2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
28 #include <linux/jiffies.h>
29 #include <asm/unaligned.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_transport.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_ioctl.h>
42 #include "scsi_priv.h"
43 #include "scsi_logging.h"
44 #include "scsi_transport_api.h"
46 #include <trace/events/scsi.h>
48 static void scsi_eh_done(struct scsi_cmnd
*scmd
);
51 * These should *probably* be handled by the host itself.
52 * Since it is allowed to sleep, it probably should.
54 #define BUS_RESET_SETTLE_TIME (10)
55 #define HOST_RESET_SETTLE_TIME (10)
57 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
58 static int scsi_try_to_abort_cmd(struct scsi_host_template
*,
61 /* called with shost->host_lock held */
62 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
64 if (atomic_read(&shost
->host_busy
) == shost
->host_failed
) {
65 trace_scsi_eh_wakeup(shost
);
66 wake_up_process(shost
->ehandler
);
67 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO
, shost
,
68 "Waking error handler thread\n"));
73 * scsi_schedule_eh - schedule EH for SCSI host
74 * @shost: SCSI host to invoke error handling on.
76 * Schedule SCSI EH without scmd.
78 void scsi_schedule_eh(struct Scsi_Host
*shost
)
82 spin_lock_irqsave(shost
->host_lock
, flags
);
84 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
85 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
86 shost
->host_eh_scheduled
++;
87 scsi_eh_wakeup(shost
);
90 spin_unlock_irqrestore(shost
->host_lock
, flags
);
92 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
94 static int scsi_host_eh_past_deadline(struct Scsi_Host
*shost
)
96 if (!shost
->last_reset
|| shost
->eh_deadline
== -1)
100 * 32bit accesses are guaranteed to be atomic
101 * (on all supported architectures), so instead
102 * of using a spinlock we can as well double check
103 * if eh_deadline has been set to 'off' during the
106 if (time_before(jiffies
, shost
->last_reset
+ shost
->eh_deadline
) &&
107 shost
->eh_deadline
> -1)
114 * scmd_eh_abort_handler - Handle command aborts
115 * @work: command to be aborted.
118 scmd_eh_abort_handler(struct work_struct
*work
)
120 struct scsi_cmnd
*scmd
=
121 container_of(work
, struct scsi_cmnd
, abort_work
.work
);
122 struct scsi_device
*sdev
= scmd
->device
;
125 if (scsi_host_eh_past_deadline(sdev
->host
)) {
126 SCSI_LOG_ERROR_RECOVERY(3,
127 scmd_printk(KERN_INFO
, scmd
,
128 "eh timeout, not aborting\n"));
130 SCSI_LOG_ERROR_RECOVERY(3,
131 scmd_printk(KERN_INFO
, scmd
,
132 "aborting command\n"));
133 rtn
= scsi_try_to_abort_cmd(sdev
->host
->hostt
, scmd
);
134 if (rtn
== SUCCESS
) {
135 set_host_byte(scmd
, DID_TIME_OUT
);
136 if (scsi_host_eh_past_deadline(sdev
->host
)) {
137 SCSI_LOG_ERROR_RECOVERY(3,
138 scmd_printk(KERN_INFO
, scmd
,
139 "eh timeout, not retrying "
140 "aborted command\n"));
141 } else if (!scsi_noretry_cmd(scmd
) &&
142 (++scmd
->retries
<= scmd
->allowed
)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING
, scmd
,
145 "retry aborted command\n"));
146 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
149 SCSI_LOG_ERROR_RECOVERY(3,
150 scmd_printk(KERN_WARNING
, scmd
,
151 "finish aborted command\n"));
152 scsi_finish_command(scmd
);
156 SCSI_LOG_ERROR_RECOVERY(3,
157 scmd_printk(KERN_INFO
, scmd
,
159 (rtn
== FAST_IO_FAIL
) ?
160 "not send" : "failed"));
164 if (!scsi_eh_scmd_add(scmd
, 0)) {
165 SCSI_LOG_ERROR_RECOVERY(3,
166 scmd_printk(KERN_WARNING
, scmd
,
167 "terminate aborted command\n"));
168 set_host_byte(scmd
, DID_TIME_OUT
);
169 scsi_finish_command(scmd
);
174 * scsi_abort_command - schedule a command abort
175 * @scmd: scmd to abort.
177 * We only need to abort commands after a command timeout
180 scsi_abort_command(struct scsi_cmnd
*scmd
)
182 struct scsi_device
*sdev
= scmd
->device
;
183 struct Scsi_Host
*shost
= sdev
->host
;
186 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
188 * Retry after abort failed, escalate to next level.
190 scmd
->eh_eflags
&= ~SCSI_EH_ABORT_SCHEDULED
;
191 SCSI_LOG_ERROR_RECOVERY(3,
192 scmd_printk(KERN_INFO
, scmd
,
193 "previous abort failed\n"));
194 BUG_ON(delayed_work_pending(&scmd
->abort_work
));
199 * Do not try a command abort if
200 * SCSI EH has already started.
202 spin_lock_irqsave(shost
->host_lock
, flags
);
203 if (scsi_host_in_recovery(shost
)) {
204 spin_unlock_irqrestore(shost
->host_lock
, flags
);
205 SCSI_LOG_ERROR_RECOVERY(3,
206 scmd_printk(KERN_INFO
, scmd
,
207 "not aborting, host in recovery\n"));
211 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
212 shost
->last_reset
= jiffies
;
213 spin_unlock_irqrestore(shost
->host_lock
, flags
);
215 scmd
->eh_eflags
|= SCSI_EH_ABORT_SCHEDULED
;
216 SCSI_LOG_ERROR_RECOVERY(3,
217 scmd_printk(KERN_INFO
, scmd
, "abort scheduled\n"));
218 queue_delayed_work(shost
->tmf_work_q
, &scmd
->abort_work
, HZ
/ 100);
223 * scsi_eh_scmd_add - add scsi cmd to error handling.
224 * @scmd: scmd to run eh on.
225 * @eh_flag: optional SCSI_EH flag.
230 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
232 struct Scsi_Host
*shost
= scmd
->device
->host
;
236 if (!shost
->ehandler
)
239 spin_lock_irqsave(shost
->host_lock
, flags
);
240 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
241 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
244 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
245 shost
->last_reset
= jiffies
;
248 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
)
249 eh_flag
&= ~SCSI_EH_CANCEL_CMD
;
250 scmd
->eh_eflags
|= eh_flag
;
251 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
252 shost
->host_failed
++;
253 scsi_eh_wakeup(shost
);
255 spin_unlock_irqrestore(shost
->host_lock
, flags
);
260 * scsi_times_out - Timeout function for normal scsi commands.
261 * @req: request that is timing out.
264 * We do not need to lock this. There is the potential for a race
265 * only in that the normal completion handling might run, but if the
266 * normal completion function determines that the timer has already
267 * fired, then it mustn't do anything.
269 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
271 struct scsi_cmnd
*scmd
= req
->special
;
272 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
273 struct Scsi_Host
*host
= scmd
->device
->host
;
275 trace_scsi_dispatch_cmd_timeout(scmd
);
276 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
278 if (host
->eh_deadline
!= -1 && !host
->last_reset
)
279 host
->last_reset
= jiffies
;
281 if (host
->transportt
->eh_timed_out
)
282 rtn
= host
->transportt
->eh_timed_out(scmd
);
283 else if (host
->hostt
->eh_timed_out
)
284 rtn
= host
->hostt
->eh_timed_out(scmd
);
286 if (rtn
== BLK_EH_NOT_HANDLED
) {
287 if (!host
->hostt
->no_async_abort
&&
288 scsi_abort_command(scmd
) == SUCCESS
)
289 return BLK_EH_NOT_HANDLED
;
291 set_host_byte(scmd
, DID_TIME_OUT
);
292 if (!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))
293 rtn
= BLK_EH_HANDLED
;
300 * scsi_block_when_processing_errors - Prevent cmds from being queued.
301 * @sdev: Device on which we are performing recovery.
304 * We block until the host is out of error recovery, and then check to
305 * see whether the host or the device is offline.
