Loading...
1/*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
3 *
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.
8 *
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11 *
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)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
19#include <linux/gfp.h>
20#include <linux/timer.h>
21#include <linux/string.h>
22#include <linux/kernel.h>
23#include <linux/freezer.h>
24#include <linux/kthread.h>
25#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
28#include <linux/jiffies.h>
29
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_common.h>
37#include <scsi/scsi_transport.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_ioctl.h>
40#include <scsi/scsi_dh.h>
41#include <scsi/sg.h>
42
43#include "scsi_priv.h"
44#include "scsi_logging.h"
45#include "scsi_transport_api.h"
46
47#include <trace/events/scsi.h>
48
49static void scsi_eh_done(struct scsi_cmnd *scmd);
50
51/*
52 * These should *probably* be handled by the host itself.
53 * Since it is allowed to sleep, it probably should.
54 */
55#define BUS_RESET_SETTLE_TIME (10)
56#define HOST_RESET_SETTLE_TIME (10)
57
58static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
59static int scsi_try_to_abort_cmd(struct scsi_host_template *,
60 struct scsi_cmnd *);
61
62/* called with shost->host_lock held */
63void scsi_eh_wakeup(struct Scsi_Host *shost)
64{
65 if (atomic_read(&shost->host_busy) == shost->host_failed) {
66 trace_scsi_eh_wakeup(shost);
67 wake_up_process(shost->ehandler);
68 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
69 "Waking error handler thread\n"));
70 }
71}
72
73/**
74 * scsi_schedule_eh - schedule EH for SCSI host
75 * @shost: SCSI host to invoke error handling on.
76 *
77 * Schedule SCSI EH without scmd.
78 */
79void scsi_schedule_eh(struct Scsi_Host *shost)
80{
81 unsigned long flags;
82
83 spin_lock_irqsave(shost->host_lock, flags);
84
85 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
86 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
87 shost->host_eh_scheduled++;
88 scsi_eh_wakeup(shost);
89 }
90
91 spin_unlock_irqrestore(shost->host_lock, flags);
92}
93EXPORT_SYMBOL_GPL(scsi_schedule_eh);
94
95static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
96{
97 if (!shost->last_reset || shost->eh_deadline == -1)
98 return 0;
99
100 /*
101 * 32bit accesses are guaranteed to be atomic
102 * (on all supported architectures), so instead
103 * of using a spinlock we can as well double check
104 * if eh_deadline has been set to 'off' during the
105 * time_before call.
106 */
107 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
108 shost->eh_deadline > -1)
109 return 0;
110
111 return 1;
112}
113
114/**
115 * scmd_eh_abort_handler - Handle command aborts
116 * @work: command to be aborted.
117 */
118void
119scmd_eh_abort_handler(struct work_struct *work)
120{
121 struct scsi_cmnd *scmd =
122 container_of(work, struct scsi_cmnd, abort_work.work);
123 struct scsi_device *sdev = scmd->device;
124 int rtn;
125
126 if (scsi_host_eh_past_deadline(sdev->host)) {
127 SCSI_LOG_ERROR_RECOVERY(3,
128 scmd_printk(KERN_INFO, scmd,
129 "eh timeout, not aborting\n"));
130 } else {
131 SCSI_LOG_ERROR_RECOVERY(3,
132 scmd_printk(KERN_INFO, scmd,
133 "aborting command\n"));
134 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
135 if (rtn == SUCCESS) {
136 set_host_byte(scmd, DID_TIME_OUT);
137 if (scsi_host_eh_past_deadline(sdev->host)) {
138 SCSI_LOG_ERROR_RECOVERY(3,
139 scmd_printk(KERN_INFO, scmd,
140 "eh timeout, not retrying "
141 "aborted command\n"));
142 } else if (!scsi_noretry_cmd(scmd) &&
143 (++scmd->retries <= scmd->allowed)) {
144 SCSI_LOG_ERROR_RECOVERY(3,
145 scmd_printk(KERN_WARNING, scmd,
146 "retry aborted command\n"));
147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
148 return;
149 } else {
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING, scmd,
152 "finish aborted command\n"));
153 scsi_finish_command(scmd);
154 return;
155 }
156 } else {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
159 "cmd abort %s\n",
160 (rtn == FAST_IO_FAIL) ?
161 "not send" : "failed"));
162 }
163 }
164
165 if (!scsi_eh_scmd_add(scmd, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING, scmd,
168 "terminate aborted command\n"));
169 set_host_byte(scmd, DID_TIME_OUT);
170 scsi_finish_command(scmd);
171 }
172}
173
174/**
175 * scsi_abort_command - schedule a command abort
176 * @scmd: scmd to abort.
177 *
178 * We only need to abort commands after a command timeout
179 */
180static int
181scsi_abort_command(struct scsi_cmnd *scmd)
182{
183 struct scsi_device *sdev = scmd->device;
184 struct Scsi_Host *shost = sdev->host;
185 unsigned long flags;
186
187 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
188 /*
189 * Retry after abort failed, escalate to next level.
190 */
191 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
192 SCSI_LOG_ERROR_RECOVERY(3,
193 scmd_printk(KERN_INFO, scmd,
194 "previous abort failed\n"));
195 BUG_ON(delayed_work_pending(&scmd->abort_work));
196 return FAILED;
197 }
198
199 /*
200 * Do not try a command abort if
201 * SCSI EH has already started.
202 */
203 spin_lock_irqsave(shost->host_lock, flags);
204 if (scsi_host_in_recovery(shost)) {
205 spin_unlock_irqrestore(shost->host_lock, flags);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_INFO, scmd,
208 "not aborting, host in recovery\n"));
209 return FAILED;
210 }
211
212 if (shost->eh_deadline != -1 && !shost->last_reset)
213 shost->last_reset = jiffies;
214 spin_unlock_irqrestore(shost->host_lock, flags);
215
216 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
217 SCSI_LOG_ERROR_RECOVERY(3,
218 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
219 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
220 return SUCCESS;
221}
222
223/**
224 * scsi_eh_scmd_add - add scsi cmd to error handling.
225 * @scmd: scmd to run eh on.
226 * @eh_flag: optional SCSI_EH flag.
227 *
228 * Return value:
229 * 0 on failure.
230 */
231int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
232{
233 struct Scsi_Host *shost = scmd->device->host;
234 unsigned long flags;
235 int ret = 0;
236
237 if (!shost->ehandler)
238 return 0;
239
240 spin_lock_irqsave(shost->host_lock, flags);
241 if (scsi_host_set_state(shost, SHOST_RECOVERY))
242 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
243 goto out_unlock;
244
245 if (shost->eh_deadline != -1 && !shost->last_reset)
246 shost->last_reset = jiffies;
247
248 ret = 1;
249 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
250 eh_flag &= ~SCSI_EH_CANCEL_CMD;
251 scmd->eh_eflags |= eh_flag;
252 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
253 shost->host_failed++;
254 scsi_eh_wakeup(shost);
255 out_unlock:
256 spin_unlock_irqrestore(shost->host_lock, flags);
257 return ret;
258}
259
260/**
261 * scsi_times_out - Timeout function for normal scsi commands.
262 * @req: request that is timing out.
263 *
264 * Notes:
265 * We do not need to lock this. There is the potential for a race
266 * only in that the normal completion handling might run, but if the
267 * normal completion function determines that the timer has already
268 * fired, then it mustn't do anything.
269 */
270enum blk_eh_timer_return scsi_times_out(struct request *req)
271{
272 struct scsi_cmnd *scmd = req->special;
273 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
274 struct Scsi_Host *host = scmd->device->host;
275
276 trace_scsi_dispatch_cmd_timeout(scmd);
277 scsi_log_completion(scmd, TIMEOUT_ERROR);
278
279 if (host->eh_deadline != -1 && !host->last_reset)
280 host->last_reset = jiffies;
281
282 if (host->transportt->eh_timed_out)
283 rtn = host->transportt->eh_timed_out(scmd);
284 else if (host->hostt->eh_timed_out)
285 rtn = host->hostt->eh_timed_out(scmd);
286
287 if (rtn == BLK_EH_NOT_HANDLED) {
288 if (!host->hostt->no_async_abort &&
289 scsi_abort_command(scmd) == SUCCESS)
290 return BLK_EH_NOT_HANDLED;
291
292 set_host_byte(scmd, DID_TIME_OUT);
293 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
294 rtn = BLK_EH_HANDLED;
295 }
296
297 return rtn;
298}
299
300/**
301 * scsi_block_when_processing_errors - Prevent cmds from being queued.
302 * @sdev: Device on which we are performing recovery.
303 *
304 * Description:
305 * We block until the host is out of error recovery, and then check to
306 * see whether the host or the device is offline.
307 *
308 * Return value:
309 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
310 */
311int scsi_block_when_processing_errors(struct scsi_device *sdev)
312{
313 int online;
314
315 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
316
317 online = scsi_device_online(sdev);
318
319 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
320 "%s: rtn: %d\n", __func__, online));
321
322 return online;
323}
324EXPORT_SYMBOL(scsi_block_when_processing_errors);
325
326#ifdef CONFIG_SCSI_LOGGING
327/**
328 * scsi_eh_prt_fail_stats - Log info on failures.
329 * @shost: scsi host being recovered.
330 * @work_q: Queue of scsi cmds to process.
331 */
332static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
333 struct list_head *work_q)
334{
335 struct scsi_cmnd *scmd;
336 struct scsi_device *sdev;
337 int total_failures = 0;
338 int cmd_failed = 0;
339 int cmd_cancel = 0;
340 int devices_failed = 0;
341
342 shost_for_each_device(sdev, shost) {
343 list_for_each_entry(scmd, work_q, eh_entry) {
344 if (scmd->device == sdev) {
345 ++total_failures;
346 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
347 ++cmd_cancel;
348 else
349 ++cmd_failed;
350 }
351 }
352
353 if (cmd_cancel || cmd_failed) {
354 SCSI_LOG_ERROR_RECOVERY(3,
355 shost_printk(KERN_INFO, shost,
356 "%s: cmds failed: %d, cancel: %d\n",
357 __func__, cmd_failed,
358 cmd_cancel));
359 cmd_cancel = 0;
360 cmd_failed = 0;
361 ++devices_failed;
362 }
363 }
364
365 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
366 "Total of %d commands on %d"
367 " devices require eh work\n",
368 total_failures, devices_failed));
369}
370#endif
371
372 /**
373 * scsi_report_lun_change - Set flag on all *other* devices on the same target
374 * to indicate that a UNIT ATTENTION is expected.
375 * @sdev: Device reporting the UNIT ATTENTION
376 */
377static void scsi_report_lun_change(struct scsi_device *sdev)
378{
379 sdev->sdev_target->expecting_lun_change = 1;
380}
381
382/**
383 * scsi_report_sense - Examine scsi sense information and log messages for
384 * certain conditions, also issue uevents for some of them.
385 * @sdev: Device reporting the sense code
386 * @sshdr: sshdr to be examined
387 */
388static void scsi_report_sense(struct scsi_device *sdev,
389 struct scsi_sense_hdr *sshdr)
390{
391 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
392
393 if (sshdr->sense_key == UNIT_ATTENTION) {
394 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
395 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
396 sdev_printk(KERN_WARNING, sdev,
397 "Inquiry data has changed");
398 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
399 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
400 scsi_report_lun_change(sdev);
401 sdev_printk(KERN_WARNING, sdev,
402 "Warning! Received an indication that the "
403 "LUN assignments on this target have "
404 "changed. The Linux SCSI layer does not "
405 "automatically remap LUN assignments.\n");
406 } else if (sshdr->asc == 0x3f)
407 sdev_printk(KERN_WARNING, sdev,
408 "Warning! Received an indication that the "
409 "operating parameters on this target have "
410 "changed. The Linux SCSI layer does not "
411 "automatically adjust these parameters.\n");
412
413 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
414 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
415 sdev_printk(KERN_WARNING, sdev,
416 "Warning! Received an indication that the "
417 "LUN reached a thin provisioning soft "
418 "threshold.\n");
419 }
420
421 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
422 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
423 sdev_printk(KERN_WARNING, sdev,
424 "Mode parameters changed");
425 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
426 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
427 sdev_printk(KERN_WARNING, sdev,
428 "Asymmetric access state changed");
429 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
430 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
431 sdev_printk(KERN_WARNING, sdev,
432 "Capacity data has changed");
433 } else if (sshdr->asc == 0x2a)
434 sdev_printk(KERN_WARNING, sdev,
435 "Parameters changed");
436 }
437
438 if (evt_type != SDEV_EVT_MAXBITS) {
439 set_bit(evt_type, sdev->pending_events);
440 schedule_work(&sdev->event_work);
441 }
442}
443
444/**
445 * scsi_check_sense - Examine scsi cmd sense
446 * @scmd: Cmd to have sense checked.
447 *
448 * Return value:
449 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
450 *
451 * Notes:
452 * When a deferred error is detected the current command has
453 * not been executed and needs retrying.
454 */
455int scsi_check_sense(struct scsi_cmnd *scmd)
456{
457 struct scsi_device *sdev = scmd->device;
458 struct scsi_sense_hdr sshdr;
459
460 if (! scsi_command_normalize_sense(scmd, &sshdr))
461 return FAILED; /* no valid sense data */
462
463 scsi_report_sense(sdev, &sshdr);
464
465 if (scsi_sense_is_deferred(&sshdr))
466 return NEEDS_RETRY;
467
468 if (sdev->handler && sdev->handler->check_sense) {
469 int rc;
470
471 rc = sdev->handler->check_sense(sdev, &sshdr);
472 if (rc != SCSI_RETURN_NOT_HANDLED)
473 return rc;
474 /* handler does not care. Drop down to default handling */
475 }
476
477 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
478 /*
479 * nasty: for mid-layer issued TURs, we need to return the
480 * actual sense data without any recovery attempt. For eh
481 * issued ones, we need to try to recover and interpret
482 */
483 return SUCCESS;
484
485 /*
486 * Previous logic looked for FILEMARK, EOM or ILI which are
487 * mainly associated with tapes and returned SUCCESS.
488 */
489 if (sshdr.response_code == 0x70) {
490 /* fixed format */
491 if (scmd->sense_buffer[2] & 0xe0)
492 return SUCCESS;
493 } else {
494 /*
495 * descriptor format: look for "stream commands sense data
496 * descriptor" (see SSC-3). Assume single sense data
497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
498 */
499 if ((sshdr.additional_length > 3) &&
500 (scmd->sense_buffer[8] == 0x4) &&
501 (scmd->sense_buffer[11] & 0xe0))
502 return SUCCESS;
503 }
504
505 switch (sshdr.sense_key) {
506 case NO_SENSE:
507 return SUCCESS;
508 case RECOVERED_ERROR:
509 return /* soft_error */ SUCCESS;
510
511 case ABORTED_COMMAND:
512 if (sshdr.asc == 0x10) /* DIF */
513 return SUCCESS;
514
515 return NEEDS_RETRY;
516 case NOT_READY:
517 case UNIT_ATTENTION:
518 /*
519 * if we are expecting a cc/ua because of a bus reset that we
520 * performed, treat this just as a retry. otherwise this is
521 * information that we should pass up to the upper-level driver
522 * so that we can deal with it there.
523 */
524 if (scmd->device->expecting_cc_ua) {
525 /*
526 * Because some device does not queue unit
527 * attentions correctly, we carefully check
528 * additional sense code and qualifier so as
529 * not to squash media change unit attention.
530 */
531 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
532 scmd->device->expecting_cc_ua = 0;
533 return NEEDS_RETRY;
534 }
535 }
536 /*
537 * we might also expect a cc/ua if another LUN on the target
538 * reported a UA with an ASC/ASCQ of 3F 0E -
539 * REPORTED LUNS DATA HAS CHANGED.
540 */
541 if (scmd->device->sdev_target->expecting_lun_change &&
542 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
543 return NEEDS_RETRY;
544 /*
545 * if the device is in the process of becoming ready, we
546 * should retry.
547 */
548 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
549 return NEEDS_RETRY;
550 /*
551 * if the device is not started, we need to wake
552 * the error handler to start the motor
553 */
554 if (scmd->device->allow_restart &&
555 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
556 return FAILED;
557 /*
558 * Pass the UA upwards for a determination in the completion
559 * functions.
560 */
561 return SUCCESS;
562
563 /* these are not supported */
564 case DATA_PROTECT:
565 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
566 /* Thin provisioning hard threshold reached */
567 set_host_byte(scmd, DID_ALLOC_FAILURE);
568 return SUCCESS;
569 }
570 case COPY_ABORTED:
571 case VOLUME_OVERFLOW:
572 case MISCOMPARE:
573 case BLANK_CHECK:
574 set_host_byte(scmd, DID_TARGET_FAILURE);
575 return SUCCESS;
576
577 case MEDIUM_ERROR:
578 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
579 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
580 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
581 set_host_byte(scmd, DID_MEDIUM_ERROR);
582 return SUCCESS;
583 }
584 return NEEDS_RETRY;
585
586 case HARDWARE_ERROR:
587 if (scmd->device->retry_hwerror)
588 return ADD_TO_MLQUEUE;
589 else
590 set_host_byte(scmd, DID_TARGET_FAILURE);
591
592 case ILLEGAL_REQUEST:
593 if (sshdr.asc == 0x20 || /* Invalid command operation code */
594 sshdr.asc == 0x21 || /* Logical block address out of range */
595 sshdr.asc == 0x24 || /* Invalid field in cdb */
596 sshdr.asc == 0x26) { /* Parameter value invalid */
597 set_host_byte(scmd, DID_TARGET_FAILURE);
598 }
599 return SUCCESS;
600
601 default:
602 return SUCCESS;
603 }
604}
605EXPORT_SYMBOL_GPL(scsi_check_sense);
606
607static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
608{
609 struct scsi_host_template *sht = sdev->host->hostt;
610 struct scsi_device *tmp_sdev;
611
612 if (!sht->track_queue_depth ||
613 sdev->queue_depth >= sdev->max_queue_depth)
614 return;
615
616 if (time_before(jiffies,
617 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
618 return;
619
620 if (time_before(jiffies,
621 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
622 return;
623
624 /*
625 * Walk all devices of a target and do
626 * ramp up on them.
627 */
628 shost_for_each_device(tmp_sdev, sdev->host) {
629 if (tmp_sdev->channel != sdev->channel ||
630 tmp_sdev->id != sdev->id ||
631 tmp_sdev->queue_depth == sdev->max_queue_depth)
632 continue;
633
634 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
635 sdev->last_queue_ramp_up = jiffies;
636 }
637}
638
639static void scsi_handle_queue_full(struct scsi_device *sdev)
640{
641 struct scsi_host_template *sht = sdev->host->hostt;
642 struct scsi_device *tmp_sdev;
643
644 if (!sht->track_queue_depth)
645 return;
646
647 shost_for_each_device(tmp_sdev, sdev->host) {
648 if (tmp_sdev->channel != sdev->channel ||
649 tmp_sdev->id != sdev->id)
650 continue;
651 /*
652 * We do not know the number of commands that were at
653 * the device when we got the queue full so we start
654 * from the highest possible value and work our way down.
