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
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>
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_driver.h>
34 #include <scsi/scsi_eh.h>
35 #include <scsi/scsi_transport.h>
36 #include <scsi/scsi_host.h>
37 #include <scsi/scsi_ioctl.h>
39 #include "scsi_priv.h"
40 #include "scsi_logging.h"
41 #include "scsi_transport_api.h"
43 #include <trace/events/scsi.h>
45 static void scsi_eh_done(struct scsi_cmnd *scmd);
47 #define SENSE_TIMEOUT (10*HZ)
50 * These should *probably* be handled by the host itself.
51 * Since it is allowed to sleep, it probably should.
53 #define BUS_RESET_SETTLE_TIME (10)
54 #define HOST_RESET_SETTLE_TIME (10)
56 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
58 /* called with shost->host_lock held */
59 void scsi_eh_wakeup(struct Scsi_Host *shost)
61 if (shost->host_busy == shost->host_failed) {
62 trace_scsi_eh_wakeup(shost);
63 wake_up_process(shost->ehandler);
64 SCSI_LOG_ERROR_RECOVERY(5,
65 printk("Waking error handler thread\n"));
70 * scsi_schedule_eh - schedule EH for SCSI host
71 * @shost: SCSI host to invoke error handling on.
73 * Schedule SCSI EH without scmd.
75 void scsi_schedule_eh(struct Scsi_Host *shost)
79 spin_lock_irqsave(shost->host_lock, flags);
81 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
82 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
83 shost->host_eh_scheduled++;
84 scsi_eh_wakeup(shost);
87 spin_unlock_irqrestore(shost->host_lock, flags);
89 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
92 * scsi_eh_scmd_add - add scsi cmd to error handling.
93 * @scmd: scmd to run eh on.
94 * @eh_flag: optional SCSI_EH flag.
99 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
101 struct Scsi_Host *shost = scmd->device->host;
105 if (!shost->ehandler)
108 spin_lock_irqsave(shost->host_lock, flags);
109 if (scsi_host_set_state(shost, SHOST_RECOVERY))
110 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
114 scmd->eh_eflags |= eh_flag;
115 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
116 shost->host_failed++;
117 scsi_eh_wakeup(shost);
119 spin_unlock_irqrestore(shost->host_lock, flags);
124 * scsi_times_out - Timeout function for normal scsi commands.
125 * @req: request that is timing out.
128 * We do not need to lock this. There is the potential for a race
129 * only in that the normal completion handling might run, but if the
130 * normal completion function determines that the timer has already
131 * fired, then it mustn't do anything.
133 enum blk_eh_timer_return scsi_times_out(struct request *req)
135 struct scsi_cmnd *scmd = req->special;
136 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
137 struct Scsi_Host *host = scmd->device->host;
139 trace_scsi_dispatch_cmd_timeout(scmd);
140 scsi_log_completion(scmd, TIMEOUT_ERROR);
142 if (host->transportt->eh_timed_out)
143 rtn = host->transportt->eh_timed_out(scmd);
144 else if (host->hostt->eh_timed_out)
145 rtn = host->hostt->eh_timed_out(scmd);
147 scmd->result |= DID_TIME_OUT << 16;
149 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
150 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
151 rtn = BLK_EH_HANDLED;
157 * scsi_block_when_processing_errors - Prevent cmds from being queued.
158 * @sdev: Device on which we are performing recovery.
161 * We block until the host is out of error recovery, and then check to
162 * see whether the host or the device is offline.
165 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
167 int scsi_block_when_processing_errors(struct scsi_device *sdev)
171 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
173 online = scsi_device_online(sdev);
175 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
180 EXPORT_SYMBOL(scsi_block_when_processing_errors);
182 #ifdef CONFIG_SCSI_LOGGING
184 * scsi_eh_prt_fail_stats - Log info on failures.
185 * @shost: scsi host being recovered.
186 * @work_q: Queue of scsi cmds to process.
188 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
189 struct list_head *work_q)
191 struct scsi_cmnd *scmd;
192 struct scsi_device *sdev;
193 int total_failures = 0;
196 int devices_failed = 0;
198 shost_for_each_device(sdev, shost) {
199 list_for_each_entry(scmd, work_q, eh_entry) {
200 if (scmd->device == sdev) {
202 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
209 if (cmd_cancel || cmd_failed) {
210 SCSI_LOG_ERROR_RECOVERY(3,
211 sdev_printk(KERN_INFO, sdev,
212 "%s: cmds failed: %d, cancel: %d\n",
213 __func__, cmd_failed,
221 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
222 " devices require eh work\n",
223 total_failures, devices_failed));
228 * scsi_check_sense - Examine scsi cmd sense
229 * @scmd: Cmd to have sense checked.
232 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
235 * When a deferred error is detected the current command has
236 * not been executed and needs retrying.
238 static int scsi_check_sense(struct scsi_cmnd *scmd)
240 struct scsi_device *sdev = scmd->device;
241 struct scsi_sense_hdr sshdr;
243 if (! scsi_command_normalize_sense(scmd, &sshdr))
244 return FAILED; /* no valid sense data */
246 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
248 * nasty: for mid-layer issued TURs, we need to return the
249 * actual sense data without any recovery attempt. For eh
250 * issued ones, we need to try to recover and interpret
254 if (scsi_sense_is_deferred(&sshdr))
257 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
258 sdev->scsi_dh_data->scsi_dh->check_sense) {
261 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
262 if (rc != SCSI_RETURN_NOT_HANDLED)
264 /* handler does not care. Drop down to default handling */
268 * Previous logic looked for FILEMARK, EOM or ILI which are
269 * mainly associated with tapes and returned SUCCESS.
271 if (sshdr.response_code == 0x70) {
273 if (scmd->sense_buffer[2] & 0xe0)
277 * descriptor format: look for "stream commands sense data
278 * descriptor" (see SSC-3). Assume single sense data
279 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
281 if ((sshdr.additional_length > 3) &&
282 (scmd->sense_buffer[8] == 0x4) &&
283 (scmd->sense_buffer[11] & 0xe0))
287 switch (sshdr.sense_key) {
290 case RECOVERED_ERROR:
291 return /* soft_error */ SUCCESS;
293 case ABORTED_COMMAND:
294 if (sshdr.asc == 0x10) /* DIF */
301 * if we are expecting a cc/ua because of a bus reset that we
302 * performed, treat this just as a retry. otherwise this is
303 * information that we should pass up to the upper-level driver
304 * so that we can deal with it there.
