treewide: Use fallthrough pseudo-keyword
[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
14216561
JB
53static void scsi_eh_done(struct scsi_cmnd *scmd);
54
1da177e4
LT
55/*
56 * These should *probably* be handled by the host itself.
57 * Since it is allowed to sleep, it probably should.
58 */
59#define BUS_RESET_SETTLE_TIME (10)
60#define HOST_RESET_SETTLE_TIME (10)
61
3eef6257 62static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
e494f6a7
HR
63static int scsi_try_to_abort_cmd(struct scsi_host_template *,
64 struct scsi_cmnd *);
3eef6257 65
1da177e4
LT
66void scsi_eh_wakeup(struct Scsi_Host *shost)
67{
f0317e88
BVA
68 lockdep_assert_held(shost->host_lock);
69
c84b023a 70 if (scsi_host_busy(shost) == shost->host_failed) {
bf816235 71 trace_scsi_eh_wakeup(shost);
3ed7a470 72 wake_up_process(shost->ehandler);
91921e01
HR
73 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
74 "Waking error handler thread\n"));
1da177e4
LT
75 }
76}
f8bbfc24
TH
77
78/**
79 * scsi_schedule_eh - schedule EH for SCSI host
80 * @shost: SCSI host to invoke error handling on.
81 *
82 * Schedule SCSI EH without scmd.
dc8875e1 83 */
f8bbfc24
TH
84void scsi_schedule_eh(struct Scsi_Host *shost)
85{
86 unsigned long flags;
87
88 spin_lock_irqsave(shost->host_lock, flags);
89
90 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
91 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
92 shost->host_eh_scheduled++;
93 scsi_eh_wakeup(shost);
94 }
95
96 spin_unlock_irqrestore(shost->host_lock, flags);
97}
98EXPORT_SYMBOL_GPL(scsi_schedule_eh);
1da177e4 99
b4562022
HR
100static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
101{
bb3b621a 102 if (!shost->last_reset || shost->eh_deadline == -1)
b4562022
HR
103 return 0;
104
76ad3e59
HR
105 /*
106 * 32bit accesses are guaranteed to be atomic
107 * (on all supported architectures), so instead
108 * of using a spinlock we can as well double check
bb3b621a 109 * if eh_deadline has been set to 'off' during the
76ad3e59
HR
110 * time_before call.
111 */
112 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
bb3b621a 113 shost->eh_deadline > -1)
b4562022
HR
114 return 0;
115
116 return 1;
117}
118
e494f6a7
HR
119/**
120 * scmd_eh_abort_handler - Handle command aborts
121 * @work: command to be aborted.
923f46f9
BVA
122 *
123 * Note: this function must be called only for a command that has timed out.
124 * Because the block layer marks a request as complete before it calls
125 * scsi_times_out(), a .scsi_done() call from the LLD for a command that has
126 * timed out do not have any effect. Hence it is safe to call
127 * scsi_finish_command() from this function.
e494f6a7
HR
128 */
129void
130scmd_eh_abort_handler(struct work_struct *work)
131{
132 struct scsi_cmnd *scmd =
133 container_of(work, struct scsi_cmnd, abort_work.work);
134 struct scsi_device *sdev = scmd->device;
e494f6a7
HR
135 int rtn;
136
e494f6a7 137 if (scsi_host_eh_past_deadline(sdev->host)) {
e494f6a7
HR
138 SCSI_LOG_ERROR_RECOVERY(3,
139 scmd_printk(KERN_INFO, scmd,
470613b4 140 "eh timeout, not aborting\n"));
e494f6a7 141 } else {
e494f6a7
HR
142 SCSI_LOG_ERROR_RECOVERY(3,
143 scmd_printk(KERN_INFO, scmd,
470613b4 144 "aborting command\n"));
e494f6a7
HR
145 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
146 if (rtn == SUCCESS) {
8922a908 147 set_host_byte(scmd, DID_TIME_OUT);
bb3b621a
RM
148 if (scsi_host_eh_past_deadline(sdev->host)) {
149 SCSI_LOG_ERROR_RECOVERY(3,
150 scmd_printk(KERN_INFO, scmd,
470613b4
HR
151 "eh timeout, not retrying "
152 "aborted command\n"));
bb3b621a 153 } else if (!scsi_noretry_cmd(scmd) &&
e494f6a7
HR
154 (++scmd->retries <= scmd->allowed)) {
155 SCSI_LOG_ERROR_RECOVERY(3,
156 scmd_printk(KERN_WARNING, scmd,
470613b4 157 "retry aborted command\n"));
e494f6a7 158 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
bb3b621a 159 return;
e494f6a7
HR
160 } else {
161 SCSI_LOG_ERROR_RECOVERY(3,
162 scmd_printk(KERN_WARNING, scmd,
470613b4 163 "finish aborted command\n"));
e494f6a7 164 scsi_finish_command(scmd);
bb3b621a 165 return;
e494f6a7 166 }
bb3b621a
RM
167 } else {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_INFO, scmd,
470613b4 170 "cmd abort %s\n",
883a030f
HR
171 (rtn == FAST_IO_FAIL) ?
172 "not send" : "failed"));
e494f6a7 173 }
e494f6a7
HR
174 }
175
a0658632 176 scsi_eh_scmd_add(scmd);
e494f6a7
HR
177}
178
179/**
180 * scsi_abort_command - schedule a command abort
181 * @scmd: scmd to abort.
182 *
183 * We only need to abort commands after a command timeout
184 */
185static int
186scsi_abort_command(struct scsi_cmnd *scmd)
187{
188 struct scsi_device *sdev = scmd->device;
189 struct Scsi_Host *shost = sdev->host;
190 unsigned long flags;
191
192 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
193 /*
194 * Retry after abort failed, escalate to next level.
195 */
196 SCSI_LOG_ERROR_RECOVERY(3,
197 scmd_printk(KERN_INFO, scmd,
470613b4 198 "previous abort failed\n"));
fcc95a76 199 BUG_ON(delayed_work_pending(&scmd->abort_work));
e494f6a7
HR
200 return FAILED;
201 }
202
e494f6a7 203 spin_lock_irqsave(shost->host_lock, flags);
bb3b621a 204 if (shost->eh_deadline != -1 && !shost->last_reset)
e494f6a7
HR
205 shost->last_reset = jiffies;
206 spin_unlock_irqrestore(shost->host_lock, flags);
207
208 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
209 SCSI_LOG_ERROR_RECOVERY(3,
470613b4 210 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
e494f6a7
HR
211 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
212 return SUCCESS;
213}
214
1da177e4 215/**
7a38dc0b 216 * scsi_eh_reset - call into ->eh_action to reset internal counters
1da177e4 217 * @scmd: scmd to run eh on.
1da177e4 218 *
7a38dc0b
HR
219 * The scsi driver might be carrying internal state about the
220 * devices, so we need to call into the driver to reset the
221 * internal state once the error handler is started.
dc8875e1 222 */
7a38dc0b
HR
223static void scsi_eh_reset(struct scsi_cmnd *scmd)
224{
225 if (!blk_rq_is_passthrough(scmd->request)) {
226 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
227 if (sdrv->eh_reset)
228 sdrv->eh_reset(scmd);
229 }
230}
231
3bd6f43f
BVA
232static void scsi_eh_inc_host_failed(struct rcu_head *head)
233{
3be8828f
BVA
234 struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu);
235 struct Scsi_Host *shost = scmd->device->host;
3bd6f43f
BVA
236 unsigned long flags;
237
238 spin_lock_irqsave(shost->host_lock, flags);
239 shost->host_failed++;
240 scsi_eh_wakeup(shost);
241 spin_unlock_irqrestore(shost->host_lock, flags);
242}
243
1da177e4
LT
244/**
245 * scsi_eh_scmd_add - add scsi cmd to error handling.
246 * @scmd: scmd to run eh on.
dc8875e1 247 */
a0658632 248void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
1da177e4
LT
249{
250 struct Scsi_Host *shost = scmd->device->host;
251 unsigned long flags;
2171b6d0 252 int ret;
1da177e4 253
2171b6d0 254 WARN_ON_ONCE(!shost->ehandler);
1da177e4
LT
255
256 spin_lock_irqsave(shost->host_lock, flags);
2171b6d0
HR
257 if (scsi_host_set_state(shost, SHOST_RECOVERY)) {
258 ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY);
259 WARN_ON_ONCE(ret);
260 }
bb3b621a 261 if (shost->eh_deadline != -1 && !shost->last_reset)
b4562022
HR
262 shost->last_reset = jiffies;
263
7a38dc0b 264 scsi_eh_reset(scmd);
1da177e4 265 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
1da177e4 266 spin_unlock_irqrestore(shost->host_lock, flags);
3bd6f43f
BVA
267 /*
268 * Ensure that all tasks observe the host state change before the
269 * host_failed change.
270 */
3be8828f 271 call_rcu(&scmd->rcu, scsi_eh_inc_host_failed);
1da177e4
LT
272}
273
1da177e4
LT
274/**
275 * scsi_times_out - Timeout function for normal scsi commands.
242f9dcb 276 * @req: request that is timing out.
1da177e4
LT
277 *
278 * Notes:
279 * We do not need to lock this. There is the potential for a race
280 * only in that the normal completion handling might run, but if the
281 * normal completion function determines that the timer has already
282 * fired, then it mustn't do anything.
dc8875e1 283 */
242f9dcb 284enum blk_eh_timer_return scsi_times_out(struct request *req)
1da177e4 285{
bed2213d 286 struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(req);
6600593c 287 enum blk_eh_timer_return rtn = BLK_EH_DONE;
0bf8c869 288 struct Scsi_Host *host = scmd->device->host;
6c5f8ce1 289
bf816235 290 trace_scsi_dispatch_cmd_timeout(scmd);
1da177e4
LT
291 scsi_log_completion(scmd, TIMEOUT_ERROR);
292
bb3b621a 293 if (host->eh_deadline != -1 && !host->last_reset)
b4562022
HR
294 host->last_reset = jiffies;
295
b6a05c82 296 if (host->hostt->eh_timed_out)
0bf8c869 297 rtn = host->hostt->eh_timed_out(scmd);
1da177e4 298
6600593c 299 if (rtn == BLK_EH_DONE) {
065990bd 300 /*
f1342709
KB
301 * Set the command to complete first in order to prevent a real
302 * completion from releasing the command while error handling
303 * is using it. If the command was already completed, then the
304 * lower level driver beat the timeout handler, and it is safe
305 * to return without escalating error recovery.
065990bd 306 *
f1342709
KB
307 * If timeout handling lost the race to a real completion, the
308 * block layer may ignore that due to a fake timeout injection,
309 * so return RESET_TIMER to allow error handling another shot
310 * at this command.
065990bd 311 */
f1342709
KB
312 if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
313 return BLK_EH_RESET_TIMER;
a0658632 314 if (scsi_abort_command(scmd) != SUCCESS) {
2171b6d0 315 set_host_byte(scmd, DID_TIME_OUT);
a0658632 316 scsi_eh_scmd_add(scmd);
2171b6d0 317 }
a33c070b 318 }
242f9dcb 319
fa990781 320 return rtn;
1da177e4
LT
321}
322
323/**
324 * scsi_block_when_processing_errors - Prevent cmds from being queued.
325 * @sdev: Device on which we are performing recovery.
326 *
327 * Description:
328 * We block until the host is out of error recovery, and then check to
329 * see whether the host or the device is offline.
