blob: 045a0868fc7b4540ac53676206ab1f5136355f13 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
3 *
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
8 *
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
11 *
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
13 * minor cleanups.
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
19#include <linux/timer.h>
20#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070022#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020023#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/interrupt.h>
25#include <linux/blkdev.h>
26#include <linux/delay.h>
27
28#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020029#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <scsi/scsi_dbg.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050033#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <scsi/scsi_host.h>
35#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#include "scsi_priv.h"
38#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070039#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define SENSE_TIMEOUT (10*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
46 */
47#define BUS_RESET_SETTLE_TIME (10)
48#define HOST_RESET_SETTLE_TIME (10)
49
50/* called with shost->host_lock held */
51void scsi_eh_wakeup(struct Scsi_Host *shost)
52{
53 if (shost->host_busy == shost->host_failed) {
James Bottomley3ed7a472005-09-19 09:50:04 -050054 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
57 }
58}
Tejun Heof8bbfc22006-05-19 21:07:05 +090059
60/**
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
63 *
64 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080065 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090066void scsi_schedule_eh(struct Scsi_Host *shost)
67{
68 unsigned long flags;
69
70 spin_lock_irqsave(shost->host_lock, flags);
71
72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
74 shost->host_eh_scheduled++;
75 scsi_eh_wakeup(shost);
76 }
77
78 spin_unlock_irqrestore(shost->host_lock, flags);
79}
80EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82/**
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
86 *
87 * Return value:
88 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080089 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070090int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
91{
92 struct Scsi_Host *shost = scmd->device->host;
93 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -050094 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
James Bottomley3ed7a472005-09-19 09:50:04 -050096 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 return 0;
98
99 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500100 if (scsi_host_set_state(shost, SHOST_RECOVERY))
101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
102 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
James Bottomley939647e2005-09-18 15:05:20 -0500104 ret = 1;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200105 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 shost->host_failed++;
108 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500109 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500111 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
114/**
115 * scsi_add_timer - Start timeout timer for a single scsi command.
116 * @scmd: scsi command that is about to start running.
117 * @timeout: amount of time to allow this command to run.
118 * @complete: timeout function to call if timer isn't canceled.
119 *
120 * Notes:
121 * This should be turned into an inline function. Each scsi command
122 * has its own timer, and as it is added to the queue, we set up the
123 * timer. When the command completes, we cancel the timer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800124 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
126 void (*complete)(struct scsi_cmnd *))
127{
128
129 /*
130 * If the clock was already running for this command, then
131 * first delete the timer. The timer handling code gets rather
132 * confused if we don't do this.
133 */
134 if (scmd->eh_timeout.function)
135 del_timer(&scmd->eh_timeout);
136
137 scmd->eh_timeout.data = (unsigned long)scmd;
138 scmd->eh_timeout.expires = jiffies + timeout;
139 scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
140
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142 " %d, (%p)\n", __FUNCTION__,
143 scmd, timeout, complete));
144
145 add_timer(&scmd->eh_timeout);
146}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148/**
149 * scsi_delete_timer - Delete/cancel timer for a given function.
150 * @scmd: Cmd that we are canceling timer for
151 *
152 * Notes:
153 * This should be turned into an inline function.
154 *
155 * Return value:
156 * 1 if we were able to detach the timer. 0 if we blew it, and the
157 * timer function has already started to run.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800158 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159int scsi_delete_timer(struct scsi_cmnd *scmd)
160{
161 int rtn;
162
163 rtn = del_timer(&scmd->eh_timeout);
164
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166 " rtn: %d\n", __FUNCTION__,
167 scmd, rtn));
168
169 scmd->eh_timeout.data = (unsigned long)NULL;
170 scmd->eh_timeout.function = NULL;
171
172 return rtn;
173}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175/**
176 * scsi_times_out - Timeout function for normal scsi commands.
177 * @scmd: Cmd that is timing out.
178 *
179 * Notes:
180 * We do not need to lock this. There is the potential for a race
181 * only in that the normal completion handling might run, but if the
182 * normal completion function determines that the timer has already
183 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800184 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185void scsi_times_out(struct scsi_cmnd *scmd)
186{
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500187 enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 scsi_log_completion(scmd, TIMEOUT_ERROR);
190
James Smartc829c392006-03-13 08:28:57 -0500191 if (scmd->device->host->transportt->eh_timed_out)
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500192 eh_timed_out = scmd->device->host->transportt->eh_timed_out;
193 else if (scmd->device->host->hostt->eh_timed_out)
194 eh_timed_out = scmd->device->host->hostt->eh_timed_out;
195 else
196 eh_timed_out = NULL;
197
198 if (eh_timed_out)
199 switch (eh_timed_out(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 case EH_HANDLED:
201 __scsi_done(scmd);
202 return;
203 case EH_RESET_TIMER:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 scsi_add_timer(scmd, scmd->timeout_per_command,
205 scsi_times_out);
206 return;
207 case EH_NOT_HANDLED:
208 break;
209 }
210
211 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
James Bottomley939647e2005-09-18 15:05:20 -0500212 scmd->result |= DID_TIME_OUT << 16;
213 __scsi_done(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215}
216
217/**
218 * scsi_block_when_processing_errors - Prevent cmds from being queued.
219 * @sdev: Device on which we are performing recovery.
220 *
221 * Description:
222 * We block until the host is out of error recovery, and then check to
223 * see whether the host or the device is offline.
224 *
225 * Return value:
226 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800227 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228int scsi_block_when_processing_errors(struct scsi_device *sdev)
229{
230 int online;
231
James Bottomley939647e2005-09-18 15:05:20 -0500232 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234 online = scsi_device_online(sdev);
235
236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__,
237 online));
238
239 return online;
240}
241EXPORT_SYMBOL(scsi_block_when_processing_errors);
242
243#ifdef CONFIG_SCSI_LOGGING
244/**
245 * scsi_eh_prt_fail_stats - Log info on failures.
246 * @shost: scsi host being recovered.
247 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800248 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
250 struct list_head *work_q)
251{
252 struct scsi_cmnd *scmd;
253 struct scsi_device *sdev;
254 int total_failures = 0;
255 int cmd_failed = 0;
256 int cmd_cancel = 0;
257 int devices_failed = 0;
258
259 shost_for_each_device(sdev, shost) {
260 list_for_each_entry(scmd, work_q, eh_entry) {
261 if (scmd->device == sdev) {
262 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200263 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 ++cmd_cancel;
265 else
266 ++cmd_failed;
267 }
268 }
269
270 if (cmd_cancel || cmd_failed) {
271 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500272 sdev_printk(KERN_INFO, sdev,
273 "%s: cmds failed: %d, cancel: %d\n",
274 __FUNCTION__, cmd_failed,
275 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 cmd_cancel = 0;
277 cmd_failed = 0;
278 ++devices_failed;
279 }
280 }
281
282 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
283 " devices require eh work\n",
284 total_failures, devices_failed));
285}
286#endif
287
288/**
289 * scsi_check_sense - Examine scsi cmd sense
290 * @scmd: Cmd to have sense checked.
