mmc: tmio: fix regression in TMIO_MMC_WRPROTECT_DISABLE handling
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / scsi / scsi_error.c
blob633c2395a92a2c918cfd068d541f99063f9470f2
1 /*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13 * minor cleanups.
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
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_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
42 #include <trace/events/scsi.h>
44 #define SENSE_TIMEOUT (10*HZ)
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
50 #define BUS_RESET_SETTLE_TIME (10)
51 #define HOST_RESET_SETTLE_TIME (10)
53 /* called with shost->host_lock held */
54 void scsi_eh_wakeup(struct Scsi_Host *shost)
56 if (shost->host_busy == shost->host_failed) {
57 trace_scsi_eh_wakeup(shost);
58 wake_up_process(shost->ehandler);
59 SCSI_LOG_ERROR_RECOVERY(5,
60 printk("Waking error handler thread\n"));
64 /**
65 * scsi_schedule_eh - schedule EH for SCSI host
66 * @shost: SCSI host to invoke error handling on.
68 * Schedule SCSI EH without scmd.
70 void scsi_schedule_eh(struct Scsi_Host *shost)
72 unsigned long flags;
74 spin_lock_irqsave(shost->host_lock, flags);
76 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
77 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
78 shost->host_eh_scheduled++;
79 scsi_eh_wakeup(shost);
82 spin_unlock_irqrestore(shost->host_lock, flags);
84 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
86 /**
87 * scsi_eh_scmd_add - add scsi cmd to error handling.
88 * @scmd: scmd to run eh on.
89 * @eh_flag: optional SCSI_EH flag.
91 * Return value:
92 * 0 on failure.
94 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
96 struct Scsi_Host *shost = scmd->device->host;
97 unsigned long flags;
98 int ret = 0;
100 if (!shost->ehandler)
101 return 0;
103 spin_lock_irqsave(shost->host_lock, flags);
104 if (scsi_host_set_state(shost, SHOST_RECOVERY))
105 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
106 goto out_unlock;
108 ret = 1;
109 scmd->eh_eflags |= eh_flag;
110 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
111 shost->host_failed++;
112 scsi_eh_wakeup(shost);
113 out_unlock:
114 spin_unlock_irqrestore(shost->host_lock, flags);
115 return ret;
119 * scsi_times_out - Timeout function for normal scsi commands.
120 * @req: request that is timing out.
122 * Notes:
123 * We do not need to lock this. There is the potential for a race
124 * only in that the normal completion handling might run, but if the
125 * normal completion function determines that the timer has already
126 * fired, then it mustn't do anything.
128 enum blk_eh_timer_return scsi_times_out(struct request *req)
130 struct scsi_cmnd *scmd = req->special;
131 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
132 struct Scsi_Host *host = scmd->device->host;
134 trace_scsi_dispatch_cmd_timeout(scmd);
135 scsi_log_completion(scmd, TIMEOUT_ERROR);
137 if (host->transportt->eh_timed_out)
138 rtn = host->transportt->eh_timed_out(scmd);
139 else if (host->hostt->eh_timed_out)
140 rtn = host->hostt->eh_timed_out(scmd);
142 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
143 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
144 scmd->result |= DID_TIME_OUT << 16;
145 rtn = BLK_EH_HANDLED;
148 return rtn;
152 * scsi_block_when_processing_errors - Prevent cmds from being queued.
153 * @sdev: Device on which we are performing recovery.
155 * Description:
156 * We block until the host is out of error recovery, and then check to
157 * see whether the host or the device is offline.
159 * Return value:
160 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
162 int scsi_block_when_processing_errors(struct scsi_device *sdev)
164 int online;
166 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
168 online = scsi_device_online(sdev);
170 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
171 online));
173 return online;
175 EXPORT_SYMBOL(scsi_block_when_processing_errors);
177 #ifdef CONFIG_SCSI_LOGGING
179 * scsi_eh_prt_fail_stats - Log info on failures.
180 * @shost: scsi host being recovered.
181 * @work_q: Queue of scsi cmds to process.
183 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
184 struct list_head *work_q)
186 struct scsi_cmnd *scmd;
187 struct scsi_device *sdev;
188 int total_failures = 0;
189 int cmd_failed = 0;
190 int cmd_cancel = 0;
191 int devices_failed = 0;
193 shost_for_each_device(sdev, shost) {
194 list_for_each_entry(scmd, work_q, eh_entry) {
195 if (scmd->device == sdev) {
196 ++total_failures;
197 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
198 ++cmd_cancel;
199 else
200 ++cmd_failed;
204 if (cmd_cancel || cmd_failed) {
205 SCSI_LOG_ERROR_RECOVERY(3,
206 sdev_printk(KERN_INFO, sdev,
207 "%s: cmds failed: %d, cancel: %d\n",
208 __func__, cmd_failed,
209 cmd_cancel));
210 cmd_cancel = 0;
211 cmd_failed = 0;
212 ++devices_failed;
216 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
217 " devices require eh work\n",
218 total_failures, devices_failed));
220 #endif
223 * scsi_check_sense - Examine scsi cmd sense
224 * @scmd: Cmd to have sense checked.
226 * Return value:
227 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
229 * Notes:
230 * When a deferred error is detected the current command has
231 * not been executed and needs retrying.
233 static int scsi_check_sense(struct scsi_cmnd *scmd)
235 struct scsi_device *sdev = scmd->device;
236 struct scsi_sense_hdr sshdr;
238 if (! scsi_command_normalize_sense(scmd, &sshdr))
239 return FAILED; /* no valid sense data */
241 if (scsi_sense_is_deferred(&sshdr))
242 return NEEDS_RETRY;
244 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
245 sdev->scsi_dh_data->scsi_dh->check_sense) {
246 int rc;
248 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
249 if (rc != SCSI_RETURN_NOT_HANDLED)
250 return rc;
251 /* handler does not care. Drop down to default handling */
255 * Previous logic looked for FILEMARK, EOM or ILI which are
256 * mainly associated with tapes and returned SUCCESS.
258 if (sshdr.response_code == 0x70) {
259 /* fixed format */
260 if (scmd->sense_buffer[2] & 0xe0)
261 return SUCCESS;
262 } else {
264 * descriptor format: look for "stream commands sense data
265 * descriptor" (see SSC-3). Assume single sense data
266 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
268 if ((sshdr.additional_length > 3) &&
269 (scmd->sense_buffer[8] == 0x4) &&
270 (scmd->sense_buffer[11] & 0xe0))
271 return SUCCESS;
274 switch (sshdr.sense_key) {
275 case NO_SENSE:
276 return SUCCESS;
277 case RECOVERED_ERROR:
278 return /* soft_error */ SUCCESS;
280 case ABORTED_COMMAND:
281 if (sshdr.asc == 0x10) /* DIF */
282 return SUCCESS;
284 return NEEDS_RETRY;
285 case NOT_READY:
286 case UNIT_ATTENTION:
288 * if we are expecting a cc/ua because of a bus reset that we
289 * performed, treat this just as a retry. otherwise this is
290 * information that we should pass up to the upper-level driver
291 * so that we can deal with it there.
293 if (scmd->device->expecting_cc_ua) {
294 scmd->device->expecting_cc_ua = 0;
295 return NEEDS_RETRY;
298 * if the device is in the process of becoming ready, we
299 * should retry.
