udf: improve error management in udf_CS0toUTF8()
[linux/fpc-iii.git] / drivers / scsi / scsi_error.c
blobc95a4e943fc6843b78d8120955318d76547a956e
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>
29 #include <asm/unaligned.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_transport.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_ioctl.h>
40 #include <scsi/sg.h>
42 #include "scsi_priv.h"
43 #include "scsi_logging.h"
44 #include "scsi_transport_api.h"
46 #include <trace/events/scsi.h>
48 static void scsi_eh_done(struct scsi_cmnd *scmd);
51 * These should *probably* be handled by the host itself.
52 * Since it is allowed to sleep, it probably should.
54 #define BUS_RESET_SETTLE_TIME (10)
55 #define HOST_RESET_SETTLE_TIME (10)
57 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
58 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
59 struct scsi_cmnd *);
61 /* called with shost->host_lock held */
62 void scsi_eh_wakeup(struct Scsi_Host *shost)
64 if (atomic_read(&shost->host_busy) == shost->host_failed) {
65 trace_scsi_eh_wakeup(shost);
66 wake_up_process(shost->ehandler);
67 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
68 "Waking error handler thread\n"));
72 /**
73 * scsi_schedule_eh - schedule EH for SCSI host
74 * @shost: SCSI host to invoke error handling on.
76 * Schedule SCSI EH without scmd.
78 void scsi_schedule_eh(struct Scsi_Host *shost)
80 unsigned long flags;
82 spin_lock_irqsave(shost->host_lock, flags);
84 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
85 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
86 shost->host_eh_scheduled++;
87 scsi_eh_wakeup(shost);
90 spin_unlock_irqrestore(shost->host_lock, flags);
92 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
94 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
96 if (!shost->last_reset || shost->eh_deadline == -1)
97 return 0;
100 * 32bit accesses are guaranteed to be atomic
101 * (on all supported architectures), so instead
102 * of using a spinlock we can as well double check
103 * if eh_deadline has been set to 'off' during the
104 * time_before call.
106 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
107 shost->eh_deadline > -1)
108 return 0;
110 return 1;
114 * scmd_eh_abort_handler - Handle command aborts
115 * @work: command to be aborted.
117 void
118 scmd_eh_abort_handler(struct work_struct *work)
120 struct scsi_cmnd *scmd =
121 container_of(work, struct scsi_cmnd, abort_work.work);
122 struct scsi_device *sdev = scmd->device;
123 int rtn;
125 if (scsi_host_eh_past_deadline(sdev->host)) {
126 SCSI_LOG_ERROR_RECOVERY(3,
127 scmd_printk(KERN_INFO, scmd,
128 "eh timeout, not aborting\n"));
129 } else {
130 SCSI_LOG_ERROR_RECOVERY(3,
131 scmd_printk(KERN_INFO, scmd,
132 "aborting command\n"));
133 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
134 if (rtn == SUCCESS) {
135 set_host_byte(scmd, DID_TIME_OUT);
136 if (scsi_host_eh_past_deadline(sdev->host)) {
137 SCSI_LOG_ERROR_RECOVERY(3,
138 scmd_printk(KERN_INFO, scmd,
139 "eh timeout, not retrying "
140 "aborted command\n"));
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 "retry aborted command\n"));
146 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
147 return;
148 } else {
149 SCSI_LOG_ERROR_RECOVERY(3,
150 scmd_printk(KERN_WARNING, scmd,
151 "finish aborted command\n"));
152 scsi_finish_command(scmd);
153 return;
155 } else {
156 SCSI_LOG_ERROR_RECOVERY(3,
157 scmd_printk(KERN_INFO, scmd,
158 "cmd abort %s\n",
159 (rtn == FAST_IO_FAIL) ?
160 "not send" : "failed"));
164 if (!scsi_eh_scmd_add(scmd, 0)) {
165 SCSI_LOG_ERROR_RECOVERY(3,
166 scmd_printk(KERN_WARNING, scmd,
167 "terminate aborted command\n"));
168 set_host_byte(scmd, DID_TIME_OUT);
169 scsi_finish_command(scmd);
174 * scsi_abort_command - schedule a command abort
175 * @scmd: scmd to abort.
177 * We only need to abort commands after a command timeout
179 static int
180 scsi_abort_command(struct scsi_cmnd *scmd)
182 struct scsi_device *sdev = scmd->device;
183 struct Scsi_Host *shost = sdev->host;
184 unsigned long flags;
186 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
188 * Retry after abort failed, escalate to next level.
190 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
191 SCSI_LOG_ERROR_RECOVERY(3,
192 scmd_printk(KERN_INFO, scmd,
193 "previous abort failed\n"));
194 BUG_ON(delayed_work_pending(&scmd->abort_work));
195 return FAILED;
199 * Do not try a command abort if
200 * SCSI EH has already started.
202 spin_lock_irqsave(shost->host_lock, flags);
203 if (scsi_host_in_recovery(shost)) {
204 spin_unlock_irqrestore(shost->host_lock, flags);
205 SCSI_LOG_ERROR_RECOVERY(3,
206 scmd_printk(KERN_INFO, scmd,
207 "not aborting, host in recovery\n"));
208 return FAILED;
211 if (shost->eh_deadline != -1 && !shost->last_reset)
212 shost->last_reset = jiffies;
213 spin_unlock_irqrestore(shost->host_lock, flags);
215 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
216 SCSI_LOG_ERROR_RECOVERY(3,
217 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
218 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
219 return SUCCESS;
223 * scsi_eh_scmd_add - add scsi cmd to error handling.
224 * @scmd: scmd to run eh on.
225 * @eh_flag: optional SCSI_EH flag.
227 * Return value:
228 * 0 on failure.
230 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
232 struct Scsi_Host *shost = scmd->device->host;
233 unsigned long flags;
234 int ret = 0;
236 if (!shost->ehandler)
237 return 0;
239 spin_lock_irqsave(shost->host_lock, flags);
240 if (scsi_host_set_state(shost, SHOST_RECOVERY))
241 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
242 goto out_unlock;
244 if (shost->eh_deadline != -1 && !shost->last_reset)
245 shost->last_reset = jiffies;
247 ret = 1;
248 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
249 eh_flag &= ~SCSI_EH_CANCEL_CMD;
250 scmd->eh_eflags |= eh_flag;
251 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
252 shost->host_failed++;
253 scsi_eh_wakeup(shost);
254 out_unlock:
255 spin_unlock_irqrestore(shost->host_lock, flags);
256 return ret;
260 * scsi_times_out - Timeout function for normal scsi commands.
261 * @req: request that is timing out.
263 * Notes:
264 * We do not need to lock this. There is the potential for a race
265 * only in that the normal completion handling might run, but if the
266 * normal completion function determines that the timer has already
267 * fired, then it mustn't do anything.
269 enum blk_eh_timer_return scsi_times_out(struct request *req)
271 struct scsi_cmnd *scmd = req->special;
272 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
273 struct Scsi_Host *host = scmd->device->host;
275 trace_scsi_dispatch_cmd_timeout(scmd);
276 scsi_log_completion(scmd, TIMEOUT_ERROR);
278 if (host->eh_deadline != -1 && !host->last_reset)
279 host->last_reset = jiffies;
281 if (host->transportt->eh_timed_out)
282 rtn = host->transportt->eh_timed_out(scmd);
283 else if (host->hostt->eh_timed_out)
284 rtn = host->hostt->eh_timed_out(scmd);
286 if (rtn == BLK_EH_NOT_HANDLED) {
287 if (!host->hostt->no_async_abort &&
288 scsi_abort_command(scmd) == SUCCESS)
289 return BLK_EH_NOT_HANDLED;
291 set_host_byte(scmd, DID_TIME_OUT);
292 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
293 rtn = BLK_EH_HANDLED;
296 return rtn;
300 * scsi_block_when_processing_errors - Prevent cmds from being queued.
301 * @sdev: Device on which we are performing recovery.
303 * Description:
304 * We block until the host is out of error recovery, and then check to
305 * see whether the host or the device is offline.
307 * Return value:
308 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
310 int scsi_block_when_processing_errors(struct scsi_device *sdev)
312 int online;
314 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
316 online = scsi_device_online(sdev);
318 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
319 "%s: rtn: %d\n", __func__, online));
321 return online;
323 EXPORT_SYMBOL(scsi_block_when_processing_errors);
325 #ifdef CONFIG_SCSI_LOGGING
327 * scsi_eh_prt_fail_stats - Log info on failures.
328 * @shost: scsi host being recovered.
329 * @work_q: Queue of scsi cmds to process.
331 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
332 struct list_head *work_q)
334 struct scsi_cmnd *scmd;
335 struct scsi_device *sdev;
336 int total_failures = 0;
337 int cmd_failed = 0;
338 int cmd_cancel = 0;
339 int devices_failed = 0;
341 shost_for_each_device(sdev, shost) {
342 list_for_each_entry(scmd, work_q, eh_entry) {
343 if (scmd->device == sdev) {
344 ++total_failures;
345 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
346 ++cmd_cancel;
347 else
348 ++cmd_failed;
352 if (cmd_cancel || cmd_failed) {
353 SCSI_LOG_ERROR_RECOVERY(3,
354 shost_printk(KERN_INFO, shost,
355 "%s: cmds failed: %d, cancel: %d\n",
356 __func__, cmd_failed,
357 cmd_cancel));
358 cmd_cancel = 0;
359 cmd_failed = 0;
360 ++devices_failed;
364 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
365 "Total of %d commands on %d"
366 " devices require eh work\n",
367 total_failures, devices_failed));
369 #endif
372 * scsi_report_lun_change - Set flag on all *other* devices on the same target
373 * to indicate that a UNIT ATTENTION is expected.
