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_transport.h>
37#include <scsi/scsi_host.h>
38#include <scsi/scsi_ioctl.h>
39
40#include "scsi_priv.h"
41#include "scsi_logging.h"
42#include "scsi_transport_api.h"
43
44#include <trace/events/scsi.h>
45
46static void scsi_eh_done(struct scsi_cmnd *scmd);
47
48/*
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
51 */
52#define BUS_RESET_SETTLE_TIME (10)
53#define HOST_RESET_SETTLE_TIME (10)
54
55static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
56static int scsi_try_to_abort_cmd(struct scsi_host_template *,
57 struct scsi_cmnd *);
58
59/* called with shost->host_lock held */
60void scsi_eh_wakeup(struct Scsi_Host *shost)
61{
62 if (shost->host_busy == shost->host_failed) {
63 trace_scsi_eh_wakeup(shost);
64 wake_up_process(shost->ehandler);
65 SCSI_LOG_ERROR_RECOVERY(5,
66 printk("Waking error handler thread\n"));
67 }
68}
69
70/**
71 * scsi_schedule_eh - schedule EH for SCSI host
72 * @shost: SCSI host to invoke error handling on.
73 *
74 * Schedule SCSI EH without scmd.
75 */
76void scsi_schedule_eh(struct Scsi_Host *shost)
77{
78 unsigned long flags;
79
80 spin_lock_irqsave(shost->host_lock, flags);
81
82 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
83 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
84 shost->host_eh_scheduled++;
85 scsi_eh_wakeup(shost);
86 }
87
88 spin_unlock_irqrestore(shost->host_lock, flags);
89}
90EXPORT_SYMBOL_GPL(scsi_schedule_eh);
91
92static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
93{
94 if (!shost->last_reset || shost->eh_deadline == -1)
95 return 0;
96
97 /*
98 * 32bit accesses are guaranteed to be atomic
99 * (on all supported architectures), so instead
100 * of using a spinlock we can as well double check
101 * if eh_deadline has been set to 'off' during the
102 * time_before call.
103 */
104 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
105 shost->eh_deadline > -1)
106 return 0;
107
108 return 1;
109}
110
111/**
112 * scmd_eh_abort_handler - Handle command aborts
113 * @work: command to be aborted.
114 */
115void
116scmd_eh_abort_handler(struct work_struct *work)
117{
118 struct scsi_cmnd *scmd =
119 container_of(work, struct scsi_cmnd, abort_work.work);
120 struct scsi_device *sdev = scmd->device;
121 int rtn;
122
123 if (scsi_host_eh_past_deadline(sdev->host)) {
124 SCSI_LOG_ERROR_RECOVERY(3,
125 scmd_printk(KERN_INFO, scmd,
126 "scmd %p eh timeout, not aborting\n",
127 scmd));
128 } else {
129 SCSI_LOG_ERROR_RECOVERY(3,
130 scmd_printk(KERN_INFO, scmd,
131 "aborting command %p\n", scmd));
132 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
133 if (rtn == SUCCESS) {
134 scmd->result |= DID_TIME_OUT << 16;
135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "scmd %p eh timeout, "
139 "not retrying aborted "
140 "command\n", scmd));
141 } else if (!scsi_noretry_cmd(scmd) &&
142 (++scmd->retries <= scmd->allowed)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING, scmd,
145 "scmd %p retry "
146 "aborted command\n", scmd));
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 "scmd %p finish "
153 "aborted command\n", scmd));
154 scsi_finish_command(scmd);
155 return;
156 }
157 } else {
158 SCSI_LOG_ERROR_RECOVERY(3,
159 scmd_printk(KERN_INFO, scmd,
160 "scmd %p abort failed, rtn %d\n",
161 scmd, rtn));
162 }
163 }
164
165 if (!scsi_eh_scmd_add(scmd, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING, scmd,
168 "scmd %p terminate "
169 "aborted command\n", scmd));
170 scmd->result |= DID_TIME_OUT << 16;
171 scsi_finish_command(scmd);
172 }
173}
174
175/**
176 * scsi_abort_command - schedule a command abort
177 * @scmd: scmd to abort.
178 *
179 * We only need to abort commands after a command timeout
180 */
181static int
182scsi_abort_command(struct scsi_cmnd *scmd)
183{
184 struct scsi_device *sdev = scmd->device;
185 struct Scsi_Host *shost = sdev->host;
186 unsigned long flags;
187
188 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
189 /*
190 * Retry after abort failed, escalate to next level.
191 */
192 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
193 SCSI_LOG_ERROR_RECOVERY(3,
194 scmd_printk(KERN_INFO, scmd,
195 "scmd %p previous abort failed\n", scmd));
196 cancel_delayed_work(&scmd->abort_work);
197 return FAILED;
198 }
199
200 /*
201 * Do not try a command abort if
202 * SCSI EH has already started.
203 */
204 spin_lock_irqsave(shost->host_lock, flags);
205 if (scsi_host_in_recovery(shost)) {
206 spin_unlock_irqrestore(shost->host_lock, flags);
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd,
209 "scmd %p not aborting, host in recovery\n",
210 scmd));
211 return FAILED;
212 }
213
214 if (shost->eh_deadline != -1 && !shost->last_reset)
215 shost->last_reset = jiffies;
216 spin_unlock_irqrestore(shost->host_lock, flags);
217
218 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
219 SCSI_LOG_ERROR_RECOVERY(3,
220 scmd_printk(KERN_INFO, scmd,
221 "scmd %p abort scheduled\n", scmd));
222 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
223 return SUCCESS;
224}
225
226/**
227 * scsi_eh_scmd_add - add scsi cmd to error handling.
228 * @scmd: scmd to run eh on.
229 * @eh_flag: optional SCSI_EH flag.
230 *
231 * Return value:
232 * 0 on failure.
233 */
234int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
235{
236 struct Scsi_Host *shost = scmd->device->host;
237 unsigned long flags;
238 int ret = 0;
239
240 if (!shost->ehandler)
241 return 0;
242
243 spin_lock_irqsave(shost->host_lock, flags);
244 if (scsi_host_set_state(shost, SHOST_RECOVERY))
245 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
246 goto out_unlock;
247
248 if (shost->eh_deadline != -1 && !shost->last_reset)
249 shost->last_reset = jiffies;
250
251 ret = 1;
252 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
253 eh_flag &= ~SCSI_EH_CANCEL_CMD;
254 scmd->eh_eflags |= eh_flag;
255 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
256 shost->host_failed++;
257 scsi_eh_wakeup(shost);
258 out_unlock:
259 spin_unlock_irqrestore(shost->host_lock, flags);
260 return ret;
261}
262
263/**
264 * scsi_times_out - Timeout function for normal scsi commands.
265 * @req: request that is timing out.
266 *
267 * Notes:
268 * We do not need to lock this. There is the potential for a race
269 * only in that the normal completion handling might run, but if the
270 * normal completion function determines that the timer has already
271 * fired, then it mustn't do anything.
272 */
273enum blk_eh_timer_return scsi_times_out(struct request *req)
274{
275 struct scsi_cmnd *scmd = req->special;
276 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
277 struct Scsi_Host *host = scmd->device->host;
278
279 trace_scsi_dispatch_cmd_timeout(scmd);
280 scsi_log_completion(scmd, TIMEOUT_ERROR);
281
282 if (host->eh_deadline != -1 && !host->last_reset)
283 host->last_reset = jiffies;
284
285 if (host->transportt->eh_timed_out)
286 rtn = host->transportt->eh_timed_out(scmd);
287 else if (host->hostt->eh_timed_out)
288 rtn = host->hostt->eh_timed_out(scmd);
289
290 if (rtn == BLK_EH_NOT_HANDLED && !host->hostt->no_async_abort)
291 if (scsi_abort_command(scmd) == SUCCESS)
292 return BLK_EH_NOT_HANDLED;
293
294 scmd->result |= DID_TIME_OUT << 16;
295
296 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
297 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
298 rtn = BLK_EH_HANDLED;
299
300 return rtn;
301}
302
303/**
304 * scsi_block_when_processing_errors - Prevent cmds from being queued.
305 * @sdev: Device on which we are performing recovery.
306 *
307 * Description:
308 * We block until the host is out of error recovery, and then check to
309 * see whether the host or the device is offline.
310 *
311 * Return value:
312 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
313 */
314int scsi_block_when_processing_errors(struct scsi_device *sdev)
315{
316 int online;
317
318 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
319
320 online = scsi_device_online(sdev);
321
322 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
323 online));
324
325 return online;
326}
327EXPORT_SYMBOL(scsi_block_when_processing_errors);
328
329#ifdef CONFIG_SCSI_LOGGING
330/**
331 * scsi_eh_prt_fail_stats - Log info on failures.
332 * @shost: scsi host being recovered.
333 * @work_q: Queue of scsi cmds to process.
334 */
335static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
336 struct list_head *work_q)
337{
338 struct scsi_cmnd *scmd;
339 struct scsi_device *sdev;
340 int total_failures = 0;
341 int cmd_failed = 0;
342 int cmd_cancel = 0;
343 int devices_failed = 0;
344
345 shost_for_each_device(sdev, shost) {
346 list_for_each_entry(scmd, work_q, eh_entry) {
347 if (scmd->device == sdev) {
348 ++total_failures;
349 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
350 ++cmd_cancel;
351 else
352 ++cmd_failed;
353 }
354 }
355
356 if (cmd_cancel || cmd_failed) {
357 SCSI_LOG_ERROR_RECOVERY(3,
358 sdev_printk(KERN_INFO, sdev,
359 "%s: cmds failed: %d, cancel: %d\n",
360 __func__, cmd_failed,
361 cmd_cancel));
362 cmd_cancel = 0;
363 cmd_failed = 0;
364 ++devices_failed;
365 }
366 }
367
368 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
369 " devices require eh work\n",
370 total_failures, devices_failed));
371}
372#endif
373
374 /**
375 * scsi_report_lun_change - Set flag on all *other* devices on the same target
376 * to indicate that a UNIT ATTENTION is expected.
377 * @sdev: Device reporting the UNIT ATTENTION
378 */
379static void scsi_report_lun_change(struct scsi_device *sdev)
380{
381 sdev->sdev_target->expecting_lun_change = 1;
382}
383
384/**
385 * scsi_report_sense - Examine scsi sense information and log messages for
386 * certain conditions, also issue uevents for some of them.
387 * @sdev: Device reporting the sense code
388 * @sshdr: sshdr to be examined
389 */
390static void scsi_report_sense(struct scsi_device *sdev,
391 struct scsi_sense_hdr *sshdr)
392{
393 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
394
395 if (sshdr->sense_key == UNIT_ATTENTION) {
396 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
397 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
398 sdev_printk(KERN_WARNING, sdev,
399 "Inquiry data has changed");
400 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
401 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
402 scsi_report_lun_change(sdev);
403 sdev_printk(KERN_WARNING, sdev,
404 "Warning! Received an indication that the "
405 "LUN assignments on this target have "
406 "changed. The Linux SCSI layer does not "
407 "automatically remap LUN assignments.\n");
408 } else if (sshdr->asc == 0x3f)
409 sdev_printk(KERN_WARNING, sdev,
410 "Warning! Received an indication that the "
411 "operating parameters on this target have "
412 "changed. The Linux SCSI layer does not "
413 "automatically adjust these parameters.\n");
414
415 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
416 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
417 sdev_printk(KERN_WARNING, sdev,
418 "Warning! Received an indication that the "
419 "LUN reached a thin provisioning soft "
420 "threshold.\n");
421 }
422
423 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
424 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
425 sdev_printk(KERN_WARNING, sdev,
426 "Mode parameters changed");
427 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
428 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
429 sdev_printk(KERN_WARNING, sdev,
430 "Capacity data has changed");
431 } else if (sshdr->asc == 0x2a)
432 sdev_printk(KERN_WARNING, sdev,
433 "Parameters changed");
434 }
435
436 if (evt_type != SDEV_EVT_MAXBITS) {
437 set_bit(evt_type, sdev->pending_events);
438 schedule_work(&sdev->event_work);
439 }
440}
441
442/**
443 * scsi_check_sense - Examine scsi cmd sense
444 * @scmd: Cmd to have sense checked.
445 *
446 * Return value:
447 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
448 *
449 * Notes:
450 * When a deferred error is detected the current command has
451 * not been executed and needs retrying.
452 */
453static int scsi_check_sense(struct scsi_cmnd *scmd)
454{
455 struct scsi_device *sdev = scmd->device;
456 struct scsi_sense_hdr sshdr;
457
458 if (! scsi_command_normalize_sense(scmd, &sshdr))
459 return FAILED; /* no valid sense data */
460
461 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
462 /*
463 * nasty: for mid-layer issued TURs, we need to return the
464 * actual sense data without any recovery attempt. For eh
465 * issued ones, we need to try to recover and interpret
466 */
467 return SUCCESS;
468
469 scsi_report_sense(sdev, &sshdr);
470
471 if (scsi_sense_is_deferred(&sshdr))
472 return NEEDS_RETRY;
473
474 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
475 sdev->scsi_dh_data->scsi_dh->check_sense) {
476 int rc;
477
478 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
479 if (rc != SCSI_RETURN_NOT_HANDLED)
480 return rc;
481 /* handler does not care. Drop down to default handling */
482 }
483
484 /*
485 * Previous logic looked for FILEMARK, EOM or ILI which are
486 * mainly associated with tapes and returned SUCCESS.
487 */
488 if (sshdr.response_code == 0x70) {
489 /* fixed format */
490 if (scmd->sense_buffer[2] & 0xe0)
491 return SUCCESS;
492 } else {
493 /*
494 * descriptor format: look for "stream commands sense data
495 * descriptor" (see SSC-3). Assume single sense data
496 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
497 */
498 if ((sshdr.additional_length > 3) &&
499 (scmd->sense_buffer[8] == 0x4) &&
500 (scmd->sense_buffer[11] & 0xe0))
501 return SUCCESS;
502 }
503
504 switch (sshdr.sense_key) {
505 case NO_SENSE:
506 return SUCCESS;
507 case RECOVERED_ERROR:
508 return /* soft_error */ SUCCESS;
509
510 case ABORTED_COMMAND:
511 if (sshdr.asc == 0x10) /* DIF */
512 return SUCCESS;
513
514 return NEEDS_RETRY;
515 case NOT_READY:
516 case UNIT_ATTENTION:
517 /*
518 * if we are expecting a cc/ua because of a bus reset that we
519 * performed, treat this just as a retry. otherwise this is
520 * information that we should pass up to the upper-level driver
521 * so that we can deal with it there.
522 */
523 if (scmd->device->expecting_cc_ua) {
524 /*
525 * Because some device does not queue unit
526 * attentions correctly, we carefully check
527 * additional sense code and qualifier so as
528 * not to squash media change unit attention.
529 */
530 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
531 scmd->device->expecting_cc_ua = 0;
532 return NEEDS_RETRY;
533 }
534 }
535 /*
536 * we might also expect a cc/ua if another LUN on the target
537 * reported a UA with an ASC/ASCQ of 3F 0E -
538 * REPORTED LUNS DATA HAS CHANGED.
539 */
540 if (scmd->device->sdev_target->expecting_lun_change &&
541 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
542 return NEEDS_RETRY;
543 /*
544 * if the device is in the process of becoming ready, we
545 * should retry.
546 */
547 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
548 return NEEDS_RETRY;
549 /*
550 * if the device is not started, we need to wake
551 * the error handler to start the motor
552 */
553 if (scmd->device->allow_restart &&
554 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
555 return FAILED;
556 /*
557 * Pass the UA upwards for a determination in the completion
558 * functions.
559 */
560 return SUCCESS;
561
562 /* these are not supported */
563 case DATA_PROTECT:
564 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
565 /* Thin provisioning hard threshold reached */
566 set_host_byte(scmd, DID_ALLOC_FAILURE);
567 return SUCCESS;
568 }
569 case COPY_ABORTED:
570 case VOLUME_OVERFLOW:
571 case MISCOMPARE:
572 case BLANK_CHECK:
573 set_host_byte(scmd, DID_TARGET_FAILURE);
574 return SUCCESS;
575
576 case MEDIUM_ERROR:
577 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
578 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
579 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
580 set_host_byte(scmd, DID_MEDIUM_ERROR);
581 return SUCCESS;
582 }
583 return NEEDS_RETRY;
584
585 case HARDWARE_ERROR:
586 if (scmd->device->retry_hwerror)
587 return ADD_TO_MLQUEUE;
588 else
589 set_host_byte(scmd, DID_TARGET_FAILURE);
590
591 case ILLEGAL_REQUEST:
592 if (sshdr.asc == 0x20 || /* Invalid command operation code */
593 sshdr.asc == 0x21 || /* Logical block address out of range */
594 sshdr.asc == 0x24 || /* Invalid field in cdb */
595 sshdr.asc == 0x26) { /* Parameter value invalid */
596 set_host_byte(scmd, DID_TARGET_FAILURE);
597 }
598 return SUCCESS;
599
600 default:
601 return SUCCESS;
602 }
603}
604
605static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
606{
607 struct scsi_host_template *sht = sdev->host->hostt;
608 struct scsi_device *tmp_sdev;
609
610 if (!sht->change_queue_depth ||
611 sdev->queue_depth >= sdev->max_queue_depth)
612 return;
613
614 if (time_before(jiffies,
615 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
616 return;
617
618 if (time_before(jiffies,
619 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
620 return;
621
622 /*
623 * Walk all devices of a target and do
624 * ramp up on them.
625 */
626 shost_for_each_device(tmp_sdev, sdev->host) {
627 if (tmp_sdev->channel != sdev->channel ||
628 tmp_sdev->id != sdev->id ||
629 tmp_sdev->queue_depth == sdev->max_queue_depth)
630 continue;
631 /*
632 * call back into LLD to increase queue_depth by one
633 * with ramp up reason code.
634 */
635 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
636 SCSI_QDEPTH_RAMP_UP);
637 sdev->last_queue_ramp_up = jiffies;
638 }
639}
640
641static void scsi_handle_queue_full(struct scsi_device *sdev)
642{
643 struct scsi_host_template *sht = sdev->host->hostt;
644 struct scsi_device *tmp_sdev;
645
646 if (!sht->change_queue_depth)
647 return;
648
649 shost_for_each_device(tmp_sdev, sdev->host) {
650 if (tmp_sdev->channel != sdev->channel ||
651 tmp_sdev->id != sdev->id)
652 continue;
653 /*
654 * We do not know the number of commands that were at
655 * the device when we got the queue full so we start
656 * from the highest possible value and work our way down.
657 */
658 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
659 SCSI_QDEPTH_QFULL);
660 }
661}
662
663/**
664 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
665 * @scmd: SCSI cmd to examine.
666 *
667 * Notes:
668 * This is *only* called when we are examining the status of commands
669 * queued during error recovery. the main difference here is that we
670 * don't allow for the possibility of retries here, and we are a lot
671 * more restrictive about what we consider acceptable.
672 */
673static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
674{
675 /*
676 * first check the host byte, to see if there is anything in there
677 * that would indicate what we need to do.
678 */
679 if (host_byte(scmd->result) == DID_RESET) {
680 /*
681 * rats. we are already in the error handler, so we now
682 * get to try and figure out what to do next. if the sense
683 * is valid, we have a pretty good idea of what to do.