308 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
310 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
314 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
316 online
= scsi_device_online(sdev
);
318 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO
, sdev
,
319 "%s: rtn: %d\n", __func__
, online
));
323 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
325 #ifdef CONFIG_SCSI_LOGGING
327 * scsi_eh_prt_fail_stats - Log info on failures.
328 * @shost: scsi host being recovered.
329 * @work_q: Queue of scsi cmds to process.
331 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
332 struct list_head
*work_q
)
334 struct scsi_cmnd
*scmd
;
335 struct scsi_device
*sdev
;
336 int total_failures
= 0;
339 int devices_failed
= 0;
341 shost_for_each_device(sdev
, shost
) {
342 list_for_each_entry(scmd
, work_q
, eh_entry
) {
343 if (scmd
->device
== sdev
) {
345 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
352 if (cmd_cancel
|| cmd_failed
) {
353 SCSI_LOG_ERROR_RECOVERY(3,
354 shost_printk(KERN_INFO
, shost
,
355 "%s: cmds failed: %d, cancel: %d\n",
356 __func__
, cmd_failed
,
364 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO
, shost
,
365 "Total of %d commands on %d"
366 " devices require eh work\n",
367 total_failures
, devices_failed
));
372 * scsi_report_lun_change - Set flag on all *other* devices on the same target
373 * to indicate that a UNIT ATTENTION is expected.
374 * @sdev: Device reporting the UNIT ATTENTION
376 static void scsi_report_lun_change(struct scsi_device
*sdev
)
378 sdev
->sdev_target
->expecting_lun_change
= 1;
382 * scsi_report_sense - Examine scsi sense information and log messages for
383 * certain conditions, also issue uevents for some of them.
384 * @sdev: Device reporting the sense code
385 * @sshdr: sshdr to be examined
387 static void scsi_report_sense(struct scsi_device
*sdev
,
388 struct scsi_sense_hdr
*sshdr
)
390 enum scsi_device_event evt_type
= SDEV_EVT_MAXBITS
; /* i.e. none */
392 if (sshdr
->sense_key
== UNIT_ATTENTION
) {
393 if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x03) {
394 evt_type
= SDEV_EVT_INQUIRY_CHANGE_REPORTED
;
395 sdev_printk(KERN_WARNING
, sdev
,
396 "Inquiry data has changed");
397 } else if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x0e) {
398 evt_type
= SDEV_EVT_LUN_CHANGE_REPORTED
;
399 scsi_report_lun_change(sdev
);
400 sdev_printk(KERN_WARNING
, sdev
,
401 "Warning! Received an indication that the "
402 "LUN assignments on this target have "
403 "changed. The Linux SCSI layer does not "
404 "automatically remap LUN assignments.\n");
405 } else if (sshdr
->asc
== 0x3f)
406 sdev_printk(KERN_WARNING
, sdev
,
407 "Warning! Received an indication that the "
408 "operating parameters on this target have "
409 "changed. The Linux SCSI layer does not "
410 "automatically adjust these parameters.\n");
412 if (sshdr
->asc
== 0x38 && sshdr
->ascq
== 0x07) {
413 evt_type
= SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
;
414 sdev_printk(KERN_WARNING
, sdev
,
415 "Warning! Received an indication that the "
416 "LUN reached a thin provisioning soft "
420 if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x01) {
421 evt_type
= SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
;
422 sdev_printk(KERN_WARNING
, sdev
,
423 "Mode parameters changed");
424 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x09) {
425 evt_type
= SDEV_EVT_CAPACITY_CHANGE_REPORTED
;
426 sdev_printk(KERN_WARNING
, sdev
,
427 "Capacity data has changed");
428 } else if (sshdr
->asc
== 0x2a)
429 sdev_printk(KERN_WARNING
, sdev
,
430 "Parameters changed");
433 if (evt_type
!= SDEV_EVT_MAXBITS
) {
434 set_bit(evt_type
, sdev
->pending_events
);
435 schedule_work(&sdev
->event_work
);
440 * scsi_check_sense - Examine scsi cmd sense
441 * @scmd: Cmd to have sense checked.
444 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
447 * When a deferred error is detected the current command has
448 * not been executed and needs retrying.
450 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
452 struct scsi_device
*sdev
= scmd
->device
;
453 struct scsi_sense_hdr sshdr
;
455 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
456 return FAILED
; /* no valid sense data */
458 scsi_report_sense(sdev
, &sshdr
);
460 if (scsi_sense_is_deferred(&sshdr
))
463 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
464 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
467 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
468 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
470 /* handler does not care. Drop down to default handling */
473 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&& scmd
->scsi_done
!= scsi_eh_done
)
475 * nasty: for mid-layer issued TURs, we need to return the
476 * actual sense data without any recovery attempt. For eh
477 * issued ones, we need to try to recover and interpret
482 * Previous logic looked for FILEMARK, EOM or ILI which are
483 * mainly associated with tapes and returned SUCCESS.
485 if (sshdr
.response_code
== 0x70) {
487 if (scmd
->sense_buffer
[2] & 0xe0)
491 * descriptor format: look for "stream commands sense data
492 * descriptor" (see SSC-3). Assume single sense data
493 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
495 if ((sshdr
.additional_length
> 3) &&
496 (scmd
->sense_buffer
[8] == 0x4) &&
497 (scmd
->sense_buffer
[11] & 0xe0))
501 switch (sshdr
.sense_key
) {
504 case RECOVERED_ERROR
:
505 return /* soft_error */ SUCCESS
;
507 case ABORTED_COMMAND
:
508 if (sshdr
.asc
== 0x10) /* DIF */
515 * if we are expecting a cc/ua because of a bus reset that we
516 * performed, treat this just as a retry. otherwise this is
517 * information that we should pass up to the upper-level driver
518 * so that we can deal with it there.
520 if (scmd
->device
->expecting_cc_ua
) {
522 * Because some device does not queue unit
523 * attentions correctly, we carefully check
524 * additional sense code and qualifier so as
525 * not to squash media change unit attention.
527 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
528 scmd
->device
->expecting_cc_ua
= 0;
533 * we might also expect a cc/ua if another LUN on the target
534 * reported a UA with an ASC/ASCQ of 3F 0E -
535 * REPORTED LUNS DATA HAS CHANGED.
537 if (scmd
->device
->sdev_target
->expecting_lun_change
&&
538 sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
541 * if the device is in the process of becoming ready, we
544 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
547 * if the device is not started, we need to wake
548 * the error handler to start the motor
550 if (scmd
->device
->allow_restart
&&
551 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
554 * Pass the UA upwards for a determination in the completion
559 /* these are not supported */
561 if (sshdr
.asc
== 0x27 && sshdr
.ascq
== 0x07) {
562 /* Thin provisioning hard threshold reached */
563 set_host_byte(scmd
, DID_ALLOC_FAILURE
);
567 case VOLUME_OVERFLOW
:
570 set_host_byte(scmd
, DID_TARGET_FAILURE
);
574 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
575 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
576 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
577 set_host_byte(scmd
, DID_MEDIUM_ERROR
);
583 if (scmd
->device
->retry_hwerror
)
584 return ADD_TO_MLQUEUE
;
586 set_host_byte(scmd
, DID_TARGET_FAILURE
);
588 case ILLEGAL_REQUEST
:
589 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
590 sshdr
.asc
== 0x21 || /* Logical block address out of range */
591 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
592 sshdr
.asc
== 0x26) { /* Parameter value invalid */
593 set_host_byte(scmd
, DID_TARGET_FAILURE
);
602 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
604 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
605 struct scsi_device
*tmp_sdev
;
607 if (!sht
->track_queue_depth
||
608 sdev
->queue_depth
>= sdev
->max_queue_depth
)
611 if (time_before(jiffies
,
612 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
615 if (time_before(jiffies
,
616 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
620 * Walk all devices of a target and do
623 shost_for_each_device(tmp_sdev
, sdev
->host
) {
624 if (tmp_sdev
->channel
!= sdev
->channel
||
625 tmp_sdev
->id
!= sdev
->id
||
626 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
629 scsi_change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1);
630 sdev
->last_queue_ramp_up
= jiffies
;
634 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
636 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
637 struct scsi_device
*tmp_sdev
;
639 if (!sht
->track_queue_depth
)
642 shost_for_each_device(tmp_sdev
, sdev
->host
) {
643 if (tmp_sdev
->channel
!= sdev
->channel
||
644 tmp_sdev
->id
!= sdev
->id
)
647 * We do not know the number of commands that were at
648 * the device when we got the queue full so we start
649 * from the highest possible value and work our way down.