374 * @sdev: Device reporting the UNIT ATTENTION
376 static void scsi_report_lun_change(struct scsi_device *sdev)
378 sdev->sdev_target->expecting_lun_change = 1;
382 * scsi_report_sense - Examine scsi sense information and log messages for
383 * certain conditions, also issue uevents for some of them.
384 * @sdev: Device reporting the sense code
385 * @sshdr: sshdr to be examined
387 static void scsi_report_sense(struct scsi_device *sdev,
388 struct scsi_sense_hdr *sshdr)
390 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
392 if (sshdr->sense_key == UNIT_ATTENTION) {
393 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
394 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
395 sdev_printk(KERN_WARNING, sdev,
396 "Inquiry data has changed");
397 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
398 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
399 scsi_report_lun_change(sdev);
400 sdev_printk(KERN_WARNING, sdev,
401 "Warning! Received an indication that the "
402 "LUN assignments on this target have "
403 "changed. The Linux SCSI layer does not "
404 "automatically remap LUN assignments.\n");
405 } else if (sshdr->asc == 0x3f)
406 sdev_printk(KERN_WARNING, sdev,
407 "Warning! Received an indication that the "
408 "operating parameters on this target have "
409 "changed. The Linux SCSI layer does not "
410 "automatically adjust these parameters.\n");
412 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
413 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
414 sdev_printk(KERN_WARNING, sdev,
415 "Warning! Received an indication that the "
416 "LUN reached a thin provisioning soft "
417 "threshold.\n");
420 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
421 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
422 sdev_printk(KERN_WARNING, sdev,
423 "Mode parameters changed");
424 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
425 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
426 sdev_printk(KERN_WARNING, sdev,
427 "Capacity data has changed");
428 } else if (sshdr->asc == 0x2a)
429 sdev_printk(KERN_WARNING, sdev,
430 "Parameters changed");
433 if (evt_type != SDEV_EVT_MAXBITS) {
434 set_bit(evt_type, sdev->pending_events);
435 schedule_work(&sdev->event_work);
440 * scsi_check_sense - Examine scsi cmd sense
441 * @scmd: Cmd to have sense checked.
443 * Return value:
444 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
446 * Notes:
447 * When a deferred error is detected the current command has
448 * not been executed and needs retrying.
450 static int scsi_check_sense(struct scsi_cmnd *scmd)
452 struct scsi_device *sdev = scmd->device;
453 struct scsi_sense_hdr sshdr;
455 if (! scsi_command_normalize_sense(scmd, &sshdr))
456 return FAILED; /* no valid sense data */
458 scsi_report_sense(sdev, &sshdr);
460 if (scsi_sense_is_deferred(&sshdr))
461 return NEEDS_RETRY;
463 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
464 sdev->scsi_dh_data->scsi_dh->check_sense) {
465 int rc;
467 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
468 if (rc != SCSI_RETURN_NOT_HANDLED)
469 return rc;
470 /* handler does not care. Drop down to default handling */
473 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
475 * nasty: for mid-layer issued TURs, we need to return the
476 * actual sense data without any recovery attempt. For eh
477 * issued ones, we need to try to recover and interpret
479 return SUCCESS;
482 * Previous logic looked for FILEMARK, EOM or ILI which are
483 * mainly associated with tapes and returned SUCCESS.
485 if (sshdr.response_code == 0x70) {
486 /* fixed format */
487 if (scmd->sense_buffer[2] & 0xe0)
488 return SUCCESS;
489 } else {
491 * descriptor format: look for "stream commands sense data
492 * descriptor" (see SSC-3). Assume single sense data
493 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
495 if ((sshdr.additional_length > 3) &&
496 (scmd->sense_buffer[8] == 0x4) &&
497 (scmd->sense_buffer[11] & 0xe0))
498 return SUCCESS;
501 switch (sshdr.sense_key) {
502 case NO_SENSE:
503 return SUCCESS;
504 case RECOVERED_ERROR:
505 return /* soft_error */ SUCCESS;
507 case ABORTED_COMMAND:
508 if (sshdr.asc == 0x10) /* DIF */
509 return SUCCESS;
511 return NEEDS_RETRY;
512 case NOT_READY:
513 case UNIT_ATTENTION:
515 * if we are expecting a cc/ua because of a bus reset that we
516 * performed, treat this just as a retry. otherwise this is
517 * information that we should pass up to the upper-level driver
518 * so that we can deal with it there.
520 if (scmd->device->expecting_cc_ua) {
522 * Because some device does not queue unit
523 * attentions correctly, we carefully check
524 * additional sense code and qualifier so as
525 * not to squash media change unit attention.
527 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
528 scmd->device->expecting_cc_ua = 0;
529 return NEEDS_RETRY;
533 * we might also expect a cc/ua if another LUN on the target
534 * reported a UA with an ASC/ASCQ of 3F 0E -
535 * REPORTED LUNS DATA HAS CHANGED.
537 if (scmd->device->sdev_target->expecting_lun_change &&
538 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
539 return NEEDS_RETRY;
541 * if the device is in the process of becoming ready, we
542 * should retry.
544 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
545 return NEEDS_RETRY;
547 * if the device is not started, we need to wake
548 * the error handler to start the motor
550 if (scmd->device->allow_restart &&
551 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
552 return FAILED;
554 * Pass the UA upwards for a determination in the completion
555 * functions.
557 return SUCCESS;
559 /* these are not supported */
560 case DATA_PROTECT:
561 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
562 /* Thin provisioning hard threshold reached */
563 set_host_byte(scmd, DID_ALLOC_FAILURE);
564 return SUCCESS;
566 case COPY_ABORTED:
567 case VOLUME_OVERFLOW:
568 case MISCOMPARE:
569 case BLANK_CHECK:
570 set_host_byte(scmd, DID_TARGET_FAILURE);
571 return SUCCESS;
573 case MEDIUM_ERROR:
574 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
575 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
576 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
577 set_host_byte(scmd, DID_MEDIUM_ERROR);
578 return SUCCESS;
580 return NEEDS_RETRY;
582 case HARDWARE_ERROR:
583 if (scmd->device->retry_hwerror)
584 return ADD_TO_MLQUEUE;
585 else
586 set_host_byte(scmd, DID_TARGET_FAILURE);
588 case ILLEGAL_REQUEST:
589 if (sshdr.asc == 0x20 || /* Invalid command operation code */
590 sshdr.asc == 0x21 || /* Logical block address out of range */
591 sshdr.asc == 0x24 || /* Invalid field in cdb */
592 sshdr.asc == 0x26) { /* Parameter value invalid */
593 set_host_byte(scmd, DID_TARGET_FAILURE);
595 return SUCCESS;
597 default:
598 return SUCCESS;
602 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
604 struct scsi_host_template *sht = sdev->host->hostt;
605 struct scsi_device *tmp_sdev;
607 if (!sht->track_queue_depth ||
608 sdev->queue_depth >= sdev->max_queue_depth)
609 return;
611 if (time_before(jiffies,
612 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
613 return;
615 if (time_before(jiffies,
616 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
617 return;
620 * Walk all devices of a target and do
621 * ramp up on them.
623 shost_for_each_device(tmp_sdev, sdev->host) {
624 if (tmp_sdev->channel != sdev->channel ||
625 tmp_sdev->id != sdev->id ||
626 tmp_sdev->queue_depth == sdev->max_queue_depth)
627 continue;
629 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
630 sdev->last_queue_ramp_up = jiffies;
634 static void scsi_handle_queue_full(struct scsi_device *sdev)
636 struct scsi_host_template *sht = sdev->host->hostt;
637 struct scsi_device *tmp_sdev;
639 if (!sht->track_queue_depth)
640 return;
642 shost_for_each_device(tmp_sdev, sdev->host) {
643 if (tmp_sdev->channel != sdev->channel ||
644 tmp_sdev->id != sdev->id)
645 continue;
647 * We do not know the number of commands that were at
648 * the device when we got the queue full so we start
649 * from the highest possible value and work our way down.
651 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
656 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
657 * @scmd: SCSI cmd to examine.
659 * Notes:
660 * This is *only* called when we are examining the status of commands
661 * queued during error recovery. the main difference here is that we
662 * don't allow for the possibility of retries here, and we are a lot
663 * more restrictive about what we consider acceptable.
665 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
668 * first check the host byte, to see if there is anything in there
669 * that would indicate what we need to do.
671 if (host_byte(scmd->result) == DID_RESET) {
673 * rats. we are already in the error handler, so we now
674 * get to try and figure out what to do next. if the sense
675 * is valid, we have a pretty good idea of what to do.
676 * if not, we mark it as FAILED.
678 return scsi_check_sense(scmd);
680 if (host_byte(scmd->result) != DID_OK)
681 return FAILED;
684 * next, check the message byte.
686 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
687 return FAILED;
690 * now, check the status byte to see if this indicates
691 * anything special.
