blob: 39ce3aba1dac5f459d668ee7ce4d15832918ec66 [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:"
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700142 " %d, (%p)\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 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,"
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700166 " rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 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
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 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",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700274 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500275 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{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700301 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 struct scsi_sense_hdr sshdr;
303
304 if (! scsi_command_normalize_sense(scmd, &sshdr))
305 return FAILED; /* no valid sense data */
306
307 if (scsi_sense_is_deferred(&sshdr))
308 return NEEDS_RETRY;
309
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700310 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
311 sdev->scsi_dh_data->scsi_dh->check_sense) {
312 int rc;
313
314 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
315 if (rc != SCSI_RETURN_NOT_HANDLED)
316 return rc;
317 /* handler does not care. Drop down to default handling */
318 }
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 /*
321 * Previous logic looked for FILEMARK, EOM or ILI which are
322 * mainly associated with tapes and returned SUCCESS.
323 */
324 if (sshdr.response_code == 0x70) {
325 /* fixed format */
326 if (scmd->sense_buffer[2] & 0xe0)
327 return SUCCESS;
328 } else {
329 /*
330 * descriptor format: look for "stream commands sense data
331 * descriptor" (see SSC-3). Assume single sense data
332 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
333 */
334 if ((sshdr.additional_length > 3) &&
335 (scmd->sense_buffer[8] == 0x4) &&
336 (scmd->sense_buffer[11] & 0xe0))
337 return SUCCESS;
338 }
339
340 switch (sshdr.sense_key) {
341 case NO_SENSE:
342 return SUCCESS;
343 case RECOVERED_ERROR:
344 return /* soft_error */ SUCCESS;
345
346 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400347 if (sshdr.asc == 0x10) /* DIF */
348 return SUCCESS;
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return NEEDS_RETRY;
351 case NOT_READY:
352 case UNIT_ATTENTION:
353 /*
354 * if we are expecting a cc/ua because of a bus reset that we
355 * performed, treat this just as a retry. otherwise this is
356 * information that we should pass up to the upper-level driver
357 * so that we can deal with it there.
358 */
359 if (scmd->device->expecting_cc_ua) {
360 scmd->device->expecting_cc_ua = 0;
361 return NEEDS_RETRY;
362 }
363 /*
364 * if the device is in the process of becoming ready, we
365 * should retry.
366 */
367 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
368 return NEEDS_RETRY;
369 /*
370 * if the device is not started, we need to wake
371 * the error handler to start the motor
372 */
373 if (scmd->device->allow_restart &&
374 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
375 return FAILED;
376 return SUCCESS;
377
378 /* these three are not supported */
379 case COPY_ABORTED:
380 case VOLUME_OVERFLOW:
381 case MISCOMPARE:
382 return SUCCESS;
383
384 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800385 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
386 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
387 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
388 return SUCCESS;
389 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 return NEEDS_RETRY;
391
392 case HARDWARE_ERROR:
393 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700394 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 else
396 return SUCCESS;
397
398 case ILLEGAL_REQUEST:
399 case BLANK_CHECK:
400 case DATA_PROTECT:
401 default:
402 return SUCCESS;
403 }
404}
405
406/**
407 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
408 * @scmd: SCSI cmd to examine.
409 *
410 * Notes:
411 * This is *only* called when we are examining the status of commands
412 * queued during error recovery. the main difference here is that we
413 * don't allow for the possibility of retries here, and we are a lot
414 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800415 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
417{
418 /*
419 * first check the host byte, to see if there is anything in there
420 * that would indicate what we need to do.
421 */
422 if (host_byte(scmd->result) == DID_RESET) {
423 /*
424 * rats. we are already in the error handler, so we now
425 * get to try and figure out what to do next. if the sense
426 * is valid, we have a pretty good idea of what to do.
427 * if not, we mark it as FAILED.
428 */
429 return scsi_check_sense(scmd);
430 }
431 if (host_byte(scmd->result) != DID_OK)
432 return FAILED;
433
434 /*
435 * next, check the message byte.
436 */
437 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
438 return FAILED;
439
440 /*
441 * now, check the status byte to see if this indicates
442 * anything special.
443 */
444 switch (status_byte(scmd->result)) {
445 case GOOD:
446 case COMMAND_TERMINATED:
447 return SUCCESS;
448 case CHECK_CONDITION:
449 return scsi_check_sense(scmd);
450 case CONDITION_GOOD:
451 case INTERMEDIATE_GOOD:
452 case INTERMEDIATE_C_GOOD:
453 /*
454 * who knows? FIXME(eric)
455 */
456 return SUCCESS;
457 case BUSY:
458 case QUEUE_FULL:
459 case RESERVATION_CONFLICT:
460 default:
461 return FAILED;
462 }
463 return FAILED;
464}
465
466/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 * scsi_eh_done - Completion function for error handling.
