_INVALID;
4997
4998	/*
4999	 * If there is nothing in the OOA queue, go ahead and free the LUN.
5000	 * If we have something in the OOA queue, we'll free it when the
5001	 * last I/O completes.
5002	 */
5003	if (TAILQ_EMPTY(&lun->ooa_queue)) {
5004		mtx_unlock(&lun->lun_lock);
5005		mtx_lock(&ctl_softc->ctl_lock);
5006		ctl_free_lun(lun);
5007		mtx_unlock(&ctl_softc->ctl_lock);
5008	} else
5009		mtx_unlock(&lun->lun_lock);
5010
5011	return (0);
5012}
5013
5014int
5015ctl_lun_inoperable(struct ctl_be_lun *be_lun)
5016{
5017	struct ctl_softc *ctl_softc;
5018	struct ctl_lun *lun;
5019
5020	ctl_softc = control_softc;
5021	lun = (struct ctl_lun *)be_lun->ctl_lun;
5022
5023	mtx_lock(&lun->lun_lock);
5024	lun->flags |= CTL_LUN_INOPERABLE;
5025	mtx_unlock(&lun->lun_lock);
5026
5027	return (0);
5028}
5029
5030int
5031ctl_lun_operable(struct ctl_be_lun *be_lun)
5032{
5033	struct ctl_softc *ctl_softc;
5034	struct ctl_lun *lun;
5035
5036	ctl_softc = control_softc;
5037	lun = (struct ctl_lun *)be_lun->ctl_lun;
5038
5039	mtx_lock(&lun->lun_lock);
5040	lun->flags &= ~CTL_LUN_INOPERABLE;
5041	mtx_unlock(&lun->lun_lock);
5042
5043	return (0);
5044}
5045
5046void
5047ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5048{
5049	struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5050
5051	mtx_lock(&lun->lun_lock);
5052	ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
5053	mtx_unlock(&lun->lun_lock);
5054}
5055
5056/*
5057 * Backend "memory move is complete" callback for requests that never
5058 * make it down to say RAIDCore's configuration code.
5059 */
5060int
5061ctl_config_move_done(union ctl_io *io)
5062{
5063	int retval;
5064
5065	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5066	KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5067	    ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5068
5069	if ((io->io_hdr.port_status != 0) &&
5070	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5071	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5072		/*
5073		 * For hardware error sense keys, the sense key
5074		 * specific value is defined to be a retry count,
5075		 * but we use it to pass back an internal FETD
5076		 * error code.  XXX KDM  Hopefully the FETD is only
5077		 * using 16 bits for an error code, since that's
5078		 * all the space we have in the sks field.
5079		 */
5080		ctl_set_internal_failure(&io->scsiio,
5081					 /*sks_valid*/ 1,
5082					 /*retry_count*/
5083					 io->io_hdr.port_status);
5084	}
5085
5086	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5087	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5088	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5089	    ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5090		/*
5091		 * XXX KDM just assuming a single pointer here, and not a
5092		 * S/G list.  If we start using S/G lists for config data,
5093		 * we'll need to know how to clean them up here as well.
5094		 */
5095		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5096			free(io->scsiio.kern_data_ptr, M_CTL);
5097		ctl_done(io);
5098		retval = CTL_RETVAL_COMPLETE;
5099	} else {
5100		/*
5101		 * XXX KDM now we need to continue data movement.  Some
5102		 * options:
5103		 * - call ctl_scsiio() again?  We don't do this for data
5104		 *   writes, because for those at least we know ahead of
5105		 *   time where the write will go and how long it is.  For
5106		 *   config writes, though, that information is largely
5107		 *   contained within the write itself, thus we need to
5108		 *   parse out the data again.
5109		 *
5110		 * - Call some other function once the data is in?
5111		 */
5112		if (ctl_debug & CTL_DEBUG_CDB_DATA)
5113			ctl_data_print(io);
5114
5115		/*
5116		 * XXX KDM call ctl_scsiio() again for now, and check flag
5117		 * bits to see whether we're allocated or not.
5118		 */
5119		retval = ctl_scsiio(&io->scsiio);
5120	}
5121	return (retval);
5122}
5123
5124/*
5125 * This gets called by a backend driver when it is done with a
5126 * data_submit method.
5127 */
5128void
5129ctl_data_submit_done(union ctl_io *io)
5130{
5131	/*
5132	 * If the IO_CONT flag is set, we need to call the supplied
5133	 * function to continue processing the I/O, instead of completing
5134	 * the I/O just yet.
5135	 *
5136	 * If there is an error, though, we don't want to keep processing.
5137	 * Instead, just send status back to the initiator.
5138	 */
5139	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5140	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5141	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5142	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5143		io->scsiio.io_cont(io);
5144		return;
5145	}
5146	ctl_done(io);
5147}
5148
5149/*
5150 * This gets called by a backend driver when it is done with a
5151 * configuration write.
5152 */
5153void
5154ctl_config_write_done(union ctl_io *io)
5155{
5156	uint8_t *buf;
5157
5158	/*
5159	 * If the IO_CONT flag is set, we need to call the supplied
5160	 * function to continue processing the I/O, instead of completing
5161	 * the I/O just yet.
5162	 *
5163	 * If there is an error, though, we don't want to keep processing.
5164	 * Instead, just send status back to the initiator.
5165	 */
5166	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5167	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5168	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5169	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5170		io->scsiio.io_cont(io);
5171		return;
5172	}
5173	/*
5174	 * Since a configuration write can be done for commands that actually
5175	 * have data allocated, like write buffer, and commands that have
5176	 * no data, like start/stop unit, we need to check here.
5177	 */
5178	if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5179		buf = io->scsiio.kern_data_ptr;
5180	else
5181		buf = NULL;
5182	ctl_done(io);
5183	if (buf)
5184		free(buf, M_CTL);
5185}
5186
5187void
5188ctl_config_read_done(union ctl_io *io)
5189{
5190	uint8_t *buf;
5191
5192	/*
5193	 * If there is some error -- we are done, skip data transfer.
5194	 */
5195	if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5196	    ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5197	     (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5198		if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5199			buf = io->scsiio.kern_data_ptr;
5200		else
5201			buf = NULL;
5202		ctl_done(io);
5203		if (buf)
5204			free(buf, M_CTL);
5205		return;
5206	}
5207
5208	/*
5209	 * If the IO_CONT flag is set, we need to call the supplied
5210	 * function to continue processing the I/O, instead of completing
5211	 * the I/O just yet.
5212	 */
5213	if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5214		io->scsiio.io_cont(io);
5215		return;
5216	}
5217
5218	ctl_datamove(io);
5219}
5220
5221/*
5222 * SCSI release command.
5223 */
5224int
5225ctl_scsi_release(struct ctl_scsiio *ctsio)
5226{
5227	int length, longid, thirdparty_id, resv_id;
5228	struct ctl_softc *ctl_softc;
5229	struct ctl_lun *lun;
5230	uint32_t residx;
5231
5232	length = 0;
5233	resv_id = 0;
5234
5235	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5236
5237	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5238	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5239	ctl_softc = control_softc;
5240
5241	switch (ctsio->cdb[0]) {
5242	case RELEASE_10: {
5243		struct scsi_release_10 *cdb;
5244
5245		cdb = (struct scsi_release_10 *)ctsio->cdb;
5246
5247		if (cdb->byte2 & SR10_LONGID)
5248			longid = 1;
5249		else
5250			thirdparty_id = cdb->thirdparty_id;
5251
5252		resv_id = cdb->resv_id;
5253		length = scsi_2btoul(cdb->length);
5254		break;
5255	}
5256	}
5257
5258
5259	/*
5260	 * XXX KDM right now, we only support LUN reservation.  We don't
5261	 * support 3rd party reservations, or extent reservations, which
5262	 * might actually need the parameter list.  If we've gotten this
5263	 * far, we've got a LUN reservation.  Anything else got kicked out
5264	 * above.  So, according to SPC, ignore the length.
5265	 */
5266	length = 0;
5267
5268	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5269	 && (length > 0)) {
5270		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5271		ctsio->kern_data_len = length;
5272		ctsio->kern_total_len = length;
5273		ctsio->kern_data_resid = 0;
5274		ctsio->kern_rel_offset = 0;
5275		ctsio->kern_sg_entries = 0;
5276		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5277		ctsio->be_move_done = ctl_config_move_done;
5278		ctl_datamove((union ctl_io *)ctsio);
5279
5280		return (CTL_RETVAL_COMPLETE);
5281	}
5282
5283	if (length > 0)
5284		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5285
5286	mtx_lock(&lun->lun_lock);
5287
5288	/*
5289	 * According to SPC, it is not an error for an intiator to attempt
5290	 * to release a reservation on a LUN that isn't reserved, or that
5291	 * is reserved by another initiator.  The reservation can only be
5292	 * released, though, by the initiator who made it or by one of
5293	 * several reset type events.
5294	 */
5295	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5296			lun->flags &= ~CTL_LUN_RESERVED;
5297
5298	mtx_unlock(&lun->lun_lock);
5299
5300	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5301		free(ctsio->kern_data_ptr, M_CTL);
5302		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5303	}
5304
5305	ctl_set_success(ctsio);
5306	ctl_done((union ctl_io *)ctsio);
5307	return (CTL_RETVAL_COMPLETE);
5308}
5309
5310int
5311ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5312{
5313	int extent, thirdparty, longid;
5314	int resv_id, length;
5315	uint64_t thirdparty_id;
5316	struct ctl_softc *ctl_softc;
5317	struct ctl_lun *lun;
5318	uint32_t residx;
5319
5320	extent = 0;
5321	thirdparty = 0;
5322	longid = 0;
5323	resv_id = 0;
5324	length = 0;
5325	thirdparty_id = 0;
5326
5327	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5328
5329	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5330	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5331	ctl_softc = control_softc;
5332
5333	switch (ctsio->cdb[0]) {
5334	case RESERVE_10: {
5335		struct scsi_reserve_10 *cdb;
5336
5337		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5338
5339		if (cdb->byte2 & SR10_LONGID)
5340			longid = 1;
5341		else
5342			thirdparty_id = cdb->thirdparty_id;
5343
5344		resv_id = cdb->resv_id;
5345		length = scsi_2btoul(cdb->length);
5346		break;
5347	}
5348	}
5349
5350	/*
5351	 * XXX KDM right now, we only support LUN reservation.  We don't
5352	 * support 3rd party reservations, or extent reservations, which
5353	 * might actually need the parameter list.  If we've gotten this
5354	 * far, we've got a LUN reservation.  Anything else got kicked out
5355	 * above.  So, according to SPC, ignore the length.
5356	 */
5357	length = 0;
5358
5359	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5360	 && (length > 0)) {
5361		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5362		ctsio->kern_data_len = length;
5363		ctsio->kern_total_len = length;
5364		ctsio->kern_data_resid = 0;
5365		ctsio->kern_rel_offset = 0;
5366		ctsio->kern_sg_entries = 0;
5367		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5368		ctsio->be_move_done = ctl_config_move_done;
5369		ctl_datamove((union ctl_io *)ctsio);
5370
5371		return (CTL_RETVAL_COMPLETE);
5372	}
5373
5374	if (length > 0)
5375		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5376
5377	mtx_lock(&lun->lun_lock);
5378	if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5379		ctl_set_reservation_conflict(ctsio);
5380		goto bailout;
5381	}
5382
5383	lun->flags |= CTL_LUN_RESERVED;
5384	lun->res_idx = residx;
5385
5386	ctl_set_success(ctsio);
5387
5388bailout:
5389	mtx_unlock(&lun->lun_lock);
5390
5391	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5392		free(ctsio->kern_data_ptr, M_CTL);
5393		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5394	}
5395
5396	ctl_done((union ctl_io *)ctsio);
5397	return (CTL_RETVAL_COMPLETE);
5398}
5399
5400int
5401ctl_start_stop(struct ctl_scsiio *ctsio)
5402{
5403	struct scsi_start_stop_unit *cdb;
5404	struct ctl_lun *lun;
5405	struct ctl_softc *ctl_softc;
5406	int retval;
5407
5408	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5409
5410	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5411	ctl_softc = control_softc;
5412	retval = 0;
5413
5414	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5415
5416	/*
5417	 * XXX KDM
5418	 * We don't support the immediate bit on a stop unit.  In order to
5419	 * do that, we would need to code up a way to know that a stop is
5420	 * pending, and hold off any new commands until it completes, one
5421	 * way or another.  Then we could accept or reject those commands
5422	 * depending on its status.  We would almost need to do the reverse
5423	 * of what we do below for an immediate start -- return the copy of
5424	 * the ctl_io to the FETD with status to send to the host (and to
5425	 * free the copy!) and then free the original I/O once the stop
5426	 * actually completes.  That way, the OOA queue mechanism can work
5427	 * to block commands that shouldn't proceed.  Another alternative
5428	 * would be to put the copy in the queue in place of the original,
5429	 * and return the original back to the caller.  That could be
5430	 * slightly safer..
5431	 */
5432	if ((cdb->byte2 & SSS_IMMED)
5433	 && ((cdb->how & SSS_START) == 0)) {
5434		ctl_set_invalid_field(ctsio,
5435				      /*sks_valid*/ 1,
5436				      /*command*/ 1,
5437				      /*field*/ 1,
5438				      /*bit_valid*/ 1,
5439				      /*bit*/ 0);
5440		ctl_done((union ctl_io *)ctsio);
5441		return (CTL_RETVAL_COMPLETE);
5442	}
5443
5444	if ((lun->flags & CTL_LUN_PR_RESERVED)
5445	 && ((cdb->how & SSS_START)==0)) {
5446		uint32_t residx;
5447
5448		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5449		if (ctl_get_prkey(lun, residx) == 0
5450		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5451
5452			ctl_set_reservation_conflict(ctsio);
5453			ctl_done((union ctl_io *)ctsio);
5454			return (CTL_RETVAL_COMPLETE);
5455		}
5456	}
5457
5458	/*
5459	 * If there is no backend on this device, we can't start or stop
5460	 * it.  In theory we shouldn't get any start/stop commands in the
5461	 * first place at this level if the LUN doesn't have a backend.
5462	 * That should get stopped by the command decode code.
5463	 */
5464	if (lun->backend == NULL) {
5465		ctl_set_invalid_opcode(ctsio);
5466		ctl_done((union ctl_io *)ctsio);
5467		return (CTL_RETVAL_COMPLETE);
5468	}
5469
5470	/*
5471	 * XXX KDM Copan-specific offline behavior.
5472	 * Figure out a reasonable way to port this?
5473	 */
5474#ifdef NEEDTOPORT
5475	mtx_lock(&lun->lun_lock);
5476
5477	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5478	 && (lun->flags & CTL_LUN_OFFLINE)) {
5479		/*
5480		 * If the LUN is offline, and the on/offline bit isn't set,
5481		 * reject the start or stop.  Otherwise, let it through.
5482		 */
5483		mtx_unlock(&lun->lun_lock);
5484		ctl_set_lun_not_ready(ctsio);
5485		ctl_done((union ctl_io *)ctsio);
5486	} else {
5487		mtx_unlock(&lun->lun_lock);
5488#endif /* NEEDTOPORT */
5489		/*
5490		 * This could be a start or a stop when we're online,
5491		 * or a stop/offline or start/online.  A start or stop when
5492		 * we're offline is covered in the case above.
5493		 */
5494		/*
5495		 * In the non-immediate case, we send the request to
5496		 * the backend and return status to the user when
5497		 * it is done.
5498		 *
5499		 * In the immediate case, we allocate a new ctl_io
5500		 * to hold a copy of the request, and send that to
5501		 * the backend.  We then set good status on the
5502		 * user's request and return it immediately.
5503		 */
5504		if (cdb->byte2 & SSS_IMMED) {
5505			union ctl_io *new_io;
5506
5507			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5508			ctl_copy_io((union ctl_io *)ctsio, new_io);
5509			retval = lun->backend->config_write(new_io);
5510			ctl_set_success(ctsio);
5511			ctl_done((union ctl_io *)ctsio);
5512		} else {
5513			retval = lun->backend->config_write(
5514				(union ctl_io *)ctsio);
5515		}
5516#ifdef NEEDTOPORT
5517	}
5518#endif
5519	return (retval);
5520}
5521
5522/*
5523 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5524 * we don't really do anything with the LBA and length fields if the user
5525 * passes them in.  Instead we'll just flush out the cache for the entire
5526 * LUN.
5527 */
5528int
5529ctl_sync_cache(struct ctl_scsiio *ctsio)
5530{
5531	struct ctl_lun *lun;
5532	struct ctl_softc *ctl_softc;
5533	uint64_t starting_lba;
5534	uint32_t block_count;
5535	int retval;
5536
5537	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5538
5539	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5540	ctl_softc = control_softc;
5541	retval = 0;
5542
5543	switch (ctsio->cdb[0]) {
5544	case SYNCHRONIZE_CACHE: {
5545		struct scsi_sync_cache *cdb;
5546		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5547
5548		starting_lba = scsi_4btoul(cdb->begin_lba);
5549		block_count = scsi_2btoul(cdb->lb_count);
5550		break;
5551	}
5552	case SYNCHRONIZE_CACHE_16: {
5553		struct scsi_sync_cache_16 *cdb;
5554		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5555
5556		starting_lba = scsi_8btou64(cdb->begin_lba);
5557		block_count = scsi_4btoul(cdb->lb_count);
5558		break;
5559	}
5560	default:
5561		ctl_set_invalid_opcode(ctsio);
5562		ctl_done((union ctl_io *)ctsio);
5563		goto bailout;
5564		break; /* NOTREACHED */
5565	}
5566
5567	/*
5568	 * We check the LBA and length, but don't do anything with them.
5569	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5570	 * get flushed.  This check will just help satisfy anyone who wants
5571	 * to see an error for an out of range LBA.
5572	 */
5573	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5574		ctl_set_lba_out_of_range(ctsio);
5575		ctl_done((union ctl_io *)ctsio);
5576		goto bailout;
5577	}
5578
5579	/*
5580	 * If this LUN has no backend, we can't flush the cache anyway.
5581	 */
5582	if (lun->backend == NULL) {
5583		ctl_set_invalid_opcode(ctsio);
5584		ctl_done((union ctl_io *)ctsio);
5585		goto bailout;
5586	}
5587
5588	/*
5589	 * Check to see whether we're configured to send the SYNCHRONIZE
5590	 * CACHE command directly to the back end.
5591	 */
5592	mtx_lock(&lun->lun_lock);
5593	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5594	 && (++(lun->sync_count) >= lun->sync_interval)) {
5595		lun->sync_count = 0;
5596		mtx_unlock(&lun->lun_lock);
5597		retval = lun->backend->config_write((union ctl_io *)ctsio);
5598	} else {
5599		mtx_unlock(&lun->lun_lock);
5600		ctl_set_success(ctsio);
5601		ctl_done((union ctl_io *)ctsio);
5602	}
5603
5604bailout:
5605
5606	return (retval);
5607}
5608
5609int
5610ctl_format(struct ctl_scsiio *ctsio)
5611{
5612	struct scsi_format *cdb;
5613	struct ctl_lun *lun;
5614	struct ctl_softc *ctl_softc;
5615	int length, defect_list_len;
5616
5617	CTL_DEBUG_PRINT(("ctl_format\n"));
5618
5619	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5620	ctl_softc = control_softc;
5621
5622	cdb = (struct scsi_format *)ctsio->cdb;
5623
5624	length = 0;
5625	if (cdb->byte2 & SF_FMTDATA) {
5626		if (cdb->byte2 & SF_LONGLIST)
5627			length = sizeof(struct scsi_format_header_long);
5628		else
5629			length = sizeof(struct scsi_format_header_short);
5630	}
5631
5632	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5633	 && (length > 0)) {
5634		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5635		ctsio->kern_data_len = length;
5636		ctsio->kern_total_len = length;
5637		ctsio->kern_data_resid = 0;
5638		ctsio->kern_rel_offset = 0;
5639		ctsio->kern_sg_entries = 0;
5640		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5641		ctsio->be_move_done = ctl_config_move_done;
5642		ctl_datamove((union ctl_io *)ctsio);
5643
5644		return (CTL_RETVAL_COMPLETE);
5645	}
5646
5647	defect_list_len = 0;
5648
5649	if (cdb->byte2 & SF_FMTDATA) {
5650		if (cdb->byte2 & SF_LONGLIST) {
5651			struct scsi_format_header_long *header;
5652
5653			header = (struct scsi_format_header_long *)
5654				ctsio->kern_data_ptr;
5655
5656			defect_list_len = scsi_4btoul(header->defect_list_len);
5657			if (defect_list_len != 0) {
5658				ctl_set_invalid_field(ctsio,
5659						      /*sks_valid*/ 1,
5660						      /*command*/ 0,
5661						      /*field*/ 2,
5662						      /*bit_valid*/ 0,
5663						      /*bit*/ 0);
5664				goto bailout;
5665			}
5666		} else {
5667			struct scsi_format_header_short *header;
5668
5669			header = (struct scsi_format_header_short *)
5670				ctsio->kern_data_ptr;
5671
5672			defect_list_len = scsi_2btoul(header->defect_list_len);
5673			if (defect_list_len != 0) {
5674				ctl_set_invalid_field(ctsio,
5675						      /*sks_valid*/ 1,
5676						      /*command*/ 0,
5677						      /*field*/ 2,
5678						      /*bit_valid*/ 0,
5679						      /*bit*/ 0);
5680				goto bailout;
5681			}
5682		}
5683	}
5684
5685	/*
5686	 * The format command will clear out the "Medium format corrupted"
5687	 * status if set by the configuration code.  That status is really
5688	 * just a way to notify the host that we have lost the media, and
5689	 * get them to issue a command that will basically make them think
5690	 * they're blowing away the media.
5691	 */
5692	mtx_lock(&lun->lun_lock);
5693	lun->flags &= ~CTL_LUN_INOPERABLE;
5694	mtx_unlock(&lun->lun_lock);
5695
5696	ctl_set_success(ctsio);
5697bailout:
5698
5699	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5700		free(ctsio->kern_data_ptr, M_CTL);
5701		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5702	}
5703
5704	ctl_done((union ctl_io *)ctsio);
5705	return (CTL_RETVAL_COMPLETE);
5706}
5707
5708int
5709ctl_read_buffer(struct ctl_scsiio *ctsio)
5710{
5711	struct scsi_read_buffer *cdb;
5712	struct ctl_lun *lun;
5713	int buffer_offset, len;
5714	static uint8_t descr[4];
5715	static uint8_t echo_descr[4] = { 0 };
5716
5717	CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5718
5719	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5720	cdb = (struct scsi_read_buffer *)ctsio->cdb;
5721
5722	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5723	    (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5724	    (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5725		ctl_set_invalid_field(ctsio,
5726				      /*sks_valid*/ 1,
5727				      /*command*/ 1,
5728				      /*field*/ 1,
5729				      /*bit_valid*/ 1,
5730				      /*bit*/ 4);
5731		ctl_done((union ctl_io *)ctsio);
5732		return (CTL_RETVAL_COMPLETE);
5733	}
5734
5735	len = scsi_3btoul(cdb->length);
5736	buffer_offset = scsi_3btoul(cdb->offset);
5737
5738	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5739		ctl_set_invalid_field(ctsio,
5740				      /*sks_valid*/ 1,
5741				      /*command*/ 1,
5742				      /*field*/ 6,
5743				      /*bit_valid*/ 0,
5744				      /*bit*/ 0);
5745		ctl_done((union ctl_io *)ctsio);
5746		return (CTL_RETVAL_COMPLETE);
5747	}
5748
5749	if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5750		descr[0] = 0;
5751		scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5752		ctsio->kern_data_ptr = descr;
5753		len = min(len, sizeof(descr));
5754	} else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5755		ctsio->kern_data_ptr = echo_descr;
5756		len = min(len, sizeof(echo_descr));
5757	} else {
5758		if (lun->write_buffer == NULL) {
5759			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5760			    M_CTL, M_WAITOK);
5761		}
5762		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5763	}
5764	ctsio->kern_data_len = len;
5765	ctsio->kern_total_len = len;
5766	ctsio->kern_data_resid = 0;
5767	ctsio->kern_rel_offset = 0;
5768	ctsio->kern_sg_entries = 0;
5769	ctl_set_success(ctsio);
5770	ctsio->be_move_done = ctl_config_move_done;
5771	ctl_datamove((union ctl_io *)ctsio);
5772	return (CTL_RETVAL_COMPLETE);
5773}
5774
5775int
5776ctl_write_buffer(struct ctl_scsiio *ctsio)
5777{
5778	struct scsi_write_buffer *cdb;
5779	struct ctl_lun *lun;
5780	int buffer_offset, len;
5781
5782	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5783
5784	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5785	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5786
5787	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5788		ctl_set_invalid_field(ctsio,
5789				      /*sks_valid*/ 1,
5790				      /*command*/ 1,
5791				      /*field*/ 1,
5792				      /*bit_valid*/ 1,
5793				      /*bit*/ 4);
5794		ctl_done((union ctl_io *)ctsio);
5795		return (CTL_RETVAL_COMPLETE);
5796	}
5797
5798	len = scsi_3btoul(cdb->length);
5799	buffer_offset = scsi_3btoul(cdb->offset);
5800
5801	if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5802		ctl_set_invalid_field(ctsio,
5803				      /*sks_valid*/ 1,
5804				      /*command*/ 1,
5805				      /*field*/ 6,
5806				      /*bit_valid*/ 0,
5807				      /*bit*/ 0);
5808		ctl_done((union ctl_io *)ctsio);
5809		return (CTL_RETVAL_COMPLETE);
5810	}
5811
5812	/*
5813	 * If we've got a kernel request that hasn't been malloced yet,
5814	 * malloc it and tell the caller the data buffer is here.
5815	 */
5816	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5817		if (lun->write_buffer == NULL) {
5818			lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5819			    M_CTL, M_WAITOK);
5820		}
5821		ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5822		ctsio->kern_data_len = len;
5823		ctsio->kern_total_len = len;
5824		ctsio->kern_data_resid = 0;
5825		ctsio->kern_rel_offset = 0;
5826		ctsio->kern_sg_entries = 0;
5827		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5828		ctsio->be_move_done = ctl_config_move_done;
5829		ctl_datamove((union ctl_io *)ctsio);
5830
5831		return (CTL_RETVAL_COMPLETE);
5832	}
5833
5834	ctl_set_success(ctsio);
5835	ctl_done((union ctl_io *)ctsio);
5836	return (CTL_RETVAL_COMPLETE);
5837}
5838
5839int
5840ctl_write_same(struct ctl_scsiio *ctsio)
5841{
5842	struct ctl_lun *lun;
5843	struct ctl_lba_len_flags *lbalen;
5844	uint64_t lba;
5845	uint32_t num_blocks;
5846	int len, retval;
5847	uint8_t byte2;
5848
5849	retval = CTL_RETVAL_COMPLETE;
5850
5851	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5852
5853	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5854
5855	switch (ctsio->cdb[0]) {
5856	case WRITE_SAME_10: {
5857		struct scsi_write_same_10 *cdb;
5858
5859		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5860
5861		lba = scsi_4btoul(cdb->addr);
5862		num_blocks = scsi_2btoul(cdb->length);
5863		byte2 = cdb->byte2;
5864		break;
5865	}
5866	case WRITE_SAME_16: {
5867		struct scsi_write_same_16 *cdb;
5868
5869		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5870
5871		lba = scsi_8btou64(cdb->addr);
5872		num_blocks = scsi_4btoul(cdb->length);
5873		byte2 = cdb->byte2;
5874		break;
5875	}
5876	default:
5877		/*
5878		 * We got a command we don't support.  This shouldn't
5879		 * happen, commands should be filtered out above us.
5880		 */
5881		ctl_set_invalid_opcode(ctsio);
5882		ctl_done((union ctl_io *)ctsio);
5883
5884		return (CTL_RETVAL_COMPLETE);
5885		break; /* NOTREACHED */
5886	}
5887
5888	/* NDOB and ANCHOR flags can be used only together with UNMAP */
5889	if ((byte2 & SWS_UNMAP) == 0 &&
5890	    (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5891		ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5892		    /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5893		ctl_done((union ctl_io *)ctsio);
5894		return (CTL_RETVAL_COMPLETE);
5895	}
5896
5897	/*
5898	 * The first check is to make sure we're in bounds, the second
5899	 * check is to catch wrap-around problems.  If the lba + num blocks
5900	 * is less than the lba, then we've wrapped around and the block
5901	 * range is invalid anyway.
5902	 */
5903	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5904	 || ((lba + num_blocks) < lba)) {
5905		ctl_set_lba_out_of_range(ctsio);
5906		ctl_done((union ctl_io *)ctsio);
5907		return (CTL_RETVAL_COMPLETE);
5908	}
5909
5910	/* Zero number of blocks means "to the last logical block" */
5911	if (num_blocks == 0) {
5912		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5913			ctl_set_invalid_field(ctsio,
5914					      /*sks_valid*/ 0,
5915					      /*command*/ 1,
5916					      /*field*/ 0,
5917					      /*bit_valid*/ 0,
5918					      /*bit*/ 0);
5919			ctl_done((union ctl_io *)ctsio);
5920			return (CTL_RETVAL_COMPLETE);
5921		}
5922		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5923	}
5924
5925	len = lun->be_lun->blocksize;
5926
5927	/*
5928	 * If we've got a kernel request that hasn't been malloced yet,
5929	 * malloc it and tell the caller the data buffer is here.
5930	 */
5931	if ((byte2 & SWS_NDOB) == 0 &&
5932	    (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5933		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5934		ctsio->kern_data_len = len;
5935		ctsio->kern_total_len = len;
5936		ctsio->kern_data_resid = 0;
5937		ctsio->kern_rel_offset = 0;
5938		ctsio->kern_sg_entries = 0;
5939		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5940		ctsio->be_move_done = ctl_config_move_done;
5941		ctl_datamove((union ctl_io *)ctsio);
5942
5943		return (CTL_RETVAL_COMPLETE);
5944	}
5945
5946	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5947	lbalen->lba = lba;
5948	lbalen->len = num_blocks;
5949	lbalen->flags = byte2;
5950	retval = lun->backend->config_write((union ctl_io *)ctsio);
5951
5952	return (retval);
5953}
5954
5955int
5956ctl_unmap(struct ctl_scsiio *ctsio)
5957{
5958	struct ctl_lun *lun;
5959	struct scsi_unmap *cdb;
5960	struct ctl_ptr_len_flags *ptrlen;
5961	struct scsi_unmap_header *hdr;
5962	struct scsi_unmap_desc *buf, *end, *endnz, *range;
5963	uint64_t lba;
5964	uint32_t num_blocks;
5965	int len, retval;
5966	uint8_t byte2;
5967
5968	retval = CTL_RETVAL_COMPLETE;
5969
5970	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5971
5972	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5973	cdb = (struct scsi_unmap *)ctsio->cdb;
5974
5975	len = scsi_2btoul(cdb->length);
5976	byte2 = cdb->byte2;
5977
5978	/*
5979	 * If we've got a kernel request that hasn't been malloced yet,
5980	 * malloc it and tell the caller the data buffer is here.
5981	 */
5982	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5983		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5984		ctsio->kern_data_len = len;
5985		ctsio->kern_total_len = len;
5986		ctsio->kern_data_resid = 0;
5987		ctsio->kern_rel_offset = 0;
5988		ctsio->kern_sg_entries = 0;
5989		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5990		ctsio->be_move_done = ctl_config_move_done;
5991		ctl_datamove((union ctl_io *)ctsio);
5992
5993		return (CTL_RETVAL_COMPLETE);
5994	}
5995
5996	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5997	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5998	if (len < sizeof (*hdr) ||
5999	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
6000	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
6001	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
6002		ctl_set_invalid_field(ctsio,
6003				      /*sks_valid*/ 0,
6004				      /*command*/ 0,
6005				      /*field*/ 0,
6006				      /*bit_valid*/ 0,
6007				      /*bit*/ 0);
6008		goto done;
6009	}
6010	len = scsi_2btoul(hdr->desc_length);
6011	buf = (struct scsi_unmap_desc *)(hdr + 1);
6012	end = buf + len / sizeof(*buf);
6013
6014	endnz = buf;
6015	for (range = buf; range < end; range++) {
6016		lba = scsi_8btou64(range->lba);
6017		num_blocks = scsi_4btoul(range->length);
6018		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
6019		 || ((lba + num_blocks) < lba)) {
6020			ctl_set_lba_out_of_range(ctsio);
6021			ctl_done((union ctl_io *)ctsio);
6022			return (CTL_RETVAL_COMPLETE);
6023		}
6024		if (num_blocks != 0)
6025			endnz = range + 1;
6026	}
6027
6028	/*
6029	 * Block backend can not handle zero last range.
6030	 * Filter it out and return if there is nothing left.
6031	 */
6032	len = (uint8_t *)endnz - (uint8_t *)buf;
6033	if (len == 0) {
6034		ctl_set_success(ctsio);
6035		goto done;
6036	}
6037
6038	mtx_lock(&lun->lun_lock);
6039	ptrlen = (struct ctl_ptr_len_flags *)
6040	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6041	ptrlen->ptr = (void *)buf;
6042	ptrlen->len = len;
6043	ptrlen->flags = byte2;
6044	ctl_check_blocked(lun);
6045	mtx_unlock(&lun->lun_lock);
6046
6047	retval = lun->backend->config_write((union ctl_io *)ctsio);
6048	return (retval);
6049
6050done:
6051	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
6052		free(ctsio->kern_data_ptr, M_CTL);
6053		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
6054	}
6055	ctl_done((union ctl_io *)ctsio);
6056	return (CTL_RETVAL_COMPLETE);
6057}
6058
6059/*
6060 * Note that this function currently doesn't actually do anything inside
6061 * CTL to enforce things if the DQue bit is turned on.
6062 *
6063 * Also note that this function can't be used in the default case, because
6064 * the DQue bit isn't set in the changeable mask for the control mode page
6065 * anyway.  This is just here as an example for how to implement a page
6066 * handler, and a placeholder in case we want to allow the user to turn
6067 * tagged queueing on and off.
6068 *
6069 * The D_SENSE bit handling is functional, however, and will turn
6070 * descriptor sense on and off for a given LUN.
6071 */
6072int
6073ctl_control_page_handler(struct ctl_scsiio *ctsio,
6074			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6075{
6076	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6077	struct ctl_lun *lun;
6078	struct ctl_softc *softc;
6079	int set_ua;
6080	uint32_t initidx;
6081
6082	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6083	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6084	set_ua = 0;
6085
6086	user_cp = (struct scsi_control_page *)page_ptr;
6087	current_cp = (struct scsi_control_page *)
6088		(page_index->page_data + (page_index->page_len *
6089		CTL_PAGE_CURRENT));
6090	saved_cp = (struct scsi_control_page *)
6091		(page_index->page_data + (page_index->page_len *
6092		CTL_PAGE_SAVED));
6093
6094	softc = control_softc;
6095
6096	mtx_lock(&lun->lun_lock);
6097	if (((current_cp->rlec & SCP_DSENSE) == 0)
6098	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6099		/*
6100		 * Descriptor sense is currently turned off and the user
6101		 * wants to turn it on.
6102		 */
6103		current_cp->rlec |= SCP_DSENSE;
6104		saved_cp->rlec |= SCP_DSENSE;
6105		lun->flags |= CTL_LUN_SENSE_DESC;
6106		set_ua = 1;
6107	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6108		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6109		/*
6110		 * Descriptor sense is currently turned on, and the user
6111		 * wants to turn it off.
6112		 */
6113		current_cp->rlec &= ~SCP_DSENSE;
6114		saved_cp->rlec &= ~SCP_DSENSE;
6115		lun->flags &= ~CTL_LUN_SENSE_DESC;
6116		set_ua = 1;
6117	}
6118	if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
6119	    (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
6120		current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6121		current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6122		saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
6123		saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
6124		set_ua = 1;
6125	}
6126	if ((current_cp->eca_and_aen & SCP_SWP) !=
6127	    (user_cp->eca_and_aen & SCP_SWP)) {
6128		current_cp->eca_and_aen &= ~SCP_SWP;
6129		current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6130		saved_cp->eca_and_aen &= ~SCP_SWP;
6131		saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
6132		set_ua = 1;
6133	}
6134	if (set_ua != 0)
6135		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6136	mtx_unlock(&lun->lun_lock);
6137
6138	return (0);
6139}
6140
6141int
6142ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
6143		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6144{
6145	struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
6146	struct ctl_lun *lun;
6147	int set_ua;
6148	uint32_t initidx;
6149
6150	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6151	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6152	set_ua = 0;
6153
6154	user_cp = (struct scsi_caching_page *)page_ptr;
6155	current_cp = (struct scsi_caching_page *)
6156		(page_index->page_data + (page_index->page_len *
6157		CTL_PAGE_CURRENT));
6158	saved_cp = (struct scsi_caching_page *)
6159		(page_index->page_data + (page_index->page_len *
6160		CTL_PAGE_SAVED));
6161
6162	mtx_lock(&lun->lun_lock);
6163	if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
6164	    (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
6165		current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6166		current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6167		saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
6168		saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
6169		set_ua = 1;
6170	}
6171	if (set_ua != 0)
6172		ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6173	mtx_unlock(&lun->lun_lock);
6174
6175	return (0);
6176}
6177
6178int
6179ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6180				struct ctl_page_index *page_index,
6181				uint8_t *page_ptr)
6182{
6183	uint8_t *c;
6184	int i;
6185
6186	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6187	ctl_time_io_secs =
6188		(c[0] << 8) |
6189		(c[1] << 0) |
6190		0;
6191	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6192	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6193	printf("page data:");
6194	for (i=0; i<8; i++)
6195		printf(" %.2x",page_ptr[i]);
6196	printf("\n");
6197	return (0);
6198}
6199
6200int
6201ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6202			       struct ctl_page_index *page_index,
6203			       int pc)
6204{
6205	struct copan_debugconf_subpage *page;
6206
6207	page = (struct copan_debugconf_subpage *)page_index->page_data +
6208		(page_index->page_len * pc);
6209
6210	switch (pc) {
6211	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6212	case SMS_PAGE_CTRL_DEFAULT >> 6:
6213	case SMS_PAGE_CTRL_SAVED >> 6:
6214		/*
6215		 * We don't update the changable or default bits for this page.
