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>
28 #include <linux/jiffies.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 #include "scsi_transport_api.h"
44 #include <trace/events/scsi.h>
46 static void scsi_eh_done(struct scsi_cmnd *scmd);
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
52 #define BUS_RESET_SETTLE_TIME (10)
53 #define HOST_RESET_SETTLE_TIME (10)
55 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
57 /* called with shost->host_lock held */
58 void scsi_eh_wakeup(struct Scsi_Host *shost)
60 if (shost->host_busy == shost->host_failed) {
61 trace_scsi_eh_wakeup(shost);
62 wake_up_process(shost->ehandler);
63 SCSI_LOG_ERROR_RECOVERY(5,
64 printk("Waking error handler thread\n"));
69 * scsi_schedule_eh - schedule EH for SCSI host
70 * @shost: SCSI host to invoke error handling on.
72 * Schedule SCSI EH without scmd.
74 void scsi_schedule_eh(struct Scsi_Host *shost)
78 spin_lock_irqsave(shost->host_lock, flags);
80 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
81 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
82 shost->host_eh_scheduled++;
83 scsi_eh_wakeup(shost);
86 spin_unlock_irqrestore(shost->host_lock, flags);
88 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
91 * scsi_eh_scmd_add - add scsi cmd to error handling.
92 * @scmd: scmd to run eh on.
93 * @eh_flag: optional SCSI_EH flag.
98 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
100 struct Scsi_Host *shost = scmd->device->host;
104 if (!shost->ehandler)
107 spin_lock_irqsave(shost->host_lock, flags);
108 if (scsi_host_set_state(shost, SHOST_RECOVERY))
109 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
113 scmd->eh_eflags |= eh_flag;
114 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
115 shost->host_failed++;
116 scsi_eh_wakeup(shost);
118 spin_unlock_irqrestore(shost->host_lock, flags);
123 * scsi_times_out - Timeout function for normal scsi commands.
124 * @req: request that is timing out.
127 * We do not need to lock this. There is the potential for a race
128 * only in that the normal completion handling might run, but if the
129 * normal completion function determines that the timer has already
130 * fired, then it mustn't do anything.
132 enum blk_eh_timer_return scsi_times_out(struct request *req)
134 struct scsi_cmnd *scmd = req->special;
135 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
136 struct Scsi_Host *host = scmd->device->host;
138 trace_scsi_dispatch_cmd_timeout(scmd);
139 scsi_log_completion(scmd, TIMEOUT_ERROR);
141 if (host->transportt->eh_timed_out)
142 rtn = host->transportt->eh_timed_out(scmd);
143 else if (host->hostt->eh_timed_out)
144 rtn = host->hostt->eh_timed_out(scmd);
146 scmd->result |= DID_TIME_OUT << 16;
148 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
149 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
150 rtn = BLK_EH_HANDLED;
156 * scsi_block_when_processing_errors - Prevent cmds from being queued.
157 * @sdev: Device on which we are performing recovery.
160 * We block until the host is out of error recovery, and then check to
161 * see whether the host or the device is offline.
164 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
166 int scsi_block_when_processing_errors(struct scsi_device *sdev)
170 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
172 online = scsi_device_online(sdev);
174 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
179 EXPORT_SYMBOL(scsi_block_when_processing_errors);
181 #ifdef CONFIG_SCSI_LOGGING
183 * scsi_eh_prt_fail_stats - Log info on failures.
184 * @shost: scsi host being recovered.
185 * @work_q: Queue of scsi cmds to process.
187 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
188 struct list_head *work_q)
190 struct scsi_cmnd *scmd;
191 struct scsi_device *sdev;
192 int total_failures = 0;
195 int devices_failed = 0;
197 shost_for_each_device(sdev, shost) {
198 list_for_each_entry(scmd, work_q, eh_entry) {
199 if (scmd->device == sdev) {
201 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
208 if (cmd_cancel || cmd_failed) {
209 SCSI_LOG_ERROR_RECOVERY(3,
210 sdev_printk(KERN_INFO, sdev,
211 "%s: cmds failed: %d, cancel: %d\n",
212 __func__, cmd_failed,
220 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
221 " devices require eh work\n",
222 total_failures, devices_failed));
227 * scsi_check_sense - Examine scsi cmd sense
228 * @scmd: Cmd to have sense checked.
231 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
234 * When a deferred error is detected the current command has
235 * not been executed and needs retrying.
237 static int scsi_check_sense(struct scsi_cmnd *scmd)
239 struct scsi_device *sdev = scmd->device;
240 struct scsi_sense_hdr sshdr;
242 if (! scsi_command_normalize_sense(scmd, &sshdr))
243 return FAILED; /* no valid sense data */
245 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
247 * nasty: for mid-layer issued TURs, we need to return the
248 * actual sense data without any recovery attempt. For eh
249 * issued ones, we need to try to recover and interpret
253 if (scsi_sense_is_deferred(&sshdr))
256 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
257 sdev->scsi_dh_data->scsi_dh->check_sense) {
260 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
261 if (rc != SCSI_RETURN_NOT_HANDLED)
263 /* handler does not care. Drop down to default handling */
267 * Previous logic looked for FILEMARK, EOM or ILI which are
268 * mainly associated with tapes and returned SUCCESS.
270 if (sshdr.response_code == 0x70) {
272 if (scmd->sense_buffer[2] & 0xe0)
276 * descriptor format: look for "stream commands sense data
277 * descriptor" (see SSC-3). Assume single sense data
278 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
280 if ((sshdr.additional_length > 3) &&
281 (scmd->sense_buffer[8] == 0x4) &&
282 (scmd->sense_buffer[11] & 0xe0))
286 switch (sshdr.sense_key) {
289 case RECOVERED_ERROR:
290 return /* soft_error */ SUCCESS;
292 case ABORTED_COMMAND:
293 if (sshdr.asc == 0x10) /* DIF */
300 * if we are expecting a cc/ua because of a bus reset that we
301 * performed, treat this just as a retry. otherwise this is
302 * information that we should pass up to the upper-level driver
303 * so that we can deal with it there.
305 if (scmd->device->expecting_cc_ua) {
307 * Because some device does not queue unit
308 * attentions correctly, we carefully check
309 * additional sense code and qualifier so as
310 * not to squash media change unit attention.