301 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
302 return NEEDS_RETRY;
304 * if the device is not started, we need to wake
305 * the error handler to start the motor
307 if (scmd->device->allow_restart &&
308 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
309 return FAILED;
311 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
312 scmd_printk(KERN_WARNING, scmd,
313 "Warning! Received an indication that the "
314 "LUN assignments on this target have "
315 "changed. The Linux SCSI layer does not "
316 "automatically remap LUN assignments.\n");
317 else if (sshdr.asc == 0x3f)
318 scmd_printk(KERN_WARNING, scmd,
319 "Warning! Received an indication that the "
320 "operating parameters on this target have "
321 "changed. The Linux SCSI layer does not "
322 "automatically adjust these parameters.\n");
325 * Pass the UA upwards for a determination in the completion
326 * functions.
328 return SUCCESS;
330 /* these are not supported */
331 case COPY_ABORTED:
332 case VOLUME_OVERFLOW:
333 case MISCOMPARE:
334 case BLANK_CHECK:
335 case DATA_PROTECT:
336 return TARGET_ERROR;
338 case MEDIUM_ERROR:
339 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
340 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
341 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
342 return TARGET_ERROR;
344 return NEEDS_RETRY;
346 case HARDWARE_ERROR:
347 if (scmd->device->retry_hwerror)
348 return ADD_TO_MLQUEUE;
349 else
350 return TARGET_ERROR;
352 case ILLEGAL_REQUEST:
353 default:
354 return SUCCESS;
358 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
360 struct scsi_host_template *sht = sdev->host->hostt;
361 struct scsi_device *tmp_sdev;
363 if (!sht->change_queue_depth ||
364 sdev->queue_depth >= sdev->max_queue_depth)
365 return;
367 if (time_before(jiffies,
368 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
369 return;
371 if (time_before(jiffies,
372 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
373 return;
376 * Walk all devices of a target and do
377 * ramp up on them.
379 shost_for_each_device(tmp_sdev, sdev->host) {
380 if (tmp_sdev->channel != sdev->channel ||
381 tmp_sdev->id != sdev->id ||
382 tmp_sdev->queue_depth == sdev->max_queue_depth)
383 continue;
385 * call back into LLD to increase queue_depth by one
386 * with ramp up reason code.
388 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
389 SCSI_QDEPTH_RAMP_UP);
390 sdev->last_queue_ramp_up = jiffies;
394 static void scsi_handle_queue_full(struct scsi_device *sdev)
396 struct scsi_host_template *sht = sdev->host->hostt;
397 struct scsi_device *tmp_sdev;
399 if (!sht->change_queue_depth)
400 return;
402 shost_for_each_device(tmp_sdev, sdev->host) {
403 if (tmp_sdev->channel != sdev->channel ||
404 tmp_sdev->id != sdev->id)
405 continue;
407 * We do not know the number of commands that were at
408 * the device when we got the queue full so we start
409 * from the highest possible value and work our way down.
411 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
412 SCSI_QDEPTH_QFULL);
417 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
418 * @scmd: SCSI cmd to examine.
420 * Notes:
421 * This is *only* called when we are examining the status of commands
422 * queued during error recovery. the main difference here is that we
423 * don't allow for the possibility of retries here, and we are a lot
424 * more restrictive about what we consider acceptable.
426 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
429 * first check the host byte, to see if there is anything in there
430 * that would indicate what we need to do.
432 if (host_byte(scmd->result) == DID_RESET) {
434 * rats. we are already in the error handler, so we now
435 * get to try and figure out what to do next. if the sense
436 * is valid, we have a pretty good idea of what to do.
437 * if not, we mark it as FAILED.
439 return scsi_check_sense(scmd);
441 if (host_byte(scmd->result) != DID_OK)
442 return FAILED;
445 * next, check the message byte.
447 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
448 return FAILED;
451 * now, check the status byte to see if this indicates
452 * anything special.
454 switch (status_byte(scmd->result)) {
455 case GOOD:
456 scsi_handle_queue_ramp_up(scmd->device);
457 case COMMAND_TERMINATED:
458 return SUCCESS;
459 case CHECK_CONDITION:
460 return scsi_check_sense(scmd);
461 case CONDITION_GOOD:
462 case INTERMEDIATE_GOOD:
463 case INTERMEDIATE_C_GOOD:
465 * who knows? FIXME(eric)
467 return SUCCESS;
468 case RESERVATION_CONFLICT:
469 if (scmd->cmnd[0] == TEST_UNIT_READY)
470 /* it is a success, we probed the device and
471 * found it */
472 return SUCCESS;
473 /* otherwise, we failed to send the command */
474 return FAILED;
475 case QUEUE_FULL:
476 scsi_handle_queue_full(scmd->device);
477 /* fall through */
478 case BUSY:
479 return NEEDS_RETRY;
480 default:
481 return FAILED;
483 return FAILED;
487 * scsi_eh_done - Completion function for error handling.
488 * @scmd: Cmd that is done.
490 static void scsi_eh_done(struct scsi_cmnd *scmd)
492 struct completion *eh_action;
494 SCSI_LOG_ERROR_RECOVERY(3,
495 printk("%s scmd: %p result: %x\n",
496 __func__, scmd, scmd->result));
498 eh_action = scmd->device->host->eh_action;
499 if (eh_action)
500 complete(eh_action);
504 * scsi_try_host_reset - ask host adapter to reset itself
505 * @scmd: SCSI cmd to send hsot reset.
507 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
509 unsigned long flags;
510 int rtn;
511 struct Scsi_Host *host = scmd->device->host;
512 struct scsi_host_template *hostt = host->hostt;
514 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
515 __func__));
517 if (!hostt->eh_host_reset_handler)
518 return FAILED;
520 rtn = hostt->eh_host_reset_handler(scmd);
522 if (rtn == SUCCESS) {
523 if (!hostt->skip_settle_delay)
524 ssleep(HOST_RESET_SETTLE_TIME);
525 spin_lock_irqsave(host->host_lock, flags);
526 scsi_report_bus_reset(host, scmd_channel(scmd));
527 spin_unlock_irqrestore(host->host_lock, flags);
530 return rtn;
534 * scsi_try_bus_reset - ask host to perform a bus reset
535 * @scmd: SCSI cmd to send bus reset.
537 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
539 unsigned long flags;
540 int rtn;
541 struct Scsi_Host *host = scmd->device->host;
542 struct scsi_host_template *hostt = host->hostt;
544 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
545 __func__));
547 if (!hostt->eh_bus_reset_handler)
548 return FAILED;
550 rtn = hostt->eh_bus_reset_handler(scmd);
552 if (rtn == SUCCESS) {
553 if (!hostt->skip_settle_delay)
554 ssleep(BUS_RESET_SETTLE_TIME);
555 spin_lock_irqsave(host->host_lock, flags);
556 scsi_report_bus_reset(host, scmd_channel(scmd));
557 spin_unlock_irqrestore(host->host_lock, flags);
560 return rtn;
563 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
565 sdev->was_reset = 1;
566 sdev->expecting_cc_ua = 1;
570 * scsi_try_target_reset - Ask host to perform a target reset
571 * @scmd: SCSI cmd used to send a target reset
573 * Notes:
574 * There is no timeout for this operation. if this operation is
575 * unreliable for a given host, then the host itself needs to put a
576 * timer on it, and set the host back to a consistent state prior to
577 * returning.
