1 // SPDX-License-Identifier: GPL-2.0-only
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
5 * SCSI error/timeout handling
6 * Initial versions: Eric Youngdale. Based upon conversations with
7 * Leonard Zubkoff and David Miller at Linux Expo,
8 * ideas originating from all over the place.
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
18 #include <linux/module.h>
19 #include <linux/sched.h>
20 #include <linux/gfp.h>
21 #include <linux/timer.h>
22 #include <linux/string.h>
23 #include <linux/kernel.h>
24 #include <linux/freezer.h>
25 #include <linux/kthread.h>
26 #include <linux/interrupt.h>
27 #include <linux/blkdev.h>
28 #include <linux/delay.h>
29 #include <linux/jiffies.h>
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_cmnd.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_device.h>
35 #include <scsi/scsi_driver.h>
36 #include <scsi/scsi_eh.h>
37 #include <scsi/scsi_common.h>
38 #include <scsi/scsi_transport.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_ioctl.h>
41 #include <scsi/scsi_dh.h>
42 #include <scsi/scsi_devinfo.h>
45 #include "scsi_priv.h"
46 #include "scsi_logging.h"
47 #include "scsi_transport_api.h"
49 #include <trace/events/scsi.h>
51 #include <linux/unaligned.h>
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
57 #define BUS_RESET_SETTLE_TIME (10)
58 #define HOST_RESET_SETTLE_TIME (10)
60 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
61 static enum scsi_disposition
scsi_try_to_abort_cmd(const struct scsi_host_template
*,
64 void scsi_eh_wakeup(struct Scsi_Host
*shost
, unsigned int busy
)
66 lockdep_assert_held(shost
->host_lock
);
68 if (busy
== shost
->host_failed
) {
69 trace_scsi_eh_wakeup(shost
);
70 wake_up_process(shost
->ehandler
);
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO
, shost
,
72 "Waking error handler thread\n"));
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
80 * Schedule SCSI EH without scmd.
82 void scsi_schedule_eh(struct Scsi_Host
*shost
)
86 spin_lock_irqsave(shost
->host_lock
, flags
);
88 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
89 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
90 shost
->host_eh_scheduled
++;
91 scsi_eh_wakeup(shost
, scsi_host_busy(shost
));
94 spin_unlock_irqrestore(shost
->host_lock
, flags
);
96 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
98 static int scsi_host_eh_past_deadline(struct Scsi_Host
*shost
)
100 if (!shost
->last_reset
|| shost
->eh_deadline
== -1)
104 * 32bit accesses are guaranteed to be atomic
105 * (on all supported architectures), so instead
106 * of using a spinlock we can as well double check
107 * if eh_deadline has been set to 'off' during the
110 if (time_before(jiffies
, shost
->last_reset
+ shost
->eh_deadline
) &&
111 shost
->eh_deadline
> -1)
117 static bool scsi_cmd_retry_allowed(struct scsi_cmnd
*cmd
)
119 if (cmd
->allowed
== SCSI_CMD_RETRIES_NO_LIMIT
)
122 return ++cmd
->retries
<= cmd
->allowed
;
125 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd
*cmd
)
127 struct scsi_device
*sdev
= cmd
->device
;
128 struct Scsi_Host
*host
= sdev
->host
;
130 if (host
->hostt
->eh_should_retry_cmd
)
131 return host
->hostt
->eh_should_retry_cmd(cmd
);
137 * scmd_eh_abort_handler - Handle command aborts
138 * @work: command to be aborted.
140 * Note: this function must be called only for a command that has timed out.
141 * Because the block layer marks a request as complete before it calls
142 * scsi_timeout(), a .scsi_done() call from the LLD for a command that has
143 * timed out do not have any effect. Hence it is safe to call
144 * scsi_finish_command() from this function.
147 scmd_eh_abort_handler(struct work_struct
*work
)
149 struct scsi_cmnd
*scmd
=
150 container_of(work
, struct scsi_cmnd
, abort_work
.work
);
151 struct scsi_device
*sdev
= scmd
->device
;
152 struct Scsi_Host
*shost
= sdev
->host
;
153 enum scsi_disposition rtn
;
156 if (scsi_host_eh_past_deadline(shost
)) {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO
, scmd
,
159 "eh timeout, not aborting\n"));
163 SCSI_LOG_ERROR_RECOVERY(3,
164 scmd_printk(KERN_INFO
, scmd
,
165 "aborting command\n"));
166 rtn
= scsi_try_to_abort_cmd(shost
->hostt
, scmd
);
167 if (rtn
!= SUCCESS
) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO
, scmd
,
171 (rtn
== FAST_IO_FAIL
) ?
172 "not send" : "failed"));
175 set_host_byte(scmd
, DID_TIME_OUT
);
176 if (scsi_host_eh_past_deadline(shost
)) {
177 SCSI_LOG_ERROR_RECOVERY(3,
178 scmd_printk(KERN_INFO
, scmd
,
179 "eh timeout, not retrying "
180 "aborted command\n"));
184 spin_lock_irqsave(shost
->host_lock
, flags
);
185 list_del_init(&scmd
->eh_entry
);
188 * If the abort succeeds, and there is no further
189 * EH action, clear the ->last_reset time.
191 if (list_empty(&shost
->eh_abort_list
) &&
192 list_empty(&shost
->eh_cmd_q
))
193 if (shost
->eh_deadline
!= -1)
194 shost
->last_reset
= 0;
196 spin_unlock_irqrestore(shost
->host_lock
, flags
);
198 if (!scsi_noretry_cmd(scmd
) &&
199 scsi_cmd_retry_allowed(scmd
) &&
200 scsi_eh_should_retry_cmd(scmd
)) {
201 SCSI_LOG_ERROR_RECOVERY(3,
202 scmd_printk(KERN_WARNING
, scmd
,
203 "retry aborted command\n"));
204 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_WARNING
, scmd
,
208 "finish aborted command\n"));
209 scsi_finish_command(scmd
);
214 spin_lock_irqsave(shost
->host_lock
, flags
);
215 list_del_init(&scmd
->eh_entry
);
216 spin_unlock_irqrestore(shost
->host_lock
, flags
);
218 scsi_eh_scmd_add(scmd
);
222 * scsi_abort_command - schedule a command abort
223 * @scmd: scmd to abort.
225 * We only need to abort commands after a command timeout
228 scsi_abort_command(struct scsi_cmnd
*scmd
)
230 struct scsi_device
*sdev
= scmd
->device
;
231 struct Scsi_Host
*shost
= sdev
->host
;
234 if (!shost
->hostt
->eh_abort_handler
) {
235 /* No abort handler, fail command directly */
239 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
241 * Retry after abort failed, escalate to next level.
243 SCSI_LOG_ERROR_RECOVERY(3,
244 scmd_printk(KERN_INFO
, scmd
,
245 "previous abort failed\n"));
246 BUG_ON(delayed_work_pending(&scmd
->abort_work
));
250 spin_lock_irqsave(shost
->host_lock
, flags
);
251 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
252 shost
->last_reset
= jiffies
;
253 BUG_ON(!list_empty(&scmd
->eh_entry
));
254 list_add_tail(&scmd
->eh_entry
, &shost
->eh_abort_list
);
255 spin_unlock_irqrestore(shost
->host_lock
, flags
);
257 scmd
->eh_eflags
|= SCSI_EH_ABORT_SCHEDULED
;
258 SCSI_LOG_ERROR_RECOVERY(3,
259 scmd_printk(KERN_INFO
, scmd
, "abort scheduled\n"));
260 queue_delayed_work(shost
->tmf_work_q
, &scmd
->abort_work
, HZ
/ 100);
265 * scsi_eh_reset - call into ->eh_action to reset internal counters
266 * @scmd: scmd to run eh on.
268 * The scsi driver might be carrying internal state about the
269 * devices, so we need to call into the driver to reset the
270 * internal state once the error handler is started.
272 static void scsi_eh_reset(struct scsi_cmnd
*scmd
)
274 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd
))) {
275 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
277 sdrv
->eh_reset(scmd
);
281 static void scsi_eh_inc_host_failed(struct rcu_head
*head
)
283 struct scsi_cmnd
*scmd
= container_of(head
, typeof(*scmd
), rcu
);
284 struct Scsi_Host
*shost
= scmd
->device
->host
;
285 unsigned int busy
= scsi_host_busy(shost
);
288 spin_lock_irqsave(shost
->host_lock
, flags
);
289 shost
->host_failed
++;
290 scsi_eh_wakeup(shost
, busy
);
291 spin_unlock_irqrestore(shost
->host_lock
, flags
);
295 * scsi_eh_scmd_add - add scsi cmd to error handling.
296 * @scmd: scmd to run eh on.