312 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
313 scmd->device->expecting_cc_ua = 0;
318 * if the device is in the process of becoming ready, we
321 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
324 * if the device is not started, we need to wake
325 * the error handler to start the motor
327 if (scmd->device->allow_restart &&
328 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
331 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
332 scmd_printk(KERN_WARNING, scmd,
333 "Warning! Received an indication that the "
334 "LUN assignments on this target have "
335 "changed. The Linux SCSI layer does not "
336 "automatically remap LUN assignments.\n");
337 else if (sshdr.asc == 0x3f)
338 scmd_printk(KERN_WARNING, scmd,
339 "Warning! Received an indication that the "
340 "operating parameters on this target have "
341 "changed. The Linux SCSI layer does not "
342 "automatically adjust these parameters.\n");
344 if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
345 scmd_printk(KERN_WARNING, scmd,
346 "Warning! Received an indication that the "
347 "LUN reached a thin provisioning soft "
351 * Pass the UA upwards for a determination in the completion
356 /* these are not supported */
358 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
359 /* Thin provisioning hard threshold reached */
360 set_host_byte(scmd, DID_ALLOC_FAILURE);
364 case VOLUME_OVERFLOW:
367 set_host_byte(scmd, DID_TARGET_FAILURE);
371 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
372 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
373 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
374 set_host_byte(scmd, DID_MEDIUM_ERROR);
380 if (scmd->device->retry_hwerror)
381 return ADD_TO_MLQUEUE;
383 set_host_byte(scmd, DID_TARGET_FAILURE);
385 case ILLEGAL_REQUEST:
386 if (sshdr.asc == 0x20 || /* Invalid command operation code */
387 sshdr.asc == 0x21 || /* Logical block address out of range */
388 sshdr.asc == 0x24 || /* Invalid field in cdb */
389 sshdr.asc == 0x26) { /* Parameter value invalid */
390 set_host_byte(scmd, DID_TARGET_FAILURE);
399 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
401 struct scsi_host_template *sht = sdev->host->hostt;
402 struct scsi_device *tmp_sdev;
404 if (!sht->change_queue_depth ||
405 sdev->queue_depth >= sdev->max_queue_depth)
408 if (time_before(jiffies,
409 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
412 if (time_before(jiffies,
413 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
417 * Walk all devices of a target and do
420 shost_for_each_device(tmp_sdev, sdev->host) {
421 if (tmp_sdev->channel != sdev->channel ||
422 tmp_sdev->id != sdev->id ||
423 tmp_sdev->queue_depth == sdev->max_queue_depth)
426 * call back into LLD to increase queue_depth by one
427 * with ramp up reason code.
429 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
430 SCSI_QDEPTH_RAMP_UP);
431 sdev->last_queue_ramp_up = jiffies;
435 static void scsi_handle_queue_full(struct scsi_device *sdev)
437 struct scsi_host_template *sht = sdev->host->hostt;
438 struct scsi_device *tmp_sdev;
440 if (!sht->change_queue_depth)
443 shost_for_each_device(tmp_sdev, sdev->host) {
444 if (tmp_sdev->channel != sdev->channel ||
445 tmp_sdev->id != sdev->id)
448 * We do not know the number of commands that were at
449 * the device when we got the queue full so we start
450 * from the highest possible value and work our way down.
452 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
458 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
459 * @scmd: SCSI cmd to examine.
462 * This is *only* called when we are examining the status of commands
463 * queued during error recovery. the main difference here is that we
464 * don't allow for the possibility of retries here, and we are a lot
465 * more restrictive about what we consider acceptable.
467 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
470 * first check the host byte, to see if there is anything in there
471 * that would indicate what we need to do.
473 if (host_byte(scmd->result) == DID_RESET) {
475 * rats. we are already in the error handler, so we now
476 * get to try and figure out what to do next. if the sense
477 * is valid, we have a pretty good idea of what to do.
478 * if not, we mark it as FAILED.
480 return scsi_check_sense(scmd);
482 if (host_byte(scmd->result) != DID_OK)
486 * next, check the message byte.
488 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
492 * now, check the status byte to see if this indicates
495 switch (status_byte(scmd->result)) {
497 scsi_handle_queue_ramp_up(scmd->device);
498 case COMMAND_TERMINATED:
500 case CHECK_CONDITION:
501 return scsi_check_sense(scmd);
503 case INTERMEDIATE_GOOD:
504 case INTERMEDIATE_C_GOOD:
506 * who knows? FIXME(eric)
509 case RESERVATION_CONFLICT:
510 if (scmd->cmnd[0] == TEST_UNIT_READY)
511 /* it is a success, we probed the device and
514 /* otherwise, we failed to send the command */
517 scsi_handle_queue_full(scmd->device);
528 * scsi_eh_done - Completion function for error handling.
529 * @scmd: Cmd that is done.
531 static void scsi_eh_done(struct scsi_cmnd *scmd)
533 struct completion *eh_action;
535 SCSI_LOG_ERROR_RECOVERY(3,
536 printk("%s scmd: %p result: %x\n",
537 __func__, scmd, scmd->result));
539 eh_action = scmd->device->host->eh_action;
545 * scsi_try_host_reset - ask host adapter to reset itself
546 * @scmd: SCSI cmd to send host reset.
548 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
552 struct Scsi_Host *host = scmd->device->host;
553 struct scsi_host_template *hostt = host->hostt;
555 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
558 if (!hostt->eh_host_reset_handler)
561 rtn = hostt->eh_host_reset_handler(scmd);
563 if (rtn == SUCCESS) {
564 if (!hostt->skip_settle_delay)
565 ssleep(HOST_RESET_SETTLE_TIME);
566 spin_lock_irqsave(host->host_lock, flags);
567 scsi_report_bus_reset(host, scmd_channel(scmd));
568 spin_unlock_irqrestore(host->host_lock, flags);
575 * scsi_try_bus_reset - ask host to perform a bus reset
576 * @scmd: SCSI cmd to send bus reset.
