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);
137 * scmd_eh_abort_handler - Handle command aborts
138 * @work: command to be aborted.
140 * Note: this function must be called only for a command that has timed out.
141 * Because the block layer marks a request as complete before it calls
142 * scsi_timeout(), a .scsi_done() call from the LLD for a command that has
143 * timed out do not have any effect. Hence it is safe to call
144 * scsi_finish_command() from this function.
147 scmd_eh_abort_handler(struct work_struct *work)
149 struct scsi_cmnd *scmd =
150 container_of(work, struct scsi_cmnd, abort_work.work);
151 struct scsi_device *sdev = scmd->device;
152 struct Scsi_Host *shost = sdev->host;
153 enum scsi_disposition rtn;
156 if (scsi_host_eh_past_deadline(shost)) {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
159 "eh timeout, not aborting\n"));
163 SCSI_LOG_ERROR_RECOVERY(3,
164 scmd_printk(KERN_INFO, scmd,
165 "aborting command\n"));
166 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
167 if (rtn != SUCCESS) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
171 (rtn == FAST_IO_FAIL) ?
172 "not send" : "failed"));
175 set_host_byte(scmd, DID_TIME_OUT);
176 if (scsi_host_eh_past_deadline(shost)) {
177 SCSI_LOG_ERROR_RECOVERY(3,
178 scmd_printk(KERN_INFO, scmd,
179 "eh timeout, not retrying "
180 "aborted command\n"));
184 spin_lock_irqsave(shost->host_lock, flags);
185 list_del_init(&scmd->eh_entry);
188 * If the abort succeeds, and there is no further
189 * EH action, clear the ->last_reset time.
191 if (list_empty(&shost->eh_abort_list) &&
192 list_empty(&shost->eh_cmd_q))
193 if (shost->eh_deadline != -1)
194 shost->last_reset = 0;
196 spin_unlock_irqrestore(shost->host_lock, flags);
198 if (!scsi_noretry_cmd(scmd) &&
199 scsi_cmd_retry_allowed(scmd) &&
200 scsi_eh_should_retry_cmd(scmd)) {
201 SCSI_LOG_ERROR_RECOVERY(3,
202 scmd_printk(KERN_WARNING, scmd,
203 "retry aborted command\n"));
204 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_WARNING, scmd,
208 "finish aborted command\n"));
209 scsi_finish_command(scmd);
214 spin_lock_irqsave(shost->host_lock, flags);
215 list_del_init(&scmd->eh_entry);
216 spin_unlock_irqrestore(shost->host_lock, flags);
218 scsi_eh_scmd_add(scmd);
222 * scsi_abort_command - schedule a command abort
223 * @scmd: scmd to abort.
225 * We only need to abort commands after a command timeout
228 scsi_abort_command(struct scsi_cmnd *scmd)
230 struct scsi_device *sdev = scmd->device;
231 struct Scsi_Host *shost = sdev->host;
234 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
236 * Retry after abort failed, escalate to next level.
238 SCSI_LOG_ERROR_RECOVERY(3,
239 scmd_printk(KERN_INFO, scmd,
240 "previous abort failed\n"));
241 BUG_ON(delayed_work_pending(&scmd->abort_work));
245 spin_lock_irqsave(shost->host_lock, flags);
246 if (shost->eh_deadline != -1 && !shost->last_reset)
247 shost->last_reset = jiffies;
248 BUG_ON(!list_empty(&scmd->eh_entry));
249 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list);
250 spin_unlock_irqrestore(shost->host_lock, flags);
252 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
253 SCSI_LOG_ERROR_RECOVERY(3,
254 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
255 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
260 * scsi_eh_reset - call into ->eh_action to reset internal counters
261 * @scmd: scmd to run eh on.
263 * The scsi driver might be carrying internal state about the
264 * devices, so we need to call into the driver to reset the
265 * internal state once the error handler is started.
267 static void scsi_eh_reset(struct scsi_cmnd *scmd)
269 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
270 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
272 sdrv->eh_reset(scmd);
276 static void scsi_eh_inc_host_failed(struct rcu_head *head)
278 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
279 struct Scsi_Host *shost = scmd->device->host;
282 spin_lock_irqsave(shost->host_lock, flags);
283 shost->host_failed++;
284 scsi_eh_wakeup(shost);
285 spin_unlock_irqrestore(shost->host_lock, flags);
289 * scsi_eh_scmd_add - add scsi cmd to error handling.
290 * @scmd: scmd to run eh on.
292 void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
294 struct Scsi_Host *shost = scmd->device->host;
298 WARN_ON_ONCE(!shost->ehandler);
300 spin_lock_irqsave(shost->host_lock, flags);
301 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
302 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
305 if (shost->eh_deadline != -1 && !shost->last_reset)
306 shost->last_reset = jiffies;
309 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
310 spin_unlock_irqrestore(shost->host_lock, flags);
312 * Ensure that all tasks observe the host state change before the
313 * host_failed change.
315 call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed);
319 * scsi_timeout - Timeout function for normal scsi commands.
320 * @req: request that is timing out.
323 * We do not need to lock this. There is the potential for a race
324 * only in that the normal completion handling might run, but if the
325 * normal completion function determines that the timer has already
326 * fired, then it mustn't do anything.
328 enum blk_eh_timer_return scsi_timeout(struct request *req)
330 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
331 struct Scsi_Host *host = scmd->device->host;
333 trace_scsi_dispatch_cmd_timeout(scmd);
334 scsi_log_completion(scmd, TIMEOUT_ERROR);
336 atomic_inc(&scmd->device->iotmo_cnt);
337 if (host->eh_deadline != -1 && !host->last_reset)
338 host->last_reset = jiffies;
340 if (host->hostt->eh_timed_out) {
341 switch (host->hostt->eh_timed_out(scmd)) {
344 case SCSI_EH_RESET_TIMER:
345 return BLK_EH_RESET_TIMER;
346 case SCSI_EH_NOT_HANDLED:
352 * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify
355 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
357 atomic_inc(&scmd->device->iodone_cnt);
358 if (scsi_abort_command(scmd) != SUCCESS) {
359 set_host_byte(scmd, DID_TIME_OUT);
360 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 "LUN assignments on this target have "
466 "changed. The Linux SCSI layer does not "
467 "automatically remap LUN assignments.\n");
468 } else if (sshdr->asc == 0x3f)
469 sdev_printk(KERN_WARNING, sdev,
470 "Operating parameters on this target have "
471 "changed. The Linux SCSI layer does not "
472 "automatically adjust these parameters.\n");
474 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
475 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
476 sdev_printk(KERN_WARNING, sdev,
477 "Warning! Received an indication that the "
478 "LUN reached a thin provisioning soft "
482 if (sshdr->asc == 0x29) {
483 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
485 * Do not print message if it is an expected side-effect
488 if (!sdev->silence_suspend)
489 sdev_printk(KERN_WARNING, sdev,
490 "Power-on or device reset occurred\n");
493 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
494 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
495 sdev_printk(KERN_WARNING, sdev,
496 "Mode parameters changed");
497 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
498 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
499 sdev_printk(KERN_WARNING, sdev,
500 "Asymmetric access state changed");
501 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
502 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
503 sdev_printk(KERN_WARNING, sdev,
504 "Capacity data has changed");
505 } else if (sshdr->asc == 0x2a)
506 sdev_printk(KERN_WARNING, sdev,
507 "Parameters changed");
510 if (evt_type != SDEV_EVT_MAXBITS) {
511 set_bit(evt_type, sdev->pending_events);
512 schedule_work(&sdev->event_work);
516 static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status)
518 cmd->result = (cmd->result & 0xffff00ff) | (status << 8);
522 * scsi_check_sense - Examine scsi cmd sense
523 * @scmd: Cmd to have sense checked.
