]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - drivers/scsi/scsi_error.c
scsi: Do not display buffer pointers in scsi_log_send()
[mirror_ubuntu-zesty-kernel.git] / drivers / scsi / scsi_error.c
CommitLineData
1da177e4
LT
1/*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
3 *
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
0bf8c869 6 * Leonard Zubkoff and David Miller at Linux Expo,
1da177e4
LT
7 * ideas originating from all over the place.
8 *
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11 *
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
0bf8c869 13 * minor cleanups.
1da177e4
LT
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
5a0e3ad6 19#include <linux/gfp.h>
1da177e4
LT
20#include <linux/timer.h>
21#include <linux/string.h>
1da177e4 22#include <linux/kernel.h>
83144186 23#include <linux/freezer.h>
c5478def 24#include <linux/kthread.h>
1da177e4
LT
25#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
fc73648a 28#include <linux/jiffies.h>
1da177e4
LT
29
30#include <scsi/scsi.h>
beb40487 31#include <scsi/scsi_cmnd.h>
1da177e4
LT
32#include <scsi/scsi_dbg.h>
33#include <scsi/scsi_device.h>
18a4d0a2 34#include <scsi/scsi_driver.h>
1da177e4 35#include <scsi/scsi_eh.h>
c829c394 36#include <scsi/scsi_transport.h>
1da177e4
LT
37#include <scsi/scsi_host.h>
38#include <scsi/scsi_ioctl.h>
176aa9d6 39#include <scsi/sg.h>
1da177e4
LT
40
41#include "scsi_priv.h"
42#include "scsi_logging.h"
79ee8304 43#include "scsi_transport_api.h"
1da177e4 44
bf816235
KT
45#include <trace/events/scsi.h>
46
14216561
JB
47static void scsi_eh_done(struct scsi_cmnd *scmd);
48
1da177e4
LT
49/*
50 * These should *probably* be handled by the host itself.
51 * Since it is allowed to sleep, it probably should.
52 */
53#define BUS_RESET_SETTLE_TIME (10)
54#define HOST_RESET_SETTLE_TIME (10)
55
3eef6257 56static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
e494f6a7
HR
57static int scsi_try_to_abort_cmd(struct scsi_host_template *,
58 struct scsi_cmnd *);
3eef6257 59
1da177e4
LT
60/* called with shost->host_lock held */
61void scsi_eh_wakeup(struct Scsi_Host *shost)
62{
74665016 63 if (atomic_read(&shost->host_busy) == shost->host_failed) {
bf816235 64 trace_scsi_eh_wakeup(shost);
3ed7a470 65 wake_up_process(shost->ehandler);
91921e01
HR
66 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
67 "Waking error handler thread\n"));
1da177e4
LT
68 }
69}
f8bbfc24
TH
70
71/**
72 * scsi_schedule_eh - schedule EH for SCSI host
73 * @shost: SCSI host to invoke error handling on.
74 *
75 * Schedule SCSI EH without scmd.
dc8875e1 76 */
f8bbfc24
TH
77void scsi_schedule_eh(struct Scsi_Host *shost)
78{
79 unsigned long flags;
80
81 spin_lock_irqsave(shost->host_lock, flags);
82
83 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
84 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
85 shost->host_eh_scheduled++;
86 scsi_eh_wakeup(shost);
87 }
88
89 spin_unlock_irqrestore(shost->host_lock, flags);
90}
91EXPORT_SYMBOL_GPL(scsi_schedule_eh);
1da177e4 92
b4562022
HR
93static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
94{
bb3b621a 95 if (!shost->last_reset || shost->eh_deadline == -1)
b4562022
HR
96 return 0;
97
76ad3e59
HR
98 /*
99 * 32bit accesses are guaranteed to be atomic
100 * (on all supported architectures), so instead
101 * of using a spinlock we can as well double check
bb3b621a 102 * if eh_deadline has been set to 'off' during the
76ad3e59
HR
103 * time_before call.
104 */
105 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
bb3b621a 106 shost->eh_deadline > -1)
b4562022
HR
107 return 0;
108
109 return 1;
110}
111
e494f6a7
HR
112/**
113 * scmd_eh_abort_handler - Handle command aborts
114 * @work: command to be aborted.
115 */
116void
117scmd_eh_abort_handler(struct work_struct *work)
118{
119 struct scsi_cmnd *scmd =
120 container_of(work, struct scsi_cmnd, abort_work.work);
121 struct scsi_device *sdev = scmd->device;
e494f6a7
HR
122 int rtn;
123
e494f6a7 124 if (scsi_host_eh_past_deadline(sdev->host)) {
e494f6a7
HR
125 SCSI_LOG_ERROR_RECOVERY(3,
126 scmd_printk(KERN_INFO, scmd,
127 "scmd %p eh timeout, not aborting\n",
128 scmd));
129 } else {
e494f6a7
HR
130 SCSI_LOG_ERROR_RECOVERY(3,
131 scmd_printk(KERN_INFO, scmd,
132 "aborting command %p\n", scmd));
133 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
134 if (rtn == SUCCESS) {
8922a908 135 set_host_byte(scmd, DID_TIME_OUT);
bb3b621a
RM
136 if (scsi_host_eh_past_deadline(sdev->host)) {
137 SCSI_LOG_ERROR_RECOVERY(3,
138 scmd_printk(KERN_INFO, scmd,
139 "scmd %p eh timeout, "
140 "not retrying aborted "
141 "command\n", scmd));
142 } else if (!scsi_noretry_cmd(scmd) &&
e494f6a7
HR
143 (++scmd->retries <= scmd->allowed)) {
144 SCSI_LOG_ERROR_RECOVERY(3,
145 scmd_printk(KERN_WARNING, scmd,
146 "scmd %p retry "
147 "aborted command\n", scmd));
148 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
bb3b621a 149 return;
e494f6a7
HR
150 } else {
151 SCSI_LOG_ERROR_RECOVERY(3,
152 scmd_printk(KERN_WARNING, scmd,
153 "scmd %p finish "
154 "aborted command\n", scmd));
155 scsi_finish_command(scmd);
bb3b621a 156 return;
e494f6a7 157 }
bb3b621a
RM
158 } else {
159 SCSI_LOG_ERROR_RECOVERY(3,
160 scmd_printk(KERN_INFO, scmd,
883a030f
HR
161 "scmd %p abort %s\n", scmd,
162 (rtn == FAST_IO_FAIL) ?
163 "not send" : "failed"));
e494f6a7 164 }
e494f6a7
HR
165 }
166
167 if (!scsi_eh_scmd_add(scmd, 0)) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_WARNING, scmd,
170 "scmd %p terminate "
171 "aborted command\n", scmd));
8922a908 172 set_host_byte(scmd, DID_TIME_OUT);
e494f6a7
HR
173 scsi_finish_command(scmd);
174 }
175}
176
177/**
178 * scsi_abort_command - schedule a command abort
179 * @scmd: scmd to abort.
180 *
181 * We only need to abort commands after a command timeout
182 */
183static int
184scsi_abort_command(struct scsi_cmnd *scmd)
185{
186 struct scsi_device *sdev = scmd->device;
187 struct Scsi_Host *shost = sdev->host;
188 unsigned long flags;
189
190 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
191 /*
192 * Retry after abort failed, escalate to next level.
193 */
7daf4804 194 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
e494f6a7
HR
195 SCSI_LOG_ERROR_RECOVERY(3,
196 scmd_printk(KERN_INFO, scmd,
197 "scmd %p previous abort failed\n", scmd));
fcc95a76 198 BUG_ON(delayed_work_pending(&scmd->abort_work));
e494f6a7
HR
199 return FAILED;
200 }
201
202 /*
203 * Do not try a command abort if
204 * SCSI EH has already started.
205 */
206 spin_lock_irqsave(shost->host_lock, flags);
207 if (scsi_host_in_recovery(shost)) {
208 spin_unlock_irqrestore(shost->host_lock, flags);
209 SCSI_LOG_ERROR_RECOVERY(3,
210 scmd_printk(KERN_INFO, scmd,
211 "scmd %p not aborting, host in recovery\n",
212 scmd));
213 return FAILED;
214 }
215
bb3b621a 216 if (shost->eh_deadline != -1 && !shost->last_reset)
e494f6a7
HR
217 shost->last_reset = jiffies;
218 spin_unlock_irqrestore(shost->host_lock, flags);
219
220 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
221 SCSI_LOG_ERROR_RECOVERY(3,
222 scmd_printk(KERN_INFO, scmd,
223 "scmd %p abort scheduled\n", scmd));
224 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
225 return SUCCESS;
226}
227
1da177e4
LT
228/**
229 * scsi_eh_scmd_add - add scsi cmd to error handling.
230 * @scmd: scmd to run eh on.
231 * @eh_flag: optional SCSI_EH flag.
232 *
233 * Return value:
234 * 0 on failure.
dc8875e1 235 */
1da177e4
LT
236int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
237{
238 struct Scsi_Host *shost = scmd->device->host;
239 unsigned long flags;
939647ee 240 int ret = 0;
1da177e4 241
3ed7a470 242 if (!shost->ehandler)
1da177e4
LT
243 return 0;
244
245 spin_lock_irqsave(shost->host_lock, flags);
939647ee
JB
246 if (scsi_host_set_state(shost, SHOST_RECOVERY))
247 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
248 goto out_unlock;
1da177e4 249
bb3b621a 250 if (shost->eh_deadline != -1 && !shost->last_reset)
b4562022
HR
251 shost->last_reset = jiffies;
252
939647ee 253 ret = 1;
e494f6a7
HR
254 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
255 eh_flag &= ~SCSI_EH_CANCEL_CMD;
3111b0d1 256 scmd->eh_eflags |= eh_flag;
1da177e4 257 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
1da177e4
LT
258 shost->host_failed++;
259 scsi_eh_wakeup(shost);
939647ee 260 out_unlock:
1da177e4 261 spin_unlock_irqrestore(shost->host_lock, flags);
939647ee 262 return ret;
1da177e4
LT
263}
264
1da177e4
LT
265/**
266 * scsi_times_out - Timeout function for normal scsi commands.
242f9dcb 267 * @req: request that is timing out.
1da177e4
LT
268 *
269 * Notes:
270 * We do not need to lock this. There is the potential for a race
271 * only in that the normal completion handling might run, but if the
272 * normal completion function determines that the timer has already
273 * fired, then it mustn't do anything.
dc8875e1 274 */
242f9dcb 275enum blk_eh_timer_return scsi_times_out(struct request *req)
1da177e4 276{
242f9dcb 277 struct scsi_cmnd *scmd = req->special;
242f9dcb 278 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
0bf8c869 279 struct Scsi_Host *host = scmd->device->host;
6c5f8ce1 280
bf816235 281 trace_scsi_dispatch_cmd_timeout(scmd);
1da177e4
LT
282 scsi_log_completion(scmd, TIMEOUT_ERROR);
283
bb3b621a 284 if (host->eh_deadline != -1 && !host->last_reset)
b4562022
HR
285 host->last_reset = jiffies;
286
0bf8c869
JJ
287 if (host->transportt->eh_timed_out)
288 rtn = host->transportt->eh_timed_out(scmd);
289 else if (host->hostt->eh_timed_out)
290 rtn = host->hostt->eh_timed_out(scmd);
1da177e4 291
a33c070b
HR
292 if (rtn == BLK_EH_NOT_HANDLED) {
293 if (!host->hostt->no_async_abort &&
294 scsi_abort_command(scmd) == SUCCESS)
e494f6a7
HR
295 return BLK_EH_NOT_HANDLED;
296
a33c070b
HR
297 set_host_byte(scmd, DID_TIME_OUT);
298 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
299 rtn = BLK_EH_HANDLED;
300 }
242f9dcb 301
fa990781 302 return rtn;
1da177e4
LT
303}
304
305/**
306 * scsi_block_when_processing_errors - Prevent cmds from being queued.
307 * @sdev: Device on which we are performing recovery.
308 *
309 * Description:
310 * We block until the host is out of error recovery, and then check to
311 * see whether the host or the device is offline.