468 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800469 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470static void scsi_eh_done(struct scsi_cmnd *scmd)
471{
Michael Reed85631672005-12-07 21:46:27 -0600472 struct completion *eh_action;
473
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100474 SCSI_LOG_ERROR_RECOVERY(3,
475 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700476 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600477
478 eh_action = scmd->device->host->eh_action;
479 if (eh_action)
480 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481}
482
483/**
Brian King292148f2007-01-30 17:51:17 -0600484 * scsi_try_host_reset - ask host adapter to reset itself
485 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800486 */
Brian King292148f2007-01-30 17:51:17 -0600487static int scsi_try_host_reset(struct scsi_cmnd *scmd)
488{
489 unsigned long flags;
490 int rtn;
491
492 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700493 __func__));
Brian King292148f2007-01-30 17:51:17 -0600494
495 if (!scmd->device->host->hostt->eh_host_reset_handler)
496 return FAILED;
497
498 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
499
500 if (rtn == SUCCESS) {
501 if (!scmd->device->host->hostt->skip_settle_delay)
502 ssleep(HOST_RESET_SETTLE_TIME);
503 spin_lock_irqsave(scmd->device->host->host_lock, flags);
504 scsi_report_bus_reset(scmd->device->host,
505 scmd_channel(scmd));
506 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
507 }
508
509 return rtn;
510}
511
512/**
513 * scsi_try_bus_reset - ask host to perform a bus reset
514 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800515 */
Brian King292148f2007-01-30 17:51:17 -0600516static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
517{
518 unsigned long flags;
519 int rtn;
520
521 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700522 __func__));
Brian King292148f2007-01-30 17:51:17 -0600523
524 if (!scmd->device->host->hostt->eh_bus_reset_handler)
525 return FAILED;
526
527 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
528
529 if (rtn == SUCCESS) {
530 if (!scmd->device->host->hostt->skip_settle_delay)
531 ssleep(BUS_RESET_SETTLE_TIME);
532 spin_lock_irqsave(scmd->device->host->host_lock, flags);
533 scsi_report_bus_reset(scmd->device->host,
534 scmd_channel(scmd));
535 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
536 }
537
538 return rtn;
539}
540
Mike Christie30bd7df2008-02-29 18:25:19 -0600541static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
542{
543 sdev->was_reset = 1;
544 sdev->expecting_cc_ua = 1;
545}
546
547/**
548 * scsi_try_target_reset - Ask host to perform a target reset
549 * @scmd: SCSI cmd used to send a target reset
550 *
551 * Notes:
552 * There is no timeout for this operation. if this operation is
553 * unreliable for a given host, then the host itself needs to put a
554 * timer on it, and set the host back to a consistent state prior to
555 * returning.
556 */
557static int scsi_try_target_reset(struct scsi_cmnd *scmd)
558{
559 unsigned long flags;
560 int rtn;
561
562 if (!scmd->device->host->hostt->eh_target_reset_handler)
563 return FAILED;
564
565 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
566 if (rtn == SUCCESS) {
567 spin_lock_irqsave(scmd->device->host->host_lock, flags);
568 __starget_for_each_device(scsi_target(scmd->device), NULL,
569 __scsi_report_device_reset);
570 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
571 }
572
573 return rtn;
574}
575
Brian King292148f2007-01-30 17:51:17 -0600576/**
577 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
578 * @scmd: SCSI cmd used to send BDR
579 *
580 * Notes:
581 * There is no timeout for this operation. if this operation is
582 * unreliable for a given host, then the host itself needs to put a
583 * timer on it, and set the host back to a consistent state prior to
584 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800585 */
Brian King292148f2007-01-30 17:51:17 -0600586static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
587{
588 int rtn;
589
590 if (!scmd->device->host->hostt->eh_device_reset_handler)
591 return FAILED;
592
593 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600594 if (rtn == SUCCESS)
595 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600596 return rtn;
597}
598
599static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
600{
601 if (!scmd->device->host->hostt->eh_abort_handler)
602 return FAILED;
603
604 return scmd->device->host->hostt->eh_abort_handler(scmd);
605}
606
607/**
608 * scsi_try_to_abort_cmd - Ask host to abort a running command.
609 * @scmd: SCSI cmd to abort from Lower Level.
610 *
611 * Notes:
612 * This function will not return until the user's completion function
613 * has been called. there is no timeout on this operation. if the
614 * author of the low-level driver wishes this operation to be timed,
615 * they can provide this facility themselves. helper functions in
616 * scsi_error.c can be supplied to make this easier to do.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800617 */
Brian King292148f2007-01-30 17:51:17 -0600618static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
619{
620 /*
621 * scsi_done was called just after the command timed out and before
622 * we had a chance to process it. (db)
623 */
624 if (scmd->serial_number == 0)
625 return SUCCESS;
626 return __scsi_try_to_abort_cmd(scmd);
627}
628
629static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
630{
631 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
632 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600633 if (scsi_try_target_reset(scmd) != SUCCESS)
634 if (scsi_try_bus_reset(scmd) != SUCCESS)
635 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600636}
637
638/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200639 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100640 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200641 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200642 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300643 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200644 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100645 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200646 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200647 * as part of the error recovery process. If @sense_bytes is 0 the command
648 * sent must be one that does not transfer any data. If @sense_bytes != 0
649 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
650 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800651 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200652void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
653 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
f59114b2005-04-17 15:00:23 -0500655 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Christoph Hellwig631c2282006-07-08 20:42:15 +0200657 /*
658 * We need saved copies of a number of fields - this is because
659 * error handling may need to overwrite these with different values
660 * to run different commands, and once error handling is complete,
661 * we will need to restore these values prior to running the actual
662 * command.
663 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200664 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300665 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200666 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200667 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200668 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200669 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400670 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400671 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200672
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400673 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300674 scmd->cmnd = ses->eh_cmnd;
675 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200676 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200677 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200678
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200679 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200680 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
681 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200682 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200683 scmd->sdb.length);
684 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200685 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200686 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200687 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200688 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200689 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200690 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200691 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200692 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300693 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200694 memcpy(scmd->cmnd, cmnd, cmnd_size);
695 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
696 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200697 }
698
699 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200700
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200701 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500703 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Christoph Hellwig631c2282006-07-08 20:42:15 +0200705 /*
706 * Zero the sense buffer. The scsi spec mandates that any
707 * untransferred sense data should be interpreted as being zero.