526 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
529 * When a deferred error is detected the current command has
530 * not been executed and needs retrying.
532 enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
534 struct scsi_device *sdev = scmd->device;
535 struct scsi_sense_hdr sshdr;
537 if (! scsi_command_normalize_sense(scmd, &sshdr))
538 return FAILED; /* no valid sense data */
540 scsi_report_sense(sdev, &sshdr);
542 if (scsi_sense_is_deferred(&sshdr))
545 if (sdev->handler && sdev->handler->check_sense) {
546 enum scsi_disposition rc;
548 rc = sdev->handler->check_sense(sdev, &sshdr);
549 if (rc != SCSI_RETURN_NOT_HANDLED)
551 /* handler does not care. Drop down to default handling */
554 if (scmd->cmnd[0] == TEST_UNIT_READY &&
555 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER)
557 * nasty: for mid-layer issued TURs, we need to return the
558 * actual sense data without any recovery attempt. For eh
559 * issued ones, we need to try to recover and interpret
564 * Previous logic looked for FILEMARK, EOM or ILI which are
565 * mainly associated with tapes and returned SUCCESS.
567 if (sshdr.response_code == 0x70) {
569 if (scmd->sense_buffer[2] & 0xe0)
573 * descriptor format: look for "stream commands sense data
574 * descriptor" (see SSC-3). Assume single sense data
575 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
577 if ((sshdr.additional_length > 3) &&
578 (scmd->sense_buffer[8] == 0x4) &&
579 (scmd->sense_buffer[11] & 0xe0))
583 switch (sshdr.sense_key) {
586 case RECOVERED_ERROR:
587 return /* soft_error */ SUCCESS;
589 case ABORTED_COMMAND:
590 if (sshdr.asc == 0x10) /* DIF */
593 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
594 return ADD_TO_MLQUEUE;
595 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
596 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
597 return ADD_TO_MLQUEUE;
603 * if we are expecting a cc/ua because of a bus reset that we
604 * performed, treat this just as a retry. otherwise this is
605 * information that we should pass up to the upper-level driver
606 * so that we can deal with it there.
608 if (scmd->device->expecting_cc_ua) {
610 * Because some device does not queue unit
611 * attentions correctly, we carefully check
612 * additional sense code and qualifier so as
613 * not to squash media change unit attention.
615 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
616 scmd->device->expecting_cc_ua = 0;
621 * we might also expect a cc/ua if another LUN on the target
622 * reported a UA with an ASC/ASCQ of 3F 0E -
623 * REPORTED LUNS DATA HAS CHANGED.
625 if (scmd->device->sdev_target->expecting_lun_change &&
626 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
629 * if the device is in the process of becoming ready, we
632 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
635 * if the device is not started, we need to wake
636 * the error handler to start the motor
638 if (scmd->device->allow_restart &&
639 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
642 * Pass the UA upwards for a determination in the completion
647 /* these are not supported */
649 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
650 /* Thin provisioning hard threshold reached */
651 set_scsi_ml_byte(scmd, SCSIML_STAT_NOSPC);
656 case VOLUME_OVERFLOW:
659 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
663 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
664 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
665 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
666 set_scsi_ml_byte(scmd, SCSIML_STAT_MED_ERROR);
672 if (scmd->device->retry_hwerror)
673 return ADD_TO_MLQUEUE;
675 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
678 case ILLEGAL_REQUEST:
679 if (sshdr.asc == 0x20 || /* Invalid command operation code */
680 sshdr.asc == 0x21 || /* Logical block address out of range */
681 sshdr.asc == 0x22 || /* Invalid function */
682 sshdr.asc == 0x24 || /* Invalid field in cdb */
683 sshdr.asc == 0x26 || /* Parameter value invalid */
684 sshdr.asc == 0x27) { /* Write protected */
685 set_scsi_ml_byte(scmd, SCSIML_STAT_TGT_FAILURE);
693 EXPORT_SYMBOL_GPL(scsi_check_sense);
695 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
697 struct scsi_host_template *sht = sdev->host->hostt;
698 struct scsi_device *tmp_sdev;
700 if (!sht->track_queue_depth ||
701 sdev->queue_depth >= sdev->max_queue_depth)
704 if (time_before(jiffies,
705 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
708 if (time_before(jiffies,
709 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
713 * Walk all devices of a target and do
716 shost_for_each_device(tmp_sdev, sdev->host) {
717 if (tmp_sdev->channel != sdev->channel ||
718 tmp_sdev->id != sdev->id ||
719 tmp_sdev->queue_depth == sdev->max_queue_depth)
722 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
723 sdev->last_queue_ramp_up = jiffies;
727 static void scsi_handle_queue_full(struct scsi_device *sdev)
729 struct scsi_host_template *sht = sdev->host->hostt;
730 struct scsi_device *tmp_sdev;
732 if (!sht->track_queue_depth)
735 shost_for_each_device(tmp_sdev, sdev->host) {
736 if (tmp_sdev->channel != sdev->channel ||
737 tmp_sdev->id != sdev->id)
740 * We do not know the number of commands that were at
741 * the device when we got the queue full so we start
742 * from the highest possible value and work our way down.
744 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
749 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
750 * @scmd: SCSI cmd to examine.
753 * This is *only* called when we are examining the status of commands
754 * queued during error recovery. the main difference here is that we
755 * don't allow for the possibility of retries here, and we are a lot
756 * more restrictive about what we consider acceptable.
758 static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
761 * first check the host byte, to see if there is anything in there
762 * that would indicate what we need to do.
764 if (host_byte(scmd->result) == DID_RESET) {
766 * rats. we are already in the error handler, so we now
767 * get to try and figure out what to do next. if the sense
768 * is valid, we have a pretty good idea of what to do.
769 * if not, we mark it as FAILED.
771 return scsi_check_sense(scmd);
773 if (host_byte(scmd->result) != DID_OK)
777 * now, check the status byte to see if this indicates
780 switch (get_status_byte(scmd)) {
782 scsi_handle_queue_ramp_up(scmd->device);
784 case SAM_STAT_COMMAND_TERMINATED:
786 case SAM_STAT_CHECK_CONDITION:
787 return scsi_check_sense(scmd);
788 case SAM_STAT_CONDITION_MET:
789 case SAM_STAT_INTERMEDIATE:
790 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
792 * who knows? FIXME(eric)
795 case SAM_STAT_RESERVATION_CONFLICT:
796 if (scmd->cmnd[0] == TEST_UNIT_READY)
797 /* it is a success, we probed the device and
800 /* otherwise, we failed to send the command */
802 case SAM_STAT_TASK_SET_FULL:
803 scsi_handle_queue_full(scmd->device);
814 * scsi_eh_done - Completion function for error handling.
