Merge tag 'for-5.17-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
[linux-block.git] / drivers / scsi / scsi_error.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
1da177e4
LT
2/*
3 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 *
5 * SCSI error/timeout handling
6 * Initial versions: Eric Youngdale. Based upon conversations with
0bf8c869 7 * Leonard Zubkoff and David Miller at Linux Expo,
1da177e4
LT
8 * ideas originating from all over the place.
9 *
10 * Restructured scsi_unjam_host and associated functions.
11 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 *
13 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
0bf8c869 14 * minor cleanups.
1da177e4
LT
15 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
16 */
17
18#include <linux/module.h>
19#include <linux/sched.h>
5a0e3ad6 20#include <linux/gfp.h>
1da177e4
LT
21#include <linux/timer.h>
22#include <linux/string.h>
1da177e4 23#include <linux/kernel.h>
83144186 24#include <linux/freezer.h>
c5478def 25#include <linux/kthread.h>
1da177e4
LT
26#include <linux/interrupt.h>
27#include <linux/blkdev.h>
28#include <linux/delay.h>
fc73648a 29#include <linux/jiffies.h>
1da177e4
LT
30
31#include <scsi/scsi.h>
beb40487 32#include <scsi/scsi_cmnd.h>
1da177e4
LT
33#include <scsi/scsi_dbg.h>
34#include <scsi/scsi_device.h>
18a4d0a2 35#include <scsi/scsi_driver.h>
1da177e4 36#include <scsi/scsi_eh.h>
7708c165 37#include <scsi/scsi_common.h>
c829c394 38#include <scsi/scsi_transport.h>
1da177e4
LT
39#include <scsi/scsi_host.h>
40#include <scsi/scsi_ioctl.h>
ee14c674 41#include <scsi/scsi_dh.h>
29cfc2ab 42#include <scsi/scsi_devinfo.h>
176aa9d6 43#include <scsi/sg.h>
1da177e4
LT
44
45#include "scsi_priv.h"
46#include "scsi_logging.h"
79ee8304 47#include "scsi_transport_api.h"
1da177e4 48
bf816235
KT
49#include <trace/events/scsi.h>
50
2908769c
DLM
51#include <asm/unaligned.h>
52
1da177e4
LT
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
3eef6257 60static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
b8e162f9
BVA
61static enum scsi_disposition scsi_try_to_abort_cmd(struct scsi_host_template *,
62 struct scsi_cmnd *);
3eef6257 63
1da177e4
LT
64void scsi_eh_wakeup(struct Scsi_Host *shost)
65{
f0317e88
BVA
66 lockdep_assert_held(shost->host_lock);
67
c84b023a 68 if (scsi_host_busy(shost) == shost->host_failed) {
bf816235 69 trace_scsi_eh_wakeup(shost);
3ed7a470 70 wake_up_process(shost->ehandler);
91921e01
HR
71 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
72 "Waking error handler thread\n"));
1da177e4
LT
73 }
74}
f8bbfc24
TH
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.
dc8875e1 81 */
f8bbfc24
TH
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);
1da177e4 97
b4562022
HR
98static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
99{
bb3b621a 100 if (!shost->last_reset || shost->eh_deadline == -1)
b4562022
HR
101 return 0;
102
76ad3e59
HR
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
bb3b621a 107 * if eh_deadline has been set to 'off' during the
76ad3e59
HR
108 * time_before call.
109 */
110 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
bb3b621a 111 shost->eh_deadline > -1)
b4562022
HR
112 return 0;
113
114 return 1;
115}
116
2a242d59
MC
117static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
118{
119 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
120 return true;
121
122 return ++cmd->retries <= cmd->allowed;
123}
124
60bee27b
MK
125static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd)
126{
127 struct scsi_device *sdev = cmd->device;
128 struct Scsi_Host *host = sdev->host;
129
130 if (host->hostt->eh_should_retry_cmd)
131 return host->hostt->eh_should_retry_cmd(cmd);
132
133 return true;
134}
135
e494f6a7
HR
136/**
137 * scmd_eh_abort_handler - Handle command aborts
138 * @work: command to be aborted.
923f46f9
BVA
139 *
140 * Note: this function must be called only for a command that has timed out.
141 * Because the block layer marks a request as complete before it calls
142 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
143 * timed out do not have any effect. Hence it is safe to call
144 * scsi_finish_command() from this function.
e494f6a7
HR
145 */
146void
147scmd_eh_abort_handler(struct work_struct *work)
148{
149 struct scsi_cmnd *scmd =
150 container_of(work, struct scsi_cmnd, abort_work.work);
151 struct scsi_device *sdev = scmd->device;
54d816d3 152 struct Scsi_Host *shost = sdev->host;
b8e162f9 153 enum scsi_disposition rtn;
5ae17501 154 unsigned long flags;
e494f6a7 155
54d816d3 156 if (scsi_host_eh_past_deadline(shost)) {
e494f6a7
HR
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
470613b4 159 "eh timeout, not aborting\n"));
54d816d3
EM
160 goto out;
161 }
162
163 SCSI_LOG_ERROR_RECOVERY(3,
e494f6a7 164 scmd_printk(KERN_INFO, scmd,
470613b4 165 "aborting command\n"));
54d816d3
EM
166 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
167 if (rtn != SUCCESS) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
170 "cmd abort %s\n",
171 (rtn == FAST_IO_FAIL) ?
172 "not send" : "failed"));
173 goto out;
174 }
175 set_host_byte(scmd, DID_TIME_OUT);
176 if (scsi_host_eh_past_deadline(shost)) {
177 SCSI_LOG_ERROR_RECOVERY(3,
178 scmd_printk(KERN_INFO, scmd,
179 "eh timeout, not retrying "
180 "aborted command\n"));
181 goto out;
e494f6a7
HR
182 }
183
54d816d3 184 spin_lock_irqsave(shost->host_lock, flags);
5ae17501 185 list_del_init(&scmd->eh_entry);
54d816d3
EM
186
187 /*
188 * If the abort succeeds, and there is no further
189 * EH action, clear the ->last_reset time.
190 */
191 if (list_empty(&shost->eh_abort_list) &&
192 list_empty(&shost->eh_cmd_q))
193 if (shost->eh_deadline != -1)
194 shost->last_reset = 0;
195
196 spin_unlock_irqrestore(shost->host_lock, flags);
197
198 if (!scsi_noretry_cmd(scmd) &&
199 scsi_cmd_retry_allowed(scmd) &&
200 scsi_eh_should_retry_cmd(scmd)) {
201 SCSI_LOG_ERROR_RECOVERY(3,
202 scmd_printk(KERN_WARNING, scmd,
203 "retry aborted command\n"));
204 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
205 } else {
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_WARNING, scmd,
208 "finish aborted command\n"));
209 scsi_finish_command(scmd);
210 }
211 return;
212
213out:
214 spin_lock_irqsave(shost->host_lock, flags);
215 list_del_init(&scmd->eh_entry);
216 spin_unlock_irqrestore(shost->host_lock, flags);
217
a0658632 218 scsi_eh_scmd_add(scmd);
e494f6a7
HR
219}
220
221/**
222 * scsi_abort_command - schedule a command abort
223 * @scmd: scmd to abort.
224 *
225 * We only need to abort commands after a command timeout
226 */
227static int
228scsi_abort_command(struct scsi_cmnd *scmd)
229{
230 struct scsi_device *sdev = scmd->device;
231 struct Scsi_Host *shost = sdev->host;
232 unsigned long flags;
233
234 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
235 /*
236 * Retry after abort failed, escalate to next level.
237 */
238 SCSI_LOG_ERROR_RECOVERY(3,
239 scmd_printk(KERN_INFO, scmd,
470613b4 240 "previous abort failed\n"));
fcc95a76 241 BUG_ON(delayed_work_pending(&scmd->abort_work));
e494f6a7
HR
242 return FAILED;
243 }
244
e494f6a7 245 spin_lock_irqsave(shost->host_lock, flags);
bb3b621a 246 if (shost->eh_deadline != -1 && !shost->last_reset)
e494f6a7 247 shost->last_reset = jiffies;
5ae17501
EM
248 BUG_ON(!list_empty(&scmd->eh_entry));
249 list_add_tail(&scmd->eh_entry, &shost->eh_abort_list);
e494f6a7
HR
250 spin_unlock_irqrestore(shost->host_lock, flags);
251
252 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
253 SCSI_LOG_ERROR_RECOVERY(3,
470613b4 254 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
e494f6a7
HR
255 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
256 return SUCCESS;
257}
258
1da177e4 259/**
7a38dc0b 260 * scsi_eh_reset - call into ->eh_action to reset internal counters
1da177e4 261 * @scmd: scmd to run eh on.
1da177e4 262 *
7a38dc0b
HR
263 * The scsi driver might be carrying internal state about the
264 * devices, so we need to call into the driver to reset the
265 * internal state once the error handler is started.
dc8875e1 266 */
7a38dc0b
HR
267static void scsi_eh_reset(struct scsi_cmnd *scmd)
268{
aa8e25e5 269 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
7a38dc0b
HR
270 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
271 if (sdrv->eh_reset)
272 sdrv->eh_reset(scmd);
273 }
274}
275
3bd6f43f
BVA
276static void scsi_eh_inc_host_failed(struct rcu_head *head)
277{
3be8828f
BVA
278 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
279 struct Scsi_Host *shost = scmd->device->host;
3bd6f43f
BVA
280 unsigned long flags;
281
282 spin_lock_irqsave(shost->host_lock, flags);
283 shost->host_failed++;
284 scsi_eh_wakeup(shost);
285 spin_unlock_irqrestore(shost->host_lock, flags);
286}
287
1da177e4
LT
288/**
289 * scsi_eh_scmd_add - add scsi cmd to error handling.
290 * @scmd: scmd to run eh on.
dc8875e1 291 */
a0658632 292void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
1da177e4
LT
293{
294 struct Scsi_Host *shost = scmd->device->host;
295 unsigned long flags;
2171b6d0 296 int ret;
1da177e4 297
2171b6d0 298 WARN_ON_ONCE(!shost->ehandler);
1da177e4
LT
299
300 spin_lock_irqsave(shost->host_lock, flags);
2171b6d0
HR
301 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
302 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
303 WARN_ON_ONCE(ret);
304 }
bb3b621a 305 if (shost->eh_deadline != -1 && !shost->last_reset)
b4562022
HR
306 shost->last_reset = jiffies;
307
7a38dc0b 308 scsi_eh_reset(scmd);
1da177e4 309 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
1da177e4 310 spin_unlock_irqrestore(shost->host_lock, flags);
3bd6f43f
BVA
311 /*
312 * Ensure that all tasks observe the host state change before the
313 * host_failed change.
314 */
3be8828f 315 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
1da177e4
LT
316}
317
1da177e4
LT
318/**
319 * scsi_times_out - Timeout function for normal scsi commands.
242f9dcb 320 * @req: request that is timing out.
1da177e4
LT
321 *
322 * Notes:
323 * We do not need to lock this. There is the potential for a race
324 * only in that the normal completion handling might run, but if the
325 * normal completion function determines that the timer has already
326 * fired, then it mustn't do anything.
dc8875e1 327 */
242f9dcb 328enum blk_eh_timer_return scsi_times_out(struct request *req)
1da177e4 329{
bed2213d 330 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
6600593c 331 enum blk_eh_timer_return rtn = BLK_EH_DONE;
0bf8c869 332 struct Scsi_Host *host = scmd->device->host;
6c5f8ce1 333
bf816235 334 trace_scsi_dispatch_cmd_timeout(scmd);
1da177e4
LT
335 scsi_log_completion(scmd, TIMEOUT_ERROR);
336
bb3b621a 337 if (host->eh_deadline != -1 && !host->last_reset)
b4562022
HR
338 host->last_reset = jiffies;
339
b6a05c82 340 if (host->hostt->eh_timed_out)
0bf8c869 341 rtn = host->hostt->eh_timed_out(scmd);
1da177e4 342
6600593c 343 if (rtn == BLK_EH_DONE) {
065990bd 344 /*
f1342709
KB
345 * Set the command to complete first in order to prevent a real
346 * completion from releasing the command while error handling
347 * is using it. If the command was already completed, then the
348 * lower level driver beat the timeout handler, and it is safe
349 * to return without escalating error recovery.
