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.
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>
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
57 #define BUS_RESET_SETTLE_TIME (10)
58 #define HOST_RESET_SETTLE_TIME (10)
60 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61 static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
64 void scsi_eh_wakeup(struct Scsi_Host *shost)
66 lockdep_assert_held(shost->host_lock);
68 if (scsi_host_busy(shost) == shost->host_failed) {
69 trace_scsi_eh_wakeup(shost);
70 wake_up_process(shost->ehandler);
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 "Waking error handler thread\n"));
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
80 * Schedule SCSI EH without scmd.
82 void scsi_schedule_eh(struct Scsi_Host *shost)
86 spin_lock_irqsave(shost->host_lock, flags);
88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
90 shost->host_eh_scheduled++;
91 scsi_eh_wakeup(shost);
94 spin_unlock_irqrestore(shost->host_lock, flags);
96 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
98 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
100 if (!shost->last_reset || shost->eh_deadline == -1)
104 * 32bit accesses are guaranteed to be atomic
105 * (on all supported architectures), so instead
106 * of using a spinlock we can as well double check
107 * if eh_deadline has been set to 'off' during the
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 shost->eh_deadline > -1)
117 static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
119 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
122 return ++cmd->retries <= cmd->allowed;
125 static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
127 struct scsi_device *sdev = cmd->device;
128 struct Scsi_Host *host = sdev->host;
130 if (host->hostt->eh_should_retry_cmd)
131 return host->hostt->eh_should_retry_cmd(cmd);
136 static void scsi_eh_complete_abort(struct scsi_cmnd *scmd, struct Scsi_Host *shost)
140 spin_lock_irqsave(shost->host_lock, flags);
141 list_del_init(&scmd->eh_entry);
143 * If the abort succeeds, and there is no further
144 * EH action, clear the ->last_reset time.
146 if (list_empty(&shost->eh_abort_list) &&
147 list_empty(&shost->eh_cmd_q))
148 if (shost->eh_deadline != -1)
149 shost->last_reset = 0;
150 spin_unlock_irqrestore(shost->host_lock, flags);
154 * scmd_eh_abort_handler - Handle command aborts
155 * @work: command to be aborted.
157 * Note: this function must be called only for a command that has timed out.
158 * Because the block layer marks a request as complete before it calls
159 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
160 * timed out do not have any effect. Hence it is safe to call
161 * scsi_finish_command() from this function.
164 scmd_eh_abort_handler(struct work_struct *work)
166 struct scsi_cmnd *scmd =
167 container_of(work, struct scsi_cmnd, abort_work.work);
168 struct scsi_device *sdev = scmd->device;
169 enum scsi_disposition rtn;
172 if (scsi_host_eh_past_deadline(sdev->host)) {
173 SCSI_LOG_ERROR_RECOVERY(3,
174 scmd_printk(KERN_INFO, scmd,
175 "eh timeout, not aborting\n"));
177 SCSI_LOG_ERROR_RECOVERY(3,
178 scmd_printk(KERN_INFO, scmd,
179 "aborting command\n"));
180 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
181 if (rtn == SUCCESS) {
182 set_host_byte(scmd, DID_TIME_OUT);
183 if (scsi_host_eh_past_deadline(sdev->host)) {
184 SCSI_LOG_ERROR_RECOVERY(3,
185 scmd_printk(KERN_INFO, scmd,
186 "eh timeout, not retrying "
187 "aborted command\n"));
188 } else if (!scsi_noretry_cmd(scmd) &&
189 scsi_cmd_retry_allowed(scmd) &&
190 scsi_eh_should_retry_cmd(scmd)) {
191 SCSI_LOG_ERROR_RECOVERY(3,
192 scmd_printk(KERN_WARNING, scmd,
193 "retry aborted command\n"));
194 scsi_eh_complete_abort(scmd, sdev->host);
195 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
198 SCSI_LOG_ERROR_RECOVERY(3,
199 scmd_printk(KERN_WARNING, scmd,
200 "finish aborted command\n"));
201 scsi_eh_complete_abort(scmd, sdev->host);
202 scsi_finish_command(scmd);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_INFO, scmd,
209 (rtn == FAST_IO_FAIL) ?
210 "not send" : "failed"));
214 spin_lock_irqsave(sdev->host->host_lock, flags);
215 list_del_init(&scmd->eh_entry);
216 spin_unlock_irqrestore(sdev->host->host_lock, flags);
217 scsi_eh_scmd_add(scmd);
221 * scsi_abort_command - schedule a command abort
222 * @scmd: scmd to abort.
224 * We only need to abort commands after a command timeout
227 scsi_abort_command(struct scsi_cmnd *scmd)
229 struct scsi_device *sdev = scmd->device;
230 struct Scsi_Host *shost = sdev->host;
233 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
235 * Retry after abort failed, escalate to next level.
237 SCSI_LOG_ERROR_RECOVERY(3,
238 scmd_printk(KERN_INFO, scmd,
239 "previous abort failed\n"));
240 BUG_ON(delayed_work_pending(&scmd->abort_work));
244 spin_lock_irqsave(shost->host_lock, flags);
245 if (shost->eh_deadline != -1 && !shost->last_reset)
246 shost->last_reset = jiffies;
247 BUG_ON(!list_empty(&scmd->eh_entry));
248 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list);
249 spin_unlock_irqrestore(shost->host_lock, flags);
251 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
252 SCSI_LOG_ERROR_RECOVERY(3,
253 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
254 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
259 * scsi_eh_reset - call into ->eh_action to reset internal counters
260 * @scmd: scmd to run eh on.
262 * The scsi driver might be carrying internal state about the
263 * devices, so we need to call into the driver to reset the
264 * internal state once the error handler is started.
266 static void scsi_eh_reset(struct scsi_cmnd *scmd)
268 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
269 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
271 sdrv->eh_reset(scmd);
275 static void scsi_eh_inc_host_failed(struct rcu_head *head)
277 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
278 struct Scsi_Host *shost = scmd->device->host;
281 spin_lock_irqsave(shost->host_lock, flags);
282 shost->host_failed++;
283 scsi_eh_wakeup(shost);
284 spin_unlock_irqrestore(shost->host_lock, flags);
288 * scsi_eh_scmd_add - add scsi cmd to error handling.
289 * @scmd: scmd to run eh on.
291 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
293 struct Scsi_Host *shost = scmd->device->host;
297 WARN_ON_ONCE(!shost->ehandler);
299 spin_lock_irqsave(shost->host_lock, flags);
300 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
301 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
304 if (shost->eh_deadline != -1 && !shost->last_reset)
305 shost->last_reset = jiffies;
308 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
309 spin_unlock_irqrestore(shost->host_lock, flags);
311 * Ensure that all tasks observe the host state change before the
312 * host_failed change.
314 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
318 * scsi_times_out - Timeout function for normal scsi commands.
319 * @req: request that is timing out.
322 * We do not need to lock this. There is the potential for a race
323 * only in that the normal completion handling might run, but if the
324 * normal completion function determines that the timer has already
325 * fired, then it mustn't do anything.
327 enum blk_eh_timer_return scsi_times_out(struct request *req)
329 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
330 enum blk_eh_timer_return rtn = BLK_EH_DONE;
331 struct Scsi_Host *host = scmd->device->host;
333 trace_scsi_dispatch_cmd_timeout(scmd);
334 scsi_log_completion(scmd, TIMEOUT_ERROR);
336 if (host->eh_deadline != -1 && !host->last_reset)
337 host->last_reset = jiffies;
339 if (host->hostt->eh_timed_out)
340 rtn = host->hostt->eh_timed_out(scmd);
342 if (rtn == BLK_EH_DONE) {
344 * Set the command to complete first in order to prevent a real
345 * completion from releasing the command while error handling
346 * is using it. If the command was already completed, then the
347 * lower level driver beat the timeout handler, and it is safe
348 * to return without escalating error recovery.
