ALSA: seq: Fix copy_from_user() call inside lock
[linux/fpc-iii.git] / drivers / scsi / scsi_error.c
blob29a67a85ee7119b06f17063177b42a152959be1d
1 /*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13 * minor cleanups.
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
28 #include <linux/jiffies.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 #include "scsi_transport_api.h"
44 #include <trace/events/scsi.h>
46 static void scsi_eh_done(struct scsi_cmnd *scmd);
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
52 #define BUS_RESET_SETTLE_TIME (10)
53 #define HOST_RESET_SETTLE_TIME (10)
55 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
56 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
57 struct scsi_cmnd *);
59 /* called with shost->host_lock held */
60 void scsi_eh_wakeup(struct Scsi_Host *shost)
62 if (atomic_read(&shost->host_busy) == shost->host_failed) {
63 trace_scsi_eh_wakeup(shost);
64 wake_up_process(shost->ehandler);
65 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
66 "Waking error handler thread\n"));
70 /**
71 * scsi_schedule_eh - schedule EH for SCSI host
72 * @shost: SCSI host to invoke error handling on.
74 * Schedule SCSI EH without scmd.
76 void scsi_schedule_eh(struct Scsi_Host *shost)
78 unsigned long flags;
80 spin_lock_irqsave(shost->host_lock, flags);
82 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
83 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
84 shost->host_eh_scheduled++;
85 scsi_eh_wakeup(shost);
88 spin_unlock_irqrestore(shost->host_lock, flags);
90 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
92 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
94 if (!shost->last_reset || shost->eh_deadline == -1)
95 return 0;
98 * 32bit accesses are guaranteed to be atomic
99 * (on all supported architectures), so instead
100 * of using a spinlock we can as well double check
101 * if eh_deadline has been set to 'off' during the
102 * time_before call.
104 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
105 shost->eh_deadline > -1)
106 return 0;
108 return 1;
112 * scmd_eh_abort_handler - Handle command aborts
113 * @work: command to be aborted.
115 void
116 scmd_eh_abort_handler(struct work_struct *work)
118 struct scsi_cmnd *scmd =
119 container_of(work, struct scsi_cmnd, abort_work.work);
120 struct scsi_device *sdev = scmd->device;
121 int rtn;
123 if (scsi_host_eh_past_deadline(sdev->host)) {
124 SCSI_LOG_ERROR_RECOVERY(3,
125 scmd_printk(KERN_INFO, scmd,
126 "scmd %p eh timeout, not aborting\n",
127 scmd));
128 } else {
129 SCSI_LOG_ERROR_RECOVERY(3,
130 scmd_printk(KERN_INFO, scmd,
131 "aborting command %p\n", scmd));
132 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
133 if (rtn == SUCCESS) {
134 set_host_byte(scmd, DID_TIME_OUT);
135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "scmd %p eh timeout, "
139 "not retrying aborted "
140 "command\n", scmd));
141 } else if (!scsi_noretry_cmd(scmd) &&
142 (++scmd->retries <= scmd->allowed)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING, scmd,
145 "scmd %p retry "
146 "aborted command\n", scmd));
147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
148 return;
149 } else {
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING, scmd,
152 "scmd %p finish "
153 "aborted command\n", scmd));
154 scsi_finish_command(scmd);
155 return;
157 } else {
158 SCSI_LOG_ERROR_RECOVERY(3,
159 scmd_printk(KERN_INFO, scmd,
160 "scmd %p abort failed, rtn %d\n",
161 scmd, rtn));
165 if (!scsi_eh_scmd_add(scmd, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING, scmd,
168 "scmd %p terminate "
169 "aborted command\n", scmd));
170 set_host_byte(scmd, DID_TIME_OUT);
171 scsi_finish_command(scmd);
176 * scsi_abort_command - schedule a command abort
177 * @scmd: scmd to abort.
179 * We only need to abort commands after a command timeout
181 static int
182 scsi_abort_command(struct scsi_cmnd *scmd)
184 struct scsi_device *sdev = scmd->device;
185 struct Scsi_Host *shost = sdev->host;
186 unsigned long flags;
188 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
190 * Retry after abort failed, escalate to next level.
192 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
193 SCSI_LOG_ERROR_RECOVERY(3,
194 scmd_printk(KERN_INFO, scmd,
195 "scmd %p previous abort failed\n", scmd));
196 BUG_ON(delayed_work_pending(&scmd->abort_work));
197 return FAILED;
201 * Do not try a command abort if
202 * SCSI EH has already started.
204 spin_lock_irqsave(shost->host_lock, flags);
205 if (scsi_host_in_recovery(shost)) {
206 spin_unlock_irqrestore(shost->host_lock, flags);
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd,
209 "scmd %p not aborting, host in recovery\n",
210 scmd));
211 return FAILED;
214 if (shost->eh_deadline != -1 && !shost->last_reset)
215 shost->last_reset = jiffies;
216 spin_unlock_irqrestore(shost->host_lock, flags);
218 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
219 SCSI_LOG_ERROR_RECOVERY(3,
220 scmd_printk(KERN_INFO, scmd,
221 "scmd %p abort scheduled\n", scmd));
222 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
223 return SUCCESS;
227 * scsi_eh_scmd_add - add scsi cmd to error handling.
228 * @scmd: scmd to run eh on.
229 * @eh_flag: optional SCSI_EH flag.
231 * Return value:
232 * 0 on failure.
234 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
236 struct Scsi_Host *shost = scmd->device->host;
237 unsigned long flags;
238 int ret = 0;
240 if (!shost->ehandler)
241 return 0;
243 spin_lock_irqsave(shost->host_lock, flags);
244 if (scsi_host_set_state(shost, SHOST_RECOVERY))
245 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
246 goto out_unlock;
248 if (shost->eh_deadline != -1 && !shost->last_reset)
249 shost->last_reset = jiffies;
251 ret = 1;
252 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
253 eh_flag &= ~SCSI_EH_CANCEL_CMD;
254 scmd->eh_eflags |= eh_flag;
255 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
256 shost->host_failed++;
257 scsi_eh_wakeup(shost);
258 out_unlock:
259 spin_unlock_irqrestore(shost->host_lock, flags);
260 return ret;
264 * scsi_times_out - Timeout function for normal scsi commands.
265 * @req: request that is timing out.
267 * Notes:
268 * We do not need to lock this. There is the potential for a race
269 * only in that the normal completion handling might run, but if the
270 * normal completion function determines that the timer has already
271 * fired, then it mustn't do anything.
273 enum blk_eh_timer_return scsi_times_out(struct request *req)
275 struct scsi_cmnd *scmd = req->special;
276 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
277 struct Scsi_Host *host = scmd->device->host;
279 trace_scsi_dispatch_cmd_timeout(scmd);
280 scsi_log_completion(scmd, TIMEOUT_ERROR);
282 if (host->eh_deadline != -1 && !host->last_reset)
283 host->last_reset = jiffies;
285 if (host->transportt->eh_timed_out)
286 rtn = host->transportt->eh_timed_out(scmd);
287 else if (host->hostt->eh_timed_out)
288 rtn = host->hostt->eh_timed_out(scmd);
290 if (rtn == BLK_EH_NOT_HANDLED) {
291 if (!host->hostt->no_async_abort &&
292 scsi_abort_command(scmd) == SUCCESS)
293 return BLK_EH_NOT_HANDLED;
295 set_host_byte(scmd, DID_TIME_OUT);
296 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
297 rtn = BLK_EH_HANDLED;
300 return rtn;
304 * scsi_block_when_processing_errors - Prevent cmds from being queued.
305 * @sdev: Device on which we are performing recovery.
307 * Description:
308 * We block until the host is out of error recovery, and then check to
309 * see whether the host or the device is offline.
311 * Return value:
312 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
314 int scsi_block_when_processing_errors(struct scsi_device *sdev)
316 int online;
318 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
320 online = scsi_device_online(sdev);
322 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
323 "%s: rtn: %d\n", __func__, online));
325 return online;
327 EXPORT_SYMBOL(scsi_block_when_processing_errors);
329 #ifdef CONFIG_SCSI_LOGGING
331 * scsi_eh_prt_fail_stats - Log info on failures.
332 * @shost: scsi host being recovered.
333 * @work_q: Queue of scsi cmds to process.
335 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
336 struct list_head *work_q)
338 struct scsi_cmnd *scmd;
339 struct scsi_device *sdev;
340 int total_failures = 0;
341 int cmd_failed = 0;
342 int cmd_cancel = 0;
343 int devices_failed = 0;
345 shost_for_each_device(sdev, shost) {
346 list_for_each_entry(scmd, work_q, eh_entry) {
347 if (scmd->device == sdev) {
348 ++total_failures;
349 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
350 ++cmd_cancel;
351 else
352 ++cmd_failed;
356 if (cmd_cancel || cmd_failed) {
357 SCSI_LOG_ERROR_RECOVERY(3,
358 sdev_printk(KERN_INFO, sdev,
359 "%s: cmds failed: %d, cancel: %d\n",
360 __func__, cmd_failed,
361 cmd_cancel));
362 cmd_cancel = 0;
363 cmd_failed = 0;
364 ++devices_failed;
368 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
369 "Total of %d commands on %d"
370 " devices require eh work\n",
371 total_failures, devices_failed));
373 #endif
376 * scsi_report_lun_change - Set flag on all *other* devices on the same target
377 * to indicate that a UNIT ATTENTION is expected.
