1 // SPDX-License-Identifier: GPL-2.0-only
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
5 * SCSI error/timeout handling
6 * Initial versions: Eric Youngdale. Based upon conversations with
7 * Leonard Zubkoff and David Miller at Linux Expo,
8 * ideas originating from all over the place.
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
18 #include <linux/module.h>
19 #include <linux/sched.h>
20 #include <linux/gfp.h>
21 #include <linux/timer.h>
22 #include <linux/string.h>
23 #include <linux/kernel.h>
24 #include <linux/freezer.h>
25 #include <linux/kthread.h>
26 #include <linux/interrupt.h>
27 #include <linux/blkdev.h>
28 #include <linux/delay.h>
29 #include <linux/jiffies.h>
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_cmnd.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_device.h>
35 #include <scsi/scsi_driver.h>
36 #include <scsi/scsi_eh.h>
37 #include <scsi/scsi_common.h>
38 #include <scsi/scsi_transport.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_ioctl.h>
41 #include <scsi/scsi_dh.h>
42 #include <scsi/scsi_devinfo.h>
45 #include "scsi_priv.h"
46 #include "scsi_logging.h"
47 #include "scsi_transport_api.h"
49 #include <trace/events/scsi.h>
51 #include <asm/unaligned.h>
53 static void scsi_eh_done(struct scsi_cmnd *scmd);
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
59 #define BUS_RESET_SETTLE_TIME (10)
60 #define HOST_RESET_SETTLE_TIME (10)
62 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
63 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
66 void scsi_eh_wakeup(struct Scsi_Host *shost)
68 lockdep_assert_held(shost->host_lock);
70 if (scsi_host_busy(shost) == shost->host_failed) {
71 trace_scsi_eh_wakeup(shost);
72 wake_up_process(shost->ehandler);
73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74 "Waking error handler thread\n"));
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
82 * Schedule SCSI EH without scmd.
84 void scsi_schedule_eh(struct Scsi_Host *shost)
88 spin_lock_irqsave(shost->host_lock, flags);
90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
92 shost->host_eh_scheduled++;
93 scsi_eh_wakeup(shost);
96 spin_unlock_irqrestore(shost->host_lock, flags);
98 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
100 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
102 if (!shost->last_reset || shost->eh_deadline == -1)
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
109 * if eh_deadline has been set to 'off' during the
112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
113 shost->eh_deadline > -1)
119 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
121 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
124 return ++cmd->retries <= cmd->allowed;
127 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
129 struct scsi_device *sdev = cmd->device;
130 struct Scsi_Host *host = sdev->host;
132 if (host->hostt->eh_should_retry_cmd)
133 return host->hostt->eh_should_retry_cmd(cmd);
138 static void scsi_eh_complete_abort(struct scsi_cmnd *scmd, struct Scsi_Host *shost)
142 spin_lock_irqsave(shost->host_lock, flags);
143 list_del_init(&scmd->eh_entry);
145 * If the abort succeeds, and there is no further
146 * EH action, clear the ->last_reset time.
148 if (list_empty(&shost->eh_abort_list) &&
149 list_empty(&shost->eh_cmd_q))
150 if (shost->eh_deadline != -1)
151 shost->last_reset = 0;
152 spin_unlock_irqrestore(shost->host_lock, flags);
156 * scmd_eh_abort_handler - Handle command aborts
157 * @work: command to be aborted.
159 * Note: this function must be called only for a command that has timed out.
160 * Because the block layer marks a request as complete before it calls
161 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
162 * timed out do not have any effect. Hence it is safe to call
163 * scsi_finish_command() from this function.
166 scmd_eh_abort_handler(struct work_struct *work)
168 struct scsi_cmnd *scmd =
169 container_of(work, struct scsi_cmnd, abort_work.work);
170 struct scsi_device *sdev = scmd->device;
171 enum scsi_disposition rtn;
174 if (scsi_host_eh_past_deadline(sdev->host)) {
175 SCSI_LOG_ERROR_RECOVERY(3,
176 scmd_printk(KERN_INFO, scmd,
177 "eh timeout, not aborting\n"));
179 SCSI_LOG_ERROR_RECOVERY(3,
180 scmd_printk(KERN_INFO, scmd,
181 "aborting command\n"));
182 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
183 if (rtn == SUCCESS) {
184 set_host_byte(scmd, DID_TIME_OUT);
185 if (scsi_host_eh_past_deadline(sdev->host)) {
186 SCSI_LOG_ERROR_RECOVERY(3,
187 scmd_printk(KERN_INFO, scmd,
188 "eh timeout, not retrying "
189 "aborted command\n"));
190 } else if (!scsi_noretry_cmd(scmd) &&
191 scsi_cmd_retry_allowed(scmd) &&
192 scsi_eh_should_retry_cmd(scmd)) {
193 SCSI_LOG_ERROR_RECOVERY(3,
194 scmd_printk(KERN_WARNING, scmd,
195 "retry aborted command\n"));
196 scsi_eh_complete_abort(scmd, sdev->host);
197 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
200 SCSI_LOG_ERROR_RECOVERY(3,
201 scmd_printk(KERN_WARNING, scmd,
202 "finish aborted command\n"));
203 scsi_eh_complete_abort(scmd, sdev->host);
204 scsi_finish_command(scmd);
208 SCSI_LOG_ERROR_RECOVERY(3,
209 scmd_printk(KERN_INFO, scmd,
211 (rtn == FAST_IO_FAIL) ?
212 "not send" : "failed"));
216 spin_lock_irqsave(sdev->host->host_lock, flags);
217 list_del_init(&scmd->eh_entry);
218 spin_unlock_irqrestore(sdev->host->host_lock, flags);
219 scsi_eh_scmd_add(scmd);
223 * scsi_abort_command - schedule a command abort
224 * @scmd: scmd to abort.
226 * We only need to abort commands after a command timeout
229 scsi_abort_command(struct scsi_cmnd *scmd)
231 struct scsi_device *sdev = scmd->device;
232 struct Scsi_Host *shost = sdev->host;
235 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
237 * Retry after abort failed, escalate to next level.
239 SCSI_LOG_ERROR_RECOVERY(3,
240 scmd_printk(KERN_INFO, scmd,
241 "previous abort failed\n"));
242 BUG_ON(delayed_work_pending(&scmd->abort_work));
246 spin_lock_irqsave(shost->host_lock, flags);
247 if (shost->eh_deadline != -1 && !shost->last_reset)
248 shost->last_reset = jiffies;
249 BUG_ON(!list_empty(&scmd->eh_entry));
250 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list);
251 spin_unlock_irqrestore(shost->host_lock, flags);
253 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
254 SCSI_LOG_ERROR_RECOVERY(3,
255 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
256 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
261 * scsi_eh_reset - call into ->eh_action to reset internal counters
262 * @scmd: scmd to run eh on.
264 * The scsi driver might be carrying internal state about the
265 * devices, so we need to call into the driver to reset the
266 * internal state once the error handler is started.
268 static void scsi_eh_reset(struct scsi_cmnd *scmd)
270 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
271 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
273 sdrv->eh_reset(scmd);
277 static void scsi_eh_inc_host_failed(struct rcu_head *head)
279 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
280 struct Scsi_Host *shost = scmd->device->host;
283 spin_lock_irqsave(shost->host_lock, flags);
284 shost->host_failed++;
285 scsi_eh_wakeup(shost);
286 spin_unlock_irqrestore(shost->host_lock, flags);
290 * scsi_eh_scmd_add - add scsi cmd to error handling.
291 * @scmd: scmd to run eh on.
293 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
295 struct Scsi_Host *shost = scmd->device->host;
299 WARN_ON_ONCE(!shost->ehandler);
301 spin_lock_irqsave(shost->host_lock, flags);
302 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
303 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
306 if (shost->eh_deadline != -1 && !shost->last_reset)
307 shost->last_reset = jiffies;
310 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
311 spin_unlock_irqrestore(shost->host_lock, flags);
313 * Ensure that all tasks observe the host state change before the
314 * host_failed change.