651 scsi_track_queue_full(tmp_sdev
, tmp_sdev
->queue_depth
- 1);
656 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
657 * @scmd: SCSI cmd to examine.
660 * This is *only* called when we are examining the status of commands
661 * queued during error recovery. the main difference here is that we
662 * don't allow for the possibility of retries here, and we are a lot
663 * more restrictive about what we consider acceptable.
665 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
668 * first check the host byte, to see if there is anything in there
669 * that would indicate what we need to do.
671 if (host_byte(scmd
->result
) == DID_RESET
) {
673 * rats. we are already in the error handler, so we now
674 * get to try and figure out what to do next. if the sense
675 * is valid, we have a pretty good idea of what to do.
676 * if not, we mark it as FAILED.
678 return scsi_check_sense(scmd
);
680 if (host_byte(scmd
->result
) != DID_OK
)
684 * next, check the message byte.
686 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
690 * now, check the status byte to see if this indicates
693 switch (status_byte(scmd
->result
)) {
695 scsi_handle_queue_ramp_up(scmd
->device
);
696 case COMMAND_TERMINATED
:
698 case CHECK_CONDITION
:
699 return scsi_check_sense(scmd
);
701 case INTERMEDIATE_GOOD
:
702 case INTERMEDIATE_C_GOOD
:
704 * who knows? FIXME(eric)
707 case RESERVATION_CONFLICT
:
708 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
709 /* it is a success, we probed the device and
712 /* otherwise, we failed to send the command */
715 scsi_handle_queue_full(scmd
->device
);
726 * scsi_eh_done - Completion function for error handling.
727 * @scmd: Cmd that is done.
729 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
731 struct completion
*eh_action
;
733 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
734 "%s result: %x\n", __func__
, scmd
->result
));
736 eh_action
= scmd
->device
->host
->eh_action
;
742 * scsi_try_host_reset - ask host adapter to reset itself
743 * @scmd: SCSI cmd to send host reset.
745 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
749 struct Scsi_Host
*host
= scmd
->device
->host
;
750 struct scsi_host_template
*hostt
= host
->hostt
;
752 SCSI_LOG_ERROR_RECOVERY(3,
753 shost_printk(KERN_INFO
, host
, "Snd Host RST\n"));
755 if (!hostt
->eh_host_reset_handler
)
758 rtn
= hostt
->eh_host_reset_handler(scmd
);
760 if (rtn
== SUCCESS
) {
761 if (!hostt
->skip_settle_delay
)
762 ssleep(HOST_RESET_SETTLE_TIME
);
763 spin_lock_irqsave(host
->host_lock
, flags
);
764 scsi_report_bus_reset(host
, scmd_channel(scmd
));
765 spin_unlock_irqrestore(host
->host_lock
, flags
);
772 * scsi_try_bus_reset - ask host to perform a bus reset
773 * @scmd: SCSI cmd to send bus reset.
775 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
779 struct Scsi_Host
*host
= scmd
->device
->host
;
780 struct scsi_host_template
*hostt
= host
->hostt
;
782 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
783 "%s: Snd Bus RST\n", __func__
));
785 if (!hostt
->eh_bus_reset_handler
)
788 rtn
= hostt
->eh_bus_reset_handler(scmd
);
790 if (rtn
== SUCCESS
) {
791 if (!hostt
->skip_settle_delay
)
792 ssleep(BUS_RESET_SETTLE_TIME
);
793 spin_lock_irqsave(host
->host_lock
, flags
);
794 scsi_report_bus_reset(host
, scmd_channel(scmd
));
795 spin_unlock_irqrestore(host
->host_lock
, flags
);
801 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
804 sdev
->expecting_cc_ua
= 1;
808 * scsi_try_target_reset - Ask host to perform a target reset
809 * @scmd: SCSI cmd used to send a target reset
812 * There is no timeout for this operation. if this operation is
813 * unreliable for a given host, then the host itself needs to put a
814 * timer on it, and set the host back to a consistent state prior to
817 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
821 struct Scsi_Host
*host
= scmd
->device
->host
;
822 struct scsi_host_template
*hostt
= host
->hostt
;
824 if (!hostt
->eh_target_reset_handler
)
827 rtn
= hostt
->eh_target_reset_handler(scmd
);
828 if (rtn
== SUCCESS
) {
829 spin_lock_irqsave(host
->host_lock
, flags
);
830 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
831 __scsi_report_device_reset
);
832 spin_unlock_irqrestore(host
->host_lock
, flags
);
839 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
840 * @scmd: SCSI cmd used to send BDR
843 * There is no timeout for this operation. if this operation is
844 * unreliable for a given host, then the host itself needs to put a
845 * timer on it, and set the host back to a consistent state prior to
848 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
851 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
853 if (!hostt
->eh_device_reset_handler
)
856 rtn
= hostt
->eh_device_reset_handler(scmd
);
858 __scsi_report_device_reset(scmd
->device
, NULL
);
863 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
864 * @hostt: SCSI driver host template
865 * @scmd: SCSI cmd used to send a target reset
868 * SUCCESS, FAILED, or FAST_IO_FAIL
871 * SUCCESS does not necessarily indicate that the command
872 * has been aborted; it only indicates that the LLDDs
873 * has cleared all references to that command.
874 * LLDDs should return FAILED only if an abort was required
875 * but could not be executed. LLDDs should return FAST_IO_FAIL
876 * if the device is temporarily unavailable (eg due to a
877 * link down on FibreChannel)
879 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
,
880 struct scsi_cmnd
*scmd
)
882 if (!hostt
->eh_abort_handler
)
885 return hostt
->eh_abort_handler(scmd
);
888 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
890 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
891 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
892 if (scsi_try_target_reset(scmd
) != SUCCESS
)
893 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
894 scsi_try_host_reset(scmd
);
898 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
899 * @scmd: SCSI command structure to hijack
900 * @ses: structure to save restore information
901 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
902 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
903 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
905 * This function is used to save a scsi command information before re-execution
906 * as part of the error recovery process. If @sense_bytes is 0 the command
907 * sent must be one that does not transfer any data. If @sense_bytes != 0
908 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
909 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
911 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
912 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
914 struct scsi_device
*sdev
= scmd
->device
;
917 * We need saved copies of a number of fields - this is because
918 * error handling may need to overwrite these with different values
919 * to run different commands, and once error handling is complete,
920 * we will need to restore these values prior to running the actual
923 ses
->cmd_len
= scmd
->cmd_len
;
924 ses
->cmnd
= scmd
->cmnd
;
925 ses
->data_direction
= scmd
->sc_data_direction
;
926 ses
->sdb
= scmd
->sdb
;
927 ses
->next_rq
= scmd
->request
->next_rq
;
928 ses
->result
= scmd
->result
;
929 ses
->underflow
= scmd
->underflow
;
930 ses
->prot_op
= scmd
->prot_op
;
932 scmd
->prot_op
= SCSI_PROT_NORMAL
;
934 scmd
->cmnd
= ses
->eh_cmnd
;
935 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
936 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
937 scmd
->request
->next_rq
= NULL
;
941 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
943 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
945 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
946 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
947 scmd
->sdb
.table
.nents
= 1;
948 scmd
->cmnd
[0] = REQUEST_SENSE
;
949 scmd
->cmnd
[4] = scmd
->sdb
.length
;
950 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
952 scmd
->sc_data_direction
= DMA_NONE
;
954 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
955 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
956 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
962 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
963 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
964 (sdev
->lun
<< 5 & 0xe0);
967 * Zero the sense buffer. The scsi spec mandates that any
968 * untransferred sense data should be interpreted as being zero.
970 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
972 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
975 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
976 * @scmd: SCSI command structure to restore
977 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
979 * Undo any damage done by above scsi_eh_prep_cmnd().