350 * If timeout handling lost the race to a real completion, the
351 * block layer may ignore that due to a fake timeout injection,
352 * so return RESET_TIMER to allow error handling another shot
355 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
356 return BLK_EH_RESET_TIMER;
357 if (scsi_abort_command(scmd) != SUCCESS) {
358 set_host_byte(scmd, DID_TIME_OUT);
359 scsi_eh_scmd_add(scmd);
367 * scsi_block_when_processing_errors - Prevent cmds from being queued.
368 * @sdev: Device on which we are performing recovery.
371 * We block until the host is out of error recovery, and then check to
372 * see whether the host or the device is offline.
375 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
377 int scsi_block_when_processing_errors(struct scsi_device *sdev)
381 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
383 online = scsi_device_online(sdev);
387 EXPORT_SYMBOL(scsi_block_when_processing_errors);
389 #ifdef CONFIG_SCSI_LOGGING
391 * scsi_eh_prt_fail_stats - Log info on failures.
392 * @shost: scsi host being recovered.
393 * @work_q: Queue of scsi cmds to process.
395 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
396 struct list_head *work_q)
398 struct scsi_cmnd *scmd;
399 struct scsi_device *sdev;
400 int total_failures = 0;
403 int devices_failed = 0;
405 shost_for_each_device(sdev, shost) {
406 list_for_each_entry(scmd, work_q, eh_entry) {
407 if (scmd->device == sdev) {
409 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
416 if (cmd_cancel || cmd_failed) {
417 SCSI_LOG_ERROR_RECOVERY(3,
418 shost_printk(KERN_INFO, shost,
419 "%s: cmds failed: %d, cancel: %d\n",
420 __func__, cmd_failed,
428 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
429 "Total of %d commands on %d"
430 " devices require eh work\n",
431 total_failures, devices_failed));
436 * scsi_report_lun_change - Set flag on all *other* devices on the same target
437 * to indicate that a UNIT ATTENTION is expected.
438 * @sdev: Device reporting the UNIT ATTENTION
440 static void scsi_report_lun_change(struct scsi_device *sdev)
442 sdev->sdev_target->expecting_lun_change = 1;
446 * scsi_report_sense - Examine scsi sense information and log messages for
447 * certain conditions, also issue uevents for some of them.
448 * @sdev: Device reporting the sense code
449 * @sshdr: sshdr to be examined
451 static void scsi_report_sense(struct scsi_device *sdev,
452 struct scsi_sense_hdr *sshdr)
454 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
456 if (sshdr->sense_key == UNIT_ATTENTION) {
457 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
458 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
459 sdev_printk(KERN_WARNING, sdev,
460 "Inquiry data has changed");
461 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
462 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
463 scsi_report_lun_change(sdev);
464 sdev_printk(KERN_WARNING, sdev,
465 "Warning! Received an indication that the "
466 "LUN assignments on this target have "
467 "changed. The Linux SCSI layer does not "
468 "automatically remap LUN assignments.\n");
469 } else if (sshdr->asc == 0x3f)
470 sdev_printk(KERN_WARNING, sdev,
471 "Warning! Received an indication that the "
472 "operating parameters on this target have "
473 "changed. The Linux SCSI layer does not "
474 "automatically adjust these parameters.\n");
476 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
477 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
478 sdev_printk(KERN_WARNING, sdev,
479 "Warning! Received an indication that the "
480 "LUN reached a thin provisioning soft "
484 if (sshdr->asc == 0x29) {
485 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
486 sdev_printk(KERN_WARNING, sdev,
487 "Power-on or device reset occurred\n");
490 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
491 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
492 sdev_printk(KERN_WARNING, sdev,
493 "Mode parameters changed");
494 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
495 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
496 sdev_printk(KERN_WARNING, sdev,
497 "Asymmetric access state changed");
498 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
499 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
500 sdev_printk(KERN_WARNING, sdev,
501 "Capacity data has changed");
502 } else if (sshdr->asc == 0x2a)
503 sdev_printk(KERN_WARNING, sdev,
504 "Parameters changed");
507 if (evt_type != SDEV_EVT_MAXBITS) {
508 set_bit(evt_type, sdev->pending_events);
509 schedule_work(&sdev->event_work);
514 * scsi_check_sense - Examine scsi cmd sense
515 * @scmd: Cmd to have sense checked.
518 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
521 * When a deferred error is detected the current command has
522 * not been executed and needs retrying.
524 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
526 struct scsi_device *sdev = scmd->device;
527 struct scsi_sense_hdr sshdr;
529 if (! scsi_command_normalize_sense(scmd, &sshdr))
530 return FAILED; /* no valid sense data */
532 scsi_report_sense(sdev, &sshdr);
534 if (scsi_sense_is_deferred(&sshdr))
537 if (sdev->handler && sdev->handler->check_sense) {
538 enum scsi_disposition rc;
540 rc = sdev->handler->check_sense(sdev, &sshdr);
541 if (rc != SCSI_RETURN_NOT_HANDLED)
543 /* handler does not care. Drop down to default handling */
546 if (scmd->cmnd[0] == TEST_UNIT_READY &&
547 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER)
549 * nasty: for mid-layer issued TURs, we need to return the
550 * actual sense data without any recovery attempt. For eh
551 * issued ones, we need to try to recover and interpret
556 * Previous logic looked for FILEMARK, EOM or ILI which are
557 * mainly associated with tapes and returned SUCCESS.
559 if (sshdr.response_code == 0x70) {
561 if (scmd->sense_buffer[2] & 0xe0)
565 * descriptor format: look for "stream commands sense data
566 * descriptor" (see SSC-3). Assume single sense data
567 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
569 if ((sshdr.additional_length > 3) &&
570 (scmd->sense_buffer[8] == 0x4) &&
571 (scmd->sense_buffer[11] & 0xe0))
575 switch (sshdr.sense_key) {
578 case RECOVERED_ERROR:
579 return /* soft_error */ SUCCESS;
581 case ABORTED_COMMAND:
582 if (sshdr.asc == 0x10) /* DIF */
585 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
586 return ADD_TO_MLQUEUE;
587 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
588 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
589 return ADD_TO_MLQUEUE;
595 * if we are expecting a cc/ua because of a bus reset that we
596 * performed, treat this just as a retry. otherwise this is
597 * information that we should pass up to the upper-level driver
598 * so that we can deal with it there.
600 if (scmd->device->expecting_cc_ua) {
602 * Because some device does not queue unit
603 * attentions correctly, we carefully check
604 * additional sense code and qualifier so as
605 * not to squash media change unit attention.
607 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
608 scmd->device->expecting_cc_ua = 0;
613 * we might also expect a cc/ua if another LUN on the target
614 * reported a UA with an ASC/ASCQ of 3F 0E -
615 * REPORTED LUNS DATA HAS CHANGED.
