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// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 *
5 * SCSI error/timeout handling
6 * Initial versions: Eric Youngdale. Based upon conversations with
7 * Leonard Zubkoff and David Miller at Linux Expo,
8 * ideas originating from all over the place.
9 *
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 *
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * minor cleanups.
15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
16 */
17
18#include <linux/module.h>
19#include <linux/sched.h>
20#include <linux/gfp.h>
21#include <linux/timer.h>
22#include <linux/string.h>
23#include <linux/kernel.h>
24#include <linux/freezer.h>
25#include <linux/kthread.h>
26#include <linux/interrupt.h>
27#include <linux/blkdev.h>
28#include <linux/delay.h>
29#include <linux/jiffies.h>
30
31#include <scsi/scsi.h>
32#include <scsi/scsi_cmnd.h>
33#include <scsi/scsi_dbg.h>
34#include <scsi/scsi_device.h>
35#include <scsi/scsi_driver.h>
36#include <scsi/scsi_eh.h>
37#include <scsi/scsi_common.h>
38#include <scsi/scsi_transport.h>
39#include <scsi/scsi_host.h>
40#include <scsi/scsi_ioctl.h>
41#include <scsi/scsi_dh.h>
42#include <scsi/scsi_devinfo.h>
43#include <scsi/sg.h>
44
45#include "scsi_priv.h"
46#include "scsi_logging.h"
47#include "scsi_transport_api.h"
48
49#include <trace/events/scsi.h>
50
51#include <asm/unaligned.h>
52
53static void scsi_eh_done(struct scsi_cmnd *scmd);
54
55/*
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
58 */
59#define BUS_RESET_SETTLE_TIME (10)
60#define HOST_RESET_SETTLE_TIME (10)
61
62static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
63static int scsi_try_to_abort_cmd(struct scsi_host_template *,
64 struct scsi_cmnd *);
65
66void scsi_eh_wakeup(struct Scsi_Host *shost)
67{
68 lockdep_assert_held(shost->host_lock);
69
70 if (scsi_host_busy(shost) == shost->host_failed) {
71 trace_scsi_eh_wakeup(shost);
72 wake_up_process(shost->ehandler);
73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74 "Waking error handler thread\n"));
75 }
76}
77
78/**
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
81 *
82 * Schedule SCSI EH without scmd.
83 */
84void scsi_schedule_eh(struct Scsi_Host *shost)
85{
86 unsigned long flags;
87
88 spin_lock_irqsave(shost->host_lock, flags);
89
90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
92 shost->host_eh_scheduled++;
93 scsi_eh_wakeup(shost);
94 }
95
96 spin_unlock_irqrestore(shost->host_lock, flags);
97}
98EXPORT_SYMBOL_GPL(scsi_schedule_eh);
99
100static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
101{
102 if (!shost->last_reset || shost->eh_deadline == -1)
103 return 0;
104
105 /*
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
109 * if eh_deadline has been set to 'off' during the
110 * time_before call.
111 */
112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
113 shost->eh_deadline > -1)
114 return 0;
115
116 return 1;
117}
118
119/**
120 * scmd_eh_abort_handler - Handle command aborts
121 * @work: command to be aborted.
122 *
123 * Note: this function must be called only for a command that has timed out.
124 * Because the block layer marks a request as complete before it calls
125 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
126 * timed out do not have any effect. Hence it is safe to call
127 * scsi_finish_command() from this function.
128 */
129void
130scmd_eh_abort_handler(struct work_struct *work)
131{
132 struct scsi_cmnd *scmd =
133 container_of(work, struct scsi_cmnd, abort_work.work);
134 struct scsi_device *sdev = scmd->device;
135 int rtn;
136
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 aborting\n"));
141 } else {
142 SCSI_LOG_ERROR_RECOVERY(3,
143 scmd_printk(KERN_INFO, scmd,
144 "aborting command\n"));
145 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
146 if (rtn == SUCCESS) {
147 set_host_byte(scmd, DID_TIME_OUT);
148 if (scsi_host_eh_past_deadline(sdev->host)) {
149 SCSI_LOG_ERROR_RECOVERY(3,
150 scmd_printk(KERN_INFO, scmd,
151 "eh timeout, not retrying "
152 "aborted command\n"));
153 } else if (!scsi_noretry_cmd(scmd) &&
154 (++scmd->retries <= scmd->allowed)) {
155 SCSI_LOG_ERROR_RECOVERY(3,
156 scmd_printk(KERN_WARNING, scmd,
157 "retry aborted command\n"));
158 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
159 return;
160 } else {
161 SCSI_LOG_ERROR_RECOVERY(3,
162 scmd_printk(KERN_WARNING, scmd,
163 "finish aborted command\n"));
164 scsi_finish_command(scmd);
165 return;
166 }
167 } else {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
170 "cmd abort %s\n",
171 (rtn == FAST_IO_FAIL) ?
172 "not send" : "failed"));
173 }
174 }
175
176 scsi_eh_scmd_add(scmd);
177}
178
179/**
180 * scsi_abort_command - schedule a command abort
181 * @scmd: scmd to abort.
182 *
183 * We only need to abort commands after a command timeout
184 */
185static int
186scsi_abort_command(struct scsi_cmnd *scmd)
187{
188 struct scsi_device *sdev = scmd->device;
189 struct Scsi_Host *shost = sdev->host;
190 unsigned long flags;
191
192 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
193 /*
194 * Retry after abort failed, escalate to next level.
195 */
196 SCSI_LOG_ERROR_RECOVERY(3,
197 scmd_printk(KERN_INFO, scmd,
198 "previous abort failed\n"));
199 BUG_ON(delayed_work_pending(&scmd->abort_work));
200 return FAILED;
201 }
202
203 spin_lock_irqsave(shost->host_lock, flags);
204 if (shost->eh_deadline != -1 && !shost->last_reset)
205 shost->last_reset = jiffies;
206 spin_unlock_irqrestore(shost->host_lock, flags);
207
208 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
209 SCSI_LOG_ERROR_RECOVERY(3,
210 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
211 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
212 return SUCCESS;
213}
214
215/**
216 * scsi_eh_reset - call into ->eh_action to reset internal counters
217 * @scmd: scmd to run eh on.
218 *
219 * The scsi driver might be carrying internal state about the
220 * devices, so we need to call into the driver to reset the
221 * internal state once the error handler is started.
222 */
223static void scsi_eh_reset(struct scsi_cmnd *scmd)
224{
225 if (!blk_rq_is_passthrough(scmd->request)) {
226 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
227 if (sdrv->eh_reset)
228 sdrv->eh_reset(scmd);
229 }
230}
231
232static void scsi_eh_inc_host_failed(struct rcu_head *head)
233{
234 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
235 struct Scsi_Host *shost = scmd->device->host;
236 unsigned long flags;
237
238 spin_lock_irqsave(shost->host_lock, flags);
239 shost->host_failed++;
240 scsi_eh_wakeup(shost);
241 spin_unlock_irqrestore(shost->host_lock, flags);
242}
243
244/**
245 * scsi_eh_scmd_add - add scsi cmd to error handling.
246 * @scmd: scmd to run eh on.
247 */
248void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
249{
250 struct Scsi_Host *shost = scmd->device->host;
251 unsigned long flags;
252 int ret;
253
254 WARN_ON_ONCE(!shost->ehandler);
255
256 spin_lock_irqsave(shost->host_lock, flags);
257 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
258 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
259 WARN_ON_ONCE(ret);
260 }
261 if (shost->eh_deadline != -1 && !shost->last_reset)
262 shost->last_reset = jiffies;
263
264 scsi_eh_reset(scmd);
265 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
266 spin_unlock_irqrestore(shost->host_lock, flags);
267 /*
268 * Ensure that all tasks observe the host state change before the
269 * host_failed change.
270 */
271 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
272}
273
274/**
275 * scsi_times_out - Timeout function for normal scsi commands.
276 * @req: request that is timing out.
277 *
278 * Notes:
279 * We do not need to lock this. There is the potential for a race
280 * only in that the normal completion handling might run, but if the
281 * normal completion function determines that the timer has already
282 * fired, then it mustn't do anything.
283 */
284enum blk_eh_timer_return scsi_times_out(struct request *req)
285{
286 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
287 enum blk_eh_timer_return rtn = BLK_EH_DONE;
288 struct Scsi_Host *host = scmd->device->host;
289
290 trace_scsi_dispatch_cmd_timeout(scmd);
291 scsi_log_completion(scmd, TIMEOUT_ERROR);
292
293 if (host->eh_deadline != -1 && !host->last_reset)
294 host->last_reset = jiffies;
295
296 if (host->hostt->eh_timed_out)
297 rtn = host->hostt->eh_timed_out(scmd);
298
299 if (rtn == BLK_EH_DONE) {
300 /*
301 * Set the command to complete first in order to prevent a real
302 * completion from releasing the command while error handling
303 * is using it. If the command was already completed, then the
304 * lower level driver beat the timeout handler, and it is safe
305 * to return without escalating error recovery.
306 *
307 * If timeout handling lost the race to a real completion, the
308 * block layer may ignore that due to a fake timeout injection,
309 * so return RESET_TIMER to allow error handling another shot
310 * at this command.
311 */
312 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
313 return BLK_EH_RESET_TIMER;
314 if (scsi_abort_command(scmd) != SUCCESS) {
315 set_host_byte(scmd, DID_TIME_OUT);
316 scsi_eh_scmd_add(scmd);
317 }
318 }
319
320 return rtn;
321}
322
323/**
324 * scsi_block_when_processing_errors - Prevent cmds from being queued.
