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>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 #include "scsi_transport_api.h"
44 #include <trace/events/scsi.h>
46 static void scsi_eh_done(struct scsi_cmnd
*scmd
);
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
52 #define BUS_RESET_SETTLE_TIME (10)
53 #define HOST_RESET_SETTLE_TIME (10)
55 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
57 /* called with shost->host_lock held */
58 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
60 if (shost
->host_busy
== shost
->host_failed
) {
61 trace_scsi_eh_wakeup(shost
);
62 wake_up_process(shost
->ehandler
);
63 SCSI_LOG_ERROR_RECOVERY(5,
64 printk("Waking error handler thread\n"));
69 * scsi_schedule_eh - schedule EH for SCSI host
70 * @shost: SCSI host to invoke error handling on.
72 * Schedule SCSI EH without scmd.
74 void scsi_schedule_eh(struct Scsi_Host
*shost
)
78 spin_lock_irqsave(shost
->host_lock
, flags
);
80 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
81 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
82 shost
->host_eh_scheduled
++;
83 scsi_eh_wakeup(shost
);
86 spin_unlock_irqrestore(shost
->host_lock
, flags
);
88 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
91 * scsi_eh_scmd_add - add scsi cmd to error handling.
92 * @scmd: scmd to run eh on.
93 * @eh_flag: optional SCSI_EH flag.
98 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
100 struct Scsi_Host
*shost
= scmd
->device
->host
;
104 if (!shost
->ehandler
)
107 spin_lock_irqsave(shost
->host_lock
, flags
);
108 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
109 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
113 scmd
->eh_eflags
|= eh_flag
;
114 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
115 shost
->host_failed
++;
116 scsi_eh_wakeup(shost
);
118 spin_unlock_irqrestore(shost
->host_lock
, flags
);
123 * scsi_times_out - Timeout function for normal scsi commands.
124 * @req: request that is timing out.
127 * We do not need to lock this. There is the potential for a race
128 * only in that the normal completion handling might run, but if the
129 * normal completion function determines that the timer has already
130 * fired, then it mustn't do anything.
132 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
134 struct scsi_cmnd
*scmd
= req
->special
;
135 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
136 struct Scsi_Host
*host
= scmd
->device
->host
;
138 trace_scsi_dispatch_cmd_timeout(scmd
);
139 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
141 if (host
->transportt
->eh_timed_out
)
142 rtn
= host
->transportt
->eh_timed_out(scmd
);
143 else if (host
->hostt
->eh_timed_out
)
144 rtn
= host
->hostt
->eh_timed_out(scmd
);
146 scmd
->result
|= DID_TIME_OUT
<< 16;
148 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
149 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
)))
150 rtn
= BLK_EH_HANDLED
;
156 * scsi_block_when_processing_errors - Prevent cmds from being queued.
157 * @sdev: Device on which we are performing recovery.
160 * We block until the host is out of error recovery, and then check to
161 * see whether the host or the device is offline.
164 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
166 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
170 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
172 online
= scsi_device_online(sdev
);
174 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
179 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
181 #ifdef CONFIG_SCSI_LOGGING
183 * scsi_eh_prt_fail_stats - Log info on failures.
184 * @shost: scsi host being recovered.
185 * @work_q: Queue of scsi cmds to process.
187 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
188 struct list_head
*work_q
)
190 struct scsi_cmnd
*scmd
;
191 struct scsi_device
*sdev
;
192 int total_failures
= 0;
195 int devices_failed
= 0;
197 shost_for_each_device(sdev
, shost
) {
198 list_for_each_entry(scmd
, work_q
, eh_entry
) {
199 if (scmd
->device
== sdev
) {
201 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
208 if (cmd_cancel
|| cmd_failed
) {
209 SCSI_LOG_ERROR_RECOVERY(3,
210 sdev_printk(KERN_INFO
, sdev
,
211 "%s: cmds failed: %d, cancel: %d\n",
212 __func__
, cmd_failed
,
220 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
221 " devices require eh work\n",
222 total_failures
, devices_failed
));
227 * scsi_report_lun_change - Set flag on all *other* devices on the same target
228 * to indicate that a UNIT ATTENTION is expected.
229 * @sdev: Device reporting the UNIT ATTENTION
231 static void scsi_report_lun_change(struct scsi_device
*sdev
)
233 sdev
->sdev_target
->expecting_lun_change
= 1;
237 * scsi_report_sense - Examine scsi sense information and log messages for
238 * certain conditions, also issue uevents for some of them.
239 * @sdev: Device reporting the sense code
240 * @sshdr: sshdr to be examined
242 static void scsi_report_sense(struct scsi_device
*sdev
,
243 struct scsi_sense_hdr
*sshdr
)
245 enum scsi_device_event evt_type
= SDEV_EVT_MAXBITS
; /* i.e. none */
247 if (sshdr
->sense_key
== UNIT_ATTENTION
) {
248 if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x03) {
249 evt_type
= SDEV_EVT_INQUIRY_CHANGE_REPORTED
;
250 sdev_printk(KERN_WARNING
, sdev
,
251 "Inquiry data has changed");
252 } else if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x0e) {
253 evt_type
= SDEV_EVT_LUN_CHANGE_REPORTED
;
254 scsi_report_lun_change(sdev
);
255 sdev_printk(KERN_WARNING
, sdev
,
256 "Warning! Received an indication that the "
257 "LUN assignments on this target have "
258 "changed. The Linux SCSI layer does not "
259 "automatically remap LUN assignments.\n");
260 } else if (sshdr
->asc
== 0x3f)
261 sdev_printk(KERN_WARNING
, sdev
,
262 "Warning! Received an indication that the "
263 "operating parameters on this target have "
264 "changed. The Linux SCSI layer does not "
265 "automatically adjust these parameters.\n");
267 if (sshdr
->asc
== 0x38 && sshdr
->ascq
== 0x07) {
268 evt_type
= SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
;
269 sdev_printk(KERN_WARNING
, sdev
,
270 "Warning! Received an indication that the "
271 "LUN reached a thin provisioning soft "
275 if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x01) {
276 evt_type
= SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
;
277 sdev_printk(KERN_WARNING
, sdev
,
278 "Mode parameters changed");
279 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x09) {
280 evt_type
= SDEV_EVT_CAPACITY_CHANGE_REPORTED
;
281 sdev_printk(KERN_WARNING
, sdev
,
282 "Capacity data has changed");
283 } else if (sshdr
->asc
== 0x2a)
284 sdev_printk(KERN_WARNING
, sdev
,
285 "Parameters changed");
288 if (evt_type
!= SDEV_EVT_MAXBITS
) {
289 set_bit(evt_type
, sdev
->pending_events
);
290 schedule_work(&sdev
->event_work
);
295 * scsi_check_sense - Examine scsi cmd sense
296 * @scmd: Cmd to have sense checked.