306 if (scmd->device->expecting_cc_ua) {
308 * Because some device does not queue unit
309 * attentions correctly, we carefully check
310 * additional sense code and qualifier so as
311 * not to squash media change unit attention.
313 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
314 scmd->device->expecting_cc_ua = 0;
319 * if the device is in the process of becoming ready, we
322 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
325 * if the device is not started, we need to wake
326 * the error handler to start the motor
328 if (scmd->device->allow_restart &&
329 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
332 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
333 scmd_printk(KERN_WARNING, scmd,
334 "Warning! Received an indication that the "
335 "LUN assignments on this target have "
336 "changed. The Linux SCSI layer does not "
337 "automatically remap LUN assignments.\n");
338 else if (sshdr.asc == 0x3f)
339 scmd_printk(KERN_WARNING, scmd,
340 "Warning! Received an indication that the "
341 "operating parameters on this target have "
342 "changed. The Linux SCSI layer does not "
343 "automatically adjust these parameters.\n");
345 if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
346 scmd_printk(KERN_WARNING, scmd,
347 "Warning! Received an indication that the "
348 "LUN reached a thin provisioning soft "
352 * Pass the UA upwards for a determination in the completion
357 /* these are not supported */
359 case VOLUME_OVERFLOW:
366 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
367 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
368 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
374 if (scmd->device->retry_hwerror)
375 return ADD_TO_MLQUEUE;
379 case ILLEGAL_REQUEST:
380 if (sshdr.asc == 0x20 || /* Invalid command operation code */
381 sshdr.asc == 0x21 || /* Logical block address out of range */
382 sshdr.asc == 0x24 || /* Invalid field in cdb */
383 sshdr.asc == 0x26) { /* Parameter value invalid */
393 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
395 struct scsi_host_template *sht = sdev->host->hostt;
396 struct scsi_device *tmp_sdev;
398 if (!sht->change_queue_depth ||
399 sdev->queue_depth >= sdev->max_queue_depth)
402 if (time_before(jiffies,
403 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
406 if (time_before(jiffies,
407 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
411 * Walk all devices of a target and do
414 shost_for_each_device(tmp_sdev, sdev->host) {
415 if (tmp_sdev->channel != sdev->channel ||
416 tmp_sdev->id != sdev->id ||
417 tmp_sdev->queue_depth == sdev->max_queue_depth)
420 * call back into LLD to increase queue_depth by one
421 * with ramp up reason code.
423 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
424 SCSI_QDEPTH_RAMP_UP);
425 sdev->last_queue_ramp_up = jiffies;
429 static void scsi_handle_queue_full(struct scsi_device *sdev)
431 struct scsi_host_template *sht = sdev->host->hostt;
432 struct scsi_device *tmp_sdev;
434 if (!sht->change_queue_depth)
437 shost_for_each_device(tmp_sdev, sdev->host) {
438 if (tmp_sdev->channel != sdev->channel ||
439 tmp_sdev->id != sdev->id)
442 * We do not know the number of commands that were at
443 * the device when we got the queue full so we start
444 * from the highest possible value and work our way down.
446 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
452 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
453 * @scmd: SCSI cmd to examine.
456 * This is *only* called when we are examining the status of commands
457 * queued during error recovery. the main difference here is that we
458 * don't allow for the possibility of retries here, and we are a lot
459 * more restrictive about what we consider acceptable.
461 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
464 * first check the host byte, to see if there is anything in there
465 * that would indicate what we need to do.
467 if (host_byte(scmd->result) == DID_RESET) {
469 * rats. we are already in the error handler, so we now
470 * get to try and figure out what to do next. if the sense
471 * is valid, we have a pretty good idea of what to do.
472 * if not, we mark it as FAILED.
474 return scsi_check_sense(scmd);
476 if (host_byte(scmd->result) != DID_OK)
480 * next, check the message byte.
482 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
486 * now, check the status byte to see if this indicates
489 switch (status_byte(scmd->result)) {
491 scsi_handle_queue_ramp_up(scmd->device);
492 case COMMAND_TERMINATED:
494 case CHECK_CONDITION:
495 return scsi_check_sense(scmd);
497 case INTERMEDIATE_GOOD:
498 case INTERMEDIATE_C_GOOD:
500 * who knows? FIXME(eric)
503 case RESERVATION_CONFLICT:
504 if (scmd->cmnd[0] == TEST_UNIT_READY)
505 /* it is a success, we probed the device and
508 /* otherwise, we failed to send the command */
511 scsi_handle_queue_full(scmd->device);
522 * scsi_eh_done - Completion function for error handling.
523 * @scmd: Cmd that is done.
525 static void scsi_eh_done(struct scsi_cmnd *scmd)
527 struct completion *eh_action;
529 SCSI_LOG_ERROR_RECOVERY(3,
530 printk("%s scmd: %p result: %x\n",
531 __func__, scmd, scmd->result));
533 eh_action = scmd->device->host->eh_action;
539 * scsi_try_host_reset - ask host adapter to reset itself
540 * @scmd: SCSI cmd to send hsot reset.
542 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
546 struct Scsi_Host *host = scmd->device->host;
547 struct scsi_host_template *hostt = host->hostt;
549 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
552 if (!hostt->eh_host_reset_handler)
555 rtn = hostt->eh_host_reset_handler(scmd);
557 if (rtn == SUCCESS) {
558 if (!hostt->skip_settle_delay)
559 ssleep(HOST_RESET_SETTLE_TIME);
560 spin_lock_irqsave(host->host_lock, flags);
561 scsi_report_bus_reset(host, scmd_channel(scmd));
562 spin_unlock_irqrestore(host->host_lock, flags);
569 * scsi_try_bus_reset - ask host to perform a bus reset
570 * @scmd: SCSI cmd to send bus reset.