291 *
292 * Return value:
293 * SUCCESS or FAILED or NEEDS_RETRY
294 *
295 * Notes:
296 * When a deferred error is detected the current command has
297 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800298 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299static int scsi_check_sense(struct scsi_cmnd *scmd)
300{
301 struct scsi_sense_hdr sshdr;
302
303 if (! scsi_command_normalize_sense(scmd, &sshdr))
304 return FAILED; /* no valid sense data */
305
306 if (scsi_sense_is_deferred(&sshdr))
307 return NEEDS_RETRY;
308
309 /*
310 * Previous logic looked for FILEMARK, EOM or ILI which are
311 * mainly associated with tapes and returned SUCCESS.
312 */
313 if (sshdr.response_code == 0x70) {
314 /* fixed format */
315 if (scmd->sense_buffer[2] & 0xe0)
316 return SUCCESS;
317 } else {
318 /*
319 * descriptor format: look for "stream commands sense data
320 * descriptor" (see SSC-3). Assume single sense data
321 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
322 */
323 if ((sshdr.additional_length > 3) &&
324 (scmd->sense_buffer[8] == 0x4) &&
325 (scmd->sense_buffer[11] & 0xe0))
326 return SUCCESS;
327 }
328
329 switch (sshdr.sense_key) {
330 case NO_SENSE:
331 return SUCCESS;
332 case RECOVERED_ERROR:
333 return /* soft_error */ SUCCESS;
334
335 case ABORTED_COMMAND:
336 return NEEDS_RETRY;
337 case NOT_READY:
338 case UNIT_ATTENTION:
339 /*
340 * if we are expecting a cc/ua because of a bus reset that we
341 * performed, treat this just as a retry. otherwise this is
342 * information that we should pass up to the upper-level driver
343 * so that we can deal with it there.
344 */
345 if (scmd->device->expecting_cc_ua) {
346 scmd->device->expecting_cc_ua = 0;
347 return NEEDS_RETRY;
348 }
349 /*
350 * if the device is in the process of becoming ready, we
351 * should retry.
352 */
353 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
354 return NEEDS_RETRY;
355 /*
356 * if the device is not started, we need to wake
357 * the error handler to start the motor
358 */
359 if (scmd->device->allow_restart &&
360 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
361 return FAILED;
362 return SUCCESS;
363
364 /* these three are not supported */
365 case COPY_ABORTED:
366 case VOLUME_OVERFLOW:
367 case MISCOMPARE:
368 return SUCCESS;
369
370 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800371 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
372 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
373 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
374 return SUCCESS;
375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return NEEDS_RETRY;
377
378 case HARDWARE_ERROR:
379 if (scmd->device->retry_hwerror)
380 return NEEDS_RETRY;
381 else
382 return SUCCESS;
383
384 case ILLEGAL_REQUEST:
385 case BLANK_CHECK:
386 case DATA_PROTECT:
387 default:
388 return SUCCESS;
389 }
390}
391
392/**
393 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
394 * @scmd: SCSI cmd to examine.
395 *
396 * Notes:
397 * This is *only* called when we are examining the status of commands
398 * queued during error recovery. the main difference here is that we
399 * don't allow for the possibility of retries here, and we are a lot
400 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800401 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
403{
404 /*
405 * first check the host byte, to see if there is anything in there
406 * that would indicate what we need to do.
407 */
408 if (host_byte(scmd->result) == DID_RESET) {
409 /*
410 * rats. we are already in the error handler, so we now
411 * get to try and figure out what to do next. if the sense
412 * is valid, we have a pretty good idea of what to do.
413 * if not, we mark it as FAILED.
414 */
415 return scsi_check_sense(scmd);
416 }
417 if (host_byte(scmd->result) != DID_OK)
418 return FAILED;
419
420 /*
421 * next, check the message byte.
422 */
423 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
424 return FAILED;
425
426 /*
427 * now, check the status byte to see if this indicates
428 * anything special.
429 */
430 switch (status_byte(scmd->result)) {
431 case GOOD:
432 case COMMAND_TERMINATED:
433 return SUCCESS;
434 case CHECK_CONDITION:
435 return scsi_check_sense(scmd);
436 case CONDITION_GOOD:
437 case INTERMEDIATE_GOOD:
438 case INTERMEDIATE_C_GOOD:
439 /*
440 * who knows? FIXME(eric)
441 */
442 return SUCCESS;
443 case BUSY:
444 case QUEUE_FULL:
445 case RESERVATION_CONFLICT:
446 default:
447 return FAILED;
448 }
449 return FAILED;
450}
451
452/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * scsi_eh_done - Completion function for error handling.