065990bd 350 *
f1342709
KB
351 * If timeout handling lost the race to a real completion, the
352 * block layer may ignore that due to a fake timeout injection,
353 * so return RESET_TIMER to allow error handling another shot
354 * at this command.
065990bd 355 */
f1342709
KB
356 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
357 return BLK_EH_RESET_TIMER;
a0658632 358 if (scsi_abort_command(scmd) != SUCCESS) {
2171b6d0 359 set_host_byte(scmd, DID_TIME_OUT);
a0658632 360 scsi_eh_scmd_add(scmd);
2171b6d0 361 }
a33c070b 362 }
242f9dcb 363
fa990781 364 return rtn;
1da177e4
LT
365}
366
367/**
368 * scsi_block_when_processing_errors - Prevent cmds from being queued.
369 * @sdev: Device on which we are performing recovery.
370 *
371 * Description:
372 * We block until the host is out of error recovery, and then check to
373 * see whether the host or the device is offline.
374 *
375 * Return value:
376 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
dc8875e1 377 */
1da177e4
LT
378int scsi_block_when_processing_errors(struct scsi_device *sdev)
379{
380 int online;
381
939647ee 382 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
1da177e4
LT
383
384 online = scsi_device_online(sdev);
385
1da177e4
LT
386 return online;
387}
388EXPORT_SYMBOL(scsi_block_when_processing_errors);
389
390#ifdef CONFIG_SCSI_LOGGING
391/**
392 * scsi_eh_prt_fail_stats - Log info on failures.
393 * @shost: scsi host being recovered.
394 * @work_q: Queue of scsi cmds to process.
dc8875e1 395 */
1da177e4
LT
396static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
397 struct list_head *work_q)
398{
399 struct scsi_cmnd *scmd;
400 struct scsi_device *sdev;
401 int total_failures = 0;
402 int cmd_failed = 0;
403 int cmd_cancel = 0;
404 int devices_failed = 0;
405
406 shost_for_each_device(sdev, shost) {
407 list_for_each_entry(scmd, work_q, eh_entry) {
408 if (scmd->device == sdev) {
409 ++total_failures;
a0658632 410 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
1da177e4 411 ++cmd_cancel;
0bf8c869 412 else
1da177e4
LT
413 ++cmd_failed;
414 }
415 }
416
417 if (cmd_cancel || cmd_failed) {
418 SCSI_LOG_ERROR_RECOVERY(3,
a3a790dc 419 shost_printk(KERN_INFO, shost,
9ccfc756 420 "%s: cmds failed: %d, cancel: %d\n",
cadbd4a5 421 __func__, cmd_failed,
9ccfc756 422 cmd_cancel));
1da177e4
LT
423 cmd_cancel = 0;
424 cmd_failed = 0;
425 ++devices_failed;
426 }
427 }
428
91921e01
HR
429 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
430 "Total of %d commands on %d"
431 " devices require eh work\n",
0bf8c869 432 total_failures, devices_failed));
1da177e4
LT
433}
434#endif
435
279afdfe
EM
436 /**
437 * scsi_report_lun_change - Set flag on all *other* devices on the same target
438 * to indicate that a UNIT ATTENTION is expected.
439 * @sdev: Device reporting the UNIT ATTENTION
440 */
441static void scsi_report_lun_change(struct scsi_device *sdev)
442{
443 sdev->sdev_target->expecting_lun_change = 1;
444}
445
446/**
447 * scsi_report_sense - Examine scsi sense information and log messages for
448 * certain conditions, also issue uevents for some of them.
449 * @sdev: Device reporting the sense code
450 * @sshdr: sshdr to be examined
451 */
452static void scsi_report_sense(struct scsi_device *sdev,
453 struct scsi_sense_hdr *sshdr)
454{
455 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
456
457 if (sshdr->sense_key == UNIT_ATTENTION) {
458 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
459 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
460 sdev_printk(KERN_WARNING, sdev,
461 "Inquiry data has changed");
462 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
463 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
464 scsi_report_lun_change(sdev);
465 sdev_printk(KERN_WARNING, sdev,
466 "Warning! Received an indication that the "
467 "LUN assignments on this target have "
468 "changed. The Linux SCSI layer does not "
469 "automatically remap LUN assignments.\n");
470 } else if (sshdr->asc == 0x3f)
471 sdev_printk(KERN_WARNING, sdev,
472 "Warning! Received an indication that the "
473 "operating parameters on this target have "
474 "changed. The Linux SCSI layer does not "
475 "automatically adjust these parameters.\n");
476
477 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
478 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
479 sdev_printk(KERN_WARNING, sdev,
480 "Warning! Received an indication that the "
481 "LUN reached a thin provisioning soft "
482 "threshold.\n");
483 }
484
cf3431bb
HR
485 if (sshdr->asc == 0x29) {
486 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
487 sdev_printk(KERN_WARNING, sdev,
488 "Power-on or device reset occurred\n");
489 }
490
279afdfe
EM
491 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
492 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
493 sdev_printk(KERN_WARNING, sdev,
494 "Mode parameters changed");
14c3e677
HR
495 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
496 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
497 sdev_printk(KERN_WARNING, sdev,
498 "Asymmetric access state changed");
279afdfe
EM
499 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
500 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
501 sdev_printk(KERN_WARNING, sdev,
502 "Capacity data has changed");
503 } else if (sshdr->asc == 0x2a)
504 sdev_printk(KERN_WARNING, sdev,
505 "Parameters changed");
506 }
507
508 if (evt_type != SDEV_EVT_MAXBITS) {
509 set_bit(evt_type, sdev->pending_events);
510 schedule_work(&sdev->event_work);
511 }
512}
513
1da177e4
LT
514/**
515 * scsi_check_sense - Examine scsi cmd sense
516 * @scmd: Cmd to have sense checked.
517 *
518 * Return value:
87f14e65 519 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1da177e4
LT
520 *
521 * Notes:
522 * When a deferred error is detected the current command has
523 * not been executed and needs retrying.
dc8875e1 524 */
b8e162f9 525enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd)
1da177e4 526{
a6a8d9f8 527 struct scsi_device *sdev = scmd->device;
1da177e4
LT
528 struct scsi_sense_hdr sshdr;
529
530 if (! scsi_command_normalize_sense(scmd, &sshdr))
531 return FAILED; /* no valid sense data */
532
279afdfe
EM
533 scsi_report_sense(sdev, &sshdr);
534
1da177e4
LT
535 if (scsi_sense_is_deferred(&sshdr))
536 return NEEDS_RETRY;
537
ee14c674 538 if (sdev->handler && sdev->handler->check_sense) {
b8e162f9 539 enum scsi_disposition rc;
a6a8d9f8 540
ee14c674 541 rc = sdev->handler->check_sense(sdev, &sshdr);
a6a8d9f8
CS
542 if (rc != SCSI_RETURN_NOT_HANDLED)
543 return rc;
544 /* handler does not care. Drop down to default handling */
545 }
546
bf23e619
BVA
547 if (scmd->cmnd[0] == TEST_UNIT_READY &&
548 scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER)
e925cc43
CH
549 /*
550 * nasty: for mid-layer issued TURs, we need to return the
551 * actual sense data without any recovery attempt. For eh
552 * issued ones, we need to try to recover and interpret
553 */
554 return SUCCESS;
555
1da177e4
LT
556 /*
557 * Previous logic looked for FILEMARK, EOM or ILI which are
558 * mainly associated with tapes and returned SUCCESS.
559 */
560 if (sshdr.response_code == 0x70) {
561 /* fixed format */
562 if (scmd->sense_buffer[2] & 0xe0)
563 return SUCCESS;
564 } else {
565 /*
566 * descriptor format: look for "stream commands sense data
567 * descriptor" (see SSC-3). Assume single sense data
568 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
569 */
570 if ((sshdr.additional_length > 3) &&
571 (scmd->sense_buffer[8] == 0x4) &&
572 (scmd->sense_buffer[11] & 0xe0))
573 return SUCCESS;
574 }
575
576 switch (sshdr.sense_key) {
577 case NO_SENSE:
578 return SUCCESS;
579 case RECOVERED_ERROR:
580 return /* soft_error */ SUCCESS;
581
582 case ABORTED_COMMAND:
511e44f4
MP
583 if (sshdr.asc == 0x10) /* DIF */
584 return SUCCESS;
585
29cfc2ab
MW
586 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
587 return ADD_TO_MLQUEUE;
c3606520
MW
588 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
589 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
590 return ADD_TO_MLQUEUE;
29cfc2ab 591
1da177e4
LT
592 return NEEDS_RETRY;
593 case NOT_READY:
594 case UNIT_ATTENTION:
595 /*
596 * if we are expecting a cc/ua because of a bus reset that we
597 * performed, treat this just as a retry. otherwise this is
598 * information that we should pass up to the upper-level driver
599 * so that we can deal with it there.
600 */
601 if (scmd->device->expecting_cc_ua) {
dfcf7775
TH
602 /*
603 * Because some device does not queue unit
604 * attentions correctly, we carefully check
605 * additional sense code and qualifier so as
606 * not to squash media change unit attention.
607 */
608 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
609 scmd->device->expecting_cc_ua = 0;
610 return NEEDS_RETRY;
611 }
1da177e4 612 }
279afdfe
EM
613 /*
614 * we might also expect a cc/ua if another LUN on the target
615 * reported a UA with an ASC/ASCQ of 3F 0E -
616 * REPORTED LUNS DATA HAS CHANGED.
617 */
618 if (scmd->device->sdev_target->expecting_lun_change &&
619 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
620 return NEEDS_RETRY;
1da177e4 621 /*
0bf8c869 622 * if the device is in the process of becoming ready, we
1da177e4
LT
623 * should retry.
624 */
625 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
626 return NEEDS_RETRY;
627 /*
628 * if the device is not started, we need to wake
629 * the error handler to start the motor
630 */
631 if (scmd->device->allow_restart &&
632 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
633 return FAILED;
02e031cb
CH
634 /*
635 * Pass the UA upwards for a determination in the completion
636 * functions.
637 */
638 return SUCCESS;
1da177e4 639
63583cca 640 /* these are not supported */
a9d6ceb8
HR
641 case DATA_PROTECT:
642 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
643 /* Thin provisioning hard threshold reached */
644 set_host_byte(scmd, DID_ALLOC_FAILURE);
645 return SUCCESS;
646 }
df561f66 647 fallthrough;
1da177e4
LT
648 case COPY_ABORTED:
649 case VOLUME_OVERFLOW:
650 case MISCOMPARE:
63583cca 651 case BLANK_CHECK:
87f14e65
HR
652 set_host_byte(scmd, DID_TARGET_FAILURE);
653 return SUCCESS;
1da177e4
LT
654
655 case MEDIUM_ERROR:
fd1b494d
LT
656 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
657 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
658 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
7e782af5 659 set_host_byte(scmd, DID_MEDIUM_ERROR);
87f14e65 660 return SUCCESS;
fd1b494d 661 }
1da177e4
LT
662 return NEEDS_RETRY;
663
664 case HARDWARE_ERROR:
665 if (scmd->device->retry_hwerror)
bb0003c1 666 return ADD_TO_MLQUEUE;
1da177e4 667 else
87f14e65 668 set_host_byte(scmd, DID_TARGET_FAILURE);
df561f66 669 fallthrough;
1da177e4
LT
670
671 case ILLEGAL_REQUEST:
47ac56db
MS
672 if (sshdr.asc == 0x20 || /* Invalid command operation code */
673 sshdr.asc == 0x21 || /* Logical block address out of range */
a8bbb2ab 674 sshdr.asc == 0x22 || /* Invalid function */
47ac56db 675 sshdr.asc == 0x24 || /* Invalid field in cdb */
d0b7a909
MW
676 sshdr.asc == 0x26 || /* Parameter value invalid */
677 sshdr.asc == 0x27) { /* Write protected */
87f14e65 678 set_host_byte(scmd, DID_TARGET_FAILURE);
47ac56db
MS
679 }
680 return SUCCESS;
681
1da177e4
LT
682 default:
683 return SUCCESS;
684 }
685}
3852e373 686EXPORT_SYMBOL_GPL(scsi_check_sense);
1da177e4 687
4a84067d
VD
688static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
689{
690 struct scsi_host_template *sht = sdev->host->hostt;
691 struct scsi_device *tmp_sdev;
692
c40ecc12 693 if (!sht->track_queue_depth ||
4a84067d
VD
694 sdev->queue_depth >= sdev->max_queue_depth)
695 return;
696
697 if (time_before(jiffies,
698 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
699 return;
700
701 if (time_before(jiffies,
702 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
703 return;
704
705 /*
706 * Walk all devices of a target and do
707 * ramp up on them.