617 if (scmd->device->sdev_target->expecting_lun_change &&
618 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
621 * if the device is in the process of becoming ready, we
624 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
627 * if the device is not started, we need to wake
628 * the error handler to start the motor
630 if (scmd->device->allow_restart &&
631 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
634 * Pass the UA upwards for a determination in the completion
639 /* these are not supported */
641 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
642 /* Thin provisioning hard threshold reached */
643 set_host_byte(scmd, DID_ALLOC_FAILURE);
648 case VOLUME_OVERFLOW:
651 set_host_byte(scmd, DID_TARGET_FAILURE);
655 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
656 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
657 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
658 set_host_byte(scmd, DID_MEDIUM_ERROR);
664 if (scmd->device->retry_hwerror)
665 return ADD_TO_MLQUEUE;
667 set_host_byte(scmd, DID_TARGET_FAILURE);
670 case ILLEGAL_REQUEST:
671 if (sshdr.asc == 0x20 || /* Invalid command operation code */
672 sshdr.asc == 0x21 || /* Logical block address out of range */
673 sshdr.asc == 0x22 || /* Invalid function */
674 sshdr.asc == 0x24 || /* Invalid field in cdb */
675 sshdr.asc == 0x26 || /* Parameter value invalid */
676 sshdr.asc == 0x27) { /* Write protected */
677 set_host_byte(scmd, DID_TARGET_FAILURE);
685 EXPORT_SYMBOL_GPL(scsi_check_sense);
687 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
689 struct scsi_host_template *sht = sdev->host->hostt;
690 struct scsi_device *tmp_sdev;
692 if (!sht->track_queue_depth ||
693 sdev->queue_depth >= sdev->max_queue_depth)
696 if (time_before(jiffies,
697 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
700 if (time_before(jiffies,
701 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
705 * Walk all devices of a target and do
708 shost_for_each_device(tmp_sdev, sdev->host) {
709 if (tmp_sdev->channel != sdev->channel ||
710 tmp_sdev->id != sdev->id ||
711 tmp_sdev->queue_depth == sdev->max_queue_depth)
714 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
715 sdev->last_queue_ramp_up = jiffies;
719 static void scsi_handle_queue_full(struct scsi_device *sdev)
721 struct scsi_host_template *sht = sdev->host->hostt;
722 struct scsi_device *tmp_sdev;
724 if (!sht->track_queue_depth)
727 shost_for_each_device(tmp_sdev, sdev->host) {
728 if (tmp_sdev->channel != sdev->channel ||
729 tmp_sdev->id != sdev->id)
732 * We do not know the number of commands that were at
733 * the device when we got the queue full so we start
734 * from the highest possible value and work our way down.
736 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
741 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
742 * @scmd: SCSI cmd to examine.
745 * This is *only* called when we are examining the status of commands
746 * queued during error recovery. the main difference here is that we
747 * don't allow for the possibility of retries here, and we are a lot
748 * more restrictive about what we consider acceptable.
750 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
753 * first check the host byte, to see if there is anything in there
754 * that would indicate what we need to do.
756 if (host_byte(scmd->result) == DID_RESET) {
758 * rats. we are already in the error handler, so we now
759 * get to try and figure out what to do next. if the sense
760 * is valid, we have a pretty good idea of what to do.
761 * if not, we mark it as FAILED.
763 return scsi_check_sense(scmd);
765 if (host_byte(scmd->result) != DID_OK)
769 * now, check the status byte to see if this indicates
772 switch (get_status_byte(scmd)) {
774 scsi_handle_queue_ramp_up(scmd->device);
776 case SAM_STAT_COMMAND_TERMINATED:
778 case SAM_STAT_CHECK_CONDITION:
779 return scsi_check_sense(scmd);
780 case SAM_STAT_CONDITION_MET:
781 case SAM_STAT_INTERMEDIATE:
782 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
784 * who knows? FIXME(eric)
787 case SAM_STAT_RESERVATION_CONFLICT:
788 if (scmd->cmnd[0] == TEST_UNIT_READY)
789 /* it is a success, we probed the device and
792 /* otherwise, we failed to send the command */
794 case SAM_STAT_TASK_SET_FULL:
795 scsi_handle_queue_full(scmd->device);
806 * scsi_eh_done - Completion function for error handling.
807 * @scmd: Cmd that is done.
809 void scsi_eh_done(struct scsi_cmnd *scmd)
811 struct completion *eh_action;
813 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
814 "%s result: %x\n", __func__, scmd->result));
816 eh_action = scmd->device->host->eh_action;
822 * scsi_try_host_reset - ask host adapter to reset itself
823 * @scmd: SCSI cmd to send host reset.
825 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
828 enum scsi_disposition rtn;
829 struct Scsi_Host *host = scmd->device->host;
830 struct scsi_host_template *hostt = host->hostt;
832 SCSI_LOG_ERROR_RECOVERY(3,
833 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
835 if (!hostt->eh_host_reset_handler)
838 rtn = hostt->eh_host_reset_handler(scmd);
840 if (rtn == SUCCESS) {
841 if (!hostt->skip_settle_delay)
842 ssleep(HOST_RESET_SETTLE_TIME);
843 spin_lock_irqsave(host->host_lock, flags);
844 scsi_report_bus_reset(host, scmd_channel(scmd));
845 spin_unlock_irqrestore(host->host_lock, flags);
852 * scsi_try_bus_reset - ask host to perform a bus reset
853 * @scmd: SCSI cmd to send bus reset.
855 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
858 enum scsi_disposition rtn;
859 struct Scsi_Host *host = scmd->device->host;
860 struct scsi_host_template *hostt = host->hostt;
862 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
863 "%s: Snd Bus RST\n", __func__));
865 if (!hostt->eh_bus_reset_handler)
868 rtn = hostt->eh_bus_reset_handler(scmd);
870 if (rtn == SUCCESS) {
871 if (!hostt->skip_settle_delay)
872 ssleep(BUS_RESET_SETTLE_TIME);
873 spin_lock_irqsave(host->host_lock, flags);
874 scsi_report_bus_reset(host, scmd_channel(scmd));
875 spin_unlock_irqrestore(host->host_lock, flags);
881 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
884 sdev->expecting_cc_ua = 1;
888 * scsi_try_target_reset - Ask host to perform a target reset
889 * @scmd: SCSI cmd used to send a target reset
892 * There is no timeout for this operation. if this operation is
893 * unreliable for a given host, then the host itself needs to put a
894 * timer on it, and set the host back to a consistent state prior to
897 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
900 enum scsi_disposition rtn;
901 struct Scsi_Host *host = scmd->device->host;
902 struct scsi_host_template *hostt = host->hostt;
904 if (!hostt->eh_target_reset_handler)
907 rtn = hostt->eh_target_reset_handler(scmd);
908 if (rtn == SUCCESS) {
909 spin_lock_irqsave(host->host_lock, flags);
910 __starget_for_each_device(scsi_target(scmd->device), NULL,
911 __scsi_report_device_reset);
912 spin_unlock_irqrestore(host->host_lock, flags);
919 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
920 * @scmd: SCSI cmd used to send BDR
923 * There is no timeout for this operation. if this operation is
924 * unreliable for a given host, then the host itself needs to put a
925 * timer on it, and set the host back to a consistent state prior to
928 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
930 enum scsi_disposition rtn;
931 struct scsi_host_template *hostt = scmd->device->host->hostt;
933 if (!hostt->eh_device_reset_handler)
936 rtn = hostt->eh_device_reset_handler(scmd);
938 __scsi_report_device_reset(scmd->device, NULL);
943 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
944 * @hostt: SCSI driver host template
945 * @scmd: SCSI cmd used to send a target reset
948 * SUCCESS, FAILED, or FAST_IO_FAIL
951 * SUCCESS does not necessarily indicate that the command
952 * has been aborted; it only indicates that the LLDDs
953 * has cleared all references to that command.