316 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
320 * scsi_times_out - Timeout function for normal scsi commands.
321 * @req: request that is timing out.
324 * We do not need to lock this. There is the potential for a race
325 * only in that the normal completion handling might run, but if the
326 * normal completion function determines that the timer has already
327 * fired, then it mustn't do anything.
329 enum blk_eh_timer_return scsi_times_out(struct request *req)
331 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
332 enum blk_eh_timer_return rtn = BLK_EH_DONE;
333 struct Scsi_Host *host = scmd->device->host;
335 trace_scsi_dispatch_cmd_timeout(scmd);
336 scsi_log_completion(scmd, TIMEOUT_ERROR);
338 if (host->eh_deadline != -1 && !host->last_reset)
339 host->last_reset = jiffies;
341 if (host->hostt->eh_timed_out)
342 rtn = host->hostt->eh_timed_out(scmd);
344 if (rtn == BLK_EH_DONE) {
346 * Set the command to complete first in order to prevent a real
347 * completion from releasing the command while error handling
348 * is using it. If the command was already completed, then the
349 * lower level driver beat the timeout handler, and it is safe
350 * to return without escalating error recovery.
352 * If timeout handling lost the race to a real completion, the
353 * block layer may ignore that due to a fake timeout injection,
354 * so return RESET_TIMER to allow error handling another shot
357 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
358 return BLK_EH_RESET_TIMER;
359 if (scsi_abort_command(scmd) != SUCCESS) {
360 set_host_byte(scmd, DID_TIME_OUT);
361 scsi_eh_scmd_add(scmd);
369 * scsi_block_when_processing_errors - Prevent cmds from being queued.
370 * @sdev: Device on which we are performing recovery.
373 * We block until the host is out of error recovery, and then check to
374 * see whether the host or the device is offline.
377 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
379 int scsi_block_when_processing_errors(struct scsi_device *sdev)
383 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
385 online = scsi_device_online(sdev);
389 EXPORT_SYMBOL(scsi_block_when_processing_errors);
391 #ifdef CONFIG_SCSI_LOGGING
393 * scsi_eh_prt_fail_stats - Log info on failures.
394 * @shost: scsi host being recovered.
395 * @work_q: Queue of scsi cmds to process.
397 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
398 struct list_head *work_q)
400 struct scsi_cmnd *scmd;
401 struct scsi_device *sdev;
402 int total_failures = 0;
405 int devices_failed = 0;
407 shost_for_each_device(sdev, shost) {
408 list_for_each_entry(scmd, work_q, eh_entry) {
409 if (scmd->device == sdev) {
411 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
418 if (cmd_cancel || cmd_failed) {
419 SCSI_LOG_ERROR_RECOVERY(3,
420 shost_printk(KERN_INFO, shost,
421 "%s: cmds failed: %d, cancel: %d\n",
422 __func__, cmd_failed,
430 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
431 "Total of %d commands on %d"
432 " devices require eh work\n",
433 total_failures, devices_failed));
438 * scsi_report_lun_change - Set flag on all *other* devices on the same target
439 * to indicate that a UNIT ATTENTION is expected.
440 * @sdev: Device reporting the UNIT ATTENTION
442 static void scsi_report_lun_change(struct scsi_device *sdev)
444 sdev->sdev_target->expecting_lun_change = 1;
448 * scsi_report_sense - Examine scsi sense information and log messages for
449 * certain conditions, also issue uevents for some of them.
450 * @sdev: Device reporting the sense code
451 * @sshdr: sshdr to be examined
453 static void scsi_report_sense(struct scsi_device *sdev,
454 struct scsi_sense_hdr *sshdr)
456 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
458 if (sshdr->sense_key == UNIT_ATTENTION) {
459 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
460 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
461 sdev_printk(KERN_WARNING, sdev,
462 "Inquiry data has changed");
463 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
464 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
465 scsi_report_lun_change(sdev);
466 sdev_printk(KERN_WARNING, sdev,
467 "Warning! Received an indication that the "
468 "LUN assignments on this target have "
469 "changed. The Linux SCSI layer does not "
470 "automatically remap LUN assignments.\n");
471 } else if (sshdr->asc == 0x3f)
472 sdev_printk(KERN_WARNING, sdev,
473 "Warning! Received an indication that the "
474 "operating parameters on this target have "
475 "changed. The Linux SCSI layer does not "
476 "automatically adjust these parameters.\n");
478 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
479 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
480 sdev_printk(KERN_WARNING, sdev,
481 "Warning! Received an indication that the "
482 "LUN reached a thin provisioning soft "
486 if (sshdr->asc == 0x29) {
487 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
489 * Do not print message if it is an expected side-effect
492 if (!sdev->silence_suspend)
493 sdev_printk(KERN_WARNING, sdev,
494 "Power-on or device reset occurred\n");
497 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
498 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
499 sdev_printk(KERN_WARNING, sdev,
500 "Mode parameters changed");
501 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
502 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
503 sdev_printk(KERN_WARNING, sdev,
504 "Asymmetric access state changed");
505 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
506 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
507 sdev_printk(KERN_WARNING, sdev,
508 "Capacity data has changed");
509 } else if (sshdr->asc == 0x2a)
510 sdev_printk(KERN_WARNING, sdev,
511 "Parameters changed");
514 if (evt_type != SDEV_EVT_MAXBITS) {
515 set_bit(evt_type, sdev->pending_events);
516 schedule_work(&sdev->event_work);
521 * scsi_check_sense - Examine scsi cmd sense
522 * @scmd: Cmd to have sense checked.
525 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
528 * When a deferred error is detected the current command has
529 * not been executed and needs retrying.
531 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
533 struct scsi_device *sdev = scmd->device;
534 struct scsi_sense_hdr sshdr;
536 if (! scsi_command_normalize_sense(scmd, &sshdr))
537 return FAILED; /* no valid sense data */
539 scsi_report_sense(sdev, &sshdr);
541 if (scsi_sense_is_deferred(&sshdr))
544 if (sdev->handler && sdev->handler->check_sense) {
545 enum scsi_disposition rc;
547 rc = sdev->handler->check_sense(sdev, &sshdr);
548 if (rc != SCSI_RETURN_NOT_HANDLED)
550 /* handler does not care. Drop down to default handling */
553 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
555 * nasty: for mid-layer issued TURs, we need to return the
556 * actual sense data without any recovery attempt. For eh
557 * issued ones, we need to try to recover and interpret
562 * Previous logic looked for FILEMARK, EOM or ILI which are
563 * mainly associated with tapes and returned SUCCESS.
565 if (sshdr.response_code == 0x70) {
567 if (scmd->sense_buffer[2] & 0xe0)
571 * descriptor format: look for "stream commands sense data
572 * descriptor" (see SSC-3). Assume single sense data
573 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
575 if ((sshdr.additional_length > 3) &&
576 (scmd->sense_buffer[8] == 0x4) &&
577 (scmd->sense_buffer[11] & 0xe0))
581 switch (sshdr.sense_key) {
584 case RECOVERED_ERROR:
585 return /* soft_error */ SUCCESS;
587 case ABORTED_COMMAND:
588 if (sshdr.asc == 0x10) /* DIF */
591 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
592 return ADD_TO_MLQUEUE;
593 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
594 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
595 return ADD_TO_MLQUEUE;
601 * if we are expecting a cc/ua because of a bus reset that we
602 * performed, treat this just as a retry. otherwise this is
603 * information that we should pass up to the upper-level driver
604 * so that we can deal with it there.
606 if (scmd->device->expecting_cc_ua) {
608 * Because some device does not queue unit
609 * attentions correctly, we carefully check
610 * additional sense code and qualifier so as
611 * not to squash media change unit attention.