693 switch (status_byte(scmd->result)) {
694 case GOOD:
695 scsi_handle_queue_ramp_up(scmd->device);
696 case COMMAND_TERMINATED:
697 return SUCCESS;
698 case CHECK_CONDITION:
699 return scsi_check_sense(scmd);
700 case CONDITION_GOOD:
701 case INTERMEDIATE_GOOD:
702 case INTERMEDIATE_C_GOOD:
704 * who knows? FIXME(eric)
706 return SUCCESS;
707 case RESERVATION_CONFLICT:
708 if (scmd->cmnd[0] == TEST_UNIT_READY)
709 /* it is a success, we probed the device and
710 * found it */
711 return SUCCESS;
712 /* otherwise, we failed to send the command */
713 return FAILED;
714 case QUEUE_FULL:
715 scsi_handle_queue_full(scmd->device);
716 /* fall through */
717 case BUSY:
718 return NEEDS_RETRY;
719 default:
720 return FAILED;
722 return FAILED;
726 * scsi_eh_done - Completion function for error handling.
727 * @scmd: Cmd that is done.
729 static void scsi_eh_done(struct scsi_cmnd *scmd)
731 struct completion *eh_action;
733 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
734 "%s result: %x\n", __func__, scmd->result));
736 eh_action = scmd->device->host->eh_action;
737 if (eh_action)
738 complete(eh_action);
742 * scsi_try_host_reset - ask host adapter to reset itself
743 * @scmd: SCSI cmd to send host reset.
745 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
747 unsigned long flags;
748 int rtn;
749 struct Scsi_Host *host = scmd->device->host;
750 struct scsi_host_template *hostt = host->hostt;
752 SCSI_LOG_ERROR_RECOVERY(3,
753 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
755 if (!hostt->eh_host_reset_handler)
756 return FAILED;
758 rtn = hostt->eh_host_reset_handler(scmd);
760 if (rtn == SUCCESS) {
761 if (!hostt->skip_settle_delay)
762 ssleep(HOST_RESET_SETTLE_TIME);
763 spin_lock_irqsave(host->host_lock, flags);
764 scsi_report_bus_reset(host, scmd_channel(scmd));
765 spin_unlock_irqrestore(host->host_lock, flags);
768 return rtn;
772 * scsi_try_bus_reset - ask host to perform a bus reset
773 * @scmd: SCSI cmd to send bus reset.
775 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
777 unsigned long flags;
778 int rtn;
779 struct Scsi_Host *host = scmd->device->host;
780 struct scsi_host_template *hostt = host->hostt;
782 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
783 "%s: Snd Bus RST\n", __func__));
785 if (!hostt->eh_bus_reset_handler)
786 return FAILED;
788 rtn = hostt->eh_bus_reset_handler(scmd);
790 if (rtn == SUCCESS) {
791 if (!hostt->skip_settle_delay)
792 ssleep(BUS_RESET_SETTLE_TIME);
793 spin_lock_irqsave(host->host_lock, flags);
794 scsi_report_bus_reset(host, scmd_channel(scmd));
795 spin_unlock_irqrestore(host->host_lock, flags);
798 return rtn;
801 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
803 sdev->was_reset = 1;
804 sdev->expecting_cc_ua = 1;
808 * scsi_try_target_reset - Ask host to perform a target reset
809 * @scmd: SCSI cmd used to send a target reset
811 * Notes:
812 * There is no timeout for this operation. if this operation is
813 * unreliable for a given host, then the host itself needs to put a
814 * timer on it, and set the host back to a consistent state prior to
815 * returning.
817 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
819 unsigned long flags;
820 int rtn;
821 struct Scsi_Host *host = scmd->device->host;
822 struct scsi_host_template *hostt = host->hostt;
824 if (!hostt->eh_target_reset_handler)
825 return FAILED;
827 rtn = hostt->eh_target_reset_handler(scmd);
828 if (rtn == SUCCESS) {
829 spin_lock_irqsave(host->host_lock, flags);
830 __starget_for_each_device(scsi_target(scmd->device), NULL,
831 __scsi_report_device_reset);
832 spin_unlock_irqrestore(host->host_lock, flags);
835 return rtn;
839 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
840 * @scmd: SCSI cmd used to send BDR
842 * Notes:
843 * There is no timeout for this operation. if this operation is
844 * unreliable for a given host, then the host itself needs to put a
845 * timer on it, and set the host back to a consistent state prior to
846 * returning.
848 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
850 int rtn;
851 struct scsi_host_template *hostt = scmd->device->host->hostt;
853 if (!hostt->eh_device_reset_handler)
854 return FAILED;
856 rtn = hostt->eh_device_reset_handler(scmd);
857 if (rtn == SUCCESS)
858 __scsi_report_device_reset(scmd->device, NULL);
859 return rtn;
863 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
864 * @hostt: SCSI driver host template
865 * @scmd: SCSI cmd used to send a target reset
867 * Return value:
868 * SUCCESS, FAILED, or FAST_IO_FAIL
870 * Notes:
871 * SUCCESS does not necessarily indicate that the command
872 * has been aborted; it only indicates that the LLDDs
873 * has cleared all references to that command.
874 * LLDDs should return FAILED only if an abort was required
875 * but could not be executed. LLDDs should return FAST_IO_FAIL
876 * if the device is temporarily unavailable (eg due to a
877 * link down on FibreChannel)
879 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
880 struct scsi_cmnd *scmd)
882 if (!hostt->eh_abort_handler)
883 return FAILED;
885 return hostt->eh_abort_handler(scmd);
888 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
890 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
891 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
892 if (scsi_try_target_reset(scmd) != SUCCESS)
893 if (scsi_try_bus_reset(scmd) != SUCCESS)
894 scsi_try_host_reset(scmd);
898 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
899 * @scmd: SCSI command structure to hijack
900 * @ses: structure to save restore information
901 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
902 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
903 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
905 * This function is used to save a scsi command information before re-execution
906 * as part of the error recovery process. If @sense_bytes is 0 the command
907 * sent must be one that does not transfer any data. If @sense_bytes != 0
908 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
909 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
911 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
912 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
914 struct scsi_device *sdev = scmd->device;
917 * We need saved copies of a number of fields - this is because
918 * error handling may need to overwrite these with different values
919 * to run different commands, and once error handling is complete,
920 * we will need to restore these values prior to running the actual
921 * command.
923 ses->cmd_len = scmd->cmd_len;
924 ses->cmnd = scmd->cmnd;
925 ses->data_direction = scmd->sc_data_direction;
926 ses->sdb = scmd->sdb;
927 ses->next_rq = scmd->request->next_rq;
928 ses->result = scmd->result;
929 ses->underflow = scmd->underflow;
930 ses->prot_op = scmd->prot_op;
932 scmd->prot_op = SCSI_PROT_NORMAL;
933 scmd->eh_eflags = 0;
934 scmd->cmnd = ses->eh_cmnd;
935 memset(scmd->cmnd, 0, BLK_MAX_CDB);
936 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
937 scmd->request->next_rq = NULL;
938 scmd->result = 0;
940 if (sense_bytes) {
941 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
942 sense_bytes);
943 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
944 scmd->sdb.length);
945 scmd->sdb.table.sgl = &ses->sense_sgl;
946 scmd->sc_data_direction = DMA_FROM_DEVICE;
947 scmd->sdb.table.nents = 1;
948 scmd->cmnd[0] = REQUEST_SENSE;
949 scmd->cmnd[4] = scmd->sdb.length;
950 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
951 } else {
952 scmd->sc_data_direction = DMA_NONE;
953 if (cmnd) {
954 BUG_ON(cmnd_size > BLK_MAX_CDB);
955 memcpy(scmd->cmnd, cmnd, cmnd_size);
956 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
960 scmd->underflow = 0;
962 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
963 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
964 (sdev->lun << 5 & 0xe0);
967 * Zero the sense buffer. The scsi spec mandates that any
968 * untransferred sense data should be interpreted as being zero.
970 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
972 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
975 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
976 * @scmd: SCSI command structure to restore
977 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
979 * Undo any damage done by above scsi_eh_prep_cmnd().
981 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
984 * Restore original data
986 scmd->cmd_len = ses->cmd_len;
987 scmd->cmnd = ses->cmnd;
988 scmd->sc_data_direction = ses->data_direction;
989 scmd->sdb = ses->sdb;
990 scmd->request->next_rq = ses->next_rq;
991 scmd->result = ses->result;
992 scmd->underflow = ses->underflow;
993 scmd->prot_op = ses->prot_op;
995 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
998 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
999 * @scmd: SCSI command structure to hijack
1000 * @cmnd: CDB to send
1001 * @cmnd_size: size in bytes of @cmnd
1002 * @timeout: timeout for this request
1003 * @sense_bytes: size of sense data to copy or 0
1005 * This function is used to send a scsi command down to a target device
1006 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1008 * Return value:
1009 * SUCCESS or FAILED or NEEDS_RETRY
1011 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1012 int cmnd_size, int timeout, unsigned sense_bytes)
1014 struct scsi_device *sdev = scmd->device;
1015 struct Scsi_Host *shost = sdev->host;
1016 DECLARE_COMPLETION_ONSTACK(done);
1017 unsigned long timeleft = timeout;
1018 struct scsi_eh_save ses;
1019 const unsigned long stall_for = msecs_to_jiffies(100);
1020 int rtn;
1022 retry:
1023 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1024 shost->eh_action = &done;
1026 scsi_log_send(scmd);
1027 scmd->scsi_done = scsi_eh_done;
1028 rtn = shost->hostt->queuecommand(shost, scmd);
1029 if (rtn) {
1030 if (timeleft > stall_for) {
1031 scsi_eh_restore_cmnd(scmd, &ses);
1032 timeleft -= stall_for;
1033 msleep(jiffies_to_msecs(stall_for));
1034 goto retry;
1036 /* signal not to enter either branch of the if () below */
1037 timeleft = 0;
1038 rtn = FAILED;
1039 } else {
1040 timeleft = wait_for_completion_timeout(&done, timeout);
1041 rtn = SUCCESS;
1044 shost->eh_action = NULL;
1046 scsi_log_completion(scmd, rtn);
1048 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1049 "%s timeleft: %ld\n",
1050 __func__, timeleft));
1053 * If there is time left scsi_eh_done got called, and we will examine
1054 * the actual status codes to see whether the command actually did
1055 * complete normally, else if we have a zero return and no time left,
1056 * the command must still be pending, so abort it and return FAILED.