579 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
581 unsigned long flags;
582 int rtn;
583 struct Scsi_Host *host = scmd->device->host;
584 struct scsi_host_template *hostt = host->hostt;
586 if (!hostt->eh_target_reset_handler)
587 return FAILED;
589 rtn = hostt->eh_target_reset_handler(scmd);
590 if (rtn == SUCCESS) {
591 spin_lock_irqsave(host->host_lock, flags);
592 __starget_for_each_device(scsi_target(scmd->device), NULL,
593 __scsi_report_device_reset);
594 spin_unlock_irqrestore(host->host_lock, flags);
597 return rtn;
601 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
602 * @scmd: SCSI cmd used to send BDR
604 * Notes:
605 * There is no timeout for this operation. if this operation is
606 * unreliable for a given host, then the host itself needs to put a
607 * timer on it, and set the host back to a consistent state prior to
608 * returning.
610 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
612 int rtn;
613 struct scsi_host_template *hostt = scmd->device->host->hostt;
615 if (!hostt->eh_device_reset_handler)
616 return FAILED;
618 rtn = hostt->eh_device_reset_handler(scmd);
619 if (rtn == SUCCESS)
620 __scsi_report_device_reset(scmd->device, NULL);
621 return rtn;
624 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
626 if (!hostt->eh_abort_handler)
627 return FAILED;
629 return hostt->eh_abort_handler(scmd);
632 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
634 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
635 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
636 if (scsi_try_target_reset(scmd) != SUCCESS)
637 if (scsi_try_bus_reset(scmd) != SUCCESS)
638 scsi_try_host_reset(scmd);
642 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
643 * @scmd: SCSI command structure to hijack
644 * @ses: structure to save restore information
645 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
646 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
647 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
649 * This function is used to save a scsi command information before re-execution
650 * as part of the error recovery process. If @sense_bytes is 0 the command
651 * sent must be one that does not transfer any data. If @sense_bytes != 0
652 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
653 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
655 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
656 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
658 struct scsi_device *sdev = scmd->device;
661 * We need saved copies of a number of fields - this is because
662 * error handling may need to overwrite these with different values
663 * to run different commands, and once error handling is complete,
664 * we will need to restore these values prior to running the actual
665 * command.
667 ses->cmd_len = scmd->cmd_len;
668 ses->cmnd = scmd->cmnd;
669 ses->data_direction = scmd->sc_data_direction;
670 ses->sdb = scmd->sdb;
671 ses->next_rq = scmd->request->next_rq;
672 ses->result = scmd->result;
673 ses->underflow = scmd->underflow;
674 ses->prot_op = scmd->prot_op;
676 scmd->prot_op = SCSI_PROT_NORMAL;
677 scmd->cmnd = ses->eh_cmnd;
678 memset(scmd->cmnd, 0, BLK_MAX_CDB);
679 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
680 scmd->request->next_rq = NULL;
682 if (sense_bytes) {
683 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
684 sense_bytes);
685 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
686 scmd->sdb.length);
687 scmd->sdb.table.sgl = &ses->sense_sgl;
688 scmd->sc_data_direction = DMA_FROM_DEVICE;
689 scmd->sdb.table.nents = 1;
690 scmd->cmnd[0] = REQUEST_SENSE;
691 scmd->cmnd[4] = scmd->sdb.length;
692 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
693 } else {
694 scmd->sc_data_direction = DMA_NONE;
695 if (cmnd) {
696 BUG_ON(cmnd_size > BLK_MAX_CDB);
697 memcpy(scmd->cmnd, cmnd, cmnd_size);
698 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
702 scmd->underflow = 0;
704 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
705 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
706 (sdev->lun << 5 & 0xe0);
709 * Zero the sense buffer. The scsi spec mandates that any
710 * untransferred sense data should be interpreted as being zero.
712 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
714 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
717 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
718 * @scmd: SCSI command structure to restore
719 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
721 * Undo any damage done by above scsi_eh_prep_cmnd().
723 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
726 * Restore original data
728 scmd->cmd_len = ses->cmd_len;
729 scmd->cmnd = ses->cmnd;
730 scmd->sc_data_direction = ses->data_direction;
731 scmd->sdb = ses->sdb;
732 scmd->request->next_rq = ses->next_rq;
733 scmd->result = ses->result;
734 scmd->underflow = ses->underflow;
735 scmd->prot_op = ses->prot_op;
737 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
740 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
741 * @scmd: SCSI command structure to hijack
742 * @cmnd: CDB to send
743 * @cmnd_size: size in bytes of @cmnd
744 * @timeout: timeout for this request
745 * @sense_bytes: size of sense data to copy or 0
747 * This function is used to send a scsi command down to a target device
748 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
750 * Return value:
751 * SUCCESS or FAILED or NEEDS_RETRY
753 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
754 int cmnd_size, int timeout, unsigned sense_bytes)
756 struct scsi_device *sdev = scmd->device;
757 struct Scsi_Host *shost = sdev->host;
758 DECLARE_COMPLETION_ONSTACK(done);
759 unsigned long timeleft;
760 struct scsi_eh_save ses;
761 int rtn;
763 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
764 shost->eh_action = &done;
766 scsi_log_send(scmd);
767 scmd->scsi_done = scsi_eh_done;
768 shost->hostt->queuecommand(shost, scmd);
770 timeleft = wait_for_completion_timeout(&done, timeout);
772 shost->eh_action = NULL;
774 scsi_log_completion(scmd, SUCCESS);
776 SCSI_LOG_ERROR_RECOVERY(3,
777 printk("%s: scmd: %p, timeleft: %ld\n",
778 __func__, scmd, timeleft));
781 * If there is time left scsi_eh_done got called, and we will
782 * examine the actual status codes to see whether the command
783 * actually did complete normally, else tell the host to forget
784 * about this command.
786 if (timeleft) {
787 rtn = scsi_eh_completed_normally(scmd);
788 SCSI_LOG_ERROR_RECOVERY(3,
789 printk("%s: scsi_eh_completed_normally %x\n",
790 __func__, rtn));
792 switch (rtn) {
793 case SUCCESS:
794 case NEEDS_RETRY:
795 case FAILED:
796 case TARGET_ERROR:
797 break;
798 case ADD_TO_MLQUEUE:
799 rtn = NEEDS_RETRY;
800 break;
801 default:
802 rtn = FAILED;
803 break;
805 } else {
806 scsi_abort_eh_cmnd(scmd);
807 rtn = FAILED;
810 scsi_eh_restore_cmnd(scmd, &ses);
811 return rtn;
815 * scsi_request_sense - Request sense data from a particular target.
816 * @scmd: SCSI cmd for request sense.
818 * Notes:
819 * Some hosts automatically obtain this information, others require
820 * that we obtain it on our own. This function will *not* return until
821 * the command either times out, or it completes.
823 static int scsi_request_sense(struct scsi_cmnd *scmd)
825 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
829 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
830 * @scmd: Original SCSI cmd that eh has finished.
831 * @done_q: Queue for processed commands.