708 */
709 shost_for_each_device(tmp_sdev, sdev->host) {
710 if (tmp_sdev->channel != sdev->channel ||
711 tmp_sdev->id != sdev->id ||
712 tmp_sdev->queue_depth == sdev->max_queue_depth)
713 continue;
c40ecc12 714
db5ed4df 715 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
4a84067d
VD
716 sdev->last_queue_ramp_up = jiffies;
717 }
718}
719
42a6a918
MC
720static void scsi_handle_queue_full(struct scsi_device *sdev)
721{
722 struct scsi_host_template *sht = sdev->host->hostt;
723 struct scsi_device *tmp_sdev;
724
c40ecc12 725 if (!sht->track_queue_depth)
42a6a918
MC
726 return;
727
728 shost_for_each_device(tmp_sdev, sdev->host) {
729 if (tmp_sdev->channel != sdev->channel ||
730 tmp_sdev->id != sdev->id)
731 continue;
732 /*
733 * We do not know the number of commands that were at
734 * the device when we got the queue full so we start
735 * from the highest possible value and work our way down.
736 */
c40ecc12 737 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
42a6a918
MC
738 }
739}
740
1da177e4
LT
741/**
742 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
743 * @scmd: SCSI cmd to examine.
744 *
745 * Notes:
746 * This is *only* called when we are examining the status of commands
747 * queued during error recovery. the main difference here is that we
748 * don't allow for the possibility of retries here, and we are a lot
749 * more restrictive about what we consider acceptable.
dc8875e1 750 */
b8e162f9 751static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd)
1da177e4
LT
752{
753 /*
754 * first check the host byte, to see if there is anything in there
755 * that would indicate what we need to do.
756 */
757 if (host_byte(scmd->result) == DID_RESET) {
758 /*
759 * rats. we are already in the error handler, so we now
760 * get to try and figure out what to do next. if the sense
761 * is valid, we have a pretty good idea of what to do.
762 * if not, we mark it as FAILED.
763 */
764 return scsi_check_sense(scmd);
765 }
766 if (host_byte(scmd->result) != DID_OK)
767 return FAILED;
768
1da177e4
LT
769 /*
770 * now, check the status byte to see if this indicates
771 * anything special.
772 */
3d45cefc
HR
773 switch (get_status_byte(scmd)) {
774 case SAM_STAT_GOOD:
4a84067d 775 scsi_handle_queue_ramp_up(scmd->device);
df561f66 776 fallthrough;
3d45cefc 777 case SAM_STAT_COMMAND_TERMINATED:
1da177e4 778 return SUCCESS;
3d45cefc 779 case SAM_STAT_CHECK_CONDITION:
1da177e4 780 return scsi_check_sense(scmd);
3d45cefc
HR
781 case SAM_STAT_CONDITION_MET:
782 case SAM_STAT_INTERMEDIATE:
783 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1da177e4
LT
784 /*
785 * who knows? FIXME(eric)
786 */
787 return SUCCESS;
3d45cefc 788 case SAM_STAT_RESERVATION_CONFLICT:
67110dfd
JB
789 if (scmd->cmnd[0] == TEST_UNIT_READY)
790 /* it is a success, we probed the device and
791 * found it */
792 return SUCCESS;
793 /* otherwise, we failed to send the command */
794 return FAILED;
3d45cefc 795 case SAM_STAT_TASK_SET_FULL:
42a6a918 796 scsi_handle_queue_full(scmd->device);
df561f66 797 fallthrough;
3d45cefc 798 case SAM_STAT_BUSY:
3eb3a928 799 return NEEDS_RETRY;
1da177e4
LT
800 default:
801 return FAILED;
802 }
803 return FAILED;
804}
805
1da177e4
LT
806/**
807 * scsi_eh_done - Completion function for error handling.
808 * @scmd: Cmd that is done.
dc8875e1 809 */
bf23e619 810void scsi_eh_done(struct scsi_cmnd *scmd)
1da177e4 811{
0bf8c869 812 struct completion *eh_action;
85631672 813
91921e01 814 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4 815 "%s result: %x\n", __func__, scmd->result));
85631672
MR
816
817 eh_action = scmd->device->host->eh_action;
818 if (eh_action)
819 complete(eh_action);
1da177e4
LT
820}
821
292148f8
BK
822/**
823 * scsi_try_host_reset - ask host adapter to reset itself
c2b3ebd0 824 * @scmd: SCSI cmd to send host reset.
dc8875e1 825 */
b8e162f9 826static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
292148f8
BK
827{
828 unsigned long flags;
b8e162f9 829 enum scsi_disposition rtn;
0bf8c869
JJ
830 struct Scsi_Host *host = scmd->device->host;
831 struct scsi_host_template *hostt = host->hostt;
292148f8 832
91921e01
HR
833 SCSI_LOG_ERROR_RECOVERY(3,
834 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
292148f8 835
0bf8c869 836 if (!hostt->eh_host_reset_handler)
292148f8
BK
837 return FAILED;
838
0bf8c869 839 rtn = hostt->eh_host_reset_handler(scmd);
292148f8
BK
840
841 if (rtn == SUCCESS) {
0bf8c869 842 if (!hostt->skip_settle_delay)
292148f8 843 ssleep(HOST_RESET_SETTLE_TIME);
0bf8c869
JJ
844 spin_lock_irqsave(host->host_lock, flags);
845 scsi_report_bus_reset(host, scmd_channel(scmd));
846 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
847 }
848
849 return rtn;
850}
851
852/**
853 * scsi_try_bus_reset - ask host to perform a bus reset
854 * @scmd: SCSI cmd to send bus reset.
dc8875e1 855 */
b8e162f9 856static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
292148f8
BK
857{
858 unsigned long flags;
b8e162f9 859 enum scsi_disposition rtn;
0bf8c869
JJ
860 struct Scsi_Host *host = scmd->device->host;
861 struct scsi_host_template *hostt = host->hostt;
292148f8 862
91921e01
HR
863 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
864 "%s: Snd Bus RST\n", __func__));
292148f8 865
0bf8c869 866 if (!hostt->eh_bus_reset_handler)
292148f8
BK
867 return FAILED;
868
0bf8c869 869 rtn = hostt->eh_bus_reset_handler(scmd);
292148f8
BK
870
871 if (rtn == SUCCESS) {
0bf8c869 872 if (!hostt->skip_settle_delay)
292148f8 873 ssleep(BUS_RESET_SETTLE_TIME);
0bf8c869
JJ
874 spin_lock_irqsave(host->host_lock, flags);
875 scsi_report_bus_reset(host, scmd_channel(scmd));
876 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
877 }
878
879 return rtn;
880}
881
30bd7df8
MC
882static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
883{
884 sdev->was_reset = 1;
885 sdev->expecting_cc_ua = 1;
886}
887
888/**
889 * scsi_try_target_reset - Ask host to perform a target reset
890 * @scmd: SCSI cmd used to send a target reset
891 *
892 * Notes:
893 * There is no timeout for this operation. if this operation is
894 * unreliable for a given host, then the host itself needs to put a
895 * timer on it, and set the host back to a consistent state prior to
896 * returning.
897 */
b8e162f9 898static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
30bd7df8
MC
899{
900 unsigned long flags;
b8e162f9 901 enum scsi_disposition rtn;
0bf8c869
JJ
902 struct Scsi_Host *host = scmd->device->host;
903 struct scsi_host_template *hostt = host->hostt;
30bd7df8 904
0bf8c869 905 if (!hostt->eh_target_reset_handler)
30bd7df8
MC
906 return FAILED;
907
0bf8c869 908 rtn = hostt->eh_target_reset_handler(scmd);
30bd7df8 909 if (rtn == SUCCESS) {
0bf8c869 910 spin_lock_irqsave(host->host_lock, flags);
30bd7df8
MC
911 __starget_for_each_device(scsi_target(scmd->device), NULL,
912 __scsi_report_device_reset);
0bf8c869 913 spin_unlock_irqrestore(host->host_lock, flags);
30bd7df8
MC
914 }
915
916 return rtn;
917}
918
292148f8
BK
919/**
920 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
921 * @scmd: SCSI cmd used to send BDR
922 *
923 * Notes:
924 * There is no timeout for this operation. if this operation is
925 * unreliable for a given host, then the host itself needs to put a
926 * timer on it, and set the host back to a consistent state prior to
927 * returning.
dc8875e1 928 */
b8e162f9 929static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
292148f8 930{
b8e162f9 931 enum scsi_disposition rtn;
0bf8c869 932 struct scsi_host_template *hostt = scmd->device->host->hostt;
292148f8 933
0bf8c869 934 if (!hostt->eh_device_reset_handler)
292148f8
BK
935 return FAILED;
936
0bf8c869 937 rtn = hostt->eh_device_reset_handler(scmd);
30bd7df8
MC
938 if (rtn == SUCCESS)
939 __scsi_report_device_reset(scmd->device, NULL);
292148f8
BK
940 return rtn;
941}
942
883a030f
HR
943/**
944 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
6583f6fb 945 * @hostt: SCSI driver host template
883a030f
HR
946 * @scmd: SCSI cmd used to send a target reset
947 *
948 * Return value:
949 * SUCCESS, FAILED, or FAST_IO_FAIL
950 *
951 * Notes:
952 * SUCCESS does not necessarily indicate that the command
953 * has been aborted; it only indicates that the LLDDs
954 * has cleared all references to that command.
955 * LLDDs should return FAILED only if an abort was required
956 * but could not be executed. LLDDs should return FAST_IO_FAIL
957 * if the device is temporarily unavailable (eg due to a
958 * link down on FibreChannel)
959 */
b8e162f9
BVA
960static enum scsi_disposition
961scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
292148f8 962{
0bf8c869 963 if (!hostt->eh_abort_handler)
292148f8
BK
964 return FAILED;
965
0bf8c869 966 return hostt->eh_abort_handler(scmd);
292148f8
BK
967}
968
292148f8
BK
969static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
970{
0bf8c869 971 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
292148f8 972 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
30bd7df8
MC
973 if (scsi_try_target_reset(scmd) != SUCCESS)
974 if (scsi_try_bus_reset(scmd) != SUCCESS)
975 scsi_try_host_reset(scmd);
292148f8
BK
976}
977
1da177e4 978/**
3b729f76 979 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
2dc611de 980 * @scmd: SCSI command structure to hijack
e1c23468 981 * @ses: structure to save restore information
55db6c1b 982 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
64a87b24 983 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
55db6c1b 984 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
2dc611de 985 *
e1c23468 986 * This function is used to save a scsi command information before re-execution
55db6c1b
BH
987 * as part of the error recovery process. If @sense_bytes is 0 the command
988 * sent must be one that does not transfer any data. If @sense_bytes != 0
989 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
990 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
dc8875e1 991 */
e1c23468
BH
992void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
993 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1da177e4 994{
f59114b7 995 struct scsi_device *sdev = scmd->device;
1da177e4 996
631c228c
CH
997 /*
998 * We need saved copies of a number of fields - this is because
999 * error handling may need to overwrite these with different values
1000 * to run different commands, and once error handling is complete,
1001 * we will need to restore these values prior to running the actual
1002 * command.