6216		 */
6217		break;
6218	case SMS_PAGE_CTRL_CURRENT >> 6:
6219		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6220		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6221		break;
6222	default:
6223#ifdef NEEDTOPORT
6224		EPRINT(0, "Invalid PC %d!!", pc);
6225#endif /* NEEDTOPORT */
6226		break;
6227	}
6228	return (0);
6229}
6230
6231
6232static int
6233ctl_do_mode_select(union ctl_io *io)
6234{
6235	struct scsi_mode_page_header *page_header;
6236	struct ctl_page_index *page_index;
6237	struct ctl_scsiio *ctsio;
6238	int control_dev, page_len;
6239	int page_len_offset, page_len_size;
6240	union ctl_modepage_info *modepage_info;
6241	struct ctl_lun *lun;
6242	int *len_left, *len_used;
6243	int retval, i;
6244
6245	ctsio = &io->scsiio;
6246	page_index = NULL;
6247	page_len = 0;
6248	retval = CTL_RETVAL_COMPLETE;
6249
6250	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6251
6252	if (lun->be_lun->lun_type != T_DIRECT)
6253		control_dev = 1;
6254	else
6255		control_dev = 0;
6256
6257	modepage_info = (union ctl_modepage_info *)
6258		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6259	len_left = &modepage_info->header.len_left;
6260	len_used = &modepage_info->header.len_used;
6261
6262do_next_page:
6263
6264	page_header = (struct scsi_mode_page_header *)
6265		(ctsio->kern_data_ptr + *len_used);
6266
6267	if (*len_left == 0) {
6268		free(ctsio->kern_data_ptr, M_CTL);
6269		ctl_set_success(ctsio);
6270		ctl_done((union ctl_io *)ctsio);
6271		return (CTL_RETVAL_COMPLETE);
6272	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6273
6274		free(ctsio->kern_data_ptr, M_CTL);
6275		ctl_set_param_len_error(ctsio);
6276		ctl_done((union ctl_io *)ctsio);
6277		return (CTL_RETVAL_COMPLETE);
6278
6279	} else if ((page_header->page_code & SMPH_SPF)
6280		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6281
6282		free(ctsio->kern_data_ptr, M_CTL);
6283		ctl_set_param_len_error(ctsio);
6284		ctl_done((union ctl_io *)ctsio);
6285		return (CTL_RETVAL_COMPLETE);
6286	}
6287
6288
6289	/*
6290	 * XXX KDM should we do something with the block descriptor?
6291	 */
6292	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6293
6294		if ((control_dev != 0)
6295		 && (lun->mode_pages.index[i].page_flags &
6296		     CTL_PAGE_FLAG_DISK_ONLY))
6297			continue;
6298
6299		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6300		    (page_header->page_code & SMPH_PC_MASK))
6301			continue;
6302
6303		/*
6304		 * If neither page has a subpage code, then we've got a
6305		 * match.
6306		 */
6307		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6308		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6309			page_index = &lun->mode_pages.index[i];
6310			page_len = page_header->page_length;
6311			break;
6312		}
6313
6314		/*
6315		 * If both pages have subpages, then the subpage numbers
6316		 * have to match.
6317		 */
6318		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6319		  && (page_header->page_code & SMPH_SPF)) {
6320			struct scsi_mode_page_header_sp *sph;
6321
6322			sph = (struct scsi_mode_page_header_sp *)page_header;
6323
6324			if (lun->mode_pages.index[i].subpage ==
6325			    sph->subpage) {
6326				page_index = &lun->mode_pages.index[i];
6327				page_len = scsi_2btoul(sph->page_length);
6328				break;
6329			}
6330		}
6331	}
6332
6333	/*
6334	 * If we couldn't find the page, or if we don't have a mode select
6335	 * handler for it, send back an error to the user.
6336	 */
6337	if ((page_index == NULL)
6338	 || (page_index->select_handler == NULL)) {
6339		ctl_set_invalid_field(ctsio,
6340				      /*sks_valid*/ 1,
6341				      /*command*/ 0,
6342				      /*field*/ *len_used,
6343				      /*bit_valid*/ 0,
6344				      /*bit*/ 0);
6345		free(ctsio->kern_data_ptr, M_CTL);
6346		ctl_done((union ctl_io *)ctsio);
6347		return (CTL_RETVAL_COMPLETE);
6348	}
6349
6350	if (page_index->page_code & SMPH_SPF) {
6351		page_len_offset = 2;
6352		page_len_size = 2;
6353	} else {
6354		page_len_size = 1;
6355		page_len_offset = 1;
6356	}
6357
6358	/*
6359	 * If the length the initiator gives us isn't the one we specify in
6360	 * the mode page header, or if they didn't specify enough data in
6361	 * the CDB to avoid truncating this page, kick out the request.
6362	 */
6363	if ((page_len != (page_index->page_len - page_len_offset -
6364			  page_len_size))
6365	 || (*len_left < page_index->page_len)) {
6366
6367
6368		ctl_set_invalid_field(ctsio,
6369				      /*sks_valid*/ 1,
6370				      /*command*/ 0,
6371				      /*field*/ *len_used + page_len_offset,
6372				      /*bit_valid*/ 0,
6373				      /*bit*/ 0);
6374		free(ctsio->kern_data_ptr, M_CTL);
6375		ctl_done((union ctl_io *)ctsio);
6376		return (CTL_RETVAL_COMPLETE);
6377	}
6378
6379	/*
6380	 * Run through the mode page, checking to make sure that the bits
6381	 * the user changed are actually legal for him to change.
6382	 */
6383	for (i = 0; i < page_index->page_len; i++) {
6384		uint8_t *user_byte, *change_mask, *current_byte;
6385		int bad_bit;
6386		int j;
6387
6388		user_byte = (uint8_t *)page_header + i;
6389		change_mask = page_index->page_data +
6390			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6391		current_byte = page_index->page_data +
6392			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6393
6394		/*
6395		 * Check to see whether the user set any bits in this byte
6396		 * that he is not allowed to set.
6397		 */
6398		if ((*user_byte & ~(*change_mask)) ==
6399		    (*current_byte & ~(*change_mask)))
6400			continue;
6401
6402		/*
6403		 * Go through bit by bit to determine which one is illegal.
6404		 */
6405		bad_bit = 0;
6406		for (j = 7; j >= 0; j--) {
6407			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6408			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6409				bad_bit = i;
6410				break;
6411			}
6412		}
6413		ctl_set_invalid_field(ctsio,
6414				      /*sks_valid*/ 1,
6415				      /*command*/ 0,
6416				      /*field*/ *len_used + i,
6417				      /*bit_valid*/ 1,
6418				      /*bit*/ bad_bit);
6419		free(ctsio->kern_data_ptr, M_CTL);
6420		ctl_done((union ctl_io *)ctsio);
6421		return (CTL_RETVAL_COMPLETE);
6422	}
6423
6424	/*
6425	 * Decrement these before we call the page handler, since we may
6426	 * end up getting called back one way or another before the handler
6427	 * returns to this context.
6428	 */
6429	*len_left -= page_index->page_len;
6430	*len_used += page_index->page_len;
6431
6432	retval = page_index->select_handler(ctsio, page_index,
6433					    (uint8_t *)page_header);
6434
6435	/*
6436	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6437	 * wait until this queued command completes to finish processing
6438	 * the mode page.  If it returns anything other than
6439	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6440	 * already set the sense information, freed the data pointer, and
6441	 * completed the io for us.
6442	 */
6443	if (retval != CTL_RETVAL_COMPLETE)
6444		goto bailout_no_done;
6445
6446	/*
6447	 * If the initiator sent us more than one page, parse the next one.
6448	 */
6449	if (*len_left > 0)
6450		goto do_next_page;
6451
6452	ctl_set_success(ctsio);
6453	free(ctsio->kern_data_ptr, M_CTL);
6454	ctl_done((union ctl_io *)ctsio);
6455
6456bailout_no_done:
6457
6458	return (CTL_RETVAL_COMPLETE);
6459
6460}
6461
6462int
6463ctl_mode_select(struct ctl_scsiio *ctsio)
6464{
6465	int param_len, pf, sp;
6466	int header_size, bd_len;
6467	int len_left, len_used;
6468	struct ctl_page_index *page_index;
6469	struct ctl_lun *lun;
6470	int control_dev, page_len;
6471	union ctl_modepage_info *modepage_info;
6472	int retval;
6473
6474	pf = 0;
6475	sp = 0;
6476	page_len = 0;
6477	len_used = 0;
6478	len_left = 0;
6479	retval = 0;
6480	bd_len = 0;
6481	page_index = NULL;
6482
6483	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6484
6485	if (lun->be_lun->lun_type != T_DIRECT)
6486		control_dev = 1;
6487	else
6488		control_dev = 0;
6489
6490	switch (ctsio->cdb[0]) {
6491	case MODE_SELECT_6: {
6492		struct scsi_mode_select_6 *cdb;
6493
6494		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6495
6496		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6497		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6498
6499		param_len = cdb->length;
6500		header_size = sizeof(struct scsi_mode_header_6);
6501		break;
6502	}
6503	case MODE_SELECT_10: {
6504		struct scsi_mode_select_10 *cdb;
6505
6506		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6507
6508		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6509		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6510
6511		param_len = scsi_2btoul(cdb->length);
6512		header_size = sizeof(struct scsi_mode_header_10);
6513		break;
6514	}
6515	default:
6516		ctl_set_invalid_opcode(ctsio);
6517		ctl_done((union ctl_io *)ctsio);
6518		return (CTL_RETVAL_COMPLETE);
6519		break; /* NOTREACHED */
6520	}
6521
6522	/*
6523	 * From SPC-3:
6524	 * "A parameter list length of zero indicates that the Data-Out Buffer
6525	 * shall be empty. This condition shall not be considered as an error."
6526	 */
6527	if (param_len == 0) {
6528		ctl_set_success(ctsio);
6529		ctl_done((union ctl_io *)ctsio);
6530		return (CTL_RETVAL_COMPLETE);
6531	}
6532
6533	/*
6534	 * Since we'll hit this the first time through, prior to
6535	 * allocation, we don't need to free a data buffer here.
6536	 */
6537	if (param_len < header_size) {
6538		ctl_set_param_len_error(ctsio);
6539		ctl_done((union ctl_io *)ctsio);
6540		return (CTL_RETVAL_COMPLETE);
6541	}
6542
6543	/*
6544	 * Allocate the data buffer and grab the user's data.  In theory,
6545	 * we shouldn't have to sanity check the parameter list length here
6546	 * because the maximum size is 64K.  We should be able to malloc
6547	 * that much without too many problems.
6548	 */
6549	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6550		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6551		ctsio->kern_data_len = param_len;
6552		ctsio->kern_total_len = param_len;
6553		ctsio->kern_data_resid = 0;
6554		ctsio->kern_rel_offset = 0;
6555		ctsio->kern_sg_entries = 0;
6556		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6557		ctsio->be_move_done = ctl_config_move_done;
6558		ctl_datamove((union ctl_io *)ctsio);
6559
6560		return (CTL_RETVAL_COMPLETE);
6561	}
6562
6563	switch (ctsio->cdb[0]) {
6564	case MODE_SELECT_6: {
6565		struct scsi_mode_header_6 *mh6;
6566
6567		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6568		bd_len = mh6->blk_desc_len;
6569		break;
6570	}
6571	case MODE_SELECT_10: {
6572		struct scsi_mode_header_10 *mh10;
6573
6574		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6575		bd_len = scsi_2btoul(mh10->blk_desc_len);
6576		break;
6577	}
6578	default:
6579		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6580		break;
6581	}
6582
6583	if (param_len < (header_size + bd_len)) {
6584		free(ctsio->kern_data_ptr, M_CTL);
6585		ctl_set_param_len_error(ctsio);
6586		ctl_done((union ctl_io *)ctsio);
6587		return (CTL_RETVAL_COMPLETE);
6588	}
6589
6590	/*
6591	 * Set the IO_CONT flag, so that if this I/O gets passed to
6592	 * ctl_config_write_done(), it'll get passed back to
6593	 * ctl_do_mode_select() for further processing, or completion if
6594	 * we're all done.
6595	 */
6596	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6597	ctsio->io_cont = ctl_do_mode_select;
6598
6599	modepage_info = (union ctl_modepage_info *)
6600		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6601
6602	memset(modepage_info, 0, sizeof(*modepage_info));
6603
6604	len_left = param_len - header_size - bd_len;
6605	len_used = header_size + bd_len;
6606
6607	modepage_info->header.len_left = len_left;
6608	modepage_info->header.len_used = len_used;
6609
6610	return (ctl_do_mode_select((union ctl_io *)ctsio));
6611}
6612
6613int
6614ctl_mode_sense(struct ctl_scsiio *ctsio)
6615{
6616	struct ctl_lun *lun;
6617	int pc, page_code, dbd, llba, subpage;
6618	int alloc_len, page_len, header_len, total_len;
6619	struct scsi_mode_block_descr *block_desc;
6620	struct ctl_page_index *page_index;
6621	int control_dev;
6622
6623	dbd = 0;
6624	llba = 0;
6625	block_desc = NULL;
6626	page_index = NULL;
6627
6628	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6629
6630	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6631
6632	if (lun->be_lun->lun_type != T_DIRECT)
6633		control_dev = 1;
6634	else
6635		control_dev = 0;
6636
6637	switch (ctsio->cdb[0]) {
6638	case MODE_SENSE_6: {
6639		struct scsi_mode_sense_6 *cdb;
6640
6641		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6642
6643		header_len = sizeof(struct scsi_mode_hdr_6);
6644		if (cdb->byte2 & SMS_DBD)
6645			dbd = 1;
6646		else
6647			header_len += sizeof(struct scsi_mode_block_descr);
6648
6649		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6650		page_code = cdb->page & SMS_PAGE_CODE;
6651		subpage = cdb->subpage;
6652		alloc_len = cdb->length;
6653		break;
6654	}
6655	case MODE_SENSE_10: {
6656		struct scsi_mode_sense_10 *cdb;
6657
6658		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6659
6660		header_len = sizeof(struct scsi_mode_hdr_10);
6661
6662		if (cdb->byte2 & SMS_DBD)
6663			dbd = 1;
6664		else
6665			header_len += sizeof(struct scsi_mode_block_descr);
6666		if (cdb->byte2 & SMS10_LLBAA)
6667			llba = 1;
6668		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6669		page_code = cdb->page & SMS_PAGE_CODE;
6670		subpage = cdb->subpage;
6671		alloc_len = scsi_2btoul(cdb->length);
6672		break;
6673	}
6674	default:
6675		ctl_set_invalid_opcode(ctsio);
6676		ctl_done((union ctl_io *)ctsio);
6677		return (CTL_RETVAL_COMPLETE);
6678		break; /* NOTREACHED */
6679	}
6680
6681	/*
6682	 * We have to make a first pass through to calculate the size of
6683	 * the pages that match the user's query.  Then we allocate enough
6684	 * memory to hold it, and actually copy the data into the buffer.
6685	 */
6686	switch (page_code) {
6687	case SMS_ALL_PAGES_PAGE: {
6688		int i;
6689
6690		page_len = 0;
6691
6692		/*
6693		 * At the moment, values other than 0 and 0xff here are
6694		 * reserved according to SPC-3.
6695		 */
6696		if ((subpage != SMS_SUBPAGE_PAGE_0)
6697		 && (subpage != SMS_SUBPAGE_ALL)) {
6698			ctl_set_invalid_field(ctsio,
6699					      /*sks_valid*/ 1,
6700					      /*command*/ 1,
6701					      /*field*/ 3,
6702					      /*bit_valid*/ 0,
6703					      /*bit*/ 0);
6704			ctl_done((union ctl_io *)ctsio);
6705			return (CTL_RETVAL_COMPLETE);
6706		}
6707
6708		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6709			if ((control_dev != 0)
6710			 && (lun->mode_pages.index[i].page_flags &
6711			     CTL_PAGE_FLAG_DISK_ONLY))
6712				continue;
6713
6714			/*
6715			 * We don't use this subpage if the user didn't
6716			 * request all subpages.
6717			 */
6718			if ((lun->mode_pages.index[i].subpage != 0)
6719			 && (subpage == SMS_SUBPAGE_PAGE_0))
6720				continue;
6721
6722#if 0
6723			printf("found page %#x len %d\n",
6724			       lun->mode_pages.index[i].page_code &
6725			       SMPH_PC_MASK,
6726			       lun->mode_pages.index[i].page_len);
6727#endif
6728			page_len += lun->mode_pages.index[i].page_len;
6729		}
6730		break;
6731	}
6732	default: {
6733		int i;
6734
6735		page_len = 0;
6736
6737		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6738			/* Look for the right page code */
6739			if ((lun->mode_pages.index[i].page_code &
6740			     SMPH_PC_MASK) != page_code)
6741				continue;
6742
6743			/* Look for the right subpage or the subpage wildcard*/
6744			if ((lun->mode_pages.index[i].subpage != subpage)
6745			 && (subpage != SMS_SUBPAGE_ALL))
6746				continue;
6747
6748			/* Make sure the page is supported for this dev type */
6749			if ((control_dev != 0)
6750			 && (lun->mode_pages.index[i].page_flags &
6751			     CTL_PAGE_FLAG_DISK_ONLY))
6752				continue;
6753
6754#if 0
6755			printf("found page %#x len %d\n",
6756			       lun->mode_pages.index[i].page_code &
6757			       SMPH_PC_MASK,
6758			       lun->mode_pages.index[i].page_len);
6759#endif
6760
6761			page_len += lun->mode_pages.index[i].page_len;
6762		}
6763
6764		if (page_len == 0) {
6765			ctl_set_invalid_field(ctsio,
6766					      /*sks_valid*/ 1,
6767					      /*command*/ 1,
6768					      /*field*/ 2,
6769					      /*bit_valid*/ 1,
6770					      /*bit*/ 5);
6771			ctl_done((union ctl_io *)ctsio);
6772			return (CTL_RETVAL_COMPLETE);
6773		}
6774		break;
6775	}
6776	}
6777
6778	total_len = header_len + page_len;
6779#if 0
6780	printf("header_len = %d, page_len = %d, total_len = %d\n",
6781	       header_len, page_len, total_len);
6782#endif
6783
6784	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6785	ctsio->kern_sg_entries = 0;
6786	ctsio->kern_data_resid = 0;
6787	ctsio->kern_rel_offset = 0;
6788	if (total_len < alloc_len) {
6789		ctsio->residual = alloc_len - total_len;
6790		ctsio->kern_data_len = total_len;
6791		ctsio->kern_total_len = total_len;
6792	} else {
6793		ctsio->residual = 0;
6794		ctsio->kern_data_len = alloc_len;
6795		ctsio->kern_total_len = alloc_len;
6796	}
6797
6798	switch (ctsio->cdb[0]) {
6799	case MODE_SENSE_6: {
6800		struct scsi_mode_hdr_6 *header;
6801
6802		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6803
6804		header->datalen = ctl_min(total_len - 1, 254);
6805		if (control_dev == 0) {
6806			header->dev_specific = 0x10; /* DPOFUA */
6807			if ((lun->flags & CTL_LUN_READONLY) ||
6808			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6809			    .eca_and_aen & SCP_SWP) != 0)
6810				    header->dev_specific |= 0x80; /* WP */
6811		}
6812		if (dbd)
6813			header->block_descr_len = 0;
6814		else
6815			header->block_descr_len =
6816				sizeof(struct scsi_mode_block_descr);
6817		block_desc = (struct scsi_mode_block_descr *)&header[1];
6818		break;
6819	}
6820	case MODE_SENSE_10: {
6821		struct scsi_mode_hdr_10 *header;
6822		int datalen;
6823
6824		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6825
6826		datalen = ctl_min(total_len - 2, 65533);
6827		scsi_ulto2b(datalen, header->datalen);
6828		if (control_dev == 0) {
6829			header->dev_specific = 0x10; /* DPOFUA */
6830			if ((lun->flags & CTL_LUN_READONLY) ||
6831			    (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6832			    .eca_and_aen & SCP_SWP) != 0)
6833				    header->dev_specific |= 0x80; /* WP */
6834		}
6835		if (dbd)
6836			scsi_ulto2b(0, header->block_descr_len);
6837		else
6838			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6839				    header->block_descr_len);
6840		block_desc = (struct scsi_mode_block_descr *)&header[1];
6841		break;
6842	}
6843	default:
6844		panic("invalid CDB type %#x", ctsio->cdb[0]);
6845		break; /* NOTREACHED */
6846	}
6847
6848	/*
6849	 * If we've got a disk, use its blocksize in the block
6850	 * descriptor.  Otherwise, just set it to 0.
6851	 */
6852	if (dbd == 0) {
6853		if (control_dev == 0)
6854			scsi_ulto3b(lun->be_lun->blocksize,
6855				    block_desc->block_len);
6856		else
6857			scsi_ulto3b(0, block_desc->block_len);
6858	}
6859
6860	switch (page_code) {
6861	case SMS_ALL_PAGES_PAGE: {
6862		int i, data_used;
6863
6864		data_used = header_len;
6865		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6866			struct ctl_page_index *page_index;
6867
6868			page_index = &lun->mode_pages.index[i];
6869
6870			if ((control_dev != 0)
6871			 && (page_index->page_flags &
6872			    CTL_PAGE_FLAG_DISK_ONLY))
6873				continue;
6874
6875			/*
6876			 * We don't use this subpage if the user didn't
6877			 * request all subpages.  We already checked (above)
6878			 * to make sure the user only specified a subpage
6879			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6880			 */
6881			if ((page_index->subpage != 0)
6882			 && (subpage == SMS_SUBPAGE_PAGE_0))
6883				continue;
6884
6885			/*
6886			 * Call the handler, if it exists, to update the
6887			 * page to the latest values.
6888			 */
6889			if (page_index->sense_handler != NULL)
6890				page_index->sense_handler(ctsio, page_index,pc);
6891
6892			memcpy(ctsio->kern_data_ptr + data_used,
6893			       page_index->page_data +
6894			       (page_index->page_len * pc),
6895			       page_index->page_len);
6896			data_used += page_index->page_len;
6897		}
6898		break;
6899	}
6900	default: {
6901		int i, data_used;
6902
6903		data_used = header_len;
6904
6905		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6906			struct ctl_page_index *page_index;
6907
6908			page_index = &lun->mode_pages.index[i];
6909
6910			/* Look for the right page code */
6911			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6912				continue;
6913
6914			/* Look for the right subpage or the subpage wildcard*/
6915			if ((page_index->subpage != subpage)
6916			 && (subpage != SMS_SUBPAGE_ALL))
6917				continue;
6918
6919			/* Make sure the page is supported for this dev type */
6920			if ((control_dev != 0)
6921			 && (page_index->page_flags &
6922			     CTL_PAGE_FLAG_DISK_ONLY))
6923				continue;
6924
6925			/*
6926			 * Call the handler, if it exists, to update the
6927			 * page to the latest values.
6928			 */
6929			if (page_index->sense_handler != NULL)
6930				page_index->sense_handler(ctsio, page_index,pc);
6931
6932			memcpy(ctsio->kern_data_ptr + data_used,
6933			       page_index->page_data +
6934			       (page_index->page_len * pc),
6935			       page_index->page_len);
6936			data_used += page_index->page_len;
6937		}
6938		break;
6939	}
6940	}
6941
6942	ctl_set_success(ctsio);
6943	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6944	ctsio->be_move_done = ctl_config_move_done;
6945	ctl_datamove((union ctl_io *)ctsio);
6946	return (CTL_RETVAL_COMPLETE);
6947}
6948
6949int
6950ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6951			       struct ctl_page_index *page_index,
6952			       int pc)
6953{
6954	struct ctl_lun *lun;
6955	struct scsi_log_param_header *phdr;
6956	uint8_t *data;
6957	uint64_t val;
6958
6959	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6960	data = page_index->page_data;
6961
6962	if (lun->backend->lun_attr != NULL &&
6963	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6964	     != UINT64_MAX) {
6965		phdr = (struct scsi_log_param_header *)data;
6966		scsi_ulto2b(0x0001, phdr->param_code);
6967		phdr->param_control = SLP_LBIN | SLP_LP;
6968		phdr->param_len = 8;
6969		data = (uint8_t *)(phdr + 1);
6970		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6971		data[4] = 0x02; /* per-pool */
6972		data += phdr->param_len;
6973	}
6974
6975	if (lun->backend->lun_attr != NULL &&
6976	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6977	     != UINT64_MAX) {
6978		phdr = (struct scsi_log_param_header *)data;
6979		scsi_ulto2b(0x0002, phdr->param_code);
6980		phdr->param_control = SLP_LBIN | SLP_LP;
6981		phdr->param_len = 8;
6982		data = (uint8_t *)(phdr + 1);
6983		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6984		data[4] = 0x01; /* per-LUN */
6985		data += phdr->param_len;
6986	}
6987
6988	if (lun->backend->lun_attr != NULL &&
6989	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6990	     != UINT64_MAX) {
6991		phdr = (struct scsi_log_param_header *)data;
6992		scsi_ulto2b(0x00f1, phdr->param_code);
6993		phdr->param_control = SLP_LBIN | SLP_LP;
6994		phdr->param_len = 8;
6995		data = (uint8_t *)(phdr + 1);
6996		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6997		data[4] = 0x02; /* per-pool */
6998		data += phdr->param_len;
6999	}
7000
7001	if (lun->backend->lun_attr != NULL &&
7002	    (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
7003	     != UINT64_MAX) {
7004		phdr = (struct scsi_log_param_header *)data;
7005		scsi_ulto2b(0x00f2, phdr->param_code);
7006		phdr->param_control = SLP_LBIN | SLP_LP;
7007		phdr->param_len = 8;
7008		data = (uint8_t *)(phdr + 1);
7009		scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
7010		data[4] = 0x02; /* per-pool */
7011		data += phdr->param_len;
7012	}
7013
7014	page_index->page_len = data - page_index->page_data;
7015	return (0);
7016}
7017
7018int
7019ctl_log_sense(struct ctl_scsiio *ctsio)
7020{
7021	struct ctl_lun *lun;
7022	int i, pc, page_code, subpage;
7023	int alloc_len, total_len;
7024	struct ctl_page_index *page_index;
7025	struct scsi_log_sense *cdb;
7026	struct scsi_log_header *header;
7027
7028	CTL_DEBUG_PRINT(("ctl_log_sense\n"));
7029
7030	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7031	cdb = (struct scsi_log_sense *)ctsio->cdb;
7032	pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
7033	page_code = cdb->page & SLS_PAGE_CODE;
7034	subpage = cdb->subpage;
7035	alloc_len = scsi_2btoul(cdb->length);
7036
7037	page_index = NULL;
7038	for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
7039		page_index = &lun->log_pages.index[i];
7040
7041		/* Look for the right page code */
7042		if ((page_index->page_code & SL_PAGE_CODE) != page_code)
7043			continue;
7044
7045		/* Look for the right subpage or the subpage wildcard*/
7046		if (page_index->subpage != subpage)
7047			continue;
7048
7049		break;
7050	}
7051	if (i >= CTL_NUM_LOG_PAGES) {
7052		ctl_set_invalid_field(ctsio,
7053				      /*sks_valid*/ 1,
7054				      /*command*/ 1,
7055				      /*field*/ 2,
7056				      /*bit_valid*/ 0,
7057				      /*bit*/ 0);
7058		ctl_done((union ctl_io *)ctsio);
7059		return (CTL_RETVAL_COMPLETE);
7060	}
7061
7062	total_len = sizeof(struct scsi_log_header) + page_index->page_len;
7063
7064	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7065	ctsio->kern_sg_entries = 0;
7066	ctsio->kern_data_resid = 0;
7067	ctsio->kern_rel_offset = 0;
7068	if (total_len < alloc_len) {
7069		ctsio->residual = alloc_len - total_len;
7070		ctsio->kern_data_len = total_len;
7071		ctsio->kern_total_len = total_len;
7072	} else {
7073		ctsio->residual = 0;
7074		ctsio->kern_data_len = alloc_len;
7075		ctsio->kern_total_len = alloc_len;
7076	}
7077
7078	header = (struct scsi_log_header *)ctsio->kern_data_ptr;
7079	header->page = page_index->page_code;
7080	if (page_index->subpage) {
7081		header->page |= SL_SPF;
7082		header->subpage = page_index->subpage;
7083	}
7084	scsi_ulto2b(page_index->page_len, header->datalen);
7085
7086	/*
7087	 * Call the handler, if it exists, to update the
7088	 * page to the latest values.
7089	 */
7090	if (page_index->sense_handler != NULL)
7091		page_index->sense_handler(ctsio, page_index, pc);
7092
7093	memcpy(header + 1, page_index->page_data, page_index->page_len);
7094
7095	ctl_set_success(ctsio);
7096	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7097	ctsio->be_move_done = ctl_config_move_done;
7098	ctl_datamove((union ctl_io *)ctsio);
7099	return (CTL_RETVAL_COMPLETE);
7100}
7101
7102int
7103ctl_read_capacity(struct ctl_scsiio *ctsio)
7104{
7105	struct scsi_read_capacity *cdb;
7106	struct scsi_read_capacity_data *data;
7107	struct ctl_lun *lun;
7108	uint32_t lba;
7109
7110	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7111
7112	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7113
7114	lba = scsi_4btoul(cdb->addr);
7115	if (((cdb->pmi & SRC_PMI) == 0)
7116	 && (lba != 0)) {
7117		ctl_set_invalid_field(/*ctsio*/ ctsio,
7118				      /*sks_valid*/ 1,
7119				      /*command*/ 1,
7120				      /*field*/ 2,
7121				      /*bit_valid*/ 0,
7122				      /*bit*/ 0);
7123		ctl_done((union ctl_io *)ctsio);
7124		return (CTL_RETVAL_COMPLETE);
7125	}
7126
7127	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7128
7129	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7130	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7131	ctsio->residual = 0;
7132	ctsio->kern_data_len = sizeof(*data);
7133	ctsio->kern_total_len = sizeof(*data);
7134	ctsio->kern_data_resid = 0;
7135	ctsio->kern_rel_offset = 0;
7136	ctsio->kern_sg_entries = 0;
7137
7138	/*
7139	 * If the maximum LBA is greater than 0xfffffffe, the user must
7140	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7141	 * serivce action set.
7142	 */
7143	if (lun->be_lun->maxlba > 0xfffffffe)
7144		scsi_ulto4b(0xffffffff, data->addr);
7145	else
7146		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7147
7148	/*
7149	 * XXX KDM this may not be 512 bytes...