572 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
576 struct Scsi_Host *host = scmd->device->host;
577 struct scsi_host_template *hostt = host->hostt;
579 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
582 if (!hostt->eh_bus_reset_handler)
585 rtn = hostt->eh_bus_reset_handler(scmd);
587 if (rtn == SUCCESS) {
588 if (!hostt->skip_settle_delay)
589 ssleep(BUS_RESET_SETTLE_TIME);
590 spin_lock_irqsave(host->host_lock, flags);
591 scsi_report_bus_reset(host, scmd_channel(scmd));
592 spin_unlock_irqrestore(host->host_lock, flags);
598 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
601 sdev->expecting_cc_ua = 1;
605 * scsi_try_target_reset - Ask host to perform a target reset
606 * @scmd: SCSI cmd used to send a target reset
609 * There is no timeout for this operation. if this operation is
610 * unreliable for a given host, then the host itself needs to put a
611 * timer on it, and set the host back to a consistent state prior to
614 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
618 struct Scsi_Host *host = scmd->device->host;
619 struct scsi_host_template *hostt = host->hostt;
621 if (!hostt->eh_target_reset_handler)
624 rtn = hostt->eh_target_reset_handler(scmd);
625 if (rtn == SUCCESS) {
626 spin_lock_irqsave(host->host_lock, flags);
627 __starget_for_each_device(scsi_target(scmd->device), NULL,
628 __scsi_report_device_reset);
629 spin_unlock_irqrestore(host->host_lock, flags);
636 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
637 * @scmd: SCSI cmd used to send BDR
640 * There is no timeout for this operation. if this operation is
641 * unreliable for a given host, then the host itself needs to put a
642 * timer on it, and set the host back to a consistent state prior to
645 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
648 struct scsi_host_template *hostt = scmd->device->host->hostt;
650 if (!hostt->eh_device_reset_handler)
653 rtn = hostt->eh_device_reset_handler(scmd);
655 __scsi_report_device_reset(scmd->device, NULL);
659 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
661 if (!hostt->eh_abort_handler)
664 return hostt->eh_abort_handler(scmd);
667 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
669 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
670 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
671 if (scsi_try_target_reset(scmd) != SUCCESS)
672 if (scsi_try_bus_reset(scmd) != SUCCESS)
673 scsi_try_host_reset(scmd);
677 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
678 * @scmd: SCSI command structure to hijack
679 * @ses: structure to save restore information
680 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
681 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
682 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
684 * This function is used to save a scsi command information before re-execution
685 * as part of the error recovery process. If @sense_bytes is 0 the command
686 * sent must be one that does not transfer any data. If @sense_bytes != 0
687 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
688 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
690 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
691 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
693 struct scsi_device *sdev = scmd->device;
696 * We need saved copies of a number of fields - this is because
697 * error handling may need to overwrite these with different values
698 * to run different commands, and once error handling is complete,
699 * we will need to restore these values prior to running the actual
702 ses->cmd_len = scmd->cmd_len;
703 ses->cmnd = scmd->cmnd;
704 ses->data_direction = scmd->sc_data_direction;
705 ses->sdb = scmd->sdb;
706 ses->next_rq = scmd->request->next_rq;
707 ses->result = scmd->result;
708 ses->underflow = scmd->underflow;
709 ses->prot_op = scmd->prot_op;
711 scmd->prot_op = SCSI_PROT_NORMAL;
712 scmd->cmnd = ses->eh_cmnd;
713 memset(scmd->cmnd, 0, BLK_MAX_CDB);
714 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
715 scmd->request->next_rq = NULL;
718 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
720 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
722 scmd->sdb.table.sgl = &ses->sense_sgl;
723 scmd->sc_data_direction = DMA_FROM_DEVICE;
724 scmd->sdb.table.nents = 1;
725 scmd->cmnd[0] = REQUEST_SENSE;
726 scmd->cmnd[4] = scmd->sdb.length;
727 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
729 scmd->sc_data_direction = DMA_NONE;
731 BUG_ON(cmnd_size > BLK_MAX_CDB);
732 memcpy(scmd->cmnd, cmnd, cmnd_size);
733 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
739 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
740 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
741 (sdev->lun << 5 & 0xe0);
744 * Zero the sense buffer. The scsi spec mandates that any
745 * untransferred sense data should be interpreted as being zero.
747 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
749 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
752 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
753 * @scmd: SCSI command structure to restore
754 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
756 * Undo any damage done by above scsi_eh_prep_cmnd().
758 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
761 * Restore original data
763 scmd->cmd_len = ses->cmd_len;
764 scmd->cmnd = ses->cmnd;
765 scmd->sc_data_direction = ses->data_direction;
766 scmd->sdb = ses->sdb;
767 scmd->request->next_rq = ses->next_rq;
768 scmd->result = ses->result;
769 scmd->underflow = ses->underflow;
770 scmd->prot_op = ses->prot_op;
772 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
775 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
776 * @scmd: SCSI command structure to hijack
778 * @cmnd_size: size in bytes of @cmnd
779 * @timeout: timeout for this request
780 * @sense_bytes: size of sense data to copy or 0
782 * This function is used to send a scsi command down to a target device
783 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
786 * SUCCESS or FAILED or NEEDS_RETRY
788 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
789 int cmnd_size, int timeout, unsigned sense_bytes)
791 struct scsi_device *sdev = scmd->device;
792 struct Scsi_Host *shost = sdev->host;
793 DECLARE_COMPLETION_ONSTACK(done);
794 unsigned long timeleft;
795 struct scsi_eh_save ses;
798 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
799 shost->eh_action = &done;
802 scmd->scsi_done = scsi_eh_done;
803 shost->hostt->queuecommand(shost, scmd);
805 timeleft = wait_for_completion_timeout(&done, timeout);
807 shost->eh_action = NULL;
809 scsi_log_completion(scmd, SUCCESS);
811 SCSI_LOG_ERROR_RECOVERY(3,
812 printk("%s: scmd: %p, timeleft: %ld\n",
813 __func__, scmd, timeleft));
816 * If there is time left scsi_eh_done got called, and we will
817 * examine the actual status codes to see whether the command
818 * actually did complete normally, else tell the host to forget
819 * about this command.
822 rtn = scsi_eh_completed_normally(scmd);
823 SCSI_LOG_ERROR_RECOVERY(3,
824 printk("%s: scsi_eh_completed_normally %x\n",
841 scsi_abort_eh_cmnd(scmd);
845 scsi_eh_restore_cmnd(scmd, &ses);
847 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
848 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
850 rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
857 * scsi_request_sense - Request sense data from a particular target.
