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 <asm/unaligned.h>
53 static void scsi_eh_done(struct scsi_cmnd
*scmd
);
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
59 #define BUS_RESET_SETTLE_TIME (10)
60 #define HOST_RESET_SETTLE_TIME (10)
62 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
63 static int scsi_try_to_abort_cmd(struct scsi_host_template
*,
66 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
68 lockdep_assert_held(shost
->host_lock
);
70 if (scsi_host_busy(shost
) == shost
->host_failed
) {
71 trace_scsi_eh_wakeup(shost
);
72 wake_up_process(shost
->ehandler
);
73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO
, shost
,
74 "Waking error handler thread\n"));
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
82 * Schedule SCSI EH without scmd.
84 void scsi_schedule_eh(struct Scsi_Host
*shost
)
88 spin_lock_irqsave(shost
->host_lock
, flags
);
90 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
91 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
92 shost
->host_eh_scheduled
++;
93 scsi_eh_wakeup(shost
);
96 spin_unlock_irqrestore(shost
->host_lock
, flags
);
98 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
100 static int scsi_host_eh_past_deadline(struct Scsi_Host
*shost
)
102 if (!shost
->last_reset
|| shost
->eh_deadline
== -1)
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
109 * if eh_deadline has been set to 'off' during the
112 if (time_before(jiffies
, shost
->last_reset
+ shost
->eh_deadline
) &&
113 shost
->eh_deadline
> -1)
120 * scmd_eh_abort_handler - Handle command aborts
121 * @work: command to be aborted.
123 * Note: this function must be called only for a command that has timed out.
124 * Because the block layer marks a request as complete before it calls
125 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
126 * timed out do not have any effect. Hence it is safe to call
127 * scsi_finish_command() from this function.
130 scmd_eh_abort_handler(struct work_struct
*work
)
132 struct scsi_cmnd
*scmd
=
133 container_of(work
, struct scsi_cmnd
, abort_work
.work
);
134 struct scsi_device
*sdev
= scmd
->device
;
137 if (scsi_host_eh_past_deadline(sdev
->host
)) {
138 SCSI_LOG_ERROR_RECOVERY(3,
139 scmd_printk(KERN_INFO
, scmd
,
140 "eh timeout, not aborting\n"));
142 SCSI_LOG_ERROR_RECOVERY(3,
143 scmd_printk(KERN_INFO
, scmd
,
144 "aborting command\n"));
145 rtn
= scsi_try_to_abort_cmd(sdev
->host
->hostt
, scmd
);
146 if (rtn
== SUCCESS
) {
147 set_host_byte(scmd
, DID_TIME_OUT
);
148 if (scsi_host_eh_past_deadline(sdev
->host
)) {
149 SCSI_LOG_ERROR_RECOVERY(3,
150 scmd_printk(KERN_INFO
, scmd
,
151 "eh timeout, not retrying "
152 "aborted command\n"));
153 } else if (!scsi_noretry_cmd(scmd
) &&
154 (++scmd
->retries
<= scmd
->allowed
)) {
155 SCSI_LOG_ERROR_RECOVERY(3,
156 scmd_printk(KERN_WARNING
, scmd
,
157 "retry aborted command\n"));
158 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
161 SCSI_LOG_ERROR_RECOVERY(3,
162 scmd_printk(KERN_WARNING
, scmd
,
163 "finish aborted command\n"));
164 scsi_finish_command(scmd
);
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO
, scmd
,
171 (rtn
== FAST_IO_FAIL
) ?
172 "not send" : "failed"));
176 scsi_eh_scmd_add(scmd
);
180 * scsi_abort_command - schedule a command abort
181 * @scmd: scmd to abort.
183 * We only need to abort commands after a command timeout
186 scsi_abort_command(struct scsi_cmnd
*scmd
)
188 struct scsi_device
*sdev
= scmd
->device
;
189 struct Scsi_Host
*shost
= sdev
->host
;
192 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
194 * Retry after abort failed, escalate to next level.
196 SCSI_LOG_ERROR_RECOVERY(3,
197 scmd_printk(KERN_INFO
, scmd
,
198 "previous abort failed\n"));
199 BUG_ON(delayed_work_pending(&scmd
->abort_work
));
203 spin_lock_irqsave(shost
->host_lock
, flags
);
204 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
205 shost
->last_reset
= jiffies
;
206 spin_unlock_irqrestore(shost
->host_lock
, flags
);
208 scmd
->eh_eflags
|= SCSI_EH_ABORT_SCHEDULED
;
209 SCSI_LOG_ERROR_RECOVERY(3,
210 scmd_printk(KERN_INFO
, scmd
, "abort scheduled\n"));
211 queue_delayed_work(shost
->tmf_work_q
, &scmd
->abort_work
, HZ
/ 100);
216 * scsi_eh_reset - call into ->eh_action to reset internal counters
217 * @scmd: scmd to run eh on.
219 * The scsi driver might be carrying internal state about the
220 * devices, so we need to call into the driver to reset the
221 * internal state once the error handler is started.
223 static void scsi_eh_reset(struct scsi_cmnd
*scmd
)
225 if (!blk_rq_is_passthrough(scmd
->request
)) {
226 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
228 sdrv
->eh_reset(scmd
);
232 static void scsi_eh_inc_host_failed(struct rcu_head
*head
)
234 struct scsi_cmnd
*scmd
= container_of(head
, typeof(*scmd
), rcu
);
235 struct Scsi_Host
*shost
= scmd
->device
->host
;
238 spin_lock_irqsave(shost
->host_lock
, flags
);
239 shost
->host_failed
++;
240 scsi_eh_wakeup(shost
);
241 spin_unlock_irqrestore(shost
->host_lock
, flags
);
245 * scsi_eh_scmd_add - add scsi cmd to error handling.
246 * @scmd: scmd to run eh on.
248 void scsi_eh_scmd_add(struct scsi_cmnd
*scmd
)
250 struct Scsi_Host
*shost
= scmd
->device
->host
;
254 WARN_ON_ONCE(!shost
->ehandler
);
256 spin_lock_irqsave(shost
->host_lock
, flags
);
257 if (scsi_host_set_state(shost
, SHOST_RECOVERY
)) {
258 ret
= scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
);
261 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
262 shost
->last_reset
= jiffies
;
265 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
266 spin_unlock_irqrestore(shost
->host_lock
, flags
);
268 * Ensure that all tasks observe the host state change before the
269 * host_failed change.
271 call_rcu(&scmd
->rcu
, scsi_eh_inc_host_failed
);
275 * scsi_times_out - Timeout function for normal scsi commands.
276 * @req: request that is timing out.
279 * We do not need to lock this. There is the potential for a race
280 * only in that the normal completion handling might run, but if the
281 * normal completion function determines that the timer has already
282 * fired, then it mustn't do anything.
284 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
286 struct scsi_cmnd
*scmd
= blk_mq_rq_to_pdu(req
);
287 enum blk_eh_timer_return rtn
= BLK_EH_DONE
;
288 struct Scsi_Host
*host
= scmd
->device
->host
;
290 trace_scsi_dispatch_cmd_timeout(scmd
);
291 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
293 if (host
->eh_deadline
!= -1 && !host
->last_reset
)
294 host
->last_reset
= jiffies
;
296 if (host
->hostt
->eh_timed_out
)
297 rtn
= host
->hostt
->eh_timed_out(scmd
);
299 if (rtn
== BLK_EH_DONE
) {
301 * Set the command to complete first in order to prevent a real
302 * completion from releasing the command while error handling
303 * is using it. If the command was already completed, then the
304 * lower level driver beat the timeout handler, and it is safe
305 * to return without escalating error recovery.