981 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
984 * Restore original data
986 scmd
->cmd_len
= ses
->cmd_len
;
987 scmd
->cmnd
= ses
->cmnd
;
988 scmd
->sc_data_direction
= ses
->data_direction
;
989 scmd
->sdb
= ses
->sdb
;
990 scmd
->request
->next_rq
= ses
->next_rq
;
991 scmd
->result
= ses
->result
;
992 scmd
->underflow
= ses
->underflow
;
993 scmd
->prot_op
= ses
->prot_op
;
995 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
998 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
999 * @scmd: SCSI command structure to hijack
1000 * @cmnd: CDB to send
1001 * @cmnd_size: size in bytes of @cmnd
1002 * @timeout: timeout for this request
1003 * @sense_bytes: size of sense data to copy or 0
1005 * This function is used to send a scsi command down to a target device
1006 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1009 * SUCCESS or FAILED or NEEDS_RETRY
1011 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
1012 int cmnd_size
, int timeout
, unsigned sense_bytes
)
1014 struct scsi_device
*sdev
= scmd
->device
;
1015 struct Scsi_Host
*shost
= sdev
->host
;
1016 DECLARE_COMPLETION_ONSTACK(done
);
1017 unsigned long timeleft
= timeout
;
1018 struct scsi_eh_save ses
;
1019 const unsigned long stall_for
= msecs_to_jiffies(100);
1023 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
1024 shost
->eh_action
= &done
;
1026 scsi_log_send(scmd
);
1027 scmd
->scsi_done
= scsi_eh_done
;
1028 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
1030 if (timeleft
> stall_for
) {
1031 scsi_eh_restore_cmnd(scmd
, &ses
);
1032 timeleft
-= stall_for
;
1033 msleep(jiffies_to_msecs(stall_for
));
1036 /* signal not to enter either branch of the if () below */
1040 timeleft
= wait_for_completion_timeout(&done
, timeout
);
1044 shost
->eh_action
= NULL
;
1046 scsi_log_completion(scmd
, rtn
);
1048 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1049 "%s timeleft: %ld\n",
1050 __func__
, timeleft
));
1053 * If there is time left scsi_eh_done got called, and we will examine
1054 * the actual status codes to see whether the command actually did
1055 * complete normally, else if we have a zero return and no time left,
1056 * the command must still be pending, so abort it and return FAILED.
1057 * If we never actually managed to issue the command, because
1058 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1059 * value above (so don't execute either branch of the if)
1062 rtn
= scsi_eh_completed_normally(scmd
);
1063 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1064 "%s: scsi_eh_completed_normally %x\n", __func__
, rtn
));
1071 case ADD_TO_MLQUEUE
:
1078 } else if (rtn
!= FAILED
) {
1079 scsi_abort_eh_cmnd(scmd
);
1083 scsi_eh_restore_cmnd(scmd
, &ses
);
1089 * scsi_request_sense - Request sense data from a particular target.
1090 * @scmd: SCSI cmd for request sense.
1093 * Some hosts automatically obtain this information, others require
1094 * that we obtain it on our own. This function will *not* return until
1095 * the command either times out, or it completes.
1097 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
1099 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
1102 static int scsi_eh_action(struct scsi_cmnd
*scmd
, int rtn
)
1104 if (scmd
->request
->cmd_type
!= REQ_TYPE_BLOCK_PC
) {
1105 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
1106 if (sdrv
->eh_action
)
1107 rtn
= sdrv
->eh_action(scmd
, rtn
);
1113 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1114 * @scmd: Original SCSI cmd that eh has finished.
1115 * @done_q: Queue for processed commands.
1118 * We don't want to use the normal command completion while we are are
1119 * still handling errors - it may cause other commands to be queued,
1120 * and that would disturb what we are doing. Thus we really want to
1121 * keep a list of pending commands for final completion, and once we
1122 * are ready to leave error handling we handle completion for real.
1124 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
1126 scmd
->device
->host
->host_failed
--;
1127 scmd
->eh_eflags
= 0;
1128 list_move_tail(&scmd
->eh_entry
, done_q
);
1130 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
1133 * scsi_eh_get_sense - Get device sense data.
1134 * @work_q: Queue of commands to process.
1135 * @done_q: Queue of processed commands.
1138 * See if we need to request sense information. if so, then get it
1139 * now, so we have a better idea of what to do.
1142 * This has the unfortunate side effect that if a shost adapter does
1143 * not automatically request sense information, we end up shutting
1144 * it down before we request it.
1146 * All drivers should request sense information internally these days,
1147 * so for now all I have to say is tough noogies if you end up in here.
1149 * XXX: Long term this code should go away, but that needs an audit of
1152 int scsi_eh_get_sense(struct list_head
*work_q
,
1153 struct list_head
*done_q
)
1155 struct scsi_cmnd
*scmd
, *next
;
1156 struct Scsi_Host
*shost
;
1159 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1160 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
1161 SCSI_SENSE_VALID(scmd
))
1164 shost
= scmd
->device
->host
;
1165 if (scsi_host_eh_past_deadline(shost
)) {
1166 SCSI_LOG_ERROR_RECOVERY(3,
1167 scmd_printk(KERN_INFO
, scmd
,
1168 "%s: skip request sense, past eh deadline\n",
1172 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1174 * don't request sense if there's no check condition
1175 * status because the error we're processing isn't one
1176 * that has a sense code (and some devices get
1177 * confused by sense requests out of the blue)
1181 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
1182 "%s: requesting sense\n",
1184 rtn
= scsi_request_sense(scmd
);
1188 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1189 "sense requested, result %x\n", scmd
->result
));
1190 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd
));
1192 rtn
= scsi_decide_disposition(scmd
);
1195 * if the result was normal, then just pass it along to the
1199 /* we don't want this command reissued, just
1200 * finished with the sense data, so set
1201 * retries to the max allowed to ensure it
1202 * won't get reissued */
1203 scmd
->retries
= scmd
->allowed
;
1204 else if (rtn
!= NEEDS_RETRY
)
1207 scsi_eh_finish_cmd(scmd
, done_q
);
1210 return list_empty(work_q
);
1212 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
1215 * scsi_eh_tur - Send TUR to device.
1216 * @scmd: &scsi_cmnd to send TUR
1219 * 0 - Device is ready. 1 - Device NOT ready.
1221 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
1223 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
1224 int retry_cnt
= 1, rtn
;
1227 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
1228 scmd
->device
->eh_timeout
, 0);
1230 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1231 "%s return: %x\n", __func__
, rtn
));
1246 * scsi_eh_test_devices - check if devices are responding from error recovery.
1247 * @cmd_list: scsi commands in error recovery.
1248 * @work_q: queue for commands which still need more error recovery
1249 * @done_q: queue for commands which are finished
1250 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1253 * Tests if devices are in a working state. Commands to devices now in
1254 * a working state are sent to the done_q while commands to devices which
1255 * are still failing to respond are returned to the work_q for more
1258 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1259 struct list_head
*work_q
,
1260 struct list_head
*done_q
, int try_stu
)
1262 struct scsi_cmnd
*scmd
, *next
;
1263 struct scsi_device
*sdev
;
1266 while (!list_empty(cmd_list
)) {
1267 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1268 sdev
= scmd
->device
;
1271 if (scsi_host_eh_past_deadline(sdev
->host
)) {
1272 /* Push items back onto work_q */
1273 list_splice_init(cmd_list
, work_q
);
1274 SCSI_LOG_ERROR_RECOVERY(3,
1275 sdev_printk(KERN_INFO
, sdev
,
1276 "%s: skip test device, past eh deadline",
1282 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1283 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1284 !scsi_eh_tur(scmd
)) ||
1287 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1288 if (scmd
->device
== sdev
) {
1291 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
))
1292 scsi_eh_finish_cmd(scmd
, done_q
);
1294 list_move_tail(&scmd
->eh_entry
, work_q
);
1297 return list_empty(work_q
);
1302 * scsi_eh_abort_cmds - abort pending commands.
1303 * @work_q: &list_head for pending commands.
1304 * @done_q: &list_head for processed commands.
1307 * Try and see whether or not it makes sense to try and abort the
1308 * running command. This only works out to be the case if we have one
1309 * command that has timed out. If the command simply failed, it makes
1310 * no sense to try and abort the command, since as far as the shost
1311 * adapter is concerned, it isn't running.