454 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800455 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456static void scsi_eh_done(struct scsi_cmnd *scmd)
457{
Michael Reed85631672005-12-07 21:46:27 -0600458 struct completion *eh_action;
459
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100460 SCSI_LOG_ERROR_RECOVERY(3,
461 printk("%s scmd: %p result: %x\n",
462 __FUNCTION__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600463
464 eh_action = scmd->device->host->eh_action;
465 if (eh_action)
466 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469/**
Brian King292148f2007-01-30 17:51:17 -0600470 * scsi_try_host_reset - ask host adapter to reset itself
471 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800472 */
Brian King292148f2007-01-30 17:51:17 -0600473static int scsi_try_host_reset(struct scsi_cmnd *scmd)
474{
475 unsigned long flags;
476 int rtn;
477
478 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
479 __FUNCTION__));
480
481 if (!scmd->device->host->hostt->eh_host_reset_handler)
482 return FAILED;
483
484 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
485
486 if (rtn == SUCCESS) {
487 if (!scmd->device->host->hostt->skip_settle_delay)
488 ssleep(HOST_RESET_SETTLE_TIME);
489 spin_lock_irqsave(scmd->device->host->host_lock, flags);
490 scsi_report_bus_reset(scmd->device->host,
491 scmd_channel(scmd));
492 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
493 }
494
495 return rtn;
496}
497
498/**
499 * scsi_try_bus_reset - ask host to perform a bus reset
500 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800501 */
Brian King292148f2007-01-30 17:51:17 -0600502static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
503{
504 unsigned long flags;
505 int rtn;
506
507 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
508 __FUNCTION__));
509
510 if (!scmd->device->host->hostt->eh_bus_reset_handler)
511 return FAILED;
512
513 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
514
515 if (rtn == SUCCESS) {
516 if (!scmd->device->host->hostt->skip_settle_delay)
517 ssleep(BUS_RESET_SETTLE_TIME);
518 spin_lock_irqsave(scmd->device->host->host_lock, flags);
519 scsi_report_bus_reset(scmd->device->host,
520 scmd_channel(scmd));
521 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
522 }
523
524 return rtn;
525}
526
527/**
528 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
529 * @scmd: SCSI cmd used to send BDR
530 *
531 * Notes:
532 * There is no timeout for this operation. if this operation is
533 * unreliable for a given host, then the host itself needs to put a
534 * timer on it, and set the host back to a consistent state prior to
535 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800536 */
Brian King292148f2007-01-30 17:51:17 -0600537static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
538{
539 int rtn;
540
541 if (!scmd->device->host->hostt->eh_device_reset_handler)
542 return FAILED;
543
544 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
545 if (rtn == SUCCESS) {
546 scmd->device->was_reset = 1;
547 scmd->device->expecting_cc_ua = 1;
548 }
549
550 return rtn;
551}
552
553static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
554{
555 if (!scmd->device->host->hostt->eh_abort_handler)
556 return FAILED;
557
558 return scmd->device->host->hostt->eh_abort_handler(scmd);
559}
560
561/**
562 * scsi_try_to_abort_cmd - Ask host to abort a running command.
563 * @scmd: SCSI cmd to abort from Lower Level.
564 *
565 * Notes:
566 * This function will not return until the user's completion function
567 * has been called. there is no timeout on this operation. if the
568 * author of the low-level driver wishes this operation to be timed,
569 * they can provide this facility themselves. helper functions in
570 * scsi_error.c can be supplied to make this easier to do.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800571 */
Brian King292148f2007-01-30 17:51:17 -0600572static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
573{
574 /*
575 * scsi_done was called just after the command timed out and before
576 * we had a chance to process it. (db)
577 */
578 if (scmd->serial_number == 0)
579 return SUCCESS;
580 return __scsi_try_to_abort_cmd(scmd);
581}
582
583static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
584{
585 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
586 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
587 if (scsi_try_bus_reset(scmd) != SUCCESS)
588 scsi_try_host_reset(scmd);
589}
590
591/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200592 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100593 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200594 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200595 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100596 * @cmnd_size: size in bytes of @cmnd
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200597 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100598 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200599 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200600 * as part of the error recovery process. If @sense_bytes is 0 the command
601 * sent must be one that does not transfer any data. If @sense_bytes != 0
602 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
603 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800604 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200605void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
606 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
f59114b2005-04-17 15:00:23 -0500608 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Christoph Hellwig631c2282006-07-08 20:42:15 +0200610 /*
611 * We need saved copies of a number of fields - this is because
612 * error handling may need to overwrite these with different values
613 * to run different commands, and once error handling is complete,
614 * we will need to restore these values prior to running the actual
615 * command.
616 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200617 ses->cmd_len = scmd->cmd_len;
618 memcpy(ses->cmnd, scmd->cmnd, sizeof(scmd->cmnd));
619 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200620 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200621 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200622 ses->result = scmd->result;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200623
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200624 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200625 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200626
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200627 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200628 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
629 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200630 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200631 scmd->sdb.length);
632 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200633 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200634 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200635 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
636 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200637 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200638 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200639 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200640 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200641 if (cmnd) {
642 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
643 memcpy(scmd->cmnd, cmnd, cmnd_size);
644 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
645 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200646 }
647
648 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200649
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200650 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500652 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Christoph Hellwig631c2282006-07-08 20:42:15 +0200654 /*
655 * Zero the sense buffer. The scsi spec mandates that any
656 * untransferred sense data should be interpreted as being zero.
657 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900658 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200659}
660EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200661
Boaz Harroshe1c23462007-10-08 16:36:45 +0200662/**
663 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
664 * @scmd: SCSI command structure to restore
665 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
666 *
667 * Undo any damage done by above scsi_prep_eh_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800668 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200669void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
670{
671 /*
672 * Restore original data
673 */
674 scmd->cmd_len = ses->cmd_len;
675 memcpy(scmd->cmnd, ses->cmnd, sizeof(scmd->cmnd));
676 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200677 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200678 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200679 scmd->result = ses->result;
680}
681EXPORT_SYMBOL(scsi_eh_restore_cmnd);
682
683/**
684 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
685 * @scmd: SCSI command structure to hijack
686 * @cmnd: CDB to send
687 * @cmnd_size: size in bytes of @cmnd
688 * @timeout: timeout for this request
689 * @sense_bytes: size of sense data to copy or 0
690 *
691 * This function is used to send a scsi command down to a target device
692 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
693 *
694 * Return value:
695 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800696 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200697static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
698 int cmnd_size, int timeout, unsigned sense_bytes)
699{
700 struct scsi_device *sdev = scmd->device;
701 struct Scsi_Host *shost = sdev->host;
702 DECLARE_COMPLETION_ONSTACK(done);
703 unsigned long timeleft;
704 unsigned long flags;
705 struct scsi_eh_save ses;
706 int rtn;
707
708 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100709 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
f59114b2005-04-17 15:00:23 -0500711 spin_lock_irqsave(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 scsi_log_send(scmd);
f59114b2005-04-17 15:00:23 -0500713 shost->hostt->queuecommand(scmd, scsi_eh_done);
714 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100716 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
f59114b2005-04-17 15:00:23 -0500718 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100720 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100722 SCSI_LOG_ERROR_RECOVERY(3,
723 printk("%s: scmd: %p, timeleft: %ld\n",
724 __FUNCTION__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100727 * If there is time left scsi_eh_done got called, and we will
728 * examine the actual status codes to see whether the command
729 * actually did complete normally, else tell the host to forget
730 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100732 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 rtn = scsi_eh_completed_normally(scmd);
734 SCSI_LOG_ERROR_RECOVERY(3,
735 printk("%s: scsi_eh_completed_normally %x\n",
736 __FUNCTION__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 switch (rtn) {
739 case SUCCESS:
740 case NEEDS_RETRY:
741 case FAILED:
742 break;
743 default:
744 rtn = FAILED;
745 break;
746 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100747 } else {
Brian King292148f2007-01-30 17:51:17 -0600748 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100749 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751
Boaz Harroshe1c23462007-10-08 16:36:45 +0200752 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return rtn;
754}
755
756/**
757 * scsi_request_sense - Request sense data from a particular target.