684 * if not, we mark it as FAILED.
685 */
686 return scsi_check_sense(scmd);
687 }
688 if (host_byte(scmd->result) != DID_OK)
689 return FAILED;
690
691 /*
692 * next, check the message byte.
693 */
694 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
695 return FAILED;
696
697 /*
698 * now, check the status byte to see if this indicates
699 * anything special.
700 */
701 switch (status_byte(scmd->result)) {
702 case GOOD:
703 scsi_handle_queue_ramp_up(scmd->device);
704 case COMMAND_TERMINATED:
705 return SUCCESS;
706 case CHECK_CONDITION:
707 return scsi_check_sense(scmd);
708 case CONDITION_GOOD:
709 case INTERMEDIATE_GOOD:
710 case INTERMEDIATE_C_GOOD:
711 /*
712 * who knows? FIXME(eric)
713 */
714 return SUCCESS;
715 case RESERVATION_CONFLICT:
716 if (scmd->cmnd[0] == TEST_UNIT_READY)
717 /* it is a success, we probed the device and
718 * found it */
719 return SUCCESS;
720 /* otherwise, we failed to send the command */
721 return FAILED;
722 case QUEUE_FULL:
723 scsi_handle_queue_full(scmd->device);
724 /* fall through */
725 case BUSY:
726 return NEEDS_RETRY;
727 default:
728 return FAILED;
729 }
730 return FAILED;
731}
732
733/**
734 * scsi_eh_done - Completion function for error handling.
735 * @scmd: Cmd that is done.
736 */
737static void scsi_eh_done(struct scsi_cmnd *scmd)
738{
739 struct completion *eh_action;
740
741 SCSI_LOG_ERROR_RECOVERY(3,
742 printk("%s scmd: %p result: %x\n",
743 __func__, scmd, scmd->result));
744
745 eh_action = scmd->device->host->eh_action;
746 if (eh_action)
747 complete(eh_action);
748}
749
750/**
751 * scsi_try_host_reset - ask host adapter to reset itself
752 * @scmd: SCSI cmd to send host reset.
753 */
754static int scsi_try_host_reset(struct scsi_cmnd *scmd)
755{
756 unsigned long flags;
757 int rtn;
758 struct Scsi_Host *host = scmd->device->host;
759 struct scsi_host_template *hostt = host->hostt;
760
761 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
762 __func__));
763
764 if (!hostt->eh_host_reset_handler)
765 return FAILED;
766
767 rtn = hostt->eh_host_reset_handler(scmd);
768
769 if (rtn == SUCCESS) {
770 if (!hostt->skip_settle_delay)
771 ssleep(HOST_RESET_SETTLE_TIME);
772 spin_lock_irqsave(host->host_lock, flags);
773 scsi_report_bus_reset(host, scmd_channel(scmd));
774 spin_unlock_irqrestore(host->host_lock, flags);
775 }
776
777 return rtn;
778}
779
780/**
781 * scsi_try_bus_reset - ask host to perform a bus reset
782 * @scmd: SCSI cmd to send bus reset.
783 */
784static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
785{
786 unsigned long flags;
787 int rtn;
788 struct Scsi_Host *host = scmd->device->host;
789 struct scsi_host_template *hostt = host->hostt;
790
791 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
792 __func__));
793
794 if (!hostt->eh_bus_reset_handler)
795 return FAILED;
796
797 rtn = hostt->eh_bus_reset_handler(scmd);
798
799 if (rtn == SUCCESS) {
800 if (!hostt->skip_settle_delay)
801 ssleep(BUS_RESET_SETTLE_TIME);
802 spin_lock_irqsave(host->host_lock, flags);
803 scsi_report_bus_reset(host, scmd_channel(scmd));
804 spin_unlock_irqrestore(host->host_lock, flags);
805 }
806
807 return rtn;
808}
809
810static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
811{
812 sdev->was_reset = 1;
813 sdev->expecting_cc_ua = 1;
814}
815
816/**
817 * scsi_try_target_reset - Ask host to perform a target reset
818 * @scmd: SCSI cmd used to send a target reset
819 *
820 * Notes:
821 * There is no timeout for this operation. if this operation is
822 * unreliable for a given host, then the host itself needs to put a
823 * timer on it, and set the host back to a consistent state prior to
824 * returning.
825 */
826static int scsi_try_target_reset(struct scsi_cmnd *scmd)
827{
828 unsigned long flags;
829 int rtn;
830 struct Scsi_Host *host = scmd->device->host;
831 struct scsi_host_template *hostt = host->hostt;
832
833 if (!hostt->eh_target_reset_handler)
834 return FAILED;
835
836 rtn = hostt->eh_target_reset_handler(scmd);
837 if (rtn == SUCCESS) {
838 spin_lock_irqsave(host->host_lock, flags);
839 __starget_for_each_device(scsi_target(scmd->device), NULL,
840 __scsi_report_device_reset);
841 spin_unlock_irqrestore(host->host_lock, flags);
842 }
843
844 return rtn;
845}
846
847/**
848 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
849 * @scmd: SCSI cmd used to send BDR
850 *
851 * Notes:
852 * There is no timeout for this operation. if this operation is
853 * unreliable for a given host, then the host itself needs to put a
854 * timer on it, and set the host back to a consistent state prior to
855 * returning.
856 */
857static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
858{
859 int rtn;
860 struct scsi_host_template *hostt = scmd->device->host->hostt;
861
862 if (!hostt->eh_device_reset_handler)
863 return FAILED;
864
865 rtn = hostt->eh_device_reset_handler(scmd);
866 if (rtn == SUCCESS)
867 __scsi_report_device_reset(scmd->device, NULL);
868 return rtn;
869}
870
871static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
872{
873 if (!hostt->eh_abort_handler)
874 return FAILED;
875
876 return hostt->eh_abort_handler(scmd);
877}
878
879static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
880{
881 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
882 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
883 if (scsi_try_target_reset(scmd) != SUCCESS)
884 if (scsi_try_bus_reset(scmd) != SUCCESS)
885 scsi_try_host_reset(scmd);
886}
887
888/**
889 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
890 * @scmd: SCSI command structure to hijack
891 * @ses: structure to save restore information
892 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
893 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
894 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
895 *
896 * This function is used to save a scsi command information before re-execution
897 * as part of the error recovery process. If @sense_bytes is 0 the command
898 * sent must be one that does not transfer any data. If @sense_bytes != 0
899 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
900 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
901 */
902void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
903 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
904{
905 struct scsi_device *sdev = scmd->device;
906
907 /*
908 * We need saved copies of a number of fields - this is because
909 * error handling may need to overwrite these with different values
910 * to run different commands, and once error handling is complete,
911 * we will need to restore these values prior to running the actual
912 * command.
913 */
914 ses->cmd_len = scmd->cmd_len;
915 ses->cmnd = scmd->cmnd;
916 ses->data_direction = scmd->sc_data_direction;
917 ses->sdb = scmd->sdb;
918 ses->next_rq = scmd->request->next_rq;
919 ses->result = scmd->result;
920 ses->underflow = scmd->underflow;
921 ses->prot_op = scmd->prot_op;
922
923 scmd->prot_op = SCSI_PROT_NORMAL;
924 scmd->eh_eflags = 0;
925 scmd->cmnd = ses->eh_cmnd;
926 memset(scmd->cmnd, 0, BLK_MAX_CDB);
927 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
928 scmd->request->next_rq = NULL;
929 scmd->result = 0;
930
931 if (sense_bytes) {
932 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
933 sense_bytes);
934 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
935 scmd->sdb.length);
936 scmd->sdb.table.sgl = &ses->sense_sgl;
937 scmd->sc_data_direction = DMA_FROM_DEVICE;
938 scmd->sdb.table.nents = 1;
939 scmd->cmnd[0] = REQUEST_SENSE;
940 scmd->cmnd[4] = scmd->sdb.length;
941 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
942 } else {
943 scmd->sc_data_direction = DMA_NONE;
944 if (cmnd) {
945 BUG_ON(cmnd_size > BLK_MAX_CDB);
946 memcpy(scmd->cmnd, cmnd, cmnd_size);
947 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
948 }
949 }
950
951 scmd->underflow = 0;
952
953 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
954 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
955 (sdev->lun << 5 & 0xe0);
956
957 /*
958 * Zero the sense buffer. The scsi spec mandates that any
959 * untransferred sense data should be interpreted as being zero.
960 */
961 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
962}
963EXPORT_SYMBOL(scsi_eh_prep_cmnd);
964
965/**
966 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
967 * @scmd: SCSI command structure to restore
968 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
969 *
970 * Undo any damage done by above scsi_eh_prep_cmnd().
971 */
972void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
973{
974 /*
975 * Restore original data
976 */
977 scmd->cmd_len = ses->cmd_len;
978 scmd->cmnd = ses->cmnd;
979 scmd->sc_data_direction = ses->data_direction;
980 scmd->sdb = ses->sdb;
981 scmd->request->next_rq = ses->next_rq;
982 scmd->result = ses->result;
983 scmd->underflow = ses->underflow;
984 scmd->prot_op = ses->prot_op;
985}
986EXPORT_SYMBOL(scsi_eh_restore_cmnd);
987
988/**
989 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
990 * @scmd: SCSI command structure to hijack
991 * @cmnd: CDB to send
992 * @cmnd_size: size in bytes of @cmnd
993 * @timeout: timeout for this request
994 * @sense_bytes: size of sense data to copy or 0
995 *
996 * This function is used to send a scsi command down to a target device
997 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
998 *
999 * Return value:
1000 * SUCCESS or FAILED or NEEDS_RETRY
1001 */
1002static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1003 int cmnd_size, int timeout, unsigned sense_bytes)
1004{
1005 struct scsi_device *sdev = scmd->device;
1006 struct Scsi_Host *shost = sdev->host;
1007 DECLARE_COMPLETION_ONSTACK(done);
1008 unsigned long timeleft = timeout;
1009 struct scsi_eh_save ses;
1010 const unsigned long stall_for = msecs_to_jiffies(100);
1011 int rtn;
1012
1013retry:
1014 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1015 shost->eh_action = &done;
1016
1017 scsi_log_send(scmd);
1018 scmd->scsi_done = scsi_eh_done;
1019 rtn = shost->hostt->queuecommand(shost, scmd);
1020 if (rtn) {
1021 if (timeleft > stall_for) {
1022 scsi_eh_restore_cmnd(scmd, &ses);
1023 timeleft -= stall_for;
1024 msleep(jiffies_to_msecs(stall_for));
1025 goto retry;
1026 }
1027 /* signal not to enter either branch of the if () below */
1028 timeleft = 0;
1029 rtn = NEEDS_RETRY;
1030 } else {
1031 timeleft = wait_for_completion_timeout(&done, timeout);
1032 }
1033
1034 shost->eh_action = NULL;
1035
1036 scsi_log_completion(scmd, rtn);
1037
1038 SCSI_LOG_ERROR_RECOVERY(3,
1039 printk("%s: scmd: %p, timeleft: %ld\n",
1040 __func__, scmd, timeleft));
1041
1042 /*
1043 * If there is time left scsi_eh_done got called, and we will examine
1044 * the actual status codes to see whether the command actually did
1045 * complete normally, else if we have a zero return and no time left,
1046 * the command must still be pending, so abort it and return FAILED.
1047 * If we never actually managed to issue the command, because
1048 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1049 * value above (so don't execute either branch of the if)
1050 */
1051 if (timeleft) {
1052 rtn = scsi_eh_completed_normally(scmd);
1053 SCSI_LOG_ERROR_RECOVERY(3,
1054 printk("%s: scsi_eh_completed_normally %x\n",
1055 __func__, rtn));
1056
1057 switch (rtn) {
1058 case SUCCESS:
1059 case NEEDS_RETRY:
1060 case FAILED:
1061 break;
1062 case ADD_TO_MLQUEUE:
1063 rtn = NEEDS_RETRY;
1064 break;
1065 default:
1066 rtn = FAILED;
1067 break;
1068 }
1069 } else if (!rtn) {
1070 scsi_abort_eh_cmnd(scmd);
1071 rtn = FAILED;
1072 }
1073
1074 scsi_eh_restore_cmnd(scmd, &ses);
1075
1076 return rtn;
1077}
1078
1079/**
1080 * scsi_request_sense - Request sense data from a particular target.
1081 * @scmd: SCSI cmd for request sense.
1082 *
1083 * Notes:
1084 * Some hosts automatically obtain this information, others require
1085 * that we obtain it on our own. This function will *not* return until
1086 * the command either times out, or it completes.
1087 */
1088static int scsi_request_sense(struct scsi_cmnd *scmd)
1089{
1090 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1091}
1092
1093static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1094{
1095 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1096 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1097 if (sdrv->eh_action)
1098 rtn = sdrv->eh_action(scmd, rtn);
1099 }
1100 return rtn;
1101}
1102
1103/**
1104 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1105 * @scmd: Original SCSI cmd that eh has finished.
1106 * @done_q: Queue for processed commands.
1107 *
1108 * Notes:
1109 * We don't want to use the normal command completion while we are are
1110 * still handling errors - it may cause other commands to be queued,
1111 * and that would disturb what we are doing. Thus we really want to
1112 * keep a list of pending commands for final completion, and once we
1113 * are ready to leave error handling we handle completion for real.
1114 */
1115void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1116{
1117 scmd->device->host->host_failed--;
1118 scmd->eh_eflags = 0;
1119 list_move_tail(&scmd->eh_entry, done_q);
1120}
1121EXPORT_SYMBOL(scsi_eh_finish_cmd);
1122
1123/**
1124 * scsi_eh_get_sense - Get device sense data.
1125 * @work_q: Queue of commands to process.
1126 * @done_q: Queue of processed commands.
1127 *
1128 * Description:
1129 * See if we need to request sense information. if so, then get it
1130 * now, so we have a better idea of what to do.
1131 *
1132 * Notes:
1133 * This has the unfortunate side effect that if a shost adapter does
1134 * not automatically request sense information, we end up shutting
1135 * it down before we request it.
1136 *
1137 * All drivers should request sense information internally these days,
1138 * so for now all I have to say is tough noogies if you end up in here.
1139 *
1140 * XXX: Long term this code should go away, but that needs an audit of
1141 * all LLDDs first.
1142 */
1143int scsi_eh_get_sense(struct list_head *work_q,
1144 struct list_head *done_q)
1145{
1146 struct scsi_cmnd *scmd, *next;
1147 struct Scsi_Host *shost;
1148 int rtn;
1149
1150 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1151 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1152 SCSI_SENSE_VALID(scmd))
1153 continue;
1154
1155 shost = scmd->device->host;
1156 if (scsi_host_eh_past_deadline(shost)) {
1157 SCSI_LOG_ERROR_RECOVERY(3,
1158 shost_printk(KERN_INFO, shost,
1159 "skip %s, past eh deadline\n",
1160 __func__));
1161 break;
1162 }
1163 if (status_byte(scmd->result) != CHECK_CONDITION)
1164 /*
1165 * don't request sense if there's no check condition
1166 * status because the error we're processing isn't one
1167 * that has a sense code (and some devices get
1168 * confused by sense requests out of the blue)
1169 */
1170 continue;
1171
1172 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1173 "%s: requesting sense\n",
1174 current->comm));
1175 rtn = scsi_request_sense(scmd);
1176 if (rtn != SUCCESS)
1177 continue;
1178
1179 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1180 " result %x\n", scmd,
1181 scmd->result));
1182 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1183
1184 rtn = scsi_decide_disposition(scmd);
1185
1186 /*
1187 * if the result was normal, then just pass it along to the
1188 * upper level.
1189 */
1190 if (rtn == SUCCESS)
1191 /* we don't want this command reissued, just
1192 * finished with the sense data, so set
1193 * retries to the max allowed to ensure it
1194 * won't get reissued */
1195 scmd->retries = scmd->allowed;
1196 else if (rtn != NEEDS_RETRY)
1197 continue;
1198
1199 scsi_eh_finish_cmd(scmd, done_q);
1200 }
1201
1202 return list_empty(work_q);
1203}
1204EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1205
1206/**
1207 * scsi_eh_tur - Send TUR to device.
1208 * @scmd: &scsi_cmnd to send TUR
1209 *
1210 * Return value:
1211 * 0 - Device is ready. 1 - Device NOT ready.
1212 */
1213static int scsi_eh_tur(struct scsi_cmnd *scmd)
1214{
1215 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1216 int retry_cnt = 1, rtn;
1217
1218retry_tur:
1219 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1220 scmd->device->eh_timeout, 0);
1221
1222 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
1223 __func__, scmd, rtn));
1224
1225 switch (rtn) {
1226 case NEEDS_RETRY:
1227 if (retry_cnt--)
1228 goto retry_tur;
1229 /*FALLTHRU*/
1230 case SUCCESS:
1231 return 0;
1232 default:
1233 return 1;
1234 }
1235}
1236
1237/**
1238 * scsi_eh_test_devices - check if devices are responding from error recovery.
1239 * @cmd_list: scsi commands in error recovery.
1240 * @work_q: queue for commands which still need more error recovery
1241 * @done_q: queue for commands which are finished
1242 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1243 *
1244 * Decription:
1245 * Tests if devices are in a working state. Commands to devices now in
1246 * a working state are sent to the done_q while commands to devices which
1247 * are still failing to respond are returned to the work_q for more
1248 * processing.
1249 **/
1250static int scsi_eh_test_devices(struct list_head *cmd_list,
1251 struct list_head *work_q,
1252 struct list_head *done_q, int try_stu)
1253{
1254 struct scsi_cmnd *scmd, *next;
1255 struct scsi_device *sdev;
1256 int finish_cmds;
1257
1258 while (!list_empty(cmd_list)) {
1259 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1260 sdev = scmd->device;
1261
1262 if (!try_stu) {
1263 if (scsi_host_eh_past_deadline(sdev->host)) {
1264 /* Push items back onto work_q */
1265 list_splice_init(cmd_list, work_q);
1266 SCSI_LOG_ERROR_RECOVERY(3,
1267 shost_printk(KERN_INFO, sdev->host,
1268 "skip %s, past eh deadline",
1269 __func__));
1270 break;
1271 }
1272 }
1273
1274 finish_cmds = !scsi_device_online(scmd->device) ||
1275 (try_stu && !scsi_eh_try_stu(scmd) &&
1276 !scsi_eh_tur(scmd)) ||
1277 !scsi_eh_tur(scmd);
1278
1279 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1280 if (scmd->device == sdev) {
1281 if (finish_cmds &&
1282 (try_stu ||
1283 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1284 scsi_eh_finish_cmd(scmd, done_q);
1285 else
1286 list_move_tail(&scmd->eh_entry, work_q);
1287 }
1288 }
1289 return list_empty(work_q);
1290}
1291
1292
1293/**
1294 * scsi_eh_abort_cmds - abort pending commands.
1295 * @work_q: &list_head for pending commands.
1296 * @done_q: &list_head for processed commands.
1297 *
1298 * Decription:
1299 * Try and see whether or not it makes sense to try and abort the
1300 * running command. This only works out to be the case if we have one
1301 * command that has timed out. If the command simply failed, it makes
1302 * no sense to try and abort the command, since as far as the shost
1303 * adapter is concerned, it isn't running.