312 *
313 * Return value:
314 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
dc8875e1 315 */
1da177e4
LT
316int scsi_block_when_processing_errors(struct scsi_device *sdev)
317{
318 int online;
319
939647ee 320 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
1da177e4
LT
321
322 online = scsi_device_online(sdev);
323
91921e01
HR
324 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
325 "%s: rtn: %d\n", __func__, online));
1da177e4
LT
326
327 return online;
328}
329EXPORT_SYMBOL(scsi_block_when_processing_errors);
330
331#ifdef CONFIG_SCSI_LOGGING
332/**
333 * scsi_eh_prt_fail_stats - Log info on failures.
334 * @shost: scsi host being recovered.
335 * @work_q: Queue of scsi cmds to process.
dc8875e1 336 */
1da177e4
LT
337static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
338 struct list_head *work_q)
339{
340 struct scsi_cmnd *scmd;
341 struct scsi_device *sdev;
342 int total_failures = 0;
343 int cmd_failed = 0;
344 int cmd_cancel = 0;
345 int devices_failed = 0;
346
347 shost_for_each_device(sdev, shost) {
348 list_for_each_entry(scmd, work_q, eh_entry) {
349 if (scmd->device == sdev) {
350 ++total_failures;
3111b0d1 351 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
1da177e4 352 ++cmd_cancel;
0bf8c869 353 else
1da177e4
LT
354 ++cmd_failed;
355 }
356 }
357
358 if (cmd_cancel || cmd_failed) {
359 SCSI_LOG_ERROR_RECOVERY(3,
a3a790dc 360 shost_printk(KERN_INFO, shost,
9ccfc756 361 "%s: cmds failed: %d, cancel: %d\n",
cadbd4a5 362 __func__, cmd_failed,
9ccfc756 363 cmd_cancel));
1da177e4
LT
364 cmd_cancel = 0;
365 cmd_failed = 0;
366 ++devices_failed;
367 }
368 }
369
91921e01
HR
370 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
371 "Total of %d commands on %d"
372 " devices require eh work\n",
0bf8c869 373 total_failures, devices_failed));
1da177e4
LT
374}
375#endif
376
279afdfe
EM
377 /**
378 * scsi_report_lun_change - Set flag on all *other* devices on the same target
379 * to indicate that a UNIT ATTENTION is expected.
380 * @sdev: Device reporting the UNIT ATTENTION
381 */
382static void scsi_report_lun_change(struct scsi_device *sdev)
383{
384 sdev->sdev_target->expecting_lun_change = 1;
385}
386
387/**
388 * scsi_report_sense - Examine scsi sense information and log messages for
389 * certain conditions, also issue uevents for some of them.
390 * @sdev: Device reporting the sense code
391 * @sshdr: sshdr to be examined
392 */
393static void scsi_report_sense(struct scsi_device *sdev,
394 struct scsi_sense_hdr *sshdr)
395{
396 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
397
398 if (sshdr->sense_key == UNIT_ATTENTION) {
399 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
400 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
401 sdev_printk(KERN_WARNING, sdev,
402 "Inquiry data has changed");
403 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
404 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
405 scsi_report_lun_change(sdev);
406 sdev_printk(KERN_WARNING, sdev,
407 "Warning! Received an indication that the "
408 "LUN assignments on this target have "
409 "changed. The Linux SCSI layer does not "
410 "automatically remap LUN assignments.\n");
411 } else if (sshdr->asc == 0x3f)
412 sdev_printk(KERN_WARNING, sdev,
413 "Warning! Received an indication that the "
414 "operating parameters on this target have "
415 "changed. The Linux SCSI layer does not "
416 "automatically adjust these parameters.\n");
417
418 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
419 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
420 sdev_printk(KERN_WARNING, sdev,
421 "Warning! Received an indication that the "
422 "LUN reached a thin provisioning soft "
423 "threshold.\n");
424 }
425
426 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
427 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
428 sdev_printk(KERN_WARNING, sdev,
429 "Mode parameters changed");
430 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
431 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
432 sdev_printk(KERN_WARNING, sdev,
433 "Capacity data has changed");
434 } else if (sshdr->asc == 0x2a)
435 sdev_printk(KERN_WARNING, sdev,
436 "Parameters changed");
437 }
438
439 if (evt_type != SDEV_EVT_MAXBITS) {
440 set_bit(evt_type, sdev->pending_events);
441 schedule_work(&sdev->event_work);
442 }
443}
444
1da177e4
LT
445/**
446 * scsi_check_sense - Examine scsi cmd sense
447 * @scmd: Cmd to have sense checked.
448 *
449 * Return value:
87f14e65 450 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1da177e4
LT
451 *
452 * Notes:
453 * When a deferred error is detected the current command has
454 * not been executed and needs retrying.
dc8875e1 455 */
1da177e4
LT
456static int scsi_check_sense(struct scsi_cmnd *scmd)
457{
a6a8d9f8 458 struct scsi_device *sdev = scmd->device;
1da177e4
LT
459 struct scsi_sense_hdr sshdr;
460
461 if (! scsi_command_normalize_sense(scmd, &sshdr))
462 return FAILED; /* no valid sense data */
463
279afdfe
EM
464 scsi_report_sense(sdev, &sshdr);
465
1da177e4
LT
466 if (scsi_sense_is_deferred(&sshdr))
467 return NEEDS_RETRY;
468
a6a8d9f8
CS
469 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
470 sdev->scsi_dh_data->scsi_dh->check_sense) {
471 int rc;
472
473 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
474 if (rc != SCSI_RETURN_NOT_HANDLED)
475 return rc;
476 /* handler does not care. Drop down to default handling */
477 }
478
e925cc43
CH
479 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
480 /*
481 * nasty: for mid-layer issued TURs, we need to return the
482 * actual sense data without any recovery attempt. For eh
483 * issued ones, we need to try to recover and interpret
484 */
485 return SUCCESS;
486
1da177e4
LT
487 /*
488 * Previous logic looked for FILEMARK, EOM or ILI which are
489 * mainly associated with tapes and returned SUCCESS.
490 */
491 if (sshdr.response_code == 0x70) {
492 /* fixed format */
493 if (scmd->sense_buffer[2] & 0xe0)
494 return SUCCESS;
495 } else {
496 /*
497 * descriptor format: look for "stream commands sense data
498 * descriptor" (see SSC-3). Assume single sense data
499 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
500 */
501 if ((sshdr.additional_length > 3) &&
502 (scmd->sense_buffer[8] == 0x4) &&
503 (scmd->sense_buffer[11] & 0xe0))
504 return SUCCESS;
505 }
506
507 switch (sshdr.sense_key) {
508 case NO_SENSE:
509 return SUCCESS;
510 case RECOVERED_ERROR:
511 return /* soft_error */ SUCCESS;
512
513 case ABORTED_COMMAND:
511e44f4
MP
514 if (sshdr.asc == 0x10) /* DIF */
515 return SUCCESS;
516
1da177e4
LT
517 return NEEDS_RETRY;
518 case NOT_READY:
519 case UNIT_ATTENTION:
520 /*
521 * if we are expecting a cc/ua because of a bus reset that we
522 * performed, treat this just as a retry. otherwise this is
523 * information that we should pass up to the upper-level driver
524 * so that we can deal with it there.
525 */
526 if (scmd->device->expecting_cc_ua) {
dfcf7775
TH
527 /*
528 * Because some device does not queue unit
529 * attentions correctly, we carefully check
530 * additional sense code and qualifier so as
531 * not to squash media change unit attention.
532 */
533 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
534 scmd->device->expecting_cc_ua = 0;
535 return NEEDS_RETRY;
536 }
1da177e4 537 }
279afdfe
EM
538 /*
539 * we might also expect a cc/ua if another LUN on the target
540 * reported a UA with an ASC/ASCQ of 3F 0E -
541 * REPORTED LUNS DATA HAS CHANGED.
542 */
543 if (scmd->device->sdev_target->expecting_lun_change &&
544 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
545 return NEEDS_RETRY;
1da177e4 546 /*
0bf8c869 547 * if the device is in the process of becoming ready, we
1da177e4
LT
548 * should retry.
549 */
550 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
551 return NEEDS_RETRY;
552 /*
553 * if the device is not started, we need to wake
554 * the error handler to start the motor
555 */
556 if (scmd->device->allow_restart &&
557 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
558 return FAILED;
02e031cb
CH
559 /*
560 * Pass the UA upwards for a determination in the completion
561 * functions.
562 */
563 return SUCCESS;
1da177e4 564
63583cca 565 /* these are not supported */
a9d6ceb8
HR
566 case DATA_PROTECT:
567 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
568 /* Thin provisioning hard threshold reached */
569 set_host_byte(scmd, DID_ALLOC_FAILURE);
570 return SUCCESS;
571 }
1da177e4
LT
572 case COPY_ABORTED:
573 case VOLUME_OVERFLOW:
574 case MISCOMPARE:
63583cca 575 case BLANK_CHECK:
87f14e65
HR
576 set_host_byte(scmd, DID_TARGET_FAILURE);
577 return SUCCESS;
1da177e4
LT
578
579 case MEDIUM_ERROR:
fd1b494d
LT
580 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
581 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
582 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
7e782af5 583 set_host_byte(scmd, DID_MEDIUM_ERROR);
87f14e65 584 return SUCCESS;
fd1b494d 585 }
1da177e4
LT
586 return NEEDS_RETRY;
587
588 case HARDWARE_ERROR:
589 if (scmd->device->retry_hwerror)
bb0003c1 590 return ADD_TO_MLQUEUE;
1da177e4 591 else
87f14e65 592 set_host_byte(scmd, DID_TARGET_FAILURE);
1da177e4
LT
593
594 case ILLEGAL_REQUEST:
47ac56db
MS
595 if (sshdr.asc == 0x20 || /* Invalid command operation code */
596 sshdr.asc == 0x21 || /* Logical block address out of range */
597 sshdr.asc == 0x24 || /* Invalid field in cdb */
598 sshdr.asc == 0x26) { /* Parameter value invalid */
87f14e65 599 set_host_byte(scmd, DID_TARGET_FAILURE);
47ac56db
MS
600 }
601 return SUCCESS;
602
1da177e4
LT
603 default:
604 return SUCCESS;
605 }
606}
607
4a84067d
VD
608static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
609{
610 struct scsi_host_template *sht = sdev->host->hostt;
611 struct scsi_device *tmp_sdev;
612
c40ecc12 613 if (!sht->track_queue_depth ||
4a84067d
VD
614 sdev->queue_depth >= sdev->max_queue_depth)
615 return;
616
617 if (time_before(jiffies,
618 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
619 return;
620
621 if (time_before(jiffies,
622 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
623 return;
624
625 /*
626 * Walk all devices of a target and do
627 * ramp up on them.
628 */
629 shost_for_each_device(tmp_sdev, sdev->host) {
630 if (tmp_sdev->channel != sdev->channel ||
631 tmp_sdev->id != sdev->id ||
632 tmp_sdev->queue_depth == sdev->max_queue_depth)
633 continue;
c40ecc12 634
db5ed4df 635 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
4a84067d
VD
636 sdev->last_queue_ramp_up = jiffies;
637 }
638}
639
42a6a918
MC
640static void scsi_handle_queue_full(struct scsi_device *sdev)
641{
642 struct scsi_host_template *sht = sdev->host->hostt;
643 struct scsi_device *tmp_sdev;
644
c40ecc12 645 if (!sht->track_queue_depth)
42a6a918
MC
646 return;
647
648 shost_for_each_device(tmp_sdev, sdev->host) {
649 if (tmp_sdev->channel != sdev->channel ||
650 tmp_sdev->id != sdev->id)
651 continue;
652 /*
653 * We do not know the number of commands that were at
654 * the device when we got the queue full so we start
655 * from the highest possible value and work our way down.
656 */
c40ecc12 657 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
42a6a918
MC
658 }
659}
660
1da177e4
LT
661/**
662 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
663 * @scmd: SCSI cmd to examine.
664 *
665 * Notes:
666 * This is *only* called when we are examining the status of commands
667 * queued during error recovery. the main difference here is that we
668 * don't allow for the possibility of retries here, and we are a lot
669 * more restrictive about what we consider acceptable.
dc8875e1 670 */
1da177e4
LT
671static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
672{
673 /*
674 * first check the host byte, to see if there is anything in there
675 * that would indicate what we need to do.