578 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
582 struct Scsi_Host *host = scmd->device->host;
583 struct scsi_host_template *hostt = host->hostt;
585 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
588 if (!hostt->eh_bus_reset_handler)
591 rtn = hostt->eh_bus_reset_handler(scmd);
593 if (rtn == SUCCESS) {
594 if (!hostt->skip_settle_delay)
595 ssleep(BUS_RESET_SETTLE_TIME);
596 spin_lock_irqsave(host->host_lock, flags);
597 scsi_report_bus_reset(host, scmd_channel(scmd));
598 spin_unlock_irqrestore(host->host_lock, flags);
604 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
607 sdev->expecting_cc_ua = 1;
611 * scsi_try_target_reset - Ask host to perform a target reset
612 * @scmd: SCSI cmd used to send a target reset
615 * There is no timeout for this operation. if this operation is
616 * unreliable for a given host, then the host itself needs to put a
617 * timer on it, and set the host back to a consistent state prior to
620 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
624 struct Scsi_Host *host = scmd->device->host;
625 struct scsi_host_template *hostt = host->hostt;
627 if (!hostt->eh_target_reset_handler)
630 rtn = hostt->eh_target_reset_handler(scmd);
631 if (rtn == SUCCESS) {
632 spin_lock_irqsave(host->host_lock, flags);
633 __starget_for_each_device(scsi_target(scmd->device), NULL,
634 __scsi_report_device_reset);
635 spin_unlock_irqrestore(host->host_lock, flags);
642 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
643 * @scmd: SCSI cmd used to send BDR
646 * There is no timeout for this operation. if this operation is
647 * unreliable for a given host, then the host itself needs to put a
648 * timer on it, and set the host back to a consistent state prior to
651 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
654 struct scsi_host_template *hostt = scmd->device->host->hostt;
656 if (!hostt->eh_device_reset_handler)
659 rtn = hostt->eh_device_reset_handler(scmd);
661 __scsi_report_device_reset(scmd->device, NULL);
665 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
667 if (!hostt->eh_abort_handler)
670 return hostt->eh_abort_handler(scmd);
673 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
675 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
676 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
677 if (scsi_try_target_reset(scmd) != SUCCESS)
678 if (scsi_try_bus_reset(scmd) != SUCCESS)
679 scsi_try_host_reset(scmd);
683 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
684 * @scmd: SCSI command structure to hijack
685 * @ses: structure to save restore information
686 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
687 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
688 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
690 * This function is used to save a scsi command information before re-execution
691 * as part of the error recovery process. If @sense_bytes is 0 the command
692 * sent must be one that does not transfer any data. If @sense_bytes != 0
693 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
694 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
696 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
697 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
699 struct scsi_device *sdev = scmd->device;
702 * We need saved copies of a number of fields - this is because
703 * error handling may need to overwrite these with different values
704 * to run different commands, and once error handling is complete,
705 * we will need to restore these values prior to running the actual
708 ses->cmd_len = scmd->cmd_len;
709 ses->cmnd = scmd->cmnd;
710 ses->data_direction = scmd->sc_data_direction;
711 ses->sdb = scmd->sdb;
712 ses->next_rq = scmd->request->next_rq;
713 ses->result = scmd->result;
714 ses->underflow = scmd->underflow;
715 ses->prot_op = scmd->prot_op;
717 scmd->prot_op = SCSI_PROT_NORMAL;
718 scmd->cmnd = ses->eh_cmnd;
719 memset(scmd->cmnd, 0, BLK_MAX_CDB);
720 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
721 scmd->request->next_rq = NULL;
724 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
726 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
728 scmd->sdb.table.sgl = &ses->sense_sgl;
729 scmd->sc_data_direction = DMA_FROM_DEVICE;
730 scmd->sdb.table.nents = 1;
731 scmd->cmnd[0] = REQUEST_SENSE;
732 scmd->cmnd[4] = scmd->sdb.length;
733 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
735 scmd->sc_data_direction = DMA_NONE;
737 BUG_ON(cmnd_size > BLK_MAX_CDB);
738 memcpy(scmd->cmnd, cmnd, cmnd_size);
739 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
745 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
746 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
747 (sdev->lun << 5 & 0xe0);
750 * Zero the sense buffer. The scsi spec mandates that any
751 * untransferred sense data should be interpreted as being zero.
753 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
755 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
758 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
759 * @scmd: SCSI command structure to restore
760 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
762 * Undo any damage done by above scsi_eh_prep_cmnd().
764 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
767 * Restore original data
769 scmd->cmd_len = ses->cmd_len;
770 scmd->cmnd = ses->cmnd;
771 scmd->sc_data_direction = ses->data_direction;
772 scmd->sdb = ses->sdb;
773 scmd->request->next_rq = ses->next_rq;
774 scmd->result = ses->result;
775 scmd->underflow = ses->underflow;
776 scmd->prot_op = ses->prot_op;
778 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
781 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
782 * @scmd: SCSI command structure to hijack
784 * @cmnd_size: size in bytes of @cmnd
785 * @timeout: timeout for this request
786 * @sense_bytes: size of sense data to copy or 0
788 * This function is used to send a scsi command down to a target device
789 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
792 * SUCCESS or FAILED or NEEDS_RETRY
794 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
795 int cmnd_size, int timeout, unsigned sense_bytes)
797 struct scsi_device *sdev = scmd->device;
798 struct Scsi_Host *shost = sdev->host;
799 DECLARE_COMPLETION_ONSTACK(done);
800 unsigned long timeleft = timeout;
801 struct scsi_eh_save ses;
802 const unsigned long stall_for = msecs_to_jiffies(100);
806 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
807 shost->eh_action = &done;
810 scmd->scsi_done = scsi_eh_done;
811 rtn = shost->hostt->queuecommand(shost, scmd);
813 if (timeleft > stall_for) {
814 scsi_eh_restore_cmnd(scmd, &ses);
815 timeleft -= stall_for;
816 msleep(jiffies_to_msecs(stall_for));
819 /* signal not to enter either branch of the if () below */
823 timeleft = wait_for_completion_timeout(&done, timeout);
826 shost->eh_action = NULL;
828 scsi_log_completion(scmd, rtn);
830 SCSI_LOG_ERROR_RECOVERY(3,
831 printk("%s: scmd: %p, timeleft: %ld\n",
832 __func__, scmd, timeleft));
835 * If there is time left scsi_eh_done got called, and we will examine
836 * the actual status codes to see whether the command actually did
837 * complete normally, else if we have a zero return and no time left,
838 * the command must still be pending, so abort it and return FAILED.