307 * If timeout handling lost the race to a real completion, the
308 * block layer may ignore that due to a fake timeout injection,
309 * so return RESET_TIMER to allow error handling another shot
312 if (test_and_set_bit(SCMD_STATE_COMPLETE
, &scmd
->state
))
313 return BLK_EH_RESET_TIMER
;
314 if (scsi_abort_command(scmd
) != SUCCESS
) {
315 set_host_byte(scmd
, DID_TIME_OUT
);
316 scsi_eh_scmd_add(scmd
);
324 * scsi_block_when_processing_errors - Prevent cmds from being queued.
325 * @sdev: Device on which we are performing recovery.
328 * We block until the host is out of error recovery, and then check to
329 * see whether the host or the device is offline.
332 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
334 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
338 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
340 online
= scsi_device_online(sdev
);
344 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
346 #ifdef CONFIG_SCSI_LOGGING
348 * scsi_eh_prt_fail_stats - Log info on failures.
349 * @shost: scsi host being recovered.
350 * @work_q: Queue of scsi cmds to process.
352 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
353 struct list_head
*work_q
)
355 struct scsi_cmnd
*scmd
;
356 struct scsi_device
*sdev
;
357 int total_failures
= 0;
360 int devices_failed
= 0;
362 shost_for_each_device(sdev
, shost
) {
363 list_for_each_entry(scmd
, work_q
, eh_entry
) {
364 if (scmd
->device
== sdev
) {
366 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
)
373 if (cmd_cancel
|| cmd_failed
) {
374 SCSI_LOG_ERROR_RECOVERY(3,
375 shost_printk(KERN_INFO
, shost
,
376 "%s: cmds failed: %d, cancel: %d\n",
377 __func__
, cmd_failed
,
385 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO
, shost
,
386 "Total of %d commands on %d"
387 " devices require eh work\n",
388 total_failures
, devices_failed
));
393 * scsi_report_lun_change - Set flag on all *other* devices on the same target
394 * to indicate that a UNIT ATTENTION is expected.
395 * @sdev: Device reporting the UNIT ATTENTION
397 static void scsi_report_lun_change(struct scsi_device
*sdev
)
399 sdev
->sdev_target
->expecting_lun_change
= 1;
403 * scsi_report_sense - Examine scsi sense information and log messages for
404 * certain conditions, also issue uevents for some of them.
405 * @sdev: Device reporting the sense code
406 * @sshdr: sshdr to be examined
408 static void scsi_report_sense(struct scsi_device
*sdev
,
409 struct scsi_sense_hdr
*sshdr
)
411 enum scsi_device_event evt_type
= SDEV_EVT_MAXBITS
; /* i.e. none */
413 if (sshdr
->sense_key
== UNIT_ATTENTION
) {
414 if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x03) {
415 evt_type
= SDEV_EVT_INQUIRY_CHANGE_REPORTED
;
416 sdev_printk(KERN_WARNING
, sdev
,
417 "Inquiry data has changed");
418 } else if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x0e) {
419 evt_type
= SDEV_EVT_LUN_CHANGE_REPORTED
;
420 scsi_report_lun_change(sdev
);
421 sdev_printk(KERN_WARNING
, sdev
,
422 "Warning! Received an indication that the "
423 "LUN assignments on this target have "
424 "changed. The Linux SCSI layer does not "
425 "automatically remap LUN assignments.\n");
426 } else if (sshdr
->asc
== 0x3f)
427 sdev_printk(KERN_WARNING
, sdev
,
428 "Warning! Received an indication that the "
429 "operating parameters on this target have "
430 "changed. The Linux SCSI layer does not "
431 "automatically adjust these parameters.\n");
433 if (sshdr
->asc
== 0x38 && sshdr
->ascq
== 0x07) {
434 evt_type
= SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
;
435 sdev_printk(KERN_WARNING
, sdev
,
436 "Warning! Received an indication that the "
437 "LUN reached a thin provisioning soft "
441 if (sshdr
->asc
== 0x29) {
442 evt_type
= SDEV_EVT_POWER_ON_RESET_OCCURRED
;
443 sdev_printk(KERN_WARNING
, sdev
,
444 "Power-on or device reset occurred\n");
447 if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x01) {
448 evt_type
= SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
;
449 sdev_printk(KERN_WARNING
, sdev
,
450 "Mode parameters changed");
451 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x06) {
452 evt_type
= SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
;
453 sdev_printk(KERN_WARNING
, sdev
,
454 "Asymmetric access state changed");
455 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x09) {
456 evt_type
= SDEV_EVT_CAPACITY_CHANGE_REPORTED
;
457 sdev_printk(KERN_WARNING
, sdev
,
458 "Capacity data has changed");
459 } else if (sshdr
->asc
== 0x2a)
460 sdev_printk(KERN_WARNING
, sdev
,
461 "Parameters changed");
464 if (evt_type
!= SDEV_EVT_MAXBITS
) {
465 set_bit(evt_type
, sdev
->pending_events
);
466 schedule_work(&sdev
->event_work
);
471 * scsi_check_sense - Examine scsi cmd sense
472 * @scmd: Cmd to have sense checked.
475 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
478 * When a deferred error is detected the current command has
479 * not been executed and needs retrying.
481 int scsi_check_sense(struct scsi_cmnd
*scmd
)
483 struct scsi_device
*sdev
= scmd
->device
;
484 struct scsi_sense_hdr sshdr
;
486 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
487 return FAILED
; /* no valid sense data */
489 scsi_report_sense(sdev
, &sshdr
);
491 if (scsi_sense_is_deferred(&sshdr
))
494 if (sdev
->handler
&& sdev
->handler
->check_sense
) {
497 rc
= sdev
->handler
->check_sense(sdev
, &sshdr
);
498 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
500 /* handler does not care. Drop down to default handling */
503 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&& scmd
->scsi_done
!= scsi_eh_done
)
505 * nasty: for mid-layer issued TURs, we need to return the
506 * actual sense data without any recovery attempt. For eh
507 * issued ones, we need to try to recover and interpret
512 * Previous logic looked for FILEMARK, EOM or ILI which are
513 * mainly associated with tapes and returned SUCCESS.
515 if (sshdr
.response_code
== 0x70) {
517 if (scmd
->sense_buffer
[2] & 0xe0)
521 * descriptor format: look for "stream commands sense data
522 * descriptor" (see SSC-3). Assume single sense data
523 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
525 if ((sshdr
.additional_length
> 3) &&
526 (scmd
->sense_buffer
[8] == 0x4) &&
527 (scmd
->sense_buffer
[11] & 0xe0))
531 switch (sshdr
.sense_key
) {
534 case RECOVERED_ERROR
:
535 return /* soft_error */ SUCCESS
;
537 case ABORTED_COMMAND
:
538 if (sshdr
.asc
== 0x10) /* DIF */
541 if (sshdr
.asc
== 0x44 && sdev
->sdev_bflags
& BLIST_RETRY_ITF
)
542 return ADD_TO_MLQUEUE
;
543 if (sshdr
.asc
== 0xc1 && sshdr
.ascq
== 0x01 &&
544 sdev
->sdev_bflags
& BLIST_RETRY_ASC_C1
)
545 return ADD_TO_MLQUEUE
;
551 * if we are expecting a cc/ua because of a bus reset that we
552 * performed, treat this just as a retry. otherwise this is
553 * information that we should pass up to the upper-level driver
554 * so that we can deal with it there.
556 if (scmd
->device
->expecting_cc_ua
) {
558 * Because some device does not queue unit
559 * attentions correctly, we carefully check
560 * additional sense code and qualifier so as
561 * not to squash media change unit attention.
563 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
564 scmd
->device
->expecting_cc_ua
= 0;
569 * we might also expect a cc/ua if another LUN on the target
570 * reported a UA with an ASC/ASCQ of 3F 0E -
571 * REPORTED LUNS DATA HAS CHANGED.