7150	 */
7151	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7152
7153	ctl_set_success(ctsio);
7154	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7155	ctsio->be_move_done = ctl_config_move_done;
7156	ctl_datamove((union ctl_io *)ctsio);
7157	return (CTL_RETVAL_COMPLETE);
7158}
7159
7160int
7161ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7162{
7163	struct scsi_read_capacity_16 *cdb;
7164	struct scsi_read_capacity_data_long *data;
7165	struct ctl_lun *lun;
7166	uint64_t lba;
7167	uint32_t alloc_len;
7168
7169	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7170
7171	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7172
7173	alloc_len = scsi_4btoul(cdb->alloc_len);
7174	lba = scsi_8btou64(cdb->addr);
7175
7176	if ((cdb->reladr & SRC16_PMI)
7177	 && (lba != 0)) {
7178		ctl_set_invalid_field(/*ctsio*/ ctsio,
7179				      /*sks_valid*/ 1,
7180				      /*command*/ 1,
7181				      /*field*/ 2,
7182				      /*bit_valid*/ 0,
7183				      /*bit*/ 0);
7184		ctl_done((union ctl_io *)ctsio);
7185		return (CTL_RETVAL_COMPLETE);
7186	}
7187
7188	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7189
7190	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7191	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7192
7193	if (sizeof(*data) < alloc_len) {
7194		ctsio->residual = alloc_len - sizeof(*data);
7195		ctsio->kern_data_len = sizeof(*data);
7196		ctsio->kern_total_len = sizeof(*data);
7197	} else {
7198		ctsio->residual = 0;
7199		ctsio->kern_data_len = alloc_len;
7200		ctsio->kern_total_len = alloc_len;
7201	}
7202	ctsio->kern_data_resid = 0;
7203	ctsio->kern_rel_offset = 0;
7204	ctsio->kern_sg_entries = 0;
7205
7206	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7207	/* XXX KDM this may not be 512 bytes... */
7208	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7209	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7210	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7211	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7212		data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7213
7214	ctl_set_success(ctsio);
7215	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7216	ctsio->be_move_done = ctl_config_move_done;
7217	ctl_datamove((union ctl_io *)ctsio);
7218	return (CTL_RETVAL_COMPLETE);
7219}
7220
7221int
7222ctl_get_lba_status(struct ctl_scsiio *ctsio)
7223{
7224	struct scsi_get_lba_status *cdb;
7225	struct scsi_get_lba_status_data *data;
7226	struct ctl_lun *lun;
7227	struct ctl_lba_len_flags *lbalen;
7228	uint64_t lba;
7229	uint32_t alloc_len, total_len;
7230	int retval;
7231
7232	CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7233
7234	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7235	cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7236	lba = scsi_8btou64(cdb->addr);
7237	alloc_len = scsi_4btoul(cdb->alloc_len);
7238
7239	if (lba > lun->be_lun->maxlba) {
7240		ctl_set_lba_out_of_range(ctsio);
7241		ctl_done((union ctl_io *)ctsio);
7242		return (CTL_RETVAL_COMPLETE);
7243	}
7244
7245	total_len = sizeof(*data) + sizeof(data->descr[0]);
7246	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7247	data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7248
7249	if (total_len < alloc_len) {
7250		ctsio->residual = alloc_len - total_len;
7251		ctsio->kern_data_len = total_len;
7252		ctsio->kern_total_len = total_len;
7253	} else {
7254		ctsio->residual = 0;
7255		ctsio->kern_data_len = alloc_len;
7256		ctsio->kern_total_len = alloc_len;
7257	}
7258	ctsio->kern_data_resid = 0;
7259	ctsio->kern_rel_offset = 0;
7260	ctsio->kern_sg_entries = 0;
7261
7262	/* Fill dummy data in case backend can't tell anything. */
7263	scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7264	scsi_u64to8b(lba, data->descr[0].addr);
7265	scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7266	    data->descr[0].length);
7267	data->descr[0].status = 0; /* Mapped or unknown. */
7268
7269	ctl_set_success(ctsio);
7270	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7271	ctsio->be_move_done = ctl_config_move_done;
7272
7273	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7274	lbalen->lba = lba;
7275	lbalen->len = total_len;
7276	lbalen->flags = 0;
7277	retval = lun->backend->config_read((union ctl_io *)ctsio);
7278	return (CTL_RETVAL_COMPLETE);
7279}
7280
7281int
7282ctl_read_defect(struct ctl_scsiio *ctsio)
7283{
7284	struct scsi_read_defect_data_10 *ccb10;
7285	struct scsi_read_defect_data_12 *ccb12;
7286	struct scsi_read_defect_data_hdr_10 *data10;
7287	struct scsi_read_defect_data_hdr_12 *data12;
7288	uint32_t alloc_len, data_len;
7289	uint8_t format;
7290
7291	CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7292
7293	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7294		ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7295		format = ccb10->format;
7296		alloc_len = scsi_2btoul(ccb10->alloc_length);
7297		data_len = sizeof(*data10);
7298	} else {
7299		ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7300		format = ccb12->format;
7301		alloc_len = scsi_4btoul(ccb12->alloc_length);
7302		data_len = sizeof(*data12);
7303	}
7304	if (alloc_len == 0) {
7305		ctl_set_success(ctsio);
7306		ctl_done((union ctl_io *)ctsio);
7307		return (CTL_RETVAL_COMPLETE);
7308	}
7309
7310	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7311	if (data_len < alloc_len) {
7312		ctsio->residual = alloc_len - data_len;
7313		ctsio->kern_data_len = data_len;
7314		ctsio->kern_total_len = data_len;
7315	} else {
7316		ctsio->residual = 0;
7317		ctsio->kern_data_len = alloc_len;
7318		ctsio->kern_total_len = alloc_len;
7319	}
7320	ctsio->kern_data_resid = 0;
7321	ctsio->kern_rel_offset = 0;
7322	ctsio->kern_sg_entries = 0;
7323
7324	if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7325		data10 = (struct scsi_read_defect_data_hdr_10 *)
7326		    ctsio->kern_data_ptr;
7327		data10->format = format;
7328		scsi_ulto2b(0, data10->length);
7329	} else {
7330		data12 = (struct scsi_read_defect_data_hdr_12 *)
7331		    ctsio->kern_data_ptr;
7332		data12->format = format;
7333		scsi_ulto2b(0, data12->generation);
7334		scsi_ulto4b(0, data12->length);
7335	}
7336
7337	ctl_set_success(ctsio);
7338	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7339	ctsio->be_move_done = ctl_config_move_done;
7340	ctl_datamove((union ctl_io *)ctsio);
7341	return (CTL_RETVAL_COMPLETE);
7342}
7343
7344int
7345ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7346{
7347	struct scsi_maintenance_in *cdb;
7348	int retval;
7349	int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os;
7350	int num_target_port_groups, num_target_ports;
7351	struct ctl_lun *lun;
7352	struct ctl_softc *softc;
7353	struct ctl_port *port;
7354	struct scsi_target_group_data *rtg_ptr;
7355	struct scsi_target_group_data_extended *rtg_ext_ptr;
7356	struct scsi_target_port_group_descriptor *tpg_desc;
7357
7358	CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7359
7360	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7361	softc = control_softc;
7362	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7363
7364	retval = CTL_RETVAL_COMPLETE;
7365
7366	switch (cdb->byte2 & STG_PDF_MASK) {
7367	case STG_PDF_LENGTH:
7368		ext = 0;
7369		break;
7370	case STG_PDF_EXTENDED:
7371		ext = 1;
7372		break;
7373	default:
7374		ctl_set_invalid_field(/*ctsio*/ ctsio,
7375				      /*sks_valid*/ 1,
7376				      /*command*/ 1,
7377				      /*field*/ 2,
7378				      /*bit_valid*/ 1,
7379				      /*bit*/ 5);
7380		ctl_done((union ctl_io *)ctsio);
7381		return(retval);
7382	}
7383
7384	if (softc->is_single)
7385		num_target_port_groups = 1;
7386	else
7387		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7388	num_target_ports = 0;
7389	mtx_lock(&softc->ctl_lock);
7390	STAILQ_FOREACH(port, &softc->port_list, links) {
7391		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7392			continue;
7393		if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS)
7394			continue;
7395		num_target_ports++;
7396	}
7397	mtx_unlock(&softc->ctl_lock);
7398
7399	if (ext)
7400		total_len = sizeof(struct scsi_target_group_data_extended);
7401	else
7402		total_len = sizeof(struct scsi_target_group_data);
7403	total_len += sizeof(struct scsi_target_port_group_descriptor) *
7404		num_target_port_groups +
7405	    sizeof(struct scsi_target_port_descriptor) *
7406		num_target_ports * num_target_port_groups;
7407
7408	alloc_len = scsi_4btoul(cdb->length);
7409
7410	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7411
7412	ctsio->kern_sg_entries = 0;
7413
7414	if (total_len < alloc_len) {
7415		ctsio->residual = alloc_len - total_len;
7416		ctsio->kern_data_len = total_len;
7417		ctsio->kern_total_len = total_len;
7418	} else {
7419		ctsio->residual = 0;
7420		ctsio->kern_data_len = alloc_len;
7421		ctsio->kern_total_len = alloc_len;
7422	}
7423	ctsio->kern_data_resid = 0;
7424	ctsio->kern_rel_offset = 0;
7425
7426	if (ext) {
7427		rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7428		    ctsio->kern_data_ptr;
7429		scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7430		rtg_ext_ptr->format_type = 0x10;
7431		rtg_ext_ptr->implicit_transition_time = 0;
7432		tpg_desc = &rtg_ext_ptr->groups[0];
7433	} else {
7434		rtg_ptr = (struct scsi_target_group_data *)
7435		    ctsio->kern_data_ptr;
7436		scsi_ulto4b(total_len - 4, rtg_ptr->length);
7437		tpg_desc = &rtg_ptr->groups[0];
7438	}
7439
7440	mtx_lock(&softc->ctl_lock);
7441	pg = softc->port_offset / CTL_MAX_PORTS;
7442	if (softc->flags & CTL_FLAG_ACTIVE_SHELF) {
7443		if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) {
7444			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7445			os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7446		} else if (lun->flags & CTL_LUN_PRIMARY_SC) {
7447			gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7448			os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7449		} else {
7450			gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7451			os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7452		}
7453	} else {
7454		gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7455		os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7456	}
7457	for (g = 0; g < num_target_port_groups; g++) {
7458		tpg_desc->pref_state = (g == pg) ? gs : os;
7459		tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP;
7460		scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7461		tpg_desc->status = TPG_IMPLICIT;
7462		pc = 0;
7463		STAILQ_FOREACH(port, &softc->port_list, links) {
7464			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7465				continue;
7466			if (ctl_map_lun_back(port->targ_port, lun->lun) >=
7467			    CTL_MAX_LUNS)
7468				continue;
7469			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
7470			scsi_ulto2b(p, tpg_desc->descriptors[pc].
7471			    relative_target_port_identifier);
7472			pc++;
7473		}
7474		tpg_desc->target_port_count = pc;
7475		tpg_desc = (struct scsi_target_port_group_descriptor *)
7476		    &tpg_desc->descriptors[pc];
7477	}
7478	mtx_unlock(&softc->ctl_lock);
7479
7480	ctl_set_success(ctsio);
7481	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7482	ctsio->be_move_done = ctl_config_move_done;
7483	ctl_datamove((union ctl_io *)ctsio);
7484	return(retval);
7485}
7486
7487int
7488ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7489{
7490	struct ctl_lun *lun;
7491	struct scsi_report_supported_opcodes *cdb;
7492	const struct ctl_cmd_entry *entry, *sentry;
7493	struct scsi_report_supported_opcodes_all *all;
7494	struct scsi_report_supported_opcodes_descr *descr;
7495	struct scsi_report_supported_opcodes_one *one;
7496	int retval;
7497	int alloc_len, total_len;
7498	int opcode, service_action, i, j, num;
7499
7500	CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7501
7502	cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7503	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7504
7505	retval = CTL_RETVAL_COMPLETE;
7506
7507	opcode = cdb->requested_opcode;
7508	service_action = scsi_2btoul(cdb->requested_service_action);
7509	switch (cdb->options & RSO_OPTIONS_MASK) {
7510	case RSO_OPTIONS_ALL:
7511		num = 0;
7512		for (i = 0; i < 256; i++) {
7513			entry = &ctl_cmd_table[i];
7514			if (entry->flags & CTL_CMD_FLAG_SA5) {
7515				for (j = 0; j < 32; j++) {
7516					sentry = &((const struct ctl_cmd_entry *)
7517					    entry->execute)[j];
7518					if (ctl_cmd_applicable(
7519					    lun->be_lun->lun_type, sentry))
7520						num++;
7521				}
7522			} else {
7523				if (ctl_cmd_applicable(lun->be_lun->lun_type,
7524				    entry))
7525					num++;
7526			}
7527		}
7528		total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7529		    num * sizeof(struct scsi_report_supported_opcodes_descr);
7530		break;
7531	case RSO_OPTIONS_OC:
7532		if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7533			ctl_set_invalid_field(/*ctsio*/ ctsio,
7534					      /*sks_valid*/ 1,
7535					      /*command*/ 1,
7536					      /*field*/ 2,
7537					      /*bit_valid*/ 1,
7538					      /*bit*/ 2);
7539			ctl_done((union ctl_io *)ctsio);
7540			return (CTL_RETVAL_COMPLETE);
7541		}
7542		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7543		break;
7544	case RSO_OPTIONS_OC_SA:
7545		if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7546		    service_action >= 32) {
7547			ctl_set_invalid_field(/*ctsio*/ ctsio,
7548					      /*sks_valid*/ 1,
7549					      /*command*/ 1,
7550					      /*field*/ 2,
7551					      /*bit_valid*/ 1,
7552					      /*bit*/ 2);
7553			ctl_done((union ctl_io *)ctsio);
7554			return (CTL_RETVAL_COMPLETE);
7555		}
7556		total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7557		break;
7558	default:
7559		ctl_set_invalid_field(/*ctsio*/ ctsio,
7560				      /*sks_valid*/ 1,
7561				      /*command*/ 1,
7562				      /*field*/ 2,
7563				      /*bit_valid*/ 1,
7564				      /*bit*/ 2);
7565		ctl_done((union ctl_io *)ctsio);
7566		return (CTL_RETVAL_COMPLETE);
7567	}
7568
7569	alloc_len = scsi_4btoul(cdb->length);
7570
7571	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7572
7573	ctsio->kern_sg_entries = 0;
7574
7575	if (total_len < alloc_len) {
7576		ctsio->residual = alloc_len - total_len;
7577		ctsio->kern_data_len = total_len;
7578		ctsio->kern_total_len = total_len;
7579	} else {
7580		ctsio->residual = 0;
7581		ctsio->kern_data_len = alloc_len;
7582		ctsio->kern_total_len = alloc_len;
7583	}
7584	ctsio->kern_data_resid = 0;
7585	ctsio->kern_rel_offset = 0;
7586
7587	switch (cdb->options & RSO_OPTIONS_MASK) {
7588	case RSO_OPTIONS_ALL:
7589		all = (struct scsi_report_supported_opcodes_all *)
7590		    ctsio->kern_data_ptr;
7591		num = 0;
7592		for (i = 0; i < 256; i++) {
7593			entry = &ctl_cmd_table[i];
7594			if (entry->flags & CTL_CMD_FLAG_SA5) {
7595				for (j = 0; j < 32; j++) {
7596					sentry = &((const struct ctl_cmd_entry *)
7597					    entry->execute)[j];
7598					if (!ctl_cmd_applicable(
7599					    lun->be_lun->lun_type, sentry))
7600						continue;
7601					descr = &all->descr[num++];
7602					descr->opcode = i;
7603					scsi_ulto2b(j, descr->service_action);
7604					descr->flags = RSO_SERVACTV;
7605					scsi_ulto2b(sentry->length,
7606					    descr->cdb_length);
7607				}
7608			} else {
7609				if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7610				    entry))
7611					continue;
7612				descr = &all->descr[num++];
7613				descr->opcode = i;
7614				scsi_ulto2b(0, descr->service_action);
7615				descr->flags = 0;
7616				scsi_ulto2b(entry->length, descr->cdb_length);
7617			}
7618		}
7619		scsi_ulto4b(
7620		    num * sizeof(struct scsi_report_supported_opcodes_descr),
7621		    all->length);
7622		break;
7623	case RSO_OPTIONS_OC:
7624		one = (struct scsi_report_supported_opcodes_one *)
7625		    ctsio->kern_data_ptr;
7626		entry = &ctl_cmd_table[opcode];
7627		goto fill_one;
7628	case RSO_OPTIONS_OC_SA:
7629		one = (struct scsi_report_supported_opcodes_one *)
7630		    ctsio->kern_data_ptr;
7631		entry = &ctl_cmd_table[opcode];
7632		entry = &((const struct ctl_cmd_entry *)
7633		    entry->execute)[service_action];
7634fill_one:
7635		if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7636			one->support = 3;
7637			scsi_ulto2b(entry->length, one->cdb_length);
7638			one->cdb_usage[0] = opcode;
7639			memcpy(&one->cdb_usage[1], entry->usage,
7640			    entry->length - 1);
7641		} else
7642			one->support = 1;
7643		break;
7644	}
7645
7646	ctl_set_success(ctsio);
7647	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7648	ctsio->be_move_done = ctl_config_move_done;
7649	ctl_datamove((union ctl_io *)ctsio);
7650	return(retval);
7651}
7652
7653int
7654ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7655{
7656	struct scsi_report_supported_tmf *cdb;
7657	struct scsi_report_supported_tmf_data *data;
7658	int retval;
7659	int alloc_len, total_len;
7660
7661	CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7662
7663	cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7664
7665	retval = CTL_RETVAL_COMPLETE;
7666
7667	total_len = sizeof(struct scsi_report_supported_tmf_data);
7668	alloc_len = scsi_4btoul(cdb->length);
7669
7670	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7671
7672	ctsio->kern_sg_entries = 0;
7673
7674	if (total_len < alloc_len) {
7675		ctsio->residual = alloc_len - total_len;
7676		ctsio->kern_data_len = total_len;
7677		ctsio->kern_total_len = total_len;
7678	} else {
7679		ctsio->residual = 0;
7680		ctsio->kern_data_len = alloc_len;
7681		ctsio->kern_total_len = alloc_len;
7682	}
7683	ctsio->kern_data_resid = 0;
7684	ctsio->kern_rel_offset = 0;
7685
7686	data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7687	data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7688	data->byte2 |= RST_ITNRS;
7689
7690	ctl_set_success(ctsio);
7691	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7692	ctsio->be_move_done = ctl_config_move_done;
7693	ctl_datamove((union ctl_io *)ctsio);
7694	return (retval);
7695}
7696
7697int
7698ctl_report_timestamp(struct ctl_scsiio *ctsio)
7699{
7700	struct scsi_report_timestamp *cdb;
7701	struct scsi_report_timestamp_data *data;
7702	struct timeval tv;
7703	int64_t timestamp;
7704	int retval;
7705	int alloc_len, total_len;
7706
7707	CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7708
7709	cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7710
7711	retval = CTL_RETVAL_COMPLETE;
7712
7713	total_len = sizeof(struct scsi_report_timestamp_data);
7714	alloc_len = scsi_4btoul(cdb->length);
7715
7716	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7717
7718	ctsio->kern_sg_entries = 0;
7719
7720	if (total_len < alloc_len) {
7721		ctsio->residual = alloc_len - total_len;
7722		ctsio->kern_data_len = total_len;
7723		ctsio->kern_total_len = total_len;
7724	} else {
7725		ctsio->residual = 0;
7726		ctsio->kern_data_len = alloc_len;
7727		ctsio->kern_total_len = alloc_len;
7728	}
7729	ctsio->kern_data_resid = 0;
7730	ctsio->kern_rel_offset = 0;
7731
7732	data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7733	scsi_ulto2b(sizeof(*data) - 2, data->length);
7734	data->origin = RTS_ORIG_OUTSIDE;
7735	getmicrotime(&tv);
7736	timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7737	scsi_ulto4b(timestamp >> 16, data->timestamp);
7738	scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7739
7740	ctl_set_success(ctsio);
7741	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7742	ctsio->be_move_done = ctl_config_move_done;
7743	ctl_datamove((union ctl_io *)ctsio);
7744	return (retval);
7745}
7746
7747int
7748ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7749{
7750	struct scsi_per_res_in *cdb;
7751	int alloc_len, total_len = 0;
7752	/* struct scsi_per_res_in_rsrv in_data; */
7753	struct ctl_lun *lun;
7754	struct ctl_softc *softc;
7755	uint64_t key;
7756
7757	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7758
7759	softc = control_softc;
7760
7761	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7762
7763	alloc_len = scsi_2btoul(cdb->length);
7764
7765	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7766
7767retry:
7768	mtx_lock(&lun->lun_lock);
7769	switch (cdb->action) {
7770	case SPRI_RK: /* read keys */
7771		total_len = sizeof(struct scsi_per_res_in_keys) +
7772			lun->pr_key_count *
7773			sizeof(struct scsi_per_res_key);
7774		break;
7775	case SPRI_RR: /* read reservation */
7776		if (lun->flags & CTL_LUN_PR_RESERVED)
7777			total_len = sizeof(struct scsi_per_res_in_rsrv);
7778		else
7779			total_len = sizeof(struct scsi_per_res_in_header);
7780		break;
7781	case SPRI_RC: /* report capabilities */
7782		total_len = sizeof(struct scsi_per_res_cap);
7783		break;
7784	case SPRI_RS: /* read full status */
7785		total_len = sizeof(struct scsi_per_res_in_header) +
7786		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7787		    lun->pr_key_count;
7788		break;
7789	default:
7790		panic("Invalid PR type %x", cdb->action);
7791	}
7792	mtx_unlock(&lun->lun_lock);
7793
7794	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7795
7796	if (total_len < alloc_len) {
7797		ctsio->residual = alloc_len - total_len;
7798		ctsio->kern_data_len = total_len;
7799		ctsio->kern_total_len = total_len;
7800	} else {
7801		ctsio->residual = 0;
7802		ctsio->kern_data_len = alloc_len;
7803		ctsio->kern_total_len = alloc_len;
7804	}
7805
7806	ctsio->kern_data_resid = 0;
7807	ctsio->kern_rel_offset = 0;
7808	ctsio->kern_sg_entries = 0;
7809
7810	mtx_lock(&lun->lun_lock);
7811	switch (cdb->action) {
7812	case SPRI_RK: { // read keys
7813        struct scsi_per_res_in_keys *res_keys;
7814		int i, key_count;
7815
7816		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7817
7818		/*
7819		 * We had to drop the lock to allocate our buffer, which
7820		 * leaves time for someone to come in with another
7821		 * persistent reservation.  (That is unlikely, though,
7822		 * since this should be the only persistent reservation
7823		 * command active right now.)
7824		 */
7825		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7826		    (lun->pr_key_count *
7827		     sizeof(struct scsi_per_res_key)))){
7828			mtx_unlock(&lun->lun_lock);
7829			free(ctsio->kern_data_ptr, M_CTL);
7830			printf("%s: reservation length changed, retrying\n",
7831			       __func__);
7832			goto retry;
7833		}
7834
7835		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7836
7837		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7838			     lun->pr_key_count, res_keys->header.length);
7839
7840		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7841			if ((key = ctl_get_prkey(lun, i)) == 0)
7842				continue;
7843
7844			/*
7845			 * We used lun->pr_key_count to calculate the
7846			 * size to allocate.  If it turns out the number of
7847			 * initiators with the registered flag set is
7848			 * larger than that (i.e. they haven't been kept in
7849			 * sync), we've got a problem.
7850			 */
7851			if (key_count >= lun->pr_key_count) {
7852#ifdef NEEDTOPORT
7853				csevent_log(CSC_CTL | CSC_SHELF_SW |
7854					    CTL_PR_ERROR,
7855					    csevent_LogType_Fault,
7856					    csevent_AlertLevel_Yellow,
7857					    csevent_FRU_ShelfController,
7858					    csevent_FRU_Firmware,
7859				        csevent_FRU_Unknown,
7860					    "registered keys %d >= key "
7861					    "count %d", key_count,
7862					    lun->pr_key_count);
7863#endif
7864				key_count++;
7865				continue;
7866			}
7867			scsi_u64to8b(key, res_keys->keys[key_count].key);
7868			key_count++;
7869		}
7870		break;
7871	}
7872	case SPRI_RR: { // read reservation
7873		struct scsi_per_res_in_rsrv *res;
7874		int tmp_len, header_only;
7875
7876		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7877
7878		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7879
7880		if (lun->flags & CTL_LUN_PR_RESERVED)
7881		{
7882			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7883			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7884				    res->header.length);
7885			header_only = 0;
7886		} else {
7887			tmp_len = sizeof(struct scsi_per_res_in_header);
7888			scsi_ulto4b(0, res->header.length);
7889			header_only = 1;
7890		}
7891
7892		/*
7893		 * We had to drop the lock to allocate our buffer, which
7894		 * leaves time for someone to come in with another
7895		 * persistent reservation.  (That is unlikely, though,
7896		 * since this should be the only persistent reservation
7897		 * command active right now.)
7898		 */
7899		if (tmp_len != total_len) {
7900			mtx_unlock(&lun->lun_lock);
7901			free(ctsio->kern_data_ptr, M_CTL);
7902			printf("%s: reservation status changed, retrying\n",
7903			       __func__);
7904			goto retry;
7905		}
7906
7907		/*
7908		 * No reservation held, so we're done.
7909		 */
7910		if (header_only != 0)
7911			break;
7912
7913		/*
7914		 * If the registration is an All Registrants type, the key
7915		 * is 0, since it doesn't really matter.
7916		 */
7917		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7918			scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7919			    res->data.reservation);
7920		}
7921		res->data.scopetype = lun->res_type;
7922		break;
7923	}
7924	case SPRI_RC:     //report capabilities
7925	{
7926		struct scsi_per_res_cap *res_cap;
7927		uint16_t type_mask;
7928
7929		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7930		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7931		res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7932		type_mask = SPRI_TM_WR_EX_AR |
7933			    SPRI_TM_EX_AC_RO |
7934			    SPRI_TM_WR_EX_RO |
7935			    SPRI_TM_EX_AC |
7936			    SPRI_TM_WR_EX |
7937			    SPRI_TM_EX_AC_AR;
7938		scsi_ulto2b(type_mask, res_cap->type_mask);
7939		break;
7940	}
7941	case SPRI_RS: { // read full status
7942		struct scsi_per_res_in_full *res_status;
7943		struct scsi_per_res_in_full_desc *res_desc;
7944		struct ctl_port *port;
7945		int i, len;
7946
7947		res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7948
7949		/*
7950		 * We had to drop the lock to allocate our buffer, which
7951		 * leaves time for someone to come in with another
7952		 * persistent reservation.  (That is unlikely, though,
7953		 * since this should be the only persistent reservation
7954		 * command active right now.)
7955		 */
7956		if (total_len < (sizeof(struct scsi_per_res_in_header) +
7957		    (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7958		     lun->pr_key_count)){
7959			mtx_unlock(&lun->lun_lock);
7960			free(ctsio->kern_data_ptr, M_CTL);
7961			printf("%s: reservation length changed, retrying\n",
7962			       __func__);
7963			goto retry;
7964		}
7965
7966		scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7967
7968		res_desc = &res_status->desc[0];
7969		for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7970			if ((key = ctl_get_prkey(lun, i)) == 0)
7971				continue;
7972
7973			scsi_u64to8b(key, res_desc->res_key.key);
7974			if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7975			    (lun->pr_res_idx == i ||
7976			     lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7977				res_desc->flags = SPRI_FULL_R_HOLDER;
7978				res_desc->scopetype = lun->res_type;
7979			}
7980			scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7981			    res_desc->rel_trgt_port_id);
7982			len = 0;
7983			port = softc->ctl_ports[
7984			    ctl_port_idx(i / CTL_MAX_INIT_PER_PORT)];
7985			if (port != NULL)
7986				len = ctl_create_iid(port,
7987				    i % CTL_MAX_INIT_PER_PORT,
7988				    res_desc->transport_id);
7989			scsi_ulto4b(len, res_desc->additional_length);
7990			res_desc = (struct scsi_per_res_in_full_desc *)
7991			    &res_desc->transport_id[len];
7992		}
7993		scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7994		    res_status->header.length);
7995		break;
7996	}
7997	default:
7998		/*
7999		 * This is a bug, because we just checked for this above,
8000		 * and should have returned an error.
8001		 */
8002		panic("Invalid PR type %x", cdb->action);
8003		break; /* NOTREACHED */
8004	}
8005	mtx_unlock(&lun->lun_lock);
8006
8007	ctl_set_success(ctsio);
8008	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8009	ctsio->be_move_done = ctl_config_move_done;
8010	ctl_datamove((union ctl_io *)ctsio);
8011	return (CTL_RETVAL_COMPLETE);
8012}
8013
8014static void
8015ctl_est_res_ua(struct ctl_lun *lun, uint32_t residx, ctl_ua_type ua)
8016{
8017	int off = lun->ctl_softc->persis_offset;
8018
8019	if (residx >= off && residx < off + CTL_MAX_INITIATORS)
8020		ctl_est_ua(lun, residx - off, ua);
8021}
8022
8023/*
8024 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
8025 * it should return.
8026 */
8027static int
8028ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
8029		uint64_t sa_res_key, uint8_t type, uint32_t residx,
8030		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
8031		struct scsi_per_res_out_parms* param)
8032{
8033	union ctl_ha_msg persis_io;
8034	int retval, i;
8035	int isc_retval;
8036
8037	retval = 0;
8038
8039	mtx_lock(&lun->lun_lock);
8040	if (sa_res_key == 0) {
8041		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8042			/* validate scope and type */
8043			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8044			     SPR_LU_SCOPE) {
8045				mtx_unlock(&lun->lun_lock);
8046				ctl_set_invalid_field(/*ctsio*/ ctsio,
8047						      /*sks_valid*/ 1,
8048						      /*command*/ 1,
8049						      /*field*/ 2,
8050						      /*bit_valid*/ 1,
8051						      /*bit*/ 4);
8052				ctl_done((union ctl_io *)ctsio);
8053				return (1);
8054			}
8055
8056		        if (type>8 || type==2 || type==4 || type==0) {
8057				mtx_unlock(&lun->lun_lock);
8058				ctl_set_invalid_field(/*ctsio*/ ctsio,
8059       	           				      /*sks_valid*/ 1,
8060						      /*command*/ 1,
8061						      /*field*/ 2,
8062						      /*bit_valid*/ 1,
8063						      /*bit*/ 0);
8064				ctl_done((union ctl_io *)ctsio);
8065				return (1);
8066		        }
8067
8068			/*
8069			 * Unregister everybody else and build UA for
8070			 * them
8071			 */
8072			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8073				if (i == residx || ctl_get_prkey(lun, i) == 0)
8074					continue;
8075
8076				ctl_clr_prkey(lun, i);
8077				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8078			}
8079			lun->pr_key_count = 1;
8080			lun->res_type = type;
8081			if (lun->res_type != SPR_TYPE_WR_EX_AR
8082			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8083				lun->pr_res_idx = residx;
8084
8085			/* send msg to other side */
8086			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8087			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8088			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8089			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8090			persis_io.pr.pr_info.res_type = type;
8091			memcpy(persis_io.pr.pr_info.sa_res_key,
8092			       param->serv_act_res_key,
8093			       sizeof(param->serv_act_res_key));
8094			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8095			     &persis_io, sizeof(persis_io), 0)) >
8096			     CTL_HA_STATUS_SUCCESS) {
8097				printf("CTL:Persis Out error returned "
8098				       "from ctl_ha_msg_send %d\n",
8099				       isc_retval);
8100			}
8101		} else {
8102			/* not all registrants */
8103			mtx_unlock(&lun->lun_lock);
8104			free(ctsio->kern_data_ptr, M_CTL);
8105			ctl_set_invalid_field(ctsio,
8106					      /*sks_valid*/ 1,
8107					      /*command*/ 0,
8108					      /*field*/ 8,
8109					      /*bit_valid*/ 0,
8110					      /*bit*/ 0);
8111			ctl_done((union ctl_io *)ctsio);
8112			return (1);
8113		}
8114	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8115		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
8116		int found = 0;
8117
8118		if (res_key == sa_res_key) {
8119			/* special case */
8120			/*
8121			 * The spec implies this is not good but doesn't
8122			 * say what to do. There are two choices either
8123			 * generate a res conflict or check condition
8124			 * with illegal field in parameter data. Since
8125			 * that is what is done when the sa_res_key is
8126			 * zero I'll take that approach since this has
8127			 * to do with the sa_res_key.
8128			 */
8129			mtx_unlock(&lun->lun_lock);
8130			free(ctsio->kern_data_ptr, M_CTL);
8131			ctl_set_invalid_field(ctsio,
8132					      /*sks_valid*/ 1,
8133					      /*command*/ 0,
8134					      /*field*/ 8,
8135					      /*bit_valid*/ 0,
8136					      /*bit*/ 0);
8137			ctl_done((union ctl_io *)ctsio);
8138			return (1);
8139		}
8140
8141		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8142			if (ctl_get_prkey(lun, i) != sa_res_key)
8143				continue;
8144
8145			found = 1;
8146			ctl_clr_prkey(lun, i);
8147			lun->pr_key_count--;
8148			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8149		}
8150		if (!found) {
8151			mtx_unlock(&lun->lun_lock);
8152			free(ctsio->kern_data_ptr, M_CTL);
8153			ctl_set_reservation_conflict(ctsio);
8154			ctl_done((union ctl_io *)ctsio);
8155			return (CTL_RETVAL_COMPLETE);
8156		}
8157		/* send msg to other side */
8158		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8159		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8160		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8161		persis_io.pr.pr_info.residx = lun->pr_res_idx;
8162		persis_io.pr.pr_info.res_type = type;
8163		memcpy(persis_io.pr.pr_info.sa_res_key,
8164		       param->serv_act_res_key,
8165		       sizeof(param->serv_act_res_key));
8166		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8167		     &persis_io, sizeof(persis_io), 0)) >
8168		     CTL_HA_STATUS_SUCCESS) {
8169			printf("CTL:Persis Out error returned from "
8170			       "ctl_ha_msg_send %d\n", isc_retval);
8171		}
8172	} else {
8173		/* Reserved but not all registrants */
8174		/* sa_res_key is res holder */
8175		if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8176			/* validate scope and type */
8177			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8178			     SPR_LU_SCOPE) {
8179				mtx_unlock(&lun->lun_lock);
8180				ctl_set_invalid_field(/*ctsio*/ ctsio,
8181						      /*sks_valid*/ 1,
8182						      /*command*/ 1,
8183						      /*field*/ 2,
8184						      /*bit_valid*/ 1,
8185						      /*bit*/ 4);
8186				ctl_done((union ctl_io *)ctsio);
8187				return (1);
8188			}
8189
8190			if (type>8 || type==2 || type==4 || type==0) {
8191				mtx_unlock(&lun->lun_lock);
8192				ctl_set_invalid_field(/*ctsio*/ ctsio,
8193						      /*sks_valid*/ 1,
8194						      /*command*/ 1,
8195						      /*field*/ 2,
8196						      /*bit_valid*/ 1,
8197						      /*bit*/ 0);
8198				ctl_done((union ctl_io *)ctsio);
8199				return (1);
8200			}
8201
8202			/*
8203			 * Do the following:
8204			 * if sa_res_key != res_key remove all
8205			 * registrants w/sa_res_key and generate UA
8206			 * for these registrants(Registrations
8207			 * Preempted) if it wasn't an exclusive
8208			 * reservation generate UA(Reservations
8209			 * Preempted) for all other registered nexuses
8210			 * if the type has changed. Establish the new
8211			 * reservation and holder. If res_key and
8212			 * sa_res_key are the same do the above
8213			 * except don't unregister the res holder.
8214			 */
8215
8216			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8217				if (i == residx || ctl_get_prkey(lun, i) == 0)
8218					continue;
8219
8220				if (sa_res_key == ctl_get_prkey(lun, i)) {
8221					ctl_clr_prkey(lun, i);
8222					lun->pr_key_count--;
8223					ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8224				} else if (type != lun->res_type
8225					&& (lun->res_type == SPR_TYPE_WR_EX_RO
8226					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
8227					ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
8228				}
8229			}
8230			lun->res_type = type;
8231			if (lun->res_type != SPR_TYPE_WR_EX_AR
8232			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8233				lun->pr_res_idx = residx;
8234			else
8235				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8236
8237			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8238			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8239			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8240			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8241			persis_io.pr.pr_info.res_type = type;
8242			memcpy(persis_io.pr.pr_info.sa_res_key,
8243			       param->serv_act_res_key,
8244			       sizeof(param->serv_act_res_key));
8245			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8246			     &persis_io, sizeof(persis_io), 0)) >
8247			     CTL_HA_STATUS_SUCCESS) {
8248				printf("CTL:Persis Out error returned "
8249				       "from ctl_ha_msg_send %d\n",
8250				       isc_retval);
8251			}
8252		} else {
8253			/*
8254			 * sa_res_key is not the res holder just
8255			 * remove registrants
8256			 */
8257			int found=0;
8258
8259			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8260				if (sa_res_key != ctl_get_prkey(lun, i))
8261					continue;
8262
8263				found = 1;
8264				ctl_clr_prkey(lun, i);
8265				lun->pr_key_count--;
8266				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8267			}
8268
8269			if (!found) {
8270				mtx_unlock(&lun->lun_lock);
8271				free(ctsio->kern_data_ptr, M_CTL);
8272				ctl_set_reservation_conflict(ctsio);
8273				ctl_done((union ctl_io *)ctsio);
8274		        	return (1);
8275			}
8276			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8277			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8278			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8279			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8280			persis_io.pr.pr_info.res_type = type;
8281			memcpy(persis_io.pr.pr_info.sa_res_key,
8282			       param->serv_act_res_key,
8283			       sizeof(param->serv_act_res_key));
8284			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8285			     &persis_io, sizeof(persis_io), 0)) >
8286			     CTL_HA_STATUS_SUCCESS) {
8287				printf("CTL:Persis Out error returned "
8288				       "from ctl_ha_msg_send %d\n",
8289				isc_retval);
8290			}
8291		}
8292	}
8293
8294	lun->PRGeneration++;
8295	mtx_unlock(&lun->lun_lock);
8296
8297	return (retval);
8298}
8299
8300static void
8301ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8302{
8303	uint64_t sa_res_key;
8304	int i;
8305
8306	sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8307
8308	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8309	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8310	 || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8311		if (sa_res_key == 0) {
8312			/*
8313			 * Unregister everybody else and build UA for
8314			 * them
8315			 */
8316			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8317				if (i == msg->pr.pr_info.residx ||
8318				    ctl_get_prkey(lun, i) == 0)
8319					continue;
8320
8321				ctl_clr_prkey(lun, i);
8322				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8323			}
8324
8325			lun->pr_key_count = 1;
8326			lun->res_type = msg->pr.pr_info.res_type;
8327			if (lun->res_type != SPR_TYPE_WR_EX_AR
8328			 && lun->res_type != SPR_TYPE_EX_AC_AR)
8329				lun->pr_res_idx = msg->pr.pr_info.residx;
8330		} else {
8331		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8332				if (sa_res_key == ctl_get_prkey(lun, i))
8333					continue;
8334
8335				ctl_clr_prkey(lun, i);
8336				lun->pr_key_count--;
8337				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8338			}
8339		}
8340	} else {
8341		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8342			if (i == msg->pr.pr_info.residx ||
8343			    ctl_get_prkey(lun, i) == 0)
8344				continue;
8345
8346			if (sa_res_key == ctl_get_prkey(lun, i)) {
8347				ctl_clr_prkey(lun, i);
8348				lun->pr_key_count--;
8349				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8350			} else if (msg->pr.pr_info.res_type != lun->res_type
8351				&& (lun->res_type == SPR_TYPE_WR_EX_RO
8352				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8353				ctl_est_res_ua(lun, i, CTL_UA_RES_RELEASE);
8354			}
8355		}
8356		lun->res_type = msg->pr.pr_info.res_type;
8357		if (lun->res_type != SPR_TYPE_WR_EX_AR
8358		 && lun->res_type != SPR_TYPE_EX_AC_AR)
8359			lun->pr_res_idx = msg->pr.pr_info.residx;
8360		else
8361			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8362	}
8363	lun->PRGeneration++;
8364
8365}
8366
8367
8368int
8369ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8370{
8371	int retval;
8372	int isc_retval;
8373	u_int32_t param_len;
8374	struct scsi_per_res_out *cdb;
8375	struct ctl_lun *lun;
8376	struct scsi_per_res_out_parms* param;
8377	struct ctl_softc *softc;
8378	uint32_t residx;
8379	uint64_t res_key, sa_res_key, key;
8380	uint8_t type;
8381	union ctl_ha_msg persis_io;
8382	int    i;
8383
8384	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8385
8386	retval = CTL_RETVAL_COMPLETE;
8387
8388	softc = control_softc;
8389
8390	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8391	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8392
8393	/*
8394	 * We only support whole-LUN scope.  The scope & type are ignored for
8395	 * register, register and ignore existing key and clear.
8396	 * We sometimes ignore scope and type on preempts too!!
8397	 * Verify reservation type here as well.