815 * @scmd: Cmd that is done.
817 void scsi_eh_done(struct scsi_cmnd *scmd)
819 struct completion *eh_action;
821 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
822 "%s result: %x\n", __func__, scmd->result));
824 eh_action = scmd->device->host->eh_action;
830 * scsi_try_host_reset - ask host adapter to reset itself
831 * @scmd: SCSI cmd to send host reset.
833 static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
836 enum scsi_disposition rtn;
837 struct Scsi_Host *host = scmd->device->host;
838 struct scsi_host_template *hostt = host->hostt;
840 SCSI_LOG_ERROR_RECOVERY(3,
841 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
843 if (!hostt->eh_host_reset_handler)
846 rtn = hostt->eh_host_reset_handler(scmd);
848 if (rtn == SUCCESS) {
849 if (!hostt->skip_settle_delay)
850 ssleep(HOST_RESET_SETTLE_TIME);
851 spin_lock_irqsave(host->host_lock, flags);
852 scsi_report_bus_reset(host, scmd_channel(scmd));
853 spin_unlock_irqrestore(host->host_lock, flags);
860 * scsi_try_bus_reset - ask host to perform a bus reset
861 * @scmd: SCSI cmd to send bus reset.
863 static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
866 enum scsi_disposition rtn;
867 struct Scsi_Host *host = scmd->device->host;
868 struct scsi_host_template *hostt = host->hostt;
870 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
871 "%s: Snd Bus RST\n", __func__));
873 if (!hostt->eh_bus_reset_handler)
876 rtn = hostt->eh_bus_reset_handler(scmd);
878 if (rtn == SUCCESS) {
879 if (!hostt->skip_settle_delay)
880 ssleep(BUS_RESET_SETTLE_TIME);
881 spin_lock_irqsave(host->host_lock, flags);
882 scsi_report_bus_reset(host, scmd_channel(scmd));
883 spin_unlock_irqrestore(host->host_lock, flags);
889 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
892 sdev->expecting_cc_ua = 1;
896 * scsi_try_target_reset - Ask host to perform a target reset
897 * @scmd: SCSI cmd used to send a target reset
900 * There is no timeout for this operation. if this operation is
901 * unreliable for a given host, then the host itself needs to put a
902 * timer on it, and set the host back to a consistent state prior to
905 static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
908 enum scsi_disposition rtn;
909 struct Scsi_Host *host = scmd->device->host;
910 struct scsi_host_template *hostt = host->hostt;
912 if (!hostt->eh_target_reset_handler)
915 rtn = hostt->eh_target_reset_handler(scmd);
916 if (rtn == SUCCESS) {
917 spin_lock_irqsave(host->host_lock, flags);
918 __starget_for_each_device(scsi_target(scmd->device), NULL,
919 __scsi_report_device_reset);
920 spin_unlock_irqrestore(host->host_lock, flags);
927 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
928 * @scmd: SCSI cmd used to send BDR
931 * There is no timeout for this operation. if this operation is
932 * unreliable for a given host, then the host itself needs to put a
933 * timer on it, and set the host back to a consistent state prior to
936 static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
938 enum scsi_disposition rtn;
939 struct scsi_host_template *hostt = scmd->device->host->hostt;
941 if (!hostt->eh_device_reset_handler)
944 rtn = hostt->eh_device_reset_handler(scmd);
946 __scsi_report_device_reset(scmd->device, NULL);
951 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
952 * @hostt: SCSI driver host template
953 * @scmd: SCSI cmd used to send a target reset
956 * SUCCESS, FAILED, or FAST_IO_FAIL
959 * SUCCESS does not necessarily indicate that the command
960 * has been aborted; it only indicates that the LLDDs
961 * has cleared all references to that command.
962 * LLDDs should return FAILED only if an abort was required
963 * but could not be executed. LLDDs should return FAST_IO_FAIL
964 * if the device is temporarily unavailable (eg due to a
965 * link down on FibreChannel)
967 static enum scsi_disposition
968 scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
970 if (!hostt->eh_abort_handler)
973 return hostt->eh_abort_handler(scmd);
976 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
978 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
979 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
980 if (scsi_try_target_reset(scmd) != SUCCESS)
981 if (scsi_try_bus_reset(scmd) != SUCCESS)
982 scsi_try_host_reset(scmd);
986 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
987 * @scmd: SCSI command structure to hijack
988 * @ses: structure to save restore information
989 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
990 * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE)
991 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
993 * This function is used to save a scsi command information before re-execution
994 * as part of the error recovery process. If @sense_bytes is 0 the command
995 * sent must be one that does not transfer any data. If @sense_bytes != 0
996 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
997 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
999 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
1000 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1002 struct scsi_device *sdev = scmd->device;
1005 * We need saved copies of a number of fields - this is because
1006 * error handling may need to overwrite these with different values
1007 * to run different commands, and once error handling is complete,
1008 * we will need to restore these values prior to running the actual
1011 ses->cmd_len = scmd->cmd_len;
1012 ses->data_direction = scmd->sc_data_direction;
1013 ses->sdb = scmd->sdb;
1014 ses->result = scmd->result;
1015 ses->resid_len = scmd->resid_len;
1016 ses->underflow = scmd->underflow;
1017 ses->prot_op = scmd->prot_op;
1018 ses->eh_eflags = scmd->eh_eflags;
1020 scmd->prot_op = SCSI_PROT_NORMAL;
1021 scmd->eh_eflags = 0;
1022 memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd));
1023 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
1024 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1026 scmd->resid_len = 0;
1029 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1031 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
1033 scmd->sdb.table.sgl = &ses->sense_sgl;
1034 scmd->sc_data_direction = DMA_FROM_DEVICE;
1035 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
1036 scmd->cmnd[0] = REQUEST_SENSE;
1037 scmd->cmnd[4] = scmd->sdb.length;
1038 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1040 scmd->sc_data_direction = DMA_NONE;
1042 BUG_ON(cmnd_size > sizeof(scmd->cmnd));
1043 memcpy(scmd->cmnd, cmnd, cmnd_size);
1044 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1048 scmd->underflow = 0;
1050 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1051 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1052 (sdev->lun << 5 & 0xe0);
1055 * Zero the sense buffer. The scsi spec mandates that any
1056 * untransferred sense data should be interpreted as being zero.
1058 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1060 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1063 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1064 * @scmd: SCSI command structure to restore
1065 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1067 * Undo any damage done by above scsi_eh_prep_cmnd().