299 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
302 * When a deferred error is detected the current command has
303 * not been executed and needs retrying.
305 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
307 struct scsi_device
*sdev
= scmd
->device
;
308 struct scsi_sense_hdr sshdr
;
310 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
311 return FAILED
; /* no valid sense data */
313 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&& scmd
->scsi_done
!= scsi_eh_done
)
315 * nasty: for mid-layer issued TURs, we need to return the
316 * actual sense data without any recovery attempt. For eh
317 * issued ones, we need to try to recover and interpret
321 scsi_report_sense(sdev
, &sshdr
);
323 if (scsi_sense_is_deferred(&sshdr
))
326 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
327 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
330 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
331 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
333 /* handler does not care. Drop down to default handling */
337 * Previous logic looked for FILEMARK, EOM or ILI which are
338 * mainly associated with tapes and returned SUCCESS.
340 if (sshdr
.response_code
== 0x70) {
342 if (scmd
->sense_buffer
[2] & 0xe0)
346 * descriptor format: look for "stream commands sense data
347 * descriptor" (see SSC-3). Assume single sense data
348 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
350 if ((sshdr
.additional_length
> 3) &&
351 (scmd
->sense_buffer
[8] == 0x4) &&
352 (scmd
->sense_buffer
[11] & 0xe0))
356 switch (sshdr
.sense_key
) {
359 case RECOVERED_ERROR
:
360 return /* soft_error */ SUCCESS
;
362 case ABORTED_COMMAND
:
363 if (sshdr
.asc
== 0x10) /* DIF */
370 * if we are expecting a cc/ua because of a bus reset that we
371 * performed, treat this just as a retry. otherwise this is
372 * information that we should pass up to the upper-level driver
373 * so that we can deal with it there.
375 if (scmd
->device
->expecting_cc_ua
) {
377 * Because some device does not queue unit
378 * attentions correctly, we carefully check
379 * additional sense code and qualifier so as
380 * not to squash media change unit attention.
382 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
383 scmd
->device
->expecting_cc_ua
= 0;
388 * we might also expect a cc/ua if another LUN on the target
389 * reported a UA with an ASC/ASCQ of 3F 0E -
390 * REPORTED LUNS DATA HAS CHANGED.
392 if (scmd
->device
->sdev_target
->expecting_lun_change
&&
393 sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
396 * if the device is in the process of becoming ready, we
399 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
402 * if the device is not started, we need to wake
403 * the error handler to start the motor
405 if (scmd
->device
->allow_restart
&&
406 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
409 * Pass the UA upwards for a determination in the completion
414 /* these are not supported */
416 if (sshdr
.asc
== 0x27 && sshdr
.ascq
== 0x07) {
417 /* Thin provisioning hard threshold reached */
418 set_host_byte(scmd
, DID_ALLOC_FAILURE
);
422 case VOLUME_OVERFLOW
:
425 set_host_byte(scmd
, DID_TARGET_FAILURE
);
429 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
430 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
431 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
432 set_host_byte(scmd
, DID_MEDIUM_ERROR
);
438 if (scmd
->device
->retry_hwerror
)
439 return ADD_TO_MLQUEUE
;
441 set_host_byte(scmd
, DID_TARGET_FAILURE
);
443 case ILLEGAL_REQUEST
:
444 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
445 sshdr
.asc
== 0x21 || /* Logical block address out of range */
446 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
447 sshdr
.asc
== 0x26) { /* Parameter value invalid */
448 set_host_byte(scmd
, DID_TARGET_FAILURE
);
457 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
459 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
460 struct scsi_device
*tmp_sdev
;
462 if (!sht
->change_queue_depth
||
463 sdev
->queue_depth
>= sdev
->max_queue_depth
)
466 if (time_before(jiffies
,
467 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
470 if (time_before(jiffies
,
471 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
475 * Walk all devices of a target and do
478 shost_for_each_device(tmp_sdev
, sdev
->host
) {
479 if (tmp_sdev
->channel
!= sdev
->channel
||
480 tmp_sdev
->id
!= sdev
->id
||
481 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
484 * call back into LLD to increase queue_depth by one
485 * with ramp up reason code.
487 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
488 SCSI_QDEPTH_RAMP_UP
);
489 sdev
->last_queue_ramp_up
= jiffies
;
493 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
495 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
496 struct scsi_device
*tmp_sdev
;
498 if (!sht
->change_queue_depth
)
501 shost_for_each_device(tmp_sdev
, sdev
->host
) {
502 if (tmp_sdev
->channel
!= sdev
->channel
||
503 tmp_sdev
->id
!= sdev
->id
)
506 * We do not know the number of commands that were at
507 * the device when we got the queue full so we start
508 * from the highest possible value and work our way down.
510 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
516 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
517 * @scmd: SCSI cmd to examine.
520 * This is *only* called when we are examining the status of commands
521 * queued during error recovery. the main difference here is that we
522 * don't allow for the possibility of retries here, and we are a lot
523 * more restrictive about what we consider acceptable.
525 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
528 * first check the host byte, to see if there is anything in there
529 * that would indicate what we need to do.
531 if (host_byte(scmd
->result
) == DID_RESET
) {
533 * rats. we are already in the error handler, so we now
534 * get to try and figure out what to do next. if the sense
535 * is valid, we have a pretty good idea of what to do.
536 * if not, we mark it as FAILED.
538 return scsi_check_sense(scmd
);
540 if (host_byte(scmd
->result
) != DID_OK
)
544 * next, check the message byte.
546 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
550 * now, check the status byte to see if this indicates
553 switch (status_byte(scmd
->result
)) {
555 scsi_handle_queue_ramp_up(scmd
->device
);
556 case COMMAND_TERMINATED
:
558 case CHECK_CONDITION
:
559 return scsi_check_sense(scmd
);
561 case INTERMEDIATE_GOOD
:
562 case INTERMEDIATE_C_GOOD
:
564 * who knows? FIXME(eric)
567 case RESERVATION_CONFLICT
:
568 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
569 /* it is a success, we probed the device and
572 /* otherwise, we failed to send the command */
575 scsi_handle_queue_full(scmd
->device
);
586 * scsi_eh_done - Completion function for error handling.