298 void scsi_eh_scmd_add(struct scsi_cmnd
*scmd
)
300 struct Scsi_Host
*shost
= scmd
->device
->host
;
304 WARN_ON_ONCE(!shost
->ehandler
);
305 WARN_ON_ONCE(!test_bit(SCMD_STATE_INFLIGHT
, &scmd
->state
));
307 spin_lock_irqsave(shost
->host_lock
, flags
);
308 if (scsi_host_set_state(shost
, SHOST_RECOVERY
)) {
309 ret
= scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
);
312 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
313 shost
->last_reset
= jiffies
;
316 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
317 spin_unlock_irqrestore(shost
->host_lock
, flags
);
319 * Ensure that all tasks observe the host state change before the
320 * host_failed change.
322 call_rcu_hurry(&scmd
->rcu
, scsi_eh_inc_host_failed
);
326 * scsi_timeout - Timeout function for normal scsi commands.
327 * @req: request that is timing out.
330 * We do not need to lock this. There is the potential for a race
331 * only in that the normal completion handling might run, but if the
332 * normal completion function determines that the timer has already
333 * fired, then it mustn't do anything.
335 enum blk_eh_timer_return
scsi_timeout(struct request
*req
)
337 struct scsi_cmnd
*scmd
= blk_mq_rq_to_pdu(req
);
338 struct Scsi_Host
*host
= scmd
->device
->host
;
340 trace_scsi_dispatch_cmd_timeout(scmd
);
341 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
343 atomic_inc(&scmd
->device
->iotmo_cnt
);
344 if (host
->eh_deadline
!= -1 && !host
->last_reset
)
345 host
->last_reset
= jiffies
;
347 if (host
->hostt
->eh_timed_out
) {
348 switch (host
->hostt
->eh_timed_out(scmd
)) {
351 case SCSI_EH_RESET_TIMER
:
352 return BLK_EH_RESET_TIMER
;
353 case SCSI_EH_NOT_HANDLED
:
359 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify
362 if (test_and_set_bit(SCMD_STATE_COMPLETE
, &scmd
->state
))
364 atomic_inc(&scmd
->device
->iodone_cnt
);
365 if (scsi_abort_command(scmd
) != SUCCESS
) {
366 set_host_byte(scmd
, DID_TIME_OUT
);
367 scsi_eh_scmd_add(scmd
);
374 * scsi_block_when_processing_errors - Prevent cmds from being queued.
375 * @sdev: Device on which we are performing recovery.
378 * We block until the host is out of error recovery, and then check to
379 * see whether the host or the device is offline.
382 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
384 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
388 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
390 online
= scsi_device_online(sdev
);
394 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
396 #ifdef CONFIG_SCSI_LOGGING
398 * scsi_eh_prt_fail_stats - Log info on failures.
399 * @shost: scsi host being recovered.
400 * @work_q: Queue of scsi cmds to process.
402 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
403 struct list_head
*work_q
)
405 struct scsi_cmnd
*scmd
;
406 struct scsi_device
*sdev
;
407 int total_failures
= 0;
410 int devices_failed
= 0;
412 shost_for_each_device(sdev
, shost
) {
413 list_for_each_entry(scmd
, work_q
, eh_entry
) {
414 if (scmd
->device
== sdev
) {
416 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
)
423 if (cmd_cancel
|| cmd_failed
) {
424 SCSI_LOG_ERROR_RECOVERY(3,
425 shost_printk(KERN_INFO
, shost
,
426 "%s: cmds failed: %d, cancel: %d\n",
427 __func__
, cmd_failed
,
435 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO
, shost
,
436 "Total of %d commands on %d"
437 " devices require eh work\n",
438 total_failures
, devices_failed
));
443 * scsi_report_lun_change - Set flag on all *other* devices on the same target
444 * to indicate that a UNIT ATTENTION is expected.
445 * @sdev: Device reporting the UNIT ATTENTION
447 static void scsi_report_lun_change(struct scsi_device
*sdev
)
449 sdev
->sdev_target
->expecting_lun_change
= 1;
453 * scsi_report_sense - Examine scsi sense information and log messages for
454 * certain conditions, also issue uevents for some of them.
455 * @sdev: Device reporting the sense code
456 * @sshdr: sshdr to be examined
458 static void scsi_report_sense(struct scsi_device
*sdev
,
459 struct scsi_sense_hdr
*sshdr
)
461 enum scsi_device_event evt_type
= SDEV_EVT_MAXBITS
; /* i.e. none */
463 if (sshdr
->sense_key
== UNIT_ATTENTION
) {
464 if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x03) {
465 evt_type
= SDEV_EVT_INQUIRY_CHANGE_REPORTED
;
466 sdev_printk(KERN_WARNING
, sdev
,
467 "Inquiry data has changed");
468 } else if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x0e) {
469 evt_type
= SDEV_EVT_LUN_CHANGE_REPORTED
;
470 scsi_report_lun_change(sdev
);
471 sdev_printk(KERN_WARNING
, sdev
,
472 "LUN assignments on this target have "
473 "changed. The Linux SCSI layer does not "
474 "automatically remap LUN assignments.\n");
475 } else if (sshdr
->asc
== 0x3f)
476 sdev_printk(KERN_WARNING
, sdev
,
477 "Operating parameters on this target have "
478 "changed. The Linux SCSI layer does not "
479 "automatically adjust these parameters.\n");
481 if (sshdr
->asc
== 0x38 && sshdr
->ascq
== 0x07) {
482 evt_type
= SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
;
483 sdev_printk(KERN_WARNING
, sdev
,
484 "Warning! Received an indication that the "
485 "LUN reached a thin provisioning soft "
489 if (sshdr
->asc
== 0x29) {
490 evt_type
= SDEV_EVT_POWER_ON_RESET_OCCURRED
;
492 * Do not print message if it is an expected side-effect
495 if (!sdev
->silence_suspend
)
496 sdev_printk(KERN_WARNING
, sdev
,
497 "Power-on or device reset occurred\n");
500 if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x01) {
501 evt_type
= SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
;
502 sdev_printk(KERN_WARNING
, sdev
,
503 "Mode parameters changed");
504 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x06) {
505 evt_type
= SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
;
506 sdev_printk(KERN_WARNING
, sdev
,
507 "Asymmetric access state changed");
508 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x09) {
509 evt_type
= SDEV_EVT_CAPACITY_CHANGE_REPORTED
;
510 sdev_printk(KERN_WARNING
, sdev
,
511 "Capacity data has changed");
512 } else if (sshdr
->asc
== 0x2a)
513 sdev_printk(KERN_WARNING
, sdev
,
514 "Parameters changed");
517 if (evt_type
!= SDEV_EVT_MAXBITS
) {
518 set_bit(evt_type
, sdev
->pending_events
);
519 schedule_work(&sdev
->event_work
);
523 static inline void set_scsi_ml_byte(struct scsi_cmnd
*cmd
, u8 status
)
525 cmd
->result
= (cmd
->result
& 0xffff00ff) | (status
<< 8);
529 * scsi_check_sense - Examine scsi cmd sense
530 * @scmd: Cmd to have sense checked.
533 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
536 * When a deferred error is detected the current command has
537 * not been executed and needs retrying.
539 enum scsi_disposition
scsi_check_sense(struct scsi_cmnd
*scmd
)
541 struct request
*req
= scsi_cmd_to_rq(scmd
);
542 struct scsi_device
*sdev
= scmd
->device
;
543 struct scsi_sense_hdr sshdr
;
545 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
546 return FAILED
; /* no valid sense data */
548 scsi_report_sense(sdev
, &sshdr
);
550 if (scsi_sense_is_deferred(&sshdr
))
553 if (sdev
->handler
&& sdev
->handler
->check_sense
) {
554 enum scsi_disposition rc
;
556 rc
= sdev
->handler
->check_sense(sdev
, &sshdr
);
557 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
559 /* handler does not care. Drop down to default handling */
562 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&&
563 scmd
->submitter
!= SUBMITTED_BY_SCSI_ERROR_HANDLER
)
565 * nasty: for mid-layer issued TURs, we need to return the
566 * actual sense data without any recovery attempt. For eh
567 * issued ones, we need to try to recover and interpret
572 * Previous logic looked for FILEMARK, EOM or ILI which are
573 * mainly associated with tapes and returned SUCCESS.
575 if (sshdr
.response_code
== 0x70) {
577 if (scmd
->sense_buffer
[2] & 0xe0)
581 * descriptor format: look for "stream commands sense data
582 * descriptor" (see SSC-3). Assume single sense data
583 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
585 if ((sshdr
.additional_length
> 3) &&
586 (scmd
->sense_buffer
[8] == 0x4) &&
587 (scmd
->sense_buffer
[11] & 0xe0))
591 switch (sshdr
.sense_key
) {
594 case RECOVERED_ERROR
:
595 return /* soft_error */ SUCCESS
;
597 case ABORTED_COMMAND
:
598 if (sshdr
.asc
== 0x10) /* DIF */
602 * Check aborts due to command duration limit policy:
603 * ABORTED COMMAND additional sense code with the
604 * COMMAND TIMEOUT BEFORE PROCESSING or
605 * COMMAND TIMEOUT DURING PROCESSING or
606 * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY
607 * additional sense code qualifiers.