1003 */
e1c23468 1004 ses->cmd_len = scmd->cmd_len;
64a87b24 1005 ses->cmnd = scmd->cmnd;
e1c23468 1006 ses->data_direction = scmd->sc_data_direction;
30b0c37b 1007 ses->sdb = scmd->sdb;
e1c23468 1008 ses->result = scmd->result;
8f8fed0c 1009 ses->resid_len = scmd->req.resid_len;
12265709 1010 ses->underflow = scmd->underflow;
db007fc5 1011 ses->prot_op = scmd->prot_op;
8e8c9d01 1012 ses->eh_eflags = scmd->eh_eflags;
631c228c 1013
db007fc5 1014 scmd->prot_op = SCSI_PROT_NORMAL;
c69e6f81 1015 scmd->eh_eflags = 0;
64a87b24
BH
1016 scmd->cmnd = ses->eh_cmnd;
1017 memset(scmd->cmnd, 0, BLK_MAX_CDB);
30b0c37b 1018 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
644373a4 1019 scmd->result = 0;
8f8fed0c 1020 scmd->req.resid_len = 0;
30b0c37b 1021
55db6c1b 1022 if (sense_bytes) {
30b0c37b
BH
1023 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
1024 sense_bytes);
e1c23468 1025 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
30b0c37b
BH
1026 scmd->sdb.length);
1027 scmd->sdb.table.sgl = &ses->sense_sgl;
55db6c1b 1028 scmd->sc_data_direction = DMA_FROM_DEVICE;
0c958ecc 1029 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
55db6c1b 1030 scmd->cmnd[0] = REQUEST_SENSE;
30b0c37b 1031 scmd->cmnd[4] = scmd->sdb.length;
55db6c1b 1032 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
631c228c 1033 } else {
631c228c 1034 scmd->sc_data_direction = DMA_NONE;
55db6c1b 1035 if (cmnd) {
64a87b24 1036 BUG_ON(cmnd_size > BLK_MAX_CDB);
55db6c1b
BH
1037 memcpy(scmd->cmnd, cmnd, cmnd_size);
1038 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1039 }
631c228c
CH
1040 }
1041
1042 scmd->underflow = 0;
631c228c 1043
55db6c1b 1044 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1da177e4 1045 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b7 1046 (sdev->lun << 5 & 0xe0);
1da177e4 1047
631c228c
CH
1048 /*
1049 * Zero the sense buffer. The scsi spec mandates that any
1050 * untransferred sense data should be interpreted as being zero.
1051 */
b80ca4f7 1052 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
e1c23468
BH
1053}
1054EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1055
1056/**
3b729f76 1057 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
e1c23468 1058 * @scmd: SCSI command structure to restore
477e608c 1059 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
e1c23468 1060 *
477e608c 1061 * Undo any damage done by above scsi_eh_prep_cmnd().
dc8875e1 1062 */
e1c23468
BH
1063void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1064{
1065 /*
1066 * Restore original data
1067 */
1068 scmd->cmd_len = ses->cmd_len;
64a87b24 1069 scmd->cmnd = ses->cmnd;
e1c23468 1070 scmd->sc_data_direction = ses->data_direction;
30b0c37b 1071 scmd->sdb = ses->sdb;
e1c23468 1072 scmd->result = ses->result;
8f8fed0c 1073 scmd->req.resid_len = ses->resid_len;
12265709 1074 scmd->underflow = ses->underflow;
db007fc5 1075 scmd->prot_op = ses->prot_op;
8e8c9d01 1076 scmd->eh_eflags = ses->eh_eflags;
e1c23468
BH
1077}
1078EXPORT_SYMBOL(scsi_eh_restore_cmnd);
631c228c 1079
e1c23468 1080/**
3b729f76 1081 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
e1c23468
BH
1082 * @scmd: SCSI command structure to hijack
1083 * @cmnd: CDB to send
1084 * @cmnd_size: size in bytes of @cmnd
1085 * @timeout: timeout for this request
1086 * @sense_bytes: size of sense data to copy or 0
1087 *
1088 * This function is used to send a scsi command down to a target device
1089 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1090 *
1091 * Return value:
1092 * SUCCESS or FAILED or NEEDS_RETRY
dc8875e1 1093 */
b8e162f9
BVA
1094static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
1095 unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes)
e1c23468
BH
1096{
1097 struct scsi_device *sdev = scmd->device;
1098 struct Scsi_Host *shost = sdev->host;
1099 DECLARE_COMPLETION_ONSTACK(done);
bbe9fb0d 1100 unsigned long timeleft = timeout, delay;
e1c23468 1101 struct scsi_eh_save ses;
fc73648a 1102 const unsigned long stall_for = msecs_to_jiffies(100);
e1c23468
BH
1103 int rtn;
1104
fc73648a 1105retry:
e1c23468 1106 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
7dfdc9a5 1107 shost->eh_action = &done;
1da177e4 1108
1da177e4 1109 scsi_log_send(scmd);
bf23e619 1110 scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
bbe9fb0d
BVA
1111
1112 /*
1113 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1114 * change the SCSI device state after we have examined it and before
1115 * .queuecommand() is called.
1116 */
1117 mutex_lock(&sdev->state_mutex);
1118 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1119 mutex_unlock(&sdev->state_mutex);
1120 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1121 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1122 SDEV_BLOCK));
1123 delay = min(timeleft, stall_for);
1124 timeleft -= delay;
1125 msleep(jiffies_to_msecs(delay));
1126 mutex_lock(&sdev->state_mutex);
1127 }
1128 if (sdev->sdev_state != SDEV_BLOCK)
1129 rtn = shost->hostt->queuecommand(shost, scmd);
1130 else
280e91b0 1131 rtn = FAILED;
bbe9fb0d
BVA
1132 mutex_unlock(&sdev->state_mutex);
1133
fc73648a
HR
1134 if (rtn) {
1135 if (timeleft > stall_for) {
1136 scsi_eh_restore_cmnd(scmd, &ses);
bf23e619 1137
fc73648a
HR
1138 timeleft -= stall_for;
1139 msleep(jiffies_to_msecs(stall_for));
1140 goto retry;
1141 }
1142 /* signal not to enter either branch of the if () below */
1143 timeleft = 0;
511833ac 1144 rtn = FAILED;
fc73648a
HR
1145 } else {
1146 timeleft = wait_for_completion_timeout(&done, timeout);
ac61d195 1147 rtn = SUCCESS;
fc73648a 1148 }
1da177e4 1149
f59114b7 1150 shost->eh_action = NULL;
1da177e4 1151
fc73648a 1152 scsi_log_completion(scmd, rtn);
1da177e4 1153
91921e01 1154 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4
HR
1155 "%s timeleft: %ld\n",
1156 __func__, timeleft));
1da177e4
LT
1157
1158 /*
fc73648a
HR
1159 * If there is time left scsi_eh_done got called, and we will examine
1160 * the actual status codes to see whether the command actually did
1161 * complete normally, else if we have a zero return and no time left,
1162 * the command must still be pending, so abort it and return FAILED.
1163 * If we never actually managed to issue the command, because
1164 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1165 * value above (so don't execute either branch of the if)
1da177e4 1166 */
7dfdc9a5 1167 if (timeleft) {
1da177e4 1168 rtn = scsi_eh_completed_normally(scmd);
91921e01
HR
1169 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1170 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
7dfdc9a5 1171
1da177e4
LT
1172 switch (rtn) {
1173 case SUCCESS:
1174 case NEEDS_RETRY:
1175 case FAILED:
1176 break;
6e883b0e
HR
1177 case ADD_TO_MLQUEUE:
1178 rtn = NEEDS_RETRY;
1179 break;
1da177e4
LT
1180 default:
1181 rtn = FAILED;
1182 break;
1183 }
511833ac 1184 } else if (rtn != FAILED) {
292148f8 1185 scsi_abort_eh_cmnd(scmd);
7dfdc9a5 1186 rtn = FAILED;
1da177e4
LT
1187 }
1188
e1c23468 1189 scsi_eh_restore_cmnd(scmd, &ses);
18a4d0a2 1190
1da177e4
LT
1191 return rtn;
1192}
1193
1194/**
1195 * scsi_request_sense - Request sense data from a particular target.
1196 * @scmd: SCSI cmd for request sense.
1197 *
1198 * Notes:
1199 * Some hosts automatically obtain this information, others require
1200 * that we obtain it on our own. This function will *not* return until
1201 * the command either times out, or it completes.
dc8875e1 1202 */
b8e162f9 1203static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd)
1da177e4 1204{
0816c925 1205 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1da177e4
LT
1206}
1207
b8e162f9
BVA
1208static enum scsi_disposition
1209scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn)
2451079b 1210{
aa8e25e5 1211 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
2451079b
JB
1212 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1213 if (sdrv->eh_action)
1214 rtn = sdrv->eh_action(scmd, rtn);
1215 }
1216 return rtn;
1217}
1218
1da177e4
LT
1219/**
1220 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1221 * @scmd: Original SCSI cmd that eh has finished.
1222 * @done_q: Queue for processed commands.
1223 *
1224 * Notes:
1225 * We don't want to use the normal command completion while we are are
1226 * still handling errors - it may cause other commands to be queued,
eb44820c 1227 * and that would disturb what we are doing. Thus we really want to
1da177e4
LT
1228 * keep a list of pending commands for final completion, and once we
1229 * are ready to leave error handling we handle completion for real.
dc8875e1 1230 */
041c5fc3 1231void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1da177e4 1232{
1da177e4
LT
1233 list_move_tail(&scmd->eh_entry, done_q);
1234}
041c5fc3 1235EXPORT_SYMBOL(scsi_eh_finish_cmd);
1da177e4
LT
1236
1237/**
1238 * scsi_eh_get_sense - Get device sense data.
1239 * @work_q: Queue of commands to process.
eb44820c 1240 * @done_q: Queue of processed commands.
1da177e4
LT
1241 *
1242 * Description:
1243 * See if we need to request sense information. if so, then get it
0bf8c869 1244 * now, so we have a better idea of what to do.
1da177e4
LT
1245 *
1246 * Notes:
1247 * This has the unfortunate side effect that if a shost adapter does
eb44820c 1248 * not automatically request sense information, we end up shutting
1da177e4
LT
1249 * it down before we request it.
1250 *
1251 * All drivers should request sense information internally these days,
1252 * so for now all I have to say is tough noogies if you end up in here.
1253 *
1254 * XXX: Long term this code should go away, but that needs an audit of
1255 * all LLDDs first.
dc8875e1 1256 */
dca84e46
DW
1257int scsi_eh_get_sense(struct list_head *work_q,
1258 struct list_head *done_q)
1da177e4 1259{
937abeaa 1260 struct scsi_cmnd *scmd, *next;
b4562022 1261 struct Scsi_Host *shost;
b8e162f9 1262 enum scsi_disposition rtn;
1da177e4 1263
709c75b5 1264 /*
1265 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1266 * should not get sense.
1267 */
937abeaa 1268 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
a0658632 1269 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1da177e4
LT
1270 SCSI_SENSE_VALID(scmd))
1271 continue;
1272
b4562022 1273 shost = scmd->device->host;
b4562022 1274 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1275 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1276 scmd_printk(KERN_INFO, scmd,
1277 "%s: skip request sense, past eh deadline\n",
1278 current->comm));
b4562022
HR
1279 break;
1280 }
d0672a03 1281 if (!scsi_status_is_check_condition(scmd->result))
d555a2ab
JB
1282 /*
1283 * don't request sense if there's no check condition
1284 * status because the error we're processing isn't one
1285 * that has a sense code (and some devices get
1286 * confused by sense requests out of the blue)
1287 */
1288 continue;
1289
3bf743e7
JG
1290 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1291 "%s: requesting sense\n",
1292 current->comm));
1da177e4
LT
1293 rtn = scsi_request_sense(scmd);
1294 if (rtn != SUCCESS)
1295 continue;
1296
91921e01 1297 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4 1298 "sense requested, result %x\n", scmd->result));
d811b848 1299 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1da177e4
LT
1300
1301 rtn = scsi_decide_disposition(scmd);
1302
1303 /*
1304 * if the result was normal, then just pass it along to the
1305 * upper level.