708 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900709 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200710}
711EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200712
Boaz Harroshe1c23462007-10-08 16:36:45 +0200713/**
714 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
715 * @scmd: SCSI command structure to restore
716 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
717 *
718 * Undo any damage done by above scsi_prep_eh_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800719 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200720void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
721{
722 /*
723 * Restore original data
724 */
725 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300726 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200727 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200728 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200729 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200730 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400731 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400732 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200733}
734EXPORT_SYMBOL(scsi_eh_restore_cmnd);
735
736/**
737 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
738 * @scmd: SCSI command structure to hijack
739 * @cmnd: CDB to send
740 * @cmnd_size: size in bytes of @cmnd
741 * @timeout: timeout for this request
742 * @sense_bytes: size of sense data to copy or 0
743 *
744 * This function is used to send a scsi command down to a target device
745 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
746 *
747 * Return value:
748 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800749 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200750static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
751 int cmnd_size, int timeout, unsigned sense_bytes)
752{
753 struct scsi_device *sdev = scmd->device;
754 struct Scsi_Host *shost = sdev->host;
755 DECLARE_COMPLETION_ONSTACK(done);
756 unsigned long timeleft;
757 unsigned long flags;
758 struct scsi_eh_save ses;
759 int rtn;
760
761 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100762 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
f59114b2005-04-17 15:00:23 -0500764 spin_lock_irqsave(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 scsi_log_send(scmd);
f59114b2005-04-17 15:00:23 -0500766 shost->hostt->queuecommand(scmd, scsi_eh_done);
767 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100769 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
f59114b2005-04-17 15:00:23 -0500771 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100773 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100775 SCSI_LOG_ERROR_RECOVERY(3,
776 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700777 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100780 * If there is time left scsi_eh_done got called, and we will
781 * examine the actual status codes to see whether the command
782 * actually did complete normally, else tell the host to forget
783 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100785 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 rtn = scsi_eh_completed_normally(scmd);
787 SCSI_LOG_ERROR_RECOVERY(3,
788 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700789 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 switch (rtn) {
792 case SUCCESS:
793 case NEEDS_RETRY:
794 case FAILED:
795 break;
796 default:
797 rtn = FAILED;
798 break;
799 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100800 } else {
Brian King292148f2007-01-30 17:51:17 -0600801 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100802 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
804
Boaz Harroshe1c23462007-10-08 16:36:45 +0200805 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return rtn;
807}
808
809/**
810 * scsi_request_sense - Request sense data from a particular target.
811 * @scmd: SCSI cmd for request sense.
812 *
813 * Notes:
814 * Some hosts automatically obtain this information, others require
815 * that we obtain it on our own. This function will *not* return until
816 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800817 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818static int scsi_request_sense(struct scsi_cmnd *scmd)
819{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200820 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821}
822
823/**
824 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
825 * @scmd: Original SCSI cmd that eh has finished.
826 * @done_q: Queue for processed commands.
827 *
828 * Notes:
829 * We don't want to use the normal command completion while we are are
830 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500831 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 * keep a list of pending commands for final completion, and once we
833 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800834 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900835void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836{
837 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200838 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 list_move_tail(&scmd->eh_entry, done_q);
840}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900841EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843/**
844 * scsi_eh_get_sense - Get device sense data.
845 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500846 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 *
848 * Description:
849 * See if we need to request sense information. if so, then get it
850 * now, so we have a better idea of what to do.
851 *
852 * Notes:
853 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500854 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 * it down before we request it.
856 *
857 * All drivers should request sense information internally these days,
858 * so for now all I have to say is tough noogies if you end up in here.
859 *
860 * XXX: Long term this code should go away, but that needs an audit of
861 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800862 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800863int scsi_eh_get_sense(struct list_head *work_q,
864 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Christoph Hellwig937abea2005-06-19 13:43:56 +0200866 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 int rtn;
868
Christoph Hellwig937abea2005-06-19 13:43:56 +0200869 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200870 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 SCSI_SENSE_VALID(scmd))
872 continue;
873
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400874 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
875 "%s: requesting sense\n",
876 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 rtn = scsi_request_sense(scmd);
878 if (rtn != SUCCESS)
879 continue;
880
881 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
882 " result %x\n", scmd,
883 scmd->result));
884 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
885
886 rtn = scsi_decide_disposition(scmd);
887
888 /*
889 * if the result was normal, then just pass it along to the
890 * upper level.
891 */
892 if (rtn == SUCCESS)
893 /* we don't want this command reissued, just
894 * finished with the sense data, so set
895 * retries to the max allowed to ensure it
896 * won't get reissued */
897 scmd->retries = scmd->allowed;
898 else if (rtn != NEEDS_RETRY)
899 continue;
900
901 scsi_eh_finish_cmd(scmd, done_q);
902 }
903
904 return list_empty(work_q);
905}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800906EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500910 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 *
912 * Return value:
913 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800914 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915static int scsi_eh_tur(struct scsi_cmnd *scmd)
916{
917 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
918 int retry_cnt = 1, rtn;
919
920retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400921 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700924 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200925
926 switch (rtn) {
927 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (retry_cnt--)
929 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200930 /*FALLTHRU*/
931 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500932 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200933 default:
934 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
937
938/**
Rob Landleyeb448202007-11-03 13:30:39 -0500939 * scsi_eh_abort_cmds - abort pending commands.
940 * @work_q: &list_head for pending commands.
941 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 *
943 * Decription:
944 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -0500945 * running command. This only works out to be the case if we have one
946 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 * no sense to try and abort the command, since as far as the shost
948 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800949 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950static int scsi_eh_abort_cmds(struct list_head *work_q,
951 struct list_head *done_q)
952{
Christoph Hellwig937abea2005-06-19 13:43:56 +0200953 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 int rtn;
955
Christoph Hellwig937abea2005-06-19 13:43:56 +0200956 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200957 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 continue;
959 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
960 "0x%p\n", current->comm,
961 scmd));
962 rtn = scsi_try_to_abort_cmd(scmd);
963 if (rtn == SUCCESS) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200964 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 if (!scsi_device_online(scmd->device) ||
966 !scsi_eh_tur(scmd)) {
967 scsi_eh_finish_cmd(scmd, done_q);
968 }
969
970 } else
971 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
972 " cmd failed:"
973 "0x%p\n",
974 current->comm,
975 scmd));
976 }
977
978 return list_empty(work_q);
979}
980
981/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500983 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 *
985 * Return value:
986 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800987 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
989{
990 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Christoph Hellwig631c2282006-07-08 20:42:15 +0200992 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -0500993 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Brian Kinged773e62007-03-29 15:25:52 -0500995 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
996 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
Brian Kinge555db92007-04-19 13:59:59 -0500997 scmd->device->timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -0500998
Christoph Hellwig631c2282006-07-08 20:42:15 +0200999 if (rtn == SUCCESS)
1000 return 0;
1001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 return 1;
1004}
1005
1006 /**
1007 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001008 * @shost: &scsi host being recovered.
1009 * @work_q: &list_head for pending commands.