676 */
677 if (host_byte(scmd->result) == DID_RESET) {
678 /*
679 * rats. we are already in the error handler, so we now
680 * get to try and figure out what to do next. if the sense
681 * is valid, we have a pretty good idea of what to do.
682 * if not, we mark it as FAILED.
683 */
684 return scsi_check_sense(scmd);
685 }
686 if (host_byte(scmd->result) != DID_OK)
687 return FAILED;
688
689 /*
690 * next, check the message byte.
691 */
692 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
693 return FAILED;
694
695 /*
696 * now, check the status byte to see if this indicates
697 * anything special.
698 */
699 switch (status_byte(scmd->result)) {
700 case GOOD:
4a84067d 701 scsi_handle_queue_ramp_up(scmd->device);
1da177e4
LT
702 case COMMAND_TERMINATED:
703 return SUCCESS;
704 case CHECK_CONDITION:
705 return scsi_check_sense(scmd);
706 case CONDITION_GOOD:
707 case INTERMEDIATE_GOOD:
708 case INTERMEDIATE_C_GOOD:
709 /*
710 * who knows? FIXME(eric)
711 */
712 return SUCCESS;
5f91bb05 713 case RESERVATION_CONFLICT:
67110dfd
JB
714 if (scmd->cmnd[0] == TEST_UNIT_READY)
715 /* it is a success, we probed the device and
716 * found it */
717 return SUCCESS;
718 /* otherwise, we failed to send the command */
719 return FAILED;
1da177e4 720 case QUEUE_FULL:
42a6a918
MC
721 scsi_handle_queue_full(scmd->device);
722 /* fall through */
723 case BUSY:
3eb3a928 724 return NEEDS_RETRY;
1da177e4
LT
725 default:
726 return FAILED;
727 }
728 return FAILED;
729}
730
1da177e4
LT
731/**
732 * scsi_eh_done - Completion function for error handling.
733 * @scmd: Cmd that is done.
dc8875e1 734 */
1da177e4
LT
735static void scsi_eh_done(struct scsi_cmnd *scmd)
736{
0bf8c869 737 struct completion *eh_action;
85631672 738
91921e01
HR
739 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
740 "%s scmd: %p result: %x\n",
cadbd4a5 741 __func__, scmd, scmd->result));
85631672
MR
742
743 eh_action = scmd->device->host->eh_action;
744 if (eh_action)
745 complete(eh_action);
1da177e4
LT
746}
747
292148f8
BK
748/**
749 * scsi_try_host_reset - ask host adapter to reset itself
c2b3ebd0 750 * @scmd: SCSI cmd to send host reset.
dc8875e1 751 */
292148f8
BK
752static int scsi_try_host_reset(struct scsi_cmnd *scmd)
753{
754 unsigned long flags;
755 int rtn;
0bf8c869
JJ
756 struct Scsi_Host *host = scmd->device->host;
757 struct scsi_host_template *hostt = host->hostt;
292148f8 758
91921e01
HR
759 SCSI_LOG_ERROR_RECOVERY(3,
760 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
292148f8 761
0bf8c869 762 if (!hostt->eh_host_reset_handler)
292148f8
BK
763 return FAILED;
764
0bf8c869 765 rtn = hostt->eh_host_reset_handler(scmd);
292148f8
BK
766
767 if (rtn == SUCCESS) {
0bf8c869 768 if (!hostt->skip_settle_delay)
292148f8 769 ssleep(HOST_RESET_SETTLE_TIME);
0bf8c869
JJ
770 spin_lock_irqsave(host->host_lock, flags);
771 scsi_report_bus_reset(host, scmd_channel(scmd));
772 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
773 }
774
775 return rtn;
776}
777
778/**
779 * scsi_try_bus_reset - ask host to perform a bus reset
780 * @scmd: SCSI cmd to send bus reset.
dc8875e1 781 */
292148f8
BK
782static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
783{
784 unsigned long flags;
785 int rtn;
0bf8c869
JJ
786 struct Scsi_Host *host = scmd->device->host;
787 struct scsi_host_template *hostt = host->hostt;
292148f8 788
91921e01
HR
789 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
790 "%s: Snd Bus RST\n", __func__));
292148f8 791
0bf8c869 792 if (!hostt->eh_bus_reset_handler)
292148f8
BK
793 return FAILED;
794
0bf8c869 795 rtn = hostt->eh_bus_reset_handler(scmd);
292148f8
BK
796
797 if (rtn == SUCCESS) {
0bf8c869 798 if (!hostt->skip_settle_delay)
292148f8 799 ssleep(BUS_RESET_SETTLE_TIME);
0bf8c869
JJ
800 spin_lock_irqsave(host->host_lock, flags);
801 scsi_report_bus_reset(host, scmd_channel(scmd));
802 spin_unlock_irqrestore(host->host_lock, flags);
292148f8
BK
803 }
804
805 return rtn;
806}
807
30bd7df8
MC
808static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
809{
810 sdev->was_reset = 1;
811 sdev->expecting_cc_ua = 1;
812}
813
814/**
815 * scsi_try_target_reset - Ask host to perform a target reset
816 * @scmd: SCSI cmd used to send a target reset
817 *
818 * Notes:
819 * There is no timeout for this operation. if this operation is
820 * unreliable for a given host, then the host itself needs to put a
821 * timer on it, and set the host back to a consistent state prior to
822 * returning.
823 */
824static int scsi_try_target_reset(struct scsi_cmnd *scmd)
825{
826 unsigned long flags;
827 int rtn;
0bf8c869
JJ
828 struct Scsi_Host *host = scmd->device->host;
829 struct scsi_host_template *hostt = host->hostt;
30bd7df8 830
0bf8c869 831 if (!hostt->eh_target_reset_handler)
30bd7df8
MC
832 return FAILED;
833
0bf8c869 834 rtn = hostt->eh_target_reset_handler(scmd);
30bd7df8 835 if (rtn == SUCCESS) {
0bf8c869 836 spin_lock_irqsave(host->host_lock, flags);
30bd7df8
MC
837 __starget_for_each_device(scsi_target(scmd->device), NULL,
838 __scsi_report_device_reset);
0bf8c869 839 spin_unlock_irqrestore(host->host_lock, flags);
30bd7df8
MC
840 }
841
842 return rtn;
843}
844
292148f8
BK
845/**
846 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
847 * @scmd: SCSI cmd used to send BDR
848 *
849 * Notes:
850 * There is no timeout for this operation. if this operation is
851 * unreliable for a given host, then the host itself needs to put a
852 * timer on it, and set the host back to a consistent state prior to
853 * returning.
dc8875e1 854 */
292148f8
BK
855static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
856{
857 int rtn;
0bf8c869 858 struct scsi_host_template *hostt = scmd->device->host->hostt;
292148f8 859
0bf8c869 860 if (!hostt->eh_device_reset_handler)
292148f8
BK
861 return FAILED;
862
0bf8c869 863 rtn = hostt->eh_device_reset_handler(scmd);
30bd7df8
MC
864 if (rtn == SUCCESS)
865 __scsi_report_device_reset(scmd->device, NULL);
292148f8
BK
866 return rtn;
867}
868
883a030f
HR
869/**
870 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
6583f6fb 871 * @hostt: SCSI driver host template
883a030f
HR
872 * @scmd: SCSI cmd used to send a target reset
873 *
874 * Return value:
875 * SUCCESS, FAILED, or FAST_IO_FAIL
876 *
877 * Notes:
878 * SUCCESS does not necessarily indicate that the command
879 * has been aborted; it only indicates that the LLDDs
880 * has cleared all references to that command.
881 * LLDDs should return FAILED only if an abort was required
882 * but could not be executed. LLDDs should return FAST_IO_FAIL
883 * if the device is temporarily unavailable (eg due to a
884 * link down on FibreChannel)
885 */
886static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
887 struct scsi_cmnd *scmd)
292148f8 888{
0bf8c869 889 if (!hostt->eh_abort_handler)
292148f8
BK
890 return FAILED;
891
0bf8c869 892 return hostt->eh_abort_handler(scmd);
292148f8
BK
893}
894
292148f8
BK
895static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
896{
0bf8c869 897 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
292148f8 898 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
30bd7df8
MC
899 if (scsi_try_target_reset(scmd) != SUCCESS)
900 if (scsi_try_bus_reset(scmd) != SUCCESS)
901 scsi_try_host_reset(scmd);
292148f8
BK
902}
903
1da177e4 904/**
3b729f76 905 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
2dc611de 906 * @scmd: SCSI command structure to hijack
e1c23468 907 * @ses: structure to save restore information
55db6c1b 908 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
64a87b24 909 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
55db6c1b 910 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
2dc611de 911 *
e1c23468 912 * This function is used to save a scsi command information before re-execution
55db6c1b
BH
913 * as part of the error recovery process. If @sense_bytes is 0 the command
914 * sent must be one that does not transfer any data. If @sense_bytes != 0
915 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
916 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
dc8875e1 917 */
e1c23468
BH
918void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
919 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
1da177e4 920{
f59114b7 921 struct scsi_device *sdev = scmd->device;
1da177e4 922
631c228c
CH
923 /*
924 * We need saved copies of a number of fields - this is because
925 * error handling may need to overwrite these with different values
926 * to run different commands, and once error handling is complete,
927 * we will need to restore these values prior to running the actual
928 * command.
929 */
e1c23468 930 ses->cmd_len = scmd->cmd_len;
64a87b24 931 ses->cmnd = scmd->cmnd;
e1c23468 932 ses->data_direction = scmd->sc_data_direction;
30b0c37b 933 ses->sdb = scmd->sdb;
6f9a35e2 934 ses->next_rq = scmd->request->next_rq;
e1c23468 935 ses->result = scmd->result;
12265709 936 ses->underflow = scmd->underflow;
db007fc5 937 ses->prot_op = scmd->prot_op;
631c228c 938
db007fc5 939 scmd->prot_op = SCSI_PROT_NORMAL;
c69e6f81 940 scmd->eh_eflags = 0;
64a87b24
BH
941 scmd->cmnd = ses->eh_cmnd;
942 memset(scmd->cmnd, 0, BLK_MAX_CDB);
30b0c37b 943 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
6f9a35e2 944 scmd->request->next_rq = NULL;
644373a4 945 scmd->result = 0;
30b0c37b 946
55db6c1b 947 if (sense_bytes) {
30b0c37b
BH
948 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
949 sense_bytes);
e1c23468 950 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
30b0c37b
BH
951 scmd->sdb.length);
952 scmd->sdb.table.sgl = &ses->sense_sgl;
55db6c1b 953 scmd->sc_data_direction = DMA_FROM_DEVICE;
30b0c37b 954 scmd->sdb.table.nents = 1;
55db6c1b 955 scmd->cmnd[0] = REQUEST_SENSE;
30b0c37b 956 scmd->cmnd[4] = scmd->sdb.length;
55db6c1b 957 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
631c228c 958 } else {
631c228c 959 scmd->sc_data_direction = DMA_NONE;
55db6c1b 960 if (cmnd) {
64a87b24 961 BUG_ON(cmnd_size > BLK_MAX_CDB);
55db6c1b
BH
962 memcpy(scmd->cmnd, cmnd, cmnd_size);
963 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
964 }
631c228c
CH
965 }
966
967 scmd->underflow = 0;
631c228c 968
55db6c1b 969 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
1da177e4 970 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b7 971 (sdev->lun << 5 & 0xe0);
1da177e4 972
631c228c
CH
973 /*
974 * Zero the sense buffer. The scsi spec mandates that any
975 * untransferred sense data should be interpreted as being zero.