655 */
656 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
657 }
658}
659
660/**
661 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
662 * @scmd: SCSI cmd to examine.
663 *
664 * Notes:
665 * This is *only* called when we are examining the status of commands
666 * queued during error recovery. the main difference here is that we
667 * don't allow for the possibility of retries here, and we are a lot
668 * more restrictive about what we consider acceptable.
669 */
670static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
671{
672 /*
673 * first check the host byte, to see if there is anything in there
674 * that would indicate what we need to do.
675 */
676 if (host_byte(scmd->result) == DID_RESET) {
677 /*
678 * rats. we are already in the error handler, so we now
679 * get to try and figure out what to do next. if the sense
680 * is valid, we have a pretty good idea of what to do.
681 * if not, we mark it as FAILED.
682 */
683 return scsi_check_sense(scmd);
684 }
685 if (host_byte(scmd->result) != DID_OK)
686 return FAILED;
687
688 /*
689 * next, check the message byte.
690 */
691 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
692 return FAILED;
693
694 /*
695 * now, check the status byte to see if this indicates
696 * anything special.
697 */
698 switch (status_byte(scmd->result)) {
699 case GOOD:
700 scsi_handle_queue_ramp_up(scmd->device);
701 case COMMAND_TERMINATED:
702 return SUCCESS;
703 case CHECK_CONDITION:
704 return scsi_check_sense(scmd);
705 case CONDITION_GOOD:
706 case INTERMEDIATE_GOOD:
707 case INTERMEDIATE_C_GOOD:
708 /*
709 * who knows? FIXME(eric)
710 */
711 return SUCCESS;
712 case RESERVATION_CONFLICT:
713 if (scmd->cmnd[0] == TEST_UNIT_READY)
714 /* it is a success, we probed the device and
715 * found it */
716 return SUCCESS;
717 /* otherwise, we failed to send the command */
718 return FAILED;
719 case QUEUE_FULL:
720 scsi_handle_queue_full(scmd->device);
721 /* fall through */
722 case BUSY:
723 return NEEDS_RETRY;
724 default:
725 return FAILED;
726 }
727 return FAILED;
728}
729
730/**
731 * scsi_eh_done - Completion function for error handling.
732 * @scmd: Cmd that is done.
733 */
734static void scsi_eh_done(struct scsi_cmnd *scmd)
735{
736 struct completion *eh_action;
737
738 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
739 "%s result: %x\n", __func__, scmd->result));
740
741 eh_action = scmd->device->host->eh_action;
742 if (eh_action)
743 complete(eh_action);
744}
745
746/**
747 * scsi_try_host_reset - ask host adapter to reset itself
748 * @scmd: SCSI cmd to send host reset.
749 */
750static int scsi_try_host_reset(struct scsi_cmnd *scmd)
751{
752 unsigned long flags;
753 int rtn;
754 struct Scsi_Host *host = scmd->device->host;
755 struct scsi_host_template *hostt = host->hostt;
756
757 SCSI_LOG_ERROR_RECOVERY(3,
758 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
759
760 if (!hostt->eh_host_reset_handler)
761 return FAILED;
762
763 rtn = hostt->eh_host_reset_handler(scmd);
764
765 if (rtn == SUCCESS) {
766 if (!hostt->skip_settle_delay)
767 ssleep(HOST_RESET_SETTLE_TIME);
768 spin_lock_irqsave(host->host_lock, flags);
769 scsi_report_bus_reset(host, scmd_channel(scmd));
770 spin_unlock_irqrestore(host->host_lock, flags);
771 }
772
773 return rtn;
774}
775
776/**
777 * scsi_try_bus_reset - ask host to perform a bus reset
778 * @scmd: SCSI cmd to send bus reset.
779 */
780static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
781{
782 unsigned long flags;
783 int rtn;
784 struct Scsi_Host *host = scmd->device->host;
785 struct scsi_host_template *hostt = host->hostt;
786
787 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
788 "%s: Snd Bus RST\n", __func__));
789
790 if (!hostt->eh_bus_reset_handler)
791 return FAILED;
792
793 rtn = hostt->eh_bus_reset_handler(scmd);
794
795 if (rtn == SUCCESS) {
796 if (!hostt->skip_settle_delay)
797 ssleep(BUS_RESET_SETTLE_TIME);
798 spin_lock_irqsave(host->host_lock, flags);
799 scsi_report_bus_reset(host, scmd_channel(scmd));
800 spin_unlock_irqrestore(host->host_lock, flags);
801 }
802
803 return rtn;
804}
805
806static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
807{
808 sdev->was_reset = 1;
809 sdev->expecting_cc_ua = 1;
810}
811
812/**
813 * scsi_try_target_reset - Ask host to perform a target reset
814 * @scmd: SCSI cmd used to send a target reset
815 *
816 * Notes:
817 * There is no timeout for this operation. if this operation is
818 * unreliable for a given host, then the host itself needs to put a
819 * timer on it, and set the host back to a consistent state prior to
820 * returning.
821 */
822static int scsi_try_target_reset(struct scsi_cmnd *scmd)
823{
824 unsigned long flags;
825 int rtn;
826 struct Scsi_Host *host = scmd->device->host;
827 struct scsi_host_template *hostt = host->hostt;
828
829 if (!hostt->eh_target_reset_handler)
830 return FAILED;
831
832 rtn = hostt->eh_target_reset_handler(scmd);
833 if (rtn == SUCCESS) {
834 spin_lock_irqsave(host->host_lock, flags);
835 __starget_for_each_device(scsi_target(scmd->device), NULL,
836 __scsi_report_device_reset);
837 spin_unlock_irqrestore(host->host_lock, flags);
838 }
839
840 return rtn;
841}
842
843/**
844 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
845 * @scmd: SCSI cmd used to send BDR
846 *
847 * Notes:
848 * There is no timeout for this operation. if this operation is
849 * unreliable for a given host, then the host itself needs to put a
850 * timer on it, and set the host back to a consistent state prior to
851 * returning.
852 */
853static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
854{
855 int rtn;
856 struct scsi_host_template *hostt = scmd->device->host->hostt;
857
858 if (!hostt->eh_device_reset_handler)
859 return FAILED;
860
861 rtn = hostt->eh_device_reset_handler(scmd);
862 if (rtn == SUCCESS)
863 __scsi_report_device_reset(scmd->device, NULL);
864 return rtn;
865}
866
867/**
868 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
869 * @hostt: SCSI driver host template
870 * @scmd: SCSI cmd used to send a target reset
871 *
872 * Return value:
873 * SUCCESS, FAILED, or FAST_IO_FAIL
874 *
875 * Notes:
876 * SUCCESS does not necessarily indicate that the command
877 * has been aborted; it only indicates that the LLDDs
878 * has cleared all references to that command.
879 * LLDDs should return FAILED only if an abort was required
880 * but could not be executed. LLDDs should return FAST_IO_FAIL
881 * if the device is temporarily unavailable (eg due to a
882 * link down on FibreChannel)
883 */
884static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
885 struct scsi_cmnd *scmd)
886{
887 if (!hostt->eh_abort_handler)
888 return FAILED;
889
890 return hostt->eh_abort_handler(scmd);
891}
892
893static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
894{
895 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
896 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
897 if (scsi_try_target_reset(scmd) != SUCCESS)
898 if (scsi_try_bus_reset(scmd) != SUCCESS)
899 scsi_try_host_reset(scmd);
900}
901
902/**
903 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
904 * @scmd: SCSI command structure to hijack
905 * @ses: structure to save restore information
906 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
907 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
908 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
909 *
910 * This function is used to save a scsi command information before re-execution
911 * as part of the error recovery process. If @sense_bytes is 0 the command
912 * sent must be one that does not transfer any data. If @sense_bytes != 0
913 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
914 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
915 */
916void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
917 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
918{
919 struct scsi_device *sdev = scmd->device;
920
921 /*
922 * We need saved copies of a number of fields - this is because
923 * error handling may need to overwrite these with different values
924 * to run different commands, and once error handling is complete,
925 * we will need to restore these values prior to running the actual
926 * command.
927 */
928 ses->cmd_len = scmd->cmd_len;
929 ses->cmnd = scmd->cmnd;
930 ses->data_direction = scmd->sc_data_direction;
931 ses->sdb = scmd->sdb;
932 ses->next_rq = scmd->request->next_rq;
933 ses->result = scmd->result;
934 ses->underflow = scmd->underflow;
935 ses->prot_op = scmd->prot_op;
936
937 scmd->prot_op = SCSI_PROT_NORMAL;
938 scmd->eh_eflags = 0;
939 scmd->cmnd = ses->eh_cmnd;
940 memset(scmd->cmnd, 0, BLK_MAX_CDB);
941 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
942 scmd->request->next_rq = NULL;
943 scmd->result = 0;
944
945 if (sense_bytes) {
946 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
947 sense_bytes);
948 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
949 scmd->sdb.length);
950 scmd->sdb.table.sgl = &ses->sense_sgl;
951 scmd->sc_data_direction = DMA_FROM_DEVICE;
952 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
953 scmd->cmnd[0] = REQUEST_SENSE;
954 scmd->cmnd[4] = scmd->sdb.length;
955 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
956 } else {
957 scmd->sc_data_direction = DMA_NONE;
958 if (cmnd) {
959 BUG_ON(cmnd_size > BLK_MAX_CDB);
960 memcpy(scmd->cmnd, cmnd, cmnd_size);
961 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
962 }
963 }
964
965 scmd->underflow = 0;
966
967 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
968 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
969 (sdev->lun << 5 & 0xe0);
970
971 /*
972 * Zero the sense buffer. The scsi spec mandates that any
973 * untransferred sense data should be interpreted as being zero.
974 */
975 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
976}
977EXPORT_SYMBOL(scsi_eh_prep_cmnd);
978
979/**
980 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
981 * @scmd: SCSI command structure to restore
982 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
983 *
984 * Undo any damage done by above scsi_eh_prep_cmnd().
985 */
986void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
987{
988 /*
989 * Restore original data
990 */
991 scmd->cmd_len = ses->cmd_len;
992 scmd->cmnd = ses->cmnd;
993 scmd->sc_data_direction = ses->data_direction;
994 scmd->sdb = ses->sdb;
995 scmd->request->next_rq = ses->next_rq;
996 scmd->result = ses->result;
997 scmd->underflow = ses->underflow;
998 scmd->prot_op = ses->prot_op;
999}
1000EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1001
1002/**
1003 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1004 * @scmd: SCSI command structure to hijack
1005 * @cmnd: CDB to send
1006 * @cmnd_size: size in bytes of @cmnd
1007 * @timeout: timeout for this request
1008 * @sense_bytes: size of sense data to copy or 0
1009 *
1010 * This function is used to send a scsi command down to a target device
1011 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1012 *
1013 * Return value:
1014 * SUCCESS or FAILED or NEEDS_RETRY
1015 */
1016static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1017 int cmnd_size, int timeout, unsigned sense_bytes)
1018{
1019 struct scsi_device *sdev = scmd->device;
1020 struct Scsi_Host *shost = sdev->host;
1021 DECLARE_COMPLETION_ONSTACK(done);
1022 unsigned long timeleft = timeout;
1023 struct scsi_eh_save ses;
1024 const unsigned long stall_for = msecs_to_jiffies(100);
1025 int rtn;
1026
1027retry:
1028 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1029 shost->eh_action = &done;
1030
1031 scsi_log_send(scmd);
1032 scmd->scsi_done = scsi_eh_done;
1033 rtn = shost->hostt->queuecommand(shost, scmd);
1034 if (rtn) {
1035 if (timeleft > stall_for) {
1036 scsi_eh_restore_cmnd(scmd, &ses);
1037 timeleft -= stall_for;
1038 msleep(jiffies_to_msecs(stall_for));
1039 goto retry;
1040 }
1041 /* signal not to enter either branch of the if () below */
1042 timeleft = 0;
1043 rtn = FAILED;
1044 } else {
1045 timeleft = wait_for_completion_timeout(&done, timeout);
1046 rtn = SUCCESS;
1047 }
1048
1049 shost->eh_action = NULL;
1050
1051 scsi_log_completion(scmd, rtn);
1052
1053 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1054 "%s timeleft: %ld\n",
1055 __func__, timeleft));
1056
1057 /*
1058 * If there is time left scsi_eh_done got called, and we will examine
1059 * the actual status codes to see whether the command actually did
1060 * complete normally, else if we have a zero return and no time left,
1061 * the command must still be pending, so abort it and return FAILED.
1062 * If we never actually managed to issue the command, because
1063 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1064 * value above (so don't execute either branch of the if)
1065 */
1066 if (timeleft) {
1067 rtn = scsi_eh_completed_normally(scmd);
1068 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1069 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1070
1071 switch (rtn) {
1072 case SUCCESS:
1073 case NEEDS_RETRY:
1074 case FAILED:
1075 break;
1076 case ADD_TO_MLQUEUE:
1077 rtn = NEEDS_RETRY;
1078 break;
1079 default:
1080 rtn = FAILED;
1081 break;
1082 }
1083 } else if (rtn != FAILED) {
1084 scsi_abort_eh_cmnd(scmd);
1085 rtn = FAILED;
1086 }
1087
1088 scsi_eh_restore_cmnd(scmd, &ses);
1089
1090 return rtn;
1091}
1092
1093/**
1094 * scsi_request_sense - Request sense data from a particular target.
1095 * @scmd: SCSI cmd for request sense.
1096 *
1097 * Notes:
1098 * Some hosts automatically obtain this information, others require
1099 * that we obtain it on our own. This function will *not* return until
1100 * the command either times out, or it completes.
1101 */
1102static int scsi_request_sense(struct scsi_cmnd *scmd)
1103{
1104 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1105}
1106
1107static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1108{
1109 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1110 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1111 if (sdrv->eh_action)
1112 rtn = sdrv->eh_action(scmd, rtn);
1113 }
1114 return rtn;
1115}
1116
1117/**
1118 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1119 * @scmd: Original SCSI cmd that eh has finished.
1120 * @done_q: Queue for processed commands.
1121 *
1122 * Notes:
1123 * We don't want to use the normal command completion while we are are
1124 * still handling errors - it may cause other commands to be queued,
1125 * and that would disturb what we are doing. Thus we really want to
1126 * keep a list of pending commands for final completion, and once we
1127 * are ready to leave error handling we handle completion for real.
1128 */
1129void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1130{
1131 scmd->eh_eflags = 0;
1132 list_move_tail(&scmd->eh_entry, done_q);
1133}
1134EXPORT_SYMBOL(scsi_eh_finish_cmd);
1135
1136/**
1137 * scsi_eh_get_sense - Get device sense data.
1138 * @work_q: Queue of commands to process.
1139 * @done_q: Queue of processed commands.
1140 *
1141 * Description:
1142 * See if we need to request sense information. if so, then get it
1143 * now, so we have a better idea of what to do.
1144 *
1145 * Notes:
1146 * This has the unfortunate side effect that if a shost adapter does
1147 * not automatically request sense information, we end up shutting
1148 * it down before we request it.
1149 *
1150 * All drivers should request sense information internally these days,
1151 * so for now all I have to say is tough noogies if you end up in here.
1152 *
1153 * XXX: Long term this code should go away, but that needs an audit of
1154 * all LLDDs first.
1155 */
1156int scsi_eh_get_sense(struct list_head *work_q,
1157 struct list_head *done_q)
1158{
1159 struct scsi_cmnd *scmd, *next;
1160 struct Scsi_Host *shost;
1161 int rtn;
1162
1163 /*
1164 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1165 * should not get sense.
1166 */
1167 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1168 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1169 (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1170 SCSI_SENSE_VALID(scmd))
1171 continue;
1172
1173 shost = scmd->device->host;
1174 if (scsi_host_eh_past_deadline(shost)) {
1175 SCSI_LOG_ERROR_RECOVERY(3,
1176 scmd_printk(KERN_INFO, scmd,
1177 "%s: skip request sense, past eh deadline\n",
1178 current->comm));
1179 break;
1180 }
1181 if (status_byte(scmd->result) != CHECK_CONDITION)
1182 /*
1183 * don't request sense if there's no check condition
1184 * status because the error we're processing isn't one
1185 * that has a sense code (and some devices get
1186 * confused by sense requests out of the blue)
1187 */
1188 continue;
1189
1190 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1191 "%s: requesting sense\n",
1192 current->comm));
1193 rtn = scsi_request_sense(scmd);
1194 if (rtn != SUCCESS)
1195 continue;
1196
1197 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1198 "sense requested, result %x\n", scmd->result));
1199 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1200
1201 rtn = scsi_decide_disposition(scmd);
1202
1203 /*
1204 * if the result was normal, then just pass it along to the
1205 * upper level.
1206 */
1207 if (rtn == SUCCESS)
1208 /* we don't want this command reissued, just
1209 * finished with the sense data, so set
1210 * retries to the max allowed to ensure it
1211 * won't get reissued */
1212 scmd->retries = scmd->allowed;
1213 else if (rtn != NEEDS_RETRY)
1214 continue;
1215
1216 scsi_eh_finish_cmd(scmd, done_q);
1217 }
1218
1219 return list_empty(work_q);
1220}
1221EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1222
1223/**
1224 * scsi_eh_tur - Send TUR to device.
1225 * @scmd: &scsi_cmnd to send TUR
1226 *
1227 * Return value:
1228 * 0 - Device is ready. 1 - Device NOT ready.
1229 */
1230static int scsi_eh_tur(struct scsi_cmnd *scmd)
1231{
1232 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1233 int retry_cnt = 1, rtn;
1234
1235retry_tur:
1236 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1237 scmd->device->eh_timeout, 0);
1238
1239 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1240 "%s return: %x\n", __func__, rtn));
1241
1242 switch (rtn) {
1243 case NEEDS_RETRY:
1244 if (retry_cnt--)
1245 goto retry_tur;
1246 /*FALLTHRU*/
1247 case SUCCESS:
1248 return 0;
1249 default:
1250 return 1;
1251 }
1252}
1253
1254/**
1255 * scsi_eh_test_devices - check if devices are responding from error recovery.
1256 * @cmd_list: scsi commands in error recovery.
1257 * @work_q: queue for commands which still need more error recovery
1258 * @done_q: queue for commands which are finished
1259 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1260 *
1261 * Decription:
1262 * Tests if devices are in a working state. Commands to devices now in
1263 * a working state are sent to the done_q while commands to devices which
1264 * are still failing to respond are returned to the work_q for more
1265 * processing.