609 if (sshdr
.asc
== 0x2e &&
610 sshdr
.ascq
>= 0x01 && sshdr
.ascq
<= 0x03) {
611 set_scsi_ml_byte(scmd
, SCSIML_STAT_DL_TIMEOUT
);
612 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
613 req
->rq_flags
|= RQF_QUIET
;
617 if (sshdr
.asc
== 0x44 && sdev
->sdev_bflags
& BLIST_RETRY_ITF
)
618 return ADD_TO_MLQUEUE
;
619 if (sshdr
.asc
== 0xc1 && sshdr
.ascq
== 0x01 &&
620 sdev
->sdev_bflags
& BLIST_RETRY_ASC_C1
)
621 return ADD_TO_MLQUEUE
;
627 * if we are expecting a cc/ua because of a bus reset that we
628 * performed, treat this just as a retry. otherwise this is
629 * information that we should pass up to the upper-level driver
630 * so that we can deal with it there.
632 if (scmd
->device
->expecting_cc_ua
) {
634 * Because some device does not queue unit
635 * attentions correctly, we carefully check
636 * additional sense code and qualifier so as
637 * not to squash media change unit attention.
639 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
640 scmd
->device
->expecting_cc_ua
= 0;
645 * we might also expect a cc/ua if another LUN on the target
646 * reported a UA with an ASC/ASCQ of 3F 0E -
647 * REPORTED LUNS DATA HAS CHANGED.
649 if (scmd
->device
->sdev_target
->expecting_lun_change
&&
650 sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
653 * if the device is in the process of becoming ready, we
656 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
659 * if the device is not started, we need to wake
660 * the error handler to start the motor
662 if (scmd
->device
->allow_restart
&&
663 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
666 * Pass the UA upwards for a determination in the completion
671 /* these are not supported */
673 if (sshdr
.asc
== 0x27 && sshdr
.ascq
== 0x07) {
674 /* Thin provisioning hard threshold reached */
675 set_scsi_ml_byte(scmd
, SCSIML_STAT_NOSPC
);
680 case VOLUME_OVERFLOW
:
683 set_scsi_ml_byte(scmd
, SCSIML_STAT_TGT_FAILURE
);
687 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
688 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
689 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
690 set_scsi_ml_byte(scmd
, SCSIML_STAT_MED_ERROR
);
696 if (scmd
->device
->retry_hwerror
)
697 return ADD_TO_MLQUEUE
;
699 set_scsi_ml_byte(scmd
, SCSIML_STAT_TGT_FAILURE
);
702 case ILLEGAL_REQUEST
:
703 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
704 sshdr
.asc
== 0x21 || /* Logical block address out of range */
705 sshdr
.asc
== 0x22 || /* Invalid function */
706 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
707 sshdr
.asc
== 0x26 || /* Parameter value invalid */
708 sshdr
.asc
== 0x27) { /* Write protected */
709 set_scsi_ml_byte(scmd
, SCSIML_STAT_TGT_FAILURE
);
714 if (sshdr
.asc
== 0x55 && sshdr
.ascq
== 0x0a) {
715 set_scsi_ml_byte(scmd
, SCSIML_STAT_DL_TIMEOUT
);
716 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
717 req
->rq_flags
|= RQF_QUIET
;
725 EXPORT_SYMBOL_GPL(scsi_check_sense
);
727 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
729 const struct scsi_host_template
*sht
= sdev
->host
->hostt
;
730 struct scsi_device
*tmp_sdev
;
732 if (!sht
->track_queue_depth
||
733 sdev
->queue_depth
>= sdev
->max_queue_depth
)
736 if (time_before(jiffies
,
737 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
740 if (time_before(jiffies
,
741 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
745 * Walk all devices of a target and do
748 shost_for_each_device(tmp_sdev
, sdev
->host
) {
749 if (tmp_sdev
->channel
!= sdev
->channel
||
750 tmp_sdev
->id
!= sdev
->id
||
751 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
754 scsi_change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1);
755 sdev
->last_queue_ramp_up
= jiffies
;
759 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
761 const struct scsi_host_template
*sht
= sdev
->host
->hostt
;
762 struct scsi_device
*tmp_sdev
;
764 if (!sht
->track_queue_depth
)
767 shost_for_each_device(tmp_sdev
, sdev
->host
) {
768 if (tmp_sdev
->channel
!= sdev
->channel
||
769 tmp_sdev
->id
!= sdev
->id
)
772 * We do not know the number of commands that were at
773 * the device when we got the queue full so we start
774 * from the highest possible value and work our way down.
776 scsi_track_queue_full(tmp_sdev
, tmp_sdev
->queue_depth
- 1);
781 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
782 * @scmd: SCSI cmd to examine.
785 * This is *only* called when we are examining the status of commands
786 * queued during error recovery. the main difference here is that we
787 * don't allow for the possibility of retries here, and we are a lot
788 * more restrictive about what we consider acceptable.
790 static enum scsi_disposition
scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
793 * first check the host byte, to see if there is anything in there
794 * that would indicate what we need to do.
796 if (host_byte(scmd
->result
) == DID_RESET
) {
798 * rats. we are already in the error handler, so we now
799 * get to try and figure out what to do next. if the sense
800 * is valid, we have a pretty good idea of what to do.
801 * if not, we mark it as FAILED.
803 return scsi_check_sense(scmd
);
805 if (host_byte(scmd
->result
) != DID_OK
)
809 * now, check the status byte to see if this indicates
812 switch (get_status_byte(scmd
)) {
814 scsi_handle_queue_ramp_up(scmd
->device
);
815 if (scmd
->sense_buffer
&& SCSI_SENSE_VALID(scmd
))
817 * If we have sense data, call scsi_check_sense() in
818 * order to set the correct SCSI ML byte (if any).
819 * No point in checking the return value, since the
820 * command has already completed successfully.
822 scsi_check_sense(scmd
);
824 case SAM_STAT_COMMAND_TERMINATED
:
826 case SAM_STAT_CHECK_CONDITION
:
827 return scsi_check_sense(scmd
);
828 case SAM_STAT_CONDITION_MET
:
829 case SAM_STAT_INTERMEDIATE
:
830 case SAM_STAT_INTERMEDIATE_CONDITION_MET
:
832 * who knows? FIXME(eric)
835 case SAM_STAT_RESERVATION_CONFLICT
:
836 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
837 /* it is a success, we probed the device and
840 /* otherwise, we failed to send the command */
842 case SAM_STAT_TASK_SET_FULL
:
843 scsi_handle_queue_full(scmd
->device
);
854 * scsi_eh_done - Completion function for error handling.
855 * @scmd: Cmd that is done.
857 void scsi_eh_done(struct scsi_cmnd
*scmd
)
859 struct completion
*eh_action
;
861 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
862 "%s result: %x\n", __func__
, scmd
->result
));
864 eh_action
= scmd
->device
->host
->eh_action
;
870 * scsi_try_host_reset - ask host adapter to reset itself
871 * @scmd: SCSI cmd to send host reset.
873 static enum scsi_disposition
scsi_try_host_reset(struct scsi_cmnd
*scmd
)
876 enum scsi_disposition rtn
;
877 struct Scsi_Host
*host
= scmd
->device
->host
;
878 const struct scsi_host_template
*hostt
= host
->hostt
;
880 SCSI_LOG_ERROR_RECOVERY(3,
881 shost_printk(KERN_INFO
, host
, "Snd Host RST\n"));
883 if (!hostt
->eh_host_reset_handler
)
886 rtn
= hostt
->eh_host_reset_handler(scmd
);
888 if (rtn
== SUCCESS
) {
889 if (!hostt
->skip_settle_delay
)
890 ssleep(HOST_RESET_SETTLE_TIME
);
891 spin_lock_irqsave(host
->host_lock
, flags
);
892 scsi_report_bus_reset(host
, scmd_channel(scmd
));
893 spin_unlock_irqrestore(host
->host_lock
, flags
);
900 * scsi_try_bus_reset - ask host to perform a bus reset
901 * @scmd: SCSI cmd to send bus reset.