325 * @sdev: Device on which we are performing recovery.
326 *
327 * Description:
328 * We block until the host is out of error recovery, and then check to
329 * see whether the host or the device is offline.
330 *
331 * Return value:
332 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
333 */
334int scsi_block_when_processing_errors(struct scsi_device *sdev)
335{
336 int online;
337
338 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
339
340 online = scsi_device_online(sdev);
341
342 return online;
343}
344EXPORT_SYMBOL(scsi_block_when_processing_errors);
345
346#ifdef CONFIG_SCSI_LOGGING
347/**
348 * scsi_eh_prt_fail_stats - Log info on failures.
349 * @shost: scsi host being recovered.
350 * @work_q: Queue of scsi cmds to process.
351 */
352static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
353 struct list_head *work_q)
354{
355 struct scsi_cmnd *scmd;
356 struct scsi_device *sdev;
357 int total_failures = 0;
358 int cmd_failed = 0;
359 int cmd_cancel = 0;
360 int devices_failed = 0;
361
362 shost_for_each_device(sdev, shost) {
363 list_for_each_entry(scmd, work_q, eh_entry) {
364 if (scmd->device == sdev) {
365 ++total_failures;
366 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
367 ++cmd_cancel;
368 else
369 ++cmd_failed;
370 }
371 }
372
373 if (cmd_cancel || cmd_failed) {
374 SCSI_LOG_ERROR_RECOVERY(3,
375 shost_printk(KERN_INFO, shost,
376 "%s: cmds failed: %d, cancel: %d\n",
377 __func__, cmd_failed,
378 cmd_cancel));
379 cmd_cancel = 0;
380 cmd_failed = 0;
381 ++devices_failed;
382 }
383 }
384
385 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
386 "Total of %d commands on %d"
387 " devices require eh work\n",
388 total_failures, devices_failed));
389}
390#endif
391
392 /**
393 * scsi_report_lun_change - Set flag on all *other* devices on the same target
394 * to indicate that a UNIT ATTENTION is expected.
395 * @sdev: Device reporting the UNIT ATTENTION
396 */
397static void scsi_report_lun_change(struct scsi_device *sdev)
398{
399 sdev->sdev_target->expecting_lun_change = 1;
400}
401
402/**
403 * scsi_report_sense - Examine scsi sense information and log messages for
404 * certain conditions, also issue uevents for some of them.
405 * @sdev: Device reporting the sense code
406 * @sshdr: sshdr to be examined
407 */
408static void scsi_report_sense(struct scsi_device *sdev,
409 struct scsi_sense_hdr *sshdr)
410{
411 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
412
413 if (sshdr->sense_key == UNIT_ATTENTION) {
414 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
415 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
416 sdev_printk(KERN_WARNING, sdev,
417 "Inquiry data has changed");
418 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
419 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
420 scsi_report_lun_change(sdev);
421 sdev_printk(KERN_WARNING, sdev,
422 "Warning! Received an indication that the "
423 "LUN assignments on this target have "
424 "changed. The Linux SCSI layer does not "
425 "automatically remap LUN assignments.\n");
426 } else if (sshdr->asc == 0x3f)
427 sdev_printk(KERN_WARNING, sdev,
428 "Warning! Received an indication that the "
429 "operating parameters on this target have "
430 "changed. The Linux SCSI layer does not "
431 "automatically adjust these parameters.\n");
432
433 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
434 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
435 sdev_printk(KERN_WARNING, sdev,
436 "Warning! Received an indication that the "
437 "LUN reached a thin provisioning soft "
438 "threshold.\n");
439 }
440
441 if (sshdr->asc == 0x29) {
442 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
443 sdev_printk(KERN_WARNING, sdev,
444 "Power-on or device reset occurred\n");
445 }
446
447 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
448 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
449 sdev_printk(KERN_WARNING, sdev,
450 "Mode parameters changed");
451 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
452 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
453 sdev_printk(KERN_WARNING, sdev,
454 "Asymmetric access state changed");
455 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
456 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
457 sdev_printk(KERN_WARNING, sdev,
458 "Capacity data has changed");
459 } else if (sshdr->asc == 0x2a)
460 sdev_printk(KERN_WARNING, sdev,
461 "Parameters changed");
462 }
463
464 if (evt_type != SDEV_EVT_MAXBITS) {
465 set_bit(evt_type, sdev->pending_events);
466 schedule_work(&sdev->event_work);
467 }
468}
469
470/**
471 * scsi_check_sense - Examine scsi cmd sense
472 * @scmd: Cmd to have sense checked.
473 *
474 * Return value:
475 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
476 *
477 * Notes:
478 * When a deferred error is detected the current command has
479 * not been executed and needs retrying.
480 */
481int scsi_check_sense(struct scsi_cmnd *scmd)
482{
483 struct scsi_device *sdev = scmd->device;
484 struct scsi_sense_hdr sshdr;
485
486 if (! scsi_command_normalize_sense(scmd, &sshdr))
487 return FAILED; /* no valid sense data */
488
489 scsi_report_sense(sdev, &sshdr);
490
491 if (scsi_sense_is_deferred(&sshdr))
492 return NEEDS_RETRY;
493
494 if (sdev->handler && sdev->handler->check_sense) {
495 int rc;
496
497 rc = sdev->handler->check_sense(sdev, &sshdr);
498 if (rc != SCSI_RETURN_NOT_HANDLED)
499 return rc;
500 /* handler does not care. Drop down to default handling */
501 }
502
503 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
504 /*
505 * nasty: for mid-layer issued TURs, we need to return the
506 * actual sense data without any recovery attempt. For eh
507 * issued ones, we need to try to recover and interpret
508 */
509 return SUCCESS;
510
511 /*
512 * Previous logic looked for FILEMARK, EOM or ILI which are
513 * mainly associated with tapes and returned SUCCESS.
514 */
515 if (sshdr.response_code == 0x70) {
516 /* fixed format */
517 if (scmd->sense_buffer[2] & 0xe0)
518 return SUCCESS;
519 } else {
520 /*
521 * descriptor format: look for "stream commands sense data
522 * descriptor" (see SSC-3). Assume single sense data
523 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
524 */
525 if ((sshdr.additional_length > 3) &&
526 (scmd->sense_buffer[8] == 0x4) &&
527 (scmd->sense_buffer[11] & 0xe0))
528 return SUCCESS;
529 }
530
531 switch (sshdr.sense_key) {
532 case NO_SENSE:
533 return SUCCESS;
534 case RECOVERED_ERROR:
535 return /* soft_error */ SUCCESS;
536
537 case ABORTED_COMMAND:
538 if (sshdr.asc == 0x10) /* DIF */
539 return SUCCESS;
540
541 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
542 return ADD_TO_MLQUEUE;
543 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
544 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
545 return ADD_TO_MLQUEUE;
546
547 return NEEDS_RETRY;
548 case NOT_READY:
549 case UNIT_ATTENTION:
550 /*
551 * if we are expecting a cc/ua because of a bus reset that we
552 * performed, treat this just as a retry. otherwise this is
553 * information that we should pass up to the upper-level driver
554 * so that we can deal with it there.
555 */
556 if (scmd->device->expecting_cc_ua) {
557 /*
558 * Because some device does not queue unit
559 * attentions correctly, we carefully check
560 * additional sense code and qualifier so as
561 * not to squash media change unit attention.
562 */
563 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
564 scmd->device->expecting_cc_ua = 0;
565 return NEEDS_RETRY;
566 }
567 }
568 /*
569 * we might also expect a cc/ua if another LUN on the target
570 * reported a UA with an ASC/ASCQ of 3F 0E -
571 * REPORTED LUNS DATA HAS CHANGED.
572 */
573 if (scmd->device->sdev_target->expecting_lun_change &&
574 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
575 return NEEDS_RETRY;
576 /*
577 * if the device is in the process of becoming ready, we
578 * should retry.
579 */
580 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
581 return NEEDS_RETRY;
582 /*
583 * if the device is not started, we need to wake
584 * the error handler to start the motor
585 */
586 if (scmd->device->allow_restart &&
587 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
588 return FAILED;
589 /*
590 * Pass the UA upwards for a determination in the completion
591 * functions.
592 */
593 return SUCCESS;
594
595 /* these are not supported */
596 case DATA_PROTECT:
597 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
598 /* Thin provisioning hard threshold reached */
599 set_host_byte(scmd, DID_ALLOC_FAILURE);
600 return SUCCESS;
601 }
602 /* FALLTHROUGH */
603 case COPY_ABORTED:
604 case VOLUME_OVERFLOW:
605 case MISCOMPARE:
606 case BLANK_CHECK:
607 set_host_byte(scmd, DID_TARGET_FAILURE);
608 return SUCCESS;
609
610 case MEDIUM_ERROR:
611 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
612 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
613 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
614 set_host_byte(scmd, DID_MEDIUM_ERROR);
615 return SUCCESS;
616 }
617 return NEEDS_RETRY;
618
619 case HARDWARE_ERROR:
620 if (scmd->device->retry_hwerror)
621 return ADD_TO_MLQUEUE;
622 else
623 set_host_byte(scmd, DID_TARGET_FAILURE);
624 /* FALLTHROUGH */
625
626 case ILLEGAL_REQUEST:
627 if (sshdr.asc == 0x20 || /* Invalid command operation code */
628 sshdr.asc == 0x21 || /* Logical block address out of range */
629 sshdr.asc == 0x22 || /* Invalid function */
630 sshdr.asc == 0x24 || /* Invalid field in cdb */
631 sshdr.asc == 0x26 || /* Parameter value invalid */
632 sshdr.asc == 0x27) { /* Write protected */
633 set_host_byte(scmd, DID_TARGET_FAILURE);
634 }
635 return SUCCESS;
636
637 default:
638 return SUCCESS;
639 }
640}
641EXPORT_SYMBOL_GPL(scsi_check_sense);
642
643static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
644{
645 struct scsi_host_template *sht = sdev->host->hostt;
646 struct scsi_device *tmp_sdev;
647
648 if (!sht->track_queue_depth ||
649 sdev->queue_depth >= sdev->max_queue_depth)
650 return;
651
652 if (time_before(jiffies,
653 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
654 return;
655
656 if (time_before(jiffies,
657 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
658 return;
659
660 /*
661 * Walk all devices of a target and do
662 * ramp up on them.