858 * @scmd: SCSI cmd for request sense.
861 * Some hosts automatically obtain this information, others require
862 * that we obtain it on our own. This function will *not* return until
863 * the command either times out, or it completes.
865 static int scsi_request_sense(struct scsi_cmnd *scmd)
867 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
871 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
872 * @scmd: Original SCSI cmd that eh has finished.
873 * @done_q: Queue for processed commands.
876 * We don't want to use the normal command completion while we are are
877 * still handling errors - it may cause other commands to be queued,
878 * and that would disturb what we are doing. Thus we really want to
879 * keep a list of pending commands for final completion, and once we
880 * are ready to leave error handling we handle completion for real.
882 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
884 scmd->device->host->host_failed--;
886 list_move_tail(&scmd->eh_entry, done_q);
888 EXPORT_SYMBOL(scsi_eh_finish_cmd);
891 * scsi_eh_get_sense - Get device sense data.
892 * @work_q: Queue of commands to process.
893 * @done_q: Queue of processed commands.
896 * See if we need to request sense information. if so, then get it
897 * now, so we have a better idea of what to do.
900 * This has the unfortunate side effect that if a shost adapter does
901 * not automatically request sense information, we end up shutting
902 * it down before we request it.
904 * All drivers should request sense information internally these days,
905 * so for now all I have to say is tough noogies if you end up in here.
907 * XXX: Long term this code should go away, but that needs an audit of
910 int scsi_eh_get_sense(struct list_head *work_q,
911 struct list_head *done_q)
913 struct scsi_cmnd *scmd, *next;
916 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
917 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
918 SCSI_SENSE_VALID(scmd))
921 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
922 "%s: requesting sense\n",
924 rtn = scsi_request_sense(scmd);
928 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
929 " result %x\n", scmd,
931 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
933 rtn = scsi_decide_disposition(scmd);
936 * if the result was normal, then just pass it along to the
940 /* we don't want this command reissued, just
941 * finished with the sense data, so set
942 * retries to the max allowed to ensure it
943 * won't get reissued */
944 scmd->retries = scmd->allowed;
945 else if (rtn != NEEDS_RETRY)
948 scsi_eh_finish_cmd(scmd, done_q);
951 return list_empty(work_q);
953 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
956 * scsi_eh_tur - Send TUR to device.
957 * @scmd: &scsi_cmnd to send TUR
960 * 0 - Device is ready. 1 - Device NOT ready.
962 static int scsi_eh_tur(struct scsi_cmnd *scmd)
964 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
965 int retry_cnt = 1, rtn;
968 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
970 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
971 __func__, scmd, rtn));
986 * scsi_eh_test_devices - check if devices are responding from error recovery.
987 * @cmd_list: scsi commands in error recovery.
988 * @work_q: queue for commands which still need more error recovery
989 * @done_q: queue for commands which are finished
990 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
993 * Tests if devices are in a working state. Commands to devices now in
994 * a working state are sent to the done_q while commands to devices which
995 * are still failing to respond are returned to the work_q for more
998 static int scsi_eh_test_devices(struct list_head *cmd_list,
999 struct list_head *work_q,
1000 struct list_head *done_q, int try_stu)
1002 struct scsi_cmnd *scmd, *next;
1003 struct scsi_device *sdev;
1006 while (!list_empty(cmd_list)) {
1007 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1008 sdev = scmd->device;
1010 finish_cmds = !scsi_device_online(scmd->device) ||
1011 (try_stu && !scsi_eh_try_stu(scmd) &&
1012 !scsi_eh_tur(scmd)) ||
1015 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1016 if (scmd->device == sdev) {
1018 scsi_eh_finish_cmd(scmd, done_q);
1020 list_move_tail(&scmd->eh_entry, work_q);
1023 return list_empty(work_q);
1028 * scsi_eh_abort_cmds - abort pending commands.
1029 * @work_q: &list_head for pending commands.
1030 * @done_q: &list_head for processed commands.
1033 * Try and see whether or not it makes sense to try and abort the
1034 * running command. This only works out to be the case if we have one
1035 * command that has timed out. If the command simply failed, it makes
1036 * no sense to try and abort the command, since as far as the shost
1037 * adapter is concerned, it isn't running.
1039 static int scsi_eh_abort_cmds(struct list_head *work_q,
1040 struct list_head *done_q)
1042 struct scsi_cmnd *scmd, *next;
1043 LIST_HEAD(check_list);
1046 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1047 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1049 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1050 "0x%p\n", current->comm,
1052 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
1053 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1054 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1055 if (rtn == FAST_IO_FAIL)
1056 scsi_eh_finish_cmd(scmd, done_q);
1058 list_move_tail(&scmd->eh_entry, &check_list);
1060 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1067 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1071 * scsi_eh_try_stu - Send START_UNIT to device.
1072 * @scmd: &scsi_cmnd to send START_UNIT
1075 * 0 - Device is ready. 1 - Device NOT ready.
1077 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1079 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1081 if (scmd->device->allow_restart) {
1082 int i, rtn = NEEDS_RETRY;
1084 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1085 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1095 * scsi_eh_stu - send START_UNIT if needed
1096 * @shost: &scsi host being recovered.
1097 * @work_q: &list_head for pending commands.
1098 * @done_q: &list_head for processed commands.
1101 * If commands are failing due to not ready, initializing command required,
1102 * try revalidating the device, which will end up sending a start unit.
1104 static int scsi_eh_stu(struct Scsi_Host *shost,
1105 struct list_head *work_q,
1106 struct list_head *done_q)
1108 struct scsi_cmnd *scmd, *stu_scmd, *next;
1109 struct scsi_device *sdev;
1111 shost_for_each_device(sdev, shost) {
1113 list_for_each_entry(scmd, work_q, eh_entry)
1114 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1115 scsi_check_sense(scmd) == FAILED ) {
1123 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1124 " 0x%p\n", current->comm, sdev));
1126 if (!scsi_eh_try_stu(stu_scmd)) {
1127 if (!scsi_device_online(sdev) ||
1128 !scsi_eh_tur(stu_scmd)) {
1129 list_for_each_entry_safe(scmd, next,
1131 if (scmd->device == sdev)
1132 scsi_eh_finish_cmd(scmd, done_q);
1136 SCSI_LOG_ERROR_RECOVERY(3,
1137 printk("%s: START_UNIT failed to sdev:"
1138 " 0x%p\n", current->comm, sdev));
1142 return list_empty(work_q);
1147 * scsi_eh_bus_device_reset - send bdr if needed
1148 * @shost: scsi host being recovered.