8398	 */
8399	type = cdb->scope_type & SPR_TYPE_MASK;
8400	if ((cdb->action == SPRO_RESERVE)
8401	 || (cdb->action == SPRO_RELEASE)) {
8402		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8403			ctl_set_invalid_field(/*ctsio*/ ctsio,
8404					      /*sks_valid*/ 1,
8405					      /*command*/ 1,
8406					      /*field*/ 2,
8407					      /*bit_valid*/ 1,
8408					      /*bit*/ 4);
8409			ctl_done((union ctl_io *)ctsio);
8410			return (CTL_RETVAL_COMPLETE);
8411		}
8412
8413		if (type>8 || type==2 || type==4 || type==0) {
8414			ctl_set_invalid_field(/*ctsio*/ ctsio,
8415					      /*sks_valid*/ 1,
8416					      /*command*/ 1,
8417					      /*field*/ 2,
8418					      /*bit_valid*/ 1,
8419					      /*bit*/ 0);
8420			ctl_done((union ctl_io *)ctsio);
8421			return (CTL_RETVAL_COMPLETE);
8422		}
8423	}
8424
8425	param_len = scsi_4btoul(cdb->length);
8426
8427	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8428		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8429		ctsio->kern_data_len = param_len;
8430		ctsio->kern_total_len = param_len;
8431		ctsio->kern_data_resid = 0;
8432		ctsio->kern_rel_offset = 0;
8433		ctsio->kern_sg_entries = 0;
8434		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8435		ctsio->be_move_done = ctl_config_move_done;
8436		ctl_datamove((union ctl_io *)ctsio);
8437
8438		return (CTL_RETVAL_COMPLETE);
8439	}
8440
8441	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8442
8443	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8444	res_key = scsi_8btou64(param->res_key.key);
8445	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8446
8447	/*
8448	 * Validate the reservation key here except for SPRO_REG_IGNO
8449	 * This must be done for all other service actions
8450	 */
8451	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8452		mtx_lock(&lun->lun_lock);
8453		if ((key = ctl_get_prkey(lun, residx)) != 0) {
8454			if (res_key != key) {
8455				/*
8456				 * The current key passed in doesn't match
8457				 * the one the initiator previously
8458				 * registered.
8459				 */
8460				mtx_unlock(&lun->lun_lock);
8461				free(ctsio->kern_data_ptr, M_CTL);
8462				ctl_set_reservation_conflict(ctsio);
8463				ctl_done((union ctl_io *)ctsio);
8464				return (CTL_RETVAL_COMPLETE);
8465			}
8466		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8467			/*
8468			 * We are not registered
8469			 */
8470			mtx_unlock(&lun->lun_lock);
8471			free(ctsio->kern_data_ptr, M_CTL);
8472			ctl_set_reservation_conflict(ctsio);
8473			ctl_done((union ctl_io *)ctsio);
8474			return (CTL_RETVAL_COMPLETE);
8475		} else if (res_key != 0) {
8476			/*
8477			 * We are not registered and trying to register but
8478			 * the register key isn't zero.
8479			 */
8480			mtx_unlock(&lun->lun_lock);
8481			free(ctsio->kern_data_ptr, M_CTL);
8482			ctl_set_reservation_conflict(ctsio);
8483			ctl_done((union ctl_io *)ctsio);
8484			return (CTL_RETVAL_COMPLETE);
8485		}
8486		mtx_unlock(&lun->lun_lock);
8487	}
8488
8489	switch (cdb->action & SPRO_ACTION_MASK) {
8490	case SPRO_REGISTER:
8491	case SPRO_REG_IGNO: {
8492
8493#if 0
8494		printf("Registration received\n");
8495#endif
8496
8497		/*
8498		 * We don't support any of these options, as we report in
8499		 * the read capabilities request (see
8500		 * ctl_persistent_reserve_in(), above).
8501		 */
8502		if ((param->flags & SPR_SPEC_I_PT)
8503		 || (param->flags & SPR_ALL_TG_PT)
8504		 || (param->flags & SPR_APTPL)) {
8505			int bit_ptr;
8506
8507			if (param->flags & SPR_APTPL)
8508				bit_ptr = 0;
8509			else if (param->flags & SPR_ALL_TG_PT)
8510				bit_ptr = 2;
8511			else /* SPR_SPEC_I_PT */
8512				bit_ptr = 3;
8513
8514			free(ctsio->kern_data_ptr, M_CTL);
8515			ctl_set_invalid_field(ctsio,
8516					      /*sks_valid*/ 1,
8517					      /*command*/ 0,
8518					      /*field*/ 20,
8519					      /*bit_valid*/ 1,
8520					      /*bit*/ bit_ptr);
8521			ctl_done((union ctl_io *)ctsio);
8522			return (CTL_RETVAL_COMPLETE);
8523		}
8524
8525		mtx_lock(&lun->lun_lock);
8526
8527		/*
8528		 * The initiator wants to clear the
8529		 * key/unregister.
8530		 */
8531		if (sa_res_key == 0) {
8532			if ((res_key == 0
8533			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8534			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8535			  && ctl_get_prkey(lun, residx) == 0)) {
8536				mtx_unlock(&lun->lun_lock);
8537				goto done;
8538			}
8539
8540			ctl_clr_prkey(lun, residx);
8541			lun->pr_key_count--;
8542
8543			if (residx == lun->pr_res_idx) {
8544				lun->flags &= ~CTL_LUN_PR_RESERVED;
8545				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8546
8547				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8548				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8549				 && lun->pr_key_count) {
8550					/*
8551					 * If the reservation is a registrants
8552					 * only type we need to generate a UA
8553					 * for other registered inits.  The
8554					 * sense code should be RESERVATIONS
8555					 * RELEASED
8556					 */
8557
8558					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8559						if (ctl_get_prkey(lun, i +
8560						    softc->persis_offset) == 0)
8561							continue;
8562						ctl_est_ua(lun, i,
8563						    CTL_UA_RES_RELEASE);
8564					}
8565				}
8566				lun->res_type = 0;
8567			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8568				if (lun->pr_key_count==0) {
8569					lun->flags &= ~CTL_LUN_PR_RESERVED;
8570					lun->res_type = 0;
8571					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8572				}
8573			}
8574			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8575			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8576			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8577			persis_io.pr.pr_info.residx = residx;
8578			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8579			     &persis_io, sizeof(persis_io), 0 )) >
8580			     CTL_HA_STATUS_SUCCESS) {
8581				printf("CTL:Persis Out error returned from "
8582				       "ctl_ha_msg_send %d\n", isc_retval);
8583			}
8584		} else /* sa_res_key != 0 */ {
8585
8586			/*
8587			 * If we aren't registered currently then increment
8588			 * the key count and set the registered flag.
8589			 */
8590			ctl_alloc_prkey(lun, residx);
8591			if (ctl_get_prkey(lun, residx) == 0)
8592				lun->pr_key_count++;
8593			ctl_set_prkey(lun, residx, sa_res_key);
8594
8595			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8596			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8597			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8598			persis_io.pr.pr_info.residx = residx;
8599			memcpy(persis_io.pr.pr_info.sa_res_key,
8600			       param->serv_act_res_key,
8601			       sizeof(param->serv_act_res_key));
8602			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8603			     &persis_io, sizeof(persis_io), 0)) >
8604			     CTL_HA_STATUS_SUCCESS) {
8605				printf("CTL:Persis Out error returned from "
8606				       "ctl_ha_msg_send %d\n", isc_retval);
8607			}
8608		}
8609		lun->PRGeneration++;
8610		mtx_unlock(&lun->lun_lock);
8611
8612		break;
8613	}
8614	case SPRO_RESERVE:
8615#if 0
8616                printf("Reserve executed type %d\n", type);
8617#endif
8618		mtx_lock(&lun->lun_lock);
8619		if (lun->flags & CTL_LUN_PR_RESERVED) {
8620			/*
8621			 * if this isn't the reservation holder and it's
8622			 * not a "all registrants" type or if the type is
8623			 * different then we have a conflict
8624			 */
8625			if ((lun->pr_res_idx != residx
8626			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8627			 || lun->res_type != type) {
8628				mtx_unlock(&lun->lun_lock);
8629				free(ctsio->kern_data_ptr, M_CTL);
8630				ctl_set_reservation_conflict(ctsio);
8631				ctl_done((union ctl_io *)ctsio);
8632				return (CTL_RETVAL_COMPLETE);
8633			}
8634			mtx_unlock(&lun->lun_lock);
8635		} else /* create a reservation */ {
8636			/*
8637			 * If it's not an "all registrants" type record
8638			 * reservation holder
8639			 */
8640			if (type != SPR_TYPE_WR_EX_AR
8641			 && type != SPR_TYPE_EX_AC_AR)
8642				lun->pr_res_idx = residx; /* Res holder */
8643			else
8644				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8645
8646			lun->flags |= CTL_LUN_PR_RESERVED;
8647			lun->res_type = type;
8648
8649			mtx_unlock(&lun->lun_lock);
8650
8651			/* send msg to other side */
8652			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8653			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8654			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8655			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8656			persis_io.pr.pr_info.res_type = type;
8657			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8658			     &persis_io, sizeof(persis_io), 0)) >
8659			     CTL_HA_STATUS_SUCCESS) {
8660				printf("CTL:Persis Out error returned from "
8661				       "ctl_ha_msg_send %d\n", isc_retval);
8662			}
8663		}
8664		break;
8665
8666	case SPRO_RELEASE:
8667		mtx_lock(&lun->lun_lock);
8668		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8669			/* No reservation exists return good status */
8670			mtx_unlock(&lun->lun_lock);
8671			goto done;
8672		}
8673		/*
8674		 * Is this nexus a reservation holder?
8675		 */
8676		if (lun->pr_res_idx != residx
8677		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8678			/*
8679			 * not a res holder return good status but
8680			 * do nothing
8681			 */
8682			mtx_unlock(&lun->lun_lock);
8683			goto done;
8684		}
8685
8686		if (lun->res_type != type) {
8687			mtx_unlock(&lun->lun_lock);
8688			free(ctsio->kern_data_ptr, M_CTL);
8689			ctl_set_illegal_pr_release(ctsio);
8690			ctl_done((union ctl_io *)ctsio);
8691			return (CTL_RETVAL_COMPLETE);
8692		}
8693
8694		/* okay to release */
8695		lun->flags &= ~CTL_LUN_PR_RESERVED;
8696		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8697		lun->res_type = 0;
8698
8699		/*
8700		 * if this isn't an exclusive access
8701		 * res generate UA for all other
8702		 * registrants.
8703		 */
8704		if (type != SPR_TYPE_EX_AC
8705		 && type != SPR_TYPE_WR_EX) {
8706			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8707				if (i == residx ||
8708				    ctl_get_prkey(lun,
8709				     i + softc->persis_offset) == 0)
8710					continue;
8711				ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8712			}
8713		}
8714		mtx_unlock(&lun->lun_lock);
8715		/* Send msg to other side */
8716		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8717		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8718		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8719		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8720		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8721			printf("CTL:Persis Out error returned from "
8722			       "ctl_ha_msg_send %d\n", isc_retval);
8723		}
8724		break;
8725
8726	case SPRO_CLEAR:
8727		/* send msg to other side */
8728
8729		mtx_lock(&lun->lun_lock);
8730		lun->flags &= ~CTL_LUN_PR_RESERVED;
8731		lun->res_type = 0;
8732		lun->pr_key_count = 0;
8733		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8734
8735		ctl_clr_prkey(lun, residx);
8736		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8737			if (ctl_get_prkey(lun, i) != 0) {
8738				ctl_clr_prkey(lun, i);
8739				ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8740			}
8741		lun->PRGeneration++;
8742		mtx_unlock(&lun->lun_lock);
8743		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8744		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8745		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8746		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8747		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8748			printf("CTL:Persis Out error returned from "
8749			       "ctl_ha_msg_send %d\n", isc_retval);
8750		}
8751		break;
8752
8753	case SPRO_PREEMPT:
8754	case SPRO_PRE_ABO: {
8755		int nretval;
8756
8757		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8758					  residx, ctsio, cdb, param);
8759		if (nretval != 0)
8760			return (CTL_RETVAL_COMPLETE);
8761		break;
8762	}
8763	default:
8764		panic("Invalid PR type %x", cdb->action);
8765	}
8766
8767done:
8768	free(ctsio->kern_data_ptr, M_CTL);
8769	ctl_set_success(ctsio);
8770	ctl_done((union ctl_io *)ctsio);
8771
8772	return (retval);
8773}
8774
8775/*
8776 * This routine is for handling a message from the other SC pertaining to
8777 * persistent reserve out. All the error checking will have been done
8778 * so only perorming the action need be done here to keep the two
8779 * in sync.
8780 */
8781static void
8782ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8783{
8784	struct ctl_lun *lun;
8785	struct ctl_softc *softc;
8786	int i;
8787	uint32_t targ_lun;
8788
8789	softc = control_softc;
8790
8791	targ_lun = msg->hdr.nexus.targ_mapped_lun;
8792	lun = softc->ctl_luns[targ_lun];
8793	mtx_lock(&lun->lun_lock);
8794	switch(msg->pr.pr_info.action) {
8795	case CTL_PR_REG_KEY:
8796		ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8797		if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8798			lun->pr_key_count++;
8799		ctl_set_prkey(lun, msg->pr.pr_info.residx,
8800		    scsi_8btou64(msg->pr.pr_info.sa_res_key));
8801		lun->PRGeneration++;
8802		break;
8803
8804	case CTL_PR_UNREG_KEY:
8805		ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8806		lun->pr_key_count--;
8807
8808		/* XXX Need to see if the reservation has been released */
8809		/* if so do we need to generate UA? */
8810		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8811			lun->flags &= ~CTL_LUN_PR_RESERVED;
8812			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8813
8814			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8815			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8816			 && lun->pr_key_count) {
8817				/*
8818				 * If the reservation is a registrants
8819				 * only type we need to generate a UA
8820				 * for other registered inits.  The
8821				 * sense code should be RESERVATIONS
8822				 * RELEASED
8823				 */
8824
8825				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8826					if (ctl_get_prkey(lun, i +
8827					    softc->persis_offset) == 0)
8828						continue;
8829
8830					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8831				}
8832			}
8833			lun->res_type = 0;
8834		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8835			if (lun->pr_key_count==0) {
8836				lun->flags &= ~CTL_LUN_PR_RESERVED;
8837				lun->res_type = 0;
8838				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8839			}
8840		}
8841		lun->PRGeneration++;
8842		break;
8843
8844	case CTL_PR_RESERVE:
8845		lun->flags |= CTL_LUN_PR_RESERVED;
8846		lun->res_type = msg->pr.pr_info.res_type;
8847		lun->pr_res_idx = msg->pr.pr_info.residx;
8848
8849		break;
8850
8851	case CTL_PR_RELEASE:
8852		/*
8853		 * if this isn't an exclusive access res generate UA for all
8854		 * other registrants.
8855		 */
8856		if (lun->res_type != SPR_TYPE_EX_AC
8857		 && lun->res_type != SPR_TYPE_WR_EX) {
8858			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8859				if (ctl_get_prkey(lun, i + softc->persis_offset) != 0)
8860					ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8861		}
8862
8863		lun->flags &= ~CTL_LUN_PR_RESERVED;
8864		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8865		lun->res_type = 0;
8866		break;
8867
8868	case CTL_PR_PREEMPT:
8869		ctl_pro_preempt_other(lun, msg);
8870		break;
8871	case CTL_PR_CLEAR:
8872		lun->flags &= ~CTL_LUN_PR_RESERVED;
8873		lun->res_type = 0;
8874		lun->pr_key_count = 0;
8875		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8876
8877		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8878			if (ctl_get_prkey(lun, i) == 0)
8879				continue;
8880			ctl_clr_prkey(lun, i);
8881			ctl_est_res_ua(lun, i, CTL_UA_REG_PREEMPT);
8882		}
8883		lun->PRGeneration++;
8884		break;
8885	}
8886
8887	mtx_unlock(&lun->lun_lock);
8888}
8889
8890int
8891ctl_read_write(struct ctl_scsiio *ctsio)
8892{
8893	struct ctl_lun *lun;
8894	struct ctl_lba_len_flags *lbalen;
8895	uint64_t lba;
8896	uint32_t num_blocks;
8897	int flags, retval;
8898	int isread;
8899
8900	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8901
8902	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8903
8904	flags = 0;
8905	retval = CTL_RETVAL_COMPLETE;
8906
8907	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8908	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8909	switch (ctsio->cdb[0]) {
8910	case READ_6:
8911	case WRITE_6: {
8912		struct scsi_rw_6 *cdb;
8913
8914		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8915
8916		lba = scsi_3btoul(cdb->addr);
8917		/* only 5 bits are valid in the most significant address byte */
8918		lba &= 0x1fffff;
8919		num_blocks = cdb->length;
8920		/*
8921		 * This is correct according to SBC-2.
8922		 */
8923		if (num_blocks == 0)
8924			num_blocks = 256;
8925		break;
8926	}
8927	case READ_10:
8928	case WRITE_10: {
8929		struct scsi_rw_10 *cdb;
8930
8931		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8932		if (cdb->byte2 & SRW10_FUA)
8933			flags |= CTL_LLF_FUA;
8934		if (cdb->byte2 & SRW10_DPO)
8935			flags |= CTL_LLF_DPO;
8936		lba = scsi_4btoul(cdb->addr);
8937		num_blocks = scsi_2btoul(cdb->length);
8938		break;
8939	}
8940	case WRITE_VERIFY_10: {
8941		struct scsi_write_verify_10 *cdb;
8942
8943		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8944		flags |= CTL_LLF_FUA;
8945		if (cdb->byte2 & SWV_DPO)
8946			flags |= CTL_LLF_DPO;
8947		lba = scsi_4btoul(cdb->addr);
8948		num_blocks = scsi_2btoul(cdb->length);
8949		break;
8950	}
8951	case READ_12:
8952	case WRITE_12: {
8953		struct scsi_rw_12 *cdb;
8954
8955		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8956		if (cdb->byte2 & SRW12_FUA)
8957			flags |= CTL_LLF_FUA;
8958		if (cdb->byte2 & SRW12_DPO)
8959			flags |= CTL_LLF_DPO;
8960		lba = scsi_4btoul(cdb->addr);
8961		num_blocks = scsi_4btoul(cdb->length);
8962		break;
8963	}
8964	case WRITE_VERIFY_12: {
8965		struct scsi_write_verify_12 *cdb;
8966
8967		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8968		flags |= CTL_LLF_FUA;
8969		if (cdb->byte2 & SWV_DPO)
8970			flags |= CTL_LLF_DPO;
8971		lba = scsi_4btoul(cdb->addr);
8972		num_blocks = scsi_4btoul(cdb->length);
8973		break;
8974	}
8975	case READ_16:
8976	case WRITE_16: {
8977		struct scsi_rw_16 *cdb;
8978
8979		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8980		if (cdb->byte2 & SRW12_FUA)
8981			flags |= CTL_LLF_FUA;
8982		if (cdb->byte2 & SRW12_DPO)
8983			flags |= CTL_LLF_DPO;
8984		lba = scsi_8btou64(cdb->addr);
8985		num_blocks = scsi_4btoul(cdb->length);
8986		break;
8987	}
8988	case WRITE_ATOMIC_16: {
8989		struct scsi_rw_16 *cdb;
8990
8991		if (lun->be_lun->atomicblock == 0) {
8992			ctl_set_invalid_opcode(ctsio);
8993			ctl_done((union ctl_io *)ctsio);
8994			return (CTL_RETVAL_COMPLETE);
8995		}
8996
8997		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8998		if (cdb->byte2 & SRW12_FUA)
8999			flags |= CTL_LLF_FUA;
9000		if (cdb->byte2 & SRW12_DPO)
9001			flags |= CTL_LLF_DPO;
9002		lba = scsi_8btou64(cdb->addr);
9003		num_blocks = scsi_4btoul(cdb->length);
9004		if (num_blocks > lun->be_lun->atomicblock) {
9005			ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
9006			    /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
9007			    /*bit*/ 0);
9008			ctl_done((union ctl_io *)ctsio);
9009			return (CTL_RETVAL_COMPLETE);
9010		}
9011		break;
9012	}
9013	case WRITE_VERIFY_16: {
9014		struct scsi_write_verify_16 *cdb;
9015
9016		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
9017		flags |= CTL_LLF_FUA;
9018		if (cdb->byte2 & SWV_DPO)
9019			flags |= CTL_LLF_DPO;
9020		lba = scsi_8btou64(cdb->addr);
9021		num_blocks = scsi_4btoul(cdb->length);
9022		break;
9023	}
9024	default:
9025		/*
9026		 * We got a command we don't support.  This shouldn't
9027		 * happen, commands should be filtered out above us.
9028		 */
9029		ctl_set_invalid_opcode(ctsio);
9030		ctl_done((union ctl_io *)ctsio);
9031
9032		return (CTL_RETVAL_COMPLETE);
9033		break; /* NOTREACHED */
9034	}
9035
9036	/*
9037	 * The first check is to make sure we're in bounds, the second
9038	 * check is to catch wrap-around problems.  If the lba + num blocks
9039	 * is less than the lba, then we've wrapped around and the block
9040	 * range is invalid anyway.
9041	 */
9042	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9043	 || ((lba + num_blocks) < lba)) {
9044		ctl_set_lba_out_of_range(ctsio);
9045		ctl_done((union ctl_io *)ctsio);
9046		return (CTL_RETVAL_COMPLETE);
9047	}
9048
9049	/*
9050	 * According to SBC-3, a transfer length of 0 is not an error.
9051	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
9052	 * translates to 256 blocks for those commands.
9053	 */
9054	if (num_blocks == 0) {
9055		ctl_set_success(ctsio);
9056		ctl_done((union ctl_io *)ctsio);
9057		return (CTL_RETVAL_COMPLETE);
9058	}
9059
9060	/* Set FUA and/or DPO if caches are disabled. */
9061	if (isread) {
9062		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9063		    SCP_RCD) != 0)
9064			flags |= CTL_LLF_FUA | CTL_LLF_DPO;
9065	} else {
9066		if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9067		    SCP_WCE) == 0)
9068			flags |= CTL_LLF_FUA;
9069	}
9070
9071	lbalen = (struct ctl_lba_len_flags *)
9072	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9073	lbalen->lba = lba;
9074	lbalen->len = num_blocks;
9075	lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
9076
9077	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9078	ctsio->kern_rel_offset = 0;
9079
9080	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
9081
9082	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9083
9084	return (retval);
9085}
9086
9087static int
9088ctl_cnw_cont(union ctl_io *io)
9089{
9090	struct ctl_scsiio *ctsio;
9091	struct ctl_lun *lun;
9092	struct ctl_lba_len_flags *lbalen;
9093	int retval;
9094
9095	ctsio = &io->scsiio;
9096	ctsio->io_hdr.status = CTL_STATUS_NONE;
9097	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
9098	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9099	lbalen = (struct ctl_lba_len_flags *)
9100	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9101	lbalen->flags &= ~CTL_LLF_COMPARE;
9102	lbalen->flags |= CTL_LLF_WRITE;
9103
9104	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
9105	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9106	return (retval);
9107}
9108
9109int
9110ctl_cnw(struct ctl_scsiio *ctsio)
9111{
9112	struct ctl_lun *lun;
9113	struct ctl_lba_len_flags *lbalen;
9114	uint64_t lba;
9115	uint32_t num_blocks;
9116	int flags, retval;
9117
9118	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9119
9120	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
9121
9122	flags = 0;
9123	retval = CTL_RETVAL_COMPLETE;
9124
9125	switch (ctsio->cdb[0]) {
9126	case COMPARE_AND_WRITE: {
9127		struct scsi_compare_and_write *cdb;
9128
9129		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
9130		if (cdb->byte2 & SRW10_FUA)
9131			flags |= CTL_LLF_FUA;
9132		if (cdb->byte2 & SRW10_DPO)
9133			flags |= CTL_LLF_DPO;
9134		lba = scsi_8btou64(cdb->addr);
9135		num_blocks = cdb->length;
9136		break;
9137	}
9138	default:
9139		/*
9140		 * We got a command we don't support.  This shouldn't
9141		 * happen, commands should be filtered out above us.
9142		 */
9143		ctl_set_invalid_opcode(ctsio);
9144		ctl_done((union ctl_io *)ctsio);
9145
9146		return (CTL_RETVAL_COMPLETE);
9147		break; /* NOTREACHED */
9148	}
9149
9150	/*
9151	 * The first check is to make sure we're in bounds, the second
9152	 * check is to catch wrap-around problems.  If the lba + num blocks
9153	 * is less than the lba, then we've wrapped around and the block
9154	 * range is invalid anyway.
9155	 */
9156	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9157	 || ((lba + num_blocks) < lba)) {
9158		ctl_set_lba_out_of_range(ctsio);
9159		ctl_done((union ctl_io *)ctsio);
9160		return (CTL_RETVAL_COMPLETE);
9161	}
9162
9163	/*
9164	 * According to SBC-3, a transfer length of 0 is not an error.
9165	 */
9166	if (num_blocks == 0) {
9167		ctl_set_success(ctsio);
9168		ctl_done((union ctl_io *)ctsio);
9169		return (CTL_RETVAL_COMPLETE);
9170	}
9171
9172	/* Set FUA if write cache is disabled. */
9173	if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
9174	    SCP_WCE) == 0)
9175		flags |= CTL_LLF_FUA;
9176
9177	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9178	ctsio->kern_rel_offset = 0;
9179
9180	/*
9181	 * Set the IO_CONT flag, so that if this I/O gets passed to
9182	 * ctl_data_submit_done(), it'll get passed back to
9183	 * ctl_ctl_cnw_cont() for further processing.
9184	 */
9185	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9186	ctsio->io_cont = ctl_cnw_cont;
9187
9188	lbalen = (struct ctl_lba_len_flags *)
9189	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9190	lbalen->lba = lba;
9191	lbalen->len = num_blocks;
9192	lbalen->flags = CTL_LLF_COMPARE | flags;
9193
9194	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9195	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9196	return (retval);
9197}
9198
9199int
9200ctl_verify(struct ctl_scsiio *ctsio)
9201{
9202	struct ctl_lun *lun;
9203	struct ctl_lba_len_flags *lbalen;
9204	uint64_t lba;
9205	uint32_t num_blocks;
9206	int bytchk, flags;
9207	int retval;
9208
9209	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9210
9211	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9212
9213	bytchk = 0;
9214	flags = CTL_LLF_FUA;
9215	retval = CTL_RETVAL_COMPLETE;
9216
9217	switch (ctsio->cdb[0]) {
9218	case VERIFY_10: {
9219		struct scsi_verify_10 *cdb;
9220
9221		cdb = (struct scsi_verify_10 *)ctsio->cdb;
9222		if (cdb->byte2 & SVFY_BYTCHK)
9223			bytchk = 1;
9224		if (cdb->byte2 & SVFY_DPO)
9225			flags |= CTL_LLF_DPO;
9226		lba = scsi_4btoul(cdb->addr);
9227		num_blocks = scsi_2btoul(cdb->length);
9228		break;
9229	}
9230	case VERIFY_12: {
9231		struct scsi_verify_12 *cdb;
9232
9233		cdb = (struct scsi_verify_12 *)ctsio->cdb;
9234		if (cdb->byte2 & SVFY_BYTCHK)
9235			bytchk = 1;
9236		if (cdb->byte2 & SVFY_DPO)
9237			flags |= CTL_LLF_DPO;
9238		lba = scsi_4btoul(cdb->addr);
9239		num_blocks = scsi_4btoul(cdb->length);
9240		break;
9241	}
9242	case VERIFY_16: {
9243		struct scsi_rw_16 *cdb;
9244
9245		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9246		if (cdb->byte2 & SVFY_BYTCHK)
9247			bytchk = 1;
9248		if (cdb->byte2 & SVFY_DPO)
9249			flags |= CTL_LLF_DPO;
9250		lba = scsi_8btou64(cdb->addr);
9251		num_blocks = scsi_4btoul(cdb->length);
9252		break;
9253	}
9254	default:
9255		/*
9256		 * We got a command we don't support.  This shouldn't
9257		 * happen, commands should be filtered out above us.
9258		 */
9259		ctl_set_invalid_opcode(ctsio);
9260		ctl_done((union ctl_io *)ctsio);
9261		return (CTL_RETVAL_COMPLETE);
9262	}
9263
9264	/*
9265	 * The first check is to make sure we're in bounds, the second
9266	 * check is to catch wrap-around problems.  If the lba + num blocks
9267	 * is less than the lba, then we've wrapped around and the block
9268	 * range is invalid anyway.
9269	 */
9270	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9271	 || ((lba + num_blocks) < lba)) {
9272		ctl_set_lba_out_of_range(ctsio);
9273		ctl_done((union ctl_io *)ctsio);
9274		return (CTL_RETVAL_COMPLETE);
9275	}
9276
9277	/*
9278	 * According to SBC-3, a transfer length of 0 is not an error.
9279	 */
9280	if (num_blocks == 0) {
9281		ctl_set_success(ctsio);
9282		ctl_done((union ctl_io *)ctsio);
9283		return (CTL_RETVAL_COMPLETE);
9284	}
9285
9286	lbalen = (struct ctl_lba_len_flags *)
9287	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9288	lbalen->lba = lba;
9289	lbalen->len = num_blocks;
9290	if (bytchk) {
9291		lbalen->flags = CTL_LLF_COMPARE | flags;
9292		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9293	} else {
9294		lbalen->flags = CTL_LLF_VERIFY | flags;
9295		ctsio->kern_total_len = 0;
9296	}
9297	ctsio->kern_rel_offset = 0;
9298
9299	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9300	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9301	return (retval);
9302}
9303
9304int
9305ctl_report_luns(struct ctl_scsiio *ctsio)
9306{
9307	struct scsi_report_luns *cdb;
9308	struct scsi_report_luns_data *lun_data;
9309	struct ctl_lun *lun, *request_lun;
9310	int num_luns, retval;
9311	uint32_t alloc_len, lun_datalen;
9312	int num_filled, well_known;
9313	uint32_t initidx, targ_lun_id, lun_id;
9314
9315	retval = CTL_RETVAL_COMPLETE;
9316	well_known = 0;
9317
9318	cdb = (struct scsi_report_luns *)ctsio->cdb;
9319
9320	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9321
9322	mtx_lock(&control_softc->ctl_lock);
9323	num_luns = control_softc->num_luns;
9324	mtx_unlock(&control_softc->ctl_lock);
9325
9326	switch (cdb->select_report) {
9327	case RPL_REPORT_DEFAULT:
9328	case RPL_REPORT_ALL:
9329		break;
9330	case RPL_REPORT_WELLKNOWN:
9331		well_known = 1;
9332		num_luns = 0;
9333		break;
9334	default:
9335		ctl_set_invalid_field(ctsio,
9336				      /*sks_valid*/ 1,
9337				      /*command*/ 1,
9338				      /*field*/ 2,
9339				      /*bit_valid*/ 0,
9340				      /*bit*/ 0);
9341		ctl_done((union ctl_io *)ctsio);
9342		return (retval);
9343		break; /* NOTREACHED */
9344	}
9345
9346	alloc_len = scsi_4btoul(cdb->length);
9347	/*
9348	 * The initiator has to allocate at least 16 bytes for this request,
9349	 * so he can at least get the header and the first LUN.  Otherwise
9350	 * we reject the request (per SPC-3 rev 14, section 6.21).
9351	 */
9352	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9353	    sizeof(struct scsi_report_luns_lundata))) {
9354		ctl_set_invalid_field(ctsio,
9355				      /*sks_valid*/ 1,
9356				      /*command*/ 1,
9357				      /*field*/ 6,
9358				      /*bit_valid*/ 0,
9359				      /*bit*/ 0);
9360		ctl_done((union ctl_io *)ctsio);
9361		return (retval);
9362	}
9363
9364	request_lun = (struct ctl_lun *)
9365		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9366
9367	lun_datalen = sizeof(*lun_data) +
9368		(num_luns * sizeof(struct scsi_report_luns_lundata));
9369
9370	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9371	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9372	ctsio->kern_sg_entries = 0;
9373
9374	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9375
9376	mtx_lock(&control_softc->ctl_lock);
9377	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9378		lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id);
9379		if (lun_id >= CTL_MAX_LUNS)
9380			continue;
9381		lun = control_softc->ctl_luns[lun_id];
9382		if (lun == NULL)
9383			continue;
9384
9385		if (targ_lun_id <= 0xff) {
9386			/*
9387			 * Peripheral addressing method, bus number 0.
9388			 */
9389			lun_data->luns[num_filled].lundata[0] =
9390				RPL_LUNDATA_ATYP_PERIPH;
9391			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9392			num_filled++;
9393		} else if (targ_lun_id <= 0x3fff) {
9394			/*
9395			 * Flat addressing method.
9396			 */
9397			lun_data->luns[num_filled].lundata[0] =
9398				RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9399			lun_data->luns[num_filled].lundata[1] =
9400				(targ_lun_id & 0xff);
9401			num_filled++;
9402		} else if (targ_lun_id <= 0xffffff) {
9403			/*
9404			 * Extended flat addressing method.
9405			 */
9406			lun_data->luns[num_filled].lundata[0] =
9407			    RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9408			scsi_ulto3b(targ_lun_id,
9409			    &lun_data->luns[num_filled].lundata[1]);
9410			num_filled++;
9411		} else {
9412			printf("ctl_report_luns: bogus LUN number %jd, "
9413			       "skipping\n", (intmax_t)targ_lun_id);
9414		}
9415		/*
9416		 * According to SPC-3, rev 14 section 6.21:
9417		 *
9418		 * "The execution of a REPORT LUNS command to any valid and
9419		 * installed logical unit shall clear the REPORTED LUNS DATA
9420		 * HAS CHANGED unit attention condition for all logical
9421		 * units of that target with respect to the requesting
9422		 * initiator. A valid and installed logical unit is one
9423		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9424		 * INQUIRY data (see 6.4.2)."
9425		 *
9426		 * If request_lun is NULL, the LUN this report luns command
9427		 * was issued to is either disabled or doesn't exist. In that
9428		 * case, we shouldn't clear any pending lun change unit
9429		 * attention.
9430		 */
9431		if (request_lun != NULL) {
9432			mtx_lock(&lun->lun_lock);
9433			ctl_clr_ua(lun, initidx, CTL_UA_RES_RELEASE);
9434			mtx_unlock(&lun->lun_lock);
9435		}
9436	}
9437	mtx_unlock(&control_softc->ctl_lock);
9438
9439	/*
9440	 * It's quite possible that we've returned fewer LUNs than we allocated
9441	 * space for.  Trim it.
9442	 */
9443	lun_datalen = sizeof(*lun_data) +
9444		(num_filled * sizeof(struct scsi_report_luns_lundata));
9445
9446	if (lun_datalen < alloc_len) {
9447		ctsio->residual = alloc_len - lun_datalen;
9448		ctsio->kern_data_len = lun_datalen;
9449		ctsio->kern_total_len = lun_datalen;
9450	} else {
9451		ctsio->residual = 0;
9452		ctsio->kern_data_len = alloc_len;
9453		ctsio->kern_total_len = alloc_len;
9454	}
9455	ctsio->kern_data_resid = 0;
9456	ctsio->kern_rel_offset = 0;
9457	ctsio->kern_sg_entries = 0;
9458
9459	/*
9460	 * We set this to the actual data length, regardless of how much
9461	 * space we actually have to return results.  If the user looks at
9462	 * this value, he'll know whether or not he allocated enough space
9463	 * and reissue the command if necessary.  We don't support well
9464	 * known logical units, so if the user asks for that, return none.
9465	 */
9466	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9467
9468	/*
9469	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9470	 * this request.
9471	 */
9472	ctl_set_success(ctsio);
9473	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9474	ctsio->be_move_done = ctl_config_move_done;
9475	ctl_datamove((union ctl_io *)ctsio);
9476	return (retval);
9477}
9478
9479int
9480ctl_request_sense(struct ctl_scsiio *ctsio)
9481{
9482	struct scsi_request_sense *cdb;
9483	struct scsi_sense_data *sense_ptr;
9484	struct ctl_lun *lun;
9485	uint32_t initidx;
9486	int have_error;
9487	scsi_sense_data_type sense_format;
9488	ctl_ua_type ua_type;
9489
9490	cdb = (struct scsi_request_sense *)ctsio->cdb;
9491
9492	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9493
9494	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9495
9496	/*
9497	 * Determine which sense format the user wants.
9498	 */
9499	if (cdb->byte2 & SRS_DESC)
9500		sense_format = SSD_TYPE_DESC;
9501	else
9502		sense_format = SSD_TYPE_FIXED;
9503
9504	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9505	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9506	ctsio->kern_sg_entries = 0;
9507
9508	/*
9509	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9510	 * larger than the largest allowed value for the length field in the
9511	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9512	 */
9513	ctsio->residual = 0;
9514	ctsio->kern_data_len = cdb->length;
9515	ctsio->kern_total_len = cdb->length;
9516
9517	ctsio->kern_data_resid = 0;
9518	ctsio->kern_rel_offset = 0;
9519	ctsio->kern_sg_entries = 0;
9520
9521	/*
9522	 * If we don't have a LUN, we don't have any pending sense.
9523	 */
9524	if (lun == NULL)
9525		goto no_sense;
9526
9527	have_error = 0;
9528	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9529	/*
9530	 * Check for pending sense, and then for pending unit attentions.
9531	 * Pending sense gets returned first, then pending unit attentions.