587 * @scmd: Cmd that is done.
589 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
591 struct completion
*eh_action
;
593 SCSI_LOG_ERROR_RECOVERY(3,
594 printk("%s scmd: %p result: %x\n",
595 __func__
, scmd
, scmd
->result
));
597 eh_action
= scmd
->device
->host
->eh_action
;
603 * scsi_try_host_reset - ask host adapter to reset itself
604 * @scmd: SCSI cmd to send host reset.
606 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
610 struct Scsi_Host
*host
= scmd
->device
->host
;
611 struct scsi_host_template
*hostt
= host
->hostt
;
613 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
616 if (!hostt
->eh_host_reset_handler
)
619 rtn
= hostt
->eh_host_reset_handler(scmd
);
621 if (rtn
== SUCCESS
) {
622 if (!hostt
->skip_settle_delay
)
623 ssleep(HOST_RESET_SETTLE_TIME
);
624 spin_lock_irqsave(host
->host_lock
, flags
);
625 scsi_report_bus_reset(host
, scmd_channel(scmd
));
626 spin_unlock_irqrestore(host
->host_lock
, flags
);
633 * scsi_try_bus_reset - ask host to perform a bus reset
634 * @scmd: SCSI cmd to send bus reset.
636 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
640 struct Scsi_Host
*host
= scmd
->device
->host
;
641 struct scsi_host_template
*hostt
= host
->hostt
;
643 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
646 if (!hostt
->eh_bus_reset_handler
)
649 rtn
= hostt
->eh_bus_reset_handler(scmd
);
651 if (rtn
== SUCCESS
) {
652 if (!hostt
->skip_settle_delay
)
653 ssleep(BUS_RESET_SETTLE_TIME
);
654 spin_lock_irqsave(host
->host_lock
, flags
);
655 scsi_report_bus_reset(host
, scmd_channel(scmd
));
656 spin_unlock_irqrestore(host
->host_lock
, flags
);
662 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
665 sdev
->expecting_cc_ua
= 1;
669 * scsi_try_target_reset - Ask host to perform a target reset
670 * @scmd: SCSI cmd used to send a target reset
673 * There is no timeout for this operation. if this operation is
674 * unreliable for a given host, then the host itself needs to put a
675 * timer on it, and set the host back to a consistent state prior to
678 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
682 struct Scsi_Host
*host
= scmd
->device
->host
;
683 struct scsi_host_template
*hostt
= host
->hostt
;
685 if (!hostt
->eh_target_reset_handler
)
688 rtn
= hostt
->eh_target_reset_handler(scmd
);
689 if (rtn
== SUCCESS
) {
690 spin_lock_irqsave(host
->host_lock
, flags
);
691 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
692 __scsi_report_device_reset
);
693 spin_unlock_irqrestore(host
->host_lock
, flags
);
700 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
701 * @scmd: SCSI cmd used to send BDR
704 * There is no timeout for this operation. if this operation is
705 * unreliable for a given host, then the host itself needs to put a
706 * timer on it, and set the host back to a consistent state prior to
709 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
712 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
714 if (!hostt
->eh_device_reset_handler
)
717 rtn
= hostt
->eh_device_reset_handler(scmd
);
719 __scsi_report_device_reset(scmd
->device
, NULL
);
723 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
, struct scsi_cmnd
*scmd
)
725 if (!hostt
->eh_abort_handler
)
728 return hostt
->eh_abort_handler(scmd
);
731 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
733 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
734 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
735 if (scsi_try_target_reset(scmd
) != SUCCESS
)
736 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
737 scsi_try_host_reset(scmd
);
741 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
742 * @scmd: SCSI command structure to hijack
743 * @ses: structure to save restore information
744 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
745 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
746 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
748 * This function is used to save a scsi command information before re-execution
749 * as part of the error recovery process. If @sense_bytes is 0 the command
750 * sent must be one that does not transfer any data. If @sense_bytes != 0
751 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
752 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
754 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
755 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
757 struct scsi_device
*sdev
= scmd
->device
;
760 * We need saved copies of a number of fields - this is because
761 * error handling may need to overwrite these with different values
762 * to run different commands, and once error handling is complete,
763 * we will need to restore these values prior to running the actual
766 ses
->cmd_len
= scmd
->cmd_len
;
767 ses
->cmnd
= scmd
->cmnd
;
768 ses
->data_direction
= scmd
->sc_data_direction
;
769 ses
->sdb
= scmd
->sdb
;
770 ses
->next_rq
= scmd
->request
->next_rq
;
771 ses
->result
= scmd
->result
;
772 ses
->underflow
= scmd
->underflow
;
773 ses
->prot_op
= scmd
->prot_op
;
775 scmd
->prot_op
= SCSI_PROT_NORMAL
;
776 scmd
->cmnd
= ses
->eh_cmnd
;
777 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
778 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
779 scmd
->request
->next_rq
= NULL
;
782 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
784 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
786 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
787 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
788 scmd
->sdb
.table
.nents
= 1;
789 scmd
->cmnd
[0] = REQUEST_SENSE
;
790 scmd
->cmnd
[4] = scmd
->sdb
.length
;
791 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
793 scmd
->sc_data_direction
= DMA_NONE
;
795 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
796 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
797 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
803 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
804 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
805 (sdev
->lun
<< 5 & 0xe0);
808 * Zero the sense buffer. The scsi spec mandates that any
809 * untransferred sense data should be interpreted as being zero.
811 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
813 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
816 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
817 * @scmd: SCSI command structure to restore
818 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
820 * Undo any damage done by above scsi_eh_prep_cmnd().