976 */
b80ca4f7 977 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
e1c23468
BH
978}
979EXPORT_SYMBOL(scsi_eh_prep_cmnd);
980
981/**
3b729f76 982 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
e1c23468 983 * @scmd: SCSI command structure to restore
477e608c 984 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
e1c23468 985 *
477e608c 986 * Undo any damage done by above scsi_eh_prep_cmnd().
dc8875e1 987 */
e1c23468
BH
988void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
989{
990 /*
991 * Restore original data
992 */
993 scmd->cmd_len = ses->cmd_len;
64a87b24 994 scmd->cmnd = ses->cmnd;
e1c23468 995 scmd->sc_data_direction = ses->data_direction;
30b0c37b 996 scmd->sdb = ses->sdb;
6f9a35e2 997 scmd->request->next_rq = ses->next_rq;
e1c23468 998 scmd->result = ses->result;
12265709 999 scmd->underflow = ses->underflow;
db007fc5 1000 scmd->prot_op = ses->prot_op;
e1c23468
BH
1001}
1002EXPORT_SYMBOL(scsi_eh_restore_cmnd);
631c228c 1003
e1c23468 1004/**
3b729f76 1005 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
e1c23468
BH
1006 * @scmd: SCSI command structure to hijack
1007 * @cmnd: CDB to send
1008 * @cmnd_size: size in bytes of @cmnd
1009 * @timeout: timeout for this request
1010 * @sense_bytes: size of sense data to copy or 0
1011 *
1012 * This function is used to send a scsi command down to a target device
1013 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1014 *
1015 * Return value:
1016 * SUCCESS or FAILED or NEEDS_RETRY
dc8875e1 1017 */
e1c23468
BH
1018static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1019 int cmnd_size, int timeout, unsigned sense_bytes)
1020{
1021 struct scsi_device *sdev = scmd->device;
1022 struct Scsi_Host *shost = sdev->host;
1023 DECLARE_COMPLETION_ONSTACK(done);
fc73648a 1024 unsigned long timeleft = timeout;
e1c23468 1025 struct scsi_eh_save ses;
fc73648a 1026 const unsigned long stall_for = msecs_to_jiffies(100);
e1c23468
BH
1027 int rtn;
1028
fc73648a 1029retry:
e1c23468 1030 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
7dfdc9a5 1031 shost->eh_action = &done;
1da177e4 1032
1da177e4 1033 scsi_log_send(scmd);
f281233d 1034 scmd->scsi_done = scsi_eh_done;
fc73648a
HR
1035 rtn = shost->hostt->queuecommand(shost, scmd);
1036 if (rtn) {
1037 if (timeleft > stall_for) {
1038 scsi_eh_restore_cmnd(scmd, &ses);
1039 timeleft -= stall_for;
1040 msleep(jiffies_to_msecs(stall_for));
1041 goto retry;
1042 }
1043 /* signal not to enter either branch of the if () below */
1044 timeleft = 0;
511833ac 1045 rtn = FAILED;
fc73648a
HR
1046 } else {
1047 timeleft = wait_for_completion_timeout(&done, timeout);
ac61d195 1048 rtn = SUCCESS;
fc73648a 1049 }
1da177e4 1050
f59114b7 1051 shost->eh_action = NULL;
1da177e4 1052
fc73648a 1053 scsi_log_completion(scmd, rtn);
1da177e4 1054
91921e01
HR
1055 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1056 "%s: scmd: %p, timeleft: %ld\n",
cadbd4a5 1057 __func__, scmd, timeleft));
1da177e4
LT
1058
1059 /*
fc73648a
HR
1060 * If there is time left scsi_eh_done got called, and we will examine
1061 * the actual status codes to see whether the command actually did
1062 * complete normally, else if we have a zero return and no time left,
1063 * the command must still be pending, so abort it and return FAILED.
1064 * If we never actually managed to issue the command, because
1065 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1066 * value above (so don't execute either branch of the if)
1da177e4 1067 */
7dfdc9a5 1068 if (timeleft) {
1da177e4 1069 rtn = scsi_eh_completed_normally(scmd);
91921e01
HR
1070 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1071 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
7dfdc9a5 1072
1da177e4
LT
1073 switch (rtn) {
1074 case SUCCESS:
1075 case NEEDS_RETRY:
1076 case FAILED:
1077 break;
6e883b0e
HR
1078 case ADD_TO_MLQUEUE:
1079 rtn = NEEDS_RETRY;
1080 break;
1da177e4
LT
1081 default:
1082 rtn = FAILED;
1083 break;
1084 }
511833ac 1085 } else if (rtn != FAILED) {
292148f8 1086 scsi_abort_eh_cmnd(scmd);
7dfdc9a5 1087 rtn = FAILED;
1da177e4
LT
1088 }
1089
e1c23468 1090 scsi_eh_restore_cmnd(scmd, &ses);
18a4d0a2 1091
1da177e4
LT
1092 return rtn;
1093}
1094
1095/**
1096 * scsi_request_sense - Request sense data from a particular target.
1097 * @scmd: SCSI cmd for request sense.
1098 *
1099 * Notes:
1100 * Some hosts automatically obtain this information, others require
1101 * that we obtain it on our own. This function will *not* return until
1102 * the command either times out, or it completes.
dc8875e1 1103 */
1da177e4
LT
1104static int scsi_request_sense(struct scsi_cmnd *scmd)
1105{
0816c925 1106 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1da177e4
LT
1107}
1108
2451079b
JB
1109static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1110{
1111 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1112 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1113 if (sdrv->eh_action)
1114 rtn = sdrv->eh_action(scmd, rtn);
1115 }
1116 return rtn;
1117}
1118
1da177e4
LT
1119/**
1120 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1121 * @scmd: Original SCSI cmd that eh has finished.
1122 * @done_q: Queue for processed commands.
1123 *
1124 * Notes:
1125 * We don't want to use the normal command completion while we are are
1126 * still handling errors - it may cause other commands to be queued,
eb44820c 1127 * and that would disturb what we are doing. Thus we really want to
1da177e4
LT
1128 * keep a list of pending commands for final completion, and once we
1129 * are ready to leave error handling we handle completion for real.
dc8875e1 1130 */
041c5fc3 1131void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1da177e4
LT
1132{
1133 scmd->device->host->host_failed--;
3111b0d1 1134 scmd->eh_eflags = 0;
1da177e4
LT
1135 list_move_tail(&scmd->eh_entry, done_q);
1136}
041c5fc3 1137EXPORT_SYMBOL(scsi_eh_finish_cmd);
1da177e4
LT
1138
1139/**
1140 * scsi_eh_get_sense - Get device sense data.
1141 * @work_q: Queue of commands to process.
eb44820c 1142 * @done_q: Queue of processed commands.
1da177e4
LT
1143 *
1144 * Description:
1145 * See if we need to request sense information. if so, then get it
0bf8c869 1146 * now, so we have a better idea of what to do.
1da177e4
LT
1147 *
1148 * Notes:
1149 * This has the unfortunate side effect that if a shost adapter does
eb44820c 1150 * not automatically request sense information, we end up shutting
1da177e4
LT
1151 * it down before we request it.
1152 *
1153 * All drivers should request sense information internally these days,
1154 * so for now all I have to say is tough noogies if you end up in here.
1155 *
1156 * XXX: Long term this code should go away, but that needs an audit of
1157 * all LLDDs first.
dc8875e1 1158 */
dca84e46
DW
1159int scsi_eh_get_sense(struct list_head *work_q,
1160 struct list_head *done_q)
1da177e4 1161{
937abeaa 1162 struct scsi_cmnd *scmd, *next;
b4562022 1163 struct Scsi_Host *shost;
1da177e4
LT
1164 int rtn;
1165
937abeaa 1166 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3111b0d1 1167 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1da177e4
LT
1168 SCSI_SENSE_VALID(scmd))
1169 continue;
1170
b4562022 1171 shost = scmd->device->host;
b4562022 1172 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1173 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1174 scmd_printk(KERN_INFO, scmd,
1175 "%s: skip request sense, past eh deadline\n",
1176 current->comm));
b4562022
HR
1177 break;
1178 }
d555a2ab
JB
1179 if (status_byte(scmd->result) != CHECK_CONDITION)
1180 /*
1181 * don't request sense if there's no check condition
1182 * status because the error we're processing isn't one
1183 * that has a sense code (and some devices get
1184 * confused by sense requests out of the blue)
1185 */
1186 continue;
1187
3bf743e7
JG
1188 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1189 "%s: requesting sense\n",
1190 current->comm));
1da177e4
LT
1191 rtn = scsi_request_sense(scmd);
1192 if (rtn != SUCCESS)
1193 continue;
1194
91921e01
HR
1195 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1196 "sense requested for %p result %x\n",
1197 scmd, scmd->result));
d811b848 1198 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1da177e4
LT
1199
1200 rtn = scsi_decide_disposition(scmd);
1201
1202 /*
1203 * if the result was normal, then just pass it along to the
1204 * upper level.
1205 */
1206 if (rtn == SUCCESS)
1207 /* we don't want this command reissued, just
1208 * finished with the sense data, so set
1209 * retries to the max allowed to ensure it
1210 * won't get reissued */
1211 scmd->retries = scmd->allowed;
1212 else if (rtn != NEEDS_RETRY)
1213 continue;
1214
1215 scsi_eh_finish_cmd(scmd, done_q);
1216 }
1217
1218 return list_empty(work_q);
1219}
dca84e46 1220EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1da177e4 1221
1da177e4
LT
1222/**
1223 * scsi_eh_tur - Send TUR to device.
eb44820c 1224 * @scmd: &scsi_cmnd to send TUR
1da177e4
LT
1225 *
1226 * Return value:
1227 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1228 */
1da177e4
LT
1229static int scsi_eh_tur(struct scsi_cmnd *scmd)
1230{
1231 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1232 int retry_cnt = 1, rtn;
1233
1234retry_tur:
0816c925
MP
1235 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1236 scmd->device->eh_timeout, 0);
1da177e4 1237
91921e01
HR
1238 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1239 "%s: scmd %p rtn %x\n", __func__, scmd, rtn));
631c228c
CH
1240
1241 switch (rtn) {
1242 case NEEDS_RETRY:
1da177e4
LT
1243 if (retry_cnt--)
1244 goto retry_tur;
631c228c
CH
1245 /*FALLTHRU*/
1246 case SUCCESS:
e47373ec 1247 return 0;
631c228c
CH
1248 default:
1249 return 1;
e47373ec 1250 }
1da177e4
LT
1251}
1252
3eef6257
DJ
1253/**
1254 * scsi_eh_test_devices - check if devices are responding from error recovery.
1255 * @cmd_list: scsi commands in error recovery.
74cf298f
RD
1256 * @work_q: queue for commands which still need more error recovery
1257 * @done_q: queue for commands which are finished
1258 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
3eef6257
DJ
1259 *
1260 * Decription:
1261 * Tests if devices are in a working state. Commands to devices now in
1262 * a working state are sent to the done_q while commands to devices which
1263 * are still failing to respond are returned to the work_q for more
1264 * processing.
1265 **/
1266static int scsi_eh_test_devices(struct list_head *cmd_list,
1267 struct list_head *work_q,
1268 struct list_head *done_q, int try_stu)
1269{
1270 struct scsi_cmnd *scmd, *next;
1271 struct scsi_device *sdev;
1272 int finish_cmds;
1273
1274 while (!list_empty(cmd_list)) {
1275 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1276 sdev = scmd->device;
1277
b4562022 1278 if (!try_stu) {
b4562022
HR
1279 if (scsi_host_eh_past_deadline(sdev->host)) {
1280 /* Push items back onto work_q */
1281 list_splice_init(cmd_list, work_q);
b4562022 1282 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1283 sdev_printk(KERN_INFO, sdev,
1284 "%s: skip test device, past eh deadline",
1285 current->comm));
b4562022
HR
1286 break;
1287 }
b4562022
HR
1288 }
1289
3eef6257
DJ
1290 finish_cmds = !scsi_device_online(scmd->device) ||
1291 (try_stu && !scsi_eh_try_stu(scmd) &&
1292 !scsi_eh_tur(scmd)) ||
1293 !scsi_eh_tur(scmd);
1294
1295 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1296 if (scmd->device == sdev) {
2451079b
JB
1297 if (finish_cmds &&
1298 (try_stu ||
1299 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
3eef6257
DJ
1300 scsi_eh_finish_cmd(scmd, done_q);
1301 else
1302 list_move_tail(&scmd->eh_entry, work_q);
1303 }
1304 }
1305 return list_empty(work_q);
1306}
1307
1308
1da177e4 1309/**
eb44820c
RL
1310 * scsi_eh_abort_cmds - abort pending commands.
1311 * @work_q: &list_head for pending commands.
1312 * @done_q: &list_head for processed commands.