330 *
331 * Return value:
332 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
dc8875e1 333 */
1da177e4
LT
334int scsi_block_when_processing_errors(struct scsi_device *sdev)
335{
336 int online;
337
939647ee 338 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
1da177e4
LT
339
340 online = scsi_device_online(sdev);
341
1da177e4
LT
342 return online;
343}
344EXPORT_SYMBOL(scsi_block_when_processing_errors);
345
346#ifdef CONFIG_SCSI_LOGGING
347/**
348 * scsi_eh_prt_fail_stats - Log info on failures.
349 * @shost: scsi host being recovered.
350 * @work_q: Queue of scsi cmds to process.
dc8875e1 351 */
1da177e4
LT
352static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
353 struct list_head *work_q)
354{
355 struct scsi_cmnd *scmd;
356 struct scsi_device *sdev;
357 int total_failures = 0;
358 int cmd_failed = 0;
359 int cmd_cancel = 0;
360 int devices_failed = 0;
361
362 shost_for_each_device(sdev, shost) {
363 list_for_each_entry(scmd, work_q, eh_entry) {
364 if (scmd->device == sdev) {
365 ++total_failures;
a0658632 366 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
1da177e4 367 ++cmd_cancel;
0bf8c869 368 else
1da177e4
LT
369 ++cmd_failed;
370 }
371 }
372
373 if (cmd_cancel || cmd_failed) {
374 SCSI_LOG_ERROR_RECOVERY(3,
a3a790dc 375 shost_printk(KERN_INFO, shost,
9ccfc756 376 "%s: cmds failed: %d, cancel: %d\n",
cadbd4a5 377 __func__, cmd_failed,
9ccfc756 378 cmd_cancel));
1da177e4
LT
379 cmd_cancel = 0;
380 cmd_failed = 0;
381 ++devices_failed;
382 }
383 }
384
91921e01
HR
385 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
386 "Total of %d commands on %d"
387 " devices require eh work\n",
0bf8c869 388 total_failures, devices_failed));
1da177e4
LT
389}
390#endif
391
279afdfe
EM
392 /**
393 * scsi_report_lun_change - Set flag on all *other* devices on the same target
394 * to indicate that a UNIT ATTENTION is expected.
395 * @sdev: Device reporting the UNIT ATTENTION
396 */
397static void scsi_report_lun_change(struct scsi_device *sdev)
398{
399 sdev->sdev_target->expecting_lun_change = 1;
400}
401
402/**
403 * scsi_report_sense - Examine scsi sense information and log messages for
404 * certain conditions, also issue uevents for some of them.
405 * @sdev: Device reporting the sense code
406 * @sshdr: sshdr to be examined
407 */
408static void scsi_report_sense(struct scsi_device *sdev,
409 struct scsi_sense_hdr *sshdr)
410{
411 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
412
413 if (sshdr->sense_key == UNIT_ATTENTION) {
414 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
415 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
416 sdev_printk(KERN_WARNING, sdev,
417 "Inquiry data has changed");
418 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
419 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
420 scsi_report_lun_change(sdev);
421 sdev_printk(KERN_WARNING, sdev,
422 "Warning! Received an indication that the "
423 "LUN assignments on this target have "
424 "changed. The Linux SCSI layer does not "
425 "automatically remap LUN assignments.\n");
426 } else if (sshdr->asc == 0x3f)
427 sdev_printk(KERN_WARNING, sdev,
428 "Warning! Received an indication that the "
429 "operating parameters on this target have "
430 "changed. The Linux SCSI layer does not "
431 "automatically adjust these parameters.\n");
432
433 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
434 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
435 sdev_printk(KERN_WARNING, sdev,
436 "Warning! Received an indication that the "
437 "LUN reached a thin provisioning soft "
438 "threshold.\n");
439 }
440
cf3431bb
HR
441 if (sshdr->asc == 0x29) {
442 evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED;
443 sdev_printk(KERN_WARNING, sdev,
444 "Power-on or device reset occurred\n");
445 }
446
279afdfe
EM
447 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
448 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
449 sdev_printk(KERN_WARNING, sdev,
450 "Mode parameters changed");
14c3e677
HR
451 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
452 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
453 sdev_printk(KERN_WARNING, sdev,
454 "Asymmetric access state changed");
279afdfe
EM
455 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
456 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
457 sdev_printk(KERN_WARNING, sdev,
458 "Capacity data has changed");
459 } else if (sshdr->asc == 0x2a)
460 sdev_printk(KERN_WARNING, sdev,
461 "Parameters changed");
462 }
463
464 if (evt_type != SDEV_EVT_MAXBITS) {
465 set_bit(evt_type, sdev->pending_events);
466 schedule_work(&sdev->event_work);
467 }
468}
469
1da177e4
LT
470/**
471 * scsi_check_sense - Examine scsi cmd sense
472 * @scmd: Cmd to have sense checked.
473 *
474 * Return value:
87f14e65 475 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1da177e4
LT
476 *
477 * Notes:
478 * When a deferred error is detected the current command has
479 * not been executed and needs retrying.
dc8875e1 480 */
3852e373 481int scsi_check_sense(struct scsi_cmnd *scmd)
1da177e4 482{
a6a8d9f8 483 struct scsi_device *sdev = scmd->device;
1da177e4
LT
484 struct scsi_sense_hdr sshdr;
485
486 if (! scsi_command_normalize_sense(scmd, &sshdr))
487 return FAILED; /* no valid sense data */
488
279afdfe
EM
489 scsi_report_sense(sdev, &sshdr);
490
1da177e4
LT
491 if (scsi_sense_is_deferred(&sshdr))
492 return NEEDS_RETRY;
493
ee14c674 494 if (sdev->handler && sdev->handler->check_sense) {
a6a8d9f8
CS
495 int rc;
496
ee14c674 497 rc = sdev->handler->check_sense(sdev, &sshdr);
a6a8d9f8
CS
498 if (rc != SCSI_RETURN_NOT_HANDLED)
499 return rc;
500 /* handler does not care. Drop down to default handling */
501 }
502
e925cc43
CH
503 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
504 /*
505 * nasty: for mid-layer issued TURs, we need to return the
506 * actual sense data without any recovery attempt. For eh
507 * issued ones, we need to try to recover and interpret
508 */
509 return SUCCESS;
510
1da177e4
LT
511 /*
512 * Previous logic looked for FILEMARK, EOM or ILI which are
513 * mainly associated with tapes and returned SUCCESS.
514 */
515 if (sshdr.response_code == 0x70) {
516 /* fixed format */
517 if (scmd->sense_buffer[2] & 0xe0)
518 return SUCCESS;
519 } else {
520 /*
521 * descriptor format: look for "stream commands sense data
522 * descriptor" (see SSC-3). Assume single sense data
523 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
524 */
525 if ((sshdr.additional_length > 3) &&
526 (scmd->sense_buffer[8] == 0x4) &&
527 (scmd->sense_buffer[11] & 0xe0))
528 return SUCCESS;
529 }
530
531 switch (sshdr.sense_key) {
532 case NO_SENSE:
533 return SUCCESS;
534 case RECOVERED_ERROR:
535 return /* soft_error */ SUCCESS;
536
537 case ABORTED_COMMAND:
511e44f4
MP
538 if (sshdr.asc == 0x10) /* DIF */
539 return SUCCESS;
540
29cfc2ab
MW
541 if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF)
542 return ADD_TO_MLQUEUE;
c3606520
MW
543 if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 &&
544 sdev->sdev_bflags & BLIST_RETRY_ASC_C1)
545 return ADD_TO_MLQUEUE;
29cfc2ab 546
1da177e4
LT
547 return NEEDS_RETRY;
548 case NOT_READY:
549 case UNIT_ATTENTION:
550 /*
551 * if we are expecting a cc/ua because of a bus reset that we
552 * performed, treat this just as a retry. otherwise this is
553 * information that we should pass up to the upper-level driver
554 * so that we can deal with it there.
555 */
556 if (scmd->device->expecting_cc_ua) {
dfcf7775
TH
557 /*
558 * Because some device does not queue unit
559 * attentions correctly, we carefully check
560 * additional sense code and qualifier so as
561 * not to squash media change unit attention.
562 */
563 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
564 scmd->device->expecting_cc_ua = 0;
565 return NEEDS_RETRY;
566 }
1da177e4 567 }
279afdfe
EM
568 /*
569 * we might also expect a cc/ua if another LUN on the target
570 * reported a UA with an ASC/ASCQ of 3F 0E -
571 * REPORTED LUNS DATA HAS CHANGED.
572 */
573 if (scmd->device->sdev_target->expecting_lun_change &&
574 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
575 return NEEDS_RETRY;
1da177e4 576 /*
0bf8c869 577 * if the device is in the process of becoming ready, we
1da177e4
LT
578 * should retry.
579 */
580 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
581 return NEEDS_RETRY;
582 /*
583 * if the device is not started, we need to wake
584 * the error handler to start the motor
585 */
586 if (scmd->device->allow_restart &&
587 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
588 return FAILED;
02e031cb
CH
589 /*
590 * Pass the UA upwards for a determination in the completion
591 * functions.
592 */
593 return SUCCESS;
1da177e4 594
63583cca 595 /* these are not supported */
a9d6ceb8
HR
596 case DATA_PROTECT:
597 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
598 /* Thin provisioning hard threshold reached */
599 set_host_byte(scmd, DID_ALLOC_FAILURE);
600 return SUCCESS;
601 }
df561f66 602 fallthrough;
1da177e4
LT
603 case COPY_ABORTED:
604 case VOLUME_OVERFLOW:
605 case MISCOMPARE:
63583cca 606 case BLANK_CHECK:
87f14e65
HR
607 set_host_byte(scmd, DID_TARGET_FAILURE);
608 return SUCCESS;
1da177e4
LT
609
610 case MEDIUM_ERROR:
fd1b494d
LT
611 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
612 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
613 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
7e782af5 614 set_host_byte(scmd, DID_MEDIUM_ERROR);
87f14e65 615 return SUCCESS;
fd1b494d 616 }
1da177e4
LT
617 return NEEDS_RETRY;
618
619 case HARDWARE_ERROR:
620 if (scmd->device->retry_hwerror)
bb0003c1 621 return ADD_TO_MLQUEUE;
1da177e4 622 else
87f14e65 623 set_host_byte(scmd, DID_TARGET_FAILURE);
df561f66 624 fallthrough;
1da177e4
LT
625
626 case ILLEGAL_REQUEST:
47ac56db
MS
627 if (sshdr.asc == 0x20 || /* Invalid command operation code */
628 sshdr.asc == 0x21 || /* Logical block address out of range */
a8bbb2ab 629 sshdr.asc == 0x22 || /* Invalid function */
47ac56db 630 sshdr.asc == 0x24 || /* Invalid field in cdb */
d0b7a909
MW
631 sshdr.asc == 0x26 || /* Parameter value invalid */
632 sshdr.asc == 0x27) { /* Write protected */
87f14e65 633 set_host_byte(scmd, DID_TARGET_FAILURE);
47ac56db
MS
634 }
635 return SUCCESS;
636
1da177e4
LT
637 default:
638 return SUCCESS;
639 }
640}
3852e373 641EXPORT_SYMBOL_GPL(scsi_check_sense);
1da177e4 642
4a84067d
VD
643static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
644{
645 struct scsi_host_template *sht = sdev->host->hostt;
646 struct scsi_device *tmp_sdev;
647
c40ecc12 648 if (!sht->track_queue_depth ||
4a84067d
VD
649 sdev->queue_depth >= sdev->max_queue_depth)
650 return;
651
652 if (time_before(jiffies,
653 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
654 return;
655
656 if (time_before(jiffies,
657 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
658 return;
659
660 /*
661 * Walk all devices of a target and do
662 * ramp up on them.