9532	 */
9533	mtx_lock(&lun->lun_lock);
9534#ifdef CTL_WITH_CA
9535	if (ctl_is_set(lun->have_ca, initidx)) {
9536		scsi_sense_data_type stored_format;
9537
9538		/*
9539		 * Check to see which sense format was used for the stored
9540		 * sense data.
9541		 */
9542		stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9543
9544		/*
9545		 * If the user requested a different sense format than the
9546		 * one we stored, then we need to convert it to the other
9547		 * format.  If we're going from descriptor to fixed format
9548		 * sense data, we may lose things in translation, depending
9549		 * on what options were used.
9550		 *
9551		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9552		 * for some reason we'll just copy it out as-is.
9553		 */
9554		if ((stored_format == SSD_TYPE_FIXED)
9555		 && (sense_format == SSD_TYPE_DESC))
9556			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9557			    &lun->pending_sense[initidx],
9558			    (struct scsi_sense_data_desc *)sense_ptr);
9559		else if ((stored_format == SSD_TYPE_DESC)
9560		      && (sense_format == SSD_TYPE_FIXED))
9561			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9562			    &lun->pending_sense[initidx],
9563			    (struct scsi_sense_data_fixed *)sense_ptr);
9564		else
9565			memcpy(sense_ptr, &lun->pending_sense[initidx],
9566			       ctl_min(sizeof(*sense_ptr),
9567			       sizeof(lun->pending_sense[initidx])));
9568
9569		ctl_clear_mask(lun->have_ca, initidx);
9570		have_error = 1;
9571	} else
9572#endif
9573	{
9574		ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9575		if (ua_type != CTL_UA_NONE)
9576			have_error = 1;
9577	}
9578	mtx_unlock(&lun->lun_lock);
9579
9580	/*
9581	 * We already have a pending error, return it.
9582	 */
9583	if (have_error != 0) {
9584		/*
9585		 * We report the SCSI status as OK, since the status of the
9586		 * request sense command itself is OK.
9587		 * We report 0 for the sense length, because we aren't doing
9588		 * autosense in this case.  We're reporting sense as
9589		 * parameter data.
9590		 */
9591		ctl_set_success(ctsio);
9592		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9593		ctsio->be_move_done = ctl_config_move_done;
9594		ctl_datamove((union ctl_io *)ctsio);
9595		return (CTL_RETVAL_COMPLETE);
9596	}
9597
9598no_sense:
9599
9600	/*
9601	 * No sense information to report, so we report that everything is
9602	 * okay.
9603	 */
9604	ctl_set_sense_data(sense_ptr,
9605			   lun,
9606			   sense_format,
9607			   /*current_error*/ 1,
9608			   /*sense_key*/ SSD_KEY_NO_SENSE,
9609			   /*asc*/ 0x00,
9610			   /*ascq*/ 0x00,
9611			   SSD_ELEM_NONE);
9612
9613	/*
9614	 * We report 0 for the sense length, because we aren't doing
9615	 * autosense in this case.  We're reporting sense as parameter data.
9616	 */
9617	ctl_set_success(ctsio);
9618	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9619	ctsio->be_move_done = ctl_config_move_done;
9620	ctl_datamove((union ctl_io *)ctsio);
9621	return (CTL_RETVAL_COMPLETE);
9622}
9623
9624int
9625ctl_tur(struct ctl_scsiio *ctsio)
9626{
9627
9628	CTL_DEBUG_PRINT(("ctl_tur\n"));
9629
9630	ctl_set_success(ctsio);
9631	ctl_done((union ctl_io *)ctsio);
9632
9633	return (CTL_RETVAL_COMPLETE);
9634}
9635
9636#ifdef notyet
9637static int
9638ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9639{
9640
9641}
9642#endif
9643
9644static int
9645ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9646{
9647	struct scsi_vpd_supported_pages *pages;
9648	int sup_page_size;
9649	struct ctl_lun *lun;
9650
9651	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9652
9653	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9654	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9655	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9656	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9657	ctsio->kern_sg_entries = 0;
9658
9659	if (sup_page_size < alloc_len) {
9660		ctsio->residual = alloc_len - sup_page_size;
9661		ctsio->kern_data_len = sup_page_size;
9662		ctsio->kern_total_len = sup_page_size;
9663	} else {
9664		ctsio->residual = 0;
9665		ctsio->kern_data_len = alloc_len;
9666		ctsio->kern_total_len = alloc_len;
9667	}
9668	ctsio->kern_data_resid = 0;
9669	ctsio->kern_rel_offset = 0;
9670	ctsio->kern_sg_entries = 0;
9671
9672	/*
9673	 * The control device is always connected.  The disk device, on the
9674	 * other hand, may not be online all the time.  Need to change this
9675	 * to figure out whether the disk device is actually online or not.
9676	 */
9677	if (lun != NULL)
9678		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9679				lun->be_lun->lun_type;
9680	else
9681		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9682
9683	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9684	/* Supported VPD pages */
9685	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9686	/* Serial Number */
9687	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9688	/* Device Identification */
9689	pages->page_list[2] = SVPD_DEVICE_ID;
9690	/* Extended INQUIRY Data */
9691	pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA;
9692	/* Mode Page Policy */
9693	pages->page_list[4] = SVPD_MODE_PAGE_POLICY;
9694	/* SCSI Ports */
9695	pages->page_list[5] = SVPD_SCSI_PORTS;
9696	/* Third-party Copy */
9697	pages->page_list[6] = SVPD_SCSI_TPC;
9698	/* Block limits */
9699	pages->page_list[7] = SVPD_BLOCK_LIMITS;
9700	/* Block Device Characteristics */
9701	pages->page_list[8] = SVPD_BDC;
9702	/* Logical Block Provisioning */
9703	pages->page_list[9] = SVPD_LBP;
9704
9705	ctl_set_success(ctsio);
9706	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9707	ctsio->be_move_done = ctl_config_move_done;
9708	ctl_datamove((union ctl_io *)ctsio);
9709	return (CTL_RETVAL_COMPLETE);
9710}
9711
9712static int
9713ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9714{
9715	struct scsi_vpd_unit_serial_number *sn_ptr;
9716	struct ctl_lun *lun;
9717	int data_len;
9718
9719	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9720
9721	data_len = 4 + CTL_SN_LEN;
9722	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9723	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9724	if (data_len < alloc_len) {
9725		ctsio->residual = alloc_len - data_len;
9726		ctsio->kern_data_len = data_len;
9727		ctsio->kern_total_len = data_len;
9728	} else {
9729		ctsio->residual = 0;
9730		ctsio->kern_data_len = alloc_len;
9731		ctsio->kern_total_len = alloc_len;
9732	}
9733	ctsio->kern_data_resid = 0;
9734	ctsio->kern_rel_offset = 0;
9735	ctsio->kern_sg_entries = 0;
9736
9737	/*
9738	 * The control device is always connected.  The disk device, on the
9739	 * other hand, may not be online all the time.  Need to change this
9740	 * to figure out whether the disk device is actually online or not.
9741	 */
9742	if (lun != NULL)
9743		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9744				  lun->be_lun->lun_type;
9745	else
9746		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9747
9748	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9749	sn_ptr->length = CTL_SN_LEN;
9750	/*
9751	 * If we don't have a LUN, we just leave the serial number as
9752	 * all spaces.
9753	 */
9754	if (lun != NULL) {
9755		strncpy((char *)sn_ptr->serial_num,
9756			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9757	} else
9758		memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9759
9760	ctl_set_success(ctsio);
9761	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9762	ctsio->be_move_done = ctl_config_move_done;
9763	ctl_datamove((union ctl_io *)ctsio);
9764	return (CTL_RETVAL_COMPLETE);
9765}
9766
9767
9768static int
9769ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9770{
9771	struct scsi_vpd_extended_inquiry_data *eid_ptr;
9772	struct ctl_lun *lun;
9773	int data_len;
9774
9775	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9776
9777	data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9778	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9779	eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9780	ctsio->kern_sg_entries = 0;
9781
9782	if (data_len < alloc_len) {
9783		ctsio->residual = alloc_len - data_len;
9784		ctsio->kern_data_len = data_len;
9785		ctsio->kern_total_len = data_len;
9786	} else {
9787		ctsio->residual = 0;
9788		ctsio->kern_data_len = alloc_len;
9789		ctsio->kern_total_len = alloc_len;
9790	}
9791	ctsio->kern_data_resid = 0;
9792	ctsio->kern_rel_offset = 0;
9793	ctsio->kern_sg_entries = 0;
9794
9795	/*
9796	 * The control device is always connected.  The disk device, on the
9797	 * other hand, may not be online all the time.
9798	 */
9799	if (lun != NULL)
9800		eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9801				     lun->be_lun->lun_type;
9802	else
9803		eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9804	eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9805	eid_ptr->page_length = data_len - 4;
9806	eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9807	eid_ptr->flags3 = SVPD_EID_V_SUP;
9808
9809	ctl_set_success(ctsio);
9810	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9811	ctsio->be_move_done = ctl_config_move_done;
9812	ctl_datamove((union ctl_io *)ctsio);
9813	return (CTL_RETVAL_COMPLETE);
9814}
9815
9816static int
9817ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9818{
9819	struct scsi_vpd_mode_page_policy *mpp_ptr;
9820	struct ctl_lun *lun;
9821	int data_len;
9822
9823	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9824
9825	data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9826	    sizeof(struct scsi_vpd_mode_page_policy_descr);
9827
9828	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9829	mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9830	ctsio->kern_sg_entries = 0;
9831
9832	if (data_len < alloc_len) {
9833		ctsio->residual = alloc_len - data_len;
9834		ctsio->kern_data_len = data_len;
9835		ctsio->kern_total_len = data_len;
9836	} else {
9837		ctsio->residual = 0;
9838		ctsio->kern_data_len = alloc_len;
9839		ctsio->kern_total_len = alloc_len;
9840	}
9841	ctsio->kern_data_resid = 0;
9842	ctsio->kern_rel_offset = 0;
9843	ctsio->kern_sg_entries = 0;
9844
9845	/*
9846	 * The control device is always connected.  The disk device, on the
9847	 * other hand, may not be online all the time.
9848	 */
9849	if (lun != NULL)
9850		mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9851				     lun->be_lun->lun_type;
9852	else
9853		mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9854	mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9855	scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9856	mpp_ptr->descr[0].page_code = 0x3f;
9857	mpp_ptr->descr[0].subpage_code = 0xff;
9858	mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9859
9860	ctl_set_success(ctsio);
9861	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9862	ctsio->be_move_done = ctl_config_move_done;
9863	ctl_datamove((union ctl_io *)ctsio);
9864	return (CTL_RETVAL_COMPLETE);
9865}
9866
9867static int
9868ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9869{
9870	struct scsi_vpd_device_id *devid_ptr;
9871	struct scsi_vpd_id_descriptor *desc;
9872	struct ctl_softc *ctl_softc;
9873	struct ctl_lun *lun;
9874	struct ctl_port *port;
9875	int data_len;
9876	uint8_t proto;
9877
9878	ctl_softc = control_softc;
9879
9880	port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9881	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9882
9883	data_len = sizeof(struct scsi_vpd_device_id) +
9884	    sizeof(struct scsi_vpd_id_descriptor) +
9885		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9886	    sizeof(struct scsi_vpd_id_descriptor) +
9887		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9888	if (lun && lun->lun_devid)
9889		data_len += lun->lun_devid->len;
9890	if (port->port_devid)
9891		data_len += port->port_devid->len;
9892	if (port->target_devid)
9893		data_len += port->target_devid->len;
9894
9895	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9896	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9897	ctsio->kern_sg_entries = 0;
9898
9899	if (data_len < alloc_len) {
9900		ctsio->residual = alloc_len - data_len;
9901		ctsio->kern_data_len = data_len;
9902		ctsio->kern_total_len = data_len;
9903	} else {
9904		ctsio->residual = 0;
9905		ctsio->kern_data_len = alloc_len;
9906		ctsio->kern_total_len = alloc_len;
9907	}
9908	ctsio->kern_data_resid = 0;
9909	ctsio->kern_rel_offset = 0;
9910	ctsio->kern_sg_entries = 0;
9911
9912	/*
9913	 * The control device is always connected.  The disk device, on the
9914	 * other hand, may not be online all the time.
9915	 */
9916	if (lun != NULL)
9917		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9918				     lun->be_lun->lun_type;
9919	else
9920		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9921	devid_ptr->page_code = SVPD_DEVICE_ID;
9922	scsi_ulto2b(data_len - 4, devid_ptr->length);
9923
9924	if (port->port_type == CTL_PORT_FC)
9925		proto = SCSI_PROTO_FC << 4;
9926	else if (port->port_type == CTL_PORT_ISCSI)
9927		proto = SCSI_PROTO_ISCSI << 4;
9928	else
9929		proto = SCSI_PROTO_SPI << 4;
9930	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9931
9932	/*
9933	 * We're using a LUN association here.  i.e., this device ID is a
9934	 * per-LUN identifier.
9935	 */
9936	if (lun && lun->lun_devid) {
9937		memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9938		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9939		    lun->lun_devid->len);
9940	}
9941
9942	/*
9943	 * This is for the WWPN which is a port association.
9944	 */
9945	if (port->port_devid) {
9946		memcpy(desc, port->port_devid->data, port->port_devid->len);
9947		desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9948		    port->port_devid->len);
9949	}
9950
9951	/*
9952	 * This is for the Relative Target Port(type 4h) identifier
9953	 */
9954	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9955	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9956	    SVPD_ID_TYPE_RELTARG;
9957	desc->length = 4;
9958	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9959	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9960	    sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9961
9962	/*
9963	 * This is for the Target Port Group(type 5h) identifier
9964	 */
9965	desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9966	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9967	    SVPD_ID_TYPE_TPORTGRP;
9968	desc->length = 4;
9969	scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1,
9970	    &desc->identifier[2]);
9971	desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9972	    sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9973
9974	/*
9975	 * This is for the Target identifier
9976	 */
9977	if (port->target_devid) {
9978		memcpy(desc, port->target_devid->data, port->target_devid->len);
9979	}
9980
9981	ctl_set_success(ctsio);
9982	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9983	ctsio->be_move_done = ctl_config_move_done;
9984	ctl_datamove((union ctl_io *)ctsio);
9985	return (CTL_RETVAL_COMPLETE);
9986}
9987
9988static int
9989ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9990{
9991	struct ctl_softc *softc = control_softc;
9992	struct scsi_vpd_scsi_ports *sp;
9993	struct scsi_vpd_port_designation *pd;
9994	struct scsi_vpd_port_designation_cont *pdc;
9995	struct ctl_lun *lun;
9996	struct ctl_port *port;
9997	int data_len, num_target_ports, iid_len, id_len, g, pg, p;
9998	int num_target_port_groups;
9999
10000	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10001
10002	if (softc->is_single)
10003		num_target_port_groups = 1;
10004	else
10005		num_target_port_groups = NUM_TARGET_PORT_GROUPS;
10006	num_target_ports = 0;
10007	iid_len = 0;
10008	id_len = 0;
10009	mtx_lock(&softc->ctl_lock);
10010	STAILQ_FOREACH(port, &softc->port_list, links) {
10011		if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10012			continue;
10013		if (lun != NULL &&
10014		    ctl_map_lun_back(port->targ_port, lun->lun) >=
10015		    CTL_MAX_LUNS)
10016			continue;
10017		num_target_ports++;
10018		if (port->init_devid)
10019			iid_len += port->init_devid->len;
10020		if (port->port_devid)
10021			id_len += port->port_devid->len;
10022	}
10023	mtx_unlock(&softc->ctl_lock);
10024
10025	data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_port_groups *
10026	    num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
10027	     sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
10028	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10029	sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
10030	ctsio->kern_sg_entries = 0;
10031
10032	if (data_len < alloc_len) {
10033		ctsio->residual = alloc_len - data_len;
10034		ctsio->kern_data_len = data_len;
10035		ctsio->kern_total_len = data_len;
10036	} else {
10037		ctsio->residual = 0;
10038		ctsio->kern_data_len = alloc_len;
10039		ctsio->kern_total_len = alloc_len;
10040	}
10041	ctsio->kern_data_resid = 0;
10042	ctsio->kern_rel_offset = 0;
10043	ctsio->kern_sg_entries = 0;
10044
10045	/*
10046	 * The control device is always connected.  The disk device, on the
10047	 * other hand, may not be online all the time.  Need to change this
10048	 * to figure out whether the disk device is actually online or not.
10049	 */
10050	if (lun != NULL)
10051		sp->device = (SID_QUAL_LU_CONNECTED << 5) |
10052				  lun->be_lun->lun_type;
10053	else
10054		sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10055
10056	sp->page_code = SVPD_SCSI_PORTS;
10057	scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
10058	    sp->page_length);
10059	pd = &sp->design[0];
10060
10061	mtx_lock(&softc->ctl_lock);
10062	pg = softc->port_offset / CTL_MAX_PORTS;
10063	for (g = 0; g < num_target_port_groups; g++) {
10064		STAILQ_FOREACH(port, &softc->port_list, links) {
10065			if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
10066				continue;
10067			if (lun != NULL &&
10068			    ctl_map_lun_back(port->targ_port, lun->lun) >=
10069			    CTL_MAX_LUNS)
10070				continue;
10071			p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS;
10072			scsi_ulto2b(p, pd->relative_port_id);
10073			if (port->init_devid && g == pg) {
10074				iid_len = port->init_devid->len;
10075				memcpy(pd->initiator_transportid,
10076				    port->init_devid->data, port->init_devid->len);
10077			} else
10078				iid_len = 0;
10079			scsi_ulto2b(iid_len, pd->initiator_transportid_length);
10080			pdc = (struct scsi_vpd_port_designation_cont *)
10081			    (&pd->initiator_transportid[iid_len]);
10082			if (port->port_devid && g == pg) {
10083				id_len = port->port_devid->len;
10084				memcpy(pdc->target_port_descriptors,
10085				    port->port_devid->data, port->port_devid->len);
10086			} else
10087				id_len = 0;
10088			scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
10089			pd = (struct scsi_vpd_port_designation *)
10090			    ((uint8_t *)pdc->target_port_descriptors + id_len);
10091		}
10092	}
10093	mtx_unlock(&softc->ctl_lock);
10094
10095	ctl_set_success(ctsio);
10096	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10097	ctsio->be_move_done = ctl_config_move_done;
10098	ctl_datamove((union ctl_io *)ctsio);
10099	return (CTL_RETVAL_COMPLETE);
10100}
10101
10102static int
10103ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
10104{
10105	struct scsi_vpd_block_limits *bl_ptr;
10106	struct ctl_lun *lun;
10107	int bs;
10108
10109	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10110
10111	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
10112	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
10113	ctsio->kern_sg_entries = 0;
10114
10115	if (sizeof(*bl_ptr) < alloc_len) {
10116		ctsio->residual = alloc_len - sizeof(*bl_ptr);
10117		ctsio->kern_data_len = sizeof(*bl_ptr);
10118		ctsio->kern_total_len = sizeof(*bl_ptr);
10119	} else {
10120		ctsio->residual = 0;
10121		ctsio->kern_data_len = alloc_len;
10122		ctsio->kern_total_len = alloc_len;
10123	}
10124	ctsio->kern_data_resid = 0;
10125	ctsio->kern_rel_offset = 0;
10126	ctsio->kern_sg_entries = 0;
10127
10128	/*
10129	 * The control device is always connected.  The disk device, on the
10130	 * other hand, may not be online all the time.  Need to change this
10131	 * to figure out whether the disk device is actually online or not.
10132	 */
10133	if (lun != NULL)
10134		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10135				  lun->be_lun->lun_type;
10136	else
10137		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10138
10139	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
10140	scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
10141	bl_ptr->max_cmp_write_len = 0xff;
10142	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
10143	if (lun != NULL) {
10144		bs = lun->be_lun->blocksize;
10145		scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
10146		if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10147			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
10148			scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
10149			if (lun->be_lun->ublockexp != 0) {
10150				scsi_ulto4b((1 << lun->be_lun->ublockexp),
10151				    bl_ptr->opt_unmap_grain);
10152				scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
10153				    bl_ptr->unmap_grain_align);
10154			}
10155		}
10156		scsi_ulto4b(lun->be_lun->atomicblock,
10157		    bl_ptr->max_atomic_transfer_length);
10158		scsi_ulto4b(0, bl_ptr->atomic_alignment);
10159		scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
10160	}
10161	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
10162
10163	ctl_set_success(ctsio);
10164	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10165	ctsio->be_move_done = ctl_config_move_done;
10166	ctl_datamove((union ctl_io *)ctsio);
10167	return (CTL_RETVAL_COMPLETE);
10168}
10169
10170static int
10171ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10172{
10173	struct scsi_vpd_block_device_characteristics *bdc_ptr;
10174	struct ctl_lun *lun;
10175	const char *value;
10176	u_int i;
10177
10178	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10179
10180	ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10181	bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10182	ctsio->kern_sg_entries = 0;
10183
10184	if (sizeof(*bdc_ptr) < alloc_len) {
10185		ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10186		ctsio->kern_data_len = sizeof(*bdc_ptr);
10187		ctsio->kern_total_len = sizeof(*bdc_ptr);
10188	} else {
10189		ctsio->residual = 0;
10190		ctsio->kern_data_len = alloc_len;
10191		ctsio->kern_total_len = alloc_len;
10192	}
10193	ctsio->kern_data_resid = 0;
10194	ctsio->kern_rel_offset = 0;
10195	ctsio->kern_sg_entries = 0;
10196
10197	/*
10198	 * The control device is always connected.  The disk device, on the
10199	 * other hand, may not be online all the time.  Need to change this
10200	 * to figure out whether the disk device is actually online or not.
10201	 */
10202	if (lun != NULL)
10203		bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10204				  lun->be_lun->lun_type;
10205	else
10206		bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10207	bdc_ptr->page_code = SVPD_BDC;
10208	scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10209	if (lun != NULL &&
10210	    (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10211		i = strtol(value, NULL, 0);
10212	else
10213		i = CTL_DEFAULT_ROTATION_RATE;
10214	scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10215	if (lun != NULL &&
10216	    (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10217		i = strtol(value, NULL, 0);
10218	else
10219		i = 0;
10220	bdc_ptr->wab_wac_ff = (i & 0x0f);
10221	bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10222
10223	ctl_set_success(ctsio);
10224	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10225	ctsio->be_move_done = ctl_config_move_done;
10226	ctl_datamove((union ctl_io *)ctsio);
10227	return (CTL_RETVAL_COMPLETE);
10228}
10229
10230static int
10231ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10232{
10233	struct scsi_vpd_logical_block_prov *lbp_ptr;
10234	struct ctl_lun *lun;
10235
10236	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10237
10238	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10239	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10240	ctsio->kern_sg_entries = 0;
10241
10242	if (sizeof(*lbp_ptr) < alloc_len) {
10243		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10244		ctsio->kern_data_len = sizeof(*lbp_ptr);
10245		ctsio->kern_total_len = sizeof(*lbp_ptr);
10246	} else {
10247		ctsio->residual = 0;
10248		ctsio->kern_data_len = alloc_len;
10249		ctsio->kern_total_len = alloc_len;
10250	}
10251	ctsio->kern_data_resid = 0;
10252	ctsio->kern_rel_offset = 0;
10253	ctsio->kern_sg_entries = 0;
10254
10255	/*
10256	 * The control device is always connected.  The disk device, on the
10257	 * other hand, may not be online all the time.  Need to change this
10258	 * to figure out whether the disk device is actually online or not.
10259	 */
10260	if (lun != NULL)
10261		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10262				  lun->be_lun->lun_type;
10263	else
10264		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10265
10266	lbp_ptr->page_code = SVPD_LBP;
10267	scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10268	lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10269	if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10270		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10271		    SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10272		lbp_ptr->prov_type = SVPD_LBP_THIN;
10273	}
10274
10275	ctl_set_success(ctsio);
10276	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10277	ctsio->be_move_done = ctl_config_move_done;
10278	ctl_datamove((union ctl_io *)ctsio);
10279	return (CTL_RETVAL_COMPLETE);
10280}
10281
10282static int
10283ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10284{
10285	struct scsi_inquiry *cdb;
10286	int alloc_len, retval;
10287
10288	cdb = (struct scsi_inquiry *)ctsio->cdb;
10289
10290	retval = CTL_RETVAL_COMPLETE;
10291
10292	alloc_len = scsi_2btoul(cdb->length);
10293
10294	switch (cdb->page_code) {
10295	case SVPD_SUPPORTED_PAGES:
10296		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10297		break;
10298	case SVPD_UNIT_SERIAL_NUMBER:
10299		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10300		break;
10301	case SVPD_DEVICE_ID:
10302		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10303		break;
10304	case SVPD_EXTENDED_INQUIRY_DATA:
10305		retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10306		break;
10307	case SVPD_MODE_PAGE_POLICY:
10308		retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10309		break;
10310	case SVPD_SCSI_PORTS:
10311		retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10312		break;
10313	case SVPD_SCSI_TPC:
10314		retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10315		break;
10316	case SVPD_BLOCK_LIMITS:
10317		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10318		break;
10319	case SVPD_BDC:
10320		retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10321		break;
10322	case SVPD_LBP:
10323		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10324		break;
10325	default:
10326		ctl_set_invalid_field(ctsio,
10327				      /*sks_valid*/ 1,
10328				      /*command*/ 1,
10329				      /*field*/ 2,
10330				      /*bit_valid*/ 0,
10331				      /*bit*/ 0);
10332		ctl_done((union ctl_io *)ctsio);
10333		retval = CTL_RETVAL_COMPLETE;
10334		break;
10335	}
10336
10337	return (retval);
10338}
10339
10340static int
10341ctl_inquiry_std(struct ctl_scsiio *ctsio)
10342{
10343	struct scsi_inquiry_data *inq_ptr;
10344	struct scsi_inquiry *cdb;
10345	struct ctl_softc *ctl_softc;
10346	struct ctl_lun *lun;
10347	char *val;
10348	uint32_t alloc_len, data_len;
10349	ctl_port_type port_type;
10350
10351	ctl_softc = control_softc;
10352
10353	/*
10354	 * Figure out whether we're talking to a Fibre Channel port or not.
10355	 * We treat the ioctl front end, and any SCSI adapters, as packetized
10356	 * SCSI front ends.
10357	 */
10358	port_type = ctl_softc->ctl_ports[
10359	    ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type;
10360	if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10361		port_type = CTL_PORT_SCSI;
10362
10363	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10364	cdb = (struct scsi_inquiry *)ctsio->cdb;
10365	alloc_len = scsi_2btoul(cdb->length);
10366
10367	/*
10368	 * We malloc the full inquiry data size here and fill it
10369	 * in.  If the user only asks for less, we'll give him
10370	 * that much.
10371	 */
10372	data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10373	ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10374	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10375	ctsio->kern_sg_entries = 0;
10376	ctsio->kern_data_resid = 0;
10377	ctsio->kern_rel_offset = 0;
10378
10379	if (data_len < alloc_len) {
10380		ctsio->residual = alloc_len - data_len;
10381		ctsio->kern_data_len = data_len;
10382		ctsio->kern_total_len = data_len;
10383	} else {
10384		ctsio->residual = 0;
10385		ctsio->kern_data_len = alloc_len;
10386		ctsio->kern_total_len = alloc_len;
10387	}
10388
10389	/*
10390	 * If we have a LUN configured, report it as connected.  Otherwise,
10391	 * report that it is offline or no device is supported, depending
10392	 * on the value of inquiry_pq_no_lun.
10393	 *
10394	 * According to the spec (SPC-4 r34), the peripheral qualifier
10395	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
10396	 *
10397	 * "A peripheral device having the specified peripheral device type
10398	 * is not connected to this logical unit. However, the device
10399	 * server is capable of supporting the specified peripheral device
10400	 * type on this logical unit."
10401	 *
10402	 * According to the same spec, the peripheral qualifier
10403	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
10404	 *
10405	 * "The device server is not capable of supporting a peripheral
10406	 * device on this logical unit. For this peripheral qualifier the
10407	 * peripheral device type shall be set to 1Fh. All other peripheral
10408	 * device type values are reserved for this peripheral qualifier."
10409	 *
10410	 * Given the text, it would seem that we probably want to report that
10411	 * the LUN is offline here.  There is no LUN connected, but we can
10412	 * support a LUN at the given LUN number.
10413	 *
10414	 * In the real world, though, it sounds like things are a little
10415	 * different:
10416	 *
10417	 * - Linux, when presented with a LUN with the offline peripheral
10418	 *   qualifier, will create an sg driver instance for it.  So when
10419	 *   you attach it to CTL, you wind up with a ton of sg driver
10420	 *   instances.  (One for every LUN that Linux bothered to probe.)
10421	 *   Linux does this despite the fact that it issues a REPORT LUNs
10422	 *   to LUN 0 to get the inventory of supported LUNs.
10423	 *
10424	 * - There is other anecdotal evidence (from Emulex folks) about
10425	 *   arrays that use the offline peripheral qualifier for LUNs that
10426	 *   are on the "passive" path in an active/passive array.
10427	 *
10428	 * So the solution is provide a hopefully reasonable default
10429	 * (return bad/no LUN) and allow the user to change the behavior
10430	 * with a tunable/sysctl variable.
10431	 */
10432	if (lun != NULL)
10433		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10434				  lun->be_lun->lun_type;
10435	else if (ctl_softc->inquiry_pq_no_lun == 0)
10436		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10437	else
10438		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10439
10440	/* RMB in byte 2 is 0 */
10441	inq_ptr->version = SCSI_REV_SPC4;
10442
10443	/*
10444	 * According to SAM-3, even if a device only supports a single
10445	 * level of LUN addressing, it should still set the HISUP bit:
10446	 *
10447	 * 4.9.1 Logical unit numbers overview
10448	 *
10449	 * All logical unit number formats described in this standard are
10450	 * hierarchical in structure even when only a single level in that
10451	 * hierarchy is used. The HISUP bit shall be set to one in the
10452	 * standard INQUIRY data (see SPC-2) when any logical unit number
10453	 * format described in this standard is used.  Non-hierarchical
10454	 * formats are outside the scope of this standard.
10455	 *
10456	 * Therefore we set the HiSup bit here.
10457	 *
10458	 * The reponse format is 2, per SPC-3.
10459	 */
10460	inq_ptr->response_format = SID_HiSup | 2;
10461
10462	inq_ptr->additional_length = data_len -
10463	    (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10464	CTL_DEBUG_PRINT(("additional_length = %d\n",
10465			 inq_ptr->additional_length));
10466
10467	inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10468	/* 16 bit addressing */
10469	if (port_type == CTL_PORT_SCSI)
10470		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10471	/* XXX set the SID_MultiP bit here if we're actually going to
10472	   respond on multiple ports */
10473	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10474
10475	/* 16 bit data bus, synchronous transfers */
10476	if (port_type == CTL_PORT_SCSI)
10477		inq_ptr->flags = SID_WBus16 | SID_Sync;
10478	/*
10479	 * XXX KDM do we want to support tagged queueing on the control
10480	 * device at all?
10481	 */
10482	if ((lun == NULL)
10483	 || (lun->be_lun->lun_type != T_PROCESSOR))
10484		inq_ptr->flags |= SID_CmdQue;
10485	/*
10486	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10487	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10488	 * name and 4 bytes for the revision.
10489	 */
10490	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10491	    "vendor")) == NULL) {
10492		strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10493	} else {
10494		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10495		strncpy(inq_ptr->vendor, val,
10496		    min(sizeof(inq_ptr->vendor), strlen(val)));
10497	}
10498	if (lun == NULL) {
10499		strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10500		    sizeof(inq_ptr->product));
10501	} else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10502		switch (lun->be_lun->lun_type) {
10503		case T_DIRECT:
10504			strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10505			    sizeof(inq_ptr->product));
10506			break;
10507		case T_PROCESSOR:
10508			strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10509			    sizeof(inq_ptr->product));
10510			break;
10511		default:
10512			strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10513			    sizeof(inq_ptr->product));
10514			break;
10515		}
10516	} else {
10517		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10518		strncpy(inq_ptr->product, val,
10519		    min(sizeof(inq_ptr->product), strlen(val)));
10520	}
10521
10522	/*
10523	 * XXX make this a macro somewhere so it automatically gets
10524	 * incremented when we make changes.
10525	 */
10526	if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10527	    "revision")) == NULL) {
10528		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10529	} else {
10530		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10531		strncpy(inq_ptr->revision, val,
10532		    min(sizeof(inq_ptr->revision), strlen(val)));
10533	}
10534
10535	/*
10536	 * For parallel SCSI, we support double transition and single
10537	 * transition clocking.  We also support QAS (Quick Arbitration
10538	 * and Selection) and Information Unit transfers on both the
10539	 * control and array devices.
10540	 */
10541	if (port_type == CTL_PORT_SCSI)
10542		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10543				    SID_SPI_IUS;
10544
10545	/* SAM-5 (no version claimed) */
10546	scsi_ulto2b(0x00A0, inq_ptr->version1);
10547	/* SPC-4 (no version claimed) */
10548	scsi_ulto2b(0x0460, inq_ptr->version2);
10549	if (port_type == CTL_PORT_FC) {
10550		/* FCP-2 ANSI INCITS.350:2003 */
10551		scsi_ulto2b(0x0917, inq_ptr->version3);
10552	} else if (port_type == CTL_PORT_SCSI) {
10553		/* SPI-4 ANSI INCITS.362:200x */
10554		scsi_ulto2b(0x0B56, inq_ptr->version3);
10555	} else if (port_type == CTL_PORT_ISCSI) {
10556		/* iSCSI (no version claimed) */
10557		scsi_ulto2b(0x0960, inq_ptr->version3);
10558	} else if (port_type == CTL_PORT_SAS) {
10559		/* SAS (no version claimed) */
10560		scsi_ulto2b(0x0BE0, inq_ptr->version3);
10561	}
10562
10563	if (lun == NULL) {
10564		/* SBC-4 (no version claimed) */
10565		scsi_ulto2b(0x0600, inq_ptr->version4);
10566	} else {
10567		switch (lun->be_lun->lun_type) {
10568		case T_DIRECT:
10569			/* SBC-4 (no version claimed) */
10570			scsi_ulto2b(0x0600, inq_ptr->version4);
10571			break;
10572		case T_PROCESSOR:
10573		default:
10574			break;
10575		}
10576	}
10577
10578	ctl_set_success(ctsio);
10579	ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10580	ctsio->be_move_done = ctl_config_move_done;
10581	ctl_datamove((union ctl_io *)ctsio);
10582	return (CTL_RETVAL_COMPLETE);
10583}
10584
10585int
10586ctl_inquiry(struct ctl_scsiio *ctsio)
10587{
10588	struct scsi_inquiry *cdb;
10589	int retval;
10590
10591	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10592
10593	cdb = (struct scsi_inquiry *)ctsio->cdb;
10594	if (cdb->byte2 & SI_EVPD)
10595		retval = ctl_inquiry_evpd(ctsio);
10596	else if (cdb->page_code == 0)
10597		retval = ctl_inquiry_std(ctsio);
10598	else {
10599		ctl_set_invalid_field(ctsio,
10600				      /*sks_valid*/ 1,
10601				      /*command*/ 1,
10602				      /*field*/ 2,
10603				      /*bit_valid*/ 0,
10604				      /*bit*/ 0);
10605		ctl_done((union ctl_io *)ctsio);
10606		return (CTL_RETVAL_COMPLETE);
10607	}
10608
10609	return (retval);
10610}
10611
10612/*
10613 * For known CDB types, parse the LBA and length.