1304 */
1305static int scsi_eh_abort_cmds(struct list_head *work_q,
1306 struct list_head *done_q)
1307{
1308 struct scsi_cmnd *scmd, *next;
1309 LIST_HEAD(check_list);
1310 int rtn;
1311 struct Scsi_Host *shost;
1312
1313 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1314 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1315 continue;
1316 shost = scmd->device->host;
1317 if (scsi_host_eh_past_deadline(shost)) {
1318 list_splice_init(&check_list, work_q);
1319 SCSI_LOG_ERROR_RECOVERY(3,
1320 shost_printk(KERN_INFO, shost,
1321 "skip %s, past eh deadline\n",
1322 __func__));
1323 return list_empty(work_q);
1324 }
1325 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1326 "0x%p\n", current->comm,
1327 scmd));
1328 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1329 if (rtn == FAILED) {
1330 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1331 " cmd failed:"
1332 "0x%p\n",
1333 current->comm,
1334 scmd));
1335 list_splice_init(&check_list, work_q);
1336 return list_empty(work_q);
1337 }
1338 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1339 if (rtn == FAST_IO_FAIL)
1340 scsi_eh_finish_cmd(scmd, done_q);
1341 else
1342 list_move_tail(&scmd->eh_entry, &check_list);
1343 }
1344
1345 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1346}
1347
1348/**
1349 * scsi_eh_try_stu - Send START_UNIT to device.
1350 * @scmd: &scsi_cmnd to send START_UNIT
1351 *
1352 * Return value:
1353 * 0 - Device is ready. 1 - Device NOT ready.
1354 */
1355static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1356{
1357 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1358
1359 if (scmd->device->allow_restart) {
1360 int i, rtn = NEEDS_RETRY;
1361
1362 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1363 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1364
1365 if (rtn == SUCCESS)
1366 return 0;
1367 }
1368
1369 return 1;
1370}
1371
1372 /**
1373 * scsi_eh_stu - send START_UNIT if needed
1374 * @shost: &scsi host being recovered.
1375 * @work_q: &list_head for pending commands.
1376 * @done_q: &list_head for processed commands.
1377 *
1378 * Notes:
1379 * If commands are failing due to not ready, initializing command required,
1380 * try revalidating the device, which will end up sending a start unit.
1381 */
1382static int scsi_eh_stu(struct Scsi_Host *shost,
1383 struct list_head *work_q,
1384 struct list_head *done_q)
1385{
1386 struct scsi_cmnd *scmd, *stu_scmd, *next;
1387 struct scsi_device *sdev;
1388
1389 shost_for_each_device(sdev, shost) {
1390 if (scsi_host_eh_past_deadline(shost)) {
1391 SCSI_LOG_ERROR_RECOVERY(3,
1392 shost_printk(KERN_INFO, shost,
1393 "skip %s, past eh deadline\n",
1394 __func__));
1395 break;
1396 }
1397 stu_scmd = NULL;
1398 list_for_each_entry(scmd, work_q, eh_entry)
1399 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1400 scsi_check_sense(scmd) == FAILED ) {
1401 stu_scmd = scmd;
1402 break;
1403 }
1404
1405 if (!stu_scmd)
1406 continue;
1407
1408 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1409 " 0x%p\n", current->comm, sdev));
1410
1411 if (!scsi_eh_try_stu(stu_scmd)) {
1412 if (!scsi_device_online(sdev) ||
1413 !scsi_eh_tur(stu_scmd)) {
1414 list_for_each_entry_safe(scmd, next,
1415 work_q, eh_entry) {
1416 if (scmd->device == sdev &&
1417 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1418 scsi_eh_finish_cmd(scmd, done_q);
1419 }
1420 }
1421 } else {
1422 SCSI_LOG_ERROR_RECOVERY(3,
1423 printk("%s: START_UNIT failed to sdev:"
1424 " 0x%p\n", current->comm, sdev));
1425 }
1426 }
1427
1428 return list_empty(work_q);
1429}
1430
1431
1432/**
1433 * scsi_eh_bus_device_reset - send bdr if needed
1434 * @shost: scsi host being recovered.
1435 * @work_q: &list_head for pending commands.
1436 * @done_q: &list_head for processed commands.
1437 *
1438 * Notes:
1439 * Try a bus device reset. Still, look to see whether we have multiple
1440 * devices that are jammed or not - if we have multiple devices, it
1441 * makes no sense to try bus_device_reset - we really would need to try
1442 * a bus_reset instead.
1443 */
1444static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1445 struct list_head *work_q,
1446 struct list_head *done_q)
1447{
1448 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1449 struct scsi_device *sdev;
1450 int rtn;
1451
1452 shost_for_each_device(sdev, shost) {
1453 if (scsi_host_eh_past_deadline(shost)) {
1454 SCSI_LOG_ERROR_RECOVERY(3,
1455 shost_printk(KERN_INFO, shost,
1456 "skip %s, past eh deadline\n",
1457 __func__));
1458 break;
1459 }
1460 bdr_scmd = NULL;
1461 list_for_each_entry(scmd, work_q, eh_entry)
1462 if (scmd->device == sdev) {
1463 bdr_scmd = scmd;
1464 break;
1465 }
1466
1467 if (!bdr_scmd)
1468 continue;
1469
1470 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1471 " 0x%p\n", current->comm,
1472 sdev));
1473 rtn = scsi_try_bus_device_reset(bdr_scmd);
1474 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1475 if (!scsi_device_online(sdev) ||
1476 rtn == FAST_IO_FAIL ||
1477 !scsi_eh_tur(bdr_scmd)) {
1478 list_for_each_entry_safe(scmd, next,
1479 work_q, eh_entry) {
1480 if (scmd->device == sdev &&
1481 scsi_eh_action(scmd, rtn) != FAILED)
1482 scsi_eh_finish_cmd(scmd,
1483 done_q);
1484 }
1485 }
1486 } else {
1487 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1488 " failed sdev:"
1489 "0x%p\n",
1490 current->comm,
1491 sdev));
1492 }
1493 }
1494
1495 return list_empty(work_q);
1496}
1497
1498/**
1499 * scsi_eh_target_reset - send target reset if needed
1500 * @shost: scsi host being recovered.
1501 * @work_q: &list_head for pending commands.
1502 * @done_q: &list_head for processed commands.
1503 *
1504 * Notes:
1505 * Try a target reset.
1506 */
1507static int scsi_eh_target_reset(struct Scsi_Host *shost,
1508 struct list_head *work_q,
1509 struct list_head *done_q)
1510{
1511 LIST_HEAD(tmp_list);
1512 LIST_HEAD(check_list);
1513
1514 list_splice_init(work_q, &tmp_list);
1515
1516 while (!list_empty(&tmp_list)) {
1517 struct scsi_cmnd *next, *scmd;
1518 int rtn;
1519 unsigned int id;
1520
1521 if (scsi_host_eh_past_deadline(shost)) {
1522 /* push back on work queue for further processing */
1523 list_splice_init(&check_list, work_q);
1524 list_splice_init(&tmp_list, work_q);
1525 SCSI_LOG_ERROR_RECOVERY(3,
1526 shost_printk(KERN_INFO, shost,
1527 "skip %s, past eh deadline\n",
1528 __func__));
1529 return list_empty(work_q);
1530 }
1531
1532 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1533 id = scmd_id(scmd);
1534
1535 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1536 "to target %d\n",
1537 current->comm, id));
1538 rtn = scsi_try_target_reset(scmd);
1539 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1540 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1541 " failed target: "
1542 "%d\n",
1543 current->comm, id));
1544 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1545 if (scmd_id(scmd) != id)
1546 continue;
1547
1548 if (rtn == SUCCESS)
1549 list_move_tail(&scmd->eh_entry, &check_list);
1550 else if (rtn == FAST_IO_FAIL)
1551 scsi_eh_finish_cmd(scmd, done_q);
1552 else
1553 /* push back on work queue for further processing */
1554 list_move(&scmd->eh_entry, work_q);
1555 }
1556 }
1557
1558 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1559}
1560
1561/**
1562 * scsi_eh_bus_reset - send a bus reset
1563 * @shost: &scsi host being recovered.
1564 * @work_q: &list_head for pending commands.
1565 * @done_q: &list_head for processed commands.
1566 */
1567static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1568 struct list_head *work_q,
1569 struct list_head *done_q)
1570{
1571 struct scsi_cmnd *scmd, *chan_scmd, *next;
1572 LIST_HEAD(check_list);
1573 unsigned int channel;
1574 int rtn;
1575
1576 /*
1577 * we really want to loop over the various channels, and do this on
1578 * a channel by channel basis. we should also check to see if any
1579 * of the failed commands are on soft_reset devices, and if so, skip
1580 * the reset.
1581 */
1582
1583 for (channel = 0; channel <= shost->max_channel; channel++) {
1584 if (scsi_host_eh_past_deadline(shost)) {
1585 list_splice_init(&check_list, work_q);
1586 SCSI_LOG_ERROR_RECOVERY(3,
1587 shost_printk(KERN_INFO, shost,
1588 "skip %s, past eh deadline\n",
1589 __func__));
1590 return list_empty(work_q);
1591 }
1592
1593 chan_scmd = NULL;
1594 list_for_each_entry(scmd, work_q, eh_entry) {
1595 if (channel == scmd_channel(scmd)) {
1596 chan_scmd = scmd;
1597 break;
1598 /*
1599 * FIXME add back in some support for
1600 * soft_reset devices.
1601 */
1602 }
1603 }
1604
1605 if (!chan_scmd)
1606 continue;
1607 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1608 " %d\n", current->comm,
1609 channel));
1610 rtn = scsi_try_bus_reset(chan_scmd);
1611 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1612 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1613 if (channel == scmd_channel(scmd)) {
1614 if (rtn == FAST_IO_FAIL)
1615 scsi_eh_finish_cmd(scmd,
1616 done_q);
1617 else
1618 list_move_tail(&scmd->eh_entry,
1619 &check_list);
1620 }
1621 }
1622 } else {
1623 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1624 " failed chan: %d\n",
1625 current->comm,
1626 channel));
1627 }
1628 }
1629 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1630}
1631
1632/**
1633 * scsi_eh_host_reset - send a host reset
1634 * @work_q: list_head for processed commands.
1635 * @done_q: list_head for processed commands.
1636 */
1637static int scsi_eh_host_reset(struct list_head *work_q,
1638 struct list_head *done_q)
1639{
1640 struct scsi_cmnd *scmd, *next;
1641 LIST_HEAD(check_list);
1642 int rtn;
1643
1644 if (!list_empty(work_q)) {
1645 scmd = list_entry(work_q->next,
1646 struct scsi_cmnd, eh_entry);
1647
1648 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1649 , current->comm));
1650
1651 rtn = scsi_try_host_reset(scmd);
1652 if (rtn == SUCCESS) {
1653 list_splice_init(work_q, &check_list);
1654 } else if (rtn == FAST_IO_FAIL) {
1655 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1656 scsi_eh_finish_cmd(scmd, done_q);
1657 }
1658 } else {
1659 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1660 " failed\n",
1661 current->comm));
1662 }
1663 }
1664 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1665}
1666
1667/**
1668 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1669 * @work_q: list_head for processed commands.
1670 * @done_q: list_head for processed commands.
1671 */
1672static void scsi_eh_offline_sdevs(struct list_head *work_q,
1673 struct list_head *done_q)
1674{
1675 struct scsi_cmnd *scmd, *next;
1676
1677 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1678 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1679 "not ready after error recovery\n");
1680 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1681 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1682 /*
1683 * FIXME: Handle lost cmds.
1684 */
1685 }
1686 scsi_eh_finish_cmd(scmd, done_q);
1687 }
1688 return;
1689}
1690
1691/**
1692 * scsi_noretry_cmd - determine if command should be failed fast
1693 * @scmd: SCSI cmd to examine.
1694 */
1695int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1696{
1697 switch (host_byte(scmd->result)) {
1698 case DID_OK:
1699 break;
1700 case DID_TIME_OUT:
1701 goto check_type;
1702 case DID_BUS_BUSY:
1703 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1704 case DID_PARITY:
1705 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1706 case DID_ERROR:
1707 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1708 status_byte(scmd->result) == RESERVATION_CONFLICT)
1709 return 0;
1710 /* fall through */
1711 case DID_SOFT_ERROR:
1712 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1713 }
1714
1715 if (status_byte(scmd->result) != CHECK_CONDITION)
1716 return 0;
1717
1718check_type:
1719 /*
1720 * assume caller has checked sense and determined
1721 * the check condition was retryable.
1722 */
1723 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1724 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1725 return 1;
1726 else
1727 return 0;
1728}
1729
1730/**
1731 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1732 * @scmd: SCSI cmd to examine.
1733 *
1734 * Notes:
1735 * This is *only* called when we are examining the status after sending
1736 * out the actual data command. any commands that are queued for error
1737 * recovery (e.g. test_unit_ready) do *not* come through here.
1738 *
1739 * When this routine returns failed, it means the error handler thread
1740 * is woken. In cases where the error code indicates an error that
1741 * doesn't require the error handler read (i.e. we don't need to
1742 * abort/reset), this function should return SUCCESS.
1743 */
1744int scsi_decide_disposition(struct scsi_cmnd *scmd)
1745{
1746 int rtn;
1747
1748 /*
1749 * if the device is offline, then we clearly just pass the result back
1750 * up to the top level.
1751 */
1752 if (!scsi_device_online(scmd->device)) {
1753 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1754 " as SUCCESS\n",
1755 __func__));
1756 return SUCCESS;
1757 }
1758
1759 /*
1760 * first check the host byte, to see if there is anything in there
1761 * that would indicate what we need to do.
1762 */
1763 switch (host_byte(scmd->result)) {
1764 case DID_PASSTHROUGH:
1765 /*
1766 * no matter what, pass this through to the upper layer.
1767 * nuke this special code so that it looks like we are saying
1768 * did_ok.
1769 */
1770 scmd->result &= 0xff00ffff;
1771 return SUCCESS;
1772 case DID_OK:
1773 /*
1774 * looks good. drop through, and check the next byte.
1775 */
1776 break;
1777 case DID_ABORT:
1778 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1779 scmd->result |= DID_TIME_OUT << 16;
1780 return SUCCESS;
1781 }
1782 case DID_NO_CONNECT:
1783 case DID_BAD_TARGET:
1784 /*
1785 * note - this means that we just report the status back
1786 * to the top level driver, not that we actually think
1787 * that it indicates SUCCESS.
1788 */
1789 return SUCCESS;
1790 /*
1791 * when the low level driver returns did_soft_error,
1792 * it is responsible for keeping an internal retry counter
1793 * in order to avoid endless loops (db)
1794 *
1795 * actually this is a bug in this function here. we should
1796 * be mindful of the maximum number of retries specified
1797 * and not get stuck in a loop.
1798 */
1799 case DID_SOFT_ERROR:
1800 goto maybe_retry;
1801 case DID_IMM_RETRY:
1802 return NEEDS_RETRY;
1803
1804 case DID_REQUEUE:
1805 return ADD_TO_MLQUEUE;
1806 case DID_TRANSPORT_DISRUPTED:
1807 /*
1808 * LLD/transport was disrupted during processing of the IO.
1809 * The transport class is now blocked/blocking,
1810 * and the transport will decide what to do with the IO
1811 * based on its timers and recovery capablilities if
1812 * there are enough retries.
1813 */
1814 goto maybe_retry;
1815 case DID_TRANSPORT_FAILFAST:
1816 /*
1817 * The transport decided to failfast the IO (most likely
1818 * the fast io fail tmo fired), so send IO directly upwards.
1819 */
1820 return SUCCESS;
1821 case DID_ERROR:
1822 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1823 status_byte(scmd->result) == RESERVATION_CONFLICT)
1824 /*
1825 * execute reservation conflict processing code
1826 * lower down
1827 */
1828 break;
1829 /* fallthrough */
1830 case DID_BUS_BUSY:
1831 case DID_PARITY:
1832 goto maybe_retry;
1833 case DID_TIME_OUT:
1834 /*
1835 * when we scan the bus, we get timeout messages for
1836 * these commands if there is no device available.
1837 * other hosts report did_no_connect for the same thing.
1838 */
1839 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1840 scmd->cmnd[0] == INQUIRY)) {
1841 return SUCCESS;
1842 } else {
1843 return FAILED;
1844 }
1845 case DID_RESET:
1846 return SUCCESS;
1847 default:
1848 return FAILED;
1849 }
1850
1851 /*
1852 * next, check the message byte.
1853 */
1854 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1855 return FAILED;
1856
1857 /*
1858 * check the status byte to see if this indicates anything special.
1859 */
1860 switch (status_byte(scmd->result)) {
1861 case QUEUE_FULL:
1862 scsi_handle_queue_full(scmd->device);
1863 /*
1864 * the case of trying to send too many commands to a
1865 * tagged queueing device.
1866 */
1867 case BUSY:
1868 /*
1869 * device can't talk to us at the moment. Should only
1870 * occur (SAM-3) when the task queue is empty, so will cause
1871 * the empty queue handling to trigger a stall in the
1872 * device.
1873 */
1874 return ADD_TO_MLQUEUE;
1875 case GOOD:
1876 if (scmd->cmnd[0] == REPORT_LUNS)
1877 scmd->device->sdev_target->expecting_lun_change = 0;
1878 scsi_handle_queue_ramp_up(scmd->device);
1879 case COMMAND_TERMINATED:
1880 return SUCCESS;
1881 case TASK_ABORTED:
1882 goto maybe_retry;
1883 case CHECK_CONDITION:
1884 rtn = scsi_check_sense(scmd);
1885 if (rtn == NEEDS_RETRY)
1886 goto maybe_retry;
1887 /* if rtn == FAILED, we have no sense information;
1888 * returning FAILED will wake the error handler thread
1889 * to collect the sense and redo the decide
1890 * disposition */
1891 return rtn;
1892 case CONDITION_GOOD:
1893 case INTERMEDIATE_GOOD:
1894 case INTERMEDIATE_C_GOOD:
1895 case ACA_ACTIVE:
1896 /*
1897 * who knows? FIXME(eric)
1898 */
1899 return SUCCESS;
1900
1901 case RESERVATION_CONFLICT:
1902 sdev_printk(KERN_INFO, scmd->device,
1903 "reservation conflict\n");
1904 set_host_byte(scmd, DID_NEXUS_FAILURE);
1905 return SUCCESS; /* causes immediate i/o error */
1906 default:
1907 return FAILED;
1908 }
1909 return FAILED;
1910
1911 maybe_retry:
1912
1913 /* we requeue for retry because the error was retryable, and
1914 * the request was not marked fast fail. Note that above,
1915 * even if the request is marked fast fail, we still requeue
1916 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1917 if ((++scmd->retries) <= scmd->allowed
1918 && !scsi_noretry_cmd(scmd)) {
1919 return NEEDS_RETRY;
1920 } else {
1921 /*
1922 * no more retries - report this one back to upper level.
1923 */
1924 return SUCCESS;
1925 }
1926}
1927
1928static void eh_lock_door_done(struct request *req, int uptodate)
1929{
1930 __blk_put_request(req->q, req);
1931}
1932
1933/**
1934 * scsi_eh_lock_door - Prevent medium removal for the specified device
1935 * @sdev: SCSI device to prevent medium removal
1936 *
1937 * Locking:
1938 * We must be called from process context.