1057 * If we never actually managed to issue the command, because
1058 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1059 * value above (so don't execute either branch of the if)
1061 if (timeleft) {
1062 rtn = scsi_eh_completed_normally(scmd);
1063 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1064 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1066 switch (rtn) {
1067 case SUCCESS:
1068 case NEEDS_RETRY:
1069 case FAILED:
1070 break;
1071 case ADD_TO_MLQUEUE:
1072 rtn = NEEDS_RETRY;
1073 break;
1074 default:
1075 rtn = FAILED;
1076 break;
1078 } else if (rtn != FAILED) {
1079 scsi_abort_eh_cmnd(scmd);
1080 rtn = FAILED;
1083 scsi_eh_restore_cmnd(scmd, &ses);
1085 return rtn;
1089 * scsi_request_sense - Request sense data from a particular target.
1090 * @scmd: SCSI cmd for request sense.
1092 * Notes:
1093 * Some hosts automatically obtain this information, others require
1094 * that we obtain it on our own. This function will *not* return until
1095 * the command either times out, or it completes.
1097 static int scsi_request_sense(struct scsi_cmnd *scmd)
1099 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1102 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1104 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1105 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1106 if (sdrv->eh_action)
1107 rtn = sdrv->eh_action(scmd, rtn);
1109 return rtn;
1113 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1114 * @scmd: Original SCSI cmd that eh has finished.
1115 * @done_q: Queue for processed commands.
1117 * Notes:
1118 * We don't want to use the normal command completion while we are are
1119 * still handling errors - it may cause other commands to be queued,
1120 * and that would disturb what we are doing. Thus we really want to
1121 * keep a list of pending commands for final completion, and once we
1122 * are ready to leave error handling we handle completion for real.
1124 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1126 scmd->device->host->host_failed--;
1127 scmd->eh_eflags = 0;
1128 list_move_tail(&scmd->eh_entry, done_q);
1130 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1133 * scsi_eh_get_sense - Get device sense data.
1134 * @work_q: Queue of commands to process.
1135 * @done_q: Queue of processed commands.
1137 * Description:
1138 * See if we need to request sense information. if so, then get it
1139 * now, so we have a better idea of what to do.
1141 * Notes:
1142 * This has the unfortunate side effect that if a shost adapter does
1143 * not automatically request sense information, we end up shutting
1144 * it down before we request it.
1146 * All drivers should request sense information internally these days,
1147 * so for now all I have to say is tough noogies if you end up in here.
1149 * XXX: Long term this code should go away, but that needs an audit of
1150 * all LLDDs first.
1152 int scsi_eh_get_sense(struct list_head *work_q,
1153 struct list_head *done_q)
1155 struct scsi_cmnd *scmd, *next;
1156 struct Scsi_Host *shost;
1157 int rtn;
1159 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1160 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1161 SCSI_SENSE_VALID(scmd))
1162 continue;
1164 shost = scmd->device->host;
1165 if (scsi_host_eh_past_deadline(shost)) {
1166 SCSI_LOG_ERROR_RECOVERY(3,
1167 scmd_printk(KERN_INFO, scmd,
1168 "%s: skip request sense, past eh deadline\n",
1169 current->comm));
1170 break;
1172 if (status_byte(scmd->result) != CHECK_CONDITION)
1174 * don't request sense if there's no check condition
1175 * status because the error we're processing isn't one
1176 * that has a sense code (and some devices get
1177 * confused by sense requests out of the blue)
1179 continue;
1181 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1182 "%s: requesting sense\n",
1183 current->comm));
1184 rtn = scsi_request_sense(scmd);
1185 if (rtn != SUCCESS)
1186 continue;
1188 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1189 "sense requested, result %x\n", scmd->result));
1190 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1192 rtn = scsi_decide_disposition(scmd);
1195 * if the result was normal, then just pass it along to the
1196 * upper level.
1198 if (rtn == SUCCESS)
1199 /* we don't want this command reissued, just
1200 * finished with the sense data, so set
1201 * retries to the max allowed to ensure it
1202 * won't get reissued */
1203 scmd->retries = scmd->allowed;
1204 else if (rtn != NEEDS_RETRY)
1205 continue;
1207 scsi_eh_finish_cmd(scmd, done_q);
1210 return list_empty(work_q);
1212 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1215 * scsi_eh_tur - Send TUR to device.
1216 * @scmd: &scsi_cmnd to send TUR
1218 * Return value:
1219 * 0 - Device is ready. 1 - Device NOT ready.
1221 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1223 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1224 int retry_cnt = 1, rtn;
1226 retry_tur:
1227 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1228 scmd->device->eh_timeout, 0);
1230 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1231 "%s return: %x\n", __func__, rtn));
1233 switch (rtn) {
1234 case NEEDS_RETRY:
1235 if (retry_cnt--)
1236 goto retry_tur;
1237 /*FALLTHRU*/
1238 case SUCCESS:
1239 return 0;
1240 default:
1241 return 1;
1246 * scsi_eh_test_devices - check if devices are responding from error recovery.
1247 * @cmd_list: scsi commands in error recovery.
1248 * @work_q: queue for commands which still need more error recovery
1249 * @done_q: queue for commands which are finished
1250 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1252 * Decription:
1253 * Tests if devices are in a working state. Commands to devices now in
1254 * a working state are sent to the done_q while commands to devices which
1255 * are still failing to respond are returned to the work_q for more
1256 * processing.
1258 static int scsi_eh_test_devices(struct list_head *cmd_list,
1259 struct list_head *work_q,
1260 struct list_head *done_q, int try_stu)
1262 struct scsi_cmnd *scmd, *next;
1263 struct scsi_device *sdev;
1264 int finish_cmds;
1266 while (!list_empty(cmd_list)) {
1267 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1268 sdev = scmd->device;
1270 if (!try_stu) {
1271 if (scsi_host_eh_past_deadline(sdev->host)) {
1272 /* Push items back onto work_q */
1273 list_splice_init(cmd_list, work_q);
1274 SCSI_LOG_ERROR_RECOVERY(3,
1275 sdev_printk(KERN_INFO, sdev,
1276 "%s: skip test device, past eh deadline",
1277 current->comm));
1278 break;
1282 finish_cmds = !scsi_device_online(scmd->device) ||
1283 (try_stu && !scsi_eh_try_stu(scmd) &&
1284 !scsi_eh_tur(scmd)) ||
1285 !scsi_eh_tur(scmd);
1287 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1288 if (scmd->device == sdev) {
1289 if (finish_cmds &&
1290 (try_stu ||
1291 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1292 scsi_eh_finish_cmd(scmd, done_q);
1293 else
1294 list_move_tail(&scmd->eh_entry, work_q);
1297 return list_empty(work_q);
1302 * scsi_eh_abort_cmds - abort pending commands.
1303 * @work_q: &list_head for pending commands.
1304 * @done_q: &list_head for processed commands.
1306 * Decription:
1307 * Try and see whether or not it makes sense to try and abort the
1308 * running command. This only works out to be the case if we have one
1309 * command that has timed out. If the command simply failed, it makes
1310 * no sense to try and abort the command, since as far as the shost
1311 * adapter is concerned, it isn't running.
1313 static int scsi_eh_abort_cmds(struct list_head *work_q,
1314 struct list_head *done_q)
1316 struct scsi_cmnd *scmd, *next;
1317 LIST_HEAD(check_list);
1318 int rtn;
1319 struct Scsi_Host *shost;
1321 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1322 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1323 continue;
1324 shost = scmd->device->host;
1325 if (scsi_host_eh_past_deadline(shost)) {
1326 list_splice_init(&check_list, work_q);
1327 SCSI_LOG_ERROR_RECOVERY(3,
1328 scmd_printk(KERN_INFO, scmd,
1329 "%s: skip aborting cmd, past eh deadline\n",
1330 current->comm));
1331 return list_empty(work_q);
1333 SCSI_LOG_ERROR_RECOVERY(3,
1334 scmd_printk(KERN_INFO, scmd,
1335 "%s: aborting cmd\n", current->comm));
1336 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1337 if (rtn == FAILED) {
1338 SCSI_LOG_ERROR_RECOVERY(3,
1339 scmd_printk(KERN_INFO, scmd,
1340 "%s: aborting cmd failed\n",
1341 current->comm));
1342 list_splice_init(&check_list, work_q);
1343 return list_empty(work_q);
1345 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1346 if (rtn == FAST_IO_FAIL)
1347 scsi_eh_finish_cmd(scmd, done_q);
1348 else
1349 list_move_tail(&scmd->eh_entry, &check_list);
1352 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1356 * scsi_eh_try_stu - Send START_UNIT to device.