1010 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 *
1012 * Notes:
1013 * If commands are failing due to not ready, initializing command required,
1014 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001015 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016static int scsi_eh_stu(struct Scsi_Host *shost,
1017 struct list_head *work_q,
1018 struct list_head *done_q)
1019{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001020 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 struct scsi_device *sdev;
1022
1023 shost_for_each_device(sdev, shost) {
1024 stu_scmd = NULL;
1025 list_for_each_entry(scmd, work_q, eh_entry)
1026 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1027 scsi_check_sense(scmd) == FAILED ) {
1028 stu_scmd = scmd;
1029 break;
1030 }
1031
1032 if (!stu_scmd)
1033 continue;
1034
1035 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1036 " 0x%p\n", current->comm, sdev));
1037
1038 if (!scsi_eh_try_stu(stu_scmd)) {
1039 if (!scsi_device_online(sdev) ||
1040 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001041 list_for_each_entry_safe(scmd, next,
1042 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 if (scmd->device == sdev)
1044 scsi_eh_finish_cmd(scmd, done_q);
1045 }
1046 }
1047 } else {
1048 SCSI_LOG_ERROR_RECOVERY(3,
1049 printk("%s: START_UNIT failed to sdev:"
1050 " 0x%p\n", current->comm, sdev));
1051 }
1052 }
1053
1054 return list_empty(work_q);
1055}
1056
1057
1058/**
1059 * scsi_eh_bus_device_reset - send bdr if needed
1060 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001061 * @work_q: &list_head for pending commands.
1062 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 *
1064 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001065 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 * devices that are jammed or not - if we have multiple devices, it
1067 * makes no sense to try bus_device_reset - we really would need to try
1068 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001069 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1071 struct list_head *work_q,
1072 struct list_head *done_q)
1073{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001074 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 struct scsi_device *sdev;
1076 int rtn;
1077
1078 shost_for_each_device(sdev, shost) {
1079 bdr_scmd = NULL;
1080 list_for_each_entry(scmd, work_q, eh_entry)
1081 if (scmd->device == sdev) {
1082 bdr_scmd = scmd;
1083 break;
1084 }
1085
1086 if (!bdr_scmd)
1087 continue;
1088
1089 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1090 " 0x%p\n", current->comm,
1091 sdev));
1092 rtn = scsi_try_bus_device_reset(bdr_scmd);
1093 if (rtn == SUCCESS) {
1094 if (!scsi_device_online(sdev) ||
1095 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001096 list_for_each_entry_safe(scmd, next,
1097 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 if (scmd->device == sdev)
1099 scsi_eh_finish_cmd(scmd,
1100 done_q);
1101 }
1102 }
1103 } else {
1104 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1105 " failed sdev:"
1106 "0x%p\n",
1107 current->comm,
1108 sdev));
1109 }
1110 }
1111
1112 return list_empty(work_q);
1113}
1114
1115/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001116 * scsi_eh_target_reset - send target reset if needed
1117 * @shost: scsi host being recovered.
1118 * @work_q: &list_head for pending commands.
1119 * @done_q: &list_head for processed commands.
1120 *
1121 * Notes:
1122 * Try a target reset.
1123 */
1124static int scsi_eh_target_reset(struct Scsi_Host *shost,
1125 struct list_head *work_q,
1126 struct list_head *done_q)
1127{
1128 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1129 unsigned int id;
1130 int rtn;
1131
1132 for (id = 0; id <= shost->max_id; id++) {
1133 tgtr_scmd = NULL;
1134 list_for_each_entry(scmd, work_q, eh_entry) {
1135 if (id == scmd_id(scmd)) {
1136 tgtr_scmd = scmd;
1137 break;
1138 }
1139 }
1140 if (!tgtr_scmd)
1141 continue;
1142
1143 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1144 "to target %d\n",
1145 current->comm, id));
1146 rtn = scsi_try_target_reset(tgtr_scmd);
1147 if (rtn == SUCCESS) {
1148 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1149 if (id == scmd_id(scmd))
1150 if (!scsi_device_online(scmd->device) ||
1151 !scsi_eh_tur(tgtr_scmd))
1152 scsi_eh_finish_cmd(scmd,
1153 done_q);
1154 }
1155 } else
1156 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1157 " failed target: "
1158 "%d\n",
1159 current->comm, id));
1160 }
1161
1162 return list_empty(work_q);
1163}
1164
1165/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001167 * @shost: &scsi host being recovered.
1168 * @work_q: &list_head for pending commands.
1169 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001170 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1172 struct list_head *work_q,
1173 struct list_head *done_q)
1174{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001175 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 unsigned int channel;
1177 int rtn;
1178
1179 /*
1180 * we really want to loop over the various channels, and do this on
1181 * a channel by channel basis. we should also check to see if any
1182 * of the failed commands are on soft_reset devices, and if so, skip
1183 * the reset.
1184 */
1185
1186 for (channel = 0; channel <= shost->max_channel; channel++) {
1187 chan_scmd = NULL;
1188 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001189 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 chan_scmd = scmd;
1191 break;
1192 /*
1193 * FIXME add back in some support for
1194 * soft_reset devices.
1195 */
1196 }
1197 }
1198
1199 if (!chan_scmd)
1200 continue;
1201 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1202 " %d\n", current->comm,
1203 channel));
1204 rtn = scsi_try_bus_reset(chan_scmd);
1205 if (rtn == SUCCESS) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001206 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001207 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 if (!scsi_device_online(scmd->device) ||
1209 !scsi_eh_tur(scmd))
1210 scsi_eh_finish_cmd(scmd,
1211 done_q);
1212 }
1213 } else {
1214 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1215 " failed chan: %d\n",
1216 current->comm,
1217 channel));
1218 }
1219 }
1220 return list_empty(work_q);
1221}
1222
1223/**
1224 * scsi_eh_host_reset - send a host reset
1225 * @work_q: list_head for processed commands.