1266 **/
1267static int scsi_eh_test_devices(struct list_head *cmd_list,
1268 struct list_head *work_q,
1269 struct list_head *done_q, int try_stu)
1270{
1271 struct scsi_cmnd *scmd, *next;
1272 struct scsi_device *sdev;
1273 int finish_cmds;
1274
1275 while (!list_empty(cmd_list)) {
1276 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1277 sdev = scmd->device;
1278
1279 if (!try_stu) {
1280 if (scsi_host_eh_past_deadline(sdev->host)) {
1281 /* Push items back onto work_q */
1282 list_splice_init(cmd_list, work_q);
1283 SCSI_LOG_ERROR_RECOVERY(3,
1284 sdev_printk(KERN_INFO, sdev,
1285 "%s: skip test device, past eh deadline",
1286 current->comm));
1287 break;
1288 }
1289 }
1290
1291 finish_cmds = !scsi_device_online(scmd->device) ||
1292 (try_stu && !scsi_eh_try_stu(scmd) &&
1293 !scsi_eh_tur(scmd)) ||
1294 !scsi_eh_tur(scmd);
1295
1296 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1297 if (scmd->device == sdev) {
1298 if (finish_cmds &&
1299 (try_stu ||
1300 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1301 scsi_eh_finish_cmd(scmd, done_q);
1302 else
1303 list_move_tail(&scmd->eh_entry, work_q);
1304 }
1305 }
1306 return list_empty(work_q);
1307}
1308
1309
1310/**
1311 * scsi_eh_abort_cmds - abort pending commands.
1312 * @work_q: &list_head for pending commands.
1313 * @done_q: &list_head for processed commands.
1314 *
1315 * Decription:
1316 * Try and see whether or not it makes sense to try and abort the
1317 * running command. This only works out to be the case if we have one
1318 * command that has timed out. If the command simply failed, it makes
1319 * no sense to try and abort the command, since as far as the shost
1320 * adapter is concerned, it isn't running.
1321 */
1322static int scsi_eh_abort_cmds(struct list_head *work_q,
1323 struct list_head *done_q)
1324{
1325 struct scsi_cmnd *scmd, *next;
1326 LIST_HEAD(check_list);
1327 int rtn;
1328 struct Scsi_Host *shost;
1329
1330 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1331 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1332 continue;
1333 shost = scmd->device->host;
1334 if (scsi_host_eh_past_deadline(shost)) {
1335 list_splice_init(&check_list, work_q);
1336 SCSI_LOG_ERROR_RECOVERY(3,
1337 scmd_printk(KERN_INFO, scmd,
1338 "%s: skip aborting cmd, past eh deadline\n",
1339 current->comm));
1340 return list_empty(work_q);
1341 }
1342 SCSI_LOG_ERROR_RECOVERY(3,
1343 scmd_printk(KERN_INFO, scmd,
1344 "%s: aborting cmd\n", current->comm));
1345 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1346 if (rtn == FAILED) {
1347 SCSI_LOG_ERROR_RECOVERY(3,
1348 scmd_printk(KERN_INFO, scmd,
1349 "%s: aborting cmd failed\n",
1350 current->comm));
1351 list_splice_init(&check_list, work_q);
1352 return list_empty(work_q);
1353 }
1354 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1355 if (rtn == FAST_IO_FAIL)
1356 scsi_eh_finish_cmd(scmd, done_q);
1357 else
1358 list_move_tail(&scmd->eh_entry, &check_list);
1359 }
1360
1361 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1362}
1363
1364/**
1365 * scsi_eh_try_stu - Send START_UNIT to device.
1366 * @scmd: &scsi_cmnd to send START_UNIT
1367 *
1368 * Return value:
1369 * 0 - Device is ready. 1 - Device NOT ready.
1370 */
1371static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1372{
1373 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1374
1375 if (scmd->device->allow_restart) {
1376 int i, rtn = NEEDS_RETRY;
1377
1378 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1379 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1380
1381 if (rtn == SUCCESS)
1382 return 0;
1383 }
1384
1385 return 1;
1386}
1387
1388 /**
1389 * scsi_eh_stu - send START_UNIT if needed
1390 * @shost: &scsi host being recovered.
1391 * @work_q: &list_head for pending commands.
1392 * @done_q: &list_head for processed commands.
1393 *
1394 * Notes:
1395 * If commands are failing due to not ready, initializing command required,
1396 * try revalidating the device, which will end up sending a start unit.
1397 */
1398static int scsi_eh_stu(struct Scsi_Host *shost,
1399 struct list_head *work_q,
1400 struct list_head *done_q)
1401{
1402 struct scsi_cmnd *scmd, *stu_scmd, *next;
1403 struct scsi_device *sdev;
1404
1405 shost_for_each_device(sdev, shost) {
1406 if (scsi_host_eh_past_deadline(shost)) {
1407 SCSI_LOG_ERROR_RECOVERY(3,
1408 sdev_printk(KERN_INFO, sdev,
1409 "%s: skip START_UNIT, past eh deadline\n",
1410 current->comm));
1411 break;
1412 }
1413 stu_scmd = NULL;
1414 list_for_each_entry(scmd, work_q, eh_entry)
1415 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1416 scsi_check_sense(scmd) == FAILED ) {
1417 stu_scmd = scmd;
1418 break;
1419 }
1420
1421 if (!stu_scmd)
1422 continue;
1423
1424 SCSI_LOG_ERROR_RECOVERY(3,
1425 sdev_printk(KERN_INFO, sdev,
1426 "%s: Sending START_UNIT\n",
1427 current->comm));
1428
1429 if (!scsi_eh_try_stu(stu_scmd)) {
1430 if (!scsi_device_online(sdev) ||
1431 !scsi_eh_tur(stu_scmd)) {
1432 list_for_each_entry_safe(scmd, next,
1433 work_q, eh_entry) {
1434 if (scmd->device == sdev &&
1435 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1436 scsi_eh_finish_cmd(scmd, done_q);
1437 }
1438 }
1439 } else {
1440 SCSI_LOG_ERROR_RECOVERY(3,
1441 sdev_printk(KERN_INFO, sdev,
1442 "%s: START_UNIT failed\n",
1443 current->comm));
1444 }
1445 }
1446
1447 return list_empty(work_q);
1448}
1449
1450
1451/**
1452 * scsi_eh_bus_device_reset - send bdr if needed
1453 * @shost: scsi host being recovered.
1454 * @work_q: &list_head for pending commands.
1455 * @done_q: &list_head for processed commands.
1456 *
1457 * Notes:
1458 * Try a bus device reset. Still, look to see whether we have multiple
1459 * devices that are jammed or not - if we have multiple devices, it
1460 * makes no sense to try bus_device_reset - we really would need to try
1461 * a bus_reset instead.
1462 */
1463static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1464 struct list_head *work_q,
1465 struct list_head *done_q)
1466{
1467 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1468 struct scsi_device *sdev;
1469 int rtn;
1470
1471 shost_for_each_device(sdev, shost) {
1472 if (scsi_host_eh_past_deadline(shost)) {
1473 SCSI_LOG_ERROR_RECOVERY(3,
1474 sdev_printk(KERN_INFO, sdev,
1475 "%s: skip BDR, past eh deadline\n",
1476 current->comm));
1477 break;
1478 }
1479 bdr_scmd = NULL;
1480 list_for_each_entry(scmd, work_q, eh_entry)
1481 if (scmd->device == sdev) {
1482 bdr_scmd = scmd;
1483 break;
1484 }
1485
1486 if (!bdr_scmd)
1487 continue;
1488
1489 SCSI_LOG_ERROR_RECOVERY(3,
1490 sdev_printk(KERN_INFO, sdev,
1491 "%s: Sending BDR\n", current->comm));
1492 rtn = scsi_try_bus_device_reset(bdr_scmd);
1493 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1494 if (!scsi_device_online(sdev) ||
1495 rtn == FAST_IO_FAIL ||
1496 !scsi_eh_tur(bdr_scmd)) {
1497 list_for_each_entry_safe(scmd, next,
1498 work_q, eh_entry) {
1499 if (scmd->device == sdev &&
1500 scsi_eh_action(scmd, rtn) != FAILED)
1501 scsi_eh_finish_cmd(scmd,
1502 done_q);
1503 }
1504 }
1505 } else {
1506 SCSI_LOG_ERROR_RECOVERY(3,
1507 sdev_printk(KERN_INFO, sdev,
1508 "%s: BDR failed\n", current->comm));
1509 }
1510 }
1511
1512 return list_empty(work_q);
1513}
1514
1515/**
1516 * scsi_eh_target_reset - send target reset if needed
1517 * @shost: scsi host being recovered.
1518 * @work_q: &list_head for pending commands.
1519 * @done_q: &list_head for processed commands.
1520 *
1521 * Notes:
1522 * Try a target reset.
1523 */
1524static int scsi_eh_target_reset(struct Scsi_Host *shost,
1525 struct list_head *work_q,
1526 struct list_head *done_q)
1527{
1528 LIST_HEAD(tmp_list);
1529 LIST_HEAD(check_list);
1530
1531 list_splice_init(work_q, &tmp_list);
1532
1533 while (!list_empty(&tmp_list)) {
1534 struct scsi_cmnd *next, *scmd;
1535 int rtn;
1536 unsigned int id;
1537
1538 if (scsi_host_eh_past_deadline(shost)) {
1539 /* push back on work queue for further processing */
1540 list_splice_init(&check_list, work_q);
1541 list_splice_init(&tmp_list, work_q);
1542 SCSI_LOG_ERROR_RECOVERY(3,
1543 shost_printk(KERN_INFO, shost,
1544 "%s: Skip target reset, past eh deadline\n",
1545 current->comm));
1546 return list_empty(work_q);
1547 }
1548
1549 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1550 id = scmd_id(scmd);
1551
1552 SCSI_LOG_ERROR_RECOVERY(3,
1553 shost_printk(KERN_INFO, shost,
1554 "%s: Sending target reset to target %d\n",
1555 current->comm, id));
1556 rtn = scsi_try_target_reset(scmd);
1557 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1558 SCSI_LOG_ERROR_RECOVERY(3,
1559 shost_printk(KERN_INFO, shost,
1560 "%s: Target reset failed"
1561 " target: %d\n",
1562 current->comm, id));
1563 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1564 if (scmd_id(scmd) != id)
1565 continue;
1566
1567 if (rtn == SUCCESS)
1568 list_move_tail(&scmd->eh_entry, &check_list);
1569 else if (rtn == FAST_IO_FAIL)
1570 scsi_eh_finish_cmd(scmd, done_q);
1571 else
1572 /* push back on work queue for further processing */
1573 list_move(&scmd->eh_entry, work_q);
1574 }
1575 }
1576
1577 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1578}
1579
1580/**
1581 * scsi_eh_bus_reset - send a bus reset
1582 * @shost: &scsi host being recovered.
1583 * @work_q: &list_head for pending commands.
1584 * @done_q: &list_head for processed commands.
1585 */
1586static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1587 struct list_head *work_q,
1588 struct list_head *done_q)
1589{
1590 struct scsi_cmnd *scmd, *chan_scmd, *next;
1591 LIST_HEAD(check_list);
1592 unsigned int channel;
1593 int rtn;
1594
1595 /*
1596 * we really want to loop over the various channels, and do this on
1597 * a channel by channel basis. we should also check to see if any
1598 * of the failed commands are on soft_reset devices, and if so, skip
1599 * the reset.
1600 */
1601
1602 for (channel = 0; channel <= shost->max_channel; channel++) {
1603 if (scsi_host_eh_past_deadline(shost)) {
1604 list_splice_init(&check_list, work_q);
1605 SCSI_LOG_ERROR_RECOVERY(3,
1606 shost_printk(KERN_INFO, shost,
1607 "%s: skip BRST, past eh deadline\n",
1608 current->comm));
1609 return list_empty(work_q);
1610 }
1611
1612 chan_scmd = NULL;
1613 list_for_each_entry(scmd, work_q, eh_entry) {
1614 if (channel == scmd_channel(scmd)) {
1615 chan_scmd = scmd;
1616 break;
1617 /*
1618 * FIXME add back in some support for
1619 * soft_reset devices.
1620 */
1621 }
1622 }
1623
1624 if (!chan_scmd)
1625 continue;
1626 SCSI_LOG_ERROR_RECOVERY(3,
1627 shost_printk(KERN_INFO, shost,
1628 "%s: Sending BRST chan: %d\n",
1629 current->comm, channel));
1630 rtn = scsi_try_bus_reset(chan_scmd);
1631 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1632 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1633 if (channel == scmd_channel(scmd)) {
1634 if (rtn == FAST_IO_FAIL)
1635 scsi_eh_finish_cmd(scmd,
1636 done_q);
1637 else
1638 list_move_tail(&scmd->eh_entry,
1639 &check_list);
1640 }
1641 }
1642 } else {
1643 SCSI_LOG_ERROR_RECOVERY(3,
1644 shost_printk(KERN_INFO, shost,
1645 "%s: BRST failed chan: %d\n",
1646 current->comm, channel));
1647 }
1648 }
1649 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1650}
1651
1652/**
1653 * scsi_eh_host_reset - send a host reset
1654 * @shost: host to be reset.
1655 * @work_q: &list_head for pending commands.
1656 * @done_q: &list_head for processed commands.
1657 */
1658static int scsi_eh_host_reset(struct Scsi_Host *shost,
1659 struct list_head *work_q,
1660 struct list_head *done_q)
1661{
1662 struct scsi_cmnd *scmd, *next;
1663 LIST_HEAD(check_list);
1664 int rtn;
1665
1666 if (!list_empty(work_q)) {
1667 scmd = list_entry(work_q->next,
1668 struct scsi_cmnd, eh_entry);
1669
1670 SCSI_LOG_ERROR_RECOVERY(3,
1671 shost_printk(KERN_INFO, shost,
1672 "%s: Sending HRST\n",
1673 current->comm));
1674
1675 rtn = scsi_try_host_reset(scmd);
1676 if (rtn == SUCCESS) {
1677 list_splice_init(work_q, &check_list);
1678 } else if (rtn == FAST_IO_FAIL) {
1679 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1680 scsi_eh_finish_cmd(scmd, done_q);
1681 }
1682 } else {
1683 SCSI_LOG_ERROR_RECOVERY(3,
1684 shost_printk(KERN_INFO, shost,
1685 "%s: HRST failed\n",
1686 current->comm));
1687 }
1688 }
1689 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1690}
1691
1692/**
1693 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1694 * @work_q: &list_head for pending commands.
1695 * @done_q: &list_head for processed commands.
1696 */
1697static void scsi_eh_offline_sdevs(struct list_head *work_q,
1698 struct list_head *done_q)
1699{
1700 struct scsi_cmnd *scmd, *next;
1701
1702 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1703 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1704 "not ready after error recovery\n");
1705 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1706 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1707 /*
1708 * FIXME: Handle lost cmds.
1709 */
1710 }
1711 scsi_eh_finish_cmd(scmd, done_q);
1712 }
1713 return;
1714}
1715
1716/**
1717 * scsi_noretry_cmd - determine if command should be failed fast
1718 * @scmd: SCSI cmd to examine.
1719 */
1720int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1721{
1722 switch (host_byte(scmd->result)) {
1723 case DID_OK:
1724 break;
1725 case DID_TIME_OUT:
1726 goto check_type;
1727 case DID_BUS_BUSY:
1728 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1729 case DID_PARITY:
1730 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1731 case DID_ERROR:
1732 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1733 status_byte(scmd->result) == RESERVATION_CONFLICT)
1734 return 0;
1735 /* fall through */
1736 case DID_SOFT_ERROR:
1737 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1738 }
1739
1740 if (status_byte(scmd->result) != CHECK_CONDITION)
1741 return 0;
1742
1743check_type:
1744 /*
1745 * assume caller has checked sense and determined
1746 * the check condition was retryable.
1747 */
1748 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1749 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1750 return 1;
1751 else
1752 return 0;
1753}
1754
1755/**
1756 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1757 * @scmd: SCSI cmd to examine.
1758 *
1759 * Notes:
1760 * This is *only* called when we are examining the status after sending
1761 * out the actual data command. any commands that are queued for error
1762 * recovery (e.g. test_unit_ready) do *not* come through here.
1763 *
1764 * When this routine returns failed, it means the error handler thread
1765 * is woken. In cases where the error code indicates an error that
1766 * doesn't require the error handler read (i.e. we don't need to
1767 * abort/reset), this function should return SUCCESS.
1768 */
1769int scsi_decide_disposition(struct scsi_cmnd *scmd)
1770{
1771 int rtn;
1772
1773 /*
1774 * if the device is offline, then we clearly just pass the result back
1775 * up to the top level.
1776 */
1777 if (!scsi_device_online(scmd->device)) {
1778 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1779 "%s: device offline - report as SUCCESS\n", __func__));
1780 return SUCCESS;
1781 }
1782
1783 /*
1784 * first check the host byte, to see if there is anything in there
1785 * that would indicate what we need to do.
1786 */
1787 switch (host_byte(scmd->result)) {
1788 case DID_PASSTHROUGH:
1789 /*
1790 * no matter what, pass this through to the upper layer.
1791 * nuke this special code so that it looks like we are saying
1792 * did_ok.
1793 */
1794 scmd->result &= 0xff00ffff;
1795 return SUCCESS;
1796 case DID_OK:
1797 /*
1798 * looks good. drop through, and check the next byte.
1799 */
1800 break;
1801 case DID_ABORT:
1802 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1803 set_host_byte(scmd, DID_TIME_OUT);
1804 return SUCCESS;
1805 }
1806 case DID_NO_CONNECT:
1807 case DID_BAD_TARGET:
1808 /*
1809 * note - this means that we just report the status back
1810 * to the top level driver, not that we actually think
1811 * that it indicates SUCCESS.
1812 */
1813 return SUCCESS;
1814 /*
1815 * when the low level driver returns did_soft_error,
1816 * it is responsible for keeping an internal retry counter
1817 * in order to avoid endless loops (db)
1818 *
1819 * actually this is a bug in this function here. we should
1820 * be mindful of the maximum number of retries specified
1821 * and not get stuck in a loop.
1822 */
1823 case DID_SOFT_ERROR:
1824 goto maybe_retry;
1825 case DID_IMM_RETRY:
1826 return NEEDS_RETRY;
1827
1828 case DID_REQUEUE:
1829 return ADD_TO_MLQUEUE;
1830 case DID_TRANSPORT_DISRUPTED:
1831 /*
1832 * LLD/transport was disrupted during processing of the IO.
1833 * The transport class is now blocked/blocking,
1834 * and the transport will decide what to do with the IO
1835 * based on its timers and recovery capablilities if
1836 * there are enough retries.
1837 */
1838 goto maybe_retry;
1839 case DID_TRANSPORT_FAILFAST:
1840 /*
1841 * The transport decided to failfast the IO (most likely
1842 * the fast io fail tmo fired), so send IO directly upwards.
1843 */
1844 return SUCCESS;
1845 case DID_ERROR:
1846 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1847 status_byte(scmd->result) == RESERVATION_CONFLICT)
1848 /*
1849 * execute reservation conflict processing code
1850 * lower down
1851 */
1852 break;
1853 /* fallthrough */
1854 case DID_BUS_BUSY:
1855 case DID_PARITY:
1856 goto maybe_retry;
1857 case DID_TIME_OUT:
1858 /*
1859 * when we scan the bus, we get timeout messages for
1860 * these commands if there is no device available.
1861 * other hosts report did_no_connect for the same thing.