1939 *
1940 * Notes:
1941 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1942 * head of the devices request queue, and continue.
1943 */
1944static void scsi_eh_lock_door(struct scsi_device *sdev)
1945{
1946 struct request *req;
1947
1948 /*
1949 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1950 * request becomes available
1951 */
1952 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1953
1954 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1955 req->cmd[1] = 0;
1956 req->cmd[2] = 0;
1957 req->cmd[3] = 0;
1958 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1959 req->cmd[5] = 0;
1960
1961 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1962
1963 req->cmd_type = REQ_TYPE_BLOCK_PC;
1964 req->cmd_flags |= REQ_QUIET;
1965 req->timeout = 10 * HZ;
1966 req->retries = 5;
1967
1968 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1969}
1970
1971/**
1972 * scsi_restart_operations - restart io operations to the specified host.
1973 * @shost: Host we are restarting.
1974 *
1975 * Notes:
1976 * When we entered the error handler, we blocked all further i/o to
1977 * this device. we need to 'reverse' this process.
1978 */
1979static void scsi_restart_operations(struct Scsi_Host *shost)
1980{
1981 struct scsi_device *sdev;
1982 unsigned long flags;
1983
1984 /*
1985 * If the door was locked, we need to insert a door lock request
1986 * onto the head of the SCSI request queue for the device. There
1987 * is no point trying to lock the door of an off-line device.
1988 */
1989 shost_for_each_device(sdev, shost) {
1990 if (scsi_device_online(sdev) && sdev->locked)
1991 scsi_eh_lock_door(sdev);
1992 }
1993
1994 /*
1995 * next free up anything directly waiting upon the host. this
1996 * will be requests for character device operations, and also for
1997 * ioctls to queued block devices.
1998 */
1999 SCSI_LOG_ERROR_RECOVERY(3,
2000 printk("scsi_eh_%d waking up host to restart\n",
2001 shost->host_no));
2002
2003 spin_lock_irqsave(shost->host_lock, flags);
2004 if (scsi_host_set_state(shost, SHOST_RUNNING))
2005 if (scsi_host_set_state(shost, SHOST_CANCEL))
2006 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2007 spin_unlock_irqrestore(shost->host_lock, flags);
2008
2009 wake_up(&shost->host_wait);
2010
2011 /*
2012 * finally we need to re-initiate requests that may be pending. we will
2013 * have had everything blocked while error handling is taking place, and
2014 * now that error recovery is done, we will need to ensure that these
2015 * requests are started.
2016 */
2017 scsi_run_host_queues(shost);
2018
2019 /*
2020 * if eh is active and host_eh_scheduled is pending we need to re-run
2021 * recovery. we do this check after scsi_run_host_queues() to allow
2022 * everything pent up since the last eh run a chance to make forward
2023 * progress before we sync again. Either we'll immediately re-run
2024 * recovery or scsi_device_unbusy() will wake us again when these
2025 * pending commands complete.
2026 */
2027 spin_lock_irqsave(shost->host_lock, flags);
2028 if (shost->host_eh_scheduled)
2029 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2030 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2031 spin_unlock_irqrestore(shost->host_lock, flags);
2032}
2033
2034/**
2035 * scsi_eh_ready_devs - check device ready state and recover if not.
2036 * @shost: host to be recovered.
2037 * @work_q: &list_head for pending commands.
2038 * @done_q: &list_head for processed commands.
2039 */
2040void scsi_eh_ready_devs(struct Scsi_Host *shost,
2041 struct list_head *work_q,
2042 struct list_head *done_q)
2043{
2044 if (!scsi_eh_stu(shost, work_q, done_q))
2045 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2046 if (!scsi_eh_target_reset(shost, work_q, done_q))
2047 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2048 if (!scsi_eh_host_reset(work_q, done_q))
2049 scsi_eh_offline_sdevs(work_q,
2050 done_q);
2051}
2052EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2053
2054/**
2055 * scsi_eh_flush_done_q - finish processed commands or retry them.
2056 * @done_q: list_head of processed commands.
2057 */
2058void scsi_eh_flush_done_q(struct list_head *done_q)
2059{
2060 struct scsi_cmnd *scmd, *next;
2061
2062 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2063 list_del_init(&scmd->eh_entry);
2064 if (scsi_device_online(scmd->device) &&
2065 !scsi_noretry_cmd(scmd) &&
2066 (++scmd->retries <= scmd->allowed)) {
2067 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
2068 " retry cmd: %p\n",
2069 current->comm,
2070 scmd));
2071 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2072 } else {
2073 /*
2074 * If just we got sense for the device (called
2075 * scsi_eh_get_sense), scmd->result is already
2076 * set, do not set DRIVER_TIMEOUT.
2077 */
2078 if (!scmd->result)
2079 scmd->result |= (DRIVER_TIMEOUT << 24);
2080 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
2081 " cmd: %p\n",
2082 current->comm, scmd));
2083 scsi_finish_command(scmd);
2084 }
2085 }
2086}
2087EXPORT_SYMBOL(scsi_eh_flush_done_q);
2088
2089/**
2090 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2091 * @shost: Host to unjam.
2092 *
2093 * Notes:
2094 * When we come in here, we *know* that all commands on the bus have
2095 * either completed, failed or timed out. we also know that no further
2096 * commands are being sent to the host, so things are relatively quiet
2097 * and we have freedom to fiddle with things as we wish.
2098 *
2099 * This is only the *default* implementation. it is possible for
2100 * individual drivers to supply their own version of this function, and
2101 * if the maintainer wishes to do this, it is strongly suggested that
2102 * this function be taken as a template and modified. this function
2103 * was designed to correctly handle problems for about 95% of the
2104 * different cases out there, and it should always provide at least a
2105 * reasonable amount of error recovery.
2106 *
2107 * Any command marked 'failed' or 'timeout' must eventually have
2108 * scsi_finish_cmd() called for it. we do all of the retry stuff
2109 * here, so when we restart the host after we return it should have an
2110 * empty queue.
2111 */
2112static void scsi_unjam_host(struct Scsi_Host *shost)
2113{
2114 unsigned long flags;
2115 LIST_HEAD(eh_work_q);
2116 LIST_HEAD(eh_done_q);
2117
2118 spin_lock_irqsave(shost->host_lock, flags);
2119 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2120 spin_unlock_irqrestore(shost->host_lock, flags);
2121
2122 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2123
2124 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2125 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2126 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2127
2128 spin_lock_irqsave(shost->host_lock, flags);
2129 if (shost->eh_deadline != -1)
2130 shost->last_reset = 0;
2131 spin_unlock_irqrestore(shost->host_lock, flags);
2132 scsi_eh_flush_done_q(&eh_done_q);
2133}
2134
2135/**
2136 * scsi_error_handler - SCSI error handler thread
2137 * @data: Host for which we are running.
2138 *
2139 * Notes:
2140 * This is the main error handling loop. This is run as a kernel thread
2141 * for every SCSI host and handles all error handling activity.
2142 */
2143int scsi_error_handler(void *data)
2144{
2145 struct Scsi_Host *shost = data;
2146
2147 /*
2148 * We use TASK_INTERRUPTIBLE so that the thread is not
2149 * counted against the load average as a running process.
2150 * We never actually get interrupted because kthread_run
2151 * disables signal delivery for the created thread.
2152 */
2153 while (!kthread_should_stop()) {
2154 set_current_state(TASK_INTERRUPTIBLE);
2155 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2156 shost->host_failed != shost->host_busy) {
2157 SCSI_LOG_ERROR_RECOVERY(1,
2158 printk("scsi_eh_%d: sleeping\n",
2159 shost->host_no));
2160 schedule();
2161 continue;
2162 }
2163
2164 __set_current_state(TASK_RUNNING);
2165 SCSI_LOG_ERROR_RECOVERY(1,
2166 printk("scsi_eh_%d: waking up %d/%d/%d\n",
2167 shost->host_no, shost->host_eh_scheduled,
2168 shost->host_failed, shost->host_busy));
2169
2170 /*
2171 * We have a host that is failing for some reason. Figure out
2172 * what we need to do to get it up and online again (if we can).
2173 * If we fail, we end up taking the thing offline.
2174 */
2175 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2176 SCSI_LOG_ERROR_RECOVERY(1,
2177 printk(KERN_ERR "Error handler scsi_eh_%d "
2178 "unable to autoresume\n",
2179 shost->host_no));
2180 continue;
2181 }
2182
2183 if (shost->transportt->eh_strategy_handler)
2184 shost->transportt->eh_strategy_handler(shost);
2185 else
2186 scsi_unjam_host(shost);
2187
2188 /*
2189 * Note - if the above fails completely, the action is to take
2190 * individual devices offline and flush the queue of any
2191 * outstanding requests that may have been pending. When we
2192 * restart, we restart any I/O to any other devices on the bus
2193 * which are still online.
2194 */
2195 scsi_restart_operations(shost);
2196 if (!shost->eh_noresume)
2197 scsi_autopm_put_host(shost);
2198 }
2199 __set_current_state(TASK_RUNNING);
2200
2201 SCSI_LOG_ERROR_RECOVERY(1,
2202 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
2203 shost->ehandler = NULL;
2204 return 0;
2205}
2206
2207/*
2208 * Function: scsi_report_bus_reset()
2209 *
2210 * Purpose: Utility function used by low-level drivers to report that
2211 * they have observed a bus reset on the bus being handled.
2212 *
2213 * Arguments: shost - Host in question
2214 * channel - channel on which reset was observed.
2215 *
2216 * Returns: Nothing
2217 *
2218 * Lock status: Host lock must be held.
2219 *
2220 * Notes: This only needs to be called if the reset is one which
2221 * originates from an unknown location. Resets originated
2222 * by the mid-level itself don't need to call this, but there
2223 * should be no harm.
2224 *
2225 * The main purpose of this is to make sure that a CHECK_CONDITION
2226 * is properly treated.
2227 */
2228void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2229{
2230 struct scsi_device *sdev;
2231
2232 __shost_for_each_device(sdev, shost) {
2233 if (channel == sdev_channel(sdev))
2234 __scsi_report_device_reset(sdev, NULL);
2235 }
2236}
2237EXPORT_SYMBOL(scsi_report_bus_reset);
2238
2239/*
2240 * Function: scsi_report_device_reset()
2241 *
2242 * Purpose: Utility function used by low-level drivers to report that
2243 * they have observed a device reset on the device being handled.
2244 *
2245 * Arguments: shost - Host in question
2246 * channel - channel on which reset was observed
2247 * target - target on which reset was observed
2248 *
2249 * Returns: Nothing
2250 *
2251 * Lock status: Host lock must be held
2252 *
2253 * Notes: This only needs to be called if the reset is one which
2254 * originates from an unknown location. Resets originated
2255 * by the mid-level itself don't need to call this, but there
2256 * should be no harm.
2257 *
2258 * The main purpose of this is to make sure that a CHECK_CONDITION
2259 * is properly treated.
2260 */
2261void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2262{
2263 struct scsi_device *sdev;
2264
2265 __shost_for_each_device(sdev, shost) {
2266 if (channel == sdev_channel(sdev) &&
2267 target == sdev_id(sdev))
2268 __scsi_report_device_reset(sdev, NULL);
2269 }
2270}
2271EXPORT_SYMBOL(scsi_report_device_reset);
2272
2273static void
2274scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2275{
2276}
2277
2278/*
2279 * Function: scsi_reset_provider
2280 *
2281 * Purpose: Send requested reset to a bus or device at any phase.
2282 *
2283 * Arguments: device - device to send reset to
2284 * flag - reset type (see scsi.h)
2285 *
2286 * Returns: SUCCESS/FAILURE.
2287 *
2288 * Notes: This is used by the SCSI Generic driver to provide
2289 * Bus/Device reset capability.
2290 */
2291int
2292scsi_reset_provider(struct scsi_device *dev, int flag)
2293{
2294 struct scsi_cmnd *scmd;
2295 struct Scsi_Host *shost = dev->host;
2296 struct request req;
2297 unsigned long flags;
2298 int rtn;
2299
2300 if (scsi_autopm_get_host(shost) < 0)
2301 return FAILED;
2302
2303 if (!get_device(&dev->sdev_gendev)) {
2304 rtn = FAILED;
2305 goto out_put_autopm_host;
2306 }
2307
2308 scmd = scsi_get_command(dev, GFP_KERNEL);
2309 blk_rq_init(NULL, &req);
2310 scmd->request = &req;
2311
2312 scmd->cmnd = req.cmd;
2313
2314 scmd->scsi_done = scsi_reset_provider_done_command;
2315 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2316
2317 scmd->cmd_len = 0;
2318
2319 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2320
2321 spin_lock_irqsave(shost->host_lock, flags);
2322 shost->tmf_in_progress = 1;
2323 spin_unlock_irqrestore(shost->host_lock, flags);
2324
2325 switch (flag) {
2326 case SCSI_TRY_RESET_DEVICE:
2327 rtn = scsi_try_bus_device_reset(scmd);
2328 if (rtn == SUCCESS)
2329 break;
2330 /* FALLTHROUGH */
2331 case SCSI_TRY_RESET_TARGET:
2332 rtn = scsi_try_target_reset(scmd);
2333 if (rtn == SUCCESS)
2334 break;
2335 /* FALLTHROUGH */
2336 case SCSI_TRY_RESET_BUS:
2337 rtn = scsi_try_bus_reset(scmd);
2338 if (rtn == SUCCESS)
2339 break;
2340 /* FALLTHROUGH */
2341 case SCSI_TRY_RESET_HOST:
2342 rtn = scsi_try_host_reset(scmd);
2343 break;
2344 default:
2345 rtn = FAILED;
2346 }
2347
2348 spin_lock_irqsave(shost->host_lock, flags);
2349 shost->tmf_in_progress = 0;
2350 spin_unlock_irqrestore(shost->host_lock, flags);
2351
2352 /*
2353 * be sure to wake up anyone who was sleeping or had their queue
2354 * suspended while we performed the TMF.
2355 */
2356 SCSI_LOG_ERROR_RECOVERY(3,
2357 printk("%s: waking up host to restart after TMF\n",
2358 __func__));
2359
2360 wake_up(&shost->host_wait);
2361
2362 scsi_run_host_queues(shost);
2363
2364 scsi_next_command(scmd);
2365out_put_autopm_host:
2366 scsi_autopm_put_host(shost);
2367 return rtn;
2368}
2369EXPORT_SYMBOL(scsi_reset_provider);
2370
2371/**
2372 * scsi_normalize_sense - normalize main elements from either fixed or
2373 * descriptor sense data format into a common format.
2374 *
2375 * @sense_buffer: byte array containing sense data returned by device
2376 * @sb_len: number of valid bytes in sense_buffer
2377 * @sshdr: pointer to instance of structure that common
2378 * elements are written to.
2379 *
2380 * Notes:
2381 * The "main elements" from sense data are: response_code, sense_key,
2382 * asc, ascq and additional_length (only for descriptor format).
2383 *
2384 * Typically this function can be called after a device has
2385 * responded to a SCSI command with the CHECK_CONDITION status.
2386 *
2387 * Return value:
2388 * 1 if valid sense data information found, else 0;
2389 */
2390int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2391 struct scsi_sense_hdr *sshdr)
2392{
2393 if (!sense_buffer || !sb_len)
2394 return 0;
2395
2396 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2397
2398 sshdr->response_code = (sense_buffer[0] & 0x7f);
2399
2400 if (!scsi_sense_valid(sshdr))
2401 return 0;
2402
2403 if (sshdr->response_code >= 0x72) {
2404 /*
2405 * descriptor format
2406 */
2407 if (sb_len > 1)
2408 sshdr->sense_key = (sense_buffer[1] & 0xf);
2409 if (sb_len > 2)
2410 sshdr->asc = sense_buffer[2];
2411 if (sb_len > 3)
2412 sshdr->ascq = sense_buffer[3];
2413 if (sb_len > 7)
2414 sshdr->additional_length = sense_buffer[7];
2415 } else {
2416 /*
2417 * fixed format
2418 */
2419 if (sb_len > 2)
2420 sshdr->sense_key = (sense_buffer[2] & 0xf);
2421 if (sb_len > 7) {
2422 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2423 sb_len : (sense_buffer[7] + 8);
2424 if (sb_len > 12)
2425 sshdr->asc = sense_buffer[12];
2426 if (sb_len > 13)
2427 sshdr->ascq = sense_buffer[13];
2428 }
2429 }
2430
2431 return 1;
2432}
2433EXPORT_SYMBOL(scsi_normalize_sense);
2434
2435int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2436 struct scsi_sense_hdr *sshdr)
2437{
2438 return scsi_normalize_sense(cmd->sense_buffer,
2439 SCSI_SENSE_BUFFERSIZE, sshdr);
2440}
2441EXPORT_SYMBOL(scsi_command_normalize_sense);
2442
2443/**
2444 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2445 * @sense_buffer: byte array of descriptor format sense data
2446 * @sb_len: number of valid bytes in sense_buffer
2447 * @desc_type: value of descriptor type to find
2448 * (e.g. 0 -> information)
2449 *
2450 * Notes:
2451 * only valid when sense data is in descriptor format
2452 *
2453 * Return value:
2454 * pointer to start of (first) descriptor if found else NULL
2455 */
2456const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2457 int desc_type)
2458{
2459 int add_sen_len, add_len, desc_len, k;
2460 const u8 * descp;
2461
2462 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2463 return NULL;
2464 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2465 return NULL;
2466 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2467 add_sen_len : (sb_len - 8);
2468 descp = &sense_buffer[8];
2469 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2470 descp += desc_len;
2471 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2472 desc_len = add_len + 2;
2473 if (descp[0] == desc_type)
2474 return descp;
2475 if (add_len < 0) // short descriptor ??
2476 break;
2477 }
2478 return NULL;
2479}
2480EXPORT_SYMBOL(scsi_sense_desc_find);
2481
2482/**
2483 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2484 * @sense_buffer: byte array of sense data
2485 * @sb_len: number of valid bytes in sense_buffer
2486 * @info_out: pointer to 64 integer where 8 or 4 byte information
2487 * field will be placed if found.
2488 *
2489 * Return value:
2490 * 1 if information field found, 0 if not found.