663 */
664 shost_for_each_device(tmp_sdev, sdev->host) {
665 if (tmp_sdev->channel != sdev->channel ||
666 tmp_sdev->id != sdev->id ||
667 tmp_sdev->queue_depth == sdev->max_queue_depth)
668 continue;
669
670 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
671 sdev->last_queue_ramp_up = jiffies;
672 }
673}
674
675static void scsi_handle_queue_full(struct scsi_device *sdev)
676{
677 struct scsi_host_template *sht = sdev->host->hostt;
678 struct scsi_device *tmp_sdev;
679
680 if (!sht->track_queue_depth)
681 return;
682
683 shost_for_each_device(tmp_sdev, sdev->host) {
684 if (tmp_sdev->channel != sdev->channel ||
685 tmp_sdev->id != sdev->id)
686 continue;
687 /*
688 * We do not know the number of commands that were at
689 * the device when we got the queue full so we start
690 * from the highest possible value and work our way down.
691 */
692 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
693 }
694}
695
696/**
697 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
698 * @scmd: SCSI cmd to examine.
699 *
700 * Notes:
701 * This is *only* called when we are examining the status of commands
702 * queued during error recovery. the main difference here is that we
703 * don't allow for the possibility of retries here, and we are a lot
704 * more restrictive about what we consider acceptable.
705 */
706static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
707{
708 /*
709 * first check the host byte, to see if there is anything in there
710 * that would indicate what we need to do.
711 */
712 if (host_byte(scmd->result) == DID_RESET) {
713 /*
714 * rats. we are already in the error handler, so we now
715 * get to try and figure out what to do next. if the sense
716 * is valid, we have a pretty good idea of what to do.
717 * if not, we mark it as FAILED.
718 */
719 return scsi_check_sense(scmd);
720 }
721 if (host_byte(scmd->result) != DID_OK)
722 return FAILED;
723
724 /*
725 * next, check the message byte.
726 */
727 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
728 return FAILED;
729
730 /*
731 * now, check the status byte to see if this indicates
732 * anything special.
733 */
734 switch (status_byte(scmd->result)) {
735 case GOOD:
736 scsi_handle_queue_ramp_up(scmd->device);
737 /* FALLTHROUGH */
738 case COMMAND_TERMINATED:
739 return SUCCESS;
740 case CHECK_CONDITION:
741 return scsi_check_sense(scmd);
742 case CONDITION_GOOD:
743 case INTERMEDIATE_GOOD:
744 case INTERMEDIATE_C_GOOD:
745 /*
746 * who knows? FIXME(eric)
747 */
748 return SUCCESS;
749 case RESERVATION_CONFLICT:
750 if (scmd->cmnd[0] == TEST_UNIT_READY)
751 /* it is a success, we probed the device and
752 * found it */
753 return SUCCESS;
754 /* otherwise, we failed to send the command */
755 return FAILED;
756 case QUEUE_FULL:
757 scsi_handle_queue_full(scmd->device);
758 /* fall through */
759 case BUSY:
760 return NEEDS_RETRY;
761 default:
762 return FAILED;
763 }
764 return FAILED;
765}
766
767/**
768 * scsi_eh_done - Completion function for error handling.
769 * @scmd: Cmd that is done.
770 */
771static void scsi_eh_done(struct scsi_cmnd *scmd)
772{
773 struct completion *eh_action;
774
775 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
776 "%s result: %x\n", __func__, scmd->result));
777
778 eh_action = scmd->device->host->eh_action;
779 if (eh_action)
780 complete(eh_action);
781}
782
783/**
784 * scsi_try_host_reset - ask host adapter to reset itself
785 * @scmd: SCSI cmd to send host reset.
786 */
787static int scsi_try_host_reset(struct scsi_cmnd *scmd)
788{
789 unsigned long flags;
790 int rtn;
791 struct Scsi_Host *host = scmd->device->host;
792 struct scsi_host_template *hostt = host->hostt;
793
794 SCSI_LOG_ERROR_RECOVERY(3,
795 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
796
797 if (!hostt->eh_host_reset_handler)
798 return FAILED;
799
800 rtn = hostt->eh_host_reset_handler(scmd);
801
802 if (rtn == SUCCESS) {
803 if (!hostt->skip_settle_delay)
804 ssleep(HOST_RESET_SETTLE_TIME);
805 spin_lock_irqsave(host->host_lock, flags);
806 scsi_report_bus_reset(host, scmd_channel(scmd));
807 spin_unlock_irqrestore(host->host_lock, flags);
808 }
809
810 return rtn;
811}
812
813/**
814 * scsi_try_bus_reset - ask host to perform a bus reset
815 * @scmd: SCSI cmd to send bus reset.
816 */
817static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
818{
819 unsigned long flags;
820 int rtn;
821 struct Scsi_Host *host = scmd->device->host;
822 struct scsi_host_template *hostt = host->hostt;
823
824 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
825 "%s: Snd Bus RST\n", __func__));
826
827 if (!hostt->eh_bus_reset_handler)
828 return FAILED;
829
830 rtn = hostt->eh_bus_reset_handler(scmd);
831
832 if (rtn == SUCCESS) {
833 if (!hostt->skip_settle_delay)
834 ssleep(BUS_RESET_SETTLE_TIME);
835 spin_lock_irqsave(host->host_lock, flags);
836 scsi_report_bus_reset(host, scmd_channel(scmd));
837 spin_unlock_irqrestore(host->host_lock, flags);
838 }
839
840 return rtn;
841}
842
843static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
844{
845 sdev->was_reset = 1;
846 sdev->expecting_cc_ua = 1;
847}
848
849/**
850 * scsi_try_target_reset - Ask host to perform a target reset
851 * @scmd: SCSI cmd used to send a target reset
852 *
853 * Notes:
854 * There is no timeout for this operation. if this operation is
855 * unreliable for a given host, then the host itself needs to put a
856 * timer on it, and set the host back to a consistent state prior to
857 * returning.
858 */
859static int scsi_try_target_reset(struct scsi_cmnd *scmd)
860{
861 unsigned long flags;
862 int rtn;
863 struct Scsi_Host *host = scmd->device->host;
864 struct scsi_host_template *hostt = host->hostt;
865
866 if (!hostt->eh_target_reset_handler)
867 return FAILED;
868
869 rtn = hostt->eh_target_reset_handler(scmd);
870 if (rtn == SUCCESS) {
871 spin_lock_irqsave(host->host_lock, flags);
872 __starget_for_each_device(scsi_target(scmd->device), NULL,
873 __scsi_report_device_reset);
874 spin_unlock_irqrestore(host->host_lock, flags);
875 }
876
877 return rtn;
878}
879
880/**
881 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
882 * @scmd: SCSI cmd used to send BDR
883 *
884 * Notes:
885 * There is no timeout for this operation. if this operation is
886 * unreliable for a given host, then the host itself needs to put a
887 * timer on it, and set the host back to a consistent state prior to
888 * returning.
889 */
890static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
891{
892 int rtn;
893 struct scsi_host_template *hostt = scmd->device->host->hostt;
894
895 if (!hostt->eh_device_reset_handler)
896 return FAILED;
897
898 rtn = hostt->eh_device_reset_handler(scmd);
899 if (rtn == SUCCESS)
900 __scsi_report_device_reset(scmd->device, NULL);
901 return rtn;
902}
903
904/**
905 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
906 * @hostt: SCSI driver host template
907 * @scmd: SCSI cmd used to send a target reset
908 *
909 * Return value:
910 * SUCCESS, FAILED, or FAST_IO_FAIL
911 *
912 * Notes:
913 * SUCCESS does not necessarily indicate that the command
914 * has been aborted; it only indicates that the LLDDs
915 * has cleared all references to that command.
916 * LLDDs should return FAILED only if an abort was required
917 * but could not be executed. LLDDs should return FAST_IO_FAIL
918 * if the device is temporarily unavailable (eg due to a
919 * link down on FibreChannel)
920 */
921static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
922 struct scsi_cmnd *scmd)
923{
924 if (!hostt->eh_abort_handler)
925 return FAILED;
926
927 return hostt->eh_abort_handler(scmd);
928}
929
930static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
931{
932 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
933 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
934 if (scsi_try_target_reset(scmd) != SUCCESS)
935 if (scsi_try_bus_reset(scmd) != SUCCESS)
936 scsi_try_host_reset(scmd);
937}
938
939/**
940 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
941 * @scmd: SCSI command structure to hijack
942 * @ses: structure to save restore information
943 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
944 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
945 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
946 *
947 * This function is used to save a scsi command information before re-execution
948 * as part of the error recovery process. If @sense_bytes is 0 the command
949 * sent must be one that does not transfer any data. If @sense_bytes != 0
950 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
951 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
952 */
953void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
954 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
955{
956 struct scsi_device *sdev = scmd->device;
957
958 /*
959 * We need saved copies of a number of fields - this is because
960 * error handling may need to overwrite these with different values
961 * to run different commands, and once error handling is complete,
962 * we will need to restore these values prior to running the actual
963 * command.