1da177e4
LT
1313 *
1314 * Decription:
1315 * Try and see whether or not it makes sense to try and abort the
eb44820c
RL
1316 * running command. This only works out to be the case if we have one
1317 * command that has timed out. If the command simply failed, it makes
1da177e4
LT
1318 * no sense to try and abort the command, since as far as the shost
1319 * adapter is concerned, it isn't running.
dc8875e1 1320 */
1da177e4
LT
1321static int scsi_eh_abort_cmds(struct list_head *work_q,
1322 struct list_head *done_q)
1323{
937abeaa 1324 struct scsi_cmnd *scmd, *next;
3eef6257 1325 LIST_HEAD(check_list);
1da177e4 1326 int rtn;
b4562022 1327 struct Scsi_Host *shost;
1da177e4 1328
937abeaa 1329 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3111b0d1 1330 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1da177e4 1331 continue;
b4562022 1332 shost = scmd->device->host;
b4562022 1333 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1334 list_splice_init(&check_list, work_q);
1335 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1336 scmd_printk(KERN_INFO, scmd,
1337 "%s: skip aborting cmd, past eh deadline\n",
1338 current->comm));
b4562022
HR
1339 return list_empty(work_q);
1340 }
91921e01 1341 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1342 scmd_printk(KERN_INFO, scmd,
1343 "%s: aborting cmd\n", current->comm));
b4562022 1344 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
6fd046f9 1345 if (rtn == FAILED) {
91921e01 1346 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1347 scmd_printk(KERN_INFO, scmd,
1348 "%s: aborting cmd failed\n",
1349 current->comm));
6fd046f9
HR
1350 list_splice_init(&check_list, work_q);
1351 return list_empty(work_q);
1352 }
1353 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1354 if (rtn == FAST_IO_FAIL)
1355 scsi_eh_finish_cmd(scmd, done_q);
1356 else
1357 list_move_tail(&scmd->eh_entry, &check_list);
1da177e4
LT
1358 }
1359
3eef6257 1360 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1361}
1362
1da177e4
LT
1363/**
1364 * scsi_eh_try_stu - Send START_UNIT to device.
eb44820c 1365 * @scmd: &scsi_cmnd to send START_UNIT
1da177e4
LT
1366 *
1367 * Return value:
1368 * 0 - Device is ready. 1 - Device NOT ready.
dc8875e1 1369 */
1da177e4
LT
1370static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1371{
1372 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1da177e4 1373
631c228c 1374 if (scmd->device->allow_restart) {
ed773e66
BK
1375 int i, rtn = NEEDS_RETRY;
1376
1377 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
9728c081 1378 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1da177e4 1379
631c228c
CH
1380 if (rtn == SUCCESS)
1381 return 0;
1382 }
1da177e4 1383
1da177e4
LT
1384 return 1;
1385}
1386
1387 /**
1388 * scsi_eh_stu - send START_UNIT if needed
eb44820c 1389 * @shost: &scsi host being recovered.
74cf298f 1390 * @work_q: &list_head for pending commands.
eb44820c 1391 * @done_q: &list_head for processed commands.
1da177e4
LT
1392 *
1393 * Notes:
1394 * If commands are failing due to not ready, initializing command required,
0bf8c869 1395 * try revalidating the device, which will end up sending a start unit.
dc8875e1 1396 */
1da177e4
LT
1397static int scsi_eh_stu(struct Scsi_Host *shost,
1398 struct list_head *work_q,
1399 struct list_head *done_q)
1400{
937abeaa 1401 struct scsi_cmnd *scmd, *stu_scmd, *next;
1da177e4
LT
1402 struct scsi_device *sdev;
1403
1404 shost_for_each_device(sdev, shost) {
b4562022 1405 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1406 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1407 sdev_printk(KERN_INFO, sdev,
1408 "%s: skip START_UNIT, past eh deadline\n",
1409 current->comm));
b4562022
HR
1410 break;
1411 }
1da177e4
LT
1412 stu_scmd = NULL;
1413 list_for_each_entry(scmd, work_q, eh_entry)
1414 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1415 scsi_check_sense(scmd) == FAILED ) {
1416 stu_scmd = scmd;
1417 break;
1418 }
1419
1420 if (!stu_scmd)
1421 continue;
1422
91921e01 1423 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1424 sdev_printk(KERN_INFO, sdev,
1425 "%s: Sending START_UNIT\n",
1426 current->comm));
1da177e4
LT
1427
1428 if (!scsi_eh_try_stu(stu_scmd)) {
1429 if (!scsi_device_online(sdev) ||
1430 !scsi_eh_tur(stu_scmd)) {
937abeaa
CH
1431 list_for_each_entry_safe(scmd, next,
1432 work_q, eh_entry) {
2451079b
JB
1433 if (scmd->device == sdev &&
1434 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1da177e4
LT
1435 scsi_eh_finish_cmd(scmd, done_q);
1436 }
1437 }
1438 } else {
1439 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1440 sdev_printk(KERN_INFO, sdev,
1441 "%s: START_UNIT failed\n",
1442 current->comm));
1da177e4
LT
1443 }
1444 }
1445
1446 return list_empty(work_q);
1447}
1448
1449
1450/**
1451 * scsi_eh_bus_device_reset - send bdr if needed
1452 * @shost: scsi host being recovered.
74cf298f 1453 * @work_q: &list_head for pending commands.
eb44820c 1454 * @done_q: &list_head for processed commands.
1da177e4
LT
1455 *
1456 * Notes:
eb44820c 1457 * Try a bus device reset. Still, look to see whether we have multiple
1da177e4
LT
1458 * devices that are jammed or not - if we have multiple devices, it
1459 * makes no sense to try bus_device_reset - we really would need to try
0bf8c869 1460 * a bus_reset instead.
dc8875e1 1461 */
1da177e4
LT
1462static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1463 struct list_head *work_q,
1464 struct list_head *done_q)
1465{
937abeaa 1466 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1da177e4
LT
1467 struct scsi_device *sdev;
1468 int rtn;
1469
1470 shost_for_each_device(sdev, shost) {
b4562022 1471 if (scsi_host_eh_past_deadline(shost)) {
b4562022 1472 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1473 sdev_printk(KERN_INFO, sdev,
1474 "%s: skip BDR, past eh deadline\n",
1475 current->comm));
b4562022
HR
1476 break;
1477 }
1da177e4
LT
1478 bdr_scmd = NULL;
1479 list_for_each_entry(scmd, work_q, eh_entry)
1480 if (scmd->device == sdev) {
1481 bdr_scmd = scmd;
1482 break;
1483 }
1484
1485 if (!bdr_scmd)
1486 continue;
1487
91921e01 1488 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1489 sdev_printk(KERN_INFO, sdev,
1490 "%s: Sending BDR\n", current->comm));
1da177e4 1491 rtn = scsi_try_bus_device_reset(bdr_scmd);
2f2eb587 1492 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1da177e4 1493 if (!scsi_device_online(sdev) ||
2f2eb587 1494 rtn == FAST_IO_FAIL ||
1da177e4 1495 !scsi_eh_tur(bdr_scmd)) {
937abeaa
CH
1496 list_for_each_entry_safe(scmd, next,
1497 work_q, eh_entry) {
2451079b
JB
1498 if (scmd->device == sdev &&
1499 scsi_eh_action(scmd, rtn) != FAILED)
1da177e4
LT
1500 scsi_eh_finish_cmd(scmd,
1501 done_q);
1502 }
1503 }
1504 } else {
91921e01 1505 SCSI_LOG_ERROR_RECOVERY(3,
a222b1e2
HR
1506 sdev_printk(KERN_INFO, sdev,
1507 "%s: BDR failed\n", current->comm));
1da177e4
LT
1508 }
1509 }
1510
1511 return list_empty(work_q);
1512}
1513
30bd7df8
MC
1514/**
1515 * scsi_eh_target_reset - send target reset if needed
1516 * @shost: scsi host being recovered.
74cf298f 1517 * @work_q: &list_head for pending commands.
30bd7df8
MC
1518 * @done_q: &list_head for processed commands.
1519 *
1520 * Notes:
1521 * Try a target reset.
1522 */
1523static int scsi_eh_target_reset(struct Scsi_Host *shost,
1524 struct list_head *work_q,
1525 struct list_head *done_q)
1526{
98db5195 1527 LIST_HEAD(tmp_list);
3eef6257 1528 LIST_HEAD(check_list);
30bd7df8 1529
98db5195
JB
1530 list_splice_init(work_q, &tmp_list);
1531
1532 while (!list_empty(&tmp_list)) {
1533 struct scsi_cmnd *next, *scmd;
1534 int rtn;
1535 unsigned int id;
b4562022 1536
b4562022 1537 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1538 /* push back on work queue for further processing */
1539 list_splice_init(&check_list, work_q);
1540 list_splice_init(&tmp_list, work_q);
1541 SCSI_LOG_ERROR_RECOVERY(3,
1542 shost_printk(KERN_INFO, shost,
a222b1e2
HR
1543 "%s: Skip target reset, past eh deadline\n",
1544 current->comm));
b4562022
HR
1545 return list_empty(work_q);
1546 }
98db5195
JB
1547
1548 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1549 id = scmd_id(scmd);
30bd7df8 1550
91921e01
HR
1551 SCSI_LOG_ERROR_RECOVERY(3,
1552 shost_printk(KERN_INFO, shost,
1553 "%s: Sending target reset to target %d\n",
1554 current->comm, id));
98db5195
JB
1555 rtn = scsi_try_target_reset(scmd);
1556 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
91921e01
HR
1557 SCSI_LOG_ERROR_RECOVERY(3,
1558 shost_printk(KERN_INFO, shost,
1559 "%s: Target reset failed"
1560 " target: %d\n",
1561 current->comm, id));
98db5195
JB
1562 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1563 if (scmd_id(scmd) != id)
1564 continue;
1565
3eef6257
DJ
1566 if (rtn == SUCCESS)
1567 list_move_tail(&scmd->eh_entry, &check_list);
1568 else if (rtn == FAST_IO_FAIL)
98db5195
JB
1569 scsi_eh_finish_cmd(scmd, done_q);
1570 else
1571 /* push back on work queue for further processing */
1572 list_move(&scmd->eh_entry, work_q);
1573 }
1574 }
30bd7df8 1575
3eef6257 1576 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
30bd7df8
MC
1577}
1578
1da177e4 1579/**
0bf8c869 1580 * scsi_eh_bus_reset - send a bus reset
eb44820c 1581 * @shost: &scsi host being recovered.
74cf298f 1582 * @work_q: &list_head for pending commands.
eb44820c 1583 * @done_q: &list_head for processed commands.
dc8875e1 1584 */
1da177e4
LT
1585static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1586 struct list_head *work_q,
1587 struct list_head *done_q)
1588{
937abeaa 1589 struct scsi_cmnd *scmd, *chan_scmd, *next;
3eef6257 1590 LIST_HEAD(check_list);
1da177e4
LT
1591 unsigned int channel;
1592 int rtn;
1593
1594 /*
1595 * we really want to loop over the various channels, and do this on
1596 * a channel by channel basis. we should also check to see if any
1597 * of the failed commands are on soft_reset devices, and if so, skip
0bf8c869 1598 * the reset.
1da177e4
LT
1599 */
1600
1601 for (channel = 0; channel <= shost->max_channel; channel++) {
b4562022 1602 if (scsi_host_eh_past_deadline(shost)) {
b4562022
HR
1603 list_splice_init(&check_list, work_q);
1604 SCSI_LOG_ERROR_RECOVERY(3,
1605 shost_printk(KERN_INFO, shost,
a222b1e2
HR
1606 "%s: skip BRST, past eh deadline\n",
1607 current->comm));
b4562022
HR
1608 return list_empty(work_q);
1609 }
b4562022 1610
1da177e4
LT
1611 chan_scmd = NULL;
1612 list_for_each_entry(scmd, work_q, eh_entry) {
422c0d61 1613 if (channel == scmd_channel(scmd)) {
1da177e4
LT
1614 chan_scmd = scmd;
1615 break;
1616 /*
1617 * FIXME add back in some support for
1618 * soft_reset devices.