2491 */
2492int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2493 u64 * info_out)
2494{
2495 int j;
2496 const u8 * ucp;
2497 u64 ull;
2498
2499 if (sb_len < 7)
2500 return 0;
2501 switch (sense_buffer[0] & 0x7f) {
2502 case 0x70:
2503 case 0x71:
2504 if (sense_buffer[0] & 0x80) {
2505 *info_out = (sense_buffer[3] << 24) +
2506 (sense_buffer[4] << 16) +
2507 (sense_buffer[5] << 8) + sense_buffer[6];
2508 return 1;
2509 } else
2510 return 0;
2511 case 0x72:
2512 case 0x73:
2513 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2514 0 /* info desc */);
2515 if (ucp && (0xa == ucp[1])) {
2516 ull = 0;
2517 for (j = 0; j < 8; ++j) {
2518 if (j > 0)
2519 ull <<= 8;
2520 ull |= ucp[4 + j];
2521 }
2522 *info_out = ull;
2523 return 1;
2524 } else
2525 return 0;
2526 default:
2527 return 0;
2528 }
2529}
2530EXPORT_SYMBOL(scsi_get_sense_info_fld);
2531
2532/**
2533 * scsi_build_sense_buffer - build sense data in a buffer
2534 * @desc: Sense format (non zero == descriptor format,
2535 * 0 == fixed format)
2536 * @buf: Where to build sense data
2537 * @key: Sense key
2538 * @asc: Additional sense code
2539 * @ascq: Additional sense code qualifier
2540 *
2541 **/
2542void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2543{
2544 if (desc) {
2545 buf[0] = 0x72; /* descriptor, current */
2546 buf[1] = key;
2547 buf[2] = asc;
2548 buf[3] = ascq;
2549 buf[7] = 0;
2550 } else {
2551 buf[0] = 0x70; /* fixed, current */
2552 buf[2] = key;
2553 buf[7] = 0xa;
2554 buf[12] = asc;
2555 buf[13] = ascq;
2556 }
2557}
2558EXPORT_SYMBOL(scsi_build_sense_buffer);
1/*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
3 *
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
8 *
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11 *
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13 * minor cleanups.
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
19#include <linux/gfp.h>
20#include <linux/timer.h>
21#include <linux/string.h>
22#include <linux/kernel.h>
23#include <linux/freezer.h>
24#include <linux/kthread.h>
25#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
28#include <linux/jiffies.h>
29
30#include <scsi/scsi.h>
31#include <scsi/scsi_cmnd.h>
32#include <scsi/scsi_dbg.h>
33#include <scsi/scsi_device.h>
34#include <scsi/scsi_driver.h>
35#include <scsi/scsi_eh.h>
36#include <scsi/scsi_common.h>
37#include <scsi/scsi_transport.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_ioctl.h>
40#include <scsi/scsi_dh.h>
41#include <scsi/sg.h>
42
43#include "scsi_priv.h"
44#include "scsi_logging.h"
45#include "scsi_transport_api.h"
46
47#include <trace/events/scsi.h>
48
49#include <asm/unaligned.h>
50
51static void scsi_eh_done(struct scsi_cmnd *scmd);
52
53/*
54 * These should *probably* be handled by the host itself.
55 * Since it is allowed to sleep, it probably should.
56 */
57#define BUS_RESET_SETTLE_TIME (10)
58#define HOST_RESET_SETTLE_TIME (10)
59
60static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
61static int scsi_try_to_abort_cmd(struct scsi_host_template *,
62 struct scsi_cmnd *);
63
64void scsi_eh_wakeup(struct Scsi_Host *shost)
65{
66 lockdep_assert_held(shost->host_lock);
67
68 if (atomic_read(&shost->host_busy) == shost->host_failed) {
69 trace_scsi_eh_wakeup(shost);
70 wake_up_process(shost->ehandler);
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 "Waking error handler thread\n"));
73 }
74}
75
76/**
77 * scsi_schedule_eh - schedule EH for SCSI host
78 * @shost: SCSI host to invoke error handling on.
79 *
80 * Schedule SCSI EH without scmd.
81 */
82void scsi_schedule_eh(struct Scsi_Host *shost)
83{
84 unsigned long flags;
85
86 spin_lock_irqsave(shost->host_lock, flags);
87
88 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
89 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
90 shost->host_eh_scheduled++;
91 scsi_eh_wakeup(shost);
92 }
93
94 spin_unlock_irqrestore(shost->host_lock, flags);
95}
96EXPORT_SYMBOL_GPL(scsi_schedule_eh);
97
98static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
99{
100 if (!shost->last_reset || shost->eh_deadline == -1)
101 return 0;
102
103 /*
104 * 32bit accesses are guaranteed to be atomic
105 * (on all supported architectures), so instead
106 * of using a spinlock we can as well double check
107 * if eh_deadline has been set to 'off' during the
108 * time_before call.
109 */
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
111 shost->eh_deadline > -1)
112 return 0;
113
114 return 1;
115}
116
117/**
118 * scmd_eh_abort_handler - Handle command aborts
119 * @work: command to be aborted.
120 *
121 * Note: this function must be called only for a command that has timed out.
122 * Because the block layer marks a request as complete before it calls
123 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
124 * timed out do not have any effect. Hence it is safe to call
125 * scsi_finish_command() from this function.
126 */
127void
128scmd_eh_abort_handler(struct work_struct *work)
129{
130 struct scsi_cmnd *scmd =
131 container_of(work, struct scsi_cmnd, abort_work.work);
132 struct scsi_device *sdev = scmd->device;
133 int rtn;
134
135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "eh timeout, not aborting\n"));
139 } else {
140 SCSI_LOG_ERROR_RECOVERY(3,
141 scmd_printk(KERN_INFO, scmd,
142 "aborting command\n"));
143 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
144 if (rtn == SUCCESS) {
145 set_host_byte(scmd, DID_TIME_OUT);
146 if (scsi_host_eh_past_deadline(sdev->host)) {
147 SCSI_LOG_ERROR_RECOVERY(3,
148 scmd_printk(KERN_INFO, scmd,
149 "eh timeout, not retrying "
150 "aborted command\n"));
151 } else if (!scsi_noretry_cmd(scmd) &&
152 (++scmd->retries <= scmd->allowed)) {
153 SCSI_LOG_ERROR_RECOVERY(3,
154 scmd_printk(KERN_WARNING, scmd,
155 "retry aborted command\n"));
156 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
157 return;
158 } else {
159 SCSI_LOG_ERROR_RECOVERY(3,
160 scmd_printk(KERN_WARNING, scmd,
161 "finish aborted command\n"));
162 scsi_finish_command(scmd);
163 return;
164 }
165 } else {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_INFO, scmd,
168 "cmd abort %s\n",
169 (rtn == FAST_IO_FAIL) ?
170 "not send" : "failed"));
171 }
172 }
173
174 scsi_eh_scmd_add(scmd);
175}
176
177/**
178 * scsi_abort_command - schedule a command abort
179 * @scmd: scmd to abort.
180 *
181 * We only need to abort commands after a command timeout
182 */
183static int
184scsi_abort_command(struct scsi_cmnd *scmd)
185{
186 struct scsi_device *sdev = scmd->device;
187 struct Scsi_Host *shost = sdev->host;
188 unsigned long flags;
189
190 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
191 /*
192 * Retry after abort failed, escalate to next level.
193 */
194 SCSI_LOG_ERROR_RECOVERY(3,
195 scmd_printk(KERN_INFO, scmd,
196 "previous abort failed\n"));
197 BUG_ON(delayed_work_pending(&scmd->abort_work));
198 return FAILED;
199 }
200
201 spin_lock_irqsave(shost->host_lock, flags);
202 if (shost->eh_deadline != -1 && !shost->last_reset)
203 shost->last_reset = jiffies;
204 spin_unlock_irqrestore(shost->host_lock, flags);
205
206 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
209 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
210 return SUCCESS;
211}
212
213/**
214 * scsi_eh_reset - call into ->eh_action to reset internal counters
215 * @scmd: scmd to run eh on.
216 *
217 * The scsi driver might be carrying internal state about the
218 * devices, so we need to call into the driver to reset the
219 * internal state once the error handler is started.
220 */
221static void scsi_eh_reset(struct scsi_cmnd *scmd)
222{
223 if (!blk_rq_is_passthrough(scmd->request)) {
224 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
225 if (sdrv->eh_reset)
226 sdrv->eh_reset(scmd);
227 }
228}
229
230static void scsi_eh_inc_host_failed(struct rcu_head *head)
231{
232 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
233 struct Scsi_Host *shost = scmd->device->host;
234 unsigned long flags;
235
236 spin_lock_irqsave(shost->host_lock, flags);
237 shost->host_failed++;
238 scsi_eh_wakeup(shost);
239 spin_unlock_irqrestore(shost->host_lock, flags);
240}
241
242/**
243 * scsi_eh_scmd_add - add scsi cmd to error handling.
244 * @scmd: scmd to run eh on.
245 */
246void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
247{
248 struct Scsi_Host *shost = scmd->device->host;
249 unsigned long flags;
250 int ret;
251
252 WARN_ON_ONCE(!shost->ehandler);
253
254 spin_lock_irqsave(shost->host_lock, flags);
255 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
256 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
257 WARN_ON_ONCE(ret);
258 }
259 if (shost->eh_deadline != -1 && !shost->last_reset)
260 shost->last_reset = jiffies;
261
262 scsi_eh_reset(scmd);
263 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
264 spin_unlock_irqrestore(shost->host_lock, flags);
265 /*
266 * Ensure that all tasks observe the host state change before the
267 * host_failed change.
268 */
269 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
270}
271
272/**
273 * scsi_times_out - Timeout function for normal scsi commands.
274 * @req: request that is timing out.
275 *
276 * Notes:
277 * We do not need to lock this. There is the potential for a race
278 * only in that the normal completion handling might run, but if the
279 * normal completion function determines that the timer has already
280 * fired, then it mustn't do anything.
281 */
282enum blk_eh_timer_return scsi_times_out(struct request *req)
283{
284 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
285 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
286 struct Scsi_Host *host = scmd->device->host;
287
288 trace_scsi_dispatch_cmd_timeout(scmd);
289 scsi_log_completion(scmd, TIMEOUT_ERROR);
290
291 if (host->eh_deadline != -1 && !host->last_reset)
292 host->last_reset = jiffies;
293
294 if (host->hostt->eh_timed_out)
295 rtn = host->hostt->eh_timed_out(scmd);
296
297 if (rtn == BLK_EH_NOT_HANDLED) {
298 if (scsi_abort_command(scmd) != SUCCESS) {
299 set_host_byte(scmd, DID_TIME_OUT);
300 scsi_eh_scmd_add(scmd);
301 }
302 }
303
304 return rtn;
305}
306
307/**
308 * scsi_block_when_processing_errors - Prevent cmds from being queued.
309 * @sdev: Device on which we are performing recovery.
310 *
311 * Description:
312 * We block until the host is out of error recovery, and then check to
313 * see whether the host or the device is offline.
314 *
315 * Return value:
316 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
317 */
318int scsi_block_when_processing_errors(struct scsi_device *sdev)
319{
320 int online;
321
322 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
323
324 online = scsi_device_online(sdev);
325
326 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
327 "%s: rtn: %d\n", __func__, online));
328
329 return online;
330}
331EXPORT_SYMBOL(scsi_block_when_processing_errors);
332
333#ifdef CONFIG_SCSI_LOGGING
334/**
335 * scsi_eh_prt_fail_stats - Log info on failures.
336 * @shost: scsi host being recovered.
337 * @work_q: Queue of scsi cmds to process.
338 */
339static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
340 struct list_head *work_q)
341{
342 struct scsi_cmnd *scmd;
343 struct scsi_device *sdev;
344 int total_failures = 0;
345 int cmd_failed = 0;
346 int cmd_cancel = 0;
347 int devices_failed = 0;
348
349 shost_for_each_device(sdev, shost) {
350 list_for_each_entry(scmd, work_q, eh_entry) {
351 if (scmd->device == sdev) {
352 ++total_failures;
353 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
354 ++cmd_cancel;
355 else
356 ++cmd_failed;
357 }
358 }
359
360 if (cmd_cancel || cmd_failed) {
361 SCSI_LOG_ERROR_RECOVERY(3,
362 shost_printk(KERN_INFO, shost,
363 "%s: cmds failed: %d, cancel: %d\n",
364 __func__, cmd_failed,
365 cmd_cancel));
366 cmd_cancel = 0;
367 cmd_failed = 0;
368 ++devices_failed;
369 }
370 }
371
372 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
373 "Total of %d commands on %d"
374 " devices require eh work\n",
375 total_failures, devices_failed));
376}
377#endif
378
379 /**
380 * scsi_report_lun_change - Set flag on all *other* devices on the same target
381 * to indicate that a UNIT ATTENTION is expected.
382 * @sdev: Device reporting the UNIT ATTENTION
383 */
384static void scsi_report_lun_change(struct scsi_device *sdev)
385{
386 sdev->sdev_target->expecting_lun_change = 1;
387}
388
389/**
390 * scsi_report_sense - Examine scsi sense information and log messages for
391 * certain conditions, also issue uevents for some of them.
392 * @sdev: Device reporting the sense code
393 * @sshdr: sshdr to be examined
394 */
395static void scsi_report_sense(struct scsi_device *sdev,
396 struct scsi_sense_hdr *sshdr)
397{
398 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
399
400 if (sshdr->sense_key == UNIT_ATTENTION) {
401 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
402 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
403 sdev_printk(KERN_WARNING, sdev,
404 "Inquiry data has changed");
405 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
406 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
407 scsi_report_lun_change(sdev);
408 sdev_printk(KERN_WARNING, sdev,
409 "Warning! Received an indication that the "
410 "LUN assignments on this target have "
411 "changed. The Linux SCSI layer does not "
412 "automatically remap LUN assignments.\n");
413 } else if (sshdr->asc == 0x3f)
414 sdev_printk(KERN_WARNING, sdev,
415 "Warning! Received an indication that the "
416 "operating parameters on this target have "
417 "changed. The Linux SCSI layer does not "
418 "automatically adjust these parameters.\n");
419
420 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
421 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
422 sdev_printk(KERN_WARNING, sdev,
423 "Warning! Received an indication that the "
424 "LUN reached a thin provisioning soft "
425 "threshold.\n");
426 }
427
428 if (sshdr->asc == 0x29) {
429 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
430 sdev_printk(KERN_WARNING, sdev,
431 "Power-on or device reset occurred\n");
432 }
433
434 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
435 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
436 sdev_printk(KERN_WARNING, sdev,
437 "Mode parameters changed");
438 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
439 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
440 sdev_printk(KERN_WARNING, sdev,
441 "Asymmetric access state changed");
442 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
443 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
444 sdev_printk(KERN_WARNING, sdev,
445 "Capacity data has changed");
446 } else if (sshdr->asc == 0x2a)
447 sdev_printk(KERN_WARNING, sdev,
448 "Parameters changed");
449 }
450
451 if (evt_type != SDEV_EVT_MAXBITS) {
452 set_bit(evt_type, sdev->pending_events);
453 schedule_work(&sdev->event_work);
454 }
455}
456
457/**
458 * scsi_check_sense - Examine scsi cmd sense
459 * @scmd: Cmd to have sense checked.
460 *
461 * Return value:
462 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
463 *
464 * Notes:
465 * When a deferred error is detected the current command has
466 * not been executed and needs retrying.
467 */
468int scsi_check_sense(struct scsi_cmnd *scmd)
469{
470 struct scsi_device *sdev = scmd->device;
471 struct scsi_sense_hdr sshdr;
472
473 if (! scsi_command_normalize_sense(scmd, &sshdr))
474 return FAILED; /* no valid sense data */
475
476 scsi_report_sense(sdev, &sshdr);
477
478 if (scsi_sense_is_deferred(&sshdr))
479 return NEEDS_RETRY;
480
481 if (sdev->handler && sdev->handler->check_sense) {
482 int rc;
483
484 rc = sdev->handler->check_sense(sdev, &sshdr);
485 if (rc != SCSI_RETURN_NOT_HANDLED)
486 return rc;
487 /* handler does not care. Drop down to default handling */
488 }
489
490 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
491 /*
492 * nasty: for mid-layer issued TURs, we need to return the
493 * actual sense data without any recovery attempt. For eh
494 * issued ones, we need to try to recover and interpret
495 */
496 return SUCCESS;
497
498 /*
499 * Previous logic looked for FILEMARK, EOM or ILI which are
500 * mainly associated with tapes and returned SUCCESS.
501 */
502 if (sshdr.response_code == 0x70) {
503 /* fixed format */
504 if (scmd->sense_buffer[2] & 0xe0)
505 return SUCCESS;
506 } else {
507 /*
508 * descriptor format: look for "stream commands sense data
509 * descriptor" (see SSC-3). Assume single sense data
510 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
511 */
512 if ((sshdr.additional_length > 3) &&
513 (scmd->sense_buffer[8] == 0x4) &&
514 (scmd->sense_buffer[11] & 0xe0))
515 return SUCCESS;
516 }
517
518 switch (sshdr.sense_key) {
519 case NO_SENSE:
520 return SUCCESS;
521 case RECOVERED_ERROR:
522 return /* soft_error */ SUCCESS;
523
524 case ABORTED_COMMAND:
525 if (sshdr.asc == 0x10) /* DIF */
526 return SUCCESS;
527
528 return NEEDS_RETRY;
529 case NOT_READY:
530 case UNIT_ATTENTION:
531 /*
532 * if we are expecting a cc/ua because of a bus reset that we
533 * performed, treat this just as a retry. otherwise this is
534 * information that we should pass up to the upper-level driver
535 * so that we can deal with it there.
536 */
537 if (scmd->device->expecting_cc_ua) {
538 /*
539 * Because some device does not queue unit
540 * attentions correctly, we carefully check
541 * additional sense code and qualifier so as
542 * not to squash media change unit attention.
543 */
544 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
545 scmd->device->expecting_cc_ua = 0;
546 return NEEDS_RETRY;
547 }
548 }
549 /*
550 * we might also expect a cc/ua if another LUN on the target
551 * reported a UA with an ASC/ASCQ of 3F 0E -
552 * REPORTED LUNS DATA HAS CHANGED.
553 */
554 if (scmd->device->sdev_target->expecting_lun_change &&
555 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
556 return NEEDS_RETRY;
557 /*
558 * if the device is in the process of becoming ready, we
559 * should retry.
560 */
561 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
562 return NEEDS_RETRY;
563 /*
564 * if the device is not started, we need to wake
565 * the error handler to start the motor
566 */
567 if (scmd->device->allow_restart &&
568 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
569 return FAILED;
570 /*
571 * Pass the UA upwards for a determination in the completion
572 * functions.
573 */
574 return SUCCESS;
575
576 /* these are not supported */
577 case DATA_PROTECT:
578 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
579 /* Thin provisioning hard threshold reached */
580 set_host_byte(scmd, DID_ALLOC_FAILURE);
581 return SUCCESS;
582 }
583 /* FALLTHROUGH */
584 case COPY_ABORTED:
585 case VOLUME_OVERFLOW:
586 case MISCOMPARE:
587 case BLANK_CHECK:
588 set_host_byte(scmd, DID_TARGET_FAILURE);
589 return SUCCESS;
590
591 case MEDIUM_ERROR:
592 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
593 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
594 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
595 set_host_byte(scmd, DID_MEDIUM_ERROR);
596 return SUCCESS;
597 }
598 return NEEDS_RETRY;
599
600 case HARDWARE_ERROR:
601 if (scmd->device->retry_hwerror)
602 return ADD_TO_MLQUEUE;
603 else
604 set_host_byte(scmd, DID_TARGET_FAILURE);
605 /* FALLTHROUGH */
606
607 case ILLEGAL_REQUEST:
608 if (sshdr.asc == 0x20 || /* Invalid command operation code */
609 sshdr.asc == 0x21 || /* Logical block address out of range */
610 sshdr.asc == 0x22 || /* Invalid function */
611 sshdr.asc == 0x24 || /* Invalid field in cdb */
612 sshdr.asc == 0x26 || /* Parameter value invalid */
613 sshdr.asc == 0x27) { /* Write protected */
614 set_host_byte(scmd, DID_TARGET_FAILURE);
615 }
616 return SUCCESS;
617
618 default:
619 return SUCCESS;
620 }
621}
622EXPORT_SYMBOL_GPL(scsi_check_sense);
623
624static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
625{
626 struct scsi_host_template *sht = sdev->host->hostt;
627 struct scsi_device *tmp_sdev;
628
629 if (!sht->track_queue_depth ||
630 sdev->queue_depth >= sdev->max_queue_depth)
631 return;
632
633 if (time_before(jiffies,
634 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
635 return;
636
637 if (time_before(jiffies,
638 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
639 return;
640
641 /*
642 * Walk all devices of a target and do
643 * ramp up on them.