10614 */
10615static int
10616ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10617{
10618	if (io->io_hdr.io_type != CTL_IO_SCSI)
10619		return (1);
10620
10621	switch (io->scsiio.cdb[0]) {
10622	case COMPARE_AND_WRITE: {
10623		struct scsi_compare_and_write *cdb;
10624
10625		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10626
10627		*lba = scsi_8btou64(cdb->addr);
10628		*len = cdb->length;
10629		break;
10630	}
10631	case READ_6:
10632	case WRITE_6: {
10633		struct scsi_rw_6 *cdb;
10634
10635		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10636
10637		*lba = scsi_3btoul(cdb->addr);
10638		/* only 5 bits are valid in the most significant address byte */
10639		*lba &= 0x1fffff;
10640		*len = cdb->length;
10641		break;
10642	}
10643	case READ_10:
10644	case WRITE_10: {
10645		struct scsi_rw_10 *cdb;
10646
10647		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10648
10649		*lba = scsi_4btoul(cdb->addr);
10650		*len = scsi_2btoul(cdb->length);
10651		break;
10652	}
10653	case WRITE_VERIFY_10: {
10654		struct scsi_write_verify_10 *cdb;
10655
10656		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10657
10658		*lba = scsi_4btoul(cdb->addr);
10659		*len = scsi_2btoul(cdb->length);
10660		break;
10661	}
10662	case READ_12:
10663	case WRITE_12: {
10664		struct scsi_rw_12 *cdb;
10665
10666		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10667
10668		*lba = scsi_4btoul(cdb->addr);
10669		*len = scsi_4btoul(cdb->length);
10670		break;
10671	}
10672	case WRITE_VERIFY_12: {
10673		struct scsi_write_verify_12 *cdb;
10674
10675		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10676
10677		*lba = scsi_4btoul(cdb->addr);
10678		*len = scsi_4btoul(cdb->length);
10679		break;
10680	}
10681	case READ_16:
10682	case WRITE_16:
10683	case WRITE_ATOMIC_16: {
10684		struct scsi_rw_16 *cdb;
10685
10686		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10687
10688		*lba = scsi_8btou64(cdb->addr);
10689		*len = scsi_4btoul(cdb->length);
10690		break;
10691	}
10692	case WRITE_VERIFY_16: {
10693		struct scsi_write_verify_16 *cdb;
10694
10695		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10696
10697		*lba = scsi_8btou64(cdb->addr);
10698		*len = scsi_4btoul(cdb->length);
10699		break;
10700	}
10701	case WRITE_SAME_10: {
10702		struct scsi_write_same_10 *cdb;
10703
10704		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10705
10706		*lba = scsi_4btoul(cdb->addr);
10707		*len = scsi_2btoul(cdb->length);
10708		break;
10709	}
10710	case WRITE_SAME_16: {
10711		struct scsi_write_same_16 *cdb;
10712
10713		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10714
10715		*lba = scsi_8btou64(cdb->addr);
10716		*len = scsi_4btoul(cdb->length);
10717		break;
10718	}
10719	case VERIFY_10: {
10720		struct scsi_verify_10 *cdb;
10721
10722		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10723
10724		*lba = scsi_4btoul(cdb->addr);
10725		*len = scsi_2btoul(cdb->length);
10726		break;
10727	}
10728	case VERIFY_12: {
10729		struct scsi_verify_12 *cdb;
10730
10731		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10732
10733		*lba = scsi_4btoul(cdb->addr);
10734		*len = scsi_4btoul(cdb->length);
10735		break;
10736	}
10737	case VERIFY_16: {
10738		struct scsi_verify_16 *cdb;
10739
10740		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10741
10742		*lba = scsi_8btou64(cdb->addr);
10743		*len = scsi_4btoul(cdb->length);
10744		break;
10745	}
10746	case UNMAP: {
10747		*lba = 0;
10748		*len = UINT64_MAX;
10749		break;
10750	}
10751	case SERVICE_ACTION_IN: {	/* GET LBA STATUS */
10752		struct scsi_get_lba_status *cdb;
10753
10754		cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10755		*lba = scsi_8btou64(cdb->addr);
10756		*len = UINT32_MAX;
10757		break;
10758	}
10759	default:
10760		return (1);
10761		break; /* NOTREACHED */
10762	}
10763
10764	return (0);
10765}
10766
10767static ctl_action
10768ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10769    bool seq)
10770{
10771	uint64_t endlba1, endlba2;
10772
10773	endlba1 = lba1 + len1 - (seq ? 0 : 1);
10774	endlba2 = lba2 + len2 - 1;
10775
10776	if ((endlba1 < lba2) || (endlba2 < lba1))
10777		return (CTL_ACTION_PASS);
10778	else
10779		return (CTL_ACTION_BLOCK);
10780}
10781
10782static int
10783ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10784{
10785	struct ctl_ptr_len_flags *ptrlen;
10786	struct scsi_unmap_desc *buf, *end, *range;
10787	uint64_t lba;
10788	uint32_t len;
10789
10790	/* If not UNMAP -- go other way. */
10791	if (io->io_hdr.io_type != CTL_IO_SCSI ||
10792	    io->scsiio.cdb[0] != UNMAP)
10793		return (CTL_ACTION_ERROR);
10794
10795	/* If UNMAP without data -- block and wait for data. */
10796	ptrlen = (struct ctl_ptr_len_flags *)
10797	    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10798	if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10799	    ptrlen->ptr == NULL)
10800		return (CTL_ACTION_BLOCK);
10801
10802	/* UNMAP with data -- check for collision. */
10803	buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10804	end = buf + ptrlen->len / sizeof(*buf);
10805	for (range = buf; range < end; range++) {
10806		lba = scsi_8btou64(range->lba);
10807		len = scsi_4btoul(range->length);
10808		if ((lba < lba2 + len2) && (lba + len > lba2))
10809			return (CTL_ACTION_BLOCK);
10810	}
10811	return (CTL_ACTION_PASS);
10812}
10813
10814static ctl_action
10815ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10816{
10817	uint64_t lba1, lba2;
10818	uint64_t len1, len2;
10819	int retval;
10820
10821	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10822		return (CTL_ACTION_ERROR);
10823
10824	retval = ctl_extent_check_unmap(io1, lba2, len2);
10825	if (retval != CTL_ACTION_ERROR)
10826		return (retval);
10827
10828	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10829		return (CTL_ACTION_ERROR);
10830
10831	return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10832}
10833
10834static ctl_action
10835ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10836{
10837	uint64_t lba1, lba2;
10838	uint64_t len1, len2;
10839
10840	if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10841		return (CTL_ACTION_ERROR);
10842	if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10843		return (CTL_ACTION_ERROR);
10844
10845	if (lba1 + len1 == lba2)
10846		return (CTL_ACTION_BLOCK);
10847	return (CTL_ACTION_PASS);
10848}
10849
10850static ctl_action
10851ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10852    union ctl_io *ooa_io)
10853{
10854	const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10855	ctl_serialize_action *serialize_row;
10856
10857	/*
10858	 * The initiator attempted multiple untagged commands at the same
10859	 * time.  Can't do that.
10860	 */
10861	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10862	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10863	 && ((pending_io->io_hdr.nexus.targ_port ==
10864	      ooa_io->io_hdr.nexus.targ_port)
10865	  && (pending_io->io_hdr.nexus.initid.id ==
10866	      ooa_io->io_hdr.nexus.initid.id))
10867	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10868		return (CTL_ACTION_OVERLAP);
10869
10870	/*
10871	 * The initiator attempted to send multiple tagged commands with
10872	 * the same ID.  (It's fine if different initiators have the same
10873	 * tag ID.)
10874	 *
10875	 * Even if all of those conditions are true, we don't kill the I/O
10876	 * if the command ahead of us has been aborted.  We won't end up
10877	 * sending it to the FETD, and it's perfectly legal to resend a
10878	 * command with the same tag number as long as the previous
10879	 * instance of this tag number has been aborted somehow.
10880	 */
10881	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10882	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10883	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10884	 && ((pending_io->io_hdr.nexus.targ_port ==
10885	      ooa_io->io_hdr.nexus.targ_port)
10886	  && (pending_io->io_hdr.nexus.initid.id ==
10887	      ooa_io->io_hdr.nexus.initid.id))
10888	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10889		return (CTL_ACTION_OVERLAP_TAG);
10890
10891	/*
10892	 * If we get a head of queue tag, SAM-3 says that we should
10893	 * immediately execute it.
10894	 *
10895	 * What happens if this command would normally block for some other
10896	 * reason?  e.g. a request sense with a head of queue tag
10897	 * immediately after a write.  Normally that would block, but this
10898	 * will result in its getting executed immediately...
10899	 *
10900	 * We currently return "pass" instead of "skip", so we'll end up
10901	 * going through the rest of the queue to check for overlapped tags.
10902	 *
10903	 * XXX KDM check for other types of blockage first??
10904	 */
10905	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10906		return (CTL_ACTION_PASS);
10907
10908	/*
10909	 * Ordered tags have to block until all items ahead of them
10910	 * have completed.  If we get called with an ordered tag, we always
10911	 * block, if something else is ahead of us in the queue.
10912	 */
10913	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10914		return (CTL_ACTION_BLOCK);
10915
10916	/*
10917	 * Simple tags get blocked until all head of queue and ordered tags
10918	 * ahead of them have completed.  I'm lumping untagged commands in
10919	 * with simple tags here.  XXX KDM is that the right thing to do?
10920	 */
10921	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10922	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10923	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10924	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10925		return (CTL_ACTION_BLOCK);
10926
10927	pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10928	ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10929
10930	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10931
10932	switch (serialize_row[pending_entry->seridx]) {
10933	case CTL_SER_BLOCK:
10934		return (CTL_ACTION_BLOCK);
10935	case CTL_SER_EXTENT:
10936		return (ctl_extent_check(ooa_io, pending_io,
10937		    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10938	case CTL_SER_EXTENTOPT:
10939		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10940		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10941			return (ctl_extent_check(ooa_io, pending_io,
10942			    (lun->serseq == CTL_LUN_SERSEQ_ON)));
10943		return (CTL_ACTION_PASS);
10944	case CTL_SER_EXTENTSEQ:
10945		if (lun->serseq != CTL_LUN_SERSEQ_OFF)
10946			return (ctl_extent_check_seq(ooa_io, pending_io));
10947		return (CTL_ACTION_PASS);
10948	case CTL_SER_PASS:
10949		return (CTL_ACTION_PASS);
10950	case CTL_SER_BLOCKOPT:
10951		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10952		    & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10953			return (CTL_ACTION_BLOCK);
10954		return (CTL_ACTION_PASS);
10955	case CTL_SER_SKIP:
10956		return (CTL_ACTION_SKIP);
10957	default:
10958		panic("invalid serialization value %d",
10959		      serialize_row[pending_entry->seridx]);
10960	}
10961
10962	return (CTL_ACTION_ERROR);
10963}
10964
10965/*
10966 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10967 * Assumptions:
10968 * - pending_io is generally either incoming, or on the blocked queue
10969 * - starting I/O is the I/O we want to start the check with.
10970 */
10971static ctl_action
10972ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10973	      union ctl_io *starting_io)
10974{
10975	union ctl_io *ooa_io;
10976	ctl_action action;
10977
10978	mtx_assert(&lun->lun_lock, MA_OWNED);
10979
10980	/*
10981	 * Run back along the OOA queue, starting with the current
10982	 * blocked I/O and going through every I/O before it on the
10983	 * queue.  If starting_io is NULL, we'll just end up returning
10984	 * CTL_ACTION_PASS.
10985	 */
10986	for (ooa_io = starting_io; ooa_io != NULL;
10987	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10988	     ooa_links)){
10989
10990		/*
10991		 * This routine just checks to see whether
10992		 * cur_blocked is blocked by ooa_io, which is ahead
10993		 * of it in the queue.  It doesn't queue/dequeue
10994		 * cur_blocked.
10995		 */
10996		action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10997		switch (action) {
10998		case CTL_ACTION_BLOCK:
10999		case CTL_ACTION_OVERLAP:
11000		case CTL_ACTION_OVERLAP_TAG:
11001		case CTL_ACTION_SKIP:
11002		case CTL_ACTION_ERROR:
11003			return (action);
11004			break; /* NOTREACHED */
11005		case CTL_ACTION_PASS:
11006			break;
11007		default:
11008			panic("invalid action %d", action);
11009			break;  /* NOTREACHED */
11010		}
11011	}
11012
11013	return (CTL_ACTION_PASS);
11014}
11015
11016/*
11017 * Assumptions:
11018 * - An I/O has just completed, and has been removed from the per-LUN OOA
11019 *   queue, so some items on the blocked queue may now be unblocked.
11020 */
11021static int
11022ctl_check_blocked(struct ctl_lun *lun)
11023{
11024	union ctl_io *cur_blocked, *next_blocked;
11025
11026	mtx_assert(&lun->lun_lock, MA_OWNED);
11027
11028	/*
11029	 * Run forward from the head of the blocked queue, checking each
11030	 * entry against the I/Os prior to it on the OOA queue to see if
11031	 * there is still any blockage.
11032	 *
11033	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11034	 * with our removing a variable on it while it is traversing the
11035	 * list.
11036	 */
11037	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11038	     cur_blocked != NULL; cur_blocked = next_blocked) {
11039		union ctl_io *prev_ooa;
11040		ctl_action action;
11041
11042		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11043							  blocked_links);
11044
11045		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11046						      ctl_ooaq, ooa_links);
11047
11048		/*
11049		 * If cur_blocked happens to be the first item in the OOA
11050		 * queue now, prev_ooa will be NULL, and the action
11051		 * returned will just be CTL_ACTION_PASS.
11052		 */
11053		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11054
11055		switch (action) {
11056		case CTL_ACTION_BLOCK:
11057			/* Nothing to do here, still blocked */
11058			break;
11059		case CTL_ACTION_OVERLAP:
11060		case CTL_ACTION_OVERLAP_TAG:
11061			/*
11062			 * This shouldn't happen!  In theory we've already
11063			 * checked this command for overlap...
11064			 */
11065			break;
11066		case CTL_ACTION_PASS:
11067		case CTL_ACTION_SKIP: {
11068			struct ctl_softc *softc;
11069			const struct ctl_cmd_entry *entry;
11070			int isc_retval;
11071
11072			/*
11073			 * The skip case shouldn't happen, this transaction
11074			 * should have never made it onto the blocked queue.
11075			 */
11076			/*
11077			 * This I/O is no longer blocked, we can remove it
11078			 * from the blocked queue.  Since this is a TAILQ
11079			 * (doubly linked list), we can do O(1) removals
11080			 * from any place on the list.
11081			 */
11082			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11083				     blocked_links);
11084			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11085
11086			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
11087				/*
11088				 * Need to send IO back to original side to
11089				 * run
11090				 */
11091				union ctl_ha_msg msg_info;
11092
11093				msg_info.hdr.original_sc =
11094					cur_blocked->io_hdr.original_sc;
11095				msg_info.hdr.serializing_sc = cur_blocked;
11096				msg_info.hdr.msg_type = CTL_MSG_R2R;
11097				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11098				     &msg_info, sizeof(msg_info), 0)) >
11099				     CTL_HA_STATUS_SUCCESS) {
11100					printf("CTL:Check Blocked error from "
11101					       "ctl_ha_msg_send %d\n",
11102					       isc_retval);
11103				}
11104				break;
11105			}
11106			entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11107			softc = control_softc;
11108
11109			/*
11110			 * Check this I/O for LUN state changes that may
11111			 * have happened while this command was blocked.
11112			 * The LUN state may have been changed by a command
11113			 * ahead of us in the queue, so we need to re-check
11114			 * for any states that can be caused by SCSI
11115			 * commands.
11116			 */
11117			if (ctl_scsiio_lun_check(softc, lun, entry,
11118						 &cur_blocked->scsiio) == 0) {
11119				cur_blocked->io_hdr.flags |=
11120				                      CTL_FLAG_IS_WAS_ON_RTR;
11121				ctl_enqueue_rtr(cur_blocked);
11122			} else
11123				ctl_done(cur_blocked);
11124			break;
11125		}
11126		default:
11127			/*
11128			 * This probably shouldn't happen -- we shouldn't
11129			 * get CTL_ACTION_ERROR, or anything else.
11130			 */
11131			break;
11132		}
11133	}
11134
11135	return (CTL_RETVAL_COMPLETE);
11136}
11137
11138/*
11139 * This routine (with one exception) checks LUN flags that can be set by
11140 * commands ahead of us in the OOA queue.  These flags have to be checked
11141 * when a command initially comes in, and when we pull a command off the
11142 * blocked queue and are preparing to execute it.  The reason we have to
11143 * check these flags for commands on the blocked queue is that the LUN
11144 * state may have been changed by a command ahead of us while we're on the
11145 * blocked queue.
11146 *
11147 * Ordering is somewhat important with these checks, so please pay
11148 * careful attention to the placement of any new checks.
11149 */
11150static int
11151ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
11152    const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11153{
11154	int retval;
11155	uint32_t residx;
11156
11157	retval = 0;
11158
11159	mtx_assert(&lun->lun_lock, MA_OWNED);
11160
11161	/*
11162	 * If this shelf is a secondary shelf controller, we have to reject
11163	 * any media access commands.
11164	 */
11165	if ((ctl_softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 &&
11166	    (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) {
11167		ctl_set_lun_standby(ctsio);
11168		retval = 1;
11169		goto bailout;
11170	}
11171
11172	if (entry->pattern & CTL_LUN_PAT_WRITE) {
11173		if (lun->flags & CTL_LUN_READONLY) {
11174			ctl_set_sense(ctsio, /*current_error*/ 1,
11175			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11176			    /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE);
11177			retval = 1;
11178			goto bailout;
11179		}
11180		if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11181		    .eca_and_aen & SCP_SWP) != 0) {
11182			ctl_set_sense(ctsio, /*current_error*/ 1,
11183			    /*sense_key*/ SSD_KEY_DATA_PROTECT,
11184			    /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11185			retval = 1;
11186			goto bailout;
11187		}
11188	}
11189
11190	/*
11191	 * Check for a reservation conflict.  If this command isn't allowed
11192	 * even on reserved LUNs, and if this initiator isn't the one who
11193	 * reserved us, reject the command with a reservation conflict.
11194	 */
11195	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
11196	if ((lun->flags & CTL_LUN_RESERVED)
11197	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11198		if (lun->res_idx != residx) {
11199			ctl_set_reservation_conflict(ctsio);
11200			retval = 1;
11201			goto bailout;
11202		}
11203	}
11204
11205	if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11206	    (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11207		/* No reservation or command is allowed. */;
11208	} else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11209	    (lun->res_type == SPR_TYPE_WR_EX ||
11210	     lun->res_type == SPR_TYPE_WR_EX_RO ||
11211	     lun->res_type == SPR_TYPE_WR_EX_AR)) {
11212		/* The command is allowed for Write Exclusive resv. */;
11213	} else {
11214		/*
11215		 * if we aren't registered or it's a res holder type
11216		 * reservation and this isn't the res holder then set a
11217		 * conflict.
11218		 */
11219		if (ctl_get_prkey(lun, residx) == 0
11220		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11221			ctl_set_reservation_conflict(ctsio);
11222			retval = 1;
11223			goto bailout;
11224		}
11225
11226	}
11227
11228	if ((lun->flags & CTL_LUN_OFFLINE)
11229	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
11230		ctl_set_lun_not_ready(ctsio);
11231		retval = 1;
11232		goto bailout;
11233	}
11234
11235	/*
11236	 * If the LUN is stopped, see if this particular command is allowed
11237	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
11238	 */
11239	if ((lun->flags & CTL_LUN_STOPPED)
11240	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11241		/* "Logical unit not ready, initializing cmd. required" */
11242		ctl_set_lun_stopped(ctsio);
11243		retval = 1;
11244		goto bailout;
11245	}
11246
11247	if ((lun->flags & CTL_LUN_INOPERABLE)
11248	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11249		/* "Medium format corrupted" */
11250		ctl_set_medium_format_corrupted(ctsio);
11251		retval = 1;
11252		goto bailout;
11253	}
11254
11255bailout:
11256	return (retval);
11257
11258}
11259
11260static void
11261ctl_failover_io(union ctl_io *io, int have_lock)
11262{
11263	ctl_set_busy(&io->scsiio);
11264	ctl_done(io);
11265}
11266
11267static void
11268ctl_failover(void)
11269{
11270	struct ctl_lun *lun;
11271	struct ctl_softc *ctl_softc;
11272	union ctl_io *next_io, *pending_io;
11273	union ctl_io *io;
11274	int lun_idx;
11275
11276	ctl_softc = control_softc;
11277
11278	mtx_lock(&ctl_softc->ctl_lock);
11279	/*
11280	 * Remove any cmds from the other SC from the rtr queue.  These
11281	 * will obviously only be for LUNs for which we're the primary.
11282	 * We can't send status or get/send data for these commands.
11283	 * Since they haven't been executed yet, we can just remove them.
11284	 * We'll either abort them or delete them below, depending on
11285	 * which HA mode we're in.
11286	 */
11287#ifdef notyet
11288	mtx_lock(&ctl_softc->queue_lock);
11289	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
11290	     io != NULL; io = next_io) {
11291		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11292		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11293			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
11294				      ctl_io_hdr, links);
11295	}
11296	mtx_unlock(&ctl_softc->queue_lock);
11297#endif
11298
11299	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
11300		lun = ctl_softc->ctl_luns[lun_idx];
11301		if (lun==NULL)
11302			continue;
11303
11304		/*
11305		 * Processor LUNs are primary on both sides.
11306		 * XXX will this always be true?
11307		 */
11308		if (lun->be_lun->lun_type == T_PROCESSOR)
11309			continue;
11310
11311		if ((lun->flags & CTL_LUN_PRIMARY_SC)
11312		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11313			printf("FAILOVER: primary lun %d\n", lun_idx);
11314		        /*
11315			 * Remove all commands from the other SC. First from the
11316			 * blocked queue then from the ooa queue. Once we have
11317			 * removed them. Call ctl_check_blocked to see if there
11318			 * is anything that can run.
11319			 */
11320			for (io = (union ctl_io *)TAILQ_FIRST(
11321			     &lun->blocked_queue); io != NULL; io = next_io) {
11322
11323		        	next_io = (union ctl_io *)TAILQ_NEXT(
11324				    &io->io_hdr, blocked_links);
11325
11326				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11327					TAILQ_REMOVE(&lun->blocked_queue,
11328						     &io->io_hdr,blocked_links);
11329					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11330					TAILQ_REMOVE(&lun->ooa_queue,
11331						     &io->io_hdr, ooa_links);
11332
11333					ctl_free_io(io);
11334				}
11335			}
11336
11337			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11338	     		     io != NULL; io = next_io) {
11339
11340		        	next_io = (union ctl_io *)TAILQ_NEXT(
11341				    &io->io_hdr, ooa_links);
11342
11343				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
11344
11345					TAILQ_REMOVE(&lun->ooa_queue,
11346						&io->io_hdr,
11347					     	ooa_links);
11348
11349					ctl_free_io(io);
11350				}
11351			}
11352			ctl_check_blocked(lun);
11353		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
11354			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11355
11356			printf("FAILOVER: primary lun %d\n", lun_idx);
11357			/*
11358			 * Abort all commands from the other SC.  We can't
11359			 * send status back for them now.  These should get
11360			 * cleaned up when they are completed or come out
11361			 * for a datamove operation.
11362			 */
11363			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
11364	     		     io != NULL; io = next_io) {
11365		        	next_io = (union ctl_io *)TAILQ_NEXT(
11366					&io->io_hdr, ooa_links);
11367
11368				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
11369					io->io_hdr.flags |= CTL_FLAG_ABORT;
11370			}
11371		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11372			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
11373
11374			printf("FAILOVER: secondary lun %d\n", lun_idx);
11375
11376			lun->flags |= CTL_LUN_PRIMARY_SC;
11377
11378			/*
11379			 * We send all I/O that was sent to this controller
11380			 * and redirected to the other side back with
11381			 * busy status, and have the initiator retry it.
11382			 * Figuring out how much data has been transferred,
11383			 * etc. and picking up where we left off would be
11384			 * very tricky.
11385			 *
11386			 * XXX KDM need to remove I/O from the blocked
11387			 * queue as well!
11388			 */
11389			for (pending_io = (union ctl_io *)TAILQ_FIRST(
11390			     &lun->ooa_queue); pending_io != NULL;
11391			     pending_io = next_io) {
11392
11393				next_io =  (union ctl_io *)TAILQ_NEXT(
11394					&pending_io->io_hdr, ooa_links);
11395
11396				pending_io->io_hdr.flags &=
11397					~CTL_FLAG_SENT_2OTHER_SC;
11398
11399				if (pending_io->io_hdr.flags &
11400				    CTL_FLAG_IO_ACTIVE) {
11401					pending_io->io_hdr.flags |=
11402						CTL_FLAG_FAILOVER;
11403				} else {
11404					ctl_set_busy(&pending_io->scsiio);
11405					ctl_done(pending_io);
11406				}
11407			}
11408
11409			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11410		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
11411			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
11412			printf("FAILOVER: secondary lun %d\n", lun_idx);
11413			/*
11414			 * if the first io on the OOA is not on the RtR queue
11415			 * add it.
11416			 */
11417			lun->flags |= CTL_LUN_PRIMARY_SC;
11418
11419			pending_io = (union ctl_io *)TAILQ_FIRST(
11420			    &lun->ooa_queue);
11421			if (pending_io==NULL) {
11422				printf("Nothing on OOA queue\n");
11423				continue;
11424			}
11425
11426			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11427			if ((pending_io->io_hdr.flags &
11428			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
11429				pending_io->io_hdr.flags |=
11430				    CTL_FLAG_IS_WAS_ON_RTR;
11431				ctl_enqueue_rtr(pending_io);
11432			}
11433#if 0
11434			else
11435			{
11436				printf("Tag 0x%04x is running\n",
11437				      pending_io->scsiio.tag_num);
11438			}
11439#endif
11440
11441			next_io = (union ctl_io *)TAILQ_NEXT(
11442			    &pending_io->io_hdr, ooa_links);
11443			for (pending_io=next_io; pending_io != NULL;
11444			     pending_io = next_io) {
11445				pending_io->io_hdr.flags &=
11446				    ~CTL_FLAG_SENT_2OTHER_SC;
11447				next_io = (union ctl_io *)TAILQ_NEXT(
11448					&pending_io->io_hdr, ooa_links);
11449				if (pending_io->io_hdr.flags &
11450				    CTL_FLAG_IS_WAS_ON_RTR) {
11451#if 0
11452				        printf("Tag 0x%04x is running\n",
11453				      		pending_io->scsiio.tag_num);
11454#endif
11455					continue;
11456				}
11457
11458				switch (ctl_check_ooa(lun, pending_io,
11459			            (union ctl_io *)TAILQ_PREV(
11460				    &pending_io->io_hdr, ctl_ooaq,
11461				    ooa_links))) {
11462
11463				case CTL_ACTION_BLOCK:
11464					TAILQ_INSERT_TAIL(&lun->blocked_queue,
11465							  &pending_io->io_hdr,
11466							  blocked_links);
11467					pending_io->io_hdr.flags |=
11468					    CTL_FLAG_BLOCKED;
11469					break;
11470				case CTL_ACTION_PASS:
11471				case CTL_ACTION_SKIP:
11472					pending_io->io_hdr.flags |=
11473					    CTL_FLAG_IS_WAS_ON_RTR;
11474					ctl_enqueue_rtr(pending_io);
11475					break;
11476				case CTL_ACTION_OVERLAP:
11477					ctl_set_overlapped_cmd(
11478					    (struct ctl_scsiio *)pending_io);
11479					ctl_done(pending_io);
11480					break;
11481				case CTL_ACTION_OVERLAP_TAG:
11482					ctl_set_overlapped_tag(
11483					    (struct ctl_scsiio *)pending_io,
11484					    pending_io->scsiio.tag_num & 0xff);
11485					ctl_done(pending_io);
11486					break;
11487				case CTL_ACTION_ERROR:
11488				default:
11489					ctl_set_internal_failure(
11490						(struct ctl_scsiio *)pending_io,
11491						0,  // sks_valid
11492						0); //retry count
11493					ctl_done(pending_io);
11494					break;
11495				}
11496			}
11497
11498			ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
11499		} else {
11500			panic("Unhandled HA mode failover, LUN flags = %#x, "
11501			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11502		}
11503	}
11504	ctl_pause_rtr = 0;
11505	mtx_unlock(&ctl_softc->ctl_lock);
11506}
11507
11508static int
11509ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11510{
11511	struct ctl_lun *lun;
11512	const struct ctl_cmd_entry *entry;
11513	uint32_t initidx, targ_lun;
11514	int retval;
11515
11516	retval = 0;
11517
11518	lun = NULL;
11519
11520	targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11521	if ((targ_lun < CTL_MAX_LUNS)
11522	 && ((lun = ctl_softc->ctl_luns[targ_lun]) != NULL)) {
11523		/*
11524		 * If the LUN is invalid, pretend that it doesn't exist.
11525		 * It will go away as soon as all pending I/O has been
11526		 * completed.
11527		 */
11528		mtx_lock(&lun->lun_lock);
11529		if (lun->flags & CTL_LUN_DISABLED) {
11530			mtx_unlock(&lun->lun_lock);
11531			lun = NULL;
11532			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11533			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11534		} else {
11535			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11536			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11537				lun->be_lun;
11538			if (lun->be_lun->lun_type == T_PROCESSOR) {
11539				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11540			}
11541
11542			/*
11543			 * Every I/O goes into the OOA queue for a
11544			 * particular LUN, and stays there until completion.
11545			 */
11546			TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11547			    ooa_links);
11548		}
11549	} else {
11550		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11551		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11552	}
11553
11554	/* Get command entry and return error if it is unsuppotyed. */
11555	entry = ctl_validate_command(ctsio);
11556	if (entry == NULL) {
11557		if (lun)
11558			mtx_unlock(&lun->lun_lock);
11559		return (retval);
11560	}
11561
11562	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11563	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11564
11565	/*
11566	 * Check to see whether we can send this command to LUNs that don't
11567	 * exist.  This should pretty much only be the case for inquiry
11568	 * and request sense.  Further checks, below, really require having
11569	 * a LUN, so we can't really check the command anymore.  Just put
11570	 * it on the rtr queue.
11571	 */
11572	if (lun == NULL) {
11573		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11574			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11575			ctl_enqueue_rtr((union ctl_io *)ctsio);
11576			return (retval);
11577		}
11578
11579		ctl_set_unsupported_lun(ctsio);
11580		ctl_done((union ctl_io *)ctsio);
11581		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11582		return (retval);
11583	} else {
11584		/*
11585		 * Make sure we support this particular command on this LUN.
11586		 * e.g., we don't support writes to the control LUN.
11587		 */
11588		if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11589			mtx_unlock(&lun->lun_lock);
11590			ctl_set_invalid_opcode(ctsio);
11591			ctl_done((union ctl_io *)ctsio);
11592			return (retval);
11593		}
11594	}
11595
11596	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11597
11598#ifdef CTL_WITH_CA
11599	/*
11600	 * If we've got a request sense, it'll clear the contingent
11601	 * allegiance condition.  Otherwise, if we have a CA condition for
11602	 * this initiator, clear it, because it sent down a command other
11603	 * than request sense.
11604	 */
11605	if ((ctsio->cdb[0] != REQUEST_SENSE)
11606	 && (ctl_is_set(lun->have_ca, initidx)))
11607		ctl_clear_mask(lun->have_ca, initidx);
11608#endif
11609
11610	/*
11611	 * If the command has this flag set, it handles its own unit
11612	 * attention reporting, we shouldn't do anything.  Otherwise we
11613	 * check for any pending unit attentions, and send them back to the
11614	 * initiator.  We only do this when a command initially comes in,
11615	 * not when we pull it off the blocked queue.
11616	 *
11617	 * According to SAM-3, section 5.3.2, the order that things get
11618	 * presented back to the host is basically unit attentions caused
11619	 * by some sort of reset event, busy status, reservation conflicts
11620	 * or task set full, and finally any other status.
11621	 *
11622	 * One issue here is that some of the unit attentions we report
11623	 * don't fall into the "reset" category (e.g. "reported luns data
11624	 * has changed").  So reporting it here, before the reservation
11625	 * check, may be technically wrong.  I guess the only thing to do
11626	 * would be to check for and report the reset events here, and then
11627	 * check for the other unit attention types after we check for a
11628	 * reservation conflict.
11629	 *
11630	 * XXX KDM need to fix this
11631	 */
11632	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11633		ctl_ua_type ua_type;
11634		scsi_sense_data_type sense_format;
11635
11636		if (lun->flags & CTL_LUN_SENSE_DESC)
11637			sense_format = SSD_TYPE_DESC;
11638		else
11639			sense_format = SSD_TYPE_FIXED;
11640
11641		ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11642		    sense_format);
11643		if (ua_type != CTL_UA_NONE) {
11644			mtx_unlock(&lun->lun_lock);
11645			ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11646			ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11647			ctsio->sense_len = SSD_FULL_SIZE;
11648			ctl_done((union ctl_io *)ctsio);
11649			return (retval);
11650		}
11651	}
11652
11653
11654	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11655		mtx_unlock(&lun->lun_lock);
11656		ctl_done((union ctl_io *)ctsio);
11657		return (retval);
11658	}
11659
11660	/*
11661	 * XXX CHD this is where we want to send IO to other side if
11662	 * this LUN is secondary on this SC. We will need to make a copy
11663	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11664	 * the copy we send as FROM_OTHER.
11665	 * We also need to stuff the address of the original IO so we can
11666	 * find it easily. Something similar will need be done on the other
11667	 * side so when we are done we can find the copy.
11668	 */
11669	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11670		union ctl_ha_msg msg_info;
11671		int isc_retval;
11672
11673		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11674
11675		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11676		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11677#if 0
11678		printf("1. ctsio %p\n", ctsio);
11679#endif
11680		msg_info.hdr.serializing_sc = NULL;
11681		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11682		msg_info.scsi.tag_num = ctsio->tag_num;
11683		msg_info.scsi.tag_type = ctsio->tag_type;
11684		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11685
11686		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11687
11688		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11689		    (void *)&msg_info, sizeof(msg_info), 0)) >
11690		    CTL_HA_STATUS_SUCCESS) {
11691			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11692			       isc_retval);
11693			printf("CTL:opcode is %x\n", ctsio->cdb[0]);
11694		} else {
11695#if 0
11696			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11697#endif
11698		}
11699
11700		/*
11701		 * XXX KDM this I/O is off the incoming queue, but hasn't
11702		 * been inserted on any other queue.  We may need to come
11703		 * up with a holding queue while we wait for serialization
11704		 * so that we have an idea of what we're waiting for from
11705		 * the other side.
11706		 */
11707		mtx_unlock(&lun->lun_lock);
11708		return (retval);
11709	}
11710
11711	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11712			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11713			      ctl_ooaq, ooa_links))) {
11714	case CTL_ACTION_BLOCK:
11715		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11716		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11717				  blocked_links);
11718		mtx_unlock(&lun->lun_lock);
11719		return (retval);
11720	case CTL_ACTION_PASS:
11721	case CTL_ACTION_SKIP:
11722		ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11723		mtx_unlock(&lun->lun_lock);
11724		ctl_enqueue_rtr((union ctl_io *)ctsio);
11725		break;
11726	case CTL_ACTION_OVERLAP:
11727		mtx_unlock(&lun->lun_lock);
11728		ctl_set_overlapped_cmd(ctsio);
11729		ctl_done((union ctl_io *)ctsio);
11730		break;
11731	case CTL_ACTION_OVERLAP_TAG:
11732		mtx_unlock(&lun->lun_lock);
11733		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11734		ctl_done((union ctl_io *)ctsio);
11735		break;
11736	case CTL_ACTION_ERROR:
11737	default:
11738		mtx_unlock(&lun->lun_lock);
11739		ctl_set_internal_failure(ctsio,
11740					 /*sks_valid*/ 0,
11741					 /*retry_count*/ 0);
11742		ctl_done((union ctl_io *)ctsio);
11743		break;
11744	}
11745	return (retval);
11746}
11747
11748const struct ctl_cmd_entry *
11749ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11750{
11751	const struct ctl_cmd_entry *entry;
11752	int service_action;
11753
11754	entry = &ctl_cmd_table[ctsio->cdb[0]];
11755	if (sa)
11756		*sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11757	if (entry->flags & CTL_CMD_FLAG_SA5) {
11758		service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11759		entry = &((const struct ctl_cmd_entry *)
11760		    entry->execute)[service_action];
11761	}
11762	return (entry);
11763}
11764
11765const struct ctl_cmd_entry *
11766ctl_validate_command(struct ctl_scsiio *ctsio)
11767{
11768	const struct ctl_cmd_entry *entry;
11769	int i, sa;
11770	uint8_t diff;
11771
11772	entry = ctl_get_cmd_entry(ctsio, &sa);
11773	if (entry->execute == NULL) {
11774		if (sa)
11775			ctl_set_invalid_field(ctsio,
11776					      /*sks_valid*/ 1,
11777					      /*command*/ 1,
11778					      /*field*/ 1,
11779					      /*bit_valid*/ 1,
11780					      /*bit*/ 4);
11781		else
11782			ctl_set_invalid_opcode(ctsio);
11783		ctl_done((union ctl_io *)ctsio);
11784		return (NULL);
11785	}
11786	KASSERT(entry->length > 0,
11787	    ("Not defined length for command 0x%02x/0x%02x",
11788	     ctsio->cdb[0], ctsio->cdb[1]));
11789	for (i = 1; i < entry->length; i++) {
11790		diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11791		if (diff == 0)
11792			continue;
11793		ctl_set_invalid_field(ctsio,
11794				      /*sks_valid*/ 1,
11795				      /*command*/ 1,
11796				      /*field*/ i,
11797				      /*bit_valid*/ 1,
11798				      /*bit*/ fls(diff) - 1);
11799		ctl_done((union ctl_io *)ctsio);
11800		return (NULL);
11801	}
11802	return (entry);
11803}
11804
11805static int
11806ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11807{
11808
11809	switch (lun_type) {
11810	case T_PROCESSOR:
11811		if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11812		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11813			return (0);
11814		break;
11815	case T_DIRECT:
11816		if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11817		    ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11818			return (0);
11819		break;
11820	default:
11821		return (0);
11822	}
11823	return (1);
11824}
11825
11826static int
11827ctl_scsiio(struct ctl_scsiio *ctsio)
11828{
11829	int retval;
11830	const struct ctl_cmd_entry *entry;
11831
11832	retval = CTL_RETVAL_COMPLETE;
11833
11834	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11835
11836	entry = ctl_get_cmd_entry(ctsio, NULL);
11837
11838	/*
11839	 * If this I/O has been aborted, just send it straight to
11840	 * ctl_done() without executing it.