378 * @sdev: Device reporting the UNIT ATTENTION
380 static void scsi_report_lun_change(struct scsi_device *sdev)
382 sdev->sdev_target->expecting_lun_change = 1;
386 * scsi_report_sense - Examine scsi sense information and log messages for
387 * certain conditions, also issue uevents for some of them.
388 * @sdev: Device reporting the sense code
389 * @sshdr: sshdr to be examined
391 static void scsi_report_sense(struct scsi_device *sdev,
392 struct scsi_sense_hdr *sshdr)
394 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
396 if (sshdr->sense_key == UNIT_ATTENTION) {
397 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
398 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
399 sdev_printk(KERN_WARNING, sdev,
400 "Inquiry data has changed");
401 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
402 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
403 scsi_report_lun_change(sdev);
404 sdev_printk(KERN_WARNING, sdev,
405 "Warning! Received an indication that the "
406 "LUN assignments on this target have "
407 "changed. The Linux SCSI layer does not "
408 "automatically remap LUN assignments.\n");
409 } else if (sshdr->asc == 0x3f)
410 sdev_printk(KERN_WARNING, sdev,
411 "Warning! Received an indication that the "
412 "operating parameters on this target have "
413 "changed. The Linux SCSI layer does not "
414 "automatically adjust these parameters.\n");
416 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
417 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
418 sdev_printk(KERN_WARNING, sdev,
419 "Warning! Received an indication that the "
420 "LUN reached a thin provisioning soft "
421 "threshold.\n");
424 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
425 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
426 sdev_printk(KERN_WARNING, sdev,
427 "Mode parameters changed");
428 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
429 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
430 sdev_printk(KERN_WARNING, sdev,
431 "Capacity data has changed");
432 } else if (sshdr->asc == 0x2a)
433 sdev_printk(KERN_WARNING, sdev,
434 "Parameters changed");
437 if (evt_type != SDEV_EVT_MAXBITS) {
438 set_bit(evt_type, sdev->pending_events);
439 schedule_work(&sdev->event_work);
444 * scsi_check_sense - Examine scsi cmd sense
445 * @scmd: Cmd to have sense checked.
447 * Return value:
448 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
450 * Notes:
451 * When a deferred error is detected the current command has
452 * not been executed and needs retrying.
454 static int scsi_check_sense(struct scsi_cmnd *scmd)
456 struct scsi_device *sdev = scmd->device;
457 struct scsi_sense_hdr sshdr;
459 if (! scsi_command_normalize_sense(scmd, &sshdr))
460 return FAILED; /* no valid sense data */
462 scsi_report_sense(sdev, &sshdr);
464 if (scsi_sense_is_deferred(&sshdr))
465 return NEEDS_RETRY;
467 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
468 sdev->scsi_dh_data->scsi_dh->check_sense) {
469 int rc;
471 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
472 if (rc != SCSI_RETURN_NOT_HANDLED)
473 return rc;
474 /* handler does not care. Drop down to default handling */
477 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
479 * nasty: for mid-layer issued TURs, we need to return the
480 * actual sense data without any recovery attempt. For eh
481 * issued ones, we need to try to recover and interpret
483 return SUCCESS;
486 * Previous logic looked for FILEMARK, EOM or ILI which are
487 * mainly associated with tapes and returned SUCCESS.
489 if (sshdr.response_code == 0x70) {
490 /* fixed format */
491 if (scmd->sense_buffer[2] & 0xe0)
492 return SUCCESS;
493 } else {
495 * descriptor format: look for "stream commands sense data
496 * descriptor" (see SSC-3). Assume single sense data
497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
499 if ((sshdr.additional_length > 3) &&
500 (scmd->sense_buffer[8] == 0x4) &&
501 (scmd->sense_buffer[11] & 0xe0))
502 return SUCCESS;
505 switch (sshdr.sense_key) {
506 case NO_SENSE:
507 return SUCCESS;
508 case RECOVERED_ERROR:
509 return /* soft_error */ SUCCESS;
511 case ABORTED_COMMAND:
512 if (sshdr.asc == 0x10) /* DIF */
513 return SUCCESS;
515 return NEEDS_RETRY;
516 case NOT_READY:
517 case UNIT_ATTENTION:
519 * if we are expecting a cc/ua because of a bus reset that we
520 * performed, treat this just as a retry. otherwise this is
521 * information that we should pass up to the upper-level driver
522 * so that we can deal with it there.
524 if (scmd->device->expecting_cc_ua) {
526 * Because some device does not queue unit
527 * attentions correctly, we carefully check
528 * additional sense code and qualifier so as
529 * not to squash media change unit attention.
531 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
532 scmd->device->expecting_cc_ua = 0;
533 return NEEDS_RETRY;
537 * we might also expect a cc/ua if another LUN on the target
538 * reported a UA with an ASC/ASCQ of 3F 0E -
539 * REPORTED LUNS DATA HAS CHANGED.
541 if (scmd->device->sdev_target->expecting_lun_change &&
542 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
543 return NEEDS_RETRY;
545 * if the device is in the process of becoming ready, we
546 * should retry.
548 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
549 return NEEDS_RETRY;
551 * if the device is not started, we need to wake
552 * the error handler to start the motor
554 if (scmd->device->allow_restart &&
555 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
556 return FAILED;
558 * Pass the UA upwards for a determination in the completion
559 * functions.
561 return SUCCESS;
563 /* these are not supported */
564 case DATA_PROTECT:
565 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
566 /* Thin provisioning hard threshold reached */
567 set_host_byte(scmd, DID_ALLOC_FAILURE);
568 return SUCCESS;
570 case COPY_ABORTED:
571 case VOLUME_OVERFLOW:
572 case MISCOMPARE:
573 case BLANK_CHECK:
574 set_host_byte(scmd, DID_TARGET_FAILURE);
575 return SUCCESS;
577 case MEDIUM_ERROR:
578 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
579 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
580 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
581 set_host_byte(scmd, DID_MEDIUM_ERROR);
582 return SUCCESS;
584 return NEEDS_RETRY;
586 case HARDWARE_ERROR:
587 if (scmd->device->retry_hwerror)
588 return ADD_TO_MLQUEUE;
589 else
590 set_host_byte(scmd, DID_TARGET_FAILURE);
592 case ILLEGAL_REQUEST:
593 if (sshdr.asc == 0x20 || /* Invalid command operation code */
594 sshdr.asc == 0x21 || /* Logical block address out of range */
595 sshdr.asc == 0x24 || /* Invalid field in cdb */
596 sshdr.asc == 0x26) { /* Parameter value invalid */
597 set_host_byte(scmd, DID_TARGET_FAILURE);
599 return SUCCESS;
601 default:
602 return SUCCESS;
606 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
608 struct scsi_host_template *sht = sdev->host->hostt;
609 struct scsi_device *tmp_sdev;
611 if (!sht->change_queue_depth ||
612 sdev->queue_depth >= sdev->max_queue_depth)
613 return;
615 if (time_before(jiffies,
616 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
617 return;
619 if (time_before(jiffies,
620 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
621 return;
624 * Walk all devices of a target and do
625 * ramp up on them.
627 shost_for_each_device(tmp_sdev, sdev->host) {
628 if (tmp_sdev->channel != sdev->channel ||
629 tmp_sdev->id != sdev->id ||
630 tmp_sdev->queue_depth == sdev->max_queue_depth)
631 continue;
633 * call back into LLD to increase queue_depth by one
634 * with ramp up reason code.
636 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
637 SCSI_QDEPTH_RAMP_UP);
638 sdev->last_queue_ramp_up = jiffies;
642 static void scsi_handle_queue_full(struct scsi_device *sdev)
644 struct scsi_host_template *sht = sdev->host->hostt;
645 struct scsi_device *tmp_sdev;
647 if (!sht->change_queue_depth)
648 return;
650 shost_for_each_device(tmp_sdev, sdev->host) {
651 if (tmp_sdev->channel != sdev->channel ||
652 tmp_sdev->id != sdev->id)
653 continue;
655 * We do not know the number of commands that were at
656 * the device when we got the queue full so we start
657 * from the highest possible value and work our way down.
659 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
660 SCSI_QDEPTH_QFULL);
665 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
666 * @scmd: SCSI cmd to examine.
668 * Notes:
669 * This is *only* called when we are examining the status of commands
670 * queued during error recovery. the main difference here is that we
671 * don't allow for the possibility of retries here, and we are a lot
672 * more restrictive about what we consider acceptable.
674 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
677 * first check the host byte, to see if there is anything in there
678 * that would indicate what we need to do.
680 if (host_byte(scmd->result) == DID_RESET) {
682 * rats. we are already in the error handler, so we now
683 * get to try and figure out what to do next. if the sense
684 * is valid, we have a pretty good idea of what to do.
685 * if not, we mark it as FAILED.
687 return scsi_check_sense(scmd);
689 if (host_byte(scmd->result) != DID_OK)
690 return FAILED;
693 * next, check the message byte.
695 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
696 return FAILED;
699 * now, check the status byte to see if this indicates
700 * anything special.