644 */
645 shost_for_each_device(tmp_sdev, sdev->host) {
646 if (tmp_sdev->channel != sdev->channel ||
647 tmp_sdev->id != sdev->id ||
648 tmp_sdev->queue_depth == sdev->max_queue_depth)
649 continue;
650
651 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
652 sdev->last_queue_ramp_up = jiffies;
653 }
654}
655
656static void scsi_handle_queue_full(struct scsi_device *sdev)
657{
658 struct scsi_host_template *sht = sdev->host->hostt;
659 struct scsi_device *tmp_sdev;
660
661 if (!sht->track_queue_depth)
662 return;
663
664 shost_for_each_device(tmp_sdev, sdev->host) {
665 if (tmp_sdev->channel != sdev->channel ||
666 tmp_sdev->id != sdev->id)
667 continue;
668 /*
669 * We do not know the number of commands that were at
670 * the device when we got the queue full so we start
671 * from the highest possible value and work our way down.
672 */
673 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
674 }
675}
676
677/**
678 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
679 * @scmd: SCSI cmd to examine.
680 *
681 * Notes:
682 * This is *only* called when we are examining the status of commands
683 * queued during error recovery. the main difference here is that we
684 * don't allow for the possibility of retries here, and we are a lot
685 * more restrictive about what we consider acceptable.
686 */
687static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
688{
689 /*
690 * first check the host byte, to see if there is anything in there
691 * that would indicate what we need to do.
692 */
693 if (host_byte(scmd->result) == DID_RESET) {
694 /*
695 * rats. we are already in the error handler, so we now
696 * get to try and figure out what to do next. if the sense
697 * is valid, we have a pretty good idea of what to do.
698 * if not, we mark it as FAILED.
699 */
700 return scsi_check_sense(scmd);
701 }
702 if (host_byte(scmd->result) != DID_OK)
703 return FAILED;
704
705 /*
706 * next, check the message byte.
707 */
708 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
709 return FAILED;
710
711 /*
712 * now, check the status byte to see if this indicates
713 * anything special.
714 */
715 switch (status_byte(scmd->result)) {
716 case GOOD:
717 scsi_handle_queue_ramp_up(scmd->device);
718 /* FALLTHROUGH */
719 case COMMAND_TERMINATED:
720 return SUCCESS;
721 case CHECK_CONDITION:
722 return scsi_check_sense(scmd);
723 case CONDITION_GOOD:
724 case INTERMEDIATE_GOOD:
725 case INTERMEDIATE_C_GOOD:
726 /*
727 * who knows? FIXME(eric)
728 */
729 return SUCCESS;
730 case RESERVATION_CONFLICT:
731 if (scmd->cmnd[0] == TEST_UNIT_READY)
732 /* it is a success, we probed the device and
733 * found it */
734 return SUCCESS;
735 /* otherwise, we failed to send the command */
736 return FAILED;
737 case QUEUE_FULL:
738 scsi_handle_queue_full(scmd->device);
739 /* fall through */
740 case BUSY:
741 return NEEDS_RETRY;
742 default:
743 return FAILED;
744 }
745 return FAILED;
746}
747
748/**
749 * scsi_eh_done - Completion function for error handling.
750 * @scmd: Cmd that is done.
751 */
752static void scsi_eh_done(struct scsi_cmnd *scmd)
753{
754 struct completion *eh_action;
755
756 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
757 "%s result: %x\n", __func__, scmd->result));
758
759 eh_action = scmd->device->host->eh_action;
760 if (eh_action)
761 complete(eh_action);
762}
763
764/**
765 * scsi_try_host_reset - ask host adapter to reset itself
766 * @scmd: SCSI cmd to send host reset.
767 */
768static int scsi_try_host_reset(struct scsi_cmnd *scmd)
769{
770 unsigned long flags;
771 int rtn;
772 struct Scsi_Host *host = scmd->device->host;
773 struct scsi_host_template *hostt = host->hostt;
774
775 SCSI_LOG_ERROR_RECOVERY(3,
776 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
777
778 if (!hostt->eh_host_reset_handler)
779 return FAILED;
780
781 rtn = hostt->eh_host_reset_handler(scmd);
782
783 if (rtn == SUCCESS) {
784 if (!hostt->skip_settle_delay)
785 ssleep(HOST_RESET_SETTLE_TIME);
786 spin_lock_irqsave(host->host_lock, flags);
787 scsi_report_bus_reset(host, scmd_channel(scmd));
788 spin_unlock_irqrestore(host->host_lock, flags);
789 }
790
791 return rtn;
792}
793
794/**
795 * scsi_try_bus_reset - ask host to perform a bus reset
796 * @scmd: SCSI cmd to send bus reset.
797 */
798static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
799{
800 unsigned long flags;
801 int rtn;
802 struct Scsi_Host *host = scmd->device->host;
803 struct scsi_host_template *hostt = host->hostt;
804
805 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
806 "%s: Snd Bus RST\n", __func__));
807
808 if (!hostt->eh_bus_reset_handler)
809 return FAILED;
810
811 rtn = hostt->eh_bus_reset_handler(scmd);
812
813 if (rtn == SUCCESS) {
814 if (!hostt->skip_settle_delay)
815 ssleep(BUS_RESET_SETTLE_TIME);
816 spin_lock_irqsave(host->host_lock, flags);
817 scsi_report_bus_reset(host, scmd_channel(scmd));
818 spin_unlock_irqrestore(host->host_lock, flags);
819 }
820
821 return rtn;
822}
823
824static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
825{
826 sdev->was_reset = 1;
827 sdev->expecting_cc_ua = 1;
828}
829
830/**
831 * scsi_try_target_reset - Ask host to perform a target reset
832 * @scmd: SCSI cmd used to send a target reset
833 *
834 * Notes:
835 * There is no timeout for this operation. if this operation is
836 * unreliable for a given host, then the host itself needs to put a
837 * timer on it, and set the host back to a consistent state prior to
838 * returning.
839 */
840static int scsi_try_target_reset(struct scsi_cmnd *scmd)
841{
842 unsigned long flags;
843 int rtn;
844 struct Scsi_Host *host = scmd->device->host;
845 struct scsi_host_template *hostt = host->hostt;
846
847 if (!hostt->eh_target_reset_handler)
848 return FAILED;
849
850 rtn = hostt->eh_target_reset_handler(scmd);
851 if (rtn == SUCCESS) {
852 spin_lock_irqsave(host->host_lock, flags);
853 __starget_for_each_device(scsi_target(scmd->device), NULL,
854 __scsi_report_device_reset);
855 spin_unlock_irqrestore(host->host_lock, flags);
856 }
857
858 return rtn;
859}
860
861/**
862 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
863 * @scmd: SCSI cmd used to send BDR
864 *
865 * Notes:
866 * There is no timeout for this operation. if this operation is
867 * unreliable for a given host, then the host itself needs to put a
868 * timer on it, and set the host back to a consistent state prior to
869 * returning.
870 */
871static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
872{
873 int rtn;
874 struct scsi_host_template *hostt = scmd->device->host->hostt;
875
876 if (!hostt->eh_device_reset_handler)
877 return FAILED;
878
879 rtn = hostt->eh_device_reset_handler(scmd);
880 if (rtn == SUCCESS)
881 __scsi_report_device_reset(scmd->device, NULL);
882 return rtn;
883}
884
885/**
886 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
887 * @hostt: SCSI driver host template
888 * @scmd: SCSI cmd used to send a target reset
889 *
890 * Return value:
891 * SUCCESS, FAILED, or FAST_IO_FAIL
892 *
893 * Notes:
894 * SUCCESS does not necessarily indicate that the command
895 * has been aborted; it only indicates that the LLDDs
896 * has cleared all references to that command.
897 * LLDDs should return FAILED only if an abort was required
898 * but could not be executed. LLDDs should return FAST_IO_FAIL
899 * if the device is temporarily unavailable (eg due to a
900 * link down on FibreChannel)
901 */
902static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
903 struct scsi_cmnd *scmd)
904{
905 if (!hostt->eh_abort_handler)
906 return FAILED;
907
908 return hostt->eh_abort_handler(scmd);
909}
910
911static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
912{
913 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
914 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
915 if (scsi_try_target_reset(scmd) != SUCCESS)
916 if (scsi_try_bus_reset(scmd) != SUCCESS)
917 scsi_try_host_reset(scmd);
918}
919
920/**
921 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
922 * @scmd: SCSI command structure to hijack
923 * @ses: structure to save restore information
924 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
925 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
926 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
927 *
928 * This function is used to save a scsi command information before re-execution
929 * as part of the error recovery process. If @sense_bytes is 0 the command
930 * sent must be one that does not transfer any data. If @sense_bytes != 0
931 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
932 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
933 */
934void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
935 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
936{
937 struct scsi_device *sdev = scmd->device;
938
939 /*
940 * We need saved copies of a number of fields - this is because
941 * error handling may need to overwrite these with different values
942 * to run different commands, and once error handling is complete,
943 * we will need to restore these values prior to running the actual
944 * command.
945 */
946 ses->cmd_len = scmd->cmd_len;
947 ses->cmnd = scmd->cmnd;
948 ses->data_direction = scmd->sc_data_direction;
949 ses->sdb = scmd->sdb;
950 ses->next_rq = scmd->request->next_rq;
951 ses->result = scmd->result;
952 ses->underflow = scmd->underflow;
953 ses->prot_op = scmd->prot_op;
954 ses->eh_eflags = scmd->eh_eflags;
955
956 scmd->prot_op = SCSI_PROT_NORMAL;
957 scmd->eh_eflags = 0;
958 scmd->cmnd = ses->eh_cmnd;
959 memset(scmd->cmnd, 0, BLK_MAX_CDB);
960 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
961 scmd->request->next_rq = NULL;
962 scmd->result = 0;
963
964 if (sense_bytes) {
965 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
966 sense_bytes);
967 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
968 scmd->sdb.length);
969 scmd->sdb.table.sgl = &ses->sense_sgl;
970 scmd->sc_data_direction = DMA_FROM_DEVICE;
971 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
972 scmd->cmnd[0] = REQUEST_SENSE;
973 scmd->cmnd[4] = scmd->sdb.length;
974 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
975 } else {
976 scmd->sc_data_direction = DMA_NONE;
977 if (cmnd) {
978 BUG_ON(cmnd_size > BLK_MAX_CDB);
979 memcpy(scmd->cmnd, cmnd, cmnd_size);
980 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
981 }
982 }
983
984 scmd->underflow = 0;
985
986 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
987 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
988 (sdev->lun << 5 & 0xe0);
989
990 /*
991 * Zero the sense buffer. The scsi spec mandates that any
992 * untransferred sense data should be interpreted as being zero.
993 */
994 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
995}
996EXPORT_SYMBOL(scsi_eh_prep_cmnd);
997
998/**
999 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
1000 * @scmd: SCSI command structure to restore
1001 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
1002 *
1003 * Undo any damage done by above scsi_eh_prep_cmnd().
1004 */
1005void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1006{
1007 /*
1008 * Restore original data
1009 */
1010 scmd->cmd_len = ses->cmd_len;
1011 scmd->cmnd = ses->cmnd;
1012 scmd->sc_data_direction = ses->data_direction;
1013 scmd->sdb = ses->sdb;
1014 scmd->request->next_rq = ses->next_rq;
1015 scmd->result = ses->result;
1016 scmd->underflow = ses->underflow;
1017 scmd->prot_op = ses->prot_op;
1018 scmd->eh_eflags = ses->eh_eflags;
1019}
1020EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1021
1022/**
1023 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1024 * @scmd: SCSI command structure to hijack
1025 * @cmnd: CDB to send
1026 * @cmnd_size: size in bytes of @cmnd
1027 * @timeout: timeout for this request
1028 * @sense_bytes: size of sense data to copy or 0
1029 *
1030 * This function is used to send a scsi command down to a target device
1031 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1032 *
1033 * Return value:
1034 * SUCCESS or FAILED or NEEDS_RETRY
1035 */
1036static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1037 int cmnd_size, int timeout, unsigned sense_bytes)
1038{
1039 struct scsi_device *sdev = scmd->device;
1040 struct Scsi_Host *shost = sdev->host;
1041 DECLARE_COMPLETION_ONSTACK(done);
1042 unsigned long timeleft = timeout;
1043 struct scsi_eh_save ses;
1044 const unsigned long stall_for = msecs_to_jiffies(100);
1045 int rtn;
1046
1047retry:
1048 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1049 shost->eh_action = &done;
1050
1051 scsi_log_send(scmd);
1052 scmd->scsi_done = scsi_eh_done;
1053 rtn = shost->hostt->queuecommand(shost, scmd);
1054 if (rtn) {
1055 if (timeleft > stall_for) {
1056 scsi_eh_restore_cmnd(scmd, &ses);
1057 timeleft -= stall_for;
1058 msleep(jiffies_to_msecs(stall_for));
1059 goto retry;
1060 }
1061 /* signal not to enter either branch of the if () below */
1062 timeleft = 0;
1063 rtn = FAILED;
1064 } else {
1065 timeleft = wait_for_completion_timeout(&done, timeout);
1066 rtn = SUCCESS;
1067 }
1068
1069 shost->eh_action = NULL;
1070
1071 scsi_log_completion(scmd, rtn);
1072
1073 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1074 "%s timeleft: %ld\n",
1075 __func__, timeleft));
1076
1077 /*
1078 * If there is time left scsi_eh_done got called, and we will examine
1079 * the actual status codes to see whether the command actually did
1080 * complete normally, else if we have a zero return and no time left,
1081 * the command must still be pending, so abort it and return FAILED.
1082 * If we never actually managed to issue the command, because
1083 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1084 * value above (so don't execute either branch of the if)
1085 */
1086 if (timeleft) {
1087 rtn = scsi_eh_completed_normally(scmd);
1088 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1089 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1090
1091 switch (rtn) {
1092 case SUCCESS:
1093 case NEEDS_RETRY:
1094 case FAILED:
1095 break;
1096 case ADD_TO_MLQUEUE:
1097 rtn = NEEDS_RETRY;
1098 break;
1099 default:
1100 rtn = FAILED;
1101 break;
1102 }
1103 } else if (rtn != FAILED) {
1104 scsi_abort_eh_cmnd(scmd);
1105 rtn = FAILED;
1106 }
1107
1108 scsi_eh_restore_cmnd(scmd, &ses);
1109
1110 return rtn;
1111}
1112
1113/**
1114 * scsi_request_sense - Request sense data from a particular target.
1115 * @scmd: SCSI cmd for request sense.
1116 *
1117 * Notes:
1118 * Some hosts automatically obtain this information, others require
1119 * that we obtain it on our own. This function will *not* return until
1120 * the command either times out, or it completes.
1121 */
1122static int scsi_request_sense(struct scsi_cmnd *scmd)
1123{
1124 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1125}
1126
1127static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1128{
1129 if (!blk_rq_is_passthrough(scmd->request)) {
1130 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1131 if (sdrv->eh_action)
1132 rtn = sdrv->eh_action(scmd, rtn);
1133 }
1134 return rtn;
1135}
1136
1137/**
1138 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1139 * @scmd: Original SCSI cmd that eh has finished.
1140 * @done_q: Queue for processed commands.
1141 *
1142 * Notes:
1143 * We don't want to use the normal command completion while we are are
1144 * still handling errors - it may cause other commands to be queued,
1145 * and that would disturb what we are doing. Thus we really want to
1146 * keep a list of pending commands for final completion, and once we
1147 * are ready to leave error handling we handle completion for real.
1148 */
1149void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1150{
1151 list_move_tail(&scmd->eh_entry, done_q);
1152}
1153EXPORT_SYMBOL(scsi_eh_finish_cmd);
1154
1155/**
1156 * scsi_eh_get_sense - Get device sense data.
1157 * @work_q: Queue of commands to process.
1158 * @done_q: Queue of processed commands.
1159 *
1160 * Description:
1161 * See if we need to request sense information. if so, then get it
1162 * now, so we have a better idea of what to do.
1163 *
1164 * Notes:
1165 * This has the unfortunate side effect that if a shost adapter does
1166 * not automatically request sense information, we end up shutting
1167 * it down before we request it.
1168 *
1169 * All drivers should request sense information internally these days,
1170 * so for now all I have to say is tough noogies if you end up in here.
1171 *
1172 * XXX: Long term this code should go away, but that needs an audit of
1173 * all LLDDs first.
1174 */
1175int scsi_eh_get_sense(struct list_head *work_q,
1176 struct list_head *done_q)
1177{
1178 struct scsi_cmnd *scmd, *next;
1179 struct Scsi_Host *shost;
1180 int rtn;
1181
1182 /*
1183 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1184 * should not get sense.
1185 */
1186 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1187 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1188 SCSI_SENSE_VALID(scmd))
1189 continue;
1190
1191 shost = scmd->device->host;
1192 if (scsi_host_eh_past_deadline(shost)) {
1193 SCSI_LOG_ERROR_RECOVERY(3,
1194 scmd_printk(KERN_INFO, scmd,
1195 "%s: skip request sense, past eh deadline\n",
1196 current->comm));
1197 break;
1198 }
1199 if (status_byte(scmd->result) != CHECK_CONDITION)
1200 /*
1201 * don't request sense if there's no check condition
1202 * status because the error we're processing isn't one
1203 * that has a sense code (and some devices get
1204 * confused by sense requests out of the blue)
1205 */
1206 continue;
1207
1208 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1209 "%s: requesting sense\n",
1210 current->comm));
1211 rtn = scsi_request_sense(scmd);
1212 if (rtn != SUCCESS)
1213 continue;
1214
1215 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1216 "sense requested, result %x\n", scmd->result));
1217 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1218
1219 rtn = scsi_decide_disposition(scmd);
1220
1221 /*
1222 * if the result was normal, then just pass it along to the
1223 * upper level.
1224 */
1225 if (rtn == SUCCESS)
1226 /* we don't want this command reissued, just
1227 * finished with the sense data, so set
1228 * retries to the max allowed to ensure it
1229 * won't get reissued */
1230 scmd->retries = scmd->allowed;
1231 else if (rtn != NEEDS_RETRY)
1232 continue;
1233
1234 scsi_eh_finish_cmd(scmd, done_q);
1235 }
1236
1237 return list_empty(work_q);
1238}
1239EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1240
1241/**
1242 * scsi_eh_tur - Send TUR to device.
1243 * @scmd: &scsi_cmnd to send TUR
1244 *
1245 * Return value:
1246 * 0 - Device is ready. 1 - Device NOT ready.