1862 */
1863 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1864 scmd->cmnd[0] == INQUIRY)) {
1865 return SUCCESS;
1866 } else {
1867 return FAILED;
1868 }
1869 case DID_RESET:
1870 return SUCCESS;
1871 default:
1872 return FAILED;
1873 }
1874
1875 /*
1876 * next, check the message byte.
1877 */
1878 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1879 return FAILED;
1880
1881 /*
1882 * check the status byte to see if this indicates anything special.
1883 */
1884 switch (status_byte(scmd->result)) {
1885 case QUEUE_FULL:
1886 scsi_handle_queue_full(scmd->device);
1887 /*
1888 * the case of trying to send too many commands to a
1889 * tagged queueing device.
1890 */
1891 case BUSY:
1892 /*
1893 * device can't talk to us at the moment. Should only
1894 * occur (SAM-3) when the task queue is empty, so will cause
1895 * the empty queue handling to trigger a stall in the
1896 * device.
1897 */
1898 return ADD_TO_MLQUEUE;
1899 case GOOD:
1900 if (scmd->cmnd[0] == REPORT_LUNS)
1901 scmd->device->sdev_target->expecting_lun_change = 0;
1902 scsi_handle_queue_ramp_up(scmd->device);
1903 case COMMAND_TERMINATED:
1904 return SUCCESS;
1905 case TASK_ABORTED:
1906 goto maybe_retry;
1907 case CHECK_CONDITION:
1908 rtn = scsi_check_sense(scmd);
1909 if (rtn == NEEDS_RETRY)
1910 goto maybe_retry;
1911 /* if rtn == FAILED, we have no sense information;
1912 * returning FAILED will wake the error handler thread
1913 * to collect the sense and redo the decide
1914 * disposition */
1915 return rtn;
1916 case CONDITION_GOOD:
1917 case INTERMEDIATE_GOOD:
1918 case INTERMEDIATE_C_GOOD:
1919 case ACA_ACTIVE:
1920 /*
1921 * who knows? FIXME(eric)
1922 */
1923 return SUCCESS;
1924
1925 case RESERVATION_CONFLICT:
1926 sdev_printk(KERN_INFO, scmd->device,
1927 "reservation conflict\n");
1928 set_host_byte(scmd, DID_NEXUS_FAILURE);
1929 return SUCCESS; /* causes immediate i/o error */
1930 default:
1931 return FAILED;
1932 }
1933 return FAILED;
1934
1935 maybe_retry:
1936
1937 /* we requeue for retry because the error was retryable, and
1938 * the request was not marked fast fail. Note that above,
1939 * even if the request is marked fast fail, we still requeue
1940 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1941 if ((++scmd->retries) <= scmd->allowed
1942 && !scsi_noretry_cmd(scmd)) {
1943 return NEEDS_RETRY;
1944 } else {
1945 /*
1946 * no more retries - report this one back to upper level.
1947 */
1948 return SUCCESS;
1949 }
1950}
1951
1952static void eh_lock_door_done(struct request *req, int uptodate)
1953{
1954 __blk_put_request(req->q, req);
1955}
1956
1957/**
1958 * scsi_eh_lock_door - Prevent medium removal for the specified device
1959 * @sdev: SCSI device to prevent medium removal
1960 *
1961 * Locking:
1962 * We must be called from process context.
1963 *
1964 * Notes:
1965 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1966 * head of the devices request queue, and continue.
1967 */
1968static void scsi_eh_lock_door(struct scsi_device *sdev)
1969{
1970 struct request *req;
1971
1972 /*
1973 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1974 * request becomes available
1975 */
1976 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1977 if (IS_ERR(req))
1978 return;
1979
1980 blk_rq_set_block_pc(req);
1981
1982 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1983 req->cmd[1] = 0;
1984 req->cmd[2] = 0;
1985 req->cmd[3] = 0;
1986 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1987 req->cmd[5] = 0;
1988
1989 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1990
1991 req->rq_flags |= RQF_QUIET;
1992 req->timeout = 10 * HZ;
1993 req->retries = 5;
1994
1995 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1996}
1997
1998/**
1999 * scsi_restart_operations - restart io operations to the specified host.
2000 * @shost: Host we are restarting.
2001 *
2002 * Notes:
2003 * When we entered the error handler, we blocked all further i/o to
2004 * this device. we need to 'reverse' this process.
2005 */
2006static void scsi_restart_operations(struct Scsi_Host *shost)
2007{
2008 struct scsi_device *sdev;
2009 unsigned long flags;
2010
2011 /*
2012 * If the door was locked, we need to insert a door lock request
2013 * onto the head of the SCSI request queue for the device. There
2014 * is no point trying to lock the door of an off-line device.
2015 */
2016 shost_for_each_device(sdev, shost) {
2017 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2018 scsi_eh_lock_door(sdev);
2019 sdev->was_reset = 0;
2020 }
2021 }
2022
2023 /*
2024 * next free up anything directly waiting upon the host. this
2025 * will be requests for character device operations, and also for
2026 * ioctls to queued block devices.
2027 */
2028 SCSI_LOG_ERROR_RECOVERY(3,
2029 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2030
2031 spin_lock_irqsave(shost->host_lock, flags);
2032 if (scsi_host_set_state(shost, SHOST_RUNNING))
2033 if (scsi_host_set_state(shost, SHOST_CANCEL))
2034 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2035 spin_unlock_irqrestore(shost->host_lock, flags);
2036
2037 wake_up(&shost->host_wait);
2038
2039 /*
2040 * finally we need to re-initiate requests that may be pending. we will
2041 * have had everything blocked while error handling is taking place, and
2042 * now that error recovery is done, we will need to ensure that these
2043 * requests are started.
2044 */
2045 scsi_run_host_queues(shost);
2046
2047 /*
2048 * if eh is active and host_eh_scheduled is pending we need to re-run
2049 * recovery. we do this check after scsi_run_host_queues() to allow
2050 * everything pent up since the last eh run a chance to make forward
2051 * progress before we sync again. Either we'll immediately re-run
2052 * recovery or scsi_device_unbusy() will wake us again when these
2053 * pending commands complete.
2054 */
2055 spin_lock_irqsave(shost->host_lock, flags);
2056 if (shost->host_eh_scheduled)
2057 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2058 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2059 spin_unlock_irqrestore(shost->host_lock, flags);
2060}
2061
2062/**
2063 * scsi_eh_ready_devs - check device ready state and recover if not.
2064 * @shost: host to be recovered.
2065 * @work_q: &list_head for pending commands.
2066 * @done_q: &list_head for processed commands.
2067 */
2068void scsi_eh_ready_devs(struct Scsi_Host *shost,
2069 struct list_head *work_q,
2070 struct list_head *done_q)
2071{
2072 if (!scsi_eh_stu(shost, work_q, done_q))
2073 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2074 if (!scsi_eh_target_reset(shost, work_q, done_q))
2075 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2076 if (!scsi_eh_host_reset(shost, work_q, done_q))
2077 scsi_eh_offline_sdevs(work_q,
2078 done_q);
2079}
2080EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2081
2082/**
2083 * scsi_eh_flush_done_q - finish processed commands or retry them.
2084 * @done_q: list_head of processed commands.
2085 */
2086void scsi_eh_flush_done_q(struct list_head *done_q)
2087{
2088 struct scsi_cmnd *scmd, *next;
2089
2090 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2091 list_del_init(&scmd->eh_entry);
2092 if (scsi_device_online(scmd->device) &&
2093 !scsi_noretry_cmd(scmd) &&
2094 (++scmd->retries <= scmd->allowed)) {
2095 SCSI_LOG_ERROR_RECOVERY(3,
2096 scmd_printk(KERN_INFO, scmd,
2097 "%s: flush retry cmd\n",
2098 current->comm));
2099 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2100 } else {
2101 /*
2102 * If just we got sense for the device (called
2103 * scsi_eh_get_sense), scmd->result is already
2104 * set, do not set DRIVER_TIMEOUT.
2105 */
2106 if (!scmd->result)
2107 scmd->result |= (DRIVER_TIMEOUT << 24);
2108 SCSI_LOG_ERROR_RECOVERY(3,
2109 scmd_printk(KERN_INFO, scmd,
2110 "%s: flush finish cmd\n",
2111 current->comm));
2112 scsi_finish_command(scmd);
2113 }
2114 }
2115}
2116EXPORT_SYMBOL(scsi_eh_flush_done_q);
2117
2118/**
2119 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2120 * @shost: Host to unjam.
2121 *
2122 * Notes:
2123 * When we come in here, we *know* that all commands on the bus have
2124 * either completed, failed or timed out. we also know that no further
2125 * commands are being sent to the host, so things are relatively quiet
2126 * and we have freedom to fiddle with things as we wish.
2127 *
2128 * This is only the *default* implementation. it is possible for
2129 * individual drivers to supply their own version of this function, and
2130 * if the maintainer wishes to do this, it is strongly suggested that
2131 * this function be taken as a template and modified. this function
2132 * was designed to correctly handle problems for about 95% of the
2133 * different cases out there, and it should always provide at least a
2134 * reasonable amount of error recovery.
2135 *
2136 * Any command marked 'failed' or 'timeout' must eventually have
2137 * scsi_finish_cmd() called for it. we do all of the retry stuff
2138 * here, so when we restart the host after we return it should have an
2139 * empty queue.
2140 */
2141static void scsi_unjam_host(struct Scsi_Host *shost)
2142{
2143 unsigned long flags;
2144 LIST_HEAD(eh_work_q);
2145 LIST_HEAD(eh_done_q);
2146
2147 spin_lock_irqsave(shost->host_lock, flags);
2148 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2149 spin_unlock_irqrestore(shost->host_lock, flags);
2150
2151 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2152
2153 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2154 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2155 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2156
2157 spin_lock_irqsave(shost->host_lock, flags);
2158 if (shost->eh_deadline != -1)
2159 shost->last_reset = 0;
2160 spin_unlock_irqrestore(shost->host_lock, flags);
2161 scsi_eh_flush_done_q(&eh_done_q);
2162}
2163
2164/**
2165 * scsi_error_handler - SCSI error handler thread
2166 * @data: Host for which we are running.
2167 *
2168 * Notes:
2169 * This is the main error handling loop. This is run as a kernel thread
2170 * for every SCSI host and handles all error handling activity.
2171 */
2172int scsi_error_handler(void *data)
2173{
2174 struct Scsi_Host *shost = data;
2175
2176 /*
2177 * We use TASK_INTERRUPTIBLE so that the thread is not
2178 * counted against the load average as a running process.
2179 * We never actually get interrupted because kthread_run
2180 * disables signal delivery for the created thread.
2181 */
2182 while (true) {
2183 /*
2184 * The sequence in kthread_stop() sets the stop flag first
2185 * then wakes the process. To avoid missed wakeups, the task
2186 * should always be in a non running state before the stop
2187 * flag is checked
2188 */
2189 set_current_state(TASK_INTERRUPTIBLE);
2190 if (kthread_should_stop())
2191 break;
2192
2193 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2194 shost->host_failed != atomic_read(&shost->host_busy)) {
2195 SCSI_LOG_ERROR_RECOVERY(1,
2196 shost_printk(KERN_INFO, shost,
2197 "scsi_eh_%d: sleeping\n",
2198 shost->host_no));
2199 schedule();
2200 continue;
2201 }
2202
2203 __set_current_state(TASK_RUNNING);
2204 SCSI_LOG_ERROR_RECOVERY(1,
2205 shost_printk(KERN_INFO, shost,
2206 "scsi_eh_%d: waking up %d/%d/%d\n",
2207 shost->host_no, shost->host_eh_scheduled,
2208 shost->host_failed,
2209 atomic_read(&shost->host_busy)));
2210
2211 /*
2212 * We have a host that is failing for some reason. Figure out
2213 * what we need to do to get it up and online again (if we can).
2214 * If we fail, we end up taking the thing offline.
2215 */
2216 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2217 SCSI_LOG_ERROR_RECOVERY(1,
2218 shost_printk(KERN_ERR, shost,
2219 "scsi_eh_%d: unable to autoresume\n",
2220 shost->host_no));
2221 continue;
2222 }
2223
2224 if (shost->transportt->eh_strategy_handler)
2225 shost->transportt->eh_strategy_handler(shost);
2226 else
2227 scsi_unjam_host(shost);
2228
2229 /* All scmds have been handled */
2230 shost->host_failed = 0;
2231
2232 /*
2233 * Note - if the above fails completely, the action is to take
2234 * individual devices offline and flush the queue of any
2235 * outstanding requests that may have been pending. When we
2236 * restart, we restart any I/O to any other devices on the bus
2237 * which are still online.
2238 */
2239 scsi_restart_operations(shost);
2240 if (!shost->eh_noresume)
2241 scsi_autopm_put_host(shost);
2242 }
2243 __set_current_state(TASK_RUNNING);
2244
2245 SCSI_LOG_ERROR_RECOVERY(1,
2246 shost_printk(KERN_INFO, shost,
2247 "Error handler scsi_eh_%d exiting\n",
2248 shost->host_no));
2249 shost->ehandler = NULL;
2250 return 0;
2251}
2252
2253/*
2254 * Function: scsi_report_bus_reset()
2255 *
2256 * Purpose: Utility function used by low-level drivers to report that
2257 * they have observed a bus reset on the bus being handled.
2258 *
2259 * Arguments: shost - Host in question
2260 * channel - channel on which reset was observed.
2261 *
2262 * Returns: Nothing
2263 *
2264 * Lock status: Host lock must be held.
2265 *
2266 * Notes: This only needs to be called if the reset is one which
2267 * originates from an unknown location. Resets originated
2268 * by the mid-level itself don't need to call this, but there
2269 * should be no harm.
2270 *
2271 * The main purpose of this is to make sure that a CHECK_CONDITION
2272 * is properly treated.
2273 */
2274void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2275{
2276 struct scsi_device *sdev;
2277
2278 __shost_for_each_device(sdev, shost) {
2279 if (channel == sdev_channel(sdev))
2280 __scsi_report_device_reset(sdev, NULL);
2281 }
2282}
2283EXPORT_SYMBOL(scsi_report_bus_reset);
2284
2285/*
2286 * Function: scsi_report_device_reset()
2287 *
2288 * Purpose: Utility function used by low-level drivers to report that
2289 * they have observed a device reset on the device being handled.
2290 *
2291 * Arguments: shost - Host in question
2292 * channel - channel on which reset was observed
2293 * target - target on which reset was observed
2294 *
2295 * Returns: Nothing
2296 *
2297 * Lock status: Host lock must be held
2298 *
2299 * Notes: This only needs to be called if the reset is one which
2300 * originates from an unknown location. Resets originated
2301 * by the mid-level itself don't need to call this, but there
2302 * should be no harm.
2303 *
2304 * The main purpose of this is to make sure that a CHECK_CONDITION
2305 * is properly treated.
2306 */
2307void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2308{
2309 struct scsi_device *sdev;
2310
2311 __shost_for_each_device(sdev, shost) {
2312 if (channel == sdev_channel(sdev) &&
2313 target == sdev_id(sdev))
2314 __scsi_report_device_reset(sdev, NULL);
2315 }
2316}
2317EXPORT_SYMBOL(scsi_report_device_reset);
2318
2319static void
2320scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2321{
2322}
2323
2324/**
2325 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2326 * @dev: scsi_device to operate on
2327 * @arg: reset type (see sg.h)
2328 */
2329int
2330scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2331{
2332 struct scsi_cmnd *scmd;
2333 struct Scsi_Host *shost = dev->host;
2334 struct request req;
2335 unsigned long flags;
2336 int error = 0, rtn, val;
2337
2338 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2339 return -EACCES;
2340
2341 error = get_user(val, arg);
2342 if (error)
2343 return error;
2344
2345 if (scsi_autopm_get_host(shost) < 0)
2346 return -EIO;
2347
2348 error = -EIO;
2349 scmd = scsi_get_command(dev, GFP_KERNEL);
2350 if (!scmd)
2351 goto out_put_autopm_host;
2352
2353 blk_rq_init(NULL, &req);
2354 scmd->request = &req;
2355
2356 scmd->cmnd = req.cmd;
2357
2358 scmd->scsi_done = scsi_reset_provider_done_command;
2359 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2360
2361 scmd->cmd_len = 0;
2362
2363 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2364
2365 spin_lock_irqsave(shost->host_lock, flags);
2366 shost->tmf_in_progress = 1;
2367 spin_unlock_irqrestore(shost->host_lock, flags);
2368
2369 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2370 case SG_SCSI_RESET_NOTHING:
2371 rtn = SUCCESS;
2372 break;
2373 case SG_SCSI_RESET_DEVICE:
2374 rtn = scsi_try_bus_device_reset(scmd);
2375 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2376 break;
2377 /* FALLTHROUGH */
2378 case SG_SCSI_RESET_TARGET:
2379 rtn = scsi_try_target_reset(scmd);
2380 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2381 break;
2382 /* FALLTHROUGH */
2383 case SG_SCSI_RESET_BUS:
2384 rtn = scsi_try_bus_reset(scmd);
2385 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2386 break;
2387 /* FALLTHROUGH */
2388 case SG_SCSI_RESET_HOST:
2389 rtn = scsi_try_host_reset(scmd);
2390 if (rtn == SUCCESS)
2391 break;
2392 default:
2393 /* FALLTHROUGH */
2394 rtn = FAILED;
2395 break;
2396 }
2397
2398 error = (rtn == SUCCESS) ? 0 : -EIO;
2399
2400 spin_lock_irqsave(shost->host_lock, flags);
2401 shost->tmf_in_progress = 0;
2402 spin_unlock_irqrestore(shost->host_lock, flags);
2403
2404 /*
2405 * be sure to wake up anyone who was sleeping or had their queue
2406 * suspended while we performed the TMF.
2407 */
2408 SCSI_LOG_ERROR_RECOVERY(3,
2409 shost_printk(KERN_INFO, shost,
2410 "waking up host to restart after TMF\n"));
2411
2412 wake_up(&shost->host_wait);
2413 scsi_run_host_queues(shost);
2414
2415 scsi_put_command(scmd);
2416
2417out_put_autopm_host:
2418 scsi_autopm_put_host(shost);
2419 return error;
2420}
2421EXPORT_SYMBOL(scsi_ioctl_reset);
2422
2423bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2424 struct scsi_sense_hdr *sshdr)
2425{
2426 return scsi_normalize_sense(cmd->sense_buffer,
2427 SCSI_SENSE_BUFFERSIZE, sshdr);
2428}
2429EXPORT_SYMBOL(scsi_command_normalize_sense);
2430
2431/**
2432 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2433 * @sense_buffer: byte array of sense data
2434 * @sb_len: number of valid bytes in sense_buffer
2435 * @info_out: pointer to 64 integer where 8 or 4 byte information
2436 * field will be placed if found.
2437 *
2438 * Return value:
2439 * 1 if information field found, 0 if not found.