1313 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
1314 struct list_head
*done_q
)
1316 struct scsi_cmnd
*scmd
, *next
;
1317 LIST_HEAD(check_list
);
1319 struct Scsi_Host
*shost
;
1321 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1322 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
1324 shost
= scmd
->device
->host
;
1325 if (scsi_host_eh_past_deadline(shost
)) {
1326 list_splice_init(&check_list
, work_q
);
1327 SCSI_LOG_ERROR_RECOVERY(3,
1328 scmd_printk(KERN_INFO
, scmd
,
1329 "%s: skip aborting cmd, past eh deadline\n",
1331 return list_empty(work_q
);
1333 SCSI_LOG_ERROR_RECOVERY(3,
1334 scmd_printk(KERN_INFO
, scmd
,
1335 "%s: aborting cmd\n", current
->comm
));
1336 rtn
= scsi_try_to_abort_cmd(shost
->hostt
, scmd
);
1337 if (rtn
== FAILED
) {
1338 SCSI_LOG_ERROR_RECOVERY(3,
1339 scmd_printk(KERN_INFO
, scmd
,
1340 "%s: aborting cmd failed\n",
1342 list_splice_init(&check_list
, work_q
);
1343 return list_empty(work_q
);
1345 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
1346 if (rtn
== FAST_IO_FAIL
)
1347 scsi_eh_finish_cmd(scmd
, done_q
);
1349 list_move_tail(&scmd
->eh_entry
, &check_list
);
1352 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1356 * scsi_eh_try_stu - Send START_UNIT to device.
1357 * @scmd: &scsi_cmnd to send START_UNIT
1360 * 0 - Device is ready. 1 - Device NOT ready.
1362 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1364 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1366 if (scmd
->device
->allow_restart
) {
1367 int i
, rtn
= NEEDS_RETRY
;
1369 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1370 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1380 * scsi_eh_stu - send START_UNIT if needed
1381 * @shost: &scsi host being recovered.
1382 * @work_q: &list_head for pending commands.
1383 * @done_q: &list_head for processed commands.
1386 * If commands are failing due to not ready, initializing command required,
1387 * try revalidating the device, which will end up sending a start unit.
1389 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1390 struct list_head
*work_q
,
1391 struct list_head
*done_q
)
1393 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1394 struct scsi_device
*sdev
;
1396 shost_for_each_device(sdev
, shost
) {
1397 if (scsi_host_eh_past_deadline(shost
)) {
1398 SCSI_LOG_ERROR_RECOVERY(3,
1399 sdev_printk(KERN_INFO
, sdev
,
1400 "%s: skip START_UNIT, past eh deadline\n",
1405 list_for_each_entry(scmd
, work_q
, eh_entry
)
1406 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1407 scsi_check_sense(scmd
) == FAILED
) {
1415 SCSI_LOG_ERROR_RECOVERY(3,
1416 sdev_printk(KERN_INFO
, sdev
,
1417 "%s: Sending START_UNIT\n",
1420 if (!scsi_eh_try_stu(stu_scmd
)) {
1421 if (!scsi_device_online(sdev
) ||
1422 !scsi_eh_tur(stu_scmd
)) {
1423 list_for_each_entry_safe(scmd
, next
,
1425 if (scmd
->device
== sdev
&&
1426 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
)
1427 scsi_eh_finish_cmd(scmd
, done_q
);
1431 SCSI_LOG_ERROR_RECOVERY(3,
1432 sdev_printk(KERN_INFO
, sdev
,
1433 "%s: START_UNIT failed\n",
1438 return list_empty(work_q
);
1443 * scsi_eh_bus_device_reset - send bdr if needed
1444 * @shost: scsi host being recovered.
1445 * @work_q: &list_head for pending commands.
1446 * @done_q: &list_head for processed commands.
1449 * Try a bus device reset. Still, look to see whether we have multiple
1450 * devices that are jammed or not - if we have multiple devices, it
1451 * makes no sense to try bus_device_reset - we really would need to try
1452 * a bus_reset instead.
1454 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1455 struct list_head
*work_q
,
1456 struct list_head
*done_q
)
1458 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1459 struct scsi_device
*sdev
;
1462 shost_for_each_device(sdev
, shost
) {
1463 if (scsi_host_eh_past_deadline(shost
)) {
1464 SCSI_LOG_ERROR_RECOVERY(3,
1465 sdev_printk(KERN_INFO
, sdev
,
1466 "%s: skip BDR, past eh deadline\n",
1471 list_for_each_entry(scmd
, work_q
, eh_entry
)
1472 if (scmd
->device
== sdev
) {
1480 SCSI_LOG_ERROR_RECOVERY(3,
1481 sdev_printk(KERN_INFO
, sdev
,
1482 "%s: Sending BDR\n", current
->comm
));
1483 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1484 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1485 if (!scsi_device_online(sdev
) ||
1486 rtn
== FAST_IO_FAIL
||
1487 !scsi_eh_tur(bdr_scmd
)) {
1488 list_for_each_entry_safe(scmd
, next
,
1490 if (scmd
->device
== sdev
&&
1491 scsi_eh_action(scmd
, rtn
) != FAILED
)
1492 scsi_eh_finish_cmd(scmd
,
1497 SCSI_LOG_ERROR_RECOVERY(3,
1498 sdev_printk(KERN_INFO
, sdev
,
1499 "%s: BDR failed\n", current
->comm
));
1503 return list_empty(work_q
);
1507 * scsi_eh_target_reset - send target reset if needed
1508 * @shost: scsi host being recovered.
1509 * @work_q: &list_head for pending commands.
1510 * @done_q: &list_head for processed commands.
1513 * Try a target reset.
1515 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1516 struct list_head
*work_q
,
1517 struct list_head
*done_q
)
1519 LIST_HEAD(tmp_list
);
1520 LIST_HEAD(check_list
);
1522 list_splice_init(work_q
, &tmp_list
);
1524 while (!list_empty(&tmp_list
)) {
1525 struct scsi_cmnd
*next
, *scmd
;
1529 if (scsi_host_eh_past_deadline(shost
)) {
1530 /* push back on work queue for further processing */
1531 list_splice_init(&check_list
, work_q
);
1532 list_splice_init(&tmp_list
, work_q
);
1533 SCSI_LOG_ERROR_RECOVERY(3,
1534 shost_printk(KERN_INFO
, shost
,
1535 "%s: Skip target reset, past eh deadline\n",
1537 return list_empty(work_q
);
1540 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1543 SCSI_LOG_ERROR_RECOVERY(3,
1544 shost_printk(KERN_INFO
, shost
,
1545 "%s: Sending target reset to target %d\n",
1546 current
->comm
, id
));
1547 rtn
= scsi_try_target_reset(scmd
);
1548 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1549 SCSI_LOG_ERROR_RECOVERY(3,
1550 shost_printk(KERN_INFO
, shost
,
1551 "%s: Target reset failed"
1553 current
->comm
, id
));
1554 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1555 if (scmd_id(scmd
) != id
)
1559 list_move_tail(&scmd
->eh_entry
, &check_list
);
1560 else if (rtn
== FAST_IO_FAIL
)
1561 scsi_eh_finish_cmd(scmd
, done_q
);
1563 /* push back on work queue for further processing */
1564 list_move(&scmd
->eh_entry
, work_q
);
1568 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1572 * scsi_eh_bus_reset - send a bus reset
1573 * @shost: &scsi host being recovered.
1574 * @work_q: &list_head for pending commands.
1575 * @done_q: &list_head for processed commands.
1577 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1578 struct list_head
*work_q
,
1579 struct list_head
*done_q
)
1581 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1582 LIST_HEAD(check_list
);
1583 unsigned int channel
;
1587 * we really want to loop over the various channels, and do this on
1588 * a channel by channel basis. we should also check to see if any
1589 * of the failed commands are on soft_reset devices, and if so, skip
1593 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1594 if (scsi_host_eh_past_deadline(shost
)) {
1595 list_splice_init(&check_list
, work_q
);
1596 SCSI_LOG_ERROR_RECOVERY(3,
1597 shost_printk(KERN_INFO
, shost
,
1598 "%s: skip BRST, past eh deadline\n",
1600 return list_empty(work_q
);
1604 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1605 if (channel
== scmd_channel(scmd
)) {
1609 * FIXME add back in some support for
1610 * soft_reset devices.