573 if (scmd
->device
->sdev_target
->expecting_lun_change
&&
574 sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
577 * if the device is in the process of becoming ready, we
580 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
583 * if the device is not started, we need to wake
584 * the error handler to start the motor
586 if (scmd
->device
->allow_restart
&&
587 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
590 * Pass the UA upwards for a determination in the completion
595 /* these are not supported */
597 if (sshdr
.asc
== 0x27 && sshdr
.ascq
== 0x07) {
598 /* Thin provisioning hard threshold reached */
599 set_host_byte(scmd
, DID_ALLOC_FAILURE
);
604 case VOLUME_OVERFLOW
:
607 set_host_byte(scmd
, DID_TARGET_FAILURE
);
611 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
612 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
613 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
614 set_host_byte(scmd
, DID_MEDIUM_ERROR
);
620 if (scmd
->device
->retry_hwerror
)
621 return ADD_TO_MLQUEUE
;
623 set_host_byte(scmd
, DID_TARGET_FAILURE
);
626 case ILLEGAL_REQUEST
:
627 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
628 sshdr
.asc
== 0x21 || /* Logical block address out of range */
629 sshdr
.asc
== 0x22 || /* Invalid function */
630 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
631 sshdr
.asc
== 0x26 || /* Parameter value invalid */
632 sshdr
.asc
== 0x27) { /* Write protected */
633 set_host_byte(scmd
, DID_TARGET_FAILURE
);
641 EXPORT_SYMBOL_GPL(scsi_check_sense
);
643 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
645 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
646 struct scsi_device
*tmp_sdev
;
648 if (!sht
->track_queue_depth
||
649 sdev
->queue_depth
>= sdev
->max_queue_depth
)
652 if (time_before(jiffies
,
653 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
656 if (time_before(jiffies
,
657 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
661 * Walk all devices of a target and do
664 shost_for_each_device(tmp_sdev
, sdev
->host
) {
665 if (tmp_sdev
->channel
!= sdev
->channel
||
666 tmp_sdev
->id
!= sdev
->id
||
667 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
670 scsi_change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1);
671 sdev
->last_queue_ramp_up
= jiffies
;
675 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
677 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
678 struct scsi_device
*tmp_sdev
;
680 if (!sht
->track_queue_depth
)
683 shost_for_each_device(tmp_sdev
, sdev
->host
) {
684 if (tmp_sdev
->channel
!= sdev
->channel
||
685 tmp_sdev
->id
!= sdev
->id
)
688 * We do not know the number of commands that were at
689 * the device when we got the queue full so we start
690 * from the highest possible value and work our way down.
692 scsi_track_queue_full(tmp_sdev
, tmp_sdev
->queue_depth
- 1);
697 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
698 * @scmd: SCSI cmd to examine.
701 * This is *only* called when we are examining the status of commands
702 * queued during error recovery. the main difference here is that we
703 * don't allow for the possibility of retries here, and we are a lot
704 * more restrictive about what we consider acceptable.
706 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
709 * first check the host byte, to see if there is anything in there
710 * that would indicate what we need to do.
712 if (host_byte(scmd
->result
) == DID_RESET
) {
714 * rats. we are already in the error handler, so we now
715 * get to try and figure out what to do next. if the sense
716 * is valid, we have a pretty good idea of what to do.
717 * if not, we mark it as FAILED.
719 return scsi_check_sense(scmd
);
721 if (host_byte(scmd
->result
) != DID_OK
)
725 * next, check the message byte.
727 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
731 * now, check the status byte to see if this indicates
734 switch (status_byte(scmd
->result
)) {
736 scsi_handle_queue_ramp_up(scmd
->device
);
738 case COMMAND_TERMINATED
:
740 case CHECK_CONDITION
:
741 return scsi_check_sense(scmd
);
743 case INTERMEDIATE_GOOD
:
744 case INTERMEDIATE_C_GOOD
:
746 * who knows? FIXME(eric)
749 case RESERVATION_CONFLICT
:
750 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
751 /* it is a success, we probed the device and
754 /* otherwise, we failed to send the command */
757 scsi_handle_queue_full(scmd
->device
);
768 * scsi_eh_done - Completion function for error handling.
769 * @scmd: Cmd that is done.
771 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
773 struct completion
*eh_action
;
775 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
776 "%s result: %x\n", __func__
, scmd
->result
));
778 eh_action
= scmd
->device
->host
->eh_action
;
784 * scsi_try_host_reset - ask host adapter to reset itself
785 * @scmd: SCSI cmd to send host reset.
787 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
791 struct Scsi_Host
*host
= scmd
->device
->host
;
792 struct scsi_host_template
*hostt
= host
->hostt
;
794 SCSI_LOG_ERROR_RECOVERY(3,
795 shost_printk(KERN_INFO
, host
, "Snd Host RST\n"));
797 if (!hostt
->eh_host_reset_handler
)
800 rtn
= hostt
->eh_host_reset_handler(scmd
);
802 if (rtn
== SUCCESS
) {
803 if (!hostt
->skip_settle_delay
)
804 ssleep(HOST_RESET_SETTLE_TIME
);
805 spin_lock_irqsave(host
->host_lock
, flags
);
806 scsi_report_bus_reset(host
, scmd_channel(scmd
));
807 spin_unlock_irqrestore(host
->host_lock
, flags
);
814 * scsi_try_bus_reset - ask host to perform a bus reset
815 * @scmd: SCSI cmd to send bus reset.
817 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
821 struct Scsi_Host
*host
= scmd
->device
->host
;
822 struct scsi_host_template
*hostt
= host
->hostt
;
824 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
825 "%s: Snd Bus RST\n", __func__
));
827 if (!hostt
->eh_bus_reset_handler
)
830 rtn
= hostt
->eh_bus_reset_handler(scmd
);
832 if (rtn
== SUCCESS
) {
833 if (!hostt
->skip_settle_delay
)
834 ssleep(BUS_RESET_SETTLE_TIME
);
835 spin_lock_irqsave(host
->host_lock
, flags
);
836 scsi_report_bus_reset(host
, scmd_channel(scmd
));
837 spin_unlock_irqrestore(host
->host_lock
, flags
);
843 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
846 sdev
->expecting_cc_ua
= 1;
850 * scsi_try_target_reset - Ask host to perform a target reset
851 * @scmd: SCSI cmd used to send a target reset
854 * There is no timeout for this operation. if this operation is
855 * unreliable for a given host, then the host itself needs to put a
856 * timer on it, and set the host back to a consistent state prior to
859 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
863 struct Scsi_Host
*host
= scmd
->device
->host
;
864 struct scsi_host_template
*hostt
= host
->hostt
;
866 if (!hostt
->eh_target_reset_handler
)
869 rtn
= hostt
->eh_target_reset_handler(scmd
);
870 if (rtn
== SUCCESS
) {
871 spin_lock_irqsave(host
->host_lock
, flags
);
872 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
873 __scsi_report_device_reset
);
874 spin_unlock_irqrestore(host
->host_lock
, flags
);
881 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
882 * @scmd: SCSI cmd used to send BDR
885 * There is no timeout for this operation. if this operation is
886 * unreliable for a given host, then the host itself needs to put a
887 * timer on it, and set the host back to a consistent state prior to
890 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
893 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
895 if (!hostt
->eh_device_reset_handler
)
898 rtn
= hostt
->eh_device_reset_handler(scmd
);
900 __scsi_report_device_reset(scmd
->device
, NULL
);
905 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
906 * @hostt: SCSI driver host template
907 * @scmd: SCSI cmd used to send a target reset
910 * SUCCESS, FAILED, or FAST_IO_FAIL
913 * SUCCESS does not necessarily indicate that the command
914 * has been aborted; it only indicates that the LLDDs
915 * has cleared all references to that command.