758 * @scmd: SCSI cmd for request sense.
759 *
760 * Notes:
761 * Some hosts automatically obtain this information, others require
762 * that we obtain it on our own. This function will *not* return until
763 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800764 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765static int scsi_request_sense(struct scsi_cmnd *scmd)
766{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200767 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
769
770/**
771 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
772 * @scmd: Original SCSI cmd that eh has finished.
773 * @done_q: Queue for processed commands.
774 *
775 * Notes:
776 * We don't want to use the normal command completion while we are are
777 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500778 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 * keep a list of pending commands for final completion, and once we
780 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800781 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900782void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
784 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200785 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 list_move_tail(&scmd->eh_entry, done_q);
787}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900788EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790/**
791 * scsi_eh_get_sense - Get device sense data.
792 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500793 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 *
795 * Description:
796 * See if we need to request sense information. if so, then get it
797 * now, so we have a better idea of what to do.
798 *
799 * Notes:
800 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500801 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 * it down before we request it.
803 *
804 * All drivers should request sense information internally these days,
805 * so for now all I have to say is tough noogies if you end up in here.
806 *
807 * XXX: Long term this code should go away, but that needs an audit of
808 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800809 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800810int scsi_eh_get_sense(struct list_head *work_q,
811 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200813 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 int rtn;
815
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200816 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200817 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 SCSI_SENSE_VALID(scmd))
819 continue;
820
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400821 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
822 "%s: requesting sense\n",
823 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 rtn = scsi_request_sense(scmd);
825 if (rtn != SUCCESS)
826 continue;
827
828 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
829 " result %x\n", scmd,
830 scmd->result));
831 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
832
833 rtn = scsi_decide_disposition(scmd);
834
835 /*
836 * if the result was normal, then just pass it along to the
837 * upper level.
838 */
839 if (rtn == SUCCESS)
840 /* we don't want this command reissued, just
841 * finished with the sense data, so set
842 * retries to the max allowed to ensure it
843 * won't get reissued */
844 scmd->retries = scmd->allowed;
845 else if (rtn != NEEDS_RETRY)
846 continue;
847
848 scsi_eh_finish_cmd(scmd, done_q);
849 }
850
851 return list_empty(work_q);
852}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800853EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500857 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 *
859 * Return value:
860 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800861 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862static int scsi_eh_tur(struct scsi_cmnd *scmd)
863{
864 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
865 int retry_cnt = 1, rtn;
866
867retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400868 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
871 __FUNCTION__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200872
873 switch (rtn) {
874 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (retry_cnt--)
876 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200877 /*FALLTHRU*/
878 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500879 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200880 default:
881 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883}
884
885/**
Rob Landleyeb448202007-11-03 13:30:39 -0500886 * scsi_eh_abort_cmds - abort pending commands.
887 * @work_q: &list_head for pending commands.
888 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 *
890 * Decription:
891 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -0500892 * running command. This only works out to be the case if we have one
893 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 * no sense to try and abort the command, since as far as the shost
895 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800896 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897static int scsi_eh_abort_cmds(struct list_head *work_q,
898 struct list_head *done_q)
899{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200900 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 int rtn;
902
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200903 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200904 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 continue;
906 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
907 "0x%p\n", current->comm,
908 scmd));
909 rtn = scsi_try_to_abort_cmd(scmd);
910 if (rtn == SUCCESS) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200911 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (!scsi_device_online(scmd->device) ||
913 !scsi_eh_tur(scmd)) {
914 scsi_eh_finish_cmd(scmd, done_q);
915 }
916
917 } else
918 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
919 " cmd failed:"
920 "0x%p\n",
921 current->comm,
922 scmd));
923 }
924
925 return list_empty(work_q);
926}
927
928/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500930 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 *
932 * Return value:
933 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800934 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
936{
937 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Christoph Hellwig631c2282006-07-08 20:42:15 +0200939 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -0500940 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Brian Kinged773e62007-03-29 15:25:52 -0500942 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
943 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
Brian Kinge555db92007-04-19 13:59:59 -0500944 scmd->device->timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -0500945
Christoph Hellwig631c2282006-07-08 20:42:15 +0200946 if (rtn == SUCCESS)
947 return 0;
948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 return 1;
951}
952
953 /**
954 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -0500955 * @shost: &scsi host being recovered.
956 * @work_q: &list_head for pending commands.
957 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 *
959 * Notes:
960 * If commands are failing due to not ready, initializing command required,
961 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800962 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963static int scsi_eh_stu(struct Scsi_Host *shost,
964 struct list_head *work_q,
965 struct list_head *done_q)
966{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200967 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 struct scsi_device *sdev;
969
970 shost_for_each_device(sdev, shost) {
971 stu_scmd = NULL;
972 list_for_each_entry(scmd, work_q, eh_entry)
973 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
974 scsi_check_sense(scmd) == FAILED ) {
975 stu_scmd = scmd;
976 break;
977 }
978
979 if (!stu_scmd)
980 continue;
981
982 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
983 " 0x%p\n", current->comm, sdev));
984
985 if (!scsi_eh_try_stu(stu_scmd)) {
986 if (!scsi_device_online(sdev) ||
987 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200988 list_for_each_entry_safe(scmd, next,
989 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (scmd->device == sdev)
991 scsi_eh_finish_cmd(scmd, done_q);
992 }
993 }
994 } else {
995 SCSI_LOG_ERROR_RECOVERY(3,
996 printk("%s: START_UNIT failed to sdev:"
997 " 0x%p\n", current->comm, sdev));
998 }
999 }
1000
1001 return list_empty(work_q);
1002}
1003
1004
1005/**
1006 * scsi_eh_bus_device_reset - send bdr if needed
1007 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001008 * @work_q: &list_head for pending commands.