954 * LLDDs should return FAILED only if an abort was required
955 * but could not be executed. LLDDs should return FAST_IO_FAIL
956 * if the device is temporarily unavailable (eg due to a
957 * link down on FibreChannel)
959 static enum scsi_disposition
960 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
962 if (!hostt->eh_abort_handler)
965 return hostt->eh_abort_handler(scmd);
968 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
970 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
971 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
972 if (scsi_try_target_reset(scmd) != SUCCESS)
973 if (scsi_try_bus_reset(scmd) != SUCCESS)
974 scsi_try_host_reset(scmd);
978 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
979 * @scmd: SCSI command structure to hijack
980 * @ses: structure to save restore information
981 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
982 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
983 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
985 * This function is used to save a scsi command information before re-execution
986 * as part of the error recovery process. If @sense_bytes is 0 the command
987 * sent must be one that does not transfer any data. If @sense_bytes != 0
988 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
989 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
991 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
992 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
994 struct scsi_device *sdev = scmd->device;
997 * We need saved copies of a number of fields - this is because
998 * error handling may need to overwrite these with different values
999 * to run different commands, and once error handling is complete,
1000 * we will need to restore these values prior to running the actual
1003 ses->cmd_len = scmd->cmd_len;
1004 ses->cmnd = scmd->cmnd;
1005 ses->data_direction = scmd->sc_data_direction;
1006 ses->sdb = scmd->sdb;
1007 ses->result = scmd->result;
1008 ses->resid_len = scmd->req.resid_len;
1009 ses->underflow = scmd->underflow;
1010 ses->prot_op = scmd->prot_op;
1011 ses->eh_eflags = scmd->eh_eflags;
1013 scmd->prot_op = SCSI_PROT_NORMAL;
1014 scmd->eh_eflags = 0;
1015 scmd->cmnd = ses->eh_cmnd;
1016 memset(scmd->cmnd, 0, BLK_MAX_CDB);
1017 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1019 scmd->req.resid_len = 0;
1022 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1024 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1026 scmd->sdb.table.sgl = &ses->sense_sgl;
1027 scmd->sc_data_direction = DMA_FROM_DEVICE;
1028 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1029 scmd->cmnd[0] = REQUEST_SENSE;
1030 scmd->cmnd[4] = scmd->sdb.length;
1031 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1033 scmd->sc_data_direction = DMA_NONE;
1035 BUG_ON(cmnd_size > BLK_MAX_CDB);
1036 memcpy(scmd->cmnd, cmnd, cmnd_size);
1037 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1041 scmd->underflow = 0;
1043 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1044 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1045 (sdev->lun << 5 & 0xe0);
1048 * Zero the sense buffer. The scsi spec mandates that any
1049 * untransferred sense data should be interpreted as being zero.
1051 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1053 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1056 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1057 * @scmd: SCSI command structure to restore
1058 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1060 * Undo any damage done by above scsi_eh_prep_cmnd().
1062 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1065 * Restore original data
1067 scmd->cmd_len = ses->cmd_len;
1068 scmd->cmnd = ses->cmnd;
1069 scmd->sc_data_direction = ses->data_direction;
1070 scmd->sdb = ses->sdb;
1071 scmd->result = ses->result;
1072 scmd->req.resid_len = ses->resid_len;
1073 scmd->underflow = ses->underflow;
1074 scmd->prot_op = ses->prot_op;
1075 scmd->eh_eflags = ses->eh_eflags;
1077 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1080 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1081 * @scmd: SCSI command structure to hijack
1082 * @cmnd: CDB to send
1083 * @cmnd_size: size in bytes of @cmnd
1084 * @timeout: timeout for this request
1085 * @sense_bytes: size of sense data to copy or 0
1087 * This function is used to send a scsi command down to a target device
1088 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1091 * SUCCESS or FAILED or NEEDS_RETRY
1093 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1094 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1096 struct scsi_device *sdev = scmd->device;
1097 struct Scsi_Host *shost = sdev->host;
1098 DECLARE_COMPLETION_ONSTACK(done);
1099 unsigned long timeleft = timeout, delay;
1100 struct scsi_eh_save ses;
1101 const unsigned long stall_for = msecs_to_jiffies(100);
1105 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1106 shost->eh_action = &done;
1108 scsi_log_send(scmd);
1109 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
1112 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1113 * change the SCSI device state after we have examined it and before
1114 * .queuecommand() is called.
1116 mutex_lock(&sdev->state_mutex);
1117 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1118 mutex_unlock(&sdev->state_mutex);
1119 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1120 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1122 delay = min(timeleft, stall_for);
1124 msleep(jiffies_to_msecs(delay));
1125 mutex_lock(&sdev->state_mutex);
1127 if (sdev->sdev_state != SDEV_BLOCK)
1128 rtn = shost->hostt->queuecommand(shost, scmd);
1131 mutex_unlock(&sdev->state_mutex);
1134 if (timeleft > stall_for) {
1135 scsi_eh_restore_cmnd(scmd, &ses);
1137 timeleft -= stall_for;
1138 msleep(jiffies_to_msecs(stall_for));
1141 /* signal not to enter either branch of the if () below */
1145 timeleft = wait_for_completion_timeout(&done, timeout);
1149 shost->eh_action = NULL;
1151 scsi_log_completion(scmd, rtn);
1153 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1154 "%s timeleft: %ld\n",
1155 __func__, timeleft));
1158 * If there is time left scsi_eh_done got called, and we will examine
1159 * the actual status codes to see whether the command actually did
1160 * complete normally, else if we have a zero return and no time left,
1161 * the command must still be pending, so abort it and return FAILED.
1162 * If we never actually managed to issue the command, because
1163 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1164 * value above (so don't execute either branch of the if)
1167 rtn = scsi_eh_completed_normally(scmd);
1168 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1169 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1176 case ADD_TO_MLQUEUE:
1183 } else if (rtn != FAILED) {
1184 scsi_abort_eh_cmnd(scmd);
1188 scsi_eh_restore_cmnd(scmd, &ses);
1194 * scsi_request_sense - Request sense data from a particular target.
1195 * @scmd: SCSI cmd for request sense.
1198 * Some hosts automatically obtain this information, others require
1199 * that we obtain it on our own. This function will *not* return until
1200 * the command either times out, or it completes.
1202 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1204 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1207 static enum scsi_disposition
1208 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1210 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
1211 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1212 if (sdrv->eh_action)
1213 rtn = sdrv->eh_action(scmd, rtn);
1219 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1220 * @scmd: Original SCSI cmd that eh has finished.
1221 * @done_q: Queue for processed commands.
1224 * We don't want to use the normal command completion while we are are
1225 * still handling errors - it may cause other commands to be queued,
1226 * and that would disturb what we are doing. Thus we really want to
1227 * keep a list of pending commands for final completion, and once we
1228 * are ready to leave error handling we handle completion for real.
1230 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1232 list_move_tail(&scmd->eh_entry, done_q);
1234 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1237 * scsi_eh_get_sense - Get device sense data.
1238 * @work_q: Queue of commands to process.
1239 * @done_q: Queue of processed commands.
1242 * See if we need to request sense information. if so, then get it
1243 * now, so we have a better idea of what to do.
1246 * This has the unfortunate side effect that if a shost adapter does
1247 * not automatically request sense information, we end up shutting
1248 * it down before we request it.
1250 * All drivers should request sense information internally these days,
1251 * so for now all I have to say is tough noogies if you end up in here.
1253 * XXX: Long term this code should go away, but that needs an audit of
1256 int scsi_eh_get_sense(struct list_head *work_q,
1257 struct list_head *done_q)
1259 struct scsi_cmnd *scmd, *next;
1260 struct Scsi_Host *shost;
1261 enum scsi_disposition rtn;
1264 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1265 * should not get sense.
1267 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1268 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1269 SCSI_SENSE_VALID(scmd))
1272 shost = scmd->device->host;
1273 if (scsi_host_eh_past_deadline(shost)) {
1274 SCSI_LOG_ERROR_RECOVERY(3,
1275 scmd_printk(KERN_INFO, scmd,
1276 "%s: skip request sense, past eh deadline\n",
1280 if (!scsi_status_is_check_condition(scmd->result))
1282 * don't request sense if there's no check condition
1283 * status because the error we're processing isn't one
1284 * that has a sense code (and some devices get
1285 * confused by sense requests out of the blue)
1289 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1290 "%s: requesting sense\n",
1292 rtn = scsi_request_sense(scmd);
1296 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1297 "sense requested, result %x\n", scmd->result));
1298 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1300 rtn = scsi_decide_disposition(scmd);
1303 * if the result was normal, then just pass it along to the
1308 * We don't want this command reissued, just finished
1309 * with the sense data, so set retries to the max
1310 * allowed to ensure it won't get reissued. If the user
1311 * has requested infinite retries, we also want to
1312 * finish this command, so force completion by setting
1313 * retries and allowed to the same value.