916 * LLDDs should return FAILED only if an abort was required
917 * but could not be executed. LLDDs should return FAST_IO_FAIL
918 * if the device is temporarily unavailable (eg due to a
919 * link down on FibreChannel)
921 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
,
922 struct scsi_cmnd
*scmd
)
924 if (!hostt
->eh_abort_handler
)
927 return hostt
->eh_abort_handler(scmd
);
930 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
932 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
933 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
934 if (scsi_try_target_reset(scmd
) != SUCCESS
)
935 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
936 scsi_try_host_reset(scmd
);
940 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
941 * @scmd: SCSI command structure to hijack
942 * @ses: structure to save restore information
943 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
944 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
945 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
947 * This function is used to save a scsi command information before re-execution
948 * as part of the error recovery process. If @sense_bytes is 0 the command
949 * sent must be one that does not transfer any data. If @sense_bytes != 0
950 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
951 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
953 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
954 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
956 struct scsi_device
*sdev
= scmd
->device
;
959 * We need saved copies of a number of fields - this is because
960 * error handling may need to overwrite these with different values
961 * to run different commands, and once error handling is complete,
962 * we will need to restore these values prior to running the actual
965 ses
->cmd_len
= scmd
->cmd_len
;
966 ses
->cmnd
= scmd
->cmnd
;
967 ses
->data_direction
= scmd
->sc_data_direction
;
968 ses
->sdb
= scmd
->sdb
;
969 ses
->result
= scmd
->result
;
970 ses
->resid_len
= scmd
->req
.resid_len
;
971 ses
->underflow
= scmd
->underflow
;
972 ses
->prot_op
= scmd
->prot_op
;
973 ses
->eh_eflags
= scmd
->eh_eflags
;
975 scmd
->prot_op
= SCSI_PROT_NORMAL
;
977 scmd
->cmnd
= ses
->eh_cmnd
;
978 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
979 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
981 scmd
->req
.resid_len
= 0;
984 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
986 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
988 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
989 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
990 scmd
->sdb
.table
.nents
= scmd
->sdb
.table
.orig_nents
= 1;
991 scmd
->cmnd
[0] = REQUEST_SENSE
;
992 scmd
->cmnd
[4] = scmd
->sdb
.length
;
993 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
995 scmd
->sc_data_direction
= DMA_NONE
;
997 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
998 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
999 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
1003 scmd
->underflow
= 0;
1005 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
1006 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
1007 (sdev
->lun
<< 5 & 0xe0);
1010 * Zero the sense buffer. The scsi spec mandates that any
1011 * untransferred sense data should be interpreted as being zero.
1013 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
1015 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
1018 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1019 * @scmd: SCSI command structure to restore
1020 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1022 * Undo any damage done by above scsi_eh_prep_cmnd().
1024 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
1027 * Restore original data
1029 scmd
->cmd_len
= ses
->cmd_len
;
1030 scmd
->cmnd
= ses
->cmnd
;
1031 scmd
->sc_data_direction
= ses
->data_direction
;
1032 scmd
->sdb
= ses
->sdb
;
1033 scmd
->result
= ses
->result
;
1034 scmd
->req
.resid_len
= ses
->resid_len
;
1035 scmd
->underflow
= ses
->underflow
;
1036 scmd
->prot_op
= ses
->prot_op
;
1037 scmd
->eh_eflags
= ses
->eh_eflags
;
1039 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
1042 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1043 * @scmd: SCSI command structure to hijack
1044 * @cmnd: CDB to send
1045 * @cmnd_size: size in bytes of @cmnd
1046 * @timeout: timeout for this request
1047 * @sense_bytes: size of sense data to copy or 0
1049 * This function is used to send a scsi command down to a target device
1050 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1053 * SUCCESS or FAILED or NEEDS_RETRY
1055 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
1056 int cmnd_size
, int timeout
, unsigned sense_bytes
)
1058 struct scsi_device
*sdev
= scmd
->device
;
1059 struct Scsi_Host
*shost
= sdev
->host
;
1060 DECLARE_COMPLETION_ONSTACK(done
);
1061 unsigned long timeleft
= timeout
, delay
;
1062 struct scsi_eh_save ses
;
1063 const unsigned long stall_for
= msecs_to_jiffies(100);
1067 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
1068 shost
->eh_action
= &done
;
1070 scsi_log_send(scmd
);
1071 scmd
->scsi_done
= scsi_eh_done
;
1074 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1075 * change the SCSI device state after we have examined it and before
1076 * .queuecommand() is called.
1078 mutex_lock(&sdev
->state_mutex
);
1079 while (sdev
->sdev_state
== SDEV_BLOCK
&& timeleft
> 0) {
1080 mutex_unlock(&sdev
->state_mutex
);
1081 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG
, sdev
,
1082 "%s: state %d <> %d\n", __func__
, sdev
->sdev_state
,
1084 delay
= min(timeleft
, stall_for
);
1086 msleep(jiffies_to_msecs(delay
));
1087 mutex_lock(&sdev
->state_mutex
);
1089 if (sdev
->sdev_state
!= SDEV_BLOCK
)
1090 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
1092 rtn
= SCSI_MLQUEUE_DEVICE_BUSY
;
1093 mutex_unlock(&sdev
->state_mutex
);
1096 if (timeleft
> stall_for
) {
1097 scsi_eh_restore_cmnd(scmd
, &ses
);
1098 timeleft
-= stall_for
;
1099 msleep(jiffies_to_msecs(stall_for
));
1102 /* signal not to enter either branch of the if () below */
1106 timeleft
= wait_for_completion_timeout(&done
, timeout
);
1110 shost
->eh_action
= NULL
;
1112 scsi_log_completion(scmd
, rtn
);
1114 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1115 "%s timeleft: %ld\n",
1116 __func__
, timeleft
));
1119 * If there is time left scsi_eh_done got called, and we will examine
1120 * the actual status codes to see whether the command actually did
1121 * complete normally, else if we have a zero return and no time left,
1122 * the command must still be pending, so abort it and return FAILED.
1123 * If we never actually managed to issue the command, because
1124 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1125 * value above (so don't execute either branch of the if)
1128 rtn
= scsi_eh_completed_normally(scmd
);
1129 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1130 "%s: scsi_eh_completed_normally %x\n", __func__
, rtn
));
1137 case ADD_TO_MLQUEUE
:
1144 } else if (rtn
!= FAILED
) {
1145 scsi_abort_eh_cmnd(scmd
);
1149 scsi_eh_restore_cmnd(scmd
, &ses
);
1155 * scsi_request_sense - Request sense data from a particular target.
1156 * @scmd: SCSI cmd for request sense.
1159 * Some hosts automatically obtain this information, others require
1160 * that we obtain it on our own. This function will *not* return until
1161 * the command either times out, or it completes.
1163 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
1165 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
1168 static int scsi_eh_action(struct scsi_cmnd
*scmd
, int rtn
)
1170 if (!blk_rq_is_passthrough(scmd
->request
)) {
1171 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
1172 if (sdrv
->eh_action
)
1173 rtn
= sdrv
->eh_action(scmd
, rtn
);
1179 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1180 * @scmd: Original SCSI cmd that eh has finished.
1181 * @done_q: Queue for processed commands.
1184 * We don't want to use the normal command completion while we are are
1185 * still handling errors - it may cause other commands to be queued,
1186 * and that would disturb what we are doing. Thus we really want to
1187 * keep a list of pending commands for final completion, and once we
1188 * are ready to leave error handling we handle completion for real.
1190 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
1192 list_move_tail(&scmd
->eh_entry
, done_q
);
1194 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
1197 * scsi_eh_get_sense - Get device sense data.
1198 * @work_q: Queue of commands to process.
1199 * @done_q: Queue of processed commands.
1202 * See if we need to request sense information. if so, then get it
1203 * now, so we have a better idea of what to do.
1206 * This has the unfortunate side effect that if a shost adapter does
1207 * not automatically request sense information, we end up shutting
1208 * it down before we request it.
1210 * All drivers should request sense information internally these days,
1211 * so for now all I have to say is tough noogies if you end up in here.
1213 * XXX: Long term this code should go away, but that needs an audit of
1216 int scsi_eh_get_sense(struct list_head
*work_q
,
1217 struct list_head
*done_q
)
1219 struct scsi_cmnd
*scmd
, *next
;
1220 struct Scsi_Host
*shost
;
1224 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1225 * should not get sense.