613 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
614 scmd->device->expecting_cc_ua = 0;
619 * we might also expect a cc/ua if another LUN on the target
620 * reported a UA with an ASC/ASCQ of 3F 0E -
621 * REPORTED LUNS DATA HAS CHANGED.
623 if (scmd->device->sdev_target->expecting_lun_change &&
624 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
627 * if the device is in the process of becoming ready, we
630 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
633 * if the device is not started, we need to wake
634 * the error handler to start the motor
636 if (scmd->device->allow_restart &&
637 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
640 * Pass the UA upwards for a determination in the completion
645 /* these are not supported */
647 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
648 /* Thin provisioning hard threshold reached */
649 set_host_byte(scmd, DID_ALLOC_FAILURE);
654 case VOLUME_OVERFLOW:
657 set_host_byte(scmd, DID_TARGET_FAILURE);
661 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
662 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
663 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
664 set_host_byte(scmd, DID_MEDIUM_ERROR);
670 if (scmd->device->retry_hwerror)
671 return ADD_TO_MLQUEUE;
673 set_host_byte(scmd, DID_TARGET_FAILURE);
676 case ILLEGAL_REQUEST:
677 if (sshdr.asc == 0x20 || /* Invalid command operation code */
678 sshdr.asc == 0x21 || /* Logical block address out of range */
679 sshdr.asc == 0x22 || /* Invalid function */
680 sshdr.asc == 0x24 || /* Invalid field in cdb */
681 sshdr.asc == 0x26 || /* Parameter value invalid */
682 sshdr.asc == 0x27) { /* Write protected */
683 set_host_byte(scmd, DID_TARGET_FAILURE);
691 EXPORT_SYMBOL_GPL(scsi_check_sense);
693 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
695 struct scsi_host_template *sht = sdev->host->hostt;
696 struct scsi_device *tmp_sdev;
698 if (!sht->track_queue_depth ||
699 sdev->queue_depth >= sdev->max_queue_depth)
702 if (time_before(jiffies,
703 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
706 if (time_before(jiffies,
707 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
711 * Walk all devices of a target and do
714 shost_for_each_device(tmp_sdev, sdev->host) {
715 if (tmp_sdev->channel != sdev->channel ||
716 tmp_sdev->id != sdev->id ||
717 tmp_sdev->queue_depth == sdev->max_queue_depth)
720 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
721 sdev->last_queue_ramp_up = jiffies;
725 static void scsi_handle_queue_full(struct scsi_device *sdev)
727 struct scsi_host_template *sht = sdev->host->hostt;
728 struct scsi_device *tmp_sdev;
730 if (!sht->track_queue_depth)
733 shost_for_each_device(tmp_sdev, sdev->host) {
734 if (tmp_sdev->channel != sdev->channel ||
735 tmp_sdev->id != sdev->id)
738 * We do not know the number of commands that were at
739 * the device when we got the queue full so we start
740 * from the highest possible value and work our way down.
742 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
747 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
748 * @scmd: SCSI cmd to examine.
751 * This is *only* called when we are examining the status of commands
752 * queued during error recovery. the main difference here is that we
753 * don't allow for the possibility of retries here, and we are a lot
754 * more restrictive about what we consider acceptable.
756 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
759 * first check the host byte, to see if there is anything in there
760 * that would indicate what we need to do.
762 if (host_byte(scmd->result) == DID_RESET) {
764 * rats. we are already in the error handler, so we now
765 * get to try and figure out what to do next. if the sense
766 * is valid, we have a pretty good idea of what to do.
767 * if not, we mark it as FAILED.
769 return scsi_check_sense(scmd);
771 if (host_byte(scmd->result) != DID_OK)
775 * now, check the status byte to see if this indicates
778 switch (get_status_byte(scmd)) {
780 scsi_handle_queue_ramp_up(scmd->device);
782 case SAM_STAT_COMMAND_TERMINATED:
784 case SAM_STAT_CHECK_CONDITION:
785 return scsi_check_sense(scmd);
786 case SAM_STAT_CONDITION_MET:
787 case SAM_STAT_INTERMEDIATE:
788 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
790 * who knows? FIXME(eric)
793 case SAM_STAT_RESERVATION_CONFLICT:
794 if (scmd->cmnd[0] == TEST_UNIT_READY)
795 /* it is a success, we probed the device and
798 /* otherwise, we failed to send the command */
800 case SAM_STAT_TASK_SET_FULL:
801 scsi_handle_queue_full(scmd->device);
812 * scsi_eh_done - Completion function for error handling.
813 * @scmd: Cmd that is done.
815 static void scsi_eh_done(struct scsi_cmnd *scmd)
817 struct completion *eh_action;
819 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
820 "%s result: %x\n", __func__, scmd->result));
822 eh_action = scmd->device->host->eh_action;
828 * scsi_try_host_reset - ask host adapter to reset itself
829 * @scmd: SCSI cmd to send host reset.
831 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
834 enum scsi_disposition rtn;
835 struct Scsi_Host *host = scmd->device->host;
836 struct scsi_host_template *hostt = host->hostt;
838 SCSI_LOG_ERROR_RECOVERY(3,
839 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
841 if (!hostt->eh_host_reset_handler)
844 rtn = hostt->eh_host_reset_handler(scmd);
846 if (rtn == SUCCESS) {
847 if (!hostt->skip_settle_delay)
848 ssleep(HOST_RESET_SETTLE_TIME);
849 spin_lock_irqsave(host->host_lock, flags);
850 scsi_report_bus_reset(host, scmd_channel(scmd));
851 spin_unlock_irqrestore(host->host_lock, flags);
858 * scsi_try_bus_reset - ask host to perform a bus reset
859 * @scmd: SCSI cmd to send bus reset.
861 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
864 enum scsi_disposition rtn;
865 struct Scsi_Host *host = scmd->device->host;
866 struct scsi_host_template *hostt = host->hostt;
868 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
869 "%s: Snd Bus RST\n", __func__));
871 if (!hostt->eh_bus_reset_handler)
874 rtn = hostt->eh_bus_reset_handler(scmd);
876 if (rtn == SUCCESS) {
877 if (!hostt->skip_settle_delay)
878 ssleep(BUS_RESET_SETTLE_TIME);
879 spin_lock_irqsave(host->host_lock, flags);
880 scsi_report_bus_reset(host, scmd_channel(scmd));
881 spin_unlock_irqrestore(host->host_lock, flags);
887 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
890 sdev->expecting_cc_ua = 1;
894 * scsi_try_target_reset - Ask host to perform a target reset
895 * @scmd: SCSI cmd used to send a target reset
898 * There is no timeout for this operation. if this operation is
899 * unreliable for a given host, then the host itself needs to put a
900 * timer on it, and set the host back to a consistent state prior to
903 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
906 enum scsi_disposition rtn;
907 struct Scsi_Host *host = scmd->device->host;
908 struct scsi_host_template *hostt = host->hostt;
910 if (!hostt->eh_target_reset_handler)
913 rtn = hostt->eh_target_reset_handler(scmd);
914 if (rtn == SUCCESS) {
915 spin_lock_irqsave(host->host_lock, flags);
916 __starget_for_each_device(scsi_target(scmd->device), NULL,
917 __scsi_report_device_reset);
918 spin_unlock_irqrestore(host->host_lock, flags);
925 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
926 * @scmd: SCSI cmd used to send BDR
929 * There is no timeout for this operation. if this operation is
930 * unreliable for a given host, then the host itself needs to put a
931 * timer on it, and set the host back to a consistent state prior to
934 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
936 enum scsi_disposition rtn;
937 struct scsi_host_template *hostt = scmd->device->host->hostt;
939 if (!hostt->eh_device_reset_handler)
942 rtn = hostt->eh_device_reset_handler(scmd);
944 __scsi_report_device_reset(scmd->device, NULL);
949 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
950 * @hostt: SCSI driver host template
951 * @scmd: SCSI cmd used to send a target reset
954 * SUCCESS, FAILED, or FAST_IO_FAIL
957 * SUCCESS does not necessarily indicate that the command
958 * has been aborted; it only indicates that the LLDDs
959 * has cleared all references to that command.