1315 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1316 scmd->retries = scmd->allowed = 1;
1318 scmd->retries = scmd->allowed;
1319 else if (rtn != NEEDS_RETRY)
1322 scsi_eh_finish_cmd(scmd, done_q);
1325 return list_empty(work_q);
1327 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1330 * scsi_eh_tur - Send TUR to device.
1331 * @scmd: &scsi_cmnd to send TUR
1334 * 0 - Device is ready. 1 - Device NOT ready.
1336 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1338 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1340 enum scsi_disposition rtn;
1343 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1344 scmd->device->eh_timeout, 0);
1346 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1347 "%s return: %x\n", __func__, rtn));
1362 * scsi_eh_test_devices - check if devices are responding from error recovery.
1363 * @cmd_list: scsi commands in error recovery.
1364 * @work_q: queue for commands which still need more error recovery
1365 * @done_q: queue for commands which are finished
1366 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1369 * Tests if devices are in a working state. Commands to devices now in
1370 * a working state are sent to the done_q while commands to devices which
1371 * are still failing to respond are returned to the work_q for more
1374 static int scsi_eh_test_devices(struct list_head *cmd_list,
1375 struct list_head *work_q,
1376 struct list_head *done_q, int try_stu)
1378 struct scsi_cmnd *scmd, *next;
1379 struct scsi_device *sdev;
1382 while (!list_empty(cmd_list)) {
1383 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1384 sdev = scmd->device;
1387 if (scsi_host_eh_past_deadline(sdev->host)) {
1388 /* Push items back onto work_q */
1389 list_splice_init(cmd_list, work_q);
1390 SCSI_LOG_ERROR_RECOVERY(3,
1391 sdev_printk(KERN_INFO, sdev,
1392 "%s: skip test device, past eh deadline",
1398 finish_cmds = !scsi_device_online(scmd->device) ||
1399 (try_stu && !scsi_eh_try_stu(scmd) &&
1400 !scsi_eh_tur(scmd)) ||
1403 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1404 if (scmd->device == sdev) {
1407 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1408 scsi_eh_finish_cmd(scmd, done_q);
1410 list_move_tail(&scmd->eh_entry, work_q);
1413 return list_empty(work_q);
1417 * scsi_eh_try_stu - Send START_UNIT to device.
1418 * @scmd: &scsi_cmnd to send START_UNIT
1421 * 0 - Device is ready. 1 - Device NOT ready.
1423 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1425 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1427 if (scmd->device->allow_restart) {
1429 enum scsi_disposition rtn = NEEDS_RETRY;
1431 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1432 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1442 * scsi_eh_stu - send START_UNIT if needed
1443 * @shost: &scsi host being recovered.
1444 * @work_q: &list_head for pending commands.
1445 * @done_q: &list_head for processed commands.
1448 * If commands are failing due to not ready, initializing command required,
1449 * try revalidating the device, which will end up sending a start unit.
1451 static int scsi_eh_stu(struct Scsi_Host *shost,
1452 struct list_head *work_q,
1453 struct list_head *done_q)
1455 struct scsi_cmnd *scmd, *stu_scmd, *next;
1456 struct scsi_device *sdev;
1458 shost_for_each_device(sdev, shost) {
1459 if (scsi_host_eh_past_deadline(shost)) {
1460 SCSI_LOG_ERROR_RECOVERY(3,
1461 sdev_printk(KERN_INFO, sdev,
1462 "%s: skip START_UNIT, past eh deadline\n",
1464 scsi_device_put(sdev);
1468 list_for_each_entry(scmd, work_q, eh_entry)
1469 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1470 scsi_check_sense(scmd) == FAILED ) {
1478 SCSI_LOG_ERROR_RECOVERY(3,
1479 sdev_printk(KERN_INFO, sdev,
1480 "%s: Sending START_UNIT\n",
1483 if (!scsi_eh_try_stu(stu_scmd)) {
1484 if (!scsi_device_online(sdev) ||
1485 !scsi_eh_tur(stu_scmd)) {
1486 list_for_each_entry_safe(scmd, next,
1488 if (scmd->device == sdev &&
1489 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1490 scsi_eh_finish_cmd(scmd, done_q);
1494 SCSI_LOG_ERROR_RECOVERY(3,
1495 sdev_printk(KERN_INFO, sdev,
1496 "%s: START_UNIT failed\n",
1501 return list_empty(work_q);
1506 * scsi_eh_bus_device_reset - send bdr if needed
1507 * @shost: scsi host being recovered.
1508 * @work_q: &list_head for pending commands.
1509 * @done_q: &list_head for processed commands.
1512 * Try a bus device reset. Still, look to see whether we have multiple
1513 * devices that are jammed or not - if we have multiple devices, it
1514 * makes no sense to try bus_device_reset - we really would need to try
1515 * a bus_reset instead.
1517 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1518 struct list_head *work_q,
1519 struct list_head *done_q)
1521 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1522 struct scsi_device *sdev;
1523 enum scsi_disposition rtn;
1525 shost_for_each_device(sdev, shost) {
1526 if (scsi_host_eh_past_deadline(shost)) {
1527 SCSI_LOG_ERROR_RECOVERY(3,
1528 sdev_printk(KERN_INFO, sdev,
1529 "%s: skip BDR, past eh deadline\n",
1531 scsi_device_put(sdev);
1535 list_for_each_entry(scmd, work_q, eh_entry)
1536 if (scmd->device == sdev) {
1544 SCSI_LOG_ERROR_RECOVERY(3,
1545 sdev_printk(KERN_INFO, sdev,
1546 "%s: Sending BDR\n", current->comm));
1547 rtn = scsi_try_bus_device_reset(bdr_scmd);
1548 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1549 if (!scsi_device_online(sdev) ||
1550 rtn == FAST_IO_FAIL ||
1551 !scsi_eh_tur(bdr_scmd)) {
1552 list_for_each_entry_safe(scmd, next,
1554 if (scmd->device == sdev &&
1555 scsi_eh_action(scmd, rtn) != FAILED)
1556 scsi_eh_finish_cmd(scmd,
1561 SCSI_LOG_ERROR_RECOVERY(3,
1562 sdev_printk(KERN_INFO, sdev,
1563 "%s: BDR failed\n", current->comm));
1567 return list_empty(work_q);
1571 * scsi_eh_target_reset - send target reset if needed
1572 * @shost: scsi host being recovered.
1573 * @work_q: &list_head for pending commands.
1574 * @done_q: &list_head for processed commands.
1577 * Try a target reset.