903 static enum scsi_disposition
scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
906 enum scsi_disposition rtn
;
907 struct Scsi_Host
*host
= scmd
->device
->host
;
908 const struct scsi_host_template
*hostt
= host
->hostt
;
910 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
911 "%s: Snd Bus RST\n", __func__
));
913 if (!hostt
->eh_bus_reset_handler
)
916 rtn
= hostt
->eh_bus_reset_handler(scmd
);
918 if (rtn
== SUCCESS
) {
919 if (!hostt
->skip_settle_delay
)
920 ssleep(BUS_RESET_SETTLE_TIME
);
921 spin_lock_irqsave(host
->host_lock
, flags
);
922 scsi_report_bus_reset(host
, scmd_channel(scmd
));
923 spin_unlock_irqrestore(host
->host_lock
, flags
);
929 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
932 sdev
->expecting_cc_ua
= 1;
936 * scsi_try_target_reset - Ask host to perform a target reset
937 * @scmd: SCSI cmd used to send a target reset
940 * There is no timeout for this operation. if this operation is
941 * unreliable for a given host, then the host itself needs to put a
942 * timer on it, and set the host back to a consistent state prior to
945 static enum scsi_disposition
scsi_try_target_reset(struct scsi_cmnd
*scmd
)
948 enum scsi_disposition rtn
;
949 struct Scsi_Host
*host
= scmd
->device
->host
;
950 const struct scsi_host_template
*hostt
= host
->hostt
;
952 if (!hostt
->eh_target_reset_handler
)
955 rtn
= hostt
->eh_target_reset_handler(scmd
);
956 if (rtn
== SUCCESS
) {
957 spin_lock_irqsave(host
->host_lock
, flags
);
958 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
959 __scsi_report_device_reset
);
960 spin_unlock_irqrestore(host
->host_lock
, flags
);
967 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
968 * @scmd: SCSI cmd used to send BDR
971 * There is no timeout for this operation. if this operation is
972 * unreliable for a given host, then the host itself needs to put a
973 * timer on it, and set the host back to a consistent state prior to
976 static enum scsi_disposition
scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
978 enum scsi_disposition rtn
;
979 const struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
981 if (!hostt
->eh_device_reset_handler
)
984 rtn
= hostt
->eh_device_reset_handler(scmd
);
986 __scsi_report_device_reset(scmd
->device
, NULL
);
991 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
992 * @hostt: SCSI driver host template
993 * @scmd: SCSI cmd used to send a target reset
996 * SUCCESS, FAILED, or FAST_IO_FAIL
999 * SUCCESS does not necessarily indicate that the command
1000 * has been aborted; it only indicates that the LLDDs
1001 * has cleared all references to that command.
1002 * LLDDs should return FAILED only if an abort was required
1003 * but could not be executed. LLDDs should return FAST_IO_FAIL
1004 * if the device is temporarily unavailable (eg due to a
1005 * link down on FibreChannel)
1007 static enum scsi_disposition
1008 scsi_try_to_abort_cmd(const struct scsi_host_template
*hostt
, struct scsi_cmnd
*scmd
)
1010 if (!hostt
->eh_abort_handler
)
1013 return hostt
->eh_abort_handler(scmd
);
1016 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
1018 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
1019 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
1020 if (scsi_try_target_reset(scmd
) != SUCCESS
)
1021 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
1022 scsi_try_host_reset(scmd
);
1026 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
1027 * @scmd: SCSI command structure to hijack
1028 * @ses: structure to save restore information
1029 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
1030 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE)
1031 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
1033 * This function is used to save a scsi command information before re-execution
1034 * as part of the error recovery process. If @sense_bytes is 0 the command
1035 * sent must be one that does not transfer any data. If @sense_bytes != 0
1036 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
1037 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
1039 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
1040 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
1042 struct scsi_device
*sdev
= scmd
->device
;
1045 * We need saved copies of a number of fields - this is because
1046 * error handling may need to overwrite these with different values
1047 * to run different commands, and once error handling is complete,
1048 * we will need to restore these values prior to running the actual
1051 ses
->cmd_len
= scmd
->cmd_len
;
1052 ses
->data_direction
= scmd
->sc_data_direction
;
1053 ses
->sdb
= scmd
->sdb
;
1054 ses
->result
= scmd
->result
;
1055 ses
->resid_len
= scmd
->resid_len
;
1056 ses
->underflow
= scmd
->underflow
;
1057 ses
->prot_op
= scmd
->prot_op
;
1058 ses
->eh_eflags
= scmd
->eh_eflags
;
1060 scmd
->prot_op
= SCSI_PROT_NORMAL
;
1061 scmd
->eh_eflags
= 0;
1062 memcpy(ses
->cmnd
, scmd
->cmnd
, sizeof(ses
->cmnd
));
1063 memset(scmd
->cmnd
, 0, sizeof(scmd
->cmnd
));
1064 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1066 scmd
->resid_len
= 0;
1069 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
1071 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
1073 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
1074 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
1075 scmd
->sdb
.table
.nents
= scmd
->sdb
.table
.orig_nents
= 1;
1076 scmd
->cmnd
[0] = REQUEST_SENSE
;
1077 scmd
->cmnd
[4] = scmd
->sdb
.length
;
1078 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
1080 scmd
->sc_data_direction
= DMA_NONE
;
1082 BUG_ON(cmnd_size
> sizeof(scmd
->cmnd
));
1083 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
1084 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
1088 scmd
->underflow
= 0;
1090 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
1091 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
1092 (sdev
->lun
<< 5 & 0xe0);
1095 * Zero the sense buffer. The scsi spec mandates that any
1096 * untransferred sense data should be interpreted as being zero.
1098 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
1100 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
1103 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1104 * @scmd: SCSI command structure to restore
1105 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1107 * Undo any damage done by above scsi_eh_prep_cmnd().
1109 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
1112 * Restore original data
1114 scmd
->cmd_len
= ses
->cmd_len
;
1115 memcpy(scmd
->cmnd
, ses
->cmnd
, sizeof(ses
->cmnd
));
1116 scmd
->sc_data_direction
= ses
->data_direction
;
1117 scmd
->sdb
= ses
->sdb
;
1118 scmd
->result
= ses
->result
;
1119 scmd
->resid_len
= ses
->resid_len
;
1120 scmd
->underflow
= ses
->underflow
;
1121 scmd
->prot_op
= ses
->prot_op
;
1122 scmd
->eh_eflags
= ses
->eh_eflags
;
1124 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
1127 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1128 * @scmd: SCSI command structure to hijack
1129 * @cmnd: CDB to send
1130 * @cmnd_size: size in bytes of @cmnd
1131 * @timeout: timeout for this request
1132 * @sense_bytes: size of sense data to copy or 0
1134 * This function is used to send a scsi command down to a target device
1135 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1138 * SUCCESS or FAILED or NEEDS_RETRY
1140 static enum scsi_disposition
scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
,
1141 unsigned char *cmnd
, int cmnd_size
, int timeout
, unsigned sense_bytes
)
1143 struct scsi_device
*sdev
= scmd
->device
;
1144 struct Scsi_Host
*shost
= sdev
->host
;
1145 DECLARE_COMPLETION_ONSTACK(done
);
1146 unsigned long timeleft
= timeout
, delay
;
1147 struct scsi_eh_save ses
;
1148 const unsigned long stall_for
= msecs_to_jiffies(100);
1152 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
1153 shost
->eh_action
= &done
;
1155 scsi_log_send(scmd
);
1156 scmd
->submitter
= SUBMITTED_BY_SCSI_ERROR_HANDLER
;
1157 scmd
->flags
|= SCMD_LAST
;
1160 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1161 * change the SCSI device state after we have examined it and before
1162 * .queuecommand() is called.
1164 mutex_lock(&sdev
->state_mutex
);
1165 while (sdev
->sdev_state
== SDEV_BLOCK
&& timeleft
> 0) {
1166 mutex_unlock(&sdev
->state_mutex
);
1167 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG
, sdev
,
1168 "%s: state %d <> %d\n", __func__
, sdev
->sdev_state
,
1170 delay
= min(timeleft
, stall_for
);
1172 msleep(jiffies_to_msecs(delay
));
1173 mutex_lock(&sdev
->state_mutex
);
1175 if (sdev
->sdev_state
!= SDEV_BLOCK
)
1176 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
1179 mutex_unlock(&sdev
->state_mutex
);
1182 if (timeleft
> stall_for
) {
1183 scsi_eh_restore_cmnd(scmd
, &ses
);
1185 timeleft
-= stall_for
;
1186 msleep(jiffies_to_msecs(stall_for
));
1189 /* signal not to enter either branch of the if () below */
1193 timeleft
= wait_for_completion_timeout(&done
, timeout
);
1197 shost
->eh_action
= NULL
;
1199 scsi_log_completion(scmd
, rtn
);
1201 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1202 "%s timeleft: %ld\n",
1203 __func__
, timeleft
));
1206 * If there is time left scsi_eh_done got called, and we will examine
1207 * the actual status codes to see whether the command actually did
1208 * complete normally, else if we have a zero return and no time left,
1209 * the command must still be pending, so abort it and return FAILED.
1210 * If we never actually managed to issue the command, because
1211 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1212 * value above (so don't execute either branch of the if)
1215 rtn
= scsi_eh_completed_normally(scmd
);
1216 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1217 "%s: scsi_eh_completed_normally %x\n", __func__
, rtn
));
1224 case ADD_TO_MLQUEUE
:
1231 } else if (rtn
!= FAILED
) {
1232 scsi_abort_eh_cmnd(scmd
);
1236 scsi_eh_restore_cmnd(scmd
, &ses
);
1242 * scsi_request_sense - Request sense data from a particular target.
1243 * @scmd: SCSI cmd for request sense.
1246 * Some hosts automatically obtain this information, others require
1247 * that we obtain it on our own. This function will *not* return until
1248 * the command either times out, or it completes.