1226 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001227 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228static int scsi_eh_host_reset(struct list_head *work_q,
1229 struct list_head *done_q)
1230{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001231 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 if (!list_empty(work_q)) {
1235 scmd = list_entry(work_q->next,
1236 struct scsi_cmnd, eh_entry);
1237
1238 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1239 , current->comm));
1240
1241 rtn = scsi_try_host_reset(scmd);
1242 if (rtn == SUCCESS) {
Christoph Hellwig937abea2005-06-19 13:43:56 +02001243 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 if (!scsi_device_online(scmd->device) ||
1245 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1246 !scsi_eh_tur(scmd))
1247 scsi_eh_finish_cmd(scmd, done_q);
1248 }
1249 } else {
1250 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1251 " failed\n",
1252 current->comm));
1253 }
1254 }
1255 return list_empty(work_q);
1256}
1257
1258/**
1259 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1260 * @work_q: list_head for processed commands.
1261 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001262 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263static void scsi_eh_offline_sdevs(struct list_head *work_q,
1264 struct list_head *done_q)
1265{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001266 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Christoph Hellwig937abea2005-06-19 13:43:56 +02001268 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001269 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1270 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001272 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 /*
1274 * FIXME: Handle lost cmds.
1275 */
1276 }
1277 scsi_eh_finish_cmd(scmd, done_q);
1278 }
1279 return;
1280}
1281
1282/**
1283 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1284 * @scmd: SCSI cmd to examine.
1285 *
1286 * Notes:
1287 * This is *only* called when we are examining the status after sending
1288 * out the actual data command. any commands that are queued for error
1289 * recovery (e.g. test_unit_ready) do *not* come through here.
1290 *
1291 * When this routine returns failed, it means the error handler thread
1292 * is woken. In cases where the error code indicates an error that
1293 * doesn't require the error handler read (i.e. we don't need to
1294 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001295 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296int scsi_decide_disposition(struct scsi_cmnd *scmd)
1297{
1298 int rtn;
1299
1300 /*
1301 * if the device is offline, then we clearly just pass the result back
1302 * up to the top level.
1303 */
1304 if (!scsi_device_online(scmd->device)) {
1305 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1306 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001307 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 return SUCCESS;
1309 }
1310
1311 /*
1312 * first check the host byte, to see if there is anything in there
1313 * that would indicate what we need to do.
1314 */
1315 switch (host_byte(scmd->result)) {
1316 case DID_PASSTHROUGH:
1317 /*
1318 * no matter what, pass this through to the upper layer.
1319 * nuke this special code so that it looks like we are saying
1320 * did_ok.
1321 */
1322 scmd->result &= 0xff00ffff;
1323 return SUCCESS;
1324 case DID_OK:
1325 /*
1326 * looks good. drop through, and check the next byte.
1327 */
1328 break;
1329 case DID_NO_CONNECT:
1330 case DID_BAD_TARGET:
1331 case DID_ABORT:
1332 /*
1333 * note - this means that we just report the status back
1334 * to the top level driver, not that we actually think
1335 * that it indicates SUCCESS.
1336 */
1337 return SUCCESS;
1338 /*
1339 * when the low level driver returns did_soft_error,
1340 * it is responsible for keeping an internal retry counter
1341 * in order to avoid endless loops (db)
1342 *
1343 * actually this is a bug in this function here. we should
1344 * be mindful of the maximum number of retries specified
1345 * and not get stuck in a loop.
1346 */
1347 case DID_SOFT_ERROR:
1348 goto maybe_retry;
1349 case DID_IMM_RETRY:
1350 return NEEDS_RETRY;
1351
bf341912005-04-12 17:49:09 -05001352 case DID_REQUEUE:
1353 return ADD_TO_MLQUEUE;
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 case DID_ERROR:
1356 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1357 status_byte(scmd->result) == RESERVATION_CONFLICT)
1358 /*
1359 * execute reservation conflict processing code
1360 * lower down
1361 */
1362 break;
1363 /* fallthrough */
1364
1365 case DID_BUS_BUSY:
1366 case DID_PARITY:
1367 goto maybe_retry;
1368 case DID_TIME_OUT:
1369 /*
1370 * when we scan the bus, we get timeout messages for
1371 * these commands if there is no device available.
1372 * other hosts report did_no_connect for the same thing.
1373 */
1374 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1375 scmd->cmnd[0] == INQUIRY)) {
1376 return SUCCESS;
1377 } else {
1378 return FAILED;
1379 }
1380 case DID_RESET:
1381 return SUCCESS;
1382 default:
1383 return FAILED;
1384 }
1385
1386 /*
1387 * next, check the message byte.
1388 */
1389 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1390 return FAILED;
1391
1392 /*
1393 * check the status byte to see if this indicates anything special.
1394 */
1395 switch (status_byte(scmd->result)) {
1396 case QUEUE_FULL:
1397 /*
1398 * the case of trying to send too many commands to a
1399 * tagged queueing device.
1400 */
1401 case BUSY:
1402 /*
1403 * device can't talk to us at the moment. Should only
1404 * occur (SAM-3) when the task queue is empty, so will cause
1405 * the empty queue handling to trigger a stall in the
1406 * device.
1407 */
1408 return ADD_TO_MLQUEUE;
1409 case GOOD:
1410 case COMMAND_TERMINATED:
1411 case TASK_ABORTED:
1412 return SUCCESS;
1413 case CHECK_CONDITION:
1414 rtn = scsi_check_sense(scmd);
1415 if (rtn == NEEDS_RETRY)
1416 goto maybe_retry;
1417 /* if rtn == FAILED, we have no sense information;
1418 * returning FAILED will wake the error handler thread
1419 * to collect the sense and redo the decide
1420 * disposition */
1421 return rtn;
1422 case CONDITION_GOOD:
1423 case INTERMEDIATE_GOOD:
1424 case INTERMEDIATE_C_GOOD:
1425 case ACA_ACTIVE:
1426 /*
1427 * who knows? FIXME(eric)
1428 */
1429 return SUCCESS;
1430
1431 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001432 sdev_printk(KERN_INFO, scmd->device,
1433 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 return SUCCESS; /* causes immediate i/o error */
1435 default:
1436 return FAILED;
1437 }
1438 return FAILED;
1439
1440 maybe_retry:
1441
1442 /* we requeue for retry because the error was retryable, and
1443 * the request was not marked fast fail. Note that above,
1444 * even if the request is marked fast fail, we still requeue
1445 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001446 if ((++scmd->retries) <= scmd->allowed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 && !blk_noretry_request(scmd->request)) {
1448 return NEEDS_RETRY;
1449 } else {
1450 /*
1451 * no more retries - report this one back to upper level.