1247 */
1248static int scsi_eh_tur(struct scsi_cmnd *scmd)
1249{
1250 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1251 int retry_cnt = 1, rtn;
1252
1253retry_tur:
1254 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1255 scmd->device->eh_timeout, 0);
1256
1257 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1258 "%s return: %x\n", __func__, rtn));
1259
1260 switch (rtn) {
1261 case NEEDS_RETRY:
1262 if (retry_cnt--)
1263 goto retry_tur;
1264 /*FALLTHRU*/
1265 case SUCCESS:
1266 return 0;
1267 default:
1268 return 1;
1269 }
1270}
1271
1272/**
1273 * scsi_eh_test_devices - check if devices are responding from error recovery.
1274 * @cmd_list: scsi commands in error recovery.
1275 * @work_q: queue for commands which still need more error recovery
1276 * @done_q: queue for commands which are finished
1277 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1278 *
1279 * Decription:
1280 * Tests if devices are in a working state. Commands to devices now in
1281 * a working state are sent to the done_q while commands to devices which
1282 * are still failing to respond are returned to the work_q for more
1283 * processing.
1284 **/
1285static int scsi_eh_test_devices(struct list_head *cmd_list,
1286 struct list_head *work_q,
1287 struct list_head *done_q, int try_stu)
1288{
1289 struct scsi_cmnd *scmd, *next;
1290 struct scsi_device *sdev;
1291 int finish_cmds;
1292
1293 while (!list_empty(cmd_list)) {
1294 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1295 sdev = scmd->device;
1296
1297 if (!try_stu) {
1298 if (scsi_host_eh_past_deadline(sdev->host)) {
1299 /* Push items back onto work_q */
1300 list_splice_init(cmd_list, work_q);
1301 SCSI_LOG_ERROR_RECOVERY(3,
1302 sdev_printk(KERN_INFO, sdev,
1303 "%s: skip test device, past eh deadline",
1304 current->comm));
1305 break;
1306 }
1307 }
1308
1309 finish_cmds = !scsi_device_online(scmd->device) ||
1310 (try_stu && !scsi_eh_try_stu(scmd) &&
1311 !scsi_eh_tur(scmd)) ||
1312 !scsi_eh_tur(scmd);
1313
1314 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1315 if (scmd->device == sdev) {
1316 if (finish_cmds &&
1317 (try_stu ||
1318 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1319 scsi_eh_finish_cmd(scmd, done_q);
1320 else
1321 list_move_tail(&scmd->eh_entry, work_q);
1322 }
1323 }
1324 return list_empty(work_q);
1325}
1326
1327/**
1328 * scsi_eh_try_stu - Send START_UNIT to device.
1329 * @scmd: &scsi_cmnd to send START_UNIT
1330 *
1331 * Return value:
1332 * 0 - Device is ready. 1 - Device NOT ready.
1333 */
1334static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1335{
1336 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1337
1338 if (scmd->device->allow_restart) {
1339 int i, rtn = NEEDS_RETRY;
1340
1341 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1342 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1343
1344 if (rtn == SUCCESS)
1345 return 0;
1346 }
1347
1348 return 1;
1349}
1350
1351 /**
1352 * scsi_eh_stu - send START_UNIT if needed
1353 * @shost: &scsi host being recovered.
1354 * @work_q: &list_head for pending commands.
1355 * @done_q: &list_head for processed commands.
1356 *
1357 * Notes:
1358 * If commands are failing due to not ready, initializing command required,
1359 * try revalidating the device, which will end up sending a start unit.
1360 */
1361static int scsi_eh_stu(struct Scsi_Host *shost,
1362 struct list_head *work_q,
1363 struct list_head *done_q)
1364{
1365 struct scsi_cmnd *scmd, *stu_scmd, *next;
1366 struct scsi_device *sdev;
1367
1368 shost_for_each_device(sdev, shost) {
1369 if (scsi_host_eh_past_deadline(shost)) {
1370 SCSI_LOG_ERROR_RECOVERY(3,
1371 sdev_printk(KERN_INFO, sdev,
1372 "%s: skip START_UNIT, past eh deadline\n",
1373 current->comm));
1374 break;
1375 }
1376 stu_scmd = NULL;
1377 list_for_each_entry(scmd, work_q, eh_entry)
1378 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1379 scsi_check_sense(scmd) == FAILED ) {
1380 stu_scmd = scmd;
1381 break;
1382 }
1383
1384 if (!stu_scmd)
1385 continue;
1386
1387 SCSI_LOG_ERROR_RECOVERY(3,
1388 sdev_printk(KERN_INFO, sdev,
1389 "%s: Sending START_UNIT\n",
1390 current->comm));
1391
1392 if (!scsi_eh_try_stu(stu_scmd)) {
1393 if (!scsi_device_online(sdev) ||
1394 !scsi_eh_tur(stu_scmd)) {
1395 list_for_each_entry_safe(scmd, next,
1396 work_q, eh_entry) {
1397 if (scmd->device == sdev &&
1398 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1399 scsi_eh_finish_cmd(scmd, done_q);
1400 }
1401 }
1402 } else {
1403 SCSI_LOG_ERROR_RECOVERY(3,
1404 sdev_printk(KERN_INFO, sdev,
1405 "%s: START_UNIT failed\n",
1406 current->comm));
1407 }
1408 }
1409
1410 return list_empty(work_q);
1411}
1412
1413
1414/**
1415 * scsi_eh_bus_device_reset - send bdr if needed
1416 * @shost: scsi host being recovered.
1417 * @work_q: &list_head for pending commands.
1418 * @done_q: &list_head for processed commands.
1419 *
1420 * Notes:
1421 * Try a bus device reset. Still, look to see whether we have multiple
1422 * devices that are jammed or not - if we have multiple devices, it
1423 * makes no sense to try bus_device_reset - we really would need to try
1424 * a bus_reset instead.
1425 */
1426static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1427 struct list_head *work_q,
1428 struct list_head *done_q)
1429{
1430 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1431 struct scsi_device *sdev;
1432 int rtn;
1433
1434 shost_for_each_device(sdev, shost) {
1435 if (scsi_host_eh_past_deadline(shost)) {
1436 SCSI_LOG_ERROR_RECOVERY(3,
1437 sdev_printk(KERN_INFO, sdev,
1438 "%s: skip BDR, past eh deadline\n",
1439 current->comm));
1440 break;
1441 }
1442 bdr_scmd = NULL;
1443 list_for_each_entry(scmd, work_q, eh_entry)
1444 if (scmd->device == sdev) {
1445 bdr_scmd = scmd;
1446 break;
1447 }
1448
1449 if (!bdr_scmd)
1450 continue;
1451
1452 SCSI_LOG_ERROR_RECOVERY(3,
1453 sdev_printk(KERN_INFO, sdev,
1454 "%s: Sending BDR\n", current->comm));
1455 rtn = scsi_try_bus_device_reset(bdr_scmd);
1456 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1457 if (!scsi_device_online(sdev) ||
1458 rtn == FAST_IO_FAIL ||
1459 !scsi_eh_tur(bdr_scmd)) {
1460 list_for_each_entry_safe(scmd, next,
1461 work_q, eh_entry) {
1462 if (scmd->device == sdev &&
1463 scsi_eh_action(scmd, rtn) != FAILED)
1464 scsi_eh_finish_cmd(scmd,
1465 done_q);
1466 }
1467 }
1468 } else {
1469 SCSI_LOG_ERROR_RECOVERY(3,
1470 sdev_printk(KERN_INFO, sdev,
1471 "%s: BDR failed\n", current->comm));
1472 }
1473 }
1474
1475 return list_empty(work_q);
1476}
1477
1478/**
1479 * scsi_eh_target_reset - send target reset if needed
1480 * @shost: scsi host being recovered.
1481 * @work_q: &list_head for pending commands.
1482 * @done_q: &list_head for processed commands.
1483 *
1484 * Notes:
1485 * Try a target reset.
1486 */
1487static int scsi_eh_target_reset(struct Scsi_Host *shost,
1488 struct list_head *work_q,
1489 struct list_head *done_q)
1490{
1491 LIST_HEAD(tmp_list);
1492 LIST_HEAD(check_list);
1493
1494 list_splice_init(work_q, &tmp_list);
1495
1496 while (!list_empty(&tmp_list)) {
1497 struct scsi_cmnd *next, *scmd;
1498 int rtn;
1499 unsigned int id;
1500
1501 if (scsi_host_eh_past_deadline(shost)) {
1502 /* push back on work queue for further processing */
1503 list_splice_init(&check_list, work_q);
1504 list_splice_init(&tmp_list, work_q);
1505 SCSI_LOG_ERROR_RECOVERY(3,
1506 shost_printk(KERN_INFO, shost,
1507 "%s: Skip target reset, past eh deadline\n",
1508 current->comm));
1509 return list_empty(work_q);
1510 }
1511
1512 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1513 id = scmd_id(scmd);
1514
1515 SCSI_LOG_ERROR_RECOVERY(3,
1516 shost_printk(KERN_INFO, shost,
1517 "%s: Sending target reset to target %d\n",
1518 current->comm, id));
1519 rtn = scsi_try_target_reset(scmd);
1520 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1521 SCSI_LOG_ERROR_RECOVERY(3,
1522 shost_printk(KERN_INFO, shost,
1523 "%s: Target reset failed"
1524 " target: %d\n",
1525 current->comm, id));
1526 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1527 if (scmd_id(scmd) != id)
1528 continue;
1529
1530 if (rtn == SUCCESS)
1531 list_move_tail(&scmd->eh_entry, &check_list);
1532 else if (rtn == FAST_IO_FAIL)
1533 scsi_eh_finish_cmd(scmd, done_q);
1534 else
1535 /* push back on work queue for further processing */
1536 list_move(&scmd->eh_entry, work_q);
1537 }
1538 }
1539
1540 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1541}
1542
1543/**
1544 * scsi_eh_bus_reset - send a bus reset
1545 * @shost: &scsi host being recovered.
1546 * @work_q: &list_head for pending commands.
1547 * @done_q: &list_head for processed commands.
1548 */
1549static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1550 struct list_head *work_q,
1551 struct list_head *done_q)
1552{
1553 struct scsi_cmnd *scmd, *chan_scmd, *next;
1554 LIST_HEAD(check_list);
1555 unsigned int channel;
1556 int rtn;
1557
1558 /*
1559 * we really want to loop over the various channels, and do this on
1560 * a channel by channel basis. we should also check to see if any
1561 * of the failed commands are on soft_reset devices, and if so, skip
1562 * the reset.
1563 */
1564
1565 for (channel = 0; channel <= shost->max_channel; channel++) {
1566 if (scsi_host_eh_past_deadline(shost)) {
1567 list_splice_init(&check_list, work_q);
1568 SCSI_LOG_ERROR_RECOVERY(3,
1569 shost_printk(KERN_INFO, shost,
1570 "%s: skip BRST, past eh deadline\n",
1571 current->comm));
1572 return list_empty(work_q);
1573 }
1574
1575 chan_scmd = NULL;
1576 list_for_each_entry(scmd, work_q, eh_entry) {
1577 if (channel == scmd_channel(scmd)) {
1578 chan_scmd = scmd;
1579 break;
1580 /*
1581 * FIXME add back in some support for
1582 * soft_reset devices.
1583 */
1584 }
1585 }
1586
1587 if (!chan_scmd)
1588 continue;
1589 SCSI_LOG_ERROR_RECOVERY(3,
1590 shost_printk(KERN_INFO, shost,
1591 "%s: Sending BRST chan: %d\n",
1592 current->comm, channel));
1593 rtn = scsi_try_bus_reset(chan_scmd);
1594 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1595 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1596 if (channel == scmd_channel(scmd)) {
1597 if (rtn == FAST_IO_FAIL)
1598 scsi_eh_finish_cmd(scmd,
1599 done_q);
1600 else
1601 list_move_tail(&scmd->eh_entry,
1602 &check_list);
1603 }
1604 }
1605 } else {
1606 SCSI_LOG_ERROR_RECOVERY(3,
1607 shost_printk(KERN_INFO, shost,
1608 "%s: BRST failed chan: %d\n",
1609 current->comm, channel));
1610 }
1611 }
1612 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1613}
1614
1615/**
1616 * scsi_eh_host_reset - send a host reset
1617 * @shost: host to be reset.
1618 * @work_q: &list_head for pending commands.
1619 * @done_q: &list_head for processed commands.
1620 */
1621static int scsi_eh_host_reset(struct Scsi_Host *shost,
1622 struct list_head *work_q,
1623 struct list_head *done_q)
1624{
1625 struct scsi_cmnd *scmd, *next;
1626 LIST_HEAD(check_list);
1627 int rtn;
1628
1629 if (!list_empty(work_q)) {
1630 scmd = list_entry(work_q->next,
1631 struct scsi_cmnd, eh_entry);
1632
1633 SCSI_LOG_ERROR_RECOVERY(3,
1634 shost_printk(KERN_INFO, shost,
1635 "%s: Sending HRST\n",
1636 current->comm));
1637
1638 rtn = scsi_try_host_reset(scmd);
1639 if (rtn == SUCCESS) {
1640 list_splice_init(work_q, &check_list);
1641 } else if (rtn == FAST_IO_FAIL) {
1642 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1643 scsi_eh_finish_cmd(scmd, done_q);
1644 }
1645 } else {
1646 SCSI_LOG_ERROR_RECOVERY(3,
1647 shost_printk(KERN_INFO, shost,
1648 "%s: HRST failed\n",
1649 current->comm));
1650 }
1651 }
1652 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1653}
1654
1655/**
1656 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1657 * @work_q: &list_head for pending commands.
1658 * @done_q: &list_head for processed commands.
1659 */
1660static void scsi_eh_offline_sdevs(struct list_head *work_q,
1661 struct list_head *done_q)
1662{
1663 struct scsi_cmnd *scmd, *next;
1664 struct scsi_device *sdev;
1665
1666 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1667 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1668 "not ready after error recovery\n");
1669 sdev = scmd->device;
1670
1671 mutex_lock(&sdev->state_mutex);
1672 scsi_device_set_state(sdev, SDEV_OFFLINE);
1673 mutex_unlock(&sdev->state_mutex);
1674
1675 scsi_eh_finish_cmd(scmd, done_q);
1676 }
1677 return;
1678}
1679
1680/**
1681 * scsi_noretry_cmd - determine if command should be failed fast
1682 * @scmd: SCSI cmd to examine.
1683 */
1684int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1685{
1686 switch (host_byte(scmd->result)) {
1687 case DID_OK:
1688 break;
1689 case DID_TIME_OUT:
1690 goto check_type;
1691 case DID_BUS_BUSY:
1692 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1693 case DID_PARITY:
1694 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1695 case DID_ERROR:
1696 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1697 status_byte(scmd->result) == RESERVATION_CONFLICT)
1698 return 0;
1699 /* fall through */
1700 case DID_SOFT_ERROR:
1701 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1702 }
1703
1704 if (status_byte(scmd->result) != CHECK_CONDITION)
1705 return 0;
1706
1707check_type:
1708 /*
1709 * assume caller has checked sense and determined
1710 * the check condition was retryable.
1711 */
1712 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1713 blk_rq_is_passthrough(scmd->request))
1714 return 1;
1715 else
1716 return 0;
1717}
1718
1719/**
1720 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1721 * @scmd: SCSI cmd to examine.
1722 *
1723 * Notes:
1724 * This is *only* called when we are examining the status after sending
1725 * out the actual data command. any commands that are queued for error
1726 * recovery (e.g. test_unit_ready) do *not* come through here.
1727 *
1728 * When this routine returns failed, it means the error handler thread
1729 * is woken. In cases where the error code indicates an error that
1730 * doesn't require the error handler read (i.e. we don't need to
1731 * abort/reset), this function should return SUCCESS.
1732 */
1733int scsi_decide_disposition(struct scsi_cmnd *scmd)
1734{
1735 int rtn;
1736
1737 /*
1738 * if the device is offline, then we clearly just pass the result back
1739 * up to the top level.
1740 */
1741 if (!scsi_device_online(scmd->device)) {
1742 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1743 "%s: device offline - report as SUCCESS\n", __func__));
1744 return SUCCESS;
1745 }
1746
1747 /*
1748 * first check the host byte, to see if there is anything in there
1749 * that would indicate what we need to do.
1750 */
1751 switch (host_byte(scmd->result)) {
1752 case DID_PASSTHROUGH:
1753 /*
1754 * no matter what, pass this through to the upper layer.
1755 * nuke this special code so that it looks like we are saying
1756 * did_ok.
1757 */
1758 scmd->result &= 0xff00ffff;
1759 return SUCCESS;
1760 case DID_OK:
1761 /*
1762 * looks good. drop through, and check the next byte.
1763 */
1764 break;
1765 case DID_ABORT:
1766 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1767 set_host_byte(scmd, DID_TIME_OUT);
1768 return SUCCESS;
1769 }
1770 /* FALLTHROUGH */
1771 case DID_NO_CONNECT:
1772 case DID_BAD_TARGET:
1773 /*
1774 * note - this means that we just report the status back
1775 * to the top level driver, not that we actually think
1776 * that it indicates SUCCESS.
1777 */
1778 return SUCCESS;
1779 case DID_SOFT_ERROR:
1780 /*
1781 * when the low level driver returns did_soft_error,
1782 * it is responsible for keeping an internal retry counter
1783 * in order to avoid endless loops (db)
1784 */
1785 goto maybe_retry;
1786 case DID_IMM_RETRY:
1787 return NEEDS_RETRY;
1788
1789 case DID_REQUEUE:
1790 return ADD_TO_MLQUEUE;
1791 case DID_TRANSPORT_DISRUPTED:
1792 /*
1793 * LLD/transport was disrupted during processing of the IO.
1794 * The transport class is now blocked/blocking,
1795 * and the transport will decide what to do with the IO
1796 * based on its timers and recovery capablilities if
1797 * there are enough retries.
1798 */
1799 goto maybe_retry;
1800 case DID_TRANSPORT_FAILFAST:
1801 /*
1802 * The transport decided to failfast the IO (most likely
1803 * the fast io fail tmo fired), so send IO directly upwards.
1804 */
1805 return SUCCESS;
1806 case DID_ERROR:
1807 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1808 status_byte(scmd->result) == RESERVATION_CONFLICT)
1809 /*
1810 * execute reservation conflict processing code
1811 * lower down
1812 */
1813 break;
1814 /* fallthrough */
1815 case DID_BUS_BUSY:
1816 case DID_PARITY:
1817 goto maybe_retry;
1818 case DID_TIME_OUT:
1819 /*
1820 * when we scan the bus, we get timeout messages for
1821 * these commands if there is no device available.
1822 * other hosts report did_no_connect for the same thing.
1823 */
1824 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1825 scmd->cmnd[0] == INQUIRY)) {
1826 return SUCCESS;
1827 } else {
1828 return FAILED;
1829 }
1830 case DID_RESET:
1831 return SUCCESS;
1832 default:
1833 return FAILED;
1834 }
1835
1836 /*
1837 * next, check the message byte.
1838 */
1839 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1840 return FAILED;
1841
1842 /*
1843 * check the status byte to see if this indicates anything special.
1844 */
1845 switch (status_byte(scmd->result)) {
1846 case QUEUE_FULL:
1847 scsi_handle_queue_full(scmd->device);
1848 /*
1849 * the case of trying to send too many commands to a
1850 * tagged queueing device.