1579 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1580 struct list_head *work_q,
1581 struct list_head *done_q)
1583 LIST_HEAD(tmp_list);
1584 LIST_HEAD(check_list);
1586 list_splice_init(work_q, &tmp_list);
1588 while (!list_empty(&tmp_list)) {
1589 struct scsi_cmnd *next, *scmd;
1590 enum scsi_disposition rtn;
1593 if (scsi_host_eh_past_deadline(shost)) {
1594 /* push back on work queue for further processing */
1595 list_splice_init(&check_list, work_q);
1596 list_splice_init(&tmp_list, work_q);
1597 SCSI_LOG_ERROR_RECOVERY(3,
1598 shost_printk(KERN_INFO, shost,
1599 "%s: Skip target reset, past eh deadline\n",
1601 return list_empty(work_q);
1604 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1607 SCSI_LOG_ERROR_RECOVERY(3,
1608 shost_printk(KERN_INFO, shost,
1609 "%s: Sending target reset to target %d\n",
1610 current->comm, id));
1611 rtn = scsi_try_target_reset(scmd);
1612 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1613 SCSI_LOG_ERROR_RECOVERY(3,
1614 shost_printk(KERN_INFO, shost,
1615 "%s: Target reset failed"
1617 current->comm, id));
1618 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1619 if (scmd_id(scmd) != id)
1623 list_move_tail(&scmd->eh_entry, &check_list);
1624 else if (rtn == FAST_IO_FAIL)
1625 scsi_eh_finish_cmd(scmd, done_q);
1627 /* push back on work queue for further processing */
1628 list_move(&scmd->eh_entry, work_q);
1632 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1636 * scsi_eh_bus_reset - send a bus reset
1637 * @shost: &scsi host being recovered.
1638 * @work_q: &list_head for pending commands.
1639 * @done_q: &list_head for processed commands.
1641 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1642 struct list_head *work_q,
1643 struct list_head *done_q)
1645 struct scsi_cmnd *scmd, *chan_scmd, *next;
1646 LIST_HEAD(check_list);
1647 unsigned int channel;
1648 enum scsi_disposition rtn;
1651 * we really want to loop over the various channels, and do this on
1652 * a channel by channel basis. we should also check to see if any
1653 * of the failed commands are on soft_reset devices, and if so, skip
1657 for (channel = 0; channel <= shost->max_channel; channel++) {
1658 if (scsi_host_eh_past_deadline(shost)) {
1659 list_splice_init(&check_list, work_q);
1660 SCSI_LOG_ERROR_RECOVERY(3,
1661 shost_printk(KERN_INFO, shost,
1662 "%s: skip BRST, past eh deadline\n",
1664 return list_empty(work_q);
1668 list_for_each_entry(scmd, work_q, eh_entry) {
1669 if (channel == scmd_channel(scmd)) {
1673 * FIXME add back in some support for
1674 * soft_reset devices.
1681 SCSI_LOG_ERROR_RECOVERY(3,
1682 shost_printk(KERN_INFO, shost,
1683 "%s: Sending BRST chan: %d\n",
1684 current->comm, channel));
1685 rtn = scsi_try_bus_reset(chan_scmd);
1686 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1687 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1688 if (channel == scmd_channel(scmd)) {
1689 if (rtn == FAST_IO_FAIL)
1690 scsi_eh_finish_cmd(scmd,
1693 list_move_tail(&scmd->eh_entry,
1698 SCSI_LOG_ERROR_RECOVERY(3,
1699 shost_printk(KERN_INFO, shost,
1700 "%s: BRST failed chan: %d\n",
1701 current->comm, channel));
1704 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1708 * scsi_eh_host_reset - send a host reset
1709 * @shost: host to be reset.
1710 * @work_q: &list_head for pending commands.
1711 * @done_q: &list_head for processed commands.
1713 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1714 struct list_head *work_q,
1715 struct list_head *done_q)
1717 struct scsi_cmnd *scmd, *next;
1718 LIST_HEAD(check_list);
1719 enum scsi_disposition rtn;
1721 if (!list_empty(work_q)) {
1722 scmd = list_entry(work_q->next,
1723 struct scsi_cmnd, eh_entry);
1725 SCSI_LOG_ERROR_RECOVERY(3,
1726 shost_printk(KERN_INFO, shost,
1727 "%s: Sending HRST\n",
1730 rtn = scsi_try_host_reset(scmd);
1731 if (rtn == SUCCESS) {
1732 list_splice_init(work_q, &check_list);
1733 } else if (rtn == FAST_IO_FAIL) {
1734 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1735 scsi_eh_finish_cmd(scmd, done_q);
1738 SCSI_LOG_ERROR_RECOVERY(3,
1739 shost_printk(KERN_INFO, shost,
1740 "%s: HRST failed\n",
1744 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1748 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1749 * @work_q: &list_head for pending commands.
1750 * @done_q: &list_head for processed commands.
1752 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1753 struct list_head *done_q)
1755 struct scsi_cmnd *scmd, *next;
1756 struct scsi_device *sdev;
1758 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1759 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1760 "not ready after error recovery\n");
1761 sdev = scmd->device;
1763 mutex_lock(&sdev->state_mutex);
1764 scsi_device_set_state(sdev, SDEV_OFFLINE);
1765 mutex_unlock(&sdev->state_mutex);
1767 scsi_eh_finish_cmd(scmd, done_q);
1773 * scsi_noretry_cmd - determine if command should be failed fast
1774 * @scmd: SCSI cmd to examine.
1776 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1778 struct request *req = scsi_cmd_to_rq(scmd);
1780 switch (host_byte(scmd->result)) {
1786 return req->cmd_flags & REQ_FAILFAST_TRANSPORT;
1788 return req->cmd_flags & REQ_FAILFAST_DEV;
1790 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1793 case DID_SOFT_ERROR:
1794 return req->cmd_flags & REQ_FAILFAST_DRIVER;
1797 if (!scsi_status_is_check_condition(scmd->result))
1802 * assume caller has checked sense and determined
1803 * the check condition was retryable.
1805 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
1812 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1813 * @scmd: SCSI cmd to examine.
1816 * This is *only* called when we are examining the status after sending
1817 * out the actual data command. any commands that are queued for error
1818 * recovery (e.g. test_unit_ready) do *not* come through here.
1820 * When this routine returns failed, it means the error handler thread
1821 * is woken. In cases where the error code indicates an error that
1822 * doesn't require the error handler read (i.e. we don't need to
1823 * abort/reset), this function should return SUCCESS.
1825 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1827 enum scsi_disposition rtn;
1830 * if the device is offline, then we clearly just pass the result back
1831 * up to the top level.
1833 if (!scsi_device_online(scmd->device)) {
1834 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1835 "%s: device offline - report as SUCCESS\n", __func__));
1840 * first check the host byte, to see if there is anything in there
1841 * that would indicate what we need to do.
1843 switch (host_byte(scmd->result)) {
1844 case DID_PASSTHROUGH:
1846 * no matter what, pass this through to the upper layer.
1847 * nuke this special code so that it looks like we are saying
1850 scmd->result &= 0xff00ffff;
1854 * looks good. drop through, and check the next byte.
1858 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1859 set_host_byte(scmd, DID_TIME_OUT);
1863 case DID_NO_CONNECT:
1864 case DID_BAD_TARGET:
1866 * note - this means that we just report the status back
1867 * to the top level driver, not that we actually think
1868 * that it indicates SUCCESS.
1871 case DID_SOFT_ERROR:
1873 * when the low level driver returns did_soft_error,
1874 * it is responsible for keeping an internal retry counter
1875 * in order to avoid endless loops (db)
1882 return ADD_TO_MLQUEUE;
1883 case DID_TRANSPORT_DISRUPTED:
1885 * LLD/transport was disrupted during processing of the IO.
1886 * The transport class is now blocked/blocking,
1887 * and the transport will decide what to do with the IO
1888 * based on its timers and recovery capablilities if
1889 * there are enough retries.
1892 case DID_TRANSPORT_FAILFAST:
1894 * The transport decided to failfast the IO (most likely
1895 * the fast io fail tmo fired), so send IO directly upwards.
1898 case DID_TRANSPORT_MARGINAL:
1900 * caller has decided not to do retries on
1901 * abort success, so send IO directly upwards
1905 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1907 * execute reservation conflict processing code
1917 * when we scan the bus, we get timeout messages for
1918 * these commands if there is no device available.