1452 */
1453 return SUCCESS;
1454 }
1455}
1456
1457/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 * scsi_eh_lock_door - Prevent medium removal for the specified device
1459 * @sdev: SCSI device to prevent medium removal
1460 *
1461 * Locking:
1462 * We must be called from process context; scsi_allocate_request()
1463 * may sleep.
1464 *
1465 * Notes:
1466 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1467 * head of the devices request queue, and continue.
1468 *
1469 * Bugs:
1470 * scsi_allocate_request() may sleep waiting for existing requests to
1471 * be processed. However, since we haven't kicked off any request
1472 * processing for this host, this may deadlock.
1473 *
1474 * If scsi_allocate_request() fails for what ever reason, we
1475 * completely forget to lock the door.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001476 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477static void scsi_eh_lock_door(struct scsi_device *sdev)
1478{
Mike Christie6e68af62005-11-11 05:30:27 -06001479 unsigned char cmnd[MAX_COMMAND_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Mike Christie6e68af62005-11-11 05:30:27 -06001481 cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1482 cmnd[1] = 0;
1483 cmnd[2] = 0;
1484 cmnd[3] = 0;
1485 cmnd[4] = SCSI_REMOVAL_PREVENT;
1486 cmnd[5] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
brking@us.ibm.combb1d1072006-01-23 15:03:22 -06001488 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
Mike Christie6e68af62005-11-11 05:30:27 -06001489 5, NULL, NULL, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490}
1491
1492
1493/**
1494 * scsi_restart_operations - restart io operations to the specified host.
1495 * @shost: Host we are restarting.
1496 *
1497 * Notes:
1498 * When we entered the error handler, we blocked all further i/o to
1499 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001500 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501static void scsi_restart_operations(struct Scsi_Host *shost)
1502{
1503 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001504 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 /*
1507 * If the door was locked, we need to insert a door lock request
1508 * onto the head of the SCSI request queue for the device. There
1509 * is no point trying to lock the door of an off-line device.
1510 */
1511 shost_for_each_device(sdev, shost) {
1512 if (scsi_device_online(sdev) && sdev->locked)
1513 scsi_eh_lock_door(sdev);
1514 }
1515
1516 /*
1517 * next free up anything directly waiting upon the host. this
1518 * will be requests for character device operations, and also for
1519 * ioctls to queued block devices.
1520 */
1521 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001522 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
James Bottomley939647e2005-09-18 15:05:20 -05001524 spin_lock_irqsave(shost->host_lock, flags);
1525 if (scsi_host_set_state(shost, SHOST_RUNNING))
1526 if (scsi_host_set_state(shost, SHOST_CANCEL))
1527 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1528 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
1530 wake_up(&shost->host_wait);
1531
1532 /*
1533 * finally we need to re-initiate requests that may be pending. we will
1534 * have had everything blocked while error handling is taking place, and
1535 * now that error recovery is done, we will need to ensure that these
1536 * requests are started.
1537 */
1538 scsi_run_host_queues(shost);
1539}
1540
1541/**
1542 * scsi_eh_ready_devs - check device ready state and recover if not.
1543 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001544 * @work_q: &list_head for pending commands.
1545 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001546 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001547void scsi_eh_ready_devs(struct Scsi_Host *shost,
1548 struct list_head *work_q,
1549 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 if (!scsi_eh_stu(shost, work_q, done_q))
1552 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001553 if (!scsi_eh_target_reset(shost, work_q, done_q))
1554 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1555 if (!scsi_eh_host_reset(work_q, done_q))
1556 scsi_eh_offline_sdevs(work_q,
1557 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001559EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
1561/**
1562 * scsi_eh_flush_done_q - finish processed commands or retry them.
1563 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001564 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001565void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
Christoph Hellwig937abea2005-06-19 13:43:56 +02001567 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Christoph Hellwig937abea2005-06-19 13:43:56 +02001569 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1570 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 if (scsi_device_online(scmd->device) &&
1572 !blk_noretry_request(scmd->request) &&
Brian King8884efa2006-02-24 17:10:04 -06001573 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1575 " retry cmd: %p\n",
1576 current->comm,
1577 scmd));
1578 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1579 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001580 /*
1581 * If just we got sense for the device (called
1582 * scsi_eh_get_sense), scmd->result is already
1583 * set, do not set DRIVER_TIMEOUT.
1584 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 if (!scmd->result)
1586 scmd->result |= (DRIVER_TIMEOUT << 24);
1587 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1588 " cmd: %p\n",
1589 current->comm, scmd));
1590 scsi_finish_command(scmd);
1591 }
1592 }
1593}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001594EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596/**
1597 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1598 * @shost: Host to unjam.
1599 *
1600 * Notes:
1601 * When we come in here, we *know* that all commands on the bus have
1602 * either completed, failed or timed out. we also know that no further
1603 * commands are being sent to the host, so things are relatively quiet
1604 * and we have freedom to fiddle with things as we wish.
1605 *
1606 * This is only the *default* implementation. it is possible for
1607 * individual drivers to supply their own version of this function, and
1608 * if the maintainer wishes to do this, it is strongly suggested that
1609 * this function be taken as a template and modified. this function
1610 * was designed to correctly handle problems for about 95% of the
1611 * different cases out there, and it should always provide at least a
1612 * reasonable amount of error recovery.
1613 *
1614 * Any command marked 'failed' or 'timeout' must eventually have
1615 * scsi_finish_cmd() called for it. we do all of the retry stuff
1616 * here, so when we restart the host after we return it should have an
1617 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001618 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619static void scsi_unjam_host(struct Scsi_Host *shost)
1620{
1621 unsigned long flags;
1622 LIST_HEAD(eh_work_q);
1623 LIST_HEAD(eh_done_q);
1624
1625 spin_lock_irqsave(shost->host_lock, flags);
1626 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1627 spin_unlock_irqrestore(shost->host_lock, flags);
1628
1629 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1630
1631 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1632 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1633 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1634
1635 scsi_eh_flush_done_q(&eh_done_q);
1636}
1637
1638/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001639 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 * @data: Host for which we are running.