1306 */
1307 if (rtn == SUCCESS)
2a242d59
MC
1308 /*
1309 * We don't want this command reissued, just finished
1310 * with the sense data, so set retries to the max
1311 * allowed to ensure it won't get reissued. If the user
1312 * has requested infinite retries, we also want to
1313 * finish this command, so force completion by setting
1314 * retries and allowed to the same value.
1315 */
1316 if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1317 scmd->retries = scmd->allowed = 1;
1318 else
1319 scmd->retries = scmd->allowed;
1da177e4
LT
1320 else if (rtn != NEEDS_RETRY)
1321 continue;
1322
1323 scsi_eh_finish_cmd(scmd, done_q);
1324 }
1325
1326 return list_empty(work_q);
1327}
dca84e46 1328EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1da177e4 1329
1da177e4
LT
1330/**
1331 * scsi_eh_tur - Send TUR to device.
eb44820c 1332 * @scmd: &scsi_cmnd to send TUR
1da177e4
LT
1333 *
1334 * Return value:
1335 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1336 */
1da177e4
LT
1337static int scsi_eh_tur(struct scsi_cmnd *scmd)
1338{
1339 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
b8e162f9
BVA
1340 int retry_cnt = 1;
1341 enum scsi_disposition rtn;
1da177e4
LT
1342
1343retry_tur:
0816c925
MP
1344 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1345 scmd->device->eh_timeout, 0);
1da177e4 1346
91921e01 1347 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4 1348 "%s return: %x\n", __func__, rtn));
631c228c
CH
1349
1350 switch (rtn) {
1351 case NEEDS_RETRY:
1da177e4
LT
1352 if (retry_cnt--)
1353 goto retry_tur;
df561f66 1354 fallthrough;
631c228c 1355 case SUCCESS:
e47373ec 1356 return 0;
631c228c
CH
1357 default:
1358 return 1;
e47373ec 1359 }
1da177e4
LT
1360}
1361
3eef6257
DJ
1362/**
1363 * scsi_eh_test_devices - check if devices are responding from error recovery.
1364 * @cmd_list: scsi commands in error recovery.
74cf298f
RD
1365 * @work_q: queue for commands which still need more error recovery
1366 * @done_q: queue for commands which are finished
1367 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
3eef6257
DJ
1368 *
1369 * Decription:
1370 * Tests if devices are in a working state. Commands to devices now in
1371 * a working state are sent to the done_q while commands to devices which
1372 * are still failing to respond are returned to the work_q for more
1373 * processing.
1374 **/
1375static int scsi_eh_test_devices(struct list_head *cmd_list,
1376 struct list_head *work_q,
1377 struct list_head *done_q, int try_stu)
1378{
1379 struct scsi_cmnd *scmd, *next;
1380 struct scsi_device *sdev;
1381 int finish_cmds;
1382
1383 while (!list_empty(cmd_list)) {
1384 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1385 sdev = scmd->device;
1386
b4562022 1387 if (!try_stu) {
b4562022
HR
1388 if (scsi_host_eh_past_deadline(sdev->host)) {
1389 /* Push items back onto work_q */
1390 list_splice_init(cmd_list, work_q);
b4562022 1391 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1392 sdev_printk(KERN_INFO, sdev,
1393 "%s: skip test device, past eh deadline",
1394 current->comm));
b4562022
HR
1395 break;
1396 }
b4562022
HR
1397 }
1398
3eef6257
DJ
1399 finish_cmds = !scsi_device_online(scmd->device) ||
1400 (try_stu && !scsi_eh_try_stu(scmd) &&
1401 !scsi_eh_tur(scmd)) ||
1402 !scsi_eh_tur(scmd);
1403
1404 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1405 if (scmd->device == sdev) {
2451079b
JB
1406 if (finish_cmds &&
1407 (try_stu ||
1408 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
3eef6257
DJ
1409 scsi_eh_finish_cmd(scmd, done_q);
1410 else
1411 list_move_tail(&scmd->eh_entry, work_q);
1412 }
1413 }
1414 return list_empty(work_q);
1415}
1416
1da177e4
LT
1417/**
1418 * scsi_eh_try_stu - Send START_UNIT to device.
eb44820c 1419 * @scmd: &scsi_cmnd to send START_UNIT
1da177e4
LT
1420 *
1421 * Return value:
1422 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1423 */
1da177e4
LT
1424static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1425{
1426 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1da177e4 1427
631c228c 1428 if (scmd->device->allow_restart) {
b8e162f9
BVA
1429 int i;
1430 enum scsi_disposition rtn = NEEDS_RETRY;
ed773e66
BK
1431
1432 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
adcc796b
CX
1433 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
1434 scmd->device->eh_timeout, 0);
1da177e4 1435
631c228c
CH
1436 if (rtn == SUCCESS)
1437 return 0;
1438 }
1da177e4 1439
1da177e4
LT
1440 return 1;
1441}
1442
1443 /**
1444 * scsi_eh_stu - send START_UNIT if needed
eb44820c 1445 * @shost: &scsi host being recovered.
74cf298f 1446 * @work_q: &list_head for pending commands.
eb44820c 1447 * @done_q: &list_head for processed commands.
1da177e4
LT
1448 *
1449 * Notes:
1450 * If commands are failing due to not ready, initializing command required,
0bf8c869 1451 * try revalidating the device, which will end up sending a start unit.
dc8875e1 1452 */
1da177e4
LT
1453static int scsi_eh_stu(struct Scsi_Host *shost,
1454 struct list_head *work_q,
1455 struct list_head *done_q)
1456{
937abeaa 1457 struct scsi_cmnd *scmd, *stu_scmd, *next;
1da177e4
LT
1458 struct scsi_device *sdev;
1459
1460 shost_for_each_device(sdev, shost) {
b4562022 1461 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1462 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1463 sdev_printk(KERN_INFO, sdev,
1464 "%s: skip START_UNIT, past eh deadline\n",
1465 current->comm));
4dea170f 1466 scsi_device_put(sdev);
b4562022
HR
1467 break;
1468 }
1da177e4
LT
1469 stu_scmd = NULL;
1470 list_for_each_entry(scmd, work_q, eh_entry)
1471 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1472 scsi_check_sense(scmd) == FAILED ) {
1473 stu_scmd = scmd;
1474 break;
1475 }
1476
1477 if (!stu_scmd)
1478 continue;
1479
91921e01 1480 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1481 sdev_printk(KERN_INFO, sdev,
1482 "%s: Sending START_UNIT\n",
1483 current->comm));
1da177e4
LT
1484
1485 if (!scsi_eh_try_stu(stu_scmd)) {
1486 if (!scsi_device_online(sdev) ||
1487 !scsi_eh_tur(stu_scmd)) {
937abeaa
CH
1488 list_for_each_entry_safe(scmd, next,
1489 work_q, eh_entry) {
2451079b
JB
1490 if (scmd->device == sdev &&
1491 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1da177e4
LT
1492 scsi_eh_finish_cmd(scmd, done_q);
1493 }
1494 }
1495 } else {
1496 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1497 sdev_printk(KERN_INFO, sdev,
1498 "%s: START_UNIT failed\n",
1499 current->comm));
1da177e4
LT
1500 }
1501 }
1502
1503 return list_empty(work_q);
1504}
1505
1506
1507/**
1508 * scsi_eh_bus_device_reset - send bdr if needed
1509 * @shost: scsi host being recovered.
74cf298f 1510 * @work_q: &list_head for pending commands.
eb44820c 1511 * @done_q: &list_head for processed commands.
1da177e4
LT
1512 *
1513 * Notes:
eb44820c 1514 * Try a bus device reset. Still, look to see whether we have multiple
1da177e4
LT
1515 * devices that are jammed or not - if we have multiple devices, it
1516 * makes no sense to try bus_device_reset - we really would need to try
0bf8c869 1517 * a bus_reset instead.
dc8875e1 1518 */
1da177e4
LT
1519static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1520 struct list_head *work_q,
1521 struct list_head *done_q)
1522{
937abeaa 1523 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1da177e4 1524 struct scsi_device *sdev;
b8e162f9 1525 enum scsi_disposition rtn;
1da177e4
LT
1526
1527 shost_for_each_device(sdev, shost) {
b4562022 1528 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1529 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1530 sdev_printk(KERN_INFO, sdev,
1531 "%s: skip BDR, past eh deadline\n",
1532 current->comm));
4dea170f 1533 scsi_device_put(sdev);
b4562022
HR
1534 break;
1535 }
1da177e4
LT
1536 bdr_scmd = NULL;
1537 list_for_each_entry(scmd, work_q, eh_entry)
1538 if (scmd->device == sdev) {
1539 bdr_scmd = scmd;
1540 break;
1541 }
1542
1543 if (!bdr_scmd)
1544 continue;
1545
91921e01 1546 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1547 sdev_printk(KERN_INFO, sdev,
1548 "%s: Sending BDR\n", current->comm));
1da177e4 1549 rtn = scsi_try_bus_device_reset(bdr_scmd);
2f2eb587 1550 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1da177e4 1551 if (!scsi_device_online(sdev) ||
2f2eb587 1552 rtn == FAST_IO_FAIL ||
1da177e4 1553 !scsi_eh_tur(bdr_scmd)) {
937abeaa
CH
1554 list_for_each_entry_safe(scmd, next,
1555 work_q, eh_entry) {
2451079b
JB
1556 if (scmd->device == sdev &&
1557 scsi_eh_action(scmd, rtn) != FAILED)
1da177e4
LT
1558 scsi_eh_finish_cmd(scmd,
1559 done_q);
1560 }
1561 }
1562 } else {
91921e01 1563 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1564 sdev_printk(KERN_INFO, sdev,
1565 "%s: BDR failed\n", current->comm));
1da177e4
LT
1566 }
1567 }
1568
1569 return list_empty(work_q);
1570}
1571
30bd7df8
MC
1572/**
1573 * scsi_eh_target_reset - send target reset if needed
1574 * @shost: scsi host being recovered.
74cf298f 1575 * @work_q: &list_head for pending commands.
30bd7df8
MC
1576 * @done_q: &list_head for processed commands.
1577 *
1578 * Notes:
1579 * Try a target reset.
1580 */
1581static int scsi_eh_target_reset(struct Scsi_Host *shost,
1582 struct list_head *work_q,
1583 struct list_head *done_q)
1584{
98db5195 1585 LIST_HEAD(tmp_list);
3eef6257 1586 LIST_HEAD(check_list);
30bd7df8 1587
98db5195
JB
1588 list_splice_init(work_q, &tmp_list);
1589
1590 while (!list_empty(&tmp_list)) {
1591 struct scsi_cmnd *next, *scmd;
b8e162f9 1592 enum scsi_disposition rtn;
98db5195 1593 unsigned int id;
b4562022 1594
b4562022 1595 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1596 /* push back on work queue for further processing */
1597 list_splice_init(&check_list, work_q);
1598 list_splice_init(&tmp_list, work_q);
1599 SCSI_LOG_ERROR_RECOVERY(3,
1600 shost_printk(KERN_INFO, shost,
a222b1e2
HR
1601 "%s: Skip target reset, past eh deadline\n",
1602 current->comm));
b4562022
HR
1603 return list_empty(work_q);
1604 }
98db5195
JB
1605
1606 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1607 id = scmd_id(scmd);
30bd7df8 1608
91921e01
HR
1609 SCSI_LOG_ERROR_RECOVERY(3,
1610 shost_printk(KERN_INFO, shost,
1611 "%s: Sending target reset to target %d\n",
1612 current->comm, id));
98db5195
JB
1613 rtn = scsi_try_target_reset(scmd);
1614 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
91921e01
HR
1615 SCSI_LOG_ERROR_RECOVERY(3,
1616 shost_printk(KERN_INFO, shost,
1617 "%s: Target reset failed"
1618 " target: %d\n",
1619 current->comm, id));
98db5195
JB
1620 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1621 if (scmd_id(scmd) != id)
1622 continue;
1623
3eef6257
DJ
1624 if (rtn == SUCCESS)
1625 list_move_tail(&scmd->eh_entry, &check_list);
1626 else if (rtn == FAST_IO_FAIL)
98db5195
JB
1627 scsi_eh_finish_cmd(scmd, done_q);
1628 else
1629 /* push back on work queue for further processing */
1630 list_move(&scmd->eh_entry, work_q);
1631 }
1632 }
30bd7df8 1633
3eef6257 1634 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
30bd7df8
MC
1635}
1636
1da177e4 1637/**
0bf8c869 1638 * scsi_eh_bus_reset - send a bus reset
eb44820c 1639 * @shost: &scsi host being recovered.