663 */
664 shost_for_each_device(tmp_sdev, sdev->host) {
665 if (tmp_sdev->channel != sdev->channel ||
666 tmp_sdev->id != sdev->id ||
667 tmp_sdev->queue_depth == sdev->max_queue_depth)
668 continue;
c40ecc12 669
db5ed4df 670 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
4a84067d
VD
671 sdev->last_queue_ramp_up = jiffies;
672 }
673}
674
42a6a918
MC
675static void scsi_handle_queue_full(struct scsi_device *sdev)
676{
677 struct scsi_host_template *sht = sdev->host->hostt;
678 struct scsi_device *tmp_sdev;
679
c40ecc12 680 if (!sht->track_queue_depth)
42a6a918
MC
681 return;
682
683 shost_for_each_device(tmp_sdev, sdev->host) {
684 if (tmp_sdev->channel != sdev->channel ||
685 tmp_sdev->id != sdev->id)
686 continue;
687 /*
688 * We do not know the number of commands that were at
689 * the device when we got the queue full so we start
690 * from the highest possible value and work our way down.
691 */
c40ecc12 692 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
42a6a918
MC
693 }
694}
695
1da177e4
LT
696/**
697 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
698 * @scmd: SCSI cmd to examine.
699 *
700 * Notes:
701 * This is *only* called when we are examining the status of commands
702 * queued during error recovery. the main difference here is that we
703 * don't allow for the possibility of retries here, and we are a lot
704 * more restrictive about what we consider acceptable.
dc8875e1 705 */
1da177e4
LT
706static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
707{
708 /*
709 * first check the host byte, to see if there is anything in there
710 * that would indicate what we need to do.
711 */
712 if (host_byte(scmd->result) == DID_RESET) {
713 /*
714 * rats. we are already in the error handler, so we now
715 * get to try and figure out what to do next. if the sense
716 * is valid, we have a pretty good idea of what to do.
717 * if not, we mark it as FAILED.
718 */
719 return scsi_check_sense(scmd);
720 }
721 if (host_byte(scmd->result) != DID_OK)
722 return FAILED;
723
724 /*
725 * next, check the message byte.
726 */
727 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
728 return FAILED;
729
730 /*
731 * now, check the status byte to see if this indicates
732 * anything special.
733 */
734 switch (status_byte(scmd->result)) {
735 case GOOD:
4a84067d 736 scsi_handle_queue_ramp_up(scmd->device);
df561f66 737 fallthrough;
1da177e4
LT
738 case COMMAND_TERMINATED:
739 return SUCCESS;
740 case CHECK_CONDITION:
741 return scsi_check_sense(scmd);
742 case CONDITION_GOOD:
743 case INTERMEDIATE_GOOD:
744 case INTERMEDIATE_C_GOOD:
745 /*
746 * who knows? FIXME(eric)
747 */
748 return SUCCESS;
5f91bb05 749 case RESERVATION_CONFLICT:
67110dfd
JB
750 if (scmd->cmnd[0] == TEST_UNIT_READY)
751 /* it is a success, we probed the device and
752 * found it */
753 return SUCCESS;
754 /* otherwise, we failed to send the command */
755 return FAILED;
1da177e4 756 case QUEUE_FULL:
42a6a918 757 scsi_handle_queue_full(scmd->device);
df561f66 758 fallthrough;
42a6a918 759 case BUSY:
3eb3a928 760 return NEEDS_RETRY;
1da177e4
LT
761 default:
762 return FAILED;
763 }
764 return FAILED;
765}
766
1da177e4
LT
767/**
768 * scsi_eh_done - Completion function for error handling.
769 * @scmd: Cmd that is done.
dc8875e1 770 */
1da177e4
LT
771static void scsi_eh_done(struct scsi_cmnd *scmd)
772{
0bf8c869 773 struct completion *eh_action;
85631672 774
91921e01 775 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4 776 "%s result: %x\n", __func__, scmd->result));
85631672
MR
777
778 eh_action = scmd->device->host->eh_action;
779 if (eh_action)
780 complete(eh_action);
1da177e4
LT
781}
782
292148f8
BK
783/**
784 * scsi_try_host_reset - ask host adapter to reset itself
c2b3ebd0 785 * @scmd: SCSI cmd to send host reset.
dc8875e1 786 */
292148f8
BK
787static int scsi_try_host_reset(struct scsi_cmnd *scmd)
788{
789 unsigned long flags;
790 int rtn;
0bf8c869
JJ
791 struct Scsi_Host *host = scmd->device->host;
792 struct scsi_host_template *hostt = host->hostt;
292148f8 793
91921e01
HR
794 SCSI_LOG_ERROR_RECOVERY(3,
795 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
292148f8 796
0bf8c869 797 if (!hostt->eh_host_reset_handler)
292148f8
BK
798 return FAILED;
799
0bf8c869 800 rtn = hostt->eh_host_reset_handler(scmd);
292148f8
BK
801
802 if (rtn == SUCCESS) {
0bf8c869 803 if (!hostt->skip_settle_delay)
292148f8 804 ssleep(HOST_RESET_SETTLE_TIME);
0bf8c869
JJ
805 spin_lock_irqsave(host->host_lock, flags);
806 scsi_report_bus_reset(host, scmd_channel(scmd));
807 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
808 }
809
810 return rtn;
811}
812
813/**
814 * scsi_try_bus_reset - ask host to perform a bus reset
815 * @scmd: SCSI cmd to send bus reset.
dc8875e1 816 */
292148f8
BK
817static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
818{
819 unsigned long flags;
820 int rtn;
0bf8c869
JJ
821 struct Scsi_Host *host = scmd->device->host;
822 struct scsi_host_template *hostt = host->hostt;
292148f8 823
91921e01
HR
824 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
825 "%s: Snd Bus RST\n", __func__));
292148f8 826
0bf8c869 827 if (!hostt->eh_bus_reset_handler)
292148f8
BK
828 return FAILED;
829
0bf8c869 830 rtn = hostt->eh_bus_reset_handler(scmd);
292148f8
BK
831
832 if (rtn == SUCCESS) {
0bf8c869 833 if (!hostt->skip_settle_delay)
292148f8 834 ssleep(BUS_RESET_SETTLE_TIME);
0bf8c869
JJ
835 spin_lock_irqsave(host->host_lock, flags);
836 scsi_report_bus_reset(host, scmd_channel(scmd));
837 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
838 }
839
840 return rtn;
841}
842
30bd7df8
MC
843static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
844{
845 sdev->was_reset = 1;
846 sdev->expecting_cc_ua = 1;
847}
848
849/**
850 * scsi_try_target_reset - Ask host to perform a target reset
851 * @scmd: SCSI cmd used to send a target reset
852 *
853 * Notes:
854 * There is no timeout for this operation. if this operation is
855 * unreliable for a given host, then the host itself needs to put a
856 * timer on it, and set the host back to a consistent state prior to
857 * returning.
858 */
859static int scsi_try_target_reset(struct scsi_cmnd *scmd)
860{
861 unsigned long flags;
862 int rtn;
0bf8c869
JJ
863 struct Scsi_Host *host = scmd->device->host;
864 struct scsi_host_template *hostt = host->hostt;
30bd7df8 865
0bf8c869 866 if (!hostt->eh_target_reset_handler)
30bd7df8
MC
867 return FAILED;
868
0bf8c869 869 rtn = hostt->eh_target_reset_handler(scmd);
30bd7df8 870 if (rtn == SUCCESS) {
0bf8c869 871 spin_lock_irqsave(host->host_lock, flags);
30bd7df8
MC
872 __starget_for_each_device(scsi_target(scmd->device), NULL,
873 __scsi_report_device_reset);
0bf8c869 874 spin_unlock_irqrestore(host->host_lock, flags);
30bd7df8
MC
875 }
876
877 return rtn;
878}
879
292148f8
BK
880/**
881 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
882 * @scmd: SCSI cmd used to send BDR
883 *
884 * Notes:
885 * There is no timeout for this operation. if this operation is
886 * unreliable for a given host, then the host itself needs to put a
887 * timer on it, and set the host back to a consistent state prior to
888 * returning.
dc8875e1 889 */
292148f8
BK
890static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
891{
892 int rtn;
0bf8c869 893 struct scsi_host_template *hostt = scmd->device->host->hostt;
292148f8 894
0bf8c869 895 if (!hostt->eh_device_reset_handler)
292148f8
BK
896 return FAILED;
897
0bf8c869 898 rtn = hostt->eh_device_reset_handler(scmd);
30bd7df8
MC
899 if (rtn == SUCCESS)
900 __scsi_report_device_reset(scmd->device, NULL);
292148f8
BK
901 return rtn;
902}
903
883a030f
HR
904/**
905 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
6583f6fb 906 * @hostt: SCSI driver host template
883a030f
HR
907 * @scmd: SCSI cmd used to send a target reset
908 *
909 * Return value:
910 * SUCCESS, FAILED, or FAST_IO_FAIL
911 *
912 * Notes:
913 * SUCCESS does not necessarily indicate that the command
914 * has been aborted; it only indicates that the LLDDs
915 * has cleared all references to that command.
916 * LLDDs should return FAILED only if an abort was required
917 * but could not be executed. LLDDs should return FAST_IO_FAIL
918 * if the device is temporarily unavailable (eg due to a
919 * link down on FibreChannel)
920 */
921static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
922 struct scsi_cmnd *scmd)
292148f8 923{
0bf8c869 924 if (!hostt->eh_abort_handler)
292148f8
BK
925 return FAILED;
926
0bf8c869 927 return hostt->eh_abort_handler(scmd);
292148f8
BK
928}
929
292148f8
BK
930static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
931{
0bf8c869 932 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
292148f8 933 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
30bd7df8
MC
934 if (scsi_try_target_reset(scmd) != SUCCESS)
935 if (scsi_try_bus_reset(scmd) != SUCCESS)
936 scsi_try_host_reset(scmd);
292148f8
BK
937}
938
1da177e4 939/**
3b729f76 940 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
2dc611de 941 * @scmd: SCSI command structure to hijack
e1c23468 942 * @ses: structure to save restore information
55db6c1b 943 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
64a87b24 944 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
55db6c1b 945 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
2dc611de 946 *
e1c23468 947 * This function is used to save a scsi command information before re-execution
55db6c1b
BH
948 * as part of the error recovery process. If @sense_bytes is 0 the command
949 * sent must be one that does not transfer any data. If @sense_bytes != 0
950 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
951 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
dc8875e1 952 */
e1c23468
BH
953void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
954 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1da177e4 955{
f59114b7 956 struct scsi_device *sdev = scmd->device;
1da177e4 957
631c228c
CH
958 /*
959 * We need saved copies of a number of fields - this is because
960 * error handling may need to overwrite these with different values
961 * to run different commands, and once error handling is complete,
962 * we will need to restore these values prior to running the actual
963 * command.