1617 SCSI_LOG_ERROR_RECOVERY(3,
1618 shost_printk(KERN_INFO
, shost
,
1619 "%s: Sending BRST chan: %d\n",
1620 current
->comm
, channel
));
1621 rtn
= scsi_try_bus_reset(chan_scmd
);
1622 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1623 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1624 if (channel
== scmd_channel(scmd
)) {
1625 if (rtn
== FAST_IO_FAIL
)
1626 scsi_eh_finish_cmd(scmd
,
1629 list_move_tail(&scmd
->eh_entry
,
1634 SCSI_LOG_ERROR_RECOVERY(3,
1635 shost_printk(KERN_INFO
, shost
,
1636 "%s: BRST failed chan: %d\n",
1637 current
->comm
, channel
));
1640 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1644 * scsi_eh_host_reset - send a host reset
1645 * @shost: host to be reset.
1646 * @work_q: &list_head for pending commands.
1647 * @done_q: &list_head for processed commands.
1649 static int scsi_eh_host_reset(struct Scsi_Host
*shost
,
1650 struct list_head
*work_q
,
1651 struct list_head
*done_q
)
1653 struct scsi_cmnd
*scmd
, *next
;
1654 LIST_HEAD(check_list
);
1657 if (!list_empty(work_q
)) {
1658 scmd
= list_entry(work_q
->next
,
1659 struct scsi_cmnd
, eh_entry
);
1661 SCSI_LOG_ERROR_RECOVERY(3,
1662 shost_printk(KERN_INFO
, shost
,
1663 "%s: Sending HRST\n",
1666 rtn
= scsi_try_host_reset(scmd
);
1667 if (rtn
== SUCCESS
) {
1668 list_splice_init(work_q
, &check_list
);
1669 } else if (rtn
== FAST_IO_FAIL
) {
1670 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1671 scsi_eh_finish_cmd(scmd
, done_q
);
1674 SCSI_LOG_ERROR_RECOVERY(3,
1675 shost_printk(KERN_INFO
, shost
,
1676 "%s: HRST failed\n",
1680 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1684 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1685 * @work_q: &list_head for pending commands.
1686 * @done_q: &list_head for processed commands.
1688 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1689 struct list_head
*done_q
)
1691 struct scsi_cmnd
*scmd
, *next
;
1693 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1694 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1695 "not ready after error recovery\n");
1696 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1697 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1699 * FIXME: Handle lost cmds.
1702 scsi_eh_finish_cmd(scmd
, done_q
);
1708 * scsi_noretry_cmd - determine if command should be failed fast
1709 * @scmd: SCSI cmd to examine.
1711 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1713 switch (host_byte(scmd
->result
)) {
1719 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1721 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1723 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1724 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1727 case DID_SOFT_ERROR
:
1728 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1731 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1736 * assume caller has checked sense and determined
1737 * the check condition was retryable.
1739 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1740 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1747 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1748 * @scmd: SCSI cmd to examine.
1751 * This is *only* called when we are examining the status after sending
1752 * out the actual data command. any commands that are queued for error
1753 * recovery (e.g. test_unit_ready) do *not* come through here.
1755 * When this routine returns failed, it means the error handler thread
1756 * is woken. In cases where the error code indicates an error that
1757 * doesn't require the error handler read (i.e. we don't need to
1758 * abort/reset), this function should return SUCCESS.
1760 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1765 * if the device is offline, then we clearly just pass the result back
1766 * up to the top level.
1768 if (!scsi_device_online(scmd
->device
)) {
1769 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO
, scmd
,
1770 "%s: device offline - report as SUCCESS\n", __func__
));
1775 * first check the host byte, to see if there is anything in there
1776 * that would indicate what we need to do.
1778 switch (host_byte(scmd
->result
)) {
1779 case DID_PASSTHROUGH
:
1781 * no matter what, pass this through to the upper layer.
1782 * nuke this special code so that it looks like we are saying
1785 scmd
->result
&= 0xff00ffff;
1789 * looks good. drop through, and check the next byte.
1793 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
1794 set_host_byte(scmd
, DID_TIME_OUT
);
1797 case DID_NO_CONNECT
:
1798 case DID_BAD_TARGET
:
1800 * note - this means that we just report the status back
1801 * to the top level driver, not that we actually think
1802 * that it indicates SUCCESS.
1806 * when the low level driver returns did_soft_error,
1807 * it is responsible for keeping an internal retry counter
1808 * in order to avoid endless loops (db)
1810 * actually this is a bug in this function here. we should
1811 * be mindful of the maximum number of retries specified
1812 * and not get stuck in a loop.
1814 case DID_SOFT_ERROR
:
1820 return ADD_TO_MLQUEUE
;
1821 case DID_TRANSPORT_DISRUPTED
:
1823 * LLD/transport was disrupted during processing of the IO.
1824 * The transport class is now blocked/blocking,
1825 * and the transport will decide what to do with the IO
1826 * based on its timers and recovery capablilities if
1827 * there are enough retries.
1830 case DID_TRANSPORT_FAILFAST
:
1832 * The transport decided to failfast the IO (most likely
1833 * the fast io fail tmo fired), so send IO directly upwards.
1837 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1838 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1840 * execute reservation conflict processing code
1850 * when we scan the bus, we get timeout messages for
1851 * these commands if there is no device available.
1852 * other hosts report did_no_connect for the same thing.
1854 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1855 scmd
->cmnd
[0] == INQUIRY
)) {
1867 * next, check the message byte.
1869 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1873 * check the status byte to see if this indicates anything special.
1875 switch (status_byte(scmd
->result
)) {
1877 scsi_handle_queue_full(scmd
->device
);
1879 * the case of trying to send too many commands to a
1880 * tagged queueing device.
1884 * device can't talk to us at the moment. Should only
1885 * occur (SAM-3) when the task queue is empty, so will cause
1886 * the empty queue handling to trigger a stall in the
1889 return ADD_TO_MLQUEUE
;
1891 if (scmd
->cmnd
[0] == REPORT_LUNS
)
1892 scmd
->device
->sdev_target
->expecting_lun_change
= 0;
1893 scsi_handle_queue_ramp_up(scmd
->device
);
1894 case COMMAND_TERMINATED
:
1898 case CHECK_CONDITION
:
1899 rtn
= scsi_check_sense(scmd
);
1900 if (rtn
== NEEDS_RETRY
)
1902 /* if rtn == FAILED, we have no sense information;
1903 * returning FAILED will wake the error handler thread
1904 * to collect the sense and redo the decide
1907 case CONDITION_GOOD
:
1908 case INTERMEDIATE_GOOD
:
1909 case INTERMEDIATE_C_GOOD
:
1912 * who knows? FIXME(eric)
1916 case RESERVATION_CONFLICT
:
1917 sdev_printk(KERN_INFO
, scmd
->device
,
1918 "reservation conflict\n");
1919 set_host_byte(scmd
, DID_NEXUS_FAILURE
);
1920 return SUCCESS
; /* causes immediate i/o error */
1928 /* we requeue for retry because the error was retryable, and
1929 * the request was not marked fast fail. Note that above,
1930 * even if the request is marked fast fail, we still requeue
1931 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1932 if ((++scmd
->retries
) <= scmd
->allowed
1933 && !scsi_noretry_cmd(scmd
)) {
1937 * no more retries - report this one back to upper level.
1943 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1945 __blk_put_request(req
->q
, req
);
1949 * scsi_eh_lock_door - Prevent medium removal for the specified device
1950 * @sdev: SCSI device to prevent medium removal
1953 * We must be called from process context.
1956 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1957 * head of the devices request queue, and continue.
1959 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1961 struct request
*req
;
1964 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1965 * request becomes available
1967 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1971 blk_rq_set_block_pc(req
);
1973 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1977 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1980 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1982 req
->cmd_flags
|= REQ_QUIET
;
1983 req
->timeout
= 10 * HZ
;
1986 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1990 * scsi_restart_operations - restart io operations to the specified host.