1250 static enum scsi_disposition
scsi_request_sense(struct scsi_cmnd
*scmd
)
1252 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
1255 static enum scsi_disposition
1256 scsi_eh_action(struct scsi_cmnd
*scmd
, enum scsi_disposition rtn
)
1258 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd
))) {
1259 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
1260 if (sdrv
->eh_action
)
1261 rtn
= sdrv
->eh_action(scmd
, rtn
);
1267 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1268 * @scmd: Original SCSI cmd that eh has finished.
1269 * @done_q: Queue for processed commands.
1272 * We don't want to use the normal command completion while we are are
1273 * still handling errors - it may cause other commands to be queued,
1274 * and that would disturb what we are doing. Thus we really want to
1275 * keep a list of pending commands for final completion, and once we
1276 * are ready to leave error handling we handle completion for real.
1278 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
1280 list_move_tail(&scmd
->eh_entry
, done_q
);
1282 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
1285 * scsi_eh_get_sense - Get device sense data.
1286 * @work_q: Queue of commands to process.
1287 * @done_q: Queue of processed commands.
1290 * See if we need to request sense information. if so, then get it
1291 * now, so we have a better idea of what to do.
1294 * This has the unfortunate side effect that if a shost adapter does
1295 * not automatically request sense information, we end up shutting
1296 * it down before we request it.
1298 * All drivers should request sense information internally these days,
1299 * so for now all I have to say is tough noogies if you end up in here.
1301 * XXX: Long term this code should go away, but that needs an audit of
1304 int scsi_eh_get_sense(struct list_head
*work_q
,
1305 struct list_head
*done_q
)
1307 struct scsi_cmnd
*scmd
, *next
;
1308 struct Scsi_Host
*shost
;
1309 enum scsi_disposition rtn
;
1312 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1313 * should not get sense.
1315 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1316 if ((scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) ||
1317 SCSI_SENSE_VALID(scmd
))
1320 shost
= scmd
->device
->host
;
1321 if (scsi_host_eh_past_deadline(shost
)) {
1322 SCSI_LOG_ERROR_RECOVERY(3,
1323 scmd_printk(KERN_INFO
, scmd
,
1324 "%s: skip request sense, past eh deadline\n",
1328 if (!scsi_status_is_check_condition(scmd
->result
))
1330 * don't request sense if there's no check condition
1331 * status because the error we're processing isn't one
1332 * that has a sense code (and some devices get
1333 * confused by sense requests out of the blue)
1337 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
1338 "%s: requesting sense\n",
1340 rtn
= scsi_request_sense(scmd
);
1344 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1345 "sense requested, result %x\n", scmd
->result
));
1346 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd
));
1348 rtn
= scsi_decide_disposition(scmd
);
1351 * if the result was normal, then just pass it along to the
1356 * We don't want this command reissued, just finished
1357 * with the sense data, so set retries to the max
1358 * allowed to ensure it won't get reissued. If the user
1359 * has requested infinite retries, we also want to
1360 * finish this command, so force completion by setting
1361 * retries and allowed to the same value.
1363 if (scmd
->allowed
== SCSI_CMD_RETRIES_NO_LIMIT
)
1364 scmd
->retries
= scmd
->allowed
= 1;
1366 scmd
->retries
= scmd
->allowed
;
1367 else if (rtn
!= NEEDS_RETRY
)
1370 scsi_eh_finish_cmd(scmd
, done_q
);
1373 return list_empty(work_q
);
1375 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
1378 * scsi_eh_tur - Send TUR to device.
1379 * @scmd: &scsi_cmnd to send TUR
1382 * 0 - Device is ready. 1 - Device NOT ready.
1384 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
1386 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
1388 enum scsi_disposition rtn
;
1391 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
1392 scmd
->device
->eh_timeout
, 0);
1394 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1395 "%s return: %x\n", __func__
, rtn
));
1410 * scsi_eh_test_devices - check if devices are responding from error recovery.
1411 * @cmd_list: scsi commands in error recovery.
1412 * @work_q: queue for commands which still need more error recovery
1413 * @done_q: queue for commands which are finished
1414 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1417 * Tests if devices are in a working state. Commands to devices now in
1418 * a working state are sent to the done_q while commands to devices which
1419 * are still failing to respond are returned to the work_q for more
1422 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1423 struct list_head
*work_q
,
1424 struct list_head
*done_q
, int try_stu
)
1426 struct scsi_cmnd
*scmd
, *next
;
1427 struct scsi_device
*sdev
;
1430 while (!list_empty(cmd_list
)) {
1431 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1432 sdev
= scmd
->device
;
1435 if (scsi_host_eh_past_deadline(sdev
->host
)) {
1436 /* Push items back onto work_q */
1437 list_splice_init(cmd_list
, work_q
);
1438 SCSI_LOG_ERROR_RECOVERY(3,
1439 sdev_printk(KERN_INFO
, sdev
,
1440 "%s: skip test device, past eh deadline",
1446 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1447 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1448 !scsi_eh_tur(scmd
)) ||
1451 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1452 if (scmd
->device
== sdev
) {
1455 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
))
1456 scsi_eh_finish_cmd(scmd
, done_q
);
1458 list_move_tail(&scmd
->eh_entry
, work_q
);
1461 return list_empty(work_q
);
1465 * scsi_eh_try_stu - Send START_UNIT to device.
1466 * @scmd: &scsi_cmnd to send START_UNIT
1469 * 0 - Device is ready. 1 - Device NOT ready.
1471 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1473 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1475 if (scmd
->device
->allow_restart
) {
1477 enum scsi_disposition rtn
= NEEDS_RETRY
;
1479 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1480 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6,
1481 scmd
->device
->eh_timeout
, 0);
1491 * scsi_eh_stu - send START_UNIT if needed
1492 * @shost: &scsi host being recovered.
1493 * @work_q: &list_head for pending commands.
1494 * @done_q: &list_head for processed commands.
1497 * If commands are failing due to not ready, initializing command required,
1498 * try revalidating the device, which will end up sending a start unit.
1500 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1501 struct list_head
*work_q
,
1502 struct list_head
*done_q
)
1504 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1505 struct scsi_device
*sdev
;
1507 shost_for_each_device(sdev
, shost
) {
1508 if (scsi_host_eh_past_deadline(shost
)) {
1509 SCSI_LOG_ERROR_RECOVERY(3,
1510 sdev_printk(KERN_INFO
, sdev
,
1511 "%s: skip START_UNIT, past eh deadline\n",
1513 scsi_device_put(sdev
);
1517 list_for_each_entry(scmd
, work_q
, eh_entry
)
1518 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1519 scsi_check_sense(scmd
) == FAILED
) {
1527 SCSI_LOG_ERROR_RECOVERY(3,
1528 sdev_printk(KERN_INFO
, sdev
,
1529 "%s: Sending START_UNIT\n",
1532 if (!scsi_eh_try_stu(stu_scmd
)) {
1533 if (!scsi_device_online(sdev
) ||
1534 !scsi_eh_tur(stu_scmd
)) {
1535 list_for_each_entry_safe(scmd
, next
,
1537 if (scmd
->device
== sdev
&&
1538 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
)
1539 scsi_eh_finish_cmd(scmd
, done_q
);
1543 SCSI_LOG_ERROR_RECOVERY(3,
1544 sdev_printk(KERN_INFO
, sdev
,
1545 "%s: START_UNIT failed\n",
1550 return list_empty(work_q
);
1555 * scsi_eh_bus_device_reset - send bdr if needed
1556 * @shost: scsi host being recovered.
1557 * @work_q: &list_head for pending commands.
1558 * @done_q: &list_head for processed commands.
1561 * Try a bus device reset. Still, look to see whether we have multiple
1562 * devices that are jammed or not - if we have multiple devices, it
1563 * makes no sense to try bus_device_reset - we really would need to try
1564 * a bus_reset instead.
1566 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1567 struct list_head
*work_q
,
1568 struct list_head
*done_q
)
1570 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1571 struct scsi_device
*sdev
;
1572 enum scsi_disposition rtn
;
1574 shost_for_each_device(sdev
, shost
) {
1575 if (scsi_host_eh_past_deadline(shost
)) {
1576 SCSI_LOG_ERROR_RECOVERY(3,
1577 sdev_printk(KERN_INFO
, sdev
,
1578 "%s: skip BDR, past eh deadline\n",
1580 scsi_device_put(sdev
);
1584 list_for_each_entry(scmd
, work_q
, eh_entry
)
1585 if (scmd
->device
== sdev
) {
1593 SCSI_LOG_ERROR_RECOVERY(3,
1594 sdev_printk(KERN_INFO
, sdev
,
1595 "%s: Sending BDR\n", current
->comm
));
1596 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1597 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1598 if (!scsi_device_online(sdev
) ||
1599 rtn
== FAST_IO_FAIL
||
1600 !scsi_eh_tur(bdr_scmd
)) {
1601 list_for_each_entry_safe(scmd
, next
,
1603 if (scmd
->device
== sdev
&&
1604 scsi_eh_action(scmd
, rtn
) != FAILED
)
1605 scsi_eh_finish_cmd(scmd
,
1610 SCSI_LOG_ERROR_RECOVERY(3,
1611 sdev_printk(KERN_INFO
, sdev
,
1612 "%s: BDR failed\n", current
->comm
));
1616 return list_empty(work_q
);
1620 * scsi_eh_target_reset - send target reset if needed
1621 * @shost: scsi host being recovered.