1227 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1228 if ((scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) ||
1229 SCSI_SENSE_VALID(scmd
))
1232 shost
= scmd
->device
->host
;
1233 if (scsi_host_eh_past_deadline(shost
)) {
1234 SCSI_LOG_ERROR_RECOVERY(3,
1235 scmd_printk(KERN_INFO
, scmd
,
1236 "%s: skip request sense, past eh deadline\n",
1240 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1242 * don't request sense if there's no check condition
1243 * status because the error we're processing isn't one
1244 * that has a sense code (and some devices get
1245 * confused by sense requests out of the blue)
1249 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
1250 "%s: requesting sense\n",
1252 rtn
= scsi_request_sense(scmd
);
1256 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1257 "sense requested, result %x\n", scmd
->result
));
1258 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd
));
1260 rtn
= scsi_decide_disposition(scmd
);
1263 * if the result was normal, then just pass it along to the
1267 /* we don't want this command reissued, just
1268 * finished with the sense data, so set
1269 * retries to the max allowed to ensure it
1270 * won't get reissued */
1271 scmd
->retries
= scmd
->allowed
;
1272 else if (rtn
!= NEEDS_RETRY
)
1275 scsi_eh_finish_cmd(scmd
, done_q
);
1278 return list_empty(work_q
);
1280 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
1283 * scsi_eh_tur - Send TUR to device.
1284 * @scmd: &scsi_cmnd to send TUR
1287 * 0 - Device is ready. 1 - Device NOT ready.
1289 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
1291 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
1292 int retry_cnt
= 1, rtn
;
1295 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
1296 scmd
->device
->eh_timeout
, 0);
1298 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1299 "%s return: %x\n", __func__
, rtn
));
1314 * scsi_eh_test_devices - check if devices are responding from error recovery.
1315 * @cmd_list: scsi commands in error recovery.
1316 * @work_q: queue for commands which still need more error recovery
1317 * @done_q: queue for commands which are finished
1318 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1321 * Tests if devices are in a working state. Commands to devices now in
1322 * a working state are sent to the done_q while commands to devices which
1323 * are still failing to respond are returned to the work_q for more
1326 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1327 struct list_head
*work_q
,
1328 struct list_head
*done_q
, int try_stu
)
1330 struct scsi_cmnd
*scmd
, *next
;
1331 struct scsi_device
*sdev
;
1334 while (!list_empty(cmd_list
)) {
1335 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1336 sdev
= scmd
->device
;
1339 if (scsi_host_eh_past_deadline(sdev
->host
)) {
1340 /* Push items back onto work_q */
1341 list_splice_init(cmd_list
, work_q
);
1342 SCSI_LOG_ERROR_RECOVERY(3,
1343 sdev_printk(KERN_INFO
, sdev
,
1344 "%s: skip test device, past eh deadline",
1350 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1351 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1352 !scsi_eh_tur(scmd
)) ||
1355 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1356 if (scmd
->device
== sdev
) {
1359 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
))
1360 scsi_eh_finish_cmd(scmd
, done_q
);
1362 list_move_tail(&scmd
->eh_entry
, work_q
);
1365 return list_empty(work_q
);
1369 * scsi_eh_try_stu - Send START_UNIT to device.
1370 * @scmd: &scsi_cmnd to send START_UNIT
1373 * 0 - Device is ready. 1 - Device NOT ready.
1375 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1377 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1379 if (scmd
->device
->allow_restart
) {
1380 int i
, rtn
= NEEDS_RETRY
;
1382 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1383 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1393 * scsi_eh_stu - send START_UNIT if needed
1394 * @shost: &scsi host being recovered.
1395 * @work_q: &list_head for pending commands.
1396 * @done_q: &list_head for processed commands.
1399 * If commands are failing due to not ready, initializing command required,
1400 * try revalidating the device, which will end up sending a start unit.
1402 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1403 struct list_head
*work_q
,
1404 struct list_head
*done_q
)
1406 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1407 struct scsi_device
*sdev
;
1409 shost_for_each_device(sdev
, shost
) {
1410 if (scsi_host_eh_past_deadline(shost
)) {
1411 SCSI_LOG_ERROR_RECOVERY(3,
1412 sdev_printk(KERN_INFO
, sdev
,
1413 "%s: skip START_UNIT, past eh deadline\n",
1418 list_for_each_entry(scmd
, work_q
, eh_entry
)
1419 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1420 scsi_check_sense(scmd
) == FAILED
) {
1428 SCSI_LOG_ERROR_RECOVERY(3,
1429 sdev_printk(KERN_INFO
, sdev
,
1430 "%s: Sending START_UNIT\n",
1433 if (!scsi_eh_try_stu(stu_scmd
)) {
1434 if (!scsi_device_online(sdev
) ||
1435 !scsi_eh_tur(stu_scmd
)) {
1436 list_for_each_entry_safe(scmd
, next
,
1438 if (scmd
->device
== sdev
&&
1439 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
)
1440 scsi_eh_finish_cmd(scmd
, done_q
);
1444 SCSI_LOG_ERROR_RECOVERY(3,
1445 sdev_printk(KERN_INFO
, sdev
,
1446 "%s: START_UNIT failed\n",
1451 return list_empty(work_q
);
1456 * scsi_eh_bus_device_reset - send bdr if needed
1457 * @shost: scsi host being recovered.
1458 * @work_q: &list_head for pending commands.
1459 * @done_q: &list_head for processed commands.
1462 * Try a bus device reset. Still, look to see whether we have multiple
1463 * devices that are jammed or not - if we have multiple devices, it
1464 * makes no sense to try bus_device_reset - we really would need to try
1465 * a bus_reset instead.
1467 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1468 struct list_head
*work_q
,
1469 struct list_head
*done_q
)
1471 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1472 struct scsi_device
*sdev
;
1475 shost_for_each_device(sdev
, shost
) {
1476 if (scsi_host_eh_past_deadline(shost
)) {
1477 SCSI_LOG_ERROR_RECOVERY(3,
1478 sdev_printk(KERN_INFO
, sdev
,
1479 "%s: skip BDR, past eh deadline\n",
1484 list_for_each_entry(scmd
, work_q
, eh_entry
)
1485 if (scmd
->device
== sdev
) {
1493 SCSI_LOG_ERROR_RECOVERY(3,
1494 sdev_printk(KERN_INFO
, sdev
,
1495 "%s: Sending BDR\n", current
->comm
));
1496 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1497 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1498 if (!scsi_device_online(sdev
) ||
1499 rtn
== FAST_IO_FAIL
||
1500 !scsi_eh_tur(bdr_scmd
)) {
1501 list_for_each_entry_safe(scmd
, next
,
1503 if (scmd
->device
== sdev
&&
1504 scsi_eh_action(scmd
, rtn
) != FAILED
)
1505 scsi_eh_finish_cmd(scmd
,
1510 SCSI_LOG_ERROR_RECOVERY(3,
1511 sdev_printk(KERN_INFO
, sdev
,
1512 "%s: BDR failed\n", current
->comm
));
1516 return list_empty(work_q
);
1520 * scsi_eh_target_reset - send target reset if needed
1521 * @shost: scsi host being recovered.
1522 * @work_q: &list_head for pending commands.
1523 * @done_q: &list_head for processed commands.
1526 * Try a target reset.