822 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
825 * Restore original data
827 scmd
->cmd_len
= ses
->cmd_len
;
828 scmd
->cmnd
= ses
->cmnd
;
829 scmd
->sc_data_direction
= ses
->data_direction
;
830 scmd
->sdb
= ses
->sdb
;
831 scmd
->request
->next_rq
= ses
->next_rq
;
832 scmd
->result
= ses
->result
;
833 scmd
->underflow
= ses
->underflow
;
834 scmd
->prot_op
= ses
->prot_op
;
836 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
839 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
840 * @scmd: SCSI command structure to hijack
842 * @cmnd_size: size in bytes of @cmnd
843 * @timeout: timeout for this request
844 * @sense_bytes: size of sense data to copy or 0
846 * This function is used to send a scsi command down to a target device
847 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
850 * SUCCESS or FAILED or NEEDS_RETRY
852 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
853 int cmnd_size
, int timeout
, unsigned sense_bytes
)
855 struct scsi_device
*sdev
= scmd
->device
;
856 struct Scsi_Host
*shost
= sdev
->host
;
857 DECLARE_COMPLETION_ONSTACK(done
);
858 unsigned long timeleft
= timeout
;
859 struct scsi_eh_save ses
;
860 const unsigned long stall_for
= msecs_to_jiffies(100);
864 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
865 shost
->eh_action
= &done
;
868 scmd
->scsi_done
= scsi_eh_done
;
869 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
871 if (timeleft
> stall_for
) {
872 scsi_eh_restore_cmnd(scmd
, &ses
);
873 timeleft
-= stall_for
;
874 msleep(jiffies_to_msecs(stall_for
));
877 /* signal not to enter either branch of the if () below */
881 timeleft
= wait_for_completion_timeout(&done
, timeout
);
884 shost
->eh_action
= NULL
;
886 scsi_log_completion(scmd
, rtn
);
888 SCSI_LOG_ERROR_RECOVERY(3,
889 printk("%s: scmd: %p, timeleft: %ld\n",
890 __func__
, scmd
, timeleft
));
893 * If there is time left scsi_eh_done got called, and we will examine
894 * the actual status codes to see whether the command actually did
895 * complete normally, else if we have a zero return and no time left,
896 * the command must still be pending, so abort it and return FAILED.
897 * If we never actually managed to issue the command, because
898 * ->queuecommand() kept returning non zero, use the rtn = FAILED
899 * value above (so don't execute either branch of the if)
902 rtn
= scsi_eh_completed_normally(scmd
);
903 SCSI_LOG_ERROR_RECOVERY(3,
904 printk("%s: scsi_eh_completed_normally %x\n",
920 scsi_abort_eh_cmnd(scmd
);
924 scsi_eh_restore_cmnd(scmd
, &ses
);
926 if (scmd
->request
->cmd_type
!= REQ_TYPE_BLOCK_PC
) {
927 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
929 rtn
= sdrv
->eh_action(scmd
, cmnd
, cmnd_size
, rtn
);
936 * scsi_request_sense - Request sense data from a particular target.
937 * @scmd: SCSI cmd for request sense.
940 * Some hosts automatically obtain this information, others require
941 * that we obtain it on our own. This function will *not* return until
942 * the command either times out, or it completes.
944 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
946 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
950 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
951 * @scmd: Original SCSI cmd that eh has finished.
952 * @done_q: Queue for processed commands.
955 * We don't want to use the normal command completion while we are are
956 * still handling errors - it may cause other commands to be queued,
957 * and that would disturb what we are doing. Thus we really want to
958 * keep a list of pending commands for final completion, and once we
959 * are ready to leave error handling we handle completion for real.
961 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
963 scmd
->device
->host
->host_failed
--;
965 list_move_tail(&scmd
->eh_entry
, done_q
);
967 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
970 * scsi_eh_get_sense - Get device sense data.
971 * @work_q: Queue of commands to process.
972 * @done_q: Queue of processed commands.
975 * See if we need to request sense information. if so, then get it
976 * now, so we have a better idea of what to do.
979 * This has the unfortunate side effect that if a shost adapter does
980 * not automatically request sense information, we end up shutting
981 * it down before we request it.
983 * All drivers should request sense information internally these days,
984 * so for now all I have to say is tough noogies if you end up in here.
986 * XXX: Long term this code should go away, but that needs an audit of
989 int scsi_eh_get_sense(struct list_head
*work_q
,
990 struct list_head
*done_q
)
992 struct scsi_cmnd
*scmd
, *next
;
995 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
996 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
997 SCSI_SENSE_VALID(scmd
))
1000 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
1001 "%s: requesting sense\n",
1003 rtn
= scsi_request_sense(scmd
);
1007 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1008 " result %x\n", scmd
,
1010 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
1012 rtn
= scsi_decide_disposition(scmd
);
1015 * if the result was normal, then just pass it along to the
1019 /* we don't want this command reissued, just
1020 * finished with the sense data, so set
1021 * retries to the max allowed to ensure it
1022 * won't get reissued */
1023 scmd
->retries
= scmd
->allowed
;
1024 else if (rtn
!= NEEDS_RETRY
)
1027 scsi_eh_finish_cmd(scmd
, done_q
);
1030 return list_empty(work_q
);
1032 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
1035 * scsi_eh_tur - Send TUR to device.
1036 * @scmd: &scsi_cmnd to send TUR
1039 * 0 - Device is ready. 1 - Device NOT ready.
1041 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
1043 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
1044 int retry_cnt
= 1, rtn
;
1047 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
1048 scmd
->device
->eh_timeout
, 0);
1050 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
1051 __func__
, scmd
, rtn
));
1066 * scsi_eh_test_devices - check if devices are responding from error recovery.
1067 * @cmd_list: scsi commands in error recovery.
1068 * @work_q: queue for commands which still need more error recovery
1069 * @done_q: queue for commands which are finished
1070 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1073 * Tests if devices are in a working state. Commands to devices now in
1074 * a working state are sent to the done_q while commands to devices which
1075 * are still failing to respond are returned to the work_q for more
1078 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1079 struct list_head
*work_q
,
1080 struct list_head
*done_q
, int try_stu
)
1082 struct scsi_cmnd
*scmd
, *next
;
1083 struct scsi_device
*sdev
;
1086 while (!list_empty(cmd_list
)) {
1087 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1088 sdev
= scmd
->device
;
1090 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1091 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1092 !scsi_eh_tur(scmd
)) ||
1095 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1096 if (scmd
->device
== sdev
) {
1098 scsi_eh_finish_cmd(scmd
, done_q
);
1100 list_move_tail(&scmd
->eh_entry
, work_q
);
1103 return list_empty(work_q
);
1108 * scsi_eh_abort_cmds - abort pending commands.
1109 * @work_q: &list_head for pending commands.
1110 * @done_q: &list_head for processed commands.
1113 * Try and see whether or not it makes sense to try and abort the
1114 * running command. This only works out to be the case if we have one
1115 * command that has timed out. If the command simply failed, it makes
1116 * no sense to try and abort the command, since as far as the shost
1117 * adapter is concerned, it isn't running.