833 * Notes:
834 * We don't want to use the normal command completion while we are are
835 * still handling errors - it may cause other commands to be queued,
836 * and that would disturb what we are doing. Thus we really want to
837 * keep a list of pending commands for final completion, and once we
838 * are ready to leave error handling we handle completion for real.
840 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
842 scmd->device->host->host_failed--;
843 scmd->eh_eflags = 0;
844 list_move_tail(&scmd->eh_entry, done_q);
846 EXPORT_SYMBOL(scsi_eh_finish_cmd);
849 * scsi_eh_get_sense - Get device sense data.
850 * @work_q: Queue of commands to process.
851 * @done_q: Queue of processed commands.
853 * Description:
854 * See if we need to request sense information. if so, then get it
855 * now, so we have a better idea of what to do.
857 * Notes:
858 * This has the unfortunate side effect that if a shost adapter does
859 * not automatically request sense information, we end up shutting
860 * it down before we request it.
862 * All drivers should request sense information internally these days,
863 * so for now all I have to say is tough noogies if you end up in here.
865 * XXX: Long term this code should go away, but that needs an audit of
866 * all LLDDs first.
868 int scsi_eh_get_sense(struct list_head *work_q,
869 struct list_head *done_q)
871 struct scsi_cmnd *scmd, *next;
872 int rtn;
874 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
875 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
876 SCSI_SENSE_VALID(scmd))
877 continue;
879 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
880 "%s: requesting sense\n",
881 current->comm));
882 rtn = scsi_request_sense(scmd);
883 if (rtn != SUCCESS)
884 continue;
886 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
887 " result %x\n", scmd,
888 scmd->result));
889 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
891 rtn = scsi_decide_disposition(scmd);
894 * if the result was normal, then just pass it along to the
895 * upper level.
897 if (rtn == SUCCESS)
898 /* we don't want this command reissued, just
899 * finished with the sense data, so set
900 * retries to the max allowed to ensure it
901 * won't get reissued */
902 scmd->retries = scmd->allowed;
903 else if (rtn != NEEDS_RETRY)
904 continue;
906 scsi_eh_finish_cmd(scmd, done_q);
909 return list_empty(work_q);
911 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
914 * scsi_eh_tur - Send TUR to device.
915 * @scmd: &scsi_cmnd to send TUR
917 * Return value:
918 * 0 - Device is ready. 1 - Device NOT ready.
920 static int scsi_eh_tur(struct scsi_cmnd *scmd)
922 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
923 int retry_cnt = 1, rtn;
925 retry_tur:
926 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
928 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
929 __func__, scmd, rtn));
931 switch (rtn) {
932 case NEEDS_RETRY:
933 if (retry_cnt--)
934 goto retry_tur;
935 /*FALLTHRU*/
936 case SUCCESS:
937 return 0;
938 default:
939 return 1;
944 * scsi_eh_abort_cmds - abort pending commands.
945 * @work_q: &list_head for pending commands.
946 * @done_q: &list_head for processed commands.
948 * Decription:
949 * Try and see whether or not it makes sense to try and abort the
950 * running command. This only works out to be the case if we have one
951 * command that has timed out. If the command simply failed, it makes
952 * no sense to try and abort the command, since as far as the shost
953 * adapter is concerned, it isn't running.
955 static int scsi_eh_abort_cmds(struct list_head *work_q,
956 struct list_head *done_q)
958 struct scsi_cmnd *scmd, *next;
959 int rtn;
961 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
962 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
963 continue;
964 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
965 "0x%p\n", current->comm,
966 scmd));
967 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
968 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
969 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
970 if (!scsi_device_online(scmd->device) ||
971 rtn == FAST_IO_FAIL ||
972 !scsi_eh_tur(scmd)) {
973 scsi_eh_finish_cmd(scmd, done_q);
975 } else
976 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
977 " cmd failed:"
978 "0x%p\n",
979 current->comm,
980 scmd));
983 return list_empty(work_q);
987 * scsi_eh_try_stu - Send START_UNIT to device.
988 * @scmd: &scsi_cmnd to send START_UNIT
990 * Return value:
991 * 0 - Device is ready. 1 - Device NOT ready.
993 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
995 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
997 if (scmd->device->allow_restart) {
998 int i, rtn = NEEDS_RETRY;
1000 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1001 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1003 if (rtn == SUCCESS)
1004 return 0;
1007 return 1;
1011 * scsi_eh_stu - send START_UNIT if needed
1012 * @shost: &scsi host being recovered.
1013 * @work_q: &list_head for pending commands.
1014 * @done_q: &list_head for processed commands.
1016 * Notes:
1017 * If commands are failing due to not ready, initializing command required,
1018 * try revalidating the device, which will end up sending a start unit.
1020 static int scsi_eh_stu(struct Scsi_Host *shost,
1021 struct list_head *work_q,
1022 struct list_head *done_q)
1024 struct scsi_cmnd *scmd, *stu_scmd, *next;
1025 struct scsi_device *sdev;
1027 shost_for_each_device(sdev, shost) {
1028 stu_scmd = NULL;
1029 list_for_each_entry(scmd, work_q, eh_entry)
1030 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1031 scsi_check_sense(scmd) == FAILED ) {
1032 stu_scmd = scmd;
1033 break;
1036 if (!stu_scmd)
1037 continue;
1039 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1040 " 0x%p\n", current->comm, sdev));
1042 if (!scsi_eh_try_stu(stu_scmd)) {
1043 if (!scsi_device_online(sdev) ||
1044 !scsi_eh_tur(stu_scmd)) {
1045 list_for_each_entry_safe(scmd, next,
1046 work_q, eh_entry) {
1047 if (scmd->device == sdev)
1048 scsi_eh_finish_cmd(scmd, done_q);
1051 } else {
1052 SCSI_LOG_ERROR_RECOVERY(3,
1053 printk("%s: START_UNIT failed to sdev:"
1054 " 0x%p\n", current->comm, sdev));
1058 return list_empty(work_q);
1063 * scsi_eh_bus_device_reset - send bdr if needed
1064 * @shost: scsi host being recovered.
1065 * @work_q: &list_head for pending commands.
1066 * @done_q: &list_head for processed commands.
1068 * Notes:
1069 * Try a bus device reset. Still, look to see whether we have multiple
1070 * devices that are jammed or not - if we have multiple devices, it
1071 * makes no sense to try bus_device_reset - we really would need to try
1072 * a bus_reset instead.
1074 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1075 struct list_head *work_q,
1076 struct list_head *done_q)
1078 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1079 struct scsi_device *sdev;
1080 int rtn;
1082 shost_for_each_device(sdev, shost) {
1083 bdr_scmd = NULL;
1084 list_for_each_entry(scmd, work_q, eh_entry)
1085 if (scmd->device == sdev) {
1086 bdr_scmd = scmd;
1087 break;
1090 if (!bdr_scmd)
1091 continue;
1093 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1094 " 0x%p\n", current->comm,
1095 sdev));
1096 rtn = scsi_try_bus_device_reset(bdr_scmd);
1097 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1098 if (!scsi_device_online(sdev) ||
1099 rtn == FAST_IO_FAIL ||
1100 !scsi_eh_tur(bdr_scmd)) {
1101 list_for_each_entry_safe(scmd, next,
1102 work_q, eh_entry) {
1103 if (scmd->device == sdev)
1104 scsi_eh_finish_cmd(scmd,
1105 done_q);
1108 } else {
1109 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1110 " failed sdev:"
1111 "0x%p\n",
1112 current->comm,
1113 sdev));
1117 return list_empty(work_q);
1121 * scsi_eh_target_reset - send target reset if needed
1122 * @shost: scsi host being recovered.