1069 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1072 * Restore original data
1074 scmd->cmd_len = ses->cmd_len;
1075 memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd));
1076 scmd->sc_data_direction = ses->data_direction;
1077 scmd->sdb = ses->sdb;
1078 scmd->result = ses->result;
1079 scmd->resid_len = ses->resid_len;
1080 scmd->underflow = ses->underflow;
1081 scmd->prot_op = ses->prot_op;
1082 scmd->eh_eflags = ses->eh_eflags;
1084 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1087 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1088 * @scmd: SCSI command structure to hijack
1089 * @cmnd: CDB to send
1090 * @cmnd_size: size in bytes of @cmnd
1091 * @timeout: timeout for this request
1092 * @sense_bytes: size of sense data to copy or 0
1094 * This function is used to send a scsi command down to a target device
1095 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1098 * SUCCESS or FAILED or NEEDS_RETRY
1100 static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1101 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
1103 struct scsi_device *sdev = scmd->device;
1104 struct Scsi_Host *shost = sdev->host;
1105 DECLARE_COMPLETION_ONSTACK(done);
1106 unsigned long timeleft = timeout, delay;
1107 struct scsi_eh_save ses;
1108 const unsigned long stall_for = msecs_to_jiffies(100);
1112 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1113 shost->eh_action = &done;
1115 scsi_log_send(scmd);
1116 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
1119 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1120 * change the SCSI device state after we have examined it and before
1121 * .queuecommand() is called.
1123 mutex_lock(&sdev->state_mutex);
1124 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1125 mutex_unlock(&sdev->state_mutex);
1126 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1127 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1129 delay = min(timeleft, stall_for);
1131 msleep(jiffies_to_msecs(delay));
1132 mutex_lock(&sdev->state_mutex);
1134 if (sdev->sdev_state != SDEV_BLOCK)
1135 rtn = shost->hostt->queuecommand(shost, scmd);
1138 mutex_unlock(&sdev->state_mutex);
1141 if (timeleft > stall_for) {
1142 scsi_eh_restore_cmnd(scmd, &ses);
1144 timeleft -= stall_for;
1145 msleep(jiffies_to_msecs(stall_for));
1148 /* signal not to enter either branch of the if () below */
1152 timeleft = wait_for_completion_timeout(&done, timeout);
1156 shost->eh_action = NULL;
1158 scsi_log_completion(scmd, rtn);
1160 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1161 "%s timeleft: %ld\n",
1162 __func__, timeleft));
1165 * If there is time left scsi_eh_done got called, and we will examine
1166 * the actual status codes to see whether the command actually did
1167 * complete normally, else if we have a zero return and no time left,
1168 * the command must still be pending, so abort it and return FAILED.
1169 * If we never actually managed to issue the command, because
1170 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1171 * value above (so don't execute either branch of the if)
1174 rtn = scsi_eh_completed_normally(scmd);
1175 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1176 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1183 case ADD_TO_MLQUEUE:
1190 } else if (rtn != FAILED) {
1191 scsi_abort_eh_cmnd(scmd);
1195 scsi_eh_restore_cmnd(scmd, &ses);
1201 * scsi_request_sense - Request sense data from a particular target.
1202 * @scmd: SCSI cmd for request sense.
1205 * Some hosts automatically obtain this information, others require
1206 * that we obtain it on our own. This function will *not* return until
1207 * the command either times out, or it completes.
1209 static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1211 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1214 static enum scsi_disposition
1215 scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
1217 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
1218 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1219 if (sdrv->eh_action)
1220 rtn = sdrv->eh_action(scmd, rtn);
1226 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1227 * @scmd: Original SCSI cmd that eh has finished.
1228 * @done_q: Queue for processed commands.
1231 * We don't want to use the normal command completion while we are are
1232 * still handling errors - it may cause other commands to be queued,
1233 * and that would disturb what we are doing. Thus we really want to
1234 * keep a list of pending commands for final completion, and once we
1235 * are ready to leave error handling we handle completion for real.
1237 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1239 list_move_tail(&scmd->eh_entry, done_q);
1241 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1244 * scsi_eh_get_sense - Get device sense data.
1245 * @work_q: Queue of commands to process.
1246 * @done_q: Queue of processed commands.
1249 * See if we need to request sense information. if so, then get it
1250 * now, so we have a better idea of what to do.
1253 * This has the unfortunate side effect that if a shost adapter does
1254 * not automatically request sense information, we end up shutting
1255 * it down before we request it.
1257 * All drivers should request sense information internally these days,
1258 * so for now all I have to say is tough noogies if you end up in here.
1260 * XXX: Long term this code should go away, but that needs an audit of
1263 int scsi_eh_get_sense(struct list_head *work_q,
1264 struct list_head *done_q)
1266 struct scsi_cmnd *scmd, *next;
1267 struct Scsi_Host *shost;
1268 enum scsi_disposition rtn;
1271 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1272 * should not get sense.
1274 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1275 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1276 SCSI_SENSE_VALID(scmd))
1279 shost = scmd->device->host;
1280 if (scsi_host_eh_past_deadline(shost)) {
1281 SCSI_LOG_ERROR_RECOVERY(3,
1282 scmd_printk(KERN_INFO, scmd,
1283 "%s: skip request sense, past eh deadline\n",
1287 if (!scsi_status_is_check_condition(scmd->result))
1289 * don't request sense if there's no check condition
1290 * status because the error we're processing isn't one
1291 * that has a sense code (and some devices get
1292 * confused by sense requests out of the blue)
1296 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1297 "%s: requesting sense\n",
1299 rtn = scsi_request_sense(scmd);
1303 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1304 "sense requested, result %x\n", scmd->result));
1305 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1307 rtn = scsi_decide_disposition(scmd);
1310 * if the result was normal, then just pass it along to the
1315 * We don't want this command reissued, just finished
1316 * with the sense data, so set retries to the max
1317 * allowed to ensure it won't get reissued. If the user
1318 * has requested infinite retries, we also want to
1319 * finish this command, so force completion by setting
1320 * retries and allowed to the same value.
1322 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1323 scmd->retries = scmd->allowed = 1;
1325 scmd->retries = scmd->allowed;
1326 else if (rtn != NEEDS_RETRY)
1329 scsi_eh_finish_cmd(scmd, done_q);
1332 return list_empty(work_q);
1334 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1337 * scsi_eh_tur - Send TUR to device.
1338 * @scmd: &scsi_cmnd to send TUR
1341 * 0 - Device is ready. 1 - Device NOT ready.
1343 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1345 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1347 enum scsi_disposition rtn;
1350 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1351 scmd->device->eh_timeout, 0);
1353 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1354 "%s return: %x\n", __func__, rtn));
1369 * scsi_eh_test_devices - check if devices are responding from error recovery.
1370 * @cmd_list: scsi commands in error recovery.
1371 * @work_q: queue for commands which still need more error recovery
1372 * @done_q: queue for commands which are finished
1373 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1376 * Tests if devices are in a working state. Commands to devices now in
1377 * a working state are sent to the done_q while commands to devices which
1378 * are still failing to respond are returned to the work_q for more
1381 static int scsi_eh_test_devices(struct list_head *cmd_list,
1382 struct list_head *work_q,
1383 struct list_head *done_q, int try_stu)
1385 struct scsi_cmnd *scmd, *next;
1386 struct scsi_device *sdev;
1389 while (!list_empty(cmd_list)) {
1390 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1391 sdev = scmd->device;
1394 if (scsi_host_eh_past_deadline(sdev->host)) {
1395 /* Push items back onto work_q */
1396 list_splice_init(cmd_list, work_q);
1397 SCSI_LOG_ERROR_RECOVERY(3,
1398 sdev_printk(KERN_INFO, sdev,
1399 "%s: skip test device, past eh deadline",
1405 finish_cmds = !scsi_device_online(scmd->device) ||
1406 (try_stu && !scsi_eh_try_stu(scmd) &&
1407 !scsi_eh_tur(scmd)) ||
1410 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1411 if (scmd->device == sdev) {
1414 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1415 scsi_eh_finish_cmd(scmd, done_q);
1417 list_move_tail(&scmd->eh_entry, work_q);
1420 return list_empty(work_q);
1424 * scsi_eh_try_stu - Send START_UNIT to device.