702 switch (status_byte(scmd->result)) {
703 case GOOD:
704 scsi_handle_queue_ramp_up(scmd->device);
705 case COMMAND_TERMINATED:
706 return SUCCESS;
707 case CHECK_CONDITION:
708 return scsi_check_sense(scmd);
709 case CONDITION_GOOD:
710 case INTERMEDIATE_GOOD:
711 case INTERMEDIATE_C_GOOD:
713 * who knows? FIXME(eric)
715 return SUCCESS;
716 case RESERVATION_CONFLICT:
717 if (scmd->cmnd[0] == TEST_UNIT_READY)
718 /* it is a success, we probed the device and
719 * found it */
720 return SUCCESS;
721 /* otherwise, we failed to send the command */
722 return FAILED;
723 case QUEUE_FULL:
724 scsi_handle_queue_full(scmd->device);
725 /* fall through */
726 case BUSY:
727 return NEEDS_RETRY;
728 default:
729 return FAILED;
731 return FAILED;
735 * scsi_eh_done - Completion function for error handling.
736 * @scmd: Cmd that is done.
738 static void scsi_eh_done(struct scsi_cmnd *scmd)
740 struct completion *eh_action;
742 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
743 "%s scmd: %p result: %x\n",
744 __func__, scmd, scmd->result));
746 eh_action = scmd->device->host->eh_action;
747 if (eh_action)
748 complete(eh_action);
752 * scsi_try_host_reset - ask host adapter to reset itself
753 * @scmd: SCSI cmd to send host reset.
755 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
757 unsigned long flags;
758 int rtn;
759 struct Scsi_Host *host = scmd->device->host;
760 struct scsi_host_template *hostt = host->hostt;
762 SCSI_LOG_ERROR_RECOVERY(3,
763 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
765 if (!hostt->eh_host_reset_handler)
766 return FAILED;
768 rtn = hostt->eh_host_reset_handler(scmd);
770 if (rtn == SUCCESS) {
771 if (!hostt->skip_settle_delay)
772 ssleep(HOST_RESET_SETTLE_TIME);
773 spin_lock_irqsave(host->host_lock, flags);
774 scsi_report_bus_reset(host, scmd_channel(scmd));
775 spin_unlock_irqrestore(host->host_lock, flags);
778 return rtn;
782 * scsi_try_bus_reset - ask host to perform a bus reset
783 * @scmd: SCSI cmd to send bus reset.
785 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
787 unsigned long flags;
788 int rtn;
789 struct Scsi_Host *host = scmd->device->host;
790 struct scsi_host_template *hostt = host->hostt;
792 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
793 "%s: Snd Bus RST\n", __func__));
795 if (!hostt->eh_bus_reset_handler)
796 return FAILED;
798 rtn = hostt->eh_bus_reset_handler(scmd);
800 if (rtn == SUCCESS) {
801 if (!hostt->skip_settle_delay)
802 ssleep(BUS_RESET_SETTLE_TIME);
803 spin_lock_irqsave(host->host_lock, flags);
804 scsi_report_bus_reset(host, scmd_channel(scmd));
805 spin_unlock_irqrestore(host->host_lock, flags);
808 return rtn;
811 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
813 sdev->was_reset = 1;
814 sdev->expecting_cc_ua = 1;
818 * scsi_try_target_reset - Ask host to perform a target reset
819 * @scmd: SCSI cmd used to send a target reset
821 * Notes:
822 * There is no timeout for this operation. if this operation is
823 * unreliable for a given host, then the host itself needs to put a
824 * timer on it, and set the host back to a consistent state prior to
825 * returning.
827 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
829 unsigned long flags;
830 int rtn;
831 struct Scsi_Host *host = scmd->device->host;
832 struct scsi_host_template *hostt = host->hostt;
834 if (!hostt->eh_target_reset_handler)
835 return FAILED;
837 rtn = hostt->eh_target_reset_handler(scmd);
838 if (rtn == SUCCESS) {
839 spin_lock_irqsave(host->host_lock, flags);
840 __starget_for_each_device(scsi_target(scmd->device), NULL,
841 __scsi_report_device_reset);
842 spin_unlock_irqrestore(host->host_lock, flags);
845 return rtn;
849 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
850 * @scmd: SCSI cmd used to send BDR
852 * Notes:
853 * There is no timeout for this operation. if this operation is
854 * unreliable for a given host, then the host itself needs to put a
855 * timer on it, and set the host back to a consistent state prior to
856 * returning.
858 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
860 int rtn;
861 struct scsi_host_template *hostt = scmd->device->host->hostt;
863 if (!hostt->eh_device_reset_handler)
864 return FAILED;
866 rtn = hostt->eh_device_reset_handler(scmd);
867 if (rtn == SUCCESS)
868 __scsi_report_device_reset(scmd->device, NULL);
869 return rtn;
872 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
874 if (!hostt->eh_abort_handler)
875 return FAILED;
877 return hostt->eh_abort_handler(scmd);
880 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
882 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
883 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
884 if (scsi_try_target_reset(scmd) != SUCCESS)
885 if (scsi_try_bus_reset(scmd) != SUCCESS)
886 scsi_try_host_reset(scmd);
890 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
891 * @scmd: SCSI command structure to hijack
892 * @ses: structure to save restore information
893 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
894 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
895 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
897 * This function is used to save a scsi command information before re-execution
898 * as part of the error recovery process. If @sense_bytes is 0 the command
899 * sent must be one that does not transfer any data. If @sense_bytes != 0
900 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
901 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
903 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
904 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
906 struct scsi_device *sdev = scmd->device;
909 * We need saved copies of a number of fields - this is because
910 * error handling may need to overwrite these with different values
911 * to run different commands, and once error handling is complete,
912 * we will need to restore these values prior to running the actual
913 * command.
915 ses->cmd_len = scmd->cmd_len;
916 ses->cmnd = scmd->cmnd;
917 ses->data_direction = scmd->sc_data_direction;
918 ses->sdb = scmd->sdb;
919 ses->next_rq = scmd->request->next_rq;
920 ses->result = scmd->result;
921 ses->underflow = scmd->underflow;
922 ses->prot_op = scmd->prot_op;
924 scmd->prot_op = SCSI_PROT_NORMAL;
925 scmd->eh_eflags = 0;
926 scmd->cmnd = ses->eh_cmnd;
927 memset(scmd->cmnd, 0, BLK_MAX_CDB);
928 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
929 scmd->request->next_rq = NULL;
930 scmd->result = 0;
932 if (sense_bytes) {
933 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
934 sense_bytes);
935 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
936 scmd->sdb.length);
937 scmd->sdb.table.sgl = &ses->sense_sgl;
938 scmd->sc_data_direction = DMA_FROM_DEVICE;
939 scmd->sdb.table.nents = 1;
940 scmd->cmnd[0] = REQUEST_SENSE;
941 scmd->cmnd[4] = scmd->sdb.length;
942 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
943 } else {
944 scmd->sc_data_direction = DMA_NONE;
945 if (cmnd) {
946 BUG_ON(cmnd_size > BLK_MAX_CDB);
947 memcpy(scmd->cmnd, cmnd, cmnd_size);
948 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
952 scmd->underflow = 0;
954 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
955 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
956 (sdev->lun << 5 & 0xe0);
959 * Zero the sense buffer. The scsi spec mandates that any
960 * untransferred sense data should be interpreted as being zero.
962 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
964 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
967 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
968 * @scmd: SCSI command structure to restore
969 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
971 * Undo any damage done by above scsi_eh_prep_cmnd().
973 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
976 * Restore original data
978 scmd->cmd_len = ses->cmd_len;
979 scmd->cmnd = ses->cmnd;
980 scmd->sc_data_direction = ses->data_direction;
981 scmd->sdb = ses->sdb;
982 scmd->request->next_rq = ses->next_rq;
983 scmd->result = ses->result;
984 scmd->underflow = ses->underflow;
985 scmd->prot_op = ses->prot_op;
987 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
990 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
991 * @scmd: SCSI command structure to hijack
992 * @cmnd: CDB to send
993 * @cmnd_size: size in bytes of @cmnd
994 * @timeout: timeout for this request
995 * @sense_bytes: size of sense data to copy or 0
997 * This function is used to send a scsi command down to a target device
998 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1000 * Return value:
1001 * SUCCESS or FAILED or NEEDS_RETRY
1003 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1004 int cmnd_size, int timeout, unsigned sense_bytes)
1006 struct scsi_device *sdev = scmd->device;
1007 struct Scsi_Host *shost = sdev->host;
1008 DECLARE_COMPLETION_ONSTACK(done);
1009 unsigned long timeleft = timeout;
1010 struct scsi_eh_save ses;
1011 const unsigned long stall_for = msecs_to_jiffies(100);
1012 int rtn;
1014 retry:
1015 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1016 shost->eh_action = &done;
1018 scsi_log_send(scmd);
1019 scmd->scsi_done = scsi_eh_done;
1020 rtn = shost->hostt->queuecommand(shost, scmd);
1021 if (rtn) {
1022 if (timeleft > stall_for) {
1023 scsi_eh_restore_cmnd(scmd, &ses);
1024 timeleft -= stall_for;
1025 msleep(jiffies_to_msecs(stall_for));
1026 goto retry;
1028 /* signal not to enter either branch of the if () below */
1029 timeleft = 0;
1030 rtn = FAILED;
1031 } else {
1032 timeleft = wait_for_completion_timeout(&done, timeout);
1033 rtn = SUCCESS;
1036 shost->eh_action = NULL;
1038 scsi_log_completion(scmd, rtn);
1040 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1041 "%s: scmd: %p, timeleft: %ld\n",
1042 __func__, scmd, timeleft));
1045 * If there is time left scsi_eh_done got called, and we will examine
1046 * the actual status codes to see whether the command actually did
1047 * complete normally, else if we have a zero return and no time left,
1048 * the command must still be pending, so abort it and return FAILED.