1991 * @shost: Host we are restarting.
1994 * When we entered the error handler, we blocked all further i/o to
1995 * this device. we need to 'reverse' this process.
1997 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1999 struct scsi_device
*sdev
;
2000 unsigned long flags
;
2003 * If the door was locked, we need to insert a door lock request
2004 * onto the head of the SCSI request queue for the device. There
2005 * is no point trying to lock the door of an off-line device.
2007 shost_for_each_device(sdev
, shost
) {
2008 if (scsi_device_online(sdev
) && sdev
->was_reset
&& sdev
->locked
) {
2009 scsi_eh_lock_door(sdev
);
2010 sdev
->was_reset
= 0;
2015 * next free up anything directly waiting upon the host. this
2016 * will be requests for character device operations, and also for
2017 * ioctls to queued block devices.
2019 SCSI_LOG_ERROR_RECOVERY(3,
2020 shost_printk(KERN_INFO
, shost
, "waking up host to restart\n"));
2022 spin_lock_irqsave(shost
->host_lock
, flags
);
2023 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
2024 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
2025 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
2026 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2028 wake_up(&shost
->host_wait
);
2031 * finally we need to re-initiate requests that may be pending. we will
2032 * have had everything blocked while error handling is taking place, and
2033 * now that error recovery is done, we will need to ensure that these
2034 * requests are started.
2036 scsi_run_host_queues(shost
);
2039 * if eh is active and host_eh_scheduled is pending we need to re-run
2040 * recovery. we do this check after scsi_run_host_queues() to allow
2041 * everything pent up since the last eh run a chance to make forward
2042 * progress before we sync again. Either we'll immediately re-run
2043 * recovery or scsi_device_unbusy() will wake us again when these
2044 * pending commands complete.
2046 spin_lock_irqsave(shost
->host_lock
, flags
);
2047 if (shost
->host_eh_scheduled
)
2048 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
2049 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
2050 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2054 * scsi_eh_ready_devs - check device ready state and recover if not.
2055 * @shost: host to be recovered.
2056 * @work_q: &list_head for pending commands.
2057 * @done_q: &list_head for processed commands.
2059 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
2060 struct list_head
*work_q
,
2061 struct list_head
*done_q
)
2063 if (!scsi_eh_stu(shost
, work_q
, done_q
))
2064 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
2065 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
2066 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
2067 if (!scsi_eh_host_reset(shost
, work_q
, done_q
))
2068 scsi_eh_offline_sdevs(work_q
,
2071 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
2074 * scsi_eh_flush_done_q - finish processed commands or retry them.
2075 * @done_q: list_head of processed commands.
2077 void scsi_eh_flush_done_q(struct list_head
*done_q
)
2079 struct scsi_cmnd
*scmd
, *next
;
2081 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
2082 list_del_init(&scmd
->eh_entry
);
2083 if (scsi_device_online(scmd
->device
) &&
2084 !scsi_noretry_cmd(scmd
) &&
2085 (++scmd
->retries
<= scmd
->allowed
)) {
2086 SCSI_LOG_ERROR_RECOVERY(3,
2087 scmd_printk(KERN_INFO
, scmd
,
2088 "%s: flush retry cmd\n",
2090 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
2093 * If just we got sense for the device (called
2094 * scsi_eh_get_sense), scmd->result is already
2095 * set, do not set DRIVER_TIMEOUT.
2098 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
2099 SCSI_LOG_ERROR_RECOVERY(3,
2100 scmd_printk(KERN_INFO
, scmd
,
2101 "%s: flush finish cmd\n",
2103 scsi_finish_command(scmd
);
2107 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
2110 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2111 * @shost: Host to unjam.
2114 * When we come in here, we *know* that all commands on the bus have
2115 * either completed, failed or timed out. we also know that no further
2116 * commands are being sent to the host, so things are relatively quiet
2117 * and we have freedom to fiddle with things as we wish.
2119 * This is only the *default* implementation. it is possible for
2120 * individual drivers to supply their own version of this function, and
2121 * if the maintainer wishes to do this, it is strongly suggested that
2122 * this function be taken as a template and modified. this function
2123 * was designed to correctly handle problems for about 95% of the
2124 * different cases out there, and it should always provide at least a
2125 * reasonable amount of error recovery.
2127 * Any command marked 'failed' or 'timeout' must eventually have
2128 * scsi_finish_cmd() called for it. we do all of the retry stuff
2129 * here, so when we restart the host after we return it should have an
2132 static void scsi_unjam_host(struct Scsi_Host
*shost
)
2134 unsigned long flags
;
2135 LIST_HEAD(eh_work_q
);
2136 LIST_HEAD(eh_done_q
);
2138 spin_lock_irqsave(shost
->host_lock
, flags
);
2139 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
2140 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2142 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
2144 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
2145 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
2146 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
2148 spin_lock_irqsave(shost
->host_lock
, flags
);
2149 if (shost
->eh_deadline
!= -1)
2150 shost
->last_reset
= 0;
2151 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2152 scsi_eh_flush_done_q(&eh_done_q
);
2156 * scsi_error_handler - SCSI error handler thread
2157 * @data: Host for which we are running.
2160 * This is the main error handling loop. This is run as a kernel thread
2161 * for every SCSI host and handles all error handling activity.
2163 int scsi_error_handler(void *data
)
2165 struct Scsi_Host
*shost
= data
;
2168 * We use TASK_INTERRUPTIBLE so that the thread is not
2169 * counted against the load average as a running process.
2170 * We never actually get interrupted because kthread_run
2171 * disables signal delivery for the created thread.
2173 while (!kthread_should_stop()) {
2174 set_current_state(TASK_INTERRUPTIBLE
);
2175 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
2176 shost
->host_failed
!= atomic_read(&shost
->host_busy
)) {
2177 SCSI_LOG_ERROR_RECOVERY(1,
2178 shost_printk(KERN_INFO
, shost
,
2179 "scsi_eh_%d: sleeping\n",
2185 __set_current_state(TASK_RUNNING
);
2186 SCSI_LOG_ERROR_RECOVERY(1,
2187 shost_printk(KERN_INFO
, shost
,
2188 "scsi_eh_%d: waking up %d/%d/%d\n",
2189 shost
->host_no
, shost
->host_eh_scheduled
,
2191 atomic_read(&shost
->host_busy
)));
2194 * We have a host that is failing for some reason. Figure out
2195 * what we need to do to get it up and online again (if we can).
2196 * If we fail, we end up taking the thing offline.
2198 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
2199 SCSI_LOG_ERROR_RECOVERY(1,
2200 shost_printk(KERN_ERR
, shost
,
2201 "scsi_eh_%d: unable to autoresume\n",
2206 if (shost
->transportt
->eh_strategy_handler
)
2207 shost
->transportt
->eh_strategy_handler(shost
);
2209 scsi_unjam_host(shost
);
2212 * Note - if the above fails completely, the action is to take
2213 * individual devices offline and flush the queue of any
2214 * outstanding requests that may have been pending. When we
2215 * restart, we restart any I/O to any other devices on the bus
2216 * which are still online.
2218 scsi_restart_operations(shost
);
2219 if (!shost
->eh_noresume
)
2220 scsi_autopm_put_host(shost
);
2222 __set_current_state(TASK_RUNNING
);
2224 SCSI_LOG_ERROR_RECOVERY(1,
2225 shost_printk(KERN_INFO
, shost
,
2226 "Error handler scsi_eh_%d exiting\n",
2228 shost
->ehandler
= NULL
;
2233 * Function: scsi_report_bus_reset()
2235 * Purpose: Utility function used by low-level drivers to report that
2236 * they have observed a bus reset on the bus being handled.
2238 * Arguments: shost - Host in question
2239 * channel - channel on which reset was observed.
2243 * Lock status: Host lock must be held.
2245 * Notes: This only needs to be called if the reset is one which
2246 * originates from an unknown location. Resets originated
2247 * by the mid-level itself don't need to call this, but there
2248 * should be no harm.
2250 * The main purpose of this is to make sure that a CHECK_CONDITION
2251 * is properly treated.