960 * LLDDs should return FAILED only if an abort was required
961 * but could not be executed. LLDDs should return FAST_IO_FAIL
962 * if the device is temporarily unavailable (eg due to a
963 * link down on FibreChannel)
965 static enum scsi_disposition
966 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
968 if (!hostt->eh_abort_handler)
971 return hostt->eh_abort_handler(scmd);
974 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
976 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
977 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
978 if (scsi_try_target_reset(scmd) != SUCCESS)
979 if (scsi_try_bus_reset(scmd) != SUCCESS)
980 scsi_try_host_reset(scmd);
984 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
985 * @scmd: SCSI command structure to hijack
986 * @ses: structure to save restore information
987 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
988 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
989 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
991 * This function is used to save a scsi command information before re-execution
992 * as part of the error recovery process. If @sense_bytes is 0 the command
993 * sent must be one that does not transfer any data. If @sense_bytes != 0
994 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
995 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
997 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
998 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1000 struct scsi_device *sdev = scmd->device;
1003 * We need saved copies of a number of fields - this is because
1004 * error handling may need to overwrite these with different values
1005 * to run different commands, and once error handling is complete,
1006 * we will need to restore these values prior to running the actual
1009 ses->cmd_len = scmd->cmd_len;
1010 ses->cmnd = scmd->cmnd;
1011 ses->data_direction = scmd->sc_data_direction;
1012 ses->sdb = scmd->sdb;
1013 ses->result = scmd->result;
1014 ses->resid_len = scmd->req.resid_len;
1015 ses->underflow = scmd->underflow;
1016 ses->prot_op = scmd->prot_op;
1017 ses->eh_eflags = scmd->eh_eflags;
1019 scmd->prot_op = SCSI_PROT_NORMAL;
1020 scmd->eh_eflags = 0;
1021 scmd->cmnd = ses->eh_cmnd;
1022 memset(scmd->cmnd, 0, BLK_MAX_CDB);
1023 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1025 scmd->req.resid_len = 0;
1028 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1030 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1032 scmd->sdb.table.sgl = &ses->sense_sgl;
1033 scmd->sc_data_direction = DMA_FROM_DEVICE;
1034 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1035 scmd->cmnd[0] = REQUEST_SENSE;
1036 scmd->cmnd[4] = scmd->sdb.length;
1037 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1039 scmd->sc_data_direction = DMA_NONE;
1041 BUG_ON(cmnd_size > BLK_MAX_CDB);
1042 memcpy(scmd->cmnd, cmnd, cmnd_size);
1043 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1047 scmd->underflow = 0;
1049 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1050 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1051 (sdev->lun << 5 & 0xe0);
1054 * Zero the sense buffer. The scsi spec mandates that any
1055 * untransferred sense data should be interpreted as being zero.
1057 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1059 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1062 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1063 * @scmd: SCSI command structure to restore
1064 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1066 * Undo any damage done by above scsi_eh_prep_cmnd().
1068 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1071 * Restore original data
1073 scmd->cmd_len = ses->cmd_len;
1074 scmd->cmnd = ses->cmnd;
1075 scmd->sc_data_direction = ses->data_direction;
1076 scmd->sdb = ses->sdb;
1077 scmd->result = ses->result;
1078 scmd->req.resid_len = ses->resid_len;
1079 scmd->underflow = ses->underflow;
1080 scmd->prot_op = ses->prot_op;
1081 scmd->eh_eflags = ses->eh_eflags;
1083 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1086 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1087 * @scmd: SCSI command structure to hijack
1088 * @cmnd: CDB to send
1089 * @cmnd_size: size in bytes of @cmnd
1090 * @timeout: timeout for this request
1091 * @sense_bytes: size of sense data to copy or 0
1093 * This function is used to send a scsi command down to a target device
1094 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1097 * SUCCESS or FAILED or NEEDS_RETRY
1099 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1100 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1102 struct scsi_device *sdev = scmd->device;
1103 struct Scsi_Host *shost = sdev->host;
1104 DECLARE_COMPLETION_ONSTACK(done);
1105 unsigned long timeleft = timeout, delay;
1106 struct scsi_eh_save ses;
1107 const unsigned long stall_for = msecs_to_jiffies(100);
1111 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1112 shost->eh_action = &done;
1114 scsi_log_send(scmd);
1115 scmd->scsi_done = scsi_eh_done;
1118 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1119 * change the SCSI device state after we have examined it and before
1120 * .queuecommand() is called.
1122 mutex_lock(&sdev->state_mutex);
1123 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1124 mutex_unlock(&sdev->state_mutex);
1125 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1126 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1128 delay = min(timeleft, stall_for);
1130 msleep(jiffies_to_msecs(delay));
1131 mutex_lock(&sdev->state_mutex);
1133 if (sdev->sdev_state != SDEV_BLOCK)
1134 rtn = shost->hostt->queuecommand(shost, scmd);
1137 mutex_unlock(&sdev->state_mutex);
1140 if (timeleft > stall_for) {
1141 scsi_eh_restore_cmnd(scmd, &ses);
1142 timeleft -= stall_for;
1143 msleep(jiffies_to_msecs(stall_for));
1146 /* signal not to enter either branch of the if () below */
1150 timeleft = wait_for_completion_timeout(&done, timeout);
1154 shost->eh_action = NULL;
1156 scsi_log_completion(scmd, rtn);
1158 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1159 "%s timeleft: %ld\n",
1160 __func__, timeleft));
1163 * If there is time left scsi_eh_done got called, and we will examine
1164 * the actual status codes to see whether the command actually did
1165 * complete normally, else if we have a zero return and no time left,
1166 * the command must still be pending, so abort it and return FAILED.
1167 * If we never actually managed to issue the command, because
1168 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1169 * value above (so don't execute either branch of the if)
1172 rtn = scsi_eh_completed_normally(scmd);
1173 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1174 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1181 case ADD_TO_MLQUEUE:
1188 } else if (rtn != FAILED) {
1189 scsi_abort_eh_cmnd(scmd);
1193 scsi_eh_restore_cmnd(scmd, &ses);
1199 * scsi_request_sense - Request sense data from a particular target.
1200 * @scmd: SCSI cmd for request sense.
1203 * Some hosts automatically obtain this information, others require
1204 * that we obtain it on our own. This function will *not* return until
1205 * the command either times out, or it completes.
1207 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1209 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1212 static enum scsi_disposition
1213 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1215 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
1216 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1217 if (sdrv->eh_action)
1218 rtn = sdrv->eh_action(scmd, rtn);
1224 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1225 * @scmd: Original SCSI cmd that eh has finished.
1226 * @done_q: Queue for processed commands.
1229 * We don't want to use the normal command completion while we are are
1230 * still handling errors - it may cause other commands to be queued,
1231 * and that would disturb what we are doing. Thus we really want to
1232 * keep a list of pending commands for final completion, and once we
1233 * are ready to leave error handling we handle completion for real.
1235 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1237 list_move_tail(&scmd->eh_entry, done_q);
1239 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1242 * scsi_eh_get_sense - Get device sense data.
1243 * @work_q: Queue of commands to process.
1244 * @done_q: Queue of processed commands.
1247 * See if we need to request sense information. if so, then get it
1248 * now, so we have a better idea of what to do.
1251 * This has the unfortunate side effect that if a shost adapter does
1252 * not automatically request sense information, we end up shutting
1253 * it down before we request it.
1255 * All drivers should request sense information internally these days,
1256 * so for now all I have to say is tough noogies if you end up in here.
1258 * XXX: Long term this code should go away, but that needs an audit of
1261 int scsi_eh_get_sense(struct list_head *work_q,
1262 struct list_head *done_q)
1264 struct scsi_cmnd *scmd, *next;
1265 struct Scsi_Host *shost;
1266 enum scsi_disposition rtn;
1269 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1270 * should not get sense.