2440 */
2441int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2442 u64 * info_out)
2443{
2444 int j;
2445 const u8 * ucp;
2446 u64 ull;
2447
2448 if (sb_len < 7)
2449 return 0;
2450 switch (sense_buffer[0] & 0x7f) {
2451 case 0x70:
2452 case 0x71:
2453 if (sense_buffer[0] & 0x80) {
2454 *info_out = (sense_buffer[3] << 24) +
2455 (sense_buffer[4] << 16) +
2456 (sense_buffer[5] << 8) + sense_buffer[6];
2457 return 1;
2458 } else
2459 return 0;
2460 case 0x72:
2461 case 0x73:
2462 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2463 0 /* info desc */);
2464 if (ucp && (0xa == ucp[1])) {
2465 ull = 0;
2466 for (j = 0; j < 8; ++j) {
2467 if (j > 0)
2468 ull <<= 8;
2469 ull |= ucp[4 + j];
2470 }
2471 *info_out = ull;
2472 return 1;
2473 } else
2474 return 0;
2475 default:
2476 return 0;
2477 }
2478}
2479EXPORT_SYMBOL(scsi_get_sense_info_fld);
1/*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
3 *
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.
8 *
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11 *
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)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
19#include <linux/gfp.h>
20#include <linux/timer.h>
21#include <linux/string.h>
22#include <linux/kernel.h>
23#include <linux/freezer.h>
24#include <linux/kthread.h>
25#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
28#include <linux/jiffies.h>
29
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_common.h>
37#include <scsi/scsi_transport.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_ioctl.h>
40#include <scsi/scsi_dh.h>
41#include <scsi/sg.h>
42
43#include "scsi_priv.h"
44#include "scsi_logging.h"
45#include "scsi_transport_api.h"
46
47#include <trace/events/scsi.h>
48
49#include <asm/unaligned.h>
50
51static void scsi_eh_done(struct scsi_cmnd *scmd);
52
53/*
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
56 */
57#define BUS_RESET_SETTLE_TIME (10)
58#define HOST_RESET_SETTLE_TIME (10)
59
60static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61static int scsi_try_to_abort_cmd(struct scsi_host_template *,
62 struct scsi_cmnd *);
63
64void scsi_eh_wakeup(struct Scsi_Host *shost)
65{
66 lockdep_assert_held(shost->host_lock);
67
68 if (atomic_read(&shost->host_busy) == shost->host_failed) {
69 trace_scsi_eh_wakeup(shost);
70 wake_up_process(shost->ehandler);
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 "Waking error handler thread\n"));
73 }
74}
75
76/**
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
79 *
80 * Schedule SCSI EH without scmd.
81 */
82void scsi_schedule_eh(struct Scsi_Host *shost)
83{
84 unsigned long flags;
85
86 spin_lock_irqsave(shost->host_lock, flags);
87
88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
90 shost->host_eh_scheduled++;
91 scsi_eh_wakeup(shost);
92 }
93
94 spin_unlock_irqrestore(shost->host_lock, flags);
95}
96EXPORT_SYMBOL_GPL(scsi_schedule_eh);
97
98static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
99{
100 if (!shost->last_reset || shost->eh_deadline == -1)
101 return 0;
102
103 /*
104 * 32bit accesses are guaranteed to be atomic
105 * (on all supported architectures), so instead
106 * of using a spinlock we can as well double check
107 * if eh_deadline has been set to 'off' during the
108 * time_before call.
109 */
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 shost->eh_deadline > -1)
112 return 0;
113
114 return 1;
115}
116
117/**
118 * scmd_eh_abort_handler - Handle command aborts
119 * @work: command to be aborted.
120 *
121 * Note: this function must be called only for a command that has timed out.
122 * Because the block layer marks a request as complete before it calls
123 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
124 * timed out do not have any effect. Hence it is safe to call
125 * scsi_finish_command() from this function.
126 */
127void
128scmd_eh_abort_handler(struct work_struct *work)
129{
130 struct scsi_cmnd *scmd =
131 container_of(work, struct scsi_cmnd, abort_work.work);
132 struct scsi_device *sdev = scmd->device;
133 int rtn;
134
135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "eh timeout, not aborting\n"));
139 } else {
140 SCSI_LOG_ERROR_RECOVERY(3,
141 scmd_printk(KERN_INFO, scmd,
142 "aborting command\n"));
143 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
144 if (rtn == SUCCESS) {
145 set_host_byte(scmd, DID_TIME_OUT);
146 if (scsi_host_eh_past_deadline(sdev->host)) {
147 SCSI_LOG_ERROR_RECOVERY(3,
148 scmd_printk(KERN_INFO, scmd,
149 "eh timeout, not retrying "
150 "aborted command\n"));
151 } else if (!scsi_noretry_cmd(scmd) &&
152 (++scmd->retries <= scmd->allowed)) {
153 SCSI_LOG_ERROR_RECOVERY(3,
154 scmd_printk(KERN_WARNING, scmd,
155 "retry aborted command\n"));
156 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
157 return;
158 } else {
159 SCSI_LOG_ERROR_RECOVERY(3,
160 scmd_printk(KERN_WARNING, scmd,
161 "finish aborted command\n"));
162 scsi_finish_command(scmd);
163 return;
164 }
165 } else {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_INFO, scmd,
168 "cmd abort %s\n",
169 (rtn == FAST_IO_FAIL) ?
170 "not send" : "failed"));
171 }
172 }
173
174 scsi_eh_scmd_add(scmd);
175}
176
177/**
178 * scsi_abort_command - schedule a command abort
179 * @scmd: scmd to abort.
180 *
181 * We only need to abort commands after a command timeout
182 */
183static int
184scsi_abort_command(struct scsi_cmnd *scmd)
185{
186 struct scsi_device *sdev = scmd->device;
187 struct Scsi_Host *shost = sdev->host;
188 unsigned long flags;
189
190 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
191 /*
192 * Retry after abort failed, escalate to next level.
193 */
194 SCSI_LOG_ERROR_RECOVERY(3,
195 scmd_printk(KERN_INFO, scmd,
196 "previous abort failed\n"));
197 BUG_ON(delayed_work_pending(&scmd->abort_work));
198 return FAILED;
199 }
200
201 spin_lock_irqsave(shost->host_lock, flags);
202 if (shost->eh_deadline != -1 && !shost->last_reset)
203 shost->last_reset = jiffies;
204 spin_unlock_irqrestore(shost->host_lock, flags);
205
206 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
209 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
210 return SUCCESS;
211}
212
213/**
214 * scsi_eh_reset - call into ->eh_action to reset internal counters
215 * @scmd: scmd to run eh on.
216 *
217 * The scsi driver might be carrying internal state about the
218 * devices, so we need to call into the driver to reset the
219 * internal state once the error handler is started.
220 */
221static void scsi_eh_reset(struct scsi_cmnd *scmd)
222{
223 if (!blk_rq_is_passthrough(scmd->request)) {
224 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
225 if (sdrv->eh_reset)
226 sdrv->eh_reset(scmd);
227 }
228}
229
230static void scsi_eh_inc_host_failed(struct rcu_head *head)
231{
232 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
233 struct Scsi_Host *shost = scmd->device->host;
234 unsigned long flags;
235
236 spin_lock_irqsave(shost->host_lock, flags);
237 shost->host_failed++;
238 scsi_eh_wakeup(shost);
239 spin_unlock_irqrestore(shost->host_lock, flags);
240}
241
242/**
243 * scsi_eh_scmd_add - add scsi cmd to error handling.
244 * @scmd: scmd to run eh on.
245 */
246void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
247{
248 struct Scsi_Host *shost = scmd->device->host;
249 unsigned long flags;
250 int ret;
251
252 WARN_ON_ONCE(!shost->ehandler);
253
254 spin_lock_irqsave(shost->host_lock, flags);
255 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
256 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
257 WARN_ON_ONCE(ret);
258 }
259 if (shost->eh_deadline != -1 && !shost->last_reset)
260 shost->last_reset = jiffies;
261
262 scsi_eh_reset(scmd);
263 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
264 spin_unlock_irqrestore(shost->host_lock, flags);
265 /*
266 * Ensure that all tasks observe the host state change before the
267 * host_failed change.
268 */
269 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
270}
271
272/**
273 * scsi_times_out - Timeout function for normal scsi commands.
274 * @req: request that is timing out.
275 *
276 * Notes:
277 * We do not need to lock this. There is the potential for a race
278 * only in that the normal completion handling might run, but if the
279 * normal completion function determines that the timer has already
280 * fired, then it mustn't do anything.
281 */
282enum blk_eh_timer_return scsi_times_out(struct request *req)
283{
284 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
285 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
286 struct Scsi_Host *host = scmd->device->host;
287
288 trace_scsi_dispatch_cmd_timeout(scmd);
289 scsi_log_completion(scmd, TIMEOUT_ERROR);
290
291 if (host->eh_deadline != -1 && !host->last_reset)
292 host->last_reset = jiffies;
293
294 if (host->hostt->eh_timed_out)
295 rtn = host->hostt->eh_timed_out(scmd);
296
297 if (rtn == BLK_EH_NOT_HANDLED) {
298 if (scsi_abort_command(scmd) != SUCCESS) {
299 set_host_byte(scmd, DID_TIME_OUT);
300 scsi_eh_scmd_add(scmd);
301 }
302 }
303
304 return rtn;
305}
306
307/**
308 * scsi_block_when_processing_errors - Prevent cmds from being queued.
309 * @sdev: Device on which we are performing recovery.
310 *
311 * Description:
312 * We block until the host is out of error recovery, and then check to
313 * see whether the host or the device is offline.
314 *
315 * Return value:
316 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
317 */
318int scsi_block_when_processing_errors(struct scsi_device *sdev)
319{
320 int online;
321
322 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
323
324 online = scsi_device_online(sdev);
325
326 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
327 "%s: rtn: %d\n", __func__, online));
328
329 return online;
330}
331EXPORT_SYMBOL(scsi_block_when_processing_errors);
332
333#ifdef CONFIG_SCSI_LOGGING
334/**
335 * scsi_eh_prt_fail_stats - Log info on failures.
336 * @shost: scsi host being recovered.
337 * @work_q: Queue of scsi cmds to process.
338 */
339static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
340 struct list_head *work_q)
341{
342 struct scsi_cmnd *scmd;
343 struct scsi_device *sdev;
344 int total_failures = 0;
345 int cmd_failed = 0;
346 int cmd_cancel = 0;
347 int devices_failed = 0;
348
349 shost_for_each_device(sdev, shost) {
350 list_for_each_entry(scmd, work_q, eh_entry) {
351 if (scmd->device == sdev) {
352 ++total_failures;
353 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
354 ++cmd_cancel;
355 else
356 ++cmd_failed;
357 }
358 }
359
360 if (cmd_cancel || cmd_failed) {
361 SCSI_LOG_ERROR_RECOVERY(3,
362 shost_printk(KERN_INFO, shost,
363 "%s: cmds failed: %d, cancel: %d\n",
364 __func__, cmd_failed,
365 cmd_cancel));
366 cmd_cancel = 0;
367 cmd_failed = 0;
368 ++devices_failed;
369 }
370 }
371
372 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
373 "Total of %d commands on %d"
374 " devices require eh work\n",
375 total_failures, devices_failed));
376}
377#endif
378
379 /**
380 * scsi_report_lun_change - Set flag on all *other* devices on the same target
381 * to indicate that a UNIT ATTENTION is expected.
382 * @sdev: Device reporting the UNIT ATTENTION
383 */
384static void scsi_report_lun_change(struct scsi_device *sdev)
385{
386 sdev->sdev_target->expecting_lun_change = 1;
387}
388
389/**
390 * scsi_report_sense - Examine scsi sense information and log messages for
391 * certain conditions, also issue uevents for some of them.
392 * @sdev: Device reporting the sense code
393 * @sshdr: sshdr to be examined
394 */
395static void scsi_report_sense(struct scsi_device *sdev,
396 struct scsi_sense_hdr *sshdr)
397{
398 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
399
400 if (sshdr->sense_key == UNIT_ATTENTION) {
401 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
402 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
403 sdev_printk(KERN_WARNING, sdev,
404 "Inquiry data has changed");
405 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
406 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
407 scsi_report_lun_change(sdev);
408 sdev_printk(KERN_WARNING, sdev,
409 "Warning! Received an indication that the "
410 "LUN assignments on this target have "
411 "changed. The Linux SCSI layer does not "
412 "automatically remap LUN assignments.\n");
413 } else if (sshdr->asc == 0x3f)
414 sdev_printk(KERN_WARNING, sdev,
415 "Warning! Received an indication that the "
416 "operating parameters on this target have "
417 "changed. The Linux SCSI layer does not "
418 "automatically adjust these parameters.\n");
419
420 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
421 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
422 sdev_printk(KERN_WARNING, sdev,
423 "Warning! Received an indication that the "
424 "LUN reached a thin provisioning soft "
425 "threshold.\n");
426 }
427
428 if (sshdr->asc == 0x29) {
429 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
430 sdev_printk(KERN_WARNING, sdev,
431 "Power-on or device reset occurred\n");
432 }
433
434 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
435 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
436 sdev_printk(KERN_WARNING, sdev,
437 "Mode parameters changed");
438 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
439 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
440 sdev_printk(KERN_WARNING, sdev,
441 "Asymmetric access state changed");
442 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
443 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
444 sdev_printk(KERN_WARNING, sdev,
445 "Capacity data has changed");
446 } else if (sshdr->asc == 0x2a)
447 sdev_printk(KERN_WARNING, sdev,
448 "Parameters changed");
449 }
450
451 if (evt_type != SDEV_EVT_MAXBITS) {
452 set_bit(evt_type, sdev->pending_events);
453 schedule_work(&sdev->event_work);
454 }
455}
456
457/**
458 * scsi_check_sense - Examine scsi cmd sense
459 * @scmd: Cmd to have sense checked.
460 *
461 * Return value:
462 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
463 *
464 * Notes:
465 * When a deferred error is detected the current command has
466 * not been executed and needs retrying.
467 */
468int scsi_check_sense(struct scsi_cmnd *scmd)
469{
470 struct scsi_device *sdev = scmd->device;
471 struct scsi_sense_hdr sshdr;
472
473 if (! scsi_command_normalize_sense(scmd, &sshdr))
474 return FAILED; /* no valid sense data */
475
476 scsi_report_sense(sdev, &sshdr);
477
478 if (scsi_sense_is_deferred(&sshdr))
479 return NEEDS_RETRY;
480
481 if (sdev->handler && sdev->handler->check_sense) {
482 int rc;
483
484 rc = sdev->handler->check_sense(sdev, &sshdr);
485 if (rc != SCSI_RETURN_NOT_HANDLED)
486 return rc;
487 /* handler does not care. Drop down to default handling */
488 }
489
490 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
491 /*
492 * nasty: for mid-layer issued TURs, we need to return the
493 * actual sense data without any recovery attempt. For eh
494 * issued ones, we need to try to recover and interpret
495 */
496 return SUCCESS;
497
498 /*
499 * Previous logic looked for FILEMARK, EOM or ILI which are
500 * mainly associated with tapes and returned SUCCESS.
501 */
502 if (sshdr.response_code == 0x70) {
503 /* fixed format */
504 if (scmd->sense_buffer[2] & 0xe0)
505 return SUCCESS;
506 } else {
507 /*
508 * descriptor format: look for "stream commands sense data
509 * descriptor" (see SSC-3). Assume single sense data
510 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
511 */
512 if ((sshdr.additional_length > 3) &&
513 (scmd->sense_buffer[8] == 0x4) &&
514 (scmd->sense_buffer[11] & 0xe0))
515 return SUCCESS;
516 }
517
518 switch (sshdr.sense_key) {
519 case NO_SENSE:
520 return SUCCESS;
521 case RECOVERED_ERROR:
522 return /* soft_error */ SUCCESS;
523
524 case ABORTED_COMMAND:
525 if (sshdr.asc == 0x10) /* DIF */
526 return SUCCESS;
527
528 return NEEDS_RETRY;
529 case NOT_READY:
530 case UNIT_ATTENTION:
531 /*
532 * if we are expecting a cc/ua because of a bus reset that we
533 * performed, treat this just as a retry. otherwise this is
534 * information that we should pass up to the upper-level driver
535 * so that we can deal with it there.
536 */
537 if (scmd->device->expecting_cc_ua) {
538 /*
539 * Because some device does not queue unit
540 * attentions correctly, we carefully check
541 * additional sense code and qualifier so as
542 * not to squash media change unit attention.
543 */
544 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
545 scmd->device->expecting_cc_ua = 0;
546 return NEEDS_RETRY;
547 }
548 }
549 /*
550 * we might also expect a cc/ua if another LUN on the target
551 * reported a UA with an ASC/ASCQ of 3F 0E -
552 * REPORTED LUNS DATA HAS CHANGED.
553 */
554 if (scmd->device->sdev_target->expecting_lun_change &&
555 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
556 return NEEDS_RETRY;
557 /*
558 * if the device is in the process of becoming ready, we
559 * should retry.
560 */
561 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
562 return NEEDS_RETRY;
563 /*
564 * if the device is not started, we need to wake
565 * the error handler to start the motor
566 */
567 if (scmd->device->allow_restart &&
568 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
569 return FAILED;
570 /*
571 * Pass the UA upwards for a determination in the completion
572 * functions.
573 */
574 return SUCCESS;
575
576 /* these are not supported */
577 case DATA_PROTECT:
578 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
579 /* Thin provisioning hard threshold reached */
580 set_host_byte(scmd, DID_ALLOC_FAILURE);
581 return SUCCESS;
582 }
583 /* FALLTHROUGH */
584 case COPY_ABORTED:
585 case VOLUME_OVERFLOW:
586 case MISCOMPARE:
587 case BLANK_CHECK:
588 set_host_byte(scmd, DID_TARGET_FAILURE);
589 return SUCCESS;
590
591 case MEDIUM_ERROR:
592 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
593 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
594 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
595 set_host_byte(scmd, DID_MEDIUM_ERROR);
596 return SUCCESS;
597 }
598 return NEEDS_RETRY;
599
600 case HARDWARE_ERROR:
601 if (scmd->device->retry_hwerror)
602 return ADD_TO_MLQUEUE;
603 else
604 set_host_byte(scmd, DID_TARGET_FAILURE);
605 /* FALLTHROUGH */
606
607 case ILLEGAL_REQUEST:
608 if (sshdr.asc == 0x20 || /* Invalid command operation code */
609 sshdr.asc == 0x21 || /* Logical block address out of range */
610 sshdr.asc == 0x22 || /* Invalid function */
611 sshdr.asc == 0x24 || /* Invalid field in cdb */
612 sshdr.asc == 0x26 || /* Parameter value invalid */
613 sshdr.asc == 0x27) { /* Write protected */
614 set_host_byte(scmd, DID_TARGET_FAILURE);
615 }
616 return SUCCESS;
617
618 default:
619 return SUCCESS;
620 }
621}
622EXPORT_SYMBOL_GPL(scsi_check_sense);
623
624static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
625{
626 struct scsi_host_template *sht = sdev->host->hostt;
627 struct scsi_device *tmp_sdev;
628
629 if (!sht->track_queue_depth ||
630 sdev->queue_depth >= sdev->max_queue_depth)
631 return;
632
633 if (time_before(jiffies,
634 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
635 return;
636
637 if (time_before(jiffies,
638 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
639 return;
640
641 /*
642 * Walk all devices of a target and do
643 * ramp up on them.