1425 * @scmd: &scsi_cmnd to send START_UNIT
1428 * 0 - Device is ready. 1 - Device NOT ready.
1430 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1432 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1434 if (scmd->device->allow_restart) {
1436 enum scsi_disposition rtn = NEEDS_RETRY;
1438 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1439 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
1440 scmd->device->eh_timeout, 0);
1450 * scsi_eh_stu - send START_UNIT if needed
1451 * @shost: &scsi host being recovered.
1452 * @work_q: &list_head for pending commands.
1453 * @done_q: &list_head for processed commands.
1456 * If commands are failing due to not ready, initializing command required,
1457 * try revalidating the device, which will end up sending a start unit.
1459 static int scsi_eh_stu(struct Scsi_Host *shost,
1460 struct list_head *work_q,
1461 struct list_head *done_q)
1463 struct scsi_cmnd *scmd, *stu_scmd, *next;
1464 struct scsi_device *sdev;
1466 shost_for_each_device(sdev, shost) {
1467 if (scsi_host_eh_past_deadline(shost)) {
1468 SCSI_LOG_ERROR_RECOVERY(3,
1469 sdev_printk(KERN_INFO, sdev,
1470 "%s: skip START_UNIT, past eh deadline\n",
1472 scsi_device_put(sdev);
1476 list_for_each_entry(scmd, work_q, eh_entry)
1477 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1478 scsi_check_sense(scmd) == FAILED ) {
1486 SCSI_LOG_ERROR_RECOVERY(3,
1487 sdev_printk(KERN_INFO, sdev,
1488 "%s: Sending START_UNIT\n",
1491 if (!scsi_eh_try_stu(stu_scmd)) {
1492 if (!scsi_device_online(sdev) ||
1493 !scsi_eh_tur(stu_scmd)) {
1494 list_for_each_entry_safe(scmd, next,
1496 if (scmd->device == sdev &&
1497 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1498 scsi_eh_finish_cmd(scmd, done_q);
1502 SCSI_LOG_ERROR_RECOVERY(3,
1503 sdev_printk(KERN_INFO, sdev,
1504 "%s: START_UNIT failed\n",
1509 return list_empty(work_q);
1514 * scsi_eh_bus_device_reset - send bdr if needed
1515 * @shost: scsi host being recovered.
1516 * @work_q: &list_head for pending commands.
1517 * @done_q: &list_head for processed commands.
1520 * Try a bus device reset. Still, look to see whether we have multiple
1521 * devices that are jammed or not - if we have multiple devices, it
1522 * makes no sense to try bus_device_reset - we really would need to try
1523 * a bus_reset instead.
1525 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1526 struct list_head *work_q,
1527 struct list_head *done_q)
1529 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1530 struct scsi_device *sdev;
1531 enum scsi_disposition rtn;
1533 shost_for_each_device(sdev, shost) {
1534 if (scsi_host_eh_past_deadline(shost)) {
1535 SCSI_LOG_ERROR_RECOVERY(3,
1536 sdev_printk(KERN_INFO, sdev,
1537 "%s: skip BDR, past eh deadline\n",
1539 scsi_device_put(sdev);
1543 list_for_each_entry(scmd, work_q, eh_entry)
1544 if (scmd->device == sdev) {
1552 SCSI_LOG_ERROR_RECOVERY(3,
1553 sdev_printk(KERN_INFO, sdev,
1554 "%s: Sending BDR\n", current->comm));
1555 rtn = scsi_try_bus_device_reset(bdr_scmd);
1556 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1557 if (!scsi_device_online(sdev) ||
1558 rtn == FAST_IO_FAIL ||
1559 !scsi_eh_tur(bdr_scmd)) {
1560 list_for_each_entry_safe(scmd, next,
1562 if (scmd->device == sdev &&
1563 scsi_eh_action(scmd, rtn) != FAILED)
1564 scsi_eh_finish_cmd(scmd,
1569 SCSI_LOG_ERROR_RECOVERY(3,
1570 sdev_printk(KERN_INFO, sdev,
1571 "%s: BDR failed\n", current->comm));
1575 return list_empty(work_q);
1579 * scsi_eh_target_reset - send target reset if needed
1580 * @shost: scsi host being recovered.
1581 * @work_q: &list_head for pending commands.
1582 * @done_q: &list_head for processed commands.
1585 * Try a target reset.
1587 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1588 struct list_head *work_q,
1589 struct list_head *done_q)
1591 LIST_HEAD(tmp_list);
1592 LIST_HEAD(check_list);
1594 list_splice_init(work_q, &tmp_list);
1596 while (!list_empty(&tmp_list)) {
1597 struct scsi_cmnd *next, *scmd;
1598 enum scsi_disposition rtn;
1601 if (scsi_host_eh_past_deadline(shost)) {
1602 /* push back on work queue for further processing */
1603 list_splice_init(&check_list, work_q);
1604 list_splice_init(&tmp_list, work_q);
1605 SCSI_LOG_ERROR_RECOVERY(3,
1606 shost_printk(KERN_INFO, shost,
1607 "%s: Skip target reset, past eh deadline\n",
1609 return list_empty(work_q);
1612 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1615 SCSI_LOG_ERROR_RECOVERY(3,
1616 shost_printk(KERN_INFO, shost,
1617 "%s: Sending target reset to target %d\n",
1618 current->comm, id));
1619 rtn = scsi_try_target_reset(scmd);
1620 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1621 SCSI_LOG_ERROR_RECOVERY(3,
1622 shost_printk(KERN_INFO, shost,
1623 "%s: Target reset failed"
1625 current->comm, id));
1626 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1627 if (scmd_id(scmd) != id)
1631 list_move_tail(&scmd->eh_entry, &check_list);
1632 else if (rtn == FAST_IO_FAIL)
1633 scsi_eh_finish_cmd(scmd, done_q);
1635 /* push back on work queue for further processing */
1636 list_move(&scmd->eh_entry, work_q);
1640 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1644 * scsi_eh_bus_reset - send a bus reset
1645 * @shost: &scsi host being recovered.
1646 * @work_q: &list_head for pending commands.
1647 * @done_q: &list_head for processed commands.
1649 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1650 struct list_head *work_q,
1651 struct list_head *done_q)
1653 struct scsi_cmnd *scmd, *chan_scmd, *next;
1654 LIST_HEAD(check_list);
1655 unsigned int channel;
1656 enum scsi_disposition rtn;
1659 * we really want to loop over the various channels, and do this on
1660 * a channel by channel basis. we should also check to see if any
1661 * of the failed commands are on soft_reset devices, and if so, skip
1665 for (channel = 0; channel <= shost->max_channel; channel++) {
1666 if (scsi_host_eh_past_deadline(shost)) {
1667 list_splice_init(&check_list, work_q);
1668 SCSI_LOG_ERROR_RECOVERY(3,
1669 shost_printk(KERN_INFO, shost,
1670 "%s: skip BRST, past eh deadline\n",
1672 return list_empty(work_q);
1676 list_for_each_entry(scmd, work_q, eh_entry) {
1677 if (channel == scmd_channel(scmd)) {
1681 * FIXME add back in some support for
1682 * soft_reset devices.