74cf298f 1640 * @work_q: &list_head for pending commands.
eb44820c 1641 * @done_q: &list_head for processed commands.
dc8875e1 1642 */
1da177e4
LT
1643static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1644 struct list_head *work_q,
1645 struct list_head *done_q)
1646{
937abeaa 1647 struct scsi_cmnd *scmd, *chan_scmd, *next;
3eef6257 1648 LIST_HEAD(check_list);
1da177e4 1649 unsigned int channel;
b8e162f9 1650 enum scsi_disposition rtn;
1da177e4
LT
1651
1652 /*
1653 * we really want to loop over the various channels, and do this on
1654 * a channel by channel basis. we should also check to see if any
1655 * of the failed commands are on soft_reset devices, and if so, skip
0bf8c869 1656 * the reset.
1da177e4
LT
1657 */
1658
1659 for (channel = 0; channel <= shost->max_channel; channel++) {
b4562022 1660 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1661 list_splice_init(&check_list, work_q);
1662 SCSI_LOG_ERROR_RECOVERY(3,
1663 shost_printk(KERN_INFO, shost,
a222b1e2
HR
1664 "%s: skip BRST, past eh deadline\n",
1665 current->comm));
b4562022
HR
1666 return list_empty(work_q);
1667 }
b4562022 1668
1da177e4
LT
1669 chan_scmd = NULL;
1670 list_for_each_entry(scmd, work_q, eh_entry) {
422c0d61 1671 if (channel == scmd_channel(scmd)) {
1da177e4
LT
1672 chan_scmd = scmd;
1673 break;
1674 /*
1675 * FIXME add back in some support for
1676 * soft_reset devices.
1677 */
1678 }
1679 }
1680
1681 if (!chan_scmd)
1682 continue;
91921e01
HR
1683 SCSI_LOG_ERROR_RECOVERY(3,
1684 shost_printk(KERN_INFO, shost,
1685 "%s: Sending BRST chan: %d\n",
1686 current->comm, channel));
1da177e4 1687 rtn = scsi_try_bus_reset(chan_scmd);
2f2eb587 1688 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
937abeaa 1689 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3eef6257
DJ
1690 if (channel == scmd_channel(scmd)) {
1691 if (rtn == FAST_IO_FAIL)
1da177e4
LT
1692 scsi_eh_finish_cmd(scmd,
1693 done_q);
3eef6257
DJ
1694 else
1695 list_move_tail(&scmd->eh_entry,
1696 &check_list);
1697 }
1da177e4
LT
1698 }
1699 } else {
91921e01
HR
1700 SCSI_LOG_ERROR_RECOVERY(3,
1701 shost_printk(KERN_INFO, shost,
1702 "%s: BRST failed chan: %d\n",
1703 current->comm, channel));
1da177e4
LT
1704 }
1705 }
3eef6257 1706 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1707}
1708
1709/**
0bf8c869 1710 * scsi_eh_host_reset - send a host reset
74cf298f
RD
1711 * @shost: host to be reset.
1712 * @work_q: &list_head for pending commands.
1713 * @done_q: &list_head for processed commands.
dc8875e1 1714 */
91921e01
HR
1715static int scsi_eh_host_reset(struct Scsi_Host *shost,
1716 struct list_head *work_q,
1da177e4
LT
1717 struct list_head *done_q)
1718{
937abeaa 1719 struct scsi_cmnd *scmd, *next;
3eef6257 1720 LIST_HEAD(check_list);
b8e162f9 1721 enum scsi_disposition rtn;
1da177e4
LT
1722
1723 if (!list_empty(work_q)) {
1724 scmd = list_entry(work_q->next,
1725 struct scsi_cmnd, eh_entry);
1726
91921e01
HR
1727 SCSI_LOG_ERROR_RECOVERY(3,
1728 shost_printk(KERN_INFO, shost,
1729 "%s: Sending HRST\n",
1730 current->comm));
1da177e4
LT
1731
1732 rtn = scsi_try_host_reset(scmd);
3eef6257
DJ
1733 if (rtn == SUCCESS) {
1734 list_splice_init(work_q, &check_list);
1735 } else if (rtn == FAST_IO_FAIL) {
937abeaa 1736 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1da177e4
LT
1737 scsi_eh_finish_cmd(scmd, done_q);
1738 }
1739 } else {
91921e01
HR
1740 SCSI_LOG_ERROR_RECOVERY(3,
1741 shost_printk(KERN_INFO, shost,
1742 "%s: HRST failed\n",
1743 current->comm));
1da177e4
LT
1744 }
1745 }
3eef6257 1746 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1da177e4
LT
1747}
1748
1749/**
1750 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
74cf298f
RD
1751 * @work_q: &list_head for pending commands.
1752 * @done_q: &list_head for processed commands.
dc8875e1 1753 */
1da177e4
LT
1754static void scsi_eh_offline_sdevs(struct list_head *work_q,
1755 struct list_head *done_q)
1756{
937abeaa 1757 struct scsi_cmnd *scmd, *next;
0db6ca8a 1758 struct scsi_device *sdev;
1da177e4 1759
937abeaa 1760 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
31765d7d
MW
1761 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1762 "not ready after error recovery\n");
0db6ca8a
BVA
1763 sdev = scmd->device;
1764
1765 mutex_lock(&sdev->state_mutex);
1766 scsi_device_set_state(sdev, SDEV_OFFLINE);
1767 mutex_unlock(&sdev->state_mutex);
1768
1da177e4
LT
1769 scsi_eh_finish_cmd(scmd, done_q);
1770 }
1771 return;
1772}
1773
4a27446f 1774/**
e494f6a7 1775 * scsi_noretry_cmd - determine if command should be failed fast
4a27446f
MC
1776 * @scmd: SCSI cmd to examine.
1777 */
1778int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1779{
aa8e25e5
BVA
1780 struct request *req = scsi_cmd_to_rq(scmd);
1781
4a27446f
MC
1782 switch (host_byte(scmd->result)) {
1783 case DID_OK:
1784 break;
e494f6a7
HR
1785 case DID_TIME_OUT:
1786 goto check_type;
4a27446f 1787 case DID_BUS_BUSY:
aa8e25e5 1788 return req->cmd_flags & REQ_FAILFAST_TRANSPORT;
4a27446f 1789 case DID_PARITY:
aa8e25e5 1790 return req->cmd_flags & REQ_FAILFAST_DEV;
4a27446f 1791 case DID_ERROR:
3d45cefc 1792 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
4a27446f 1793 return 0;
df561f66 1794 fallthrough;
4a27446f 1795 case DID_SOFT_ERROR:
aa8e25e5 1796 return req->cmd_flags & REQ_FAILFAST_DRIVER;
4a27446f
MC
1797 }
1798
d0672a03 1799 if (!scsi_status_is_check_condition(scmd->result))
e494f6a7 1800 return 0;
4a27446f 1801
e494f6a7
HR
1802check_type:
1803 /*
1804 * assume caller has checked sense and determined
1805 * the check condition was retryable.
1806 */
aa8e25e5 1807 if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(req))
e494f6a7 1808 return 1;
342c81ee
DLM
1809
1810 return 0;
4a27446f
MC
1811}
1812
1da177e4
LT
1813/**
1814 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1815 * @scmd: SCSI cmd to examine.
1816 *
1817 * Notes:
1818 * This is *only* called when we are examining the status after sending
1819 * out the actual data command. any commands that are queued for error
1820 * recovery (e.g. test_unit_ready) do *not* come through here.
1821 *
1822 * When this routine returns failed, it means the error handler thread
1823 * is woken. In cases where the error code indicates an error that
1824 * doesn't require the error handler read (i.e. we don't need to
1825 * abort/reset), this function should return SUCCESS.
dc8875e1 1826 */
b8e162f9 1827enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd)
1da177e4 1828{
b8e162f9 1829 enum scsi_disposition rtn;
1da177e4
LT
1830
1831 /*
1832 * if the device is offline, then we clearly just pass the result back
1833 * up to the top level.
1834 */
1835 if (!scsi_device_online(scmd->device)) {
91921e01
HR
1836 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1837 "%s: device offline - report as SUCCESS\n", __func__));
1da177e4
LT
1838 return SUCCESS;
1839 }
1840
1841 /*
1842 * first check the host byte, to see if there is anything in there
1843 * that would indicate what we need to do.
1844 */
1845 switch (host_byte(scmd->result)) {
1846 case DID_PASSTHROUGH:
1847 /*
1848 * no matter what, pass this through to the upper layer.
1849 * nuke this special code so that it looks like we are saying
1850 * did_ok.
1851 */
1852 scmd->result &= 0xff00ffff;
1853 return SUCCESS;
1854 case DID_OK:
1855 /*
1856 * looks good. drop through, and check the next byte.
1857 */
1858 break;
e494f6a7
HR
1859 case DID_ABORT:
1860 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
8922a908 1861 set_host_byte(scmd, DID_TIME_OUT);
e494f6a7
HR
1862 return SUCCESS;
1863 }
df561f66 1864 fallthrough;
1da177e4
LT
1865 case DID_NO_CONNECT:
1866 case DID_BAD_TARGET:
1da177e4
LT
1867 /*
1868 * note - this means that we just report the status back
1869 * to the top level driver, not that we actually think
1870 * that it indicates SUCCESS.
1871 */
1872 return SUCCESS;
ad95028a 1873 case DID_SOFT_ERROR:
1da177e4
LT
1874 /*
1875 * when the low level driver returns did_soft_error,
0bf8c869 1876 * it is responsible for keeping an internal retry counter
1da177e4 1877 * in order to avoid endless loops (db)
1da177e4 1878 */
1da177e4
LT
1879 goto maybe_retry;
1880 case DID_IMM_RETRY:
1881 return NEEDS_RETRY;
1882
bf341919 1883 case DID_REQUEUE:
1884 return ADD_TO_MLQUEUE;
a4dfaa6f
MC
1885 case DID_TRANSPORT_DISRUPTED:
1886 /*
1887 * LLD/transport was disrupted during processing of the IO.
1888 * The transport class is now blocked/blocking,
1889 * and the transport will decide what to do with the IO
939c2288
MC
1890 * based on its timers and recovery capablilities if
1891 * there are enough retries.
a4dfaa6f 1892 */
939c2288 1893 goto maybe_retry;
a4dfaa6f
MC
1894 case DID_TRANSPORT_FAILFAST:
1895 /*
1896 * The transport decided to failfast the IO (most likely
1897 * the fast io fail tmo fired), so send IO directly upwards.
1898 */
1899 return SUCCESS;
962c8dcd
MK
1900 case DID_TRANSPORT_MARGINAL:
1901 /*
1902 * caller has decided not to do retries on
1903 * abort success, so send IO directly upwards
1904 */
1905 return SUCCESS;
1da177e4 1906 case DID_ERROR:
3d45cefc 1907 if (get_status_byte(scmd) == SAM_STAT_RESERVATION_CONFLICT)
1da177e4
LT
1908 /*
1909 * execute reservation conflict processing code
1910 * lower down
1911 */
1912 break;
df561f66 1913 fallthrough;
1da177e4
LT
1914 case DID_BUS_BUSY:
1915 case DID_PARITY:
1916 goto maybe_retry;
1917 case DID_TIME_OUT:
1918 /*
1919 * when we scan the bus, we get timeout messages for
1920 * these commands if there is no device available.