1119 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
1120 struct list_head
*done_q
)
1122 struct scsi_cmnd
*scmd
, *next
;
1123 LIST_HEAD(check_list
);
1126 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1127 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
1129 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1130 "0x%p\n", current
->comm
,
1132 rtn
= scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
);
1133 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1134 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
1135 if (rtn
== FAST_IO_FAIL
)
1136 scsi_eh_finish_cmd(scmd
, done_q
);
1138 list_move_tail(&scmd
->eh_entry
, &check_list
);
1140 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1147 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1151 * scsi_eh_try_stu - Send START_UNIT to device.
1152 * @scmd: &scsi_cmnd to send START_UNIT
1155 * 0 - Device is ready. 1 - Device NOT ready.
1157 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1159 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1161 if (scmd
->device
->allow_restart
) {
1162 int i
, rtn
= NEEDS_RETRY
;
1164 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1165 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1175 * scsi_eh_stu - send START_UNIT if needed
1176 * @shost: &scsi host being recovered.
1177 * @work_q: &list_head for pending commands.
1178 * @done_q: &list_head for processed commands.
1181 * If commands are failing due to not ready, initializing command required,
1182 * try revalidating the device, which will end up sending a start unit.
1184 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1185 struct list_head
*work_q
,
1186 struct list_head
*done_q
)
1188 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1189 struct scsi_device
*sdev
;
1191 shost_for_each_device(sdev
, shost
) {
1193 list_for_each_entry(scmd
, work_q
, eh_entry
)
1194 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1195 scsi_check_sense(scmd
) == FAILED
) {
1203 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1204 " 0x%p\n", current
->comm
, sdev
));
1206 if (!scsi_eh_try_stu(stu_scmd
)) {
1207 if (!scsi_device_online(sdev
) ||
1208 !scsi_eh_tur(stu_scmd
)) {
1209 list_for_each_entry_safe(scmd
, next
,
1211 if (scmd
->device
== sdev
)
1212 scsi_eh_finish_cmd(scmd
, done_q
);
1216 SCSI_LOG_ERROR_RECOVERY(3,
1217 printk("%s: START_UNIT failed to sdev:"
1218 " 0x%p\n", current
->comm
, sdev
));
1222 return list_empty(work_q
);
1227 * scsi_eh_bus_device_reset - send bdr if needed
1228 * @shost: scsi host being recovered.
1229 * @work_q: &list_head for pending commands.
1230 * @done_q: &list_head for processed commands.
1233 * Try a bus device reset. Still, look to see whether we have multiple
1234 * devices that are jammed or not - if we have multiple devices, it
1235 * makes no sense to try bus_device_reset - we really would need to try
1236 * a bus_reset instead.
1238 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1239 struct list_head
*work_q
,
1240 struct list_head
*done_q
)
1242 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1243 struct scsi_device
*sdev
;
1246 shost_for_each_device(sdev
, shost
) {
1248 list_for_each_entry(scmd
, work_q
, eh_entry
)
1249 if (scmd
->device
== sdev
) {
1257 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1258 " 0x%p\n", current
->comm
,
1260 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1261 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1262 if (!scsi_device_online(sdev
) ||
1263 rtn
== FAST_IO_FAIL
||
1264 !scsi_eh_tur(bdr_scmd
)) {
1265 list_for_each_entry_safe(scmd
, next
,
1267 if (scmd
->device
== sdev
)
1268 scsi_eh_finish_cmd(scmd
,
1273 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1281 return list_empty(work_q
);
1285 * scsi_eh_target_reset - send target reset if needed
1286 * @shost: scsi host being recovered.
1287 * @work_q: &list_head for pending commands.
1288 * @done_q: &list_head for processed commands.
1291 * Try a target reset.
1293 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1294 struct list_head
*work_q
,
1295 struct list_head
*done_q
)
1297 LIST_HEAD(tmp_list
);
1298 LIST_HEAD(check_list
);
1300 list_splice_init(work_q
, &tmp_list
);
1302 while (!list_empty(&tmp_list
)) {
1303 struct scsi_cmnd
*next
, *scmd
;
1307 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1310 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1312 current
->comm
, id
));
1313 rtn
= scsi_try_target_reset(scmd
);
1314 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1315 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1318 current
->comm
, id
));
1319 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1320 if (scmd_id(scmd
) != id
)
1324 list_move_tail(&scmd
->eh_entry
, &check_list
);
1325 else if (rtn
== FAST_IO_FAIL
)
1326 scsi_eh_finish_cmd(scmd
, done_q
);
1328 /* push back on work queue for further processing */
1329 list_move(&scmd
->eh_entry
, work_q
);
1333 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1337 * scsi_eh_bus_reset - send a bus reset
1338 * @shost: &scsi host being recovered.
1339 * @work_q: &list_head for pending commands.
1340 * @done_q: &list_head for processed commands.
1342 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1343 struct list_head
*work_q
,
1344 struct list_head
*done_q
)
1346 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1347 LIST_HEAD(check_list
);
1348 unsigned int channel
;
1352 * we really want to loop over the various channels, and do this on
1353 * a channel by channel basis. we should also check to see if any
1354 * of the failed commands are on soft_reset devices, and if so, skip
1358 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1360 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1361 if (channel
== scmd_channel(scmd
)) {
1365 * FIXME add back in some support for
1366 * soft_reset devices.
1373 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1374 " %d\n", current
->comm
,
1376 rtn
= scsi_try_bus_reset(chan_scmd
);
1377 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1378 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1379 if (channel
== scmd_channel(scmd
)) {
1380 if (rtn
== FAST_IO_FAIL
)
1381 scsi_eh_finish_cmd(scmd
,
1384 list_move_tail(&scmd
->eh_entry
,
1389 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1390 " failed chan: %d\n",
1395 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1399 * scsi_eh_host_reset - send a host reset
1400 * @work_q: list_head for processed commands.
1401 * @done_q: list_head for processed commands.
1403 static int scsi_eh_host_reset(struct list_head
*work_q
,
1404 struct list_head
*done_q
)
1406 struct scsi_cmnd
*scmd
, *next
;
1407 LIST_HEAD(check_list
);
1410 if (!list_empty(work_q
)) {
1411 scmd
= list_entry(work_q
->next
,
1412 struct scsi_cmnd
, eh_entry
);
1414 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1417 rtn
= scsi_try_host_reset(scmd
);
1418 if (rtn
== SUCCESS
) {
1419 list_splice_init(work_q
, &check_list
);
1420 } else if (rtn
== FAST_IO_FAIL
) {
1421 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1422 scsi_eh_finish_cmd(scmd
, done_q
);
1425 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1430 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1434 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1435 * @work_q: list_head for processed commands.
1436 * @done_q: list_head for processed commands.