1049 * If we never actually managed to issue the command, because
1050 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1051 * value above (so don't execute either branch of the if)
1053 if (timeleft) {
1054 rtn = scsi_eh_completed_normally(scmd);
1055 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1056 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1058 switch (rtn) {
1059 case SUCCESS:
1060 case NEEDS_RETRY:
1061 case FAILED:
1062 break;
1063 case ADD_TO_MLQUEUE:
1064 rtn = NEEDS_RETRY;
1065 break;
1066 default:
1067 rtn = FAILED;
1068 break;
1070 } else if (rtn != FAILED) {
1071 scsi_abort_eh_cmnd(scmd);
1072 rtn = FAILED;
1075 scsi_eh_restore_cmnd(scmd, &ses);
1077 return rtn;
1081 * scsi_request_sense - Request sense data from a particular target.
1082 * @scmd: SCSI cmd for request sense.
1084 * Notes:
1085 * Some hosts automatically obtain this information, others require
1086 * that we obtain it on our own. This function will *not* return until
1087 * the command either times out, or it completes.
1089 static int scsi_request_sense(struct scsi_cmnd *scmd)
1091 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1094 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1096 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1097 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1098 if (sdrv->eh_action)
1099 rtn = sdrv->eh_action(scmd, rtn);
1101 return rtn;
1105 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1106 * @scmd: Original SCSI cmd that eh has finished.
1107 * @done_q: Queue for processed commands.
1109 * Notes:
1110 * We don't want to use the normal command completion while we are are
1111 * still handling errors - it may cause other commands to be queued,
1112 * and that would disturb what we are doing. Thus we really want to
1113 * keep a list of pending commands for final completion, and once we
1114 * are ready to leave error handling we handle completion for real.
1116 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1118 scmd->eh_eflags = 0;
1119 list_move_tail(&scmd->eh_entry, done_q);
1121 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1124 * scsi_eh_get_sense - Get device sense data.
1125 * @work_q: Queue of commands to process.
1126 * @done_q: Queue of processed commands.
1128 * Description:
1129 * See if we need to request sense information. if so, then get it
1130 * now, so we have a better idea of what to do.
1132 * Notes:
1133 * This has the unfortunate side effect that if a shost adapter does
1134 * not automatically request sense information, we end up shutting
1135 * it down before we request it.
1137 * All drivers should request sense information internally these days,
1138 * so for now all I have to say is tough noogies if you end up in here.
1140 * XXX: Long term this code should go away, but that needs an audit of
1141 * all LLDDs first.
1143 int scsi_eh_get_sense(struct list_head *work_q,
1144 struct list_head *done_q)
1146 struct scsi_cmnd *scmd, *next;
1147 struct Scsi_Host *shost;
1148 int rtn;
1150 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1151 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1152 SCSI_SENSE_VALID(scmd))
1153 continue;
1155 shost = scmd->device->host;
1156 if (scsi_host_eh_past_deadline(shost)) {
1157 SCSI_LOG_ERROR_RECOVERY(3,
1158 shost_printk(KERN_INFO, shost,
1159 "skip %s, past eh deadline\n",
1160 __func__));
1161 break;
1163 if (status_byte(scmd->result) != CHECK_CONDITION)
1165 * don't request sense if there's no check condition
1166 * status because the error we're processing isn't one
1167 * that has a sense code (and some devices get
1168 * confused by sense requests out of the blue)
1170 continue;
1172 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1173 "%s: requesting sense\n",
1174 current->comm));
1175 rtn = scsi_request_sense(scmd);
1176 if (rtn != SUCCESS)
1177 continue;
1179 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1180 "sense requested for %p result %x\n",
1181 scmd, scmd->result));
1182 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1184 rtn = scsi_decide_disposition(scmd);
1187 * if the result was normal, then just pass it along to the
1188 * upper level.
1190 if (rtn == SUCCESS)
1191 /* we don't want this command reissued, just
1192 * finished with the sense data, so set
1193 * retries to the max allowed to ensure it
1194 * won't get reissued */
1195 scmd->retries = scmd->allowed;
1196 else if (rtn != NEEDS_RETRY)
1197 continue;
1199 scsi_eh_finish_cmd(scmd, done_q);
1202 return list_empty(work_q);
1204 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1207 * scsi_eh_tur - Send TUR to device.
1208 * @scmd: &scsi_cmnd to send TUR
1210 * Return value:
1211 * 0 - Device is ready. 1 - Device NOT ready.
1213 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1215 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1216 int retry_cnt = 1, rtn;
1218 retry_tur:
1219 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1220 scmd->device->eh_timeout, 0);
1222 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1223 "%s: scmd %p rtn %x\n", __func__, scmd, rtn));
1225 switch (rtn) {
1226 case NEEDS_RETRY:
1227 if (retry_cnt--)
1228 goto retry_tur;
1229 /*FALLTHRU*/
1230 case SUCCESS:
1231 return 0;
1232 default:
1233 return 1;
1238 * scsi_eh_test_devices - check if devices are responding from error recovery.
1239 * @cmd_list: scsi commands in error recovery.
1240 * @work_q: queue for commands which still need more error recovery
1241 * @done_q: queue for commands which are finished
1242 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1244 * Decription:
1245 * Tests if devices are in a working state. Commands to devices now in
1246 * a working state are sent to the done_q while commands to devices which
1247 * are still failing to respond are returned to the work_q for more
1248 * processing.
1250 static int scsi_eh_test_devices(struct list_head *cmd_list,
1251 struct list_head *work_q,
1252 struct list_head *done_q, int try_stu)
1254 struct scsi_cmnd *scmd, *next;
1255 struct scsi_device *sdev;
1256 int finish_cmds;
1258 while (!list_empty(cmd_list)) {
1259 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1260 sdev = scmd->device;
1262 if (!try_stu) {
1263 if (scsi_host_eh_past_deadline(sdev->host)) {
1264 /* Push items back onto work_q */
1265 list_splice_init(cmd_list, work_q);
1266 SCSI_LOG_ERROR_RECOVERY(3,
1267 shost_printk(KERN_INFO, sdev->host,
1268 "skip %s, past eh deadline",
1269 __func__));
1270 break;
1274 finish_cmds = !scsi_device_online(scmd->device) ||
1275 (try_stu && !scsi_eh_try_stu(scmd) &&
1276 !scsi_eh_tur(scmd)) ||
1277 !scsi_eh_tur(scmd);
1279 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1280 if (scmd->device == sdev) {
1281 if (finish_cmds &&
1282 (try_stu ||
1283 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1284 scsi_eh_finish_cmd(scmd, done_q);
1285 else
1286 list_move_tail(&scmd->eh_entry, work_q);
1289 return list_empty(work_q);
1294 * scsi_eh_abort_cmds - abort pending commands.
1295 * @work_q: &list_head for pending commands.
1296 * @done_q: &list_head for processed commands.
1298 * Decription:
1299 * Try and see whether or not it makes sense to try and abort the
1300 * running command. This only works out to be the case if we have one
1301 * command that has timed out. If the command simply failed, it makes
1302 * no sense to try and abort the command, since as far as the shost
1303 * adapter is concerned, it isn't running.
1305 static int scsi_eh_abort_cmds(struct list_head *work_q,
1306 struct list_head *done_q)
1308 struct scsi_cmnd *scmd, *next;
1309 LIST_HEAD(check_list);
1310 int rtn;
1311 struct Scsi_Host *shost;
1313 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1314 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1315 continue;
1316 shost = scmd->device->host;
1317 if (scsi_host_eh_past_deadline(shost)) {
1318 list_splice_init(&check_list, work_q);
1319 SCSI_LOG_ERROR_RECOVERY(3,
1320 shost_printk(KERN_INFO, shost,
1321 "skip %s, past eh deadline\n",
1322 __func__));
1323 return list_empty(work_q);
1325 SCSI_LOG_ERROR_RECOVERY(3,
1326 shost_printk(KERN_INFO, shost,
1327 "%s: aborting cmd: 0x%p\n",
1328 current->comm, scmd));
1329 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1330 if (rtn == FAILED) {
1331 SCSI_LOG_ERROR_RECOVERY(3,
1332 shost_printk(KERN_INFO, shost,
1333 "%s: aborting cmd failed: 0x%p\n",
1334 current->comm, scmd));
1335 list_splice_init(&check_list, work_q);
1336 return list_empty(work_q);
1338 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1339 if (rtn == FAST_IO_FAIL)
1340 scsi_eh_finish_cmd(scmd, done_q);
1341 else
1342 list_move_tail(&scmd->eh_entry, &check_list);
1345 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1349 * scsi_eh_try_stu - Send START_UNIT to device.