1689 SCSI_LOG_ERROR_RECOVERY(3,
1690 shost_printk(KERN_INFO, shost,
1691 "%s: Sending BRST chan: %d\n",
1692 current->comm, channel));
1693 rtn = scsi_try_bus_reset(chan_scmd);
1694 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1695 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1696 if (channel == scmd_channel(scmd)) {
1697 if (rtn == FAST_IO_FAIL)
1698 scsi_eh_finish_cmd(scmd,
1701 list_move_tail(&scmd->eh_entry,
1706 SCSI_LOG_ERROR_RECOVERY(3,
1707 shost_printk(KERN_INFO, shost,
1708 "%s: BRST failed chan: %d\n",
1709 current->comm, channel));
1712 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1716 * scsi_eh_host_reset - send a host reset
1717 * @shost: host to be reset.
1718 * @work_q: &list_head for pending commands.
1719 * @done_q: &list_head for processed commands.
1721 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1722 struct list_head *work_q,
1723 struct list_head *done_q)
1725 struct scsi_cmnd *scmd, *next;
1726 LIST_HEAD(check_list);
1727 enum scsi_disposition rtn;
1729 if (!list_empty(work_q)) {
1730 scmd = list_entry(work_q->next,
1731 struct scsi_cmnd, eh_entry);
1733 SCSI_LOG_ERROR_RECOVERY(3,
1734 shost_printk(KERN_INFO, shost,
1735 "%s: Sending HRST\n",
1738 rtn = scsi_try_host_reset(scmd);
1739 if (rtn == SUCCESS) {
1740 list_splice_init(work_q, &check_list);
1741 } else if (rtn == FAST_IO_FAIL) {
1742 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1743 scsi_eh_finish_cmd(scmd, done_q);
1746 SCSI_LOG_ERROR_RECOVERY(3,
1747 shost_printk(KERN_INFO, shost,
1748 "%s: HRST failed\n",
1752 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1756 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1757 * @work_q: &list_head for pending commands.
1758 * @done_q: &list_head for processed commands.
1760 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1761 struct list_head *done_q)
1763 struct scsi_cmnd *scmd, *next;
1764 struct scsi_device *sdev;
1766 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1767 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1768 "not ready after error recovery\n");
1769 sdev = scmd->device;
1771 mutex_lock(&sdev->state_mutex);
1772 scsi_device_set_state(sdev, SDEV_OFFLINE);
1773 mutex_unlock(&sdev->state_mutex);
1775 scsi_eh_finish_cmd(scmd, done_q);
1781 * scsi_noretry_cmd - determine if command should be failed fast
1782 * @scmd: SCSI cmd to examine.
1784 bool scsi_noretry_cmd(struct scsi_cmnd *scmd)
1786 struct request *req = scsi_cmd_to_rq(scmd);
1788 switch (host_byte(scmd->result)) {
1794 return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT);
1796 return !!(req->cmd_flags & REQ_FAILFAST_DEV);
1798 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1801 case DID_SOFT_ERROR:
1802 return !!(req->cmd_flags & REQ_FAILFAST_DRIVER);
1805 if (!scsi_status_is_check_condition(scmd->result))
1810 * assume caller has checked sense and determined
1811 * the check condition was retryable.
1813 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
1820 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1821 * @scmd: SCSI cmd to examine.
1824 * This is *only* called when we are examining the status after sending
1825 * out the actual data command. any commands that are queued for error
1826 * recovery (e.g. test_unit_ready) do *not* come through here.
1828 * When this routine returns failed, it means the error handler thread
1829 * is woken. In cases where the error code indicates an error that
1830 * doesn't require the error handler read (i.e. we don't need to
1831 * abort/reset), this function should return SUCCESS.
1833 enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1835 enum scsi_disposition rtn;
1838 * if the device is offline, then we clearly just pass the result back
1839 * up to the top level.
1841 if (!scsi_device_online(scmd->device)) {
1842 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1843 "%s: device offline - report as SUCCESS\n", __func__));
1848 * first check the host byte, to see if there is anything in there
1849 * that would indicate what we need to do.
1851 switch (host_byte(scmd->result)) {
1852 case DID_PASSTHROUGH:
1854 * no matter what, pass this through to the upper layer.
1855 * nuke this special code so that it looks like we are saying
1858 scmd->result &= 0xff00ffff;
1862 * looks good. drop through, and check the next byte.
1866 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1867 set_host_byte(scmd, DID_TIME_OUT);
1871 case DID_NO_CONNECT:
1872 case DID_BAD_TARGET:
1874 * note - this means that we just report the status back
1875 * to the top level driver, not that we actually think
1876 * that it indicates SUCCESS.
1879 case DID_SOFT_ERROR:
1881 * when the low level driver returns did_soft_error,
1882 * it is responsible for keeping an internal retry counter
1883 * in order to avoid endless loops (db)
1890 return ADD_TO_MLQUEUE;
1891 case DID_TRANSPORT_DISRUPTED:
1893 * LLD/transport was disrupted during processing of the IO.
1894 * The transport class is now blocked/blocking,
1895 * and the transport will decide what to do with the IO
1896 * based on its timers and recovery capablilities if
1897 * there are enough retries.
1900 case DID_TRANSPORT_FAILFAST:
1902 * The transport decided to failfast the IO (most likely
1903 * the fast io fail tmo fired), so send IO directly upwards.
1906 case DID_TRANSPORT_MARGINAL:
1908 * caller has decided not to do retries on
1909 * abort success, so send IO directly upwards
1913 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1915 * execute reservation conflict processing code
1925 * when we scan the bus, we get timeout messages for
1926 * these commands if there is no device available.
1927 * other hosts report did_no_connect for the same thing.
1929 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1930 scmd->cmnd[0] == INQUIRY)) {
1942 * check the status byte to see if this indicates anything special.
1944 switch (get_status_byte(scmd)) {
1945 case SAM_STAT_TASK_SET_FULL:
1946 scsi_handle_queue_full(scmd->device);
1948 * the case of trying to send too many commands to a
1949 * tagged queueing device.
1954 * device can't talk to us at the moment. Should only
1955 * occur (SAM-3) when the task queue is empty, so will cause
1956 * the empty queue handling to trigger a stall in the
1959 return ADD_TO_MLQUEUE;
1961 if (scmd->cmnd[0] == REPORT_LUNS)
1962 scmd->device->sdev_target->expecting_lun_change = 0;
1963 scsi_handle_queue_ramp_up(scmd->device);
1965 case SAM_STAT_COMMAND_TERMINATED:
1967 case SAM_STAT_TASK_ABORTED:
1969 case SAM_STAT_CHECK_CONDITION:
1970 rtn = scsi_check_sense(scmd);
1971 if (rtn == NEEDS_RETRY)
1973 /* if rtn == FAILED, we have no sense information;
1974 * returning FAILED will wake the error handler thread
1975 * to collect the sense and redo the decide
1978 case SAM_STAT_CONDITION_MET:
1979 case SAM_STAT_INTERMEDIATE:
1980 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1981 case SAM_STAT_ACA_ACTIVE:
1983 * who knows? FIXME(eric)
1987 case SAM_STAT_RESERVATION_CONFLICT:
1988 sdev_printk(KERN_INFO, scmd->device,
1989 "reservation conflict\n");
1990 set_scsi_ml_byte(scmd, SCSIML_STAT_RESV_CONFLICT);
1991 return SUCCESS; /* causes immediate i/o error */
1997 /* we requeue for retry because the error was retryable, and
1998 * the request was not marked fast fail. Note that above,
1999 * even if the request is marked fast fail, we still requeue
2000 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2001 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
2005 * no more retries - report this one back to upper level.