1123 * @work_q: &list_head for pending commands.
1124 * @done_q: &list_head for processed commands.
1126 * Notes:
1127 * Try a target reset.
1129 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1130 struct list_head *work_q,
1131 struct list_head *done_q)
1133 LIST_HEAD(tmp_list);
1135 list_splice_init(work_q, &tmp_list);
1137 while (!list_empty(&tmp_list)) {
1138 struct scsi_cmnd *next, *scmd;
1139 int rtn;
1140 unsigned int id;
1142 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1143 id = scmd_id(scmd);
1145 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1146 "to target %d\n",
1147 current->comm, id));
1148 rtn = scsi_try_target_reset(scmd);
1149 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1150 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1151 " failed target: "
1152 "%d\n",
1153 current->comm, id));
1154 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1155 if (scmd_id(scmd) != id)
1156 continue;
1158 if ((rtn == SUCCESS || rtn == FAST_IO_FAIL)
1159 && (!scsi_device_online(scmd->device) ||
1160 rtn == FAST_IO_FAIL || !scsi_eh_tur(scmd)))
1161 scsi_eh_finish_cmd(scmd, done_q);
1162 else
1163 /* push back on work queue for further processing */
1164 list_move(&scmd->eh_entry, work_q);
1168 return list_empty(work_q);
1172 * scsi_eh_bus_reset - send a bus reset
1173 * @shost: &scsi host being recovered.
1174 * @work_q: &list_head for pending commands.
1175 * @done_q: &list_head for processed commands.
1177 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1178 struct list_head *work_q,
1179 struct list_head *done_q)
1181 struct scsi_cmnd *scmd, *chan_scmd, *next;
1182 unsigned int channel;
1183 int rtn;
1186 * we really want to loop over the various channels, and do this on
1187 * a channel by channel basis. we should also check to see if any
1188 * of the failed commands are on soft_reset devices, and if so, skip
1189 * the reset.
1192 for (channel = 0; channel <= shost->max_channel; channel++) {
1193 chan_scmd = NULL;
1194 list_for_each_entry(scmd, work_q, eh_entry) {
1195 if (channel == scmd_channel(scmd)) {
1196 chan_scmd = scmd;
1197 break;
1199 * FIXME add back in some support for
1200 * soft_reset devices.
1205 if (!chan_scmd)
1206 continue;
1207 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1208 " %d\n", current->comm,
1209 channel));
1210 rtn = scsi_try_bus_reset(chan_scmd);
1211 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1212 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1213 if (channel == scmd_channel(scmd))
1214 if (!scsi_device_online(scmd->device) ||
1215 rtn == FAST_IO_FAIL ||
1216 !scsi_eh_tur(scmd))
1217 scsi_eh_finish_cmd(scmd,
1218 done_q);
1220 } else {
1221 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1222 " failed chan: %d\n",
1223 current->comm,
1224 channel));
1227 return list_empty(work_q);
1231 * scsi_eh_host_reset - send a host reset
1232 * @work_q: list_head for processed commands.
1233 * @done_q: list_head for processed commands.
1235 static int scsi_eh_host_reset(struct list_head *work_q,
1236 struct list_head *done_q)
1238 struct scsi_cmnd *scmd, *next;
1239 int rtn;
1241 if (!list_empty(work_q)) {
1242 scmd = list_entry(work_q->next,
1243 struct scsi_cmnd, eh_entry);
1245 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1246 , current->comm));
1248 rtn = scsi_try_host_reset(scmd);
1249 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1250 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1251 if (!scsi_device_online(scmd->device) ||
1252 rtn == FAST_IO_FAIL ||
1253 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1254 !scsi_eh_tur(scmd))
1255 scsi_eh_finish_cmd(scmd, done_q);
1257 } else {
1258 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1259 " failed\n",
1260 current->comm));
1263 return list_empty(work_q);
1267 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1268 * @work_q: list_head for processed commands.
1269 * @done_q: list_head for processed commands.
1271 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1272 struct list_head *done_q)
1274 struct scsi_cmnd *scmd, *next;
1276 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1277 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1278 "not ready after error recovery\n");
1279 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1280 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1282 * FIXME: Handle lost cmds.
1285 scsi_eh_finish_cmd(scmd, done_q);
1287 return;
1291 * scsi_noretry_cmd - determinte if command should be failed fast
1292 * @scmd: SCSI cmd to examine.
1294 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1296 switch (host_byte(scmd->result)) {
1297 case DID_OK:
1298 break;
1299 case DID_BUS_BUSY:
1300 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1301 case DID_PARITY:
1302 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1303 case DID_ERROR:
1304 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1305 status_byte(scmd->result) == RESERVATION_CONFLICT)
1306 return 0;
1307 /* fall through */
1308 case DID_SOFT_ERROR:
1309 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1312 switch (status_byte(scmd->result)) {
1313 case CHECK_CONDITION:
1315 * assume caller has checked sense and determinted
1316 * the check condition was retryable.
1318 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1319 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1320 return 1;
1323 return 0;
1327 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1328 * @scmd: SCSI cmd to examine.
1330 * Notes:
1331 * This is *only* called when we are examining the status after sending
1332 * out the actual data command. any commands that are queued for error
1333 * recovery (e.g. test_unit_ready) do *not* come through here.
1335 * When this routine returns failed, it means the error handler thread
1336 * is woken. In cases where the error code indicates an error that
1337 * doesn't require the error handler read (i.e. we don't need to
1338 * abort/reset), this function should return SUCCESS.
1340 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1342 int rtn;
1345 * if the device is offline, then we clearly just pass the result back
1346 * up to the top level.
1348 if (!scsi_device_online(scmd->device)) {
1349 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1350 " as SUCCESS\n",
1351 __func__));
1352 return SUCCESS;
1356 * first check the host byte, to see if there is anything in there
1357 * that would indicate what we need to do.
1359 switch (host_byte(scmd->result)) {
1360 case DID_PASSTHROUGH:
1362 * no matter what, pass this through to the upper layer.
1363 * nuke this special code so that it looks like we are saying
1364 * did_ok.
1366 scmd->result &= 0xff00ffff;
1367 return SUCCESS;
1368 case DID_OK:
1370 * looks good. drop through, and check the next byte.
1372 break;
1373 case DID_NO_CONNECT:
1374 case DID_BAD_TARGET:
1375 case DID_ABORT:
1377 * note - this means that we just report the status back
1378 * to the top level driver, not that we actually think
1379 * that it indicates SUCCESS.
1381 return SUCCESS;
1383 * when the low level driver returns did_soft_error,
1384 * it is responsible for keeping an internal retry counter
1385 * in order to avoid endless loops (db)
1387 * actually this is a bug in this function here. we should
1388 * be mindful of the maximum number of retries specified
1389 * and not get stuck in a loop.