1528 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1529 struct list_head
*work_q
,
1530 struct list_head
*done_q
)
1532 LIST_HEAD(tmp_list
);
1533 LIST_HEAD(check_list
);
1535 list_splice_init(work_q
, &tmp_list
);
1537 while (!list_empty(&tmp_list
)) {
1538 struct scsi_cmnd
*next
, *scmd
;
1542 if (scsi_host_eh_past_deadline(shost
)) {
1543 /* push back on work queue for further processing */
1544 list_splice_init(&check_list
, work_q
);
1545 list_splice_init(&tmp_list
, work_q
);
1546 SCSI_LOG_ERROR_RECOVERY(3,
1547 shost_printk(KERN_INFO
, shost
,
1548 "%s: Skip target reset, past eh deadline\n",
1550 return list_empty(work_q
);
1553 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1556 SCSI_LOG_ERROR_RECOVERY(3,
1557 shost_printk(KERN_INFO
, shost
,
1558 "%s: Sending target reset to target %d\n",
1559 current
->comm
, id
));
1560 rtn
= scsi_try_target_reset(scmd
);
1561 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1562 SCSI_LOG_ERROR_RECOVERY(3,
1563 shost_printk(KERN_INFO
, shost
,
1564 "%s: Target reset failed"
1566 current
->comm
, id
));
1567 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1568 if (scmd_id(scmd
) != id
)
1572 list_move_tail(&scmd
->eh_entry
, &check_list
);
1573 else if (rtn
== FAST_IO_FAIL
)
1574 scsi_eh_finish_cmd(scmd
, done_q
);
1576 /* push back on work queue for further processing */
1577 list_move(&scmd
->eh_entry
, work_q
);
1581 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1585 * scsi_eh_bus_reset - send a bus reset
1586 * @shost: &scsi host being recovered.
1587 * @work_q: &list_head for pending commands.
1588 * @done_q: &list_head for processed commands.
1590 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1591 struct list_head
*work_q
,
1592 struct list_head
*done_q
)
1594 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1595 LIST_HEAD(check_list
);
1596 unsigned int channel
;
1600 * we really want to loop over the various channels, and do this on
1601 * a channel by channel basis. we should also check to see if any
1602 * of the failed commands are on soft_reset devices, and if so, skip
1606 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1607 if (scsi_host_eh_past_deadline(shost
)) {
1608 list_splice_init(&check_list
, work_q
);
1609 SCSI_LOG_ERROR_RECOVERY(3,
1610 shost_printk(KERN_INFO
, shost
,
1611 "%s: skip BRST, past eh deadline\n",
1613 return list_empty(work_q
);
1617 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1618 if (channel
== scmd_channel(scmd
)) {
1622 * FIXME add back in some support for
1623 * soft_reset devices.
1630 SCSI_LOG_ERROR_RECOVERY(3,
1631 shost_printk(KERN_INFO
, shost
,
1632 "%s: Sending BRST chan: %d\n",
1633 current
->comm
, channel
));
1634 rtn
= scsi_try_bus_reset(chan_scmd
);
1635 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1636 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1637 if (channel
== scmd_channel(scmd
)) {
1638 if (rtn
== FAST_IO_FAIL
)
1639 scsi_eh_finish_cmd(scmd
,
1642 list_move_tail(&scmd
->eh_entry
,
1647 SCSI_LOG_ERROR_RECOVERY(3,
1648 shost_printk(KERN_INFO
, shost
,
1649 "%s: BRST failed chan: %d\n",
1650 current
->comm
, channel
));
1653 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1657 * scsi_eh_host_reset - send a host reset
1658 * @shost: host to be reset.
1659 * @work_q: &list_head for pending commands.
1660 * @done_q: &list_head for processed commands.
1662 static int scsi_eh_host_reset(struct Scsi_Host
*shost
,
1663 struct list_head
*work_q
,
1664 struct list_head
*done_q
)
1666 struct scsi_cmnd
*scmd
, *next
;
1667 LIST_HEAD(check_list
);
1670 if (!list_empty(work_q
)) {
1671 scmd
= list_entry(work_q
->next
,
1672 struct scsi_cmnd
, eh_entry
);
1674 SCSI_LOG_ERROR_RECOVERY(3,
1675 shost_printk(KERN_INFO
, shost
,
1676 "%s: Sending HRST\n",
1679 rtn
= scsi_try_host_reset(scmd
);
1680 if (rtn
== SUCCESS
) {
1681 list_splice_init(work_q
, &check_list
);
1682 } else if (rtn
== FAST_IO_FAIL
) {
1683 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1684 scsi_eh_finish_cmd(scmd
, done_q
);
1687 SCSI_LOG_ERROR_RECOVERY(3,
1688 shost_printk(KERN_INFO
, shost
,
1689 "%s: HRST failed\n",
1693 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1697 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1698 * @work_q: &list_head for pending commands.
1699 * @done_q: &list_head for processed commands.
1701 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1702 struct list_head
*done_q
)
1704 struct scsi_cmnd
*scmd
, *next
;
1705 struct scsi_device
*sdev
;
1707 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1708 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1709 "not ready after error recovery\n");
1710 sdev
= scmd
->device
;
1712 mutex_lock(&sdev
->state_mutex
);
1713 scsi_device_set_state(sdev
, SDEV_OFFLINE
);
1714 mutex_unlock(&sdev
->state_mutex
);
1716 scsi_eh_finish_cmd(scmd
, done_q
);
1722 * scsi_noretry_cmd - determine if command should be failed fast
1723 * @scmd: SCSI cmd to examine.
1725 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1727 switch (host_byte(scmd
->result
)) {
1733 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1735 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1737 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1738 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1741 case DID_SOFT_ERROR
:
1742 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1745 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1750 * assume caller has checked sense and determined
1751 * the check condition was retryable.
1753 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1754 blk_rq_is_passthrough(scmd
->request
))
1761 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1762 * @scmd: SCSI cmd to examine.
1765 * This is *only* called when we are examining the status after sending
1766 * out the actual data command. any commands that are queued for error
1767 * recovery (e.g. test_unit_ready) do *not* come through here.
1769 * When this routine returns failed, it means the error handler thread
1770 * is woken. In cases where the error code indicates an error that
1771 * doesn't require the error handler read (i.e. we don't need to
1772 * abort/reset), this function should return SUCCESS.
1774 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1779 * if the device is offline, then we clearly just pass the result back
1780 * up to the top level.
1782 if (!scsi_device_online(scmd
->device
)) {
1783 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO
, scmd
,
1784 "%s: device offline - report as SUCCESS\n", __func__
));
1789 * first check the host byte, to see if there is anything in there
1790 * that would indicate what we need to do.
1792 switch (host_byte(scmd
->result
)) {
1793 case DID_PASSTHROUGH
:
1795 * no matter what, pass this through to the upper layer.
1796 * nuke this special code so that it looks like we are saying
1799 scmd
->result
&= 0xff00ffff;
1803 * looks good. drop through, and check the next byte.
1807 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
1808 set_host_byte(scmd
, DID_TIME_OUT
);
1812 case DID_NO_CONNECT
:
1813 case DID_BAD_TARGET
:
1815 * note - this means that we just report the status back
1816 * to the top level driver, not that we actually think
1817 * that it indicates SUCCESS.
1820 case DID_SOFT_ERROR
:
1822 * when the low level driver returns did_soft_error,
1823 * it is responsible for keeping an internal retry counter
1824 * in order to avoid endless loops (db)
1831 return ADD_TO_MLQUEUE
;
1832 case DID_TRANSPORT_DISRUPTED
:
1834 * LLD/transport was disrupted during processing of the IO.
1835 * The transport class is now blocked/blocking,
1836 * and the transport will decide what to do with the IO
1837 * based on its timers and recovery capablilities if
1838 * there are enough retries.
1841 case DID_TRANSPORT_FAILFAST
:
1843 * The transport decided to failfast the IO (most likely
1844 * the fast io fail tmo fired), so send IO directly upwards.
1848 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1849 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1851 * execute reservation conflict processing code
1861 * when we scan the bus, we get timeout messages for
1862 * these commands if there is no device available.
1863 * other hosts report did_no_connect for the same thing.
1865 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1866 scmd
->cmnd
[0] == INQUIRY
)) {
1878 * next, check the message byte.
1880 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1884 * check the status byte to see if this indicates anything special.
1886 switch (status_byte(scmd
->result
)) {
1888 scsi_handle_queue_full(scmd
->device
);
1890 * the case of trying to send too many commands to a
1891 * tagged queueing device.