1622 * @work_q: &list_head for pending commands.
1623 * @done_q: &list_head for processed commands.
1626 * Try a target reset.
1628 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1629 struct list_head
*work_q
,
1630 struct list_head
*done_q
)
1632 LIST_HEAD(tmp_list
);
1633 LIST_HEAD(check_list
);
1635 list_splice_init(work_q
, &tmp_list
);
1637 while (!list_empty(&tmp_list
)) {
1638 struct scsi_cmnd
*next
, *scmd
;
1639 enum scsi_disposition rtn
;
1642 if (scsi_host_eh_past_deadline(shost
)) {
1643 /* push back on work queue for further processing */
1644 list_splice_init(&check_list
, work_q
);
1645 list_splice_init(&tmp_list
, work_q
);
1646 SCSI_LOG_ERROR_RECOVERY(3,
1647 shost_printk(KERN_INFO
, shost
,
1648 "%s: Skip target reset, past eh deadline\n",
1650 return list_empty(work_q
);
1653 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1656 SCSI_LOG_ERROR_RECOVERY(3,
1657 shost_printk(KERN_INFO
, shost
,
1658 "%s: Sending target reset to target %d\n",
1659 current
->comm
, id
));
1660 rtn
= scsi_try_target_reset(scmd
);
1661 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1662 SCSI_LOG_ERROR_RECOVERY(3,
1663 shost_printk(KERN_INFO
, shost
,
1664 "%s: Target reset failed"
1666 current
->comm
, id
));
1667 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1668 if (scmd_id(scmd
) != id
)
1672 list_move_tail(&scmd
->eh_entry
, &check_list
);
1673 else if (rtn
== FAST_IO_FAIL
)
1674 scsi_eh_finish_cmd(scmd
, done_q
);
1676 /* push back on work queue for further processing */
1677 list_move(&scmd
->eh_entry
, work_q
);
1681 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1685 * scsi_eh_bus_reset - send a bus reset
1686 * @shost: &scsi host being recovered.
1687 * @work_q: &list_head for pending commands.
1688 * @done_q: &list_head for processed commands.
1690 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1691 struct list_head
*work_q
,
1692 struct list_head
*done_q
)
1694 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1695 LIST_HEAD(check_list
);
1696 unsigned int channel
;
1697 enum scsi_disposition rtn
;
1700 * we really want to loop over the various channels, and do this on
1701 * a channel by channel basis. we should also check to see if any
1702 * of the failed commands are on soft_reset devices, and if so, skip
1706 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1707 if (scsi_host_eh_past_deadline(shost
)) {
1708 list_splice_init(&check_list
, work_q
);
1709 SCSI_LOG_ERROR_RECOVERY(3,
1710 shost_printk(KERN_INFO
, shost
,
1711 "%s: skip BRST, past eh deadline\n",
1713 return list_empty(work_q
);
1717 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1718 if (channel
== scmd_channel(scmd
)) {
1722 * FIXME add back in some support for
1723 * soft_reset devices.
1730 SCSI_LOG_ERROR_RECOVERY(3,
1731 shost_printk(KERN_INFO
, shost
,
1732 "%s: Sending BRST chan: %d\n",
1733 current
->comm
, channel
));
1734 rtn
= scsi_try_bus_reset(chan_scmd
);
1735 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1736 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1737 if (channel
== scmd_channel(scmd
)) {
1738 if (rtn
== FAST_IO_FAIL
)
1739 scsi_eh_finish_cmd(scmd
,
1742 list_move_tail(&scmd
->eh_entry
,
1747 SCSI_LOG_ERROR_RECOVERY(3,
1748 shost_printk(KERN_INFO
, shost
,
1749 "%s: BRST failed chan: %d\n",
1750 current
->comm
, channel
));
1753 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1757 * scsi_eh_host_reset - send a host reset
1758 * @shost: host to be reset.
1759 * @work_q: &list_head for pending commands.
1760 * @done_q: &list_head for processed commands.
1762 static int scsi_eh_host_reset(struct Scsi_Host
*shost
,
1763 struct list_head
*work_q
,
1764 struct list_head
*done_q
)
1766 struct scsi_cmnd
*scmd
, *next
;
1767 LIST_HEAD(check_list
);
1768 enum scsi_disposition rtn
;
1770 if (!list_empty(work_q
)) {
1771 scmd
= list_entry(work_q
->next
,
1772 struct scsi_cmnd
, eh_entry
);
1774 SCSI_LOG_ERROR_RECOVERY(3,
1775 shost_printk(KERN_INFO
, shost
,
1776 "%s: Sending HRST\n",
1779 rtn
= scsi_try_host_reset(scmd
);
1780 if (rtn
== SUCCESS
) {
1781 list_splice_init(work_q
, &check_list
);
1782 } else if (rtn
== FAST_IO_FAIL
) {
1783 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1784 scsi_eh_finish_cmd(scmd
, done_q
);
1787 SCSI_LOG_ERROR_RECOVERY(3,
1788 shost_printk(KERN_INFO
, shost
,
1789 "%s: HRST failed\n",
1793 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1797 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1798 * @work_q: &list_head for pending commands.
1799 * @done_q: &list_head for processed commands.
1801 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1802 struct list_head
*done_q
)
1804 struct scsi_cmnd
*scmd
, *next
;
1805 struct scsi_device
*sdev
;
1807 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1808 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1809 "not ready after error recovery\n");
1810 sdev
= scmd
->device
;
1812 mutex_lock(&sdev
->state_mutex
);
1813 scsi_device_set_state(sdev
, SDEV_OFFLINE
);
1814 mutex_unlock(&sdev
->state_mutex
);
1816 scsi_eh_finish_cmd(scmd
, done_q
);
1822 * scsi_noretry_cmd - determine if command should be failed fast
1823 * @scmd: SCSI cmd to examine.
1825 bool scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1827 struct request
*req
= scsi_cmd_to_rq(scmd
);
1829 switch (host_byte(scmd
->result
)) {
1835 return !!(req
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1837 return !!(req
->cmd_flags
& REQ_FAILFAST_DEV
);
1839 if (get_status_byte(scmd
) == SAM_STAT_RESERVATION_CONFLICT
)
1842 case DID_SOFT_ERROR
:
1843 return !!(req
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1846 /* Never retry commands aborted due to a duration limit timeout */
1847 if (scsi_ml_byte(scmd
->result
) == SCSIML_STAT_DL_TIMEOUT
)
1850 if (!scsi_status_is_check_condition(scmd
->result
))
1855 * assume caller has checked sense and determined
1856 * the check condition was retryable.
1858 if (req
->cmd_flags
& REQ_FAILFAST_DEV
|| blk_rq_is_passthrough(req
))
1865 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1866 * @scmd: SCSI cmd to examine.
1869 * This is *only* called when we are examining the status after sending
1870 * out the actual data command. any commands that are queued for error
1871 * recovery (e.g. test_unit_ready) do *not* come through here.
1873 * When this routine returns failed, it means the error handler thread
1874 * is woken. In cases where the error code indicates an error that
1875 * doesn't require the error handler read (i.e. we don't need to
1876 * abort/reset), this function should return SUCCESS.
1878 enum scsi_disposition
scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1880 enum scsi_disposition rtn
;
1883 * if the device is offline, then we clearly just pass the result back
1884 * up to the top level.
1886 if (!scsi_device_online(scmd
->device
)) {
1887 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO
, scmd
,
1888 "%s: device offline - report as SUCCESS\n", __func__
));
1893 * first check the host byte, to see if there is anything in there
1894 * that would indicate what we need to do.
1896 switch (host_byte(scmd
->result
)) {
1897 case DID_PASSTHROUGH
:
1899 * no matter what, pass this through to the upper layer.
1900 * nuke this special code so that it looks like we are saying
1903 scmd
->result
&= 0xff00ffff;
1907 * looks good. drop through, and check the next byte.
1911 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
1912 set_host_byte(scmd
, DID_TIME_OUT
);
1916 case DID_NO_CONNECT
:
1917 case DID_BAD_TARGET
:
1919 * note - this means that we just report the status back
1920 * to the top level driver, not that we actually think
1921 * that it indicates SUCCESS.
1924 case DID_SOFT_ERROR
:
1926 * when the low level driver returns did_soft_error,
1927 * it is responsible for keeping an internal retry counter
1928 * in order to avoid endless loops (db)
1935 return ADD_TO_MLQUEUE
;
1936 case DID_TRANSPORT_DISRUPTED
:
1938 * LLD/transport was disrupted during processing of the IO.
1939 * The transport class is now blocked/blocking,
1940 * and the transport will decide what to do with the IO
1941 * based on its timers and recovery capablilities if
1942 * there are enough retries.