1438 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1439 struct list_head
*done_q
)
1441 struct scsi_cmnd
*scmd
, *next
;
1443 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1444 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1445 "not ready after error recovery\n");
1446 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1447 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1449 * FIXME: Handle lost cmds.
1452 scsi_eh_finish_cmd(scmd
, done_q
);
1458 * scsi_noretry_cmd - determinte if command should be failed fast
1459 * @scmd: SCSI cmd to examine.
1461 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1463 switch (host_byte(scmd
->result
)) {
1467 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1469 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1471 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1472 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1475 case DID_SOFT_ERROR
:
1476 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1479 switch (status_byte(scmd
->result
)) {
1480 case CHECK_CONDITION
:
1482 * assume caller has checked sense and determinted
1483 * the check condition was retryable.
1485 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1486 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1494 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1495 * @scmd: SCSI cmd to examine.
1498 * This is *only* called when we are examining the status after sending
1499 * out the actual data command. any commands that are queued for error
1500 * recovery (e.g. test_unit_ready) do *not* come through here.
1502 * When this routine returns failed, it means the error handler thread
1503 * is woken. In cases where the error code indicates an error that
1504 * doesn't require the error handler read (i.e. we don't need to
1505 * abort/reset), this function should return SUCCESS.
1507 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1512 * if the device is offline, then we clearly just pass the result back
1513 * up to the top level.
1515 if (!scsi_device_online(scmd
->device
)) {
1516 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1523 * first check the host byte, to see if there is anything in there
1524 * that would indicate what we need to do.
1526 switch (host_byte(scmd
->result
)) {
1527 case DID_PASSTHROUGH
:
1529 * no matter what, pass this through to the upper layer.
1530 * nuke this special code so that it looks like we are saying
1533 scmd
->result
&= 0xff00ffff;
1537 * looks good. drop through, and check the next byte.
1540 case DID_NO_CONNECT
:
1541 case DID_BAD_TARGET
:
1544 * note - this means that we just report the status back
1545 * to the top level driver, not that we actually think
1546 * that it indicates SUCCESS.
1550 * when the low level driver returns did_soft_error,
1551 * it is responsible for keeping an internal retry counter
1552 * in order to avoid endless loops (db)
1554 * actually this is a bug in this function here. we should
1555 * be mindful of the maximum number of retries specified
1556 * and not get stuck in a loop.
1558 case DID_SOFT_ERROR
:
1564 return ADD_TO_MLQUEUE
;
1565 case DID_TRANSPORT_DISRUPTED
:
1567 * LLD/transport was disrupted during processing of the IO.
1568 * The transport class is now blocked/blocking,
1569 * and the transport will decide what to do with the IO
1570 * based on its timers and recovery capablilities if
1571 * there are enough retries.
1574 case DID_TRANSPORT_FAILFAST
:
1576 * The transport decided to failfast the IO (most likely
1577 * the fast io fail tmo fired), so send IO directly upwards.
1581 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1582 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1584 * execute reservation conflict processing code
1594 * when we scan the bus, we get timeout messages for
1595 * these commands if there is no device available.
1596 * other hosts report did_no_connect for the same thing.
1598 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1599 scmd
->cmnd
[0] == INQUIRY
)) {
1611 * next, check the message byte.
1613 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1617 * check the status byte to see if this indicates anything special.
1619 switch (status_byte(scmd
->result
)) {
1621 scsi_handle_queue_full(scmd
->device
);
1623 * the case of trying to send too many commands to a
1624 * tagged queueing device.
1628 * device can't talk to us at the moment. Should only
1629 * occur (SAM-3) when the task queue is empty, so will cause
1630 * the empty queue handling to trigger a stall in the
1633 return ADD_TO_MLQUEUE
;
1635 if (scmd
->cmnd
[0] == REPORT_LUNS
)
1636 scmd
->device
->sdev_target
->expecting_lun_change
= 0;
1637 scsi_handle_queue_ramp_up(scmd
->device
);
1638 case COMMAND_TERMINATED
:
1642 case CHECK_CONDITION
:
1643 rtn
= scsi_check_sense(scmd
);
1644 if (rtn
== NEEDS_RETRY
)
1646 /* if rtn == FAILED, we have no sense information;
1647 * returning FAILED will wake the error handler thread
1648 * to collect the sense and redo the decide
1651 case CONDITION_GOOD
:
1652 case INTERMEDIATE_GOOD
:
1653 case INTERMEDIATE_C_GOOD
:
1656 * who knows? FIXME(eric)
1660 case RESERVATION_CONFLICT
:
1661 sdev_printk(KERN_INFO
, scmd
->device
,
1662 "reservation conflict\n");
1663 set_host_byte(scmd
, DID_NEXUS_FAILURE
);
1664 return SUCCESS
; /* causes immediate i/o error */
1672 /* we requeue for retry because the error was retryable, and
1673 * the request was not marked fast fail. Note that above,
1674 * even if the request is marked fast fail, we still requeue
1675 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1676 if ((++scmd
->retries
) <= scmd
->allowed
1677 && !scsi_noretry_cmd(scmd
)) {
1681 * no more retries - report this one back to upper level.
1687 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1689 __blk_put_request(req
->q
, req
);
1693 * scsi_eh_lock_door - Prevent medium removal for the specified device
1694 * @sdev: SCSI device to prevent medium removal
1697 * We must be called from process context.
1700 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1701 * head of the devices request queue, and continue.
1703 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1705 struct request
*req
;
1708 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1709 * request becomes available
1711 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1713 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1717 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1720 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1722 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1723 req
->cmd_flags
|= REQ_QUIET
;
1724 req
->timeout
= 10 * HZ
;
1727 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1731 * scsi_restart_operations - restart io operations to the specified host.
1732 * @shost: Host we are restarting.
1735 * When we entered the error handler, we blocked all further i/o to
1736 * this device. we need to 'reverse' this process.
1738 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1740 struct scsi_device
*sdev
;
1741 unsigned long flags
;
1744 * If the door was locked, we need to insert a door lock request
1745 * onto the head of the SCSI request queue for the device. There
1746 * is no point trying to lock the door of an off-line device.
1748 shost_for_each_device(sdev
, shost
) {
1749 if (scsi_device_online(sdev
) && sdev
->locked
)
1750 scsi_eh_lock_door(sdev
);
1754 * next free up anything directly waiting upon the host. this
1755 * will be requests for character device operations, and also for
1756 * ioctls to queued block devices.