1149 * @work_q: &list_head for pending commands.
1150 * @done_q: &list_head for processed commands.
1153 * Try a bus device reset. Still, look to see whether we have multiple
1154 * devices that are jammed or not - if we have multiple devices, it
1155 * makes no sense to try bus_device_reset - we really would need to try
1156 * a bus_reset instead.
1158 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1159 struct list_head *work_q,
1160 struct list_head *done_q)
1162 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1163 struct scsi_device *sdev;
1166 shost_for_each_device(sdev, shost) {
1168 list_for_each_entry(scmd, work_q, eh_entry)
1169 if (scmd->device == sdev) {
1177 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1178 " 0x%p\n", current->comm,
1180 rtn = scsi_try_bus_device_reset(bdr_scmd);
1181 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1182 if (!scsi_device_online(sdev) ||
1183 rtn == FAST_IO_FAIL ||
1184 !scsi_eh_tur(bdr_scmd)) {
1185 list_for_each_entry_safe(scmd, next,
1187 if (scmd->device == sdev)
1188 scsi_eh_finish_cmd(scmd,
1193 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1201 return list_empty(work_q);
1205 * scsi_eh_target_reset - send target reset if needed
1206 * @shost: scsi host being recovered.
1207 * @work_q: &list_head for pending commands.
1208 * @done_q: &list_head for processed commands.
1211 * Try a target reset.
1213 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1214 struct list_head *work_q,
1215 struct list_head *done_q)
1217 LIST_HEAD(tmp_list);
1218 LIST_HEAD(check_list);
1220 list_splice_init(work_q, &tmp_list);
1222 while (!list_empty(&tmp_list)) {
1223 struct scsi_cmnd *next, *scmd;
1227 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1230 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1232 current->comm, id));
1233 rtn = scsi_try_target_reset(scmd);
1234 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1235 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1238 current->comm, id));
1239 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1240 if (scmd_id(scmd) != id)
1244 list_move_tail(&scmd->eh_entry, &check_list);
1245 else if (rtn == FAST_IO_FAIL)
1246 scsi_eh_finish_cmd(scmd, done_q);
1248 /* push back on work queue for further processing */
1249 list_move(&scmd->eh_entry, work_q);
1253 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1257 * scsi_eh_bus_reset - send a bus reset
1258 * @shost: &scsi host being recovered.
1259 * @work_q: &list_head for pending commands.
1260 * @done_q: &list_head for processed commands.
1262 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1263 struct list_head *work_q,
1264 struct list_head *done_q)
1266 struct scsi_cmnd *scmd, *chan_scmd, *next;
1267 LIST_HEAD(check_list);
1268 unsigned int channel;
1272 * we really want to loop over the various channels, and do this on
1273 * a channel by channel basis. we should also check to see if any
1274 * of the failed commands are on soft_reset devices, and if so, skip
1278 for (channel = 0; channel <= shost->max_channel; channel++) {
1280 list_for_each_entry(scmd, work_q, eh_entry) {
1281 if (channel == scmd_channel(scmd)) {
1285 * FIXME add back in some support for
1286 * soft_reset devices.
1293 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1294 " %d\n", current->comm,
1296 rtn = scsi_try_bus_reset(chan_scmd);
1297 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1298 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1299 if (channel == scmd_channel(scmd)) {
1300 if (rtn == FAST_IO_FAIL)
1301 scsi_eh_finish_cmd(scmd,
1304 list_move_tail(&scmd->eh_entry,
1309 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1310 " failed chan: %d\n",
1315 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1319 * scsi_eh_host_reset - send a host reset
1320 * @work_q: list_head for processed commands.
1321 * @done_q: list_head for processed commands.
1323 static int scsi_eh_host_reset(struct list_head *work_q,
1324 struct list_head *done_q)
1326 struct scsi_cmnd *scmd, *next;
1327 LIST_HEAD(check_list);
1330 if (!list_empty(work_q)) {
1331 scmd = list_entry(work_q->next,
1332 struct scsi_cmnd, eh_entry);
1334 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1337 rtn = scsi_try_host_reset(scmd);
1338 if (rtn == SUCCESS) {
1339 list_splice_init(work_q, &check_list);
1340 } else if (rtn == FAST_IO_FAIL) {
1341 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1342 scsi_eh_finish_cmd(scmd, done_q);
1345 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1350 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1354 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1355 * @work_q: list_head for processed commands.
1356 * @done_q: list_head for processed commands.
1358 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1359 struct list_head *done_q)
1361 struct scsi_cmnd *scmd, *next;
1363 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1364 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1365 "not ready after error recovery\n");
1366 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1367 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1369 * FIXME: Handle lost cmds.
1372 scsi_eh_finish_cmd(scmd, done_q);
1378 * scsi_noretry_cmd - determinte if command should be failed fast
1379 * @scmd: SCSI cmd to examine.
1381 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1383 switch (host_byte(scmd->result)) {
1387 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1389 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1391 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1392 status_byte(scmd->result) == RESERVATION_CONFLICT)
1395 case DID_SOFT_ERROR:
1396 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1399 switch (status_byte(scmd->result)) {
1400 case CHECK_CONDITION:
1402 * assume caller has checked sense and determinted
1403 * the check condition was retryable.
1405 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1406 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1414 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1415 * @scmd: SCSI cmd to examine.
1418 * This is *only* called when we are examining the status after sending
1419 * out the actual data command. any commands that are queued for error
1420 * recovery (e.g. test_unit_ready) do *not* come through here.
1422 * When this routine returns failed, it means the error handler thread
1423 * is woken. In cases where the error code indicates an error that
1424 * doesn't require the error handler read (i.e. we don't need to
1425 * abort/reset), this function should return SUCCESS.
1427 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1432 * if the device is offline, then we clearly just pass the result back
1433 * up to the top level.
1435 if (!scsi_device_online(scmd->device)) {
1436 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1443 * first check the host byte, to see if there is anything in there
1444 * that would indicate what we need to do.