839 * If we never actually managed to issue the command, because
840 * ->queuecommand() kept returning non zero, use the rtn = FAILED
841 * value above (so don't execute either branch of the if)
844 rtn = scsi_eh_completed_normally(scmd);
845 SCSI_LOG_ERROR_RECOVERY(3,
846 printk("%s: scsi_eh_completed_normally %x\n",
862 scsi_abort_eh_cmnd(scmd);
866 scsi_eh_restore_cmnd(scmd, &ses);
868 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
869 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
871 rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
878 * scsi_request_sense - Request sense data from a particular target.
879 * @scmd: SCSI cmd for request sense.
882 * Some hosts automatically obtain this information, others require
883 * that we obtain it on our own. This function will *not* return until
884 * the command either times out, or it completes.
886 static int scsi_request_sense(struct scsi_cmnd *scmd)
888 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
892 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
893 * @scmd: Original SCSI cmd that eh has finished.
894 * @done_q: Queue for processed commands.
897 * We don't want to use the normal command completion while we are are
898 * still handling errors - it may cause other commands to be queued,
899 * and that would disturb what we are doing. Thus we really want to
900 * keep a list of pending commands for final completion, and once we
901 * are ready to leave error handling we handle completion for real.
903 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
905 scmd->device->host->host_failed--;
907 list_move_tail(&scmd->eh_entry, done_q);
909 EXPORT_SYMBOL(scsi_eh_finish_cmd);
912 * scsi_eh_get_sense - Get device sense data.
913 * @work_q: Queue of commands to process.
914 * @done_q: Queue of processed commands.
917 * See if we need to request sense information. if so, then get it
918 * now, so we have a better idea of what to do.
921 * This has the unfortunate side effect that if a shost adapter does
922 * not automatically request sense information, we end up shutting
923 * it down before we request it.
925 * All drivers should request sense information internally these days,
926 * so for now all I have to say is tough noogies if you end up in here.
928 * XXX: Long term this code should go away, but that needs an audit of
931 int scsi_eh_get_sense(struct list_head *work_q,
932 struct list_head *done_q)
934 struct scsi_cmnd *scmd, *next;
937 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
938 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
939 SCSI_SENSE_VALID(scmd))
942 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
943 "%s: requesting sense\n",
945 rtn = scsi_request_sense(scmd);
949 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
950 " result %x\n", scmd,
952 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
954 rtn = scsi_decide_disposition(scmd);
957 * if the result was normal, then just pass it along to the
961 /* we don't want this command reissued, just
962 * finished with the sense data, so set
963 * retries to the max allowed to ensure it
964 * won't get reissued */
965 scmd->retries = scmd->allowed;
966 else if (rtn != NEEDS_RETRY)
969 scsi_eh_finish_cmd(scmd, done_q);
972 return list_empty(work_q);
974 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
977 * scsi_eh_tur - Send TUR to device.
978 * @scmd: &scsi_cmnd to send TUR
981 * 0 - Device is ready. 1 - Device NOT ready.
983 static int scsi_eh_tur(struct scsi_cmnd *scmd)
985 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
986 int retry_cnt = 1, rtn;
989 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
990 scmd->device->eh_timeout, 0);
992 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
993 __func__, scmd, rtn));
1008 * scsi_eh_test_devices - check if devices are responding from error recovery.
1009 * @cmd_list: scsi commands in error recovery.
1010 * @work_q: queue for commands which still need more error recovery
1011 * @done_q: queue for commands which are finished
1012 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1015 * Tests if devices are in a working state. Commands to devices now in
1016 * a working state are sent to the done_q while commands to devices which
1017 * are still failing to respond are returned to the work_q for more
1020 static int scsi_eh_test_devices(struct list_head *cmd_list,
1021 struct list_head *work_q,
1022 struct list_head *done_q, int try_stu)
1024 struct scsi_cmnd *scmd, *next;
1025 struct scsi_device *sdev;
1028 while (!list_empty(cmd_list)) {
1029 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1030 sdev = scmd->device;
1032 finish_cmds = !scsi_device_online(scmd->device) ||
1033 (try_stu && !scsi_eh_try_stu(scmd) &&
1034 !scsi_eh_tur(scmd)) ||
1037 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1038 if (scmd->device == sdev) {
1040 scsi_eh_finish_cmd(scmd, done_q);
1042 list_move_tail(&scmd->eh_entry, work_q);
1045 return list_empty(work_q);
1050 * scsi_eh_abort_cmds - abort pending commands.
1051 * @work_q: &list_head for pending commands.
1052 * @done_q: &list_head for processed commands.
1055 * Try and see whether or not it makes sense to try and abort the
1056 * running command. This only works out to be the case if we have one
1057 * command that has timed out. If the command simply failed, it makes
1058 * no sense to try and abort the command, since as far as the shost
1059 * adapter is concerned, it isn't running.
1061 static int scsi_eh_abort_cmds(struct list_head *work_q,
1062 struct list_head *done_q)
1064 struct scsi_cmnd *scmd, *next;
1065 LIST_HEAD(check_list);
1068 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1069 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1071 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1072 "0x%p\n", current->comm,
1074 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
1075 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1076 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1077 if (rtn == FAST_IO_FAIL)
1078 scsi_eh_finish_cmd(scmd, done_q);
1080 list_move_tail(&scmd->eh_entry, &check_list);
1082 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1089 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1093 * scsi_eh_try_stu - Send START_UNIT to device.
1094 * @scmd: &scsi_cmnd to send START_UNIT
1097 * 0 - Device is ready. 1 - Device NOT ready.
1099 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1101 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1103 if (scmd->device->allow_restart) {
1104 int i, rtn = NEEDS_RETRY;
1106 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1107 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1117 * scsi_eh_stu - send START_UNIT if needed
1118 * @shost: &scsi host being recovered.
1119 * @work_q: &list_head for pending commands.
1120 * @done_q: &list_head for processed commands.
1123 * If commands are failing due to not ready, initializing command required,
1124 * try revalidating the device, which will end up sending a start unit.