964 */
965 ses->cmd_len = scmd->cmd_len;
966 ses->cmnd = scmd->cmnd;
967 ses->data_direction = scmd->sc_data_direction;
968 ses->sdb = scmd->sdb;
969 ses->result = scmd->result;
970 ses->resid_len = scmd->req.resid_len;
971 ses->underflow = scmd->underflow;
972 ses->prot_op = scmd->prot_op;
973 ses->eh_eflags = scmd->eh_eflags;
974
975 scmd->prot_op = SCSI_PROT_NORMAL;
976 scmd->eh_eflags = 0;
977 scmd->cmnd = ses->eh_cmnd;
978 memset(scmd->cmnd, 0, BLK_MAX_CDB);
979 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
980 scmd->result = 0;
981 scmd->req.resid_len = 0;
982
983 if (sense_bytes) {
984 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
985 sense_bytes);
986 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
987 scmd->sdb.length);
988 scmd->sdb.table.sgl = &ses->sense_sgl;
989 scmd->sc_data_direction = DMA_FROM_DEVICE;
990 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
991 scmd->cmnd[0] = REQUEST_SENSE;
992 scmd->cmnd[4] = scmd->sdb.length;
993 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
994 } else {
995 scmd->sc_data_direction = DMA_NONE;
996 if (cmnd) {
997 BUG_ON(cmnd_size > BLK_MAX_CDB);
998 memcpy(scmd->cmnd, cmnd, cmnd_size);
999 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1000 }
1001 }
1002
1003 scmd->underflow = 0;
1004
1005 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1006 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
1007 (sdev->lun << 5 & 0xe0);
1008
1009 /*
1010 * Zero the sense buffer. The scsi spec mandates that any
1011 * untransferred sense data should be interpreted as being zero.
1012 */
1013 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1014}
1015EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1016
1017/**
1018 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1019 * @scmd: SCSI command structure to restore
1020 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1021 *
1022 * Undo any damage done by above scsi_eh_prep_cmnd().
1023 */
1024void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1025{
1026 /*
1027 * Restore original data
1028 */
1029 scmd->cmd_len = ses->cmd_len;
1030 scmd->cmnd = ses->cmnd;
1031 scmd->sc_data_direction = ses->data_direction;
1032 scmd->sdb = ses->sdb;
1033 scmd->result = ses->result;
1034 scmd->req.resid_len = ses->resid_len;
1035 scmd->underflow = ses->underflow;
1036 scmd->prot_op = ses->prot_op;
1037 scmd->eh_eflags = ses->eh_eflags;
1038}
1039EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1040
1041/**
1042 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1043 * @scmd: SCSI command structure to hijack
1044 * @cmnd: CDB to send
1045 * @cmnd_size: size in bytes of @cmnd
1046 * @timeout: timeout for this request
1047 * @sense_bytes: size of sense data to copy or 0
1048 *
1049 * This function is used to send a scsi command down to a target device
1050 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1051 *
1052 * Return value:
1053 * SUCCESS or FAILED or NEEDS_RETRY
1054 */
1055static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1056 int cmnd_size, int timeout, unsigned sense_bytes)
1057{
1058 struct scsi_device *sdev = scmd->device;
1059 struct Scsi_Host *shost = sdev->host;
1060 DECLARE_COMPLETION_ONSTACK(done);
1061 unsigned long timeleft = timeout, delay;
1062 struct scsi_eh_save ses;
1063 const unsigned long stall_for = msecs_to_jiffies(100);
1064 int rtn;
1065
1066retry:
1067 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1068 shost->eh_action = &done;
1069
1070 scsi_log_send(scmd);
1071 scmd->scsi_done = scsi_eh_done;
1072
1073 /*
1074 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1075 * change the SCSI device state after we have examined it and before
1076 * .queuecommand() is called.
1077 */
1078 mutex_lock(&sdev->state_mutex);
1079 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1080 mutex_unlock(&sdev->state_mutex);
1081 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1082 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1083 SDEV_BLOCK));
1084 delay = min(timeleft, stall_for);
1085 timeleft -= delay;
1086 msleep(jiffies_to_msecs(delay));
1087 mutex_lock(&sdev->state_mutex);
1088 }
1089 if (sdev->sdev_state != SDEV_BLOCK)
1090 rtn = shost->hostt->queuecommand(shost, scmd);
1091 else
1092 rtn = SCSI_MLQUEUE_DEVICE_BUSY;
1093 mutex_unlock(&sdev->state_mutex);
1094
1095 if (rtn) {
1096 if (timeleft > stall_for) {
1097 scsi_eh_restore_cmnd(scmd, &ses);
1098 timeleft -= stall_for;
1099 msleep(jiffies_to_msecs(stall_for));
1100 goto retry;
1101 }
1102 /* signal not to enter either branch of the if () below */
1103 timeleft = 0;
1104 rtn = FAILED;
1105 } else {
1106 timeleft = wait_for_completion_timeout(&done, timeout);
1107 rtn = SUCCESS;
1108 }
1109
1110 shost->eh_action = NULL;
1111
1112 scsi_log_completion(scmd, rtn);
1113
1114 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1115 "%s timeleft: %ld\n",
1116 __func__, timeleft));
1117
1118 /*
1119 * If there is time left scsi_eh_done got called, and we will examine
1120 * the actual status codes to see whether the command actually did
1121 * complete normally, else if we have a zero return and no time left,
1122 * the command must still be pending, so abort it and return FAILED.
1123 * If we never actually managed to issue the command, because
1124 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1125 * value above (so don't execute either branch of the if)
1126 */
1127 if (timeleft) {
1128 rtn = scsi_eh_completed_normally(scmd);
1129 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1130 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1131
1132 switch (rtn) {
1133 case SUCCESS:
1134 case NEEDS_RETRY:
1135 case FAILED:
1136 break;
1137 case ADD_TO_MLQUEUE:
1138 rtn = NEEDS_RETRY;
1139 break;
1140 default:
1141 rtn = FAILED;
1142 break;
1143 }
1144 } else if (rtn != FAILED) {
1145 scsi_abort_eh_cmnd(scmd);
1146 rtn = FAILED;
1147 }
1148
1149 scsi_eh_restore_cmnd(scmd, &ses);
1150
1151 return rtn;
1152}
1153
1154/**
1155 * scsi_request_sense - Request sense data from a particular target.
1156 * @scmd: SCSI cmd for request sense.
1157 *
1158 * Notes:
1159 * Some hosts automatically obtain this information, others require
1160 * that we obtain it on our own. This function will *not* return until
1161 * the command either times out, or it completes.
1162 */
1163static int scsi_request_sense(struct scsi_cmnd *scmd)
1164{
1165 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1166}
1167
1168static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1169{
1170 if (!blk_rq_is_passthrough(scmd->request)) {
1171 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1172 if (sdrv->eh_action)
1173 rtn = sdrv->eh_action(scmd, rtn);
1174 }
1175 return rtn;
1176}
1177
1178/**
1179 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1180 * @scmd: Original SCSI cmd that eh has finished.
1181 * @done_q: Queue for processed commands.
1182 *
1183 * Notes:
1184 * We don't want to use the normal command completion while we are are
1185 * still handling errors - it may cause other commands to be queued,
1186 * and that would disturb what we are doing. Thus we really want to
1187 * keep a list of pending commands for final completion, and once we
1188 * are ready to leave error handling we handle completion for real.
1189 */
1190void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1191{
1192 list_move_tail(&scmd->eh_entry, done_q);
1193}
1194EXPORT_SYMBOL(scsi_eh_finish_cmd);
1195
1196/**
1197 * scsi_eh_get_sense - Get device sense data.
1198 * @work_q: Queue of commands to process.
1199 * @done_q: Queue of processed commands.
1200 *
1201 * Description:
1202 * See if we need to request sense information. if so, then get it
1203 * now, so we have a better idea of what to do.
1204 *
1205 * Notes:
1206 * This has the unfortunate side effect that if a shost adapter does
1207 * not automatically request sense information, we end up shutting
1208 * it down before we request it.
1209 *
1210 * All drivers should request sense information internally these days,
1211 * so for now all I have to say is tough noogies if you end up in here.
1212 *
1213 * XXX: Long term this code should go away, but that needs an audit of
1214 * all LLDDs first.
1215 */
1216int scsi_eh_get_sense(struct list_head *work_q,
1217 struct list_head *done_q)
1218{
1219 struct scsi_cmnd *scmd, *next;
1220 struct Scsi_Host *shost;
1221 int rtn;
1222
1223 /*
1224 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1225 * should not get sense.