1357 * @scmd: &scsi_cmnd to send START_UNIT
1359 * Return value:
1360 * 0 - Device is ready. 1 - Device NOT ready.
1362 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1364 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1366 if (scmd->device->allow_restart) {
1367 int i, rtn = NEEDS_RETRY;
1369 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1370 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1372 if (rtn == SUCCESS)
1373 return 0;
1376 return 1;
1380 * scsi_eh_stu - send START_UNIT if needed
1381 * @shost: &scsi host being recovered.
1382 * @work_q: &list_head for pending commands.
1383 * @done_q: &list_head for processed commands.
1385 * Notes:
1386 * If commands are failing due to not ready, initializing command required,
1387 * try revalidating the device, which will end up sending a start unit.
1389 static int scsi_eh_stu(struct Scsi_Host *shost,
1390 struct list_head *work_q,
1391 struct list_head *done_q)
1393 struct scsi_cmnd *scmd, *stu_scmd, *next;
1394 struct scsi_device *sdev;
1396 shost_for_each_device(sdev, shost) {
1397 if (scsi_host_eh_past_deadline(shost)) {
1398 SCSI_LOG_ERROR_RECOVERY(3,
1399 sdev_printk(KERN_INFO, sdev,
1400 "%s: skip START_UNIT, past eh deadline\n",
1401 current->comm));
1402 break;
1404 stu_scmd = NULL;
1405 list_for_each_entry(scmd, work_q, eh_entry)
1406 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1407 scsi_check_sense(scmd) == FAILED ) {
1408 stu_scmd = scmd;
1409 break;
1412 if (!stu_scmd)
1413 continue;
1415 SCSI_LOG_ERROR_RECOVERY(3,
1416 sdev_printk(KERN_INFO, sdev,
1417 "%s: Sending START_UNIT\n",
1418 current->comm));
1420 if (!scsi_eh_try_stu(stu_scmd)) {
1421 if (!scsi_device_online(sdev) ||
1422 !scsi_eh_tur(stu_scmd)) {
1423 list_for_each_entry_safe(scmd, next,
1424 work_q, eh_entry) {
1425 if (scmd->device == sdev &&
1426 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1427 scsi_eh_finish_cmd(scmd, done_q);
1430 } else {
1431 SCSI_LOG_ERROR_RECOVERY(3,
1432 sdev_printk(KERN_INFO, sdev,
1433 "%s: START_UNIT failed\n",
1434 current->comm));
1438 return list_empty(work_q);
1443 * scsi_eh_bus_device_reset - send bdr if needed
1444 * @shost: scsi host being recovered.
1445 * @work_q: &list_head for pending commands.
1446 * @done_q: &list_head for processed commands.
1448 * Notes:
1449 * Try a bus device reset. Still, look to see whether we have multiple
1450 * devices that are jammed or not - if we have multiple devices, it
1451 * makes no sense to try bus_device_reset - we really would need to try
1452 * a bus_reset instead.
1454 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1455 struct list_head *work_q,
1456 struct list_head *done_q)
1458 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1459 struct scsi_device *sdev;
1460 int rtn;
1462 shost_for_each_device(sdev, shost) {
1463 if (scsi_host_eh_past_deadline(shost)) {
1464 SCSI_LOG_ERROR_RECOVERY(3,
1465 sdev_printk(KERN_INFO, sdev,
1466 "%s: skip BDR, past eh deadline\n",
1467 current->comm));
1468 break;
1470 bdr_scmd = NULL;
1471 list_for_each_entry(scmd, work_q, eh_entry)
1472 if (scmd->device == sdev) {
1473 bdr_scmd = scmd;
1474 break;
1477 if (!bdr_scmd)
1478 continue;
1480 SCSI_LOG_ERROR_RECOVERY(3,
1481 sdev_printk(KERN_INFO, sdev,
1482 "%s: Sending BDR\n", current->comm));
1483 rtn = scsi_try_bus_device_reset(bdr_scmd);
1484 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1485 if (!scsi_device_online(sdev) ||
1486 rtn == FAST_IO_FAIL ||
1487 !scsi_eh_tur(bdr_scmd)) {
1488 list_for_each_entry_safe(scmd, next,
1489 work_q, eh_entry) {
1490 if (scmd->device == sdev &&
1491 scsi_eh_action(scmd, rtn) != FAILED)
1492 scsi_eh_finish_cmd(scmd,
1493 done_q);
1496 } else {
1497 SCSI_LOG_ERROR_RECOVERY(3,
1498 sdev_printk(KERN_INFO, sdev,
1499 "%s: BDR failed\n", current->comm));
1503 return list_empty(work_q);
1507 * scsi_eh_target_reset - send target reset if needed
1508 * @shost: scsi host being recovered.
1509 * @work_q: &list_head for pending commands.
1510 * @done_q: &list_head for processed commands.
1512 * Notes:
1513 * Try a target reset.
1515 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1516 struct list_head *work_q,
1517 struct list_head *done_q)
1519 LIST_HEAD(tmp_list);
1520 LIST_HEAD(check_list);
1522 list_splice_init(work_q, &tmp_list);
1524 while (!list_empty(&tmp_list)) {
1525 struct scsi_cmnd *next, *scmd;
1526 int rtn;
1527 unsigned int id;
1529 if (scsi_host_eh_past_deadline(shost)) {
1530 /* push back on work queue for further processing */
1531 list_splice_init(&check_list, work_q);
1532 list_splice_init(&tmp_list, work_q);
1533 SCSI_LOG_ERROR_RECOVERY(3,
1534 shost_printk(KERN_INFO, shost,
1535 "%s: Skip target reset, past eh deadline\n",
1536 current->comm));
1537 return list_empty(work_q);
1540 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1541 id = scmd_id(scmd);
1543 SCSI_LOG_ERROR_RECOVERY(3,
1544 shost_printk(KERN_INFO, shost,
1545 "%s: Sending target reset to target %d\n",
1546 current->comm, id));
1547 rtn = scsi_try_target_reset(scmd);
1548 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1549 SCSI_LOG_ERROR_RECOVERY(3,
1550 shost_printk(KERN_INFO, shost,
1551 "%s: Target reset failed"
1552 " target: %d\n",
1553 current->comm, id));
1554 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1555 if (scmd_id(scmd) != id)
1556 continue;
1558 if (rtn == SUCCESS)
1559 list_move_tail(&scmd->eh_entry, &check_list);
1560 else if (rtn == FAST_IO_FAIL)
1561 scsi_eh_finish_cmd(scmd, done_q);
1562 else
1563 /* push back on work queue for further processing */
1564 list_move(&scmd->eh_entry, work_q);
1568 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1572 * scsi_eh_bus_reset - send a bus reset
1573 * @shost: &scsi host being recovered.
1574 * @work_q: &list_head for pending commands.
1575 * @done_q: &list_head for processed commands.
1577 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1578 struct list_head *work_q,
1579 struct list_head *done_q)
1581 struct scsi_cmnd *scmd, *chan_scmd, *next;
1582 LIST_HEAD(check_list);
1583 unsigned int channel;
1584 int rtn;
1587 * we really want to loop over the various channels, and do this on
1588 * a channel by channel basis. we should also check to see if any
1589 * of the failed commands are on soft_reset devices, and if so, skip
1590 * the reset.
1593 for (channel = 0; channel <= shost->max_channel; channel++) {
1594 if (scsi_host_eh_past_deadline(shost)) {
1595 list_splice_init(&check_list, work_q);
1596 SCSI_LOG_ERROR_RECOVERY(3,
1597 shost_printk(KERN_INFO, shost,
1598 "%s: skip BRST, past eh deadline\n",
1599 current->comm));
1600 return list_empty(work_q);
1603 chan_scmd = NULL;
1604 list_for_each_entry(scmd, work_q, eh_entry) {
1605 if (channel == scmd_channel(scmd)) {
1606 chan_scmd = scmd;
1607 break;
1609 * FIXME add back in some support for
1610 * soft_reset devices.
1615 if (!chan_scmd)
1616 continue;
1617 SCSI_LOG_ERROR_RECOVERY(3,
1618 shost_printk(KERN_INFO, shost,
1619 "%s: Sending BRST chan: %d\n",
1620 current->comm, channel));
1621 rtn = scsi_try_bus_reset(chan_scmd);
1622 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1623 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1624 if (channel == scmd_channel(scmd)) {
1625 if (rtn == FAST_IO_FAIL)
1626 scsi_eh_finish_cmd(scmd,
1627 done_q);
1628 else
1629 list_move_tail(&scmd->eh_entry,
1630 &check_list);
1633 } else {
1634 SCSI_LOG_ERROR_RECOVERY(3,
1635 shost_printk(KERN_INFO, shost,
1636 "%s: BRST failed chan: %d\n",
1637 current->comm, channel));
1640 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1644 * scsi_eh_host_reset - send a host reset
1645 * @shost: host to be reset.
1646 * @work_q: &list_head for pending commands.
1647 * @done_q: &list_head for processed commands.