1619 */
1620 }
1621 }
1622
1623 if (!chan_scmd)
1624 continue;
91921e01
HR
1625 SCSI_LOG_ERROR_RECOVERY(3,
1626 shost_printk(KERN_INFO, shost,
1627 "%s: Sending BRST chan: %d\n",
1628 current->comm, channel));
1da177e4 1629 rtn = scsi_try_bus_reset(chan_scmd);
2f2eb587 1630 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
937abeaa 1631 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
3eef6257
DJ
1632 if (channel == scmd_channel(scmd)) {
1633 if (rtn == FAST_IO_FAIL)
1da177e4
LT
1634 scsi_eh_finish_cmd(scmd,
1635 done_q);
3eef6257
DJ
1636 else
1637 list_move_tail(&scmd->eh_entry,
1638 &check_list);
1639 }
1da177e4
LT
1640 }
1641 } else {
91921e01
HR
1642 SCSI_LOG_ERROR_RECOVERY(3,
1643 shost_printk(KERN_INFO, shost,
1644 "%s: BRST failed chan: %d\n",
1645 current->comm, channel));
1da177e4
LT
1646 }
1647 }
3eef6257 1648 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1da177e4
LT
1649}
1650
1651/**
0bf8c869 1652 * scsi_eh_host_reset - send a host reset
74cf298f
RD
1653 * @shost: host to be reset.
1654 * @work_q: &list_head for pending commands.
1655 * @done_q: &list_head for processed commands.
dc8875e1 1656 */
91921e01
HR
1657static int scsi_eh_host_reset(struct Scsi_Host *shost,
1658 struct list_head *work_q,
1da177e4
LT
1659 struct list_head *done_q)
1660{
937abeaa 1661 struct scsi_cmnd *scmd, *next;
3eef6257 1662 LIST_HEAD(check_list);
1da177e4 1663 int rtn;
1da177e4
LT
1664
1665 if (!list_empty(work_q)) {
1666 scmd = list_entry(work_q->next,
1667 struct scsi_cmnd, eh_entry);
1668
91921e01
HR
1669 SCSI_LOG_ERROR_RECOVERY(3,
1670 shost_printk(KERN_INFO, shost,
1671 "%s: Sending HRST\n",
1672 current->comm));
1da177e4
LT
1673
1674 rtn = scsi_try_host_reset(scmd);
3eef6257
DJ
1675 if (rtn == SUCCESS) {
1676 list_splice_init(work_q, &check_list);
1677 } else if (rtn == FAST_IO_FAIL) {
937abeaa 1678 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1da177e4
LT
1679 scsi_eh_finish_cmd(scmd, done_q);
1680 }
1681 } else {
91921e01
HR
1682 SCSI_LOG_ERROR_RECOVERY(3,
1683 shost_printk(KERN_INFO, shost,
1684 "%s: HRST failed\n",
1685 current->comm));
1da177e4
LT
1686 }
1687 }
3eef6257 1688 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1da177e4
LT
1689}
1690
1691/**
1692 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
74cf298f
RD
1693 * @work_q: &list_head for pending commands.
1694 * @done_q: &list_head for processed commands.
dc8875e1 1695 */
1da177e4
LT
1696static void scsi_eh_offline_sdevs(struct list_head *work_q,
1697 struct list_head *done_q)
1698{
937abeaa 1699 struct scsi_cmnd *scmd, *next;
1da177e4 1700
937abeaa 1701 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
31765d7d
MW
1702 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1703 "not ready after error recovery\n");
1da177e4 1704 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
3111b0d1 1705 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1da177e4
LT
1706 /*
1707 * FIXME: Handle lost cmds.
1708 */
1709 }
1710 scsi_eh_finish_cmd(scmd, done_q);
1711 }
1712 return;
1713}
1714
4a27446f 1715/**
e494f6a7 1716 * scsi_noretry_cmd - determine if command should be failed fast
4a27446f
MC
1717 * @scmd: SCSI cmd to examine.
1718 */
1719int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1720{
1721 switch (host_byte(scmd->result)) {
1722 case DID_OK:
1723 break;
e494f6a7
HR
1724 case DID_TIME_OUT:
1725 goto check_type;
4a27446f 1726 case DID_BUS_BUSY:
33659ebb 1727 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
4a27446f 1728 case DID_PARITY:
33659ebb 1729 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
4a27446f
MC
1730 case DID_ERROR:
1731 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1732 status_byte(scmd->result) == RESERVATION_CONFLICT)
1733 return 0;
1734 /* fall through */
1735 case DID_SOFT_ERROR:
33659ebb 1736 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
4a27446f
MC
1737 }
1738
e494f6a7
HR
1739 if (status_byte(scmd->result) != CHECK_CONDITION)
1740 return 0;
4a27446f 1741
e494f6a7
HR
1742check_type:
1743 /*
1744 * assume caller has checked sense and determined
1745 * the check condition was retryable.
1746 */
1747 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1748 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1749 return 1;
1750 else
1751 return 0;
4a27446f
MC
1752}
1753
1da177e4
LT
1754/**
1755 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1756 * @scmd: SCSI cmd to examine.
1757 *
1758 * Notes:
1759 * This is *only* called when we are examining the status after sending
1760 * out the actual data command. any commands that are queued for error
1761 * recovery (e.g. test_unit_ready) do *not* come through here.
1762 *
1763 * When this routine returns failed, it means the error handler thread
1764 * is woken. In cases where the error code indicates an error that
1765 * doesn't require the error handler read (i.e. we don't need to
1766 * abort/reset), this function should return SUCCESS.
dc8875e1 1767 */
1da177e4
LT
1768int scsi_decide_disposition(struct scsi_cmnd *scmd)
1769{
1770 int rtn;
1771
1772 /*
1773 * if the device is offline, then we clearly just pass the result back
1774 * up to the top level.
1775 */
1776 if (!scsi_device_online(scmd->device)) {
91921e01
HR
1777 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1778 "%s: device offline - report as SUCCESS\n", __func__));
1da177e4
LT
1779 return SUCCESS;
1780 }
1781
1782 /*
1783 * first check the host byte, to see if there is anything in there
1784 * that would indicate what we need to do.
1785 */
1786 switch (host_byte(scmd->result)) {
1787 case DID_PASSTHROUGH:
1788 /*
1789 * no matter what, pass this through to the upper layer.
1790 * nuke this special code so that it looks like we are saying
1791 * did_ok.
1792 */
1793 scmd->result &= 0xff00ffff;
1794 return SUCCESS;
1795 case DID_OK:
1796 /*
1797 * looks good. drop through, and check the next byte.
1798 */
1799 break;
e494f6a7
HR
1800 case DID_ABORT:
1801 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
8922a908 1802 set_host_byte(scmd, DID_TIME_OUT);
e494f6a7
HR
1803 return SUCCESS;
1804 }
1da177e4
LT
1805 case DID_NO_CONNECT:
1806 case DID_BAD_TARGET:
1da177e4
LT
1807 /*
1808 * note - this means that we just report the status back
1809 * to the top level driver, not that we actually think
1810 * that it indicates SUCCESS.
1811 */
1812 return SUCCESS;
1813 /*
1814 * when the low level driver returns did_soft_error,
0bf8c869 1815 * it is responsible for keeping an internal retry counter
1da177e4
LT
1816 * in order to avoid endless loops (db)
1817 *
1818 * actually this is a bug in this function here. we should
1819 * be mindful of the maximum number of retries specified
1820 * and not get stuck in a loop.
1821 */
1822 case DID_SOFT_ERROR:
1823 goto maybe_retry;
1824 case DID_IMM_RETRY:
1825 return NEEDS_RETRY;
1826
bf341919
JB
1827 case DID_REQUEUE:
1828 return ADD_TO_MLQUEUE;
a4dfaa6f
MC
1829 case DID_TRANSPORT_DISRUPTED:
1830 /*
1831 * LLD/transport was disrupted during processing of the IO.
1832 * The transport class is now blocked/blocking,
1833 * and the transport will decide what to do with the IO
939c2288
MC
1834 * based on its timers and recovery capablilities if
1835 * there are enough retries.
a4dfaa6f 1836 */
939c2288 1837 goto maybe_retry;
a4dfaa6f
MC
1838 case DID_TRANSPORT_FAILFAST:
1839 /*
1840 * The transport decided to failfast the IO (most likely
1841 * the fast io fail tmo fired), so send IO directly upwards.
1842 */
1843 return SUCCESS;
1da177e4
LT
1844 case DID_ERROR:
1845 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1846 status_byte(scmd->result) == RESERVATION_CONFLICT)
1847 /*
1848 * execute reservation conflict processing code
1849 * lower down
1850 */
1851 break;
1852 /* fallthrough */
1da177e4
LT
1853 case DID_BUS_BUSY:
1854 case DID_PARITY:
1855 goto maybe_retry;
1856 case DID_TIME_OUT:
1857 /*
1858 * when we scan the bus, we get timeout messages for
1859 * these commands if there is no device available.
1860 * other hosts report did_no_connect for the same thing.
1861 */
1862 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1863 scmd->cmnd[0] == INQUIRY)) {
1864 return SUCCESS;
1865 } else {
1866 return FAILED;
1867 }
1868 case DID_RESET:
1869 return SUCCESS;
1870 default:
1871 return FAILED;
1872 }
1873
1874 /*
1875 * next, check the message byte.
1876 */
1877 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1878 return FAILED;
1879
1880 /*
1881 * check the status byte to see if this indicates anything special.
1882 */
1883 switch (status_byte(scmd->result)) {
1884 case QUEUE_FULL:
42a6a918 1885 scsi_handle_queue_full(scmd->device);
1da177e4
LT
1886 /*
1887 * the case of trying to send too many commands to a
1888 * tagged queueing device.
1889 */
1890 case BUSY:
1891 /*
1892 * device can't talk to us at the moment. Should only
1893 * occur (SAM-3) when the task queue is empty, so will cause
1894 * the empty queue handling to trigger a stall in the
1895 * device.
1896 */
1897 return ADD_TO_MLQUEUE;
1898 case GOOD:
279afdfe
EM
1899 if (scmd->cmnd[0] == REPORT_LUNS)
1900 scmd->device->sdev_target->expecting_lun_change = 0;
4a84067d 1901 scsi_handle_queue_ramp_up(scmd->device);
1da177e4 1902 case COMMAND_TERMINATED:
1da177e4 1903 return SUCCESS;
a9b589d9
VB
1904 case TASK_ABORTED:
1905 goto maybe_retry;
1da177e4
LT
1906 case CHECK_CONDITION:
1907 rtn = scsi_check_sense(scmd);
1908 if (rtn == NEEDS_RETRY)
1909 goto maybe_retry;
1910 /* if rtn == FAILED, we have no sense information;
1911 * returning FAILED will wake the error handler thread
1912 * to collect the sense and redo the decide
1913 * disposition */
1914 return rtn;
1915 case CONDITION_GOOD:
1916 case INTERMEDIATE_GOOD:
1917 case INTERMEDIATE_C_GOOD:
1918 case ACA_ACTIVE:
1919 /*
1920 * who knows? FIXME(eric)
1921 */
1922 return SUCCESS;
1923
1924 case RESERVATION_CONFLICT:
9ccfc756
JB
1925 sdev_printk(KERN_INFO, scmd->device,
1926 "reservation conflict\n");
2082ebc4 1927 set_host_byte(scmd, DID_NEXUS_FAILURE);
1da177e4
LT
1928 return SUCCESS; /* causes immediate i/o error */
1929 default:
1930 return FAILED;
1931 }
1932 return FAILED;
1933
1934 maybe_retry:
1935
1936 /* we requeue for retry because the error was retryable, and
1937 * the request was not marked fast fail. Note that above,
1938 * even if the request is marked fast fail, we still requeue
1939 * for queue congestion conditions (QUEUE_FULL or BUSY) */
8884efab 1940 if ((++scmd->retries) <= scmd->allowed
4a27446f 1941 && !scsi_noretry_cmd(scmd)) {
1da177e4
LT
1942 return NEEDS_RETRY;
1943 } else {
1944 /*
1945 * no more retries - report this one back to upper level.