1641 *
1642 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001643 * This is the main error handling loop. This is run as a kernel thread
1644 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001645 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646int scsi_error_handler(void *data)
1647{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001648 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001651 * We use TASK_INTERRUPTIBLE so that the thread is not
1652 * counted against the load average as a running process.
1653 * We never actually get interrupted because kthread_run
1654 * disables singal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001656 set_current_state(TASK_INTERRUPTIBLE);
1657 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001658 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001659 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001660 SCSI_LOG_ERROR_RECOVERY(1,
1661 printk("Error handler scsi_eh_%d sleeping\n",
1662 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001663 schedule();
1664 set_current_state(TASK_INTERRUPTIBLE);
1665 continue;
1666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
James Bottomley3ed7a472005-09-19 09:50:04 -05001668 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001669 SCSI_LOG_ERROR_RECOVERY(1,
1670 printk("Error handler scsi_eh_%d waking up\n",
1671 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 /*
1674 * We have a host that is failing for some reason. Figure out
1675 * what we need to do to get it up and online again (if we can).
1676 * If we fail, we end up taking the thing offline.
1677 */
Christoph Hellwig9227c332006-04-01 19:21:04 +02001678 if (shost->transportt->eh_strategy_handler)
1679 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 else
1681 scsi_unjam_host(shost);
1682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 /*
1684 * Note - if the above fails completely, the action is to take
1685 * individual devices offline and flush the queue of any
1686 * outstanding requests that may have been pending. When we
1687 * restart, we restart any I/O to any other devices on the bus
1688 * which are still online.
1689 */
1690 scsi_restart_operations(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001691 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001693 __set_current_state(TASK_RUNNING);
1694
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001695 SCSI_LOG_ERROR_RECOVERY(1,
1696 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001697 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 return 0;
1699}
1700
1701/*
1702 * Function: scsi_report_bus_reset()
1703 *
1704 * Purpose: Utility function used by low-level drivers to report that
1705 * they have observed a bus reset on the bus being handled.
1706 *
1707 * Arguments: shost - Host in question
1708 * channel - channel on which reset was observed.
1709 *
1710 * Returns: Nothing
1711 *
1712 * Lock status: Host lock must be held.
1713 *
1714 * Notes: This only needs to be called if the reset is one which
1715 * originates from an unknown location. Resets originated
1716 * by the mid-level itself don't need to call this, but there
1717 * should be no harm.
1718 *
1719 * The main purpose of this is to make sure that a CHECK_CONDITION
1720 * is properly treated.
1721 */
1722void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1723{
1724 struct scsi_device *sdev;
1725
1726 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001727 if (channel == sdev_channel(sdev))
1728 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
1730}
1731EXPORT_SYMBOL(scsi_report_bus_reset);
1732
1733/*
1734 * Function: scsi_report_device_reset()
1735 *
1736 * Purpose: Utility function used by low-level drivers to report that
1737 * they have observed a device reset on the device being handled.
1738 *
1739 * Arguments: shost - Host in question
1740 * channel - channel on which reset was observed
1741 * target - target on which reset was observed
1742 *
1743 * Returns: Nothing
1744 *
1745 * Lock status: Host lock must be held
1746 *
1747 * Notes: This only needs to be called if the reset is one which
1748 * originates from an unknown location. Resets originated
1749 * by the mid-level itself don't need to call this, but there
1750 * should be no harm.
1751 *
1752 * The main purpose of this is to make sure that a CHECK_CONDITION
1753 * is properly treated.
1754 */
1755void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1756{
1757 struct scsi_device *sdev;
1758
1759 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001760 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001761 target == sdev_id(sdev))
1762 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
1764}
1765EXPORT_SYMBOL(scsi_report_device_reset);
1766
1767static void
1768scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1769{
1770}
1771
1772/*
1773 * Function: scsi_reset_provider
1774 *
1775 * Purpose: Send requested reset to a bus or device at any phase.
1776 *
1777 * Arguments: device - device to send reset to
1778 * flag - reset type (see scsi.h)
1779 *
1780 * Returns: SUCCESS/FAILURE.
1781 *
1782 * Notes: This is used by the SCSI Generic driver to provide
1783 * Bus/Device reset capability.
1784 */
1785int
1786scsi_reset_provider(struct scsi_device *dev, int flag)
1787{
1788 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
James Smartd7a1bb02006-03-08 14:50:12 -05001789 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001791 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 int rtn;
1793
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001794 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 scmd->request = &req;
1796 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001797
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001798 scmd->cmnd = req.cmd;
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001801 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803 scmd->cmd_len = 0;
1804
1805 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 init_timer(&scmd->eh_timeout);
1808
James Smartd7a1bb02006-03-08 14:50:12 -05001809 spin_lock_irqsave(shost->host_lock, flags);
1810 shost->tmf_in_progress = 1;
1811 spin_unlock_irqrestore(shost->host_lock, flags);
1812
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 switch (flag) {
1814 case SCSI_TRY_RESET_DEVICE:
1815 rtn = scsi_try_bus_device_reset(scmd);
1816 if (rtn == SUCCESS)
1817 break;
1818 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001819 case SCSI_TRY_RESET_TARGET:
1820 rtn = scsi_try_target_reset(scmd);
1821 if (rtn == SUCCESS)
1822 break;
1823 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 case SCSI_TRY_RESET_BUS:
1825 rtn = scsi_try_bus_reset(scmd);
1826 if (rtn == SUCCESS)
1827 break;
1828 /* FALLTHROUGH */
1829 case SCSI_TRY_RESET_HOST:
1830 rtn = scsi_try_host_reset(scmd);
1831 break;
1832 default:
1833 rtn = FAILED;
1834 }
1835
James Smartd7a1bb02006-03-08 14:50:12 -05001836 spin_lock_irqsave(shost->host_lock, flags);
1837 shost->tmf_in_progress = 0;
1838 spin_unlock_irqrestore(shost->host_lock, flags);
1839
1840 /*
1841 * be sure to wake up anyone who was sleeping or had their queue
1842 * suspended while we performed the TMF.