964 */
e1c23468 965 ses->cmd_len = scmd->cmd_len;
64a87b24 966 ses->cmnd = scmd->cmnd;
e1c23468 967 ses->data_direction = scmd->sc_data_direction;
30b0c37b 968 ses->sdb = scmd->sdb;
e1c23468 969 ses->result = scmd->result;
8f8fed0c 970 ses->resid_len = scmd->req.resid_len;
12265709 971 ses->underflow = scmd->underflow;
db007fc5 972 ses->prot_op = scmd->prot_op;
8e8c9d01 973 ses->eh_eflags = scmd->eh_eflags;
631c228c 974
db007fc5 975 scmd->prot_op = SCSI_PROT_NORMAL;
c69e6f81 976 scmd->eh_eflags = 0;
64a87b24
BH
977 scmd->cmnd = ses->eh_cmnd;
978 memset(scmd->cmnd, 0, BLK_MAX_CDB);
30b0c37b 979 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
644373a4 980 scmd->result = 0;
8f8fed0c 981 scmd->req.resid_len = 0;
30b0c37b 982
55db6c1b 983 if (sense_bytes) {
30b0c37b
BH
984 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
985 sense_bytes);
e1c23468 986 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
30b0c37b
BH
987 scmd->sdb.length);
988 scmd->sdb.table.sgl = &ses->sense_sgl;
55db6c1b 989 scmd->sc_data_direction = DMA_FROM_DEVICE;
0c958ecc 990 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
55db6c1b 991 scmd->cmnd[0] = REQUEST_SENSE;
30b0c37b 992 scmd->cmnd[4] = scmd->sdb.length;
55db6c1b 993 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
631c228c 994 } else {
631c228c 995 scmd->sc_data_direction = DMA_NONE;
55db6c1b 996 if (cmnd) {
64a87b24 997 BUG_ON(cmnd_size > BLK_MAX_CDB);
55db6c1b
BH
998 memcpy(scmd->cmnd, cmnd, cmnd_size);
999 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
1000 }
631c228c
CH
1001 }
1002
1003 scmd->underflow = 0;
631c228c 1004
55db6c1b 1005 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1da177e4 1006 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b7 1007 (sdev->lun << 5 & 0xe0);
1da177e4 1008
631c228c
CH
1009 /*
1010 * Zero the sense buffer. The scsi spec mandates that any
1011 * untransferred sense data should be interpreted as being zero.
1012 */
b80ca4f7 1013 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
e1c23468
BH
1014}
1015EXPORT_SYMBOL(scsi_eh_prep_cmnd);
1016
1017/**
3b729f76 1018 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
e1c23468 1019 * @scmd: SCSI command structure to restore
477e608c 1020 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
e1c23468 1021 *
477e608c 1022 * Undo any damage done by above scsi_eh_prep_cmnd().
dc8875e1 1023 */
e1c23468
BH
1024void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
1025{
1026 /*
1027 * Restore original data
1028 */
1029 scmd->cmd_len = ses->cmd_len;
64a87b24 1030 scmd->cmnd = ses->cmnd;
e1c23468 1031 scmd->sc_data_direction = ses->data_direction;
30b0c37b 1032 scmd->sdb = ses->sdb;
e1c23468 1033 scmd->result = ses->result;
8f8fed0c 1034 scmd->req.resid_len = ses->resid_len;
12265709 1035 scmd->underflow = ses->underflow;
db007fc5 1036 scmd->prot_op = ses->prot_op;
8e8c9d01 1037 scmd->eh_eflags = ses->eh_eflags;
e1c23468
BH
1038}
1039EXPORT_SYMBOL(scsi_eh_restore_cmnd);
631c228c 1040
e1c23468 1041/**
3b729f76 1042 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
e1c23468
BH
1043 * @scmd: SCSI command structure to hijack
1044 * @cmnd: CDB to send
1045 * @cmnd_size: size in bytes of @cmnd
1046 * @timeout: timeout for this request
1047 * @sense_bytes: size of sense data to copy or 0
1048 *
1049 * This function is used to send a scsi command down to a target device
1050 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1051 *
1052 * Return value:
1053 * SUCCESS or FAILED or NEEDS_RETRY
dc8875e1 1054 */
e1c23468
BH
1055static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1056 int cmnd_size, int timeout, unsigned sense_bytes)
1057{
1058 struct scsi_device *sdev = scmd->device;
1059 struct Scsi_Host *shost = sdev->host;
1060 DECLARE_COMPLETION_ONSTACK(done);
bbe9fb0d 1061 unsigned long timeleft = timeout, delay;
e1c23468 1062 struct scsi_eh_save ses;
fc73648a 1063 const unsigned long stall_for = msecs_to_jiffies(100);
e1c23468
BH
1064 int rtn;
1065
fc73648a 1066retry:
e1c23468 1067 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
7dfdc9a5 1068 shost->eh_action = &done;
1da177e4 1069
1da177e4 1070 scsi_log_send(scmd);
f281233d 1071 scmd->scsi_done = scsi_eh_done;
bbe9fb0d
BVA
1072
1073 /*
1074 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1075 * change the SCSI device state after we have examined it and before
1076 * .queuecommand() is called.
1077 */
1078 mutex_lock(&sdev->state_mutex);
1079 while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1080 mutex_unlock(&sdev->state_mutex);
1081 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1082 "%s: state %d <> %d\n", __func__, sdev->sdev_state,
1083 SDEV_BLOCK));
1084 delay = min(timeleft, stall_for);
1085 timeleft -= delay;
1086 msleep(jiffies_to_msecs(delay));
1087 mutex_lock(&sdev->state_mutex);
1088 }
1089 if (sdev->sdev_state != SDEV_BLOCK)
1090 rtn = shost->hostt->queuecommand(shost, scmd);
1091 else
1092 rtn = SCSI_MLQUEUE_DEVICE_BUSY;
1093 mutex_unlock(&sdev->state_mutex);
1094
fc73648a
HR
1095 if (rtn) {
1096 if (timeleft > stall_for) {
1097 scsi_eh_restore_cmnd(scmd, &ses);
1098 timeleft -= stall_for;
1099 msleep(jiffies_to_msecs(stall_for));
1100 goto retry;
1101 }
1102 /* signal not to enter either branch of the if () below */
1103 timeleft = 0;
511833ac 1104 rtn = FAILED;
fc73648a
HR
1105 } else {
1106 timeleft = wait_for_completion_timeout(&done, timeout);
ac61d195 1107 rtn = SUCCESS;
fc73648a 1108 }
1da177e4 1109
f59114b7 1110 shost->eh_action = NULL;
1da177e4 1111
fc73648a 1112 scsi_log_completion(scmd, rtn);
1da177e4 1113
91921e01 1114 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4
HR
1115 "%s timeleft: %ld\n",
1116 __func__, timeleft));
1da177e4
LT
1117
1118 /*
fc73648a
HR
1119 * If there is time left scsi_eh_done got called, and we will examine
1120 * the actual status codes to see whether the command actually did
1121 * complete normally, else if we have a zero return and no time left,
1122 * the command must still be pending, so abort it and return FAILED.
1123 * If we never actually managed to issue the command, because
1124 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1125 * value above (so don't execute either branch of the if)
1da177e4 1126 */
7dfdc9a5 1127 if (timeleft) {
1da177e4 1128 rtn = scsi_eh_completed_normally(scmd);
91921e01
HR
1129 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1130 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
7dfdc9a5 1131
1da177e4
LT
1132 switch (rtn) {
1133 case SUCCESS:
1134 case NEEDS_RETRY:
1135 case FAILED:
1136 break;
6e883b0e
HR
1137 case ADD_TO_MLQUEUE:
1138 rtn = NEEDS_RETRY;
1139 break;
1da177e4
LT
1140 default:
1141 rtn = FAILED;
1142 break;
1143 }
511833ac 1144 } else if (rtn != FAILED) {
292148f8 1145 scsi_abort_eh_cmnd(scmd);
7dfdc9a5 1146 rtn = FAILED;
1da177e4
LT
1147 }
1148
e1c23468 1149 scsi_eh_restore_cmnd(scmd, &ses);
18a4d0a2 1150
1da177e4
LT
1151 return rtn;
1152}
1153
1154/**
1155 * scsi_request_sense - Request sense data from a particular target.
1156 * @scmd: SCSI cmd for request sense.
1157 *
1158 * Notes:
1159 * Some hosts automatically obtain this information, others require
1160 * that we obtain it on our own. This function will *not* return until
1161 * the command either times out, or it completes.
dc8875e1 1162 */
1da177e4
LT
1163static int scsi_request_sense(struct scsi_cmnd *scmd)
1164{
0816c925 1165 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1da177e4
LT
1166}
1167
2451079b
JB
1168static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1169{
57292b58 1170 if (!blk_rq_is_passthrough(scmd->request)) {
2451079b
JB
1171 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1172 if (sdrv->eh_action)
1173 rtn = sdrv->eh_action(scmd, rtn);
1174 }
1175 return rtn;
1176}
1177
1da177e4
LT
1178/**
1179 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1180 * @scmd: Original SCSI cmd that eh has finished.
1181 * @done_q: Queue for processed commands.
1182 *
1183 * Notes:
1184 * We don't want to use the normal command completion while we are are
1185 * still handling errors - it may cause other commands to be queued,
eb44820c 1186 * and that would disturb what we are doing. Thus we really want to
1da177e4
LT
1187 * keep a list of pending commands for final completion, and once we
1188 * are ready to leave error handling we handle completion for real.
dc8875e1 1189 */
041c5fc3 1190void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1da177e4 1191{
1da177e4
LT
1192 list_move_tail(&scmd->eh_entry, done_q);
1193}
041c5fc3 1194EXPORT_SYMBOL(scsi_eh_finish_cmd);
1da177e4
LT
1195
1196/**
1197 * scsi_eh_get_sense - Get device sense data.
1198 * @work_q: Queue of commands to process.
eb44820c 1199 * @done_q: Queue of processed commands.
1da177e4
LT
1200 *
1201 * Description:
1202 * See if we need to request sense information. if so, then get it
0bf8c869 1203 * now, so we have a better idea of what to do.
1da177e4
LT
1204 *
1205 * Notes:
1206 * This has the unfortunate side effect that if a shost adapter does
eb44820c 1207 * not automatically request sense information, we end up shutting
1da177e4
LT
1208 * it down before we request it.
1209 *
1210 * All drivers should request sense information internally these days,
1211 * so for now all I have to say is tough noogies if you end up in here.
1212 *
1213 * XXX: Long term this code should go away, but that needs an audit of
1214 * all LLDDs first.
dc8875e1 1215 */
dca84e46
DW
1216int scsi_eh_get_sense(struct list_head *work_q,
1217 struct list_head *done_q)
1da177e4 1218{
937abeaa 1219 struct scsi_cmnd *scmd, *next;
b4562022 1220 struct Scsi_Host *shost;
1da177e4
LT
1221 int rtn;
1222
709c75b5 1223 /*
1224 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1225 * should not get sense.
1226 */
937abeaa 1227 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
a0658632 1228 if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
1da177e4
LT
1229 SCSI_SENSE_VALID(scmd))
1230 continue;
1231
b4562022 1232 shost = scmd->device->host;
b4562022 1233 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1234 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1235 scmd_printk(KERN_INFO, scmd,
1236 "%s: skip request sense, past eh deadline\n",
1237 current->comm));
b4562022
HR
1238 break;
1239 }
d555a2ab
JB
1240 if (status_byte(scmd->result) != CHECK_CONDITION)
1241 /*
1242 * don't request sense if there's no check condition
1243 * status because the error we're processing isn't one
1244 * that has a sense code (and some devices get
1245 * confused by sense requests out of the blue)
1246 */
1247 continue;
1248
3bf743e7
JG
1249 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1250 "%s: requesting sense\n",
1251 current->comm));
1da177e4
LT
1252 rtn = scsi_request_sense(scmd);
1253 if (rtn != SUCCESS)
1254 continue;
1255
91921e01 1256 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4 1257 "sense requested, result %x\n", scmd->result));
d811b848 1258 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1da177e4
LT
1259
1260 rtn = scsi_decide_disposition(scmd);
1261
1262 /*
1263 * if the result was normal, then just pass it along to the
1264 * upper level.