1946 */
1947 return SUCCESS;
1948 }
1949}
1950
f078727b
FT
1951static void eh_lock_door_done(struct request *req, int uptodate)
1952{
1953 __blk_put_request(req->q, req);
1954}
1955
1da177e4
LT
1956/**
1957 * scsi_eh_lock_door - Prevent medium removal for the specified device
1958 * @sdev: SCSI device to prevent medium removal
1959 *
1960 * Locking:
91bc31fb 1961 * We must be called from process context.
1da177e4
LT
1962 *
1963 * Notes:
1964 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1965 * head of the devices request queue, and continue.
dc8875e1 1966 */
1da177e4
LT
1967static void scsi_eh_lock_door(struct scsi_device *sdev)
1968{
f078727b 1969 struct request *req;
1da177e4 1970
91bc31fb
JB
1971 /*
1972 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1973 * request becomes available
1974 */
f078727b 1975 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
a492f075 1976 if (IS_ERR(req))
eb571eea 1977 return;
1da177e4 1978
f27b087b
JA
1979 blk_rq_set_block_pc(req);
1980
f078727b
FT
1981 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1982 req->cmd[1] = 0;
1983 req->cmd[2] = 0;
1984 req->cmd[3] = 0;
1985 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1986 req->cmd[5] = 0;
1da177e4 1987
f078727b
FT
1988 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1989
f078727b
FT
1990 req->cmd_flags |= REQ_QUIET;
1991 req->timeout = 10 * HZ;
1992 req->retries = 5;
1993
1994 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1995}
1da177e4
LT
1996
1997/**
1998 * scsi_restart_operations - restart io operations to the specified host.
1999 * @shost: Host we are restarting.
2000 *
2001 * Notes:
2002 * When we entered the error handler, we blocked all further i/o to
2003 * this device. we need to 'reverse' this process.
dc8875e1 2004 */
1da177e4
LT
2005static void scsi_restart_operations(struct Scsi_Host *shost)
2006{
2007 struct scsi_device *sdev;
939647ee 2008 unsigned long flags;
1da177e4
LT
2009
2010 /*
2011 * If the door was locked, we need to insert a door lock request
2012 * onto the head of the SCSI request queue for the device. There
2013 * is no point trying to lock the door of an off-line device.
2014 */
2015 shost_for_each_device(sdev, shost) {
48379270 2016 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
1da177e4 2017 scsi_eh_lock_door(sdev);
48379270
CH
2018 sdev->was_reset = 0;
2019 }
1da177e4
LT
2020 }
2021
2022 /*
2023 * next free up anything directly waiting upon the host. this
2024 * will be requests for character device operations, and also for
2025 * ioctls to queued block devices.
2026 */
b4562022 2027 SCSI_LOG_ERROR_RECOVERY(3,
91921e01 2028 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
1da177e4 2029
939647ee
JB
2030 spin_lock_irqsave(shost->host_lock, flags);
2031 if (scsi_host_set_state(shost, SHOST_RUNNING))
2032 if (scsi_host_set_state(shost, SHOST_CANCEL))
2033 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2034 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2035
2036 wake_up(&shost->host_wait);
2037
2038 /*
2039 * finally we need to re-initiate requests that may be pending. we will
2040 * have had everything blocked while error handling is taking place, and
2041 * now that error recovery is done, we will need to ensure that these
2042 * requests are started.
2043 */
2044 scsi_run_host_queues(shost);
57fc2e33
DW
2045
2046 /*
2047 * if eh is active and host_eh_scheduled is pending we need to re-run
2048 * recovery. we do this check after scsi_run_host_queues() to allow
2049 * everything pent up since the last eh run a chance to make forward
2050 * progress before we sync again. Either we'll immediately re-run
2051 * recovery or scsi_device_unbusy() will wake us again when these
2052 * pending commands complete.
2053 */
2054 spin_lock_irqsave(shost->host_lock, flags);
2055 if (shost->host_eh_scheduled)
2056 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2057 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2058 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2059}
2060
2061/**
2062 * scsi_eh_ready_devs - check device ready state and recover if not.
74cf298f
RD
2063 * @shost: host to be recovered.
2064 * @work_q: &list_head for pending commands.
eb44820c 2065 * @done_q: &list_head for processed commands.
dc8875e1 2066 */
dca84e46
DW
2067void scsi_eh_ready_devs(struct Scsi_Host *shost,
2068 struct list_head *work_q,
2069 struct list_head *done_q)
1da177e4
LT
2070{
2071 if (!scsi_eh_stu(shost, work_q, done_q))
2072 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
30bd7df8
MC
2073 if (!scsi_eh_target_reset(shost, work_q, done_q))
2074 if (!scsi_eh_bus_reset(shost, work_q, done_q))
91921e01 2075 if (!scsi_eh_host_reset(shost, work_q, done_q))
30bd7df8
MC
2076 scsi_eh_offline_sdevs(work_q,
2077 done_q);
1da177e4 2078}
dca84e46 2079EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1da177e4
LT
2080
2081/**
2082 * scsi_eh_flush_done_q - finish processed commands or retry them.
2083 * @done_q: list_head of processed commands.
dc8875e1 2084 */
041c5fc3 2085void scsi_eh_flush_done_q(struct list_head *done_q)
1da177e4 2086{
937abeaa 2087 struct scsi_cmnd *scmd, *next;
1da177e4 2088
937abeaa
CH
2089 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2090 list_del_init(&scmd->eh_entry);
1da177e4 2091 if (scsi_device_online(scmd->device) &&
4a27446f 2092 !scsi_noretry_cmd(scmd) &&
8884efab 2093 (++scmd->retries <= scmd->allowed)) {
91921e01
HR
2094 SCSI_LOG_ERROR_RECOVERY(3,
2095 scmd_printk(KERN_INFO, scmd,
2096 "%s: flush retry cmd: %p\n",
2097 current->comm, scmd));
1da177e4
LT
2098 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2099 } else {
793698ce
PM
2100 /*
2101 * If just we got sense for the device (called
2102 * scsi_eh_get_sense), scmd->result is already
2103 * set, do not set DRIVER_TIMEOUT.
2104 */
1da177e4
LT
2105 if (!scmd->result)
2106 scmd->result |= (DRIVER_TIMEOUT << 24);
91921e01
HR
2107 SCSI_LOG_ERROR_RECOVERY(3,
2108 scmd_printk(KERN_INFO, scmd,
2109 "%s: flush finish cmd: %p\n",
2110 current->comm, scmd));
1da177e4
LT
2111 scsi_finish_command(scmd);
2112 }
2113 }
2114}
041c5fc3 2115EXPORT_SYMBOL(scsi_eh_flush_done_q);
1da177e4
LT
2116
2117/**
2118 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2119 * @shost: Host to unjam.
2120 *
2121 * Notes:
2122 * When we come in here, we *know* that all commands on the bus have
2123 * either completed, failed or timed out. we also know that no further
2124 * commands are being sent to the host, so things are relatively quiet
2125 * and we have freedom to fiddle with things as we wish.
2126 *
2127 * This is only the *default* implementation. it is possible for
2128 * individual drivers to supply their own version of this function, and
2129 * if the maintainer wishes to do this, it is strongly suggested that
2130 * this function be taken as a template and modified. this function
2131 * was designed to correctly handle problems for about 95% of the
2132 * different cases out there, and it should always provide at least a
2133 * reasonable amount of error recovery.
2134 *
2135 * Any command marked 'failed' or 'timeout' must eventually have
2136 * scsi_finish_cmd() called for it. we do all of the retry stuff
2137 * here, so when we restart the host after we return it should have an
2138 * empty queue.
dc8875e1 2139 */
1da177e4
LT
2140static void scsi_unjam_host(struct Scsi_Host *shost)
2141{
2142 unsigned long flags;
2143 LIST_HEAD(eh_work_q);
2144 LIST_HEAD(eh_done_q);
2145
2146 spin_lock_irqsave(shost->host_lock, flags);
2147 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2148 spin_unlock_irqrestore(shost->host_lock, flags);
2149
2150 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2151
2152 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2153 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2154 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2155
b4562022 2156 spin_lock_irqsave(shost->host_lock, flags);
bb3b621a 2157 if (shost->eh_deadline != -1)
b4562022
HR
2158 shost->last_reset = 0;
2159 spin_unlock_irqrestore(shost->host_lock, flags);
1da177e4
LT
2160 scsi_eh_flush_done_q(&eh_done_q);
2161}
2162
2163/**
ad42eb1b 2164 * scsi_error_handler - SCSI error handler thread
1da177e4
LT
2165 * @data: Host for which we are running.
2166 *
2167 * Notes:
ad42eb1b
CH
2168 * This is the main error handling loop. This is run as a kernel thread
2169 * for every SCSI host and handles all error handling activity.
dc8875e1 2170 */
1da177e4
LT
2171int scsi_error_handler(void *data)
2172{
ad42eb1b 2173 struct Scsi_Host *shost = data;
1da177e4 2174
1da177e4 2175 /*
ad42eb1b
CH
2176 * We use TASK_INTERRUPTIBLE so that the thread is not
2177 * counted against the load average as a running process.
2178 * We never actually get interrupted because kthread_run
c03264a7 2179 * disables signal delivery for the created thread.
1da177e4 2180 */
3ed7a470 2181 while (!kthread_should_stop()) {
b9d5c6b7 2182 set_current_state(TASK_INTERRUPTIBLE);
ee7863bc 2183 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
74665016 2184 shost->host_failed != atomic_read(&shost->host_busy)) {
ad42eb1b 2185 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2186 shost_printk(KERN_INFO, shost,
2187 "scsi_eh_%d: sleeping\n",
2188 shost->host_no));
3ed7a470 2189 schedule();
3ed7a470
JB
2190 continue;
2191 }
1da177e4 2192
3ed7a470 2193 __set_current_state(TASK_RUNNING);
ad42eb1b 2194 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2195 shost_printk(KERN_INFO, shost,
2196 "scsi_eh_%d: waking up %d/%d/%d\n",
2197 shost->host_no, shost->host_eh_scheduled,
74665016
CH
2198 shost->host_failed,
2199 atomic_read(&shost->host_busy)));
1da177e4 2200
1da177e4
LT
2201 /*
2202 * We have a host that is failing for some reason. Figure out
2203 * what we need to do to get it up and online again (if we can).
2204 * If we fail, we end up taking the thing offline.
2205 */
ae0751ff 2206 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
bc4f2401 2207 SCSI_LOG_ERROR_RECOVERY(1,
a222b1e2
HR
2208 shost_printk(KERN_ERR, shost,
2209 "scsi_eh_%d: unable to autoresume\n",
2210 shost->host_no));
bc4f2401
AS
2211 continue;
2212 }
2213
9227c33d
CH
2214 if (shost->transportt->eh_strategy_handler)
2215 shost->transportt->eh_strategy_handler(shost);
1da177e4
LT
2216 else
2217 scsi_unjam_host(shost);
2218
1da177e4
LT
2219 /*
2220 * Note - if the above fails completely, the action is to take
2221 * individual devices offline and flush the queue of any
2222 * outstanding requests that may have been pending. When we
2223 * restart, we restart any I/O to any other devices on the bus
2224 * which are still online.
2225 */
2226 scsi_restart_operations(shost);
ae0751ff
LM
2227 if (!shost->eh_noresume)
2228 scsi_autopm_put_host(shost);
1da177e4 2229 }
461a0ffb
SR
2230 __set_current_state(TASK_RUNNING);
2231
ad42eb1b 2232 SCSI_LOG_ERROR_RECOVERY(1,
91921e01
HR
2233 shost_printk(KERN_INFO, shost,
2234 "Error handler scsi_eh_%d exiting\n",
2235 shost->host_no));
3ed7a470 2236 shost->ehandler = NULL;
1da177e4
LT
2237 return 0;
2238}
2239
2240/*
2241 * Function: scsi_report_bus_reset()
2242 *
2243 * Purpose: Utility function used by low-level drivers to report that
2244 * they have observed a bus reset on the bus being handled.
2245 *
2246 * Arguments: shost - Host in question
2247 * channel - channel on which reset was observed.
2248 *
2249 * Returns: Nothing
2250 *
2251 * Lock status: Host lock must be held.
2252 *
2253 * Notes: This only needs to be called if the reset is one which
2254 * originates from an unknown location. Resets originated
2255 * by the mid-level itself don't need to call this, but there
2256 * should be no harm.