1226 */
1227 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1228 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1229 SCSI_SENSE_VALID(scmd))
1230 continue;
1231
1232 shost = scmd->device->host;
1233 if (scsi_host_eh_past_deadline(shost)) {
1234 SCSI_LOG_ERROR_RECOVERY(3,
1235 scmd_printk(KERN_INFO, scmd,
1236 "%s: skip request sense, past eh deadline\n",
1237 current->comm));
1238 break;
1239 }
1240 if (status_byte(scmd->result) != CHECK_CONDITION)
1241 /*
1242 * don't request sense if there's no check condition
1243 * status because the error we're processing isn't one
1244 * that has a sense code (and some devices get
1245 * confused by sense requests out of the blue)
1246 */
1247 continue;
1248
1249 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1250 "%s: requesting sense\n",
1251 current->comm));
1252 rtn = scsi_request_sense(scmd);
1253 if (rtn != SUCCESS)
1254 continue;
1255
1256 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1257 "sense requested, result %x\n", scmd->result));
1258 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1259
1260 rtn = scsi_decide_disposition(scmd);
1261
1262 /*
1263 * if the result was normal, then just pass it along to the
1264 * upper level.
1265 */
1266 if (rtn == SUCCESS)
1267 /* we don't want this command reissued, just
1268 * finished with the sense data, so set
1269 * retries to the max allowed to ensure it
1270 * won't get reissued */
1271 scmd->retries = scmd->allowed;
1272 else if (rtn != NEEDS_RETRY)
1273 continue;
1274
1275 scsi_eh_finish_cmd(scmd, done_q);
1276 }
1277
1278 return list_empty(work_q);
1279}
1280EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1281
1282/**
1283 * scsi_eh_tur - Send TUR to device.
1284 * @scmd: &scsi_cmnd to send TUR
1285 *
1286 * Return value:
1287 * 0 - Device is ready. 1 - Device NOT ready.
1288 */
1289static int scsi_eh_tur(struct scsi_cmnd *scmd)
1290{
1291 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1292 int retry_cnt = 1, rtn;
1293
1294retry_tur:
1295 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1296 scmd->device->eh_timeout, 0);
1297
1298 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1299 "%s return: %x\n", __func__, rtn));
1300
1301 switch (rtn) {
1302 case NEEDS_RETRY:
1303 if (retry_cnt--)
1304 goto retry_tur;
1305 /*FALLTHRU*/
1306 case SUCCESS:
1307 return 0;
1308 default:
1309 return 1;
1310 }
1311}
1312
1313/**
1314 * scsi_eh_test_devices - check if devices are responding from error recovery.
1315 * @cmd_list: scsi commands in error recovery.
1316 * @work_q: queue for commands which still need more error recovery
1317 * @done_q: queue for commands which are finished
1318 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1319 *
1320 * Decription:
1321 * Tests if devices are in a working state. Commands to devices now in
1322 * a working state are sent to the done_q while commands to devices which
1323 * are still failing to respond are returned to the work_q for more
1324 * processing.
1325 **/
1326static int scsi_eh_test_devices(struct list_head *cmd_list,
1327 struct list_head *work_q,
1328 struct list_head *done_q, int try_stu)
1329{
1330 struct scsi_cmnd *scmd, *next;
1331 struct scsi_device *sdev;
1332 int finish_cmds;
1333
1334 while (!list_empty(cmd_list)) {
1335 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1336 sdev = scmd->device;
1337
1338 if (!try_stu) {
1339 if (scsi_host_eh_past_deadline(sdev->host)) {
1340 /* Push items back onto work_q */
1341 list_splice_init(cmd_list, work_q);
1342 SCSI_LOG_ERROR_RECOVERY(3,
1343 sdev_printk(KERN_INFO, sdev,
1344 "%s: skip test device, past eh deadline",
1345 current->comm));
1346 break;
1347 }
1348 }
1349
1350 finish_cmds = !scsi_device_online(scmd->device) ||
1351 (try_stu && !scsi_eh_try_stu(scmd) &&
1352 !scsi_eh_tur(scmd)) ||
1353 !scsi_eh_tur(scmd);
1354
1355 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1356 if (scmd->device == sdev) {
1357 if (finish_cmds &&
1358 (try_stu ||
1359 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1360 scsi_eh_finish_cmd(scmd, done_q);
1361 else
1362 list_move_tail(&scmd->eh_entry, work_q);
1363 }
1364 }
1365 return list_empty(work_q);
1366}
1367
1368/**
1369 * scsi_eh_try_stu - Send START_UNIT to device.
1370 * @scmd: &scsi_cmnd to send START_UNIT
1371 *
1372 * Return value:
1373 * 0 - Device is ready. 1 - Device NOT ready.
1374 */
1375static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1376{
1377 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1378
1379 if (scmd->device->allow_restart) {
1380 int i, rtn = NEEDS_RETRY;
1381
1382 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1383 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1384
1385 if (rtn == SUCCESS)
1386 return 0;
1387 }
1388
1389 return 1;
1390}
1391
1392 /**
1393 * scsi_eh_stu - send START_UNIT if needed
1394 * @shost: &scsi host being recovered.
1395 * @work_q: &list_head for pending commands.
1396 * @done_q: &list_head for processed commands.
1397 *
1398 * Notes:
1399 * If commands are failing due to not ready, initializing command required,
1400 * try revalidating the device, which will end up sending a start unit.
1401 */
1402static int scsi_eh_stu(struct Scsi_Host *shost,
1403 struct list_head *work_q,
1404 struct list_head *done_q)
1405{
1406 struct scsi_cmnd *scmd, *stu_scmd, *next;
1407 struct scsi_device *sdev;
1408
1409 shost_for_each_device(sdev, shost) {
1410 if (scsi_host_eh_past_deadline(shost)) {
1411 SCSI_LOG_ERROR_RECOVERY(3,
1412 sdev_printk(KERN_INFO, sdev,
1413 "%s: skip START_UNIT, past eh deadline\n",
1414 current->comm));
1415 break;
1416 }
1417 stu_scmd = NULL;
1418 list_for_each_entry(scmd, work_q, eh_entry)
1419 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1420 scsi_check_sense(scmd) == FAILED ) {
1421 stu_scmd = scmd;
1422 break;
1423 }
1424
1425 if (!stu_scmd)
1426 continue;
1427
1428 SCSI_LOG_ERROR_RECOVERY(3,
1429 sdev_printk(KERN_INFO, sdev,
1430 "%s: Sending START_UNIT\n",
1431 current->comm));
1432
1433 if (!scsi_eh_try_stu(stu_scmd)) {
1434 if (!scsi_device_online(sdev) ||
1435 !scsi_eh_tur(stu_scmd)) {
1436 list_for_each_entry_safe(scmd, next,
1437 work_q, eh_entry) {
1438 if (scmd->device == sdev &&
1439 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1440 scsi_eh_finish_cmd(scmd, done_q);
1441 }
1442 }
1443 } else {
1444 SCSI_LOG_ERROR_RECOVERY(3,
1445 sdev_printk(KERN_INFO, sdev,
1446 "%s: START_UNIT failed\n",
1447 current->comm));
1448 }
1449 }
1450
1451 return list_empty(work_q);
1452}
1453
1454
1455/**
1456 * scsi_eh_bus_device_reset - send bdr if needed
1457 * @shost: scsi host being recovered.
1458 * @work_q: &list_head for pending commands.
1459 * @done_q: &list_head for processed commands.
1460 *
1461 * Notes:
1462 * Try a bus device reset. Still, look to see whether we have multiple
1463 * devices that are jammed or not - if we have multiple devices, it
1464 * makes no sense to try bus_device_reset - we really would need to try
1465 * a bus_reset instead.
1466 */
1467static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1468 struct list_head *work_q,
1469 struct list_head *done_q)
1470{
1471 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1472 struct scsi_device *sdev;
1473 int rtn;
1474
1475 shost_for_each_device(sdev, shost) {
1476 if (scsi_host_eh_past_deadline(shost)) {
1477 SCSI_LOG_ERROR_RECOVERY(3,
1478 sdev_printk(KERN_INFO, sdev,
1479 "%s: skip BDR, past eh deadline\n",
1480 current->comm));
1481 break;
1482 }
1483 bdr_scmd = NULL;
1484 list_for_each_entry(scmd, work_q, eh_entry)
1485 if (scmd->device == sdev) {
1486 bdr_scmd = scmd;
1487 break;
1488 }
1489
1490 if (!bdr_scmd)
1491 continue;
1492
1493 SCSI_LOG_ERROR_RECOVERY(3,
1494 sdev_printk(KERN_INFO, sdev,
1495 "%s: Sending BDR\n", current->comm));
1496 rtn = scsi_try_bus_device_reset(bdr_scmd);
1497 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1498 if (!scsi_device_online(sdev) ||
1499 rtn == FAST_IO_FAIL ||
1500 !scsi_eh_tur(bdr_scmd)) {
1501 list_for_each_entry_safe(scmd, next,
1502 work_q, eh_entry) {
1503 if (scmd->device == sdev &&
1504 scsi_eh_action(scmd, rtn) != FAILED)
1505 scsi_eh_finish_cmd(scmd,
1506 done_q);
1507 }
1508 }
1509 } else {
1510 SCSI_LOG_ERROR_RECOVERY(3,
1511 sdev_printk(KERN_INFO, sdev,
1512 "%s: BDR failed\n", current->comm));
1513 }
1514 }
1515
1516 return list_empty(work_q);
1517}
1518
1519/**
1520 * scsi_eh_target_reset - send target reset if needed
1521 * @shost: scsi host being recovered.
1522 * @work_q: &list_head for pending commands.
1523 * @done_q: &list_head for processed commands.
1524 *
1525 * Notes:
1526 * Try a target reset.