1350 * @scmd: &scsi_cmnd to send START_UNIT
1352 * Return value:
1353 * 0 - Device is ready. 1 - Device NOT ready.
1355 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1357 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1359 if (scmd->device->allow_restart) {
1360 int i, rtn = NEEDS_RETRY;
1362 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1363 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1365 if (rtn == SUCCESS)
1366 return 0;
1369 return 1;
1373 * scsi_eh_stu - send START_UNIT if needed
1374 * @shost: &scsi host being recovered.
1375 * @work_q: &list_head for pending commands.
1376 * @done_q: &list_head for processed commands.
1378 * Notes:
1379 * If commands are failing due to not ready, initializing command required,
1380 * try revalidating the device, which will end up sending a start unit.
1382 static int scsi_eh_stu(struct Scsi_Host *shost,
1383 struct list_head *work_q,
1384 struct list_head *done_q)
1386 struct scsi_cmnd *scmd, *stu_scmd, *next;
1387 struct scsi_device *sdev;
1389 shost_for_each_device(sdev, shost) {
1390 if (scsi_host_eh_past_deadline(shost)) {
1391 SCSI_LOG_ERROR_RECOVERY(3,
1392 shost_printk(KERN_INFO, shost,
1393 "skip %s, past eh deadline\n",
1394 __func__));
1395 break;
1397 stu_scmd = NULL;
1398 list_for_each_entry(scmd, work_q, eh_entry)
1399 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1400 scsi_check_sense(scmd) == FAILED ) {
1401 stu_scmd = scmd;
1402 break;
1405 if (!stu_scmd)
1406 continue;
1408 SCSI_LOG_ERROR_RECOVERY(3,
1409 shost_printk(KERN_INFO, shost,
1410 "%s: Sending START_UNIT to sdev: 0x%p\n",
1411 current->comm, sdev));
1413 if (!scsi_eh_try_stu(stu_scmd)) {
1414 if (!scsi_device_online(sdev) ||
1415 !scsi_eh_tur(stu_scmd)) {
1416 list_for_each_entry_safe(scmd, next,
1417 work_q, eh_entry) {
1418 if (scmd->device == sdev &&
1419 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1420 scsi_eh_finish_cmd(scmd, done_q);
1423 } else {
1424 SCSI_LOG_ERROR_RECOVERY(3,
1425 shost_printk(KERN_INFO, shost,
1426 "%s: START_UNIT failed to sdev:"
1427 " 0x%p\n", current->comm, sdev));
1431 return list_empty(work_q);
1436 * scsi_eh_bus_device_reset - send bdr if needed
1437 * @shost: scsi host being recovered.
1438 * @work_q: &list_head for pending commands.
1439 * @done_q: &list_head for processed commands.
1441 * Notes:
1442 * Try a bus device reset. Still, look to see whether we have multiple
1443 * devices that are jammed or not - if we have multiple devices, it
1444 * makes no sense to try bus_device_reset - we really would need to try
1445 * a bus_reset instead.
1447 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1448 struct list_head *work_q,
1449 struct list_head *done_q)
1451 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1452 struct scsi_device *sdev;
1453 int rtn;
1455 shost_for_each_device(sdev, shost) {
1456 if (scsi_host_eh_past_deadline(shost)) {
1457 SCSI_LOG_ERROR_RECOVERY(3,
1458 shost_printk(KERN_INFO, shost,
1459 "skip %s, past eh deadline\n",
1460 __func__));
1461 break;
1463 bdr_scmd = NULL;
1464 list_for_each_entry(scmd, work_q, eh_entry)
1465 if (scmd->device == sdev) {
1466 bdr_scmd = scmd;
1467 break;
1470 if (!bdr_scmd)
1471 continue;
1473 SCSI_LOG_ERROR_RECOVERY(3,
1474 shost_printk(KERN_INFO, shost,
1475 "%s: Sending BDR sdev: 0x%p\n",
1476 current->comm, sdev));
1477 rtn = scsi_try_bus_device_reset(bdr_scmd);
1478 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1479 if (!scsi_device_online(sdev) ||
1480 rtn == FAST_IO_FAIL ||
1481 !scsi_eh_tur(bdr_scmd)) {
1482 list_for_each_entry_safe(scmd, next,
1483 work_q, eh_entry) {
1484 if (scmd->device == sdev &&
1485 scsi_eh_action(scmd, rtn) != FAILED)
1486 scsi_eh_finish_cmd(scmd,
1487 done_q);
1490 } else {
1491 SCSI_LOG_ERROR_RECOVERY(3,
1492 shost_printk(KERN_INFO, shost,
1493 "%s: BDR failed sdev: 0x%p\n",
1494 current->comm, sdev));
1498 return list_empty(work_q);
1502 * scsi_eh_target_reset - send target reset if needed
1503 * @shost: scsi host being recovered.
1504 * @work_q: &list_head for pending commands.
1505 * @done_q: &list_head for processed commands.
1507 * Notes:
1508 * Try a target reset.
1510 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1511 struct list_head *work_q,
1512 struct list_head *done_q)
1514 LIST_HEAD(tmp_list);
1515 LIST_HEAD(check_list);
1517 list_splice_init(work_q, &tmp_list);
1519 while (!list_empty(&tmp_list)) {
1520 struct scsi_cmnd *next, *scmd;
1521 int rtn;
1522 unsigned int id;
1524 if (scsi_host_eh_past_deadline(shost)) {
1525 /* push back on work queue for further processing */
1526 list_splice_init(&check_list, work_q);
1527 list_splice_init(&tmp_list, work_q);
1528 SCSI_LOG_ERROR_RECOVERY(3,
1529 shost_printk(KERN_INFO, shost,
1530 "skip %s, past eh deadline\n",
1531 __func__));
1532 return list_empty(work_q);
1535 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1536 id = scmd_id(scmd);
1538 SCSI_LOG_ERROR_RECOVERY(3,
1539 shost_printk(KERN_INFO, shost,
1540 "%s: Sending target reset to target %d\n",
1541 current->comm, id));
1542 rtn = scsi_try_target_reset(scmd);
1543 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1544 SCSI_LOG_ERROR_RECOVERY(3,
1545 shost_printk(KERN_INFO, shost,
1546 "%s: Target reset failed"
1547 " target: %d\n",
1548 current->comm, id));
1549 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1550 if (scmd_id(scmd) != id)
1551 continue;
1553 if (rtn == SUCCESS)
1554 list_move_tail(&scmd->eh_entry, &check_list);
1555 else if (rtn == FAST_IO_FAIL)
1556 scsi_eh_finish_cmd(scmd, done_q);
1557 else
1558 /* push back on work queue for further processing */
1559 list_move(&scmd->eh_entry, work_q);
1563 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1567 * scsi_eh_bus_reset - send a bus reset
1568 * @shost: &scsi host being recovered.
1569 * @work_q: &list_head for pending commands.
1570 * @done_q: &list_head for processed commands.
1572 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1573 struct list_head *work_q,
1574 struct list_head *done_q)
1576 struct scsi_cmnd *scmd, *chan_scmd, *next;
1577 LIST_HEAD(check_list);
1578 unsigned int channel;
1579 int rtn;
1582 * we really want to loop over the various channels, and do this on
1583 * a channel by channel basis. we should also check to see if any
1584 * of the failed commands are on soft_reset devices, and if so, skip
1585 * the reset.
1588 for (channel = 0; channel <= shost->max_channel; channel++) {
1589 if (scsi_host_eh_past_deadline(shost)) {
1590 list_splice_init(&check_list, work_q);
1591 SCSI_LOG_ERROR_RECOVERY(3,
1592 shost_printk(KERN_INFO, shost,
1593 "skip %s, past eh deadline\n",
1594 __func__));
1595 return list_empty(work_q);
1598 chan_scmd = NULL;
1599 list_for_each_entry(scmd, work_q, eh_entry) {
1600 if (channel == scmd_channel(scmd)) {
1601 chan_scmd = scmd;
1602 break;
1604 * FIXME add back in some support for
1605 * soft_reset devices.
1610 if (!chan_scmd)
1611 continue;
1612 SCSI_LOG_ERROR_RECOVERY(3,
1613 shost_printk(KERN_INFO, shost,
1614 "%s: Sending BRST chan: %d\n",
1615 current->comm, channel));
1616 rtn = scsi_try_bus_reset(chan_scmd);
1617 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1618 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1619 if (channel == scmd_channel(scmd)) {
1620 if (rtn == FAST_IO_FAIL)
1621 scsi_eh_finish_cmd(scmd,
1622 done_q);
1623 else
1624 list_move_tail(&scmd->eh_entry,
1625 &check_list);
1628 } else {
1629 SCSI_LOG_ERROR_RECOVERY(3,
1630 shost_printk(KERN_INFO, shost,
1631 "%s: BRST failed chan: %d\n",
1632 current->comm, channel));
1635 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1639 * scsi_eh_host_reset - send a host reset
1640 * @shost: host to be reset.
1641 * @work_q: &list_head for pending commands.
1642 * @done_q: &list_head for processed commands.