1126 static int scsi_eh_stu(struct Scsi_Host *shost,
1127 struct list_head *work_q,
1128 struct list_head *done_q)
1130 struct scsi_cmnd *scmd, *stu_scmd, *next;
1131 struct scsi_device *sdev;
1133 shost_for_each_device(sdev, shost) {
1135 list_for_each_entry(scmd, work_q, eh_entry)
1136 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1137 scsi_check_sense(scmd) == FAILED ) {
1145 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1146 " 0x%p\n", current->comm, sdev));
1148 if (!scsi_eh_try_stu(stu_scmd)) {
1149 if (!scsi_device_online(sdev) ||
1150 !scsi_eh_tur(stu_scmd)) {
1151 list_for_each_entry_safe(scmd, next,
1153 if (scmd->device == sdev)
1154 scsi_eh_finish_cmd(scmd, done_q);
1158 SCSI_LOG_ERROR_RECOVERY(3,
1159 printk("%s: START_UNIT failed to sdev:"
1160 " 0x%p\n", current->comm, sdev));
1164 return list_empty(work_q);
1169 * scsi_eh_bus_device_reset - send bdr if needed
1170 * @shost: scsi host being recovered.
1171 * @work_q: &list_head for pending commands.
1172 * @done_q: &list_head for processed commands.
1175 * Try a bus device reset. Still, look to see whether we have multiple
1176 * devices that are jammed or not - if we have multiple devices, it
1177 * makes no sense to try bus_device_reset - we really would need to try
1178 * a bus_reset instead.
1180 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1181 struct list_head *work_q,
1182 struct list_head *done_q)
1184 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1185 struct scsi_device *sdev;
1188 shost_for_each_device(sdev, shost) {
1190 list_for_each_entry(scmd, work_q, eh_entry)
1191 if (scmd->device == sdev) {
1199 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1200 " 0x%p\n", current->comm,
1202 rtn = scsi_try_bus_device_reset(bdr_scmd);
1203 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1204 if (!scsi_device_online(sdev) ||
1205 rtn == FAST_IO_FAIL ||
1206 !scsi_eh_tur(bdr_scmd)) {
1207 list_for_each_entry_safe(scmd, next,
1209 if (scmd->device == sdev)
1210 scsi_eh_finish_cmd(scmd,
1215 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1223 return list_empty(work_q);
1227 * scsi_eh_target_reset - send target reset if needed
1228 * @shost: scsi host being recovered.
1229 * @work_q: &list_head for pending commands.
1230 * @done_q: &list_head for processed commands.
1233 * Try a target reset.
1235 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1236 struct list_head *work_q,
1237 struct list_head *done_q)
1239 LIST_HEAD(tmp_list);
1240 LIST_HEAD(check_list);
1242 list_splice_init(work_q, &tmp_list);
1244 while (!list_empty(&tmp_list)) {
1245 struct scsi_cmnd *next, *scmd;
1249 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1252 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1254 current->comm, id));
1255 rtn = scsi_try_target_reset(scmd);
1256 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1257 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1260 current->comm, id));
1261 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1262 if (scmd_id(scmd) != id)
1266 list_move_tail(&scmd->eh_entry, &check_list);
1267 else if (rtn == FAST_IO_FAIL)
1268 scsi_eh_finish_cmd(scmd, done_q);
1270 /* push back on work queue for further processing */
1271 list_move(&scmd->eh_entry, work_q);
1275 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1279 * scsi_eh_bus_reset - send a bus reset
1280 * @shost: &scsi host being recovered.
1281 * @work_q: &list_head for pending commands.
1282 * @done_q: &list_head for processed commands.
1284 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1285 struct list_head *work_q,
1286 struct list_head *done_q)
1288 struct scsi_cmnd *scmd, *chan_scmd, *next;
1289 LIST_HEAD(check_list);
1290 unsigned int channel;
1294 * we really want to loop over the various channels, and do this on
1295 * a channel by channel basis. we should also check to see if any
1296 * of the failed commands are on soft_reset devices, and if so, skip
1300 for (channel = 0; channel <= shost->max_channel; channel++) {
1302 list_for_each_entry(scmd, work_q, eh_entry) {
1303 if (channel == scmd_channel(scmd)) {
1307 * FIXME add back in some support for
1308 * soft_reset devices.
1315 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1316 " %d\n", current->comm,
1318 rtn = scsi_try_bus_reset(chan_scmd);
1319 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1320 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1321 if (channel == scmd_channel(scmd)) {
1322 if (rtn == FAST_IO_FAIL)
1323 scsi_eh_finish_cmd(scmd,
1326 list_move_tail(&scmd->eh_entry,
1331 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1332 " failed chan: %d\n",
1337 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1341 * scsi_eh_host_reset - send a host reset
1342 * @work_q: list_head for processed commands.
1343 * @done_q: list_head for processed commands.
1345 static int scsi_eh_host_reset(struct list_head *work_q,
1346 struct list_head *done_q)
1348 struct scsi_cmnd *scmd, *next;
1349 LIST_HEAD(check_list);
1352 if (!list_empty(work_q)) {
1353 scmd = list_entry(work_q->next,
1354 struct scsi_cmnd, eh_entry);
1356 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1359 rtn = scsi_try_host_reset(scmd);
1360 if (rtn == SUCCESS) {
1361 list_splice_init(work_q, &check_list);
1362 } else if (rtn == FAST_IO_FAIL) {
1363 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1364 scsi_eh_finish_cmd(scmd, done_q);
1367 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1372 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1376 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1377 * @work_q: list_head for processed commands.
1378 * @done_q: list_head for processed commands.
1380 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1381 struct list_head *done_q)
1383 struct scsi_cmnd *scmd, *next;
1385 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1386 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1387 "not ready after error recovery\n");
1388 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1389 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1391 * FIXME: Handle lost cmds.
1394 scsi_eh_finish_cmd(scmd, done_q);
1400 * scsi_noretry_cmd - determinte if command should be failed fast
1401 * @scmd: SCSI cmd to examine.
1403 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1405 switch (host_byte(scmd->result)) {
1409 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1411 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1413 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1414 status_byte(scmd->result) == RESERVATION_CONFLICT)
1417 case DID_SOFT_ERROR:
1418 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1421 switch (status_byte(scmd->result)) {
1422 case CHECK_CONDITION:
1424 * assume caller has checked sense and determinted
1425 * the check condition was retryable.
1427 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1428 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1436 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1437 * @scmd: SCSI cmd to examine.
1440 * This is *only* called when we are examining the status after sending
1441 * out the actual data command. any commands that are queued for error
1442 * recovery (e.g. test_unit_ready) do *not* come through here.
1444 * When this routine returns failed, it means the error handler thread
1445 * is woken. In cases where the error code indicates an error that
1446 * doesn't require the error handler read (i.e. we don't need to
1447 * abort/reset), this function should return SUCCESS.