1945 case DID_TRANSPORT_FAILFAST
:
1947 * The transport decided to failfast the IO (most likely
1948 * the fast io fail tmo fired), so send IO directly upwards.
1951 case DID_TRANSPORT_MARGINAL
:
1953 * caller has decided not to do retries on
1954 * abort success, so send IO directly upwards
1958 if (get_status_byte(scmd
) == SAM_STAT_RESERVATION_CONFLICT
)
1960 * execute reservation conflict processing code
1970 * when we scan the bus, we get timeout messages for
1971 * these commands if there is no device available.
1972 * other hosts report did_no_connect for the same thing.
1974 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1975 scmd
->cmnd
[0] == INQUIRY
)) {
1987 * check the status byte to see if this indicates anything special.
1989 switch (get_status_byte(scmd
)) {
1990 case SAM_STAT_TASK_SET_FULL
:
1991 scsi_handle_queue_full(scmd
->device
);
1993 * the case of trying to send too many commands to a
1994 * tagged queueing device.
1999 * device can't talk to us at the moment. Should only
2000 * occur (SAM-3) when the task queue is empty, so will cause
2001 * the empty queue handling to trigger a stall in the
2004 return ADD_TO_MLQUEUE
;
2006 if (scmd
->cmnd
[0] == REPORT_LUNS
)
2007 scmd
->device
->sdev_target
->expecting_lun_change
= 0;
2008 scsi_handle_queue_ramp_up(scmd
->device
);
2009 if (scmd
->sense_buffer
&& SCSI_SENSE_VALID(scmd
))
2011 * If we have sense data, call scsi_check_sense() in
2012 * order to set the correct SCSI ML byte (if any).
2013 * No point in checking the return value, since the
2014 * command has already completed successfully.
2016 scsi_check_sense(scmd
);
2018 case SAM_STAT_COMMAND_TERMINATED
:
2020 case SAM_STAT_TASK_ABORTED
:
2022 case SAM_STAT_CHECK_CONDITION
:
2023 rtn
= scsi_check_sense(scmd
);
2024 if (rtn
== NEEDS_RETRY
)
2026 /* if rtn == FAILED, we have no sense information;
2027 * returning FAILED will wake the error handler thread
2028 * to collect the sense and redo the decide
2031 case SAM_STAT_CONDITION_MET
:
2032 case SAM_STAT_INTERMEDIATE
:
2033 case SAM_STAT_INTERMEDIATE_CONDITION_MET
:
2034 case SAM_STAT_ACA_ACTIVE
:
2036 * who knows? FIXME(eric)
2040 case SAM_STAT_RESERVATION_CONFLICT
:
2041 sdev_printk(KERN_INFO
, scmd
->device
,
2042 "reservation conflict\n");
2043 set_scsi_ml_byte(scmd
, SCSIML_STAT_RESV_CONFLICT
);
2044 return SUCCESS
; /* causes immediate i/o error */
2050 /* we requeue for retry because the error was retryable, and
2051 * the request was not marked fast fail. Note that above,
2052 * even if the request is marked fast fail, we still requeue
2053 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2054 if (scsi_cmd_retry_allowed(scmd
) && !scsi_noretry_cmd(scmd
)) {
2058 * no more retries - report this one back to upper level.
2064 static enum rq_end_io_ret
eh_lock_door_done(struct request
*req
,
2065 blk_status_t status
)
2067 blk_mq_free_request(req
);
2068 return RQ_END_IO_NONE
;
2072 * scsi_eh_lock_door - Prevent medium removal for the specified device
2073 * @sdev: SCSI device to prevent medium removal
2076 * We must be called from process context.
2079 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2080 * head of the devices request queue, and continue.
2082 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
2084 struct scsi_cmnd
*scmd
;
2085 struct request
*req
;
2087 req
= scsi_alloc_request(sdev
->request_queue
, REQ_OP_DRV_IN
, 0);
2090 scmd
= blk_mq_rq_to_pdu(req
);
2092 scmd
->cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
2096 scmd
->cmnd
[4] = SCSI_REMOVAL_PREVENT
;
2098 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
2101 req
->rq_flags
|= RQF_QUIET
;
2102 req
->timeout
= 10 * HZ
;
2103 req
->end_io
= eh_lock_door_done
;
2105 blk_execute_rq_nowait(req
, true);
2109 * scsi_restart_operations - restart io operations to the specified host.
2110 * @shost: Host we are restarting.
2113 * When we entered the error handler, we blocked all further i/o to
2114 * this device. we need to 'reverse' this process.
2116 static void scsi_restart_operations(struct Scsi_Host
*shost
)
2118 struct scsi_device
*sdev
;
2119 unsigned long flags
;
2122 * If the door was locked, we need to insert a door lock request
2123 * onto the head of the SCSI request queue for the device. There
2124 * is no point trying to lock the door of an off-line device.
2126 shost_for_each_device(sdev
, shost
) {
2127 if (scsi_device_online(sdev
) && sdev
->was_reset
&& sdev
->locked
) {
2128 scsi_eh_lock_door(sdev
);
2129 sdev
->was_reset
= 0;
2134 * next free up anything directly waiting upon the host. this
2135 * will be requests for character device operations, and also for
2136 * ioctls to queued block devices.
2138 SCSI_LOG_ERROR_RECOVERY(3,
2139 shost_printk(KERN_INFO
, shost
, "waking up host to restart\n"));
2141 spin_lock_irqsave(shost
->host_lock
, flags
);
2142 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
2143 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
2144 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
2145 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2147 wake_up(&shost
->host_wait
);
2150 * finally we need to re-initiate requests that may be pending. we will
2151 * have had everything blocked while error handling is taking place, and
2152 * now that error recovery is done, we will need to ensure that these
2153 * requests are started.
2155 scsi_run_host_queues(shost
);
2158 * if eh is active and host_eh_scheduled is pending we need to re-run
2159 * recovery. we do this check after scsi_run_host_queues() to allow
2160 * everything pent up since the last eh run a chance to make forward
2161 * progress before we sync again. Either we'll immediately re-run
2162 * recovery or scsi_device_unbusy() will wake us again when these
2163 * pending commands complete.
2165 spin_lock_irqsave(shost
->host_lock
, flags
);
2166 if (shost
->host_eh_scheduled
)
2167 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
2168 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
2169 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2173 * scsi_eh_ready_devs - check device ready state and recover if not.
2174 * @shost: host to be recovered.
2175 * @work_q: &list_head for pending commands.
2176 * @done_q: &list_head for processed commands.
2178 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
2179 struct list_head
*work_q
,
2180 struct list_head
*done_q
)
2182 if (!scsi_eh_stu(shost
, work_q
, done_q
))
2183 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
2184 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
2185 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
2186 if (!scsi_eh_host_reset(shost
, work_q
, done_q
))
2187 scsi_eh_offline_sdevs(work_q
,
2190 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
2193 * scsi_eh_flush_done_q - finish processed commands or retry them.
2194 * @done_q: list_head of processed commands.
2196 void scsi_eh_flush_done_q(struct list_head
*done_q
)
2198 struct scsi_cmnd
*scmd
, *next
;
2200 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
2201 struct scsi_device
*sdev
= scmd
->device
;
2203 list_del_init(&scmd
->eh_entry
);
2204 if (scsi_device_online(sdev
) && !scsi_noretry_cmd(scmd
) &&
2205 scsi_cmd_retry_allowed(scmd
) &&
2206 scsi_eh_should_retry_cmd(scmd
)) {
2207 SCSI_LOG_ERROR_RECOVERY(3,
2208 scmd_printk(KERN_INFO
, scmd
,
2209 "%s: flush retry cmd\n",
2211 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
2212 blk_mq_kick_requeue_list(sdev
->request_queue
);
2215 * If just we got sense for the device (called
2216 * scsi_eh_get_sense), scmd->result is already
2217 * set, do not set DID_TIME_OUT.
2219 if (!scmd
->result
&&
2220 !(scmd
->flags
& SCMD_FORCE_EH_SUCCESS
))
2221 scmd
->result
|= (DID_TIME_OUT
<< 16);
2222 SCSI_LOG_ERROR_RECOVERY(3,
2223 scmd_printk(KERN_INFO
, scmd
,
2224 "%s: flush finish cmd\n",
2226 scsi_finish_command(scmd
);
2230 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
2233 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2234 * @shost: Host to unjam.
2237 * When we come in here, we *know* that all commands on the bus have
2238 * either completed, failed or timed out. we also know that no further
2239 * commands are being sent to the host, so things are relatively quiet
2240 * and we have freedom to fiddle with things as we wish.
2242 * This is only the *default* implementation. it is possible for
2243 * individual drivers to supply their own version of this function, and
2244 * if the maintainer wishes to do this, it is strongly suggested that
2245 * this function be taken as a template and modified. this function
2246 * was designed to correctly handle problems for about 95% of the
2247 * different cases out there, and it should always provide at least a
2248 * reasonable amount of error recovery.