1644 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1645 struct list_head *work_q,
1646 struct list_head *done_q)
1648 struct scsi_cmnd *scmd, *next;
1649 LIST_HEAD(check_list);
1650 int rtn;
1652 if (!list_empty(work_q)) {
1653 scmd = list_entry(work_q->next,
1654 struct scsi_cmnd, eh_entry);
1656 SCSI_LOG_ERROR_RECOVERY(3,
1657 shost_printk(KERN_INFO, shost,
1658 "%s: Sending HRST\n",
1659 current->comm));
1661 rtn = scsi_try_host_reset(scmd);
1662 if (rtn == SUCCESS) {
1663 list_splice_init(work_q, &check_list);
1664 } else if (rtn == FAST_IO_FAIL) {
1665 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1666 scsi_eh_finish_cmd(scmd, done_q);
1668 } else {
1669 SCSI_LOG_ERROR_RECOVERY(3,
1670 shost_printk(KERN_INFO, shost,
1671 "%s: HRST failed\n",
1672 current->comm));
1675 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1679 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1680 * @work_q: &list_head for pending commands.
1681 * @done_q: &list_head for processed commands.
1683 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1684 struct list_head *done_q)
1686 struct scsi_cmnd *scmd, *next;
1688 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1689 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1690 "not ready after error recovery\n");
1691 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1692 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1694 * FIXME: Handle lost cmds.
1697 scsi_eh_finish_cmd(scmd, done_q);
1699 return;
1703 * scsi_noretry_cmd - determine if command should be failed fast
1704 * @scmd: SCSI cmd to examine.
1706 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1708 switch (host_byte(scmd->result)) {
1709 case DID_OK:
1710 break;
1711 case DID_TIME_OUT:
1712 goto check_type;
1713 case DID_BUS_BUSY:
1714 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1715 case DID_PARITY:
1716 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1717 case DID_ERROR:
1718 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1719 status_byte(scmd->result) == RESERVATION_CONFLICT)
1720 return 0;
1721 /* fall through */
1722 case DID_SOFT_ERROR:
1723 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1726 if (status_byte(scmd->result) != CHECK_CONDITION)
1727 return 0;
1729 check_type:
1731 * assume caller has checked sense and determined
1732 * the check condition was retryable.
1734 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1735 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1736 return 1;
1737 else
1738 return 0;
1742 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1743 * @scmd: SCSI cmd to examine.
1745 * Notes:
1746 * This is *only* called when we are examining the status after sending
1747 * out the actual data command. any commands that are queued for error
1748 * recovery (e.g. test_unit_ready) do *not* come through here.
1750 * When this routine returns failed, it means the error handler thread
1751 * is woken. In cases where the error code indicates an error that
1752 * doesn't require the error handler read (i.e. we don't need to
1753 * abort/reset), this function should return SUCCESS.
1755 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1757 int rtn;
1760 * if the device is offline, then we clearly just pass the result back
1761 * up to the top level.
1763 if (!scsi_device_online(scmd->device)) {
1764 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1765 "%s: device offline - report as SUCCESS\n", __func__));
1766 return SUCCESS;
1770 * first check the host byte, to see if there is anything in there
1771 * that would indicate what we need to do.
1773 switch (host_byte(scmd->result)) {
1774 case DID_PASSTHROUGH:
1776 * no matter what, pass this through to the upper layer.
1777 * nuke this special code so that it looks like we are saying
1778 * did_ok.
1780 scmd->result &= 0xff00ffff;
1781 return SUCCESS;
1782 case DID_OK:
1784 * looks good. drop through, and check the next byte.
1786 break;
1787 case DID_ABORT:
1788 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1789 set_host_byte(scmd, DID_TIME_OUT);
1790 return SUCCESS;
1792 case DID_NO_CONNECT:
1793 case DID_BAD_TARGET:
1795 * note - this means that we just report the status back
1796 * to the top level driver, not that we actually think
1797 * that it indicates SUCCESS.
1799 return SUCCESS;
1801 * when the low level driver returns did_soft_error,
1802 * it is responsible for keeping an internal retry counter
1803 * in order to avoid endless loops (db)
1805 * actually this is a bug in this function here. we should
1806 * be mindful of the maximum number of retries specified
1807 * and not get stuck in a loop.
1809 case DID_SOFT_ERROR:
1810 goto maybe_retry;
1811 case DID_IMM_RETRY:
1812 return NEEDS_RETRY;
1814 case DID_REQUEUE:
1815 return ADD_TO_MLQUEUE;
1816 case DID_TRANSPORT_DISRUPTED:
1818 * LLD/transport was disrupted during processing of the IO.
1819 * The transport class is now blocked/blocking,
1820 * and the transport will decide what to do with the IO
1821 * based on its timers and recovery capablilities if
1822 * there are enough retries.
1824 goto maybe_retry;
1825 case DID_TRANSPORT_FAILFAST:
1827 * The transport decided to failfast the IO (most likely
1828 * the fast io fail tmo fired), so send IO directly upwards.
1830 return SUCCESS;
1831 case DID_ERROR:
1832 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1833 status_byte(scmd->result) == RESERVATION_CONFLICT)
1835 * execute reservation conflict processing code
1836 * lower down
1838 break;
1839 /* fallthrough */
1840 case DID_BUS_BUSY:
1841 case DID_PARITY:
1842 goto maybe_retry;
1843 case DID_TIME_OUT:
1845 * when we scan the bus, we get timeout messages for
1846 * these commands if there is no device available.
1847 * other hosts report did_no_connect for the same thing.
1849 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1850 scmd->cmnd[0] == INQUIRY)) {
1851 return SUCCESS;
1852 } else {
1853 return FAILED;
1855 case DID_RESET:
1856 return SUCCESS;
1857 default:
1858 return FAILED;
1862 * next, check the message byte.
1864 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1865 return FAILED;
1868 * check the status byte to see if this indicates anything special.
1870 switch (status_byte(scmd->result)) {
1871 case QUEUE_FULL:
1872 scsi_handle_queue_full(scmd->device);
1874 * the case of trying to send too many commands to a
1875 * tagged queueing device.
1877 case BUSY:
1879 * device can't talk to us at the moment. Should only
1880 * occur (SAM-3) when the task queue is empty, so will cause
1881 * the empty queue handling to trigger a stall in the
1882 * device.
1884 return ADD_TO_MLQUEUE;
1885 case GOOD:
1886 if (scmd->cmnd[0] == REPORT_LUNS)
1887 scmd->device->sdev_target->expecting_lun_change = 0;
1888 scsi_handle_queue_ramp_up(scmd->device);
1889 case COMMAND_TERMINATED:
1890 return SUCCESS;
1891 case TASK_ABORTED:
1892 goto maybe_retry;
1893 case CHECK_CONDITION:
1894 rtn = scsi_check_sense(scmd);
1895 if (rtn == NEEDS_RETRY)
1896 goto maybe_retry;
1897 /* if rtn == FAILED, we have no sense information;
1898 * returning FAILED will wake the error handler thread
1899 * to collect the sense and redo the decide
1900 * disposition */
1901 return rtn;
1902 case CONDITION_GOOD:
1903 case INTERMEDIATE_GOOD:
1904 case INTERMEDIATE_C_GOOD:
1905 case ACA_ACTIVE:
1907 * who knows? FIXME(eric)
1909 return SUCCESS;
1911 case RESERVATION_CONFLICT:
1912 sdev_printk(KERN_INFO, scmd->device,
1913 "reservation conflict\n");
1914 set_host_byte(scmd, DID_NEXUS_FAILURE);
1915 return SUCCESS; /* causes immediate i/o error */
1916 default:
1917 return FAILED;
1919 return FAILED;
1921 maybe_retry:
1923 /* we requeue for retry because the error was retryable, and
1924 * the request was not marked fast fail. Note that above,
1925 * even if the request is marked fast fail, we still requeue
1926 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1927 if ((++scmd->retries) <= scmd->allowed
1928 && !scsi_noretry_cmd(scmd)) {
1929 return NEEDS_RETRY;
1930 } else {
1932 * no more retries - report this one back to upper level.
1934 return SUCCESS;
1938 static void eh_lock_door_done(struct request *req, int uptodate)
1940 __blk_put_request(req->q, req);
1944 * scsi_eh_lock_door - Prevent medium removal for the specified device
1945 * @sdev: SCSI device to prevent medium removal
1947 * Locking:
1948 * We must be called from process context.
1950 * Notes:
1951 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1952 * head of the devices request queue, and continue.
1954 static void scsi_eh_lock_door(struct scsi_device *sdev)
1956 struct request *req;
1959 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1960 * request becomes available
1962 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1963 if (IS_ERR(req))
1964 return;
1966 blk_rq_set_block_pc(req);
1968 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1969 req->cmd[1] = 0;
1970 req->cmd[2] = 0;
1971 req->cmd[3] = 0;
1972 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1973 req->cmd[5] = 0;
1975 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1977 req->cmd_flags |= REQ_QUIET;
1978 req->timeout = 10 * HZ;
1979 req->retries = 5;
1981 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1985 * scsi_restart_operations - restart io operations to the specified host.
1986 * @shost: Host we are restarting.