1758 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1761 spin_lock_irqsave(shost
->host_lock
, flags
);
1762 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1763 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1764 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1765 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1767 wake_up(&shost
->host_wait
);
1770 * finally we need to re-initiate requests that may be pending. we will
1771 * have had everything blocked while error handling is taking place, and
1772 * now that error recovery is done, we will need to ensure that these
1773 * requests are started.
1775 scsi_run_host_queues(shost
);
1778 * if eh is active and host_eh_scheduled is pending we need to re-run
1779 * recovery. we do this check after scsi_run_host_queues() to allow
1780 * everything pent up since the last eh run a chance to make forward
1781 * progress before we sync again. Either we'll immediately re-run
1782 * recovery or scsi_device_unbusy() will wake us again when these
1783 * pending commands complete.
1785 spin_lock_irqsave(shost
->host_lock
, flags
);
1786 if (shost
->host_eh_scheduled
)
1787 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
1788 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
1789 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1793 * scsi_eh_ready_devs - check device ready state and recover if not.
1794 * @shost: host to be recovered.
1795 * @work_q: &list_head for pending commands.
1796 * @done_q: &list_head for processed commands.
1798 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1799 struct list_head
*work_q
,
1800 struct list_head
*done_q
)
1802 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1803 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1804 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1805 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1806 if (!scsi_eh_host_reset(work_q
, done_q
))
1807 scsi_eh_offline_sdevs(work_q
,
1810 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1813 * scsi_eh_flush_done_q - finish processed commands or retry them.
1814 * @done_q: list_head of processed commands.
1816 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1818 struct scsi_cmnd
*scmd
, *next
;
1820 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1821 list_del_init(&scmd
->eh_entry
);
1822 if (scsi_device_online(scmd
->device
) &&
1823 !scsi_noretry_cmd(scmd
) &&
1824 (++scmd
->retries
<= scmd
->allowed
)) {
1825 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1829 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1832 * If just we got sense for the device (called
1833 * scsi_eh_get_sense), scmd->result is already
1834 * set, do not set DRIVER_TIMEOUT.
1837 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1838 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1840 current
->comm
, scmd
));
1841 scsi_finish_command(scmd
);
1845 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1848 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1849 * @shost: Host to unjam.
1852 * When we come in here, we *know* that all commands on the bus have
1853 * either completed, failed or timed out. we also know that no further
1854 * commands are being sent to the host, so things are relatively quiet
1855 * and we have freedom to fiddle with things as we wish.
1857 * This is only the *default* implementation. it is possible for
1858 * individual drivers to supply their own version of this function, and
1859 * if the maintainer wishes to do this, it is strongly suggested that
1860 * this function be taken as a template and modified. this function
1861 * was designed to correctly handle problems for about 95% of the
1862 * different cases out there, and it should always provide at least a
1863 * reasonable amount of error recovery.
1865 * Any command marked 'failed' or 'timeout' must eventually have
1866 * scsi_finish_cmd() called for it. we do all of the retry stuff
1867 * here, so when we restart the host after we return it should have an
1870 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1872 unsigned long flags
;
1873 LIST_HEAD(eh_work_q
);
1874 LIST_HEAD(eh_done_q
);
1876 spin_lock_irqsave(shost
->host_lock
, flags
);
1877 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1878 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1880 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1882 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1883 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1884 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1886 scsi_eh_flush_done_q(&eh_done_q
);
1890 * scsi_error_handler - SCSI error handler thread
1891 * @data: Host for which we are running.
1894 * This is the main error handling loop. This is run as a kernel thread
1895 * for every SCSI host and handles all error handling activity.
1897 int scsi_error_handler(void *data
)
1899 struct Scsi_Host
*shost
= data
;
1902 * We use TASK_INTERRUPTIBLE so that the thread is not
1903 * counted against the load average as a running process.
1904 * We never actually get interrupted because kthread_run
1905 * disables signal delivery for the created thread.
1907 while (!kthread_should_stop()) {
1908 set_current_state(TASK_INTERRUPTIBLE
);
1909 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1910 shost
->host_failed
!= shost
->host_busy
) {
1911 SCSI_LOG_ERROR_RECOVERY(1,
1912 printk("Error handler scsi_eh_%d sleeping\n",
1918 __set_current_state(TASK_RUNNING
);
1919 SCSI_LOG_ERROR_RECOVERY(1,
1920 printk("Error handler scsi_eh_%d waking up\n",
1924 * We have a host that is failing for some reason. Figure out
1925 * what we need to do to get it up and online again (if we can).
1926 * If we fail, we end up taking the thing offline.
1928 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
1929 SCSI_LOG_ERROR_RECOVERY(1,
1930 printk(KERN_ERR
"Error handler scsi_eh_%d "
1931 "unable to autoresume\n",
1936 if (shost
->transportt
->eh_strategy_handler
)
1937 shost
->transportt
->eh_strategy_handler(shost
);
1939 scsi_unjam_host(shost
);
1942 * Note - if the above fails completely, the action is to take
1943 * individual devices offline and flush the queue of any
1944 * outstanding requests that may have been pending. When we
1945 * restart, we restart any I/O to any other devices on the bus
1946 * which are still online.
1948 scsi_restart_operations(shost
);
1949 if (!shost
->eh_noresume
)
1950 scsi_autopm_put_host(shost
);
1952 __set_current_state(TASK_RUNNING
);
1954 SCSI_LOG_ERROR_RECOVERY(1,
1955 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1956 shost
->ehandler
= NULL
;
1961 * Function: scsi_report_bus_reset()
1963 * Purpose: Utility function used by low-level drivers to report that
1964 * they have observed a bus reset on the bus being handled.
1966 * Arguments: shost - Host in question
1967 * channel - channel on which reset was observed.
1971 * Lock status: Host lock must be held.
1973 * Notes: This only needs to be called if the reset is one which
1974 * originates from an unknown location. Resets originated
1975 * by the mid-level itself don't need to call this, but there
1976 * should be no harm.
1978 * The main purpose of this is to make sure that a CHECK_CONDITION
1979 * is properly treated.
1981 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1983 struct scsi_device
*sdev
;
1985 __shost_for_each_device(sdev
, shost
) {
1986 if (channel
== sdev_channel(sdev
))
1987 __scsi_report_device_reset(sdev
, NULL
);
1990 EXPORT_SYMBOL(scsi_report_bus_reset
);
1993 * Function: scsi_report_device_reset()
1995 * Purpose: Utility function used by low-level drivers to report that
1996 * they have observed a device reset on the device being handled.