2250 * Any command marked 'failed' or 'timeout' must eventually have
2251 * scsi_finish_cmd() called for it. we do all of the retry stuff
2252 * here, so when we restart the host after we return it should have an
2255 static void scsi_unjam_host(struct Scsi_Host
*shost
)
2257 unsigned long flags
;
2258 LIST_HEAD(eh_work_q
);
2259 LIST_HEAD(eh_done_q
);
2261 spin_lock_irqsave(shost
->host_lock
, flags
);
2262 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
2263 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2265 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
2267 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
2268 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
2270 spin_lock_irqsave(shost
->host_lock
, flags
);
2271 if (shost
->eh_deadline
!= -1)
2272 shost
->last_reset
= 0;
2273 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2274 scsi_eh_flush_done_q(&eh_done_q
);
2278 * scsi_error_handler - SCSI error handler thread
2279 * @data: Host for which we are running.
2282 * This is the main error handling loop. This is run as a kernel thread
2283 * for every SCSI host and handles all error handling activity.
2285 int scsi_error_handler(void *data
)
2287 struct Scsi_Host
*shost
= data
;
2290 * We use TASK_INTERRUPTIBLE so that the thread is not
2291 * counted against the load average as a running process.
2292 * We never actually get interrupted because kthread_run
2293 * disables signal delivery for the created thread.
2297 * The sequence in kthread_stop() sets the stop flag first
2298 * then wakes the process. To avoid missed wakeups, the task
2299 * should always be in a non running state before the stop
2302 set_current_state(TASK_INTERRUPTIBLE
);
2303 if (kthread_should_stop())
2306 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
2307 shost
->host_failed
!= scsi_host_busy(shost
)) {
2308 SCSI_LOG_ERROR_RECOVERY(1,
2309 shost_printk(KERN_INFO
, shost
,
2310 "scsi_eh_%d: sleeping\n",
2316 __set_current_state(TASK_RUNNING
);
2317 SCSI_LOG_ERROR_RECOVERY(1,
2318 shost_printk(KERN_INFO
, shost
,
2319 "scsi_eh_%d: waking up %d/%d/%d\n",
2320 shost
->host_no
, shost
->host_eh_scheduled
,
2322 scsi_host_busy(shost
)));
2325 * We have a host that is failing for some reason. Figure out
2326 * what we need to do to get it up and online again (if we can).
2327 * If we fail, we end up taking the thing offline.
2329 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
2330 SCSI_LOG_ERROR_RECOVERY(1,
2331 shost_printk(KERN_ERR
, shost
,
2332 "scsi_eh_%d: unable to autoresume\n",
2337 if (shost
->transportt
->eh_strategy_handler
)
2338 shost
->transportt
->eh_strategy_handler(shost
);
2340 scsi_unjam_host(shost
);
2342 /* All scmds have been handled */
2343 shost
->host_failed
= 0;
2346 * Note - if the above fails completely, the action is to take
2347 * individual devices offline and flush the queue of any
2348 * outstanding requests that may have been pending. When we
2349 * restart, we restart any I/O to any other devices on the bus
2350 * which are still online.
2352 scsi_restart_operations(shost
);
2353 if (!shost
->eh_noresume
)
2354 scsi_autopm_put_host(shost
);
2356 __set_current_state(TASK_RUNNING
);
2358 SCSI_LOG_ERROR_RECOVERY(1,
2359 shost_printk(KERN_INFO
, shost
,
2360 "Error handler scsi_eh_%d exiting\n",
2362 shost
->ehandler
= NULL
;
2367 * Function: scsi_report_bus_reset()
2369 * Purpose: Utility function used by low-level drivers to report that
2370 * they have observed a bus reset on the bus being handled.
2372 * Arguments: shost - Host in question
2373 * channel - channel on which reset was observed.
2377 * Lock status: Host lock must be held.
2379 * Notes: This only needs to be called if the reset is one which
2380 * originates from an unknown location. Resets originated
2381 * by the mid-level itself don't need to call this, but there
2382 * should be no harm.
2384 * The main purpose of this is to make sure that a CHECK_CONDITION
2385 * is properly treated.
2387 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
2389 struct scsi_device
*sdev
;
2391 __shost_for_each_device(sdev
, shost
) {
2392 if (channel
== sdev_channel(sdev
))
2393 __scsi_report_device_reset(sdev
, NULL
);
2396 EXPORT_SYMBOL(scsi_report_bus_reset
);
2399 * Function: scsi_report_device_reset()
2401 * Purpose: Utility function used by low-level drivers to report that
2402 * they have observed a device reset on the device being handled.
2404 * Arguments: shost - Host in question
2405 * channel - channel on which reset was observed
2406 * target - target on which reset was observed
2410 * Lock status: Host lock must be held
2412 * Notes: This only needs to be called if the reset is one which
2413 * originates from an unknown location. Resets originated
2414 * by the mid-level itself don't need to call this, but there
2415 * should be no harm.
2417 * The main purpose of this is to make sure that a CHECK_CONDITION
2418 * is properly treated.
2420 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
2422 struct scsi_device
*sdev
;
2424 __shost_for_each_device(sdev
, shost
) {
2425 if (channel
== sdev_channel(sdev
) &&
2426 target
== sdev_id(sdev
))
2427 __scsi_report_device_reset(sdev
, NULL
);
2430 EXPORT_SYMBOL(scsi_report_device_reset
);
2433 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2434 * @dev: scsi_device to operate on
2435 * @arg: reset type (see sg.h)
2438 scsi_ioctl_reset(struct scsi_device
*dev
, int __user
*arg
)
2440 struct scsi_cmnd
*scmd
;
2441 struct Scsi_Host
*shost
= dev
->host
;
2443 unsigned long flags
;
2445 enum scsi_disposition rtn
;
2447 if (!capable(CAP_SYS_ADMIN
) || !capable(CAP_SYS_RAWIO
))
2450 error
= get_user(val
, arg
);
2454 if (scsi_autopm_get_host(shost
) < 0)
2458 rq
= kzalloc(sizeof(struct request
) + sizeof(struct scsi_cmnd
) +
2459 shost
->hostt
->cmd_size
, GFP_KERNEL
);
2461 goto out_put_autopm_host
;
2462 blk_rq_init(NULL
, rq
);
2464 scmd
= (struct scsi_cmnd
*)(rq
+ 1);
2465 scsi_init_command(dev
, scmd
);
2467 scmd
->submitter
= SUBMITTED_BY_SCSI_RESET_IOCTL
;
2468 scmd
->flags
|= SCMD_LAST
;
2469 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2473 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2475 spin_lock_irqsave(shost
->host_lock
, flags
);
2476 shost
->tmf_in_progress
= 1;
2477 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2479 switch (val
& ~SG_SCSI_RESET_NO_ESCALATE
) {
2480 case SG_SCSI_RESET_NOTHING
:
2483 case SG_SCSI_RESET_DEVICE
:
2484 rtn
= scsi_try_bus_device_reset(scmd
);
2485 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2488 case SG_SCSI_RESET_TARGET
:
2489 rtn
= scsi_try_target_reset(scmd
);
2490 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2493 case SG_SCSI_RESET_BUS
:
2494 rtn
= scsi_try_bus_reset(scmd
);
2495 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2498 case SG_SCSI_RESET_HOST
:
2499 rtn
= scsi_try_host_reset(scmd
);
2508 error
= (rtn
== SUCCESS
) ? 0 : -EIO
;
2510 spin_lock_irqsave(shost
->host_lock
, flags
);
2511 shost
->tmf_in_progress
= 0;
2512 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2515 * be sure to wake up anyone who was sleeping or had their queue
2516 * suspended while we performed the TMF.
2518 SCSI_LOG_ERROR_RECOVERY(3,
2519 shost_printk(KERN_INFO
, shost
,
2520 "waking up host to restart after TMF\n"));
2522 wake_up(&shost
->host_wait
);
2523 scsi_run_host_queues(shost
);
2527 out_put_autopm_host
:
2528 scsi_autopm_put_host(shost
);
2532 bool scsi_command_normalize_sense(const struct scsi_cmnd
*cmd
,
2533 struct scsi_sense_hdr
*sshdr
)
2535 return scsi_normalize_sense(cmd
->sense_buffer
,
2536 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2538 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2541 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2542 * @sense_buffer: byte array of sense data
2543 * @sb_len: number of valid bytes in sense_buffer
2544 * @info_out: pointer to 64 integer where 8 or 4 byte information
2545 * field will be placed if found.
2548 * true if information field found, false if not found.
2550 bool scsi_get_sense_info_fld(const u8
*sense_buffer
, int sb_len
,
2557 switch (sense_buffer
[0] & 0x7f) {
2560 if (sense_buffer
[0] & 0x80) {
2561 *info_out
= get_unaligned_be32(&sense_buffer
[3]);
2567 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2569 if (ucp
&& (0xa == ucp
[1])) {
2570 *info_out
= get_unaligned_be64(&ucp
[4]);
2578 EXPORT_SYMBOL(scsi_get_sense_info_fld
);