1009 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 *
1011 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001012 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 * devices that are jammed or not - if we have multiple devices, it
1014 * makes no sense to try bus_device_reset - we really would need to try
1015 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001016 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1018 struct list_head *work_q,
1019 struct list_head *done_q)
1020{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001021 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 struct scsi_device *sdev;
1023 int rtn;
1024
1025 shost_for_each_device(sdev, shost) {
1026 bdr_scmd = NULL;
1027 list_for_each_entry(scmd, work_q, eh_entry)
1028 if (scmd->device == sdev) {
1029 bdr_scmd = scmd;
1030 break;
1031 }
1032
1033 if (!bdr_scmd)
1034 continue;
1035
1036 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1037 " 0x%p\n", current->comm,
1038 sdev));
1039 rtn = scsi_try_bus_device_reset(bdr_scmd);
1040 if (rtn == SUCCESS) {
1041 if (!scsi_device_online(sdev) ||
1042 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001043 list_for_each_entry_safe(scmd, next,
1044 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 if (scmd->device == sdev)
1046 scsi_eh_finish_cmd(scmd,
1047 done_q);
1048 }
1049 }
1050 } else {
1051 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1052 " failed sdev:"
1053 "0x%p\n",
1054 current->comm,
1055 sdev));
1056 }
1057 }
1058
1059 return list_empty(work_q);
1060}
1061
1062/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001064 * @shost: &scsi host being recovered.
1065 * @work_q: &list_head for pending commands.
1066 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001067 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1069 struct list_head *work_q,
1070 struct list_head *done_q)
1071{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001072 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 unsigned int channel;
1074 int rtn;
1075
1076 /*
1077 * we really want to loop over the various channels, and do this on
1078 * a channel by channel basis. we should also check to see if any
1079 * of the failed commands are on soft_reset devices, and if so, skip
1080 * the reset.
1081 */
1082
1083 for (channel = 0; channel <= shost->max_channel; channel++) {
1084 chan_scmd = NULL;
1085 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001086 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 chan_scmd = scmd;
1088 break;
1089 /*
1090 * FIXME add back in some support for
1091 * soft_reset devices.
1092 */
1093 }
1094 }
1095
1096 if (!chan_scmd)
1097 continue;
1098 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1099 " %d\n", current->comm,
1100 channel));
1101 rtn = scsi_try_bus_reset(chan_scmd);
1102 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001103 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001104 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (!scsi_device_online(scmd->device) ||
1106 !scsi_eh_tur(scmd))
1107 scsi_eh_finish_cmd(scmd,
1108 done_q);
1109 }
1110 } else {
1111 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1112 " failed chan: %d\n",
1113 current->comm,
1114 channel));
1115 }
1116 }
1117 return list_empty(work_q);
1118}
1119
1120/**
1121 * scsi_eh_host_reset - send a host reset
1122 * @work_q: list_head for processed commands.
1123 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001124 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125static int scsi_eh_host_reset(struct list_head *work_q,
1126 struct list_head *done_q)
1127{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001128 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
1131 if (!list_empty(work_q)) {
1132 scmd = list_entry(work_q->next,
1133 struct scsi_cmnd, eh_entry);
1134
1135 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1136 , current->comm));
1137
1138 rtn = scsi_try_host_reset(scmd);
1139 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001140 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 if (!scsi_device_online(scmd->device) ||
1142 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1143 !scsi_eh_tur(scmd))
1144 scsi_eh_finish_cmd(scmd, done_q);
1145 }
1146 } else {
1147 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1148 " failed\n",
1149 current->comm));
1150 }
1151 }
1152 return list_empty(work_q);
1153}
1154
1155/**
1156 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1157 * @work_q: list_head for processed commands.
1158 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001159 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160static void scsi_eh_offline_sdevs(struct list_head *work_q,
1161 struct list_head *done_q)
1162{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001163 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001165 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001166 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1167 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001169 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 /*
1171 * FIXME: Handle lost cmds.
1172 */
1173 }
1174 scsi_eh_finish_cmd(scmd, done_q);
1175 }
1176 return;
1177}
1178
1179/**
1180 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1181 * @scmd: SCSI cmd to examine.
1182 *
1183 * Notes:
1184 * This is *only* called when we are examining the status after sending
1185 * out the actual data command. any commands that are queued for error
1186 * recovery (e.g. test_unit_ready) do *not* come through here.
1187 *
1188 * When this routine returns failed, it means the error handler thread
1189 * is woken. In cases where the error code indicates an error that
1190 * doesn't require the error handler read (i.e. we don't need to
1191 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001192 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193int scsi_decide_disposition(struct scsi_cmnd *scmd)
1194{
1195 int rtn;
1196
1197 /*
1198 * if the device is offline, then we clearly just pass the result back
1199 * up to the top level.
1200 */
1201 if (!scsi_device_online(scmd->device)) {
1202 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1203 " as SUCCESS\n",
1204 __FUNCTION__));
1205 return SUCCESS;
1206 }
1207
1208 /*
1209 * first check the host byte, to see if there is anything in there
1210 * that would indicate what we need to do.
1211 */
1212 switch (host_byte(scmd->result)) {
1213 case DID_PASSTHROUGH:
1214 /*
1215 * no matter what, pass this through to the upper layer.
1216 * nuke this special code so that it looks like we are saying
1217 * did_ok.
1218 */
1219 scmd->result &= 0xff00ffff;
1220 return SUCCESS;
1221 case DID_OK:
1222 /*
1223 * looks good. drop through, and check the next byte.
1224 */
1225 break;
1226 case DID_NO_CONNECT:
1227 case DID_BAD_TARGET:
1228 case DID_ABORT:
1229 /*
1230 * note - this means that we just report the status back
1231 * to the top level driver, not that we actually think
1232 * that it indicates SUCCESS.
1233 */
1234 return SUCCESS;
1235 /*
1236 * when the low level driver returns did_soft_error,
1237 * it is responsible for keeping an internal retry counter
1238 * in order to avoid endless loops (db)
1239 *
1240 * actually this is a bug in this function here. we should
1241 * be mindful of the maximum number of retries specified
1242 * and not get stuck in a loop.
1243 */
1244 case DID_SOFT_ERROR:
1245 goto maybe_retry;
1246 case DID_IMM_RETRY:
1247 return NEEDS_RETRY;
1248
bf341912005-04-12 17:49:09 -05001249 case DID_REQUEUE:
1250 return ADD_TO_MLQUEUE;
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 case DID_ERROR:
1253 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1254 status_byte(scmd->result) == RESERVATION_CONFLICT)
1255 /*
1256 * execute reservation conflict processing code
1257 * lower down
1258 */
1259 break;
1260 /* fallthrough */
1261
1262 case DID_BUS_BUSY:
1263 case DID_PARITY:
1264 goto maybe_retry;
1265 case DID_TIME_OUT:
1266 /*
1267 * when we scan the bus, we get timeout messages for
1268 * these commands if there is no device available.
1269 * other hosts report did_no_connect for the same thing.