1265 */
1266 if (rtn == SUCCESS)
1267 /* we don't want this command reissued, just
1268 * finished with the sense data, so set
1269 * retries to the max allowed to ensure it
1270 * won't get reissued */
1271 scmd->retries = scmd->allowed;
1272 else if (rtn != NEEDS_RETRY)
1273 continue;
1274
1275 scsi_eh_finish_cmd(scmd, done_q);
1276 }
1277
1278 return list_empty(work_q);
1279}
dca84e46 1280EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1da177e4 1281
1da177e4
LT
1282/**
1283 * scsi_eh_tur - Send TUR to device.
eb44820c 1284 * @scmd: &scsi_cmnd to send TUR
1da177e4
LT
1285 *
1286 * Return value:
1287 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1288 */
1da177e4
LT
1289static int scsi_eh_tur(struct scsi_cmnd *scmd)
1290{
1291 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1292 int retry_cnt = 1, rtn;
1293
1294retry_tur:
0816c925
MP
1295 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1296 scmd->device->eh_timeout, 0);
1da177e4 1297
91921e01 1298 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
470613b4 1299 "%s return: %x\n", __func__, rtn));
631c228c
CH
1300
1301 switch (rtn) {
1302 case NEEDS_RETRY:
1da177e4
LT
1303 if (retry_cnt--)
1304 goto retry_tur;
df561f66 1305 fallthrough;
631c228c 1306 case SUCCESS:
e47373ec 1307 return 0;
631c228c
CH
1308 default:
1309 return 1;
e47373ec 1310 }
1da177e4
LT
1311}
1312
3eef6257
DJ
1313/**
1314 * scsi_eh_test_devices - check if devices are responding from error recovery.
1315 * @cmd_list: scsi commands in error recovery.
74cf298f
RD
1316 * @work_q: queue for commands which still need more error recovery
1317 * @done_q: queue for commands which are finished
1318 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
3eef6257
DJ
1319 *
1320 * Decription:
1321 * Tests if devices are in a working state. Commands to devices now in
1322 * a working state are sent to the done_q while commands to devices which
1323 * are still failing to respond are returned to the work_q for more
1324 * processing.
1325 **/
1326static int scsi_eh_test_devices(struct list_head *cmd_list,
1327 struct list_head *work_q,
1328 struct list_head *done_q, int try_stu)
1329{
1330 struct scsi_cmnd *scmd, *next;
1331 struct scsi_device *sdev;
1332 int finish_cmds;
1333
1334 while (!list_empty(cmd_list)) {
1335 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1336 sdev = scmd->device;
1337
b4562022 1338 if (!try_stu) {
b4562022
HR
1339 if (scsi_host_eh_past_deadline(sdev->host)) {
1340 /* Push items back onto work_q */
1341 list_splice_init(cmd_list, work_q);
b4562022 1342 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1343 sdev_printk(KERN_INFO, sdev,
1344 "%s: skip test device, past eh deadline",
1345 current->comm));
b4562022
HR
1346 break;
1347 }
b4562022
HR
1348 }
1349
3eef6257
DJ
1350 finish_cmds = !scsi_device_online(scmd->device) ||
1351 (try_stu && !scsi_eh_try_stu(scmd) &&
1352 !scsi_eh_tur(scmd)) ||
1353 !scsi_eh_tur(scmd);
1354
1355 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1356 if (scmd->device == sdev) {
2451079b
JB
1357 if (finish_cmds &&
1358 (try_stu ||
1359 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
3eef6257
DJ
1360 scsi_eh_finish_cmd(scmd, done_q);
1361 else
1362 list_move_tail(&scmd->eh_entry, work_q);
1363 }
1364 }
1365 return list_empty(work_q);
1366}
1367
1da177e4
LT
1368/**
1369 * scsi_eh_try_stu - Send START_UNIT to device.
eb44820c 1370 * @scmd: &scsi_cmnd to send START_UNIT
1da177e4
LT
1371 *
1372 * Return value:
1373 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1374 */
1da177e4
LT
1375static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1376{
1377 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1da177e4 1378
631c228c 1379 if (scmd->device->allow_restart) {
ed773e66
BK
1380 int i, rtn = NEEDS_RETRY;
1381
1382 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
9728c081 1383 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1da177e4 1384
631c228c
CH
1385 if (rtn == SUCCESS)
1386 return 0;
1387 }
1da177e4 1388
1da177e4
LT
1389 return 1;
1390}
1391
1392 /**
1393 * scsi_eh_stu - send START_UNIT if needed
eb44820c 1394 * @shost: &scsi host being recovered.
74cf298f 1395 * @work_q: &list_head for pending commands.
eb44820c 1396 * @done_q: &list_head for processed commands.
1da177e4
LT
1397 *
1398 * Notes:
1399 * If commands are failing due to not ready, initializing command required,
0bf8c869 1400 * try revalidating the device, which will end up sending a start unit.
dc8875e1 1401 */
1da177e4
LT
1402static int scsi_eh_stu(struct Scsi_Host *shost,
1403 struct list_head *work_q,
1404 struct list_head *done_q)
1405{
937abeaa 1406 struct scsi_cmnd *scmd, *stu_scmd, *next;
1da177e4
LT
1407 struct scsi_device *sdev;
1408
1409 shost_for_each_device(sdev, shost) {
b4562022 1410 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1411 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1412 sdev_printk(KERN_INFO, sdev,
1413 "%s: skip START_UNIT, past eh deadline\n",
1414 current->comm));
4dea170f 1415 scsi_device_put(sdev);
b4562022
HR
1416 break;
1417 }
1da177e4
LT
1418 stu_scmd = NULL;
1419 list_for_each_entry(scmd, work_q, eh_entry)
1420 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1421 scsi_check_sense(scmd) == FAILED ) {
1422 stu_scmd = scmd;
1423 break;
1424 }
1425
1426 if (!stu_scmd)
1427 continue;
1428
91921e01 1429 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1430 sdev_printk(KERN_INFO, sdev,
1431 "%s: Sending START_UNIT\n",
1432 current->comm));
1da177e4
LT
1433
1434 if (!scsi_eh_try_stu(stu_scmd)) {
1435 if (!scsi_device_online(sdev) ||
1436 !scsi_eh_tur(stu_scmd)) {
937abeaa
CH
1437 list_for_each_entry_safe(scmd, next,
1438 work_q, eh_entry) {
2451079b
JB
1439 if (scmd->device == sdev &&
1440 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1da177e4
LT
1441 scsi_eh_finish_cmd(scmd, done_q);
1442 }
1443 }
1444 } else {
1445 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1446 sdev_printk(KERN_INFO, sdev,
1447 "%s: START_UNIT failed\n",
1448 current->comm));
1da177e4
LT
1449 }
1450 }
1451
1452 return list_empty(work_q);
1453}
1454
1455
1456/**
1457 * scsi_eh_bus_device_reset - send bdr if needed
1458 * @shost: scsi host being recovered.
74cf298f 1459 * @work_q: &list_head for pending commands.
eb44820c 1460 * @done_q: &list_head for processed commands.
1da177e4
LT
1461 *
1462 * Notes:
eb44820c 1463 * Try a bus device reset. Still, look to see whether we have multiple
1da177e4
LT
1464 * devices that are jammed or not - if we have multiple devices, it
1465 * makes no sense to try bus_device_reset - we really would need to try
0bf8c869 1466 * a bus_reset instead.
dc8875e1 1467 */
1da177e4
LT
1468static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1469 struct list_head *work_q,
1470 struct list_head *done_q)
1471{
937abeaa 1472 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1da177e4
LT
1473 struct scsi_device *sdev;
1474 int rtn;
1475
1476 shost_for_each_device(sdev, shost) {
b4562022 1477 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1478 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1479 sdev_printk(KERN_INFO, sdev,
1480 "%s: skip BDR, past eh deadline\n",
1481 current->comm));
4dea170f 1482 scsi_device_put(sdev);
b4562022
HR
1483 break;
1484 }
1da177e4
LT
1485 bdr_scmd = NULL;
1486 list_for_each_entry(scmd, work_q, eh_entry)
1487 if (scmd->device == sdev) {
1488 bdr_scmd = scmd;
1489 break;
1490 }
1491
1492 if (!bdr_scmd)
1493 continue;
1494
91921e01 1495 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1496 sdev_printk(KERN_INFO, sdev,
1497 "%s: Sending BDR\n", current->comm));
1da177e4 1498 rtn = scsi_try_bus_device_reset(bdr_scmd);
2f2eb587 1499 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1da177e4 1500 if (!scsi_device_online(sdev) ||
2f2eb587 1501 rtn == FAST_IO_FAIL ||
1da177e4 1502 !scsi_eh_tur(bdr_scmd)) {
937abeaa
CH
1503 list_for_each_entry_safe(scmd, next,
1504 work_q, eh_entry) {
2451079b
JB
1505 if (scmd->device == sdev &&
1506 scsi_eh_action(scmd, rtn) != FAILED)
1da177e4
LT
1507 scsi_eh_finish_cmd(scmd,
1508 done_q);
1509 }
1510 }
1511 } else {
91921e01 1512 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1513 sdev_printk(KERN_INFO, sdev,
1514 "%s: BDR failed\n", current->comm));
1da177e4
LT
1515 }
1516 }
1517
1518 return list_empty(work_q);
1519}
1520
30bd7df8
MC
1521/**
1522 * scsi_eh_target_reset - send target reset if needed
1523 * @shost: scsi host being recovered.
74cf298f 1524 * @work_q: &list_head for pending commands.
30bd7df8
MC
1525 * @done_q: &list_head for processed commands.
1526 *
1527 * Notes:
1528 * Try a target reset.
1529 */
1530static int scsi_eh_target_reset(struct Scsi_Host *shost,
1531 struct list_head *work_q,
1532 struct list_head *done_q)
1533{
98db5195 1534 LIST_HEAD(tmp_list);
3eef6257 1535 LIST_HEAD(check_list);
30bd7df8 1536
98db5195
JB
1537 list_splice_init(work_q, &tmp_list);
1538
1539 while (!list_empty(&tmp_list)) {
1540 struct scsi_cmnd *next, *scmd;
1541 int rtn;
1542 unsigned int id;
b4562022 1543
b4562022 1544 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1545 /* push back on work queue for further processing */
1546 list_splice_init(&check_list, work_q);
1547 list_splice_init(&tmp_list, work_q);
1548 SCSI_LOG_ERROR_RECOVERY(3,
1549 shost_printk(KERN_INFO, shost,
a222b1e2
HR
1550 "%s: Skip target reset, past eh deadline\n",
1551 current->comm));
b4562022
HR
1552 return list_empty(work_q);
1553 }
98db5195
JB
1554
1555 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1556 id = scmd_id(scmd);
30bd7df8 1557
91921e01
HR
1558 SCSI_LOG_ERROR_RECOVERY(3,
1559 shost_printk(KERN_INFO, shost,
1560 "%s: Sending target reset to target %d\n",
1561 current->comm, id));
98db5195
JB
1562 rtn = scsi_try_target_reset(scmd);
1563 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
91921e01
HR
1564 SCSI_LOG_ERROR_RECOVERY(3,
1565 shost_printk(KERN_INFO, shost,
1566 "%s: Target reset failed"
1567 " target: %d\n",
1568 current->comm, id));
98db5195
JB
1569 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1570 if (scmd_id(scmd) != id)
1571 continue;
1572
3eef6257
DJ
1573 if (rtn == SUCCESS)
1574 list_move_tail(&scmd->eh_entry, &check_list);
1575 else if (rtn == FAST_IO_FAIL)
98db5195
JB
1576 scsi_eh_finish_cmd(scmd, done_q);
1577 else
1578 /* push back on work queue for further processing */
1579 list_move(&scmd->eh_entry, work_q);
1580 }
1581 }
30bd7df8 1582
3eef6257 1583 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
30bd7df8
MC
1584}
1585
1da177e4 1586/**
0bf8c869 1587 * scsi_eh_bus_reset - send a bus reset
eb44820c 1588 * @shost: &scsi host being recovered.