11841	 */
11842	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11843		ctl_done((union ctl_io *)ctsio);
11844		goto bailout;
11845	}
11846
11847	/*
11848	 * All the checks should have been handled by ctl_scsiio_precheck().
11849	 * We should be clear now to just execute the I/O.
11850	 */
11851	retval = entry->execute(ctsio);
11852
11853bailout:
11854	return (retval);
11855}
11856
11857/*
11858 * Since we only implement one target right now, a bus reset simply resets
11859 * our single target.
11860 */
11861static int
11862ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11863{
11864	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11865}
11866
11867static int
11868ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11869		 ctl_ua_type ua_type)
11870{
11871	struct ctl_lun *lun;
11872	int retval;
11873
11874	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11875		union ctl_ha_msg msg_info;
11876
11877		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11878		msg_info.hdr.nexus = io->io_hdr.nexus;
11879		if (ua_type==CTL_UA_TARG_RESET)
11880			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11881		else
11882			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11883		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11884		msg_info.hdr.original_sc = NULL;
11885		msg_info.hdr.serializing_sc = NULL;
11886		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11887		    (void *)&msg_info, sizeof(msg_info), 0)) {
11888		}
11889	}
11890	retval = 0;
11891
11892	mtx_lock(&ctl_softc->ctl_lock);
11893	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11894		retval += ctl_lun_reset(lun, io, ua_type);
11895	mtx_unlock(&ctl_softc->ctl_lock);
11896
11897	return (retval);
11898}
11899
11900/*
11901 * The LUN should always be set.  The I/O is optional, and is used to
11902 * distinguish between I/Os sent by this initiator, and by other
11903 * initiators.  We set unit attention for initiators other than this one.
11904 * SAM-3 is vague on this point.  It does say that a unit attention should
11905 * be established for other initiators when a LUN is reset (see section
11906 * 5.7.3), but it doesn't specifically say that the unit attention should
11907 * be established for this particular initiator when a LUN is reset.  Here
11908 * is the relevant text, from SAM-3 rev 8:
11909 *
11910 * 5.7.2 When a SCSI initiator port aborts its own tasks
11911 *
11912 * When a SCSI initiator port causes its own task(s) to be aborted, no
11913 * notification that the task(s) have been aborted shall be returned to
11914 * the SCSI initiator port other than the completion response for the
11915 * command or task management function action that caused the task(s) to
11916 * be aborted and notification(s) associated with related effects of the
11917 * action (e.g., a reset unit attention condition).
11918 *
11919 * XXX KDM for now, we're setting unit attention for all initiators.
11920 */
11921static int
11922ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11923{
11924	union ctl_io *xio;
11925#if 0
11926	uint32_t initidx;
11927#endif
11928#ifdef CTL_WITH_CA
11929	int i;
11930#endif
11931
11932	mtx_lock(&lun->lun_lock);
11933	/*
11934	 * Run through the OOA queue and abort each I/O.
11935	 */
11936#if 0
11937	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11938#endif
11939	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11940	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11941		xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11942	}
11943
11944	/*
11945	 * This version sets unit attention for every
11946	 */
11947#if 0
11948	initidx = ctl_get_initindex(&io->io_hdr.nexus);
11949	ctl_est_ua_all(lun, initidx, ua_type);
11950#else
11951	ctl_est_ua_all(lun, -1, ua_type);
11952#endif
11953
11954	/*
11955	 * A reset (any kind, really) clears reservations established with
11956	 * RESERVE/RELEASE.  It does not clear reservations established
11957	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11958	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11959	 * reservations made with the RESERVE/RELEASE commands, because
11960	 * those commands are obsolete in SPC-3.
11961	 */
11962	lun->flags &= ~CTL_LUN_RESERVED;
11963
11964#ifdef CTL_WITH_CA
11965	for (i = 0; i < CTL_MAX_INITIATORS; i++)
11966		ctl_clear_mask(lun->have_ca, i);
11967#endif
11968	mtx_unlock(&lun->lun_lock);
11969
11970	return (0);
11971}
11972
11973static void
11974ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11975    int other_sc)
11976{
11977	union ctl_io *xio;
11978
11979	mtx_assert(&lun->lun_lock, MA_OWNED);
11980
11981	/*
11982	 * Run through the OOA queue and attempt to find the given I/O.
11983	 * The target port, initiator ID, tag type and tag number have to
11984	 * match the values that we got from the initiator.  If we have an
11985	 * untagged command to abort, simply abort the first untagged command
11986	 * we come to.  We only allow one untagged command at a time of course.
11987	 */
11988	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11989	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11990
11991		if ((targ_port == UINT32_MAX ||
11992		     targ_port == xio->io_hdr.nexus.targ_port) &&
11993		    (init_id == UINT32_MAX ||
11994		     init_id == xio->io_hdr.nexus.initid.id)) {
11995			if (targ_port != xio->io_hdr.nexus.targ_port ||
11996			    init_id != xio->io_hdr.nexus.initid.id)
11997				xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11998			xio->io_hdr.flags |= CTL_FLAG_ABORT;
11999			if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12000				union ctl_ha_msg msg_info;
12001
12002				msg_info.hdr.nexus = xio->io_hdr.nexus;
12003				msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12004				msg_info.task.tag_num = xio->scsiio.tag_num;
12005				msg_info.task.tag_type = xio->scsiio.tag_type;
12006				msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12007				msg_info.hdr.original_sc = NULL;
12008				msg_info.hdr.serializing_sc = NULL;
12009				ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12010				    (void *)&msg_info, sizeof(msg_info), 0);
12011			}
12012		}
12013	}
12014}
12015
12016static int
12017ctl_abort_task_set(union ctl_io *io)
12018{
12019	struct ctl_softc *softc = control_softc;
12020	struct ctl_lun *lun;
12021	uint32_t targ_lun;
12022
12023	/*
12024	 * Look up the LUN.
12025	 */
12026	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12027	mtx_lock(&softc->ctl_lock);
12028	if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
12029		lun = softc->ctl_luns[targ_lun];
12030	else {
12031		mtx_unlock(&softc->ctl_lock);
12032		return (1);
12033	}
12034
12035	mtx_lock(&lun->lun_lock);
12036	mtx_unlock(&softc->ctl_lock);
12037	if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12038		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12039		    io->io_hdr.nexus.initid.id,
12040		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12041	} else { /* CTL_TASK_CLEAR_TASK_SET */
12042		ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12043		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12044	}
12045	mtx_unlock(&lun->lun_lock);
12046	return (0);
12047}
12048
12049static int
12050ctl_i_t_nexus_reset(union ctl_io *io)
12051{
12052	struct ctl_softc *softc = control_softc;
12053	struct ctl_lun *lun;
12054	uint32_t initidx, residx;
12055
12056	initidx = ctl_get_initindex(&io->io_hdr.nexus);
12057	residx = ctl_get_resindex(&io->io_hdr.nexus);
12058	mtx_lock(&softc->ctl_lock);
12059	STAILQ_FOREACH(lun, &softc->lun_list, links) {
12060		mtx_lock(&lun->lun_lock);
12061		ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12062		    io->io_hdr.nexus.initid.id,
12063		    (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12064#ifdef CTL_WITH_CA
12065		ctl_clear_mask(lun->have_ca, initidx);
12066#endif
12067		if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
12068			lun->flags &= ~CTL_LUN_RESERVED;
12069		ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12070		mtx_unlock(&lun->lun_lock);
12071	}
12072	mtx_unlock(&softc->ctl_lock);
12073	return (0);
12074}
12075
12076static int
12077ctl_abort_task(union ctl_io *io)
12078{
12079	union ctl_io *xio;
12080	struct ctl_lun *lun;
12081	struct ctl_softc *ctl_softc;
12082#if 0
12083	struct sbuf sb;
12084	char printbuf[128];
12085#endif
12086	int found;
12087	uint32_t targ_lun;
12088
12089	ctl_softc = control_softc;
12090	found = 0;
12091
12092	/*
12093	 * Look up the LUN.
12094	 */
12095	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12096	mtx_lock(&ctl_softc->ctl_lock);
12097	if ((targ_lun < CTL_MAX_LUNS)
12098	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12099		lun = ctl_softc->ctl_luns[targ_lun];
12100	else {
12101		mtx_unlock(&ctl_softc->ctl_lock);
12102		return (1);
12103	}
12104
12105#if 0
12106	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12107	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12108#endif
12109
12110	mtx_lock(&lun->lun_lock);
12111	mtx_unlock(&ctl_softc->ctl_lock);
12112	/*
12113	 * Run through the OOA queue and attempt to find the given I/O.
12114	 * The target port, initiator ID, tag type and tag number have to
12115	 * match the values that we got from the initiator.  If we have an
12116	 * untagged command to abort, simply abort the first untagged command
12117	 * we come to.  We only allow one untagged command at a time of course.
12118	 */
12119#if 0
12120	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
12121#endif
12122	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12123	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12124#if 0
12125		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12126
12127		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12128			    lun->lun, xio->scsiio.tag_num,
12129			    xio->scsiio.tag_type,
12130			    (xio->io_hdr.blocked_links.tqe_prev
12131			    == NULL) ? "" : " BLOCKED",
12132			    (xio->io_hdr.flags &
12133			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12134			    (xio->io_hdr.flags &
12135			    CTL_FLAG_ABORT) ? " ABORT" : "",
12136			    (xio->io_hdr.flags &
12137			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12138		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12139		sbuf_finish(&sb);
12140		printf("%s\n", sbuf_data(&sb));
12141#endif
12142
12143		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
12144		 && (xio->io_hdr.nexus.initid.id ==
12145		     io->io_hdr.nexus.initid.id)) {
12146			/*
12147			 * If the abort says that the task is untagged, the
12148			 * task in the queue must be untagged.  Otherwise,
12149			 * we just check to see whether the tag numbers
12150			 * match.  This is because the QLogic firmware
12151			 * doesn't pass back the tag type in an abort
12152			 * request.
12153			 */
12154#if 0
12155			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12156			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12157			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12158#endif
12159			/*
12160			 * XXX KDM we've got problems with FC, because it
12161			 * doesn't send down a tag type with aborts.  So we
12162			 * can only really go by the tag number...
12163			 * This may cause problems with parallel SCSI.
12164			 * Need to figure that out!!
12165			 */
12166			if (xio->scsiio.tag_num == io->taskio.tag_num) {
12167				xio->io_hdr.flags |= CTL_FLAG_ABORT;
12168				found = 1;
12169				if ((io->io_hdr.flags &
12170				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
12171				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12172					union ctl_ha_msg msg_info;
12173
12174					io->io_hdr.flags |=
12175					                CTL_FLAG_SENT_2OTHER_SC;
12176					msg_info.hdr.nexus = io->io_hdr.nexus;
12177					msg_info.task.task_action =
12178						CTL_TASK_ABORT_TASK;
12179					msg_info.task.tag_num =
12180						io->taskio.tag_num;
12181					msg_info.task.tag_type =
12182						io->taskio.tag_type;
12183					msg_info.hdr.msg_type =
12184						CTL_MSG_MANAGE_TASKS;
12185					msg_info.hdr.original_sc = NULL;
12186					msg_info.hdr.serializing_sc = NULL;
12187#if 0
12188					printf("Sent Abort to other side\n");
12189#endif
12190					if (CTL_HA_STATUS_SUCCESS !=
12191					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12192		    				(void *)&msg_info,
12193						sizeof(msg_info), 0)) {
12194					}
12195				}
12196#if 0
12197				printf("ctl_abort_task: found I/O to abort\n");
12198#endif
12199				break;
12200			}
12201		}
12202	}
12203	mtx_unlock(&lun->lun_lock);
12204
12205	if (found == 0) {
12206		/*
12207		 * This isn't really an error.  It's entirely possible for
12208		 * the abort and command completion to cross on the wire.
12209		 * This is more of an informative/diagnostic error.
12210		 */
12211#if 0
12212		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12213		       "%d:%d:%d:%d tag %d type %d\n",
12214		       io->io_hdr.nexus.initid.id,
12215		       io->io_hdr.nexus.targ_port,
12216		       io->io_hdr.nexus.targ_target.id,
12217		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12218		       io->taskio.tag_type);
12219#endif
12220	}
12221	return (0);
12222}
12223
12224static void
12225ctl_run_task(union ctl_io *io)
12226{
12227	struct ctl_softc *ctl_softc = control_softc;
12228	int retval = 1;
12229	const char *task_desc;
12230
12231	CTL_DEBUG_PRINT(("ctl_run_task\n"));
12232
12233	KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12234	    ("ctl_run_task: Unextected io_type %d\n",
12235	     io->io_hdr.io_type));
12236
12237	task_desc = ctl_scsi_task_string(&io->taskio);
12238	if (task_desc != NULL) {
12239#ifdef NEEDTOPORT
12240		csevent_log(CSC_CTL | CSC_SHELF_SW |
12241			    CTL_TASK_REPORT,
12242			    csevent_LogType_Trace,
12243			    csevent_Severity_Information,
12244			    csevent_AlertLevel_Green,
12245			    csevent_FRU_Firmware,
12246			    csevent_FRU_Unknown,
12247			    "CTL: received task: %s",task_desc);
12248#endif
12249	} else {
12250#ifdef NEEDTOPORT
12251		csevent_log(CSC_CTL | CSC_SHELF_SW |
12252			    CTL_TASK_REPORT,
12253			    csevent_LogType_Trace,
12254			    csevent_Severity_Information,
12255			    csevent_AlertLevel_Green,
12256			    csevent_FRU_Firmware,
12257			    csevent_FRU_Unknown,
12258			    "CTL: received unknown task "
12259			    "type: %d (%#x)",
12260			    io->taskio.task_action,
12261			    io->taskio.task_action);
12262#endif
12263	}
12264	switch (io->taskio.task_action) {
12265	case CTL_TASK_ABORT_TASK:
12266		retval = ctl_abort_task(io);
12267		break;
12268	case CTL_TASK_ABORT_TASK_SET:
12269	case CTL_TASK_CLEAR_TASK_SET:
12270		retval = ctl_abort_task_set(io);
12271		break;
12272	case CTL_TASK_CLEAR_ACA:
12273		break;
12274	case CTL_TASK_I_T_NEXUS_RESET:
12275		retval = ctl_i_t_nexus_reset(io);
12276		break;
12277	case CTL_TASK_LUN_RESET: {
12278		struct ctl_lun *lun;
12279		uint32_t targ_lun;
12280
12281		targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12282		mtx_lock(&ctl_softc->ctl_lock);
12283		if ((targ_lun < CTL_MAX_LUNS)
12284		 && (ctl_softc->ctl_luns[targ_lun] != NULL))
12285			lun = ctl_softc->ctl_luns[targ_lun];
12286		else {
12287			mtx_unlock(&ctl_softc->ctl_lock);
12288			retval = 1;
12289			break;
12290		}
12291
12292		if (!(io->io_hdr.flags &
12293		    CTL_FLAG_FROM_OTHER_SC)) {
12294			union ctl_ha_msg msg_info;
12295
12296			io->io_hdr.flags |=
12297				CTL_FLAG_SENT_2OTHER_SC;
12298			msg_info.hdr.msg_type =
12299				CTL_MSG_MANAGE_TASKS;
12300			msg_info.hdr.nexus = io->io_hdr.nexus;
12301			msg_info.task.task_action =
12302				CTL_TASK_LUN_RESET;
12303			msg_info.hdr.original_sc = NULL;
12304			msg_info.hdr.serializing_sc = NULL;
12305			if (CTL_HA_STATUS_SUCCESS !=
12306			    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
12307			    (void *)&msg_info,
12308			    sizeof(msg_info), 0)) {
12309			}
12310		}
12311
12312		retval = ctl_lun_reset(lun, io,
12313				       CTL_UA_LUN_RESET);
12314		mtx_unlock(&ctl_softc->ctl_lock);
12315		break;
12316	}
12317	case CTL_TASK_TARGET_RESET:
12318		retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET);
12319		break;
12320	case CTL_TASK_BUS_RESET:
12321		retval = ctl_bus_reset(ctl_softc, io);
12322		break;
12323	case CTL_TASK_PORT_LOGIN:
12324		break;
12325	case CTL_TASK_PORT_LOGOUT:
12326		break;
12327	default:
12328		printf("ctl_run_task: got unknown task management event %d\n",
12329		       io->taskio.task_action);
12330		break;
12331	}
12332	if (retval == 0)
12333		io->io_hdr.status = CTL_SUCCESS;
12334	else
12335		io->io_hdr.status = CTL_ERROR;
12336	ctl_done(io);
12337}
12338
12339/*
12340 * For HA operation.  Handle commands that come in from the other
12341 * controller.
12342 */
12343static void
12344ctl_handle_isc(union ctl_io *io)
12345{
12346	int free_io;
12347	struct ctl_lun *lun;
12348	struct ctl_softc *ctl_softc;
12349	uint32_t targ_lun;
12350
12351	ctl_softc = control_softc;
12352
12353	targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12354	lun = ctl_softc->ctl_luns[targ_lun];
12355
12356	switch (io->io_hdr.msg_type) {
12357	case CTL_MSG_SERIALIZE:
12358		free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12359		break;
12360	case CTL_MSG_R2R: {
12361		const struct ctl_cmd_entry *entry;
12362
12363		/*
12364		 * This is only used in SER_ONLY mode.
12365		 */
12366		free_io = 0;
12367		entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12368		mtx_lock(&lun->lun_lock);
12369		if (ctl_scsiio_lun_check(ctl_softc, lun,
12370		    entry, (struct ctl_scsiio *)io) != 0) {
12371			mtx_unlock(&lun->lun_lock);
12372			ctl_done(io);
12373			break;
12374		}
12375		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12376		mtx_unlock(&lun->lun_lock);
12377		ctl_enqueue_rtr(io);
12378		break;
12379	}
12380	case CTL_MSG_FINISH_IO:
12381		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
12382			free_io = 0;
12383			ctl_done(io);
12384		} else {
12385			free_io = 1;
12386			mtx_lock(&lun->lun_lock);
12387			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12388				     ooa_links);
12389			ctl_check_blocked(lun);
12390			mtx_unlock(&lun->lun_lock);
12391		}
12392		break;
12393	case CTL_MSG_PERS_ACTION:
12394		ctl_hndl_per_res_out_on_other_sc(
12395			(union ctl_ha_msg *)&io->presio.pr_msg);
12396		free_io = 1;
12397		break;
12398	case CTL_MSG_BAD_JUJU:
12399		free_io = 0;
12400		ctl_done(io);
12401		break;
12402	case CTL_MSG_DATAMOVE:
12403		/* Only used in XFER mode */
12404		free_io = 0;
12405		ctl_datamove_remote(io);
12406		break;
12407	case CTL_MSG_DATAMOVE_DONE:
12408		/* Only used in XFER mode */
12409		free_io = 0;
12410		io->scsiio.be_move_done(io);
12411		break;
12412	default:
12413		free_io = 1;
12414		printf("%s: Invalid message type %d\n",
12415		       __func__, io->io_hdr.msg_type);
12416		break;
12417	}
12418	if (free_io)
12419		ctl_free_io(io);
12420
12421}
12422
12423
12424/*
12425 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12426 * there is no match.
12427 */
12428static ctl_lun_error_pattern
12429ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12430{
12431	const struct ctl_cmd_entry *entry;
12432	ctl_lun_error_pattern filtered_pattern, pattern;
12433
12434	pattern = desc->error_pattern;
12435
12436	/*
12437	 * XXX KDM we need more data passed into this function to match a
12438	 * custom pattern, and we actually need to implement custom pattern
12439	 * matching.
12440	 */
12441	if (pattern & CTL_LUN_PAT_CMD)
12442		return (CTL_LUN_PAT_CMD);
12443
12444	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12445		return (CTL_LUN_PAT_ANY);
12446
12447	entry = ctl_get_cmd_entry(ctsio, NULL);
12448
12449	filtered_pattern = entry->pattern & pattern;
12450
12451	/*
12452	 * If the user requested specific flags in the pattern (e.g.
12453	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12454	 * flags.
12455	 *
12456	 * If the user did not specify any flags, it doesn't matter whether
12457	 * or not the command supports the flags.
12458	 */
12459	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12460	     (pattern & ~CTL_LUN_PAT_MASK))
12461		return (CTL_LUN_PAT_NONE);
12462
12463	/*
12464	 * If the user asked for a range check, see if the requested LBA
12465	 * range overlaps with this command's LBA range.
12466	 */
12467	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12468		uint64_t lba1;
12469		uint64_t len1;
12470		ctl_action action;
12471		int retval;
12472
12473		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12474		if (retval != 0)
12475			return (CTL_LUN_PAT_NONE);
12476
12477		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12478					      desc->lba_range.len, FALSE);
12479		/*
12480		 * A "pass" means that the LBA ranges don't overlap, so
12481		 * this doesn't match the user's range criteria.
12482		 */
12483		if (action == CTL_ACTION_PASS)
12484			return (CTL_LUN_PAT_NONE);
12485	}
12486
12487	return (filtered_pattern);
12488}
12489
12490static void
12491ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12492{
12493	struct ctl_error_desc *desc, *desc2;
12494
12495	mtx_assert(&lun->lun_lock, MA_OWNED);
12496
12497	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12498		ctl_lun_error_pattern pattern;
12499		/*
12500		 * Check to see whether this particular command matches
12501		 * the pattern in the descriptor.
12502		 */
12503		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12504		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12505			continue;
12506
12507		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12508		case CTL_LUN_INJ_ABORTED:
12509			ctl_set_aborted(&io->scsiio);
12510			break;
12511		case CTL_LUN_INJ_MEDIUM_ERR:
12512			ctl_set_medium_error(&io->scsiio);
12513			break;
12514		case CTL_LUN_INJ_UA:
12515			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12516			 * OCCURRED */
12517			ctl_set_ua(&io->scsiio, 0x29, 0x00);
12518			break;
12519		case CTL_LUN_INJ_CUSTOM:
12520			/*
12521			 * We're assuming the user knows what he is doing.
12522			 * Just copy the sense information without doing
12523			 * checks.
12524			 */
12525			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12526			      ctl_min(sizeof(desc->custom_sense),
12527				      sizeof(io->scsiio.sense_data)));
12528			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12529			io->scsiio.sense_len = SSD_FULL_SIZE;
12530			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12531			break;
12532		case CTL_LUN_INJ_NONE:
12533		default:
12534			/*
12535			 * If this is an error injection type we don't know
12536			 * about, clear the continuous flag (if it is set)
12537			 * so it will get deleted below.
12538			 */
12539			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12540			break;
12541		}
12542		/*
12543		 * By default, each error injection action is a one-shot
12544		 */
12545		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12546			continue;
12547
12548		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12549
12550		free(desc, M_CTL);
12551	}
12552}
12553
12554#ifdef CTL_IO_DELAY
12555static void
12556ctl_datamove_timer_wakeup(void *arg)
12557{
12558	union ctl_io *io;
12559
12560	io = (union ctl_io *)arg;
12561
12562	ctl_datamove(io);
12563}
12564#endif /* CTL_IO_DELAY */
12565
12566void
12567ctl_datamove(union ctl_io *io)
12568{
12569	void (*fe_datamove)(union ctl_io *io);
12570
12571	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12572
12573	CTL_DEBUG_PRINT(("ctl_datamove\n"));
12574
12575#ifdef CTL_TIME_IO
12576	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12577		char str[256];
12578		char path_str[64];
12579		struct sbuf sb;
12580
12581		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12582		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12583
12584		sbuf_cat(&sb, path_str);
12585		switch (io->io_hdr.io_type) {
12586		case CTL_IO_SCSI:
12587			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12588			sbuf_printf(&sb, "\n");
12589			sbuf_cat(&sb, path_str);
12590			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12591				    io->scsiio.tag_num, io->scsiio.tag_type);
12592			break;
12593		case CTL_IO_TASK:
12594			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12595				    "Tag Type: %d\n", io->taskio.task_action,
12596				    io->taskio.tag_num, io->taskio.tag_type);
12597			break;
12598		default:
12599			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12600			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12601			break;
12602		}
12603		sbuf_cat(&sb, path_str);
12604		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12605			    (intmax_t)time_uptime - io->io_hdr.start_time);
12606		sbuf_finish(&sb);
12607		printf("%s", sbuf_data(&sb));
12608	}
12609#endif /* CTL_TIME_IO */
12610
12611#ifdef CTL_IO_DELAY
12612	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12613		struct ctl_lun *lun;
12614
12615		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12616
12617		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12618	} else {
12619		struct ctl_lun *lun;
12620
12621		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12622		if ((lun != NULL)
12623		 && (lun->delay_info.datamove_delay > 0)) {
12624			struct callout *callout;
12625
12626			callout = (struct callout *)&io->io_hdr.timer_bytes;
12627			callout_init(callout, /*mpsafe*/ 1);
12628			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12629			callout_reset(callout,
12630				      lun->delay_info.datamove_delay * hz,
12631				      ctl_datamove_timer_wakeup, io);
12632			if (lun->delay_info.datamove_type ==
12633			    CTL_DELAY_TYPE_ONESHOT)
12634				lun->delay_info.datamove_delay = 0;
12635			return;
12636		}
12637	}
12638#endif
12639
12640	/*
12641	 * This command has been aborted.  Set the port status, so we fail
12642	 * the data move.
12643	 */
12644	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12645		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12646		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12647		       io->io_hdr.nexus.targ_port,
12648		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12649		       io->io_hdr.nexus.targ_lun);
12650		io->io_hdr.port_status = 31337;
12651		/*
12652		 * Note that the backend, in this case, will get the
12653		 * callback in its context.  In other cases it may get
12654		 * called in the frontend's interrupt thread context.
12655		 */
12656		io->scsiio.be_move_done(io);
12657		return;
12658	}
12659
12660	/* Don't confuse frontend with zero length data move. */
12661	if (io->scsiio.kern_data_len == 0) {
12662		io->scsiio.be_move_done(io);
12663		return;
12664	}
12665
12666	/*
12667	 * If we're in XFER mode and this I/O is from the other shelf
12668	 * controller, we need to send the DMA to the other side to
12669	 * actually transfer the data to/from the host.  In serialize only
12670	 * mode the transfer happens below CTL and ctl_datamove() is only
12671	 * called on the machine that originally received the I/O.
12672	 */
12673	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12674	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12675		union ctl_ha_msg msg;
12676		uint32_t sg_entries_sent;
12677		int do_sg_copy;
12678		int i;
12679
12680		memset(&msg, 0, sizeof(msg));
12681		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12682		msg.hdr.original_sc = io->io_hdr.original_sc;
12683		msg.hdr.serializing_sc = io;
12684		msg.hdr.nexus = io->io_hdr.nexus;
12685		msg.dt.flags = io->io_hdr.flags;
12686		/*
12687		 * We convert everything into a S/G list here.  We can't
12688		 * pass by reference, only by value between controllers.
12689		 * So we can't pass a pointer to the S/G list, only as many
12690		 * S/G entries as we can fit in here.  If it's possible for
12691		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12692		 * then we need to break this up into multiple transfers.
12693		 */
12694		if (io->scsiio.kern_sg_entries == 0) {
12695			msg.dt.kern_sg_entries = 1;
12696			/*
12697			 * If this is in cached memory, flush the cache
12698			 * before we send the DMA request to the other
12699			 * controller.  We want to do this in either the
12700			 * read or the write case.  The read case is
12701			 * straightforward.  In the write case, we want to
12702			 * make sure nothing is in the local cache that
12703			 * could overwrite the DMAed data.
12704			 */
12705			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12706				/*
12707				 * XXX KDM use bus_dmamap_sync() here.
12708				 */
12709			}
12710
12711			/*
12712			 * Convert to a physical address if this is a
12713			 * virtual address.
12714			 */
12715			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12716				msg.dt.sg_list[0].addr =
12717					io->scsiio.kern_data_ptr;
12718			} else {
12719				/*
12720				 * XXX KDM use busdma here!
12721				 */
12722#if 0
12723				msg.dt.sg_list[0].addr = (void *)
12724					vtophys(io->scsiio.kern_data_ptr);
12725#endif
12726			}
12727
12728			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12729			do_sg_copy = 0;
12730		} else {
12731			struct ctl_sg_entry *sgl;
12732
12733			do_sg_copy = 1;
12734			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12735			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12736			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12737				/*
12738				 * XXX KDM use bus_dmamap_sync() here.
12739				 */
12740			}
12741		}
12742
12743		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12744		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12745		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12746		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12747		msg.dt.sg_sequence = 0;
12748
12749		/*
12750		 * Loop until we've sent all of the S/G entries.  On the
12751		 * other end, we'll recompose these S/G entries into one
12752		 * contiguous list before passing it to the
12753		 */
12754		for (sg_entries_sent = 0; sg_entries_sent <
12755		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12756			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12757				sizeof(msg.dt.sg_list[0])),
12758				msg.dt.kern_sg_entries - sg_entries_sent);
12759
12760			if (do_sg_copy != 0) {
12761				struct ctl_sg_entry *sgl;
12762				int j;
12763
12764				sgl = (struct ctl_sg_entry *)
12765					io->scsiio.kern_data_ptr;
12766				/*
12767				 * If this is in cached memory, flush the cache
12768				 * before we send the DMA request to the other
12769				 * controller.  We want to do this in either
12770				 * the * read or the write case.  The read
12771				 * case is straightforward.  In the write
12772				 * case, we want to make sure nothing is
12773				 * in the local cache that could overwrite
12774				 * the DMAed data.
12775				 */
12776
12777				for (i = sg_entries_sent, j = 0;
12778				     i < msg.dt.cur_sg_entries; i++, j++) {
12779					if ((io->io_hdr.flags &
12780					     CTL_FLAG_NO_DATASYNC) == 0) {
12781						/*
12782						 * XXX KDM use bus_dmamap_sync()
12783						 */
12784					}
12785					if ((io->io_hdr.flags &
12786					     CTL_FLAG_BUS_ADDR) == 0) {
12787						/*
12788						 * XXX KDM use busdma.
12789						 */
12790#if 0
12791						msg.dt.sg_list[j].addr =(void *)
12792						       vtophys(sgl[i].addr);
12793#endif
12794					} else {
12795						msg.dt.sg_list[j].addr =
12796							sgl[i].addr;
12797					}
12798					msg.dt.sg_list[j].len = sgl[i].len;
12799				}
12800			}
12801
12802			sg_entries_sent += msg.dt.cur_sg_entries;
12803			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12804				msg.dt.sg_last = 1;
12805			else
12806				msg.dt.sg_last = 0;
12807
12808			/*
12809			 * XXX KDM drop and reacquire the lock here?
12810			 */
12811			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12812			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12813				/*
12814				 * XXX do something here.
12815				 */
12816			}
12817
12818			msg.dt.sent_sg_entries = sg_entries_sent;
12819		}
12820		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12821		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12822			ctl_failover_io(io, /*have_lock*/ 0);
12823
12824	} else {
12825
12826		/*
12827		 * Lookup the fe_datamove() function for this particular
12828		 * front end.
12829		 */
12830		fe_datamove =
12831		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12832
12833		fe_datamove(io);
12834	}
12835}
12836
12837static void
12838ctl_send_datamove_done(union ctl_io *io, int have_lock)
12839{
12840	union ctl_ha_msg msg;
12841	int isc_status;
12842
12843	memset(&msg, 0, sizeof(msg));
12844
12845	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12846	msg.hdr.original_sc = io;
12847	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12848	msg.hdr.nexus = io->io_hdr.nexus;
12849	msg.hdr.status = io->io_hdr.status;
12850	msg.scsi.tag_num = io->scsiio.tag_num;
12851	msg.scsi.tag_type = io->scsiio.tag_type;
12852	msg.scsi.scsi_status = io->scsiio.scsi_status;
12853	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12854	       sizeof(io->scsiio.sense_data));
12855	msg.scsi.sense_len = io->scsiio.sense_len;
12856	msg.scsi.sense_residual = io->scsiio.sense_residual;
12857	msg.scsi.fetd_status = io->io_hdr.port_status;
12858	msg.scsi.residual = io->scsiio.residual;
12859	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12860
12861	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12862		ctl_failover_io(io, /*have_lock*/ have_lock);
12863		return;
12864	}
12865
12866	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12867	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12868		/* XXX do something if this fails */
12869	}
12870
12871}
12872
12873/*
12874 * The DMA to the remote side is done, now we need to tell the other side
12875 * we're done so it can continue with its data movement.
12876 */
12877static void
12878ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12879{
12880	union ctl_io *io;
12881
12882	io = rq->context;
12883
12884	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12885		printf("%s: ISC DMA write failed with error %d", __func__,
12886		       rq->ret);
12887		ctl_set_internal_failure(&io->scsiio,
12888					 /*sks_valid*/ 1,
12889					 /*retry_count*/ rq->ret);
12890	}
12891
12892	ctl_dt_req_free(rq);
12893
12894	/*
12895	 * In this case, we had to malloc the memory locally.  Free it.
12896	 */
12897	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12898		int i;
12899		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12900			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12901	}
12902	/*
12903	 * The data is in local and remote memory, so now we need to send
12904	 * status (good or back) back to the other side.
12905	 */
12906	ctl_send_datamove_done(io, /*have_lock*/ 0);
12907}
12908
12909/*
12910 * We've moved the data from the host/controller into local memory.  Now we
12911 * need to push it over to the remote controller's memory.
12912 */
12913static int
12914ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12915{
12916	int retval;
12917
12918	retval = 0;
12919
12920	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12921					  ctl_datamove_remote_write_cb);
12922
12923	return (retval);
12924}
12925
12926static void
12927ctl_datamove_remote_write(union ctl_io *io)
12928{
12929	int retval;
12930	void (*fe_datamove)(union ctl_io *io);
12931
12932	/*
12933	 * - Get the data from the host/HBA into local memory.
12934	 * - DMA memory from the local controller to the remote controller.
12935	 * - Send status back to the remote controller.
12936	 */
12937
12938	retval = ctl_datamove_remote_sgl_setup(io);
12939	if (retval != 0)
12940		return;
12941
12942	/* Switch the pointer over so the FETD knows what to do */
12943	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12944
12945	/*
12946	 * Use a custom move done callback, since we need to send completion
12947	 * back to the other controller, not to the backend on this side.
12948	 */
12949	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
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	return;
12956
12957}
12958
12959static int
12960ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12961{
12962#if 0
12963	char str[256];
12964	char path_str[64];
12965	struct sbuf sb;
12966#endif
12967
12968	/*
12969	 * In this case, we had to malloc the memory locally.  Free it.
12970	 */
12971	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12972		int i;
12973		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12974			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12975	}
12976
12977#if 0
12978	scsi_path_string(io, path_str, sizeof(path_str));
12979	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12980	sbuf_cat(&sb, path_str);
12981	scsi_command_string(&io->scsiio, NULL, &sb);
12982	sbuf_printf(&sb, "\n");
12983	sbuf_cat(&sb, path_str);
12984	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12985		    io->scsiio.tag_num, io->scsiio.tag_type);
12986	sbuf_cat(&sb, path_str);
12987	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12988		    io->io_hdr.flags, io->io_hdr.status);
12989	sbuf_finish(&sb);
12990	printk("%s", sbuf_data(&sb));
12991#endif
12992
12993
12994	/*
12995	 * The read is done, now we need to send status (good or bad) back
12996	 * to the other side.
12997	 */
12998	ctl_send_datamove_done(io, /*have_lock*/ 0);
12999
13000	return (0);
13001}
13002
13003static void
13004ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
13005{
13006	union ctl_io *io;
13007	void (*fe_datamove)(union ctl_io *io);
13008
13009	io = rq->context;
13010
13011	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
13012		printf("%s: ISC DMA read failed with error %d", __func__,
13013		       rq->ret);
13014		ctl_set_internal_failure(&io->scsiio,
13015					 /*sks_valid*/ 1,
13016					 /*retry_count*/ rq->ret);
13017	}
13018
13019	ctl_dt_req_free(rq);
13020
13021	/* Switch the pointer over so the FETD knows what to do */
13022	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
13023
13024	/*
13025	 * Use a custom move done callback, since we need to send completion
13026	 * back to the other controller, not to the backend on this side.