1527 */
1528static int scsi_eh_target_reset(struct Scsi_Host *shost,
1529 struct list_head *work_q,
1530 struct list_head *done_q)
1531{
1532 LIST_HEAD(tmp_list);
1533 LIST_HEAD(check_list);
1534
1535 list_splice_init(work_q, &tmp_list);
1536
1537 while (!list_empty(&tmp_list)) {
1538 struct scsi_cmnd *next, *scmd;
1539 int rtn;
1540 unsigned int id;
1541
1542 if (scsi_host_eh_past_deadline(shost)) {
1543 /* push back on work queue for further processing */
1544 list_splice_init(&check_list, work_q);
1545 list_splice_init(&tmp_list, work_q);
1546 SCSI_LOG_ERROR_RECOVERY(3,
1547 shost_printk(KERN_INFO, shost,
1548 "%s: Skip target reset, past eh deadline\n",
1549 current->comm));
1550 return list_empty(work_q);
1551 }
1552
1553 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1554 id = scmd_id(scmd);
1555
1556 SCSI_LOG_ERROR_RECOVERY(3,
1557 shost_printk(KERN_INFO, shost,
1558 "%s: Sending target reset to target %d\n",
1559 current->comm, id));
1560 rtn = scsi_try_target_reset(scmd);
1561 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1562 SCSI_LOG_ERROR_RECOVERY(3,
1563 shost_printk(KERN_INFO, shost,
1564 "%s: Target reset failed"
1565 " target: %d\n",
1566 current->comm, id));
1567 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1568 if (scmd_id(scmd) != id)
1569 continue;
1570
1571 if (rtn == SUCCESS)
1572 list_move_tail(&scmd->eh_entry, &check_list);
1573 else if (rtn == FAST_IO_FAIL)
1574 scsi_eh_finish_cmd(scmd, done_q);
1575 else
1576 /* push back on work queue for further processing */
1577 list_move(&scmd->eh_entry, work_q);
1578 }
1579 }
1580
1581 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1582}
1583
1584/**
1585 * scsi_eh_bus_reset - send a bus reset
1586 * @shost: &scsi host being recovered.
1587 * @work_q: &list_head for pending commands.
1588 * @done_q: &list_head for processed commands.
1589 */
1590static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1591 struct list_head *work_q,
1592 struct list_head *done_q)
1593{
1594 struct scsi_cmnd *scmd, *chan_scmd, *next;
1595 LIST_HEAD(check_list);
1596 unsigned int channel;
1597 int rtn;
1598
1599 /*
1600 * we really want to loop over the various channels, and do this on
1601 * a channel by channel basis. we should also check to see if any
1602 * of the failed commands are on soft_reset devices, and if so, skip
1603 * the reset.
1604 */
1605
1606 for (channel = 0; channel <= shost->max_channel; channel++) {
1607 if (scsi_host_eh_past_deadline(shost)) {
1608 list_splice_init(&check_list, work_q);
1609 SCSI_LOG_ERROR_RECOVERY(3,
1610 shost_printk(KERN_INFO, shost,
1611 "%s: skip BRST, past eh deadline\n",
1612 current->comm));
1613 return list_empty(work_q);
1614 }
1615
1616 chan_scmd = NULL;
1617 list_for_each_entry(scmd, work_q, eh_entry) {
1618 if (channel == scmd_channel(scmd)) {
1619 chan_scmd = scmd;
1620 break;
1621 /*
1622 * FIXME add back in some support for
1623 * soft_reset devices.
1624 */
1625 }
1626 }
1627
1628 if (!chan_scmd)
1629 continue;
1630 SCSI_LOG_ERROR_RECOVERY(3,
1631 shost_printk(KERN_INFO, shost,
1632 "%s: Sending BRST chan: %d\n",
1633 current->comm, channel));
1634 rtn = scsi_try_bus_reset(chan_scmd);
1635 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1636 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1637 if (channel == scmd_channel(scmd)) {
1638 if (rtn == FAST_IO_FAIL)
1639 scsi_eh_finish_cmd(scmd,
1640 done_q);
1641 else
1642 list_move_tail(&scmd->eh_entry,
1643 &check_list);
1644 }
1645 }
1646 } else {
1647 SCSI_LOG_ERROR_RECOVERY(3,
1648 shost_printk(KERN_INFO, shost,
1649 "%s: BRST failed chan: %d\n",
1650 current->comm, channel));
1651 }
1652 }
1653 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1654}
1655
1656/**
1657 * scsi_eh_host_reset - send a host reset
1658 * @shost: host to be reset.
1659 * @work_q: &list_head for pending commands.
1660 * @done_q: &list_head for processed commands.
1661 */
1662static int scsi_eh_host_reset(struct Scsi_Host *shost,
1663 struct list_head *work_q,
1664 struct list_head *done_q)
1665{
1666 struct scsi_cmnd *scmd, *next;
1667 LIST_HEAD(check_list);
1668 int rtn;
1669
1670 if (!list_empty(work_q)) {
1671 scmd = list_entry(work_q->next,
1672 struct scsi_cmnd, eh_entry);
1673
1674 SCSI_LOG_ERROR_RECOVERY(3,
1675 shost_printk(KERN_INFO, shost,
1676 "%s: Sending HRST\n",
1677 current->comm));
1678
1679 rtn = scsi_try_host_reset(scmd);
1680 if (rtn == SUCCESS) {
1681 list_splice_init(work_q, &check_list);
1682 } else if (rtn == FAST_IO_FAIL) {
1683 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1684 scsi_eh_finish_cmd(scmd, done_q);
1685 }
1686 } else {
1687 SCSI_LOG_ERROR_RECOVERY(3,
1688 shost_printk(KERN_INFO, shost,
1689 "%s: HRST failed\n",
1690 current->comm));
1691 }
1692 }
1693 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1694}
1695
1696/**
1697 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1698 * @work_q: &list_head for pending commands.
1699 * @done_q: &list_head for processed commands.
1700 */
1701static void scsi_eh_offline_sdevs(struct list_head *work_q,
1702 struct list_head *done_q)
1703{
1704 struct scsi_cmnd *scmd, *next;
1705 struct scsi_device *sdev;
1706
1707 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1708 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1709 "not ready after error recovery\n");
1710 sdev = scmd->device;
1711
1712 mutex_lock(&sdev->state_mutex);
1713 scsi_device_set_state(sdev, SDEV_OFFLINE);
1714 mutex_unlock(&sdev->state_mutex);
1715
1716 scsi_eh_finish_cmd(scmd, done_q);
1717 }
1718 return;
1719}
1720
1721/**
1722 * scsi_noretry_cmd - determine if command should be failed fast
1723 * @scmd: SCSI cmd to examine.
1724 */
1725int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1726{
1727 switch (host_byte(scmd->result)) {
1728 case DID_OK:
1729 break;
1730 case DID_TIME_OUT:
1731 goto check_type;
1732 case DID_BUS_BUSY:
1733 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1734 case DID_PARITY:
1735 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1736 case DID_ERROR:
1737 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1738 status_byte(scmd->result) == RESERVATION_CONFLICT)
1739 return 0;
1740 /* fall through */
1741 case DID_SOFT_ERROR:
1742 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1743 }
1744
1745 if (status_byte(scmd->result) != CHECK_CONDITION)
1746 return 0;
1747
1748check_type:
1749 /*
1750 * assume caller has checked sense and determined
1751 * the check condition was retryable.
1752 */
1753 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1754 blk_rq_is_passthrough(scmd->request))
1755 return 1;
1756 else
1757 return 0;
1758}
1759
1760/**
1761 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1762 * @scmd: SCSI cmd to examine.
1763 *
1764 * Notes:
1765 * This is *only* called when we are examining the status after sending
1766 * out the actual data command. any commands that are queued for error
1767 * recovery (e.g. test_unit_ready) do *not* come through here.
1768 *
1769 * When this routine returns failed, it means the error handler thread
1770 * is woken. In cases where the error code indicates an error that
1771 * doesn't require the error handler read (i.e. we don't need to
1772 * abort/reset), this function should return SUCCESS.
1773 */
1774int scsi_decide_disposition(struct scsi_cmnd *scmd)
1775{
1776 int rtn;
1777
1778 /*
1779 * if the device is offline, then we clearly just pass the result back
1780 * up to the top level.
1781 */
1782 if (!scsi_device_online(scmd->device)) {
1783 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1784 "%s: device offline - report as SUCCESS\n", __func__));
1785 return SUCCESS;
1786 }
1787
1788 /*
1789 * first check the host byte, to see if there is anything in there
1790 * that would indicate what we need to do.
1791 */
1792 switch (host_byte(scmd->result)) {
1793 case DID_PASSTHROUGH:
1794 /*
1795 * no matter what, pass this through to the upper layer.
1796 * nuke this special code so that it looks like we are saying
1797 * did_ok.
1798 */
1799 scmd->result &= 0xff00ffff;
1800 return SUCCESS;
1801 case DID_OK:
1802 /*
1803 * looks good. drop through, and check the next byte.
1804 */
1805 break;
1806 case DID_ABORT:
1807 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1808 set_host_byte(scmd, DID_TIME_OUT);
1809 return SUCCESS;
1810 }
1811 /* FALLTHROUGH */
1812 case DID_NO_CONNECT:
1813 case DID_BAD_TARGET:
1814 /*
1815 * note - this means that we just report the status back
1816 * to the top level driver, not that we actually think
1817 * that it indicates SUCCESS.