1270 */
1271 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1272 scmd->cmnd[0] == INQUIRY)) {
1273 return SUCCESS;
1274 } else {
1275 return FAILED;
1276 }
1277 case DID_RESET:
1278 return SUCCESS;
1279 default:
1280 return FAILED;
1281 }
1282
1283 /*
1284 * next, check the message byte.
1285 */
1286 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1287 return FAILED;
1288
1289 /*
1290 * check the status byte to see if this indicates anything special.
1291 */
1292 switch (status_byte(scmd->result)) {
1293 case QUEUE_FULL:
1294 /*
1295 * the case of trying to send too many commands to a
1296 * tagged queueing device.
1297 */
1298 case BUSY:
1299 /*
1300 * device can't talk to us at the moment. Should only
1301 * occur (SAM-3) when the task queue is empty, so will cause
1302 * the empty queue handling to trigger a stall in the
1303 * device.
1304 */
1305 return ADD_TO_MLQUEUE;
1306 case GOOD:
1307 case COMMAND_TERMINATED:
1308 case TASK_ABORTED:
1309 return SUCCESS;
1310 case CHECK_CONDITION:
1311 rtn = scsi_check_sense(scmd);
1312 if (rtn == NEEDS_RETRY)
1313 goto maybe_retry;
1314 /* if rtn == FAILED, we have no sense information;
1315 * returning FAILED will wake the error handler thread
1316 * to collect the sense and redo the decide
1317 * disposition */
1318 return rtn;
1319 case CONDITION_GOOD:
1320 case INTERMEDIATE_GOOD:
1321 case INTERMEDIATE_C_GOOD:
1322 case ACA_ACTIVE:
1323 /*
1324 * who knows? FIXME(eric)
1325 */
1326 return SUCCESS;
1327
1328 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001329 sdev_printk(KERN_INFO, scmd->device,
1330 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 return SUCCESS; /* causes immediate i/o error */
1332 default:
1333 return FAILED;
1334 }
1335 return FAILED;
1336
1337 maybe_retry:
1338
1339 /* we requeue for retry because the error was retryable, and
1340 * the request was not marked fast fail. Note that above,
1341 * even if the request is marked fast fail, we still requeue
1342 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001343 if ((++scmd->retries) <= scmd->allowed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 && !blk_noretry_request(scmd->request)) {
1345 return NEEDS_RETRY;
1346 } else {
1347 /*
1348 * no more retries - report this one back to upper level.
1349 */
1350 return SUCCESS;
1351 }
1352}
1353
1354/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 * scsi_eh_lock_door - Prevent medium removal for the specified device
1356 * @sdev: SCSI device to prevent medium removal
1357 *
1358 * Locking:
1359 * We must be called from process context; scsi_allocate_request()
1360 * may sleep.
1361 *
1362 * Notes:
1363 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1364 * head of the devices request queue, and continue.
1365 *
1366 * Bugs:
1367 * scsi_allocate_request() may sleep waiting for existing requests to
1368 * be processed. However, since we haven't kicked off any request
1369 * processing for this host, this may deadlock.
1370 *
1371 * If scsi_allocate_request() fails for what ever reason, we
1372 * completely forget to lock the door.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001373 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374static void scsi_eh_lock_door(struct scsi_device *sdev)
1375{
Mike Christie6e68af62005-11-11 05:30:27 -06001376 unsigned char cmnd[MAX_COMMAND_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Mike Christie6e68af62005-11-11 05:30:27 -06001378 cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1379 cmnd[1] = 0;
1380 cmnd[2] = 0;
1381 cmnd[3] = 0;
1382 cmnd[4] = SCSI_REMOVAL_PREVENT;
1383 cmnd[5] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
brking@us.ibm.combb1d1072006-01-23 15:03:22 -06001385 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
Mike Christie6e68af62005-11-11 05:30:27 -06001386 5, NULL, NULL, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
1389
1390/**
1391 * scsi_restart_operations - restart io operations to the specified host.
1392 * @shost: Host we are restarting.
1393 *
1394 * Notes:
1395 * When we entered the error handler, we blocked all further i/o to
1396 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001397 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398static void scsi_restart_operations(struct Scsi_Host *shost)
1399{
1400 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001401 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 /*
1404 * If the door was locked, we need to insert a door lock request
1405 * onto the head of the SCSI request queue for the device. There
1406 * is no point trying to lock the door of an off-line device.
1407 */
1408 shost_for_each_device(sdev, shost) {
1409 if (scsi_device_online(sdev) && sdev->locked)
1410 scsi_eh_lock_door(sdev);
1411 }
1412
1413 /*
1414 * next free up anything directly waiting upon the host. this
1415 * will be requests for character device operations, and also for
1416 * ioctls to queued block devices.
1417 */
1418 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1419 __FUNCTION__));
1420
James Bottomley939647e2005-09-18 15:05:20 -05001421 spin_lock_irqsave(shost->host_lock, flags);
1422 if (scsi_host_set_state(shost, SHOST_RUNNING))
1423 if (scsi_host_set_state(shost, SHOST_CANCEL))
1424 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1425 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427 wake_up(&shost->host_wait);
1428
1429 /*
1430 * finally we need to re-initiate requests that may be pending. we will
1431 * have had everything blocked while error handling is taking place, and
1432 * now that error recovery is done, we will need to ensure that these
1433 * requests are started.
1434 */
1435 scsi_run_host_queues(shost);
1436}
1437
1438/**
1439 * scsi_eh_ready_devs - check device ready state and recover if not.
1440 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001441 * @work_q: &list_head for pending commands.
1442 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001443 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001444void scsi_eh_ready_devs(struct Scsi_Host *shost,
1445 struct list_head *work_q,
1446 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447{
1448 if (!scsi_eh_stu(shost, work_q, done_q))
1449 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1450 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1451 if (!scsi_eh_host_reset(work_q, done_q))
1452 scsi_eh_offline_sdevs(work_q, done_q);
1453}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001454EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456/**
1457 * scsi_eh_flush_done_q - finish processed commands or retry them.
1458 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001459 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001460void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001462 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001464 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1465 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 if (scsi_device_online(scmd->device) &&
1467 !blk_noretry_request(scmd->request) &&
Brian King8884efa2006-02-24 17:10:04 -06001468 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1470 " retry cmd: %p\n",
1471 current->comm,
1472 scmd));
1473 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1474 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001475 /*
1476 * If just we got sense for the device (called
1477 * scsi_eh_get_sense), scmd->result is already
1478 * set, do not set DRIVER_TIMEOUT.