644 */
645 shost_for_each_device(tmp_sdev, sdev->host) {
646 if (tmp_sdev->channel != sdev->channel ||
647 tmp_sdev->id != sdev->id ||
648 tmp_sdev->queue_depth == sdev->max_queue_depth)
649 continue;
650
651 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
652 sdev->last_queue_ramp_up = jiffies;
653 }
654}
655
656static void scsi_handle_queue_full(struct scsi_device *sdev)
657{
658 struct scsi_host_template *sht = sdev->host->hostt;
659 struct scsi_device *tmp_sdev;
660
661 if (!sht->track_queue_depth)
662 return;
663
664 shost_for_each_device(tmp_sdev, sdev->host) {
665 if (tmp_sdev->channel != sdev->channel ||
666 tmp_sdev->id != sdev->id)
667 continue;
668 /*
669 * We do not know the number of commands that were at
670 * the device when we got the queue full so we start
671 * from the highest possible value and work our way down.
672 */
673 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
674 }
675}
676
677/**
678 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
679 * @scmd: SCSI cmd to examine.
680 *
681 * Notes:
682 * This is *only* called when we are examining the status of commands
683 * queued during error recovery. the main difference here is that we
684 * don't allow for the possibility of retries here, and we are a lot
685 * more restrictive about what we consider acceptable.
686 */
687static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
688{
689 /*
690 * first check the host byte, to see if there is anything in there
691 * that would indicate what we need to do.
692 */
693 if (host_byte(scmd->result) == DID_RESET) {
694 /*
695 * rats. we are already in the error handler, so we now
696 * get to try and figure out what to do next. if the sense
697 * is valid, we have a pretty good idea of what to do.
698 * if not, we mark it as FAILED.
699 */
700 return scsi_check_sense(scmd);
701 }
702 if (host_byte(scmd->result) != DID_OK)
703 return FAILED;
704
705 /*
706 * next, check the message byte.
707 */
708 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
709 return FAILED;
710
711 /*
712 * now, check the status byte to see if this indicates
713 * anything special.
714 */
715 switch (status_byte(scmd->result)) {
716 case GOOD:
717 scsi_handle_queue_ramp_up(scmd->device);
718 /* FALLTHROUGH */
719 case COMMAND_TERMINATED:
720 return SUCCESS;
721 case CHECK_CONDITION:
722 return scsi_check_sense(scmd);
723 case CONDITION_GOOD:
724 case INTERMEDIATE_GOOD:
725 case INTERMEDIATE_C_GOOD:
726 /*
727 * who knows? FIXME(eric)
728 */
729 return SUCCESS;
730 case RESERVATION_CONFLICT:
731 if (scmd->cmnd[0] == TEST_UNIT_READY)
732 /* it is a success, we probed the device and
733 * found it */
734 return SUCCESS;
735 /* otherwise, we failed to send the command */
736 return FAILED;
737 case QUEUE_FULL:
738 scsi_handle_queue_full(scmd->device);
739 /* fall through */
740 case BUSY:
741 return NEEDS_RETRY;
742 default:
743 return FAILED;
744 }
745 return FAILED;
746}
747
748/**
749 * scsi_eh_done - Completion function for error handling.
750 * @scmd: Cmd that is done.
751 */
752static void scsi_eh_done(struct scsi_cmnd *scmd)
753{
754 struct completion *eh_action;
755
756 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
757 "%s result: %x\n", __func__, scmd->result));
758
759 eh_action = scmd->device->host->eh_action;
760 if (eh_action)
761 complete(eh_action);
762}
763
764/**
765 * scsi_try_host_reset - ask host adapter to reset itself
766 * @scmd: SCSI cmd to send host reset.
767 */
768static int scsi_try_host_reset(struct scsi_cmnd *scmd)
769{
770 unsigned long flags;
771 int rtn;
772 struct Scsi_Host *host = scmd->device->host;
773 struct scsi_host_template *hostt = host->hostt;
774
775 SCSI_LOG_ERROR_RECOVERY(3,
776 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
777
778 if (!hostt->eh_host_reset_handler)
779 return FAILED;
780
781 rtn = hostt->eh_host_reset_handler(scmd);
782
783 if (rtn == SUCCESS) {
784 if (!hostt->skip_settle_delay)
785 ssleep(HOST_RESET_SETTLE_TIME);
786 spin_lock_irqsave(host->host_lock, flags);
787 scsi_report_bus_reset(host, scmd_channel(scmd));
788 spin_unlock_irqrestore(host->host_lock, flags);
789 }
790
791 return rtn;
792}
793
794/**
795 * scsi_try_bus_reset - ask host to perform a bus reset
796 * @scmd: SCSI cmd to send bus reset.
797 */
798static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
799{
800 unsigned long flags;
801 int rtn;
802 struct Scsi_Host *host = scmd->device->host;
803 struct scsi_host_template *hostt = host->hostt;
804
805 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
806 "%s: Snd Bus RST\n", __func__));
807
808 if (!hostt->eh_bus_reset_handler)
809 return FAILED;
810
811 rtn = hostt->eh_bus_reset_handler(scmd);
812
813 if (rtn == SUCCESS) {
814 if (!hostt->skip_settle_delay)
815 ssleep(BUS_RESET_SETTLE_TIME);
816 spin_lock_irqsave(host->host_lock, flags);
817 scsi_report_bus_reset(host, scmd_channel(scmd));
818 spin_unlock_irqrestore(host->host_lock, flags);
819 }
820
821 return rtn;
822}
823
824static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
825{
826 sdev->was_reset = 1;
827 sdev->expecting_cc_ua = 1;
828}
829
830/**
831 * scsi_try_target_reset - Ask host to perform a target reset
832 * @scmd: SCSI cmd used to send a target reset
833 *
834 * Notes:
835 * There is no timeout for this operation. if this operation is
836 * unreliable for a given host, then the host itself needs to put a
837 * timer on it, and set the host back to a consistent state prior to
838 * returning.
839 */
840static int scsi_try_target_reset(struct scsi_cmnd *scmd)
841{
842 unsigned long flags;
843 int rtn;
844 struct Scsi_Host *host = scmd->device->host;
845 struct scsi_host_template *hostt = host->hostt;
846
847 if (!hostt->eh_target_reset_handler)
848 return FAILED;
849
850 rtn = hostt->eh_target_reset_handler(scmd);
851 if (rtn == SUCCESS) {
852 spin_lock_irqsave(host->host_lock, flags);
853 __starget_for_each_device(scsi_target(scmd->device), NULL,
854 __scsi_report_device_reset);
855 spin_unlock_irqrestore(host->host_lock, flags);
856 }
857
858 return rtn;
859}
860
861/**
862 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
863 * @scmd: SCSI cmd used to send BDR
864 *
865 * Notes:
866 * There is no timeout for this operation. if this operation is
867 * unreliable for a given host, then the host itself needs to put a
868 * timer on it, and set the host back to a consistent state prior to
869 * returning.
870 */
871static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
872{
873 int rtn;
874 struct scsi_host_template *hostt = scmd->device->host->hostt;
875
876 if (!hostt->eh_device_reset_handler)
877 return FAILED;
878
879 rtn = hostt->eh_device_reset_handler(scmd);
880 if (rtn == SUCCESS)
881 __scsi_report_device_reset(scmd->device, NULL);
882 return rtn;
883}
884
885/**
886 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
887 * @hostt: SCSI driver host template
888 * @scmd: SCSI cmd used to send a target reset
889 *
890 * Return value:
891 * SUCCESS, FAILED, or FAST_IO_FAIL
892 *
893 * Notes:
894 * SUCCESS does not necessarily indicate that the command
895 * has been aborted; it only indicates that the LLDDs
896 * has cleared all references to that command.
897 * LLDDs should return FAILED only if an abort was required
898 * but could not be executed. LLDDs should return FAST_IO_FAIL
899 * if the device is temporarily unavailable (eg due to a
900 * link down on FibreChannel)
901 */
902static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
903 struct scsi_cmnd *scmd)
904{
905 if (!hostt->eh_abort_handler)
906 return FAILED;
907
908 return hostt->eh_abort_handler(scmd);
909}
910
911static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
912{
913 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
914 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
915 if (scsi_try_target_reset(scmd) != SUCCESS)
916 if (scsi_try_bus_reset(scmd) != SUCCESS)
917 scsi_try_host_reset(scmd);
918}
919
920/**
921 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
922 * @scmd: SCSI command structure to hijack
923 * @ses: structure to save restore information
924 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
925 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
926 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
927 *
928 * This function is used to save a scsi command information before re-execution
929 * as part of the error recovery process. If @sense_bytes is 0 the command
930 * sent must be one that does not transfer any data. If @sense_bytes != 0
931 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
932 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
933 */
934void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
935 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
936{
937 struct scsi_device *sdev = scmd->device;
938
939 /*
940 * We need saved copies of a number of fields - this is because
941 * error handling may need to overwrite these with different values
942 * to run different commands, and once error handling is complete,
943 * we will need to restore these values prior to running the actual
944 * command.
945 */
946 ses->cmd_len = scmd->cmd_len;
947 ses->cmnd = scmd->cmnd;
948 ses->data_direction = scmd->sc_data_direction;
949 ses->sdb = scmd->sdb;
950 ses->next_rq = scmd->request->next_rq;
951 ses->result = scmd->result;
952 ses->underflow = scmd->underflow;
953 ses->prot_op = scmd->prot_op;
954 ses->eh_eflags = scmd->eh_eflags;
955
956 scmd->prot_op = SCSI_PROT_NORMAL;
957 scmd->eh_eflags = 0;
958 scmd->cmnd = ses->eh_cmnd;
959 memset(scmd->cmnd, 0, BLK_MAX_CDB);
960 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
961 scmd->request->next_rq = NULL;
962 scmd->result = 0;
963
964 if (sense_bytes) {
965 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
966 sense_bytes);
967 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
968 scmd->sdb.length);
969 scmd->sdb.table.sgl = &ses->sense_sgl;
970 scmd->sc_data_direction = DMA_FROM_DEVICE;
971 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
972 scmd->cmnd[0] = REQUEST_SENSE;
973 scmd->cmnd[4] = scmd->sdb.length;
974 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
975 } else {
976 scmd->sc_data_direction = DMA_NONE;
977 if (cmnd) {
978 BUG_ON(cmnd_size > BLK_MAX_CDB);
979 memcpy(scmd->cmnd, cmnd, cmnd_size);
980 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
981 }
982 }
983
984 scmd->underflow = 0;
985
986 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
987 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
988 (sdev->lun << 5 & 0xe0);
989
990 /*
991 * Zero the sense buffer. The scsi spec mandates that any
992 * untransferred sense data should be interpreted as being zero.
993 */
994 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
995}
996EXPORT_SYMBOL(scsi_eh_prep_cmnd);
997
998/**
999 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1000 * @scmd: SCSI command structure to restore
1001 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1002 *
1003 * Undo any damage done by above scsi_eh_prep_cmnd().
1004 */
1005void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1006{
1007 /*
1008 * Restore original data
1009 */
1010 scmd->cmd_len = ses->cmd_len;
1011 scmd->cmnd = ses->cmnd;
1012 scmd->sc_data_direction = ses->data_direction;
1013 scmd->sdb = ses->sdb;
1014 scmd->request->next_rq = ses->next_rq;
1015 scmd->result = ses->result;
1016 scmd->underflow = ses->underflow;
1017 scmd->prot_op = ses->prot_op;
1018 scmd->eh_eflags = ses->eh_eflags;
1019}
1020EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1021
1022/**
1023 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1024 * @scmd: SCSI command structure to hijack
1025 * @cmnd: CDB to send
1026 * @cmnd_size: size in bytes of @cmnd
1027 * @timeout: timeout for this request
1028 * @sense_bytes: size of sense data to copy or 0
1029 *
1030 * This function is used to send a scsi command down to a target device
1031 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1032 *
1033 * Return value:
1034 * SUCCESS or FAILED or NEEDS_RETRY
1035 */
1036static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1037 int cmnd_size, int timeout, unsigned sense_bytes)
1038{
1039 struct scsi_device *sdev = scmd->device;
1040 struct Scsi_Host *shost = sdev->host;
1041 DECLARE_COMPLETION_ONSTACK(done);
1042 unsigned long timeleft = timeout;
1043 struct scsi_eh_save ses;
1044 const unsigned long stall_for = msecs_to_jiffies(100);
1045 int rtn;
1046
1047retry:
1048 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1049 shost->eh_action = &done;
1050
1051 scsi_log_send(scmd);
1052 scmd->scsi_done = scsi_eh_done;
1053 rtn = shost->hostt->queuecommand(shost, scmd);
1054 if (rtn) {
1055 if (timeleft > stall_for) {
1056 scsi_eh_restore_cmnd(scmd, &ses);
1057 timeleft -= stall_for;
1058 msleep(jiffies_to_msecs(stall_for));
1059 goto retry;
1060 }
1061 /* signal not to enter either branch of the if () below */
1062 timeleft = 0;
1063 rtn = FAILED;
1064 } else {
1065 timeleft = wait_for_completion_timeout(&done, timeout);
1066 rtn = SUCCESS;
1067 }
1068
1069 shost->eh_action = NULL;
1070
1071 scsi_log_completion(scmd, rtn);
1072
1073 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1074 "%s timeleft: %ld\n",
1075 __func__, timeleft));
1076
1077 /*
1078 * If there is time left scsi_eh_done got called, and we will examine
1079 * the actual status codes to see whether the command actually did
1080 * complete normally, else if we have a zero return and no time left,
1081 * the command must still be pending, so abort it and return FAILED.
1082 * If we never actually managed to issue the command, because
1083 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1084 * value above (so don't execute either branch of the if)
1085 */
1086 if (timeleft) {
1087 rtn = scsi_eh_completed_normally(scmd);
1088 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1089 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1090
1091 switch (rtn) {
1092 case SUCCESS:
1093 case NEEDS_RETRY:
1094 case FAILED:
1095 break;
1096 case ADD_TO_MLQUEUE:
1097 rtn = NEEDS_RETRY;
1098 break;
1099 default:
1100 rtn = FAILED;
1101 break;
1102 }
1103 } else if (rtn != FAILED) {
1104 scsi_abort_eh_cmnd(scmd);
1105 rtn = FAILED;
1106 }
1107
1108 scsi_eh_restore_cmnd(scmd, &ses);
1109
1110 return rtn;
1111}
1112
1113/**
1114 * scsi_request_sense - Request sense data from a particular target.
1115 * @scmd: SCSI cmd for request sense.
1116 *
1117 * Notes:
1118 * Some hosts automatically obtain this information, others require
1119 * that we obtain it on our own. This function will *not* return until
1120 * the command either times out, or it completes.
1121 */
1122static int scsi_request_sense(struct scsi_cmnd *scmd)
1123{
1124 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1125}
1126
1127static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1128{
1129 if (!blk_rq_is_passthrough(scmd->request)) {
1130 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1131 if (sdrv->eh_action)
1132 rtn = sdrv->eh_action(scmd, rtn);
1133 }
1134 return rtn;
1135}
1136
1137/**
1138 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1139 * @scmd: Original SCSI cmd that eh has finished.
1140 * @done_q: Queue for processed commands.
1141 *
1142 * Notes:
1143 * We don't want to use the normal command completion while we are are
1144 * still handling errors - it may cause other commands to be queued,
1145 * and that would disturb what we are doing. Thus we really want to
1146 * keep a list of pending commands for final completion, and once we
1147 * are ready to leave error handling we handle completion for real.
1148 */
1149void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1150{
1151 list_move_tail(&scmd->eh_entry, done_q);
1152}
1153EXPORT_SYMBOL(scsi_eh_finish_cmd);
1154
1155/**
1156 * scsi_eh_get_sense - Get device sense data.
1157 * @work_q: Queue of commands to process.
1158 * @done_q: Queue of processed commands.
1159 *
1160 * Description:
1161 * See if we need to request sense information. if so, then get it
1162 * now, so we have a better idea of what to do.
1163 *
1164 * Notes:
1165 * This has the unfortunate side effect that if a shost adapter does
1166 * not automatically request sense information, we end up shutting
1167 * it down before we request it.
1168 *
1169 * All drivers should request sense information internally these days,
1170 * so for now all I have to say is tough noogies if you end up in here.
1171 *
1172 * XXX: Long term this code should go away, but that needs an audit of
1173 * all LLDDs first.
1174 */
1175int scsi_eh_get_sense(struct list_head *work_q,
1176 struct list_head *done_q)
1177{
1178 struct scsi_cmnd *scmd, *next;
1179 struct Scsi_Host *shost;
1180 int rtn;
1181
1182 /*
1183 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1184 * should not get sense.
1185 */
1186 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1187 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1188 SCSI_SENSE_VALID(scmd))
1189 continue;
1190
1191 shost = scmd->device->host;
1192 if (scsi_host_eh_past_deadline(shost)) {
1193 SCSI_LOG_ERROR_RECOVERY(3,
1194 scmd_printk(KERN_INFO, scmd,
1195 "%s: skip request sense, past eh deadline\n",
1196 current->comm));
1197 break;
1198 }
1199 if (status_byte(scmd->result) != CHECK_CONDITION)
1200 /*
1201 * don't request sense if there's no check condition
1202 * status because the error we're processing isn't one
1203 * that has a sense code (and some devices get
1204 * confused by sense requests out of the blue)
1205 */
1206 continue;
1207
1208 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1209 "%s: requesting sense\n",
1210 current->comm));
1211 rtn = scsi_request_sense(scmd);
1212 if (rtn != SUCCESS)
1213 continue;
1214
1215 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1216 "sense requested, result %x\n", scmd->result));
1217 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1218
1219 rtn = scsi_decide_disposition(scmd);
1220
1221 /*
1222 * if the result was normal, then just pass it along to the
1223 * upper level.
1224 */
1225 if (rtn == SUCCESS)
1226 /* we don't want this command reissued, just
1227 * finished with the sense data, so set
1228 * retries to the max allowed to ensure it
1229 * won't get reissued */
1230 scmd->retries = scmd->allowed;
1231 else if (rtn != NEEDS_RETRY)
1232 continue;
1233
1234 scsi_eh_finish_cmd(scmd, done_q);
1235 }
1236
1237 return list_empty(work_q);
1238}
1239EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1240
1241/**
1242 * scsi_eh_tur - Send TUR to device.
1243 * @scmd: &scsi_cmnd to send TUR
1244 *
1245 * Return value:
1246 * 0 - Device is ready. 1 - Device NOT ready.
1247 */
1248static int scsi_eh_tur(struct scsi_cmnd *scmd)
1249{
1250 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1251 int retry_cnt = 1, rtn;
1252
1253retry_tur:
1254 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1255 scmd->device->eh_timeout, 0);
1256
1257 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1258 "%s return: %x\n", __func__, rtn));
1259
1260 switch (rtn) {
1261 case NEEDS_RETRY:
1262 if (retry_cnt--)
1263 goto retry_tur;
1264 /*FALLTHRU*/
1265 case SUCCESS:
1266 return 0;
1267 default:
1268 return 1;
1269 }
1270}
1271
1272/**
1273 * scsi_eh_test_devices - check if devices are responding from error recovery.