1818 */
1819 return SUCCESS;
1820 case DID_SOFT_ERROR:
1821 /*
1822 * when the low level driver returns did_soft_error,
1823 * it is responsible for keeping an internal retry counter
1824 * in order to avoid endless loops (db)
1825 */
1826 goto maybe_retry;
1827 case DID_IMM_RETRY:
1828 return NEEDS_RETRY;
1829
1830 case DID_REQUEUE:
1831 return ADD_TO_MLQUEUE;
1832 case DID_TRANSPORT_DISRUPTED:
1833 /*
1834 * LLD/transport was disrupted during processing of the IO.
1835 * The transport class is now blocked/blocking,
1836 * and the transport will decide what to do with the IO
1837 * based on its timers and recovery capablilities if
1838 * there are enough retries.
1839 */
1840 goto maybe_retry;
1841 case DID_TRANSPORT_FAILFAST:
1842 /*
1843 * The transport decided to failfast the IO (most likely
1844 * the fast io fail tmo fired), so send IO directly upwards.
1845 */
1846 return SUCCESS;
1847 case DID_ERROR:
1848 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1849 status_byte(scmd->result) == RESERVATION_CONFLICT)
1850 /*
1851 * execute reservation conflict processing code
1852 * lower down
1853 */
1854 break;
1855 /* fallthrough */
1856 case DID_BUS_BUSY:
1857 case DID_PARITY:
1858 goto maybe_retry;
1859 case DID_TIME_OUT:
1860 /*
1861 * when we scan the bus, we get timeout messages for
1862 * these commands if there is no device available.
1863 * other hosts report did_no_connect for the same thing.
1864 */
1865 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1866 scmd->cmnd[0] == INQUIRY)) {
1867 return SUCCESS;
1868 } else {
1869 return FAILED;
1870 }
1871 case DID_RESET:
1872 return SUCCESS;
1873 default:
1874 return FAILED;
1875 }
1876
1877 /*
1878 * next, check the message byte.
1879 */
1880 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1881 return FAILED;
1882
1883 /*
1884 * check the status byte to see if this indicates anything special.
1885 */
1886 switch (status_byte(scmd->result)) {
1887 case QUEUE_FULL:
1888 scsi_handle_queue_full(scmd->device);
1889 /*
1890 * the case of trying to send too many commands to a
1891 * tagged queueing device.
1892 */
1893 /* FALLTHROUGH */
1894 case BUSY:
1895 /*
1896 * device can't talk to us at the moment. Should only
1897 * occur (SAM-3) when the task queue is empty, so will cause
1898 * the empty queue handling to trigger a stall in the
1899 * device.
1900 */
1901 return ADD_TO_MLQUEUE;
1902 case GOOD:
1903 if (scmd->cmnd[0] == REPORT_LUNS)
1904 scmd->device->sdev_target->expecting_lun_change = 0;
1905 scsi_handle_queue_ramp_up(scmd->device);
1906 /* FALLTHROUGH */
1907 case COMMAND_TERMINATED:
1908 return SUCCESS;
1909 case TASK_ABORTED:
1910 goto maybe_retry;
1911 case CHECK_CONDITION:
1912 rtn = scsi_check_sense(scmd);
1913 if (rtn == NEEDS_RETRY)
1914 goto maybe_retry;
1915 /* if rtn == FAILED, we have no sense information;
1916 * returning FAILED will wake the error handler thread
1917 * to collect the sense and redo the decide
1918 * disposition */
1919 return rtn;
1920 case CONDITION_GOOD:
1921 case INTERMEDIATE_GOOD:
1922 case INTERMEDIATE_C_GOOD:
1923 case ACA_ACTIVE:
1924 /*
1925 * who knows? FIXME(eric)
1926 */
1927 return SUCCESS;
1928
1929 case RESERVATION_CONFLICT:
1930 sdev_printk(KERN_INFO, scmd->device,
1931 "reservation conflict\n");
1932 set_host_byte(scmd, DID_NEXUS_FAILURE);
1933 return SUCCESS; /* causes immediate i/o error */
1934 default:
1935 return FAILED;
1936 }
1937 return FAILED;
1938
1939maybe_retry:
1940
1941 /* we requeue for retry because the error was retryable, and
1942 * the request was not marked fast fail. Note that above,
1943 * even if the request is marked fast fail, we still requeue
1944 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1945 if ((++scmd->retries) <= scmd->allowed
1946 && !scsi_noretry_cmd(scmd)) {
1947 return NEEDS_RETRY;
1948 } else {
1949 /*
1950 * no more retries - report this one back to upper level.
1951 */
1952 return SUCCESS;
1953 }
1954}
1955
1956static void eh_lock_door_done(struct request *req, blk_status_t status)
1957{
1958 blk_put_request(req);
1959}
1960
1961/**
1962 * scsi_eh_lock_door - Prevent medium removal for the specified device
1963 * @sdev: SCSI device to prevent medium removal
1964 *
1965 * Locking:
1966 * We must be called from process context.
1967 *
1968 * Notes:
1969 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1970 * head of the devices request queue, and continue.
1971 */
1972static void scsi_eh_lock_door(struct scsi_device *sdev)
1973{
1974 struct request *req;
1975 struct scsi_request *rq;
1976
1977 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0);
1978 if (IS_ERR(req))
1979 return;
1980 rq = scsi_req(req);
1981
1982 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1983 rq->cmd[1] = 0;
1984 rq->cmd[2] = 0;
1985 rq->cmd[3] = 0;
1986 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1987 rq->cmd[5] = 0;
1988 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1989
1990 req->rq_flags |= RQF_QUIET;
1991 req->timeout = 10 * HZ;
1992 rq->retries = 5;
1993
1994 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1995}
1996
1997/**
1998 * scsi_restart_operations - restart io operations to the specified host.
1999 * @shost: Host we are restarting.
2000 *
2001 * Notes:
2002 * When we entered the error handler, we blocked all further i/o to
2003 * this device. we need to 'reverse' this process.
2004 */
2005static void scsi_restart_operations(struct Scsi_Host *shost)
2006{
2007 struct scsi_device *sdev;
2008 unsigned long flags;
2009
2010 /*
2011 * If the door was locked, we need to insert a door lock request
2012 * onto the head of the SCSI request queue for the device. There
2013 * is no point trying to lock the door of an off-line device.
2014 */
2015 shost_for_each_device(sdev, shost) {
2016 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2017 scsi_eh_lock_door(sdev);
2018 sdev->was_reset = 0;
2019 }
2020 }
2021
2022 /*
2023 * next free up anything directly waiting upon the host. this
2024 * will be requests for character device operations, and also for
2025 * ioctls to queued block devices.
2026 */
2027 SCSI_LOG_ERROR_RECOVERY(3,
2028 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2029
2030 spin_lock_irqsave(shost->host_lock, flags);
2031 if (scsi_host_set_state(shost, SHOST_RUNNING))
2032 if (scsi_host_set_state(shost, SHOST_CANCEL))
2033 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2034 spin_unlock_irqrestore(shost->host_lock, flags);
2035
2036 wake_up(&shost->host_wait);
2037
2038 /*
2039 * finally we need to re-initiate requests that may be pending. we will
2040 * have had everything blocked while error handling is taking place, and
2041 * now that error recovery is done, we will need to ensure that these
2042 * requests are started.
2043 */
2044 scsi_run_host_queues(shost);
2045
2046 /*
2047 * if eh is active and host_eh_scheduled is pending we need to re-run
2048 * recovery. we do this check after scsi_run_host_queues() to allow
2049 * everything pent up since the last eh run a chance to make forward
2050 * progress before we sync again. Either we'll immediately re-run
2051 * recovery or scsi_device_unbusy() will wake us again when these
2052 * pending commands complete.
2053 */
2054 spin_lock_irqsave(shost->host_lock, flags);
2055 if (shost->host_eh_scheduled)
2056 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2057 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2058 spin_unlock_irqrestore(shost->host_lock, flags);
2059}
2060
2061/**
2062 * scsi_eh_ready_devs - check device ready state and recover if not.
2063 * @shost: host to be recovered.
2064 * @work_q: &list_head for pending commands.
2065 * @done_q: &list_head for processed commands.
2066 */
2067void scsi_eh_ready_devs(struct Scsi_Host *shost,
2068 struct list_head *work_q,
2069 struct list_head *done_q)
2070{
2071 if (!scsi_eh_stu(shost, work_q, done_q))
2072 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2073 if (!scsi_eh_target_reset(shost, work_q, done_q))
2074 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2075 if (!scsi_eh_host_reset(shost, work_q, done_q))
2076 scsi_eh_offline_sdevs(work_q,
2077 done_q);
2078}
2079EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2080
2081/**
2082 * scsi_eh_flush_done_q - finish processed commands or retry them.
2083 * @done_q: list_head of processed commands.
2084 */
2085void scsi_eh_flush_done_q(struct list_head *done_q)
2086{
2087 struct scsi_cmnd *scmd, *next;
2088
2089 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2090 list_del_init(&scmd->eh_entry);
2091 if (scsi_device_online(scmd->device) &&
2092 !scsi_noretry_cmd(scmd) &&
2093 (++scmd->retries <= scmd->allowed)) {
2094 SCSI_LOG_ERROR_RECOVERY(3,
2095 scmd_printk(KERN_INFO, scmd,
2096 "%s: flush retry cmd\n",
2097 current->comm));
2098 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2099 } else {
2100 /*
2101 * If just we got sense for the device (called
2102 * scsi_eh_get_sense), scmd->result is already
2103 * set, do not set DRIVER_TIMEOUT.