1272 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1273 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1274 SCSI_SENSE_VALID(scmd))
1277 shost = scmd->device->host;
1278 if (scsi_host_eh_past_deadline(shost)) {
1279 SCSI_LOG_ERROR_RECOVERY(3,
1280 scmd_printk(KERN_INFO, scmd,
1281 "%s: skip request sense, past eh deadline\n",
1285 if (!scsi_status_is_check_condition(scmd->result))
1287 * don't request sense if there's no check condition
1288 * status because the error we're processing isn't one
1289 * that has a sense code (and some devices get
1290 * confused by sense requests out of the blue)
1294 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1295 "%s: requesting sense\n",
1297 rtn = scsi_request_sense(scmd);
1301 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1302 "sense requested, result %x\n", scmd->result));
1303 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1305 rtn = scsi_decide_disposition(scmd);
1308 * if the result was normal, then just pass it along to the
1313 * We don't want this command reissued, just finished
1314 * with the sense data, so set retries to the max
1315 * allowed to ensure it won't get reissued. If the user
1316 * has requested infinite retries, we also want to
1317 * finish this command, so force completion by setting
1318 * retries and allowed to the same value.
1320 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1321 scmd->retries = scmd->allowed = 1;
1323 scmd->retries = scmd->allowed;
1324 else if (rtn != NEEDS_RETRY)
1327 scsi_eh_finish_cmd(scmd, done_q);
1330 return list_empty(work_q);
1332 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1335 * scsi_eh_tur - Send TUR to device.
1336 * @scmd: &scsi_cmnd to send TUR
1339 * 0 - Device is ready. 1 - Device NOT ready.
1341 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1343 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1345 enum scsi_disposition rtn;
1348 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1349 scmd->device->eh_timeout, 0);
1351 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1352 "%s return: %x\n", __func__, rtn));
1367 * scsi_eh_test_devices - check if devices are responding from error recovery.
1368 * @cmd_list: scsi commands in error recovery.
1369 * @work_q: queue for commands which still need more error recovery
1370 * @done_q: queue for commands which are finished
1371 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1374 * Tests if devices are in a working state. Commands to devices now in
1375 * a working state are sent to the done_q while commands to devices which
1376 * are still failing to respond are returned to the work_q for more
1379 static int scsi_eh_test_devices(struct list_head *cmd_list,
1380 struct list_head *work_q,
1381 struct list_head *done_q, int try_stu)
1383 struct scsi_cmnd *scmd, *next;
1384 struct scsi_device *sdev;
1387 while (!list_empty(cmd_list)) {
1388 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1389 sdev = scmd->device;
1392 if (scsi_host_eh_past_deadline(sdev->host)) {
1393 /* Push items back onto work_q */
1394 list_splice_init(cmd_list, work_q);
1395 SCSI_LOG_ERROR_RECOVERY(3,
1396 sdev_printk(KERN_INFO, sdev,
1397 "%s: skip test device, past eh deadline",
1403 finish_cmds = !scsi_device_online(scmd->device) ||
1404 (try_stu && !scsi_eh_try_stu(scmd) &&
1405 !scsi_eh_tur(scmd)) ||
1408 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1409 if (scmd->device == sdev) {
1412 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1413 scsi_eh_finish_cmd(scmd, done_q);
1415 list_move_tail(&scmd->eh_entry, work_q);
1418 return list_empty(work_q);
1422 * scsi_eh_try_stu - Send START_UNIT to device.
1423 * @scmd: &scsi_cmnd to send START_UNIT
1426 * 0 - Device is ready. 1 - Device NOT ready.
1428 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1430 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1432 if (scmd->device->allow_restart) {
1434 enum scsi_disposition rtn = NEEDS_RETRY;
1436 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1437 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1447 * scsi_eh_stu - send START_UNIT if needed
1448 * @shost: &scsi host being recovered.
1449 * @work_q: &list_head for pending commands.
1450 * @done_q: &list_head for processed commands.
1453 * If commands are failing due to not ready, initializing command required,
1454 * try revalidating the device, which will end up sending a start unit.
1456 static int scsi_eh_stu(struct Scsi_Host *shost,
1457 struct list_head *work_q,
1458 struct list_head *done_q)
1460 struct scsi_cmnd *scmd, *stu_scmd, *next;
1461 struct scsi_device *sdev;
1463 shost_for_each_device(sdev, shost) {
1464 if (scsi_host_eh_past_deadline(shost)) {
1465 SCSI_LOG_ERROR_RECOVERY(3,
1466 sdev_printk(KERN_INFO, sdev,
1467 "%s: skip START_UNIT, past eh deadline\n",
1469 scsi_device_put(sdev);
1473 list_for_each_entry(scmd, work_q, eh_entry)
1474 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1475 scsi_check_sense(scmd) == FAILED ) {
1483 SCSI_LOG_ERROR_RECOVERY(3,
1484 sdev_printk(KERN_INFO, sdev,
1485 "%s: Sending START_UNIT\n",
1488 if (!scsi_eh_try_stu(stu_scmd)) {
1489 if (!scsi_device_online(sdev) ||
1490 !scsi_eh_tur(stu_scmd)) {
1491 list_for_each_entry_safe(scmd, next,
1493 if (scmd->device == sdev &&
1494 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1495 scsi_eh_finish_cmd(scmd, done_q);
1499 SCSI_LOG_ERROR_RECOVERY(3,
1500 sdev_printk(KERN_INFO, sdev,
1501 "%s: START_UNIT failed\n",
1506 return list_empty(work_q);
1511 * scsi_eh_bus_device_reset - send bdr if needed
1512 * @shost: scsi host being recovered.
1513 * @work_q: &list_head for pending commands.
1514 * @done_q: &list_head for processed commands.
1517 * Try a bus device reset. Still, look to see whether we have multiple
1518 * devices that are jammed or not - if we have multiple devices, it
1519 * makes no sense to try bus_device_reset - we really would need to try
1520 * a bus_reset instead.
1522 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1523 struct list_head *work_q,
1524 struct list_head *done_q)
1526 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1527 struct scsi_device *sdev;
1528 enum scsi_disposition rtn;
1530 shost_for_each_device(sdev, shost) {
1531 if (scsi_host_eh_past_deadline(shost)) {
1532 SCSI_LOG_ERROR_RECOVERY(3,
1533 sdev_printk(KERN_INFO, sdev,
1534 "%s: skip BDR, past eh deadline\n",
1536 scsi_device_put(sdev);
1540 list_for_each_entry(scmd, work_q, eh_entry)
1541 if (scmd->device == sdev) {
1549 SCSI_LOG_ERROR_RECOVERY(3,
1550 sdev_printk(KERN_INFO, sdev,
1551 "%s: Sending BDR\n", current->comm));
1552 rtn = scsi_try_bus_device_reset(bdr_scmd);
1553 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1554 if (!scsi_device_online(sdev) ||
1555 rtn == FAST_IO_FAIL ||
1556 !scsi_eh_tur(bdr_scmd)) {
1557 list_for_each_entry_safe(scmd, next,
1559 if (scmd->device == sdev &&
1560 scsi_eh_action(scmd, rtn) != FAILED)
1561 scsi_eh_finish_cmd(scmd,
1566 SCSI_LOG_ERROR_RECOVERY(3,
1567 sdev_printk(KERN_INFO, sdev,
1568 "%s: BDR failed\n", current->comm));
1572 return list_empty(work_q);
1576 * scsi_eh_target_reset - send target reset if needed
1577 * @shost: scsi host being recovered.
1578 * @work_q: &list_head for pending commands.
1579 * @done_q: &list_head for processed commands.
1582 * Try a target reset.