1921 * other hosts report did_no_connect for the same thing.
1922 */
1923 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1924 scmd->cmnd[0] == INQUIRY)) {
1925 return SUCCESS;
1926 } else {
1927 return FAILED;
1928 }
1929 case DID_RESET:
1930 return SUCCESS;
1931 default:
1932 return FAILED;
1933 }
1934
1da177e4
LT
1935 /*
1936 * check the status byte to see if this indicates anything special.
1937 */
3d45cefc
HR
1938 switch (get_status_byte(scmd)) {
1939 case SAM_STAT_TASK_SET_FULL:
42a6a918 1940 scsi_handle_queue_full(scmd->device);
1da177e4
LT
1941 /*
1942 * the case of trying to send too many commands to a
1943 * tagged queueing device.
1944 */
df561f66 1945 fallthrough;
3d45cefc 1946 case SAM_STAT_BUSY:
1da177e4
LT
1947 /*
1948 * device can't talk to us at the moment. Should only
1949 * occur (SAM-3) when the task queue is empty, so will cause
1950 * the empty queue handling to trigger a stall in the
1951 * device.
1952 */
1953 return ADD_TO_MLQUEUE;
3d45cefc 1954 case SAM_STAT_GOOD:
279afdfe
EM
1955 if (scmd->cmnd[0] == REPORT_LUNS)
1956 scmd->device->sdev_target->expecting_lun_change = 0;
4a84067d 1957 scsi_handle_queue_ramp_up(scmd->device);
df561f66 1958 fallthrough;
3d45cefc 1959 case SAM_STAT_COMMAND_TERMINATED:
1da177e4 1960 return SUCCESS;
3d45cefc 1961 case SAM_STAT_TASK_ABORTED:
a9b589d9 1962 goto maybe_retry;
3d45cefc 1963 case SAM_STAT_CHECK_CONDITION:
1da177e4
LT
1964 rtn = scsi_check_sense(scmd);
1965 if (rtn == NEEDS_RETRY)
1966 goto maybe_retry;
1967 /* if rtn == FAILED, we have no sense information;
1968 * returning FAILED will wake the error handler thread
1969 * to collect the sense and redo the decide
1970 * disposition */
1971 return rtn;
3d45cefc
HR
1972 case SAM_STAT_CONDITION_MET:
1973 case SAM_STAT_INTERMEDIATE:
1974 case SAM_STAT_INTERMEDIATE_CONDITION_MET:
1975 case SAM_STAT_ACA_ACTIVE:
1da177e4
LT
1976 /*
1977 * who knows? FIXME(eric)
1978 */
1979 return SUCCESS;
1980
3d45cefc 1981 case SAM_STAT_RESERVATION_CONFLICT:
9ccfc756
JB
1982 sdev_printk(KERN_INFO, scmd->device,
1983 "reservation conflict\n");
2082ebc4 1984 set_host_byte(scmd, DID_NEXUS_FAILURE);
1da177e4
LT
1985 return SUCCESS; /* causes immediate i/o error */
1986 default:
1987 return FAILED;
1988 }
1989 return FAILED;
1990
8ef7fe4b 1991maybe_retry:
1da177e4
LT
1992
1993 /* we requeue for retry because the error was retryable, and
1994 * the request was not marked fast fail. Note that above,
1995 * even if the request is marked fast fail, we still requeue
1996 * for queue congestion conditions (QUEUE_FULL or BUSY) */
2a242d59 1997 if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
1da177e4
LT
1998 return NEEDS_RETRY;
1999 } else {
2000 /*
2001 * no more retries - report this one back to upper level.
2002 */
2003 return SUCCESS;
2004 }
2005}
2006
2a842aca 2007static void eh_lock_door_done(struct request *req, blk_status_t status)
f078727b 2008{
0bf6d96c 2009 blk_mq_free_request(req);
f078727b
FT
2010}
2011
1da177e4
LT
2012/**
2013 * scsi_eh_lock_door - Prevent medium removal for the specified device
2014 * @sdev: SCSI device to prevent medium removal
2015 *
2016 * Locking:
91bc31fb 2017 * We must be called from process context.
1da177e4
LT
2018 *
2019 * Notes:
2020 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
2021 * head of the devices request queue, and continue.
dc8875e1 2022 */
1da177e4
LT
2023static void scsi_eh_lock_door(struct scsi_device *sdev)
2024{
f078727b 2025 struct request *req;
82ed4db4 2026 struct scsi_request *rq;
1da177e4 2027
68ec3b81 2028 req = scsi_alloc_request(sdev->request_queue, REQ_OP_DRV_IN, 0);
a492f075 2029 if (IS_ERR(req))
eb571eea 2030 return;
82ed4db4 2031 rq = scsi_req(req);
1da177e4 2032
82ed4db4
CH
2033 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
2034 rq->cmd[1] = 0;
2035 rq->cmd[2] = 0;
2036 rq->cmd[3] = 0;
2037 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
2038 rq->cmd[5] = 0;
2039 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
f078727b 2040
e8064021 2041 req->rq_flags |= RQF_QUIET;
f078727b 2042 req->timeout = 10 * HZ;
64c7f1d1 2043 rq->retries = 5;
f078727b 2044
b84ba30b 2045 blk_execute_rq_nowait(req, true, eh_lock_door_done);
f078727b 2046}
1da177e4
LT
2047
2048/**
2049 * scsi_restart_operations - restart io operations to the specified host.
2050 * @shost: Host we are restarting.
2051 *
2052 * Notes:
2053 * When we entered the error handler, we blocked all further i/o to
2054 * this device. we need to 'reverse' this process.
dc8875e1 2055 */
1da177e4
LT
2056static void scsi_restart_operations(struct Scsi_Host *shost)
2057{
2058 struct scsi_device *sdev;
939647ee 2059 unsigned long flags;
1da177e4
LT
2060
2061 /*
2062 * If the door was locked, we need to insert a door lock request
2063 * onto the head of the SCSI request queue for the device. There
2064 * is no point trying to lock the door of an off-line device.
2065 */
2066 shost_for_each_device(sdev, shost) {
48379270 2067 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1da177e4 2068 scsi_eh_lock_door(sdev);
48379270
CH
2069 sdev->was_reset = 0;
2070 }
1da177e4
LT
2071 }
2072
2073 /*
2074 * next free up anything directly waiting upon the host. this
2075 * will be requests for character device operations, and also for
2076 * ioctls to queued block devices.
2077 */
b4562022 2078 SCSI_LOG_ERROR_RECOVERY(3,
91921e01 2079 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1da177e4 2080
939647ee
JB
2081 spin_lock_irqsave(shost->host_lock, flags);
2082 if (scsi_host_set_state(shost, SHOST_RUNNING))
2083 if (scsi_host_set_state(shost, SHOST_CANCEL))
2084 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2085 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2086
2087 wake_up(&shost->host_wait);
2088
2089 /*
2090 * finally we need to re-initiate requests that may be pending. we will
2091 * have had everything blocked while error handling is taking place, and
2092 * now that error recovery is done, we will need to ensure that these
2093 * requests are started.
2094 */
2095 scsi_run_host_queues(shost);
57fc2e33
DW
2096
2097 /*
2098 * if eh is active and host_eh_scheduled is pending we need to re-run
2099 * recovery. we do this check after scsi_run_host_queues() to allow
2100 * everything pent up since the last eh run a chance to make forward
2101 * progress before we sync again. Either we'll immediately re-run
2102 * recovery or scsi_device_unbusy() will wake us again when these
2103 * pending commands complete.
2104 */
2105 spin_lock_irqsave(shost->host_lock, flags);
2106 if (shost->host_eh_scheduled)
2107 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2108 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2109 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2110}
2111
2112/**
2113 * scsi_eh_ready_devs - check device ready state and recover if not.
74cf298f
RD
2114 * @shost: host to be recovered.
2115 * @work_q: &list_head for pending commands.
eb44820c 2116 * @done_q: &list_head for processed commands.
dc8875e1 2117 */
dca84e46
DW
2118void scsi_eh_ready_devs(struct Scsi_Host *shost,
2119 struct list_head *work_q,
2120 struct list_head *done_q)
1da177e4
LT
2121{
2122 if (!scsi_eh_stu(shost, work_q, done_q))
2123 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
30bd7df8
MC
2124 if (!scsi_eh_target_reset(shost, work_q, done_q))
2125 if (!scsi_eh_bus_reset(shost, work_q, done_q))
91921e01 2126 if (!scsi_eh_host_reset(shost, work_q, done_q))
30bd7df8
MC
2127 scsi_eh_offline_sdevs(work_q,
2128 done_q);
1da177e4 2129}
dca84e46 2130EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1da177e4
LT
2131
2132/**
2133 * scsi_eh_flush_done_q - finish processed commands or retry them.
2134 * @done_q: list_head of processed commands.
dc8875e1 2135 */
041c5fc3 2136void scsi_eh_flush_done_q(struct list_head *done_q)
1da177e4 2137{
937abeaa 2138 struct scsi_cmnd *scmd, *next;
1da177e4 2139
937abeaa
CH
2140 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2141 list_del_init(&scmd->eh_entry);
1da177e4 2142 if (scsi_device_online(scmd->device) &&
60bee27b
MK
2143 !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd) &&
2144 scsi_eh_should_retry_cmd(scmd)) {
91921e01
HR
2145 SCSI_LOG_ERROR_RECOVERY(3,
2146 scmd_printk(KERN_INFO, scmd,
470613b4
HR
2147 "%s: flush retry cmd\n",
2148 current->comm));
1da177e4
LT
2149 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2150 } else {
793698ce
PM
2151 /*
2152 * If just we got sense for the device (called
2153 * scsi_eh_get_sense), scmd->result is already
4bd51e54 2154 * set, do not set DID_TIME_OUT.
793698ce 2155 */
1da177e4 2156 if (!scmd->result)
4bd51e54 2157 scmd->result |= (DID_TIME_OUT << 16);
91921e01
HR
2158 SCSI_LOG_ERROR_RECOVERY(3,
2159 scmd_printk(KERN_INFO, scmd,
470613b4
HR
2160 "%s: flush finish cmd\n",
2161 current->comm));
1da177e4
LT
2162 scsi_finish_command(scmd);
2163 }
2164 }
2165}
041c5fc3 2166EXPORT_SYMBOL(scsi_eh_flush_done_q);
1da177e4
LT
2167
2168/**
2169 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2170 * @shost: Host to unjam.
2171 *
2172 * Notes:
2173 * When we come in here, we *know* that all commands on the bus have
2174 * either completed, failed or timed out. we also know that no further
2175 * commands are being sent to the host, so things are relatively quiet
2176 * and we have freedom to fiddle with things as we wish.
2177 *
2178 * This is only the *default* implementation. it is possible for
2179 * individual drivers to supply their own version of this function, and
2180 * if the maintainer wishes to do this, it is strongly suggested that
2181 * this function be taken as a template and modified. this function
2182 * was designed to correctly handle problems for about 95% of the
2183 * different cases out there, and it should always provide at least a
2184 * reasonable amount of error recovery.
2185 *
2186 * Any command marked 'failed' or 'timeout' must eventually have
2187 * scsi_finish_cmd() called for it. we do all of the retry stuff
2188 * here, so when we restart the host after we return it should have an
2189 * empty queue.
dc8875e1 2190 */
1da177e4
LT
2191static void scsi_unjam_host(struct Scsi_Host *shost)
2192{
2193 unsigned long flags;
2194 LIST_HEAD(eh_work_q);
2195 LIST_HEAD(eh_done_q);
2196
2197 spin_lock_irqsave(shost->host_lock, flags);
2198 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2199 spin_unlock_irqrestore(shost->host_lock, flags);
2200
2201 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2202
2203 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
a0658632 2204 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1da177e4 2205
b4562022 2206 spin_lock_irqsave(shost->host_lock, flags);
bb3b621a 2207 if (shost->eh_deadline != -1)
b4562022
HR
2208 shost->last_reset = 0;
2209 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2210 scsi_eh_flush_done_q(&eh_done_q);
2211}
2212
2213/**
ad42eb1b 2214 * scsi_error_handler - SCSI error handler thread
1da177e4
LT
2215 * @data: Host for which we are running.