1479 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 if (!scmd->result)
1481 scmd->result |= (DRIVER_TIMEOUT << 24);
1482 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1483 " cmd: %p\n",
1484 current->comm, scmd));
1485 scsi_finish_command(scmd);
1486 }
1487 }
1488}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001489EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
1491/**
1492 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1493 * @shost: Host to unjam.
1494 *
1495 * Notes:
1496 * When we come in here, we *know* that all commands on the bus have
1497 * either completed, failed or timed out. we also know that no further
1498 * commands are being sent to the host, so things are relatively quiet
1499 * and we have freedom to fiddle with things as we wish.
1500 *
1501 * This is only the *default* implementation. it is possible for
1502 * individual drivers to supply their own version of this function, and
1503 * if the maintainer wishes to do this, it is strongly suggested that
1504 * this function be taken as a template and modified. this function
1505 * was designed to correctly handle problems for about 95% of the
1506 * different cases out there, and it should always provide at least a
1507 * reasonable amount of error recovery.
1508 *
1509 * Any command marked 'failed' or 'timeout' must eventually have
1510 * scsi_finish_cmd() called for it. we do all of the retry stuff
1511 * here, so when we restart the host after we return it should have an
1512 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001513 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514static void scsi_unjam_host(struct Scsi_Host *shost)
1515{
1516 unsigned long flags;
1517 LIST_HEAD(eh_work_q);
1518 LIST_HEAD(eh_done_q);
1519
1520 spin_lock_irqsave(shost->host_lock, flags);
1521 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1522 spin_unlock_irqrestore(shost->host_lock, flags);
1523
1524 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1525
1526 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1527 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1528 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1529
1530 scsi_eh_flush_done_q(&eh_done_q);
1531}
1532
1533/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001534 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 * @data: Host for which we are running.
1536 *
1537 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001538 * This is the main error handling loop. This is run as a kernel thread
1539 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001540 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541int scsi_error_handler(void *data)
1542{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001543 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001546 * We use TASK_INTERRUPTIBLE so that the thread is not
1547 * counted against the load average as a running process.
1548 * We never actually get interrupted because kthread_run
1549 * disables singal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001551 set_current_state(TASK_INTERRUPTIBLE);
1552 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001553 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001554 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001555 SCSI_LOG_ERROR_RECOVERY(1,
1556 printk("Error handler scsi_eh_%d sleeping\n",
1557 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001558 schedule();
1559 set_current_state(TASK_INTERRUPTIBLE);
1560 continue;
1561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
James Bottomley3ed7a472005-09-19 09:50:04 -05001563 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001564 SCSI_LOG_ERROR_RECOVERY(1,
1565 printk("Error handler scsi_eh_%d waking up\n",
1566 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 /*
1569 * We have a host that is failing for some reason. Figure out
1570 * what we need to do to get it up and online again (if we can).
1571 * If we fail, we end up taking the thing offline.
1572 */
Christoph Hellwig9227c332006-04-01 19:21:04 +02001573 if (shost->transportt->eh_strategy_handler)
1574 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 else
1576 scsi_unjam_host(shost);
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 /*
1579 * Note - if the above fails completely, the action is to take
1580 * individual devices offline and flush the queue of any
1581 * outstanding requests that may have been pending. When we
1582 * restart, we restart any I/O to any other devices on the bus
1583 * which are still online.
1584 */
1585 scsi_restart_operations(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001586 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001588 __set_current_state(TASK_RUNNING);
1589
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001590 SCSI_LOG_ERROR_RECOVERY(1,
1591 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001592 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 return 0;
1594}
1595
1596/*
1597 * Function: scsi_report_bus_reset()
1598 *
1599 * Purpose: Utility function used by low-level drivers to report that
1600 * they have observed a bus reset on the bus being handled.
1601 *
1602 * Arguments: shost - Host in question
1603 * channel - channel on which reset was observed.
1604 *
1605 * Returns: Nothing
1606 *
1607 * Lock status: Host lock must be held.
1608 *
1609 * Notes: This only needs to be called if the reset is one which
1610 * originates from an unknown location. Resets originated
1611 * by the mid-level itself don't need to call this, but there
1612 * should be no harm.
1613 *
1614 * The main purpose of this is to make sure that a CHECK_CONDITION
1615 * is properly treated.
1616 */
1617void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1618{
1619 struct scsi_device *sdev;
1620
1621 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001622 if (channel == sdev_channel(sdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 sdev->was_reset = 1;
1624 sdev->expecting_cc_ua = 1;
1625 }
1626 }
1627}
1628EXPORT_SYMBOL(scsi_report_bus_reset);
1629
1630/*
1631 * Function: scsi_report_device_reset()
1632 *
1633 * Purpose: Utility function used by low-level drivers to report that
1634 * they have observed a device reset on the device being handled.
1635 *
1636 * Arguments: shost - Host in question
1637 * channel - channel on which reset was observed
1638 * target - target on which reset was observed
1639 *
1640 * Returns: Nothing
1641 *
1642 * Lock status: Host lock must be held
1643 *
1644 * Notes: This only needs to be called if the reset is one which
1645 * originates from an unknown location. Resets originated
1646 * by the mid-level itself don't need to call this, but there
1647 * should be no harm.
1648 *
1649 * The main purpose of this is to make sure that a CHECK_CONDITION
1650 * is properly treated.
1651 */
1652void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1653{
1654 struct scsi_device *sdev;
1655
1656 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001657 if (channel == sdev_channel(sdev) &&
1658 target == sdev_id(sdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 sdev->was_reset = 1;
1660 sdev->expecting_cc_ua = 1;
1661 }
1662 }
1663}
1664EXPORT_SYMBOL(scsi_report_device_reset);
1665
1666static void
1667scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1668{
1669}
1670
1671/*
1672 * Function: scsi_reset_provider
1673 *
1674 * Purpose: Send requested reset to a bus or device at any phase.
1675 *
1676 * Arguments: device - device to send reset to
1677 * flag - reset type (see scsi.h)
1678 *
1679 * Returns: SUCCESS/FAILURE.
1680 *
1681 * Notes: This is used by the SCSI Generic driver to provide
1682 * Bus/Device reset capability.