2011 static enum rq_end_io_ret eh_lock_door_done(struct request *req,
2012 blk_status_t status)
2014 blk_mq_free_request(req);
2015 return RQ_END_IO_NONE;
2019 * scsi_eh_lock_door - Prevent medium removal for the specified device
2020 * @sdev: SCSI device to prevent medium removal
2023 * We must be called from process context.
2026 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2027 * head of the devices request queue, and continue.
2029 static void scsi_eh_lock_door(struct scsi_device *sdev)
2031 struct scsi_cmnd *scmd;
2032 struct request *req;
2034 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
2037 scmd = blk_mq_rq_to_pdu(req);
2039 scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL;
2043 scmd->cmnd[4] = SCSI_REMOVAL_PREVENT;
2045 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
2048 req->rq_flags |= RQF_QUIET;
2049 req->timeout = 10 * HZ;
2050 req->end_io = eh_lock_door_done;
2052 blk_execute_rq_nowait(req, true);
2056 * scsi_restart_operations - restart io operations to the specified host.
2057 * @shost: Host we are restarting.
2060 * When we entered the error handler, we blocked all further i/o to
2061 * this device. we need to 'reverse' this process.
2063 static void scsi_restart_operations(struct Scsi_Host *shost)
2065 struct scsi_device *sdev;
2066 unsigned long flags;
2069 * If the door was locked, we need to insert a door lock request
2070 * onto the head of the SCSI request queue for the device. There
2071 * is no point trying to lock the door of an off-line device.
2073 shost_for_each_device(sdev, shost) {
2074 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2075 scsi_eh_lock_door(sdev);
2076 sdev->was_reset = 0;
2081 * next free up anything directly waiting upon the host. this
2082 * will be requests for character device operations, and also for
2083 * ioctls to queued block devices.
2085 SCSI_LOG_ERROR_RECOVERY(3,
2086 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2088 spin_lock_irqsave(shost->host_lock, flags);
2089 if (scsi_host_set_state(shost, SHOST_RUNNING))
2090 if (scsi_host_set_state(shost, SHOST_CANCEL))
2091 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2092 spin_unlock_irqrestore(shost->host_lock, flags);
2094 wake_up(&shost->host_wait);
2097 * finally we need to re-initiate requests that may be pending. we will
2098 * have had everything blocked while error handling is taking place, and
2099 * now that error recovery is done, we will need to ensure that these
2100 * requests are started.
2102 scsi_run_host_queues(shost);
2105 * if eh is active and host_eh_scheduled is pending we need to re-run
2106 * recovery. we do this check after scsi_run_host_queues() to allow
2107 * everything pent up since the last eh run a chance to make forward
2108 * progress before we sync again. Either we'll immediately re-run
2109 * recovery or scsi_device_unbusy() will wake us again when these
2110 * pending commands complete.
2112 spin_lock_irqsave(shost->host_lock, flags);
2113 if (shost->host_eh_scheduled)
2114 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2115 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2116 spin_unlock_irqrestore(shost->host_lock, flags);
2120 * scsi_eh_ready_devs - check device ready state and recover if not.
2121 * @shost: host to be recovered.
2122 * @work_q: &list_head for pending commands.
2123 * @done_q: &list_head for processed commands.
2125 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2126 struct list_head *work_q,
2127 struct list_head *done_q)
2129 if (!scsi_eh_stu(shost, work_q, done_q))
2130 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2131 if (!scsi_eh_target_reset(shost, work_q, done_q))
2132 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2133 if (!scsi_eh_host_reset(shost, work_q, done_q))
2134 scsi_eh_offline_sdevs(work_q,
2137 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2140 * scsi_eh_flush_done_q - finish processed commands or retry them.
2141 * @done_q: list_head of processed commands.
2143 void scsi_eh_flush_done_q(struct list_head *done_q)
2145 struct scsi_cmnd *scmd, *next;
2147 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2148 list_del_init(&scmd->eh_entry);
2149 if (scsi_device_online(scmd->device) &&
2150 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2151 scsi_eh_should_retry_cmd(scmd)) {
2152 SCSI_LOG_ERROR_RECOVERY(3,
2153 scmd_printk(KERN_INFO, scmd,
2154 "%s: flush retry cmd\n",
2156 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2159 * If just we got sense for the device (called
2160 * scsi_eh_get_sense), scmd->result is already
2161 * set, do not set DID_TIME_OUT.
2164 scmd->result |= (DID_TIME_OUT << 16);
2165 SCSI_LOG_ERROR_RECOVERY(3,
2166 scmd_printk(KERN_INFO, scmd,
2167 "%s: flush finish cmd\n",
2169 scsi_finish_command(scmd);
2173 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2176 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2177 * @shost: Host to unjam.
2180 * When we come in here, we *know* that all commands on the bus have
2181 * either completed, failed or timed out. we also know that no further
2182 * commands are being sent to the host, so things are relatively quiet
2183 * and we have freedom to fiddle with things as we wish.
2185 * This is only the *default* implementation. it is possible for
2186 * individual drivers to supply their own version of this function, and
2187 * if the maintainer wishes to do this, it is strongly suggested that
2188 * this function be taken as a template and modified. this function
2189 * was designed to correctly handle problems for about 95% of the
2190 * different cases out there, and it should always provide at least a
2191 * reasonable amount of error recovery.
2193 * Any command marked 'failed' or 'timeout' must eventually have
2194 * scsi_finish_cmd() called for it. we do all of the retry stuff
2195 * here, so when we restart the host after we return it should have an
2198 static void scsi_unjam_host(struct Scsi_Host *shost)
2200 unsigned long flags;
2201 LIST_HEAD(eh_work_q);
2202 LIST_HEAD(eh_done_q);
2204 spin_lock_irqsave(shost->host_lock, flags);
2205 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2206 spin_unlock_irqrestore(shost->host_lock, flags);
2208 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2210 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2211 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2213 spin_lock_irqsave(shost->host_lock, flags);
2214 if (shost->eh_deadline != -1)
2215 shost->last_reset = 0;
2216 spin_unlock_irqrestore(shost->host_lock, flags);
2217 scsi_eh_flush_done_q(&eh_done_q);
2221 * scsi_error_handler - SCSI error handler thread
2222 * @data: Host for which we are running.
2225 * This is the main error handling loop. This is run as a kernel thread
2226 * for every SCSI host and handles all error handling activity.