1649 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1650 struct list_head *work_q,
1651 struct list_head *done_q)
1653 struct scsi_cmnd *scmd, *next;
1654 LIST_HEAD(check_list);
1655 int rtn;
1657 if (!list_empty(work_q)) {
1658 scmd = list_entry(work_q->next,
1659 struct scsi_cmnd, eh_entry);
1661 SCSI_LOG_ERROR_RECOVERY(3,
1662 shost_printk(KERN_INFO, shost,
1663 "%s: Sending HRST\n",
1664 current->comm));
1666 rtn = scsi_try_host_reset(scmd);
1667 if (rtn == SUCCESS) {
1668 list_splice_init(work_q, &check_list);
1669 } else if (rtn == FAST_IO_FAIL) {
1670 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1671 scsi_eh_finish_cmd(scmd, done_q);
1673 } else {
1674 SCSI_LOG_ERROR_RECOVERY(3,
1675 shost_printk(KERN_INFO, shost,
1676 "%s: HRST failed\n",
1677 current->comm));
1680 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1684 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1685 * @work_q: &list_head for pending commands.
1686 * @done_q: &list_head for processed commands.
1688 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1689 struct list_head *done_q)
1691 struct scsi_cmnd *scmd, *next;
1693 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1694 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1695 "not ready after error recovery\n");
1696 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1697 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1699 * FIXME: Handle lost cmds.
1702 scsi_eh_finish_cmd(scmd, done_q);
1704 return;
1708 * scsi_noretry_cmd - determine if command should be failed fast
1709 * @scmd: SCSI cmd to examine.
1711 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1713 switch (host_byte(scmd->result)) {
1714 case DID_OK:
1715 break;
1716 case DID_TIME_OUT:
1717 goto check_type;
1718 case DID_BUS_BUSY:
1719 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1720 case DID_PARITY:
1721 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1722 case DID_ERROR:
1723 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1724 status_byte(scmd->result) == RESERVATION_CONFLICT)
1725 return 0;
1726 /* fall through */
1727 case DID_SOFT_ERROR:
1728 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1731 if (status_byte(scmd->result) != CHECK_CONDITION)
1732 return 0;
1734 check_type:
1736 * assume caller has checked sense and determined
1737 * the check condition was retryable.
1739 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1740 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1741 return 1;
1742 else
1743 return 0;
1747 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1748 * @scmd: SCSI cmd to examine.
1750 * Notes:
1751 * This is *only* called when we are examining the status after sending
1752 * out the actual data command. any commands that are queued for error
1753 * recovery (e.g. test_unit_ready) do *not* come through here.
1755 * When this routine returns failed, it means the error handler thread
1756 * is woken. In cases where the error code indicates an error that
1757 * doesn't require the error handler read (i.e. we don't need to
1758 * abort/reset), this function should return SUCCESS.
1760 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1762 int rtn;
1765 * if the device is offline, then we clearly just pass the result back
1766 * up to the top level.
1768 if (!scsi_device_online(scmd->device)) {
1769 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1770 "%s: device offline - report as SUCCESS\n", __func__));
1771 return SUCCESS;
1775 * first check the host byte, to see if there is anything in there
1776 * that would indicate what we need to do.
1778 switch (host_byte(scmd->result)) {
1779 case DID_PASSTHROUGH:
1781 * no matter what, pass this through to the upper layer.
1782 * nuke this special code so that it looks like we are saying
1783 * did_ok.
1785 scmd->result &= 0xff00ffff;
1786 return SUCCESS;
1787 case DID_OK:
1789 * looks good. drop through, and check the next byte.
1791 break;
1792 case DID_ABORT:
1793 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1794 set_host_byte(scmd, DID_TIME_OUT);
1795 return SUCCESS;
1797 case DID_NO_CONNECT:
1798 case DID_BAD_TARGET:
1800 * note - this means that we just report the status back
1801 * to the top level driver, not that we actually think
1802 * that it indicates SUCCESS.
1804 return SUCCESS;
1806 * when the low level driver returns did_soft_error,
1807 * it is responsible for keeping an internal retry counter
1808 * in order to avoid endless loops (db)
1810 * actually this is a bug in this function here. we should
1811 * be mindful of the maximum number of retries specified
1812 * and not get stuck in a loop.
1814 case DID_SOFT_ERROR:
1815 goto maybe_retry;
1816 case DID_IMM_RETRY:
1817 return NEEDS_RETRY;
1819 case DID_REQUEUE:
1820 return ADD_TO_MLQUEUE;
1821 case DID_TRANSPORT_DISRUPTED:
1823 * LLD/transport was disrupted during processing of the IO.
1824 * The transport class is now blocked/blocking,
1825 * and the transport will decide what to do with the IO
1826 * based on its timers and recovery capablilities if
1827 * there are enough retries.
1829 goto maybe_retry;
1830 case DID_TRANSPORT_FAILFAST:
1832 * The transport decided to failfast the IO (most likely
1833 * the fast io fail tmo fired), so send IO directly upwards.
1835 return SUCCESS;
1836 case DID_ERROR:
1837 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1838 status_byte(scmd->result) == RESERVATION_CONFLICT)
1840 * execute reservation conflict processing code
1841 * lower down
1843 break;
1844 /* fallthrough */
1845 case DID_BUS_BUSY:
1846 case DID_PARITY:
1847 goto maybe_retry;
1848 case DID_TIME_OUT:
1850 * when we scan the bus, we get timeout messages for
1851 * these commands if there is no device available.
1852 * other hosts report did_no_connect for the same thing.
1854 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1855 scmd->cmnd[0] == INQUIRY)) {
1856 return SUCCESS;
1857 } else {
1858 return FAILED;
1860 case DID_RESET:
1861 return SUCCESS;
1862 default:
1863 return FAILED;
1867 * next, check the message byte.
1869 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1870 return FAILED;
1873 * check the status byte to see if this indicates anything special.
1875 switch (status_byte(scmd->result)) {
1876 case QUEUE_FULL:
1877 scsi_handle_queue_full(scmd->device);
1879 * the case of trying to send too many commands to a
1880 * tagged queueing device.
1882 case BUSY:
1884 * device can't talk to us at the moment. Should only
1885 * occur (SAM-3) when the task queue is empty, so will cause
1886 * the empty queue handling to trigger a stall in the
1887 * device.
1889 return ADD_TO_MLQUEUE;
1890 case GOOD:
1891 if (scmd->cmnd[0] == REPORT_LUNS)
1892 scmd->device->sdev_target->expecting_lun_change = 0;
1893 scsi_handle_queue_ramp_up(scmd->device);
1894 case COMMAND_TERMINATED:
1895 return SUCCESS;
1896 case TASK_ABORTED:
1897 goto maybe_retry;
1898 case CHECK_CONDITION:
1899 rtn = scsi_check_sense(scmd);
1900 if (rtn == NEEDS_RETRY)
1901 goto maybe_retry;
1902 /* if rtn == FAILED, we have no sense information;
1903 * returning FAILED will wake the error handler thread
1904 * to collect the sense and redo the decide
1905 * disposition */
1906 return rtn;
1907 case CONDITION_GOOD:
1908 case INTERMEDIATE_GOOD:
1909 case INTERMEDIATE_C_GOOD:
1910 case ACA_ACTIVE:
1912 * who knows? FIXME(eric)
1914 return SUCCESS;
1916 case RESERVATION_CONFLICT:
1917 sdev_printk(KERN_INFO, scmd->device,
1918 "reservation conflict\n");
1919 set_host_byte(scmd, DID_NEXUS_FAILURE);
1920 return SUCCESS; /* causes immediate i/o error */
1921 default:
1922 return FAILED;
1924 return FAILED;
1926 maybe_retry:
1928 /* we requeue for retry because the error was retryable, and
1929 * the request was not marked fast fail. Note that above,
1930 * even if the request is marked fast fail, we still requeue
1931 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1932 if ((++scmd->retries) <= scmd->allowed
1933 && !scsi_noretry_cmd(scmd)) {
1934 return NEEDS_RETRY;
1935 } else {
1937 * no more retries - report this one back to upper level.
1939 return SUCCESS;
1943 static void eh_lock_door_done(struct request *req, int uptodate)
1945 __blk_put_request(req->q, req);
1949 * scsi_eh_lock_door - Prevent medium removal for the specified device
1950 * @sdev: SCSI device to prevent medium removal
1952 * Locking:
1953 * We must be called from process context.
1955 * Notes:
1956 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1957 * head of the devices request queue, and continue.
1959 static void scsi_eh_lock_door(struct scsi_device *sdev)
1961 struct request *req;
1964 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1965 * request becomes available
1967 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1968 if (IS_ERR(req))
1969 return;
1971 blk_rq_set_block_pc(req);
1973 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1974 req->cmd[1] = 0;
1975 req->cmd[2] = 0;
1976 req->cmd[3] = 0;
1977 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1978 req->cmd[5] = 0;
1980 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1982 req->cmd_flags |= REQ_QUIET;
1983 req->timeout = 10 * HZ;
1984 req->retries = 5;
1986 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1990 * scsi_restart_operations - restart io operations to the specified host.
1991 * @shost: Host we are restarting.
1993 * Notes:
1994 * When we entered the error handler, we blocked all further i/o to
1995 * this device. we need to 'reverse' this process.