1274 * @cmd_list: scsi commands in error recovery.
1275 * @work_q: queue for commands which still need more error recovery
1276 * @done_q: queue for commands which are finished
1277 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1278 *
1279 * Decription:
1280 * Tests if devices are in a working state. Commands to devices now in
1281 * a working state are sent to the done_q while commands to devices which
1282 * are still failing to respond are returned to the work_q for more
1283 * processing.
1284 **/
1285static int scsi_eh_test_devices(struct list_head *cmd_list,
1286 struct list_head *work_q,
1287 struct list_head *done_q, int try_stu)
1288{
1289 struct scsi_cmnd *scmd, *next;
1290 struct scsi_device *sdev;
1291 int finish_cmds;
1292
1293 while (!list_empty(cmd_list)) {
1294 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1295 sdev = scmd->device;
1296
1297 if (!try_stu) {
1298 if (scsi_host_eh_past_deadline(sdev->host)) {
1299 /* Push items back onto work_q */
1300 list_splice_init(cmd_list, work_q);
1301 SCSI_LOG_ERROR_RECOVERY(3,
1302 sdev_printk(KERN_INFO, sdev,
1303 "%s: skip test device, past eh deadline",
1304 current->comm));
1305 break;
1306 }
1307 }
1308
1309 finish_cmds = !scsi_device_online(scmd->device) ||
1310 (try_stu && !scsi_eh_try_stu(scmd) &&
1311 !scsi_eh_tur(scmd)) ||
1312 !scsi_eh_tur(scmd);
1313
1314 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1315 if (scmd->device == sdev) {
1316 if (finish_cmds &&
1317 (try_stu ||
1318 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1319 scsi_eh_finish_cmd(scmd, done_q);
1320 else
1321 list_move_tail(&scmd->eh_entry, work_q);
1322 }
1323 }
1324 return list_empty(work_q);
1325}
1326
1327/**
1328 * scsi_eh_try_stu - Send START_UNIT to device.
1329 * @scmd: &scsi_cmnd to send START_UNIT
1330 *
1331 * Return value:
1332 * 0 - Device is ready. 1 - Device NOT ready.
1333 */
1334static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1335{
1336 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1337
1338 if (scmd->device->allow_restart) {
1339 int i, rtn = NEEDS_RETRY;
1340
1341 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1342 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1343
1344 if (rtn == SUCCESS)
1345 return 0;
1346 }
1347
1348 return 1;
1349}
1350
1351 /**
1352 * scsi_eh_stu - send START_UNIT if needed
1353 * @shost: &scsi host being recovered.
1354 * @work_q: &list_head for pending commands.
1355 * @done_q: &list_head for processed commands.
1356 *
1357 * Notes:
1358 * If commands are failing due to not ready, initializing command required,
1359 * try revalidating the device, which will end up sending a start unit.
1360 */
1361static int scsi_eh_stu(struct Scsi_Host *shost,
1362 struct list_head *work_q,
1363 struct list_head *done_q)
1364{
1365 struct scsi_cmnd *scmd, *stu_scmd, *next;
1366 struct scsi_device *sdev;
1367
1368 shost_for_each_device(sdev, shost) {
1369 if (scsi_host_eh_past_deadline(shost)) {
1370 SCSI_LOG_ERROR_RECOVERY(3,
1371 sdev_printk(KERN_INFO, sdev,
1372 "%s: skip START_UNIT, past eh deadline\n",
1373 current->comm));
1374 break;
1375 }
1376 stu_scmd = NULL;
1377 list_for_each_entry(scmd, work_q, eh_entry)
1378 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1379 scsi_check_sense(scmd) == FAILED ) {
1380 stu_scmd = scmd;
1381 break;
1382 }
1383
1384 if (!stu_scmd)
1385 continue;
1386
1387 SCSI_LOG_ERROR_RECOVERY(3,
1388 sdev_printk(KERN_INFO, sdev,
1389 "%s: Sending START_UNIT\n",
1390 current->comm));
1391
1392 if (!scsi_eh_try_stu(stu_scmd)) {
1393 if (!scsi_device_online(sdev) ||
1394 !scsi_eh_tur(stu_scmd)) {
1395 list_for_each_entry_safe(scmd, next,
1396 work_q, eh_entry) {
1397 if (scmd->device == sdev &&
1398 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1399 scsi_eh_finish_cmd(scmd, done_q);
1400 }
1401 }
1402 } else {
1403 SCSI_LOG_ERROR_RECOVERY(3,
1404 sdev_printk(KERN_INFO, sdev,
1405 "%s: START_UNIT failed\n",
1406 current->comm));
1407 }
1408 }
1409
1410 return list_empty(work_q);
1411}
1412
1413
1414/**
1415 * scsi_eh_bus_device_reset - send bdr if needed
1416 * @shost: scsi host being recovered.
1417 * @work_q: &list_head for pending commands.
1418 * @done_q: &list_head for processed commands.
1419 *
1420 * Notes:
1421 * Try a bus device reset. Still, look to see whether we have multiple
1422 * devices that are jammed or not - if we have multiple devices, it
1423 * makes no sense to try bus_device_reset - we really would need to try
1424 * a bus_reset instead.
1425 */
1426static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1427 struct list_head *work_q,
1428 struct list_head *done_q)
1429{
1430 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1431 struct scsi_device *sdev;
1432 int rtn;
1433
1434 shost_for_each_device(sdev, shost) {
1435 if (scsi_host_eh_past_deadline(shost)) {
1436 SCSI_LOG_ERROR_RECOVERY(3,
1437 sdev_printk(KERN_INFO, sdev,
1438 "%s: skip BDR, past eh deadline\n",
1439 current->comm));
1440 break;
1441 }
1442 bdr_scmd = NULL;
1443 list_for_each_entry(scmd, work_q, eh_entry)
1444 if (scmd->device == sdev) {
1445 bdr_scmd = scmd;
1446 break;
1447 }
1448
1449 if (!bdr_scmd)
1450 continue;
1451
1452 SCSI_LOG_ERROR_RECOVERY(3,
1453 sdev_printk(KERN_INFO, sdev,
1454 "%s: Sending BDR\n", current->comm));
1455 rtn = scsi_try_bus_device_reset(bdr_scmd);
1456 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1457 if (!scsi_device_online(sdev) ||
1458 rtn == FAST_IO_FAIL ||
1459 !scsi_eh_tur(bdr_scmd)) {
1460 list_for_each_entry_safe(scmd, next,
1461 work_q, eh_entry) {
1462 if (scmd->device == sdev &&
1463 scsi_eh_action(scmd, rtn) != FAILED)
1464 scsi_eh_finish_cmd(scmd,
1465 done_q);
1466 }
1467 }
1468 } else {
1469 SCSI_LOG_ERROR_RECOVERY(3,
1470 sdev_printk(KERN_INFO, sdev,
1471 "%s: BDR failed\n", current->comm));
1472 }
1473 }
1474
1475 return list_empty(work_q);
1476}
1477
1478/**
1479 * scsi_eh_target_reset - send target reset if needed
1480 * @shost: scsi host being recovered.
1481 * @work_q: &list_head for pending commands.
1482 * @done_q: &list_head for processed commands.
1483 *
1484 * Notes:
1485 * Try a target reset.
1486 */
1487static int scsi_eh_target_reset(struct Scsi_Host *shost,
1488 struct list_head *work_q,
1489 struct list_head *done_q)
1490{
1491 LIST_HEAD(tmp_list);
1492 LIST_HEAD(check_list);
1493
1494 list_splice_init(work_q, &tmp_list);
1495
1496 while (!list_empty(&tmp_list)) {
1497 struct scsi_cmnd *next, *scmd;
1498 int rtn;
1499 unsigned int id;
1500
1501 if (scsi_host_eh_past_deadline(shost)) {
1502 /* push back on work queue for further processing */
1503 list_splice_init(&check_list, work_q);
1504 list_splice_init(&tmp_list, work_q);
1505 SCSI_LOG_ERROR_RECOVERY(3,
1506 shost_printk(KERN_INFO, shost,
1507 "%s: Skip target reset, past eh deadline\n",
1508 current->comm));
1509 return list_empty(work_q);
1510 }
1511
1512 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1513 id = scmd_id(scmd);
1514
1515 SCSI_LOG_ERROR_RECOVERY(3,
1516 shost_printk(KERN_INFO, shost,
1517 "%s: Sending target reset to target %d\n",
1518 current->comm, id));
1519 rtn = scsi_try_target_reset(scmd);
1520 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1521 SCSI_LOG_ERROR_RECOVERY(3,
1522 shost_printk(KERN_INFO, shost,
1523 "%s: Target reset failed"
1524 " target: %d\n",
1525 current->comm, id));
1526 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1527 if (scmd_id(scmd) != id)
1528 continue;
1529
1530 if (rtn == SUCCESS)
1531 list_move_tail(&scmd->eh_entry, &check_list);
1532 else if (rtn == FAST_IO_FAIL)
1533 scsi_eh_finish_cmd(scmd, done_q);
1534 else
1535 /* push back on work queue for further processing */
1536 list_move(&scmd->eh_entry, work_q);
1537 }
1538 }
1539
1540 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1541}
1542
1543/**
1544 * scsi_eh_bus_reset - send a bus reset
1545 * @shost: &scsi host being recovered.
1546 * @work_q: &list_head for pending commands.
1547 * @done_q: &list_head for processed commands.
1548 */
1549static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1550 struct list_head *work_q,
1551 struct list_head *done_q)
1552{
1553 struct scsi_cmnd *scmd, *chan_scmd, *next;
1554 LIST_HEAD(check_list);
1555 unsigned int channel;
1556 int rtn;
1557
1558 /*
1559 * we really want to loop over the various channels, and do this on
1560 * a channel by channel basis. we should also check to see if any
1561 * of the failed commands are on soft_reset devices, and if so, skip
1562 * the reset.
1563 */
1564
1565 for (channel = 0; channel <= shost->max_channel; channel++) {
1566 if (scsi_host_eh_past_deadline(shost)) {
1567 list_splice_init(&check_list, work_q);
1568 SCSI_LOG_ERROR_RECOVERY(3,
1569 shost_printk(KERN_INFO, shost,
1570 "%s: skip BRST, past eh deadline\n",
1571 current->comm));
1572 return list_empty(work_q);
1573 }
1574
1575 chan_scmd = NULL;
1576 list_for_each_entry(scmd, work_q, eh_entry) {
1577 if (channel == scmd_channel(scmd)) {
1578 chan_scmd = scmd;
1579 break;
1580 /*
1581 * FIXME add back in some support for
1582 * soft_reset devices.
1583 */
1584 }
1585 }
1586
1587 if (!chan_scmd)
1588 continue;
1589 SCSI_LOG_ERROR_RECOVERY(3,
1590 shost_printk(KERN_INFO, shost,
1591 "%s: Sending BRST chan: %d\n",
1592 current->comm, channel));
1593 rtn = scsi_try_bus_reset(chan_scmd);
1594 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1595 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1596 if (channel == scmd_channel(scmd)) {
1597 if (rtn == FAST_IO_FAIL)
1598 scsi_eh_finish_cmd(scmd,
1599 done_q);
1600 else
1601 list_move_tail(&scmd->eh_entry,
1602 &check_list);
1603 }
1604 }
1605 } else {
1606 SCSI_LOG_ERROR_RECOVERY(3,
1607 shost_printk(KERN_INFO, shost,
1608 "%s: BRST failed chan: %d\n",
1609 current->comm, channel));
1610 }
1611 }
1612 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1613}
1614
1615/**
1616 * scsi_eh_host_reset - send a host reset
1617 * @shost: host to be reset.
1618 * @work_q: &list_head for pending commands.
1619 * @done_q: &list_head for processed commands.
1620 */
1621static int scsi_eh_host_reset(struct Scsi_Host *shost,
1622 struct list_head *work_q,
1623 struct list_head *done_q)
1624{
1625 struct scsi_cmnd *scmd, *next;
1626 LIST_HEAD(check_list);
1627 int rtn;
1628
1629 if (!list_empty(work_q)) {
1630 scmd = list_entry(work_q->next,
1631 struct scsi_cmnd, eh_entry);
1632
1633 SCSI_LOG_ERROR_RECOVERY(3,
1634 shost_printk(KERN_INFO, shost,
1635 "%s: Sending HRST\n",
1636 current->comm));
1637
1638 rtn = scsi_try_host_reset(scmd);
1639 if (rtn == SUCCESS) {
1640 list_splice_init(work_q, &check_list);
1641 } else if (rtn == FAST_IO_FAIL) {
1642 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1643 scsi_eh_finish_cmd(scmd, done_q);
1644 }
1645 } else {
1646 SCSI_LOG_ERROR_RECOVERY(3,
1647 shost_printk(KERN_INFO, shost,
1648 "%s: HRST failed\n",
1649 current->comm));
1650 }
1651 }
1652 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1653}
1654
1655/**
1656 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1657 * @work_q: &list_head for pending commands.
1658 * @done_q: &list_head for processed commands.
1659 */
1660static void scsi_eh_offline_sdevs(struct list_head *work_q,
1661 struct list_head *done_q)
1662{
1663 struct scsi_cmnd *scmd, *next;
1664 struct scsi_device *sdev;
1665
1666 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1667 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1668 "not ready after error recovery\n");
1669 sdev = scmd->device;
1670
1671 mutex_lock(&sdev->state_mutex);
1672 scsi_device_set_state(sdev, SDEV_OFFLINE);
1673 mutex_unlock(&sdev->state_mutex);
1674
1675 scsi_eh_finish_cmd(scmd, done_q);
1676 }
1677 return;
1678}
1679
1680/**
1681 * scsi_noretry_cmd - determine if command should be failed fast
1682 * @scmd: SCSI cmd to examine.
1683 */
1684int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1685{
1686 switch (host_byte(scmd->result)) {
1687 case DID_OK:
1688 break;
1689 case DID_TIME_OUT:
1690 goto check_type;
1691 case DID_BUS_BUSY:
1692 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1693 case DID_PARITY:
1694 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1695 case DID_ERROR:
1696 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1697 status_byte(scmd->result) == RESERVATION_CONFLICT)
1698 return 0;
1699 /* fall through */
1700 case DID_SOFT_ERROR:
1701 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1702 }
1703
1704 if (status_byte(scmd->result) != CHECK_CONDITION)
1705 return 0;
1706
1707check_type:
1708 /*
1709 * assume caller has checked sense and determined
1710 * the check condition was retryable.
1711 */
1712 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1713 blk_rq_is_passthrough(scmd->request))
1714 return 1;
1715 else
1716 return 0;
1717}
1718
1719/**
1720 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1721 * @scmd: SCSI cmd to examine.
1722 *
1723 * Notes:
1724 * This is *only* called when we are examining the status after sending
1725 * out the actual data command. any commands that are queued for error
1726 * recovery (e.g. test_unit_ready) do *not* come through here.
1727 *
1728 * When this routine returns failed, it means the error handler thread
1729 * is woken. In cases where the error code indicates an error that
1730 * doesn't require the error handler read (i.e. we don't need to
1731 * abort/reset), this function should return SUCCESS.
1732 */
1733int scsi_decide_disposition(struct scsi_cmnd *scmd)
1734{
1735 int rtn;
1736
1737 /*
1738 * if the device is offline, then we clearly just pass the result back
1739 * up to the top level.
1740 */
1741 if (!scsi_device_online(scmd->device)) {
1742 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1743 "%s: device offline - report as SUCCESS\n", __func__));
1744 return SUCCESS;
1745 }
1746
1747 /*
1748 * first check the host byte, to see if there is anything in there
1749 * that would indicate what we need to do.
1750 */
1751 switch (host_byte(scmd->result)) {
1752 case DID_PASSTHROUGH:
1753 /*
1754 * no matter what, pass this through to the upper layer.
1755 * nuke this special code so that it looks like we are saying
1756 * did_ok.
1757 */
1758 scmd->result &= 0xff00ffff;
1759 return SUCCESS;
1760 case DID_OK:
1761 /*
1762 * looks good. drop through, and check the next byte.
1763 */
1764 break;
1765 case DID_ABORT:
1766 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1767 set_host_byte(scmd, DID_TIME_OUT);
1768 return SUCCESS;
1769 }
1770 /* FALLTHROUGH */
1771 case DID_NO_CONNECT:
1772 case DID_BAD_TARGET:
1773 /*
1774 * note - this means that we just report the status back
1775 * to the top level driver, not that we actually think
1776 * that it indicates SUCCESS.
1777 */
1778 return SUCCESS;
1779 case DID_SOFT_ERROR:
1780 /*
1781 * when the low level driver returns did_soft_error,
1782 * it is responsible for keeping an internal retry counter
1783 * in order to avoid endless loops (db)
1784 */
1785 goto maybe_retry;
1786 case DID_IMM_RETRY:
1787 return NEEDS_RETRY;
1788
1789 case DID_REQUEUE:
1790 return ADD_TO_MLQUEUE;
1791 case DID_TRANSPORT_DISRUPTED:
1792 /*
1793 * LLD/transport was disrupted during processing of the IO.
1794 * The transport class is now blocked/blocking,
1795 * and the transport will decide what to do with the IO
1796 * based on its timers and recovery capablilities if
1797 * there are enough retries.
1798 */
1799 goto maybe_retry;
1800 case DID_TRANSPORT_FAILFAST:
1801 /*
1802 * The transport decided to failfast the IO (most likely
1803 * the fast io fail tmo fired), so send IO directly upwards.
1804 */
1805 return SUCCESS;
1806 case DID_ERROR:
1807 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1808 status_byte(scmd->result) == RESERVATION_CONFLICT)
1809 /*
1810 * execute reservation conflict processing code
1811 * lower down
1812 */
1813 break;
1814 /* fallthrough */
1815 case DID_BUS_BUSY:
1816 case DID_PARITY:
1817 goto maybe_retry;
1818 case DID_TIME_OUT:
1819 /*
1820 * when we scan the bus, we get timeout messages for
1821 * these commands if there is no device available.
1822 * other hosts report did_no_connect for the same thing.
1823 */
1824 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1825 scmd->cmnd[0] == INQUIRY)) {
1826 return SUCCESS;
1827 } else {
1828 return FAILED;
1829 }
1830 case DID_RESET:
1831 return SUCCESS;
1832 default:
1833 return FAILED;
1834 }
1835
1836 /*
1837 * next, check the message byte.
1838 */
1839 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1840 return FAILED;
1841
1842 /*
1843 * check the status byte to see if this indicates anything special.
1844 */
1845 switch (status_byte(scmd->result)) {
1846 case QUEUE_FULL:
1847 scsi_handle_queue_full(scmd->device);
1848 /*
1849 * the case of trying to send too many commands to a
1850 * tagged queueing device.
1851 */
1852 /* FALLTHROUGH */
1853 case BUSY:
1854 /*
1855 * device can't talk to us at the moment. Should only
1856 * occur (SAM-3) when the task queue is empty, so will cause
1857 * the empty queue handling to trigger a stall in the
1858 * device.