1446 switch (host_byte(scmd->result)) {
1447 case DID_PASSTHROUGH:
1449 * no matter what, pass this through to the upper layer.
1450 * nuke this special code so that it looks like we are saying
1453 scmd->result &= 0xff00ffff;
1457 * looks good. drop through, and check the next byte.
1460 case DID_NO_CONNECT:
1461 case DID_BAD_TARGET:
1464 * note - this means that we just report the status back
1465 * to the top level driver, not that we actually think
1466 * that it indicates SUCCESS.
1470 * when the low level driver returns did_soft_error,
1471 * it is responsible for keeping an internal retry counter
1472 * in order to avoid endless loops (db)
1474 * actually this is a bug in this function here. we should
1475 * be mindful of the maximum number of retries specified
1476 * and not get stuck in a loop.
1478 case DID_SOFT_ERROR:
1484 return ADD_TO_MLQUEUE;
1485 case DID_TRANSPORT_DISRUPTED:
1487 * LLD/transport was disrupted during processing of the IO.
1488 * The transport class is now blocked/blocking,
1489 * and the transport will decide what to do with the IO
1490 * based on its timers and recovery capablilities if
1491 * there are enough retries.
1494 case DID_TRANSPORT_FAILFAST:
1496 * The transport decided to failfast the IO (most likely
1497 * the fast io fail tmo fired), so send IO directly upwards.
1501 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1502 status_byte(scmd->result) == RESERVATION_CONFLICT)
1504 * execute reservation conflict processing code
1514 * when we scan the bus, we get timeout messages for
1515 * these commands if there is no device available.
1516 * other hosts report did_no_connect for the same thing.
1518 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1519 scmd->cmnd[0] == INQUIRY)) {
1531 * next, check the message byte.
1533 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1537 * check the status byte to see if this indicates anything special.
1539 switch (status_byte(scmd->result)) {
1541 scsi_handle_queue_full(scmd->device);
1543 * the case of trying to send too many commands to a
1544 * tagged queueing device.
1548 * device can't talk to us at the moment. Should only
1549 * occur (SAM-3) when the task queue is empty, so will cause
1550 * the empty queue handling to trigger a stall in the
1553 return ADD_TO_MLQUEUE;
1555 scsi_handle_queue_ramp_up(scmd->device);
1556 case COMMAND_TERMINATED:
1560 case CHECK_CONDITION:
1561 rtn = scsi_check_sense(scmd);
1562 if (rtn == NEEDS_RETRY)
1564 else if (rtn == TARGET_ERROR) {
1566 * Need to modify host byte to signal a
1567 * permanent target failure
1569 set_host_byte(scmd, DID_TARGET_FAILURE);
1572 /* if rtn == FAILED, we have no sense information;
1573 * returning FAILED will wake the error handler thread
1574 * to collect the sense and redo the decide
1577 case CONDITION_GOOD:
1578 case INTERMEDIATE_GOOD:
1579 case INTERMEDIATE_C_GOOD:
1582 * who knows? FIXME(eric)
1586 case RESERVATION_CONFLICT:
1587 sdev_printk(KERN_INFO, scmd->device,
1588 "reservation conflict\n");
1589 set_host_byte(scmd, DID_NEXUS_FAILURE);
1590 return SUCCESS; /* causes immediate i/o error */
1598 /* we requeue for retry because the error was retryable, and
1599 * the request was not marked fast fail. Note that above,
1600 * even if the request is marked fast fail, we still requeue
1601 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1602 if ((++scmd->retries) <= scmd->allowed
1603 && !scsi_noretry_cmd(scmd)) {
1607 * no more retries - report this one back to upper level.
1613 static void eh_lock_door_done(struct request *req, int uptodate)
1615 __blk_put_request(req->q, req);
1619 * scsi_eh_lock_door - Prevent medium removal for the specified device
1620 * @sdev: SCSI device to prevent medium removal
1623 * We must be called from process context.
1626 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1627 * head of the devices request queue, and continue.
1629 static void scsi_eh_lock_door(struct scsi_device *sdev)
1631 struct request *req;
1634 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1635 * request becomes available
1637 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1639 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1643 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1646 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1648 req->cmd_type = REQ_TYPE_BLOCK_PC;
1649 req->cmd_flags |= REQ_QUIET;
1650 req->timeout = 10 * HZ;
1653 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1657 * scsi_restart_operations - restart io operations to the specified host.
1658 * @shost: Host we are restarting.
1661 * When we entered the error handler, we blocked all further i/o to
1662 * this device. we need to 'reverse' this process.
1664 static void scsi_restart_operations(struct Scsi_Host *shost)
1666 struct scsi_device *sdev;
1667 unsigned long flags;
1670 * If the door was locked, we need to insert a door lock request
1671 * onto the head of the SCSI request queue for the device. There
1672 * is no point trying to lock the door of an off-line device.
1674 shost_for_each_device(sdev, shost) {
1675 if (scsi_device_online(sdev) && sdev->locked)
1676 scsi_eh_lock_door(sdev);
1680 * next free up anything directly waiting upon the host. this
1681 * will be requests for character device operations, and also for
1682 * ioctls to queued block devices.
1684 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1687 spin_lock_irqsave(shost->host_lock, flags);
1688 if (scsi_host_set_state(shost, SHOST_RUNNING))
1689 if (scsi_host_set_state(shost, SHOST_CANCEL))
1690 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1691 spin_unlock_irqrestore(shost->host_lock, flags);
1693 wake_up(&shost->host_wait);
1696 * finally we need to re-initiate requests that may be pending. we will
1697 * have had everything blocked while error handling is taking place, and
1698 * now that error recovery is done, we will need to ensure that these
1699 * requests are started.
1701 scsi_run_host_queues(shost);
1704 * if eh is active and host_eh_scheduled is pending we need to re-run
1705 * recovery. we do this check after scsi_run_host_queues() to allow
1706 * everything pent up since the last eh run a chance to make forward
1707 * progress before we sync again. Either we'll immediately re-run
1708 * recovery or scsi_device_unbusy() will wake us again when these
1709 * pending commands complete.
1711 spin_lock_irqsave(shost->host_lock, flags);
1712 if (shost->host_eh_scheduled)
1713 if (scsi_host_set_state(shost, SHOST_RECOVERY))
1714 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
1715 spin_unlock_irqrestore(shost->host_lock, flags);
1719 * scsi_eh_ready_devs - check device ready state and recover if not.