1896 * device can't talk to us at the moment. Should only
1897 * occur (SAM-3) when the task queue is empty, so will cause
1898 * the empty queue handling to trigger a stall in the
1901 return ADD_TO_MLQUEUE
;
1903 if (scmd
->cmnd
[0] == REPORT_LUNS
)
1904 scmd
->device
->sdev_target
->expecting_lun_change
= 0;
1905 scsi_handle_queue_ramp_up(scmd
->device
);
1907 case COMMAND_TERMINATED
:
1911 case CHECK_CONDITION
:
1912 rtn
= scsi_check_sense(scmd
);
1913 if (rtn
== NEEDS_RETRY
)
1915 /* if rtn == FAILED, we have no sense information;
1916 * returning FAILED will wake the error handler thread
1917 * to collect the sense and redo the decide
1920 case CONDITION_GOOD
:
1921 case INTERMEDIATE_GOOD
:
1922 case INTERMEDIATE_C_GOOD
:
1925 * who knows? FIXME(eric)
1929 case RESERVATION_CONFLICT
:
1930 sdev_printk(KERN_INFO
, scmd
->device
,
1931 "reservation conflict\n");
1932 set_host_byte(scmd
, DID_NEXUS_FAILURE
);
1933 return SUCCESS
; /* causes immediate i/o error */
1941 /* we requeue for retry because the error was retryable, and
1942 * the request was not marked fast fail. Note that above,
1943 * even if the request is marked fast fail, we still requeue
1944 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1945 if ((++scmd
->retries
) <= scmd
->allowed
1946 && !scsi_noretry_cmd(scmd
)) {
1950 * no more retries - report this one back to upper level.
1956 static void eh_lock_door_done(struct request
*req
, blk_status_t status
)
1958 blk_put_request(req
);
1962 * scsi_eh_lock_door - Prevent medium removal for the specified device
1963 * @sdev: SCSI device to prevent medium removal
1966 * We must be called from process context.
1969 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1970 * head of the devices request queue, and continue.
1972 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1974 struct request
*req
;
1975 struct scsi_request
*rq
;
1977 req
= blk_get_request(sdev
->request_queue
, REQ_OP_SCSI_IN
, 0);
1982 rq
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1986 rq
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1988 rq
->cmd_len
= COMMAND_SIZE(rq
->cmd
[0]);
1990 req
->rq_flags
|= RQF_QUIET
;
1991 req
->timeout
= 10 * HZ
;
1994 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1998 * scsi_restart_operations - restart io operations to the specified host.
1999 * @shost: Host we are restarting.
2002 * When we entered the error handler, we blocked all further i/o to
2003 * this device. we need to 'reverse' this process.
2005 static void scsi_restart_operations(struct Scsi_Host
*shost
)
2007 struct scsi_device
*sdev
;
2008 unsigned long flags
;
2011 * If the door was locked, we need to insert a door lock request
2012 * onto the head of the SCSI request queue for the device. There
2013 * is no point trying to lock the door of an off-line device.
2015 shost_for_each_device(sdev
, shost
) {
2016 if (scsi_device_online(sdev
) && sdev
->was_reset
&& sdev
->locked
) {
2017 scsi_eh_lock_door(sdev
);
2018 sdev
->was_reset
= 0;
2023 * next free up anything directly waiting upon the host. this
2024 * will be requests for character device operations, and also for
2025 * ioctls to queued block devices.
2027 SCSI_LOG_ERROR_RECOVERY(3,
2028 shost_printk(KERN_INFO
, shost
, "waking up host to restart\n"));
2030 spin_lock_irqsave(shost
->host_lock
, flags
);
2031 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
2032 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
2033 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
2034 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2036 wake_up(&shost
->host_wait
);
2039 * finally we need to re-initiate requests that may be pending. we will
2040 * have had everything blocked while error handling is taking place, and
2041 * now that error recovery is done, we will need to ensure that these
2042 * requests are started.
2044 scsi_run_host_queues(shost
);
2047 * if eh is active and host_eh_scheduled is pending we need to re-run
2048 * recovery. we do this check after scsi_run_host_queues() to allow
2049 * everything pent up since the last eh run a chance to make forward
2050 * progress before we sync again. Either we'll immediately re-run
2051 * recovery or scsi_device_unbusy() will wake us again when these
2052 * pending commands complete.
2054 spin_lock_irqsave(shost
->host_lock
, flags
);
2055 if (shost
->host_eh_scheduled
)
2056 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
2057 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
2058 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2062 * scsi_eh_ready_devs - check device ready state and recover if not.
2063 * @shost: host to be recovered.
2064 * @work_q: &list_head for pending commands.
2065 * @done_q: &list_head for processed commands.
2067 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
2068 struct list_head
*work_q
,
2069 struct list_head
*done_q
)
2071 if (!scsi_eh_stu(shost
, work_q
, done_q
))
2072 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
2073 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
2074 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
2075 if (!scsi_eh_host_reset(shost
, work_q
, done_q
))
2076 scsi_eh_offline_sdevs(work_q
,
2079 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
2082 * scsi_eh_flush_done_q - finish processed commands or retry them.
2083 * @done_q: list_head of processed commands.
2085 void scsi_eh_flush_done_q(struct list_head
*done_q
)
2087 struct scsi_cmnd
*scmd
, *next
;
2089 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
2090 list_del_init(&scmd
->eh_entry
);
2091 if (scsi_device_online(scmd
->device
) &&
2092 !scsi_noretry_cmd(scmd
) &&
2093 (++scmd
->retries
<= scmd
->allowed
)) {
2094 SCSI_LOG_ERROR_RECOVERY(3,
2095 scmd_printk(KERN_INFO
, scmd
,
2096 "%s: flush retry cmd\n",
2098 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
2101 * If just we got sense for the device (called
2102 * scsi_eh_get_sense), scmd->result is already
2103 * set, do not set DRIVER_TIMEOUT.
2106 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
2107 SCSI_LOG_ERROR_RECOVERY(3,
2108 scmd_printk(KERN_INFO
, scmd
,
2109 "%s: flush finish cmd\n",
2111 scsi_finish_command(scmd
);
2115 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
2118 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2119 * @shost: Host to unjam.
2122 * When we come in here, we *know* that all commands on the bus have
2123 * either completed, failed or timed out. we also know that no further
2124 * commands are being sent to the host, so things are relatively quiet
2125 * and we have freedom to fiddle with things as we wish.
2127 * This is only the *default* implementation. it is possible for
2128 * individual drivers to supply their own version of this function, and
2129 * if the maintainer wishes to do this, it is strongly suggested that
2130 * this function be taken as a template and modified. this function
2131 * was designed to correctly handle problems for about 95% of the
2132 * different cases out there, and it should always provide at least a
2133 * reasonable amount of error recovery.
2135 * Any command marked 'failed' or 'timeout' must eventually have
2136 * scsi_finish_cmd() called for it. we do all of the retry stuff
2137 * here, so when we restart the host after we return it should have an
2140 static void scsi_unjam_host(struct Scsi_Host
*shost
)
2142 unsigned long flags
;
2143 LIST_HEAD(eh_work_q
);
2144 LIST_HEAD(eh_done_q
);
2146 spin_lock_irqsave(shost
->host_lock
, flags
);
2147 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
2148 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2150 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
2152 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
2153 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
2155 spin_lock_irqsave(shost
->host_lock
, flags
);
2156 if (shost
->eh_deadline
!= -1)
2157 shost
->last_reset
= 0;
2158 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2159 scsi_eh_flush_done_q(&eh_done_q
);
2163 * scsi_error_handler - SCSI error handler thread
2164 * @data: Host for which we are running.
2167 * This is the main error handling loop. This is run as a kernel thread
2168 * for every SCSI host and handles all error handling activity.
2170 int scsi_error_handler(void *data
)
2172 struct Scsi_Host
*shost
= data
;
2175 * We use TASK_INTERRUPTIBLE so that the thread is not
2176 * counted against the load average as a running process.