1683 */
1684int
1685scsi_reset_provider(struct scsi_device *dev, int flag)
1686{
1687 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
James Smartd7a1bb02006-03-08 14:50:12 -05001688 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001690 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 int rtn;
1692
1693 scmd->request = &req;
1694 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001695
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1697
1698 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001699 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701 scmd->cmd_len = 0;
1702
1703 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 init_timer(&scmd->eh_timeout);
1706
James Smartd7a1bb02006-03-08 14:50:12 -05001707 spin_lock_irqsave(shost->host_lock, flags);
1708 shost->tmf_in_progress = 1;
1709 spin_unlock_irqrestore(shost->host_lock, flags);
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 switch (flag) {
1712 case SCSI_TRY_RESET_DEVICE:
1713 rtn = scsi_try_bus_device_reset(scmd);
1714 if (rtn == SUCCESS)
1715 break;
1716 /* FALLTHROUGH */
1717 case SCSI_TRY_RESET_BUS:
1718 rtn = scsi_try_bus_reset(scmd);
1719 if (rtn == SUCCESS)
1720 break;
1721 /* FALLTHROUGH */
1722 case SCSI_TRY_RESET_HOST:
1723 rtn = scsi_try_host_reset(scmd);
1724 break;
1725 default:
1726 rtn = FAILED;
1727 }
1728
James Smartd7a1bb02006-03-08 14:50:12 -05001729 spin_lock_irqsave(shost->host_lock, flags);
1730 shost->tmf_in_progress = 0;
1731 spin_unlock_irqrestore(shost->host_lock, flags);
1732
1733 /*
1734 * be sure to wake up anyone who was sleeping or had their queue
1735 * suspended while we performed the TMF.
1736 */
1737 SCSI_LOG_ERROR_RECOVERY(3,
1738 printk("%s: waking up host to restart after TMF\n",
1739 __FUNCTION__));
1740
1741 wake_up(&shost->host_wait);
1742
1743 scsi_run_host_queues(shost);
1744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 scsi_next_command(scmd);
1746 return rtn;
1747}
1748EXPORT_SYMBOL(scsi_reset_provider);
1749
1750/**
1751 * scsi_normalize_sense - normalize main elements from either fixed or
1752 * descriptor sense data format into a common format.
1753 *
1754 * @sense_buffer: byte array containing sense data returned by device
1755 * @sb_len: number of valid bytes in sense_buffer
1756 * @sshdr: pointer to instance of structure that common
1757 * elements are written to.
1758 *
1759 * Notes:
1760 * The "main elements" from sense data are: response_code, sense_key,
1761 * asc, ascq and additional_length (only for descriptor format).
1762 *
1763 * Typically this function can be called after a device has
1764 * responded to a SCSI command with the CHECK_CONDITION status.
1765 *
1766 * Return value:
1767 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001768 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1770 struct scsi_sense_hdr *sshdr)
1771{
James Bottomley33aa6872005-08-28 11:31:14 -05001772 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 return 0;
1774
1775 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1776
1777 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05001778
1779 if (!scsi_sense_valid(sshdr))
1780 return 0;
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 if (sshdr->response_code >= 0x72) {
1783 /*
1784 * descriptor format
1785 */
1786 if (sb_len > 1)
1787 sshdr->sense_key = (sense_buffer[1] & 0xf);
1788 if (sb_len > 2)
1789 sshdr->asc = sense_buffer[2];
1790 if (sb_len > 3)
1791 sshdr->ascq = sense_buffer[3];
1792 if (sb_len > 7)
1793 sshdr->additional_length = sense_buffer[7];
1794 } else {
1795 /*
1796 * fixed format
1797 */
1798 if (sb_len > 2)
1799 sshdr->sense_key = (sense_buffer[2] & 0xf);
1800 if (sb_len > 7) {
1801 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1802 sb_len : (sense_buffer[7] + 8);
1803 if (sb_len > 12)
1804 sshdr->asc = sense_buffer[12];
1805 if (sb_len > 13)
1806 sshdr->ascq = sense_buffer[13];
1807 }
1808 }
1809
1810 return 1;
1811}
1812EXPORT_SYMBOL(scsi_normalize_sense);
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1815 struct scsi_sense_hdr *sshdr)
1816{
1817 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09001818 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819}
1820EXPORT_SYMBOL(scsi_command_normalize_sense);
1821
1822/**
Rob Landleyeb448202007-11-03 13:30:39 -05001823 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 * @sense_buffer: byte array of descriptor format sense data
1825 * @sb_len: number of valid bytes in sense_buffer
1826 * @desc_type: value of descriptor type to find
1827 * (e.g. 0 -> information)
1828 *
1829 * Notes:
1830 * only valid when sense data is in descriptor format
1831 *
1832 * Return value:
1833 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001834 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1836 int desc_type)
1837{
1838 int add_sen_len, add_len, desc_len, k;
1839 const u8 * descp;
1840
1841 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1842 return NULL;
1843 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1844 return NULL;
1845 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1846 add_sen_len : (sb_len - 8);
1847 descp = &sense_buffer[8];
1848 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1849 descp += desc_len;
1850 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1851 desc_len = add_len + 2;
1852 if (descp[0] == desc_type)
1853 return descp;
1854 if (add_len < 0) // short descriptor ??
1855 break;
1856 }
1857 return NULL;
1858}
1859EXPORT_SYMBOL(scsi_sense_desc_find);
1860
1861/**
Rob Landleyeb448202007-11-03 13:30:39 -05001862 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 * @sense_buffer: byte array of sense data
1864 * @sb_len: number of valid bytes in sense_buffer
1865 * @info_out: pointer to 64 integer where 8 or 4 byte information
1866 * field will be placed if found.
1867 *
1868 * Return value:
1869 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001870 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1872 u64 * info_out)
1873{
1874 int j;
1875 const u8 * ucp;
1876 u64 ull;
1877
1878 if (sb_len < 7)
1879 return 0;
1880 switch (sense_buffer[0] & 0x7f) {
1881 case 0x70:
1882 case 0x71:
1883 if (sense_buffer[0] & 0x80) {
1884 *info_out = (sense_buffer[3] << 24) +
1885 (sense_buffer[4] << 16) +
1886 (sense_buffer[5] << 8) + sense_buffer[6];
1887 return 1;
1888 } else
1889 return 0;
1890 case 0x72:
1891 case 0x73:
1892 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1893 0 /* info desc */);
1894 if (ucp && (0xa == ucp[1])) {
1895 ull = 0;
1896 for (j = 0; j < 8; ++j) {
1897 if (j > 0)
1898 ull <<= 8;
1899 ull |= ucp[4 + j];
1900 }
1901 *info_out = ull;
1902 return 1;
1903 } else
1904 return 0;
1905 default:
1906 return 0;
1907 }
1908}
1909EXPORT_SYMBOL(scsi_get_sense_info_fld);