74cf298f 1589 * @work_q: &list_head for pending commands.
eb44820c 1590 * @done_q: &list_head for processed commands.
dc8875e1 1591 */
1da177e4
LT
1592static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1593 struct list_head *work_q,
1594 struct list_head *done_q)
1595{
937abeaa 1596 struct scsi_cmnd *scmd, *chan_scmd, *next;
3eef6257 1597 LIST_HEAD(check_list);
1da177e4
LT
1598 unsigned int channel;
1599 int rtn;
1600
1601 /*
1602 * we really want to loop over the various channels, and do this on
1603 * a channel by channel basis. we should also check to see if any
1604 * of the failed commands are on soft_reset devices, and if so, skip
0bf8c869 1605 * the reset.
1da177e4
LT
1606 */
1607
1608 for (channel = 0; channel <= shost->max_channel; channel++) {
b4562022 1609 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1610 list_splice_init(&check_list, work_q);
1611 SCSI_LOG_ERROR_RECOVERY(3,
1612 shost_printk(KERN_INFO, shost,
a222b1e2
HR
1613 "%s: skip BRST, past eh deadline\n",
1614 current->comm));
b4562022
HR
1615 return list_empty(work_q);
1616 }
b4562022 1617
1da177e4
LT
1618 chan_scmd = NULL;
1619 list_for_each_entry(scmd, work_q, eh_entry) {
422c0d61 1620 if (channel == scmd_channel(scmd)) {
1da177e4
LT
1621 chan_scmd = scmd;
1622 break;
1623 /*
1624 * FIXME add back in some support for
1625 * soft_reset devices.
1626 */
1627 }
1628 }
1629
1630 if (!chan_scmd)
1631 continue;
91921e01
HR
1632 SCSI_LOG_ERROR_RECOVERY(3,
1633 shost_printk(KERN_INFO, shost,
1634 "%s: Sending BRST chan: %d\n",
1635 current->comm, channel));
1da177e4 1636 rtn = scsi_try_bus_reset(chan_scmd);
2f2eb587 1637 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
937abeaa 1638 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3eef6257
DJ
1639 if (channel == scmd_channel(scmd)) {
1640 if (rtn == FAST_IO_FAIL)
1da177e4
LT
1641 scsi_eh_finish_cmd(scmd,
1642 done_q);
3eef6257
DJ
1643 else
1644 list_move_tail(&scmd->eh_entry,
1645 &check_list);
1646 }
1da177e4
LT
1647 }
1648 } else {
91921e01
HR
1649 SCSI_LOG_ERROR_RECOVERY(3,
1650 shost_printk(KERN_INFO, shost,
1651 "%s: BRST failed chan: %d\n",
1652 current->comm, channel));
1da177e4
LT
1653 }
1654 }
3eef6257 1655 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1656}
1657
1658/**
0bf8c869 1659 * scsi_eh_host_reset - send a host reset
74cf298f
RD
1660 * @shost: host to be reset.
1661 * @work_q: &list_head for pending commands.
1662 * @done_q: &list_head for processed commands.
dc8875e1 1663 */
91921e01
HR
1664static int scsi_eh_host_reset(struct Scsi_Host *shost,
1665 struct list_head *work_q,
1da177e4
LT
1666 struct list_head *done_q)
1667{
937abeaa 1668 struct scsi_cmnd *scmd, *next;
3eef6257 1669 LIST_HEAD(check_list);
1da177e4 1670 int rtn;
1da177e4
LT
1671
1672 if (!list_empty(work_q)) {
1673 scmd = list_entry(work_q->next,
1674 struct scsi_cmnd, eh_entry);
1675
91921e01
HR
1676 SCSI_LOG_ERROR_RECOVERY(3,
1677 shost_printk(KERN_INFO, shost,
1678 "%s: Sending HRST\n",
1679 current->comm));
1da177e4
LT
1680
1681 rtn = scsi_try_host_reset(scmd);
3eef6257
DJ
1682 if (rtn == SUCCESS) {
1683 list_splice_init(work_q, &check_list);
1684 } else if (rtn == FAST_IO_FAIL) {
937abeaa 1685 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1da177e4
LT
1686 scsi_eh_finish_cmd(scmd, done_q);
1687 }
1688 } else {
91921e01
HR
1689 SCSI_LOG_ERROR_RECOVERY(3,
1690 shost_printk(KERN_INFO, shost,
1691 "%s: HRST failed\n",
1692 current->comm));
1da177e4
LT
1693 }
1694 }
3eef6257 1695 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1da177e4
LT
1696}
1697
1698/**
1699 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
74cf298f
RD
1700 * @work_q: &list_head for pending commands.
1701 * @done_q: &list_head for processed commands.
dc8875e1 1702 */
1da177e4
LT
1703static void scsi_eh_offline_sdevs(struct list_head *work_q,
1704 struct list_head *done_q)
1705{
937abeaa 1706 struct scsi_cmnd *scmd, *next;
0db6ca8a 1707 struct scsi_device *sdev;
1da177e4 1708
937abeaa 1709 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
31765d7d
MW
1710 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1711 "not ready after error recovery\n");
0db6ca8a
BVA
1712 sdev = scmd->device;
1713
1714 mutex_lock(&sdev->state_mutex);
1715 scsi_device_set_state(sdev, SDEV_OFFLINE);
1716 mutex_unlock(&sdev->state_mutex);
1717
1da177e4
LT
1718 scsi_eh_finish_cmd(scmd, done_q);
1719 }
1720 return;
1721}
1722
4a27446f 1723/**
e494f6a7 1724 * scsi_noretry_cmd - determine if command should be failed fast
4a27446f
MC
1725 * @scmd: SCSI cmd to examine.
1726 */
1727int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1728{
1729 switch (host_byte(scmd->result)) {
1730 case DID_OK:
1731 break;
e494f6a7
HR
1732 case DID_TIME_OUT:
1733 goto check_type;
4a27446f 1734 case DID_BUS_BUSY:
33659ebb 1735 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
4a27446f 1736 case DID_PARITY:
33659ebb 1737 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
4a27446f
MC
1738 case DID_ERROR:
1739 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1740 status_byte(scmd->result) == RESERVATION_CONFLICT)
1741 return 0;
df561f66 1742 fallthrough;
4a27446f 1743 case DID_SOFT_ERROR:
33659ebb 1744 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
4a27446f
MC
1745 }
1746
e494f6a7
HR
1747 if (status_byte(scmd->result) != CHECK_CONDITION)
1748 return 0;
4a27446f 1749
e494f6a7
HR
1750check_type:
1751 /*
1752 * assume caller has checked sense and determined
1753 * the check condition was retryable.
1754 */
1755 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
57292b58 1756 blk_rq_is_passthrough(scmd->request))
e494f6a7
HR
1757 return 1;
1758 else
1759 return 0;
4a27446f
MC
1760}
1761
1da177e4
LT
1762/**
1763 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1764 * @scmd: SCSI cmd to examine.
1765 *
1766 * Notes:
1767 * This is *only* called when we are examining the status after sending
1768 * out the actual data command. any commands that are queued for error
1769 * recovery (e.g. test_unit_ready) do *not* come through here.
1770 *
1771 * When this routine returns failed, it means the error handler thread
1772 * is woken. In cases where the error code indicates an error that
1773 * doesn't require the error handler read (i.e. we don't need to
1774 * abort/reset), this function should return SUCCESS.
dc8875e1 1775 */
1da177e4
LT
1776int scsi_decide_disposition(struct scsi_cmnd *scmd)
1777{
1778 int rtn;
1779
1780 /*
1781 * if the device is offline, then we clearly just pass the result back
1782 * up to the top level.
1783 */
1784 if (!scsi_device_online(scmd->device)) {
91921e01
HR
1785 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1786 "%s: device offline - report as SUCCESS\n", __func__));
1da177e4
LT
1787 return SUCCESS;
1788 }
1789
1790 /*
1791 * first check the host byte, to see if there is anything in there
1792 * that would indicate what we need to do.
1793 */
1794 switch (host_byte(scmd->result)) {
1795 case DID_PASSTHROUGH:
1796 /*
1797 * no matter what, pass this through to the upper layer.
1798 * nuke this special code so that it looks like we are saying
1799 * did_ok.
1800 */
1801 scmd->result &= 0xff00ffff;
1802 return SUCCESS;
1803 case DID_OK:
1804 /*
1805 * looks good. drop through, and check the next byte.
1806 */
1807 break;
e494f6a7
HR
1808 case DID_ABORT:
1809 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
8922a908 1810 set_host_byte(scmd, DID_TIME_OUT);
e494f6a7
HR
1811 return SUCCESS;
1812 }
df561f66 1813 fallthrough;
1da177e4
LT
1814 case DID_NO_CONNECT:
1815 case DID_BAD_TARGET:
1da177e4
LT
1816 /*
1817 * note - this means that we just report the status back
1818 * to the top level driver, not that we actually think
1819 * that it indicates SUCCESS.
1820 */
1821 return SUCCESS;
ad95028a 1822 case DID_SOFT_ERROR:
1da177e4
LT
1823 /*
1824 * when the low level driver returns did_soft_error,
0bf8c869 1825 * it is responsible for keeping an internal retry counter
1da177e4 1826 * in order to avoid endless loops (db)
1da177e4 1827 */
1da177e4
LT
1828 goto maybe_retry;
1829 case DID_IMM_RETRY:
1830 return NEEDS_RETRY;
1831
bf341919 1832 case DID_REQUEUE:
1833 return ADD_TO_MLQUEUE;
a4dfaa6f
MC
1834 case DID_TRANSPORT_DISRUPTED:
1835 /*
1836 * LLD/transport was disrupted during processing of the IO.
1837 * The transport class is now blocked/blocking,
1838 * and the transport will decide what to do with the IO
939c2288
MC
1839 * based on its timers and recovery capablilities if
1840 * there are enough retries.
a4dfaa6f 1841 */
939c2288 1842 goto maybe_retry;
a4dfaa6f
MC
1843 case DID_TRANSPORT_FAILFAST:
1844 /*
1845 * The transport decided to failfast the IO (most likely
1846 * the fast io fail tmo fired), so send IO directly upwards.
1847 */
1848 return SUCCESS;
1da177e4
LT
1849 case DID_ERROR:
1850 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1851 status_byte(scmd->result) == RESERVATION_CONFLICT)
1852 /*
1853 * execute reservation conflict processing code
1854 * lower down
1855 */
1856 break;
df561f66 1857 fallthrough;
1da177e4
LT
1858 case DID_BUS_BUSY:
1859 case DID_PARITY:
1860 goto maybe_retry;
1861 case DID_TIME_OUT:
1862 /*
1863 * when we scan the bus, we get timeout messages for
1864 * these commands if there is no device available.
1865 * other hosts report did_no_connect for the same thing.
1866 */
1867 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1868 scmd->cmnd[0] == INQUIRY)) {
1869 return SUCCESS;
1870 } else {
1871 return FAILED;
1872 }
1873 case DID_RESET:
1874 return SUCCESS;
1875 default:
1876 return FAILED;
1877 }
1878
1879 /*
1880 * next, check the message byte.
1881 */
1882 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1883 return FAILED;
1884
1885 /*
1886 * check the status byte to see if this indicates anything special.