2104 */
2105 if (!scmd->result)
2106 scmd->result |= (DRIVER_TIMEOUT << 24);
2107 SCSI_LOG_ERROR_RECOVERY(3,
2108 scmd_printk(KERN_INFO, scmd,
2109 "%s: flush finish cmd\n",
2110 current->comm));
2111 scsi_finish_command(scmd);
2112 }
2113 }
2114}
2115EXPORT_SYMBOL(scsi_eh_flush_done_q);
2116
2117/**
2118 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2119 * @shost: Host to unjam.
2120 *
2121 * Notes:
2122 * When we come in here, we *know* that all commands on the bus have
2123 * either completed, failed or timed out. we also know that no further
2124 * commands are being sent to the host, so things are relatively quiet
2125 * and we have freedom to fiddle with things as we wish.
2126 *
2127 * This is only the *default* implementation. it is possible for
2128 * individual drivers to supply their own version of this function, and
2129 * if the maintainer wishes to do this, it is strongly suggested that
2130 * this function be taken as a template and modified. this function
2131 * was designed to correctly handle problems for about 95% of the
2132 * different cases out there, and it should always provide at least a
2133 * reasonable amount of error recovery.
2134 *
2135 * Any command marked 'failed' or 'timeout' must eventually have
2136 * scsi_finish_cmd() called for it. we do all of the retry stuff
2137 * here, so when we restart the host after we return it should have an
2138 * empty queue.
2139 */
2140static void scsi_unjam_host(struct Scsi_Host *shost)
2141{
2142 unsigned long flags;
2143 LIST_HEAD(eh_work_q);
2144 LIST_HEAD(eh_done_q);
2145
2146 spin_lock_irqsave(shost->host_lock, flags);
2147 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2148 spin_unlock_irqrestore(shost->host_lock, flags);
2149
2150 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2151
2152 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2153 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2154
2155 spin_lock_irqsave(shost->host_lock, flags);
2156 if (shost->eh_deadline != -1)
2157 shost->last_reset = 0;
2158 spin_unlock_irqrestore(shost->host_lock, flags);
2159 scsi_eh_flush_done_q(&eh_done_q);
2160}
2161
2162/**
2163 * scsi_error_handler - SCSI error handler thread
2164 * @data: Host for which we are running.
2165 *
2166 * Notes:
2167 * This is the main error handling loop. This is run as a kernel thread
2168 * for every SCSI host and handles all error handling activity.
2169 */
2170int scsi_error_handler(void *data)
2171{
2172 struct Scsi_Host *shost = data;
2173
2174 /*
2175 * We use TASK_INTERRUPTIBLE so that the thread is not
2176 * counted against the load average as a running process.
2177 * We never actually get interrupted because kthread_run
2178 * disables signal delivery for the created thread.
2179 */
2180 while (true) {
2181 /*
2182 * The sequence in kthread_stop() sets the stop flag first
2183 * then wakes the process. To avoid missed wakeups, the task
2184 * should always be in a non running state before the stop
2185 * flag is checked
2186 */
2187 set_current_state(TASK_INTERRUPTIBLE);
2188 if (kthread_should_stop())
2189 break;
2190
2191 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2192 shost->host_failed != scsi_host_busy(shost)) {
2193 SCSI_LOG_ERROR_RECOVERY(1,
2194 shost_printk(KERN_INFO, shost,
2195 "scsi_eh_%d: sleeping\n",
2196 shost->host_no));
2197 schedule();
2198 continue;
2199 }
2200
2201 __set_current_state(TASK_RUNNING);
2202 SCSI_LOG_ERROR_RECOVERY(1,
2203 shost_printk(KERN_INFO, shost,
2204 "scsi_eh_%d: waking up %d/%d/%d\n",
2205 shost->host_no, shost->host_eh_scheduled,
2206 shost->host_failed,
2207 scsi_host_busy(shost)));
2208
2209 /*
2210 * We have a host that is failing for some reason. Figure out
2211 * what we need to do to get it up and online again (if we can).
2212 * If we fail, we end up taking the thing offline.
2213 */
2214 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2215 SCSI_LOG_ERROR_RECOVERY(1,
2216 shost_printk(KERN_ERR, shost,
2217 "scsi_eh_%d: unable to autoresume\n",
2218 shost->host_no));
2219 continue;
2220 }
2221
2222 if (shost->transportt->eh_strategy_handler)
2223 shost->transportt->eh_strategy_handler(shost);
2224 else
2225 scsi_unjam_host(shost);
2226
2227 /* All scmds have been handled */
2228 shost->host_failed = 0;
2229
2230 /*
2231 * Note - if the above fails completely, the action is to take
2232 * individual devices offline and flush the queue of any
2233 * outstanding requests that may have been pending. When we
2234 * restart, we restart any I/O to any other devices on the bus
2235 * which are still online.
2236 */
2237 scsi_restart_operations(shost);
2238 if (!shost->eh_noresume)
2239 scsi_autopm_put_host(shost);
2240 }
2241 __set_current_state(TASK_RUNNING);
2242
2243 SCSI_LOG_ERROR_RECOVERY(1,
2244 shost_printk(KERN_INFO, shost,
2245 "Error handler scsi_eh_%d exiting\n",
2246 shost->host_no));
2247 shost->ehandler = NULL;
2248 return 0;
2249}
2250
2251/*
2252 * Function: scsi_report_bus_reset()
2253 *
2254 * Purpose: Utility function used by low-level drivers to report that
2255 * they have observed a bus reset on the bus being handled.
2256 *
2257 * Arguments: shost - Host in question
2258 * channel - channel on which reset was observed.
2259 *
2260 * Returns: Nothing
2261 *
2262 * Lock status: Host lock must be held.
2263 *
2264 * Notes: This only needs to be called if the reset is one which
2265 * originates from an unknown location. Resets originated
2266 * by the mid-level itself don't need to call this, but there
2267 * should be no harm.
2268 *
2269 * The main purpose of this is to make sure that a CHECK_CONDITION
2270 * is properly treated.
2271 */
2272void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2273{
2274 struct scsi_device *sdev;
2275
2276 __shost_for_each_device(sdev, shost) {
2277 if (channel == sdev_channel(sdev))
2278 __scsi_report_device_reset(sdev, NULL);
2279 }
2280}
2281EXPORT_SYMBOL(scsi_report_bus_reset);
2282
2283/*
2284 * Function: scsi_report_device_reset()
2285 *
2286 * Purpose: Utility function used by low-level drivers to report that
2287 * they have observed a device reset on the device being handled.
2288 *
2289 * Arguments: shost - Host in question
2290 * channel - channel on which reset was observed
2291 * target - target on which reset was observed
2292 *
2293 * Returns: Nothing
2294 *
2295 * Lock status: Host lock must be held
2296 *
2297 * Notes: This only needs to be called if the reset is one which
2298 * originates from an unknown location. Resets originated
2299 * by the mid-level itself don't need to call this, but there
2300 * should be no harm.
2301 *
2302 * The main purpose of this is to make sure that a CHECK_CONDITION
2303 * is properly treated.
2304 */
2305void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2306{
2307 struct scsi_device *sdev;
2308
2309 __shost_for_each_device(sdev, shost) {
2310 if (channel == sdev_channel(sdev) &&
2311 target == sdev_id(sdev))
2312 __scsi_report_device_reset(sdev, NULL);
2313 }
2314}
2315EXPORT_SYMBOL(scsi_report_device_reset);
2316
2317static void
2318scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2319{
2320}
2321
2322/**
2323 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2324 * @dev: scsi_device to operate on
2325 * @arg: reset type (see sg.h)
2326 */
2327int
2328scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2329{
2330 struct scsi_cmnd *scmd;
2331 struct Scsi_Host *shost = dev->host;
2332 struct request *rq;
2333 unsigned long flags;
2334 int error = 0, rtn, val;
2335
2336 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2337 return -EACCES;
2338
2339 error = get_user(val, arg);
2340 if (error)
2341 return error;
2342
2343 if (scsi_autopm_get_host(shost) < 0)
2344 return -EIO;
2345
2346 error = -EIO;
2347 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2348 shost->hostt->cmd_size, GFP_KERNEL);
2349 if (!rq)
2350 goto out_put_autopm_host;
2351 blk_rq_init(NULL, rq);
2352
2353 scmd = (struct scsi_cmnd *)(rq + 1);
2354 scsi_init_command(dev, scmd);
2355 scmd->request = rq;
2356 scmd->cmnd = scsi_req(rq)->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 /* FALLTHROUGH */
2393 default:
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 kfree(rq);
2417
2418out_put_autopm_host:
2419 scsi_autopm_put_host(shost);
2420 return error;
2421}
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 * true if information field found, false if not found.
2440 */
2441bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2442 u64 *info_out)
2443{
2444 const u8 * ucp;
2445
2446 if (sb_len < 7)
2447 return false;
2448 switch (sense_buffer[0] & 0x7f) {
2449 case 0x70:
2450 case 0x71:
2451 if (sense_buffer[0] & 0x80) {
2452 *info_out = get_unaligned_be32(&sense_buffer[3]);
2453 return true;
2454 }
2455 return false;
2456 case 0x72:
2457 case 0x73:
2458 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2459 0 /* info desc */);
2460 if (ucp && (0xa == ucp[1])) {
2461 *info_out = get_unaligned_be64(&ucp[4]);
2462 return true;
2463 }
2464 return false;
2465 default:
2466 return false;
2467 }
2468}
2469EXPORT_SYMBOL(scsi_get_sense_info_fld);