1988 * Notes:
1989 * When we entered the error handler, we blocked all further i/o to
1990 * this device. we need to 'reverse' this process.
1992 static void scsi_restart_operations(struct Scsi_Host *shost)
1994 struct scsi_device *sdev;
1995 unsigned long flags;
1998 * If the door was locked, we need to insert a door lock request
1999 * onto the head of the SCSI request queue for the device. There
2000 * is no point trying to lock the door of an off-line device.
2002 shost_for_each_device(sdev, shost) {
2003 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2004 scsi_eh_lock_door(sdev);
2005 sdev->was_reset = 0;
2010 * next free up anything directly waiting upon the host. this
2011 * will be requests for character device operations, and also for
2012 * ioctls to queued block devices.
2014 SCSI_LOG_ERROR_RECOVERY(3,
2015 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2017 spin_lock_irqsave(shost->host_lock, flags);
2018 if (scsi_host_set_state(shost, SHOST_RUNNING))
2019 if (scsi_host_set_state(shost, SHOST_CANCEL))
2020 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2021 spin_unlock_irqrestore(shost->host_lock, flags);
2023 wake_up(&shost->host_wait);
2026 * finally we need to re-initiate requests that may be pending. we will
2027 * have had everything blocked while error handling is taking place, and
2028 * now that error recovery is done, we will need to ensure that these
2029 * requests are started.
2031 scsi_run_host_queues(shost);
2034 * if eh is active and host_eh_scheduled is pending we need to re-run
2035 * recovery. we do this check after scsi_run_host_queues() to allow
2036 * everything pent up since the last eh run a chance to make forward
2037 * progress before we sync again. Either we'll immediately re-run
2038 * recovery or scsi_device_unbusy() will wake us again when these
2039 * pending commands complete.
2041 spin_lock_irqsave(shost->host_lock, flags);
2042 if (shost->host_eh_scheduled)
2043 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2044 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2045 spin_unlock_irqrestore(shost->host_lock, flags);
2049 * scsi_eh_ready_devs - check device ready state and recover if not.
2050 * @shost: host to be recovered.
2051 * @work_q: &list_head for pending commands.
2052 * @done_q: &list_head for processed commands.
2054 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2055 struct list_head *work_q,
2056 struct list_head *done_q)
2058 if (!scsi_eh_stu(shost, work_q, done_q))
2059 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2060 if (!scsi_eh_target_reset(shost, work_q, done_q))
2061 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2062 if (!scsi_eh_host_reset(shost, work_q, done_q))
2063 scsi_eh_offline_sdevs(work_q,
2064 done_q);
2066 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2069 * scsi_eh_flush_done_q - finish processed commands or retry them.
2070 * @done_q: list_head of processed commands.
2072 void scsi_eh_flush_done_q(struct list_head *done_q)
2074 struct scsi_cmnd *scmd, *next;
2076 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2077 list_del_init(&scmd->eh_entry);
2078 if (scsi_device_online(scmd->device) &&
2079 !scsi_noretry_cmd(scmd) &&
2080 (++scmd->retries <= scmd->allowed)) {
2081 SCSI_LOG_ERROR_RECOVERY(3,
2082 scmd_printk(KERN_INFO, scmd,
2083 "%s: flush retry cmd: %p\n",
2084 current->comm, scmd));
2085 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2086 } else {
2088 * If just we got sense for the device (called
2089 * scsi_eh_get_sense), scmd->result is already
2090 * set, do not set DRIVER_TIMEOUT.
2092 if (!scmd->result)
2093 scmd->result |= (DRIVER_TIMEOUT << 24);
2094 SCSI_LOG_ERROR_RECOVERY(3,
2095 scmd_printk(KERN_INFO, scmd,
2096 "%s: flush finish cmd: %p\n",
2097 current->comm, scmd));
2098 scsi_finish_command(scmd);
2102 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2105 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2106 * @shost: Host to unjam.
2108 * Notes:
2109 * When we come in here, we *know* that all commands on the bus have
2110 * either completed, failed or timed out. we also know that no further
2111 * commands are being sent to the host, so things are relatively quiet
2112 * and we have freedom to fiddle with things as we wish.
2114 * This is only the *default* implementation. it is possible for
2115 * individual drivers to supply their own version of this function, and
2116 * if the maintainer wishes to do this, it is strongly suggested that
2117 * this function be taken as a template and modified. this function
2118 * was designed to correctly handle problems for about 95% of the
2119 * different cases out there, and it should always provide at least a
2120 * reasonable amount of error recovery.
2122 * Any command marked 'failed' or 'timeout' must eventually have
2123 * scsi_finish_cmd() called for it. we do all of the retry stuff
2124 * here, so when we restart the host after we return it should have an
2125 * empty queue.
2127 static void scsi_unjam_host(struct Scsi_Host *shost)
2129 unsigned long flags;
2130 LIST_HEAD(eh_work_q);
2131 LIST_HEAD(eh_done_q);
2133 spin_lock_irqsave(shost->host_lock, flags);
2134 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2135 spin_unlock_irqrestore(shost->host_lock, flags);
2137 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2139 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2140 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2141 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2143 spin_lock_irqsave(shost->host_lock, flags);
2144 if (shost->eh_deadline != -1)
2145 shost->last_reset = 0;
2146 spin_unlock_irqrestore(shost->host_lock, flags);
2147 scsi_eh_flush_done_q(&eh_done_q);
2151 * scsi_error_handler - SCSI error handler thread
2152 * @data: Host for which we are running.
2154 * Notes:
2155 * This is the main error handling loop. This is run as a kernel thread
2156 * for every SCSI host and handles all error handling activity.
2158 int scsi_error_handler(void *data)
2160 struct Scsi_Host *shost = data;
2163 * We use TASK_INTERRUPTIBLE so that the thread is not
2164 * counted against the load average as a running process.
2165 * We never actually get interrupted because kthread_run
2166 * disables signal delivery for the created thread.
2168 while (true) {
2170 * The sequence in kthread_stop() sets the stop flag first
2171 * then wakes the process. To avoid missed wakeups, the task
2172 * should always be in a non running state before the stop
2173 * flag is checked
2175 set_current_state(TASK_INTERRUPTIBLE);
2176 if (kthread_should_stop())
2177 break;
2179 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2180 shost->host_failed != atomic_read(&shost->host_busy)) {
2181 SCSI_LOG_ERROR_RECOVERY(1,
2182 shost_printk(KERN_INFO, shost,
2183 "scsi_eh_%d: sleeping\n",
2184 shost->host_no));
2185 schedule();
2186 continue;
2189 __set_current_state(TASK_RUNNING);
2190 SCSI_LOG_ERROR_RECOVERY(1,
2191 shost_printk(KERN_INFO, shost,
2192 "scsi_eh_%d: waking up %d/%d/%d\n",
2193 shost->host_no, shost->host_eh_scheduled,
2194 shost->host_failed,
2195 atomic_read(&shost->host_busy)));
2198 * We have a host that is failing for some reason. Figure out
2199 * what we need to do to get it up and online again (if we can).
2200 * If we fail, we end up taking the thing offline.
2202 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2203 SCSI_LOG_ERROR_RECOVERY(1,
2204 printk(KERN_ERR "Error handler scsi_eh_%d "
2205 "unable to autoresume\n",
2206 shost->host_no));
2207 continue;
2210 if (shost->transportt->eh_strategy_handler)
2211 shost->transportt->eh_strategy_handler(shost);
2212 else
2213 scsi_unjam_host(shost);
2215 /* All scmds have been handled */
2216 shost->host_failed = 0;
2219 * Note - if the above fails completely, the action is to take
2220 * individual devices offline and flush the queue of any
2221 * outstanding requests that may have been pending. When we
2222 * restart, we restart any I/O to any other devices on the bus
2223 * which are still online.
2225 scsi_restart_operations(shost);
2226 if (!shost->eh_noresume)
2227 scsi_autopm_put_host(shost);
2229 __set_current_state(TASK_RUNNING);
2231 SCSI_LOG_ERROR_RECOVERY(1,
2232 shost_printk(KERN_INFO, shost,
2233 "Error handler scsi_eh_%d exiting\n",
2234 shost->host_no));
2235 shost->ehandler = NULL;
2236 return 0;
2240 * Function: scsi_report_bus_reset()
2242 * Purpose: Utility function used by low-level drivers to report that
2243 * they have observed a bus reset on the bus being handled.
2245 * Arguments: shost - Host in question
2246 * channel - channel on which reset was observed.
2248 * Returns: Nothing
2250 * Lock status: Host lock must be held.
2252 * Notes: This only needs to be called if the reset is one which
2253 * originates from an unknown location. Resets originated
2254 * by the mid-level itself don't need to call this, but there
2255 * should be no harm.
2257 * The main purpose of this is to make sure that a CHECK_CONDITION
2258 * is properly treated.
2260 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2262 struct scsi_device *sdev;
2264 __shost_for_each_device(sdev, shost) {
2265 if (channel == sdev_channel(sdev))
2266 __scsi_report_device_reset(sdev, NULL);
2269 EXPORT_SYMBOL(scsi_report_bus_reset);
2272 * Function: scsi_report_device_reset()
2274 * Purpose: Utility function used by low-level drivers to report that
2275 * they have observed a device reset on the device being handled.