1887 */
1888 switch (status_byte(scmd->result)) {
1889 case QUEUE_FULL:
42a6a918 1890 scsi_handle_queue_full(scmd->device);
1da177e4
LT
1891 /*
1892 * the case of trying to send too many commands to a
1893 * tagged queueing device.
1894 */
df561f66 1895 fallthrough;
1da177e4
LT
1896 case BUSY:
1897 /*
1898 * device can't talk to us at the moment. Should only
1899 * occur (SAM-3) when the task queue is empty, so will cause
1900 * the empty queue handling to trigger a stall in the
1901 * device.
1902 */
1903 return ADD_TO_MLQUEUE;
1904 case GOOD:
279afdfe
EM
1905 if (scmd->cmnd[0] == REPORT_LUNS)
1906 scmd->device->sdev_target->expecting_lun_change = 0;
4a84067d 1907 scsi_handle_queue_ramp_up(scmd->device);
df561f66 1908 fallthrough;
1da177e4 1909 case COMMAND_TERMINATED:
1da177e4 1910 return SUCCESS;
a9b589d9
VB
1911 case TASK_ABORTED:
1912 goto maybe_retry;
1da177e4
LT
1913 case CHECK_CONDITION:
1914 rtn = scsi_check_sense(scmd);
1915 if (rtn == NEEDS_RETRY)
1916 goto maybe_retry;
1917 /* if rtn == FAILED, we have no sense information;
1918 * returning FAILED will wake the error handler thread
1919 * to collect the sense and redo the decide
1920 * disposition */
1921 return rtn;
1922 case CONDITION_GOOD:
1923 case INTERMEDIATE_GOOD:
1924 case INTERMEDIATE_C_GOOD:
1925 case ACA_ACTIVE:
1926 /*
1927 * who knows? FIXME(eric)
1928 */
1929 return SUCCESS;
1930
1931 case RESERVATION_CONFLICT:
9ccfc756
JB
1932 sdev_printk(KERN_INFO, scmd->device,
1933 "reservation conflict\n");
2082ebc4 1934 set_host_byte(scmd, DID_NEXUS_FAILURE);
1da177e4
LT
1935 return SUCCESS; /* causes immediate i/o error */
1936 default:
1937 return FAILED;
1938 }
1939 return FAILED;
1940
8ef7fe4b 1941maybe_retry:
1da177e4
LT
1942
1943 /* we requeue for retry because the error was retryable, and
1944 * the request was not marked fast fail. Note that above,
1945 * even if the request is marked fast fail, we still requeue
1946 * for queue congestion conditions (QUEUE_FULL or BUSY) */
8884efab 1947 if ((++scmd->retries) <= scmd->allowed
4a27446f 1948 && !scsi_noretry_cmd(scmd)) {
1da177e4
LT
1949 return NEEDS_RETRY;
1950 } else {
1951 /*
1952 * no more retries - report this one back to upper level.
1953 */
1954 return SUCCESS;
1955 }
1956}
1957
2a842aca 1958static void eh_lock_door_done(struct request *req, blk_status_t status)
f078727b 1959{
92bc5a24 1960 blk_put_request(req);
f078727b
FT
1961}
1962
1da177e4
LT
1963/**
1964 * scsi_eh_lock_door - Prevent medium removal for the specified device
1965 * @sdev: SCSI device to prevent medium removal
1966 *
1967 * Locking:
91bc31fb 1968 * We must be called from process context.
1da177e4
LT
1969 *
1970 * Notes:
1971 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1972 * head of the devices request queue, and continue.
dc8875e1 1973 */
1da177e4
LT
1974static void scsi_eh_lock_door(struct scsi_device *sdev)
1975{
f078727b 1976 struct request *req;
82ed4db4 1977 struct scsi_request *rq;
1da177e4 1978
ff005a06 1979 req = blk_get_request(sdev->request_queue, REQ_OP_SCSI_IN, 0);
a492f075 1980 if (IS_ERR(req))
eb571eea 1981 return;
82ed4db4 1982 rq = scsi_req(req);
1da177e4 1983
82ed4db4
CH
1984 rq->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1985 rq->cmd[1] = 0;
1986 rq->cmd[2] = 0;
1987 rq->cmd[3] = 0;
1988 rq->cmd[4] = SCSI_REMOVAL_PREVENT;
1989 rq->cmd[5] = 0;
1990 rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
f078727b 1991
e8064021 1992 req->rq_flags |= RQF_QUIET;
f078727b 1993 req->timeout = 10 * HZ;
64c7f1d1 1994 rq->retries = 5;
f078727b
FT
1995
1996 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1997}
1da177e4
LT
1998
1999/**
2000 * scsi_restart_operations - restart io operations to the specified host.
2001 * @shost: Host we are restarting.
2002 *
2003 * Notes:
2004 * When we entered the error handler, we blocked all further i/o to
2005 * this device. we need to 'reverse' this process.
dc8875e1 2006 */
1da177e4
LT
2007static void scsi_restart_operations(struct Scsi_Host *shost)
2008{
2009 struct scsi_device *sdev;
939647ee 2010 unsigned long flags;
1da177e4
LT
2011
2012 /*
2013 * If the door was locked, we need to insert a door lock request
2014 * onto the head of the SCSI request queue for the device. There
2015 * is no point trying to lock the door of an off-line device.
2016 */
2017 shost_for_each_device(sdev, shost) {
48379270 2018 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1da177e4 2019 scsi_eh_lock_door(sdev);
48379270
CH
2020 sdev->was_reset = 0;
2021 }
1da177e4
LT
2022 }
2023
2024 /*
2025 * next free up anything directly waiting upon the host. this
2026 * will be requests for character device operations, and also for
2027 * ioctls to queued block devices.
2028 */
b4562022 2029 SCSI_LOG_ERROR_RECOVERY(3,
91921e01 2030 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1da177e4 2031
939647ee
JB
2032 spin_lock_irqsave(shost->host_lock, flags);
2033 if (scsi_host_set_state(shost, SHOST_RUNNING))
2034 if (scsi_host_set_state(shost, SHOST_CANCEL))
2035 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2036 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2037
2038 wake_up(&shost->host_wait);
2039
2040 /*
2041 * finally we need to re-initiate requests that may be pending. we will
2042 * have had everything blocked while error handling is taking place, and
2043 * now that error recovery is done, we will need to ensure that these
2044 * requests are started.
2045 */
2046 scsi_run_host_queues(shost);
57fc2e33
DW
2047
2048 /*
2049 * if eh is active and host_eh_scheduled is pending we need to re-run
2050 * recovery. we do this check after scsi_run_host_queues() to allow
2051 * everything pent up since the last eh run a chance to make forward
2052 * progress before we sync again. Either we'll immediately re-run
2053 * recovery or scsi_device_unbusy() will wake us again when these
2054 * pending commands complete.
2055 */
2056 spin_lock_irqsave(shost->host_lock, flags);
2057 if (shost->host_eh_scheduled)
2058 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2059 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2060 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2061}
2062
2063/**
2064 * scsi_eh_ready_devs - check device ready state and recover if not.
74cf298f
RD
2065 * @shost: host to be recovered.
2066 * @work_q: &list_head for pending commands.
eb44820c 2067 * @done_q: &list_head for processed commands.
dc8875e1 2068 */
dca84e46
DW
2069void scsi_eh_ready_devs(struct Scsi_Host *shost,
2070 struct list_head *work_q,
2071 struct list_head *done_q)
1da177e4
LT
2072{
2073 if (!scsi_eh_stu(shost, work_q, done_q))
2074 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
30bd7df8
MC
2075 if (!scsi_eh_target_reset(shost, work_q, done_q))
2076 if (!scsi_eh_bus_reset(shost, work_q, done_q))
91921e01 2077 if (!scsi_eh_host_reset(shost, work_q, done_q))
30bd7df8
MC
2078 scsi_eh_offline_sdevs(work_q,
2079 done_q);
1da177e4 2080}
dca84e46 2081EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1da177e4
LT
2082
2083/**
2084 * scsi_eh_flush_done_q - finish processed commands or retry them.
2085 * @done_q: list_head of processed commands.
dc8875e1 2086 */
041c5fc3 2087void scsi_eh_flush_done_q(struct list_head *done_q)
1da177e4 2088{
937abeaa 2089 struct scsi_cmnd *scmd, *next;
1da177e4 2090
937abeaa
CH
2091 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2092 list_del_init(&scmd->eh_entry);
1da177e4 2093 if (scsi_device_online(scmd->device) &&
4a27446f 2094 !scsi_noretry_cmd(scmd) &&
8884efab 2095 (++scmd->retries <= scmd->allowed)) {
91921e01
HR
2096 SCSI_LOG_ERROR_RECOVERY(3,
2097 scmd_printk(KERN_INFO, scmd,
470613b4
HR
2098 "%s: flush retry cmd\n",
2099 current->comm));
1da177e4
LT
2100 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2101 } else {
793698ce
PM
2102 /*
2103 * If just we got sense for the device (called
2104 * scsi_eh_get_sense), scmd->result is already
2105 * set, do not set DRIVER_TIMEOUT.
2106 */
1da177e4
LT
2107 if (!scmd->result)
2108 scmd->result |= (DRIVER_TIMEOUT << 24);
91921e01
HR
2109 SCSI_LOG_ERROR_RECOVERY(3,
2110 scmd_printk(KERN_INFO, scmd,
470613b4
HR
2111 "%s: flush finish cmd\n",
2112 current->comm));
1da177e4
LT
2113 scsi_finish_command(scmd);
2114 }
2115 }
2116}
041c5fc3 2117EXPORT_SYMBOL(scsi_eh_flush_done_q);
1da177e4
LT
2118
2119/**
2120 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2121 * @shost: Host to unjam.
2122 *
2123 * Notes:
2124 * When we come in here, we *know* that all commands on the bus have
2125 * either completed, failed or timed out. we also know that no further
2126 * commands are being sent to the host, so things are relatively quiet
2127 * and we have freedom to fiddle with things as we wish.
2128 *
2129 * This is only the *default* implementation. it is possible for
2130 * individual drivers to supply their own version of this function, and
2131 * if the maintainer wishes to do this, it is strongly suggested that
2132 * this function be taken as a template and modified. this function
2133 * was designed to correctly handle problems for about 95% of the
2134 * different cases out there, and it should always provide at least a
2135 * reasonable amount of error recovery.
2136 *
2137 * Any command marked 'failed' or 'timeout' must eventually have
2138 * scsi_finish_cmd() called for it. we do all of the retry stuff
2139 * here, so when we restart the host after we return it should have an
2140 * empty queue.
dc8875e1 2141 */
1da177e4
LT
2142static void scsi_unjam_host(struct Scsi_Host *shost)
2143{
2144 unsigned long flags;
2145 LIST_HEAD(eh_work_q);
2146 LIST_HEAD(eh_done_q);
2147
2148 spin_lock_irqsave(shost->host_lock, flags);
2149 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2150 spin_unlock_irqrestore(shost->host_lock, flags);
2151
2152 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2153
2154 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
a0658632 2155 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1da177e4 2156
b4562022 2157 spin_lock_irqsave(shost->host_lock, flags);
bb3b621a 2158 if (shost->eh_deadline != -1)
b4562022
HR
2159 shost->last_reset = 0;
2160 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2161 scsi_eh_flush_done_q(&eh_done_q);
2162}
2163
2164/**
ad42eb1b 2165 * scsi_error_handler - SCSI error handler thread
1da177e4
LT
2166 * @data: Host for which we are running.