1584 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1585 struct list_head *work_q,
1586 struct list_head *done_q)
1588 LIST_HEAD(tmp_list);
1589 LIST_HEAD(check_list);
1591 list_splice_init(work_q, &tmp_list);
1593 while (!list_empty(&tmp_list)) {
1594 struct scsi_cmnd *next, *scmd;
1595 enum scsi_disposition rtn;
1598 if (scsi_host_eh_past_deadline(shost)) {
1599 /* push back on work queue for further processing */
1600 list_splice_init(&check_list, work_q);
1601 list_splice_init(&tmp_list, work_q);
1602 SCSI_LOG_ERROR_RECOVERY(3,
1603 shost_printk(KERN_INFO, shost,
1604 "%s: Skip target reset, past eh deadline\n",
1606 return list_empty(work_q);
1609 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1612 SCSI_LOG_ERROR_RECOVERY(3,
1613 shost_printk(KERN_INFO, shost,
1614 "%s: Sending target reset to target %d\n",
1615 current->comm, id));
1616 rtn = scsi_try_target_reset(scmd);
1617 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1618 SCSI_LOG_ERROR_RECOVERY(3,
1619 shost_printk(KERN_INFO, shost,
1620 "%s: Target reset failed"
1622 current->comm, id));
1623 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1624 if (scmd_id(scmd) != id)
1628 list_move_tail(&scmd->eh_entry, &check_list);
1629 else if (rtn == FAST_IO_FAIL)
1630 scsi_eh_finish_cmd(scmd, done_q);
1632 /* push back on work queue for further processing */
1633 list_move(&scmd->eh_entry, work_q);
1637 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1641 * scsi_eh_bus_reset - send a bus reset
1642 * @shost: &scsi host being recovered.
1643 * @work_q: &list_head for pending commands.
1644 * @done_q: &list_head for processed commands.
1646 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1647 struct list_head *work_q,
1648 struct list_head *done_q)
1650 struct scsi_cmnd *scmd, *chan_scmd, *next;
1651 LIST_HEAD(check_list);
1652 unsigned int channel;
1653 enum scsi_disposition rtn;
1656 * we really want to loop over the various channels, and do this on
1657 * a channel by channel basis. we should also check to see if any
1658 * of the failed commands are on soft_reset devices, and if so, skip
1662 for (channel = 0; channel <= shost->max_channel; channel++) {
1663 if (scsi_host_eh_past_deadline(shost)) {
1664 list_splice_init(&check_list, work_q);
1665 SCSI_LOG_ERROR_RECOVERY(3,
1666 shost_printk(KERN_INFO, shost,
1667 "%s: skip BRST, past eh deadline\n",
1669 return list_empty(work_q);
1673 list_for_each_entry(scmd, work_q, eh_entry) {
1674 if (channel == scmd_channel(scmd)) {
1678 * FIXME add back in some support for
1679 * soft_reset devices.
1686 SCSI_LOG_ERROR_RECOVERY(3,
1687 shost_printk(KERN_INFO, shost,
1688 "%s: Sending BRST chan: %d\n",
1689 current->comm, channel));
1690 rtn = scsi_try_bus_reset(chan_scmd);
1691 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1692 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1693 if (channel == scmd_channel(scmd)) {
1694 if (rtn == FAST_IO_FAIL)
1695 scsi_eh_finish_cmd(scmd,
1698 list_move_tail(&scmd->eh_entry,
1703 SCSI_LOG_ERROR_RECOVERY(3,
1704 shost_printk(KERN_INFO, shost,
1705 "%s: BRST failed chan: %d\n",
1706 current->comm, channel));
1709 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1713 * scsi_eh_host_reset - send a host reset
1714 * @shost: host to be reset.
1715 * @work_q: &list_head for pending commands.
1716 * @done_q: &list_head for processed commands.
1718 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1719 struct list_head *work_q,
1720 struct list_head *done_q)
1722 struct scsi_cmnd *scmd, *next;
1723 LIST_HEAD(check_list);
1724 enum scsi_disposition rtn;
1726 if (!list_empty(work_q)) {
1727 scmd = list_entry(work_q->next,
1728 struct scsi_cmnd, eh_entry);
1730 SCSI_LOG_ERROR_RECOVERY(3,
1731 shost_printk(KERN_INFO, shost,
1732 "%s: Sending HRST\n",
1735 rtn = scsi_try_host_reset(scmd);
1736 if (rtn == SUCCESS) {
1737 list_splice_init(work_q, &check_list);
1738 } else if (rtn == FAST_IO_FAIL) {
1739 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1740 scsi_eh_finish_cmd(scmd, done_q);
1743 SCSI_LOG_ERROR_RECOVERY(3,
1744 shost_printk(KERN_INFO, shost,
1745 "%s: HRST failed\n",
1749 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1753 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1754 * @work_q: &list_head for pending commands.
1755 * @done_q: &list_head for processed commands.
1757 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1758 struct list_head *done_q)
1760 struct scsi_cmnd *scmd, *next;
1761 struct scsi_device *sdev;
1763 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1764 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1765 "not ready after error recovery\n");
1766 sdev = scmd->device;
1768 mutex_lock(&sdev->state_mutex);
1769 scsi_device_set_state(sdev, SDEV_OFFLINE);
1770 mutex_unlock(&sdev->state_mutex);
1772 scsi_eh_finish_cmd(scmd, done_q);
1778 * scsi_noretry_cmd - determine if command should be failed fast
1779 * @scmd: SCSI cmd to examine.
1781 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1783 struct request *req = scsi_cmd_to_rq(scmd);
1785 switch (host_byte(scmd->result)) {
1791 return req->cmd_flags & REQ_FAILFAST_TRANSPORT;
1793 return req->cmd_flags & REQ_FAILFAST_DEV;
1795 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1798 case DID_SOFT_ERROR:
1799 return req->cmd_flags & REQ_FAILFAST_DRIVER;
1802 if (!scsi_status_is_check_condition(scmd->result))
1807 * assume caller has checked sense and determined
1808 * the check condition was retryable.
1810 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
1817 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1818 * @scmd: SCSI cmd to examine.
1821 * This is *only* called when we are examining the status after sending
1822 * out the actual data command. any commands that are queued for error
1823 * recovery (e.g. test_unit_ready) do *not* come through here.
1825 * When this routine returns failed, it means the error handler thread
1826 * is woken. In cases where the error code indicates an error that
1827 * doesn't require the error handler read (i.e. we don't need to
1828 * abort/reset), this function should return SUCCESS.
1830 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1832 enum scsi_disposition rtn;
1835 * if the device is offline, then we clearly just pass the result back
1836 * up to the top level.
1838 if (!scsi_device_online(scmd->device)) {
1839 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1840 "%s: device offline - report as SUCCESS\n", __func__));
1845 * first check the host byte, to see if there is anything in there
1846 * that would indicate what we need to do.
1848 switch (host_byte(scmd->result)) {
1849 case DID_PASSTHROUGH:
1851 * no matter what, pass this through to the upper layer.
1852 * nuke this special code so that it looks like we are saying
1855 scmd->result &= 0xff00ffff;
1859 * looks good. drop through, and check the next byte.
1863 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1864 set_host_byte(scmd, DID_TIME_OUT);
1868 case DID_NO_CONNECT:
1869 case DID_BAD_TARGET:
1871 * note - this means that we just report the status back
1872 * to the top level driver, not that we actually think
1873 * that it indicates SUCCESS.
1876 case DID_SOFT_ERROR:
1878 * when the low level driver returns did_soft_error,
1879 * it is responsible for keeping an internal retry counter
1880 * in order to avoid endless loops (db)
1887 return ADD_TO_MLQUEUE;
1888 case DID_TRANSPORT_DISRUPTED:
1890 * LLD/transport was disrupted during processing of the IO.
1891 * The transport class is now blocked/blocking,
1892 * and the transport will decide what to do with the IO
1893 * based on its timers and recovery capablilities if
1894 * there are enough retries.
1897 case DID_TRANSPORT_FAILFAST:
1899 * The transport decided to failfast the IO (most likely
1900 * the fast io fail tmo fired), so send IO directly upwards.