1843 */
1844 SCSI_LOG_ERROR_RECOVERY(3,
1845 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001846 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001847
1848 wake_up(&shost->host_wait);
1849
1850 scsi_run_host_queues(shost);
1851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 scsi_next_command(scmd);
1853 return rtn;
1854}
1855EXPORT_SYMBOL(scsi_reset_provider);
1856
1857/**
1858 * scsi_normalize_sense - normalize main elements from either fixed or
1859 * descriptor sense data format into a common format.
1860 *
1861 * @sense_buffer: byte array containing sense data returned by device
1862 * @sb_len: number of valid bytes in sense_buffer
1863 * @sshdr: pointer to instance of structure that common
1864 * elements are written to.
1865 *
1866 * Notes:
1867 * The "main elements" from sense data are: response_code, sense_key,
1868 * asc, ascq and additional_length (only for descriptor format).
1869 *
1870 * Typically this function can be called after a device has
1871 * responded to a SCSI command with the CHECK_CONDITION status.
1872 *
1873 * Return value:
1874 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001875 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1877 struct scsi_sense_hdr *sshdr)
1878{
James Bottomley33aa6872005-08-28 11:31:14 -05001879 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 return 0;
1881
1882 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1883
1884 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05001885
1886 if (!scsi_sense_valid(sshdr))
1887 return 0;
1888
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 if (sshdr->response_code >= 0x72) {
1890 /*
1891 * descriptor format
1892 */
1893 if (sb_len > 1)
1894 sshdr->sense_key = (sense_buffer[1] & 0xf);
1895 if (sb_len > 2)
1896 sshdr->asc = sense_buffer[2];
1897 if (sb_len > 3)
1898 sshdr->ascq = sense_buffer[3];
1899 if (sb_len > 7)
1900 sshdr->additional_length = sense_buffer[7];
1901 } else {
1902 /*
1903 * fixed format
1904 */
1905 if (sb_len > 2)
1906 sshdr->sense_key = (sense_buffer[2] & 0xf);
1907 if (sb_len > 7) {
1908 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1909 sb_len : (sense_buffer[7] + 8);
1910 if (sb_len > 12)
1911 sshdr->asc = sense_buffer[12];
1912 if (sb_len > 13)
1913 sshdr->ascq = sense_buffer[13];
1914 }
1915 }
1916
1917 return 1;
1918}
1919EXPORT_SYMBOL(scsi_normalize_sense);
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1922 struct scsi_sense_hdr *sshdr)
1923{
1924 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09001925 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926}
1927EXPORT_SYMBOL(scsi_command_normalize_sense);
1928
1929/**
Rob Landleyeb448202007-11-03 13:30:39 -05001930 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 * @sense_buffer: byte array of descriptor format sense data
1932 * @sb_len: number of valid bytes in sense_buffer
1933 * @desc_type: value of descriptor type to find
1934 * (e.g. 0 -> information)
1935 *
1936 * Notes:
1937 * only valid when sense data is in descriptor format
1938 *
1939 * Return value:
1940 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001941 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1943 int desc_type)
1944{
1945 int add_sen_len, add_len, desc_len, k;
1946 const u8 * descp;
1947
1948 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1949 return NULL;
1950 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1951 return NULL;
1952 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1953 add_sen_len : (sb_len - 8);
1954 descp = &sense_buffer[8];
1955 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1956 descp += desc_len;
1957 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1958 desc_len = add_len + 2;
1959 if (descp[0] == desc_type)
1960 return descp;
1961 if (add_len < 0) // short descriptor ??
1962 break;
1963 }
1964 return NULL;
1965}
1966EXPORT_SYMBOL(scsi_sense_desc_find);
1967
1968/**
Rob Landleyeb448202007-11-03 13:30:39 -05001969 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 * @sense_buffer: byte array of sense data
1971 * @sb_len: number of valid bytes in sense_buffer
1972 * @info_out: pointer to 64 integer where 8 or 4 byte information
1973 * field will be placed if found.
1974 *
1975 * Return value:
1976 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001977 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1979 u64 * info_out)
1980{
1981 int j;
1982 const u8 * ucp;
1983 u64 ull;
1984
1985 if (sb_len < 7)
1986 return 0;
1987 switch (sense_buffer[0] & 0x7f) {
1988 case 0x70:
1989 case 0x71:
1990 if (sense_buffer[0] & 0x80) {
1991 *info_out = (sense_buffer[3] << 24) +
1992 (sense_buffer[4] << 16) +
1993 (sense_buffer[5] << 8) + sense_buffer[6];
1994 return 1;
1995 } else
1996 return 0;
1997 case 0x72:
1998 case 0x73:
1999 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2000 0 /* info desc */);
2001 if (ucp && (0xa == ucp[1])) {
2002 ull = 0;
2003 for (j = 0; j < 8; ++j) {
2004 if (j > 0)
2005 ull <<= 8;
2006 ull |= ucp[4 + j];
2007 }
2008 *info_out = ull;
2009 return 1;
2010 } else
2011 return 0;
2012 default:
2013 return 0;
2014 }
2015}
2016EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002017
2018/**
2019 * scsi_build_sense_buffer - build sense data in a buffer
2020 * @desc: Sense format (non zero == descriptor format,
2021 * 0 == fixed format)
2022 * @buf: Where to build sense data
2023 * @key: Sense key
2024 * @asc: Additional sense code
2025 * @ascq: Additional sense code qualifier
2026 *
2027 **/
2028void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2029{
2030 if (desc) {
2031 buf[0] = 0x72; /* descriptor, current */
2032 buf[1] = key;
2033 buf[2] = asc;
2034 buf[3] = ascq;
2035 buf[7] = 0;
2036 } else {
2037 buf[0] = 0x70; /* fixed, current */
2038 buf[2] = key;
2039 buf[7] = 0xa;
2040 buf[12] = asc;
2041 buf[13] = ascq;
2042 }
2043}
2044EXPORT_SYMBOL(scsi_build_sense_buffer);