1851 */
1852 /* FALLTHROUGH */
1853 case BUSY:
1854 /*
1855 * device can't talk to us at the moment. Should only
1856 * occur (SAM-3) when the task queue is empty, so will cause
1857 * the empty queue handling to trigger a stall in the
1858 * device.
1859 */
1860 return ADD_TO_MLQUEUE;
1861 case GOOD:
1862 if (scmd->cmnd[0] == REPORT_LUNS)
1863 scmd->device->sdev_target->expecting_lun_change = 0;
1864 scsi_handle_queue_ramp_up(scmd->device);
1865 /* FALLTHROUGH */
1866 case COMMAND_TERMINATED:
1867 return SUCCESS;
1868 case TASK_ABORTED:
1869 goto maybe_retry;
1870 case CHECK_CONDITION:
1871 rtn = scsi_check_sense(scmd);
1872 if (rtn == NEEDS_RETRY)
1873 goto maybe_retry;
1874 /* if rtn == FAILED, we have no sense information;
1875 * returning FAILED will wake the error handler thread
1876 * to collect the sense and redo the decide
1877 * disposition */
1878 return rtn;
1879 case CONDITION_GOOD:
1880 case INTERMEDIATE_GOOD:
1881 case INTERMEDIATE_C_GOOD:
1882 case ACA_ACTIVE:
1883 /*
1884 * who knows? FIXME(eric)
1885 */
1886 return SUCCESS;
1887
1888 case RESERVATION_CONFLICT:
1889 sdev_printk(KERN_INFO, scmd->device,
1890 "reservation conflict\n");
1891 set_host_byte(scmd, DID_NEXUS_FAILURE);
1892 return SUCCESS; /* causes immediate i/o error */
1893 default:
1894 return FAILED;
1895 }
1896 return FAILED;
1897
1898maybe_retry:
1899
1900 /* we requeue for retry because the error was retryable, and
1901 * the request was not marked fast fail. Note that above,
1902 * even if the request is marked fast fail, we still requeue
1903 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1904 if ((++scmd->retries) <= scmd->allowed
1905 && !scsi_noretry_cmd(scmd)) {
1906 return NEEDS_RETRY;
1907 } else {
1908 /*
1909 * no more retries - report this one back to upper level.
1910 */
1911 return SUCCESS;
1912 }
1913}
1914
1915static void eh_lock_door_done(struct request *req, blk_status_t status)
1916{
1917 __blk_put_request(req->q, req);
1918}
1919
1920/**
1921 * scsi_eh_lock_door - Prevent medium removal for the specified device
1922 * @sdev: SCSI device to prevent medium removal
1923 *
1924 * Locking:
1925 * We must be called from process context.
1926 *
1927 * Notes:
1928 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1929 * head of the devices request queue, and continue.
1930 */
1931static void scsi_eh_lock_door(struct scsi_device *sdev)
1932{
1933 struct request *req;
1934 struct scsi_request *rq;
1935
1936 /*
1937 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1938 * request becomes available
1939 */
1940 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, GFP_KERNEL);
1941 if (IS_ERR(req))
1942 return;
1943 rq = scsi_req(req);
1944
1945 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1946 rq->cmd[1] = 0;
1947 rq->cmd[2] = 0;
1948 rq->cmd[3] = 0;
1949 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1950 rq->cmd[5] = 0;
1951 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
1952
1953 req->rq_flags |= RQF_QUIET;
1954 req->timeout = 10 * HZ;
1955 rq->retries = 5;
1956
1957 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1958}
1959
1960/**
1961 * scsi_restart_operations - restart io operations to the specified host.
1962 * @shost: Host we are restarting.
1963 *
1964 * Notes:
1965 * When we entered the error handler, we blocked all further i/o to
1966 * this device. we need to 'reverse' this process.
1967 */
1968static void scsi_restart_operations(struct Scsi_Host *shost)
1969{
1970 struct scsi_device *sdev;
1971 unsigned long flags;
1972
1973 /*
1974 * If the door was locked, we need to insert a door lock request
1975 * onto the head of the SCSI request queue for the device. There
1976 * is no point trying to lock the door of an off-line device.
1977 */
1978 shost_for_each_device(sdev, shost) {
1979 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1980 scsi_eh_lock_door(sdev);
1981 sdev->was_reset = 0;
1982 }
1983 }
1984
1985 /*
1986 * next free up anything directly waiting upon the host. this
1987 * will be requests for character device operations, and also for
1988 * ioctls to queued block devices.
1989 */
1990 SCSI_LOG_ERROR_RECOVERY(3,
1991 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1992
1993 spin_lock_irqsave(shost->host_lock, flags);
1994 if (scsi_host_set_state(shost, SHOST_RUNNING))
1995 if (scsi_host_set_state(shost, SHOST_CANCEL))
1996 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1997 spin_unlock_irqrestore(shost->host_lock, flags);
1998
1999 wake_up(&shost->host_wait);
2000
2001 /*
2002 * finally we need to re-initiate requests that may be pending. we will
2003 * have had everything blocked while error handling is taking place, and
2004 * now that error recovery is done, we will need to ensure that these
2005 * requests are started.
2006 */
2007 scsi_run_host_queues(shost);
2008
2009 /*
2010 * if eh is active and host_eh_scheduled is pending we need to re-run
2011 * recovery. we do this check after scsi_run_host_queues() to allow
2012 * everything pent up since the last eh run a chance to make forward
2013 * progress before we sync again. Either we'll immediately re-run
2014 * recovery or scsi_device_unbusy() will wake us again when these
2015 * pending commands complete.
2016 */
2017 spin_lock_irqsave(shost->host_lock, flags);
2018 if (shost->host_eh_scheduled)
2019 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2020 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2021 spin_unlock_irqrestore(shost->host_lock, flags);
2022}
2023
2024/**
2025 * scsi_eh_ready_devs - check device ready state and recover if not.
2026 * @shost: host to be recovered.
2027 * @work_q: &list_head for pending commands.
2028 * @done_q: &list_head for processed commands.
2029 */
2030void scsi_eh_ready_devs(struct Scsi_Host *shost,
2031 struct list_head *work_q,
2032 struct list_head *done_q)
2033{
2034 if (!scsi_eh_stu(shost, work_q, done_q))
2035 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2036 if (!scsi_eh_target_reset(shost, work_q, done_q))
2037 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2038 if (!scsi_eh_host_reset(shost, work_q, done_q))
2039 scsi_eh_offline_sdevs(work_q,
2040 done_q);
2041}
2042EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2043
2044/**
2045 * scsi_eh_flush_done_q - finish processed commands or retry them.
2046 * @done_q: list_head of processed commands.
2047 */
2048void scsi_eh_flush_done_q(struct list_head *done_q)
2049{
2050 struct scsi_cmnd *scmd, *next;
2051
2052 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2053 list_del_init(&scmd->eh_entry);
2054 if (scsi_device_online(scmd->device) &&
2055 !scsi_noretry_cmd(scmd) &&
2056 (++scmd->retries <= scmd->allowed)) {
2057 SCSI_LOG_ERROR_RECOVERY(3,
2058 scmd_printk(KERN_INFO, scmd,
2059 "%s: flush retry cmd\n",
2060 current->comm));
2061 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2062 } else {
2063 /*
2064 * If just we got sense for the device (called
2065 * scsi_eh_get_sense), scmd->result is already
2066 * set, do not set DRIVER_TIMEOUT.
2067 */
2068 if (!scmd->result)
2069 scmd->result |= (DRIVER_TIMEOUT << 24);
2070 SCSI_LOG_ERROR_RECOVERY(3,
2071 scmd_printk(KERN_INFO, scmd,
2072 "%s: flush finish cmd\n",
2073 current->comm));
2074 scsi_finish_command(scmd);
2075 }
2076 }
2077}
2078EXPORT_SYMBOL(scsi_eh_flush_done_q);
2079
2080/**
2081 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2082 * @shost: Host to unjam.
2083 *
2084 * Notes:
2085 * When we come in here, we *know* that all commands on the bus have
2086 * either completed, failed or timed out. we also know that no further
2087 * commands are being sent to the host, so things are relatively quiet
2088 * and we have freedom to fiddle with things as we wish.
2089 *
2090 * This is only the *default* implementation. it is possible for
2091 * individual drivers to supply their own version of this function, and
2092 * if the maintainer wishes to do this, it is strongly suggested that
2093 * this function be taken as a template and modified. this function
2094 * was designed to correctly handle problems for about 95% of the
2095 * different cases out there, and it should always provide at least a
2096 * reasonable amount of error recovery.
2097 *
2098 * Any command marked 'failed' or 'timeout' must eventually have
2099 * scsi_finish_cmd() called for it. we do all of the retry stuff
2100 * here, so when we restart the host after we return it should have an
2101 * empty queue.
2102 */
2103static void scsi_unjam_host(struct Scsi_Host *shost)
2104{
2105 unsigned long flags;
2106 LIST_HEAD(eh_work_q);
2107 LIST_HEAD(eh_done_q);
2108
2109 spin_lock_irqsave(shost->host_lock, flags);
2110 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2111 spin_unlock_irqrestore(shost->host_lock, flags);
2112
2113 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2114
2115 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2116 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2117
2118 spin_lock_irqsave(shost->host_lock, flags);
2119 if (shost->eh_deadline != -1)
2120 shost->last_reset = 0;
2121 spin_unlock_irqrestore(shost->host_lock, flags);
2122 scsi_eh_flush_done_q(&eh_done_q);
2123}
2124
2125/**
2126 * scsi_error_handler - SCSI error handler thread
2127 * @data: Host for which we are running.
2128 *
2129 * Notes:
2130 * This is the main error handling loop. This is run as a kernel thread
2131 * for every SCSI host and handles all error handling activity.
2132 */
2133int scsi_error_handler(void *data)
2134{
2135 struct Scsi_Host *shost = data;
2136
2137 /*
2138 * We use TASK_INTERRUPTIBLE so that the thread is not
2139 * counted against the load average as a running process.
2140 * We never actually get interrupted because kthread_run
2141 * disables signal delivery for the created thread.
2142 */
2143 while (true) {
2144 /*
2145 * The sequence in kthread_stop() sets the stop flag first
2146 * then wakes the process. To avoid missed wakeups, the task
2147 * should always be in a non running state before the stop
2148 * flag is checked
2149 */
2150 set_current_state(TASK_INTERRUPTIBLE);
2151 if (kthread_should_stop())
2152 break;
2153
2154 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2155 shost->host_failed != atomic_read(&shost->host_busy)) {
2156 SCSI_LOG_ERROR_RECOVERY(1,
2157 shost_printk(KERN_INFO, shost,
2158 "scsi_eh_%d: sleeping\n",
2159 shost->host_no));
2160 schedule();
2161 continue;
2162 }
2163
2164 __set_current_state(TASK_RUNNING);
2165 SCSI_LOG_ERROR_RECOVERY(1,
2166 shost_printk(KERN_INFO, shost,
2167 "scsi_eh_%d: waking up %d/%d/%d\n",
2168 shost->host_no, shost->host_eh_scheduled,
2169 shost->host_failed,
2170 atomic_read(&shost->host_busy)));
2171
2172 /*
2173 * We have a host that is failing for some reason. Figure out
2174 * what we need to do to get it up and online again (if we can).
2175 * If we fail, we end up taking the thing offline.
2176 */
2177 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2178 SCSI_LOG_ERROR_RECOVERY(1,
2179 shost_printk(KERN_ERR, shost,
2180 "scsi_eh_%d: unable to autoresume\n",
2181 shost->host_no));
2182 continue;
2183 }
2184
2185 if (shost->transportt->eh_strategy_handler)
2186 shost->transportt->eh_strategy_handler(shost);
2187 else
2188 scsi_unjam_host(shost);
2189
2190 /* All scmds have been handled */
2191 shost->host_failed = 0;
2192
2193 /*
2194 * Note - if the above fails completely, the action is to take
2195 * individual devices offline and flush the queue of any
2196 * outstanding requests that may have been pending. When we
2197 * restart, we restart any I/O to any other devices on the bus
2198 * which are still online.
2199 */
2200 scsi_restart_operations(shost);
2201 if (!shost->eh_noresume)
2202 scsi_autopm_put_host(shost);
2203 }
2204 __set_current_state(TASK_RUNNING);
2205
2206 SCSI_LOG_ERROR_RECOVERY(1,
2207 shost_printk(KERN_INFO, shost,
2208 "Error handler scsi_eh_%d exiting\n",
2209 shost->host_no));
2210 shost->ehandler = NULL;
2211 return 0;
2212}
2213
2214/*
2215 * Function: scsi_report_bus_reset()
2216 *
2217 * Purpose: Utility function used by low-level drivers to report that
2218 * they have observed a bus reset on the bus being handled.
2219 *
2220 * Arguments: shost - Host in question
2221 * channel - channel on which reset was observed.
2222 *
2223 * Returns: Nothing
2224 *
2225 * Lock status: Host lock must be held.
2226 *
2227 * Notes: This only needs to be called if the reset is one which
2228 * originates from an unknown location. Resets originated
2229 * by the mid-level itself don't need to call this, but there
2230 * should be no harm.
2231 *
2232 * The main purpose of this is to make sure that a CHECK_CONDITION
2233 * is properly treated.
2234 */
2235void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2236{
2237 struct scsi_device *sdev;
2238
2239 __shost_for_each_device(sdev, shost) {
2240 if (channel == sdev_channel(sdev))
2241 __scsi_report_device_reset(sdev, NULL);
2242 }
2243}
2244EXPORT_SYMBOL(scsi_report_bus_reset);
2245
2246/*
2247 * Function: scsi_report_device_reset()
2248 *
2249 * Purpose: Utility function used by low-level drivers to report that
2250 * they have observed a device reset on the device being handled.
2251 *
2252 * Arguments: shost - Host in question
2253 * channel - channel on which reset was observed
2254 * target - target on which reset was observed
2255 *
2256 * Returns: Nothing
2257 *
2258 * Lock status: Host lock must be held
2259 *
2260 * Notes: This only needs to be called if the reset is one which
2261 * originates from an unknown location. Resets originated
2262 * by the mid-level itself don't need to call this, but there
2263 * should be no harm.
2264 *
2265 * The main purpose of this is to make sure that a CHECK_CONDITION
2266 * is properly treated.
2267 */
2268void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2269{
2270 struct scsi_device *sdev;
2271
2272 __shost_for_each_device(sdev, shost) {
2273 if (channel == sdev_channel(sdev) &&
2274 target == sdev_id(sdev))
2275 __scsi_report_device_reset(sdev, NULL);
2276 }
2277}
2278EXPORT_SYMBOL(scsi_report_device_reset);
2279
2280static void
2281scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2282{
2283}
2284
2285/**
2286 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2287 * @dev: scsi_device to operate on
2288 * @arg: reset type (see sg.h)
2289 */
2290int
2291scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2292{
2293 struct scsi_cmnd *scmd;
2294 struct Scsi_Host *shost = dev->host;
2295 struct request *rq;
2296 unsigned long flags;
2297 int error = 0, rtn, val;
2298
2299 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2300 return -EACCES;
2301
2302 error = get_user(val, arg);
2303 if (error)
2304 return error;
2305
2306 if (scsi_autopm_get_host(shost) < 0)
2307 return -EIO;
2308
2309 error = -EIO;
2310 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2311 shost->hostt->cmd_size, GFP_KERNEL);
2312 if (!rq)
2313 goto out_put_autopm_host;
2314 blk_rq_init(NULL, rq);
2315
2316 scmd = (struct scsi_cmnd *)(rq + 1);
2317 scsi_init_command(dev, scmd);
2318 scmd->request = rq;
2319 scmd->cmnd = scsi_req(rq)->cmd;
2320
2321 scmd->scsi_done = scsi_reset_provider_done_command;
2322 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2323
2324 scmd->cmd_len = 0;
2325
2326 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2327
2328 spin_lock_irqsave(shost->host_lock, flags);
2329 shost->tmf_in_progress = 1;
2330 spin_unlock_irqrestore(shost->host_lock, flags);
2331
2332 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2333 case SG_SCSI_RESET_NOTHING:
2334 rtn = SUCCESS;
2335 break;
2336 case SG_SCSI_RESET_DEVICE:
2337 rtn = scsi_try_bus_device_reset(scmd);
2338 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2339 break;
2340 /* FALLTHROUGH */
2341 case SG_SCSI_RESET_TARGET:
2342 rtn = scsi_try_target_reset(scmd);
2343 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2344 break;
2345 /* FALLTHROUGH */
2346 case SG_SCSI_RESET_BUS:
2347 rtn = scsi_try_bus_reset(scmd);
2348 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2349 break;
2350 /* FALLTHROUGH */
2351 case SG_SCSI_RESET_HOST:
2352 rtn = scsi_try_host_reset(scmd);
2353 if (rtn == SUCCESS)
2354 break;
2355 /* FALLTHROUGH */
2356 default:
2357 rtn = FAILED;
2358 break;
2359 }
2360
2361 error = (rtn == SUCCESS) ? 0 : -EIO;
2362
2363 spin_lock_irqsave(shost->host_lock, flags);
2364 shost->tmf_in_progress = 0;
2365 spin_unlock_irqrestore(shost->host_lock, flags);
2366
2367 /*
2368 * be sure to wake up anyone who was sleeping or had their queue
2369 * suspended while we performed the TMF.
2370 */
2371 SCSI_LOG_ERROR_RECOVERY(3,
2372 shost_printk(KERN_INFO, shost,
2373 "waking up host to restart after TMF\n"));
2374
2375 wake_up(&shost->host_wait);
2376 scsi_run_host_queues(shost);
2377
2378 scsi_put_command(scmd);
2379 kfree(rq);
2380
2381out_put_autopm_host:
2382 scsi_autopm_put_host(shost);
2383 return error;
2384}
2385EXPORT_SYMBOL(scsi_ioctl_reset);
2386
2387bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2388 struct scsi_sense_hdr *sshdr)
2389{
2390 return scsi_normalize_sense(cmd->sense_buffer,
2391 SCSI_SENSE_BUFFERSIZE, sshdr);
2392}
2393EXPORT_SYMBOL(scsi_command_normalize_sense);
2394
2395/**
2396 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2397 * @sense_buffer: byte array of sense data
2398 * @sb_len: number of valid bytes in sense_buffer
2399 * @info_out: pointer to 64 integer where 8 or 4 byte information
2400 * field will be placed if found.
2401 *
2402 * Return value:
2403 * true if information field found, false if not found.
2404 */
2405bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2406 u64 *info_out)
2407{
2408 const u8 * ucp;
2409
2410 if (sb_len < 7)
2411 return false;
2412 switch (sense_buffer[0] & 0x7f) {
2413 case 0x70:
2414 case 0x71:
2415 if (sense_buffer[0] & 0x80) {
2416 *info_out = get_unaligned_be32(&sense_buffer[3]);
2417 return true;
2418 }
2419 return false;
2420 case 0x72:
2421 case 0x73:
2422 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2423 0 /* info desc */);
2424 if (ucp && (0xa == ucp[1])) {
2425 *info_out = get_unaligned_be64(&ucp[4]);
2426 return true;
2427 }
2428 return false;
2429 default:
2430 return false;
2431 }
2432}
2433EXPORT_SYMBOL(scsi_get_sense_info_fld);