2167 *
2168 * Notes:
ad42eb1b
CH
2169 * This is the main error handling loop. This is run as a kernel thread
2170 * for every SCSI host and handles all error handling activity.
dc8875e1 2171 */
1da177e4
LT
2172int scsi_error_handler(void *data)
2173{
ad42eb1b 2174 struct Scsi_Host *shost = data;
1da177e4 2175
1da177e4 2176 /*
ad42eb1b
CH
2177 * We use TASK_INTERRUPTIBLE so that the thread is not
2178 * counted against the load average as a running process.
2179 * We never actually get interrupted because kthread_run
c03264a7 2180 * disables signal delivery for the created thread.
1da177e4 2181 */
537b604c
MH
2182 while (true) {
2183 /*
2184 * The sequence in kthread_stop() sets the stop flag first
2185 * then wakes the process. To avoid missed wakeups, the task
2186 * should always be in a non running state before the stop
2187 * flag is checked
2188 */
b9d5c6b7 2189 set_current_state(TASK_INTERRUPTIBLE);
537b604c
MH
2190 if (kthread_should_stop())
2191 break;
2192
ee7863bc 2193 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
c84b023a 2194 shost->host_failed != scsi_host_busy(shost)) {
ad42eb1b 2195 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2196 shost_printk(KERN_INFO, shost,
2197 "scsi_eh_%d: sleeping\n",
2198 shost->host_no));
3ed7a470 2199 schedule();
3ed7a470
JB
2200 continue;
2201 }
1da177e4 2202
3ed7a470 2203 __set_current_state(TASK_RUNNING);
ad42eb1b 2204 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2205 shost_printk(KERN_INFO, shost,
2206 "scsi_eh_%d: waking up %d/%d/%d\n",
2207 shost->host_no, shost->host_eh_scheduled,
74665016 2208 shost->host_failed,
c84b023a 2209 scsi_host_busy(shost)));
1da177e4 2210
1da177e4
LT
2211 /*
2212 * We have a host that is failing for some reason. Figure out
2213 * what we need to do to get it up and online again (if we can).
2214 * If we fail, we end up taking the thing offline.
2215 */
ae0751ff 2216 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
bc4f2401 2217 SCSI_LOG_ERROR_RECOVERY(1,
a222b1e2
HR
2218 shost_printk(KERN_ERR, shost,
2219 "scsi_eh_%d: unable to autoresume\n",
2220 shost->host_no));
bc4f2401
AS
2221 continue;
2222 }
2223
9227c33d
CH
2224 if (shost->transportt->eh_strategy_handler)
2225 shost->transportt->eh_strategy_handler(shost);
1da177e4
LT
2226 else
2227 scsi_unjam_host(shost);
2228
72d8c36e
WF
2229 /* All scmds have been handled */
2230 shost->host_failed = 0;
2231
1da177e4
LT
2232 /*
2233 * Note - if the above fails completely, the action is to take
2234 * individual devices offline and flush the queue of any
2235 * outstanding requests that may have been pending. When we
2236 * restart, we restart any I/O to any other devices on the bus
2237 * which are still online.
2238 */
2239 scsi_restart_operations(shost);
ae0751ff
LM
2240 if (!shost->eh_noresume)
2241 scsi_autopm_put_host(shost);
1da177e4 2242 }
461a0ffb
SR
2243 __set_current_state(TASK_RUNNING);
2244
ad42eb1b 2245 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2246 shost_printk(KERN_INFO, shost,
2247 "Error handler scsi_eh_%d exiting\n",
2248 shost->host_no));
3ed7a470 2249 shost->ehandler = NULL;
1da177e4
LT
2250 return 0;
2251}
2252
2253/*
2254 * Function: scsi_report_bus_reset()
2255 *
2256 * Purpose: Utility function used by low-level drivers to report that
2257 * they have observed a bus reset on the bus being handled.
2258 *
2259 * Arguments: shost - Host in question
2260 * channel - channel on which reset was observed.
2261 *
2262 * Returns: Nothing
2263 *
2264 * Lock status: Host lock must be held.
2265 *
2266 * Notes: This only needs to be called if the reset is one which
2267 * originates from an unknown location. Resets originated
2268 * by the mid-level itself don't need to call this, but there
2269 * should be no harm.
2270 *
2271 * The main purpose of this is to make sure that a CHECK_CONDITION
2272 * is properly treated.
2273 */
2274void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2275{
2276 struct scsi_device *sdev;
2277
2278 __shost_for_each_device(sdev, shost) {
30bd7df8
MC
2279 if (channel == sdev_channel(sdev))
2280 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2281 }
2282}
2283EXPORT_SYMBOL(scsi_report_bus_reset);
2284
2285/*
2286 * Function: scsi_report_device_reset()
2287 *
2288 * Purpose: Utility function used by low-level drivers to report that
2289 * they have observed a device reset on the device being handled.
2290 *
2291 * Arguments: shost - Host in question
2292 * channel - channel on which reset was observed
2293 * target - target on which reset was observed
2294 *
2295 * Returns: Nothing
2296 *
2297 * Lock status: Host lock must be held
2298 *
2299 * Notes: This only needs to be called if the reset is one which
2300 * originates from an unknown location. Resets originated
2301 * by the mid-level itself don't need to call this, but there
2302 * should be no harm.
2303 *
2304 * The main purpose of this is to make sure that a CHECK_CONDITION
2305 * is properly treated.
2306 */
2307void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2308{
2309 struct scsi_device *sdev;
2310
2311 __shost_for_each_device(sdev, shost) {
422c0d61 2312 if (channel == sdev_channel(sdev) &&
30bd7df8
MC
2313 target == sdev_id(sdev))
2314 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2315 }
2316}
2317EXPORT_SYMBOL(scsi_report_device_reset);
2318
2319static void
2320scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2321{
2322}
2323
176aa9d6
CH
2324/**
2325 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2326 * @dev: scsi_device to operate on
2327 * @arg: reset type (see sg.h)
1da177e4
LT
2328 */
2329int
176aa9d6 2330scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
1da177e4 2331{
bc4f2401 2332 struct scsi_cmnd *scmd;
d7a1bb0a 2333 struct Scsi_Host *shost = dev->host;
e9c787e6 2334 struct request *rq;
d7a1bb0a 2335 unsigned long flags;
176aa9d6
CH
2336 int error = 0, rtn, val;
2337
2338 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2339 return -EACCES;
2340
2341 error = get_user(val, arg);
2342 if (error)
2343 return error;
1da177e4 2344
bc4f2401 2345 if (scsi_autopm_get_host(shost) < 0)
176aa9d6 2346 return -EIO;
bc4f2401 2347
176aa9d6 2348 error = -EIO;
e9c787e6
CH
2349 rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) +
2350 shost->hostt->cmd_size, GFP_KERNEL);
2351 if (!rq)
95eeb5f5 2352 goto out_put_autopm_host;
e9c787e6 2353 blk_rq_init(NULL, rq);
95eeb5f5 2354
e9c787e6
CH
2355 scmd = (struct scsi_cmnd *)(rq + 1);
2356 scsi_init_command(dev, scmd);
2357 scmd->request = rq;
82ed4db4 2358 scmd->cmnd = scsi_req(rq)->cmd;
64a87b24 2359
1da177e4 2360 scmd->scsi_done = scsi_reset_provider_done_command;
30b0c37b 2361 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1da177e4
LT
2362
2363 scmd->cmd_len = 0;
2364
2365 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1da177e4 2366
d7a1bb0a
JS
2367 spin_lock_irqsave(shost->host_lock, flags);
2368 shost->tmf_in_progress = 1;
2369 spin_unlock_irqrestore(shost->host_lock, flags);
2370
176aa9d6
CH
2371 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2372 case SG_SCSI_RESET_NOTHING:
2373 rtn = SUCCESS;
2374 break;
2375 case SG_SCSI_RESET_DEVICE:
1da177e4 2376 rtn = scsi_try_bus_device_reset(scmd);
176aa9d6 2377 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
1da177e4 2378 break;
df561f66 2379 fallthrough;
176aa9d6 2380 case SG_SCSI_RESET_TARGET:
30bd7df8 2381 rtn = scsi_try_target_reset(scmd);
176aa9d6 2382 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
30bd7df8 2383 break;
df561f66 2384 fallthrough;
176aa9d6 2385 case SG_SCSI_RESET_BUS:
1da177e4 2386 rtn = scsi_try_bus_reset(scmd);
176aa9d6 2387 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
1da177e4 2388 break;
df561f66 2389 fallthrough;
176aa9d6 2390 case SG_SCSI_RESET_HOST:
1da177e4 2391 rtn = scsi_try_host_reset(scmd);
176aa9d6
CH
2392 if (rtn == SUCCESS)
2393 break;
df561f66 2394 fallthrough;
3bf2ff67 2395 default:
1da177e4 2396 rtn = FAILED;
176aa9d6 2397 break;
1da177e4
LT
2398 }
2399
176aa9d6
CH
2400 error = (rtn == SUCCESS) ? 0 : -EIO;
2401
d7a1bb0a
JS
2402 spin_lock_irqsave(shost->host_lock, flags);
2403 shost->tmf_in_progress = 0;
2404 spin_unlock_irqrestore(shost->host_lock, flags);
2405
2406 /*
2407 * be sure to wake up anyone who was sleeping or had their queue
2408 * suspended while we performed the TMF.
2409 */
2410 SCSI_LOG_ERROR_RECOVERY(3,
91921e01
HR
2411 shost_printk(KERN_INFO, shost,
2412 "waking up host to restart after TMF\n"));
d7a1bb0a
JS
2413
2414 wake_up(&shost->host_wait);
d7a1bb0a
JS
2415 scsi_run_host_queues(shost);
2416
e9c787e6 2417 kfree(rq);
0f121dd8 2418
04796336 2419out_put_autopm_host:
bc4f2401 2420 scsi_autopm_put_host(shost);
176aa9d6 2421 return error;
1da177e4 2422}
1da177e4 2423
4753cbc0
HR
2424bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2425 struct scsi_sense_hdr *sshdr)
1da177e4
LT
2426{
2427 return scsi_normalize_sense(cmd->sense_buffer,
b80ca4f7 2428 SCSI_SENSE_BUFFERSIZE, sshdr);
1da177e4
LT
2429}
2430EXPORT_SYMBOL(scsi_command_normalize_sense);
2431
1da177e4 2432/**
eb44820c 2433 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1da177e4
LT
2434 * @sense_buffer: byte array of sense data
2435 * @sb_len: number of valid bytes in sense_buffer
2436 * @info_out: pointer to 64 integer where 8 or 4 byte information
2437 * field will be placed if found.
2438 *
2439 * Return value:
2908769c 2440 * true if information field found, false if not found.
dc8875e1 2441 */
2908769c
DLM
2442bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
2443 u64 *info_out)
1da177e4 2444{
1da177e4 2445 const u8 * ucp;
1da177e4
LT
2446
2447 if (sb_len < 7)
2908769c 2448 return false;
1da177e4
LT
2449 switch (sense_buffer[0] & 0x7f) {
2450 case 0x70:
2451 case 0x71:
2452 if (sense_buffer[0] & 0x80) {
2908769c
DLM
2453 *info_out = get_unaligned_be32(&sense_buffer[3]);
2454 return true;
2455 }
2456 return false;
1da177e4
LT
2457 case 0x72:
2458 case 0x73:
2459 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2460 0 /* info desc */);
2461 if (ucp && (0xa == ucp[1])) {
2908769c
DLM
2462 *info_out = get_unaligned_be64(&ucp[4]);
2463 return true;
2464 }
2465 return false;
1da177e4 2466 default:
2908769c 2467 return false;
1da177e4
LT
2468 }
2469}
2470EXPORT_SYMBOL(scsi_get_sense_info_fld);