2253 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
2255 struct scsi_device
*sdev
;
2257 __shost_for_each_device(sdev
, shost
) {
2258 if (channel
== sdev_channel(sdev
))
2259 __scsi_report_device_reset(sdev
, NULL
);
2262 EXPORT_SYMBOL(scsi_report_bus_reset
);
2265 * Function: scsi_report_device_reset()
2267 * Purpose: Utility function used by low-level drivers to report that
2268 * they have observed a device reset on the device being handled.
2270 * Arguments: shost - Host in question
2271 * channel - channel on which reset was observed
2272 * target - target on which reset was observed
2276 * Lock status: Host lock must be held
2278 * Notes: This only needs to be called if the reset is one which
2279 * originates from an unknown location. Resets originated
2280 * by the mid-level itself don't need to call this, but there
2281 * should be no harm.
2283 * The main purpose of this is to make sure that a CHECK_CONDITION
2284 * is properly treated.
2286 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
2288 struct scsi_device
*sdev
;
2290 __shost_for_each_device(sdev
, shost
) {
2291 if (channel
== sdev_channel(sdev
) &&
2292 target
== sdev_id(sdev
))
2293 __scsi_report_device_reset(sdev
, NULL
);
2296 EXPORT_SYMBOL(scsi_report_device_reset
);
2299 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
2304 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2305 * @dev: scsi_device to operate on
2306 * @arg: reset type (see sg.h)
2309 scsi_ioctl_reset(struct scsi_device
*dev
, int __user
*arg
)
2311 struct scsi_cmnd
*scmd
;
2312 struct Scsi_Host
*shost
= dev
->host
;
2314 unsigned long flags
;
2315 int error
= 0, rtn
, val
;
2317 if (!capable(CAP_SYS_ADMIN
) || !capable(CAP_SYS_RAWIO
))
2320 error
= get_user(val
, arg
);
2324 if (scsi_autopm_get_host(shost
) < 0)
2328 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
2330 goto out_put_autopm_host
;
2332 blk_rq_init(NULL
, &req
);
2333 scmd
->request
= &req
;
2335 scmd
->cmnd
= req
.cmd
;
2337 scmd
->scsi_done
= scsi_reset_provider_done_command
;
2338 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2342 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2344 spin_lock_irqsave(shost
->host_lock
, flags
);
2345 shost
->tmf_in_progress
= 1;
2346 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2348 switch (val
& ~SG_SCSI_RESET_NO_ESCALATE
) {
2349 case SG_SCSI_RESET_NOTHING
:
2352 case SG_SCSI_RESET_DEVICE
:
2353 rtn
= scsi_try_bus_device_reset(scmd
);
2354 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2357 case SG_SCSI_RESET_TARGET
:
2358 rtn
= scsi_try_target_reset(scmd
);
2359 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2362 case SG_SCSI_RESET_BUS
:
2363 rtn
= scsi_try_bus_reset(scmd
);
2364 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2367 case SG_SCSI_RESET_HOST
:
2368 rtn
= scsi_try_host_reset(scmd
);
2377 error
= (rtn
== SUCCESS
) ? 0 : -EIO
;
2379 spin_lock_irqsave(shost
->host_lock
, flags
);
2380 shost
->tmf_in_progress
= 0;
2381 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2384 * be sure to wake up anyone who was sleeping or had their queue
2385 * suspended while we performed the TMF.
2387 SCSI_LOG_ERROR_RECOVERY(3,
2388 shost_printk(KERN_INFO
, shost
,
2389 "waking up host to restart after TMF\n"));
2391 wake_up(&shost
->host_wait
);
2392 scsi_run_host_queues(shost
);
2394 scsi_put_command(scmd
);
2396 out_put_autopm_host
:
2397 scsi_autopm_put_host(shost
);
2400 EXPORT_SYMBOL(scsi_ioctl_reset
);
2403 * scsi_normalize_sense - normalize main elements from either fixed or
2404 * descriptor sense data format into a common format.
2406 * @sense_buffer: byte array containing sense data returned by device
2407 * @sb_len: number of valid bytes in sense_buffer
2408 * @sshdr: pointer to instance of structure that common
2409 * elements are written to.
2412 * The "main elements" from sense data are: response_code, sense_key,
2413 * asc, ascq and additional_length (only for descriptor format).
2415 * Typically this function can be called after a device has
2416 * responded to a SCSI command with the CHECK_CONDITION status.
2419 * true if valid sense data information found, else false;
2421 bool scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
2422 struct scsi_sense_hdr
*sshdr
)
2424 if (!sense_buffer
|| !sb_len
)
2427 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
2429 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
2431 if (!scsi_sense_valid(sshdr
))
2434 if (sshdr
->response_code
>= 0x72) {
2439 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
2441 sshdr
->asc
= sense_buffer
[2];
2443 sshdr
->ascq
= sense_buffer
[3];
2445 sshdr
->additional_length
= sense_buffer
[7];
2451 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2453 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2454 sb_len
: (sense_buffer
[7] + 8);
2456 sshdr
->asc
= sense_buffer
[12];
2458 sshdr
->ascq
= sense_buffer
[13];
2464 EXPORT_SYMBOL(scsi_normalize_sense
);
2466 bool scsi_command_normalize_sense(const struct scsi_cmnd
*cmd
,
2467 struct scsi_sense_hdr
*sshdr
)
2469 return scsi_normalize_sense(cmd
->sense_buffer
,
2470 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2472 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2475 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2476 * @sense_buffer: byte array of descriptor format sense data
2477 * @sb_len: number of valid bytes in sense_buffer
2478 * @desc_type: value of descriptor type to find
2479 * (e.g. 0 -> information)
2482 * only valid when sense data is in descriptor format
2485 * pointer to start of (first) descriptor if found else NULL
2487 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2490 int add_sen_len
, add_len
, desc_len
, k
;
2493 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2495 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2497 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2498 add_sen_len
: (sb_len
- 8);
2499 descp
= &sense_buffer
[8];
2500 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2502 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2503 desc_len
= add_len
+ 2;
2504 if (descp
[0] == desc_type
)
2506 if (add_len
< 0) // short descriptor ??
2511 EXPORT_SYMBOL(scsi_sense_desc_find
);
2514 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2515 * @sense_buffer: byte array of sense data
2516 * @sb_len: number of valid bytes in sense_buffer
2517 * @info_out: pointer to 64 integer where 8 or 4 byte information
2518 * field will be placed if found.
2521 * 1 if information field found, 0 if not found.
2523 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2532 switch (sense_buffer
[0] & 0x7f) {
2535 if (sense_buffer
[0] & 0x80) {
2536 *info_out
= (sense_buffer
[3] << 24) +
2537 (sense_buffer
[4] << 16) +
2538 (sense_buffer
[5] << 8) + sense_buffer
[6];
2544 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2546 if (ucp
&& (0xa == ucp
[1])) {
2548 for (j
= 0; j
< 8; ++j
) {
2561 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2564 * scsi_build_sense_buffer - build sense data in a buffer
2565 * @desc: Sense format (non zero == descriptor format,
2566 * 0 == fixed format)
2567 * @buf: Where to build sense data
2569 * @asc: Additional sense code
2570 * @ascq: Additional sense code qualifier
2573 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2576 buf
[0] = 0x72; /* descriptor, current */
2582 buf
[0] = 0x70; /* fixed, current */
2589 EXPORT_SYMBOL(scsi_build_sense_buffer
);
2592 * scsi_set_sense_information - set the information field in a
2593 * formatted sense data buffer
2594 * @buf: Where to build sense data
2595 * @info: 64-bit information value to be set
2598 void scsi_set_sense_information(u8
*buf
, u64 info
)
2600 if ((buf
[0] & 0x7f) == 0x72) {
2604 ucp
= (char *)scsi_sense_desc_find(buf
, len
+ 8, 0);
2607 ucp
= buf
+ 8 + len
;
2611 ucp
[2] = 0x80; /* Valid bit */
2613 put_unaligned_be64(info
, &ucp
[4]);
2614 } else if ((buf
[0] & 0x7f) == 0x70) {
2616 put_unaligned_be64(info
, &buf
[3]);
2619 EXPORT_SYMBOL(scsi_set_sense_information
);