1449 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1454 * if the device is offline, then we clearly just pass the result back
1455 * up to the top level.
1457 if (!scsi_device_online(scmd->device)) {
1458 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1465 * first check the host byte, to see if there is anything in there
1466 * that would indicate what we need to do.
1468 switch (host_byte(scmd->result)) {
1469 case DID_PASSTHROUGH:
1471 * no matter what, pass this through to the upper layer.
1472 * nuke this special code so that it looks like we are saying
1475 scmd->result &= 0xff00ffff;
1479 * looks good. drop through, and check the next byte.
1482 case DID_NO_CONNECT:
1483 case DID_BAD_TARGET:
1486 * note - this means that we just report the status back
1487 * to the top level driver, not that we actually think
1488 * that it indicates SUCCESS.
1492 * when the low level driver returns did_soft_error,
1493 * it is responsible for keeping an internal retry counter
1494 * in order to avoid endless loops (db)
1496 * actually this is a bug in this function here. we should
1497 * be mindful of the maximum number of retries specified
1498 * and not get stuck in a loop.
1500 case DID_SOFT_ERROR:
1506 return ADD_TO_MLQUEUE;
1507 case DID_TRANSPORT_DISRUPTED:
1509 * LLD/transport was disrupted during processing of the IO.
1510 * The transport class is now blocked/blocking,
1511 * and the transport will decide what to do with the IO
1512 * based on its timers and recovery capablilities if
1513 * there are enough retries.
1516 case DID_TRANSPORT_FAILFAST:
1518 * The transport decided to failfast the IO (most likely
1519 * the fast io fail tmo fired), so send IO directly upwards.
1523 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1524 status_byte(scmd->result) == RESERVATION_CONFLICT)
1526 * execute reservation conflict processing code
1536 * when we scan the bus, we get timeout messages for
1537 * these commands if there is no device available.
1538 * other hosts report did_no_connect for the same thing.
1540 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1541 scmd->cmnd[0] == INQUIRY)) {
1553 * next, check the message byte.
1555 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1559 * check the status byte to see if this indicates anything special.
1561 switch (status_byte(scmd->result)) {
1563 scsi_handle_queue_full(scmd->device);
1565 * the case of trying to send too many commands to a
1566 * tagged queueing device.
1570 * device can't talk to us at the moment. Should only
1571 * occur (SAM-3) when the task queue is empty, so will cause
1572 * the empty queue handling to trigger a stall in the
1575 return ADD_TO_MLQUEUE;
1577 scsi_handle_queue_ramp_up(scmd->device);
1578 case COMMAND_TERMINATED:
1582 case CHECK_CONDITION:
1583 rtn = scsi_check_sense(scmd);
1584 if (rtn == NEEDS_RETRY)
1586 /* if rtn == FAILED, we have no sense information;
1587 * returning FAILED will wake the error handler thread
1588 * to collect the sense and redo the decide
1591 case CONDITION_GOOD:
1592 case INTERMEDIATE_GOOD:
1593 case INTERMEDIATE_C_GOOD:
1596 * who knows? FIXME(eric)
1600 case RESERVATION_CONFLICT:
1601 sdev_printk(KERN_INFO, scmd->device,
1602 "reservation conflict\n");
1603 set_host_byte(scmd, DID_NEXUS_FAILURE);
1604 return SUCCESS; /* causes immediate i/o error */
1612 /* we requeue for retry because the error was retryable, and
1613 * the request was not marked fast fail. Note that above,
1614 * even if the request is marked fast fail, we still requeue
1615 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1616 if ((++scmd->retries) <= scmd->allowed
1617 && !scsi_noretry_cmd(scmd)) {
1621 * no more retries - report this one back to upper level.
1627 static void eh_lock_door_done(struct request *req, int uptodate)
1629 __blk_put_request(req->q, req);
1633 * scsi_eh_lock_door - Prevent medium removal for the specified device
1634 * @sdev: SCSI device to prevent medium removal
1637 * We must be called from process context.
1640 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1641 * head of the devices request queue, and continue.
1643 static void scsi_eh_lock_door(struct scsi_device *sdev)
1645 struct request *req;
1648 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1649 * request becomes available
1651 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1653 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1657 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1660 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1662 req->cmd_type = REQ_TYPE_BLOCK_PC;
1663 req->cmd_flags |= REQ_QUIET;
1664 req->timeout = 10 * HZ;
1667 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1671 * scsi_restart_operations - restart io operations to the specified host.
1672 * @shost: Host we are restarting.
1675 * When we entered the error handler, we blocked all further i/o to
1676 * this device. we need to 'reverse' this process.
1678 static void scsi_restart_operations(struct Scsi_Host *shost)
1680 struct scsi_device *sdev;
1681 unsigned long flags;
1684 * If the door was locked, we need to insert a door lock request
1685 * onto the head of the SCSI request queue for the device. There
1686 * is no point trying to lock the door of an off-line device.
1688 shost_for_each_device(sdev, shost) {
1689 if (scsi_device_online(sdev) && sdev->locked)
1690 scsi_eh_lock_door(sdev);
1694 * next free up anything directly waiting upon the host. this
1695 * will be requests for character device operations, and also for
1696 * ioctls to queued block devices.
1698 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1701 spin_lock_irqsave(shost->host_lock, flags);
1702 if (scsi_host_set_state(shost, SHOST_RUNNING))
1703 if (scsi_host_set_state(shost, SHOST_CANCEL))
1704 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1705 spin_unlock_irqrestore(shost->host_lock, flags);
1707 wake_up(&shost->host_wait);
1710 * finally we need to re-initiate requests that may be pending. we will
1711 * have had everything blocked while error handling is taking place, and
1712 * now that error recovery is done, we will need to ensure that these
1713 * requests are started.
1715 scsi_run_host_queues(shost);
1718 * if eh is active and host_eh_scheduled is pending we need to re-run
1719 * recovery. we do this check after scsi_run_host_queues() to allow
1720 * everything pent up since the last eh run a chance to make forward
1721 * progress before we sync again. Either we'll immediately re-run
1722 * recovery or scsi_device_unbusy() will wake us again when these
1723 * pending commands complete.