1997 static void scsi_restart_operations(struct Scsi_Host *shost)
1999 struct scsi_device *sdev;
2000 unsigned long flags;
2003 * If the door was locked, we need to insert a door lock request
2004 * onto the head of the SCSI request queue for the device. There
2005 * is no point trying to lock the door of an off-line device.
2007 shost_for_each_device(sdev, shost) {
2008 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2009 scsi_eh_lock_door(sdev);
2010 sdev->was_reset = 0;
2015 * next free up anything directly waiting upon the host. this
2016 * will be requests for character device operations, and also for
2017 * ioctls to queued block devices.
2019 SCSI_LOG_ERROR_RECOVERY(3,
2020 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2022 spin_lock_irqsave(shost->host_lock, flags);
2023 if (scsi_host_set_state(shost, SHOST_RUNNING))
2024 if (scsi_host_set_state(shost, SHOST_CANCEL))
2025 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2026 spin_unlock_irqrestore(shost->host_lock, flags);
2028 wake_up(&shost->host_wait);
2031 * finally we need to re-initiate requests that may be pending. we will
2032 * have had everything blocked while error handling is taking place, and
2033 * now that error recovery is done, we will need to ensure that these
2034 * requests are started.
2036 scsi_run_host_queues(shost);
2039 * if eh is active and host_eh_scheduled is pending we need to re-run
2040 * recovery. we do this check after scsi_run_host_queues() to allow
2041 * everything pent up since the last eh run a chance to make forward
2042 * progress before we sync again. Either we'll immediately re-run
2043 * recovery or scsi_device_unbusy() will wake us again when these
2044 * pending commands complete.
2046 spin_lock_irqsave(shost->host_lock, flags);
2047 if (shost->host_eh_scheduled)
2048 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2049 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2050 spin_unlock_irqrestore(shost->host_lock, flags);
2054 * scsi_eh_ready_devs - check device ready state and recover if not.
2055 * @shost: host to be recovered.
2056 * @work_q: &list_head for pending commands.
2057 * @done_q: &list_head for processed commands.
2059 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2060 struct list_head *work_q,
2061 struct list_head *done_q)
2063 if (!scsi_eh_stu(shost, work_q, done_q))
2064 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2065 if (!scsi_eh_target_reset(shost, work_q, done_q))
2066 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2067 if (!scsi_eh_host_reset(shost, work_q, done_q))
2068 scsi_eh_offline_sdevs(work_q,
2069 done_q);
2071 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2074 * scsi_eh_flush_done_q - finish processed commands or retry them.
2075 * @done_q: list_head of processed commands.
2077 void scsi_eh_flush_done_q(struct list_head *done_q)
2079 struct scsi_cmnd *scmd, *next;
2081 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2082 list_del_init(&scmd->eh_entry);
2083 if (scsi_device_online(scmd->device) &&
2084 !scsi_noretry_cmd(scmd) &&
2085 (++scmd->retries <= scmd->allowed)) {
2086 SCSI_LOG_ERROR_RECOVERY(3,
2087 scmd_printk(KERN_INFO, scmd,
2088 "%s: flush retry cmd\n",
2089 current->comm));
2090 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2091 } else {
2093 * If just we got sense for the device (called
2094 * scsi_eh_get_sense), scmd->result is already
2095 * set, do not set DRIVER_TIMEOUT.
2097 if (!scmd->result)
2098 scmd->result |= (DRIVER_TIMEOUT << 24);
2099 SCSI_LOG_ERROR_RECOVERY(3,
2100 scmd_printk(KERN_INFO, scmd,
2101 "%s: flush finish cmd\n",
2102 current->comm));
2103 scsi_finish_command(scmd);
2107 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2110 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2111 * @shost: Host to unjam.
2113 * Notes:
2114 * When we come in here, we *know* that all commands on the bus have
2115 * either completed, failed or timed out. we also know that no further
2116 * commands are being sent to the host, so things are relatively quiet
2117 * and we have freedom to fiddle with things as we wish.
2119 * This is only the *default* implementation. it is possible for
2120 * individual drivers to supply their own version of this function, and
2121 * if the maintainer wishes to do this, it is strongly suggested that
2122 * this function be taken as a template and modified. this function
2123 * was designed to correctly handle problems for about 95% of the
2124 * different cases out there, and it should always provide at least a
2125 * reasonable amount of error recovery.
2127 * Any command marked 'failed' or 'timeout' must eventually have
2128 * scsi_finish_cmd() called for it. we do all of the retry stuff
2129 * here, so when we restart the host after we return it should have an
2130 * empty queue.
2132 static void scsi_unjam_host(struct Scsi_Host *shost)
2134 unsigned long flags;
2135 LIST_HEAD(eh_work_q);
2136 LIST_HEAD(eh_done_q);
2138 spin_lock_irqsave(shost->host_lock, flags);
2139 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2140 spin_unlock_irqrestore(shost->host_lock, flags);
2142 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2144 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2145 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2146 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2148 spin_lock_irqsave(shost->host_lock, flags);
2149 if (shost->eh_deadline != -1)
2150 shost->last_reset = 0;
2151 spin_unlock_irqrestore(shost->host_lock, flags);
2152 scsi_eh_flush_done_q(&eh_done_q);
2156 * scsi_error_handler - SCSI error handler thread
2157 * @data: Host for which we are running.
2159 * Notes:
2160 * This is the main error handling loop. This is run as a kernel thread
2161 * for every SCSI host and handles all error handling activity.
2163 int scsi_error_handler(void *data)
2165 struct Scsi_Host *shost = data;
2168 * We use TASK_INTERRUPTIBLE so that the thread is not
2169 * counted against the load average as a running process.
2170 * We never actually get interrupted because kthread_run
2171 * disables signal delivery for the created thread.
2173 while (!kthread_should_stop()) {
2174 set_current_state(TASK_INTERRUPTIBLE);
2175 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2176 shost->host_failed != atomic_read(&shost->host_busy)) {
2177 SCSI_LOG_ERROR_RECOVERY(1,
2178 shost_printk(KERN_INFO, shost,
2179 "scsi_eh_%d: sleeping\n",
2180 shost->host_no));
2181 schedule();
2182 continue;
2185 __set_current_state(TASK_RUNNING);
2186 SCSI_LOG_ERROR_RECOVERY(1,
2187 shost_printk(KERN_INFO, shost,
2188 "scsi_eh_%d: waking up %d/%d/%d\n",
2189 shost->host_no, shost->host_eh_scheduled,
2190 shost->host_failed,
2191 atomic_read(&shost->host_busy)));
2194 * We have a host that is failing for some reason. Figure out
2195 * what we need to do to get it up and online again (if we can).
2196 * If we fail, we end up taking the thing offline.
2198 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2199 SCSI_LOG_ERROR_RECOVERY(1,
2200 shost_printk(KERN_ERR, shost,
2201 "scsi_eh_%d: unable to autoresume\n",
2202 shost->host_no));
2203 continue;
2206 if (shost->transportt->eh_strategy_handler)
2207 shost->transportt->eh_strategy_handler(shost);
2208 else
2209 scsi_unjam_host(shost);
2212 * Note - if the above fails completely, the action is to take
2213 * individual devices offline and flush the queue of any
2214 * outstanding requests that may have been pending. When we
2215 * restart, we restart any I/O to any other devices on the bus
2216 * which are still online.
2218 scsi_restart_operations(shost);
2219 if (!shost->eh_noresume)
2220 scsi_autopm_put_host(shost);
2222 __set_current_state(TASK_RUNNING);
2224 SCSI_LOG_ERROR_RECOVERY(1,
2225 shost_printk(KERN_INFO, shost,
2226 "Error handler scsi_eh_%d exiting\n",
2227 shost->host_no));
2228 shost->ehandler = NULL;
2229 return 0;
2233 * Function: scsi_report_bus_reset()
2235 * Purpose: Utility function used by low-level drivers to report that
2236 * they have observed a bus reset on the bus being handled.
2238 * Arguments: shost - Host in question
2239 * channel - channel on which reset was observed.
2241 * Returns: Nothing
2243 * Lock status: Host lock must be held.
2245 * Notes: This only needs to be called if the reset is one which
2246 * originates from an unknown location. Resets originated
2247 * by the mid-level itself don't need to call this, but there
2248 * should be no harm.
2250 * The main purpose of this is to make sure that a CHECK_CONDITION
2251 * is properly treated.
2253 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2255 struct scsi_device *sdev;
2257 __shost_for_each_device(sdev, shost) {
2258 if (channel == sdev_channel(sdev))
2259 __scsi_report_device_reset(sdev, NULL);
2262 EXPORT_SYMBOL(scsi_report_bus_reset);
2265 * Function: scsi_report_device_reset()
2267 * Purpose: Utility function used by low-level drivers to report that
2268 * they have observed a device reset on the device being handled.
2270 * Arguments: shost - Host in question
2271 * channel - channel on which reset was observed
2272 * target - target on which reset was observed
2274 * Returns: Nothing
2276 * Lock status: Host lock must be held
2278 * Notes: This only needs to be called if the reset is one which
2279 * originates from an unknown location. Resets originated
2280 * by the mid-level itself don't need to call this, but there
2281 * should be no harm.
2283 * The main purpose of this is to make sure that a CHECK_CONDITION
2284 * is properly treated.