1903 case DID_TRANSPORT_MARGINAL:
1905 * caller has decided not to do retries on
1906 * abort success, so send IO directly upwards
1910 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1912 * execute reservation conflict processing code
1922 * when we scan the bus, we get timeout messages for
1923 * these commands if there is no device available.
1924 * other hosts report did_no_connect for the same thing.
1926 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1927 scmd->cmnd[0] == INQUIRY)) {
1939 * check the status byte to see if this indicates anything special.
1941 switch (get_status_byte(scmd)) {
1942 case SAM_STAT_TASK_SET_FULL:
1943 scsi_handle_queue_full(scmd->device);
1945 * the case of trying to send too many commands to a
1946 * tagged queueing device.
1951 * device can't talk to us at the moment. Should only
1952 * occur (SAM-3) when the task queue is empty, so will cause
1953 * the empty queue handling to trigger a stall in the
1956 return ADD_TO_MLQUEUE;
1958 if (scmd->cmnd[0] == REPORT_LUNS)
1959 scmd->device->sdev_target->expecting_lun_change = 0;
1960 scsi_handle_queue_ramp_up(scmd->device);
1962 case SAM_STAT_COMMAND_TERMINATED:
1964 case SAM_STAT_TASK_ABORTED:
1966 case SAM_STAT_CHECK_CONDITION:
1967 rtn = scsi_check_sense(scmd);
1968 if (rtn == NEEDS_RETRY)
1970 /* if rtn == FAILED, we have no sense information;
1971 * returning FAILED will wake the error handler thread
1972 * to collect the sense and redo the decide
1975 case SAM_STAT_CONDITION_MET:
1976 case SAM_STAT_INTERMEDIATE:
1977 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1978 case SAM_STAT_ACA_ACTIVE:
1980 * who knows? FIXME(eric)
1984 case SAM_STAT_RESERVATION_CONFLICT:
1985 sdev_printk(KERN_INFO, scmd->device,
1986 "reservation conflict\n");
1987 set_host_byte(scmd, DID_NEXUS_FAILURE);
1988 return SUCCESS; /* causes immediate i/o error */
1996 /* we requeue for retry because the error was retryable, and
1997 * the request was not marked fast fail. Note that above,
1998 * even if the request is marked fast fail, we still requeue
1999 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2000 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
2004 * no more retries - report this one back to upper level.
2010 static void eh_lock_door_done(struct request *req, blk_status_t status)
2012 blk_put_request(req);
2016 * scsi_eh_lock_door - Prevent medium removal for the specified device
2017 * @sdev: SCSI device to prevent medium removal
2020 * We must be called from process context.
2023 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2024 * head of the devices request queue, and continue.
2026 static void scsi_eh_lock_door(struct scsi_device *sdev)
2028 struct request *req;
2029 struct scsi_request *rq;
2031 req = blk_get_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2036 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
2040 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2042 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
2044 req->rq_flags |= RQF_QUIET;
2045 req->timeout = 10 * HZ;
2048 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done);
2052 * scsi_restart_operations - restart io operations to the specified host.
2053 * @shost: Host we are restarting.
2056 * When we entered the error handler, we blocked all further i/o to
2057 * this device. we need to 'reverse' this process.
2059 static void scsi_restart_operations(struct Scsi_Host *shost)
2061 struct scsi_device *sdev;
2062 unsigned long flags;
2065 * If the door was locked, we need to insert a door lock request
2066 * onto the head of the SCSI request queue for the device. There
2067 * is no point trying to lock the door of an off-line device.
2069 shost_for_each_device(sdev, shost) {
2070 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2071 scsi_eh_lock_door(sdev);
2072 sdev->was_reset = 0;
2077 * next free up anything directly waiting upon the host. this
2078 * will be requests for character device operations, and also for
2079 * ioctls to queued block devices.
2081 SCSI_LOG_ERROR_RECOVERY(3,
2082 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2084 spin_lock_irqsave(shost->host_lock, flags);
2085 if (scsi_host_set_state(shost, SHOST_RUNNING))
2086 if (scsi_host_set_state(shost, SHOST_CANCEL))
2087 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2088 spin_unlock_irqrestore(shost->host_lock, flags);
2090 wake_up(&shost->host_wait);
2093 * finally we need to re-initiate requests that may be pending. we will
2094 * have had everything blocked while error handling is taking place, and
2095 * now that error recovery is done, we will need to ensure that these
2096 * requests are started.
2098 scsi_run_host_queues(shost);
2101 * if eh is active and host_eh_scheduled is pending we need to re-run
2102 * recovery. we do this check after scsi_run_host_queues() to allow
2103 * everything pent up since the last eh run a chance to make forward
2104 * progress before we sync again. Either we'll immediately re-run
2105 * recovery or scsi_device_unbusy() will wake us again when these
2106 * pending commands complete.
2108 spin_lock_irqsave(shost->host_lock, flags);
2109 if (shost->host_eh_scheduled)
2110 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2111 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2112 spin_unlock_irqrestore(shost->host_lock, flags);
2116 * scsi_eh_ready_devs - check device ready state and recover if not.
2117 * @shost: host to be recovered.
2118 * @work_q: &list_head for pending commands.
2119 * @done_q: &list_head for processed commands.
2121 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2122 struct list_head *work_q,
2123 struct list_head *done_q)
2125 if (!scsi_eh_stu(shost, work_q, done_q))
2126 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2127 if (!scsi_eh_target_reset(shost, work_q, done_q))
2128 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2129 if (!scsi_eh_host_reset(shost, work_q, done_q))
2130 scsi_eh_offline_sdevs(work_q,
2133 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2136 * scsi_eh_flush_done_q - finish processed commands or retry them.
2137 * @done_q: list_head of processed commands.
2139 void scsi_eh_flush_done_q(struct list_head *done_q)
2141 struct scsi_cmnd *scmd, *next;
2143 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2144 list_del_init(&scmd->eh_entry);
2145 if (scsi_device_online(scmd->device) &&
2146 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2147 scsi_eh_should_retry_cmd(scmd)) {
2148 SCSI_LOG_ERROR_RECOVERY(3,
2149 scmd_printk(KERN_INFO, scmd,
2150 "%s: flush retry cmd\n",
2152 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2155 * If just we got sense for the device (called
2156 * scsi_eh_get_sense), scmd->result is already
2157 * set, do not set DID_TIME_OUT.
2160 scmd->result |= (DID_TIME_OUT << 16);
2161 SCSI_LOG_ERROR_RECOVERY(3,
2162 scmd_printk(KERN_INFO, scmd,
2163 "%s: flush finish cmd\n",
2165 scsi_finish_command(scmd);
2169 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2172 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2173 * @shost: Host to unjam.
2176 * When we come in here, we *know* that all commands on the bus have
2177 * either completed, failed or timed out. we also know that no further
2178 * commands are being sent to the host, so things are relatively quiet
2179 * and we have freedom to fiddle with things as we wish.
2181 * This is only the *default* implementation. it is possible for
2182 * individual drivers to supply their own version of this function, and
2183 * if the maintainer wishes to do this, it is strongly suggested that
2184 * this function be taken as a template and modified. this function
2185 * was designed to correctly handle problems for about 95% of the
2186 * different cases out there, and it should always provide at least a
2187 * reasonable amount of error recovery.
2189 * Any command marked 'failed' or 'timeout' must eventually have
2190 * scsi_finish_cmd() called for it. we do all of the retry stuff
2191 * here, so when we restart the host after we return it should have an
2194 static void scsi_unjam_host(struct Scsi_Host *shost)
2196 unsigned long flags;
2197 LIST_HEAD(eh_work_q);
2198 LIST_HEAD(eh_done_q);
2200 spin_lock_irqsave(shost->host_lock, flags);
2201 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2202 spin_unlock_irqrestore(shost->host_lock, flags);
2204 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2206 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2207 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2209 spin_lock_irqsave(shost->host_lock, flags);
2210 if (shost->eh_deadline != -1)
2211 shost->last_reset = 0;
2212 spin_unlock_irqrestore(shost->host_lock, flags);
2213 scsi_eh_flush_done_q(&eh_done_q);
2217 * scsi_error_handler - SCSI error handler thread
2218 * @data: Host for which we are running.