1725 spin_lock_irqsave(shost->host_lock, flags);
1726 if (shost->host_eh_scheduled)
1727 if (scsi_host_set_state(shost, SHOST_RECOVERY))
1728 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
1729 spin_unlock_irqrestore(shost->host_lock, flags);
1733 * scsi_eh_ready_devs - check device ready state and recover if not.
1734 * @shost: host to be recovered.
1735 * @work_q: &list_head for pending commands.
1736 * @done_q: &list_head for processed commands.
1738 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1739 struct list_head *work_q,
1740 struct list_head *done_q)
1742 if (!scsi_eh_stu(shost, work_q, done_q))
1743 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1744 if (!scsi_eh_target_reset(shost, work_q, done_q))
1745 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1746 if (!scsi_eh_host_reset(work_q, done_q))
1747 scsi_eh_offline_sdevs(work_q,
1750 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1753 * scsi_eh_flush_done_q - finish processed commands or retry them.
1754 * @done_q: list_head of processed commands.
1756 void scsi_eh_flush_done_q(struct list_head *done_q)
1758 struct scsi_cmnd *scmd, *next;
1760 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1761 list_del_init(&scmd->eh_entry);
1762 if (scsi_device_online(scmd->device) &&
1763 !scsi_noretry_cmd(scmd) &&
1764 (++scmd->retries <= scmd->allowed)) {
1765 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1769 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1772 * If just we got sense for the device (called
1773 * scsi_eh_get_sense), scmd->result is already
1774 * set, do not set DRIVER_TIMEOUT.
1777 scmd->result |= (DRIVER_TIMEOUT << 24);
1778 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1780 current->comm, scmd));
1781 scsi_finish_command(scmd);
1785 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1788 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1789 * @shost: Host to unjam.
1792 * When we come in here, we *know* that all commands on the bus have
1793 * either completed, failed or timed out. we also know that no further
1794 * commands are being sent to the host, so things are relatively quiet
1795 * and we have freedom to fiddle with things as we wish.
1797 * This is only the *default* implementation. it is possible for
1798 * individual drivers to supply their own version of this function, and
1799 * if the maintainer wishes to do this, it is strongly suggested that
1800 * this function be taken as a template and modified. this function
1801 * was designed to correctly handle problems for about 95% of the
1802 * different cases out there, and it should always provide at least a
1803 * reasonable amount of error recovery.
1805 * Any command marked 'failed' or 'timeout' must eventually have
1806 * scsi_finish_cmd() called for it. we do all of the retry stuff
1807 * here, so when we restart the host after we return it should have an
1810 static void scsi_unjam_host(struct Scsi_Host *shost)
1812 unsigned long flags;
1813 LIST_HEAD(eh_work_q);
1814 LIST_HEAD(eh_done_q);
1816 spin_lock_irqsave(shost->host_lock, flags);
1817 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1818 spin_unlock_irqrestore(shost->host_lock, flags);
1820 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1822 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1823 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1824 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1826 scsi_eh_flush_done_q(&eh_done_q);
1830 * scsi_error_handler - SCSI error handler thread
1831 * @data: Host for which we are running.
1834 * This is the main error handling loop. This is run as a kernel thread
1835 * for every SCSI host and handles all error handling activity.
1837 int scsi_error_handler(void *data)
1839 struct Scsi_Host *shost = data;
1842 * We use TASK_INTERRUPTIBLE so that the thread is not
1843 * counted against the load average as a running process.
1844 * We never actually get interrupted because kthread_run
1845 * disables signal delivery for the created thread.
1847 while (!kthread_should_stop()) {
1848 set_current_state(TASK_INTERRUPTIBLE);
1849 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1850 shost->host_failed != shost->host_busy) {
1851 SCSI_LOG_ERROR_RECOVERY(1,
1852 printk("Error handler scsi_eh_%d sleeping\n",
1858 __set_current_state(TASK_RUNNING);
1859 SCSI_LOG_ERROR_RECOVERY(1,
1860 printk("Error handler scsi_eh_%d waking up\n",
1864 * We have a host that is failing for some reason. Figure out
1865 * what we need to do to get it up and online again (if we can).
1866 * If we fail, we end up taking the thing offline.
1868 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
1869 SCSI_LOG_ERROR_RECOVERY(1,
1870 printk(KERN_ERR "Error handler scsi_eh_%d "
1871 "unable to autoresume\n",
1876 if (shost->transportt->eh_strategy_handler)
1877 shost->transportt->eh_strategy_handler(shost);
1879 scsi_unjam_host(shost);
1882 * Note - if the above fails completely, the action is to take
1883 * individual devices offline and flush the queue of any
1884 * outstanding requests that may have been pending. When we
1885 * restart, we restart any I/O to any other devices on the bus
1886 * which are still online.
1888 scsi_restart_operations(shost);
1889 if (!shost->eh_noresume)
1890 scsi_autopm_put_host(shost);
1892 __set_current_state(TASK_RUNNING);
1894 SCSI_LOG_ERROR_RECOVERY(1,
1895 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1896 shost->ehandler = NULL;
1901 * Function: scsi_report_bus_reset()
1903 * Purpose: Utility function used by low-level drivers to report that
1904 * they have observed a bus reset on the bus being handled.
1906 * Arguments: shost - Host in question
1907 * channel - channel on which reset was observed.
1911 * Lock status: Host lock must be held.
1913 * Notes: This only needs to be called if the reset is one which
1914 * originates from an unknown location. Resets originated
1915 * by the mid-level itself don't need to call this, but there
1916 * should be no harm.
1918 * The main purpose of this is to make sure that a CHECK_CONDITION
1919 * is properly treated.
1921 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1923 struct scsi_device *sdev;
1925 __shost_for_each_device(sdev, shost) {
1926 if (channel == sdev_channel(sdev))
1927 __scsi_report_device_reset(sdev, NULL);
1930 EXPORT_SYMBOL(scsi_report_bus_reset);
1933 * Function: scsi_report_device_reset()
1935 * Purpose: Utility function used by low-level drivers to report that
1936 * they have observed a device reset on the device being handled.
1938 * Arguments: shost - Host in question
1939 * channel - channel on which reset was observed
1940 * target - target on which reset was observed
1944 * Lock status: Host lock must be held
1946 * Notes: This only needs to be called if the reset is one which
1947 * originates from an unknown location. Resets originated
1948 * by the mid-level itself don't need to call this, but there
1949 * should be no harm.