1391 case DID_SOFT_ERROR:
1392 goto maybe_retry;
1393 case DID_IMM_RETRY:
1394 return NEEDS_RETRY;
1396 case DID_REQUEUE:
1397 return ADD_TO_MLQUEUE;
1398 case DID_TRANSPORT_DISRUPTED:
1400 * LLD/transport was disrupted during processing of the IO.
1401 * The transport class is now blocked/blocking,
1402 * and the transport will decide what to do with the IO
1403 * based on its timers and recovery capablilities if
1404 * there are enough retries.
1406 goto maybe_retry;
1407 case DID_TRANSPORT_FAILFAST:
1409 * The transport decided to failfast the IO (most likely
1410 * the fast io fail tmo fired), so send IO directly upwards.
1412 return SUCCESS;
1413 case DID_ERROR:
1414 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1415 status_byte(scmd->result) == RESERVATION_CONFLICT)
1417 * execute reservation conflict processing code
1418 * lower down
1420 break;
1421 /* fallthrough */
1422 case DID_BUS_BUSY:
1423 case DID_PARITY:
1424 goto maybe_retry;
1425 case DID_TIME_OUT:
1427 * when we scan the bus, we get timeout messages for
1428 * these commands if there is no device available.
1429 * other hosts report did_no_connect for the same thing.
1431 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1432 scmd->cmnd[0] == INQUIRY)) {
1433 return SUCCESS;
1434 } else {
1435 return FAILED;
1437 case DID_RESET:
1438 return SUCCESS;
1439 default:
1440 return FAILED;
1444 * next, check the message byte.
1446 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1447 return FAILED;
1450 * check the status byte to see if this indicates anything special.
1452 switch (status_byte(scmd->result)) {
1453 case QUEUE_FULL:
1454 scsi_handle_queue_full(scmd->device);
1456 * the case of trying to send too many commands to a
1457 * tagged queueing device.
1459 case BUSY:
1461 * device can't talk to us at the moment. Should only
1462 * occur (SAM-3) when the task queue is empty, so will cause
1463 * the empty queue handling to trigger a stall in the
1464 * device.
1466 return ADD_TO_MLQUEUE;
1467 case GOOD:
1468 scsi_handle_queue_ramp_up(scmd->device);
1469 case COMMAND_TERMINATED:
1470 return SUCCESS;
1471 case TASK_ABORTED:
1472 goto maybe_retry;
1473 case CHECK_CONDITION:
1474 rtn = scsi_check_sense(scmd);
1475 if (rtn == NEEDS_RETRY)
1476 goto maybe_retry;
1477 else if (rtn == TARGET_ERROR) {
1479 * Need to modify host byte to signal a
1480 * permanent target failure
1482 scmd->result |= (DID_TARGET_FAILURE << 16);
1483 rtn = SUCCESS;
1485 /* if rtn == FAILED, we have no sense information;
1486 * returning FAILED will wake the error handler thread
1487 * to collect the sense and redo the decide
1488 * disposition */
1489 return rtn;
1490 case CONDITION_GOOD:
1491 case INTERMEDIATE_GOOD:
1492 case INTERMEDIATE_C_GOOD:
1493 case ACA_ACTIVE:
1495 * who knows? FIXME(eric)
1497 return SUCCESS;
1499 case RESERVATION_CONFLICT:
1500 sdev_printk(KERN_INFO, scmd->device,
1501 "reservation conflict\n");
1502 scmd->result |= (DID_NEXUS_FAILURE << 16);
1503 return SUCCESS; /* causes immediate i/o error */
1504 default:
1505 return FAILED;
1507 return FAILED;
1509 maybe_retry:
1511 /* we requeue for retry because the error was retryable, and
1512 * the request was not marked fast fail. Note that above,
1513 * even if the request is marked fast fail, we still requeue
1514 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1515 if ((++scmd->retries) <= scmd->allowed
1516 && !scsi_noretry_cmd(scmd)) {
1517 return NEEDS_RETRY;
1518 } else {
1520 * no more retries - report this one back to upper level.
1522 return SUCCESS;
1526 static void eh_lock_door_done(struct request *req, int uptodate)
1528 __blk_put_request(req->q, req);
1532 * scsi_eh_lock_door - Prevent medium removal for the specified device
1533 * @sdev: SCSI device to prevent medium removal
1535 * Locking:
1536 * We must be called from process context.
1538 * Notes:
1539 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1540 * head of the devices request queue, and continue.
1542 static void scsi_eh_lock_door(struct scsi_device *sdev)
1544 struct request *req;
1547 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1548 * request becomes available
1550 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1552 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1553 req->cmd[1] = 0;
1554 req->cmd[2] = 0;
1555 req->cmd[3] = 0;
1556 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1557 req->cmd[5] = 0;
1559 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1561 req->cmd_type = REQ_TYPE_BLOCK_PC;
1562 req->cmd_flags |= REQ_QUIET;
1563 req->timeout = 10 * HZ;
1564 req->retries = 5;
1566 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1570 * scsi_restart_operations - restart io operations to the specified host.
1571 * @shost: Host we are restarting.
1573 * Notes:
1574 * When we entered the error handler, we blocked all further i/o to
1575 * this device. we need to 'reverse' this process.
1577 static void scsi_restart_operations(struct Scsi_Host *shost)
1579 struct scsi_device *sdev;
1580 unsigned long flags;
1583 * If the door was locked, we need to insert a door lock request
1584 * onto the head of the SCSI request queue for the device. There
1585 * is no point trying to lock the door of an off-line device.
1587 shost_for_each_device(sdev, shost) {
1588 if (scsi_device_online(sdev) && sdev->locked)
1589 scsi_eh_lock_door(sdev);
1593 * next free up anything directly waiting upon the host. this
1594 * will be requests for character device operations, and also for
1595 * ioctls to queued block devices.
1597 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1598 __func__));
1600 spin_lock_irqsave(shost->host_lock, flags);
1601 if (scsi_host_set_state(shost, SHOST_RUNNING))
1602 if (scsi_host_set_state(shost, SHOST_CANCEL))
1603 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1604 spin_unlock_irqrestore(shost->host_lock, flags);
1606 wake_up(&shost->host_wait);
1609 * finally we need to re-initiate requests that may be pending. we will
1610 * have had everything blocked while error handling is taking place, and
1611 * now that error recovery is done, we will need to ensure that these
1612 * requests are started.
1614 scsi_run_host_queues(shost);
1618 * scsi_eh_ready_devs - check device ready state and recover if not.
1619 * @shost: host to be recovered.
1620 * @work_q: &list_head for pending commands.
1621 * @done_q: &list_head for processed commands.
1623 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1624 struct list_head *work_q,
1625 struct list_head *done_q)
1627 if (!scsi_eh_stu(shost, work_q, done_q))
1628 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1629 if (!scsi_eh_target_reset(shost, work_q, done_q))
1630 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1631 if (!scsi_eh_host_reset(work_q, done_q))
1632 scsi_eh_offline_sdevs(work_q,
1633 done_q);
1635 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1638 * scsi_eh_flush_done_q - finish processed commands or retry them.
1639 * @done_q: list_head of processed commands.