1919 * other hosts report did_no_connect for the same thing.
1921 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1922 scmd->cmnd[0] == INQUIRY)) {
1934 * check the status byte to see if this indicates anything special.
1936 switch (get_status_byte(scmd)) {
1937 case SAM_STAT_TASK_SET_FULL:
1938 scsi_handle_queue_full(scmd->device);
1940 * the case of trying to send too many commands to a
1941 * tagged queueing device.
1946 * device can't talk to us at the moment. Should only
1947 * occur (SAM-3) when the task queue is empty, so will cause
1948 * the empty queue handling to trigger a stall in the
1951 return ADD_TO_MLQUEUE;
1953 if (scmd->cmnd[0] == REPORT_LUNS)
1954 scmd->device->sdev_target->expecting_lun_change = 0;
1955 scsi_handle_queue_ramp_up(scmd->device);
1957 case SAM_STAT_COMMAND_TERMINATED:
1959 case SAM_STAT_TASK_ABORTED:
1961 case SAM_STAT_CHECK_CONDITION:
1962 rtn = scsi_check_sense(scmd);
1963 if (rtn == NEEDS_RETRY)
1965 /* if rtn == FAILED, we have no sense information;
1966 * returning FAILED will wake the error handler thread
1967 * to collect the sense and redo the decide
1970 case SAM_STAT_CONDITION_MET:
1971 case SAM_STAT_INTERMEDIATE:
1972 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1973 case SAM_STAT_ACA_ACTIVE:
1975 * who knows? FIXME(eric)
1979 case SAM_STAT_RESERVATION_CONFLICT:
1980 sdev_printk(KERN_INFO, scmd->device,
1981 "reservation conflict\n");
1982 set_host_byte(scmd, DID_NEXUS_FAILURE);
1983 return SUCCESS; /* causes immediate i/o error */
1991 /* we requeue for retry because the error was retryable, and
1992 * the request was not marked fast fail. Note that above,
1993 * even if the request is marked fast fail, we still requeue
1994 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1995 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
1999 * no more retries - report this one back to upper level.
2005 static void eh_lock_door_done(struct request *req, blk_status_t status)
2007 blk_mq_free_request(req);
2011 * scsi_eh_lock_door - Prevent medium removal for the specified device
2012 * @sdev: SCSI device to prevent medium removal
2015 * We must be called from process context.
2018 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2019 * head of the devices request queue, and continue.
2021 static void scsi_eh_lock_door(struct scsi_device *sdev)
2023 struct request *req;
2024 struct scsi_request *rq;
2026 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2031 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
2035 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2037 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
2039 req->rq_flags |= RQF_QUIET;
2040 req->timeout = 10 * HZ;
2043 blk_execute_rq_nowait(NULL, req, 1, eh_lock_door_done);
2047 * scsi_restart_operations - restart io operations to the specified host.
2048 * @shost: Host we are restarting.
2051 * When we entered the error handler, we blocked all further i/o to
2052 * this device. we need to 'reverse' this process.
2054 static void scsi_restart_operations(struct Scsi_Host *shost)
2056 struct scsi_device *sdev;
2057 unsigned long flags;
2060 * If the door was locked, we need to insert a door lock request
2061 * onto the head of the SCSI request queue for the device. There
2062 * is no point trying to lock the door of an off-line device.
2064 shost_for_each_device(sdev, shost) {
2065 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2066 scsi_eh_lock_door(sdev);
2067 sdev->was_reset = 0;
2072 * next free up anything directly waiting upon the host. this
2073 * will be requests for character device operations, and also for
2074 * ioctls to queued block devices.
2076 SCSI_LOG_ERROR_RECOVERY(3,
2077 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2079 spin_lock_irqsave(shost->host_lock, flags);
2080 if (scsi_host_set_state(shost, SHOST_RUNNING))
2081 if (scsi_host_set_state(shost, SHOST_CANCEL))
2082 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2083 spin_unlock_irqrestore(shost->host_lock, flags);
2085 wake_up(&shost->host_wait);
2088 * finally we need to re-initiate requests that may be pending. we will
2089 * have had everything blocked while error handling is taking place, and
2090 * now that error recovery is done, we will need to ensure that these
2091 * requests are started.
2093 scsi_run_host_queues(shost);
2096 * if eh is active and host_eh_scheduled is pending we need to re-run
2097 * recovery. we do this check after scsi_run_host_queues() to allow
2098 * everything pent up since the last eh run a chance to make forward
2099 * progress before we sync again. Either we'll immediately re-run
2100 * recovery or scsi_device_unbusy() will wake us again when these
2101 * pending commands complete.
2103 spin_lock_irqsave(shost->host_lock, flags);
2104 if (shost->host_eh_scheduled)
2105 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2106 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2107 spin_unlock_irqrestore(shost->host_lock, flags);
2111 * scsi_eh_ready_devs - check device ready state and recover if not.
2112 * @shost: host to be recovered.
2113 * @work_q: &list_head for pending commands.
2114 * @done_q: &list_head for processed commands.
2116 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2117 struct list_head *work_q,
2118 struct list_head *done_q)
2120 if (!scsi_eh_stu(shost, work_q, done_q))
2121 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2122 if (!scsi_eh_target_reset(shost, work_q, done_q))
2123 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2124 if (!scsi_eh_host_reset(shost, work_q, done_q))
2125 scsi_eh_offline_sdevs(work_q,
2128 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2131 * scsi_eh_flush_done_q - finish processed commands or retry them.
2132 * @done_q: list_head of processed commands.
2134 void scsi_eh_flush_done_q(struct list_head *done_q)
2136 struct scsi_cmnd *scmd, *next;
2138 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2139 list_del_init(&scmd->eh_entry);
2140 if (scsi_device_online(scmd->device) &&
2141 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2142 scsi_eh_should_retry_cmd(scmd)) {
2143 SCSI_LOG_ERROR_RECOVERY(3,
2144 scmd_printk(KERN_INFO, scmd,
2145 "%s: flush retry cmd\n",
2147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2150 * If just we got sense for the device (called
2151 * scsi_eh_get_sense), scmd->result is already
2152 * set, do not set DID_TIME_OUT.
2155 scmd->result |= (DID_TIME_OUT << 16);
2156 SCSI_LOG_ERROR_RECOVERY(3,
2157 scmd_printk(KERN_INFO, scmd,
2158 "%s: flush finish cmd\n",
2160 scsi_finish_command(scmd);
2164 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2167 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2168 * @shost: Host to unjam.
2171 * When we come in here, we *know* that all commands on the bus have
2172 * either completed, failed or timed out. we also know that no further
2173 * commands are being sent to the host, so things are relatively quiet
2174 * and we have freedom to fiddle with things as we wish.
2176 * This is only the *default* implementation. it is possible for
2177 * individual drivers to supply their own version of this function, and
2178 * if the maintainer wishes to do this, it is strongly suggested that
2179 * this function be taken as a template and modified. this function
2180 * was designed to correctly handle problems for about 95% of the
2181 * different cases out there, and it should always provide at least a
2182 * reasonable amount of error recovery.
2184 * Any command marked 'failed' or 'timeout' must eventually have
2185 * scsi_finish_cmd() called for it. we do all of the retry stuff
2186 * here, so when we restart the host after we return it should have an
2189 static void scsi_unjam_host(struct Scsi_Host *shost)
2191 unsigned long flags;
2192 LIST_HEAD(eh_work_q);
2193 LIST_HEAD(eh_done_q);
2195 spin_lock_irqsave(shost->host_lock, flags);
2196 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2197 spin_unlock_irqrestore(shost->host_lock, flags);
2199 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2201 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2202 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2204 spin_lock_irqsave(shost->host_lock, flags);
2205 if (shost->eh_deadline != -1)
2206 shost->last_reset = 0;
2207 spin_unlock_irqrestore(shost->host_lock, flags);
2208 scsi_eh_flush_done_q(&eh_done_q);
2212 * scsi_error_handler - SCSI error handler thread
2213 * @data: Host for which we are running.