1951 * The main purpose of this is to make sure that a CHECK_CONDITION
1952 * is properly treated.
1954 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1956 struct scsi_device *sdev;
1958 __shost_for_each_device(sdev, shost) {
1959 if (channel == sdev_channel(sdev) &&
1960 target == sdev_id(sdev))
1961 __scsi_report_device_reset(sdev, NULL);
1964 EXPORT_SYMBOL(scsi_report_device_reset);
1967 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1972 * Function: scsi_reset_provider
1974 * Purpose: Send requested reset to a bus or device at any phase.
1976 * Arguments: device - device to send reset to
1977 * flag - reset type (see scsi.h)
1979 * Returns: SUCCESS/FAILURE.
1981 * Notes: This is used by the SCSI Generic driver to provide
1982 * Bus/Device reset capability.
1985 scsi_reset_provider(struct scsi_device *dev, int flag)
1987 struct scsi_cmnd *scmd;
1988 struct Scsi_Host *shost = dev->host;
1990 unsigned long flags;
1993 if (scsi_autopm_get_host(shost) < 0)
1996 scmd = scsi_get_command(dev, GFP_KERNEL);
1997 blk_rq_init(NULL, &req);
1998 scmd->request = &req;
2000 scmd->cmnd = req.cmd;
2002 scmd->scsi_done = scsi_reset_provider_done_command;
2003 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2007 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2009 spin_lock_irqsave(shost->host_lock, flags);
2010 shost->tmf_in_progress = 1;
2011 spin_unlock_irqrestore(shost->host_lock, flags);
2014 case SCSI_TRY_RESET_DEVICE:
2015 rtn = scsi_try_bus_device_reset(scmd);
2019 case SCSI_TRY_RESET_TARGET:
2020 rtn = scsi_try_target_reset(scmd);
2024 case SCSI_TRY_RESET_BUS:
2025 rtn = scsi_try_bus_reset(scmd);
2029 case SCSI_TRY_RESET_HOST:
2030 rtn = scsi_try_host_reset(scmd);
2036 spin_lock_irqsave(shost->host_lock, flags);
2037 shost->tmf_in_progress = 0;
2038 spin_unlock_irqrestore(shost->host_lock, flags);
2041 * be sure to wake up anyone who was sleeping or had their queue
2042 * suspended while we performed the TMF.
2044 SCSI_LOG_ERROR_RECOVERY(3,
2045 printk("%s: waking up host to restart after TMF\n",
2048 wake_up(&shost->host_wait);
2050 scsi_run_host_queues(shost);
2052 scsi_next_command(scmd);
2053 scsi_autopm_put_host(shost);
2056 EXPORT_SYMBOL(scsi_reset_provider);
2059 * scsi_normalize_sense - normalize main elements from either fixed or
2060 * descriptor sense data format into a common format.
2062 * @sense_buffer: byte array containing sense data returned by device
2063 * @sb_len: number of valid bytes in sense_buffer
2064 * @sshdr: pointer to instance of structure that common
2065 * elements are written to.
2068 * The "main elements" from sense data are: response_code, sense_key,
2069 * asc, ascq and additional_length (only for descriptor format).
2071 * Typically this function can be called after a device has
2072 * responded to a SCSI command with the CHECK_CONDITION status.
2075 * 1 if valid sense data information found, else 0;
2077 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2078 struct scsi_sense_hdr *sshdr)
2080 if (!sense_buffer || !sb_len)
2083 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2085 sshdr->response_code = (sense_buffer[0] & 0x7f);
2087 if (!scsi_sense_valid(sshdr))
2090 if (sshdr->response_code >= 0x72) {
2095 sshdr->sense_key = (sense_buffer[1] & 0xf);
2097 sshdr->asc = sense_buffer[2];
2099 sshdr->ascq = sense_buffer[3];
2101 sshdr->additional_length = sense_buffer[7];
2107 sshdr->sense_key = (sense_buffer[2] & 0xf);
2109 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2110 sb_len : (sense_buffer[7] + 8);
2112 sshdr->asc = sense_buffer[12];
2114 sshdr->ascq = sense_buffer[13];
2120 EXPORT_SYMBOL(scsi_normalize_sense);
2122 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2123 struct scsi_sense_hdr *sshdr)
2125 return scsi_normalize_sense(cmd->sense_buffer,
2126 SCSI_SENSE_BUFFERSIZE, sshdr);
2128 EXPORT_SYMBOL(scsi_command_normalize_sense);
2131 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2132 * @sense_buffer: byte array of descriptor format sense data
2133 * @sb_len: number of valid bytes in sense_buffer
2134 * @desc_type: value of descriptor type to find
2135 * (e.g. 0 -> information)
2138 * only valid when sense data is in descriptor format
2141 * pointer to start of (first) descriptor if found else NULL
2143 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2146 int add_sen_len, add_len, desc_len, k;
2149 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2151 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2153 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2154 add_sen_len : (sb_len - 8);
2155 descp = &sense_buffer[8];
2156 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2158 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2159 desc_len = add_len + 2;
2160 if (descp[0] == desc_type)
2162 if (add_len < 0) // short descriptor ??
2167 EXPORT_SYMBOL(scsi_sense_desc_find);
2170 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2171 * @sense_buffer: byte array of sense data
2172 * @sb_len: number of valid bytes in sense_buffer
2173 * @info_out: pointer to 64 integer where 8 or 4 byte information
2174 * field will be placed if found.
2177 * 1 if information field found, 0 if not found.
2179 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2188 switch (sense_buffer[0] & 0x7f) {
2191 if (sense_buffer[0] & 0x80) {
2192 *info_out = (sense_buffer[3] << 24) +
2193 (sense_buffer[4] << 16) +
2194 (sense_buffer[5] << 8) + sense_buffer[6];
2200 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2202 if (ucp && (0xa == ucp[1])) {
2204 for (j = 0; j < 8; ++j) {
2217 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2220 * scsi_build_sense_buffer - build sense data in a buffer
2221 * @desc: Sense format (non zero == descriptor format,
2222 * 0 == fixed format)
2223 * @buf: Where to build sense data
2225 * @asc: Additional sense code
2226 * @ascq: Additional sense code qualifier
2229 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2232 buf[0] = 0x72; /* descriptor, current */
2238 buf[0] = 0x70; /* fixed, current */
2245 EXPORT_SYMBOL(scsi_build_sense_buffer);