2221 * This is the main error handling loop. This is run as a kernel thread
2222 * for every SCSI host and handles all error handling activity.
2224 int scsi_error_handler(void *data)
2226 struct Scsi_Host *shost = data;
2229 * We use TASK_INTERRUPTIBLE so that the thread is not
2230 * counted against the load average as a running process.
2231 * We never actually get interrupted because kthread_run
2232 * disables signal delivery for the created thread.
2236 * The sequence in kthread_stop() sets the stop flag first
2237 * then wakes the process. To avoid missed wakeups, the task
2238 * should always be in a non running state before the stop
2241 set_current_state(TASK_INTERRUPTIBLE);
2242 if (kthread_should_stop())
2245 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2246 shost->host_failed != scsi_host_busy(shost)) {
2247 SCSI_LOG_ERROR_RECOVERY(1,
2248 shost_printk(KERN_INFO, shost,
2249 "scsi_eh_%d: sleeping\n",
2255 __set_current_state(TASK_RUNNING);
2256 SCSI_LOG_ERROR_RECOVERY(1,
2257 shost_printk(KERN_INFO, shost,
2258 "scsi_eh_%d: waking up %d/%d/%d\n",
2259 shost->host_no, shost->host_eh_scheduled,
2261 scsi_host_busy(shost)));
2264 * We have a host that is failing for some reason. Figure out
2265 * what we need to do to get it up and online again (if we can).
2266 * If we fail, we end up taking the thing offline.
2268 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2269 SCSI_LOG_ERROR_RECOVERY(1,
2270 shost_printk(KERN_ERR, shost,
2271 "scsi_eh_%d: unable to autoresume\n",
2276 if (shost->transportt->eh_strategy_handler)
2277 shost->transportt->eh_strategy_handler(shost);
2279 scsi_unjam_host(shost);
2281 /* All scmds have been handled */
2282 shost->host_failed = 0;
2285 * Note - if the above fails completely, the action is to take
2286 * individual devices offline and flush the queue of any
2287 * outstanding requests that may have been pending. When we
2288 * restart, we restart any I/O to any other devices on the bus
2289 * which are still online.
2291 scsi_restart_operations(shost);
2292 if (!shost->eh_noresume)
2293 scsi_autopm_put_host(shost);
2295 __set_current_state(TASK_RUNNING);
2297 SCSI_LOG_ERROR_RECOVERY(1,
2298 shost_printk(KERN_INFO, shost,
2299 "Error handler scsi_eh_%d exiting\n",
2301 shost->ehandler = NULL;
2306 * Function: scsi_report_bus_reset()
2308 * Purpose: Utility function used by low-level drivers to report that
2309 * they have observed a bus reset on the bus being handled.
2311 * Arguments: shost - Host in question
2312 * channel - channel on which reset was observed.
2316 * Lock status: Host lock must be held.
2318 * Notes: This only needs to be called if the reset is one which
2319 * originates from an unknown location. Resets originated
2320 * by the mid-level itself don't need to call this, but there
2321 * should be no harm.
2323 * The main purpose of this is to make sure that a CHECK_CONDITION
2324 * is properly treated.
2326 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2328 struct scsi_device *sdev;
2330 __shost_for_each_device(sdev, shost) {
2331 if (channel == sdev_channel(sdev))
2332 __scsi_report_device_reset(sdev, NULL);
2335 EXPORT_SYMBOL(scsi_report_bus_reset);
2338 * Function: scsi_report_device_reset()
2340 * Purpose: Utility function used by low-level drivers to report that
2341 * they have observed a device reset on the device being handled.
2343 * Arguments: shost - Host in question
2344 * channel - channel on which reset was observed
2345 * target - target on which reset was observed
2349 * Lock status: Host lock must be held
2351 * Notes: This only needs to be called if the reset is one which
2352 * originates from an unknown location. Resets originated
2353 * by the mid-level itself don't need to call this, but there
2354 * should be no harm.
2356 * The main purpose of this is to make sure that a CHECK_CONDITION
2357 * is properly treated.
2359 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2361 struct scsi_device *sdev;
2363 __shost_for_each_device(sdev, shost) {
2364 if (channel == sdev_channel(sdev) &&
2365 target == sdev_id(sdev))
2366 __scsi_report_device_reset(sdev, NULL);
2369 EXPORT_SYMBOL(scsi_report_device_reset);
2372 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2377 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2378 * @dev: scsi_device to operate on
2379 * @arg: reset type (see sg.h)
2382 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2384 struct scsi_cmnd *scmd;
2385 struct Scsi_Host *shost = dev->host;
2387 unsigned long flags;
2389 enum scsi_disposition rtn;
2391 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2394 error = get_user(val, arg);
2398 if (scsi_autopm_get_host(shost) < 0)
2402 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2403 shost->hostt->cmd_size, GFP_KERNEL);
2405 goto out_put_autopm_host;
2406 blk_rq_init(NULL, rq);
2408 scmd = (struct scsi_cmnd *)(rq + 1);
2409 scsi_init_command(dev, scmd);
2410 scmd->cmnd = scsi_req(rq)->cmd;
2412 scmd->scsi_done = scsi_reset_provider_done_command;
2413 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2417 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2419 spin_lock_irqsave(shost->host_lock, flags);
2420 shost->tmf_in_progress = 1;
2421 spin_unlock_irqrestore(shost->host_lock, flags);
2423 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2424 case SG_SCSI_RESET_NOTHING:
2427 case SG_SCSI_RESET_DEVICE:
2428 rtn = scsi_try_bus_device_reset(scmd);
2429 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2432 case SG_SCSI_RESET_TARGET:
2433 rtn = scsi_try_target_reset(scmd);
2434 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2437 case SG_SCSI_RESET_BUS:
2438 rtn = scsi_try_bus_reset(scmd);
2439 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2442 case SG_SCSI_RESET_HOST:
2443 rtn = scsi_try_host_reset(scmd);
2452 error = (rtn == SUCCESS) ? 0 : -EIO;
2454 spin_lock_irqsave(shost->host_lock, flags);
2455 shost->tmf_in_progress = 0;
2456 spin_unlock_irqrestore(shost->host_lock, flags);
2459 * be sure to wake up anyone who was sleeping or had their queue
2460 * suspended while we performed the TMF.
2462 SCSI_LOG_ERROR_RECOVERY(3,
2463 shost_printk(KERN_INFO, shost,
2464 "waking up host to restart after TMF\n"));
2466 wake_up(&shost->host_wait);
2467 scsi_run_host_queues(shost);
2471 out_put_autopm_host:
2472 scsi_autopm_put_host(shost);
2476 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2477 struct scsi_sense_hdr *sshdr)
2479 return scsi_normalize_sense(cmd->sense_buffer,
2480 SCSI_SENSE_BUFFERSIZE, sshdr);
2482 EXPORT_SYMBOL(scsi_command_normalize_sense);
2485 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2486 * @sense_buffer: byte array of sense data
2487 * @sb_len: number of valid bytes in sense_buffer
2488 * @info_out: pointer to 64 integer where 8 or 4 byte information
2489 * field will be placed if found.
2492 * true if information field found, false if not found.
2494 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2501 switch (sense_buffer[0] & 0x7f) {
2504 if (sense_buffer[0] & 0x80) {
2505 *info_out = get_unaligned_be32(&sense_buffer[3]);
2511 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2513 if (ucp && (0xa == ucp[1])) {
2514 *info_out = get_unaligned_be64(&ucp[4]);
2522 EXPORT_SYMBOL(scsi_get_sense_info_fld);