1720 * @shost: host to be recovered.
1721 * @work_q: &list_head for pending commands.
1722 * @done_q: &list_head for processed commands.
1724 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1725 struct list_head *work_q,
1726 struct list_head *done_q)
1728 if (!scsi_eh_stu(shost, work_q, done_q))
1729 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1730 if (!scsi_eh_target_reset(shost, work_q, done_q))
1731 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1732 if (!scsi_eh_host_reset(work_q, done_q))
1733 scsi_eh_offline_sdevs(work_q,
1736 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1739 * scsi_eh_flush_done_q - finish processed commands or retry them.
1740 * @done_q: list_head of processed commands.
1742 void scsi_eh_flush_done_q(struct list_head *done_q)
1744 struct scsi_cmnd *scmd, *next;
1746 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1747 list_del_init(&scmd->eh_entry);
1748 if (scsi_device_online(scmd->device) &&
1749 !scsi_noretry_cmd(scmd) &&
1750 (++scmd->retries <= scmd->allowed)) {
1751 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1755 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1758 * If just we got sense for the device (called
1759 * scsi_eh_get_sense), scmd->result is already
1760 * set, do not set DRIVER_TIMEOUT.
1763 scmd->result |= (DRIVER_TIMEOUT << 24);
1764 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1766 current->comm, scmd));
1767 scsi_finish_command(scmd);
1771 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1774 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1775 * @shost: Host to unjam.
1778 * When we come in here, we *know* that all commands on the bus have
1779 * either completed, failed or timed out. we also know that no further
1780 * commands are being sent to the host, so things are relatively quiet
1781 * and we have freedom to fiddle with things as we wish.
1783 * This is only the *default* implementation. it is possible for
1784 * individual drivers to supply their own version of this function, and
1785 * if the maintainer wishes to do this, it is strongly suggested that
1786 * this function be taken as a template and modified. this function
1787 * was designed to correctly handle problems for about 95% of the
1788 * different cases out there, and it should always provide at least a
1789 * reasonable amount of error recovery.
1791 * Any command marked 'failed' or 'timeout' must eventually have
1792 * scsi_finish_cmd() called for it. we do all of the retry stuff
1793 * here, so when we restart the host after we return it should have an
1796 static void scsi_unjam_host(struct Scsi_Host *shost)
1798 unsigned long flags;
1799 LIST_HEAD(eh_work_q);
1800 LIST_HEAD(eh_done_q);
1802 spin_lock_irqsave(shost->host_lock, flags);
1803 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1804 spin_unlock_irqrestore(shost->host_lock, flags);
1806 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1808 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1809 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1810 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1812 scsi_eh_flush_done_q(&eh_done_q);
1816 * scsi_error_handler - SCSI error handler thread
1817 * @data: Host for which we are running.
1820 * This is the main error handling loop. This is run as a kernel thread
1821 * for every SCSI host and handles all error handling activity.
1823 int scsi_error_handler(void *data)
1825 struct Scsi_Host *shost = data;
1828 * We use TASK_INTERRUPTIBLE so that the thread is not
1829 * counted against the load average as a running process.
1830 * We never actually get interrupted because kthread_run
1831 * disables signal delivery for the created thread.
1833 while (!kthread_should_stop()) {
1834 set_current_state(TASK_INTERRUPTIBLE);
1835 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1836 shost->host_failed != shost->host_busy) {
1837 SCSI_LOG_ERROR_RECOVERY(1,
1838 printk("Error handler scsi_eh_%d sleeping\n",
1844 __set_current_state(TASK_RUNNING);
1845 SCSI_LOG_ERROR_RECOVERY(1,
1846 printk("Error handler scsi_eh_%d waking up\n",
1850 * We have a host that is failing for some reason. Figure out
1851 * what we need to do to get it up and online again (if we can).
1852 * If we fail, we end up taking the thing offline.
1854 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
1855 SCSI_LOG_ERROR_RECOVERY(1,
1856 printk(KERN_ERR "Error handler scsi_eh_%d "
1857 "unable to autoresume\n",
1862 if (shost->transportt->eh_strategy_handler)
1863 shost->transportt->eh_strategy_handler(shost);
1865 scsi_unjam_host(shost);
1868 * Note - if the above fails completely, the action is to take
1869 * individual devices offline and flush the queue of any
1870 * outstanding requests that may have been pending. When we
1871 * restart, we restart any I/O to any other devices on the bus
1872 * which are still online.
1874 scsi_restart_operations(shost);
1875 if (!shost->eh_noresume)
1876 scsi_autopm_put_host(shost);
1878 __set_current_state(TASK_RUNNING);
1880 SCSI_LOG_ERROR_RECOVERY(1,
1881 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1882 shost->ehandler = NULL;
1887 * Function: scsi_report_bus_reset()
1889 * Purpose: Utility function used by low-level drivers to report that
1890 * they have observed a bus reset on the bus being handled.
1892 * Arguments: shost - Host in question
1893 * channel - channel on which reset was observed.
1897 * Lock status: Host lock must be held.
1899 * Notes: This only needs to be called if the reset is one which
1900 * originates from an unknown location. Resets originated
1901 * by the mid-level itself don't need to call this, but there
1902 * should be no harm.
1904 * The main purpose of this is to make sure that a CHECK_CONDITION
1905 * is properly treated.
1907 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1909 struct scsi_device *sdev;
1911 __shost_for_each_device(sdev, shost) {
1912 if (channel == sdev_channel(sdev))
1913 __scsi_report_device_reset(sdev, NULL);
1916 EXPORT_SYMBOL(scsi_report_bus_reset);
1919 * Function: scsi_report_device_reset()
1921 * Purpose: Utility function used by low-level drivers to report that
1922 * they have observed a device reset on the device being handled.
1924 * Arguments: shost - Host in question
1925 * channel - channel on which reset was observed
1926 * target - target on which reset was observed
1930 * Lock status: Host lock must be held
1932 * Notes: This only needs to be called if the reset is one which
1933 * originates from an unknown location. Resets originated
1934 * by the mid-level itself don't need to call this, but there
1935 * should be no harm.
1937 * The main purpose of this is to make sure that a CHECK_CONDITION
1938 * is properly treated.