2216 * This is the main error handling loop. This is run as a kernel thread
2217 * for every SCSI host and handles all error handling activity.
2219 int scsi_error_handler(void *data)
2221 struct Scsi_Host *shost = data;
2224 * We use TASK_INTERRUPTIBLE so that the thread is not
2225 * counted against the load average as a running process.
2226 * We never actually get interrupted because kthread_run
2227 * disables signal delivery for the created thread.
2231 * The sequence in kthread_stop() sets the stop flag first
2232 * then wakes the process. To avoid missed wakeups, the task
2233 * should always be in a non running state before the stop
2236 set_current_state(TASK_INTERRUPTIBLE);
2237 if (kthread_should_stop())
2240 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2241 shost->host_failed != scsi_host_busy(shost)) {
2242 SCSI_LOG_ERROR_RECOVERY(1,
2243 shost_printk(KERN_INFO, shost,
2244 "scsi_eh_%d: sleeping\n",
2250 __set_current_state(TASK_RUNNING);
2251 SCSI_LOG_ERROR_RECOVERY(1,
2252 shost_printk(KERN_INFO, shost,
2253 "scsi_eh_%d: waking up %d/%d/%d\n",
2254 shost->host_no, shost->host_eh_scheduled,
2256 scsi_host_busy(shost)));
2259 * We have a host that is failing for some reason. Figure out
2260 * what we need to do to get it up and online again (if we can).
2261 * If we fail, we end up taking the thing offline.
2263 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2264 SCSI_LOG_ERROR_RECOVERY(1,
2265 shost_printk(KERN_ERR, shost,
2266 "scsi_eh_%d: unable to autoresume\n",
2271 if (shost->transportt->eh_strategy_handler)
2272 shost->transportt->eh_strategy_handler(shost);
2274 scsi_unjam_host(shost);
2276 /* All scmds have been handled */
2277 shost->host_failed = 0;
2280 * Note - if the above fails completely, the action is to take
2281 * individual devices offline and flush the queue of any
2282 * outstanding requests that may have been pending. When we
2283 * restart, we restart any I/O to any other devices on the bus
2284 * which are still online.
2286 scsi_restart_operations(shost);
2287 if (!shost->eh_noresume)
2288 scsi_autopm_put_host(shost);
2290 __set_current_state(TASK_RUNNING);
2292 SCSI_LOG_ERROR_RECOVERY(1,
2293 shost_printk(KERN_INFO, shost,
2294 "Error handler scsi_eh_%d exiting\n",
2296 shost->ehandler = NULL;
2301 * Function: scsi_report_bus_reset()
2303 * Purpose: Utility function used by low-level drivers to report that
2304 * they have observed a bus reset on the bus being handled.
2306 * Arguments: shost - Host in question
2307 * channel - channel on which reset was observed.
2311 * Lock status: Host lock must be held.
2313 * Notes: This only needs to be called if the reset is one which
2314 * originates from an unknown location. Resets originated
2315 * by the mid-level itself don't need to call this, but there
2316 * should be no harm.
2318 * The main purpose of this is to make sure that a CHECK_CONDITION
2319 * is properly treated.
2321 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2323 struct scsi_device *sdev;
2325 __shost_for_each_device(sdev, shost) {
2326 if (channel == sdev_channel(sdev))
2327 __scsi_report_device_reset(sdev, NULL);
2330 EXPORT_SYMBOL(scsi_report_bus_reset);
2333 * Function: scsi_report_device_reset()
2335 * Purpose: Utility function used by low-level drivers to report that
2336 * they have observed a device reset on the device being handled.
2338 * Arguments: shost - Host in question
2339 * channel - channel on which reset was observed
2340 * target - target on which reset was observed
2344 * Lock status: Host lock must be held
2346 * Notes: This only needs to be called if the reset is one which
2347 * originates from an unknown location. Resets originated
2348 * by the mid-level itself don't need to call this, but there
2349 * should be no harm.
2351 * The main purpose of this is to make sure that a CHECK_CONDITION
2352 * is properly treated.
2354 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2356 struct scsi_device *sdev;
2358 __shost_for_each_device(sdev, shost) {
2359 if (channel == sdev_channel(sdev) &&
2360 target == sdev_id(sdev))
2361 __scsi_report_device_reset(sdev, NULL);
2364 EXPORT_SYMBOL(scsi_report_device_reset);
2367 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2368 * @dev: scsi_device to operate on
2369 * @arg: reset type (see sg.h)
2372 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2374 struct scsi_cmnd *scmd;
2375 struct Scsi_Host *shost = dev->host;
2377 unsigned long flags;
2379 enum scsi_disposition rtn;
2381 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2384 error = get_user(val, arg);
2388 if (scsi_autopm_get_host(shost) < 0)
2392 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2393 shost->hostt->cmd_size, GFP_KERNEL);
2395 goto out_put_autopm_host;
2396 blk_rq_init(NULL, rq);
2398 scmd = (struct scsi_cmnd *)(rq + 1);
2399 scsi_init_command(dev, scmd);
2400 scmd->cmnd = scsi_req(rq)->cmd;
2402 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL;
2403 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2407 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2409 spin_lock_irqsave(shost->host_lock, flags);
2410 shost->tmf_in_progress = 1;
2411 spin_unlock_irqrestore(shost->host_lock, flags);
2413 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2414 case SG_SCSI_RESET_NOTHING:
2417 case SG_SCSI_RESET_DEVICE:
2418 rtn = scsi_try_bus_device_reset(scmd);
2419 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2422 case SG_SCSI_RESET_TARGET:
2423 rtn = scsi_try_target_reset(scmd);
2424 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2427 case SG_SCSI_RESET_BUS:
2428 rtn = scsi_try_bus_reset(scmd);
2429 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2432 case SG_SCSI_RESET_HOST:
2433 rtn = scsi_try_host_reset(scmd);
2442 error = (rtn == SUCCESS) ? 0 : -EIO;
2444 spin_lock_irqsave(shost->host_lock, flags);
2445 shost->tmf_in_progress = 0;
2446 spin_unlock_irqrestore(shost->host_lock, flags);
2449 * be sure to wake up anyone who was sleeping or had their queue
2450 * suspended while we performed the TMF.
2452 SCSI_LOG_ERROR_RECOVERY(3,
2453 shost_printk(KERN_INFO, shost,
2454 "waking up host to restart after TMF\n"));
2456 wake_up(&shost->host_wait);
2457 scsi_run_host_queues(shost);
2461 out_put_autopm_host:
2462 scsi_autopm_put_host(shost);
2466 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2467 struct scsi_sense_hdr *sshdr)
2469 return scsi_normalize_sense(cmd->sense_buffer,
2470 SCSI_SENSE_BUFFERSIZE, sshdr);
2472 EXPORT_SYMBOL(scsi_command_normalize_sense);
2475 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2476 * @sense_buffer: byte array of sense data
2477 * @sb_len: number of valid bytes in sense_buffer
2478 * @info_out: pointer to 64 integer where 8 or 4 byte information
2479 * field will be placed if found.
2482 * true if information field found, false if not found.
2484 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2491 switch (sense_buffer[0] & 0x7f) {
2494 if (sense_buffer[0] & 0x80) {
2495 *info_out = get_unaligned_be32(&sense_buffer[3]);
2501 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2503 if (ucp && (0xa == ucp[1])) {
2504 *info_out = get_unaligned_be64(&ucp[4]);
2512 EXPORT_SYMBOL(scsi_get_sense_info_fld);