2257 *
2258 * The main purpose of this is to make sure that a CHECK_CONDITION
2259 * is properly treated.
2260 */
2261void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2262{
2263 struct scsi_device *sdev;
2264
2265 __shost_for_each_device(sdev, shost) {
30bd7df8
MC
2266 if (channel == sdev_channel(sdev))
2267 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2268 }
2269}
2270EXPORT_SYMBOL(scsi_report_bus_reset);
2271
2272/*
2273 * Function: scsi_report_device_reset()
2274 *
2275 * Purpose: Utility function used by low-level drivers to report that
2276 * they have observed a device reset on the device being handled.
2277 *
2278 * Arguments: shost - Host in question
2279 * channel - channel on which reset was observed
2280 * target - target on which reset was observed
2281 *
2282 * Returns: Nothing
2283 *
2284 * Lock status: Host lock must be held
2285 *
2286 * Notes: This only needs to be called if the reset is one which
2287 * originates from an unknown location. Resets originated
2288 * by the mid-level itself don't need to call this, but there
2289 * should be no harm.
2290 *
2291 * The main purpose of this is to make sure that a CHECK_CONDITION
2292 * is properly treated.
2293 */
2294void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2295{
2296 struct scsi_device *sdev;
2297
2298 __shost_for_each_device(sdev, shost) {
422c0d61 2299 if (channel == sdev_channel(sdev) &&
30bd7df8
MC
2300 target == sdev_id(sdev))
2301 __scsi_report_device_reset(sdev, NULL);
1da177e4
LT
2302 }
2303}
2304EXPORT_SYMBOL(scsi_report_device_reset);
2305
2306static void
2307scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2308{
2309}
2310
176aa9d6
CH
2311/**
2312 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2313 * @dev: scsi_device to operate on
2314 * @arg: reset type (see sg.h)
1da177e4
LT
2315 */
2316int
176aa9d6 2317scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
1da177e4 2318{
bc4f2401 2319 struct scsi_cmnd *scmd;
d7a1bb0a 2320 struct Scsi_Host *shost = dev->host;
1da177e4 2321 struct request req;
d7a1bb0a 2322 unsigned long flags;
176aa9d6
CH
2323 int error = 0, rtn, val;
2324
2325 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2326 return -EACCES;
2327
2328 error = get_user(val, arg);
2329 if (error)
2330 return error;
1da177e4 2331
bc4f2401 2332 if (scsi_autopm_get_host(shost) < 0)
176aa9d6 2333 return -EIO;
bc4f2401 2334
176aa9d6 2335 error = -EIO;
bc4f2401 2336 scmd = scsi_get_command(dev, GFP_KERNEL);
0f121dd8 2337 if (!scmd)
95eeb5f5 2338 goto out_put_autopm_host;
95eeb5f5 2339
4f54eec8 2340 blk_rq_init(NULL, &req);
1da177e4 2341 scmd->request = &req;
b4edcbca 2342
64a87b24
BH
2343 scmd->cmnd = req.cmd;
2344
1da177e4 2345 scmd->scsi_done = scsi_reset_provider_done_command;
30b0c37b 2346 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1da177e4
LT
2347
2348 scmd->cmd_len = 0;
2349
2350 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1da177e4 2351
d7a1bb0a
JS
2352 spin_lock_irqsave(shost->host_lock, flags);
2353 shost->tmf_in_progress = 1;
2354 spin_unlock_irqrestore(shost->host_lock, flags);
2355
176aa9d6
CH
2356 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2357 case SG_SCSI_RESET_NOTHING:
2358 rtn = SUCCESS;
2359 break;
2360 case SG_SCSI_RESET_DEVICE:
1da177e4 2361 rtn = scsi_try_bus_device_reset(scmd);
176aa9d6 2362 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
1da177e4
LT
2363 break;
2364 /* FALLTHROUGH */
176aa9d6 2365 case SG_SCSI_RESET_TARGET:
30bd7df8 2366 rtn = scsi_try_target_reset(scmd);
176aa9d6 2367 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
30bd7df8
MC
2368 break;
2369 /* FALLTHROUGH */
176aa9d6 2370 case SG_SCSI_RESET_BUS:
1da177e4 2371 rtn = scsi_try_bus_reset(scmd);
176aa9d6 2372 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
1da177e4
LT
2373 break;
2374 /* FALLTHROUGH */
176aa9d6 2375 case SG_SCSI_RESET_HOST:
1da177e4 2376 rtn = scsi_try_host_reset(scmd);
176aa9d6
CH
2377 if (rtn == SUCCESS)
2378 break;
1da177e4 2379 default:
176aa9d6 2380 /* FALLTHROUGH */
1da177e4 2381 rtn = FAILED;
176aa9d6 2382 break;
1da177e4
LT
2383 }
2384
176aa9d6
CH
2385 error = (rtn == SUCCESS) ? 0 : -EIO;
2386
d7a1bb0a
JS
2387 spin_lock_irqsave(shost->host_lock, flags);
2388 shost->tmf_in_progress = 0;
2389 spin_unlock_irqrestore(shost->host_lock, flags);
2390
2391 /*
2392 * be sure to wake up anyone who was sleeping or had their queue
2393 * suspended while we performed the TMF.
2394 */
2395 SCSI_LOG_ERROR_RECOVERY(3,
91921e01
HR
2396 shost_printk(KERN_INFO, shost,
2397 "waking up host to restart after TMF\n"));
d7a1bb0a
JS
2398
2399 wake_up(&shost->host_wait);
d7a1bb0a
JS
2400 scsi_run_host_queues(shost);
2401
0f121dd8
CH
2402 scsi_put_command(scmd);
2403
04796336 2404out_put_autopm_host:
bc4f2401 2405 scsi_autopm_put_host(shost);
176aa9d6 2406 return error;
1da177e4 2407}
176aa9d6 2408EXPORT_SYMBOL(scsi_ioctl_reset);
1da177e4
LT
2409
2410/**
2411 * scsi_normalize_sense - normalize main elements from either fixed or
2412 * descriptor sense data format into a common format.
2413 *
2414 * @sense_buffer: byte array containing sense data returned by device
2415 * @sb_len: number of valid bytes in sense_buffer
2416 * @sshdr: pointer to instance of structure that common
2417 * elements are written to.
2418 *
2419 * Notes:
2420 * The "main elements" from sense data are: response_code, sense_key,
2421 * asc, ascq and additional_length (only for descriptor format).
2422 *
2423 * Typically this function can be called after a device has
2424 * responded to a SCSI command with the CHECK_CONDITION status.
2425 *
2426 * Return value:
4753cbc0 2427 * true if valid sense data information found, else false;
dc8875e1 2428 */
4753cbc0
HR
2429bool scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2430 struct scsi_sense_hdr *sshdr)
1da177e4 2431{
33aa687d 2432 if (!sense_buffer || !sb_len)
4753cbc0 2433 return false;
1da177e4
LT
2434
2435 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2436
2437 sshdr->response_code = (sense_buffer[0] & 0x7f);
33aa687d
JB
2438
2439 if (!scsi_sense_valid(sshdr))
4753cbc0 2440 return false;
33aa687d 2441
1da177e4
LT
2442 if (sshdr->response_code >= 0x72) {
2443 /*
2444 * descriptor format
2445 */
2446 if (sb_len > 1)
2447 sshdr->sense_key = (sense_buffer[1] & 0xf);
2448 if (sb_len > 2)
2449 sshdr->asc = sense_buffer[2];
2450 if (sb_len > 3)
2451 sshdr->ascq = sense_buffer[3];
2452 if (sb_len > 7)
2453 sshdr->additional_length = sense_buffer[7];
2454 } else {
0bf8c869 2455 /*
1da177e4
LT
2456 * fixed format
2457 */
2458 if (sb_len > 2)
2459 sshdr->sense_key = (sense_buffer[2] & 0xf);
2460 if (sb_len > 7) {
2461 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2462 sb_len : (sense_buffer[7] + 8);
2463 if (sb_len > 12)
2464 sshdr->asc = sense_buffer[12];
2465 if (sb_len > 13)
2466 sshdr->ascq = sense_buffer[13];
2467 }
2468 }
2469
4753cbc0 2470 return true;
1da177e4
LT
2471}
2472EXPORT_SYMBOL(scsi_normalize_sense);
2473
4753cbc0
HR
2474bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2475 struct scsi_sense_hdr *sshdr)
1da177e4
LT
2476{
2477 return scsi_normalize_sense(cmd->sense_buffer,
b80ca4f7 2478 SCSI_SENSE_BUFFERSIZE, sshdr);
1da177e4
LT
2479}
2480EXPORT_SYMBOL(scsi_command_normalize_sense);
2481
2482/**
eb44820c 2483 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1da177e4
LT
2484 * @sense_buffer: byte array of descriptor format sense data
2485 * @sb_len: number of valid bytes in sense_buffer
2486 * @desc_type: value of descriptor type to find
2487 * (e.g. 0 -> information)
2488 *
2489 * Notes:
2490 * only valid when sense data is in descriptor format
2491 *
2492 * Return value:
2493 * pointer to start of (first) descriptor if found else NULL
dc8875e1 2494 */
1da177e4
LT
2495const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2496 int desc_type)
2497{
2498 int add_sen_len, add_len, desc_len, k;
2499 const u8 * descp;
2500
2501 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2502 return NULL;
2503 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2504 return NULL;
2505 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2506 add_sen_len : (sb_len - 8);
2507 descp = &sense_buffer[8];
2508 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2509 descp += desc_len;
2510 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2511 desc_len = add_len + 2;
2512 if (descp[0] == desc_type)
2513 return descp;
2514 if (add_len < 0) // short descriptor ??
2515 break;
2516 }
2517 return NULL;
2518}
2519EXPORT_SYMBOL(scsi_sense_desc_find);
2520
2521/**
eb44820c 2522 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1da177e4
LT
2523 * @sense_buffer: byte array of sense data
2524 * @sb_len: number of valid bytes in sense_buffer
2525 * @info_out: pointer to 64 integer where 8 or 4 byte information
2526 * field will be placed if found.
2527 *
2528 * Return value:
2529 * 1 if information field found, 0 if not found.
dc8875e1 2530 */
1da177e4
LT
2531int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2532 u64 * info_out)
2533{
2534 int j;
2535 const u8 * ucp;
2536 u64 ull;
2537
2538 if (sb_len < 7)
2539 return 0;
2540 switch (sense_buffer[0] & 0x7f) {
2541 case 0x70:
2542 case 0x71:
2543 if (sense_buffer[0] & 0x80) {
2544 *info_out = (sense_buffer[3] << 24) +
2545 (sense_buffer[4] << 16) +
2546 (sense_buffer[5] << 8) + sense_buffer[6];
2547 return 1;
2548 } else
2549 return 0;
2550 case 0x72:
2551 case 0x73:
2552 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2553 0 /* info desc */);
2554 if (ucp && (0xa == ucp[1])) {
2555 ull = 0;
2556 for (j = 0; j < 8; ++j) {
2557 if (j > 0)
2558 ull <<= 8;
2559 ull |= ucp[4 + j];
2560 }
2561 *info_out = ull;
2562 return 1;
2563 } else
2564 return 0;
2565 default:
2566 return 0;
2567 }
2568}
2569EXPORT_SYMBOL(scsi_get_sense_info_fld);
3bc6a261
FT
2570
2571/**
2572 * scsi_build_sense_buffer - build sense data in a buffer
2573 * @desc: Sense format (non zero == descriptor format,
2574 * 0 == fixed format)
2575 * @buf: Where to build sense data
2576 * @key: Sense key
2577 * @asc: Additional sense code
2578 * @ascq: Additional sense code qualifier
2579 *
2580 **/
2581void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2582{
2583 if (desc) {
2584 buf[0] = 0x72; /* descriptor, current */
2585 buf[1] = key;
2586 buf[2] = asc;
2587 buf[3] = ascq;
2588 buf[7] = 0;
2589 } else {
2590 buf[0] = 0x70; /* fixed, current */
2591 buf[2] = key;
2592 buf[7] = 0xa;
2593 buf[12] = asc;
2594 buf[13] = ascq;
2595 }
2596}
2597EXPORT_SYMBOL(scsi_build_sense_buffer);