2228 int scsi_error_handler(void *data)
2230 struct Scsi_Host *shost = data;
2233 * We use TASK_INTERRUPTIBLE so that the thread is not
2234 * counted against the load average as a running process.
2235 * We never actually get interrupted because kthread_run
2236 * disables signal delivery for the created thread.
2240 * The sequence in kthread_stop() sets the stop flag first
2241 * then wakes the process. To avoid missed wakeups, the task
2242 * should always be in a non running state before the stop
2245 set_current_state(TASK_INTERRUPTIBLE);
2246 if (kthread_should_stop())
2249 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2250 shost->host_failed != scsi_host_busy(shost)) {
2251 SCSI_LOG_ERROR_RECOVERY(1,
2252 shost_printk(KERN_INFO, shost,
2253 "scsi_eh_%d: sleeping\n",
2259 __set_current_state(TASK_RUNNING);
2260 SCSI_LOG_ERROR_RECOVERY(1,
2261 shost_printk(KERN_INFO, shost,
2262 "scsi_eh_%d: waking up %d/%d/%d\n",
2263 shost->host_no, shost->host_eh_scheduled,
2265 scsi_host_busy(shost)));
2268 * We have a host that is failing for some reason. Figure out
2269 * what we need to do to get it up and online again (if we can).
2270 * If we fail, we end up taking the thing offline.
2272 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2273 SCSI_LOG_ERROR_RECOVERY(1,
2274 shost_printk(KERN_ERR, shost,
2275 "scsi_eh_%d: unable to autoresume\n",
2280 if (shost->transportt->eh_strategy_handler)
2281 shost->transportt->eh_strategy_handler(shost);
2283 scsi_unjam_host(shost);
2285 /* All scmds have been handled */
2286 shost->host_failed = 0;
2289 * Note - if the above fails completely, the action is to take
2290 * individual devices offline and flush the queue of any
2291 * outstanding requests that may have been pending. When we
2292 * restart, we restart any I/O to any other devices on the bus
2293 * which are still online.
2295 scsi_restart_operations(shost);
2296 if (!shost->eh_noresume)
2297 scsi_autopm_put_host(shost);
2299 __set_current_state(TASK_RUNNING);
2301 SCSI_LOG_ERROR_RECOVERY(1,
2302 shost_printk(KERN_INFO, shost,
2303 "Error handler scsi_eh_%d exiting\n",
2305 shost->ehandler = NULL;
2310 * Function: scsi_report_bus_reset()
2312 * Purpose: Utility function used by low-level drivers to report that
2313 * they have observed a bus reset on the bus being handled.
2315 * Arguments: shost - Host in question
2316 * channel - channel on which reset was observed.
2320 * Lock status: Host lock must be held.
2322 * Notes: This only needs to be called if the reset is one which
2323 * originates from an unknown location. Resets originated
2324 * by the mid-level itself don't need to call this, but there
2325 * should be no harm.
2327 * The main purpose of this is to make sure that a CHECK_CONDITION
2328 * is properly treated.
2330 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2332 struct scsi_device *sdev;
2334 __shost_for_each_device(sdev, shost) {
2335 if (channel == sdev_channel(sdev))
2336 __scsi_report_device_reset(sdev, NULL);
2339 EXPORT_SYMBOL(scsi_report_bus_reset);
2342 * Function: scsi_report_device_reset()
2344 * Purpose: Utility function used by low-level drivers to report that
2345 * they have observed a device reset on the device being handled.
2347 * Arguments: shost - Host in question
2348 * channel - channel on which reset was observed
2349 * target - target on which reset was observed
2353 * Lock status: Host lock must be held
2355 * Notes: This only needs to be called if the reset is one which
2356 * originates from an unknown location. Resets originated
2357 * by the mid-level itself don't need to call this, but there
2358 * should be no harm.
2360 * The main purpose of this is to make sure that a CHECK_CONDITION
2361 * is properly treated.
2363 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2365 struct scsi_device *sdev;
2367 __shost_for_each_device(sdev, shost) {
2368 if (channel == sdev_channel(sdev) &&
2369 target == sdev_id(sdev))
2370 __scsi_report_device_reset(sdev, NULL);
2373 EXPORT_SYMBOL(scsi_report_device_reset);
2376 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2377 * @dev: scsi_device to operate on
2378 * @arg: reset type (see sg.h)
2381 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2383 struct scsi_cmnd *scmd;
2384 struct Scsi_Host *shost = dev->host;
2386 unsigned long flags;
2388 enum scsi_disposition rtn;
2390 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2393 error = get_user(val, arg);
2397 if (scsi_autopm_get_host(shost) < 0)
2401 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2402 shost->hostt->cmd_size, GFP_KERNEL);
2404 goto out_put_autopm_host;
2405 blk_rq_init(NULL, rq);
2407 scmd = (struct scsi_cmnd *)(rq + 1);
2408 scsi_init_command(dev, scmd);
2410 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL;
2411 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2415 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2417 spin_lock_irqsave(shost->host_lock, flags);
2418 shost->tmf_in_progress = 1;
2419 spin_unlock_irqrestore(shost->host_lock, flags);
2421 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2422 case SG_SCSI_RESET_NOTHING:
2425 case SG_SCSI_RESET_DEVICE:
2426 rtn = scsi_try_bus_device_reset(scmd);
2427 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2430 case SG_SCSI_RESET_TARGET:
2431 rtn = scsi_try_target_reset(scmd);
2432 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2435 case SG_SCSI_RESET_BUS:
2436 rtn = scsi_try_bus_reset(scmd);
2437 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2440 case SG_SCSI_RESET_HOST:
2441 rtn = scsi_try_host_reset(scmd);
2450 error = (rtn == SUCCESS) ? 0 : -EIO;
2452 spin_lock_irqsave(shost->host_lock, flags);
2453 shost->tmf_in_progress = 0;
2454 spin_unlock_irqrestore(shost->host_lock, flags);
2457 * be sure to wake up anyone who was sleeping or had their queue
2458 * suspended while we performed the TMF.
2460 SCSI_LOG_ERROR_RECOVERY(3,
2461 shost_printk(KERN_INFO, shost,
2462 "waking up host to restart after TMF\n"));
2464 wake_up(&shost->host_wait);
2465 scsi_run_host_queues(shost);
2469 out_put_autopm_host:
2470 scsi_autopm_put_host(shost);
2474 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2475 struct scsi_sense_hdr *sshdr)
2477 return scsi_normalize_sense(cmd->sense_buffer,
2478 SCSI_SENSE_BUFFERSIZE, sshdr);
2480 EXPORT_SYMBOL(scsi_command_normalize_sense);
2483 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2484 * @sense_buffer: byte array of sense data
2485 * @sb_len: number of valid bytes in sense_buffer
2486 * @info_out: pointer to 64 integer where 8 or 4 byte information
2487 * field will be placed if found.
2490 * true if information field found, false if not found.
2492 bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2499 switch (sense_buffer[0] & 0x7f) {
2502 if (sense_buffer[0] & 0x80) {
2503 *info_out = get_unaligned_be32(&sense_buffer[3]);
2509 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2511 if (ucp && (0xa == ucp[1])) {
2512 *info_out = get_unaligned_be64(&ucp[4]);
2520 EXPORT_SYMBOL(scsi_get_sense_info_fld);