1641 void scsi_eh_flush_done_q(struct list_head *done_q)
1643 struct scsi_cmnd *scmd, *next;
1645 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1646 list_del_init(&scmd->eh_entry);
1647 if (scsi_device_online(scmd->device) &&
1648 !scsi_noretry_cmd(scmd) &&
1649 (++scmd->retries <= scmd->allowed)) {
1650 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1651 " retry cmd: %p\n",
1652 current->comm,
1653 scmd));
1654 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1655 } else {
1657 * If just we got sense for the device (called
1658 * scsi_eh_get_sense), scmd->result is already
1659 * set, do not set DRIVER_TIMEOUT.
1661 if (!scmd->result)
1662 scmd->result |= (DRIVER_TIMEOUT << 24);
1663 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1664 " cmd: %p\n",
1665 current->comm, scmd));
1666 scsi_finish_command(scmd);
1670 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1673 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1674 * @shost: Host to unjam.
1676 * Notes:
1677 * When we come in here, we *know* that all commands on the bus have
1678 * either completed, failed or timed out. we also know that no further
1679 * commands are being sent to the host, so things are relatively quiet
1680 * and we have freedom to fiddle with things as we wish.
1682 * This is only the *default* implementation. it is possible for
1683 * individual drivers to supply their own version of this function, and
1684 * if the maintainer wishes to do this, it is strongly suggested that
1685 * this function be taken as a template and modified. this function
1686 * was designed to correctly handle problems for about 95% of the
1687 * different cases out there, and it should always provide at least a
1688 * reasonable amount of error recovery.
1690 * Any command marked 'failed' or 'timeout' must eventually have
1691 * scsi_finish_cmd() called for it. we do all of the retry stuff
1692 * here, so when we restart the host after we return it should have an
1693 * empty queue.
1695 static void scsi_unjam_host(struct Scsi_Host *shost)
1697 unsigned long flags;
1698 LIST_HEAD(eh_work_q);
1699 LIST_HEAD(eh_done_q);
1701 spin_lock_irqsave(shost->host_lock, flags);
1702 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1703 spin_unlock_irqrestore(shost->host_lock, flags);
1705 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1707 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1708 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1709 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1711 scsi_eh_flush_done_q(&eh_done_q);
1715 * scsi_error_handler - SCSI error handler thread
1716 * @data: Host for which we are running.
1718 * Notes:
1719 * This is the main error handling loop. This is run as a kernel thread
1720 * for every SCSI host and handles all error handling activity.
1722 int scsi_error_handler(void *data)
1724 struct Scsi_Host *shost = data;
1727 * We use TASK_INTERRUPTIBLE so that the thread is not
1728 * counted against the load average as a running process.
1729 * We never actually get interrupted because kthread_run
1730 * disables signal delivery for the created thread.
1732 set_current_state(TASK_INTERRUPTIBLE);
1733 while (!kthread_should_stop()) {
1734 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1735 shost->host_failed != shost->host_busy) {
1736 SCSI_LOG_ERROR_RECOVERY(1,
1737 printk("Error handler scsi_eh_%d sleeping\n",
1738 shost->host_no));
1739 schedule();
1740 set_current_state(TASK_INTERRUPTIBLE);
1741 continue;
1744 __set_current_state(TASK_RUNNING);
1745 SCSI_LOG_ERROR_RECOVERY(1,
1746 printk("Error handler scsi_eh_%d waking up\n",
1747 shost->host_no));
1750 * We have a host that is failing for some reason. Figure out
1751 * what we need to do to get it up and online again (if we can).
1752 * If we fail, we end up taking the thing offline.
1754 if (scsi_autopm_get_host(shost) != 0) {
1755 SCSI_LOG_ERROR_RECOVERY(1,
1756 printk(KERN_ERR "Error handler scsi_eh_%d "
1757 "unable to autoresume\n",
1758 shost->host_no));
1759 continue;
1762 if (shost->transportt->eh_strategy_handler)
1763 shost->transportt->eh_strategy_handler(shost);
1764 else
1765 scsi_unjam_host(shost);
1768 * Note - if the above fails completely, the action is to take
1769 * individual devices offline and flush the queue of any
1770 * outstanding requests that may have been pending. When we
1771 * restart, we restart any I/O to any other devices on the bus
1772 * which are still online.
1774 scsi_restart_operations(shost);
1775 scsi_autopm_put_host(shost);
1776 set_current_state(TASK_INTERRUPTIBLE);
1778 __set_current_state(TASK_RUNNING);
1780 SCSI_LOG_ERROR_RECOVERY(1,
1781 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1782 shost->ehandler = NULL;
1783 return 0;
1787 * Function: scsi_report_bus_reset()
1789 * Purpose: Utility function used by low-level drivers to report that
1790 * they have observed a bus reset on the bus being handled.
1792 * Arguments: shost - Host in question
1793 * channel - channel on which reset was observed.
1795 * Returns: Nothing
1797 * Lock status: Host lock must be held.
1799 * Notes: This only needs to be called if the reset is one which
1800 * originates from an unknown location. Resets originated
1801 * by the mid-level itself don't need to call this, but there
1802 * should be no harm.
1804 * The main purpose of this is to make sure that a CHECK_CONDITION
1805 * is properly treated.
1807 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1809 struct scsi_device *sdev;
1811 __shost_for_each_device(sdev, shost) {
1812 if (channel == sdev_channel(sdev))
1813 __scsi_report_device_reset(sdev, NULL);
1816 EXPORT_SYMBOL(scsi_report_bus_reset);
1819 * Function: scsi_report_device_reset()
1821 * Purpose: Utility function used by low-level drivers to report that
1822 * they have observed a device reset on the device being handled.
1824 * Arguments: shost - Host in question
1825 * channel - channel on which reset was observed
1826 * target - target on which reset was observed
1828 * Returns: Nothing
1830 * Lock status: Host lock must be held
1832 * Notes: This only needs to be called if the reset is one which
1833 * originates from an unknown location. Resets originated
1834 * by the mid-level itself don't need to call this, but there
1835 * should be no harm.
1837 * The main purpose of this is to make sure that a CHECK_CONDITION
1838 * is properly treated.
1840 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1842 struct scsi_device *sdev;
1844 __shost_for_each_device(sdev, shost) {
1845 if (channel == sdev_channel(sdev) &&
1846 target == sdev_id(sdev))
1847 __scsi_report_device_reset(sdev, NULL);
1850 EXPORT_SYMBOL(scsi_report_device_reset);
1852 static void
1853 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1858 * Function: scsi_reset_provider
1860 * Purpose: Send requested reset to a bus or device at any phase.
1862 * Arguments: device - device to send reset to
1863 * flag - reset type (see scsi.h)
1865 * Returns: SUCCESS/FAILURE.
1867 * Notes: This is used by the SCSI Generic driver to provide
1868 * Bus/Device reset capability.