2177 * We never actually get interrupted because kthread_run
2178 * disables signal delivery for the created thread.
2182 * The sequence in kthread_stop() sets the stop flag first
2183 * then wakes the process. To avoid missed wakeups, the task
2184 * should always be in a non running state before the stop
2187 set_current_state(TASK_INTERRUPTIBLE
);
2188 if (kthread_should_stop())
2191 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
2192 shost
->host_failed
!= scsi_host_busy(shost
)) {
2193 SCSI_LOG_ERROR_RECOVERY(1,
2194 shost_printk(KERN_INFO
, shost
,
2195 "scsi_eh_%d: sleeping\n",
2201 __set_current_state(TASK_RUNNING
);
2202 SCSI_LOG_ERROR_RECOVERY(1,
2203 shost_printk(KERN_INFO
, shost
,
2204 "scsi_eh_%d: waking up %d/%d/%d\n",
2205 shost
->host_no
, shost
->host_eh_scheduled
,
2207 scsi_host_busy(shost
)));
2210 * We have a host that is failing for some reason. Figure out
2211 * what we need to do to get it up and online again (if we can).
2212 * If we fail, we end up taking the thing offline.
2214 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
2215 SCSI_LOG_ERROR_RECOVERY(1,
2216 shost_printk(KERN_ERR
, shost
,
2217 "scsi_eh_%d: unable to autoresume\n",
2222 if (shost
->transportt
->eh_strategy_handler
)
2223 shost
->transportt
->eh_strategy_handler(shost
);
2225 scsi_unjam_host(shost
);
2227 /* All scmds have been handled */
2228 shost
->host_failed
= 0;
2231 * Note - if the above fails completely, the action is to take
2232 * individual devices offline and flush the queue of any
2233 * outstanding requests that may have been pending. When we
2234 * restart, we restart any I/O to any other devices on the bus
2235 * which are still online.
2237 scsi_restart_operations(shost
);
2238 if (!shost
->eh_noresume
)
2239 scsi_autopm_put_host(shost
);
2241 __set_current_state(TASK_RUNNING
);
2243 SCSI_LOG_ERROR_RECOVERY(1,
2244 shost_printk(KERN_INFO
, shost
,
2245 "Error handler scsi_eh_%d exiting\n",
2247 shost
->ehandler
= NULL
;
2252 * Function: scsi_report_bus_reset()
2254 * Purpose: Utility function used by low-level drivers to report that
2255 * they have observed a bus reset on the bus being handled.
2257 * Arguments: shost - Host in question
2258 * channel - channel on which reset was observed.
2262 * Lock status: Host lock must be held.
2264 * Notes: This only needs to be called if the reset is one which
2265 * originates from an unknown location. Resets originated
2266 * by the mid-level itself don't need to call this, but there
2267 * should be no harm.
2269 * The main purpose of this is to make sure that a CHECK_CONDITION
2270 * is properly treated.
2272 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
2274 struct scsi_device
*sdev
;
2276 __shost_for_each_device(sdev
, shost
) {
2277 if (channel
== sdev_channel(sdev
))
2278 __scsi_report_device_reset(sdev
, NULL
);
2281 EXPORT_SYMBOL(scsi_report_bus_reset
);
2284 * Function: scsi_report_device_reset()
2286 * Purpose: Utility function used by low-level drivers to report that
2287 * they have observed a device reset on the device being handled.
2289 * Arguments: shost - Host in question
2290 * channel - channel on which reset was observed
2291 * target - target on which reset was observed
2295 * Lock status: Host lock must be held
2297 * Notes: This only needs to be called if the reset is one which
2298 * originates from an unknown location. Resets originated
2299 * by the mid-level itself don't need to call this, but there
2300 * should be no harm.
2302 * The main purpose of this is to make sure that a CHECK_CONDITION
2303 * is properly treated.
2305 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
2307 struct scsi_device
*sdev
;
2309 __shost_for_each_device(sdev
, shost
) {
2310 if (channel
== sdev_channel(sdev
) &&
2311 target
== sdev_id(sdev
))
2312 __scsi_report_device_reset(sdev
, NULL
);
2315 EXPORT_SYMBOL(scsi_report_device_reset
);
2318 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
2323 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2324 * @dev: scsi_device to operate on
2325 * @arg: reset type (see sg.h)
2328 scsi_ioctl_reset(struct scsi_device
*dev
, int __user
*arg
)
2330 struct scsi_cmnd
*scmd
;
2331 struct Scsi_Host
*shost
= dev
->host
;
2333 unsigned long flags
;
2334 int error
= 0, rtn
, val
;
2336 if (!capable(CAP_SYS_ADMIN
) || !capable(CAP_SYS_RAWIO
))
2339 error
= get_user(val
, arg
);
2343 if (scsi_autopm_get_host(shost
) < 0)
2347 rq
= kzalloc(sizeof(struct request
) + sizeof(struct scsi_cmnd
) +
2348 shost
->hostt
->cmd_size
, GFP_KERNEL
);
2350 goto out_put_autopm_host
;
2351 blk_rq_init(NULL
, rq
);
2353 scmd
= (struct scsi_cmnd
*)(rq
+ 1);
2354 scsi_init_command(dev
, scmd
);
2356 scmd
->cmnd
= scsi_req(rq
)->cmd
;
2358 scmd
->scsi_done
= scsi_reset_provider_done_command
;
2359 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2363 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2365 spin_lock_irqsave(shost
->host_lock
, flags
);
2366 shost
->tmf_in_progress
= 1;
2367 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2369 switch (val
& ~SG_SCSI_RESET_NO_ESCALATE
) {
2370 case SG_SCSI_RESET_NOTHING
:
2373 case SG_SCSI_RESET_DEVICE
:
2374 rtn
= scsi_try_bus_device_reset(scmd
);
2375 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2378 case SG_SCSI_RESET_TARGET
:
2379 rtn
= scsi_try_target_reset(scmd
);
2380 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2383 case SG_SCSI_RESET_BUS
:
2384 rtn
= scsi_try_bus_reset(scmd
);
2385 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2388 case SG_SCSI_RESET_HOST
:
2389 rtn
= scsi_try_host_reset(scmd
);
2398 error
= (rtn
== SUCCESS
) ? 0 : -EIO
;
2400 spin_lock_irqsave(shost
->host_lock
, flags
);
2401 shost
->tmf_in_progress
= 0;
2402 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2405 * be sure to wake up anyone who was sleeping or had their queue
2406 * suspended while we performed the TMF.
2408 SCSI_LOG_ERROR_RECOVERY(3,
2409 shost_printk(KERN_INFO
, shost
,
2410 "waking up host to restart after TMF\n"));
2412 wake_up(&shost
->host_wait
);
2413 scsi_run_host_queues(shost
);
2415 scsi_put_command(scmd
);
2418 out_put_autopm_host
:
2419 scsi_autopm_put_host(shost
);
2423 bool scsi_command_normalize_sense(const struct scsi_cmnd
*cmd
,
2424 struct scsi_sense_hdr
*sshdr
)
2426 return scsi_normalize_sense(cmd
->sense_buffer
,
2427 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2429 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2432 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2433 * @sense_buffer: byte array of sense data
2434 * @sb_len: number of valid bytes in sense_buffer
2435 * @info_out: pointer to 64 integer where 8 or 4 byte information
2436 * field will be placed if found.
2439 * true if information field found, false if not found.
2441 bool scsi_get_sense_info_fld(const u8
*sense_buffer
, int sb_len
,
2448 switch (sense_buffer
[0] & 0x7f) {
2451 if (sense_buffer
[0] & 0x80) {
2452 *info_out
= get_unaligned_be32(&sense_buffer
[3]);
2458 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2460 if (ucp
&& (0xa == ucp
[1])) {
2461 *info_out
= get_unaligned_be64(&ucp
[4]);
2469 EXPORT_SYMBOL(scsi_get_sense_info_fld
);