1940 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1942 struct scsi_device *sdev;
1944 __shost_for_each_device(sdev, shost) {
1945 if (channel == sdev_channel(sdev) &&
1946 target == sdev_id(sdev))
1947 __scsi_report_device_reset(sdev, NULL);
1950 EXPORT_SYMBOL(scsi_report_device_reset);
1953 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1958 * Function: scsi_reset_provider
1960 * Purpose: Send requested reset to a bus or device at any phase.
1962 * Arguments: device - device to send reset to
1963 * flag - reset type (see scsi.h)
1965 * Returns: SUCCESS/FAILURE.
1967 * Notes: This is used by the SCSI Generic driver to provide
1968 * Bus/Device reset capability.
1971 scsi_reset_provider(struct scsi_device *dev, int flag)
1973 struct scsi_cmnd *scmd;
1974 struct Scsi_Host *shost = dev->host;
1976 unsigned long flags;
1979 if (scsi_autopm_get_host(shost) < 0)
1982 scmd = scsi_get_command(dev, GFP_KERNEL);
1983 blk_rq_init(NULL, &req);
1984 scmd->request = &req;
1986 scmd->cmnd = req.cmd;
1988 scmd->scsi_done = scsi_reset_provider_done_command;
1989 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1993 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1995 spin_lock_irqsave(shost->host_lock, flags);
1996 shost->tmf_in_progress = 1;
1997 spin_unlock_irqrestore(shost->host_lock, flags);
2000 case SCSI_TRY_RESET_DEVICE:
2001 rtn = scsi_try_bus_device_reset(scmd);
2005 case SCSI_TRY_RESET_TARGET:
2006 rtn = scsi_try_target_reset(scmd);
2010 case SCSI_TRY_RESET_BUS:
2011 rtn = scsi_try_bus_reset(scmd);
2015 case SCSI_TRY_RESET_HOST:
2016 rtn = scsi_try_host_reset(scmd);
2022 spin_lock_irqsave(shost->host_lock, flags);
2023 shost->tmf_in_progress = 0;
2024 spin_unlock_irqrestore(shost->host_lock, flags);
2027 * be sure to wake up anyone who was sleeping or had their queue
2028 * suspended while we performed the TMF.
2030 SCSI_LOG_ERROR_RECOVERY(3,
2031 printk("%s: waking up host to restart after TMF\n",
2034 wake_up(&shost->host_wait);
2036 scsi_run_host_queues(shost);
2038 scsi_next_command(scmd);
2039 scsi_autopm_put_host(shost);
2042 EXPORT_SYMBOL(scsi_reset_provider);
2045 * scsi_normalize_sense - normalize main elements from either fixed or
2046 * descriptor sense data format into a common format.
2048 * @sense_buffer: byte array containing sense data returned by device
2049 * @sb_len: number of valid bytes in sense_buffer
2050 * @sshdr: pointer to instance of structure that common
2051 * elements are written to.
2054 * The "main elements" from sense data are: response_code, sense_key,
2055 * asc, ascq and additional_length (only for descriptor format).
2057 * Typically this function can be called after a device has
2058 * responded to a SCSI command with the CHECK_CONDITION status.
2061 * 1 if valid sense data information found, else 0;
2063 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2064 struct scsi_sense_hdr *sshdr)
2066 if (!sense_buffer || !sb_len)
2069 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2071 sshdr->response_code = (sense_buffer[0] & 0x7f);
2073 if (!scsi_sense_valid(sshdr))
2076 if (sshdr->response_code >= 0x72) {
2081 sshdr->sense_key = (sense_buffer[1] & 0xf);
2083 sshdr->asc = sense_buffer[2];
2085 sshdr->ascq = sense_buffer[3];
2087 sshdr->additional_length = sense_buffer[7];
2093 sshdr->sense_key = (sense_buffer[2] & 0xf);
2095 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2096 sb_len : (sense_buffer[7] + 8);
2098 sshdr->asc = sense_buffer[12];
2100 sshdr->ascq = sense_buffer[13];
2106 EXPORT_SYMBOL(scsi_normalize_sense);
2108 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2109 struct scsi_sense_hdr *sshdr)
2111 return scsi_normalize_sense(cmd->sense_buffer,
2112 SCSI_SENSE_BUFFERSIZE, sshdr);
2114 EXPORT_SYMBOL(scsi_command_normalize_sense);
2117 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2118 * @sense_buffer: byte array of descriptor format sense data
2119 * @sb_len: number of valid bytes in sense_buffer
2120 * @desc_type: value of descriptor type to find
2121 * (e.g. 0 -> information)
2124 * only valid when sense data is in descriptor format
2127 * pointer to start of (first) descriptor if found else NULL
2129 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2132 int add_sen_len, add_len, desc_len, k;
2135 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2137 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2139 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2140 add_sen_len : (sb_len - 8);
2141 descp = &sense_buffer[8];
2142 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2144 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2145 desc_len = add_len + 2;
2146 if (descp[0] == desc_type)
2148 if (add_len < 0) // short descriptor ??
2153 EXPORT_SYMBOL(scsi_sense_desc_find);
2156 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2157 * @sense_buffer: byte array of sense data
2158 * @sb_len: number of valid bytes in sense_buffer
2159 * @info_out: pointer to 64 integer where 8 or 4 byte information
2160 * field will be placed if found.
2163 * 1 if information field found, 0 if not found.
2165 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2174 switch (sense_buffer[0] & 0x7f) {
2177 if (sense_buffer[0] & 0x80) {
2178 *info_out = (sense_buffer[3] << 24) +
2179 (sense_buffer[4] << 16) +
2180 (sense_buffer[5] << 8) + sense_buffer[6];
2186 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2188 if (ucp && (0xa == ucp[1])) {
2190 for (j = 0; j < 8; ++j) {
2203 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2206 * scsi_build_sense_buffer - build sense data in a buffer
2207 * @desc: Sense format (non zero == descriptor format,
2208 * 0 == fixed format)
2209 * @buf: Where to build sense data
2211 * @asc: Additional sense code
2212 * @ascq: Additional sense code qualifier
2215 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2218 buf[0] = 0x72; /* descriptor, current */
2224 buf[0] = 0x70; /* fixed, current */
2231 EXPORT_SYMBOL(scsi_build_sense_buffer);