2277 * Arguments: shost - Host in question
2278 * channel - channel on which reset was observed
2279 * target - target on which reset was observed
2281 * Returns: Nothing
2283 * Lock status: Host lock must be held
2285 * Notes: This only needs to be called if the reset is one which
2286 * originates from an unknown location. Resets originated
2287 * by the mid-level itself don't need to call this, but there
2288 * should be no harm.
2290 * The main purpose of this is to make sure that a CHECK_CONDITION
2291 * is properly treated.
2293 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2295 struct scsi_device *sdev;
2297 __shost_for_each_device(sdev, shost) {
2298 if (channel == sdev_channel(sdev) &&
2299 target == sdev_id(sdev))
2300 __scsi_report_device_reset(sdev, NULL);
2303 EXPORT_SYMBOL(scsi_report_device_reset);
2305 static void
2306 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2311 * Function: scsi_reset_provider
2313 * Purpose: Send requested reset to a bus or device at any phase.
2315 * Arguments: device - device to send reset to
2316 * flag - reset type (see scsi.h)
2318 * Returns: SUCCESS/FAILURE.
2320 * Notes: This is used by the SCSI Generic driver to provide
2321 * Bus/Device reset capability.
2324 scsi_reset_provider(struct scsi_device *dev, int flag)
2326 struct scsi_cmnd *scmd;
2327 struct Scsi_Host *shost = dev->host;
2328 struct request req;
2329 unsigned long flags;
2330 int rtn;
2332 if (scsi_autopm_get_host(shost) < 0)
2333 return FAILED;
2335 if (!get_device(&dev->sdev_gendev)) {
2336 rtn = FAILED;
2337 goto out_put_autopm_host;
2340 scmd = scsi_get_command(dev, GFP_KERNEL);
2341 if (!scmd) {
2342 rtn = FAILED;
2343 put_device(&dev->sdev_gendev);
2344 goto out_put_autopm_host;
2347 blk_rq_init(NULL, &req);
2348 scmd->request = &req;
2350 scmd->cmnd = req.cmd;
2352 scmd->scsi_done = scsi_reset_provider_done_command;
2353 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2355 scmd->cmd_len = 0;
2357 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2359 spin_lock_irqsave(shost->host_lock, flags);
2360 shost->tmf_in_progress = 1;
2361 spin_unlock_irqrestore(shost->host_lock, flags);
2363 switch (flag) {
2364 case SCSI_TRY_RESET_DEVICE:
2365 rtn = scsi_try_bus_device_reset(scmd);
2366 if (rtn == SUCCESS)
2367 break;
2368 /* FALLTHROUGH */
2369 case SCSI_TRY_RESET_TARGET:
2370 rtn = scsi_try_target_reset(scmd);
2371 if (rtn == SUCCESS)
2372 break;
2373 /* FALLTHROUGH */
2374 case SCSI_TRY_RESET_BUS:
2375 rtn = scsi_try_bus_reset(scmd);
2376 if (rtn == SUCCESS)
2377 break;
2378 /* FALLTHROUGH */
2379 case SCSI_TRY_RESET_HOST:
2380 rtn = scsi_try_host_reset(scmd);
2381 break;
2382 default:
2383 rtn = FAILED;
2386 spin_lock_irqsave(shost->host_lock, flags);
2387 shost->tmf_in_progress = 0;
2388 spin_unlock_irqrestore(shost->host_lock, flags);
2391 * be sure to wake up anyone who was sleeping or had their queue
2392 * suspended while we performed the TMF.
2394 SCSI_LOG_ERROR_RECOVERY(3,
2395 shost_printk(KERN_INFO, shost,
2396 "waking up host to restart after TMF\n"));
2398 wake_up(&shost->host_wait);
2400 scsi_run_host_queues(shost);
2402 scsi_next_command(scmd);
2403 out_put_autopm_host:
2404 scsi_autopm_put_host(shost);
2405 return rtn;
2407 EXPORT_SYMBOL(scsi_reset_provider);
2410 * scsi_normalize_sense - normalize main elements from either fixed or
2411 * descriptor sense data format into a common format.
2413 * @sense_buffer: byte array containing sense data returned by device
2414 * @sb_len: number of valid bytes in sense_buffer
2415 * @sshdr: pointer to instance of structure that common
2416 * elements are written to.
2418 * Notes:
2419 * The "main elements" from sense data are: response_code, sense_key,
2420 * asc, ascq and additional_length (only for descriptor format).
2422 * Typically this function can be called after a device has
2423 * responded to a SCSI command with the CHECK_CONDITION status.
2425 * Return value:
2426 * 1 if valid sense data information found, else 0;
2428 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2429 struct scsi_sense_hdr *sshdr)
2431 if (!sense_buffer || !sb_len)
2432 return 0;
2434 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2436 sshdr->response_code = (sense_buffer[0] & 0x7f);
2438 if (!scsi_sense_valid(sshdr))
2439 return 0;
2441 if (sshdr->response_code >= 0x72) {
2443 * descriptor format
2445 if (sb_len > 1)
2446 sshdr->sense_key = (sense_buffer[1] & 0xf);
2447 if (sb_len > 2)
2448 sshdr->asc = sense_buffer[2];
2449 if (sb_len > 3)
2450 sshdr->ascq = sense_buffer[3];
2451 if (sb_len > 7)
2452 sshdr->additional_length = sense_buffer[7];
2453 } else {
2455 * fixed format
2457 if (sb_len > 2)
2458 sshdr->sense_key = (sense_buffer[2] & 0xf);
2459 if (sb_len > 7) {
2460 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2461 sb_len : (sense_buffer[7] + 8);
2462 if (sb_len > 12)
2463 sshdr->asc = sense_buffer[12];
2464 if (sb_len > 13)
2465 sshdr->ascq = sense_buffer[13];
2469 return 1;
2471 EXPORT_SYMBOL(scsi_normalize_sense);
2473 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2474 struct scsi_sense_hdr *sshdr)
2476 return scsi_normalize_sense(cmd->sense_buffer,
2477 SCSI_SENSE_BUFFERSIZE, sshdr);
2479 EXPORT_SYMBOL(scsi_command_normalize_sense);
2482 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2483 * @sense_buffer: byte array of descriptor format sense data
2484 * @sb_len: number of valid bytes in sense_buffer
2485 * @desc_type: value of descriptor type to find
2486 * (e.g. 0 -> information)
2488 * Notes:
2489 * only valid when sense data is in descriptor format
2491 * Return value:
2492 * pointer to start of (first) descriptor if found else NULL
2494 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2495 int desc_type)
2497 int add_sen_len, add_len, desc_len, k;
2498 const u8 * descp;
2500 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2501 return NULL;
2502 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2503 return NULL;
2504 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2505 add_sen_len : (sb_len - 8);
2506 descp = &sense_buffer[8];
2507 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2508 descp += desc_len;
2509 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2510 desc_len = add_len + 2;
2511 if (descp[0] == desc_type)
2512 return descp;
2513 if (add_len < 0) // short descriptor ??
2514 break;
2516 return NULL;
2518 EXPORT_SYMBOL(scsi_sense_desc_find);
2521 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2522 * @sense_buffer: byte array of sense data
2523 * @sb_len: number of valid bytes in sense_buffer
2524 * @info_out: pointer to 64 integer where 8 or 4 byte information
2525 * field will be placed if found.
2527 * Return value:
2528 * 1 if information field found, 0 if not found.
2530 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2531 u64 * info_out)
2533 int j;
2534 const u8 * ucp;
2535 u64 ull;
2537 if (sb_len < 7)
2538 return 0;
2539 switch (sense_buffer[0] & 0x7f) {
2540 case 0x70:
2541 case 0x71:
2542 if (sense_buffer[0] & 0x80) {
2543 *info_out = (sense_buffer[3] << 24) +
2544 (sense_buffer[4] << 16) +
2545 (sense_buffer[5] << 8) + sense_buffer[6];
2546 return 1;
2547 } else
2548 return 0;
2549 case 0x72:
2550 case 0x73:
2551 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2552 0 /* info desc */);
2553 if (ucp && (0xa == ucp[1])) {
2554 ull = 0;
2555 for (j = 0; j < 8; ++j) {
2556 if (j > 0)
2557 ull <<= 8;
2558 ull |= ucp[4 + j];
2560 *info_out = ull;
2561 return 1;
2562 } else
2563 return 0;
2564 default:
2565 return 0;
2568 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2571 * scsi_build_sense_buffer - build sense data in a buffer
2572 * @desc: Sense format (non zero == descriptor format,
2573 * 0 == fixed format)
2574 * @buf: Where to build sense data
2575 * @key: Sense key
2576 * @asc: Additional sense code
2577 * @ascq: Additional sense code qualifier
2580 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2582 if (desc) {
2583 buf[0] = 0x72; /* descriptor, current */
2584 buf[1] = key;
2585 buf[2] = asc;
2586 buf[3] = ascq;
2587 buf[7] = 0;
2588 } else {
2589 buf[0] = 0x70; /* fixed, current */
2590 buf[2] = key;
2591 buf[7] = 0xa;
2592 buf[12] = asc;
2593 buf[13] = ascq;
2596 EXPORT_SYMBOL(scsi_build_sense_buffer);