1871 scsi_reset_provider(struct scsi_device *dev, int flag)
1873 struct scsi_cmnd *scmd;
1874 struct Scsi_Host *shost = dev->host;
1875 struct request req;
1876 unsigned long flags;
1877 int rtn;
1879 if (scsi_autopm_get_host(shost) < 0)
1880 return FAILED;
1882 scmd = scsi_get_command(dev, GFP_KERNEL);
1883 blk_rq_init(NULL, &req);
1884 scmd->request = &req;
1886 scmd->cmnd = req.cmd;
1888 scmd->scsi_done = scsi_reset_provider_done_command;
1889 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1891 scmd->cmd_len = 0;
1893 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1895 spin_lock_irqsave(shost->host_lock, flags);
1896 shost->tmf_in_progress = 1;
1897 spin_unlock_irqrestore(shost->host_lock, flags);
1899 switch (flag) {
1900 case SCSI_TRY_RESET_DEVICE:
1901 rtn = scsi_try_bus_device_reset(scmd);
1902 if (rtn == SUCCESS)
1903 break;
1904 /* FALLTHROUGH */
1905 case SCSI_TRY_RESET_TARGET:
1906 rtn = scsi_try_target_reset(scmd);
1907 if (rtn == SUCCESS)
1908 break;
1909 /* FALLTHROUGH */
1910 case SCSI_TRY_RESET_BUS:
1911 rtn = scsi_try_bus_reset(scmd);
1912 if (rtn == SUCCESS)
1913 break;
1914 /* FALLTHROUGH */
1915 case SCSI_TRY_RESET_HOST:
1916 rtn = scsi_try_host_reset(scmd);
1917 break;
1918 default:
1919 rtn = FAILED;
1922 spin_lock_irqsave(shost->host_lock, flags);
1923 shost->tmf_in_progress = 0;
1924 spin_unlock_irqrestore(shost->host_lock, flags);
1927 * be sure to wake up anyone who was sleeping or had their queue
1928 * suspended while we performed the TMF.
1930 SCSI_LOG_ERROR_RECOVERY(3,
1931 printk("%s: waking up host to restart after TMF\n",
1932 __func__));
1934 wake_up(&shost->host_wait);
1936 scsi_run_host_queues(shost);
1938 scsi_next_command(scmd);
1939 scsi_autopm_put_host(shost);
1940 return rtn;
1942 EXPORT_SYMBOL(scsi_reset_provider);
1945 * scsi_normalize_sense - normalize main elements from either fixed or
1946 * descriptor sense data format into a common format.
1948 * @sense_buffer: byte array containing sense data returned by device
1949 * @sb_len: number of valid bytes in sense_buffer
1950 * @sshdr: pointer to instance of structure that common
1951 * elements are written to.
1953 * Notes:
1954 * The "main elements" from sense data are: response_code, sense_key,
1955 * asc, ascq and additional_length (only for descriptor format).
1957 * Typically this function can be called after a device has
1958 * responded to a SCSI command with the CHECK_CONDITION status.
1960 * Return value:
1961 * 1 if valid sense data information found, else 0;
1963 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1964 struct scsi_sense_hdr *sshdr)
1966 if (!sense_buffer || !sb_len)
1967 return 0;
1969 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1971 sshdr->response_code = (sense_buffer[0] & 0x7f);
1973 if (!scsi_sense_valid(sshdr))
1974 return 0;
1976 if (sshdr->response_code >= 0x72) {
1978 * descriptor format
1980 if (sb_len > 1)
1981 sshdr->sense_key = (sense_buffer[1] & 0xf);
1982 if (sb_len > 2)
1983 sshdr->asc = sense_buffer[2];
1984 if (sb_len > 3)
1985 sshdr->ascq = sense_buffer[3];
1986 if (sb_len > 7)
1987 sshdr->additional_length = sense_buffer[7];
1988 } else {
1990 * fixed format
1992 if (sb_len > 2)
1993 sshdr->sense_key = (sense_buffer[2] & 0xf);
1994 if (sb_len > 7) {
1995 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1996 sb_len : (sense_buffer[7] + 8);
1997 if (sb_len > 12)
1998 sshdr->asc = sense_buffer[12];
1999 if (sb_len > 13)
2000 sshdr->ascq = sense_buffer[13];
2004 return 1;
2006 EXPORT_SYMBOL(scsi_normalize_sense);
2008 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2009 struct scsi_sense_hdr *sshdr)
2011 return scsi_normalize_sense(cmd->sense_buffer,
2012 SCSI_SENSE_BUFFERSIZE, sshdr);
2014 EXPORT_SYMBOL(scsi_command_normalize_sense);
2017 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2018 * @sense_buffer: byte array of descriptor format sense data
2019 * @sb_len: number of valid bytes in sense_buffer
2020 * @desc_type: value of descriptor type to find
2021 * (e.g. 0 -> information)
2023 * Notes:
2024 * only valid when sense data is in descriptor format
2026 * Return value:
2027 * pointer to start of (first) descriptor if found else NULL
2029 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2030 int desc_type)
2032 int add_sen_len, add_len, desc_len, k;
2033 const u8 * descp;
2035 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2036 return NULL;
2037 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2038 return NULL;
2039 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2040 add_sen_len : (sb_len - 8);
2041 descp = &sense_buffer[8];
2042 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2043 descp += desc_len;
2044 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2045 desc_len = add_len + 2;
2046 if (descp[0] == desc_type)
2047 return descp;
2048 if (add_len < 0) // short descriptor ??
2049 break;
2051 return NULL;
2053 EXPORT_SYMBOL(scsi_sense_desc_find);
2056 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2057 * @sense_buffer: byte array of sense data
2058 * @sb_len: number of valid bytes in sense_buffer
2059 * @info_out: pointer to 64 integer where 8 or 4 byte information
2060 * field will be placed if found.
2062 * Return value:
2063 * 1 if information field found, 0 if not found.
2065 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2066 u64 * info_out)
2068 int j;
2069 const u8 * ucp;
2070 u64 ull;
2072 if (sb_len < 7)
2073 return 0;
2074 switch (sense_buffer[0] & 0x7f) {
2075 case 0x70:
2076 case 0x71:
2077 if (sense_buffer[0] & 0x80) {
2078 *info_out = (sense_buffer[3] << 24) +
2079 (sense_buffer[4] << 16) +
2080 (sense_buffer[5] << 8) + sense_buffer[6];
2081 return 1;
2082 } else
2083 return 0;
2084 case 0x72:
2085 case 0x73:
2086 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2087 0 /* info desc */);
2088 if (ucp && (0xa == ucp[1])) {
2089 ull = 0;
2090 for (j = 0; j < 8; ++j) {
2091 if (j > 0)
2092 ull <<= 8;
2093 ull |= ucp[4 + j];
2095 *info_out = ull;
2096 return 1;
2097 } else
2098 return 0;
2099 default:
2100 return 0;
2103 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2106 * scsi_build_sense_buffer - build sense data in a buffer
2107 * @desc: Sense format (non zero == descriptor format,
2108 * 0 == fixed format)
2109 * @buf: Where to build sense data
2110 * @key: Sense key
2111 * @asc: Additional sense code
2112 * @ascq: Additional sense code qualifier
2115 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2117 if (desc) {
2118 buf[0] = 0x72; /* descriptor, current */
2119 buf[1] = key;
2120 buf[2] = asc;
2121 buf[3] = ascq;
2122 buf[7] = 0;
2123 } else {
2124 buf[0] = 0x70; /* fixed, current */
2125 buf[2] = key;
2126 buf[7] = 0xa;
2127 buf[12] = asc;
2128 buf[13] = ascq;
2131 EXPORT_SYMBOL(scsi_build_sense_buffer);