1859 */
1860 return ADD_TO_MLQUEUE;
1861 case GOOD:
1862 if (scmd->cmnd[0] == REPORT_LUNS)
1863 scmd->device->sdev_target->expecting_lun_change = 0;
1864 scsi_handle_queue_ramp_up(scmd->device);
1865 /* FALLTHROUGH */
1866 case COMMAND_TERMINATED:
1867 return SUCCESS;
1868 case TASK_ABORTED:
1869 goto maybe_retry;
1870 case CHECK_CONDITION:
1871 rtn = scsi_check_sense(scmd);
1872 if (rtn == NEEDS_RETRY)
1873 goto maybe_retry;
1874 /* if rtn == FAILED, we have no sense information;
1875 * returning FAILED will wake the error handler thread
1876 * to collect the sense and redo the decide
1877 * disposition */
1878 return rtn;
1879 case CONDITION_GOOD:
1880 case INTERMEDIATE_GOOD:
1881 case INTERMEDIATE_C_GOOD:
1882 case ACA_ACTIVE:
1883 /*
1884 * who knows? FIXME(eric)
1885 */
1886 return SUCCESS;
1887
1888 case RESERVATION_CONFLICT:
1889 sdev_printk(KERN_INFO, scmd->device,
1890 "reservation conflict\n");
1891 set_host_byte(scmd, DID_NEXUS_FAILURE);
1892 return SUCCESS; /* causes immediate i/o error */
1893 default:
1894 return FAILED;
1895 }
1896 return FAILED;
1897
1898maybe_retry:
1899
1900 /* we requeue for retry because the error was retryable, and
1901 * the request was not marked fast fail. Note that above,
1902 * even if the request is marked fast fail, we still requeue
1903 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1904 if ((++scmd->retries) <= scmd->allowed
1905 && !scsi_noretry_cmd(scmd)) {
1906 return NEEDS_RETRY;
1907 } else {
1908 /*
1909 * no more retries - report this one back to upper level.
1910 */
1911 return SUCCESS;
1912 }
1913}
1914
1915static void eh_lock_door_done(struct request *req, blk_status_t status)
1916{
1917 __blk_put_request(req->q, req);
1918}
1919
1920/**
1921 * scsi_eh_lock_door - Prevent medium removal for the specified device
1922 * @sdev: SCSI device to prevent medium removal
1923 *
1924 * Locking:
1925 * We must be called from process context.
1926 *
1927 * Notes:
1928 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1929 * head of the devices request queue, and continue.
1930 */
1931static void scsi_eh_lock_door(struct scsi_device *sdev)
1932{
1933 struct request *req;
1934 struct scsi_request *rq;
1935
1936 /*
1937 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1938 * request becomes available
1939 */
1940 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL);
1941 if (IS_ERR(req))
1942 return;
1943 rq = scsi_req(req);
1944
1945 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1946 rq->cmd[1] = 0;
1947 rq->cmd[2] = 0;
1948 rq->cmd[3] = 0;
1949 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1950 rq->cmd[5] = 0;
1951 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1952
1953 req->rq_flags |= RQF_QUIET;
1954 req->timeout = 10 * HZ;
1955 rq->retries = 5;
1956
1957 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1958}
1959
1960/**
1961 * scsi_restart_operations - restart io operations to the specified host.
1962 * @shost: Host we are restarting.
1963 *
1964 * Notes:
1965 * When we entered the error handler, we blocked all further i/o to
1966 * this device. we need to 'reverse' this process.
1967 */
1968static void scsi_restart_operations(struct Scsi_Host *shost)
1969{
1970 struct scsi_device *sdev;
1971 unsigned long flags;
1972
1973 /*
1974 * If the door was locked, we need to insert a door lock request
1975 * onto the head of the SCSI request queue for the device. There
1976 * is no point trying to lock the door of an off-line device.
1977 */
1978 shost_for_each_device(sdev, shost) {
1979 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1980 scsi_eh_lock_door(sdev);
1981 sdev->was_reset = 0;
1982 }
1983 }
1984
1985 /*
1986 * next free up anything directly waiting upon the host. this
1987 * will be requests for character device operations, and also for
1988 * ioctls to queued block devices.
1989 */
1990 SCSI_LOG_ERROR_RECOVERY(3,
1991 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1992
1993 spin_lock_irqsave(shost->host_lock, flags);
1994 if (scsi_host_set_state(shost, SHOST_RUNNING))
1995 if (scsi_host_set_state(shost, SHOST_CANCEL))
1996 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1997 spin_unlock_irqrestore(shost->host_lock, flags);
1998
1999 wake_up(&shost->host_wait);
2000
2001 /*
2002 * finally we need to re-initiate requests that may be pending. we will
2003 * have had everything blocked while error handling is taking place, and
2004 * now that error recovery is done, we will need to ensure that these
2005 * requests are started.
2006 */
2007 scsi_run_host_queues(shost);
2008
2009 /*
2010 * if eh is active and host_eh_scheduled is pending we need to re-run
2011 * recovery. we do this check after scsi_run_host_queues() to allow
2012 * everything pent up since the last eh run a chance to make forward
2013 * progress before we sync again. Either we'll immediately re-run
2014 * recovery or scsi_device_unbusy() will wake us again when these
2015 * pending commands complete.
2016 */
2017 spin_lock_irqsave(shost->host_lock, flags);
2018 if (shost->host_eh_scheduled)
2019 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2020 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2021 spin_unlock_irqrestore(shost->host_lock, flags);
2022}
2023
2024/**
2025 * scsi_eh_ready_devs - check device ready state and recover if not.
2026 * @shost: host to be recovered.
2027 * @work_q: &list_head for pending commands.
2028 * @done_q: &list_head for processed commands.
2029 */
2030void scsi_eh_ready_devs(struct Scsi_Host *shost,
2031 struct list_head *work_q,
2032 struct list_head *done_q)
2033{
2034 if (!scsi_eh_stu(shost, work_q, done_q))
2035 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2036 if (!scsi_eh_target_reset(shost, work_q, done_q))
2037 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2038 if (!scsi_eh_host_reset(shost, work_q, done_q))
2039 scsi_eh_offline_sdevs(work_q,
2040 done_q);
2041}
2042EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2043
2044/**
2045 * scsi_eh_flush_done_q - finish processed commands or retry them.
2046 * @done_q: list_head of processed commands.
2047 */
2048void scsi_eh_flush_done_q(struct list_head *done_q)
2049{
2050 struct scsi_cmnd *scmd, *next;
2051
2052 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2053 list_del_init(&scmd->eh_entry);
2054 if (scsi_device_online(scmd->device) &&
2055 !scsi_noretry_cmd(scmd) &&
2056 (++scmd->retries <= scmd->allowed)) {
2057 SCSI_LOG_ERROR_RECOVERY(3,
2058 scmd_printk(KERN_INFO, scmd,
2059 "%s: flush retry cmd\n",
2060 current->comm));
2061 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2062 } else {
2063 /*
2064 * If just we got sense for the device (called
2065 * scsi_eh_get_sense), scmd->result is already
2066 * set, do not set DRIVER_TIMEOUT.
2067 */
2068 if (!scmd->result)
2069 scmd->result |= (DRIVER_TIMEOUT << 24);
2070 SCSI_LOG_ERROR_RECOVERY(3,
2071 scmd_printk(KERN_INFO, scmd,
2072 "%s: flush finish cmd\n",
2073 current->comm));
2074 scsi_finish_command(scmd);
2075 }
2076 }
2077}
2078EXPORT_SYMBOL(scsi_eh_flush_done_q);
2079
2080/**
2081 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2082 * @shost: Host to unjam.
2083 *
2084 * Notes:
2085 * When we come in here, we *know* that all commands on the bus have
2086 * either completed, failed or timed out. we also know that no further
2087 * commands are being sent to the host, so things are relatively quiet
2088 * and we have freedom to fiddle with things as we wish.
2089 *
2090 * This is only the *default* implementation. it is possible for
2091 * individual drivers to supply their own version of this function, and
2092 * if the maintainer wishes to do this, it is strongly suggested that
2093 * this function be taken as a template and modified. this function
2094 * was designed to correctly handle problems for about 95% of the
2095 * different cases out there, and it should always provide at least a
2096 * reasonable amount of error recovery.
2097 *
2098 * Any command marked 'failed' or 'timeout' must eventually have
2099 * scsi_finish_cmd() called for it. we do all of the retry stuff
2100 * here, so when we restart the host after we return it should have an
2101 * empty queue.
2102 */
2103static void scsi_unjam_host(struct Scsi_Host *shost)
2104{
2105 unsigned long flags;
2106 LIST_HEAD(eh_work_q);
2107 LIST_HEAD(eh_done_q);
2108
2109 spin_lock_irqsave(shost->host_lock, flags);
2110 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2111 spin_unlock_irqrestore(shost->host_lock, flags);
2112
2113 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2114
2115 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2116 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2117
2118 spin_lock_irqsave(shost->host_lock, flags);
2119 if (shost->eh_deadline != -1)
2120 shost->last_reset = 0;
2121 spin_unlock_irqrestore(shost->host_lock, flags);
2122 scsi_eh_flush_done_q(&eh_done_q);
2123}
2124
2125/**
2126 * scsi_error_handler - SCSI error handler thread
2127 * @data: Host for which we are running.
2128 *
2129 * Notes:
2130 * This is the main error handling loop. This is run as a kernel thread
2131 * for every SCSI host and handles all error handling activity.
2132 */
2133int scsi_error_handler(void *data)
2134{
2135 struct Scsi_Host *shost = data;
2136
2137 /*
2138 * We use TASK_INTERRUPTIBLE so that the thread is not
2139 * counted against the load average as a running process.
2140 * We never actually get interrupted because kthread_run
2141 * disables signal delivery for the created thread.
2142 */
2143 while (true) {
2144 /*
2145 * The sequence in kthread_stop() sets the stop flag first
2146 * then wakes the process. To avoid missed wakeups, the task
2147 * should always be in a non running state before the stop
2148 * flag is checked
2149 */
2150 set_current_state(TASK_INTERRUPTIBLE);
2151 if (kthread_should_stop())
2152 break;
2153
2154 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2155 shost->host_failed != atomic_read(&shost->host_busy)) {
2156 SCSI_LOG_ERROR_RECOVERY(1,
2157 shost_printk(KERN_INFO, shost,
2158 "scsi_eh_%d: sleeping\n",
2159 shost->host_no));
2160 schedule();
2161 continue;
2162 }
2163
2164 __set_current_state(TASK_RUNNING);
2165 SCSI_LOG_ERROR_RECOVERY(1,
2166 shost_printk(KERN_INFO, shost,
2167 "scsi_eh_%d: waking up %d/%d/%d\n",
2168 shost->host_no, shost->host_eh_scheduled,
2169 shost->host_failed,
2170 atomic_read(&shost->host_busy)));
2171
2172 /*
2173 * We have a host that is failing for some reason. Figure out
2174 * what we need to do to get it up and online again (if we can).
2175 * If we fail, we end up taking the thing offline.
2176 */
2177 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2178 SCSI_LOG_ERROR_RECOVERY(1,
2179 shost_printk(KERN_ERR, shost,
2180 "scsi_eh_%d: unable to autoresume\n",
2181 shost->host_no));
2182 continue;
2183 }
2184
2185 if (shost->transportt->eh_strategy_handler)
2186 shost->transportt->eh_strategy_handler(shost);
2187 else
2188 scsi_unjam_host(shost);
2189
2190 /* All scmds have been handled */
2191 shost->host_failed = 0;
2192
2193 /*
2194 * Note - if the above fails completely, the action is to take
2195 * individual devices offline and flush the queue of any
2196 * outstanding requests that may have been pending. When we
2197 * restart, we restart any I/O to any other devices on the bus
2198 * which are still online.
2199 */
2200 scsi_restart_operations(shost);
2201 if (!shost->eh_noresume)
2202 scsi_autopm_put_host(shost);
2203 }
2204 __set_current_state(TASK_RUNNING);
2205
2206 SCSI_LOG_ERROR_RECOVERY(1,
2207 shost_printk(KERN_INFO, shost,
2208 "Error handler scsi_eh_%d exiting\n",
2209 shost->host_no));
2210 shost->ehandler = NULL;
2211 return 0;
2212}
2213
2214/*
2215 * Function: scsi_report_bus_reset()
2216 *
2217 * Purpose: Utility function used by low-level drivers to report that
2218 * they have observed a bus reset on the bus being handled.
2219 *
2220 * Arguments: shost - Host in question
2221 * channel - channel on which reset was observed.
2222 *
2223 * Returns: Nothing
2224 *
2225 * Lock status: Host lock must be held.
2226 *
2227 * Notes: This only needs to be called if the reset is one which
2228 * originates from an unknown location. Resets originated
2229 * by the mid-level itself don't need to call this, but there
2230 * should be no harm.
2231 *
2232 * The main purpose of this is to make sure that a CHECK_CONDITION
2233 * is properly treated.
2234 */
2235void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2236{
2237 struct scsi_device *sdev;
2238
2239 __shost_for_each_device(sdev, shost) {
2240 if (channel == sdev_channel(sdev))
2241 __scsi_report_device_reset(sdev, NULL);
2242 }
2243}
2244EXPORT_SYMBOL(scsi_report_bus_reset);
2245
2246/*
2247 * Function: scsi_report_device_reset()
2248 *
2249 * Purpose: Utility function used by low-level drivers to report that
2250 * they have observed a device reset on the device being handled.
2251 *
2252 * Arguments: shost - Host in question
2253 * channel - channel on which reset was observed
2254 * target - target on which reset was observed
2255 *
2256 * Returns: Nothing
2257 *
2258 * Lock status: Host lock must be held
2259 *
2260 * Notes: This only needs to be called if the reset is one which
2261 * originates from an unknown location. Resets originated
2262 * by the mid-level itself don't need to call this, but there
2263 * should be no harm.
2264 *
2265 * The main purpose of this is to make sure that a CHECK_CONDITION
2266 * is properly treated.
2267 */
2268void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2269{
2270 struct scsi_device *sdev;
2271
2272 __shost_for_each_device(sdev, shost) {
2273 if (channel == sdev_channel(sdev) &&
2274 target == sdev_id(sdev))
2275 __scsi_report_device_reset(sdev, NULL);
2276 }
2277}
2278EXPORT_SYMBOL(scsi_report_device_reset);
2279
2280static void
2281scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2282{
2283}
2284
2285/**
2286 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2287 * @dev: scsi_device to operate on
2288 * @arg: reset type (see sg.h)
2289 */
2290int
2291scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2292{
2293 struct scsi_cmnd *scmd;
2294 struct Scsi_Host *shost = dev->host;
2295 struct request *rq;
2296 unsigned long flags;
2297 int error = 0, rtn, val;
2298
2299 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2300 return -EACCES;
2301
2302 error = get_user(val, arg);
2303 if (error)
2304 return error;
2305
2306 if (scsi_autopm_get_host(shost) < 0)
2307 return -EIO;
2308
2309 error = -EIO;
2310 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2311 shost->hostt->cmd_size, GFP_KERNEL);
2312 if (!rq)
2313 goto out_put_autopm_host;
2314 blk_rq_init(NULL, rq);
2315
2316 scmd = (struct scsi_cmnd *)(rq + 1);
2317 scsi_init_command(dev, scmd);
2318 scmd->request = rq;
2319 scmd->cmnd = scsi_req(rq)->cmd;
2320
2321 scmd->scsi_done = scsi_reset_provider_done_command;
2322 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2323
2324 scmd->cmd_len = 0;
2325
2326 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2327
2328 spin_lock_irqsave(shost->host_lock, flags);
2329 shost->tmf_in_progress = 1;
2330 spin_unlock_irqrestore(shost->host_lock, flags);
2331
2332 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2333 case SG_SCSI_RESET_NOTHING:
2334 rtn = SUCCESS;
2335 break;
2336 case SG_SCSI_RESET_DEVICE:
2337 rtn = scsi_try_bus_device_reset(scmd);
2338 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2339 break;
2340 /* FALLTHROUGH */
2341 case SG_SCSI_RESET_TARGET:
2342 rtn = scsi_try_target_reset(scmd);
2343 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2344 break;
2345 /* FALLTHROUGH */
2346 case SG_SCSI_RESET_BUS:
2347 rtn = scsi_try_bus_reset(scmd);
2348 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2349 break;
2350 /* FALLTHROUGH */
2351 case SG_SCSI_RESET_HOST:
2352 rtn = scsi_try_host_reset(scmd);
2353 if (rtn == SUCCESS)
2354 break;
2355 /* FALLTHROUGH */
2356 default:
2357 rtn = FAILED;
2358 break;
2359 }
2360
2361 error = (rtn == SUCCESS) ? 0 : -EIO;
2362
2363 spin_lock_irqsave(shost->host_lock, flags);
2364 shost->tmf_in_progress = 0;
2365 spin_unlock_irqrestore(shost->host_lock, flags);
2366
2367 /*
2368 * be sure to wake up anyone who was sleeping or had their queue
2369 * suspended while we performed the TMF.
2370 */
2371 SCSI_LOG_ERROR_RECOVERY(3,
2372 shost_printk(KERN_INFO, shost,
2373 "waking up host to restart after TMF\n"));
2374
2375 wake_up(&shost->host_wait);
2376 scsi_run_host_queues(shost);
2377
2378 scsi_put_command(scmd);
2379 kfree(rq);
2380
2381out_put_autopm_host:
2382 scsi_autopm_put_host(shost);
2383 return error;
2384}
2385EXPORT_SYMBOL(scsi_ioctl_reset);
2386
2387bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2388 struct scsi_sense_hdr *sshdr)
2389{
2390 return scsi_normalize_sense(cmd->sense_buffer,
2391 SCSI_SENSE_BUFFERSIZE, sshdr);
2392}
2393EXPORT_SYMBOL(scsi_command_normalize_sense);
2394
2395/**
2396 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2397 * @sense_buffer: byte array of sense data
2398 * @sb_len: number of valid bytes in sense_buffer
2399 * @info_out: pointer to 64 integer where 8 or 4 byte information
2400 * field will be placed if found.
2401 *
2402 * Return value:
2403 * true if information field found, false if not found.
2404 */
2405bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2406 u64 *info_out)
2407{
2408 const u8 * ucp;
2409
2410 if (sb_len < 7)
2411 return false;
2412 switch (sense_buffer[0] & 0x7f) {
2413 case 0x70:
2414 case 0x71:
2415 if (sense_buffer[0] & 0x80) {
2416 *info_out = get_unaligned_be32(&sense_buffer[3]);
2417 return true;
2418 }
2419 return false;
2420 case 0x72:
2421 case 0x73:
2422 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2423 0 /* info desc */);
2424 if (ucp && (0xa == ucp[1])) {
2425 *info_out = get_unaligned_be64(&ucp[4]);
2426 return true;
2427 }
2428 return false;
2429 default:
2430 return false;
2431 }
2432}
2433EXPORT_SYMBOL(scsi_get_sense_info_fld);