2286 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2288 struct scsi_device *sdev;
2290 __shost_for_each_device(sdev, shost) {
2291 if (channel == sdev_channel(sdev) &&
2292 target == sdev_id(sdev))
2293 __scsi_report_device_reset(sdev, NULL);
2296 EXPORT_SYMBOL(scsi_report_device_reset);
2298 static void
2299 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2304 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2305 * @dev: scsi_device to operate on
2306 * @arg: reset type (see sg.h)
2309 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2311 struct scsi_cmnd *scmd;
2312 struct Scsi_Host *shost = dev->host;
2313 struct request req;
2314 unsigned long flags;
2315 int error = 0, rtn, val;
2317 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2318 return -EACCES;
2320 error = get_user(val, arg);
2321 if (error)
2322 return error;
2324 if (scsi_autopm_get_host(shost) < 0)
2325 return -EIO;
2327 error = -EIO;
2328 scmd = scsi_get_command(dev, GFP_KERNEL);
2329 if (!scmd)
2330 goto out_put_autopm_host;
2332 blk_rq_init(NULL, &req);
2333 scmd->request = &req;
2335 scmd->cmnd = req.cmd;
2337 scmd->scsi_done = scsi_reset_provider_done_command;
2338 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2340 scmd->cmd_len = 0;
2342 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2344 spin_lock_irqsave(shost->host_lock, flags);
2345 shost->tmf_in_progress = 1;
2346 spin_unlock_irqrestore(shost->host_lock, flags);
2348 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2349 case SG_SCSI_RESET_NOTHING:
2350 rtn = SUCCESS;
2351 break;
2352 case SG_SCSI_RESET_DEVICE:
2353 rtn = scsi_try_bus_device_reset(scmd);
2354 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2355 break;
2356 /* FALLTHROUGH */
2357 case SG_SCSI_RESET_TARGET:
2358 rtn = scsi_try_target_reset(scmd);
2359 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2360 break;
2361 /* FALLTHROUGH */
2362 case SG_SCSI_RESET_BUS:
2363 rtn = scsi_try_bus_reset(scmd);
2364 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2365 break;
2366 /* FALLTHROUGH */
2367 case SG_SCSI_RESET_HOST:
2368 rtn = scsi_try_host_reset(scmd);
2369 if (rtn == SUCCESS)
2370 break;
2371 default:
2372 /* FALLTHROUGH */
2373 rtn = FAILED;
2374 break;
2377 error = (rtn == SUCCESS) ? 0 : -EIO;
2379 spin_lock_irqsave(shost->host_lock, flags);
2380 shost->tmf_in_progress = 0;
2381 spin_unlock_irqrestore(shost->host_lock, flags);
2384 * be sure to wake up anyone who was sleeping or had their queue
2385 * suspended while we performed the TMF.
2387 SCSI_LOG_ERROR_RECOVERY(3,
2388 shost_printk(KERN_INFO, shost,
2389 "waking up host to restart after TMF\n"));
2391 wake_up(&shost->host_wait);
2392 scsi_run_host_queues(shost);
2394 scsi_put_command(scmd);
2396 out_put_autopm_host:
2397 scsi_autopm_put_host(shost);
2398 return error;
2400 EXPORT_SYMBOL(scsi_ioctl_reset);
2403 * scsi_normalize_sense - normalize main elements from either fixed or
2404 * descriptor sense data format into a common format.
2406 * @sense_buffer: byte array containing sense data returned by device
2407 * @sb_len: number of valid bytes in sense_buffer
2408 * @sshdr: pointer to instance of structure that common
2409 * elements are written to.
2411 * Notes:
2412 * The "main elements" from sense data are: response_code, sense_key,
2413 * asc, ascq and additional_length (only for descriptor format).
2415 * Typically this function can be called after a device has
2416 * responded to a SCSI command with the CHECK_CONDITION status.
2418 * Return value:
2419 * true if valid sense data information found, else false;
2421 bool scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2422 struct scsi_sense_hdr *sshdr)
2424 if (!sense_buffer || !sb_len)
2425 return false;
2427 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2429 sshdr->response_code = (sense_buffer[0] & 0x7f);
2431 if (!scsi_sense_valid(sshdr))
2432 return false;
2434 if (sshdr->response_code >= 0x72) {
2436 * descriptor format
2438 if (sb_len > 1)
2439 sshdr->sense_key = (sense_buffer[1] & 0xf);
2440 if (sb_len > 2)
2441 sshdr->asc = sense_buffer[2];
2442 if (sb_len > 3)
2443 sshdr->ascq = sense_buffer[3];
2444 if (sb_len > 7)
2445 sshdr->additional_length = sense_buffer[7];
2446 } else {
2448 * fixed format
2450 if (sb_len > 2)
2451 sshdr->sense_key = (sense_buffer[2] & 0xf);
2452 if (sb_len > 7) {
2453 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2454 sb_len : (sense_buffer[7] + 8);
2455 if (sb_len > 12)
2456 sshdr->asc = sense_buffer[12];
2457 if (sb_len > 13)
2458 sshdr->ascq = sense_buffer[13];
2462 return true;
2464 EXPORT_SYMBOL(scsi_normalize_sense);
2466 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2467 struct scsi_sense_hdr *sshdr)
2469 return scsi_normalize_sense(cmd->sense_buffer,
2470 SCSI_SENSE_BUFFERSIZE, sshdr);
2472 EXPORT_SYMBOL(scsi_command_normalize_sense);
2475 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2476 * @sense_buffer: byte array of descriptor format sense data
2477 * @sb_len: number of valid bytes in sense_buffer
2478 * @desc_type: value of descriptor type to find
2479 * (e.g. 0 -> information)
2481 * Notes:
2482 * only valid when sense data is in descriptor format
2484 * Return value:
2485 * pointer to start of (first) descriptor if found else NULL
2487 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2488 int desc_type)
2490 int add_sen_len, add_len, desc_len, k;
2491 const u8 * descp;
2493 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2494 return NULL;
2495 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2496 return NULL;
2497 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2498 add_sen_len : (sb_len - 8);
2499 descp = &sense_buffer[8];
2500 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2501 descp += desc_len;
2502 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2503 desc_len = add_len + 2;
2504 if (descp[0] == desc_type)
2505 return descp;
2506 if (add_len < 0) // short descriptor ??
2507 break;
2509 return NULL;
2511 EXPORT_SYMBOL(scsi_sense_desc_find);
2514 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2515 * @sense_buffer: byte array of sense data
2516 * @sb_len: number of valid bytes in sense_buffer
2517 * @info_out: pointer to 64 integer where 8 or 4 byte information
2518 * field will be placed if found.
2520 * Return value:
2521 * 1 if information field found, 0 if not found.
2523 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2524 u64 * info_out)
2526 int j;
2527 const u8 * ucp;
2528 u64 ull;
2530 if (sb_len < 7)
2531 return 0;
2532 switch (sense_buffer[0] & 0x7f) {
2533 case 0x70:
2534 case 0x71:
2535 if (sense_buffer[0] & 0x80) {
2536 *info_out = (sense_buffer[3] << 24) +
2537 (sense_buffer[4] << 16) +
2538 (sense_buffer[5] << 8) + sense_buffer[6];
2539 return 1;
2540 } else
2541 return 0;
2542 case 0x72:
2543 case 0x73:
2544 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2545 0 /* info desc */);
2546 if (ucp && (0xa == ucp[1])) {
2547 ull = 0;
2548 for (j = 0; j < 8; ++j) {
2549 if (j > 0)
2550 ull <<= 8;
2551 ull |= ucp[4 + j];
2553 *info_out = ull;
2554 return 1;
2555 } else
2556 return 0;
2557 default:
2558 return 0;
2561 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2564 * scsi_build_sense_buffer - build sense data in a buffer
2565 * @desc: Sense format (non zero == descriptor format,
2566 * 0 == fixed format)
2567 * @buf: Where to build sense data
2568 * @key: Sense key
2569 * @asc: Additional sense code
2570 * @ascq: Additional sense code qualifier
2573 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2575 if (desc) {
2576 buf[0] = 0x72; /* descriptor, current */
2577 buf[1] = key;
2578 buf[2] = asc;
2579 buf[3] = ascq;
2580 buf[7] = 0;
2581 } else {
2582 buf[0] = 0x70; /* fixed, current */
2583 buf[2] = key;
2584 buf[7] = 0xa;
2585 buf[12] = asc;
2586 buf[13] = ascq;
2589 EXPORT_SYMBOL(scsi_build_sense_buffer);
2592 * scsi_set_sense_information - set the information field in a
2593 * formatted sense data buffer
2594 * @buf: Where to build sense data
2595 * @info: 64-bit information value to be set
2598 void scsi_set_sense_information(u8 *buf, u64 info)
2600 if ((buf[0] & 0x7f) == 0x72) {
2601 u8 *ucp, len;
2603 len = buf[7];
2604 ucp = (char *)scsi_sense_desc_find(buf, len + 8, 0);
2605 if (!ucp) {
2606 buf[7] = len + 0xa;
2607 ucp = buf + 8 + len;
2609 ucp[0] = 0;
2610 ucp[1] = 0xa;
2611 ucp[2] = 0x80; /* Valid bit */
2612 ucp[3] = 0;
2613 put_unaligned_be64(info, &ucp[4]);
2614 } else if ((buf[0] & 0x7f) == 0x70) {
2615 buf[0] |= 0x80;
2616 put_unaligned_be64(info, &buf[3]);
2619 EXPORT_SYMBOL(scsi_set_sense_information);