2216 *
2217 * Notes:
ad42eb1b
CH
2218 * This is the main error handling loop. This is run as a kernel thread
2219 * for every SCSI host and handles all error handling activity.
dc8875e1 2220 */
1da177e4
LT
2221int scsi_error_handler(void *data)
2222{
ad42eb1b 2223 struct Scsi_Host *shost = data;
1da177e4 2224
1da177e4 2225 /*
ad42eb1b
CH
2226 * We use TASK_INTERRUPTIBLE so that the thread is not
2227 * counted against the load average as a running process.
2228 * We never actually get interrupted because kthread_run
c03264a7 2229 * disables signal delivery for the created thread.
1da177e4 2230 */
537b604c
MH
2231 while (true) {
2232 /*
2233 * The sequence in kthread_stop() sets the stop flag first
2234 * then wakes the process. To avoid missed wakeups, the task
2235 * should always be in a non running state before the stop
2236 * flag is checked
2237 */
b9d5c6b7 2238 set_current_state(TASK_INTERRUPTIBLE);
537b604c
MH
2239 if (kthread_should_stop())
2240 break;
2241
ee7863bc 2242 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
c84b023a 2243 shost->host_failed != scsi_host_busy(shost)) {
ad42eb1b 2244 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2245 shost_printk(KERN_INFO, shost,
2246 "scsi_eh_%d: sleeping\n",
2247 shost->host_no));
3ed7a470 2248 schedule();
3ed7a470
JB
2249 continue;
2250 }
1da177e4 2251
3ed7a470 2252 __set_current_state(TASK_RUNNING);
ad42eb1b 2253 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2254 shost_printk(KERN_INFO, shost,
2255 "scsi_eh_%d: waking up %d/%d/%d\n",
2256 shost->host_no, shost->host_eh_scheduled,
74665016 2257 shost->host_failed,
c84b023a 2258 scsi_host_busy(shost)));
1da177e4 2259
1da177e4
LT
2260 /*
2261 * We have a host that is failing for some reason. Figure out
2262 * what we need to do to get it up and online again (if we can).
2263 * If we fail, we end up taking the thing offline.
2264 */
ae0751ff 2265 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
bc4f2401 2266 SCSI_LOG_ERROR_RECOVERY(1,
a222b1e2
HR
2267 shost_printk(KERN_ERR, shost,
2268 "scsi_eh_%d: unable to autoresume\n",
2269 shost->host_no));
bc4f2401
AS
2270 continue;
2271 }
2272
9227c33d
CH
2273 if (shost->transportt->eh_strategy_handler)
2274 shost->transportt->eh_strategy_handler(shost);
1da177e4
LT
2275 else
2276 scsi_unjam_host(shost);
2277
72d8c36e
WF
2278 /* All scmds have been handled */
2279 shost->host_failed = 0;
2280
1da177e4
LT
2281 /*
2282 * Note - if the above fails completely, the action is to take
2283 * individual devices offline and flush the queue of any
2284 * outstanding requests that may have been pending. When we
2285 * restart, we restart any I/O to any other devices on the bus
2286 * which are still online.
2287 */
2288 scsi_restart_operations(shost);
ae0751ff
LM
2289 if (!shost->eh_noresume)
2290 scsi_autopm_put_host(shost);
1da177e4 2291 }
461a0ffb
SR
2292 __set_current_state(TASK_RUNNING);
2293
ad42eb1b 2294 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2295 shost_printk(KERN_INFO, shost,
2296 "Error handler scsi_eh_%d exiting\n",
2297 shost->host_no));
3ed7a470 2298 shost->ehandler = NULL;
1da177e4
LT
2299 return 0;
2300}
2301
2302/*
2303 * Function: scsi_report_bus_reset()
2304 *
2305 * Purpose: Utility function used by low-level drivers to report that
2306 * they have observed a bus reset on the bus being handled.
2307 *
2308 * Arguments: shost - Host in question
2309 * channel - channel on which reset was observed.
2310 *
2311 * Returns: Nothing
2312 *
2313 * Lock status: Host lock must be held.
2314 *
2315 * Notes: This only needs to be called if the reset is one which
2316 * originates from an unknown location. Resets originated
2317 * by the mid-level itself don't need to call this, but there
2318 * should be no harm.
2319 *
2320 * The main purpose of this is to make sure that a CHECK_CONDITION
2321 * is properly treated.
2322 */
2323void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2324{
2325 struct scsi_device *sdev;
2326
2327 __shost_for_each_device(sdev, shost) {
30bd7df8
MC
2328 if (channel == sdev_channel(sdev))
2329 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2330 }
2331}
2332EXPORT_SYMBOL(scsi_report_bus_reset);
2333
2334/*
2335 * Function: scsi_report_device_reset()
2336 *
2337 * Purpose: Utility function used by low-level drivers to report that
2338 * they have observed a device reset on the device being handled.
2339 *
2340 * Arguments: shost - Host in question
2341 * channel - channel on which reset was observed
2342 * target - target on which reset was observed
2343 *
2344 * Returns: Nothing
2345 *
2346 * Lock status: Host lock must be held
2347 *
2348 * Notes: This only needs to be called if the reset is one which
2349 * originates from an unknown location. Resets originated
2350 * by the mid-level itself don't need to call this, but there
2351 * should be no harm.
2352 *
2353 * The main purpose of this is to make sure that a CHECK_CONDITION
2354 * is properly treated.
2355 */
2356void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2357{
2358 struct scsi_device *sdev;
2359
2360 __shost_for_each_device(sdev, shost) {
422c0d61 2361 if (channel == sdev_channel(sdev) &&
30bd7df8
MC
2362 target == sdev_id(sdev))
2363 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2364 }
2365}
2366EXPORT_SYMBOL(scsi_report_device_reset);
2367
176aa9d6
CH
2368/**
2369 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2370 * @dev: scsi_device to operate on
2371 * @arg: reset type (see sg.h)
1da177e4
LT
2372 */
2373int
176aa9d6 2374scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
1da177e4 2375{
bc4f2401 2376 struct scsi_cmnd *scmd;
d7a1bb0a 2377 struct Scsi_Host *shost = dev->host;
e9c787e6 2378 struct request *rq;
d7a1bb0a 2379 unsigned long flags;
b8e162f9
BVA
2380 int error = 0, val;
2381 enum scsi_disposition rtn;
176aa9d6
CH
2382
2383 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2384 return -EACCES;
2385
2386 error = get_user(val, arg);
2387 if (error)
2388 return error;
1da177e4 2389
bc4f2401 2390 if (scsi_autopm_get_host(shost) < 0)
176aa9d6 2391 return -EIO;
bc4f2401 2392
176aa9d6 2393 error = -EIO;
e9c787e6
CH
2394 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2395 shost->hostt->cmd_size, GFP_KERNEL);
2396 if (!rq)
95eeb5f5 2397 goto out_put_autopm_host;
e9c787e6 2398 blk_rq_init(NULL, rq);
95eeb5f5 2399
e9c787e6
CH
2400 scmd = (struct scsi_cmnd *)(rq + 1);
2401 scsi_init_command(dev, scmd);
82ed4db4 2402 scmd->cmnd = scsi_req(rq)->cmd;
64a87b24 2403
bf23e619 2404 scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL;
30b0c37b 2405 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1da177e4
LT
2406
2407 scmd->cmd_len = 0;
2408
2409 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1da177e4 2410
d7a1bb0a
JS
2411 spin_lock_irqsave(shost->host_lock, flags);
2412 shost->tmf_in_progress = 1;
2413 spin_unlock_irqrestore(shost->host_lock, flags);
2414
176aa9d6
CH
2415 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2416 case SG_SCSI_RESET_NOTHING:
2417 rtn = SUCCESS;
2418 break;
2419 case SG_SCSI_RESET_DEVICE:
1da177e4 2420 rtn = scsi_try_bus_device_reset(scmd);
176aa9d6 2421 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
1da177e4 2422 break;
df561f66 2423 fallthrough;
176aa9d6 2424 case SG_SCSI_RESET_TARGET:
30bd7df8 2425 rtn = scsi_try_target_reset(scmd);
176aa9d6 2426 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
30bd7df8 2427 break;
df561f66 2428 fallthrough;
176aa9d6 2429 case SG_SCSI_RESET_BUS:
1da177e4 2430 rtn = scsi_try_bus_reset(scmd);
176aa9d6 2431 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
1da177e4 2432 break;
df561f66 2433 fallthrough;
176aa9d6 2434 case SG_SCSI_RESET_HOST:
1da177e4 2435 rtn = scsi_try_host_reset(scmd);
176aa9d6
CH
2436 if (rtn == SUCCESS)
2437 break;
df561f66 2438 fallthrough;
3bf2ff67 2439 default:
1da177e4 2440 rtn = FAILED;
176aa9d6 2441 break;
1da177e4
LT
2442 }
2443
176aa9d6
CH
2444 error = (rtn == SUCCESS) ? 0 : -EIO;
2445
d7a1bb0a
JS
2446 spin_lock_irqsave(shost->host_lock, flags);
2447 shost->tmf_in_progress = 0;
2448 spin_unlock_irqrestore(shost->host_lock, flags);
2449
2450 /*
2451 * be sure to wake up anyone who was sleeping or had their queue
2452 * suspended while we performed the TMF.
2453 */
2454 SCSI_LOG_ERROR_RECOVERY(3,
91921e01
HR
2455 shost_printk(KERN_INFO, shost,
2456 "waking up host to restart after TMF\n"));
d7a1bb0a
JS
2457
2458 wake_up(&shost->host_wait);
d7a1bb0a
JS
2459 scsi_run_host_queues(shost);
2460
e9c787e6 2461 kfree(rq);
0f121dd8 2462
04796336 2463out_put_autopm_host:
bc4f2401 2464 scsi_autopm_put_host(shost);
176aa9d6 2465 return error;
1da177e4 2466}
1da177e4 2467
4753cbc0
HR
2468bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2469 struct scsi_sense_hdr *sshdr)
1da177e4
LT
2470{
2471 return scsi_normalize_sense(cmd->sense_buffer,
b80ca4f7 2472 SCSI_SENSE_BUFFERSIZE, sshdr);
1da177e4
LT
2473}
2474EXPORT_SYMBOL(scsi_command_normalize_sense);
2475
1da177e4 2476/**
eb44820c 2477 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1da177e4
LT
2478 * @sense_buffer: byte array of sense data
2479 * @sb_len: number of valid bytes in sense_buffer
2480 * @info_out: pointer to 64 integer where 8 or 4 byte information
2481 * field will be placed if found.
2482 *
2483 * Return value:
2908769c 2484 * true if information field found, false if not found.
dc8875e1 2485 */
2908769c
DLM
2486bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2487 u64 *info_out)
1da177e4 2488{
1da177e4 2489 const u8 * ucp;
1da177e4
LT
2490
2491 if (sb_len < 7)
2908769c 2492 return false;
1da177e4
LT
2493 switch (sense_buffer[0] & 0x7f) {
2494 case 0x70:
2495 case 0x71:
2496 if (sense_buffer[0] & 0x80) {
2908769c
DLM
2497 *info_out = get_unaligned_be32(&sense_buffer[3]);
2498 return true;
2499 }
2500 return false;
1da177e4
LT
2501 case 0x72:
2502 case 0x73:
2503 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2504 0 /* info desc */);
2505 if (ucp && (0xa == ucp[1])) {
2908769c
DLM
2506 *info_out = get_unaligned_be64(&ucp[4]);
2507 return true;
2508 }
2509 return false;
1da177e4 2510 default:
2908769c 2511 return false;
1da177e4
LT
2512 }
2513}
2514EXPORT_SYMBOL(scsi_get_sense_info_fld);