1998 * Arguments: shost - Host in question
1999 * channel - channel on which reset was observed
2000 * target - target on which reset was observed
2004 * Lock status: Host lock must be held
2006 * Notes: This only needs to be called if the reset is one which
2007 * originates from an unknown location. Resets originated
2008 * by the mid-level itself don't need to call this, but there
2009 * should be no harm.
2011 * The main purpose of this is to make sure that a CHECK_CONDITION
2012 * is properly treated.
2014 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
2016 struct scsi_device
*sdev
;
2018 __shost_for_each_device(sdev
, shost
) {
2019 if (channel
== sdev_channel(sdev
) &&
2020 target
== sdev_id(sdev
))
2021 __scsi_report_device_reset(sdev
, NULL
);
2024 EXPORT_SYMBOL(scsi_report_device_reset
);
2027 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
2032 * Function: scsi_reset_provider
2034 * Purpose: Send requested reset to a bus or device at any phase.
2036 * Arguments: device - device to send reset to
2037 * flag - reset type (see scsi.h)
2039 * Returns: SUCCESS/FAILURE.
2041 * Notes: This is used by the SCSI Generic driver to provide
2042 * Bus/Device reset capability.
2045 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
2047 struct scsi_cmnd
*scmd
;
2048 struct Scsi_Host
*shost
= dev
->host
;
2050 unsigned long flags
;
2053 if (scsi_autopm_get_host(shost
) < 0)
2056 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
2057 blk_rq_init(NULL
, &req
);
2058 scmd
->request
= &req
;
2060 scmd
->cmnd
= req
.cmd
;
2062 scmd
->scsi_done
= scsi_reset_provider_done_command
;
2063 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2067 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2069 spin_lock_irqsave(shost
->host_lock
, flags
);
2070 shost
->tmf_in_progress
= 1;
2071 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2074 case SCSI_TRY_RESET_DEVICE
:
2075 rtn
= scsi_try_bus_device_reset(scmd
);
2079 case SCSI_TRY_RESET_TARGET
:
2080 rtn
= scsi_try_target_reset(scmd
);
2084 case SCSI_TRY_RESET_BUS
:
2085 rtn
= scsi_try_bus_reset(scmd
);
2089 case SCSI_TRY_RESET_HOST
:
2090 rtn
= scsi_try_host_reset(scmd
);
2096 spin_lock_irqsave(shost
->host_lock
, flags
);
2097 shost
->tmf_in_progress
= 0;
2098 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2101 * be sure to wake up anyone who was sleeping or had their queue
2102 * suspended while we performed the TMF.
2104 SCSI_LOG_ERROR_RECOVERY(3,
2105 printk("%s: waking up host to restart after TMF\n",
2108 wake_up(&shost
->host_wait
);
2110 scsi_run_host_queues(shost
);
2112 scsi_next_command(scmd
);
2113 scsi_autopm_put_host(shost
);
2116 EXPORT_SYMBOL(scsi_reset_provider
);
2119 * scsi_normalize_sense - normalize main elements from either fixed or
2120 * descriptor sense data format into a common format.
2122 * @sense_buffer: byte array containing sense data returned by device
2123 * @sb_len: number of valid bytes in sense_buffer
2124 * @sshdr: pointer to instance of structure that common
2125 * elements are written to.
2128 * The "main elements" from sense data are: response_code, sense_key,
2129 * asc, ascq and additional_length (only for descriptor format).
2131 * Typically this function can be called after a device has
2132 * responded to a SCSI command with the CHECK_CONDITION status.
2135 * 1 if valid sense data information found, else 0;
2137 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
2138 struct scsi_sense_hdr
*sshdr
)
2140 if (!sense_buffer
|| !sb_len
)
2143 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
2145 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
2147 if (!scsi_sense_valid(sshdr
))
2150 if (sshdr
->response_code
>= 0x72) {
2155 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
2157 sshdr
->asc
= sense_buffer
[2];
2159 sshdr
->ascq
= sense_buffer
[3];
2161 sshdr
->additional_length
= sense_buffer
[7];
2167 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2169 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2170 sb_len
: (sense_buffer
[7] + 8);
2172 sshdr
->asc
= sense_buffer
[12];
2174 sshdr
->ascq
= sense_buffer
[13];
2180 EXPORT_SYMBOL(scsi_normalize_sense
);
2182 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2183 struct scsi_sense_hdr
*sshdr
)
2185 return scsi_normalize_sense(cmd
->sense_buffer
,
2186 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2188 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2191 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2192 * @sense_buffer: byte array of descriptor format sense data
2193 * @sb_len: number of valid bytes in sense_buffer
2194 * @desc_type: value of descriptor type to find
2195 * (e.g. 0 -> information)
2198 * only valid when sense data is in descriptor format
2201 * pointer to start of (first) descriptor if found else NULL
2203 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2206 int add_sen_len
, add_len
, desc_len
, k
;
2209 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2211 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2213 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2214 add_sen_len
: (sb_len
- 8);
2215 descp
= &sense_buffer
[8];
2216 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2218 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2219 desc_len
= add_len
+ 2;
2220 if (descp
[0] == desc_type
)
2222 if (add_len
< 0) // short descriptor ??
2227 EXPORT_SYMBOL(scsi_sense_desc_find
);
2230 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2231 * @sense_buffer: byte array of sense data
2232 * @sb_len: number of valid bytes in sense_buffer
2233 * @info_out: pointer to 64 integer where 8 or 4 byte information
2234 * field will be placed if found.
2237 * 1 if information field found, 0 if not found.
2239 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2248 switch (sense_buffer
[0] & 0x7f) {
2251 if (sense_buffer
[0] & 0x80) {
2252 *info_out
= (sense_buffer
[3] << 24) +
2253 (sense_buffer
[4] << 16) +
2254 (sense_buffer
[5] << 8) + sense_buffer
[6];
2260 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2262 if (ucp
&& (0xa == ucp
[1])) {
2264 for (j
= 0; j
< 8; ++j
) {
2277 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2280 * scsi_build_sense_buffer - build sense data in a buffer
2281 * @desc: Sense format (non zero == descriptor format,
2282 * 0 == fixed format)
2283 * @buf: Where to build sense data
2285 * @asc: Additional sense code
2286 * @ascq: Additional sense code qualifier
2289 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2292 buf
[0] = 0x72; /* descriptor, current */
2298 buf
[0] = 0x70; /* fixed, current */
2305 EXPORT_SYMBOL(scsi_build_sense_buffer
);