13027	 */
13028	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
13029
13030	/* XXX KDM add checks like the ones in ctl_datamove? */
13031
13032	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
13033
13034	fe_datamove(io);
13035}
13036
13037static int
13038ctl_datamove_remote_sgl_setup(union ctl_io *io)
13039{
13040	struct ctl_sg_entry *local_sglist, *remote_sglist;
13041	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
13042	struct ctl_softc *softc;
13043	int retval;
13044	int i;
13045
13046	retval = 0;
13047	softc = control_softc;
13048
13049	local_sglist = io->io_hdr.local_sglist;
13050	local_dma_sglist = io->io_hdr.local_dma_sglist;
13051	remote_sglist = io->io_hdr.remote_sglist;
13052	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13053
13054	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
13055		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
13056			local_sglist[i].len = remote_sglist[i].len;
13057
13058			/*
13059			 * XXX Detect the situation where the RS-level I/O
13060			 * redirector on the other side has already read the
13061			 * data off of the AOR RS on this side, and
13062			 * transferred it to remote (mirror) memory on the
13063			 * other side.  Since we already have the data in
13064			 * memory here, we just need to use it.
13065			 *
13066			 * XXX KDM this can probably be removed once we
13067			 * get the cache device code in and take the
13068			 * current AOR implementation out.
13069			 */
13070#ifdef NEEDTOPORT
13071			if ((remote_sglist[i].addr >=
13072			     (void *)vtophys(softc->mirr->addr))
13073			 && (remote_sglist[i].addr <
13074			     ((void *)vtophys(softc->mirr->addr) +
13075			     CacheMirrorOffset))) {
13076				local_sglist[i].addr = remote_sglist[i].addr -
13077					CacheMirrorOffset;
13078				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13079				     CTL_FLAG_DATA_IN)
13080					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
13081			} else {
13082				local_sglist[i].addr = remote_sglist[i].addr +
13083					CacheMirrorOffset;
13084			}
13085#endif
13086#if 0
13087			printf("%s: local %p, remote %p, len %d\n",
13088			       __func__, local_sglist[i].addr,
13089			       remote_sglist[i].addr, local_sglist[i].len);
13090#endif
13091		}
13092	} else {
13093		uint32_t len_to_go;
13094
13095		/*
13096		 * In this case, we don't have automatically allocated
13097		 * memory for this I/O on this controller.  This typically
13098		 * happens with internal CTL I/O -- e.g. inquiry, mode
13099		 * sense, etc.  Anything coming from RAIDCore will have
13100		 * a mirror area available.
13101		 */
13102		len_to_go = io->scsiio.kern_data_len;
13103
13104		/*
13105		 * Clear the no datasync flag, we have to use malloced
13106		 * buffers.
13107		 */
13108		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
13109
13110		/*
13111		 * The difficult thing here is that the size of the various
13112		 * S/G segments may be different than the size from the
13113		 * remote controller.  That'll make it harder when DMAing
13114		 * the data back to the other side.
13115		 */
13116		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
13117		     sizeof(io->io_hdr.remote_sglist[0])) &&
13118		     (len_to_go > 0); i++) {
13119			local_sglist[i].len = ctl_min(len_to_go, 131072);
13120			CTL_SIZE_8B(local_dma_sglist[i].len,
13121				    local_sglist[i].len);
13122			local_sglist[i].addr =
13123				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
13124
13125			local_dma_sglist[i].addr = local_sglist[i].addr;
13126
13127			if (local_sglist[i].addr == NULL) {
13128				int j;
13129
13130				printf("malloc failed for %zd bytes!",
13131				       local_dma_sglist[i].len);
13132				for (j = 0; j < i; j++) {
13133					free(local_sglist[j].addr, M_CTL);
13134				}
13135				ctl_set_internal_failure(&io->scsiio,
13136							 /*sks_valid*/ 1,
13137							 /*retry_count*/ 4857);
13138				retval = 1;
13139				goto bailout_error;
13140
13141			}
13142			/* XXX KDM do we need a sync here? */
13143
13144			len_to_go -= local_sglist[i].len;
13145		}
13146		/*
13147		 * Reset the number of S/G entries accordingly.  The
13148		 * original number of S/G entries is available in
13149		 * rem_sg_entries.
13150		 */
13151		io->scsiio.kern_sg_entries = i;
13152
13153#if 0
13154		printf("%s: kern_sg_entries = %d\n", __func__,
13155		       io->scsiio.kern_sg_entries);
13156		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13157			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
13158			       local_sglist[i].addr, local_sglist[i].len,
13159			       local_dma_sglist[i].len);
13160#endif
13161	}
13162
13163
13164	return (retval);
13165
13166bailout_error:
13167
13168	ctl_send_datamove_done(io, /*have_lock*/ 0);
13169
13170	return (retval);
13171}
13172
13173static int
13174ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
13175			 ctl_ha_dt_cb callback)
13176{
13177	struct ctl_ha_dt_req *rq;
13178	struct ctl_sg_entry *remote_sglist, *local_sglist;
13179	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
13180	uint32_t local_used, remote_used, total_used;
13181	int retval;
13182	int i, j;
13183
13184	retval = 0;
13185
13186	rq = ctl_dt_req_alloc();
13187
13188	/*
13189	 * If we failed to allocate the request, and if the DMA didn't fail
13190	 * anyway, set busy status.  This is just a resource allocation
13191	 * failure.
13192	 */
13193	if ((rq == NULL)
13194	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
13195		ctl_set_busy(&io->scsiio);
13196
13197	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
13198
13199		if (rq != NULL)
13200			ctl_dt_req_free(rq);
13201
13202		/*
13203		 * The data move failed.  We need to return status back
13204		 * to the other controller.  No point in trying to DMA
13205		 * data to the remote controller.
13206		 */
13207
13208		ctl_send_datamove_done(io, /*have_lock*/ 0);
13209
13210		retval = 1;
13211
13212		goto bailout;
13213	}
13214
13215	local_sglist = io->io_hdr.local_sglist;
13216	local_dma_sglist = io->io_hdr.local_dma_sglist;
13217	remote_sglist = io->io_hdr.remote_sglist;
13218	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
13219	local_used = 0;
13220	remote_used = 0;
13221	total_used = 0;
13222
13223	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
13224		rq->ret = CTL_HA_STATUS_SUCCESS;
13225		rq->context = io;
13226		callback(rq);
13227		goto bailout;
13228	}
13229
13230	/*
13231	 * Pull/push the data over the wire from/to the other controller.
13232	 * This takes into account the possibility that the local and
13233	 * remote sglists may not be identical in terms of the size of
13234	 * the elements and the number of elements.
13235	 *
13236	 * One fundamental assumption here is that the length allocated for
13237	 * both the local and remote sglists is identical.  Otherwise, we've
13238	 * essentially got a coding error of some sort.
13239	 */
13240	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13241		int isc_ret;
13242		uint32_t cur_len, dma_length;
13243		uint8_t *tmp_ptr;
13244
13245		rq->id = CTL_HA_DATA_CTL;
13246		rq->command = command;
13247		rq->context = io;
13248
13249		/*
13250		 * Both pointers should be aligned.  But it is possible
13251		 * that the allocation length is not.  They should both
13252		 * also have enough slack left over at the end, though,
13253		 * to round up to the next 8 byte boundary.
13254		 */
13255		cur_len = ctl_min(local_sglist[i].len - local_used,
13256				  remote_sglist[j].len - remote_used);
13257
13258		/*
13259		 * In this case, we have a size issue and need to decrease
13260		 * the size, except in the case where we actually have less
13261		 * than 8 bytes left.  In that case, we need to increase
13262		 * the DMA length to get the last bit.
13263		 */
13264		if ((cur_len & 0x7) != 0) {
13265			if (cur_len > 0x7) {
13266				cur_len = cur_len - (cur_len & 0x7);
13267				dma_length = cur_len;
13268			} else {
13269				CTL_SIZE_8B(dma_length, cur_len);
13270			}
13271
13272		} else
13273			dma_length = cur_len;
13274
13275		/*
13276		 * If we had to allocate memory for this I/O, instead of using
13277		 * the non-cached mirror memory, we'll need to flush the cache
13278		 * before trying to DMA to the other controller.
13279		 *
13280		 * We could end up doing this multiple times for the same
13281		 * segment if we have a larger local segment than remote
13282		 * segment.  That shouldn't be an issue.
13283		 */
13284		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
13285			/*
13286			 * XXX KDM use bus_dmamap_sync() here.
13287			 */
13288		}
13289
13290		rq->size = dma_length;
13291
13292		tmp_ptr = (uint8_t *)local_sglist[i].addr;
13293		tmp_ptr += local_used;
13294
13295		/* Use physical addresses when talking to ISC hardware */
13296		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13297			/* XXX KDM use busdma */
13298#if 0
13299			rq->local = vtophys(tmp_ptr);
13300#endif
13301		} else
13302			rq->local = tmp_ptr;
13303
13304		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13305		tmp_ptr += remote_used;
13306		rq->remote = tmp_ptr;
13307
13308		rq->callback = NULL;
13309
13310		local_used += cur_len;
13311		if (local_used >= local_sglist[i].len) {
13312			i++;
13313			local_used = 0;
13314		}
13315
13316		remote_used += cur_len;
13317		if (remote_used >= remote_sglist[j].len) {
13318			j++;
13319			remote_used = 0;
13320		}
13321		total_used += cur_len;
13322
13323		if (total_used >= io->scsiio.kern_data_len)
13324			rq->callback = callback;
13325
13326		if ((rq->size & 0x7) != 0) {
13327			printf("%s: warning: size %d is not on 8b boundary\n",
13328			       __func__, rq->size);
13329		}
13330		if (((uintptr_t)rq->local & 0x7) != 0) {
13331			printf("%s: warning: local %p not on 8b boundary\n",
13332			       __func__, rq->local);
13333		}
13334		if (((uintptr_t)rq->remote & 0x7) != 0) {
13335			printf("%s: warning: remote %p not on 8b boundary\n",
13336			       __func__, rq->local);
13337		}
13338#if 0
13339		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
13340		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13341		       rq->local, rq->remote, rq->size);
13342#endif
13343
13344		isc_ret = ctl_dt_single(rq);
13345		if (isc_ret == CTL_HA_STATUS_WAIT)
13346			continue;
13347
13348		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
13349			rq->ret = CTL_HA_STATUS_SUCCESS;
13350		} else {
13351			rq->ret = isc_ret;
13352		}
13353		callback(rq);
13354		goto bailout;
13355	}
13356
13357bailout:
13358	return (retval);
13359
13360}
13361
13362static void
13363ctl_datamove_remote_read(union ctl_io *io)
13364{
13365	int retval;
13366	int i;
13367
13368	/*
13369	 * This will send an error to the other controller in the case of a
13370	 * failure.
13371	 */
13372	retval = ctl_datamove_remote_sgl_setup(io);
13373	if (retval != 0)
13374		return;
13375
13376	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13377					  ctl_datamove_remote_read_cb);
13378	if ((retval != 0)
13379	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
13380		/*
13381		 * Make sure we free memory if there was an error..  The
13382		 * ctl_datamove_remote_xfer() function will send the
13383		 * datamove done message, or call the callback with an
13384		 * error if there is a problem.
13385		 */
13386		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13387			free(io->io_hdr.local_sglist[i].addr, M_CTL);
13388	}
13389
13390	return;
13391}
13392
13393/*
13394 * Process a datamove request from the other controller.  This is used for
13395 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13396 * first.  Once that is complete, the data gets DMAed into the remote
13397 * controller's memory.  For reads, we DMA from the remote controller's
13398 * memory into our memory first, and then move it out to the FETD.
13399 */
13400static void
13401ctl_datamove_remote(union ctl_io *io)
13402{
13403	struct ctl_softc *softc;
13404
13405	softc = control_softc;
13406
13407	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
13408
13409	/*
13410	 * Note that we look for an aborted I/O here, but don't do some of
13411	 * the other checks that ctl_datamove() normally does.
13412	 * We don't need to run the datamove delay code, since that should
13413	 * have been done if need be on the other controller.
13414	 */
13415	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13416		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
13417		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
13418		       io->io_hdr.nexus.targ_port,
13419		       io->io_hdr.nexus.targ_target.id,
13420		       io->io_hdr.nexus.targ_lun);
13421		io->io_hdr.port_status = 31338;
13422		ctl_send_datamove_done(io, /*have_lock*/ 0);
13423		return;
13424	}
13425
13426	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
13427		ctl_datamove_remote_write(io);
13428	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
13429		ctl_datamove_remote_read(io);
13430	} else {
13431		union ctl_ha_msg msg;
13432		struct scsi_sense_data *sense;
13433		uint8_t sks[3];
13434		int retry_count;
13435
13436		memset(&msg, 0, sizeof(msg));
13437
13438		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
13439		msg.hdr.status = CTL_SCSI_ERROR;
13440		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
13441
13442		retry_count = 4243;
13443
13444		sense = &msg.scsi.sense_data;
13445		sks[0] = SSD_SCS_VALID;
13446		sks[1] = (retry_count >> 8) & 0xff;
13447		sks[2] = retry_count & 0xff;
13448
13449		/* "Internal target failure" */
13450		scsi_set_sense_data(sense,
13451				    /*sense_format*/ SSD_TYPE_NONE,
13452				    /*current_error*/ 1,
13453				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
13454				    /*asc*/ 0x44,
13455				    /*ascq*/ 0x00,
13456				    /*type*/ SSD_ELEM_SKS,
13457				    /*size*/ sizeof(sks),
13458				    /*data*/ sks,
13459				    SSD_ELEM_NONE);
13460
13461		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
13462		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13463			ctl_failover_io(io, /*have_lock*/ 1);
13464			return;
13465		}
13466
13467		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
13468		    CTL_HA_STATUS_SUCCESS) {
13469			/* XXX KDM what to do if this fails? */
13470		}
13471		return;
13472	}
13473
13474}
13475
13476static int
13477ctl_process_done(union ctl_io *io)
13478{
13479	struct ctl_lun *lun;
13480	struct ctl_softc *ctl_softc = control_softc;
13481	void (*fe_done)(union ctl_io *io);
13482	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
13483
13484	CTL_DEBUG_PRINT(("ctl_process_done\n"));
13485
13486	fe_done =
13487	    control_softc->ctl_ports[targ_port]->fe_done;
13488
13489#ifdef CTL_TIME_IO
13490	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13491		char str[256];
13492		char path_str[64];
13493		struct sbuf sb;
13494
13495		ctl_scsi_path_string(io, path_str, sizeof(path_str));
13496		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13497
13498		sbuf_cat(&sb, path_str);
13499		switch (io->io_hdr.io_type) {
13500		case CTL_IO_SCSI:
13501			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13502			sbuf_printf(&sb, "\n");
13503			sbuf_cat(&sb, path_str);
13504			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13505				    io->scsiio.tag_num, io->scsiio.tag_type);
13506			break;
13507		case CTL_IO_TASK:
13508			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13509				    "Tag Type: %d\n", io->taskio.task_action,
13510				    io->taskio.tag_num, io->taskio.tag_type);
13511			break;
13512		default:
13513			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13514			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
13515			break;
13516		}
13517		sbuf_cat(&sb, path_str);
13518		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13519			    (intmax_t)time_uptime - io->io_hdr.start_time);
13520		sbuf_finish(&sb);
13521		printf("%s", sbuf_data(&sb));
13522	}
13523#endif /* CTL_TIME_IO */
13524
13525	switch (io->io_hdr.io_type) {
13526	case CTL_IO_SCSI:
13527		break;
13528	case CTL_IO_TASK:
13529		if (bootverbose || (ctl_debug & CTL_DEBUG_INFO))
13530			ctl_io_error_print(io, NULL);
13531		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
13532			ctl_free_io(io);
13533		else
13534			fe_done(io);
13535		return (CTL_RETVAL_COMPLETE);
13536	default:
13537		panic("ctl_process_done: invalid io type %d\n",
13538		      io->io_hdr.io_type);
13539		break; /* NOTREACHED */
13540	}
13541
13542	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13543	if (lun == NULL) {
13544		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13545				 io->io_hdr.nexus.targ_mapped_lun));
13546		goto bailout;
13547	}
13548
13549	mtx_lock(&lun->lun_lock);
13550
13551	/*
13552	 * Check to see if we have any errors to inject here.  We only
13553	 * inject errors for commands that don't already have errors set.
13554	 */
13555	if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
13556	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13557	    ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13558		ctl_inject_error(lun, io);
13559
13560	/*
13561	 * XXX KDM how do we treat commands that aren't completed
13562	 * successfully?
13563	 *
13564	 * XXX KDM should we also track I/O latency?
13565	 */
13566	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13567	    io->io_hdr.io_type == CTL_IO_SCSI) {
13568#ifdef CTL_TIME_IO
13569		struct bintime cur_bt;
13570#endif
13571		int type;
13572
13573		if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13574		    CTL_FLAG_DATA_IN)
13575			type = CTL_STATS_READ;
13576		else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13577		    CTL_FLAG_DATA_OUT)
13578			type = CTL_STATS_WRITE;
13579		else
13580			type = CTL_STATS_NO_IO;
13581
13582		lun->stats.ports[targ_port].bytes[type] +=
13583		    io->scsiio.kern_total_len;
13584		lun->stats.ports[targ_port].operations[type]++;
13585#ifdef CTL_TIME_IO
13586		bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13587		   &io->io_hdr.dma_bt);
13588		lun->stats.ports[targ_port].num_dmas[type] +=
13589		    io->io_hdr.num_dmas;
13590		getbintime(&cur_bt);
13591		bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13592		bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13593#endif
13594	}
13595
13596	/*
13597	 * Remove this from the OOA queue.
13598	 */
13599	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13600
13601	/*
13602	 * Run through the blocked queue on this LUN and see if anything
13603	 * has become unblocked, now that this transaction is done.
13604	 */
13605	ctl_check_blocked(lun);
13606
13607	/*
13608	 * If the LUN has been invalidated, free it if there is nothing
13609	 * left on its OOA queue.
13610	 */
13611	if ((lun->flags & CTL_LUN_INVALID)
13612	 && TAILQ_EMPTY(&lun->ooa_queue)) {
13613		mtx_unlock(&lun->lun_lock);
13614		mtx_lock(&ctl_softc->ctl_lock);
13615		ctl_free_lun(lun);
13616		mtx_unlock(&ctl_softc->ctl_lock);
13617	} else
13618		mtx_unlock(&lun->lun_lock);
13619
13620bailout:
13621
13622	/*
13623	 * If this command has been aborted, make sure we set the status
13624	 * properly.  The FETD is responsible for freeing the I/O and doing
13625	 * whatever it needs to do to clean up its state.
13626	 */
13627	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13628		ctl_set_task_aborted(&io->scsiio);
13629
13630	/*
13631	 * If enabled, print command error status.
13632	 * We don't print UAs unless debugging was enabled explicitly.
13633	 */
13634	do {
13635		if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)
13636			break;
13637		if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0)
13638			break;
13639		if ((ctl_debug & CTL_DEBUG_INFO) == 0 &&
13640		    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) &&
13641		     (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13642			int error_code, sense_key, asc, ascq;
13643
13644			scsi_extract_sense_len(&io->scsiio.sense_data,
13645			    io->scsiio.sense_len, &error_code, &sense_key,
13646			    &asc, &ascq, /*show_errors*/ 0);
13647			if (sense_key == SSD_KEY_UNIT_ATTENTION)
13648				break;
13649		}
13650
13651		ctl_io_error_print(io, NULL);
13652	} while (0);
13653
13654	/*
13655	 * Tell the FETD or the other shelf controller we're done with this
13656	 * command.  Note that only SCSI commands get to this point.  Task
13657	 * management commands are completed above.
13658	 *
13659	 * We only send status to the other controller if we're in XFER
13660	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13661	 * received the I/O (from CTL's perspective), and so the status is
13662	 * generated there.
13663	 *
13664	 * XXX KDM if we hold the lock here, we could cause a deadlock
13665	 * if the frontend comes back in in this context to queue
13666	 * something.
13667	 */
13668	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13669	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13670		union ctl_ha_msg msg;
13671
13672		memset(&msg, 0, sizeof(msg));
13673		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13674		msg.hdr.original_sc = io->io_hdr.original_sc;
13675		msg.hdr.nexus = io->io_hdr.nexus;
13676		msg.hdr.status = io->io_hdr.status;
13677		msg.scsi.scsi_status = io->scsiio.scsi_status;
13678		msg.scsi.tag_num = io->scsiio.tag_num;
13679		msg.scsi.tag_type = io->scsiio.tag_type;
13680		msg.scsi.sense_len = io->scsiio.sense_len;
13681		msg.scsi.sense_residual = io->scsiio.sense_residual;
13682		msg.scsi.residual = io->scsiio.residual;
13683		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13684		       sizeof(io->scsiio.sense_data));
13685		/*
13686		 * We copy this whether or not this is an I/O-related
13687		 * command.  Otherwise, we'd have to go and check to see
13688		 * whether it's a read/write command, and it really isn't
13689		 * worth it.
13690		 */
13691		memcpy(&msg.scsi.lbalen,
13692		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13693		       sizeof(msg.scsi.lbalen));
13694
13695		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13696				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13697			/* XXX do something here */
13698		}
13699
13700		ctl_free_io(io);
13701	} else
13702		fe_done(io);
13703
13704	return (CTL_RETVAL_COMPLETE);
13705}
13706
13707#ifdef CTL_WITH_CA
13708/*
13709 * Front end should call this if it doesn't do autosense.  When the request
13710 * sense comes back in from the initiator, we'll dequeue this and send it.
13711 */
13712int
13713ctl_queue_sense(union ctl_io *io)
13714{
13715	struct ctl_lun *lun;
13716	struct ctl_softc *ctl_softc;
13717	uint32_t initidx, targ_lun;
13718
13719	ctl_softc = control_softc;
13720
13721	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13722
13723	/*
13724	 * LUN lookup will likely move to the ctl_work_thread() once we
13725	 * have our new queueing infrastructure (that doesn't put things on
13726	 * a per-LUN queue initially).  That is so that we can handle
13727	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13728	 * can't deal with that right now.
13729	 */
13730	mtx_lock(&ctl_softc->ctl_lock);
13731
13732	/*
13733	 * If we don't have a LUN for this, just toss the sense
13734	 * information.
13735	 */
13736	targ_lun = io->io_hdr.nexus.targ_lun;
13737	targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun);
13738	if ((targ_lun < CTL_MAX_LUNS)
13739	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13740		lun = ctl_softc->ctl_luns[targ_lun];
13741	else
13742		goto bailout;
13743
13744	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13745
13746	mtx_lock(&lun->lun_lock);
13747	/*
13748	 * Already have CA set for this LUN...toss the sense information.
13749	 */
13750	if (ctl_is_set(lun->have_ca, initidx)) {
13751		mtx_unlock(&lun->lun_lock);
13752		goto bailout;
13753	}
13754
13755	memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13756	       ctl_min(sizeof(lun->pending_sense[initidx]),
13757	       sizeof(io->scsiio.sense_data)));
13758	ctl_set_mask(lun->have_ca, initidx);
13759	mtx_unlock(&lun->lun_lock);
13760
13761bailout:
13762	mtx_unlock(&ctl_softc->ctl_lock);
13763
13764	ctl_free_io(io);
13765
13766	return (CTL_RETVAL_COMPLETE);
13767}
13768#endif
13769
13770/*
13771 * Primary command inlet from frontend ports.  All SCSI and task I/O
13772 * requests must go through this function.
13773 */
13774int
13775ctl_queue(union ctl_io *io)
13776{
13777	struct ctl_softc *ctl_softc;
13778
13779	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13780
13781	ctl_softc = control_softc;
13782
13783#ifdef CTL_TIME_IO
13784	io->io_hdr.start_time = time_uptime;
13785	getbintime(&io->io_hdr.start_bt);
13786#endif /* CTL_TIME_IO */
13787
13788	/* Map FE-specific LUN ID into global one. */
13789	io->io_hdr.nexus.targ_mapped_lun =
13790	    ctl_map_lun(io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun);
13791
13792	switch (io->io_hdr.io_type) {
13793	case CTL_IO_SCSI:
13794	case CTL_IO_TASK:
13795		if (ctl_debug & CTL_DEBUG_CDB)
13796			ctl_io_print(io);
13797		ctl_enqueue_incoming(io);
13798		break;
13799	default:
13800		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13801		return (EINVAL);
13802	}
13803
13804	return (CTL_RETVAL_COMPLETE);
13805}
13806
13807#ifdef CTL_IO_DELAY
13808static void
13809ctl_done_timer_wakeup(void *arg)
13810{
13811	union ctl_io *io;
13812
13813	io = (union ctl_io *)arg;
13814	ctl_done(io);
13815}
13816#endif /* CTL_IO_DELAY */
13817
13818void
13819ctl_done(union ctl_io *io)
13820{
13821	struct ctl_softc *ctl_softc;
13822
13823	ctl_softc = control_softc;
13824
13825	/*
13826	 * Enable this to catch duplicate completion issues.
13827	 */
13828#if 0
13829	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13830		printf("%s: type %d msg %d cdb %x iptl: "
13831		       "%d:%d:%d:%d tag 0x%04x "
13832		       "flag %#x status %x\n",
13833			__func__,
13834			io->io_hdr.io_type,
13835			io->io_hdr.msg_type,
13836			io->scsiio.cdb[0],
13837			io->io_hdr.nexus.initid.id,
13838			io->io_hdr.nexus.targ_port,
13839			io->io_hdr.nexus.targ_target.id,
13840			io->io_hdr.nexus.targ_lun,
13841			(io->io_hdr.io_type ==
13842			CTL_IO_TASK) ?
13843			io->taskio.tag_num :
13844			io->scsiio.tag_num,
13845		        io->io_hdr.flags,
13846			io->io_hdr.status);
13847	} else
13848		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13849#endif
13850
13851	/*
13852	 * This is an internal copy of an I/O, and should not go through
13853	 * the normal done processing logic.
13854	 */
13855	if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13856		return;
13857
13858	/*
13859	 * We need to send a msg to the serializing shelf to finish the IO
13860	 * as well.  We don't send a finish message to the other shelf if
13861	 * this is a task management command.  Task management commands
13862	 * aren't serialized in the OOA queue, but rather just executed on
13863	 * both shelf controllers for commands that originated on that
13864	 * controller.
13865	 */
13866	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13867	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13868		union ctl_ha_msg msg_io;
13869
13870		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13871		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13872		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13873		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13874		}
13875		/* continue on to finish IO */
13876	}
13877#ifdef CTL_IO_DELAY
13878	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13879		struct ctl_lun *lun;
13880
13881		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13882
13883		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13884	} else {
13885		struct ctl_lun *lun;
13886
13887		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13888
13889		if ((lun != NULL)
13890		 && (lun->delay_info.done_delay > 0)) {
13891			struct callout *callout;
13892
13893			callout = (struct callout *)&io->io_hdr.timer_bytes;
13894			callout_init(callout, /*mpsafe*/ 1);
13895			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13896			callout_reset(callout,
13897				      lun->delay_info.done_delay * hz,
13898				      ctl_done_timer_wakeup, io);
13899			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13900				lun->delay_info.done_delay = 0;
13901			return;
13902		}
13903	}
13904#endif /* CTL_IO_DELAY */
13905
13906	ctl_enqueue_done(io);
13907}
13908
13909int
13910ctl_isc(struct ctl_scsiio *ctsio)
13911{
13912	struct ctl_lun *lun;
13913	int retval;
13914
13915	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13916
13917	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13918
13919	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13920
13921	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13922
13923	return (retval);
13924}
13925
13926
13927static void
13928ctl_work_thread(void *arg)
13929{
13930	struct ctl_thread *thr = (struct ctl_thread *)arg;
13931	struct ctl_softc *softc = thr->ctl_softc;
13932	union ctl_io *io;
13933	int retval;
13934
13935	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13936
13937	for (;;) {
13938		retval = 0;
13939
13940		/*
13941		 * We handle the queues in this order:
13942		 * - ISC
13943		 * - done queue (to free up resources, unblock other commands)
13944		 * - RtR queue
13945		 * - incoming queue
13946		 *
13947		 * If those queues are empty, we break out of the loop and
13948		 * go to sleep.
13949		 */
13950		mtx_lock(&thr->queue_lock);
13951		io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13952		if (io != NULL) {
13953			STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13954			mtx_unlock(&thr->queue_lock);
13955			ctl_handle_isc(io);
13956			continue;
13957		}
13958		io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13959		if (io != NULL) {
13960			STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13961			/* clear any blocked commands, call fe_done */
13962			mtx_unlock(&thr->queue_lock);
13963			retval = ctl_process_done(io);
13964			continue;
13965		}
13966		io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13967		if (io != NULL) {
13968			STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13969			mtx_unlock(&thr->queue_lock);
13970			if (io->io_hdr.io_type == CTL_IO_TASK)
13971				ctl_run_task(io);
13972			else
13973				ctl_scsiio_precheck(softc, &io->scsiio);
13974			continue;
13975		}
13976		if (!ctl_pause_rtr) {
13977			io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13978			if (io != NULL) {
13979				STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13980				mtx_unlock(&thr->queue_lock);
13981				retval = ctl_scsiio(&io->scsiio);
13982				if (retval != CTL_RETVAL_COMPLETE)
13983					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13984				continue;
13985			}
13986		}
13987
13988		/* Sleep until we have something to do. */
13989		mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13990	}
13991}
13992
13993static void
13994ctl_lun_thread(void *arg)
13995{
13996	struct ctl_softc *softc = (struct ctl_softc *)arg;
13997	struct ctl_be_lun *be_lun;
13998	int retval;
13999
14000	CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
14001
14002	for (;;) {
14003		retval = 0;
14004		mtx_lock(&softc->ctl_lock);
14005		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
14006		if (be_lun != NULL) {
14007			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
14008			mtx_unlock(&softc->ctl_lock);
14009			ctl_create_lun(be_lun);
14010			continue;
14011		}
14012
14013		/* Sleep until we have something to do. */
14014		mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
14015		    PDROP | PRIBIO, "-", 0);
14016	}
14017}
14018
14019static void
14020ctl_thresh_thread(void *arg)
14021{
14022	struct ctl_softc *softc = (struct ctl_softc *)arg;
14023	struct ctl_lun *lun;
14024	struct ctl_be_lun *be_lun;
14025	struct scsi_da_rw_recovery_page *rwpage;
14026	struct ctl_logical_block_provisioning_page *page;
14027	const char *attr;
14028	uint64_t thres, val;
14029	int i, e;
14030
14031	CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
14032
14033	for (;;) {
14034		mtx_lock(&softc->ctl_lock);
14035		STAILQ_FOREACH(lun, &softc->lun_list, links) {
14036			be_lun = lun->be_lun;
14037			if ((lun->flags & CTL_LUN_DISABLED) ||
14038			    (lun->flags & CTL_LUN_OFFLINE) ||
14039			    lun->backend->lun_attr == NULL)
14040				continue;
14041			rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
14042			if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
14043				continue;
14044			e = 0;
14045			page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
14046			for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
14047				if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
14048					continue;
14049				thres = scsi_4btoul(page->descr[i].count);
14050				thres <<= CTL_LBP_EXPONENT;
14051				switch (page->descr[i].resource) {
14052				case 0x01:
14053					attr = "blocksavail";
14054					break;
14055				case 0x02:
14056					attr = "blocksused";
14057					break;
14058				case 0xf1:
14059					attr = "poolblocksavail";
14060					break;
14061				case 0xf2:
14062					attr = "poolblocksused";
14063					break;
14064				default:
14065					continue;
14066				}
14067				mtx_unlock(&softc->ctl_lock); // XXX
14068				val = lun->backend->lun_attr(
14069				    lun->be_lun->be_lun, attr);
14070				mtx_lock(&softc->ctl_lock);
14071				if (val == UINT64_MAX)
14072					continue;
14073				if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
14074				    == SLBPPD_ARMING_INC)
14075					e |= (val >= thres);
14076				else
14077					e |= (val <= thres);
14078			}
14079			mtx_lock(&lun->lun_lock);
14080			if (e) {
14081				if (lun->lasttpt == 0 ||
14082				    time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
14083					lun->lasttpt = time_uptime;
14084					ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
14085				}
14086			} else {
14087				lun->lasttpt = 0;
14088				ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
14089			}
14090			mtx_unlock(&lun->lun_lock);
14091		}
14092		mtx_unlock(&softc->ctl_lock);
14093		pause("-", CTL_LBP_PERIOD * hz);
14094	}
14095}
14096
14097static void
14098ctl_enqueue_incoming(union ctl_io *io)
14099{
14100	struct ctl_softc *softc = control_softc;
14101	struct ctl_thread *thr;
14102	u_int idx;
14103
14104	idx = (io->io_hdr.nexus.targ_port * 127 +
14105	       io->io_hdr.nexus.initid.id) % worker_threads;
14106	thr = &softc->threads[idx];
14107	mtx_lock(&thr->queue_lock);
14108	STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
14109	mtx_unlock(&thr->queue_lock);
14110	wakeup(thr);
14111}
14112
14113static void
14114ctl_enqueue_rtr(union ctl_io *io)
14115{
14116	struct ctl_softc *softc = control_softc;
14117	struct ctl_thread *thr;
14118
14119	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14120	mtx_lock(&thr->queue_lock);
14121	STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
14122	mtx_unlock(&thr->queue_lock);
14123	wakeup(thr);
14124}
14125
14126static void
14127ctl_enqueue_done(union ctl_io *io)
14128{
14129	struct ctl_softc *softc = control_softc;
14130	struct ctl_thread *thr;
14131
14132	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14133	mtx_lock(&thr->queue_lock);
14134	STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
14135	mtx_unlock(&thr->queue_lock);
14136	wakeup(thr);
14137}
14138
14139static void
14140ctl_enqueue_isc(union ctl_io *io)
14141{
14142	struct ctl_softc *softc = control_softc;
14143	struct ctl_thread *thr;
14144
14145	thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
14146	mtx_lock(&thr->queue_lock);
14147	STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
14148	mtx_unlock(&thr->queue_lock);
14149	wakeup(thr);
14150}
14151
14152/* Initialization and failover */
14153
14154void
14155ctl_init_isc_msg(void)
14156{
14157	printf("CTL: Still calling this thing\n");
14158}
14159
14160/*
14161 * Init component
14162 * 	Initializes component into configuration defined by bootMode
14163 *	(see hasc-sv.c)
14164 *  	returns hasc_Status:
14165 * 		OK
14166 *		ERROR - fatal error
14167 */
14168static ctl_ha_comp_status
14169ctl_isc_init(struct ctl_ha_component *c)
14170{
14171	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14172
14173	c->status = ret;
14174	return ret;
14175}
14176
14177/* Start component
14178 * 	Starts component in state requested. If component starts successfully,
14179 *	it must set its own state to the requestrd state
14180 *	When requested state is HASC_STATE_HA, the component may refine it
14181 * 	by adding _SLAVE or _MASTER flags.
14182 *	Currently allowed state transitions are:
14183 *	UNKNOWN->HA		- initial startup
14184 *	UNKNOWN->SINGLE - initial startup when no parter detected
14185 *	HA->SINGLE		- failover
14186 * returns ctl_ha_comp_status:
14187 * 		OK	- component successfully started in requested state
14188 *		FAILED  - could not start the requested state, failover may
14189 * 			  be possible
14190 *		ERROR	- fatal error detected, no future startup possible
14191 */
14192static ctl_ha_comp_status
14193ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
14194{
14195	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
14196
14197	printf("%s: go\n", __func__);
14198
14199	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
14200	if (c->state == CTL_HA_STATE_UNKNOWN ) {
14201		control_softc->is_single = 0;
14202		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
14203		    != CTL_HA_STATUS_SUCCESS) {
14204			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
14205			ret = CTL_HA_COMP_STATUS_ERROR;
14206		}
14207	} else if (CTL_HA_STATE_IS_HA(c->state)
14208		&& CTL_HA_STATE_IS_SINGLE(state)){
14209		// HA->SINGLE transition
14210	        ctl_failover();
14211		control_softc->is_single = 1;
14212	} else {
14213		printf("ctl_isc_start:Invalid state transition %X->%X\n",
14214		       c->state, state);
14215		ret = CTL_HA_COMP_STATUS_ERROR;
14216	}
14217	if (CTL_HA_STATE_IS_SINGLE(state))
14218		control_softc->is_single = 1;
14219
14220	c->state = state;
14221	c->status = ret;
14222	return ret;
14223}
14224
14225/*
14226 * Quiesce component
14227 * The component must clear any error conditions (set status to OK) and
14228 * prepare itself to another Start call
14229 * returns ctl_ha_comp_status:
14230 * 	OK
14231 *	ERROR
14232 */
14233static ctl_ha_comp_status
14234ctl_isc_quiesce(struct ctl_ha_component *c)
14235{
14236	int ret = CTL_HA_COMP_STATUS_OK;
14237
14238	ctl_pause_rtr = 1;
14239	c->status = ret;
14240	return ret;
14241}
14242
14243struct ctl_ha_component ctl_ha_component_ctlisc =
14244{
14245	.name = "CTL ISC",
14246	.state = CTL_HA_STATE_UNKNOWN,
14247	.init = ctl_isc_init,
14248	.start = ctl_isc_start,
14249	.quiesce = ctl_isc_quiesce
14250};
14251
14252/*
14253 *  vim: ts=8
14254 */
14255